Proceeding Book of International Tax Conference 2022 2.0 (2)_compressed (1)

Proceeding Book of
International Tax Conference

"G20 Indonesian Presidency:
Achieving a Stronger and More
Sustainable World Recovery through
International and
National Tax Reform"
7th December 2022

Ministry of Finance Republic of Indonesia
Jakarta, Indonesia
Proceeding Book of
International Tax
Conference
"G20 Indonesian Presidency: Achieving a Stronger and
More Sustainable World Recovery through International
and National Tax Reform"
Copyright © 2023 Directorate General of Tax
Ministry of Finance Republic of Indonesia, Jakarta
Committee
Advisor Material Division

Director of Tax Dissemination, Services, Head of Section of Dissemination
and Public Relations Materials
Member: Dewi Herawati, Adhi Triana
Team Leader Ahmad Nugraha, Lidya Lexy Marleta,
Head of Subdirectorate of Tax Niky Adam, Martha Noviaditya, Fernando
Dissemination Bakhtiar Simanjuntak
Vice Team Leader I Publication Division

Head of Subdirectorate of Public Head of Section of Website Management
Relations Member: Niko Haditia Samosir, Erwan
Muslim Yusuf Raja, Bagas Satria
Vice Team Leader II Pamungkas, Arif Miftahur Roza,
Head of Subdirectorate of Tax Services Vibratoriano Heri Ramadani, Gema
Anugrah Ramadhan
Secretary
Head of Section of Tax Cervices and Operator dan Multimedia Division
Consultation Support Head of Section of External Relations
Member: Rima Budiarti, Sella Anggraini Member: Arnoldus Yan Pranata, Renzo
Harahap, Rifka Juliana, Irma Angelina Ilham Ryandhi, Ilham Fauzi, Yogi Bayu
S., Raden Roro Rai Sukhufi Dewi Avian, Al Ma’rufi Syafruddin, Bintang
Fatimatuzahra
Event Division
Head of Section of Documentation and Logistic and Consumption Division
Library, Head of Section of Head of Subdivision of Directorate
Dissemination Support Administration
Member: Yusuf A’zis Izmi Rosyid, Reza Member: Kartika Rakhmi Putri, Lesgia
Meiladi, Anika Yusman, Muhammad Arif Sistrivani, Lukas Adityanesa Hartono,
Yusuf Hasibuan, Iga Mawarni Ariyanto Ragil Novitasari, Dian Wahyu Hidayat,
Shekhy Arrissa, Dewi Wulandari
Cooperation Division
Head of Section of International Call for Paper Editor
Cooperation Zulhendrizal, Rizmy Otlani Novastria,
Member: Dessy Christianty, Haifa Rizki Aulia Harahap, Meita Wulandari,
Ghassani, Yusuf Abdillah Ikhsan Anika Yusman, Sosro Dahana,
Mohammad Arief Iqbal, Deddy Dwi
Purnomo, Martha Noviaditya, Shekhy Purwanti, Yulianti Abbas, Selvi, Agustine
Arrissa, Ayu Amaliah Indira, Dhimas, Dwianika, Darussalam, Gatot
Wisnu Mahendra, Hesti Luh Soepriyanto, Zulaikha, Danny Septriadi,
Suprobowati, Niky Adam, Praditya Janu Meinie Susanty, Andi Nugroho Suryo
Wisaksono, Panji Nugroho, Hanung Kuncoro, Christine Tjen, Bawono
Adittya Aristyatama, Franxis Erika Kristiaji, Gusti Ayu Indah Ratnasari,
Murtiasari, Denty Tresna Mutiara, Adella Ahmad Komara, Futu Faturay, Arnaldo
Septikarina, Imaduddin Zauki, Rian Purba, Arifin Rosid, Lasmin, Sriadi
Ramdani, Darmawan Sidiq, Pinurba Setyanto, Melki Ferdian, Carolina Candri
Anandita, Rizky Ikhsan, Rahadian Noor, Prihandinisari, Niken Evi Suryani, I Putu
Dwi Langgeng Santoso, Faisal, Angga Sudiana, Himawan Saputro, Abdul Haris
Sukma Dhaniswara, Boy Valentin Purba, Muhammadi, Budi Susila, Matondang
Nova Sari Dewi, Seno Andri Prasetyo, Elsa Siburian, Muh Dularif, Lury Sofyan,
Ester Tri Whisnu Kartika, Kadek Pasek, Nora Dwi Prasasti Asih, Subagio Efendi,
Parwata, Ferga Aristama, Ardian Unggul A.K.S Bayunanto, Firman Tatariyanto,
Nurkusuma, Ni Putu Purnami Jayanti Gorga Parlaungan, Ika Retnaningtyas,
Judiana Manihuruk, Lenny Erna Meliani,
Call for Paper Reviewer Prima Kusuma Sumantri, Primadona
Pardomuan Robinson Sihombing, Nur Harahap, Yari Yuhariprasetia, Wishnu
Cahyonowati, Yudhi Dharma Nauly, Ni agung baroto, Zulhendrizal, Nidya
Wayan Rustiarini, Puji Harto, Wing Hapsari, Bawadi
Wahyu Winarno, Christina Juliana, Dyah
Author:
Arifin Rosid, Muhammad Dahlan, Fajar Surya Putra, Yeni Farida, Anies Said
Basalamah, Herru Widiatmani,... [et. al.]
ISBN:
Publisher:
Ministry of Finance
Jl. Dr. Wahidin Raya No. 1 Jakarta Pusat
Indonesia
Redaction:
Jl. Gatot Subroto, Kav. 40-42,
Jakarta 12190
Directorate General of Tax, Ministry of
Finance
First Edition, 2023

PREFACE
We are very grateful to God Almighty for His endless blessing that the International Tax
Conference 2022 was held successfully on December 5-7, 2022. This tax conference
has been held every two years since 2018 and this year is the third tax conference and
the first international call for papers and the 1st International Tax Conference held by
the Directorate General of Taxes and supported by our generous donors Australian
Taxation Office (ATO) and Deutsche Gesellschaft für Internationale Zusammenarbeit
(GIZ).
The theme of this conference is “G20 Indonesian Presidency: Achieving a Stronger and
More Sustainable World Recovery through International and National Tax Reform". Tax
reform is the key to improving state revenues. Moreover, global changes such as digital
technology that are increasingly dominating and climate change must be responded to
appropriately. Domestically, the existence of digital technology has also changed how
people interact and transact. Thus, the tax system must be strengthened by reforming
the tax system in Indonesia. The conference was a great opportunity to share initial
successes (and mishaps) from which each of us can learn.
We hope that our conference can answer some of the current challenges that are still
being faced by several countries in this recovery period as the impact of Covid-19. This
event surely would not have been a success without the solid cooperation between the
Australian Taxation Office (ATO) and Deutsche Gesellschaft für Internationale
Zusammenarbeit (GIZ).
Finally, we would like to thank all participants and presenters who have been supporting
the International Tax Conference 2022. We have a total of 217 applicants and 172
papers submitted. We also proudly inform you that this number is higher than the last
two years, which was only 118 papers submitted. All papers submitted are impressive
and contain research regarding policy recommendations that are very useful for the
Directorate General of Taxes in creating a better tax system in the future. After a process
of double-blind review, there are 35 best papers selected to be presented in the
conference-parallel session on December 5, 2022. The 35 best papers’ authors have
agreed and committed to submit their full paper in the proceedings.
Besides the parallel session of 35 Best Papers, the conference included altogether
about a keynote session and panel sessions in which there were invaluable
presentations from Yon Arsal, Ph.D., Assistant Minister for Tax Compliance, MoF
Indonesia; Hector Thompson, Deputy Commissioner, International at the Australian
Taxation Office (ATO); and Dr. Machfud Sidik, M.Sc., Taxation and Fiscal Policy Expert
of the Deutsche Gesellschaft für Internationale Zusammenarbeit (GIZ).
1
The last, we sincerely hope that these proceedings will bring a lot of benefits to all the
participants and readers. We welcome any suggestions and constructive feedback to
improve the next tax conference and proceedings, and we look forward to seeing you
again.
Jakarta, Februari 2023
Director of Tax
Dissemination, Services,
and Public Relations
2
TABLE OF CONTENT
PREFACE........................................................................................................................................ 1
TABLE OF CONTENT ...................................................................................................................... 3
PREDICTING FIRMS’ TAXPAYING BEHAVIOUR USINGARTIFICIAL NEURAL NETWORKS: THE CASE
OF INDONESIA .............................................................................................................................. 5
DOES GOVERNMENT EFFECTIVENESS MODERATE THE RELATIONSHIP BETWEEN REGULATORY
QUALITY AND TAX COMPLEXITY?: A TALE OF A HUNDRED NATIONS ........................................ 39
CORRECTIONS TO DOMESTIC TRANSFER PRICING POLICIES A ZERO-SUM-GAME OF NATIONAL
TAX REVENUE?: CASE STUDY OF KPP PRATAMA IN DKI JAKARTA REGION ................................ 55
REVIEW OF THE EMPLOYEE PERFORMANCE EVALUATION SYSTEM AT THE DIRECTORATE
GENERAL OF TAXES: ITS EFFECTIVENESS AND EMPLOYEE SATISFACTION LEVEL....................... 74
THE EFFECT OF THE QUALITY OF TAX DETERMINATION, WARNING LETTER, AND DISTRESS
WARRANT ON TAX ARREARS DISBURSEMENT ............................................................................. 1
OKUNS LAW, PHILLIPS CURVE AND ITS EFFECT ON THE GROWTH OF INCOME TAX ARTICLE 21
PAYMENTS DURING COVID-19 PANDEMIC ................................................................................ 18
THE IMPACT OF SATISFACTION AND IMPORTANCE PERCEPTION ON THE APPROWEB’S
TRIGGER DATA FOLLOW-UP: AN EMPIRICAL STUDY BASED ON ADMINISTRATIVE DATA ......... 34
A SUSTAINABLE APPROACH ON TAX AUDIT INSPECTION: HOW TAX AGGRESSIVE STRATEGIES
CAN BE DETECTED FROM FEE OF INDEPENDENT AUDITOR ....................................................... 45
EXIT TAX ADOPTION TO PROTECT INDONESIA’S TAX BASE: ARE WE READY? ........................... 54
MACHINE LEARNING: CLASSIFIYING TAXPAYER’S SUPERVISING ZONE BASED ON THE STREET
ADDRESS USING NATURAL LANGUAGE PROCESSING ALGORITHM ........................................... 70
EVALUATION OF CORPORATE INCOME TAX RATES REDUCTION POLICY APPLIED TO THE
GENERAL EQUILIBRIUM APPROACH AND THE DEGREE OF SELF-FINANCING METHOD: CASE
STUDY IN INDONESIA.................................................................................................................. 84
DETERMINANTS OF TAX MORALE (EMPIRICAL STUDY ON MSME IN JAVA)............................... 96
INDONESIA’S TAX REVENUE: .................................................................................................... 107
TOWARD SEAMLESS TAXATION................................................................................................ 120
THROUGH A SPLIT PAYMENT MECHANISM.............................................................................. 120
DOES GIVING CARROT WORK EFFECTIVELY? A STUDY OF TAX RELAXATION IN INDONESIA ... 132
RECONSTRUCTION CONCEPT OF THE MEANING OF PERMANENT ESTABLISHMENT PHYSICAL
PRESENCE ................................................................................................................................. 146
THE TAXATION ASPECT REVIEW OF CASH POOLING BASED ON INDONESIAN REGULATIONS 160
3
ASSESSING TAXPAYERS’ ABILITY TO PAY: A MACHINE LEARNING APPROACH ........................ 174
FINAL TAX REGRESSION ............................................................................................................ 187
SIMILARITY INDEX FOR ESTIMATING TAX REFORM PROGRAM AND ITS IMPACT ON TA
REVENUE: INDONESIA CASE STUDY.......................................................................................... 206
THE IMPACT OF THE ALTERNATIVE TAX BASE MEASUREMENT POLICY ON THE VAT REVENUE
PERFORMANCE IN THE INDONESIAN AGRICULTURAL SECTOR .................................................. 12
FORECASTING VALUE-ADDED TAX (VAT) REVENUE USING AUTOREGRESSIVE INTEGRATED
MOVING AVERAGE (ARIMA) BOX-JENKINS METHOD ................................................................... 1
USING DATA MINING TO DETECT FRAUD PATTERNSFROM TAXPAYER’S FINANCIAL STATEMENT ...... 18
APPLICATION OF DATA MINING TO TAXPAYERS ISSUING FICTITIOUS TAX INVOICE USING
CLASSIFICATION TECHNIQUES .................................................................................................... 34
IMPACT STREET EARNINGS ON TAX AVOIDANCE ....................................................................... 50
SELF-SERVICE ANALYTICS AND DATA PIPELINE IN TAX ANALYTICS ............................................ 63
USABILITY MEASUREMENT OF E-BUPOT INSTANSI PEMERINTAH APPLICATIONS (CASE STUDIES
IN THE GENERAL ELECTION SUPERVISORY AGENCY).................................................................. 76
IMPACT OF AUTOMOTIVE TAX INCENTIVES DURING COVID-19: EVIDENCE FROM INDONESIA 92
NATURAL LANGUAGE PROCESSING ANALYSIS OF APPEAL DECISIONS IN TRANSFER PRICING
TAX DISPUTES IN INDONESIA ................................................................................................... 106
DO OPERATIONAL RISKS AND CREDIT RISKS MAKE BANKS MORE AGGRESSIVE TO AVOID
TAXES? EMPIRICAL STUDY OF BANKING COMPANIES ON THE INDONESIA STOCK EXCHANGE
.................................................................................................................................................. 122
EXAMINING FIRMS’ FINANCING OPTIONS: DO THIN CAPITALISATION RULES MATTER? ........ 135
TAXATION, DIGITAL FINANCIAL SERVICES, AND THEIR IMPACTS ON FINANCIAL INCLUSION.... 26
4
PREDICTING FIRMS’ TAXPAYING BEHAVIOUR
USINGARTIFICIAL NEURAL NETWORKS: THE CASE
OF INDONESIA
Arifin Rosid
Directorate General of Taxation, Jakarta. Email: arifin.rosid@pajak.go.id
Faculty of Economics and Business, Universitas Indonesia, Depok. Email: arifin.rosid@ui.ac.id
ABSTRACT
Big data and sophisticated analytics might help tax authorities extract actionable data insights. In
response, this paper employs an Artificial Neural Networks (ANN) model to predict and discover the
determinants of firms’ taxpaying behaviour. Examining 538,254 firm-level administrative data across
fiscal years 2014 and 2019, this study is the first to apply ANN to exploit the taxpaying behaviour of
Indonesian firms. Multi-Layer Perceptron Neural Network-based models were trained to predict three
categories of taxpaying measurement—i.e, Corporate Tax Turnover Ratio (CTTOR)—across varying
magnitudes of annual turnover.The models predicted the firms' taxpaying behaviour with an average
accuracy rate above 92%. This study also reveals heterogeneous channels responsible for firms’
taxpaying behaviour across groups. The findings demonstrate other business income and positive
fiscal adjustment to be significant predictors of taxpaying behaviour for small and medium firms. In
contrast, operating profit margin, other business expenses, and negative fiscal adjustment are
prominent predictors for large corporations. The results will assist decision-makers in tax
administrations about potential areas of misreporting, enabling them to develop evidence-based and
effective policy actions.
Keywords: corporate taxpayers, tax compliance, taxpaying behaviour, Artificial Neural Networks
Big data dan analisis data yang canggih dapat membantu otoritas pajak menghasilkan informasi yang
berhasil guna. Sebagai respon, tulisan ini menggunakan model Jaringan Syaraf Tiruan (JST) untuk
memprediksi dan menemukan faktor-faktor penentu perilaku pembayaran oleh Wajib Pajak Badan.
Menguji data dari 538.254 Surat Pemberitahuan Tahunan (SPT) PPh Badan sejak tahun pajak 2014
sampai dengan 2019, studi ini adalah yang pertama menerapkan JST untuk mengeksploitasi perilaku
membayar pajak dari wajib pajak badan di Indonesia. Model berbasis Multi-Layer Perceptron Neural
Network dilatih untuk memprediksi tiga kategori besaran pembayaran pajak—yaitu, Rasio Perputaran
Pajak Perusahaan (Corporate Tax Turnover Ratio/CTTOR)—pada beberapa kategori perusahaan
berdasarkan besaran omzet tahunan. Model dalam studi ini berhasil memprediksi perilaku
pembayaran pajak perusahaan dengan tingkat akurasi rata-rata di atas 92%. Riset ini juga
menemukan variasi beberapa faktor yang menentukan perilaku pembayaran pajak dari grup
perusahaan yang berbeda. Studi ini menemukan bahwa pendapatan dari luar usaha dan
penyesuaian fiskal positif adalah prediktor kunci dari perilaku membayar pajak untuk perusahaan
kecil dan menengah. Sebaliknya, margin laba operasi, biaya dari luar usaha, dan penyesuaian fiskal
negatif adalah prediktor yang signifikan untuk perusahaan besar. Hasil studi ini dapat membantu para
pengambil keputusan di otoritas perpajakan menunjukkan beberapa area yang berpotensi terjadi
penghindaran pajak, yang dapat digunakan untuk mengembangkan kebijakan dan rencana aksi
berbasis bukti yang efektif.
Kata kunci: wajib pajak badan, kepatuhan pajak, perilaku membayar pajak, Jaringan Syaraf Tiruan (JST)
1. INTRODUCTION
For a nation to meet its objectives for sustainable development, taxation is essential
(Bird, 2010). Infrastructure, maintaining economic growth, and eradicating poverty are
5
just a few development initiatives that employ tax revenue. However, in many countries,
tax revenue mobilisation remains significantly below the levels required to support
sustainable development objectives—i.e., 15% of GDP is an often-quoted yearly target
(Prichard et al., 2019).
Due to the complexity of tax and the time and resources needed to monitor and
examine the tax returns of both individuals and firms, tax noncompliance is challenging
to detect. As a result, one of the most important goals of tax authorities worldwide is to
quantify and identify taxpaying behaviour (Pérez López et al., 2019). In this sense, the
way tax authorities handle and evaluate the data at their disposal may be improved by
big data and advanced analytics (Brondolo et al., 2022). Big data and sophisticated
analytics might help tax authorities extract actionable insights from the information they
already have while also supplying them with new tools to strengthen enforcement and
discover tax fraud, evasion, and avoidance (Dom et al., 2022). Revenue bodies need to
be aware of the potential for a fresh wave of ‘innovative technologies’ to reshape tax
administration (ADB, 2022). These technologies are gaining popularity in an increasing
number of revenue bodies worldwide. Among these is artificial intelligence (ADB, 2022).
In recent years artificial intelligence has evolved into a technology that enables the
administration of massive datasets and the application of algorithms that, despite their
complicated structure, provide results that can be comprehended (Pérez López et al.,
2019). Two tasks performed by the tax authorities that may be accomplished using
artificial intelligence (AI) are tax audits and tax collecting (Huang, 2018). This can be
accomplished by the utilisation of work automation using computer-controlled tools and
the use of AI. Several approaches may adopt AI to enhance services for tech-savvy
taxpayers. Additionally, applying AI in taxes might help tax authorities analyse risks, and
spot unusual commercial practises (Wang & Wang, 2020). Thus, the development of
new analytical tools has dramatically improved the efficiency and efficacy of tax
administration, such as the enhanced identification of potential non-compliance via
better risk assessment modeling and employing advanced analytical approaches
(OECD, 2020a).
Big data and neural networks are often linked. This is due to the common
association of neural networks with challenging or impractical modelling tasks that
cannot be accomplished by other kinds of models (Cook, 2020). For this reason, this
paper aims to contribute through research on applying neural network models to tax
administrative data to facilitate the identification of tax evasion by quantifying taxpayers'
propensity to commit underreporting behaviour. With this objective in mind, one of the
machine learning prediction techniques for supervised learning—the Artificial Neural
Networks (ANN) model—is used. ANN method is excellent at solving random, ill-defined
problems, highly non-linear, with many distinct and complex variables (Graupe, 2013).
In the last ten years, ANN has become a powerful and essential class of machine
learning technologies (Cook, 2020). It is a non-parametric modeling tool that can perform
the mapping of complex functions with sufficient accuracy (Zhang et al., 1999). ANN is
widely used in cross-disciplinary research and tasks. For example, in finance, Sánchez-
Serrano et al. (2020) used the ANN approach to create a predictive model of special
audit opinion for consolidated financial statements. This study can predict audit opinion
with an accuracy rate of 83%. Artificial intelligence approaches in finance are starting to
be widely used, considering that behaviour is often non-linear and full of uncertainty
(Bahrammirzaee, 2010). In the field of education, for example, Aryadoust and Baghaei
(2016) examine the relationship between reading ability, lexical knowledge, and
grammar of a group of students who use English as a foreign language. In this study,
ANN accurately classified approximately 78% of students.
6
Several studies built on artificial intelligence (AI) in the form of harmony search
optimization algorithms, support vector machines, genetic algorithms, decision trees,
logistic regressions, and neural networks to discover tax evasion behaviour (Goumagias
et al., 2012; Lin et al., 2012; González & Velásquez, 2013; Warner et al., 2015;
Rahimikia et al. 2017). Specifically, the ANN approach has been applied in various tax
studies across jurisdictions (Chen et al., 2011; Lin et al., 2012; Rahimikia et al., 2017;
Jupri & Sarno, 2018; Jang, 2019; Pérez López et al., 2019). For instance, Lin et al.
(2012) used an artificial intelligence approach to detect tax evasion in Taiwan. In Chile,
González and Velásquez (2013) used decision trees, neural networks, and Bayesian
networks to spot fraud tendencies for audited taxpayers. Rahimikia et al. (2017)
combined multilayer perceptron neural networks with various classification algorithms to
identify corporate tax evasion in Iran. Pérez López et al. (2019) contribute to identifying
tax fraud for personal income tax returns in Spain by using ANN as an advanced
prediction technique with an efficiency rate of 84.3%. However, to the best of the author's
knowledge, no empirical studies have used the ANN approach to predict the taxpaying
behaviour of firms in Indonesia. As Saragih et al. (2022) argue, the potential advantage
of an AI application for modernising Indonesia's tax administration system is that it would
facilitate enforcement by including AI in tax audits to monitor taxpayers and in tax
services to enhance the effectiveness of the tax authority. Subsequently, this paper is
the first to exploit how neural network algorithms can be adopted to predict Indonesian
firms' taxpaying behaviour and identify their significant predictors. In doing so, this study
compared four different groups based on the size of firms' annual turnover.
The results are very encouraging. This study supports the notion that artificial
intelligence approaches are superior to conventional statistical methods for addressing
various issues, particularly those involving nonlinear patterns (see, for example,
Bahrammirzaee, 2010; Chen et al., 2011; Lin et al., 2012; Aryadoust and Baghaei, 2016;
Sánchez-Serrano et al., 2020). This study concludes that the ANN approach accurately
predicts the taxpaying behaviour of Indonesian firms with different annual turnovers, with
an accuracy rate above 92%. The implementation of artificial intelligence also allows this
study to identify heterogeneous channels responsible for firms' taxpaying behaviour
across groups. As Brondolo et al. (2022) posit, particularly in establishing compliance
risk management strategy, with the use of supervised machine learning, the algorithm
may reveal complex data patterns associated with successful case outcomes while
deemphasizing those that were not. The findings would be of benefit since the DGT has
currently not yet incorporated artificial intelligence in its operations (ADB, 2022). The
present study raises the possibility of generating actionable data insights and identifying
areas of misreporting at strategic levels.
This article contributes at two levels. First, at the taxation literature level, this paper
adds empirical knowledge about how artificial intelligence approaches can be applied in
taxation studies. The practical implications of this study are essential because it is the
first study to try to predict the taxpaying behaviour of Indonesian firms using
administrative data. Second, on a practical level, the findings of this paper help
Indonesia's tax authorities identify several unique, influential factors and provide
actionable data insights that lead to income taxpaying behaviour.
This paper consists of six main parts. Section 2 describes the conceptual framework
of this study. Section 3 describes the data and the analytical approach, while Section 4
describes the results and discussions. Section 5 concludes, and Section 6 discusses
the practical implications and limitations.
7
2. CONCEPTUAL FRAMEWORK
2.1. Institutional Settings
The Indonesian government launched and implemented major and radical tax
reform in 1983 by introducing self-assessment to its tax system as a strategic response
to the threat to the national budget. The primary goal of the reform was to increase non-
oil-related tax revenue to decrease the government's dependency on oil money (Heij,
2001). At that time, Indonesia's tax system underwent significant modifications that
brought it into compliance with worldwide best practices (Alm, 2019). Most of these
structural elements are still in place today.
However, the tax system has shown that it cannot generate optimal revenue
collection, partly because it has developed through time in a fragmented, ad hoc fashion
with little apparent attention given to how the elements of the system need to fit together
(Alm, 2019). Although these issues are not unique to Indonesia (see, for example, IMF,
2014; World Bank, 2016), the lessons from these Indonesian tax reform initiatives are
often vague, perhaps even unaddressed, and specific (Alm, 2019). For this reason, the
Indonesian tax authority—the Directorate General of Taxation (DGT)—is still dealing
with increasingly challenging and complex issues in collecting tax revenue, and both the
growth and ratio of tax collection showed a decreasing trend (DGT, 2021). In the 2019
fiscal year, Indonesia's tax ratio was 11.6% (ADB, 2022). This figure is much lower than
the average tax ratio for OECD countries, which is 33.8%, followed by Latin America
and the Caribbean (23.0%), Asia-Pacific (18.7%), and Africa (16.8%) (ADB, 2022).
From 2017 to 2021, income taxes in Indonesia contribute more than 55% of national
tax revenue. Income tax revenue is divided into two categories: (i) oil and gas and (ii)
non-oil and gas income taxes. Most income tax revenue comes from non-oil and gas
income taxes (92%), comprising several types of income tax based on related sections
of income tax law. Graph 1 shows that corporations contribute the most considerable
portion of income tax revenue from 2017 to 2021, ranging from 26.9% to 34.3%. The
Covid pandemic resulted in a decline in corporation tax income in 2020, followed by a
rise in 2021. In 2020, as Graph 1 shows, the contribution of Article 25 of Corporations
and Article 22 on Imports reached the lowest points due to the pandemic and tax
incentives provided for impacted industries.
Graph 1: Type of income taxes in Indonesia and their contribution to total

Source: Adapted from the DGT's internal data
8
International experience indicates that a small number of large corporations (usually
less than 1%) account for 60%–70% of domestic tax revenues, whereas a significant
number of small enterprises contribute for less than 5%–10%. Between these two
categories, medium-sized firms contribute about 20%–30% of domestic tax revenues
(York, 2011). In most nations, this distribution of corporate taxpayers is identical (York,
2011).
Due to its continual low tax ratio, DGT (2015) has identified several difficulties and
is primarily concerned with persistent compliance problems. Regarding this, while
artificial intelligence is currently not yet implemented in the DGT (ADB, 2022), the DGT
is concerned with establishing a data-driven organisation to improve tax compliance
(DGT, 2022). It is envisaged that by being a data-driven organisation, DGT's ambition
to become a trusted partner for nation-building and to collect state revenues via effective
tax administration may be accomplished (DGT, 2022). Indeed, successful digitalisation
initiatives understand that technology is a tool for solving a specific issue rather than a
goal in and of itself (Dom et al., 2022).
2.2. Tax Avoidance and Taxpaying Behaviour

Before attempting to analyse corporate taxpaying behaviour, one needs to
comprehend tax compliance first. There are several ways to examine tax compliance,
some of which can lead to more definitional concerns. The definitions include a variety
of topics and are broad. Consequently, the definition of compliant behaviour is not
universally agreed upon (Devos, 2014). Following Rosid et al. (2018), this study
emphasises the benefit of adopting an operational approach instead of a conceptual
approach. This paper defines tax compliance as 'taxpayers' willingness to correctly
report tax liability in accordance with the prevailing tax law' (OECD, 2014). Thus, in
contrast to forced willingness (ex post), this concept is more concerned with voluntary
willingness (ex ante) of taxpaying behaviour.1
Any arrangement or transaction that lowers a company's tax obligation is
considered tax avoidance (Dyreng et al., 2008). While various ways to measure tax
avoidance (Gebhart, 2017), Hanlon & Heitzman (2010) define tax avoidance as any tax
reduction resulting from a transaction directly affecting the company's tax burden.
Nevertheless, OECD finds it challenging to define the term 'tax avoidance'—and
distinguish it from 'tax evasion', but it generally refers to tax planning undertaken by
taxpayers with the intention of reducing their tax liabilities.2 While such planning may be
considered legal, it is typically contrary to the objectives of tax laws. Slemrod (2016)
opines that the dividing line between illegal tax evasion and legal tax evasion is blurry.
For this reason, this study does not emphasise the legality of tax avoidance but rather
how much the company reduces its tax payments.
In this sphere, effective tax rates (ETR) may be extensively used to gauge a firm's
tax burden (Dyreng et al., 2008; Gebhart, 2017). ETR—calculated by using a measure
of pre-tax earnings or cash flow as the denominator and specific estimates of the tax
due as the nominator—is also one of the commonly used metrics of tax evasion in this
area (Hanlon & Heitzman, 2010). This proxy represents the typical tax rate per revenue
unit or cash flow. Thus, ETR is a potent predictor of the efficacy of a company's tax
planning activities (Mills et al., 1998).
1
It is important to note that this definition purposefully omits the registration, filing, and income reporting elements
owing to the practical and legal challenges in acquiring the data.
2
See, the OECD Centre for Tax Policy and Administration's Glossary of Tax Terms at http://www.oecd.org/ctp/
glossaryoftaxterms.htm, accessed 25 July 2022.
9
However, it is worth emphasising that ETR may only be able to capture non-
conforming tax avoidance activities but ignore conforming tax avoidance activities
(Badertscher et al., 2019). This is because conforming tax avoidance is more difficult to
detect because it reduces book income and tax income at the same time.3 When using
a measurement such as ETR, this will change the numerator and denominator at the
same time—preventing the company from being detected as tax evaders. In addition,
conforming tax avoidance is more difficult to detect because there are not many
measurements related to conforming tax avoidance (Badertscher et al., 2019; Hanlon &
Heitzman, 2010). Regarding this, the noncompliance rate for firms concerning their size
seems to be 'U-shaped, with medium-sized enterprises among the group of large
corporations showing the lowest noncompliance rate (Slemrod, 2007).
For this reason—and for practicality—to gauge the level of tax avoidance, we
instead chose to utilise a different metric adopted by the Indonesian tax authority: the
corporate tax to turnover ratio (CTTOR). CTTOR is a tax payment measure based on
declared income tax payable scaled by annual turnover. This practical measure refers
to the Director General of Taxation Circular Letter Number SE-02/PJ/2016 concerning
the Benchmark Behavioural Model (BBM). By doing so, this paper provides actionable
policy implications in addition to giving the Indonesian tax authorities an objective'
empirical indicator' of the taxpaying behaviour of corporate income taxpayers.
Relating to this, the recently completed Voluntary Disclosure Program (VDP)
provides some relevant indicators of persistent non-compliance behaviour.4 As of the
completion of the VDP's implementation on 30 June 2022, taxpayers had disclosed
assets totalling IDR 594.82 trillion. They had paid a total of IDR 61.01 trillion in tax
liabilities derived from those assets in the form of income tax. 5 Both under the tax
amnesty and VDP schemes, the value of assets that were declared and the ransom
money that was paid were the strong indication of underreporting behaviour among
Indonesian taxpayers. In this sense, one justification for prevalent tax non-compliance
in Indonesia might be the outdated information technology being utilised in tax
administration, which does not reflect the most recent computer technology being used
in other countries to track taxpayers and their incomes (Alm, 2019).
2.3. Artificial Neural Networks (ANN)

ANN is one of the most popular artificial intelligence methods in various disciplines
(Cook, 2020), including finance (Bahrammirzaee, 2010). It comprises mathematical
approaches typically applied in prediction and classification studies (Aryadoust &
Baghaei, 2016). ANN is a frequently utilised technique for predictive data mining
analysis because of its accuracy, adaptability, and simplicity. Particularly in instances
where the underlying processes are sophisticated (IBM, 2021).6 ANN is grounded on
3
Two strategies—nonconforming and conforming tax avoidance—can help businesses lower their income tax
obligations. Companies that use non-conforming tax avoidance strategies decrease their income tax liability but not
their book income. Companies that use conforming tax avoidance lower their income tax obligations by taking actions
that lower both their taxable and book earnings (Badertscher et al. 2019).
4
Note that the Indonesian government has just introduced a tax amnesty program quite recently, i.e., in 2016. A tax
amnesty is a ‘forgiveness’ period during which people are given a chance to pay back taxes that have gone unpaid
without having to worry about facing the financial penalties and/or criminal punishment that come with being found
guilty of tax evasion. In the 2016 tax amnesty program, a total of IDR 4.87 trillion in unreported assets were reported to
the tax authorities, with about three-quarters of these disclosed assets being domestic assets. See,
https://setkab.go.id/realisasi-tax-amnesty-menkeu-tebusan-rp130-triliun-deklarasi-rp4-8134-triliun-dan-repatriasi-rp146-
triliun/, accessed 24 July 2022.
5
See, https://www.kemenkeu.go.id/publikasi/berita/pps-berakhir-menkeu-harta-yang-diungkap-rp594-82-triliun-dengan-
pph-rp61-01-triliun/, accessed 26 July 2022.
6
Data mining is the exploration and analysis of vast amounts of data using automated or semiautomatic methods to
spot meaningful patterns and rules (Fayyad et al., 1996).
10
the idea that the specific link between independent and dependent variables can be
calculated using non-linear mathematical functions (Aryadoust & Baghaei, 2016). Thus,
the real benefit of ANN is pattern identification and classification due to its non-linear,
non-parametric nature of adaptive learning (Zhang et al., 1999).7 Table 1 summarises
the benefits and drawbacks of several classification algorithms.
Generally, as Table 1 suggests, neural network models outperform other linear and
non-linear models in terms of accuracy and predictive capability (Murorunkwere et al.,
2022). From a quantitative standpoint, neural networks often consist of optimal
combinations that enable more accurate predictions and estimates than other models
(Pérez López et al., 2019). Without requiring the researcher to make prior assumptions
about the connections between the dependent and independent variables, a neural
network may mimic various statistical models. Instead, the learning process determines
how the connections will take shape (IBM, 2021).
Table 1: Overview of benefits and drawbacks of the classification algorithms

Advantages Disadvantages
Logistic  Models are often very accurate  It can only provide linear solutions
regression  Works well on small datasets  Problems with high collinearity of the
 Predicts probabilities input variables
 Easy to interpret, in particular, the
influence of each input variable
Linear  Typically, very fast building the model  More restrictive assumptions than other
discriminant  Works well on small datasets methods (e.g., logistic regression)
analysis  Optimal if data assumptions are  Usually, needs data preparation
fulfilled  Sensitive to outliers
 Only applicable to linear problems
Decision  Robust to outliers  Can be computationally
trees  Model and decision rules are easy to expensive to train
understand  Large trees tend to overfitting
 Can handle different data types  Most of the time it does not find the
 Fast in prediction and no assumptions optimal solution
on variable distributions needed.  Prefers variables with many categories
Thus, less effort for data preparation or numerical data
 Can handle missing values
Neural  Good performance on large datasets  Training can be computationally
networks  Very good at allowing nonlinear expensive
relations and can generate very  Results and effects of input variables
complex decision boundaries are hard to interpret (black box
 Non-parametric algorithm)
 No distribution assumptions needed  Tends to overfitting and does not
 Can handle noisy data always find the optimal solution
 Often outperforms other methods  Can only process continuous input
variables
Source: Adapted from Wendler & Gröttrup (2021, pp 761-762)
The findings of the neural network should generally resemble those of the linear
regression model if there is a linear connection between the dependent and independent
variables. The neural network will automatically get close to the ‘correct’ model structure
if a non-linear connection would be more suitable (IBM, 2021). In this sense, in addition
to being crucial for traditional statistical decision theory, posterior probabilities are also
crucial for many managerial decision-making scenarios (Zhang et al., 1999). Although
7
The discussion of ANN involves a complex mathematical model. Readers can refer to Ripley (1996) and Haykin
(1999) for a comprehensive discussion of theoretical and mathematical models. Readers interested in the practical
aspects of ANNs might learn more in 'IBM SPSS Neural Networks 28' (IBM, 2021).
11
there are other methods for estimating posterior probability, ANN is the only one
currently known to directly estimate posterior probabilities without knowledge of the
underlying group population distributions (Zhang et al., 1999).
The general ANN model is a three-layered structure of linked nodes: the input, the
hidden, and the output units. The nodes between input and output layers might create
one or more hidden layers. As illustrated in Figure 1, every neuron in one layer has a
connection to every other subsequent layer, but neurons corresponding to the same
layer have no interactions between them (Cook, 2020). The input layer gets sensory
input, the hidden layer executes the information processing, and the output layer
generates the class label or anticipates continuous values.8
(a) ANN architecture (b) ANN activation node
Figure 1: ANN architecture and neural network activation node

Source: Adapted from Cook (2020)
The primary purpose of the input layer is to disperse the data supplied to the neural
network for further processing. The hidden and output layer nodes process the signals
by adding synaptic weights or processing factors. Each layer has an extra node known
as bias, which adds a new term to the output of all the nodes in the layer. All inputs to a
node are weighted, integrated, and processed by a function known as a transfer function
or activation function, which regulates the output flow from that node to facilitate
communication with all nodes in the following layer (Pérez López et al., 2019).
The following equations represent the nett sum of the weighted inputs entering a
node j and the output activation function that translates a neuron's weighted input to its
1
output activation (often the sigmoid function): 𝑆𝑗 = ∑𝑛𝑖=1 𝑋𝑖 𝑊𝑖𝑗 and 𝑂𝑗 = 𝑠 . An
1+𝑒 𝑗
activation function processes a neuron's inputs x0, x1, . . . , xn with their associated
weights w0, w1, . . . ,wn and the result is the neuron's output, which corresponds to:
𝑛
𝑂𝑢𝑡𝑝𝑢𝑡 = 𝑓(𝑤1 𝑥1 + 𝑤2 𝑥2 + ⋯ + 𝑤𝑛 𝑥𝑛 + 𝑏𝑖𝑎𝑠) = 𝑓(∑ 𝑥𝑖 𝑤𝑖 + 𝑏𝑖𝑎𝑠)

𝑖=1
8
This structure is also known as 'feedforward architecture' because the links in the network move forward from the
input layer to the output layer without a feedback loop. The input layer consists of predictors, the hidden layer consists
of units that cannot be observed (unobservable), and the output layer contains the response. The unit in the output is a
collection of several functions from the unit in the hidden layer (IBM 2021).
12
where wi and xi are the weight vector and input vector, respectively, f is the activation
function used on the sum of products of each input and its corresponding weight and the
bias (Murorunkwere et al., 2022).9 Without an activation function, every ANN is merely
a basic linear function (Agostinelli et al., 2015).
Even though traditional linear equations are simple and quick to solve, they are
restricted in their complexity and lack the capacity to learn and find intricate data
mappings. In response, ANN makes it simple to manage large datasets and, despite the
complexity of their algorithms, provide easily interpretable results—which is why they
are widely used in the financial industry, marketing, forecasting, and increasingly in risk
assessment and fraud detection (Murorunkwere et al., 2022). In addition to facilitating
the categorisation of each group of taxpayers as low, moderate, and high tax payments,
the neural network exposes firms' propensity for taxpaying behaviour based on the
variables under study.
3. RESEARCH METHODS
3.1. Empirical Data
Information on taxable income and tax liability are required for quantifying company
tax compliance behaviour (Salihu et al., 2013). A key strength of the present study was
access to a large set of actual data. This study uses administrative data from fiscal years
2014 to 2019, consisting of 538,254 useable anonymous corporate tax records. This
firm-level data offers a unique period and natural setting covering six fiscal years prior
to the Covid-19 pandemic—i.e., when the economy plummed significantly (OECD,
2020b). As shown earlier in Graph 1, during the pandemic, the contribution of corporate
income tax reached the lowest point. During these unprecedented times, the economy
plummeted substantially, with more than three-quarters of tax administrations
experiencing a decline (OECD, 2022). One of the key advantages of adopting six years
period is that it may prevent year-to-year volatility in the variables under investigation.
Table 2 details of the variables.
Table 2: Variables under study

Variable name Code Description
1 Gross profit margin GPM Annual turnover minus cost of goods sold, scaled by
annual turnover
2 Operating profit margin OPM Annual turnover minus cost of goods sold and
operating expenses, scaled by annual turnover
3 Other income ratio OIR Other business income scaled by annual turnover
4 Other expense ratio OER Other business expense scaled by annual turnover
5 Positive fiscal adjustment ratio PFAR Positive fiscal adjustment scaled by annual turnover
6 Negative fiscal adjustment ratio NFAR Negative fiscal adjustment scaled by annual turnover
7 Corporate Tax Turnover Ratio CTTOR Income tax payable scaled by annual turnover
Accordingly, this study relies heavily on Corporate Annual Income Tax Return Form
1771, 1771-I, and 1771-III. These data include detailed information on corporate income
reporting, particularly from the perspective of the income statement—i.e., annual
turnover (part 1a Form 1771-I), cost of goods sold (part 1b Form 1771-I), operating
9
The neural network will dynamically select whether the model is linear or non-linear in this process. Because of this
flexibility, the synaptic weights of the neural network are not intuitive. If the interpretation of the linear connection
between the dependent and independent variables is the primary objective, a standard statistical model approach is
preferable (IBM, 2021). The estimation of synaptic weights is based solely on training data and is therefore generally
not used to interpret ANN test results. That is, although the data is partitioned into three categories, training, testing,
and holdout, the estimation of synaptic weights is based solely on training data (IBM, 2021).
13
expenses (part 1c Form 1771-I), other business income (part 1e Form 1771-I), other
business expense (part 1f Form 1771-I), total positive fiscal adjustment (part 5m Form
1771-I),10 total negative fiscal adjustment (part 6e Form 1771-I), taxable income (part 8
Form 1771-I), and income tax payable (part 4 Form 1771).
Following Rosid and Ariyani (2022), this paper reduced the original data to obtain a
useable sample by removing observations that fit into these criteria. First, we do not
include businesses with an annual turnover of less than IDR 5 billion.11 Second, this
analysis excludes entities with a CTTOR of greater than one or negative.12 Third, this
study excludes firms that are liable to schedular-final incomes taxes. These include firms
in the construction service sector, real estate businesses, representative offices,
shipping and air transportation enterprises, and financial brokerage firms. Finally, it is
essential to note that this usable sample does not consider the impact of either tax audit
or tax oversight activities, as well as the type of tax offices where the taxpayers
administered. This study also does not consider the impact of loss carryforwards (LCF)
on the total taxable income. Only 2.9% of firms within the usable sample report LCF.
Thus, in this scenario, the inclusion should be inconsequential.
This research divides the data into four size-based categories. The reasons are
twofold. First, the way the Indonesian tax authorities administer its taxpayers are size
based—i.e., Primary Tax Office (PTO), Medium Tax Office (MTO), and Large Tax Office
(LTO). Second, businesses of various sizes exhibit various compliance habits (Slemrod,
2007). Following Indonesia Government Regulation number 7 of 2021, the categories
are as follows: i) between IDR 5 billion to 15 billion—categorised as ‘small firms’; ii)
between IDR 15 billion to 50 billion—categorised as ‘medium firms’; iii) between IDR 50
billion to 100 billion—categorised as ‘medium-large firms’; iv) more than IDR 100
billion—categorised as ‘large firms.’13
3.2. Analytical Approach

This study builds on an applied research approach. Applied research intends to
address specific practical questions or provide answers to real issues (Neuman, 2011).14
Using the perspective of the Cross-Industry Standard Process for Data Mining (CRISP-
DM), this study is in the modelling phase.15 In this phase, data mining algorithms are
constructed to extract knowledge from the data. The modelling process generates a
model or group of models accurately representing the learned information. In this sense,
intuitively, taxpaying behaviour appears to have a great deal to do with non-linear
variables. In non-linear areas, the degree of accuracy of contemporary approaches in
10
In this context, the terms ‘positive’ and ‘negative’ relate to the consequence of the adjustment on taxable income;
hence, positive denotes an increase in taxable income, and vice versa.
11
This analysis substitute IDR 5 billion thresholds for IDR 4.8 billion for the eligibility test out of practicality. Note that
firms with an annual turnover of less than IDR 4.8 billion are eligible to benefit from a final income tax. As governed by
Government Regulation Number 46 of 2013, beginning in January 2014, taxpayers with annual sales below IDR 4.8
billion are taxed at a rate of one percent of monthly turnover. The rate is then reduced 50%—at a half percent rate from
monthly turnover—starting from July 2018 based on Government Regulation Number 23 of 2018.
12
The reason is that to measure taxpaying behaviour, income tax liability and annual turnover values need to be
positive. Consequently, a negative number would suggest an error. On the other hand, the CTTOR value surpassing
one would be illogical given that the maximum tax rate (applied to taxable income) is 25%.
13
Note that according to Government Regulation number 7 of 2021 regarding Ease, Protection, and Empowerment of
Cooperatives and Micro, Small, and Medium Enterprises, businesses with annual turnover between IDR 2 billion to 15
billion are categorised as ‘small’ businesses, while businesses with turnover between IDR 15 billion to 50 billion are
categorised as ‘medium’ businesses. As of 31 July 2022, IDR one billion equals to roughly USD 67,417.
14
The implication is that applied research rarely has a strong relationship with activities to build, test, or relate theory in
depth (Neuman 2011).
15
See, https://www.ibm.com/docs/en/spss-modeler/18.2.0?topic=dm-crisp-help-overview, accessed 15 July 2022. Note
that iteration and adjusting in reaction to real-world data are key components of the CRISP-DM strategy, which is being
used in Canada and New Zealand (OECD, 2016).
14
artificial intelligence tends to be superior to classic statistical methods (Bahrammirzaee,
2010).16 Based on these considerations—and referring to the research aim, the author
employs an Artificial Neural Network (ANN) model as a primary approach.
More specifically, this paper uses the Multilayer Perceptron (MLP) module from IBM
SPSS.17 The concept of MLP in ANN departs from backpropagation learning error—an
algorithm that is most often used in ANN (Pérez López et al., 2019). The MLP approach
is more popularly used in ANN than other approaches (Zhang et al., 1999). A study
conducted by Sánchez-Serrano et al. (2020) also found that the MLP method produces
a higher level of accuracy than the Radial Basis Function (RBF) approach.18
Following Bekesiene et al. (2021), the model used in this study randomly classifies the
data into three groups: (i) 60% training data, (ii) 20% testing data, and (iii) 20% holdout
data—or commonly called 60%-20%-20% format.19 In this format, the training data is
used to find the weights and then to build the model, while the testing data is used to
find errors and prevent overtraining during training mode. Finally, the holdout data is
used to validate the model (IBM 2021).20
There are seven independent variables used to build the model: : (i) gross profit
margin, (ii) operating profit margin, (iii) other business income ratio, (iv) other business
expense ratio, (v) positive fiscal adjustment ratio, (vi) negative fiscal adjustment ratio,
and (vii) types of annual income tax return. The dependent variable in the model is an
ordinal variable representing the three level of CTTOR: (i) low CTTOR (< 0.59%); (ii)
moderate CTTOR (0.59% to 1.19%); (iii) high CTTOR (>1.19%).
4. RESULTS AND DISCUSSIONS

4.1. Descriptive Statistics
The descriptive statistics for the seven variables, which provide the means,
medians, standard deviations, variance, minimum and maximum values, form the basis
of this study. Table 3 presents the results.
Table 3: Descriptive Statistics and Relationships between Tested Variables

Panel A. Descriptive statistics (n = 538,254)
Mean Median SD Variance Min. Max.
Gross profit margin (%) 20.83 15.17 20.04 401.50 -98.60 100.00
Operating profit margin (%) 4.49 3.02 11.57 133.86 -175.64 100.00
Other income ratio (%) 1.20 0.04 8.62 74.38 -92.59 4,308.28
Other expense ratio (%) 1.59 0.01 9.53 90.87 -70.42 4,332.31
Positive fiscal adjustment ratio (%) 7.24 0.22 21.78 474.50 -104.07 100.00
Negative fiscal adjustment ratio (%) 0.60 0.00 4.51 20.33 -189.45 100.00
CTTOR (%) 0.89 0.42 1.35 1.82 0.00 23.37
Panel B. Descriptive statistics (n = 538,226)
CTTOR Category Total
16
The linear regression model has a fixed model structure and a set of assumptions that are applied before learning
from the data. On the other hand, neural networks may estimate numerous statistical models without needing us to to
postulate beforehand specific link between the dependent and independent variables—because the form of the
relationship is determined during the model's learning process (IBM, 2021).
17
There are two ANN-based prediction application modules within IBM SPSS: MLP and Radial Basis Function (RBF)
(IBM 2021). MLP and RBF are the two most widely used neural network architectures (Ripley 1996).
18
It should be noted that the MLP method is not always more accurate. For example, Jupri and Sarno (2018) compared
four classification algorithms C4.5, Support Vector Machine (SVM), K-Nearest Neighbor (KNN) and MLP to classify the
level of compliance of taxpayers and concluded that the C4.5 algorithm is a more accurate classification algorithm.
19
The study by Bekesiene et al. (2021) found that the 60%-20%-20% partition format is more optimal than the 50%-
30%-20% and the 70%-20%-10% format.
20
The partition format that is quite common is 70% for training data and 30% for testing data (70%-30%). This
formation is usually adopted for relatively few observations (e.g., less than 1,000) and therefore does not have the data
allocated to validate the model—i.e., the holdout data.
15
Types of Low CTTOR Moderate CTTOR High CTTOR
tax return Observ. % Observ. % Observ. % Observ. %
Nil 105,606 33.9% 1,294 1.5% 1,868 1.3% 108,768 20.2%
Underpaid 182,612 58.7% 81,791 94.3% 131,708 93.9% 396,111 73.6%
Overpaid 22,978 7.4% 3,653 4.2% 6,716 4.8% 33,347 6.2%
311,196 100.0% 86,738 100.0% 140,292 100.0% 538,226 100.0%
Panel C. Relationship between variables under study (n = 538,254)
GPM OPM OIR OER PFAR NFAR CTTOR
GPM 1 .476** .023** .080** .065** .065** .349**
OPM .476** 1 -.095** .018** .072** .045** .410**
OIR .023** -.095** 1 .698** .020** .108** .074**
OER .080** .018** .698** 1 .038** .073** .004**
PFAR .065** .072** .020** .038** 1 .098** -.120**
NFAR .065** .045** .108** .073** .098** 1 .011**
CTTOR .349** .410** .074** .004** -.120** .011** 1
**. Correlation is significant at the 0.01 level (2-tailed).
Note: GPM = gross profit margin; OPM = operating profit margin; OIR = other income ratio; OER = other expense
ratio; PFAR = positive fiscal adjustment ratio; NFAR = negative fiscal adjustment ratio; CTTOR = corporate tax
turnover ratio. For types of tax returns, 28 observations are missing.
The Pearson's correlation coefficient was further employed to assess the

relationships between the variables under study. Panel B of Table 3 presents the
correlation coefficient (r) of the tested variables. The strongest positive significant
correlation was identified between the relationship between the variable other income
ratio and other expense ratio (OIR, r = 0.698,  < 0.001). The strength of the relationship
between these two variables is somewhat high because it has an r value of 0.69
(Schober et al., 2018). A significant negative correlation—although weak—was identified
between the relationship between the variable positive fiscal adjustment ratio and
CTTOR (PFAR, r = -0.120,  < 0.001). The correlational relationships among other
variables have a coefficient that varies with the tendency of a weak relationship because
it has a value of < 0.40. 21 Graph 2 facilitates the visualisation of the associations
between the level of CTTOR and business sector and firms’ size.
Panel A: Business sector (n = 538,254) Panel B: Annual turnover category (n = 538,254)
21
According to Schober et al. (2018), the correlation coefficient 0.40-0.69 indicates a moderate correlation, 0.70-0.89
indicates a strong correlation, and 0.90-1.00 indicates a powerful correlation.
16
A = Agriculture, B = Mining and quarrying, C = Manufacturing, D = Electricity and gas, E = Water supply,
sewerage, etc., G = Wholesale and retail trade, H = Transportation and storage, I = Accommodation and food
service, J = Information and communication, K = Financial and insurance, L = Others
Note: Panel A depicts a biplot displaying three levels of CTTOR and how they relate to specific business sector. The
proximity between the two legends suggests a commonality or trend. In panel A, firms in the wholesale and retail
trade sector, for instance, tend to have a low CTTOR. In contrast, businesses in the transportation, storage,
mining, and quarrying industries often have a high CTTOR. In Panel B, for instance, small firms have a stronger
tendency to report a low CTTOR, but large corporations are inclined to have a high CTTOR.
Graph 2: Visual Associations: CTTOR, Business Sectors, and Annual Turnover
For details on the number of observations and the percentage of firms in three
categories of CTTOR for each business sector under study, see Appendix 1. For details
on the share of CTTOR categories by business sectors and by size for 2014 to 2019,
see Appendix 2.
4.2. Preliminary Analysis: ANOVA and Graphical Evidence
To see whether firms that vary in annual turnover categories also differ considerably
in terms of variables under study—i.e., their gross profit margin, operating profit margin,
other income ratios, other expense ratios, positive fiscal adjustment ratios, and negative
fiscal adjustment ratios, this study employs preliminary analysis. This initial step builds
on an analysis of variance (ANOVA) and provides visual evidence to see whether firms
in four annual turnover categories also vary substantially in terms of the variables under
investigation.
Table 4 describes that small firms have the highest GPM compared to other
categories (M = 24, SD = 20.6). Large firms, on the other hand, have the lowest GPM
(M = 16.23, SD = 17.62). However, in terms of operational profitability, the OPM value
for large firms is the highest (M = 5.03, SD = 11.91), followed by the OPM value of small
firms (M = 4.77, SD = 11.69). Medium-large and large firms virtually have the same OPM
values (M = 3.99, SD = 11.34 for medium firms; M = 3.96, SD = 11.15 for the rest).
Additionally, large firms report the largest portion of other income ratio (M = 1.75, SD =
6.44), whereas small firms show the lowest (M = 0.99, SD = 6.74). It is evident that the
annual turnover and other income ratio are positively correlated—i.e., the higher the
turnover, the greater the OIR.
Further, the value of other expense ratio reported by large businesses is
likewise the largest (M = 2.54, SD = 7.00), while the ratio reported by small firms
is the lowest (M = 1.13, SD = 8.75). It is evident that there is a positive correlation
between annual turnover and other expense ratio, meaning that the bigger the
turnover, the higher the OER. In terms of positive fiscal adjustment ratio, small
firms show the largest (M = 9.50, SD = 25.78), followed by medium-large firms (M
= 6.48, SD = 20.30) and medium firms (M = 5.74, SD = 18.70). In contrast, large
firms report the lowest (M = 4.79, SD = 15.55). Nevertheless, this group shows
the largest negative fiscal adjustment ratio (M = 1.07, SD = 5.17). As
Table 4 presents, annual turnover positively correlates with the negative fiscal
adjustment ratio, with a larger turnover resulting in a higher NFAR.
Table 4: ANOVA Results of Variables Under Study by Annual Turnover

GPM (%) OPM (%)
Annual turnover Observ. Mean SD Mean SD
Small firms (IDR 5B to 15B) 224,181 24.00 20.62 4.78 11.69
Medium firms (IDR 15B to 50B) 166,171 20.41 20.16 4.00 11.35
Medium-large firms (IDR 50B to 100B) 61,572 16.84 18.59 3.96 11.16
Large firms (> IDR 100B) 86,330 16.20 17.62 5.04 11.92
17
Total 538,254 20.83 20.04 4.49 11.57
OIR (%) OER (%)
Large firms (> IDR 100B) 86,330 1.75 6.44 2.54 7.00
Total 538,254 1.20 8.62 1.59 9.53
PFAR (%) NFAR (%)
Large firms (> IDR 100B) 86,330 4.79 15.55 1.07 5.17
Total 538,254 7.24 21.78 0.60 4.51
Note: GPM = gross profit margin; OPM = operating profit margin; OIR = other income ratio; OER = other expense ratio;
PFAR = positive fiscal adjustment ratio; NFAR = negative fiscal adjustment ratio; CTTOR = corporate tax turnover
ratio. All ANOVA results in this table are statistically signigicant at  < 0.001.
Graph 3 plots the mean value of each variable evaluated throughout the fiscal
year 2014 to 2019. It can be noticed that, in general, there are parallel trends in
each panel. For instance, in terms of gross profit margin, it can be observed that
the small firms continuously reported the most significant figure compared to
other groups throughout the six-year observation period. Conversely, large
firms consistently had the lowest GPM from 2014 to 2019. Further, as described
earlier in
Table 4, large firms have the lowest figure in terms of negative fiscal adjustment
ratio. The advantage of providing the graphical illustration in Graph 3 is that the similar
tendencies seem independent from the year of observation since the order are depicted
consistently.
Panel A. Gross profit margin Panel B. Operating profit margin
Panel C. Other business income ratio Panel D. Other business expense ratio
18
Panel E. Positive fiscal adjustment ratio Panel F. Negative fiscal adjustment ratio
Note: These graphs show the results of a two-way ANOVA describing plots of the various means of the studied
variables for groups with differing annual turnover for fiscal year 2014-2019. The vertical axis depicts the scale
of estimated marginal mean scale. The horizontal axis represents fiscal years. All differences are statistically
significant at the 0.01 level.
Graph 3: The Difference of Variable Means Among Four Groups
4.3. Artificial Neural Networks (ANN)

The primary objective of this study is to examine how accurately the multilayer
perceptron (MLP) Neural Networks approach predicts the category of taxpaying
behaviour of Indonesian firms by analyzing the tax record data of Indonesian firms for
six fiscal years.
Table 5 shows the number of observations used to build the ANN model in four
datasets. As indicated earlier, following Bekesiene et al. (2021), the administrative data
in this study were divided into three groups with the following proportions: (i) training
data 60% (n = 323,394); (ii) testing 20% (n = 107,297); and (iii) holdout data 20% (n
=107,534). There are 28 observations that were excluded by the application in the
analysis. The table suggests that the largest observation is for firms with annual turnover
between IDR 5 billion to IDR 15 billion (42%) while the smallest proportion is firms with
annual turnover between IDR 50 billion to IDR 100 billion (11%).
Table 5: Case Processing Summary

Small firms Medium firms Medium-large firms Large firms
(IDR 5 billion to 15 (IDR 15 billion to 50 (IDR 50 billion to (More than IDR
billion) billion) 100 billion) 100 billion)
19
Sample N Percent N Percent N Percent N Percent
Training 135,140 60.3% 99,779 60.0% 36,816 59.8% 51,659 59.8%
Testing 44,634 19.9% 33,167 20.0% 12,236 19.9% 17,260 20.0%
Holdout 44,393 19.8% 33,215 20.0% 12,517 20.3% 17,409 20.2%
Valid 224,167 100.0% 166,161 100.0% 61,569 100.0% 86,328 100.0%
Excluded 14 10 3 2
Total 224,181 166,171 61,572 86,330
Note: the 60%-20%-20% proportion is not precise due to the excluded observations and rounding. The excluded
observations are due to missing data in the types of the annual tax return.
Table 6 describes the number of neurons in each layer and seven independent
variables used in the analysis (input layer): (i) annual tax return type, (ii) gross profit
margin, (iii) operating profit margin, (iv) other income ratio, (v) other expense ratio, (vi)
positive fiscal adjustment ratio, (vii) negative fiscal adjustment ratio. In this analysis, one
categorical variable is included in the factors group, and six scale variables are included
in the covariates category, comprising seven inputs.
Table 6: Network Information
Input Layer Factors 1 Annual tax return type
Covariates 1 Gross profit margin (%)
2 Operating profit margin (%)
3 Other income ratio (%)
4 Other expense ratio (%)
5 Positive fiscal adj. ratio (%)
6 Negative fiscal adj. ratio (%)
Number of Unitsa 9
Rescaling Method for Covariates Standardized
Hidden Layer(s) Number of Hidden Layers 2
Number of Units in Hidden Layer 1a 7
Number of Units in Hidden Layer 2a 5
Activation Function Sigmoid
Output Layer Dependent Variables 1 Three CTTOR Category
Number of Units 3
Activation Function Softmax
Error Function Cross-entropy
Note: a. Excluding the bias unit. This network information summary is identical for all four grups in this study.
Small firms (n = 224,181) Medium firms (n = 166,171)
20
Medium-large firms (n = 61,572) Large firms (n = 86,330)
Note: These graphs show the ANN models in this study. There are seven independent variables used in the input layer.
Each of the model has two hidden layers. There are seven and five nodes in the first and second hidden layers
respectively. The output layer has three units that represent the category of taxpaying behaviour. The activation
of the hidden layer function in this analysis uses a Sigmoid, while the output layer uses Softmax.
Graph 4: Structure of Neural Networks for the Model Prediction
21
The automatic architecture feature of the application shows that there are two
hidden layers in the model.22 There are seven and five nodes in the first and second
hidden layers, respectively. The output layer has three units that represent the category
of taxpaying behaviour. The activation of the hidden layer function in this analysis uses
a Sigmoid, while the output layer uses Softmax. Cross-entropy is used as an error
function since the Softmax method is used as an activation function. Graph 4 depicts
the network diagrams visualising the number of nodes in the input, hidden, and output
layers.
The size of the input nodes in Graph 4 represents the extent of the influence of the
related independent variables on the dependent variable. Larger rectangles imply a
more substantial influence of the related independent variable on the output. For
example, the annual tax return type and operating profit margin nodes seem to have a
weak influence on the CTTOR categories. These impacts will be discussed in detail
later.
Next, Table 7 presents a summary of information related to the training (and testing)
and validation results on the holdout sample. The value of the cross-entropy error is also
presented for both the training sample and the testing sample—it shows the value of the
error function minimized by the ANN model during the training phase. The reduced
cross-entropy error values for the testing sample compared to the training samples for
all groups indicate no overfitting of the training data in the network models.
Table 7: ANN Model Summary

Small firms Medium Medium-large Large firms
firms firms
Training Cross Entropy Error 42,037.16 22,814.93 9,211.94 10,952.95
Percent Incorrect Predictions 10.9% 6.4% 7.7% 6.8%
Stopping Rule Used Max. number Max. Max. number Max. number of
of epochs number of of epochs epochs (100)
(100) epochs (100) exceeded
exceeded (100) exceeded
exceeded
Training Time 0:00:10,35 0:00:07,97 0:00:03,24 0:00:04,56
Testing Cross Entropy Error 13,972.8 7,444.7 3,078.6 3,750.5
Percent Incorrect Predictions 10.8% 6.4% 7.6% 7.0%
Holdout Percent Incorrect Predictions 10.9% 6.4% 6.9% 7.1%
Dependent Variable: CTTOR Category
These results justify the role of sample testing in preventing overtraining. Based on
the information in Table 7, the proportion of incorrect predictions from the training
samples was low—i.e., 10.9%, 6.4%, 7.7%, and 6.8% for group small, medium, medium-
large, and large firms, respectively. Likewise, the proportion of incorrect predictions from
the testing samples was 10.8%, 6.4%, 7.6%, and 7.0% for small, medium, medium-
large, and large firms, respectively. For the holdout sample, the proportion of incorrect
predictions is 10.9%, 6.4%, 6.9%, and 7.1%, respectively for each group. These results
suggest that the models have performed well predicting taxpaying behaviour.
Table 8 provides information about the accuracy (i.e., confusion matrix) of the ANN
model for training, testing, and holdout data samples. The formula for sensitivity (i.e.,
22
Hidden layers allow the ANN to emulate non-linear patterns more accurately in the data. Without a hidden layer, the
ANN will behave like an ordinary linear model which cannot detect non-linear patterns (Aryadoust & Baghaei, 2016).
22
𝑇𝑃
recall or true positive rate) is 𝑇𝑃+𝐹𝑁
x 100%. The formula for the specifity (i.e., true
𝑇𝑁
negative rate) is 𝑇𝑁+𝐹𝑃
x 100%, while the formula for the accuracy of the model is
𝑇𝑁+𝑇𝑃
x 100%. Each group has a relatively high accuracy rate of 89.1% to 93.6%
𝑇𝑁+𝐹𝑃+𝑇𝑃+𝐹𝑁
in aggregate. For instance, the model accurately predicts 93.8% firms have low CTTOR
for all training, testing, and holdout data for small firms. For the moderate CTTOR
category, the model predicts with an accuracy of 72.6%, 72.7%, and 72.2%, respectively
for the training, testing, and holdout data. For firms with high CTTOR, the model predicts
with an accuracy of 89.3%, 89.7%, and 89.8% respectively for training, testing, and
holdout data.
Table 8: Accuracy of classification

Panel A
Predicted for small firms Predicted for medium firms
(IDR 5 billion to 15 billion) (IDR 15 billion to 50 billion)
Sample Observed Low Moderate High Percent Low Moderate High Percent
Correct Correct
Training Low 75,006 3,103 1,847 93.8% 53,201 1,211 1,398 95.3%
Moderate 3,472 16,862 2,876 72.6% 1,300 13,690 1,319 83.9%
High 947 2,461 28,566 89.3% 474 703 26,483 95.7%
Percent 58.8% 16.6% 24.6% 89.1% 55.1% 15.6% 29.3% 93.6%
Testing Low 24,762 1,024 604 93.8% 17,716 384 431 95.6%
Moderate 1,139 5,586 954 72.7% 418 4,520 467 83.6%
High 333 760 9,472 89.7% 183 230 8,818 95.5%
Percent 58.8% 16.5% 24.7% 89.2% 55.2% 15.5% 29.3% 93.6%
Holdout Low 24,675 1,027 599 93.8% 17,804 396 430 95.6%
Moderate 1,174 5,546 961 72.2% 455 4,602 433 83.8%
High 270 796 9,345 89.8% 156 258 8,681 95.4%
Percent 58.8% 16.6% 24.6% 89.1% 55.4% 15.8% 28.7% 93.6%
Panel B
Predicted for medium-large firms Predicted for large firms
(IDR 50 billion to 100 billion) (More than IDR 100 billion)
Sample Observed Low Moderate High Percent Low Moderate High Percent
Correct Correct
Training Low 21,191 335 472 96.3% 27,356 763 1,030 93.8%
Moderate 947 3,819 499 72.5% 634 6,064 616 82.9%
High 315 252 8,986 94.1% 275 182 14,739 97.0%
Percent 61.0% 12.0% 27.0% 92.3% 54.7% 13.6% 31.7% 93.2%
Testing Low 7,121 114 169 96.2% 9,162 299 316 93.7%
Moderate 308 1,270 159 73.1% 206 2,000 232 82.0%
High 104 70 2,921 94.4% 84 63 4,898 97.1%
Percent 61.6% 11.9% 26.6% 92.4% 54.8% 13.7% 31.6% 93.0%
Holdout Low 7,182 106 136 96.7% 9,177 296 352 93.4%
Moderate 299 1,331 139 75.2% 240 1,995 206 81.7%
High 104 81 3,139 94.4% 75 72 4,996 97.1%
Percent 60.6% 12.1% 27.3% 93.1% 54.5% 13.6% 31.9% 92.9%
Dependent Variable: CTTOR Category (i.e., low, moderate, high).
More specifically, in the training data of small firms, 75,006 firms are correctly
classified as to have low CTTOR, with 3,103 firms and 1,847 firms are falsely predicted
as to have moderate and high CTTOR respectively. In the moderate CTTOR group, the
model correctly predicts 16,862 firms, with 3,472 firms and 2,876 firms are falsely
classified as low and high CTTOR, respectively. Furthermore, in the high CTTOR
category, the model accurately predicts 28,566 firms, whereas 947 and 2,461 firms,
respectively, are incorrectly labelled as low and moderate CTTOR categories. This
model results in an overall accuracy rate of 89.1%, 89.2%, and 89.1% for training,
23
testing, and holdout sample in the group A. For the medium firms, the model correctly
predicts 95.3%, 95.6%, and 95.6% firms with moderate CTTOR for training, testing, and
holdout data, respectively. For the moderate CTTOR category, the group B model
predicts with an accuracy of 83.9%, 83.6%, and 83.8% for training, testing, and holdout
data, respectively.
For companies with a high CTTOR, the model predicts with an accuracy of 93.6%
for all training, testing, and holdout data. In medium firms’ category training data, 53,201
entities are accurately identified as having a low CTTOR, whereas 1,211 firms and 1,398
firms are incorrectly categorised as having a moderate and high CTTOR, respectively.
In the moderate CTTOR category, the model accurately predicts 13,690 businesses,
whereas 1,300 and 1,319 entities are incorrectly labelled as low and high CTTOR,
respectively. In addition, the model successfully predicts 26,483 enterprises in the high
CTTOR group, but 474 and 703 firms are misclassified in the low and moderate CTTOR
categories, respectively. This model gives medium firms’ category a total accuracy of
93.6% for all training, testing, and holdout sample data.
Similarly, as Panel B of Table 8 shows, the models for medium-large and large firms
also demonstrate high accuracy of prediction. For instance, the model in the medium-
large category results in an overall accuracy rate of 92.3%, 92.4%, and 93.1% for
training, testing, and holdout sample, respectively. Similarly, the ANN model gives large
firms’ category a total accuracy of 93.2%, 93%, and 92.9% for training, testing, and
holdout sample data, respectively.
The dependent variable in the present study comprised three levels of CTTOR: low
(1), moderate (3), and high (3). Specificity, also known as the real negative rate, is the
percentage of high-CTTOR corporations that were accurately categorised into Group 3.
Sensitivity is the fraction of actual low-CTTOR firms that were correctly placed into
Group 1 (Fawcett, 2006). When a high-CTTOR is misclassified as 1 or 2, or a low-
CTTOR is incorrectly classified as 2 or 3 (a false negative), an error has been made (a
false positive; Swets, 1988). As Graph 5 depicts, these indices are likewise visually
shown in a Receiver Operating Characteristic (ROC) curve for the dependent variable.
With a diagonal or no-discrimination line connecting the bottom left corner of the
graph to the diagonally opposite corner, ROC shows 1-specificity on the x-axis versus
sensitivity on the y-axis. Poor classification is shown by points below the no-
discrimination line, whereas successful classification is indicated by points above the
no-discrimination line (Fawcett, 2006). The formula for sensitivity (i.e., recall or true
𝑇𝑃
positive rate) is 𝑇𝑃+𝐹𝑁 x 100%. The formula for the specifity (i.e., true negative rate) is
𝑇𝑁 𝑇𝑁+𝑇𝑃
𝑇𝑁+𝐹𝑃
x 100%, while the formula for the model’s accuracy is 𝑇𝑁+𝐹𝑃+𝑇𝑃+𝐹𝑁 x 100%.
Graph 5 describes the area under the curve (AUC) values range in this study range
between 0.941 and 0.988, indicating the prediction results are very accurate (see
Appendices 3 and 4 for the cumulative gain charts and the lift charts).
Further, Table 9 shows the assessment of the independent variables in the ANN
models, which is measured by relative importance and normalized importance. The
independent variable importance in the ANN models suggests that each group has
heterogenous predictors. This study focuses on the three most important predictors.23
For instance, as shown in Table 9, for small firms, the variable positive fiscal adjustment
ratio has the highest score (0.259; normalized importance = 100%), followed by other
23
As a predictive model, the ANN can be combined with the Decision Tree model for more complete analysis (IBM,
2021). One of the popular approaches in the Decision Tree model is Chi-Squared Automatic Interaction Detection
(CHAID). In the CHAID approach, only three levels of the tree—which in this case refers to the independent variable—
are generated (IBM, 2017). Based on these considerations, the author only focuses on the three most important
variables.
24
income ratio (0.197; normalized importance = 76.1%) and operating profit margin (0.138;
normalized importance = 53.2%). Differently, for medium firms, the variable negative
fiscal adjustment ratio has the highest score (0.179; normalized importance = 100%),
followed closely by postive fiscal adjustment ratio (0.178; normalized importance =
99.4%) and other income ratio (0.178; normalized importance = 99.3%).
Panel A: ROC Curves

Note: These charts show the Receiver Operating Characteristics (ROC) curves. The ROC curve is a two-dimensional
representation of classification performance (Fawcett 2006). It provides an overview of the sensitivity and
specificity levels based on a combination of training and sampling data. A 45-degree diagonal line from the bottom
left to the top right represents the no-discrimination line. A point below the no-discrimination line indicates an
inaccurate classification, and a point above the no-discrimination line indicates an effective classification result
(Fawcett 2006).
Panel B: Area Under the Curve

CTTOR Low CTTOR 0.970 0.974 0.981 0.983
Category Moderate CTTOR 0.941 0.965 0.963 0.973
High CTTOR 0.977 0.981 0.984 0.988
Note: The area under the curve (AUC) values range in this study range between 0.941 and 0.988. In general, the AUC
values are classified into five categories: 0.50–0.60 (fail); 0.60–0.70 (less accurate); 0.70–0.80 (fairly accurate);
0.80–0.90 (accurate); and 0.90–1.00 (very accurate). Thus, the AUC value of the study shows that this prediction
result is very accurate.
Graph 5: ROC Curves and Area Under the Curve
25
Meanwhile, for medium-large firms, the variable other expense ratio has the highest
score (0.222; normalized importance = 100%), followed closely by negative fiscal
adjustment ratio (0.207; normalized importance = 93.0%) and operating profit margin
(0.201; normalized importance = 90.5%). Finally, for large firms, the variable other
expense ratio has the highest score (0.208; normalized importance = 100%), followed
closely by operating profit margin (0.195; normalized importance = 93.9%) and negative
fiscal adjustment ratio (0.191; normalized importance = 91.9%).
Interestingly, while the three most influential factors appear to be somewhat varied,
the two variables with the lowest influence are similar—i.e., the gross profit margin and
the annual tax return type. For instance, for small firms, gross profit margin has an
importance score of 0.070 (normalized importance = 26.9%) and the annual tax return
type has importance score of 0.110 (normalized importance = 42.6%). For medium firms,
gross profit margin has importance score of 0.031 (normalized importance = 17.1%),
and the annual tax return type has an importance score of 0.098 (normalized importance
= 54.5%). The same patterns apply to both medium-large and large firms. These results
imply that the prediction of taxpaying behaviour of Indonesian firms does not depend on
the gross profit margin and the type of annual tax return.
Table 9: Independent Variable Importance

Panel A
Small firms Medium firms
(IDR 5 billion to 15 billion) (IDR 15 billion to 50 billion)
Importance Normalized Importance Normalized
Importance Importance
Annual tax return type 0.110 42.6% 0.098 54.5%
Gross profit margin (%) 0.070 26.9% 0.031 17.1%
Operating profit margin (%) 0.138 53.2% 0.172 95.8%
Other income ratio (%) 0.197 76.1% 0.178 99.3%
Other expense ratio (%) 0.114 43.9% 0.165 92.2%
Positive fiscal adj. ratio (%) 0.259 100.0% 0.178 99.4%
Negative fiscal adj. ratio (%) 0.113 43.7% 0.179 100.0%
Panel B
Medium-large firms Large firms
(IDR 50 billion to 100 billion) (More than IDR 100 billion)
Importance Normalized Importance Normalized
Importance Importance
Annual tax return type 0.029 13.2% 0.036 17.4%
Gross profit margin (%) 0.045 20.1% 0.027 13.1%
Operating profit margin (%) 0.201 90.5% 0.195 93.9%
Other income ratio (%) 0.176 79.3% 0.178 85.6%
Other expense ratio (%) 0.222 100.0% 0.208 100.0%
Positive fiscal adj. ratio (%) 0.120 53.8% 0.166 79.9%
Negative fiscal adj. ratio (%) 0.207 93.0% 0.191 91.9%
For easier comprehension, the results described in Table 9 are diagrammatically

represented in Graph 6. As shown in the graph, for instance, the variable positive fiscal
adjustment ratio, other income ratio, and OPM are the most three influential variables in
predicting the CTTOR category for small firms. The variable positive fiscal adjustment
ratio, negative fiscal adjustment, and other income ratio are the three most important
variables in predicting the CTTOR category for medium firms. It is interesting to recall
that medium-large and large firms have similar three most important factors: other
expense ratio, negative fiscal adjustment ratio, and operating profit margin. Also, as
26
Graph 6 shown, the two variables with the lowest influence for all four groups are the
gross profit margin and annual tax return type variables.
The findings indicate that for small firms, positive fiscal adjustment ratio, other
income ratio, and OPM have stronger relationships with taxpaying behaviour than other
income ratio and negative fiscal adjustment ratio. Businesses in this category reported
the highest mean value for positive fiscal adjustment ratio—i.e., 9.5%, meaning that
companies in this category will increase their taxable revenue by IDR 95.000 for every
IDR one million annual turnover. This number is higher than the population's average
(7.24%). Firms in this category, in contrast, reported the lowest other income ratio, at
only 1% of turnover, meaning that firms in this category will report other business income
of IDR 1,000 for every IDR one million in annual turnover. This proportion is lower than
the population average (1.2%), with firms with an annual turnover of more than IDR 100
billion reporting the largest other income ratio (1.7%).
Note: The graph dispalys the result of ANN's analysis separately for four groups of firms. The value of IVI indicates
that the variable positive fiscal adjustment is of the most significant importance in the prediction of taxpaying
behaviour for firms with annual turnover between IDR 5 billion and IDR 15 billion, while the negative fiscal
adjustment is of the most significant importance for firms with annual turnover between IDR 55 billion and IDR
50 billion. For firms with an annual turnover of more than IDR 50 billion, the variable other business expense is
the most important independent variable for predicting taxpaying behaviour.
Graph 6: Independent Variable Importance (IVI) by Annual Turnover Category
Further, operating profit margin is the third most crucial predictor for firms in this
category. This group's OPM is within the range of 4.78%—slightly higher than the
27
population’s mean of 4.49%—meaning that businesses in this category declared
operating profits of IDR 47,800 for every IDR one million in sales. These findings suggest
that part 5 Form 1771-I (positive fiscal adjustment), part 1e Form 1771-I (other business
income), and part 1c Form 1771-I (operating expense) are potential areas of
misreporting for small businesses when it comes to annual income tax returns.24
The best predictors of taxpaying behaviour for medium firms are negative fiscal
adjustment ratio, positive fiscal adjustment ratio, and other income ratio. Negative fiscal
adjustment ratio for firms in this category is 0.52%, which indicates that entities in this
group reduced their taxable income of IDR 5,200 for every IDR one million sales. This
percentage falls just short of the population mean of 0.60%. It should be noted that there
is a positive correlation between negative fiscal adjustment ratio and turnover for
businesses in this category—i.e., the negative fiscal adjusment increases with more
turnover. Like for small firms, positive fiscal adjustment ratio is also a significant predictor
of taxpaying behaviour for medium firms. Medium firms make average positive fiscal
adjustments of IDR 57,400 for each IDR one million sales (5.74%). This number is
slightly below the population average of 7.24%. Further, the predictor comparable to that
of small firms is other income ratio. Medium firms reported an OIR of 1.13%, indicating
that medium firms report additional business income of IDR 11,300 for every IDR one
million turnover. These findings imply that potential areas of misreporting for medium
firms are part 6e Form 1771-I (negative fiscal adjustment items), part 5m Form 1771-I
(positive fiscal adjustment items), and part 1e Form 1771-I (other business income).
The best predictors of taxpaying behaviour for medium-large firms are other
expense ratio, negative fiscal adjustment ratio, and operating profit margin. Other
expense ratio for medium-large firms is 1.95%, meaning that entities in this category
claimed additional business expenses of IDR 19,500 for every IDR one million sales.
This percentage is higher than the population mean of 1.59%. It should be noted that
there is a positive correlation between other expense ratio and turnover—i.e., the
additional business expense increases with more turnover. Further, medium-large firms
make average negative fiscal adjustments of IDR 6,700 for each IDR one million sales
(0.60%). This number is slightly higher than the population average of 0.67%. The
predictor comparable to that of small firms is operating profit margin. Medium-large firms
reported an OPM of 3.96%, indicating that medium-large firms report operational profits
of IDR 39,600 for every IDR one million turnover. These results suggest that part 1c
Form 1771-I (operating expense), part 1e Form 1771-I (other business expenditure),
and part 6e Form 1771-I (negative fiscal adjustment) are possible areas of misreporting
for medium-large enterprises.
Interestingly, the most prominent predictors for large corporations appear to
resemble those of medium-large firms: other expense ratio, operating profit margin, and
negative fiscal adjustment ratio. Large firms reported the highest portion of other
expense ratio among other categories. Other expense ratio for large firms is 2.54%,
meaning that large firms claimed additional business expenses of IDR 25,400 for every
IDR one million sales. This ratio is 60% higher than the population mean of 1.59%. The
predictor comparable to that of medium-large firms—and small firms—is the operating
profit margin. Large corporations reported an OPM of 5.04%, indicating that large firms
report operational profits of IDR 50,400 for every IDR one million turnover. Further, large
firms make average negative fiscal adjustments of IDR 10,700 for each IDR one million
sales (1.07%). This number is the highest among other categories as the relationship
between negative fiscal adjustment and turnover is positive. These findings imply that
24
Note that gross profit margin is the least important predictor for firms in this group. Operating expense is the single
feature that distinguishes operating profit margin from gross profit margin.
28
potential areas of misreporting for large corporations are part 1e Form 1771-I (other
business expense), part 1c Form 1771-I (operating expense) and part 6e Form 1771-I
(negative fiscal adjustment). Interestingly, while other business income is a prominent
predictor for small and medium firms, additional business expense is crucial for medium-
large and large firms. A possible explanation for this might be that large firms have more
income visibility and therefore have less control over misreporting of income streams.
For this reason, the potential areas of misreporting are expenses over which large firms
have more control. In contrast, small and medium firms have less visibility in income and
therefore have more control over misreporting concerning income streams.
5. CONCLUSIONS
This study utilised an ANN model to predict the taxpaying behaviour of Indonesian
firms. Initially, this paper examined whether firms in four size-based categories—i.e.,
small, medium, medium-large, and large firms—also differ considerably in terms of their
gross profit margin, operating profit margin, other income ratios, other expense ratios,
positive fiscal adjustment ratios, and negative fiscal adjustment ratios. This study found
that firms that vary in annual turnover categories also differ considerably in the variables
under study. The results suggest that small firms have the highest gross profit margin
compared to other groups. However, in terms of operational profitability, large firms
reported the highest. Large firms also declared the most considerable portion of other
business income and the largest share of other business expenses. Interestingly, small
firms show the most significant positive fiscal adjustment ratio, while large corporations
declare otherwise. In contrast, large firms report the largest negative fiscal adjustment
ratio.
This study supports the notion that artificial intelligence approaches is superior to
that of conventional statistical methods for addressing various issues, particularly those
involving nonlinear patterns (see, for example, Bahrammirzaee, 2010; Chen et al., 2011;
Lin et al., 2012; Aryadoust and Baghaei, 2016; Sánchez-Serrano et al., 2020). This
paper demonstrated that the ANN models accurately predict the taxpaying behaviour of
Indonesian firms across four groups. The classification accuracy rate was high, with an
overall 92.2% accuracy in categorising the firms into the low, moderate, and high
CTTOR. This study used the MLP module from ANN and built on a model with a 60%–
20%–20% formation. The results also showed that the channels responsible for
taxpaying behaviour vary for different groups. For small firms, the three most influential
factors in predicting taxpaying behaviour are the positive fiscal adjustment ratio, other
income ratio, and OPM. This means that the extent to which firms positively adjust their
commercial, the presence of other business income, and the share of the cost of goods
sales are critical factors for small firms. The accuracy rate, significant predictors, and
areas of concern for all groups can be summarised in Table 10, as follows.
Table 10: Summary: Accuracy Rate, Predictors, and Areas of Concern

Prominent predictors by firms’ category
(IDR 5B to 15B) (IDR 15B to 50B) (IDR 50B to 100B) (> IDR 100B)
Accuracy rate* 89.1% 93.6% 93.1% 92.9%
Annual return type No No No No
GPM No No No No
OPM Yes No Yes Yes
OIR Yes Yes No No
OER No No Yes Yes
PFAR Yes Yes No No
NFAR No Yes Yes Yes
29
Areas of concern Part 1c Form Part 1e Form Part 1c Form 1771-I, part 1e Form
within annual 1771-I, part 1e 1771-I, part 5m 1771-I, and part 6e Form 1771-I
income tax return Form 1771-I, and Form 1771-I, and
part 5 Form 1771- 6e Form 1771-I
I
Note: * accuracy rate based on holdout sample; GPM = gross profit margin; OPM = operating profit margin; OIR =
other income ratio; OER = other expense ratio; PFAR = positive fiscal adjustment ratio; NFAR = negative fiscal
adjustment ratio
These findings, while preliminary, suggest that it is possible to minimise the

monitoring time and expense since the model is applied to a national-level big data set
and directly exhibits several areas of concern in the annual tax returns. The findings
would assist decision-makers in tax administrations about potential areas of
misreporting, enabling them to develop evidence-based and effective policy actions.
6. IMPLICATIONS AND LIMITATIONS

6.1. Implications
Although further research is required to delve deeply into the effectiveness of
machine learning in comprehending taxpaying behaviour, this work has revealed new
insights into how potential areas of misreporting can be identified with minimal cost.
Several practical implications for the Indonesian tax authorities have been shown by
illustrating how various variables affect taxpaying behaviour across different groups,
including their areas of concern in the annual income tax returns (as presented in Table
10).
This study provides findings relevant to making targeted strategic decisions—for
example, the results of this analysis can be used to segment groups of firms deemed
necessary to increase their level of tax payment, focusing on their variable of interests.
These may support the development of effective treatment strategies to improve
compliance rates. Particularly in developing a compliance risk management plan, with
supervised machine learning, the case selection algorithm may highlight complex data
patterns linked to successful case outcomes while deemphasizing those that did not
(Brondolo et al., 2022). It would be of assistance because the DGT has not yet used
artificial intelligence in its operations (ADB, 2022). As Dom et al. (2022) argue, it would
be intriguing to get insights into the possibilities of more advanced emerging
technologies, like blockchain and artificial intelligence, in lower-capacity settings when
their usage becomes more widespread.
6.2. Limitations
Due to practical constraints, this study makes no distinction between legal tax
avoidance and illegal tax evasion. In addition to using limited variables, this article does
not address the impact of enforcement or monitoring activities. The findings also raise
intriguing questions regarding the nature and the causal relationships between the
predictors and the conforming and non-conforming tax avoidance among Indonesian
firms. Unfortunately, this study cannot provide further detailed information on, for
example, why—or how—factors such as positive and negative fiscal adjustment,
operating profit margin, other business income, and other business expenses are the
most critical factors in determining taxpaying behaviour, while in contrast, gross profit
margin and the type of tax return are negligible. Studies with different empirical data and
analytical approaches are needed to answer those questions. Therefore, a further study
with more focus on the causal inference among the identified predictors and
underreporting behaviour is suggested.
30
Acknowledgment
I thank the Assistant to the Minister of Finance for Tax Compliance, Yon Arsal; Director of Tax
Potential Revenue and Compliance DGT, Ihsan Priyawibawa; Deputy Director of Policy Impact
DGT, Eureka Putra; as well as all members of Taxpayer Behaviour Studies team for their support
and assistance. The views expressed here are those of the author and do not necessarily reflect
those of the individuals or organisations acknowledged here.
31
REFERENCE
Agostinelli, F., Hoffman, M., Sadowski, P., & Baldi, P. (2015). Learning Activation
Functions to Improve Deep Neural Networks. arXiv 2014, arXiv:1412.6830
Alm, J. (2019). Can Indonesia Reform its Tax System. Tulane Economics Working Paper
Series Working Paper 1906.
Aryadoust, V., & Baghaei, P. (2016). Does EFL Readers' Lexical And Grammatical
Knowledge Predict Their Reading Ability? Insights From A Perceptron Artificial
Neural Network Study. Educational Assessment, vol. 21, no. 2, pp. 135–156.
Asian Development Bank (ADB) (2022). A Comparative Analysis of Tax Administration
in Asia and the Pacific. Asian Development Bank 6 ADB Avenue, Mandaluyong
City, 1550 Metro Manila, Philippines.
Castellón González, P., & Velásquez, J. D. (2013). Characterization And Detection Of
Taxpayers With False Invoices Using Data Mining Techniques. Expert Systems
with Applications 40 (5):1427–1436.
Badertscher, B. A., Katz, S. P., Rego, S. O., & Wilson, R. J. (2019). Conforming Tax
Avoidance and Capital Market Pressure. The Accounting Review, 94(6), 1–30.
Bahrammirzaee, A. (2010). A Comparative Survey of Artificial Intelligence Applications
in Finance: Artificial Neural Networks, Expert Systems and Hybrid Intelligent
Systems. Neural Computing and Applications, vol. 19, no. 8, pp. 1165-1195.
Bekesiene, S., Smaliukiene, R., & Vaicaitiene, R. (2021). Using Artificial Neural
Networks in Predicting the Level of Stress Among Military Conscripts.
Mathematics, vol. 9, no. 6, articles 626.
Bird, R. M. (2010). Taxation and Development. World Bank, Washington, DC. © World
Bank. https://openknowledge.worldbank.com/handle/10986/10150 License: CC
BY 3.0 Unported.
Brondolo, J., Chooi, A., Schloss, T., & Siouclis, A. (2022). Compliance Risk
Management: Developing Compliance Improvement Plans. Technical Notes and
Manuals (International Monetary Fund).
Chen, J.H., Su, M.C., Chen, C.Y., Hsu, F.H., & Wu, C.C. (2011). Application of Neural
Networks for Detecting Erroneous Tax Reports from Construction Companies.
Automation in Construction, vol. 20, no. 7, pp. 935–939.
Cook, T. R. (2020). Neural Networks. In P. Fuleky (Ed.), Macroeconomic Forecasting in
the Era of Big Data: Theory and Practice.
Devos, K. (2014). Factors Influencing Individual Taxpayer Compliance Behaviour,
Springer, Dordrecht, Netherlands.
Dom, R., Custers, A., Davenport, S., & Prichard, W. (2022). Innovations in Tax
Compliance: Building Trust, Navigating Politics, and Tailoring Reform.
Washington, DC: World Bank. doi:10.1596/978-1-4648-1755-7. License:
Creative Commons Attribution CC BY 3.0 IGO.
Dyreng, S. D., Hanlon, M., & Maydew, E. L. (2008). Long-Run Corporate Tax Avoidance.
The Accounting Review, 83(1), 61–82.
Directorate General of Taxation (DGT) (2015). Strategic Plan of the Directorate General
of Taxation Years 2015–2019, Decree of Director General of Taxation Number
95/PJ/2015, Jakarta: DGT.
Directorate General of Taxation (DGT) (2020). Buku Statistik Perpajakan 2020 (in
Bahasa).
Directorate General of Taxation (DGT) 2021. Annual Report 2020: Constantly optimizing
opportunities amid challenging times, Jakarta: DGT.
32
Directorate General of Taxation (DGT) (2022). CRMBI Langkah Awal Menuju Data
Driven Organization. Jakarta: Direktorat Jenderal Pajak (in Bahasa).
Directorate General of Taxes (DGT) (2021). Tax statistics book 2020, Directorate
General of Taxes, Jakarta.
Fawcett, T. (2006). An introduction to ROC analysis. Pattern Recognition Letters, vol.
27, no. 8, pp. 861–874.
Fayyad, U., G. Piatetsky-Shapiro, and P. Smyth. (1996). From Data Mining to
Knowledge Discovery in Databases. AI Magazine, 17(3), 37.
https://doi.org/10.1609/aimag.v17i3.1230
Gebhart, M. S. (2017). Measuring Corporate Tax Avoidance - An Analysis of Different
Measures. Junior Management Science, 2(2), 43–60.
doi:10.5282/jums/v2i2pp43-60
Goumagias, N. D., Hristu-Varsakelis, D., and Saraidaris, A. (2012). A Decision Support
Model for Tax Revenue Collection in Greece. Decision Support Systems 53
(1):76–96.
Graupe, D. (2013). Principles of Artificial Neural Networks. World Scientific Publishing
Co. Pte. Ltd, 27 Warren Street, Suite 401-402, Hackensack, NJ 07601
Hanlon, M., & Heitzman, S. (2010). A Review of Tax Research. Journal of Accounting
and Economics, 50(2–3), 127–178.
Huang, Z. (2018). Discussion on the Development of Artificial Intelligence in
Taxation. American Journal of Industrial and Business Management, 8, 1817-
1824. doi: 10.4236/ajibm.2018.88123.
Haykin, S.S. (1999). Neural Networks: A Comprehensive Foundation, 2nd edition,
Macmillan College Publishing, New York.
Heij, G. (2001). The 1981-83 Indonesian Income Tax Reform Process: Who Pulled The
Strings? Bulletin of Indonesian Economic Studies, vol. 37, pp. 233251.
International Business Machines (IBM). (2017). IBM SPSS Decision Trees 25, IBM
Corporation, accessed June 15, 2021,<ftp://public.dhe.ibm.com/software/
analytics/spss/documentation/statistics/25.0/en/client/Manuals/IBM_SPSS_Dec
ision_Trees.pdf>.
International Business Machines (IBM). (2021). IBM SPSS Neural Networks 28, IBM
Corporation, accessed 14 July 2022,<https://www.ibm.com/docs/en/
SSLVMB_28.0.0/pdf/ IBM_SPSS_Neural_Network.pdf.>.
International Monetary Fund (IMF). (2014). Tax Policy And Administration – Setting A
Strategy For The Coming Years. Washington, D.C.: IMF.
Jang, S.B. (2019). A Design of A Tax Prediction System Based on Artificial Neural
Network'. Paper presented at the International Conference on Platform
Technology and Service (PlatCon), Jeju, 28-30 January.
Jupri, M., & Sarno, R. (2018). Taxpayer Compliance Classification Using C4.5, SVM,
KNN, Naive Bayes And MLP. Paper presented at the International Conference
on Information and Communications Technology (ICOIACT), Yogyakarta, 6-7
March.
Mills, L., Erickson, M. M., & Maydew, E. L. (1998). Investments in Tax Planning. The
Journal of the American Taxation Association; Sarasota Vol. 20, Issue 1, (Spring
1998): 1–20.
Murorunkwere, B. F., Tuyishimire, O., Haughton, D., & Nzabanita, J. (2022). Fraud
Detection Using Neural Networks: A Case Study of Income Tax. 14 (6):168
Lin, C.H., Lin, I.C., Wu, C.H., Yang, Y.C., & Roan, J. (2012). The Application of Decision
Tree and Artificial Neural Network to Income Tax Audit: The Examples of Profit-
33
Seeking Enterprise Income Tax and Individual Income Tax in Taiwan. Journal of
the Chinese Institute of Engineers, vol. 35, no. 4, pp. 401–411.
Neuman, W.L. (2011). Social Research Methods: Qualitative And Quantitative
Approaches, 7th edition, Allyn & Bacon, Massachusetts.
OECD (2014). Measures of Tax Compliance Outcomes: A Practical Guide, OECD
Publishing, Paris, available http://dx.doi.org/10.1787/9789264223233-en.
OECD (2016). Advanced Analytics for Better Tax Administration: Putting Data to Work,
OECD Publishing, Paris. http://dx.doi.org/10.1787/9789264256453-en
OECD (2020a). Tax Administration 3.0: The Digital Transformation of Tax
Administration, OECD, Paris. http://www.oecd.org/tax/forum-on-tax-
administration/publications-and-products/tax-administration-3-0-the-digital-
transformation-of-tax-administration.htm
OECD (2020b). Tax and Fiscal Policy in Response to the Coronavirus Crisis:
Strengthening Confidence and Resilience, Organization for Economic Co-
operation and Development, accessed 21 July
2022,<http://www.oecd.org/coronavirus/en/>.
OECD. (2022). Tax Administration 2022: Comparative Information on OECD and other
Advanced and Emerging Economies, OECD Publishing, Paris,
https://doi.org/10.1787/1e797131-en.
Pérez López, C., Delgado Rodríguez, M. J., & de Lucas Santos, S. (2019). Tax Fraud
Detection through Neural Networks: An Application Using a Sample of Personal
Income Taxpayers. Future Internet 11 (4).
Prichard, W., Custers, A., Dom, R. et al. (2019). Innovations in Tax Compliance:
conceptual framework. Innov Tax Compl Concept Framew. https:// doi. org/ 10.
1596/ 1813- 9450- 9032
Rahimikia, E., Mohammadi, S., Rahmani, T., & Ghazanfari, M. (2017). Detecting
Corporate Tax Evasion Using A Hybrid Intelligent System: A Case Study of Iran.
International Journal of Accounting Information Systems 25:1–17.
Ripley, B.D. (1996). Pattern Recognition And Neural Networks. Cambridge University
Pres, Cambridge.
Rosid, A., Evans, C., & Tran-Nam, B. (2018). Tax Non-Compliance and Perceptions of
Corruption: Policy Implications for Developing Countries. Bulletin of Indonesian
Economic Studies, 54(1), 25–60. doi:10.1080/00074918.2017.1364349
Rosid, A., & Ariyani, F. (2022). Does ‘Information Reporting’ Really Matter for Tax
Compliance? The Case of Indonesia (May 19, 2022). Available at SSRN:
https://ssrn.com/abstract=4114000 or http://dx.doi.org/10.2139/ssrn.4114000
Salihu, I. A., Obid, S. N. S., & Annuar, H. A. (2013) Measures of Corporate Tax
Avoidance: Empirical Evidence from An Emerging Economy. International
Journal of Business and Society, 14(3):412–427,.
Sánchez-Serrano, J.R., Alaminos, D., Garcia-Lagos, F., & Callejón-Gil, A.M. (2020).
Predicting Audit Opinion In Consolidated Financial Statements With Artificial
Neural Networks. Mathematics, vol. 8, no. 8, articles 1288.
Saragih, A. H., Reyhani, Q., Setyowati, M. S., & Hendrawan, A. (2022). The Potential of
An Artificial Intelligence (AI) Application for the Tax Administration System’s
Modernization: The Case of Indonesia. Artificial Intelligence and Law.
doi:10.1007/s10506-022-09321-y
Schober, P., Boer, C., & Schwarte, L.A. (2018). Correlation Coefficients: Appropriate
Use and Interpretation. Anesthesia & Analgesia, vol. 126, no. 5, pp. 1763–1768.
Slemrod, J. (2007). Cheating Ourselves: The Economics of Tax Evasion. The Journal of
Economic Perspectives, Vol. 21, No. 1 (Winter, 2007), pp. 25–48
34
Slemrod, J. (2016). Tax Compliance and Tax Evasion. In The New Palgrave Dictionary
of Economics (pp. 1-6). London: Palgrave Macmillan UK.
Swets, J. A. (1988). Measuring the Accuracy of Diagnostic Systems. Science, 240,
1285–1293.
Vochozka, M., Rowland, Z., Suler, P., & Marousek, J. (2020). The Influence of the
International Price of Oil on the Value of the EUR/USD Exchange Rate. Journal
of Competitiveness, 12(2), 167–190. https://doi.org/10.7441/joc.2020.02.10
Wang, Y., Wang, P. (2020). New Personal Tax Collection Management System Based
on Artifcial Intelligence and Its Application in The Middle Class. J Phys Conf Ser
1574:1–8. https://doi.org/10.1088/1742- 6596/1574/1/012105
Warner, G., Wijesinghe, S., Marques, U., Badar, O., Rosen, J., Hemberg, E., % O’Reilly,
U.-M. (2015). Modeling Tax Evasion with Genetic Algorithms. Economics of
Governance 16 (2):165-178
Wendler, T., & Gröttrup, S. (2021). Data Mining with SPSS Modeler: Theory, Exercises
and Solutions. Springer Nature Switzerland AG 2016,
https://doi.org/10.1007/978-3-030-54338-9
World Bank (2016). Resilience through Reform. Indonesia Economic Quarterly. Jakarta,
Indonesia: The World Bank.
York, S. (2011). A Risk-Based Approach to Large Businesses. In M. S. Khwaja, R.
Awasthi, & J. Loeprick (Eds.), Risk-Based Tax Audits: Approaches and Country
Experiences. The Office of the Publisher, The World Bank 818 H Street NW,
Washington, DC 20433.
Zhang, G., Hu, M.Y., Patuwo, B.E., & Indro, D.C. (1999). Artificial Neural Networks In
Bankruptcy Prediction: General Framework and Cross-Validation Analysis',
European Journal of Operational Research, vol. 116, no. 1, pp. 16–32.
35
APPENDICES
Appendix 1. CTTOR categories by business sectors
CTTOR category
Business sector Total
Low Moderate High
Agriculture 5,958 56% 1,148 11% 3,601 34% 10,707 100%
Mining and quarrying 3,223 45% 760 11% 3,222 45% 7,205 100%
Manufacturing 46,512 49% 20,068 21% 27,816 29% 94,396 100%
Electricity and gas 1,613 68% 132 6% 642 27% 2,387 100%
Water supply, sewerage, etc. 739 43% 198 12% 772 45% 1,709 100%
Wholesale and retail trade 192,364 63% 48,451 16% 63,520 21% 304,335 100%
Transportation and storage 11,257 45% 3,384 13% 10,538 42% 25,179 100%
Accomm. & food service 6,752 47% 2,692 19% 4,881 34% 14,325 100%
Information and 3,074 39% 1,080 14% 3,716 47% 7,870 100%
communication
Financial and insurance 6,835 58% 972 8% 3,879 33% 11,686 100%
Others 32,869 56% 7,853 13% 17,705 30% 58,427 100%
311,196 58% 86,738 16% 140,292 26% 538,226 100%
Appendix 2. CTTOR categories by business sectors and by size for 2014 to 2019
Note: A = Agriculture, B = Mining and quarrying, C = Manufacturing, D = Electricity and

gas, E = Water supply, sewerage, etc., G = Wholesale and retail trade, H =
Transportation and storage, I = Accommodation and food service, J = Information
and communication, K = Financial and insurance, L = Others.
This graph displays the proportion of businesses in each industry by the three CTTOR
categories: low (<0.59%), moderate(0.59% to 1.19%), and high (>1.19%). The graph
36
is classified into four categories, based on annual turnover size. For instance, the
graphs show that firms in the wholesale and trade sector (G) have the lowest
percentage of firms having high CTTOR—i.e., only 20 percent—across categories.
Appendix 3: Cumulative gain chart

Note: The graph depicts the cumulative gains, which is the presentence of accurate classifications provided by the ANN
model versus accurate classifications that may occur by chance (i.e. without using the model)). For example, the
third point on the curve for the moderate CTTOR category in Group A is at (30%, 92%), meaning that if the
network scores a dataset and sorts all of the cases by predicted pseudo-probability of failure, it would be expected
that the top 30% would contain approximately 92% of all of the cases that actually fall into the moderate CTTOR
category. Gain is a measure of the effectiveness of a classification model, calculated as the proportion of correct
predictions made with the model, versus the percentage of right predictions gained without a model (baseline)
(baseline). The more above the basline a curve sits, the bigger the gain. A bigger overall gain signals better
performance.
37
Appendix 4: Lift chart
Note: Lift charts, similar to gain charts, are visual tools for assessing the performance of classification models.
However, in contrast to the confusion matrix that tests models for the entire data, the gain or lift chart analyses
model performance in a segment of the data. A lift chart leverages a piece of information to present a clear
visual of the benefit of using a model in contrast to not using a model. For instance, in Group A, the factors from
the gains diagram are utilised to compute the lift factor (i.e., the benefit): the lift at 92% for the moderate
CTTOR category is 92%/30% = 3.1.
38
DOES GOVERNMENT EFFECTIVENESS MODERATE
THE RELATIONSHIP BETWEEN REGULATORY
QUALITY AND TAX COMPLEXITY?: A TALE OF A
HUNDRED NATIONS
Muhammad Dahlan
Direktorat Jenderal Pajak, Ph.D. Candidate at Cardiff University, United Kingdom. Email:
Dahlanm@cardiff.ac.uk
ABSTRACT
This research aims to analyze the relationship between regulatory quality and tax complexity
worldwide and whether this relationship is moderated by government effectiveness. The hypothesis
is that robust regulations would promote simplicity and efficiency in the tax system. This study
incorporates 100 countries as a sample, derived from the tax complexity index developed in 2016.
Further, a moderated regression using SPSS PROCESS Macro is used to test the hypotheses and
generate findings. This study shows that regulatory quality relates to the tax complexity in the
countries, and the effectiveness of the government indeed moderates this relationship. The tax
authority may use the findings to address the importance of the quality of regulations in shaping an
efficient tax system.
Keywords: tax complexity, regulatory quality, government effectiveness, moderation, governance index
Penelitian ini berujuan untuk mengetahui hubungan antara kualitas pembentukan peraturan dan
kompleksitas pajak dan untuk menguji apakah hubungan tersebut di moderasi oleh variable
efektivitas pemerintahan. Hipotesis yang dibangun adalah bahwa pengaturan peraturan yang
berkualitas akan mendorong efisiensi dan kesederhanaan system pajak. Studi ini menggunakan
sampel 100 negara yang diperoleh dari penelitian tentang index kompleksitas pajak. Selanjutnya,
pengujian regresi dengan menggunakan variabel moderator diperoleh dengan aplikasi SPSS
PROCESS. Hasil dari penelitian didapatkan kesimpulan bahwa kualitas peraturan berpengaruh
terhadap kompleksitas system pajak dan pengaruh ini semakin diperbesar dengan adanya variabel
efektivitas pemerintahan di suatu negara. Otoritas pajak dapat menggunakan hasil penelitian ini
untuk memberikan informasi tambahan mengenai pentingnya kualitas peraturan dalam
menciptakan system pajak yang lebih efisien.
Kata kunci: kompleksitas pajak, kualitas peraturan, efektivitas pemerintahan, moderasi, indeks tata kelola
1. INTRODUCTION
Benjamin Franklin once claimed that nothing can be said to be certain in this world
except death and taxes. However, in modern society, taxes are not only inevitable but
also complex. Hence, tax complexity arises as a trade-off between tax system design
and reform regarding tax fairness and efficiency faced by the taxpayers (Carnes &
Cuccia, 1996; Kaplow, 1994; Stantcheva, 2021). The perceived tax complexity may be
correlated with uncertainty in the tax system and the frustration of the taxpayer (Abeler
& Jäger, 2015; Krause, 2000). Also, the perception of fairness for the taxpayer is
beneficial in increasing the level of tax compliance (Beck et al., 1991). Thus, complexity
would lead to the non-compliance behavior of the taxpayers.
39
Concerns about tax complexity emerged in some tax authorities, such as the United
States (Ingraham & Karlinsky, 2005), the United Kingdom (Budak & James, 2018),
Australia (McKerchar, 2005), and Asia regions (Deloitte, 2014). In addition, the
consequences of tax system complexity may discourage foreign direct investment
(Müller & Voget, 2012), negatively impact economic growth (Collier et al., 2018), and
encourage tax avoidance (Budak & James, 2018). As a result, tax complexity is
considered a worldwide issue and needs a comparative perspective to address
(Freudenberg et al., 2012; McKerchar, 2005).
As we can notice from previous research, the complexity of the tax system is
addressed in many ways. For example, from the perspective of compliance cost (J. B.
Slemrod & Blumenthal, 1996), multi-facet of the tax system (J. Slemrod, 2005), and a
single country assessment (Saad, 2014). As such, a cross-country level of measurement
is needed to investigate the tax complexity around the world.
Hoppe et al. (2021) developed a tax complexity index (TCI) to capture the
complexity of corporate income tax that Multinational Enterprises (MNEs) need to face
in some countries. The latent variables used in TCI were developed initially based on
Hoppe et al. (2018) study, which consists of tax code complexity (meaning that
complexity derives from the tax laws) and tax framework complexity (complexity is a
result of the legislative and administrative burden of the tax system). Later, the
dimensions of tax complexity are constructed to build a TCI from cross-country data.
Therefore, TCI developed by Hoppe et al. (2021) is the most comprehensive approach
to measure the complexity level among tax authorities worldwide.
TCI can range from 0 (tax system is not complicated) to one (an extremely complex
tax system). A quick result of the ten countries with the most efficient tax system is
presented in Graph 1 and the ten most complicated tax systems in Graph 2.
0,00 0,05 0,10 0,15 0,20 0,25 0,30

Jersey
Nicaragua
Estonia
Mauritius
Yemen, Rep.
Hong Kong SAR, China
Singapore
Liechtenstein
Bulgaria
Lithuania
Graph 1. Ten Countries with the Most Efficient Tax System

Source: Elaborated from the Tax Complexity Index (Hoppe et al., 2021)
40
0,42 0,44 0,46 0,48 0,50 0,52 0,54
Philippines
Croatia
Tanzania
Ghana
Indonesia
Zimbabwe
Albania
Egypt, Arab Rep.
Colombia
Brazil
Graph 2. Ten Countries with the Most Complex Tax System

Source: Elaborated from the Tax Complexity Index (Hoppe et al., 2021)
Table 1 shows that Jersey is a jurisdiction with the most efficient (perceived as least
complex) tax system, noting the lowest indices in the TCI, while Brazil has the most
complex tax system according to the index. Indonesia shared a similar score with Ghana
and is included in one of the countries with the most complicated tax system among a
hundred nations in the study. Furthermore, one of the indicators in TCI is the tax
framework complexity, which measures the legal process within the tax system. The
regulation practices are essential in shaping the policy objectives of a nation (World
Bank, 2017), particularly in the context of the tax system. A complicated tax system may
arise when the laws and regulations are poorly developed and implemented. Thus, it is
worth investigating the impact of regulatory quality on tax complexity.
This study intends to contribute to the debate on the critical notion of countries'
regulatory quality, specifically in relation to the development of the tax system.
Regulatory quality captures the perception of the government's ability to define,
formulate, and implement appropriate policies and regulations that promote private
business development (Kaufmann et al., 2011). Here, in the context of tax policy
development, which correlates with the MNEs' activities in the tax jurisdictions. The
value of countries' regulatory quality is defined by the World Bank's worldwide
governance indicators (WGI), which was initiated by Kaufmann et al. (2011) study. The
study also corroborates one of the WGI indicators, government effectiveness, as
moderating variable in the relationship between regulatory quality as the independent
variable and tax complexity as a dependent variable. According to the WGI, government
effectiveness constitutes the quality of public services, policy formulation, and
implementation in a country. It also measures the degree of government commitment to
enacting such a policy. The author believes that countries with excellent regulation
quality would have a less-complex tax system, and this relationship is likely to happen
in well-govern countries.
1. LITERATURE REVIEW AND HYPOTHESIS

In literature, tax complexity has been explored and defined in many tax jurisdictions.
From the perspective of the US individual taxpayers, J. Slemrod (1989) studied that tax
system simplification would reduce the non-compliance behavior of the individual
taxpayers. From the aspects of the theoretical framework, Cooper (1993) and Evans &
Tran-Nam (2010) investigated a novel issue around income tax simplification and the
conceptual background of the tax complexity, specifically in the Australian tax system.
41
Cooper (1993) stated that tax complexity resulted from adopting a complex tax base.
The claim would raise the political aspect as the basis of the sources of tax complexity.
Theoretically, according to Cooper (1993), there are many sources of complexity in the
tax system. Tax complexities are derived from the interests of bureaucrats (government
and legislators) to maximize their control power, size-wise and budget-wise. Also, the
interests of the political groups in shaping the tax policymaking and the interests of tax
professionals to exploit tax laws on behalf of their clients are essential in explaining the
sources of complexity. Lastly, the taxpayers' interest in complying with the regulations
intentionally, or worst, evading taxes, might be crucial to the tax system's complexity.
In addition, Evans & Tran-Nam (2010) discussed that tax changes and reforms were
critical in the tax system simplification, which can be done by measuring the compliance
costs faced by the taxpayers on a regular basis. The research has similar value to
Cooper (1993) on the role of government in managing tax simplification and complexity.
From the perspective of the tax practitioners, Hoppe & Sureth-Sloane (2018)
investigated the perspective of tax consultants in 108 countries on what drives tax
complexity. The researchers used a qualitative content analysis (QCA) and cluster
analysis in analyzing the datasets and found that the tax code was the main driver of the
tax system complexity. Other factors that were also important were changes in tax
regulations and tax officers' inconsistency in applying tax laws, particularly during the
audits.
From a tax complexity measurement standpoint, J. B. Slemrod & Blumenthal (1996)
measured that higher tax compliance costs were the main effect of the complexity in the
US tax system. Their study was based on a survey among 1,329 US largest
corporations. It revealed that, sometimes, tax law reforms added complexity to the tax
system. However, Eichfelder & Hechtner (2018) challenged the measurement of tax
compliance costs due to the survey method. They claimed that the 'questionnaire
framing effects' might lead to bias in cost estimation and did not necessarily reflect the
tax complexity. Another research by Eichfelder & Vaillancourt (2014) found that tax
compliance burdens were the drivers of tax law complexity. The drivers can be seen in
the mean of the number of different taxes, the complicated language of tax regulations,
and the intricate tax calculations. Thus, the tax system is complicated: narratively and
quantitatively.
In relation to the governance's impact on tax complexity, Ajaz & Ahmad (2010)
studied that complexity in the tax system leads to the abuse of power from the tax
officials and drives corruption. Also, complexity in paying taxes causes corruption among
taxpayers. The research incorporated panel data from 25 developing countries and used
the Generalized Method of Moments (GMM) to answer the research questions. They
found that institutional governance (as represented by the governance index) was
statistically significant in reducing the complexity and increasing tax revenue. Thus,
government effectiveness, accountability, stability, the rule of law, and control of
corruption are essential in the tax revenue collection for developing countries.
Another piece of literature exploring the role of governance in the tax system was
Epaphra & Massawe (2017), who examined the institutional authority and its role in the
tax revenue policy using panel data analysis of 30 African countries. The Random
Effects analysis showed that excellent governance, as demonstrated by the
government's effectiveness, regulatory quality, the rule of law, and accountability, are
determinants for tax revenue generation. The findings suggest that increasing
institutional capacity reduces corruption and increases tax efficiency (or minimizes tax
complexity). The result of this research is supported by a study by Arif & Rawat (2018),
who also found that enhancing governance quality leads to the reduction of corruption
42
rate, broadens the tax base, and reduces complexity in the tax administration. The study
addressed the issue in the emerging economies using the principal component analysis
and multi-factor analysis.
Based on the explanation above and following the tax complexity index by Hoppe et
al. (2021) and the worldwide governance index by Kaufmann et al. (2011), two
hypotheses using a hundred economies as research objects are developed in this study
as follows:
Hypothesis 1
Ho = countries' regulatory quality does not directly impact the tax complexity
Ha = countries' regulatory quality does directly impact the tax complexity
Hypothesis 2
Ho = government effectiveness does not moderate the relationship between
regulatory quality and tax complexity
Ha = government effectiveness does moderate the relationship between regulatory
quality and tax complexity
2. RESEARCH METHODOLOGY
This study uses documentary research as the basis for the exploration of concern
variables (Ahmed, 2010; Scott, 2014). Documentary sources are identified in the form
of statistics of tax complexity index developed by other researchers (Hoppe et al., 2018,
2021) and institutional publications of WGI by the World Bank (Kaufmann et al., 2011).
The data will then be analyzed quantitatively. The secondary data as the basis of
documentary research comes from the tax complexity index (TCI) for the dependent
variable. This study also incorporates the worldwide governance indicators (WGI) for the
regulatory quality as an independent variable and government effectiveness as a
moderated variable. The data covers a hundred countries following the lists from TCI.
The rationale is to cover a cross-country analysis to capture a more reliable result that
can be justified accordingly.
TCI scores span from zero to one as an indicator of the tax system's complexity.
Zero means the system is perfectly efficient and not complex, while the value of one
defines the perfect complexity of the tax system. However, based on the TCI indices, no
single country has a score of zero or one. Thus, perfection in the tax system does not
exist. On the other hand, regulatory quality and government effectiveness range
between -2.5 (considered weak governance, both regulatory and effectiveness) to +2.5
(perceived as strong governance indicators). Policymakers and scholars have
addressed the use of WGI. For example, the US government used WGI to target the
millions of dollars of grants to foreign countries (Thomas, 2010). In addition, according
to the World Bank (2006), “Other donor governments, such as the Netherlands, also rely
on the Worldwide Governance Indicators to monitor the quality of governance.”
For the regression model, this study incorporates the regression with moderating
effect. The model is constructed using Hayes' moderation model (Hayes, 2018) as
follows:
43
W
X Y
Figure 1. A Simple Moderation Model

Source: Hayes (2018)
In a moderation model, the relationship between X and Y (the effect of X on Y) is

determined as related to the presence of moderator W, as shown in Figure 1. Testing
the moderation effect on the linear regression assumes the linearity between X and Y
and that the relationship is linearly moderated by W. thus, the estimation model of Y can
be presented in the form of the model as follows:
Y = 𝛂 + b1X + b2W + b3XW + 𝛆
A linear moderation relationship is examined with the inference of the value of b3,
which constitutes the regression weight for XW (the interaction term). Suppose the value
of b3 is not zero. In that case, the X's effect on Y varies to the degree of W. the affirmation
of the moderation effect can be tested using simple slope analysis or the Johnson-
Neyman technique (Hayes, 2018).
Concerning this study, the model can be adjusted according to the variables
incorporated and expressed in an alternative form:
Govern
RegQual TCI
Figure 2. A Moderation Model in the Study

Source: Author's Analysis Based on Hayes (2018)
TCI = 𝛂 + b1RegQual + b2Govern + b3RegQual*Govern + 𝛆

explanations:
TCI = the tax complexity scores ranges: from 0 to 1
RegQual = the regulatory quality indicator of a country ranges: from -2.5 to
+2.5
Govern = the government effectiveness indicator of a country ranges: from
-2.5 to +2.5
𝜀 = error terms
𝛼 = the Y-intercept of the model
Subsequently, SPSS version 27 will be used to analyze the datasets. The

PROCESS system is added to the SPSS to support the moderation analysis into a
single calculation and was developed by Hayes (2012).
44
3. RESULTS AND DISCUSSIONS
This part discusses the analysis results using SPSS PROCESS related to the
moderation effect of government effectiveness on regulatory quality and tax complexity
among a hundred nations. In addition, a brief discussion of tax complexity in Indonesia
according to the data will also be addressed.
3.1. Results
4.1.1 Descriptive Statistics and Correlations
To start the analysis, the author conducts a descriptive analysis to present the
spread of the data for all variables. Table 1 shows the descriptive numbers of regulation
quality, government effectiveness, and tax complexity index. The data indicates that the
tax system's complexity score ranged from 0.19 to .53, with a mean of .366. The lowest
score (meaning the tax jurisdiction with the least complex system) belongs to Jersey,
and the highest score (the most complicated tax system) goes to Brazil. In addition, the
institutional governance indicators, as shown by the regulatory quality and effectiveness
of the government process, virtually share similar values ranging from -1.8 to 2.2, with a
mean of around 0.4. Singapore has the most excellent quality regulations, while
Venezuela shares the country with the least regulatory quality. Regarding effectiveness,
Singapore once again hits the highest mark as a country with the most effective
governmental process. At the same time, the Republic of Yemen acts as a nation with
the most ineffective government.
Table 1. Descriptive Statistics

Source: Author's Analysis Using SPSS v.27
Variables Obs missing Mean Std. Dev Min Max
TCI 100 0 0.366 0.70 0.19 0.53
RegQual 100 0 0.437 0.95 -1.99 2.18
Govern 100 0 0.441 0.91 -1.82 2.20
Next, before analyzing the correlation, a normality test is conducted to determine

whether the data is approximately normally distributed, and parametric analysis can be
addressed. The Kolmogorov-Smirnov (K-S) test will determine whether the underlying
data follows a normal distribution. The null hypothesis is that the data is generally
distributed at the 0.05 level of significance. The K-S result is presented in Table 2.
Table 2. Tests of Normality (K-S)

Variables statistics df Sig. Decision
TCI 0.057 100 0.200 normally distributed
RegQual 0.081 100 0.106 normally distributed
Govern 0.082 100 0.092 normally distributed
n = 100, p is sihnificant at 0.05 level
The result of the K-S analysis confirms the normal distribution of the variables in the
study. As shown by the p-value above 0.05 for three variables, we can conclude that
normality can be assumed for the dataset, and appropriate parametric analysis can be
conducted afterwards (Marshall & Samuels, 2017). Next, Table 3 illustrates Pearson's
correlation among variables. The data shows that there is a strong correlation between
regulatory quality and government effectiveness, as indicated by a value of 0.948.
45
Table 3. Correlations Among Variables
Variables TCI RegQual Govern
TCI 1
RegQual -0.387 1
Govern -0.348 0.948 1
The correlation between two variables is considered significant if the Pearson value
is more than 0.70 (Boslaugh, 2012). As such, the correlation between variables would
diminish the quality of the moderation model in this research. To overcome such
problems, some researchers proposed using the mean centering method of the predictor
variables. The mean-centred means subtracting the variable's mean from each value or
case to reduce multicollinearity problems in multiple regression (Aiken et al., 1991;
Eveland Jr, 1997; Hayes, 2020; Jaccard et al., 2003). Therefore, this study uses this
approach to solve the correlation problems between regulatory quality and the
effectiveness of the government process. SPSS PROCESS provides a one-click feature
of the mean-centered of variables in the system. Also, the classical regression
assumptions regarding the residual’s independence, normality of residuals, and
heteroscedasticity are already met in this study before conducting the moderated
analysis.
4.1.2 Moderated Regression Analysis

To test and answer hypotheses 1 and 2, the author used PROCESS macro in SPSS
version 27, following Model 1 developed by Hayes (2012). SPSS PROCESS could
generally estimate the direct effect of regulatory quality on tax complexity and test the
moderated effect of government effectiveness. Further, the bootstrap samples provided
by the PROCESS are elaborated in the model to evaluate the direct significance with
95% confidence intervals (Scarpi et al., 2019). If zero is outside the confidence intervals
(zero is not within the Lower-Level Confidence Interval – LLCI and Upper-Level
Confidence Interval – ULCI), the model has a significant outcome. The result of the
analysis is presented in Table 4.
Table 4. Direct and Moderation Effects
Source: Author's Analysis Using SPSS PROCESS Macro
Dependent variable: TCI
Model Summary
R2 SE F df1 df2 p
0.2098 0.0040 8.4979 3.00 96.00 0.000
Model per Variables
coeff t p LLCI ULCI
constant 0.3811 45.0614 0.000 0.3643 0.3979
RegQual -0.0483 -2.2821 0.0247 -0.0904 -0.0063
(direct
effect)
Govern 0.0209 0.9486 0.3452 -0.0228 0.0647
Interaction -0.0181 -2.6323 0.0099 -0.0318 -0.0045
(RegQual*Gover
n)
46
R2 change F df1 df2 p
X*W 0.057 6.9290 1.00 96.00 0.0099
(moderation
effect)
Based on Table 4, we can conclude that the overall model has a significant level
with the p-value <0.001 and R2 equal to 0.2098. The value means that the model in this
study can explain approximately 20.98% of the tax complexity model. Moreover,
hypothesis 1 is tested using a direct effect of RegQual on TCI and can be summarized
as b = -0.0483, t(96) = -2.2821, p = 0.0247, LLCI = -0.0904, and ULCI = -0.0063. The
conclusion of direct effect analysis suggests a statistically significant effect of regulatory
quality on tax complexity (p-value is less than 0.05, and zero is outside the value of LLCI
and ULCI). Because the t value is negative, the slope of the regression line would be a
reverse relationship. The increasing score of regulatory quality would decrease the tax
complexity index. Thus, the more excellent the quality of countries' regulation is, the
more efficient the tax system in those countries, as indicated by the low scores of tax
complexity. As a result, Ho in hypothesis 1 is rejected, and we can accept the Ha, which
means that countries' regulatory quality directly impacts the tax system's complexities.
Additionally, to determine the moderation effect of government effectiveness, we
can look at the interaction result under the unconditional interaction of X*W. the result
reveals that the interaction between X (regulatory quality) dan W (government
effectiveness) is significant, with a p-value less than 5% significance level. The
interaction can be written as b = -0.0181, t(96) = -2.6323, p = 0.0099, LLCI = -0.0318,
and ULCI = -0.0045. The interpretation of the result is that there is a statistically
significant effect of the moderation variable of government effectiveness, as indicated
by the low p-value (less than 5%), and zero is not within ULCI and LLCI. Also, there is a
change in the value of R2 by 5.7% due to the moderation effect of the government
effectiveness. Thus, we can reject Ho and accept Ha in hypothesis 2 because
government effectiveness does moderate the relationship between regulatory quality
and tax complexity.
3.2. Discussions
This study aims to explore the impact of the government's ability to formulate sound
policies and regulations on the complexities of a tax system and to observe the
moderating variable of regulations' effectiveness. Firstly, the finding reveals that
countries' ability to provide sound rules correlates significantly with the tax complexity
faced by the taxpayers. The result is consistent with previous research that governance
is essential in addressing tax complexity (Ajaz & Ahmad, 2010, Epaphra & Massawe,
2017, Arif & Rawat, 2018). That is, countries with improved regulatory quality promote
simplification in the tax system and provide efficient tax policies and administrations.
Secondly, the study findings show that countries' public services quality (as shown in
government effectiveness) moderates the first finding.
Consequently, countries with excellent regulations and simplified tax systems have
top-notch public services and keep the highest standard of public qualities. Lastly, the
degree of moderation effect can be seen with the simple slopes to understand the
moderating effect of government effectiveness. The overall moderation model of this
study is captured in Figure 3. Further, it shows the degree of significance and slope of
regression for the main effects of regulatory quality on tax complexity. It also presents
the moderated effect of government effectiveness on the relationship between
independent and dependent variables.
47
Govern
p = 0.0099
t = -2.6263
RegQual p = 0.0247 TCI
t = -2.2821
Figure 3. A Moderation Model With p and t Values
In addition to the moderation effect, conditional effects of the moderator are also
tested using simple slopes. The three categories of the moderator are assessed: below
the mean (-1SD), equal to the mean (0SD), and above the mean (+1SD) (Buchanan,
2015). As illustrated in Graph 3, the moderating effects appear in all three categories of
the moderator (low, average, and high), which shows that the effectiveness of
government is indeed moderating the relationship between regulatory quality and
countries' tax complexities. Examination of the interaction plots shows an illustrative
effect that increasing the quality of regulations (both formulation and implementation)
would decrease the value of tax complexity at every level of countries' government
effectiveness. Here, the meaning is clear: if a country wants to have an efficient and less
complex tax system, the government needs to formulate and implement sound policies
and regulations that encourage corporate taxpayers to develop their businesses (pro-
investment tax policies). The rationale behind this conclusion is that the tax complexity,
as proposed by Hoppe et al. (2021), is examined from the perspective of MNEs
(corporate taxpayers).
Graph 3. Interaction Plots of Regulatory Quality and Tax Complexity as Moderated by

Government Effectiveness
Moreover, a cluster analysis is corroborated to group countries based on similarity

from the value of regulatory and complexity to validate the interaction plots. Figure 4
illustrates the countries' clusters using the Twostep Cluster Analysis provided in the
SPSS. The model summary provides two groups with good quality.
48
Figure 4. Twostep Cluster Ouput
Graph 4. Cluster Output of Regulatory Quality and Tax Complexity

According to the cluster output, there are two clusters for the association between
regulatory and complexity. Firstly, the cluster consists of case numbers, for example, 11
(Pakistan), 30 (Indonesia), 31 (Lebanon), 47 (Mexico), and 69 (Russian Federation).
Secondly, the cluster with cases numbers 49 (Qatar), 63 (Luxembourg), 65 (Malta), 80
(New Zealand), and 95 (Ireland). The first clusters are countries categorized as low- and
middle-income countries and have slightly poorly regulated policies and complex tax
systems (as indicated with high TCI scores and low regulatory values). Conversely, the
second cluster is formed by the high-income nations, which also have high quality in the
regulations development and less complex tax system. Thus, the findings provide robust
empirical evidence that quality in the countries' regulations shapes efficient tax laws and
frameworks. This result supports the study by Ajaz & Ahmad (2010), Hoppe et al. (2021),
and Jalilian et al. (2007). They found that the quality of governance was critical for the
countries' development, reducing the complexity of the tax policy and tackling corruption.
Also, this study confirms the research by Evans (2012), who stated that regulatory
reform is essential in managing tax complexity and helping taxpayers reduce the
compliance burden.
49
From the perspective of Indonesia, which includes one of the countries with the
most complicated tax system, as shown in Graph 5, the finding is essential in managing
tax policy and administration by the Indonesian tax authority, the Directorate General of
Taxes. Graph 5 provides precise data that Indonesia has the lowest quality in all aspects
of variables in this study compared to the average value of the overall sample in this
study and ASEAN countries. The areas of complexity include filing different kinds of
direct and indirect taxes, system disruptions or errors during the filing process, transfer
pricing regulations, and intricate audit process (Adila, 2015; Hoppe et al., 2019, 2021;
Hoppe & Sureth-Sloane, 2018; Wicaksono, 2016).
Moreover, to reduce and manage the burden of complexity, the institutional capacity
to reform the direction of tax administration is crucial in providing certainty and increasing
compliance. The recently enacted law, the Tax Regulations Harmonization Law (HPP
Law), takes an essential step to reduce the regulatory burden in the Indonesian tax
system that might lead to the reduction of tax complexity in the coming years. The HPP
Law is considered simple, efficient, promotes certainty, and supports the nation's
interests (Koesmoeljana, 2021). Also, the core tax system reform initiated by the DGT
might put an inspiring vision to simplify the Indonesian tax system and increase DGT's
capacity to manage complexity. The administrative and legislative process of tax
regulatory formulation is essential to reduce the complexity and foster taxpayers'
compliance (Evans & Tran-Nam, 2010).
Further, the increased attention from global tax institutions plays a vital role in
simplifying tax systems (Harpaz, 2014). Harmonizing tax code initiatives proposed by
the OECD in the form of the Base Erosion and Profit Shifting (BEPS) action plan is a
way of achieving global equality in the tax treatment (OECD, 2018). Thus, reducing tax
complexity, not only in Indonesia but also on a global scale, is not a fantasy. Also, an
efficient global tax system could be achieved in the foreseeable future.
0,6
0,5
0,4
0,3
0,2
0,1
0
-0,1 Tax Complexity Regulatory Quality Government
Effectiveness
-0,2
Indonesia Average (n=100) ASEAN
Graph 5. Comparison of Variables for Indonesia

Source: Author's Analysis
4. CONCLUSIONS
This study tested the main effect of regulatory quality on the complexity of the tax
system and the moderation effect of government effectiveness on such relationships.
The analysis's examination using SPSS PROCESS Macro found that the quality of
formulated regulation directly impacts the tax complexity (p = <0.05, t = -2.2821). Also,
the government effectiveness moderates the relationship (p = <0.05, t = -2.6263) in
every level of moderation (low, average, and high). In brief, the findings generally draw
50
on the tax code and tax framework complexity using a hundred nations as samples using
the index developed by Hoppe et al. (2021). Two clusters are also presented in the study
to capture further the importance of regulatory quality in reducing the complexity of the
tax system. A regulatory reform might be essential in managing the efficient tax system
(as indicated by the reduction in complexity).
Global coordination is also essential in reducing tax complexities, promoting
fairness, and enhancing efficiency. Tax complexity is not only an issue of a tax authority.
Thus, harmonization and coordination need to be done globally to foster tax compliance,
simplify system, and support global economy.

Some limitations may be drawn from this study. First, using a single year might not
be sufficient to generalize the findings. This is due to the limited data on the cross-
country tax complexity index in the literature and only one research connected to the
development of the index. Thus, future studies could incorporate a panel data analysis
to increase the robustness, assuming that there will be a multi-year index available.
Second, this study uses only two indicators instead of all indicators in the worldwide
governance index. Hence, limiting the government's ability to capture good governance
in the study. Lastly, the complexity index captures the MNEs' view of the tax system.
Future research may incorporate individual taxpayers' or tax practitioners' perceptions
of the tax complexity.
Other than limitations, this research contributes to the development of determinants
of tax complexity, particularly related to governance across many countries.
Furthermore, the findings of this study could be used by tax authorities to mitigate and
address the issue of the complexity of tax laws. So, efficient tax policies and
administrations will be achieved worldwide and promote future compliance of the
taxpayers.
Acknowledgement
This paper is written explicitly for the DGT's Taxation Call for Paper 2022 and conducted during
the author’s Ph.D. study at the Cardiff University, funded by the Indonesia Endowment Fund for
Education (LPDP). Also, the author has no conflict of interest in this research.
REFERENCES
Abeler, J., & Jäger, S. (2015). Complex tax incentives. American Economic Journal:
Economic Policy, 7(3), 1–28.
Adila, R. (2015). Sistem Administrasi Pajak RI Paling Rumit se-Asean. Okezone.
https://economy.okezone.com/read/2015/03/29/20/1126002/sistem-administrasi-
pajak-ri-paling-rumit-se-asean
Ahmed, J. U. (2010). Documentary research method: New dimensions. Indus Journal
of Management & Social Sciences, 4(1), 1–14.
Aiken, L. S., West, S. G., & Reno, R. R. (1991). Multiple regression: Testing and
interpreting interactions. Sage.
Ajaz, T., & Ahmad, E. (2010). The effect of corruption and governance on tax revenues.
The Pakistan Development Review, 405–417.
Arif, I., & Rawat, A. S. (2018). Corruption, governance, and tax revenue: evidence from
EAGLE countries. Journal of Transnational Management, 23(2–3), 119–133.
51
Beck, P. J., Davis, J. S., & Jung, W.-O. (1991). Experimental evidence on taxpayer
reporting under uncertainty. Accounting Review, 535–558.
Boslaugh, S. (2012). Statistics in a nutshell: A desktop quick reference. “ O’Reilly Media,
Inc.”
Buchanan, Erin. M. (2015). SPSS - Moderation Analyses with Simple Slopes + Process.
Statistics of DOOM.
Budak, T., & James, S. (2018). The level of tax complexity: A comparative analysis
between the UK and Turkey based on the OTS Index. Int’l Tax J., 44, 23.
Carnes, G. A., & Cuccia, A. D. (1996). An analysis of the effect of tax complexity and
its perceived justification on equity judgments. The Journal of the American
Taxation Association, 18(2), 40.
Collier, R., Kari, S., Ropponen, O., Simmler, M., & Todtenhaup, M. (2018). Dissecting
the EU’s Recent Anti-Tax Avoidance Measures: Merits and Problems. EconPol
Policy Report.
Cooper, G. S. (1993). Themes and issues in tax simplification. Austl. Tax F., 10, 417.
Deloitte. (2014). 2014 Asia Pacific Tax Complexity Survey.
https://www2.deloitte.com/cn/en/pages/tax/articles/2014-asia-pacific-tax-
complexity-survey.html
Eichfelder, S., & Hechtner, F. (2018). Tax compliance costs: Cost burden and cost
reliability. Public Finance Review, 46(5), 764–792.
Eichfelder, S., & Vaillancourt, F. (2014). Tax compliance costs: A review of cost burdens
and cost structures. Available at SSRN 2535664.
Epaphra, M., & Massawe, J. (2017). Corruption, governance, and tax revenues in Africa.
Business and Economic Horizons (BEH), 13(1232-2019–715), 439–467.
Evans, C. (2012). Tax governance issues: managing system complexity. Economic
Papers: A Journal of Applied Economics and Policy, 31(1), 30–35.
Evans, C., & Tran-Nam, B. (2010). Controlling tax complexity: Rhetoric or reality.
Australia’s Future Tax System: The Prospects After Henry, Evans C, Krever R and
Mellor P (Eds) Thomson Reuters, Sydney, 439–463.
Eveland Jr, W. P. (1997). Interactions and nonlinearity in mass communication:
Connecting theory and methodology. Journalism & Mass Communication
Quarterly, 74(2), 400–416.
Freudenberg, B., Tran-Nam, B., Karlinsky, S., & Gupta, R. (2012). A comparative
analysis of tax advisers’ perception of small business tax law complexity: United
States, Australia, and New Zealand. Australian Tax Forum, 27(4), 677–718.
Harpaz, J. (2014). How To Simplify Taxes In The Most Complicated Country In The
World. Forbes. https://www.forbes.com/sites/joeharpaz/2014/09/30/how-to-
simplify-taxes-in-the-most-complicated-country-in-the-world/?sh=7aecc6766c72
Hayes, A. F. (2012). PROCESS: A versatile computational tool for observed variable
mediation, moderation, and conditional process modeling. University of Kansas,
KS.
Hayes, A. F. (2018). Partial, conditional, and moderated mediation: Quantification,
inference, and interpretation. Communication Monographs, 85(1), 4–40.
Hayes, A. F. (2020). Statistical methods for communication science. Routledge.
Hoppe, T., Schanz, D., Sturm, S., & Sureth, C. (2019). Measuring tax complexity across
countries: A survey study on MNCs. Arqus Discussion Paper., 245.
52
Hoppe, T., Schanz, D., Sturm, S., & Sureth-Sloane, C. (2018). What are the Drivers of
Tax Complexity for MNCs? Global Evidence. Intertax, 46(8/9).
Hoppe, T., Schanz, D., Sturm, S., & Sureth-Sloane, C. (2021). The Tax Complexity
Index–A survey-based country measure of tax code and framework complexity.
European Accounting Review, 1–35.
Hoppe, T., & Sureth-Sloane, C. (2018). What are the drivers of tax complexity for MNCs?
Global evidence. Intertax, 46(8/9).
Ingraham, L. R., & Karlinsky, S. S. (2005). Tax professionals’ perceptions of small
business tax law complexity. Tax Notes, 107(1), 79.
Jaccard, J., Turrisi, R., & Jaccard, J. (2003). Interaction effects in multiple regression
(Issue 72). Sage.
Jalilian, H., Kirkpatrick, C., & Parker, D. (2007). The impact of regulation on economic
growth in developing countries: A cross-country analysis. World Development,
35(1), 87–103.
Kaplow, L. (1994). Accuracy, complexity, and the income tax. National Bureau of
Economic Research Cambridge, Mass., USA.
Kaufmann, D., Kraay, A., & Mastruzzi, M. (2011). The worldwide governance indicators:
Methodology and analytical issues1. Hague Journal on the Rule of Law, 3(2), 220–
246.
Koesmoeljana, B. (2021). Wide-ranging changes in Indonesia under Tax Regulation
Harmonization Law. https://www.ey.com/en_id/tax-service/wide-ranging-taxation-
changes-in-indonesia-under-tax-regulation-
Krause, K. (2000). Tax complexity: Problem or opportunity? Public Finance Review,
28(5), 395–414.
Marshall, E., & Samuels, P. (2017). Checking normality for parametric tests.
ResearchGate.
McKerchar, M. (2005). The impact of income tax complexity of practitioners in Australia.
Austl. Tax F., 20, 529.
Müller, C., & Voget, J. (2012). Tax complexity and foreign direct investment. Available
at SSRN 3051912.
OECD. (2018). OECD/G20 Inclusive Framework on BEPS. OECD.
https://www.oecd.org/tax/beps/flyer-inclusive-framework-on-beps.pdf
Saad, N. (2014). Tax knowledge, tax complexity, and tax compliance: Taxpayers’ view.
Procedia-Social and Behavioral Sciences, 109, 1069–1075.
Scarpi, D., Mason, M., & Raggiotto, F. (2019). To Rome with love: A moderated
mediation model in Roman heritage consumption. Tourism Management, 71, 389–
401.
Scott, J. (2014). A matter of record: Documentary sources in social research. John
Wiley & Sons.
Slemrod, J. (1989). Complexity, Compliance Costs, and Tax Evasion. In J.A. Roth, and
J.T. Scholz (Eds.). In Taxpayer Compliance. Philadelphia: University of
Pennsylvania Press.
Slemrod, J. (2005). The etiology of tax complexity: Evidence from US state income tax
systems. Public Finance Review, 33(3), 279–299.
Slemrod, J. B., & Blumenthal, M. (1996). The income tax compliance cost of big
business. Public Finance Quarterly, 24(4), 411–438.
53
Stantcheva, S. (2021). Understanding tax policy: How do people reason? The Quarterly
Journal of Economics, 136(4), 2309–2369.
Thomas, M. A. (2010). What do the worldwide governance indicators measure? The
European Journal of Development Research, 22(1), 31–54.
Wicaksono, P. E. (2016). Prosedur yang Rumit Jadi Pemicu Orang RI Malas Bayar
Pajak. Liputan6. https://www.liputan6.com/bisnis/read/2652666/prosedur-yang-
rumit-jadi-pemicu-orang-ri-malas-bayar-pajak
World Bank. (2017). A Joint Offering of the Governance and Trade & Competitiveness
Global Practices.
54
CORRECTIONS TO DOMESTIC TRANSFER PRICING
POLICIES A ZERO-SUM-GAME OF NATIONAL TAX
REVENUE?: CASE STUDY OF KPP PRATAMA IN DKI
JAKARTA REGION
[Fajar Surya Putra]a, [Yeni Farida]b

a [Directorate General of Taxation, Ministry of Finance of Republic of Indonesia, Jakarta, Indonesia]
Email: [pajar.putra9@gmail.com]
b [Fiscal Policy Agency, Ministry of Finance of Republic of Indonesia, Jakarta, Indonesia]
Email: [kursiman53@gmail.com]
ABSTRACT
Domestic transfer pricing correction by tax officers does not significantly impact national tax revenue.
Tax revenue increasing due to domestic transfer pricing correction will be accompanied by tax
revenue declining from the affiliate parties in Indonesia (Setiawan, 2014). This condition is known as
the zero-sum-game condition. This research aims to examine zero-sum-game conditions due to
domestic transfer pricing correction. The zero-sum-game condition is examined by predicting the
corresponding adjustment by the affiliated parties. This research uses the case study method
because the researcher does not have a domestic transfer pricing correction population.
Keywords: domestic transfer pricing, zero-sum-game, a corresponding adjustment
Sebagian peneliti dan praktisi perpajakan memandang koreksi atas kebijakan harga transfer
domestik tidak berdampak signifikan bagi penerimaan pajak nasional. Penambahan penerimaan
pajak akibat koreksi kebijakan harga transfer domestik akan diiringi dengan penurunan penerimaan
pajak dari pihak afiliasi di Indonesia (Setiawan, 2014). Kondisi tersebut dikenal dengan kondisi zero-
sum-game. Riset ini bertujuan meneliti kondisi zero-sum-game pada koreksi kebijakan harga transfer
domestik oleh Direktorat Jenderal Pajak. Penelitian ini menggunakan pendekatan studi kasus karena
peneliti tidak mengetahui populasi koreksi kebijakan harga transfer domestik di Indonesia. Proses
pembuktian aspek zero-sum-game dilakukan dengan menggunakan proyeksi corresponding
adjustment yang akan dilakukan oleh pihak afiliasi Wajib Pajak.
Kata kunci: kebijakan harga transfer domestik, zero-sum-game, corresponding adjustment
1. INTRODUCTION
1.1. Background
Compared to multinational transfer pricing studies, domestic transfer pricing has
attracted little attention among academics and tax practitioners. Setiawan (2014) claims
that the risk of domestic transfer pricing for state revenue is near nil because a tax
reduction for one company will increase taxes for other companies in the same country.
In contrast to Setiawan (2014), Supriyadi (2022) assesses the risk of domestic transfer
pricing to be the same as that of foreign transfer pricing. This risk arises when one of
the affiliated parties gets tax facilities, one is subject to Final Income Tax, or one of the
affiliated parties has compensation for fiscal losses.
Sitanggang and Firmansyah (2021), in their research, even found that a public
company did not disclose transfer pricing transactions related to compensation for key
management. This research will discuss the risks of domestic transfer pricing
transactions between domestic affiliates. The primary considerations of researchers
raising the issue of domestic transfer pricing include the low level of domestic transfer
55
pricing research in Indonesia and the significant transfer pricing risk that must be
managed by the Directorate General of Taxes (DGT). The total transfer pricing
transactions based on taxpayer declaration data on corporate income tax returns is IDR
5,800 trillion (Asmadi in Wildan, 2022). This amount does not include transfer pricing
transactions not declared by the Taxpayer.
The researcher used a case study at one of the Primary Tax Service Offices (KPP
Pratama) in DKI Jakarta Province. The consideration for using the case study
methodology is that the researcher needs data on the population monitoring domestic
transfer pricing transactions. The researcher chose the sample at KPP Pratama
because taxpayers at KPP Pratama tend not to disclose information related to affiliated
transactions. Researchers took three research samples of Request for Explanation of
Data and/or Information (SP2DK).
2. THEORETICAL FRAMEWORK AND HYPOTHESIS

DEVELOPMENT
2.1. Special Relations, Affiliate Transactions, and Transfer Pricing
The term special relationship currently only exists in the scope of taxation after the
Statement of Accounting Standard 07 (Revised 2015) Disclosures of Related Parties
(from now on referred to as PSAK 07) no longer uses the term. A special tax relationship
is regulated in Article 18, paragraph (4) of the Income Tax Law and Article 2, paragraph
(2) of the Value Added Tax Law (VAT Law). A special relationship is considered to exist
if:
a. Taxpayers have direct and indirect equity participation of at least 25% in other
taxpayers, a relationship between taxpayers with the participation of at least 25% in
two taxpayers or more, or a relationship between two taxpayers or more, the last
mentioned,
b. Taxpayers control other Taxpayers, or two or more Taxpayers are under the same
control either directly or indirectly, or
c. There is a family relationship, either blood or marriage, in a straight line and/or one
degree sideways.
Further explanation regarding the control aspect is explained in Article 8, paragraph
4, letter b of Government Regulation 94 of 2010 concerning the Calculation of Taxable
Income and Payment of Income Tax in the Current Year. Direct management control is
considered to exist if there are similarities in managing two or more Taxpayers. For
example, Mr. A is the Director of PT X and the Main Director of PT Y, so PT X and PT
Y are considered to have a special relationship. Included in the definition of mastery is
if the management of different companies have family relations either by blood or by
marriage in a straight line and/or one degree sideways. For example, Mr. A served as
Director of PT X, and Mrs. C, as Mr. A's wife, became Director of PT Y, PT X, and PT Y
are considered to have a special relationship.
The government refined the definition of special relationship in Minister of Finance
Regulation Number 22/PMK.03/2020 concerning Procedures for Implementing Transfer
Price Agreements (Advance Pricing Agreement) (from now on referred to as PMK-
22/PMK.03/2020). These improvements can be seen, among others, in the provisions
of Article 1 point 13, Article 1 point 15, and Article 4 PMK-22/PMK.03/2020. Article 1
number 13 PMK-22/PMK.03/2020 defines affiliated parties as parties that have a special
relationship with each other. Article 1 number 13 links the concept of a special
relationship in tax regulations with affiliated transactions in PSAK 07. Article 1 number
15 broadens the scope of related transactions by adding the definition of related
transactions as transactions made between parties who are not related but are affiliated
56
parties of one party or both parties to the transaction, determining the counterparty of
the transaction, and the transaction price. Article 4 PMK-22/PMK.03/2020 expands the
element of management control by adding the following two conditions:
a. the parties who are commercially or financially known or claim to be in the same
business group, or
b. one party claims to have a special relationship with the other party.
Even though the government has prepared various regulations relating to related
transactions, Ardianto and Dyan (2018) consider that applying PSAK 7 is more
appropriate to explain related transactions than tax regulations. Based on that study, the
researcher cites the provisions in paragraph 9 of PSAK 07 regarding the definition of
affiliated transactions. Based on paragraph 9 of PSAK 07, affiliated transactions are
defined as transfers of resources, services, or obligations between the reporting entity
and related parties, regardless of whether a price is charged. The related parties in
PSAK 07 are broadly divided into two groups, namely people or close family members
and other entities. A person or close family member is considered to have an affiliate
relationship with the reporting entity if they have control or joint control, have significant
influence, or are key management personnel of the reporting entity or parent entity of
the reporting entity. An entity is considered to have a special relationship if it fulfills the
following conditions:
a. the entity and the reporting entity are members of the same business group,
b. the entity is an associate or joint venture of a business group, and the reporting entity
is a member,
c. the entity and the reporting entity are joint ventures of the same third party,
d. the entity is a joint venture of the third party entity, and the reporting entity is an
associate entity of the same third party,
e. the entity is a post-employment benefit plan for employee benefits from a reporting
entity or an entity related to the reporting entity. If the reporting entity is the entity
administering the program, then the sponsoring entity is also related to the reporting
entity,
f. an entity that is controlled or jointly controlled by a person or close family member
who has a relationship with the reporting entity,
g. a person or close family member with control or joint control of the entity who has
significant influence over the entity or is a member of the entity's key management
personnel, and
h. the entity, or a member of a group of which it is a part, provides critical management
personnel services to the reporting entity or the parent of the reporting entity.
Based on the explanation above, the researcher sees that the scope of affiliated
relationships in PSAK 07 (Revised 2015) is relatively broader than tax regulations. The
definition of affiliate in PSAK 07 accommodates sponsoring entities and joint ventures
that are not regulated in tax regulations. However, PSAK 07 as accounting standards
and tax regulations as tax standards certainly have different objectives. PSAK 07
mitigates the revenue loss aspect and the early recognition of revenue from affiliated
parties, such as in the case of Enron's early recognition of revenue from related
transactions.
Transactions between affiliated parties then give rise to a practice known as transfer
pricing. Article 1 point 8 Regulation of the Director General of Taxes number PER-
32/PJ/2011 concerning Amendments to Regulation of the Director General of Taxes
number PER-43/PJ/2010 concerning Application of the Principles of Fairness and
Business Usability in Transactions between Taxpayers and Parties with Special
Relationships (from now on referred to as PER-32/PJ/2011) defines transfer pricing as
57
determining prices in transactions between parties that have special relationships.
Kurniawan (2015) defines transfer pricing as a company policy in determining the
transfer price of a transaction with parties affected by special relationships. Taxpayers
must apply the principles of fairness and business custom for every transaction with
related parties. This principle of fairness and business commonality can be interpreted
that the price or profit between related parties must be the same as or within the price
or profit range of a party with no special relationship so that transactions between
taxpayers and affiliated parties have reflected fair market value. The steps for
implementing the Principles of Fairness and Business Usability based on Article 3
paragraph (2) of the Director General of Taxes Regulation number PER-32/PJ/2011
include:
a. comparability analysis and comparison determination
a transaction is considered comparable if there is no material or significant difference
in conditions that could affect the price or profit. In the case of a difference in
conditions, adjustments can be made to the difference in conditions. Taxpayers must
use internal comparison data if internal and external comparison data have the same
level of comparability.
b. selection of appropriate transfer pricing methods
the transfer pricing method permitted in Article 11 paragraph (2) of the Director
General of Taxes Regulation number 32/PJ/2011 consists of the comparable
uncontrolled price (CUP) method, the resale price method (RPM), the cost plus
method (CPM), the profit split method (PSM), or the transactional net margin method
(TNMM). Each method has specific characteristics and conditions to be implemented.
Taxpayers using the CUP method are used when the goods or services have identical
characteristics, and the conditions of related transactions have a high degree of
comparability with independent parties. The RPM method is used when the functional
analysis results reflect a high degree of comparability even though the goods or
services have different characteristics and the party reselling does not provide
significant added value to the goods or services being traded. The CPM method is
used when the transactions carried out are service provision transactions, sales of
semi-finished goods to affiliated parties, joint facility agreement contracts, and long-
term buy and supply agreements. The PSM method is used when transactions
between independent parties are closely related, or there are special intangibles
between affiliated parties. Taxpayers use the TNMM method when one of the
affiliated parties makes a special contribution or one of the parties to a transaction
makes a complex transaction and has transactions related to one another.
c. applying the principles of fairness and customary business
fair price or fair profit can be in the form of a single price or profit or an acceptable
price range or fair profit.
d. the documentation of steps to determine fair prices or profits should apply to the
applicable laws and regulations provisions.
The four steps above also apply to particular transactions, such as transactions for
the provision of services and transactions for the use of intangible assets. Precisely for
these transactions, DGT added the following provisions:
a. service transactions
a service transaction between two related parties is considered to exist if the delivery
or acquisition of services occurs, the economic benefits can add value, and the value
of the service transaction between related parties is equal to the value of the service
transaction between independent parties.
b. transaction utilization and transfer of intangible assets.
58
transactions for the utilization and transfer of intangible assets comply with the
principles of fairness and business practice if transactions for utilization of intangible
assets occur, there are economic or commercial benefits, and the value of
transactions for utilization of intangible assets is the same as transactions with
independent parties.
2.2. Domestic Transfer Pricing Regulatory Framework
Transfer pricing issues arise from transactions between several affiliated companies
in one country (OECD, 2022). However, in the OECD Transfer Pricing Guidelines for
Multinational Enterprises and Tax Administrations, the OECD does not include this issue
in their transfer pricing guidelines. The issue of domestic transfer pricing appears in
Article 2, paragraph (2) of the Director General of Taxes Regulation number PER-
32/PJ/2011. The article explains that applying the principle of fairness and customary
business in domestic affiliated transactions is only applied to related transactions that
take advantage of differences in tariffs. Among others, due to the imposition of final or
non-final tax, the treatment of the imposition of sales tax on luxury goods, or transactions
conducted with Taxpayers of Oil and Gas Cooperation Contract Contractors.
The term "among others" in Article 2 paragraph (2) of the Director General of Tax
Regulation number PER-32/PJ/2011 is dynamic because it only illustrates transactions
with the possibility of a tariff difference between domestic affiliated parties.
2.3. Other Countries’ Domestic Transfer Pricing Regulations
One of the countries with special regulations related to domestic transfer pricing is
India. The expansion of regulations regarding domestic transfer pricing is regulated in
The Finance Act 2012 (Price Waterhouse Cooper, 2013). Domestic transfer pricing
regulations in India regulate, among other things, sales and purchase transactions
between taxpayers with tax holiday facilities with closely connected entities and
expenses based on section 40A(2) of The Finance Act 2012.
This article regulates domestic parties required to implement domestic transfer
pricing provisions. Section 40A(2) of The Finance Act 2012 is almost the same as the
provisions of Article 18 paragraph (4) of the Income Tax Law and Article 2 paragraph (2)
of the Value Added Tax Law. One of the differences between section 40A(2) and Article
18 paragraph (4) of the Income Tax Law is the difference in terminology and the
difference in the percentage of ownership, which is considered to have a substantial
effect. At first glance, Section 40A(2) of the Finance Act seems to directly adopt
commercial accounting provisions, both in terms of the use of terms such as sister
company, investor company, and investee company, as well as in terms of ownership,
which sets a lower limit for control at 20% shareholding. Price Waterhouse Cooper (2013)
illustrates the scope of The Finance Act 2012 with a sample company, company B.
Company B must apply domestic transfer pricing provisions when transacting with
company owners with substantial interests, entities, and individuals. Substantial interest
or substantial interest in The Finance Act 2012 is in line with the provisions on
commercial accounting, characterized by shared ownership of 20%. Other domestic
transfer pricing transactions are between company B and its sister companies and
company B and associated entities/subsidiaries/joint ventures owned by company B. In
addition, transactions between company B and companies that the Directors of
Company B substantially own include transactions with members of the director's family.
Company B's dealings with the Directors of Company B, the Directors of the company it
owns, and shareholders with substantial interests, including their family members.
Payments to Directors are also included in one of the scopes of domestic transfer pricing.
59
This study uses a qualitative research methodology with a case study approach.
According to Yin in Prihatsanti, Suryanto, and Hendriani (2018), the case study
approach is used because of several considerations. First, the research focuses on
answering the "how" and "why" aspects. Second, the researcher cannot manipulate the
behavior of the research object. Third, the researcher wants to cover contextual
conditions because he believes this is irrelevant to the studied matter. Fourth, the
boundaries between phenomenon and context need to be clarified. In this case, the
researcher did not have SP2DK population data related to domestic transfer pricing, so
the researcher used the case study method at one of the KPP Pratama in DKI Jakarta.
Competence related to the implementation of transfer pricing as a research source.
The researcher uses literature references related to transfer pricing, such as books,
regulations, and DGT policies regarding transfer pricing. Researchers also used primary
documentation such as SP2DK, minutes of providing information (BAPK), minutes of
requests for explanations for data and/or information (BA P4DK), and reports on the
results of requests for explanations for requests for data and/or information (LHP2DK)
as well as interviews with informants. The resource persons are staff at the Special
Transaction Examination Sub-Directorate and the Head of Supervision Section I.
Researchers took three SP2DK as samples. The SP2DK sample will be analyzed
based on the regulatory framework and information from sources to see whether the
monitoring and correction of transfer pricing results affect state revenues or are they just
a zero-sum-game condition, as explained by Setiawan (2014).
Researchers use a descriptive approach. for each sample with a description focus
on the following aspects:
a. disclosure of related transactions on corporate income tax returns,
b. sample position on compliance risk management transfer pricing (CRM TP),
c. taxpayer mode and transfer pricing scheme,
d. state revenue from the correction of domestic transfer pricing, and
e. state revenue after the corresponding adjustment.
4. RESULT AND DISCUSSION

4.1. Overview of the Research Sample
The research sample consists of three taxpayers PT M, PT N, and PT V. PT V's
notable relationship transactions originate from the charging of interest costs between
PT V and its affiliated parties. The interest rate on PT V affiliate loans reaches more than
38%. PT V's Account Representative (AR) then sends SP2DK relating to charging
interest on PT V's affiliated loans. AR uses Jakarta Interbank Offered Rate (JIBOR) data
as a reasonable interest rate because AR PT V does not know PT V's credit risk as a
factor of adjustment to the data JIBOR rate.
The following research sample is PT N. PT N's notable relationship transactions
originate from the imposition of intra-group service fees, which include marketing
services, office administration services, and supervision services. Initially, AR PT N did
not know that the service fee transaction between PT N and its counterparty was a
related party transaction. AR only found out that the transaction was related when the
Taxpayer's attorney gave the fee contract and supporting documents to AR PT N during
the counseling.
Between honorarium recipients in Period 21 PPh SPT and PT M's majority
shareholder. PT M also demonstrated the practice of related transactions. In this case,
PT M compensated family members who became the owner of shares and managed PT
60
M. However, PT M did not declare the transaction as an affiliated transaction. AR is
aware of the existence of the transaction after suspecting that there is a similarity in the
last name. AR then checked the Population and Civil Registry data (Dukcapil Data)
relating to family status between honoree recipients and PT M's majority shareholder,
and AR's suspicions were proven correct. A summary relating to the general description
of the research sample can be seen in Table 1.
Table 1 Summary of the General Description of the Research Sample

Source: Researchers processed
No. Tax SP2DK Tax LHP2DK Types of Affiliate Transactions
payer Number year
1 PT V SP2DK- 2020 Already The interest rate charged for the loan exceeds
592 the fair interest rate. There is no difference in
tax rates between PT V and affiliated parties.
2 PT N SP2DK- 2019 Already Intra-group service fees in the form of office
296 administration services, sales services, and
supervision services between PT N and
affiliated parties who are taxpayers subject to a
Final Income Tax of 0.5%.
3 PT M SP2DK- 2018 Already Compensation for family members who own
595 shares as well as management.
4.2. SP2DK PT N 2019 Tax Year

AR PT N supervises compliance with PT N's tax obligations for the 2019 tax year
based on differences in the number of purchases on the Corporate Income Tax Return
and the acquisition data on the Periodic VAT Return. AR then analyzed financial
statements and found indications that taxpayers had not fulfilled their tax obligations as
follows:
a. The indicated Taxpayer has not made Article 21 withholding income tax for
supervision services for IDR 1,125,000,000,
b. The indicated Taxpayer has not made Article 23 withholding income tax for marketing
services for IDR 2,250,000,000, and
c. Taxpayers need to explain the sandstone cost of IDR 900,000,000 and the
operational cost of IDR 1,125,000,000.
Table 2 PT N Financial Statements

Source: The researcher processed
Sales 33.346.792.280
Cost of Goods Sold 26.414.716.359 -
Gross Profit 6.932.075.921
Insurance costs 36.785.318
Marketing costs 2.250.000.000
Analysis costs 6.050.000
Sandstone costs 900.000.000
Operational costs 1.125.000.000
Supervision costs 1.125.000.000
Salary and benefits costs 41.106.720
Medical costs 777.734
Rental costs 5.005.000
Miscellaneous expense 78.201.850
Total operational expense 5.570.426.622 -
EBIT 1.361.649.299
61
Interest income 6.152.759
Interest tax (1.230.552)
Bank administration fee (893.500)
Net income outside of business 4.028.707 +
Earning before tax 1.365.678.006
Income tax 315.709.562 -
Net income 1.049.968.444
PT N does not disclose affiliated transactions in Special Attachments 3A, 3A-1, and
3A-2 on the 2019 corporate income tax return. AR PT N does not find any PT N transfer
pricing risk in CRM TP data. PT N's AR suspected that PT N's notable relationship
transactions with its affiliated parties in CRM TP were not detected due to the absence
of tax documents between PT N and its affiliated parties. PT N affiliates are not Taxable
Entrepreneurs (PKP), so they do not issue tax invoices. PT N also does not withhold
Article 23 Income Tax because PT N affiliates are Taxpayers subject to Final Income
Tax. This regulation is based on Government Regulation Number 23 of 2018 concerning
Income Tax on Income from Business Received or Obtained by Taxpayers Having
Certain Gross Turnover (from now on referred to as PP 23 of 2018).
AR PT N then submitted initial findings to the Taxpayer via SP2DK-296 dated
February 21, 2022, and the Taxpayer provided a written response as follows:
a. The Taxpayer explains that the difference in purchase data originates from the
purchase of non-taxable goods and completes the response with supporting
documents in the form of purchase documents, delivery documents, and proof
of payment to suppliers,
b. Taxpayers do not withhold PPh Article 21 because Corporate Taxpayers carry
out supervision services, namely PT N1, taxpayers also do not withhold PPh
Article 23 because PT N1 has a Certificate of Fulfilling Criteria as a Taxpayer
Based on Government Regulation Number 23 of 2018,
c. Taxpayers do not deduct PPh Article 23 for marketing services from PT N2 and
PT N3 because the two companies have a Certificate of Fulfilling the Criteria as
Taxpayer Based on Government Regulation Number 23 of 2018,
d. the sandstone cost account represents marketing costs to PT X, and the
operational cost represents the cost of office administration services performed
by PT N4.
AR conducted research on data on PT N1, PT N2, PT N3, PT N4, PT X, PT N
suppliers, and PT N customers in 2019 with the following research results:
a. PT N only has one primary customer, but PT N has 3 (three) Taxpayers who
perform the role of marketing services, namely PT X, PT N2, and PT N3,
b. PT N has two leading suppliers, and
c. PT N does not have fixed assets in the form of an office or mining goods storage
site but has office administration services with a tremendous value, PT N records
office rental fees of Rp. 5,005,000 and office administrative service fees paid to
PT N4 of Rp. 1,125,000,000.
62
Supplier 1 (PKP) Supervision Services Customer
PT N1 (non-PKP) (PKP)
Marketing service 1
PT X (non-PKP)
PT N Marketing service 2
PT N2 (non-PKP)
Marketing service 3
Supplier 2 (PKP) Office administration services
PT N3 (non-PKP)
PT N4 (non PKP)
Purchase stage Managerial stage Sales stage
Figure 1 Indication of the PT N transaction anomaly
Source: Researchers processed
service flow
the flow of goods from suppliers to customers
The next oddity was during the analysis of document contracts between PT N and
PT N1, PT N2, PT N3, and PT N4. All of contract format between PT N and PT N1 to
PT N4 consists of 4 articles with details in Article 1 concerning contract details, Article 2
concerning commissions/remuneration and banking data, Article 3 regarding dispute
resolution, and Article 4 concerning contract closing provisions. The basis for distributing
remuneration/commission between PT N and PT N1 to PT N4 is totally same which
based on the number of shipping barges of IDR 75,000,000/barge shipping of mining
goods. For the odd transaction between PT N and PT N1 to PT N4, AR then traces the
relationship between these companies.
PT N with PT N1 and PT N2 have a special relationship based on the management
similarity between the three companies. The Director of PT N (from now on referred to
as Dir. N) serves as Commissioner of PT N1 and Director of PT N2. In addition to serving
as management of PT N2, the Director of PT N also has a 40% share ownership in PT
N2. PT N and PT N4 also had a special relationship as they have similar management,
namely the Director of PT N1 (from now on referred to as Dir. N1). The director serves
as commissioner of PT N4, so between PT N1, PT N, and PT N4, there is a division of
management that makes the three companies have a relationship special. The special
relationship between PT N and PT N3 also stems from the management division
between PT N, PT N2, and PT N3. The Director of PT N2 (from now on referred to as
Dir.N2) serves as the Director of PT N3, so PT N, PT N2, and PT N3 have a special
relationship.
A special relationship also stems from a controlling relationship. In this case, the
holding company PT N, namely N Holding, has subsidiaries PT N5 and PT N6. N Holding
has a 70% stake in PT N5 and PT N6. PT N5 owns 90% of PT N's shares, while the
Main Director of PT N6 (from now on referred to as Dir. N6) is a shareholder and
administrator of PT N3 (40% of shares) and PT N4 (60% of shares).
63
N Holding
70%
Shareholding PT N6
PT N5
Management similarity 70% Pres. Dir.: Kom. N
Service flow
PT N2 PT N PT N3
Director: Dir. N2 Director: Dir. N Director: Dir. N2
Director: Dir. N
90%
PT N1 PT N4
Director: Dir. N1 Director: Dir. N6
40%
Figure 2 Pattern of Special Relationship Bussines Group PT N

Sumber: The researcher processed
After the pattern of the special relationship between PT N and the service providers
is proven. By the provisions of Article 14 paragraph (2) of the Regulation of the Director
General of Taxes Number PER-32/PJ/2011, AR PT N then conducts tests on aspects
of the existence of service delivery, aspects of economic or commercial benefits of
service delivery, and lastly the aspect of fair value for service delivery.The process of
testing these aspects is carried out through a process of requesting information from
taxpayers by using a question guide based on Appendix I of the Director General of
Taxes Circular Number SE-50/PJ/2013 concerning Technical Instructions for
Examination of Taxpayers with Special Relationships. Based on the data at the BAPK
between AR and PT N's attorney, it is known that PT N did not prepare transfer pricing
documentation (TP Doc).
Marketing services are provided by three different companies, namely PT X (an
independent party), PT N2, and PT N3. Taxpayers in 2019 have one primary customer;
the first sales transaction was carried out on July 8, 2019. The marketing services work
contract between PT N with PT N2 and PT N3 was entirely signed on January 7, 2019.
64
Transactions between PT N and PT X (parties independent) were carried out without a
contractual agreement. AR conducts tests related to the background needs of marketing
service providers and the qualifications of marketing service providers. The Taxpayer's
proxy explains the background to the need for a marketing service provider because PT
N is a new Taxpayer, so it requires another company capable of improving its
performance. The Taxpayer's attorney also explained that the selection of service
providers is entirely in the decision of the Director of PT N (Dir. N), and the appointment
process is carried out directly without any tender process for service providers.
The next question related to the existence aspect of the service (service has been
rendered) is the qualification aspect of the marketing service provider company. AR
asked questions related to the company's qualifications, segregation of duties on
marketing services, and the basis of remuneration/compensation. In addition, AR also
confirmed whether or not there were dedicated persons in the form of marketing staff
from PT N2 and PT N3, as well as evidence that PT N2 and PT N3 had provided
marketing services for PT N. The Taxpayer's attorney explained that PT N2 and PT N3
do not have special qualifications related to marketing services, and there is no
separation of duties between PT N2, PT N3, and PT X. The basis of
remuneration/compensation for PT N2 and PT N3 is IDR 75,000,000/barge shipment,
both large-capacity and small-capacity barges. For the availability of a dedicated person,
the Taxpayer's attorney explained that PT N2 and PT N3 did not provide a dedicated
person for PT N. The Taxpayer also explained to PT N's AR that the basis for the invoice
was only a sales invoice without any evidence/other documents related to the delivery
of marketing services by PT N2 and PT N3.
Based on the BAPK results, AR compared the Taxpayer's statement with the data
on the SIDJP. Based on data from SIDJP, PT N2 is classified as a transportation
management service business field with the number of employees on the Periodic
Income Tax Return 21 of 2019 is nil and the total assets in Appendix 1A of the 2019
Corporate Income Tax Return are nil. PT N2 was registered as a taxpayer on July 18,
2019. PT N3 has a business field classification as a business consulting service and
business brokerage with assets in Appendix 1A of the Corporate Income Tax Return
only in the form of air conditioners and printing machines and has a workforce of 16
people.
In connection with the process of providing supervision services by PT N1, the
Taxpayer's attorney admitted that he needed to understand in detail what kind of
supervision was carried out by PT N1. The Taxpayer's attorney only knows that the
supervision services are related to loading and unloading goods. Questions and
answers relating to the service that has been rendered aspect were answered in the
same way by the Taxpayer's attorney as the answers in the cases of PT N2 and PT N3.
Based on the search results on SIDJP, it is known that the PT N1 taxpayer has a
business field classification as a rental service and civil construction machinery and
lease. PT N1 does not have the assets in Appendix 1A of the Corporate Income Tax
Return and only has three employees.
The last intra-group service transaction is an office administration service
transaction provided by PT N4. In addition to the same questions asked to PT N2 and
PT N3, PT N's AR also asked about the background of the need for office administration
services because PT N was known to only be a virtual rent office and did not have many
employees. The Taxpayer's attorney did not answer this and only explained that PT N4
carried out activities in the form of administration and filing of company documents and
PT N's financial reports for the subsequent financial statements to be finalized by the
Taxpayer's Attorney. PT N4 does not have assets based on Appendix 1A of the
65
Corporate Income Tax Return and does not have a workforce based on Periodic Income
Tax Return Article 21.
Table 3 Summary of Aspect Testing Analysis Service Has Been Rendered

No. Question Aspect Marketing Supervision Office
Services Services Administration
Services
1. Service needs The company The company Taxpayers do
background is still new and is still newly not explain
requires established the
marketing and requires background of
services supervision the need for
services services
2. Tasks Distributions There is not N/A N/A
any
3. Service providers There is not There is not There is not
qualifications any any any
4. Basis of remuneration/ Per barge Per barge Per barge
Compensation delivery of delivery of delivery of
goods goods goods
5. Availability of There is not There is not There is not
exceptional employees any any any
for PT N
6. Proof of service has Sales invoice Sales invoice Sales invoice
been provided only only only
7. Ownership of assets to Air conditioning There is not There is not
support the operations of and printing any any
service provider press *)
companies
8. The number of workers 16 employee*) Three There is not
in labor supply employees any
companies
*) assets in the form of air conditioners and printing machines, as well as human
resources, are only owned by PT N3, while PT N2 has neither assets nor human
resources.
The AR N test results align with the results of Simamora and Hermawan's research
(2017), which explains that most intra-group service dispute cases are related to the
failure of taxpayers to provide evidence that services have been provided. Based on the
test results above, AR PT N and Section Head AR PT N ask the Taxpayer to make a
positive fiscal correction entirely for intra-group service costs because the service
transaction fails to fulfill the service aspect that has been provided. The Taxpayer then
conveys to AR PT N and Section Head AR PT N that the Taxpayer is pleased to make
a positive fiscal correction on all office administration and supervision service fees.
Nevertheless, the Taxpayer still thinks that marketing services have been carried out by
PT N2 and PT N3 and asks AR to make adjustments to the next stage, namely the
fairness of the transaction price between PT N2 and PT N3, and PT N. In this regard,
AR uses internal comparison data as transactions between PT N and PT X as
comparison data. AR chose to use the CPM method to determine the fair price of
transactions by the provisions of the Director General of Taxes Regulation number
32/PJ/2011. AR uses PT X's Gross Profit Margin data, which is 56.03%, to calculate the
66
fair price of the transaction between PT N and PT N2, and PT N3. Based on data from
SIDJP, it is known that PT N2 and PT N3 have a total cost of IDR 56,744,192 and IDR
480,895,037 so the fair price of the transaction is IDR 88,538,026 for PT N2 and IDR
588,177,781, so the Taxpayer must make a positive fiscal correction of IDR
1,573,284,194 for marketing services transactions with PT N2 and PT N3.
PT N subsequently corrected the 2019 Corporate Income Tax Return for IDR
850,001,428. For the 2019 Corporate Income Tax Return correction transaction, AR
subsequently issued a 2019 Corporate Income Tax STP number 00039/106/19/XXX/22
for IDR 196,787,561. That is, the total revenue for the state treasury was IDR
1,046,788,989.
The next researcher simulated if PT N1, PT N2, PT N3, and PT N4 made a
corresponding adjustment. The four companies are taxpayers subject to final income tax
based on Government Regulation 23 of 2018 concerning income tax from businesses
received or earned by taxpayers with inevitable gross turnover. For supervision service
and office administration service transactions with PT N1 and PT N4, all positive fiscal
corrections were carried out so that the corresponding adjustment value for the
transaction was IDR 2,250,000,000 and the tax refunded by the state was 0.5% x
2,250,000,000 or IDR 11,250,000. The taxpayers of PT N2 and PT N3 are partially
corrected for IDR 1,573,284,194 so that the tax value that the state must return in the
event of a corresponding adjustment is IDR 1,573,284,194 x 0.5% or IDR 7,866,421.
The state received an additional revenue of IDR 1,046,788,989 and a potential loss of
IDR 19,116,420, so the value of net state revenue from domestic transfer pricing
supervision activities amounted to IDR 1,027,672,568 or in other words, in the case of
PT N there is no zero-sum game.
4.3. SP2DK PT M Tax Year 2018 (CRM TP)

PT M is the holding company of the M Group. The most significant income for
taxpayers comes from divestment and dividends from subsidiaries of the M Group.
Dividends from subsidiaries include deemed dividends from equity participation jointly
with PT R, a member company of the M Group.
The main issue related to the supervision of PT M is Article 9 of Government
Regulation 94 of 2010. It is related to gains on foreign exchange for taxpayers who
receive final income, which is not taxable objects. The existence of honorarium payment
transactions to family members of the management of PT M. Taxpayers transfer shares
associated entity, namely PT M1, with a gain on the transfer of IDR 94,307,637,763. PT
M also received dividend income from its subsidiary for IDR 39,730,782,539, gain on
foreign exchange of IDR 6,773,515,605, and gain on equity participation in the company
P2P Lending and deposit interest of IDR 1,841,970,338. PT M does not declare any
related transactions in Appendix VI and Appendix 3A of the Corporate Income Tax
Return. The financial statements of PT M researchers are presented in Table 4.
Table 4 PT M Financial Statement

Profits from the divestment of shares 94.307.637.763
Dividend 39.730.782.539
Bank interest income and P2Plending 1.841.970.338
Other income 50.743
Gain on foreign exchange 6.773.515.605
Total Revenue 142.653.956.988
Salary expense 287.553.619
Insurance fee 45.000.000
67
Miscellaneous expense 26.685.648
Legal and professional fees 8.980.000
Honorarium fees 31.021.428.571
Health costs 2.143.500
Total cost 31.391.791.338
Profit before tax 111.262.165.550
Negative Fiscal Correction 43.379.722.607
Positive Fiscal Correction -
Taxable income 67.882.443.043
Income tax 16.970.610.761
Net profit 94.291.554.889
PT M1 owns the capital ownership structure of PT M at 22.22%, PT M2 at 32.56%,

and the Main Director of PT M, namely Pres.Dir M, at 45.22%. PT M1 and PT M2 are
members of the M Group and have a special relationship with PT M. The Managing
Director of M places his family members as administrators of several M Group
companies. The second child of the Managing Director of M, namely Dir. M2, became
the Director of PT M. similarly, the first child of the Pres.Dir M, namely Pres.Dir M1 as
the President Director of PT M4, became a company whose shares were transferred in
2018. In simple terms, the corporate structure of the M Group is shown in Figure 3.
66,3%
PT M2
PT M1 32,56%
PT M Pres.Dir
M
Pres. Dir : Pres.Dir M
22,22% 18,7%
31,5% Pres.Dir
PT M4 M1
Pres.Dir : Pres.Dir. M1 1,8%
Figure 3 Company Affiliation Relations in the M Group

Source: researcher processed
PT M, PT M1, and members of the M Group who own PT M4 transferred shares of

PT M4 to other parties. PT M recorded a profit of IDR 94,307,637,763 from the
divestment transaction. PT M charges an honorarium fee to the CEO of M1 and Mrs.
President Director M, the wife of President Director M of IDR 7,000,000,000. After finding
the similarity in Pres.Dir M, Pres.Dir M1, and the wife of President Director of M, AR
realized that the honorarium was paid to a family member of Pres.Dir M.
AR then indicated the honorarium payment as a fee for the personal needs of the
Taxpayer or his dependents by the provisions of Article 9 paragraph (1) letter i of the
Income Tax Law.
The Taxpayer's attorney then provides evidence and supporting documents in the
form of a Directors Circular Decision regarding honorees related to the transfer of PT
68
M4 shares. PT M Directors Circular Decree does not regulate the basis of remuneration
and performance indicators of honorarium recipients. AR PT M then diverted suspicion
from Article 9 paragraph (1) letter i of the Income Tax Law to an alleged particular
transaction. AR PT M asked for evidence that Pres.Dir M1 and the wife President
Director of M, rendered his services and expertise to PT M in the context of the
divestment of PT M4 shares to a Taxpayer's Attorney. PT M's attorney did not provide
data to PT M's AR regarding the aspect of providing services that had been carried out
by the Pres.Dir M1 and the wife of President Director M. However, he is willing to make
adjustments in the form of a price fairness correction of 50% of the honorarium for
Pres.Dir M1 and the wife of President Director M. Taxpayer's attorney assumes that the
services have been done by the President Director M1 and the wife of Director M.
However, PT M did not make transfer pricing documentation for the transaction. PT M
subsequently made a positive fiscal correction of IDR 3,500,000,000 for the honorarium
expenses. As a result, PT M's Corporate Income Tax from the domestic transfer pricing
correction increased by IDR 875,000,000. In addition, the sanction based on Article 8
paragraph (2) increased to IDR 207,900,000. Consequently, state revenue from PT M's
domestic transfer pricing correction became IDR 1,082,900,000.
In the case of PT M, the issue of domestic transfer pricing is not material because,
at first glance, the Article 21 PPh rate for individuals of 30% is much higher than the
Corporate Income Tax rate of 25%. That is partially true because the components of the
distribution of money to the Pres.Dir M1 and the wife of the President Director M is the
distribution of dividends from PT M to shareholders (in this case, his family members).
So PT M is still obliged to deduct PPh Article 4 paragraph (2) of the individual dividend
by 10%. Therefore there is an issue of a 5% difference in tariffs in the case of PT M.
Table 5 Comparison Simulation of Before and After Domestic TP corrections

Source: researcher processed
No. Tax Type Taxes Payable
a. Initial conditions
1 Corporate Income Tax Rp16.970.610.761 (See Table 4)
2 Income Tax 21 Rp2.100.000.000 (30% x 7.000.000.000)
3 Income Tax 4 par.(2) Rp0
Total Income Tax Payable Rp.19.070.610.761
b. condition after correction of domestic TP
1 Corporate Income Tax Initial taxable income: 67.882.443.043
Correction honor fee: 3.500.000.000
Final taxable income: 71.382.443.043
Corporate Income Tax: 17.845.610.761
2 Income Tax 21 Rp1.050.000.000 (30% x 3.500.000.000)
3 Income Tax 4 par.(2) Rp350.000.000 (10% x 3.500.000.000)
Total Income Tax Payable Rp19.245.610.761
Difference post-correction conditions Rp175.000.000
Based on the calculations in Table 5 above, it is again proven that the domestic
transfer pricing correction is not a zero-sum game correction. This correction generates
an additional potential for state revenue of IDR 382,900,000 originating from a potential
of IDR 175,000,000 (if the Taxpayer makes a corresponding adjustment) plus notice of
tax Article 8 paragraph (2) of IDR 207,900,000.
69
4.4. SP2DK PT V Tax Year 2020
PT V is the holding company of group V. PT V's primary revenue comes from
dividend income and the transfer of shares of subsidiaries, associates, or joint ventures
controlled by PT V. PT V, as the holding company of group V, does not prepare
consolidated financial statements and only makes financial statements for himself as a
reporting entity. PT V's tax attorney explained that there were no consolidated financial
statements because one family controlled PT V's majority shares, and the family did not
need consolidated financial statements.
PT V 2020 sold its share ownership in PT V1 of 33.3% to PT Z for IDR
45,676,929,773. PT V also recorded dividend income from one of the associated entities
owned by PT V of IDR 10,328,551,974. Taxpayers recorded an interest expense
component of IDR 2,820,600,000 and investment costs for PT V1 shares of IDR
35,187,000,000. The financial statements of PT V researchers show in Table 6.
Table 6 PT V Financial Statement

Share divestment income 45.676.929.773
Dividend 10.328.551.974
Deposit Interest Income 15.676.164
Total income 56.021.157.911
Salary expense 134.377.036
Bank Administration fee 490.000
Permit cost 800.000
Investment costs 35.187.000.000
Loan interest costs 2.820.600.000
Consultant fee 400.000.000
Total cost 38.543.367.036
Profit before tax 17.477.790.875
Negative Fiscal Correction 10.344.228.138
Positive Fiscal Correction -
Taxable income 7.133.562.737
income tax 1.569.383.802
Net profit 5.564.178.935
Based on the data in Table 6, the biggest cost structure for taxpayers comes from
investment costs and interest costs. The investment cost is the cost of acquiring PT V1's
share ownership. The value matched with changes in PT V1's company deed related to
the amount of authorized capital and the amount of paid-up capital of PT V1 so that AR
does not make any corrections to PT V1's investment costs. The interest expense of
IDR 2,820,600,000 came from PT V's loan to PT V2. The interest expense is related to
PT V's loan transaction to PT V2 in 2018, amounting to IDR 8,945,000,000. PT V paid
off part of the loan in 2019, so the balance owed to PT V2 as of December 31, 2019,
amounted to IDR 6,395,000,000. PT V does not recognize interest expense on an
accrual basis in 2018 and 2019 and does not disclose debt transactions to PT V2 in
Attachment VI and Special Attachment 3A of the Corporate Income Tax Return.
PT V's notable relationship transactions were also not detected in the DGT's CRM
TP, even though there was proof of withholding Article 23 PPh. Hence, AR PT V
suspects that the PT V case and its affiliated parties did not appear in CRM TP risk
because PT V and PT V2 are at the same tax rate. PT V reduced the share of PT
70
ownership in PT V2 to 24.12% from the value that should have been recorded, 99%.
The affiliation scheme of researchers is shown in Figure 4.
Information : 99% 1%
X Ltd
PT V
Hongkong
33,3%
Ownership relationship
PT V1 PT V2
Debt and interest transactions

Figure 4 Group Interest and Special Relationship Transaction Scheme
Based on the indication of a special relationship above, AR PT V made SP2DK-

592 and asked about the fairness of charging PT V's loan interest costs to PT V2. AR
uses JIBOR data as the fair interest rate. AR uses the JIBOR interest rate because AR
does not know the level of financing risk owned by PT V. The JIBOR interest rate in
2020 is 4.57 %/year.
PT V's proxy agreed to a correction based on the fairness of interest rates but did
not agree if the reasonableness of interest rates was measured using JIBOR interest
rate data. Proxy of PT V expressed willingness to correct the interest rate if using the
interest rate of one of the private banks in Indonesia of 10.75% in 2018, 11.25% in 2019,
and 10.5% in 2020. With this reasonable interest rate, the Taxpayer, on a self-
assessment basis, corrects the 2020 Corporate Income Tax Return. The Taxpayer
makes a positive fiscal correction of interest expense of IDR 468,100,000. The
correction came from calculating interest expense using the interest rate at one of the
private banks amounting to IDR 2,352,500,000 compared to the previous interest
expense of IDR 2,820,600,000.
Regarding the corresponding adjustment, the amount of negative fiscal
correction made by PT V will be the same as that of positive fiscal correction. However,
in the case of PT V, it also does not result in a zero-sum game because there is a
potential for corporate income tax bills Article 8 paragraph (2) of the Law KUP to PT V
for IDR 19,105,162. In addition, the corresponding adjustment by PT V2 will be one of
the checkpoints because PT V2 will be overpaid. AR PT V has suspicions that the
location of the transfer pricing transaction is actually in PT V2 because X Ltd, as the
holder of 1% shares in PT V2, has a legal domicile in Hong Kong, which has a corporate
income tax rate of only 16.5% (Delloite, 2020). That is in line with one of the three basic
interest cost transfer pricing schemes, according to DGT (2016), where the group places
a high amount of interest in countries with high tax rates. Even though it only controls
1% of PT V2's shares, X Ltd has a special relationship with PT V2 and PT V because X
ltd controls PT V2 technologically. Based on data from the World Intellectual Property
Organization (WIPO), X Ltd is the trademark owner for products produced by PT V2.
The loan extension from X Ltd by PT V2 to PT V1 can also be used in PT V2's transfer
pricing document as material for analyzing the substance and loan requirements of PT
V2 to X Ltd.
71
5. CONCLUSION
The domestic transfer pricing correction case is not a zero-sum-game condition
for national tax revenues, for the PT N case, additional state revenue from the domestic
transfer pricing correction even reached more than IDR 1,000,000,000 (including the
sanctions in Article 8 paragraph (2) of the KUP Law). The domestic transfer pricing
regulations in Article 2 paragraph (2) emphasize the importance of correcting domestic
transfer pricing in cases of difference in rates or compensation for losses. However, in
the case of the same tax rates or as if they were the same in the cases of PT M and PT
V, there is still additional state revenue even with the condition after the corresponding
adjustment was made.
Tax practitioners, especially AR and the DGT Tax Examination Team, should be
quick to make corrections to domestic transfer pricing because, in the case of correction
of intra-group services, a non-zero-sum game condition is found for national revenue.

Based on the research results, all taxpayers in the research sample did not
disclose their special relationship. DGT needs to make an inventory of related transact
are not declared by taxpayers, especially taxpayers at KPP Pratama. DGT also needs
to add variables to the CRM TP component for related party transactions not
accompanied by tax documents, such as tax invoices and proof of withholding.
This study has limitations because it does not explain the sample relating to related
transactions in the context of buying and selling and the use of intangible taxable goods
between taxpayers and domestic affiliates.
AWARD
My most profound appreciation goes to Mr. Sigit Sugiharto, Staff of the
Examination Section for Transfer Pricing and Other Special Transactions, and Mr.
Hendri Senopati, Head of Supervision Section I, KPP Pratama Jakarta Gambir Dua.
BIBLIOGRAPHY
Ardianto, A. dan Dyan, R.(2018). Strategi Diversifikasi, Transfer Pricing dan Beban
Pajak. Jurnal Keuangan dan Perbankan, 14(2), 45-53
Deloitte.(2021).Hongkong Tax and Investment Guide 2021. Hongkong: Deloitte Touche
Tohmatsu. Diakses pada tanggal 13 Agustus 2022, dari
https://www2.deloitte.com/content/dam/Deloitte/cn/Documents/tax/deloitte-cn-tax-
hk-tax-guide-en-210928.pdf
Directorat General of Taxes.(2016). Panduan Analisis Risiko Transfer Pricing di Seksi
Pengawasan dan Konsultasi. Jakarta: Direktorat Potensi Kepatuhan dan
Penerimaan. Tersedia dari Tax Knowledge Base Direktorat Jenderal Pajak.
Director General of Taxes.(2011). Regulation Number PER-32/PJ/2011 number PER-
32/PJ/2011 concerning Amendments to Regulation of the Director General of Taxes
number PER-43/PJ/2010 concerning Application of the Principles of Fairness and
Business Usability in Transactions between Taxpayers and Parties with Special
Relationships.Jakarta:DGT
Director General of Taxes.(2013). Director General of Taxes Circular Number SE-
50/PJ/2013 concerning Technical Instructions for Examination of Taxpayers with
Special Relationships.Jakarta:DGT
Doly, Taripar.(2016).Transfer Pricing Dalam Negeri. Diakses 11 Agustus 2022, dari
https://nusahati.com/2016/05/transfer-pricing-dalam-negeri/
72
Indonesian Chartered Accountant.(2015). The Statement of Accounting Standard 07
(Revised 2015) Disclosures of Related Parties.Jakarta:IAI
Finance Minister.(2020). Minister of Finance Regulation Number 22/PMK.03/2020
concerning Procedures for Implementing Transfer Price Agreements (Advance
Pricing Agreement). Jakarta: Ministry of Finance
Kurniawan, Anang Mury.(2015). Buku Pintar Transfer Pricing Untuk Kepentingan Pajak.
Yogyakarta: Andy
Organization of Economic Co-operation and Development. (2022). OECD Transfer
Pricing Guidelines for Multinational Enterprises and Tax Administrations 2022, OECD
Publishing, Paris, https://doi.org/10.1787/0e655865-en
PWC.(2013).Domestic Transfer Pricing Navigating New Challenges. Diakses pada 26
Juli 2022, from https://www.pwc.in/assets/pdfs/publications/2013/domestic-transfer-
pricing-navigating-new-challenges-booklet.pdf
Prihatsanti, Unika, Suryanto, Hendriani dan Wiwin.(2018).Menggunakan Studi Kasus
sebagai Metode Ilmiah dalam Psikologi.Buletin Psikologi, 26(2), hal. 126-136.
https://10.22146/buletinpsikologi.38895
Republic of Indonesia.(2021). Law of the Republic of Indonesia Number 7 of 1983
concerning Income Tax as amended several times last by Law of the Republic of
Indonesia Number 7 of 2021.Jakarta: State Secretary
concerning General Provisions and Tax procedures as amended several times last
by Law of the Republic of Indonesia Number 7 of 2021.Jakarta: State Secretary
concerning Value Added Tax as amended several times last by Law of the Republic
of Indonesia Number 7 of 2021.Jakarta: State Secretary
Republic of Indonesia.(2010). Government Regulation Number 94 of 2010 concerning
Calculation of Taxable Income and Payment of Income Tax in the Current
Year.Jakarta:State Secretary
Republic of Indonesia.(2018). Government Regulation Number 23 of 2018. Income Tax
on Income from Business Received or Obtained by Taxpayers Who Have a Certain
Gross Circulation Sales Number.Jakarta:State Secretary
Setiawan, Hadi. (2013, September). Transfer Pricing dan Risikonya Terhadap
Penerimaan Negara. Buletin Info Risiko Fiskal (IRF) Edisi III tahun 2013. Diakses dari
https://djppr.kemenkeu.go.id/dukungan/publikasiViewer/6234
Simamora, Beatrice Eka Putri P&Hermawan, Ancella Anitawati.(2017).Transfer Pricing
Analysis on Intra-Group Services and the Related Transfer Pricing Disputes from
Indonesian Tax Perspectives. Advances in Economics, Business, and Management
Research (AEBMR), 55(6), pages 24-28. https://doi.org/10.2991/iac-17.2018.5
Sitanggang, Raymondo dan Firmansyah, Amri.(2021).Transaksi dengan Pihak Berelasi
dan Praktik Transfer Pricing di Indonesia. Jurnal Pajak dan Keuangan Negara PKN
STAN, 2(2), hal.34-52
Supriyadi. (Presenter). (2021, Oktober). Transaksi Afiliasi Domestik Berisiko Transfer
Pricing [Video Pembelajaran]. Jakarta: Kemenkeu Learning Center. Diakses pada 15
Juli 2022,https://klc2.kemenkeu.go.id/knowledge/transaksi-afiliasi-domestik-
berisiko-transfer-pricing-5bc5cfd9
Wildan, Muhammad. (2022). DJP Janji Seriusi Penanganan Transfer Pricing, Seperti
Apa? Diakses 15 Juli 2022, dari https://news.ddtc.co.id/djp-janji-seriusi-penanganan-
transfer-pricing-seperti-apa-37257
73
REVIEW OF THE EMPLOYEE PERFORMANCE
EVALUATION SYSTEM AT THE DIRECTORATE
GENERAL OF TAXES: ITS EFFECTIVENESS AND
EMPLOYEE SATISFACTION LEVEL
Anies Said Basalamaha, Herru Widiatmantib
aLeadership and Managerial Training Center, Financial Education and Training Agency, Ministry of
Finance, Indonesia. Email: asbasalamah@yahoo.com
bLeadership and Managerial Training Center, Financial Education and Training Agency, Ministry of
Finance, Indonesia. Email: widiatmanti@gmail.com
ABSTRACT
This quantitative research seeks to determine the effectiveness of the Directorate General of Taxes
(DG Tax) performance evaluation system and measure how satisfied the employees are, and to prove
differences among different categories of employees. This research is an extension of the research
by Widiatmanti (2020) towards 1,587 DG Tax employees. ANOVA and EFA were used to analyze the
data in addition to descriptive data analysis. Generally respondents feel the system is quite effective,
and they are quite satisfied, except for three variables that the majority of respondents strongly
disagree and disagree that relate to performance status of SABCD and the level 1 and level 2 ranking
sessions. ANOVA method indicates that there are indeed differences among different categories of
employees, while EFA method reveals that despite seven insignificant variables, none of them
indicates the need to change DG Tax Regulation 12/PJ/2018, including its SABCD performance
status. However, the need for change or adjustments arises because there are provisions at higher
level of regulations: Government Regulation and two ministerial decrees. Therefore, we suggest that
the use of forced rank in the SABCD performance status is abolished and be replaced mainly due to
excellent (S) and poor (D) performance.
Keywords: Performance Evaluation System; DG Tax; Exploratory Factor Analysis (EFA)
Penelitian kuantitatif ini bertujuan untuk mengetahui efektivitas sistem evaluasi kinerja Direktorat
Jenderal Pajak (DJP) dan mengukur seberapa puas pegawai, serta membuktikan perbedaan antara
kategori pegawai. Penelitian ini merupakan pengembangan dari penelitian Widiatmanti (2020)
terhadap 1.587 pegawai Ditjen Pajak menggunakan metode ANOVA dan EFA untuk menganalisis
data di samping analisis data deskriptif. Secara umum responden merasa sistem ini cukup efektif dan
cukup puas, kecuali terhadap tiga variabel yang mayoritas responden sangat tidak setuju dan tidak
setuju. Dua variabel berhubungan dengan status kinerja SABCD, dan satu lainnya berkaitan dengan
sidang pemeringkatan tahap 1 dan tahap 2. Metode ANOVA menunjukkan bahwa memang ada
perbedaan antara kategori pegawai, sedangkan metode EFA mengungkapkan bahwa meskipun ada
tujuh variabel yang tidak signifikan, tidak ada satupun variabel yang menunjukkan perlunya mengubah
Peraturan Ditjen Pajak 12/PJ/2018, termasuk status kinerja SABCD-nya. Namun, kebutuhan akan
perubahan atau penyesuaian itu muncul karena ada ketentuan di tingkat peraturan yang lebih tinggi:
Peraturan Pemerintah dan dua keputusan menteri. Oleh karena itu, kami menyarankan agar
penggunaan peringkat dengan angka tertentu untuk setiap kategori (forced ranked) dalam status
kinerja SABCD dihapuskan dan diganti semata-mata berkaitan dengan kinerja dari yang sangat baik
(S) sampai yang buruk (D).
Kata kunci: Sistem Evaluasi Kinerja; Direktorat Jenderal Pajak; Exploratory Factor Analysis (EFA)
74
1. INTRODUCTION
Law No. 5 of 2014 concerning State Civil Apparatus, and Government Regulation
No. 11 of 2017 concerning Management of Civil Servants which has been amended by
Government Regulation No. 17 of 2020, have significantly changed the management of
civil servants, especially with regard to career management, career development,
competency development, career patterns, transfers to other units, and promotions. This
change is related to the application of the merit system to improve civil servants’
competencies, performance, and professionalism. In implementing the merit system, an
objective, measurable, accountable, participatory and transparent employee
performance assessment is needed so that professional, agile, high-integrity and high-
performing civil servants can be achieved. Moreover, government has also enacted
Government Regulation No. 30 of 2019 concerning Civil Servants Performance
Assessment which regulates, among other things, the substance of performance
appraisal consisting of work behavior assessment and performance appraisal, weighting
method of employee performance standards, appraisers and performance appraisal
teams, performance information systems, assessment procedures, performance ratings,
performance reporting, performance rewards, punishments and objections.
One of the applications in the ministerial level can be seen in the Directorate
General of Taxes (DG Tax) of the Ministry of Finance (MoF). Its employees receive not
only the highest performance allowance compared to other employees at MoF, but also
to all other civil servants. Performance appraisal at DG Tax as regulated in the Director
General of Taxes Regulation No. PER-12/PJ/2018 concerning Performance
Management within the DG Tax is using the e-performance application, the same as the
performance measurement applicable to other echelon I units within MoF. However,
employee performance rating (1 to 5) and the employee performance status (SABCD
where S represents the best and D represents the worst) are only applicable to the DG
Tax employees. Table 1 depicts not only the five categories for each status but also
shows the existence of a forced distribution. The performance status of employees at
DG Tax is one of the criteria in providing performance allowances and promotions
(Widiatmanti, 2020). According to the DG Tax Regulation No. 12 of 2018 as indicated
by Widiatmanti (2020), in addition to providing categories to the best employees in order
to give awards to them, there are also other purposes such as for employee
management, development of a conducive and competitive work climate, means of
effective communication and the establishment of a harmonious relationship between
subordinates and superiors, increasing employee job satisfaction, and developing an
effective work culture so that employees are able to make optimum contributions.
Table 1 Performance Status Categories of DG Tax Employees

Source: Widiatmanti (2020)
Conversion Status
Categories of
Employee Performance Status Employee
Employee
Percentage Performance
Performance Status
Achievement (%)
15% Best Performance S (15%) 100,0
A (20%) 97.5
70% Average B (30%) 95.0
C (20%) 92.5
75
15% Below Average D (15%) 90.0
The implementation of performance appraisal using forced distribution which has

an impact on employee performance allowance and promotion raises the reluctance and
anxiety of superiors who assess and rank, in addition to being perceived or even
expressed as unfair by employees who are assessed and rated. From the authors' initial
observations with several employees, especially those who involved in the Leadership
Development Program training activities at the Center for Education, Training and
Development of Human Resources, some employees told how difficult they were when
facing employees during the ranking process, in addition to the difficulty in motivating
employees affected by forced distribution for each office unit. Moreover, according to
them, performance status also turned out to be one of the criteria for organizations and
leaders in transfering employees from one unit to another. These phenomena not only
contradictory to the purpose of the DG Tax Regulation No. 12 of 2018 as indicated earlier,
but also lead to employees dissatisfaction, both those who assess and those who are
assessed, regarding the determination of the distribution per category of employee
performance status. This dissatisfaction inspires the authors to conduct research to
answer the following research questions: what is the perception of DG Tax employees
regarding the current performance appraisal model? If there are few things need to be
changed, what are they?
2. LITERATURE REVIEW
Performance evaluation is not only a systematic evaluation of the performance of
employees but also understands their abilities so that it can be used to plan further
career development for the employees concerned. Torrington et al. (2020) define job
evaluation as a “formal, systematic process to determine the relative worth of jobs within
an organization,” and it can be done in several ways or methods: ranking method that is
appropriate in small organizations, classification method or job grading that is often used
in public sector organizations in which jobs are categorized into groups, and factor-
comparison method that combines the ranking and point factor methods using
quantitative data. The latter is the most widely used one (Torrington et al., 2020).
According to Mathis et al. (2017), there are many methods that can be used to
evaluate performance, either one method for all jobs and employees, different methods
for different groups of employees, or a combination method. Rating scale, comparative
methods which compare the performance levels of employees against one another that
include ranking and forced distribution, and narrative methods such as critical events or
other essay methods are tools for evaluating performance. Another tool is to use
psychologists or assessment center to assess the potentials of employees (Mathis et
al., 2017; Dessler, 2020; Torrington et al., 2020).
Research on the effectiveness of performance evaluation is inconclusive. Sharma
et al. (2016) found that accuracy and fairness are two factors that contribute
performance management systems (PMS) effectiveness. Dewettinck & van Dijk (2012)
also found that fairness mediates the relationship between characteristics of employee
PMS and their effectiveness. Torrington et al. (2020) indicate that “performance
management must have credibility with employees,” while Blackman et al. (2018) argue
that to be effective, PMS have to be operated as part of the organization’s core business.
The research results are also inconclusive. Mathis et al. (2017) quotes a report that
indicates “86% of firms actively use performance management, around 70% of those
surveyed thought that the process was not positive, and 29% believed that is was unfair.
76
A mere 3% of companies plan to alter their current approaches.” Evita et al. (2017)
indicates that the Graphic Rating Scale and employee daily work reports are considered
ineffective because many employees consider both methods are merely a formality,
subjective, and there are no clear and measurable standards as well as feedback on
employee performance achievements. As a result, employees feel uncomfortable and
unmotivated at work.
Research by Harbi et al. (2017) links performance appraisals with Saudi Arabian
culture in the form of wasta personal relationships (literally, wasta can be interpreted as
connections or favoritism) which cause people who do not have close relationships to
feel unfairly treated and lead to negative outcomes. They also show how employees
begin to reject this wasta norms and adopt alternative values related to the notion of
organizational justice and individual egalitarianism (Harbi et al., 2017). Bias by raters is
also indicated by Mathis et al. (2017) as one of the major sources of error in the
performance appraisal process.
This research is an extension of the authors’ quantitative and qualitative research
(2022) using questionnaires as attached. Although the results show that in general
respondents are satisfied with the system with most of the modes are 3 indicating
respondents are agree, and the system is perceived to be quite effective, respondents
who chose Strongly Disagree and Disagree for the negative tones of the questions were
quite a lot. The interview also reveals that all reviewees hoped that the SABCD
performance status is abolished since it causes injustice, reduces employee motivation,
reduces collaboration efforts, and creates organizational uproar (Widiatmanti, 2020). To
know which factors from X1 to X38 need to replace or abolish as suggested by
interviewees, this study was done using the theoretical framework shown in Figure 1.
The analysis used to achieve the first research objective is the exploratory factor
analysis (EFA), while the ANOVA test was used to fulfill the second objective.
The population in this study was 45,948 all DG Tax employees (Biro SDM, 2020)
that spread over the work units of the Head Office, 34 Regional Tax Offices, 352 Tax
Offices and 204 Tax Counseling and Consulting Offices. The questionnaire was sent
using Google Forms. However, the number of respondents who filled out the
questionnaires was only 1,684 or 3.665% response rate which might be small compared
to the average internet survey. Nevertheless, using the Slovin formula, the level of
achieved tolerable error is 2.392% which is more conservative than 5% commonly used.
In other words, 1,684 samples returned is higher than the minimum sample according
to Slovin formula for 5% tolerable error. However, since there are few respondents who
filled the questionnaire twice or even three times, only 1,587 respondents were
processed.
4. ANALYSIS AND DISCUSSIONS

4.1. Descriptive Data
Of the 1,587 respondents, 1,068 are male (67.30%) and 519 are female (32.70%);
1,247 respondents are married (78,58%), 307 respondents are single (19.34%) and 33
respondents are either widows or widowers (2.08%). Meanwhile, based on their age,
56 respondents are over 55 years old (3.53%); 422 respondents are over 45 up to 55
years (26.59%), 453 respondents are over 35 up to 45 years (28.54%), 460 respondents
are over 25 up to 35 years (28.99%), 186 respondents are over 20 to 25 years (11.72%),
and 10 respondents (0.63%) aged up to 20 years.
77
As for the respondents’ positions, 882 respondents are staffs (55.58%), 193
respondents are functional officers (12.16%), 444 respondents are echelon IV or
supervisors (27.98%), 63 respondents are echelon III or administrators (3.97%), and 5
respondents are echelon II (0.32%). Based on their educational level, 508 respondents
are high school graduates or equivalent (32.01%), 209 respondents are Diploma I
(13.17%), 194 respondents are Diploma III graduates (12.22%), 623 respondents are
S1 or Diploma IV graduates (39.26%), 43 respondents are S2 graduates (2.71%), and
10 respondents are S3 graduates (0.36%).
78
DG Tax Performance
Affected by:
EMPLOYEE PERFORMANCE
Measured by:
PERFORMANCE EVALUATION MODEL BASED ON DG TAX REGULATION

No. 12/PJ/2018
(Dis)Satisfied Employees
Phenomenon
WHY?
Due to the Effectiveness of

Performance Appraisal? Due to its
1. Accuracy. 1. Regulation?
2. Fairness. 2. Components?
3. Credibility.
3. Stages?
4. As part of the organization’s core
business. 4. Model?
5. Impartial. 5. Information?
6. Unbiased.
Quantitative
Is It Effective? Is It Satisfactory?
Analysis
1. Mapping of Performance Measurement Problems

2. Which Areas of the Regulation No 12 of 2018 Need to be Improved?
Figure 1. Research Theoretical Framework

Source: Adapted from Widiatmanti (2020)
In terms of the length of time the respondents worked, 208 respondents have up to
five years of service (13.11%), 277 respondents have five to ten years of service
(17.45%), 270 respondents have more than ten to 15 years of service (17.01%), 537
respondents have more than 15 to 20 years of service (33.84%), 214 respondents have
79
more than 20 to 30 years of service (13.48%), and 81 respondents have more than 30
years of service (5.01 %). Combining this characteristic with the respondents’ age that
only ten respondents whose age are up to 20 years, most respondents are
representative enough in a sense that they are able to distinguish between the old DG
Tax performance evaluation method and the new one.
Few respondents seem to hide their identities, such as echelon III officer with the
length of service of less than ten years which is impossible to achieve at DG Tax, or with
age of up to 20 years but having years of service of more than 5 years. Other respondent
chose 3 on the age meaning his or her age is 25 – 35 years, yet he or she chose 6 for
tenure, meaning he or she has already work for more than 30 years. However, since
EFA method does not distinguish whether the respondents are male or female, officials
or staff and so on, the 1,587 respondents were continued to be processed since all of
them filled out the questionnaire completely. However, for the purpose of distinguishing
among the categories of respondents, these weird respondents will be dropped from the
calculation.
To achieve the objectives of this research, the questionnaire sent was divided into
four groups. The first group relates to the performance appraisal variables, the second
one relates to the performance appraisal model variables, the third group relates to the
effectiveness of the performance appraisal model variables, and the fourth group relates
to the performance appraisal regulatory variables. Table 1 of the Appendix details the
answers to the questionnaire along with each median and mode. As can be seen in
Table 1 of the Appendix, all variables were responded by both those who strongly agreed
(4) and those who strongly disagreed (1). This shows that statistically their answers
support the author's initial observations because the questions in the questionnaire are
partly derived from these initial observations. For example, for question A2 that asks
whether respondents are satisfied with the results of the performance appraisal they get,
although the majority of respondents chose satisfied (3 or agree) and very satisfied (4
or strongly agree), 84 respondents chose very disagree (1) and 220 chose disagree (2).
Similarly, although most respondents chose agree (3) and strongly agree (4) in almost
all questions, there are still many respondents who chose disagree (2) and strongly
disagree (1).
Three questions that the majority of respondents chose strongly disagree (1) and
disagree (2) are the accuracy of using the normal curve for the performance status of
SABCD employees (question C8), level 1 and level 2 ranking sessions are publicly
informed to employees (question C11), and the performance status of SABCD motivates
employees to improve group performance (question D8) with median of 2, respectively
and mode of 2, 2, and 3, respectively. As with question D8 that performance status
(performance categories SABCD) motivates employee to improve group performance,
even though the median is also 2 which means disagree, the number who chose Agree
is still more than those who chose Strongly Disagree, Disagree and Strongly Agree with
the mode 3 which means agree. For all other variables, the modes and medians are 3
which means Agree. The only variable that both median and mode are 4 is question B3
namely performance is assessed within a certain period of time (see Table 1 of the
Appendix). As shown in Table 1 of the Appendix, all variables were responded, both by
those who chose strongly agree (4) and those who chose strongly disagree (1) with
almost all medians and modes are 3.
Descriptive statistics also shows similar results (Widiatmanti, 2020). Using the
criteria as shown in Table 2, for group A a total value of 51,107 was obtained, namely
the total who answered 1 multiplied by 1 plus the total who answered 2 multiplied by 2
plus the total who answered 3 multiplied by 3 plus the total who answered 4 multiplied
80
by 4, or (1,308 X 1) + (3.447 X 2) + (7.903 X 3) + (4.799 X 4). With a total of 1,587
respondents, an average value of 32.20 (i.e., 51,107:1,587) is obtained. Referring to
Table-2, the value of 32.20 is in the criteria of "Quite Satisfactory." Meanwhile for group
B a total score of 29,913 was obtained so that an average value of 18.85 was also
included in the criteria of "Quite Satisfactory." Meanwhile for group C a total value of
48,716 was obtained so that an average value of 30.70 was also included in the criteria
of "Quite Satisfactory." As for group D, the total score is 42,878 so that an average value
of 27.02 is obtained which is included in the criteria of "Quite Effective." This indicates
that there is still room for improvement for DG Tax so that employee dissatisfaction can
be reduced or even eliminated as well as increasing the effectiveness of performance
appraisal at DG Tax.
Table 2 Criteria for Performance Appraisal Variables

Total Score of Total Score of Total Score of Total Score of
Criteria Criteria
Group A Group B Group C Group D
11 – 19.3 6 – 10.5 11 – 19.3 Unsatisfactory 10 – 17.5 Ineffective

19.4 – 27.7 10.6 – 15.1 19.4 – 27.7 Less Satisfactory 17.6 – 25.1 Less Effective
27.8 – 36.1 15.2 – 19.7 27.8 – 36.1 Quite Satisfactory 25.2 – 32.5 Quite Effective
36.2 – 44.5 19.8 – 24.3 36.2 – 44.5 Satisfactory 32.6 – 40.1 Effective
Total Answers ∑ choices 1 + ∑ choices 2 + ∑ choices 3 + ∑ choices 4

Average = =
1.587 1.587
Another descriptive analysis that can be explored is to compare whether or not the
population means differ among various types of DG Tax employees who participated in
the survey. As indicated earlier, few respondents seem to hide their identities. As a result,
categories of tenure and age were not be processed in differentiating population means
among various types of DG Tax employees. Using ANOVA, it is found that the lowest
and highest average variables are 2.268 and 3.487, respectively. It can be seen from
Table 3 that the p-value is 0, so it can be concluded that there are significant differences
in the population average among the five types of DG Tax employees, i.e., staffs,
functional officers, echelon IV, echelon III and echelon II. This means that on average
the five types of employees at DG Tax gave different answers.
4.2. Validity and Reliability Tests

To assure that exploratory factor analysis (EFA) is the correct measuring instrument,
several tests should be carried out before this method is used as a “requirement” that
EFA is a valid tool. The first validity test is using Pearson correlation test to assess
whether or not data is correlated with one another. Of the 1,587 respondents who
completed the questionnaire, all variables have significant correlations because each
has a p-value of < .001. Therefore, in further testing, no variables need to be eliminated
as a treatment (Hair et al., 2019).
Table 3 ANOVA for Different Types of Respondents

Source: Survey results.
P-
Categories SS df MS F F crit
value
81
Gender 11857.45 38 312.04 434.03 0.000 1.405027
Marital Status 7710.60 38 202.91 278.17 0.000 1.405027
Positions 7161.43 38 188.46 255.17 0.000 1.405027
Educational Levels 5991.38 38 157.67 211.74 0.000 1.405027
Places of Work 4840.38 38 127.38 152.27 0.000 1.405027
Normality test in multivariate analysis is a basic assumption where if the variation

from the normal distribution is large enough, then all statistical methods used will be
invalid (Hair et al., 2019). In this study we used the Shapiro-Wilks test which is one of
the two most popular methods (Hair et al., 2019). Using Jamovi program, the lowest
Shapiro-Wilks value is 0.718 and the highest is 0.908 but each variable has a low p
value of < .001. Therefore, following the opinion of Zaiontz (2017) that testing whether
the data follows a normal distribution multivariate analysis is difficult, and for large
samples such as in this study, it usually relies on the multivariate central limit theorem,
i.e., for a certain set of “random vectors X1, X2, …, Xk that are independent and
identically distributed, then the sample mean vector, X̄, is approximately multivariate
normally distributed for sufficiently large samples” (Zaiontz, 2017). Accordingly, all of the
38 variables remain being processed.
Reliability testing carried out on 38 variables results in a Cronbach alpha value of
0.975 with the lowest value of 0.974 and the highest of 0.976, all of which is greater than
the lowest acceptable reliability limit of 0.6 or 0.7 (Hair et al., 2019).
An autocorrelation test was conducted using the Durbin–Watson test to determine
whether there is a relationship between the elements of a series of observations in order
to eliminate the effect of standard errors. From the respondents' data, the d value is 1.97.
Since tables in statistical text books generally contain Durbin-Watson tables for k of 1 to
5 while this study employs 38 variables, we use the rule of thumb that if the value of the
Durbin–Watson test is between 1.5 and 2.5 (Hutcheson & Sofroniou, 1999) or close to
2 (McClave et al., 2018), it means there is no autocorrelation. As stated by McClave et
al. (2018), autocorrelation exists when the d value is less than 2 (positive correlation) or
greater than 2 (negative correlation), close to 0 (very strong positive correlation) or close
to 4 (very strong negative correlation). Thus, it can be concluded that this study shows
there is no autocorrelation since d statistic 2.5 > 1.97 > 1.5 and tends to close to 2
(Hutcheson & Sofroniou, 1999; McClave et al., 2018).
Multicollinearity test was used to determine whether or not there is a linear
relationship among variables in the regression model that influence each other. One of
the two methods usually used is variance inflation factor (VIF) or the tolerance value
(Hair et al., 2019) where FIV = (1 / tolerance). The results show that the highest VIF
value is 2.02 which is far from 10 as the limit, and the lowest tolerance value is 0.495
which is much greater than 0.1 of the lowest limit value that according to Hair et al. (2019)
are considered as common cutoff thresholds (see Table 2 in Appendix for the detail
results). A VIF value of more than 5 indicates a high multicollinearity, a VIF value of up
to 5 means moderate multicollinearity, and a VIF value of 1 means there is no
multicollinearity (Hair et al., 2019).
Heteroscedasticity test was conducted to determine whether there are symptoms
where the probability distribution of deviation is not the same for all observations, and to
determine the fulfillment of the homoscedasticity assumption (Hair et al., 2019). Using
Jamovi we conducted the Levene test as one the most common tests (Hair et al., 2019;
Navarro & Foxcroft, 2018). The results showed small p-values as predicted by Grace-
Martin (2021) and Navarro & Foxcroft (2018) for large samples as with this research.
Out of 38 variables, only 17 variables have p-values of 0.05 or above as the cutoff
82
threshold. Even when we conducted Welch one-way test as a remedy (Navarro &
Foxcroft, 2018), the results are quite the same, only 19 variables have p-values of 0.05
or above. Since it will be 50% of the variables that need to be excluded from the
calculation, we tend to neglect the Levene test for large samples as suggested by Grace-
Martin (2021).
To provide certainty that the methods used in this study will provide consistent
results, testing of the questionnaire was carried out using the expert panel method
(Schindler, 2019) that added and improved the questions in the questionnaire
(Widiatmanti, 2020). This study did not test the questionnaire because it is the extension
of the study of Widiatmanti (2020).
4.3. Exploratory Factor Analysis (EFA)

Based on validity and reliability tests discussed earlier, none of the variables were
deleted since all of them meet the criteria except for the low p-values in normality and
Levene tests that can be neglected for big samples such as in this research (Zaiontz,
2017; Grace-Martin (2021). As such, the EFA testing can be carried out. In addition, with
the large number of variables (38 variables) and big samples (1,587 respondents) in this
study, the minimum ratio of 20 respondents for every 1 variable (Hair et al., 2019) is also
met. Using Jamovi program, testing of these 38 variables was conducted using the
following steps:
1. Determining which variables significantly correlate with each other. In EFA, this kind
of correlation is extremely important, and only variables with significant correlation will
be processed in the next stage.
Bartlett test of sphericity and Kaiser-Meyer-Olkin test measure of sampling adequacy
(MSA) were performed, resulting in the overall MSA of 0.978 with the smallest MSA
of 0.940 and the largest 0.992 (see Table 3 of the Appendix). According to Hair et al.
(2019), an MSA of 0.8 or more is very good, 0.7 or more is good, 0.6 or more is
mediocre, 0.5 or more is bad, and less than 0.5 is unacceptable. Since there are no
variables with individual MSA below 0.5, no further variables were dropped for further
testing stage. As for Bartlett’s test for sphericity, the p-value is < .001, indicating that
in general this test is adequate.
2. Determining the value of communalities which indicate whether or not a variable is
dominant in a set of variables.
As pointed out by Hair et al. (2019), “[h]igh communality values indicate that a large
amount of the variance in a variable has been extracted by the factor solution. Small
communalities show that a substantial portion of the variable’s variance is not accounted for
by the factors.” The most reliable criteria according to Hair et al. (2019) for variables
between 20 to 50 is when the communality values are above 0.40.
Although Jamovi does not provide a feature to determine communality values, the program
calculates the uniqueness values which when subtracted from 1 will result in communalities
since uniqueness equals to “1 – communality” (Navarro & Foxcroft, 2019). Using the
criteria mentioned above, several variables that are included in the factor loading but
have a communalities value of 0.4 or less will be excluded from EFA. As a result,
variables B4 and B5 will be dropped for further testing since it has communality value
of 0.123 and 0.106, respectively (see the bold and italic numbers in Table 4 of the
Appendix).
3. Determining what factors are considered dominant from the series of variables under
study.
The dominance of these factors is related to the magnitude of the eigenvalue of each
variable, which in Jamovi program is set to be 1. The number of components formed
83
in the Jamovi program are sorted from those with the largest eigenvalue to the
smallest value, and the number of factors stopped at the eigenvalue of 1. From the
data being processed, this research acquires three factors (see Table 4 of the
Appendix) since the fourth factor has an eigenvalue value of 0.64650 which is lower
than 1.
4. Determining the amount of the loading factor for each variable, which indicates the
level of correspondence between the variable and the factor. The greater the loading,
the more representative the factor will be.
In Jamovi program, the loading factor is automatically set to 0.3. In this study, the
correlation is significant when the loading is 0.5 or more. It is therefore none of the
values in “Factors” columns in Table 4 of the Appendix are less than 0.5.
From Table 4 of the Appendix we can conclude that five variables are not significant
since their loading factors are below 0.5, namely B6 regarding guidance from
supervisors, C4 regarding performance appraisal information from supervisors, C6
regarding coaching from supervisors on aspects need to improve, D1 regarding
alignment between KPI and organizational goals, and D4 regarding employee
development. As for B4 and B5, these two variables were already be excluded since
their communality values are less than 0.40 as the threshold for variables between 20
to 50 (Hair et al., 2019).
Table 4 of the Appendix shows that if DG Tax employees were not considered male
or female, married or not, echelon officers or not, or whatever their ages and educations
are, there are three dominant factors affecting performance evaluation with 31 variables
that significantly dominant. The first factor is related to variables A1 to B2, D2, D5 and
D6. The eleven variables that relate to the performance appraisal in the A group are all
significant variables. Combine with the results described in section 4.1. Descriptive Data,
these eleven variables are considered effective and satisfactory with the average of 32.2
indicating quite effective, and the medians and modes of 3 respectively indicating
respondents generally Agree.
As for the performance appraisal model variables in group B, only three variables
are significant. Two variables, namely B1 (feedback from supervisor regarding
performance) and B2 (performance assessment using a rating scale), are in the first
factor and B3 in the third factor. As with variables B4 and B5, EFA method considers
these variables are insignificant since their communality values are below 0.4 (Hair et
al., 2019). It also means that although most respondents chose 3 (agree) and 4 (very
agree) for this B4 (see Table 1 of the Appendix), EFA method points out that
performance assessment does not have to be done using technical and managerial
ability assessment in order to be considered fair by employees or as indicated by
Sharma et al. (2016) and by Dewettinck & van Dijk (2012). DG Tax Regulation No.
12/PJ/2018 is not as bad as wasta in Harbi et al. (2017) research that create rejection
from employees. This is also true for the B5 variable, which means that performance
appraisal does not necessarily have to be done using the self-assessment method,
although most respondents considered so (see Table 1 of the Appendix).
On the other hand, since Minister of State Apparatus and Bureaucratic Reform has
enacted regulation No. 38 of 2017 regarding Competency Standards for State Civil
Apparatus both for technical and managerial skills, despite its insignificance according
to EFA method, DG Tax Regulation No. 12/PJ/2018 with regard to technical competency
evaluation needs to be changed so that performance evaluation would be better and
considered fair by employees if it is conducted based on, among other things, technical
and managerial competencies of every employee. This is also true for the variable B6
(providing guidance, i.e., coaching, mentoring and counseling on the results of employee
84
performance). Although EFA method considers this variable insignificant (see Table 1
of the Appendix), since coaching for performance is mandatory for every manager at
MoF as regulated in the ministerial decree No. 590/KMK.01/2016, then every manager
is still obliged to do so, especially for employees whose performance is not include in
the S and/or A categories. However, this would be very difficult for every manager if the
employee’s original performance is good but is forced to be excluded from category S,
A, B and/or C due to a forced ranked system. As a result, this forced ranked system
needs to be reconsidered. If this forced rank mechanism relates to allowance or annual
budget, then the revised category of SABCD could be mainly due to excellent (S) and
poor (D) performance, while the allowance depends on the availability of budget, similar
to other units within the MoF.
As for the effectiveness of the performance appraisal model represented by C
variables, only variables C4 (informing performance appraisal by supervisor) and C6
(coaching by supervisor for aspects need to improve) are considered insignificant.
Variable C4 was proposed by Widiatmanti (2020) in order to measure the fairness of DG
Tax PMS and therefore is not enacted yet in the DG Tax Regulation No. 12/PJ/2018.
Since according to EFA this variable is considered insignificant, we do not propose to
be included in the revision of DG Tax Regulation No. 12/PJ/2018. As for variable C6,
although insignificant according to EFA method, most manager have already coached
their subordinates as indicated in Appendix 1 who chose 3 (agree) and 4 (very agree)
with median and mode of 3, respectively (see Table 1 of the Appendix). In addition, as
regulated in the ministerial decree No. 590/KMK.01/2016, this is an obligation for every
manager in the MoF, especially for employees whose performance are not S nor A
categories.
Lastly, for the performance appraisal regulatory variables represented by D
variables, only variables D1 (aligning KPI with organizational goals) and D4 (aligning
development that employees get with their competency gap) are considered insignificant.
Variable D1 and D4 were proposed by Widiatmanti (2020) in accordance with Blackman
et al. (2018) who argue that PMS have to be operated as part of the organization’s core
business to be effective, and with the Government Regulation No. 17 of 2020 that link
the merit system with the competency development. As with variable C4, these two
variables are also not enacted yet in the DG Tax Regulation No. 12/PJ/2018. Since
according to EFA these variables are considered insignificant, we do not propose
variable D1 to be included in the revision of DG Tax Regulation No. 12/PJ/2018. Yet, in
order to comply with Government Regulation No. 17 of 2020, we propose that the
revision to DG Tax Regulation No. 12/PJ/2018 includes this D4 variable.
The second factor relates to variables A9 (performance appraisal with rating scale
according to the role in the organization is objective), C7 to C11 (performance status
SABCD, and Stage 1 and Stage 2 Rating Sessions), D3 (performance appraisal is able
to distinguish between competent and incompetent employees), and D5 to D10 (the
performance appraisal regulatory variables). Variable A9 can be included in either first
factor or the second one. However, in the first sector it has bigger loading factor, i.e.,
0.613 vs. 0.551. This means that correlation value of variable A9 with factor 1 is bigger
than that of variable A9 with factor 2, indicating that the sample variation in factor 1 can be
better explained by using variable A9 in the straight-line model than that in factor 2.
As described earlier, the effectiveness of the performance appraisal model
represented by C variables, all variables are considered significant except variables C4
(informing performance appraisal by supervisor) and C6 (coaching by supervisor for
aspects need to improve). While variable C4 can be neglected since it was proposed by
Widiatmanti (2020) and it is not enacted yet in the DG Tax Regulation No. 12/PJ/2018,
85
variable C6 relates to ministerial decree No. 590/KMK. 01/2016 that instructs every
manager in the MoF to coach his or her subordinates. As such, this variable still
important at DG Tax, especially for managers whose subordinates’ performances are
not S nor A categories.
The last factor relates to variable B3 (performance is assessed quarterly,
semiannually or annually), C1 to C3 and C5 (effectiveness of the performance appraisal
model). Other variables in group B are considered insignificant according to EFA method,
i.e., B4, B5, and B6 described earlier. Only variable B6 (providing guidance, i.e.,
coaching, mentoring and counseling on the results of employee performance) that
should be kept since it relates to ministerial decree No. 590/KMK. 01/2016 which every
manager must obey, especially whose employees performances are not S nor A
categories.
The exploratory factor analysis calculated above shows that statistically the DG Tax
Regulation No. 12/PJ/2018 is effective and satisfactory so that it does not need to be
changed or replaced. However, due to regulations No. 38 of 2017 of the Minister of State
Apparatus and Bureaucratic Reform that asks government units to cover both technical
and managerial skills, Government Regulation No. 17 of 2020 that link the merit system
with the competency development, as well as ministerial decree No. 590/KMK. 01/2016,
we suggest that DG Tax Regulation No. 12/PJ/2018 need to be adjusted. Employees
who are categorized as Cs and Ds in the SABCD performance status are obliged to
participate in the training program. This adjustment can also be made by issuing a
Circular Letter or other directive from the Director General of Tax that asks every
supervisor to assign his or her subordinates participate in the training program without
changing or replacing the original DG Tax Regulation No. 12/PJ/2018. However, since
employees who are categorized as Cs or Ds are not necessarily due to bad
performances but because of forced ranked, we suggest that forced rank in the DG Tax
Regulation No. 12/PJ/2018 is abolished. The revised category of SABCD could be
mainly due to excellent (S) and poor (D) performance, and the annual allowance
depends on the availability of budget, similar to other units within the MoF.
5. CONCLUSIONS AND RECOMMENDATIONS

This research shows that most DG Tax employees participated in this research
generally agree and strongly agree with DG Tax Regulation No. 12/PJ/2018 regarding
performance evaluation system. Although there are some employees who chose
disagree and very disagree in every question in the questionnaire, in general this
regulation is considered quite effective and quite satisfactory. However, the ANOVA
method used in this research reveals that there are differences among the different
categories of DG Tax employees indicating that its effectiveness and satisfaction are
different between men and women category, between married and unmarried
employees, among different educational levels, and among different ranks and positions.
Out of 38 questions (variables), only three variables that the majority of respondents
chose strongly disagree (1) and disagree (2). Two variables relate to performance status
of SABCD, and the other one relates to the level 1 and level 2 ranking sessions. Using
EFA method, these three variables are considered significant. This indicate that DG Tax
should reconsider the use of forced rank in the SABCD performance status since this
method although motivates employees to work better than the others (variable C7), the
majority of respondents disagree and very disagree that this this method motivates them
improving group performance (variable D8). Therefore, we suggest that the use of forced
rank in the SABCD performance status is abolished and be replaced mainly due to
excellent (S) and poor (D) performance. If the forced rank mechanism is due to the
86
limited annual budget, then we suggest that the provision of additional performance
allowances is adjusted to the availability of the budget, similar to other units within the
MoF.
The EFA method also shows that seven variables are insignificant in the DG Tax
performance evaluation system according to DG Tax Regulation No. 12/PJ/ 2018,
namely variables B4 (performance assessment must be carried out using technical and
managerial ability assessment), B5 (performance appraisal is done using self-appraisal
method), B6 (supervisor always provides guidance (coaching, mentoring and counseling)
on the results of my performance), C4 (the results of the performance appraisal are
always informed by superior), C6 (superior coaches and tells what aspects need to
improve), D1 (performance appraisal at DGT is aligned between the KPI and the
organization's goals) and D4 (the development that employee gets from the institution
is always in accordance with the competency gap). This indicates that the DG Tax
Regulation No. 12/PJ/2018 does not need to be changed, especially with regards to the
above variables. However, there are three regulations which according to the authors
make the DG Tax Regulation No. 12/PJ/2018 needs to change, namely regulations No.
38 of 2017 of the Minister of State Apparatus and Bureaucratic Reform that asks
government units to cover both technical and managerial skills, Government Regulation
No. 17 of 2020 that link the merit system with the competency development, and the
ministerial decree No. 590/KMK. 01/2016 that asks every manager in the MoF to coach
his or her subordinates with regard to performance. As such, we suggest that DG Tax
Regulation No. 12/PJ/2018 be amended or modified to suit the three regulations as well
as to abolish the forced rank mechanism.
6. LIMITATION OF THIS RESEARCH

This research has several limitations. Although the number of samples is considered
big and above the minimum samples according to Slovin formula for 5% tolerable error,
the proportion is relatively low compared to the total DG Tax employees of 45,948 (Biro
SDM, 2020). Had the sample been more than 1,587, the results might have been
different. In addition, this research cut the category of employees with regard to tenure
and age since there are some respondents who filled the questionnaire suspiciously.
Had these respondents were singled out instead the two categories, the results might
have been different.
BIBLIOGRAPHY
Biro Sumber Daya Manusia Kementerian Keuangan.
(2020). Data Pegawai per 1 Juni 2020.
www.sdm.kemenkeu.go.id. Accessed on June 1, 2020.
Blackman, D., Buick, F. & O’Donnell, M. (2018) “Why Performance Management Should
Not be Like Dieting.” Australian Journal of Public Administration, Vol. 76, pp. 524–
28.
Dessler, G. (2020). Human Resource Management. Pearson Education, Inc.
Dewettinck, K. & van Dijk, H. (2012) Linking Belgian Employee Performance
Management System Characteristics with Performance Management System
Effectiveness: Exploring the Mediating Role of Fairness. International Journal of
Human Resource Management, Vol. 24, pp. 806–25.
87
Evita, S. N., Muizu, M. O. Z. & Atmodjo, R. T.W. (2017).
Penilaian Kinerja Karyawan dengan Menggunakan
Metode Behaviorally Anchor Rating Scale dan
Management By Objectives (Studi Kasus Pada PT.
Qwords Company International). Pekbis Jurnal, Vol.9,
No.1, Maret 2017, pp. 18-32.
Grace-Martin, K., 2021. The Problem with Using Tests
for Statistical Assumptions. May 12, 2021.
https://www.theanalysisfactor.com/the-problem-with-
tests-for-statistical-assum-ptions/. Accessed on
October 20, 2021.
Hair, J. F. J., Black, W. C., Babin, B. J., Anderson, R. E.,
Black, W. C., & Anderson, R. E. (2019). Multivariate
Data Analysis. Cencage Learning.
Harbi, S. A., Thursfield, D., & Bright, D. (2017). Culture,
Wasta and Perceptions of Performance Appraisal in
Saudi Arabia. The International Journal of Human
Resource Management, 28(19), pp. 2792–2810.
Hutcheson, G.D. & Sofroniou, N., 1999. The Multivariate
Social Scientist: Introductory Statistics Using
Generalized Linear Models. Sage Publications Ltd.
Law of the Republic of Indonesia Number 5 of 2014 concerning State Civil Apparatus.
Mathis, R. L., Jackson, J. H., Valentine, S. R., & Meglich, P. (2017). Human Resource
Management. Cengage Learning.
McClave, J.T., Benson, P.G. and Sincich, T., 2018. Statis-tics for Business and
Economics, 13th ed. USA: Pear-son Education, Inc.
Navarro, D.J. & Foxcroft, D.R., 2019. Learning Statis-tics with Jamovi: A Tutorial for
Psychology Students and Other Beginners. https://www.learnstatswith-jamovi.com/.
Downloaded April 1, 2021.
Regulation of the Government of the Republic of Indonesia Number 11 of 2017
concerning Management of the State Civil Apparatus.
Regulation of the Government of the Republic of Indonesia Number 17 of 2020
concerning Replacement of the Regulation of the Government of the Republic of
Indonesia Number 11 of 2017 concerning Management of the State Civil Apparatus.
Regulation of the Minister of State Apparatus and Bureaucratic Reform Number 38 of
2017 concerning Competency Standards for State Civil Apparatus Positions.
Schindler, P. S. (2019). Business Research Methods. McGraw-Hill Education.
Sharma, N. P., Sharma, T., & Agarwal, M. N. (2016). Measuring employee perception
of performance management system effectiveness. Employee Relations, Vol. 38,
pp. 224–47.
Torrington, D., Hall, L., Taylor, S., & Atkinson, C. (2020). Human Resource Management.
88
Pearson.
Widiatmanti, H. (2020). Model Sistem Penilaian Kinerja Pegawai DJP. Prosiding Kajian
Akademis Pusdiklat Pengembangan Sumber Daya Manusia Kementerian Keuangan,
p. 89.
Zaiontz, C., 2017. Multivariate Central Limit Theorem. https://www.real-
statistics.com/multivariate-statistics/multivariate-normal-distribution/ multivariate-
central-limit-theorem/. Accessed on April 22, 2021.
89
APPENDIXES
Table 1 Questionnaire Results and their Medians and Modes
Vari- Respondents Choices
ables Statements 1 2 3 4 Medians Modes
A1 My performance has been appraised based on the work I have done 85 209 768 525 3 3
A2 I am satisfied with the results of my performance appraisal 84 220 742 541 3 3
A3 Evaluation of employee performance has shown the quantity of work/activities objectively 116 318 745 408 3 3
A4 Evaluation of employee performance has shown the quality of work/activities objectively 101 330 748 408 3 3
Performance appraisal is carried out according to the work plan that has been agreed with
A5 67 196 773 551 3 3
superior
A6 Employee performance appraisal has been used as a basis for competency development 124 339 702 422 3 3
A7 Employee performance appraisal has been used as a basis for career development 119 352 697 419 3 3
Performance appraisal has become an objective performance criterion in accordance with
A8 131 391 682 383 3 3
organizational goals
A9 Performance appraisal with rating scale according to the role in the organization is objective 223 410 618 336 3 3
Performance appraisal is carried out by comparing the work results with minimum work
A10 110 371 732 374 3 3
standards using clear parameters
My supervisor has transparently provided information on the results of the performance
A11 149 309 687 442 3 3
appraisal
B1 Performance is assessed based on the work plan, and is always given feedback by superior 65 257 824 441 3 3
Currently, performance is assessed using a rating scale from the lowest to the highest (scoring
B2 41 134 780 632 3 3
scale 0 to 100)
B3 Performance is assessed within a certain period of time (quarterly, semiannually or annually) 16 50 666 855 4 4
B4 Performance assessment must be carried out using technical and managerial ability assessment 93 271 694 529 3 3
B5 I will be happy if the performance appraisal is done by self-appraisal. 92 329 731 435 3 3
My supervisor always provides guidance (coaching, mentoring and counseling) on the results
B6 57 273 752 505 3 3
of my performance
My performance planning set out in the KPI is in line with my main role in achieving
C1 31 186 910 460 3 3
organizational goals
I can easily get information about the results of the KPI achievements through the e-
C2 47 163 780 597 3 3
performance application
C3 The SIKKA application for inputting Work Targets is very easy and practical 25 154 807 601 3 3
C4 The results of the performance appraisal are always informed by superior 119 378 701 389 3 3
C5 Performance appraisal guidelines and parameters are clear and understandable 91 347 754 395 3 3
C6 My superior coaches and tells what aspects need to improve 85 343 740 419 3 3
C7 The categorization of performance status S, A, B, C, D motivates me to work better than others 0 368 537 332 3 3
C8 Using normal curve for ranking employee performance status S, A, B, C, D is correct 450 477 445 215 2 2
The Stage 1 Rating Session by the Chairperson and the Raters held in the office always goes well
C9 189 487 667 244 3 3
and does not cause turmoil
The Stage 2 rating session by the Chairperson and the Raters held in echelon 2 units always
C10 171 486 698 232 3 3
goes well and does not cause turmoil
C11 The results of the Stage 1 and Stage 2 Rating Sessions have been publicly informed to employees 346 539 499 203 2 2
D1 Performance appraisal at DGT is aligned between the KPI and the organization's goals 66 250 820 451 3 3
D2 The performance appraisal results improve knowledge and skills to be even better at work 75 236 730 546 3 3
Performance appraisal is able to distinguish between competent and incompetent employees
D3 216 509 639 223 3 3
at work
So far, the development that I get from the institution is always in accordance with the
D4 106 444 785 252 3 3
competency gap
D5 Performance appraisal has objective criteria and shows high consistency in its assessment 134 444 726 283 3 3
D6 Performance appraisal system results are accurate and reliable 177 514 650 246 3 3
D7 Performance appraisal are accepted by all employees (superiors and subordinates) 152 462 712 261 3 3
Employee performance status (performance categories S, A, B, C, D) motivates me to improve
D8 437 407 477 266 2 3
group performance
D9 Performance appraisal is easy to understand, simple and uncomplicated in its application 197 457 663 270 3 3
D10 Performance appraisal is transparent and informative for employees 226 484 618 259 3 3
90
Table 2 VIF and Tolerance Values
Source: Survey Results.
Variables VIF Tolerance Variables VIF Tolerance Variables VIF Tolerance Variables VIF Tolerance
A1 1.90 0.526 A11 1.45 0.690 C4 1.45 0.690 D2 1.44 0.694
A2 1.84 0.543 B1 1.57 0.637 C5 1.49 0.671 D3 1.50 0.667
A3 1.74 0.575 B2 1.44 0.694 C6 1.51 0.662 D4 1.48 0.676
A4 1.80 0.556 B3 1.36 0.735 C7 1.52 0.658 D5 1.72 0.581
A5 1.56 0.641 B4 1.14 0.877 C8 1.54 0.649 D6 1.81 0.552
A6 1.62 0.617 B5 1.13 0.885 C9 2.00 0.500 D7 1.59 0.629
A7 1.59 0.629 B6 1.38 0.725 C10 2.02 0.495 D8 1.49 0.671
A8 1.68 0.595 C1 1.37 0.730 C11 1.46 0.685 D9 1.54 0.649
A9 1.54 0.649 C2 1.35 0.741 D1 1.42 0.704 D10 1.60 0.625
A10 1.57 0.637 C3 1.32 0.758
Table 3 MSA Test Results

Variables MSA Variables MSA Variables MSA Variables MSA Variables MSA
A1 0.968 A9 0.990 B6 0.968 C8 0.969 D5 0.986
A2 0.967 A10 0.991 C1 0.979 C9 0.944 D6 0.983
A3 0.983 A11 0.983 C2 0.962 C10 0.940 D7 0.992
A4 0.981 B1 0.987 C3 0.966 C11 0.983 D8 0.969
A5 0.989 B2 0.974 C4 0.975 D1 0.986 D9 0.980
A6 0.984 B3 0.960 C5 0.983 D2 0.989 D10 0.980
A7 0.977 B4 0.955 C6 0.969 D3 0.987
A8 0.986 B5 0.951 C7 0.960 D4 0.987
1
Table 4 Component Matrix of Exploratory Factor Analysis
Factors Factors
Variables Communalities Variables Communalities
1 2 3 1 2 3
A1 0.794 0.773 C3 0.671 0.508
A2 0.767 0.725 C4 0.533
A3 0.745 0.755 C5 0.574 0.656
A4 0.770 0.772 C6 0.554
A5 0.709 0.689 C7 0.679 0.568
A6 0.687 0.693 C8 0.768 0.675
A7 0.618 0.624 C9 0.679 0.625
A8 0.677 0.738 C10 0.687 0.630
A9 0.613 0.551 0.706 C11 0.683 0.603
A10 0.658 0.699 D1 0.569
A11 0.562 0.587 D2 0.538 0.587
B1 0.610 0.662 D3 0.570 0.624
B2 0.549 0.518 D4 0.579
B3 0.501 0.422 D5 0.544 0.560 0.719
B4 0.123 D6 0.539 0.60 0.751
B5 0.106 D7 0.594 0.698
B6 0.454 D8 0.728 0.627
C1 0.615 0.534 D9 0.579 0.619
C2 0.685 0.533 D10 0.619 0.678
Notes: Extraction Method: Principal Axis Factoring; Rotation Method: Varimax
2
THE EFFECT OF THE QUALITY OF TAX
DETERMINATION, WARNING LETTER, AND DISTRESS
WARRANT ON TAX ARREARS DISBURSEMENT
Rizki Puput Fathonaha, Nuramalia Hasanahb, Petrolis Nusa Perdanac

a Facultyof Economics Universitas Negeri Jakarta, Indonesia. Email: rizkipuputfathonah@gmail.com
b Faculty of Economics Universitas Negeri Jakarta, Indonesia. Email: nuramalia@unj.ac.id
c Faculty of Economics Universitas Negeri Jakarta, Indonesia. Email: petrolis98@unj.ac.id
ABSTRACT
The purpose of this study is to find out the effect of the quality of tax determination, warning letter,
and distress warrant on tax arrears disbursement at Kalideres Tax Office. This study uses secondary
data. The sample of this study is monthly time series data consisting of reports of tax arrears
disbursement, reports of request for reduced sanctions, warning letter, and Distress Warrant which
had been issued by Kalideres Tax Office in 2017-2020. The sample selection technique is saturated
sampling with total 48 data sample. This study uses multiple linear regression analysis and
processed using SPSS program. The result of the test have proven that the quality of tax
determination has significant effect on tax arrears disbursement. Meanwhile warning letter and
distress warrant has no effect on tax arrears disbursement.
Keywords: The Quality of Tax Determination, Warning Letter, Distress Warrant, Tax Arrears Disbursement
Latar belakang penelitian ini adalah jumlah tunggakan pajak yang selalu meningkat karena
kurangnya kesadaran Wajib Pajak untuk membayar utang pajak. Sehingga petugas pajak berupaya
untuk mencairkan tunggakan pajak. Penelitian ini bertujuan untuk menguji pengaruh kualitas
penetapan pajak, surat teguran, dan surat paksa terhadap pencairan tunggakan pajak di KPP
Pratama Jakarta Kalideres tahun 2017-2020. Penelitian ini menggunakan data sekunder. Sampel
dalam penelitian ini adalah data bulanan yang terdiri dari laporan pencairan tunggakan pajak, laporan
permohonan pengurangan sanksi/keberatan/banding, laporan surat teguran, dan laporan surat
paksa pada KPP Pratama Jakarta Kalideres tahun 2017-2020. Teknik pengambilan sampel adalah
samping jenuh yang berjumlah 48 data. Penelitian ini menggunakan metode analisis regresi linear
berganda dan diolah menggunakan program SPSS. Hasil penelitian menunjukkan bahwa kualitas
penetapan pajak berpengaruh terhadap pencairan tunggakan pajak. Sedangkan surat teguran dan
surat paksa tidak berpengaruh terhadap pencairan tunggakan pajak.
Kata kunci: Kualitas Penetapan Pajak, Surat Teguran, Surat Paksa, Pencairan Tunggakan Pajak
1. INTRODUCTION
1.1. Research Background
The government always tries to create prosperity for the people through national
development in all areas such as the economics, public health, education, infrastructure,
technology, and so on. National development requires a huge cost. The government
uses eight sources of revenue to funding national development such as taxes, natural
resources, customs and excise, levies, donations, profits from State-Owned Enterprises,
and other sources (Purwanto, 2017:94).
Among the eight sources of revenue, taxes have a very important role because taxes
have the largest amount of revenue. According to the Central Bureau of Statistics (BPS),
the amount of tax revenue in 2017 was Rp 1,343,529.80 billion, the amount of tax
revenue in 2018 was Rp 1,518,789.80 billion, the amount of tax revenue in 2019 was Rp
1
1,546,141.90 billion, and the amount of tax revenue in 2020 was Rp 1,404. 507.50 billion.
The data shows that tax revenue increased in 2017-2019, then decreased in 2020. Even
though it has decreased, the amount of tax revenue is still greater than non-tax revenue.
This analysis reflect that taxes has a great impact to state revenues.
The government tries to increase tax revenue by the implementation of Self
Assessment System. This system trust the taxpayers to determining their taxes
independently (Rochmawati, 2015). Through the implementation of this system, the
government invites the public to participate in tax collection. So the awareness of tax
payers can be increase. But in reality, there are still many violations by taxpayers like
avoiding paying taxes, late for paying taxes, and pay underpaid taxes.
There is a phenomenon of the increase of tax arrears every year at Kalideres Tax
Office as can be seen from the data below:
Table 1 Tax Arrears at Kalideres Tax Office 2017-2020

Source: Kalideres Tax Office (2022)
Year Tax Arrears (Rp) Increase/Decrease (%)
2017 411.701.687.599 -
2018 590.003.933.132 43%
2019 758.075.410.740 28%
2020 1.375.923.454.434 82%
From the table above, the tax arrears at Kalideres Tax Office in 2017-2020 period
always increases. Tax arrears in 2017 reached Rp 411,701,687,599. Then in 2018 the
tax arrears increased 43% to Rp 590,003,933,132. In 2019, tax arrears increased 28%
to Rp 758,075,410,740. In 2020, the amount of tax arrears increased 82% to Rp
1,375,923,454,434. The data shows that the taxpayers has a low compliance for paying
their tax. The problem of tax arrears must be resolved by the government, especially
Directorate General of Taxes because it will affect the state income. Tax authorities must
take firm action against uncooperative taxpayers as an effort to optimize the tax arrears
disbursement.
Tax determination can affect the effectiveness of tax arrears disbursement
(Angraeny, 2017:4754). The tax audit process will be issuing Tax Determination Letter if
there is an error when the taxpayer filling out the Notice Letter (SPT) or the Taxpayer is
not reported a fiscal data to Tax Office. The quality of tax determination is considered
have a good quality if the amount of tax determination does not change, either increasing
or decreasing, even though the taxpayers request a reduction of their tax sanctions.
Tax collection is an action by the tax authorities to forced taxpayers for pay off their
tax obligation. The first step in tax collection is the issuance of Warning Letter to remind
the Taxpayer to fulfill their tax obligation. Warning Letter is issued 7 days after the due
date and sent using a courier service to the Taxpayer’s address.
Then if the tax obligation has not been paid after the issuance of Warning Letter, a
Distress Warrant will be issued to force the Taxpayer to pay off their the tax obligation
and tax collection fees. Distress Warrant is issued 21 days after the issuance of Warning
Letter. Bailiff will be come to taxpayers address and giving the Distress Warrant directly
to taxpayers. If the tax obligation has not been paid off within 2 x 24 hours, a Confiscation
Letter will be issued. The bailiff will seize the assets and sell the confiscated assets
without having to wait for a court decision.
2
With the phenomenon of the increase of tax arrears, the author is interested to
choose tax arrears disbursement as a theme of this research. This research used
Kalideres Tax Office as an object with 2017-2020 as observation period, so this study
was entitled "The Effect of the Quality of Tax Determination, Warning Letter, and Distress
Warrant on Tax Arrears Disbursement (Case Study at Kalideres Tax Office in 2017-
2020)”.

DEVELOPMENT
2.1. Theory Review
2.1.1. Theory of Absolute Obligation
According to Absolute Obligation Theory, the state is an organization consists of a
group of individuals (society) who live together and settle in an area. The society trust
the state to carry out the public interest to create prosperity for the people. To realize the
prosperity in society, the government can make decisions in various fields including
taxation.
According to this theory, the state has the right to collect taxes from its people, and
the people has to pay taxes as a sign of their devotion to the state (Mardiasmo, 2018:6).
Tax will be used to fund state expenditures to increase people's prosperity. In order for
the government to carry out its duties in carrying out the public interest, the people must
be aware to pay taxes voluntarily.
The government choose the Directorate General of Taxes to organize tax collection
in Indonesia. In collecting taxes, tax officer must determine a good quality of tax. The
quality of tax determination is considered good if the tax determination does not change
although the taxpayer request to reduce sanctions or submits an appeal. To produce a
good quality of tax determination, the tax officer must be full of consideration in tax audit
and append an accurate data so a consistent tax decision can be issued.
When the taxpayer violates tax regulation and creates tax arrears, the government
has the right to do tax collection. Tax collection begins with issuing Warning Letter for
giving a warning to Taxpayer. The issuance of Warning Letter is expected to make tax
obligation can be paid off immediately. Then if the Taxpayer still not paying their tax
obligation after the issuance of Warning Letter, a Distress Warrant will be issued to forced
Taxpayer to pay their tax obligation
2.1.2. Tax Arrears Disbursement

According to Sa'diyah (2020:7), tax arrears is a taxes that have not been paid to the
state after the due date. Nainggolan (2015:9) said the definition of tax arrears
disbursement is all forms of disbursement related to tax arrears consists of payments,
write-offs, transfers, or objections. Waluyo (2013:64) said tax arrears disbursement
occurs if the taxpayer has paid off their tax arrears, the amount of tax receivables is
reduced due to an appeal that is granted, and the write-off of tax receivables. Meanwhile,
Angraeny (2017) said the definition of tax arrears disbursement can have two meanings.
First, tax arrears disbursement because of settlement. The settlement can be made in
cash, through book-entry or by selling the assets that have been confiscated. Second,
tax arrears disbursement because of write-offs. The tax arrears are written off because
the taxpayer has bankrupt and unable to pay their tax obligation.
3
2.1.3. The Quality of Tax Determination
Based on Law No 16 in 2009, Tax Determination Letter is not issued for all Notice
Letter, but only issued if Taxpayer makes a mistake when filling out the Notice Letter or
not reported complete fiscal data to Tax Office. A tax determination letter is a written
determination consist of the amount of tax obligation for a certain type and year issued
by tax authorities for the taxpayer whose name is listed on Tax Determination Letter
(Waluyo, 2013:52). Tax determination letter is part of passive tax collection as a result
of tax audit with the function to notifying or correcting the amount of tax obligation,
carrying out billing, imposing sanctions, and returning surplus taxes (Suandy, 2016:169).
Tax authorities must be issued a good quality of tax determination. According to
Hidayat and Cheisviyanny (2013), tax determination is considered good if the amount of
the tax determination does not change, whether it increases or decreases, even though
the taxpayer request to reduce or eliminate sanctions, filed objections and appealed to
the tax authorities. This indicates that the tax audit has been preceded by careful
consideration with accurate data. If the amount of tax is change, then the tax
determination is declared to have poor quality because it is considered inconsistent.
2.1.4. Warning Letter

PMK No.189/PMK.03/2020 Article 1 Paragraph (14) said Warning Letter or other
similar letters are issued by government to warn Taxpayers to pay off their tax
obligations. Meanwhile, according to Article 3 paragraph (2), the tax authorities issued
Warning Letter if Taxpayer does not pay off their tax obligation after 7 days have passed
since the due date of payment.
Warning Letter is the first step in active billing to make the Taxpayer pay their tax
obligation according to Notice of Tax Collection (STP), Notice of Tax Underpayment
Assessment (SKPKB), and the Notice of Additional Tax Underpayment Assessment
(SKBKBT). Warning Letter is issued if Taxpayer does not pay off their tax obligation
within 7 days after due date of payment. Warning Letter is still persuasive because only
have to remind the Taxpayer and has not been charged a billing fee.
Warning Letters are delivered directly to Taxpayer’s address or sent using shipping
services (Pohan, 2017:233). Warning Letter is not issued if the government approved
the submition of postponement in paying taxes and the officer has carried out the billing
immediately and at once.
2.1.5. Distress Warrant

PMK No.189/PMK.03/2020 Article 1 Paragraph (15) said Distress Warrant is an
order to pay tax obligation and tax collection fees. Widiasfita (2018:44) said Distress
Warrant has the same legal force as a Judge's decision in civil cases which cannot be
appealed again to the superior Judge, has definite legal force, has a dual function to
collect taxes (central and regional taxes) and collect non-tax (collection fees).
The issuance of Distress Warrant is the second step in active billing which is issued
when the Taxpayer does not pay their tax obligation within 21 days after the issuance of
Warning Letter. Distress Warrant is tougher than Warning Letter because it is coercive.
Distress Warrant has executorial power and definite legal force so the active billing
process can be carried out without having to wait for a court decision. The payment of
tax obligation and tax collection fees must be made within 2x24 hours so the tax
collection does not proceed to the issuance of a Confiscation Letter. Mardiasmo (2018)
said the issuance of Distress Warrant is carried out if the tax obligation has not been paid
after the issuance of Warning Letter, the tax officer has made a billing immediately and
at the same time, and the Taxpayer does not fulfill the requirement to postponement of
tax payment.
4
According to PMK No.189/PMK.03/2020 Article 13 Paragraph (1), Tax Bailiff send
and read the Distress Warrant directly to Taxpayer and giving a copy of Distress Warrant
to taxpayer. According to Widiasfita (2018), the procedure to inform Distress Warrant
includes:
1. The tax authorities is giving distress warrant to tax bailiff.
2. The bailiff visits the taxpayer’s address or location of taxpayer's business activities
and then reads the distress warrant in front of taxpayer.
3. The bailiffs and taxpayers sign the Notification Report of Announcement of Distress
Warrant (BAPSP) as a proof that distress warrant has been notified to taxpayer.
4. The bailiff giving a copy of distress warrant to the Taxpayer.
5. The bailiff prepares a report about the notifying of distress warrant when the tax
collection has been completed.
2.2. Hypothesis Development

2.2.1. The Effect of the Quality of Tax Determination on Tax Arrears Disbursement
Tax determination letter is a written determination containing the amount of tax of a
certain type and year issued by the tax authorities for the taxpayer whose name is stated
in the letter (Waluyo, 2013:52). Tax determination is considered to have a good quality
if the amount of tax determination has not changed, either increased or decreased, even
though the taxpayer has request a reduction or eliminate sanctions, filed an objection
and appealed to the authorities (Hidayat & Cheisviyanny, 2013). If the tax determination
have a good quality, tax arrears disbursement will be increase. And if the tax
determination have a poor quality, the tax arrears disbursement will be decrease. If the
amount of tax determination is changes, then the tax determination is declared to have
poor quality because it is inconsistent.
Research by Redyanza (2018) and Sa’diyah (2020) found the quality of tax
determination has a positive effect on tax arrears disbursement. Meanwhile Supriadi
(2020) found that the quality of tax determination has a negative effect on tax arrears
disbursement. From this description, the author is giving a hypothesis:
H1: There is an effect of the Quality of Tax Determination on Tax Arrears Disbursement
2.2.2. The Effect of Warning Letter on Tax Arrears Disbursement

Warning Letter is the first step in active tax collection to warning the Taxpayer to pay
their tax obligation according to Notice of Tax Collection (STP), Notice of Tax
Underpayment Assessment (SKPKB), and the Notice of Additional Tax Underpayment
Assessment (SKBKBT). Warning Letter is issued 7 days after the due date of payment
(Fatmandika et al., 2016).
If Warning Letter can make the Taxpayer aware about their tax obligation and the
Taxpayer immediately pay their tax, then tax arrears disbursement will be increases.
Meanwhile, if Warning Letter cannot make the Taxpayer aware about their tax obligation
so that tax obligation is still not paid by Taxpayer, then tax arrears disbursement will be
decrease.
Research by Angraeny (2017) revealed Warning Letter affect tax arrears
disbursement. But this result is contrast to Silooy (2017), Maisyaroh et al. (2019), and
Purwanto (2017) who found Warning Letter had no significant effect on tax arrears
disbursement. From this description, the author is giving a hypothesis:
H2: There is an effect of Warning Letter on Tax Arrears Disbursement
2.2.3. The Effect of Distress Warrant on Tax Arrears Disbursement

Based on PMK No.189/PMK.03/2020 Article 1 Paragraph (15) reveals Distress
Warrant is an order for Taxpayer to pay their tax obligation and tax collection fees.
Distress Warrant issued when tax payable is not paid within 21 days after the issuance
5
of Warning Letter. If the tax obligation still not paid after the issuance of distress warrant,
the confiscation letter would be issued.
If Distress Warrant can make the Taxpayer aware about their tax obligations and the
Taxpayer immediately pay their tax, then tax arrears disbursement will be increases.
Meanwhile, if Distress Warrant cannot make the Taxpayer aware about their tax
obligations so that tax obligation is still not paid by Taxpayer, then tax arrears
disbursement will be decrease.
Research by Angraeny (2017) found Distress Warrant affect Tax Arrears
Disbursement. But this result is contrast to Silooy (2017), Maisyaroh et al. (2019), and
Purwanto (2017) who found Distress Warrant had no effect on tax arrears disbursement.
From this description, the author is giving a hypothesis:
H3: There is an effect of Distress Warrant on the Tax Arrears Disbursement
Based on the explanation, the theoretical framework is following below:
Figure 1 Theoretical Framework

Source: Data processed by the author (2022)
The unit of analysis in this research is Kalideres Tax Office. The population in this
research are monthly data consist report of tax arrears disbursement, reports of requests
for reduced sanctions/objections/appeals, reports of warning letter, and reports of
distress warrant at Kalideres Tax Office in 2017-2020. The sampling technique is non-
probability sampling spesifically saturated sampling. So all members of population are
used as research samples. The amount of sample in this research is 48 samples (4 years
x 12 months). The research method is quantitative method and analyzed using multiple
linear regression analysis and processed using SPSS program.
3.1. Tax Arrears Disbursement

According to Nainggolan (2015:9), tax arrears disbursement is all forms of
disbursement related to tax arrears that paid to the state treasury which can be a
payments, write-offs, transfers, or objections. Tax arrears disbursement can be seen
from the amount of tax payable after the due of payment based on Notice of Tax
Collection (STP), Notice of Tax Underpayment Assessment (SKPKB), and the Notice of
Additional Tax Underpayment Assessment (SKBKBT). The percentage of tax arrears
disbursement is measured by comparising the amount of tax arrears disbursement and
the amount of tax arrears (Widiasfita, 2018:51). The data of this research is tax arrears
disbursement data from Kalideres Tax Office in 2017-2020 period. Tax arrears
disbursement formula is following below:
Tax Arrears Disbursement
Tax Arrears Disbursement = Tax Arrears
x 100%
6
3.2. The Quality of Tax Determination
The quality of tax determination is considered good if the amount of tax
determination does not change although the taxpayer request a reduction of their tax
sanctions, request an objections and request an appeals to tax authorities (Hidayat &
Cheisviyanny, 2013:3). The quality of tax determination is calculated, processed and
grouped manually using a categorization table because the tax authorities doesn’t have
the quality of tax determination data. The percentage of the quality of tax determination
is calculated by comparing the amount of tax that must be paid after taxpayer request a
reduction of sanctions, objections, or appeal (Supriadi & Hidyatullah, 2020:76). The data
of this research is the list of requests for reduction sanctions, objections, and appeals
from Kalideres Tax Office in 2017-2020. The quality of tax determination formula is
following below:
(Proposed tax value − Accepted tax value)

The Quality of Tax Determination = x 100%
The amount of proposed tax value
After obtained the presentage using the formula above, then the data is grouped by
categorization to determine the quality of tax determination which is presented in the
following table:
Table 2 Categorization of The Quality of Tax Determination

Source: Hidayat dan Cheisviyanny (2013)
No. Percentage The Quality of Tax Determination
1. 00,00 – 10,99 Very Bad
2. 20,00 – 30,99 Not Good
3. 40,00 – 50,99 Less Good
4. 60,00 – 70,99 Good
5. 80,00 – 100 Very Good
3.3. Warning Letter

PMK No.189/PMK.03/2020 Article 1 Paragraph (14) said Warning Letter or other
similar letters issued by Tax Authorities to warn Taxpayers to pay off their tax obligations.
The percentage of Warning Letter is calculated by comparing the realization and target
of warning letters payment in rupiah (Widiasfita, 2018:52). The data of this research is
the list of Warning Letters from Kalideres Tax Office in 2017-2020. The formula for
calculating Warning Letter is following below:
Realization of Warning Letter Payment

Warning Letter = Target of Warning Letter Payment
x 100%
3.4. Distress Warrant

PMK No.189/PMK.03/2020 Article 1 Paragraph (15) said Distress Warrant is an order
for Taxpayer to pay their tax obligation and tax collection fees. The percentage of
Distress Warrant is calculated by comparing the realization and target of Distress
Warrants payment in rupiah (Widiasfita, 2018:53). The data of this research is the list of
Distress Warrants from Kalideres Tax Office in 2017-2020. The formula for calculating
Distress Warrant is following below:
Realization of Distress Warrant Payment

Distress Warrant = Target of Distress Warrant Payment
x 100%
7
4. RESULT AND ANALYSIS
4.1. Research Result
4.1.1. Descriptive Statistics Test
Descriptive statistic test is used to describe the variables by calculating the
maximum, minimum, average and standard deviation values.
Table 3 Descriptive Statistic Test

Source: SPSS 25, Data processed by the author (2022)
N Minimum Maximum Mean Std. Deviation
PTP 48 ,00249 ,13887 ,0319653 ,02843900
KPP 48 ,00000 1,00000 ,7754645 ,33006001
ST 48 ,00000 9,09738 1,0569908 1,93140644
SP 48 ,00000 61,88293 1,9628525 8,89296684
Valid N (listwise) 48
Tax Arrears Disbursement (Y) has the lowest value 0.00249 in May 2017. The
highest value of Tax Arrears Disbursement is 0.13887 in May 2018. The average value
is 0.0319653 and the standard deviation is 0.028439.
The Quality of Tax Disbursement (X1) has the lowest value 0.00 in April 2017, May
2017, August 2017, May 2020, and September 2020. The Quality of Tax Determination
has the highest value 1.00 for 10 months in January 2017 , March 2017, July 2017,
September 2017, June 2018, October 2018, May 2019, June 2019, March 2020 and April
2020. The average value is 0.7754645 and the standard deviation is 0.33006001.
Warning Letter (X2) has the lowest value 0.00 in June 2018. The highest value of
Warning Letter is 9.09738 in December 2018. The average value is 1.0569908 and the
standard deviation is 1.93140644.
Distress Warrant (X3) has the lowest value 0.00 in January 2017, June 2018,
December 2018, March 2020, April 2020, May 2020 and December 2020. The highest
value of Distress Warrant is 61.8829 in December 2019 .The average value is 1.9628525
and the standard deviation is 8.89296684.
4.1.2. Normality Test

The normality test aims to determine if the data is distributed normally or not normal.
To detect the normality, the author used the Kolmogorov-Smirnov test.
Table 4 Kolmogorov-Smirnov Test Before Data Transformation

PTP KPP ST SP
N 48 48 48 48
Normal Parametersa,b Mean ,0320 ,7755 1,0570 1,9629
8
Std. Deviation ,02844 ,33006 1,93141 8,89297
Most Extreme Differences Absolute ,150 ,255 ,318 ,413
Positive ,140 ,248 ,318 ,405
Negative -,150 -,255 -,292 -,413
Test Statistic ,150 ,255 ,318 ,413
Asymp. Sig. (2-tailed) ,009c ,000c ,000c ,000c
Based on table 4, the Tax Arrears Disbursement, The Quality of Tax Determination,
Warning Letter, and Distress Warrant have a significance value less than 0.05, which is
indicating that all data is not normally distributed. Therefore it is necessary to transform
the data to make the data normally distributed. Tax Arrears Disbursement, Warning
Letters, and Distress Warrants are transformed into logarithms 10 or LG10 (x).
Meanwhile, the Quality of Tax Determination is transformed into the logarithms 10 or
LG10 (k-x). The results of Kolmogorov-Smirnov test after data transformation is following
below:
Table 5 Kolmogorov-Smirnov Test After Transformation

Log10_PTP Log10_KPP Log10_ST Log10_SP
N 48 38 47 40
Normal Parametersa,b Mean -1,6725 -1,0513 -,4621 -,4080
Std. Deviation ,42436 ,83469 ,73090 ,73438
Most Extreme Absolute ,121 ,105 ,097 ,077

Differences
Positive ,086 ,104 ,065 ,077
Negative -,121 -,105 -,097 -,056
Test Statistic ,121 ,105 ,097 ,077
Asymp. Sig. (2-tailed) ,076c ,200c,d ,200c,d ,200c,d
Based on table 5 above, Tax Arrears Disbursement, The Quality of Tax

Determination, Warning Letter, and Distress Warrant has a significance value greater
than 0.05, so the data is normally distributed. After transforming the data, there are outlier
data in this research. The Quality of Tax Determination data is decreased from 48 into
38 data, Warning Letter data is decreased from 48 into 47 data, and Distress Warrants
data is decreased from 48 data into 40 data.
9
4.1.3. Multicollinearity Test
The multicollinearity test aims to find out whether there are similarities between the
independent variables. The multicollinearity is detected using tolerance value and VIF
value. If the tolerance value is more than 0.10 and the VIF value is less than 10, it means
there is no multicollinearity. If the tolerance value is less than 0.10 and the VIF value is
more than 10, it means there is a multicollinearity in this research. The results of
multicollinearity test is following below:
Table 6 Multicollinearity Test

Unstandardized Standardized Collinearity
Coefficients Coefficients Statistics
Model B Std. Error Beta t Sig. Tolerance VIF
1 (Constant) -1,899 ,141 -13,457 ,000
Log10_KPP -,181 ,085 -,389 -2,128 ,042 ,852 1,173
Log10_ST -,080 ,127 -,124 -,628 ,535 ,728 1,373
Log10_SP -,101 ,107 -,185 -,946 ,352 ,746 1,341
Based on table 6 above, all independent variables have tolerance values more than
0.10 and VIF values less than 10, so there is no multicollinearity in this research.
4.1.4. Autocorrelation test

The autocorrelation test aims to find out whether there is a high correlation between
residuals. The autocorrelation is detected using Runs Test. If the significance is more
than 0.05, there is no autocorrelation problem. If the significance is less than 0.05, there
is an autocorrelation problem.
Table 7 Autocorrelation Test With Runs Test

Unstandardized Residual
Test Valuea ,02972
Cases < Test Value 17
Cases >= Test Value 17
Total Cases 34
Number of Runs 13
Z -1,567
Asymp. Sig. (2-tailed) ,117
a. Median
10
Based on the table 7 above, the autocorrelation test using the Runs Test obtained a
significance value 0.117 which is more than 0.05, so there is no autocorrelation problem
in the regression model.
4.1.5. Heteroscedasticity Test

The heteroscedasticity test is detected using a scatterplot graph. If the dots don’t
form a clear pattern and spread above and below the number 0 on the Y axis, it means
there is no heteroscedasticity.
Figure 2 The result of Scatterplot Graph Test

After heteroscedasticity test, the points on the graph don’t have a clear pattern and
relatively spread above and below the number 0 on the Y axis, so there is no
heteroscedasticity in this research.
Besides using the scatterplot graph, heteroscedasticity test also can be detected
using the Glejser test. If the significance value is more than 0.05, it means there is no
heteroscedasticity. The heteroscedasticity test using glejser test can be seen from the
table below:
Table 8 Glejser Test

Unstandardized Coefficients Standardized Coefficients
Model B Std. Error Beta t Sig.
1 (Constant) ,324 ,073 4,458 ,000
Log10_KPP ,025 ,044 ,112 ,576 ,569
Log10_ST -,041 ,066 -,130 -,619 ,540
Log10_SP ,018 ,055 ,069 ,333 ,742
a. Dependent Variable: Abs_RES
Based on the table 8 above, the significance value of all independent variables is
more than 0.05, which means there is no heteroscedasticity in this research
11
4.1.6. Multiple Linear Regression Analysis
Multiple linear regression analysis aims to determine the significant effect of the
independent variables and the dependent variable.
Table 9 Multiple Linear Regression Analysis

1 (Constant) -1,899 ,141 -13,457 ,000
Log10_KPP -,181 ,085 -,389 -2,128 ,042
Log10_ST -,080 ,127 -,124 -,628 ,535
Log10_SP -,101 ,107 -,185 -,946 ,352
a. Dependent Variable: Log10_PTP
Based on the table 9 above, the equations in regression model is following below:
Y = -1,899 - 0,181 X1 - 0,080 X2 - 0,101 X3 + ε
Description:
1. The constant value is -1.899. It is indicates that if independent variable is assumed
to be zero, then Tax Arrears Disbursement (Y) will be decreases into 1.899.
2. The Quality of Tax Determination (X1) has a negative sign regression coefficient
0.181. It is indicates that if the other independent variables are assumed to be
constant and the Quality of Tax Determination increases one level, then the Tax
Arrears Disbursement (Y) will be decreases 0.181.
3. Warning Letter (X2) has a negative sign regression coefficient 0.080, which means
if the other independent variables are assumed to be constant and Warning Letter
increases one level, then Tax Arrears Disbursement (Y) will be decreases 0.080.
4. Distress Warrant (X3) has a negative sign regression coefficient 0.101, which means
if the other independent variables are assumed to be constant and Distress Warrant
increases one level, then Tax Arrears Disbursement (Y) will be decreases 0.101.
4.1.7. T Test
T Test aims to determine the effect of the independent variables on the dependent
variable individually. T Test is detected by comparing the significance value with 0.05. If
the significance value is more than 0.05, there is no effect of the independent variables
on dependent variables. If the significance value is less than 0.05, there is an effect of
the independent variables on dependent variables. The results of T Test can be seen
from the table below:
Table 10 T Test Result

1 (Constant) -1,899 ,141 -13,457 ,000
12
Log10_KPP -,181 ,085 -,389 -2,128 ,042
Log10_ST -,080 ,127 -,124 -,628 ,535
Log10_SP -,101 ,107 -,185 -,946 ,352
Based on the table 10 above, the conclusions of T Test is following below:

1. The first hypothesis (H1) is there is an effect of the Quality of Tax Determination on
Tax Arrears Disbursement at Kalideres Tax Office. After the T test, it is known that
the Quality of Tax Determination (KPP), has a significance value 0.042 which is less
than 0.05 (0.042 <0.05). This means that there is an effect of the Quality of Tax
Determination on Tax Arrears Disbursement so the first hypothesis (H1) is accepted.
2. The second hypothesis (H2) is there is an effect of Warning Letter on Tax Arrears
Disbursement at Kalideres Tax Office. After the T test it is known that Warning Letter
(ST) has a significance value 0.535 which is more than 0.05 (0.535 > 0.05). This
means that there is no effect of Warning Letter on Tax Arrears Disbursement so the
second hypothesis (H2) is rejected.
3. The third hypothesis (H3) is there is an effect of Distress Warrant on Tax Arrears
Disbursement at Kalideres Tax Office. After the T test it is known that Distress
Warrant (SP) has a significance value 0.352 was obtained which is more than 0.05
(0.352 > 0.05). This means that there is no effect of Distress Warrant Tax Arrears
Disbursement so the third hypothesis (H3) is rejected
4.1.8.Determination Coefficient Test (R2)

Determination coefficient test aims to calculate the level of contribution of the
independent variables in explaining the dependent variable which is indicated by the
value of Adjusted R Square. The results of determination coefficient test can be seen
from the table below:
Tabel 11 Determination Coefficient Test

Model R R Square Adjusted R Square Std. Error of the Estimate
1 ,380a ,145 ,059 ,38475
Based on the table 11 above, Adjusted R Square value is 0.059. This means the
Quality of Tax Determination, Warning Letter and Distress Warrant is able to influenced
Tax Arrears Disbursement 5.9%. Meanwhile the remaining value 94.1% is influenced by
other variables.
4.2. Research Analysis

4.2.1. The Effect of the Quality of Tax Determination on Tax Arrears Disbursement
The first hypothesis (H1) is there is an effect of the Quality of Tax Determination on
Tax Arrears Disbursement at Kalideres Tax Office. The results of testing the first
hypothesis is there is an effect of the Quality of Tax Determination on Tax Arrears
Disbursement at Kalideres Tax Office. The results of this research support the
hypothesis proposed by the author, which means the first hypothesis is accepted.
The results of this research prove that the quality of tax determination can affect tax
arrears disbursement. If the quality of tax determination is good, the determination is
carrying out through a proper tax audits so it affects the tax arrears disbursement. On
13
the other hand, if the quality of tax determination is not good, the determination is not
carried out through a proper tax audit, which makes it difficult to withdraw the tax arrears.
The function of tax determination letter is to notify and correct the amount of tax,
giving sanctions, collect taxes, and return excessive taxes (Suandy, 2011:169). Tax
determination letter is part of passive billing so if the tax authorities doesn’t collect tax
with tax determination letters, then the taxpayer will be delay in paying taxes. In addition,
the sanctions in the tax determination letter will be forcing the taxpayer to pay off the tax
obligation.
The quality of tax determination is categorized as good if the amount of
determination does not change although the Taxpayer request to reduce or eliminate
sanctions, request for objections and request an appealed to tax authorities (Hidayat &
Cheisviyanny, 2013:3). Based on the data from the list of requests for reduction
sanctions, objections, and appeals every month at Kalideres Tax Office, the authors
calculate and grouped the data using categorization to find out the quality of tax
determination.
Table 12 The Quality of Tax Determination at Kalideres Tax Office in 2017-2020

Source: Data processed by the author (2022)
No The Quality of Tax Determination Amount Percentage
1 Very Good (80 - 100%) 32 67%
2 Good (60 - 80%) 6 13%
3 Less Good (40 - 60%) 3 6%
4 Not Good (20 - 40%) 2 4%
5 Very Bad (0 - 20%) 5 10%
Total 48 100%
Based on the table 12 above, it is known that 67% of the quality of tax determination
is classified as very good, with total 32 months. There are 13% of the quality of tax
determination is classified as good, with total 6 months. There are 6% of the quality of
tax determination is classified as less good, with total 3 months. There are 4% of the
quality of tax determination is classified as not good, with total 2 months. And finally there
is 10% of the quality of tax determination is classified as very bad, with total 5 months.
The data reflects that the quality of tax determination at Kalideres Tax Office in 2017-
2020 is very good because there are more than 50% of the quality of tax determination
which is classified as very good. The performance of the tax authorities in issuing a tax
determination at Kalideres Tax Office is good and consistent because supported by
accurate data and affect the tax arrears disbursement.
The results of this research is support Redyanza (2018) and Sa'diyah (2020) who
found that the quality of tax determination has a positive and significant effect on tax
arrears disbursement. However, the results of this research is contradictive with Supriadi
and Hidyatullah (2020) who found that the quality of tax determination has a negative
effect on tax arrears disbursement.
4.2.2. The Effect of Warning Letter on Tax Arrears Disbursement

The second hypothesis (H2) is there is an effect of Warning Letter on Tax Arrears
Disbursement at Kalideres Tax Office. The results of testing the second hypothesis is
Warning Letter has no effect on Tax Arrears Disbursement at Kalideres Tax Office. The
14
results of this research don’t support the hypothesis proposed by the author, which
means the second hypothesis is rejected.
The results of this research prove that Warning Letter still not able to increase the
awareness of taxpayers and not able to maximize the tax arrears disbursement. It’s
because the taxpayers still not pay their tax obligation although Warning Letter is issued.
The factors that influenced this results are the taxpayer's financial condition is not in good
condition, the taxpayer refused Warning Letter, the taxpayer has low awareness of
paying off their tax obligation, and Warning Letter is not delivered to taxpayer’s address
(Wahdi et al., 2018:115).
One of the factors that causes Warning Letter to have no effect on tax arrears
disbursement is the Taxpayer's financial condition is currently difficult. So even though
the tax authorities has issued a Warning Letter, the Taxpayer is unable to pay their tax
obligation. In addition, Warning Letter is persuasive and there is no legal sanctions for
uncooperative taxpayers. Taxpayers still think Warning Letter is not very important and
not urgent, so they tend to ignore and delay the payment of tax obligations.
According to Maisyaroh et al. (2019:606), another factor that causes Warning Letter
to have no effect on tax arrears disbursement is Warning Letter not delivered to
Taxpayers’s address because the address is unclear or the Taxpayer has changed
address without informing the Tax Office. Then Warning Letter will be returned to Tax
Office so there is no realization of the issuance of warning letter. Furthermore, the tax
officer will verify to confirm the address and update the taxpayer data. After the data is
updated, the tax authorities will be issue a Warning Letter again so the number of
Warning Letter in the next period increases but the probability of realization is low.
The results of this research is support Maisyaroh et al. (2019), Silooy (2017) and
Purwanto (2017) who found that Warning Letter had no effect on Tax Arrears
Disbursement. However, the results of this research is contradictive with Angraeny (2017)
who found that Warning Letter has a significant effect on Tax Arrears Disbursement.
4.2.3. The Effect of Distress Warrant on Tax Arrears Disbursement

The third hypothesis (H3) is there is an effect of Distress Warrant on Tax Arrears
Disbursement at Kalideres Tax Office. The results of testing the third hypothesis is
Distress Warrant has no effect on Tax Arrears Disbursement at Kalideres Tax Office.
The results of this research don’t support the hypothesis proposed by the author, which
means the third hypothesis is rejected.
The results of this research prove that Distress Warrant still not able to increase the
awareness of Taxpayer and not able to maximize the tax arrears disbursement. It’s
because the taxpayers still not pay their tax obligation although Distress Warrant is
issued. The factors that influenced this results are taxpayer's financial conditions is
currently difficult, Taxpayers refuse Distress Warrant, The Bailiffs cannot find Taxpayers
address because the address is unclear or Taxpayers changed the addresses, and the
amount of Tax Bailiffs is lack and limited (Wahdi et al., 2018:115).
The factors that causes Distress Warrant to has no effect on tax arrears
disbursement is taxpayer's financial condition is not good. This makes the Taxpayer
unable to pay off the tax obligation even though the Bailiff issued and gave Distress
Warrant to Taxpayer.
The next factor is taxpayers rejected Distress Warrant and delay paying the tax
obligation because the taxpayer is currently submitting an objection or appeal against
the amount of tax arrears to reduce tax sanctions or cancel the Notice of Tax Collection
(STP) or Notice of Tax Assessment (SKP). So even though the bailiff has notified
Distress Warrant, the taxpayer refuse to pay their tax obligation until a decision is
obtained on taxpayer's request.
15
Notification of Distress Warrant is carried out by the Bailiff by visiting the Taxpayer’s
address and read the Distress Warrant in front of Taxpayer. However, in practice the
bailiffs often cannot find the taxpayer’s address because the address is unclear or the
address has changed without notified Tax Office. This causes the tax arrears
disbursement by issuing Distress Warrant cannot be carried out.
According to Maisyaroh et al. (2019:606), Human Resources (HR) problem can be
a factor that causes Distress Warrant is difficult to delivered to Taxpayer. The amount of
bailiffs at Kalideres Tax Office is only two officers. Meanwhile, the amount of Distress
Warrant is quite a lot. Based on PMK No.189/PMK.03/2020 Article 2 Paragraph (4),
beside being tasked to notifying Distress Warrant, the Bailiff also has other duties, like
carrying out billing actions, confiscations, and taking hostages. The tasks that must be
carried out by Bailiff causes Distress Warrant becomes less intensive so the Tax Arrears
Disbursement cannot run optimally because it takes a long time.
The results of this research is support Maisyaroh et al. (2019), Silooy (2017) and
Purwanto (2017) who found that Distress Warrant had no effect on Tax Arrears
Disbursement. However, the results of this research is contradictive with Angraeny (2017)
who found that the Distress Warrant has a significant effect on Tax Arrears Disbursement.
5. CONCLUSION
The conclusion in this study is the quality of tax determination affects the tax arrears
disbursement at Kalideres Tax Office. The quality of tax determination that classified as
good reflects that the performance of tax authorities in issuing tax determination is good
because it is consistent. Meanwhile Warning Letter and Distress Warrant had no effect
on tax arrears disbursement at Kalideres Tax Office. The results of this study indicate
that Warning Letter and Distress Warrant are not able to raise awareness of Taxpayer
and not able to maximize tax arrears disbursement. It is because Taxpayer still not paid
their tax obligation after tax authorities issued Warning Letter and Distress Warrant.
6. IMPLICATION, LIMITATION, AND SUGGESTION

6.1. Implications
This research contributes to enriching knowledge and literature, especially in the
field of taxation and obtaining empirical evidence about the factors that influence the tax
arrears disbursement.
This research contributes to KPP Pratama Jakarta Kalideres as evaluation material
so the improvements can be made and the amount tax arrears disbursement can be
increase. The improvements are carrying out a proper tax audits with accordance
procedures to produce a good quality of tax determination, organize intensive tax
collection, organize counseling about tax for Taxpayer, increasing the number of bailiffs
in Tax Office so the tax collection actions can be more effective and efficient, and
requiring taxpayers to update their home address periodically so Warning Letter and
Distress Warrant can be sent to Taxpayer's address.
This research contributes to the government as an suggestion to increase public
awareness and public compliance in paying taxes. For example holding intensive
socialization and counseling about the importance of paying taxes, making advertising
about procedures of tax reporting in media.
6.2. Limitations
The limitations in this research are following below:
1. The observation period used in this research is relatively short, that is only four years
(2017-2020) so it will be affects the accuracy of the research results.
2. The unit of analysis used in this research is limited to Kalideres Tax Office.
16
3. This research only uses three independent variables, namely the Quality of Tax
Determination, Warning Letter, and Distress Warrant.
6.3. Suggestions
The suggestions in this research that can be a consideration for further research are
following below:
1. Add a longer observation period to obtain more accurate results.
2. Expanding the unit of analysis by using other Tax Office to obtain broader results.
3. Add more independent variables that have not been used in this research which may
affect the tax arrears disbursement such as Confiscation Letter, administrative
sanctions, tax audits, taxpayer compliance, etc.
BIBLIOGRAPHY
Angraeny, D. M. (2017). Pengaruh Penerbitan Surat Teguran, Surat Paksa, dan Surat
Perintah Melakukan Penyitaan terhadap Pencairan Tunggakan Pajak Oleh Wajib
Pajak Badan. JOM Fekon, 4(2), 4752–4765.
Fatmandika, D., Susilo, H., & Agusti, R. R. (2016). Pengaruh Surat Teguran terhadap
Pencairan Tunggakan Pajak dengan Surat Paksa sebagai Variabel Intervening.
Jurnal Perpajakan (JEJAK), 9(1), 1–8.
https://www.perpajakan.studentjournal.ub.ac.id
Hidayat, R., & Cheisviyanny, C. (2013). Pengaruh Kualitas Penetapan Pajak dan
Tindakan Penagihan Aktif terhadap Pencairan Tunggakan Pajak. Jurnal WRA, 1(1),
1–20.
Maisyaroh, U., Harimurti, F., & Suharno. (2019). Pengaruh Surat Teguran, Surat Paksa,
dan Surat Perintah Melaksanakan Penyitaan Terhadap Pencairan Tunggakan
Pajak Penghasilan Orang Pribadi. Jurnal Akuntansi Dan Sistem Teknologi
Informasi, 15(4), 599–608.
Nainggolan, Y. D. (2015). Pengaruh Penagihan Aktif dengan Surat Teguran dan Surat
Paksa terhadap Pencairan Tunggakan Pajak di Kantor Pelayanan Pajak Madya
Pekanbaru. Jom FEKON, 2(2), 1–15.
Pohan, C. A. (2017). Pembahasan Komprehensif Pengantar Perpajakan Edisi 2 Teori
dan Konsep Hukum Pajak (2nd ed.). Jakarta: Mitra Wacana Media.
Purwanto. (2017). The Effect of Tax Collection with Warning Letter and Distress Warrant
to Tax Arrears Disbursement. Jurnal Aset (Akuntansi Riset), 9(2), 93–104.
http://ejournal.upi.edu/index.php/aset
Ratnawati, J., & Hernawati, R. I. (2015). Dasar-Dasar Perpajakan (1st ed.). Yogyakarta:
Deepublish.
Redyanza, & Siti Khairani. (2018). Pengaruh Kualitas Penetapan Pajak, Pemeriksaan
Pajak, Tindakan Penagihan Aktif Pajak Terhadap Pencairan Tunggakan Pajak. 1–
11.
Rochmawati, M. (2015). Pengaruh Kualitas Penetapan Pajak, Pemeriksaan Pajak, dan
Surat Paksa terhadap Pencairan Tunggakan Pajak. Jurnal FEKON, 2(2), 1–11.
Sa’diyah, N. A., & Idayati, F. (2020). Pengaruh Kualitas Penetapan Pajak dan Tindakan
Penagihan Aktif terhadap Pencairan Tunggakan Pajak KPP Pratama Surabaya.
Jurnal Ilmu Dan Riset Akuntansi, 9(16), 1–18.
Silooy, R. W. (2017). Pengaruh Tindakan Penagihan Aktif Terhadap Pencairan
Tunggakan Pajak. Jurnal SOSOQ, 5(1), 80–98.
Suandy, E. (2011). Perencanaan Pajak (Edisi 5). Jakarta: Salemba Empat.
Suandy, E. (2016). Hukum Pajak (Edisi 7). Jakarta: Salemba Empat.
Supriadi, & Hidyatullah, M. (2020). Pengaruh Kualitas Penetapan Pajak dan Tingkat
Penagihan Aktif terhadap Pencairan Tunggakan Pajak (Studi Kasus Pada KPP
17
Pratama Surabaya Mulyorejo Tahun 2015-2018). Jurnal Liability, 2(2), 73–86.
https://journal.uwks.ac.id/index.php/liability
Wahdi, N., Wijayanti, R., & Danang. (2018). Efektivitas Penagihan Pajak dengan Surat
Teguran, Surat Paksa, dan Penyitaan dan Kontribusinya Terhadap Penerimaan
Pajak di KPP Pratama Semarang Tengah Satu. Dinamika Sosial Budaya, 20(2),
106–119. http://journals.usm.ac.id/index.php/jdsb
Waluyo. (2013). Perpajakan Indonesia (Edisi 11). Jakarta: Salemba Empat.
Widiasfita, S. (2018). Pengaruh Kualitas Penetapan Pajak, Surat Teguran, dan Surat
Sita terhadap Pencairan Tunggakan Pajak. Universitas Negeri Jakarta.
OKUNS LAW, PHILLIPS CURVE AND ITS EFFECT ON
18
THE GROWTH OF INCOME TAX ARTICLE 21
PAYMENTS DURING COVID-19 PANDEMIC
Galih Ardin
The National Graduate Institute for Policy Studies, Japan. Email: mef19502@grips.ac.jp
ABSTRACT
The COVID-19 pandemic has changed the macroeconomics face not only in Indonesia but also in
other countries in the world. Restrictions on social activities, lockdowns, a decline in aggregate
demand and supply as well as a drop in export and import activities have triggered a decline in
economic growth, rising unemployment rate, and falling inflation rates leading to deflation. Logically,
the turmoil in macroeconomic indicators will affect tax revenues, especially Article 21 Income Tax.
Through the Okus Law and Phillips Curve approach, the author tries to examine the relationship
between unemployment rates, economic growth, and inflation rates during the pandemic on income
tax payments Article 21. Based on the results tested using the Ordinary Least Squares (OLS)
method, it is known that the unemployment rate does not have a significant effect on economic
growth and inflation rates in Indonesia. In addition, economic growth also does not affect the
payment of Income Tax Article 21. However, there is an interesting thing where the inflation rate
has a positive effect on the payment of Income Tax Article 21 in Indonesia.
Keywords: okuns, phillips curve, covid-19, income tax article 21, macroeconomics
Pandemi covid 19 telah mengubah tatanan dan wajah makroekonomi tidak hanya di Indonesia tetapi
juga negara – negara lain di dunia. Pembatasan kegiatan sosial, lockdown, penurunan permintaan
dan penawaran agregat serta anjloknya kegiatan ekspor dan impor telah memicu terjadinya
penurunan pertumbuhan ekonomi, naiknya angka pengangguran hingga merosotnya angka inflasi
yang mengarah kepada deflasi. Secara logis, gejolak indikator makro ekonomi tersebut akan
berpengaruh terhadap penerimaan perpajakan khusunya PPh Pasal 21. Melalui pendekatan hukum
Okus dan Phillips Curve, penulis berusaha meneliti hubungan antara tingkat pengangguran,
pertumbuhan ekonomi dan tingkat inflasi pada masa pandemi terhadap setoran PPh Pasal 21.
Berdasarkan hasil pengujian dengan metode Ordinary Least Squares (OLS) diketahui bahwa
tingkat pengangguran tidak mempunyai pengaruh signifikan terhadap pertumbuhan ekonomi dan
tingkat inflasi di Indonesia. Selain itu, pertumbuhan ekonomi juga tidak berpengaruh terhadap
pembayaran PPh Pasal 21. Namun demikian, ada hal yang menarik dimana, tingkat inflasi
berpengaruh positif terhadap setoran PPh Pasal 21 di Indonesia.
Kata kunci: okuns, phillips curve, covid-19, PPh 21, makroekonomi
1. INTRODUCTION
1.1. Background
More than two years of the COVID-19 pandemic has hit Indonesia and other
countries around the world. Based on data from the Covid-19 Countermeasures Task
Force (2022), it is known that until the end of June 2022 the number of positive confirmed
cases of Covid-19 in Indonesia reached 6,06 people. Of these, 5.89 million were
declared cured and 156,622 of them died.
Unfortunately, the Covid-19 pandemic does not only attack the health sector. The
economic and financial sectors were also devastated due to the Covid-19 pandemic.
Badan Pusat Statistik (2022) noted that at the time of the Covid-19 pandemic, the
country's growth rate contracted by 1.26%. In fact, in the same period the previous year,
economic growth in Indonesia was able to reach 5.06%.
The same condition was happened in the employment sector. BPS (2022) reported
that in August 2020, or the beginning of the entry of the coronavirus-19, the
19
unemployment rate in the country reached 7.08%. Compared to the same period the
previous year, the unemployment rate in Indonesia in the first quarter of 2020
experienced a sharp increase where in the first quarter of 2019 the number of
unemployment in Indonesia is 5,23%.
This condition continues in 2021. In the first and third quarters of 2021, the
unemployment rate in Indonesia was still in the range of 6,26% and 6,49%. However,
better results are experienced in 2022 where in the first quarter of 2022, the
unemployment rate in Indonesia is 5.83% (BPS, 2022).
Theslowdown in economic growth and the increase in unemployment during the
pandemic was caused by a decrease in aggregate demand and supply as well as social
restrictions on community activities as an effort to prevent the spread of the COVID-19
virus (Ministry of Finance, 2021).
Although at this time the covid-19 pandemic conditions are relatively under control,
the economic growth rate has begun to improve and the unemployment rate has begun
to decline, it is not impossible that one day similar conditions will be repeated.
Therefore, a study is needed to map the relationship between economic growths,
unemployment rate and tax revenue in Indonesia. This mapping is important to give the
government an idea of the economic topography in the country.
Basically, there have been several studies that discuss the impact of the COVID-19
pandemic on the workforce and unemployment. For example, Ngadi, Meillianna and
Purba (2020) found that during the Covid-19 pandemic, the termination rate reached
15.6% of the total formal workforce. Another study conducted by Fahri, Jalil and
Kasnelly found that the covid-19 pandemic has an influence on the unemployment rate.
However, unfortunately there is no research that specifically discusses the effect of
economic growth, unemployment rate and inflation rate on tax revenues, especially
article 21 income tax revenues in pandemic period.
Therefore, through Okun's law and Philips Curve approaches, the I tries to find out
the relationship between economic growth and the unemployment rate in Indonesia and
its effect on growth receipt of Income Tax Article 21.
The study consists of five parts. The first part discusses the background and reasons
for conducting the research. The second part talks about the theoretical framework and
the development of hypotheses. The third section describes the research methodology.
Furthermore, the results of the discussion are explained in the fourth section and the
conclusions and suggestions will be reviewed in the fifth part.
2. TEORETIS FRAMEWORK AND HYPOTHESIS DEVELOPMENT

2.1. Theoretical framework
Okuns' law has long been accepted by economists as one of the references
discussing the relationship between economic growth and the unemployment rate.
Okun's' law states that there is a relationship between economic growth and the
unemployment rate in USA where if there is an increase in economic growth by 3% it
will have an impact on decrease in the unemployment rate by 1% (Prachowny, 1993).
On the other hand, the Phillips curve is used by economists to explain the relationship
between the unemployment rate and the inflation rate. In other words, the lower the
unemployment rate, the higher the inflation rate. (Dritsaki & Dritsaki, 2013)
In subsequent developments, Okun’s Law and Phillips Curve continued to undergo
improvements and refinements. As done by Soylu, Çakmak, & Okur (2018) who
investigated the relationship between economic growth and unemployment in some
Eastern European countries for the period 1992-2014 in the context of Okun's law. The
results of research by Soylu et al (2018) show that unemployment has a positive effect
on economic growth.
20
Next, Yelwa, Okoroafor, & Awe (2015) study the same relationship by adding
variables for the period between 1987 and 2012. The results are no different from
previous studies, namely that inflation shows a negative influence on output.
Research conducted by Darman (2013) with time series data analysis method using
Ordinary Least Square (OLS) method. The study concluded that Okun's law proved
invalid in the Indonesian economy because it was different from the original coefficient
of Okun's law. The degree of significance of the value of the Okun coefficient is quite
small empirically. This shows that the unemployment rate in Indonesia is not affected by
changes in real output.
Kanyarat (2010) analyzed the Phillips curve for Thailand using the OLS method using
two categories of inflation (quarterly and annual). In his research, he showed that before
the 1997 economic crisis there was no relationship between inflation and unemployment.
This negative relationship between inflation and unemployment only occurred after the
1997 economic crisis which had provided a strong structural shock to the capacity of the
economy and the financial sector.
On the other hand, Amir (2007) analyzed the effect of inflation and economic growth
on unemployment in Indonesia in the period 1980-2005 using graphical analysis and
regression of the ANOVA method. The dependent variable is the unemployment rate and
the independent variable is inflation. The result is that there is a negative but insignificant
(no influence) relationship between inflation and economic growth to unemployment both
statistically and graphically. This is thought to be because inflation in Indonesia is more
likely to be caused by an increase in production costs, such as an increase in fuel oil
(BBM), not due to an increase in demand.
2.2. Hypothesis framework

As explained in the introduction, through this paper the author wants to examine the
relationship between economic growth and inflation to the unemployment rate and its
effect on the growth of Income Tax Article 21 payments.
Therefore, in order to fulfilling the purpose of the study, there are three research
questions to be answered in this study: First, how does the unemployment rate affect
economic growth? Second, how does the unemployment rate have a correlation with
the inflation rate? Third, how does economic growth and inflation rate affect the growth
of Income Tax Article 21 payments?
To answer the first research question, I used Okun's Law approach with the initial
hypothesis that the unemployment rate has no effect on economic growth (H0) and the
alternative hypothesis (H1) is Okun's law occurs in Indonesia
Meanwhile, to answer the second research question, I use the Phillips Curve
approach with the unemployment rate hypothesis not affecting the inflation rate as the
initial hypothesis (H0) and the unemployment rate hypothesis has no effect on the
inflation rate as an alternative hypothesis (H1).
Furthermore, to answer the third research question, the author uses the hypothesis
that economic growth and inflation rates do not affect the growth of Article 21 income tax
deposits in Indonesia as the initial hypothesis (H0) and the hypothesis that economic
growth and inflation rates affect the growth of Article 21 income tax deposits in Indonesia
as an alternative hypothesis (H 1).
There are four main data that will be used in this study, namely: GDP growth data,
unemployment rate, inflation rate and growth in Article 21 income tax payments. Data on
GDP growth, inflation rate and unemployment are taken from the World Bank data with
the consideration that the World Bank provides historical data with a relatively longer
period of time, making it easier to perform regression analysis.
21
In addition, World Bank’s data also provides data on economic growth, inflation and
the level of other countries, making it easier to compare Indonesia's economic conditions
with other countries. For Income Tax Article 21 payment data, I use officially published
DGT data.
In testing the hypothesis of the first research question, I used linear regression with
a difference version approach as follows:
Yt – Yt-1 = α + β (Ut – Ut-1) + εt
Where Y t – Y t-1 is economic growth, α is interception, β is the coefficient, (U t – U t-

1) is the change in the unemployment rate and ε is the term error.
Furthermore, in order to examine the hypothesis of the second research question,

the authors used linear regression with the following equation:
Yi = β 0 + β1X1i + ui, i=1, . . ., n
Where Y i is the inflation rate, β0 is the intercept, β1 is the coefficient, X1i is the
unemployment rate and ui is the term error.
Next, in testing the hypothesis of the third research question, the authors used linear
regression with the following equation:
Yi = β 0 + β 1 X 1i + β2X2i + ui, i=1, . . ., n
Where Y i is the growth of Income Tax payments Article 21 year on year (yoy), β0 is
the intercept, β1 is the coefficient for the economic growth variable, X 1i is the economic
growth variable (GDP growth), β2 is the coefficient for the inflation variable, X2i is the
inflation variable, while ui is a term error.
In simple, the relationship between these variables can be described in the following
figure
Figure 1 : Relationship Between Variables in Research
Unemploy Economic
Research Question 1 Growth
ment Rate
Yt – Yt-1 = α + β (Ut – Ut-1) + εt
Unemploy Inflation
Research Question 2 ment Rate Rate
Yi = β 0 + β1X1i + ui
Economic
Growth Change
Research Question 3 in PPh 21
Inflation Payment
Rate
22
Yi = β 0 + β 1 X 1i + β2X2i + ui
Due to limited time and resources, the writing team limited this study to variables of
economic growth, inflation, unemployment rate from 1992 to 2021 in Indonesia.
4. RESULTS AND DISCUSSION

4.1. Overview of economic growth in Indonesia
According data from the World Bank (2022), it is known that basically economic
growth in Indonesia is running quite well. The highest economic growth was achieved in
1968 where in that year Indonesia experienced a growth of 10.92%. However, in 1998
Indonesia experienced the lowest economic growth of -13.13%. This economic growth
gradually increased in the following year as the trade balance improved to 0.79% in 1999
and 4.92% in 2000. In fact, when the subprime mortgage crisis occurred in 2009,
economic growth in Indonesia was still able to grow by 4.63%. Details about economic
growth in Indonesia can be seen in chart 1 as follows:
Chart 1: Ekonomic Growth in Indonesia in 1961 – 2021

15,00%
10,00%
Economic Growth
5,00%
0,00%
1961
1976
1964
1967
1970
1973
1979
1982
1985
1988
1991
1994
1997
2000
2003
2006
2009
2012
2015
2018
2021
-5,00%
-10,00%
-15,00%
Year
Growth
Source: World Bank, 2022.
Compared to others ASEAN countries and OECD member, economic growth in

Indonesia is relatively lower than those in the period from 1961 to 1968. However, in the
period from 1968 to 1998, Indonesia's economic growth showed an increase and was
consistently in a higher position than the OECD countries and several ASEAN countries.
In that period, only Singapore and Thailand had higher economic growth than Indonesia.
However, when the economic crisis occurred in 1998, economic growth in Indonesia
experienced the deepest contraction compared to other countries. In the following years
after the economic crisis, economic growth in Indonesia was relatively more stable than
other countries. Details on the economic growth of ASEAN countries and OECD member
countries can be seen in chart 2 as follows:
23
Figure 2: Economic Growth in ASEAN andOECD Countries in 1961 – 2021
25,00%
20,00%
15,00%
10,00%
Economic Growth
5,00%
0,00%
1961
1964
1967
1970
1973
1976
1979
1982
1985
1988
1991
1994
1997
2000
2003
2006
2009
2012
2015
2018
2021
-5,00%
-10,00%
-15,00%
-20,00%
-25,00%
Year
Indonesia Malaysia Singapore

Brunei Darussalam Vietnam Lao PDR
Myanmar Thailand Philippines
OECD
4.2. Overview of the unemployment rate in Indonesia

Based on World Bank data (2022) it also known that in 1991 the unemployment rate
in Indonesia was generally low at 2.62%. The unemployment rate in Indonesia had
experienced a spike in 1994 and 1998. In fact, it experienced its peak in 2007 where at
that time the unemployment rate in Indonesia reached 8.06%. This unfavorable condition
gradually improved in the next period so that Indonesia could reduce the unemployment
rate to touch the level of 3.62% in 2019.
In 2020 the unemployment rate in Indonesia again increased to 4.11%. This is
nothing but due to the COVID-19 pandemic that has hit Indonesia and other countries in
the world. The details of the development of the unemployment rate in Indonesia can be
seen in chart 3 as follows:
24
Chart 3: Unemployment rate in Indonesia in the period 1991 - 2021
9,00%
8,00%
Unemployment Rate
7,00%
6,00%
5,00%
4,00%
3,00%
2,00%
1,00%
0,00%
Year
Unemployment Rate
According data of the World Bank (2022), it is known that the unemployment rate in
Indonesia is generally higher when compared to others ASEAN countries. However,
when compared to OECD countries, the unemployment rate in Indonesia is relatively
lower. The unemployment rate in Indonesia is higher than that of OECD countries only
in the period from 2004 to 2009. We can see details about the unemployment rate in
chart 4 as follows:
Figure 4: Unemployment Rate in ASEAN andOECD Countries in 1961 – 2021

10,00%
Unemployment Rate
8,00%
6,00%
4,00%
2,00%
0,00%
1991 1993 1995 1997 1999 2001 2003 2005 2007 2009 2011 2013 2015 2017 2019 2021
Year
Indonesia Malaysia Brunei Darussalam

Singapore Philippines Myanmar
Lao PDR Cambodia Vietnam
Thailand OECD members
25
4.3. Overview of inflation in Indonesia
Based on the World Bank report (2022) it is known that during the period from 1990
to 2020 inflation in Indonesia was generally less than 10%. However, in 1998 the inflation
rate in Indonesia reached a peak of 58.45% and in 1999 the inflation rate in Indonesia
reached 20.48%. This high inflation rate gradually fell the following year until it touched
3.69% in 2020. The lowest inflation rate in Indonesia was experienced in 2019 at 3.03%.
Fluctuations in the inflation rate in Indonesia can be seen in chart 5 as follows:
Chart 5: Inflation rate in Indonesia in the period 1991 – 2021

80,00%
Inflation Rate
60,00%
40,00%
20,00%
0,00%
Year
Series 1
If we compare with other countries, then we can know that the inflation rate in
Indonesia is generally higher than that of ASEAN countries and OECD countries.
However, the inflation rate in Indonesia is generally lower than that of Laos and Myanmar
as illustrated in Figure 6 as follows:
Chart 6: Inflation Rate in ASEAN and OECD Countries In 1991 - 2021

150,00%
100,00%
Inflation Rate
50,00%
0,00%
1991 1993 1995 1997 1999 2001 2003 2005 2007 2009 2011 2013 2015 2017 2019 2021
-50,00%
Year
Indonesia Malaysia Singapore

Brunei Darussalam Philippines Thailand
Vietnam Lao PDR Myanmar
Cambodia OECD members
4.4. Overview of Income Tax Deposit Article 21 in Indonesia

Based on DGT data, it is known that the payment of Income Tax Article 21 in
Indonesia generally shows a good trend. Nominally, the amount of PPh Article 21
payments continues to increase. In fact, in 2019 the amount of Income Tax Article 21
26
paid by taxpayers reached 146 trillion Rupiah. This number decreased slightly in 2020
to 136 trillion rupiah as a result of the Covid-19 pandemic.
However, the amount of PPh Article 21 deposit in 2020 is still better than the amount
of PPh article 21 deposit in 2018 of 135 trillion rupiah. To get an idea of the trend of PPh
Article 21 payments in Indonesia, we can see the chart as follows:
Graph 7: Value and Growth of Income Tax Payments Article 21

in the Period 1991 - 2021
160.000.000.000.000 1200%
140.000.000.000.000 1000%
120.000.000.000.000
800%
100.000.000.000.000
600%
80.000.000.000.000
400%
60.000.000.000.000
200%
40.000.000.000.000
20.000.000.000.000 0%
- -200%
Setoran PPh Pasal 21 Growth yoy
Source: processed author
Based on the chart above, we can see that nominally the payment of Income Tax
Article 21 has continued to increase since 1991. However, if we look at year on year
growth , we will find that the growth of Income Tax Article 21 payments had experienced
a drastic spike in 1997, 2002 and 2008. This could be due to the payment of employee
lay-off severance pay as a result of the economic crisis, political crisis and subprime
mortgage crisis in those years. To prove this, in the next section the author will analyze
the relationship between the payment of Income Tax Article 21 with economic growth
and inflation.
4.5. The relationship between economic growth and unemployment in

Indonesia (Okun’s Law)
Okuns' law states that there is a negative relationship between the unemployment
rate and economic growth. Therefore, in examining the law of ocons in Indonesia, the
author uses the initial hypothesis (Ho) that changes in economic growth have no effect
on the growth of the unemployment rate in Indonesia.
To test this hypothesis, the authors used time series data on changes in the
unemployment rate and changes in economic growth from 1992 to 2020 in Indonesia
with the following results:
27
Table 1 : Results of Testing the Unemployment Rate with Economic Growth
Based on these tests we can see that changes in the unemployment rate are
negatively correlated to changes in GDP with a coefficient of –2,07. This means that any
1% increase in the unemployment rate change will reduce economic growth by 2.07%.
However, in the Table we can also see that the regression has a t-statistic of – 1,26 or
lower than its critical value of 1,96. This can be interpreted to mean that statistically
changes in the unemployment rate have no significant effect on economic growth.
In addition, the test results also showed an R-square value of 0,0555 which means
that the vari abel change in the unemployment rate can only explain by 5,55% of the
changes that occur in the economic growth rate. Mathematically, the relationship
between the unemployment rate and economic growth can be explained as follows:
Yt – Yt-1 = α + β (Ut – Ut-1) + εt (i.i.d)
Yt – Yt-1 = 0,01 – 2,07 (Ut – Ut-1) + εt
Where Yt – Y t-1 is the change in the economic growth rate, Ut – U t-1 is the change
in the unemployment rate and εt is the term error.
Chart 8: Scatter Plot of Economic Growth by Unemployment Rate
The fact that the unemployment rate does not have a significant effect on the
economic growth rate is in line with the findings of Astuti (2016) which states that there
28
is a negative relationship between changes in economic growth rates and changes in the
unemployment rate in Indonesia but the negative relationship isnot statistically
significant. This shows that economic growth in Indonesia is not strong enough to
improve the welfare of its population.
4.6. The Relationship between Inflation and Unemployment in Indonesia

(Phillips Curve)
The Phillips curve explains that there is a negative relationship between the
unemployment rate and inflation. That is, a year with a low unemployment rate generally
has a high inflation rate. Conversely, years with high unemployment rates tend to have
low inflation rates.
However, based on research on the unemployment rate and inflation in Indonesia
using the regression method, it is known that in Indonesia the unemployment rate and
inflation rate do not have a significant correlation. This is proven by the lower t-statistic
value than the critical value of 1,96 (confidence level 95%). In addition, the p-value in the
regression test also shows a number of 0,064 or greater than the critical value of 0,05.
In theory, this led to the initial hypothesis that there was no correlation between the
unemployment rate and the inflation rate to be plausible or statistically acceptable.
Furthermore, the results of this test indirectly say that the condition of the Phillips curve
does not occur in Indonesia as shown by the following graph:
Graph 9: Scatter Plot of Inflation Rate with Unemployment Rate
The relationship between the inflation rate and the unemployment rate was
previously discussed by Sukirno. Sukirno (2019) argues that the inflation rate that occurs
will result in an increase in the interest rate (loans). An increase in the interest rate will
reduce investment to develop productive sectors. This will affect the high number of
unemployed due to low employment opportunities as a result of low investment. This is
in line with the findings of the test results which state that in Indonesia, there is a positive
correlation between inflation and the unemployment rate.
Furthermore, the test results also stated that the value of R-squared was 0,0353. We
can translate this fact as the assumption that the variable unemployment rate can only
explain 3,53 for the inflation that occurs in Indonesia. The rest are still many factors that
affect the inflation rate in Indonesia. Mathematically, the relationship between the
inflation rate and the unemployment rate is as follows:
29
Yi = β0 + β1X1i + ui (i.i.d.)
Yi = 0,2 + 1,2X + ui
Table 2 : Testing Results of Unemployment Rate by Inflation Rate
4.7. The Relationship Between Unemployment, Economic Growth And Inflation

Against Income Tax Deposits Article 21
In order to answer the third research question, namely determining the relationship
between article 21 income tax payments, economic growth and inflation rates,
researchers used data on the growth of Article 21 income tax payments (year on year)
since 1991. In addition, researchers also use economic growth and inflation data
provided by the World Bank.
Table 3: Results of Testing the Growth of Income Tax Payments Article

21 with the Rate of Inflation and Economic Growth
Based on the test results, it is known that the variable GDP growth has a t-statistical
value of 1,88 or lower than the critical value of 1,96. This means that the initial hypothesis
that economic growth has no effect on the growth of Article 21 income tax payments is
acceptable even though economic growth has a coefficient of 15,8.
On the other hand, the test results also found that the inflation variable had a t-
statistic of 1,98 or greater than the critical value of 1,96. This t-statistical value shows
that the initial hypothesis that the inflation rate had no effect on the growth of Article 21
income tax deposits was rejected while explaining that the inflation rate had an effect on
the growth of Article 21 income tax payments.
30
Furthermore, the R-squared value in this analysis is 0,0477. This means that the
economic growth variable and the inflation variable can only account for 4,7% of the
changes that occurred in the income tax article 21 payment. Apart from the inflation rate
and economic growth, there are still various other factors that affect the payment of
Income Tax Article 21.
Mathematically, the relationship between the growth of Income Tax payments in
Article 21 and economic growth and the inflation rate is described in the following
equation:
Yi = β0 + β1 X 1i + β2X2i + ui
Y i = -0,3 + 15,8X1i + 7,2X2i + ui
Where Yi is the rate of economic growth, X1i is the rate of economic growth, X2i is the
rate of inflation and ui is the error terms.
Based on the data, facts and test results above, we can presumably know that
economic growth does not directly affect the growth of income tax Article 21 payments.
On the other hand, the inflation rate affects the growth of Article 21 income tax payments.
This shows that the labor market in Indonesia is elastic to rising prices. Any increase in
the price of goods will be responded to by an increase in wages and salaries of
employees. This could be due to the strength of trade unions asking for an increase in
the regional minimum wage.
5. CONCLUSION
Based on the regression test of changes in economic growth with changes in the
unemployment rate, it is known that there is no significant relationship between the
unemployment rate and economic growth in Indonesia. This shows that statistically, the
law of ocons does not occur in Indonesia.
Based on the regression test of the inflation rate with the unemployment rate, it is
also known that there is no significant relationship between the inflation rate and the
unemployment rate. This also answers the second research question that the phillips
curve does not apply in Indonesia.
Based on the regression test between the growth of Article 21 income tax payments
and the inflation rate and economic growth, it is known that economic growth does not
have a direct effect on the growth of Article 21 income tax payments. However, based
on the test results, it is known that the inflation rate has a positive correlation with the
growth of Article 21 income tax deposits in Indonesia.
Based on research, it is known that economic growth does not directly affect the
growth of Income Tax Article 21 payments. Therefore, the growth of Regional Gross
Domestic Product (GRDP) should not be used by the DGT Regional Office in determining
the target of monitoring the period payment (PPM) of Income Tax Article 21.
Based on the results of the study, it is also known that regional inflation provided by
BPS and Bank Indonesia can be used by the DGT Regional Office as an indicator of
growth in Income Tax Article 21 payments because based on the test results, it is known
that the inflation rate has a positive effect on the growth of Article 21 income tax deposits.
Therefore, a more in-depth study is carried out on what factors statistically affect the
payment of Income Tax Article 21 so that Period Payment Supervision (PPM) can be
evaluated periodically.
31
The authors realize that there are still shortcomings in both the process, methods
and results presented in this study. Therefore, the author encourages further research
to be carried out in knowing the relationship between economic growth, inflation rate
and unemployment rate to receipts tax by using more complete data, more persistent
variables and more comprehensive methods.
References
Amir, A. (2007). The Effect of Inflation and Economic Growth on Unemployment in
Indonesia. Journal of Inflation and Unemployment, 127 – 137.
Astuti, P. B. (2016). Analysis of the Philips Curve and Okun Law in Indonesia 1986-2016.
Business Focus : Management and Accounting Assessment Media, 15(1), 72-
91. doi:https://doi.org/10.32639/fokusbisnis.v15i1.72
Central Bureau of Statistics. (2022, June 13). [2010 Series] 4. GDP Growth Rate by
Expenditure (Percent), 2019. Retrieved from the Central Bureau of Statistics:
https://www.bps.go.id/indicator/169/108/4/-seri-2010-4-laju-pertumbuhan-pdb-
menurut-pengeluaran.html
Central Bureau of Statistics. (2022, June 13). Number and Percentage of Working and
Unemployed Population 2021-2022. Retrieved from the Central Bureau of
Statistics: https://www.bps.go.id/indicator/6/1953/1/jumlah-dan-persentase-
penduduk-bekerja-dan-pengangguran.html
Darman, D. (2013). The Effect of Economic Growth on the Unemployment Rate: An
Analysis of Okun's Law. The Winners, 14(1), 1-12. doi:10.21512/tw.v14i1.639
Dritsaki, C., & Dritsaki, M. (2013). Phillips curve inflation and unemployment: an empirical
research for Greece. Journal of Computional Economics and Econometries, 3(1),
27-42. doi:10.1504/IJCEE.2013.056265
Kanyarat, B. (2010). The Phillips curve for Thailand. EcoMod.
Ministry of Finance of the Republic of Indonesia. (2021). Recording the Covid-19
Pandemic and understanding the Hard Work of state budget guards. Jakarta:
Ministry of Finance.
Ngadi, Meliana, R., & Purba, y. A. (2020). The impact of covid-19 pandemic on layoffs
and workers' incomes in Indonesia. Indonesian Journal of Population , 43-48.
Prachowny, M. F. (1993). Okun's Law: Theoretical Foundations and Revised Estimates.
The MIT Press, 331-336. doi:https://doi.org/10.2307/2109440
Covid-19 Response Task Force. (2022, May 13). Scatter data. Taken back from covid-
19: https://covid19.go.id/
Soylu, Ö. B., Çakmak, İ., & Okur, F. (2018). Economic growth and unemployment issue:
Panel data analysis in Eastern European Countries. Journal of International
Studies, 11(1), 93-107. doi::10.14254/2071-8330.2018/11-1/7
Sukirno, S. (2019). Macroeconomics : introductory theory. Jakarta: Rajawali Pers.
The World Bank. (2022, July 14). GDP growth (annual %) - Indonesia. Taken back from
The Wolrd Bank:
https://data.worldbank.org/indicator/NY.GDP.MKTP.KD.ZG?locations=ID
The World Bank. (2022, July 14). Inflation, consumer prices (annual %) - Indonesia.
Diambil kembali dari The World Bank:
https://data.worldbank.org/indicator/FP.CPI.TOTL.ZG?locations=ID
The World Bank. (2022, July 14). Unemployment, total (% of total labor force) (modeled
ILO estimate) - Indonesia. Diambil kembali dari The World Bank:
https://data.worldbank.org/indicator/SL.UEM.TOTL.ZS?locations=ID
32
Yelwa, M., Okoroafor, D. O., & Awe, E. (2015). Analysis of the Relationship between
Inflation, Unemployment and Economic Growth in Nigeria: 1987-2012. Applied
Economics and Finance, 2(3), 102-109. doi:10.11114/aef.v2i3.943
33
THE IMPACT OF SATISFACTION AND IMPORTANCE
PERCEPTION ON THE APPROWEB’S TRIGGER DATA
FOLLOW-UP: AN EMPIRICAL STUDY BASED ON
ADMINISTRATIVE DATA
Galih Ardin
a Directorate General of Taxes of the Ministry of Finance of the Republic of Indonesia,
Email: galih.ardin@kemenkeu.go.id
ABSTRACT
Approweb has long been used by the Directorate General of Taxes as one of the reliable tools in
supervising taxpayer tax compliance. In its development, Approweb continues to undergo
improvements and refinements. The experience and perception of the Account Representatives are
an important component in Approweb's improvement efforts. This study aims to map the relationship
between account representatives' perception of the convenience and importance of Approweb to the
success rate of trigger data. Using the SEM (Structural Equation Modelling) method, we tried to
capture the dynamics of relationships between the variables studied. We found that the level of AR
satisfaction on Approweb Trigger Data has a positive impact on the follow up of those data. We also
found that, statistically, the level of importance perception on the trigger data has no effect on the
follow up of those data.
Keywords: taxation, tax system, compliance, tax data, tax structure
Approweb telah lama digunakan oleh Direktorat Jenderal Pajak sebagai salah satu alat yang handal
dalam mengawasi kepatuhan perpajakan wajib pajak. Dalam perkembangannya, Approweb terus
mengalami perbaikan dan penyempurnaan. Pengalaman dan persepsi Account Representative
merupakan komponen penting dalam upaya perbaikan Approweb. Kajian ini bertujuan untuk
memetakan hubungan antara persepsi account representative atas aspek kemudahan dan
kepentingan dari approweb terhadap success rate data pemicu. Dengan menggunakan metode SEM
(Structural Equation Modelling), penulis mencoba menangkap dinamika hubungan antar variable
yang diteliti. Berdasarkan hasil pengujian diketahui bahwa tingkat kepuasan AR terhadap aplikasi
Approweb berpegaruh positif terhadap success rate. Di sisi lain, secara statistik tingkat persepsi
kepentingan terhadap aplikasi Approweb tidak berpengaruh terhadap tingkat success rate data
pemicu.
Kata kunci: perpajakan, sistem perpajakan, kepatuhan, data perpajakan, struktur perpajakan
1. INTRODUCTION
1.1. Background
For a long time, Taxation Web-Based Profile Application or Approweb has been
known as one of the tools used by the Directorate General of Taxes (DGT) to supervise
the fulfillment of tax obligations from taxpayers. Since its launch in 2012, there have been
many changes, additions and improvements made to this application.
In October 2020, DGT has completed a survey on the satisfaction of the use of
Approweb by Account Representative as respondents. There are several interesting
facts revealed by the survey. One of them, it is known that 70% of the respondents stated
that it is necessary to improve Approweb’s data accuracy. Furthermore, based on this
survey, it’s also know that according to the user perspective, the accuracy of Trigger
Data has a higher level of importance rather than level of satisfaction (DGT, 2020).
On the other hand, based on statistical data on Approweb, it is known that the follow-
ed up trigger data of Approweb from 2016 to 2020 is no more than 16%. In another words,
34
we can say that the success rate of trigger data was under 16% in average. This
condition leads to the assumption that the low success rate of the trigger data is
influenced by the low perception of AR towards the satisfaction of the trigger data in
Approweb.
Basically, there have been several studies that review the use of Approweb. Santoso
and Setiawan (2017), for example, mentioned that performance expectations, business
expectations, social influences, perceptual security controls, pleasant perceptions
positively affect the intention to use Approweb. Furthermore, research conducted by
Lawata, Nasaruddin and Sari (2020) found that at the level of tax office and usefulness
has a positive and significant effect on Approweb trigger data. While, perspective of
easiness has a positive effect but insignificant effect on trigger data in the Approweb.
Unfortunately, there have been no specific studies discussing the effect of account
representative perceptions on approweb success rates at the national level.
Based on those aforementioned facts, the I am interested in examining the Account
Representative's perception of Approweb Trigger data on the success rate of those data.
In examining the hypothesis, I using the SEM (Structural Equation Modelling) method.
This study contributed to two levels. First, this study contributes on the knowledge
about how the SEM method is applied in the field of taxation. Second, this study also
contributes by providing policymakers with an overview of what factors affect the success
rate of trigger data in tax compliance supervision. Thus, policy makers can take
corrective actions on the Approweb application which will ultimately increase the
effectiveness and efficiency of Approweb as a tool used in tax compliance supervision
activities.
The study consisted of four main parts. The first part will discuss the introduction
and background of the research. The second part discusses the theoretical framework
and hypotheses developed. The third part will discuss the data and research
methodology used. The fourth part will talk about the discussion and results of the
research. While the fifth part will discuss conclusions and suggestions.

DEVELOPMENT
There are two research questions to be answered in this study as follows: First, what
are the characteristics and perceptions of AR to Approweb trigger data? Second, does
AR's perception of the satisfaction level, importance, level of benefit and accuracy of
trigger data affect the level of follow-up trigger data?
To answer those research questions, I employed the hypotheses as follows:
H0: The level of satisfaction, importance, level of benefit and accuracy of the trigger data
have no effect on the rate of disbursement of the trigger data.
H1: The level of satisfaction, importance, level of benefit and accuracy of the trigger data
affect the rate of disbursement of trigger data.
3. RESEARCH DATA AND METHODOLOGY

In conducting this study, there were two types of data used. First, survey data on the
experience of using Approweb, and second, statistical data of follow-up Approweb trigger
data. Meanwhile, in terms of methods, I used the SEM (Structural Equation Modelling)
method using the SPSS application in determining the relationship between approweb
user perceptions and success rates.
Hox and Bechger (1999) argued that SEM is a powerful tool that can combine
complex models with several latent variables. In addition, by using SEM researchers can
35
specifically confirm models that affect factor analysis, regression models, and complex
models.
There were seven variables used in this study. Six variables are independent
variables and one dependent variable. Independent variables were obtained through a
survey that has been conducted in 2020 which includes:
a) Variable perception of AR satisfaction with the benefits of Approweb,
b) Variable perception of AR satisfaction towards ease Approweb,
c) Variable perception of AR satisfaction to accuracy Approweb,
d) Variables of perceived importance of AR to the benefits of Approweb,
e) Variables of perceived importance of AR towards the ease of Approweb
f) Variables perception of AR importance to Approweb accuracy.
If we look at the variables above, then we will be able to find that there are basically
two large groups of independent variables in this study, namely the satisfaction variable
and the interest variable. Therefore, in this study, the writing team used the SEM
(structural equation modelling) approach because the variables 'Satisfaction' and
'Interest' are latent variables that cannot be observed directly.

4.1. History of Approweb
The emergence of the Approweb application within DGT cannot be separated from
the DGT's goal to create a Taxpayer profiling application that makes it easier for AR to
explore the potential of taxpayers. Through the Letter of the Director General of Taxes
Number S-142 / PJ / 2008, the Director General of Taxes for the first time, provided
instructions on creating a tax payers profile of in Ms. Excel application. This letter was
triggered by the uniformity regarding the form, method, and content of the Taxpayer
profile. Before the issuance of this letter there were various forms, methods and contents
of the Taxpayer profile developed by each work unit such as profiling using the Ms. Word,
Ms. Excel, Ms. Power Point and even Client-Server applications.
In its implementation, the application of the Director General of Taxes Letter Number
S-142 / PJ / 2008 experienced many obstacles, especially in the problem of taxpayer
data. To fill in the taxpayer profile in the form of Ms. Excel, AR must fill in various
applications such as Taxpayer Masterfile data, revenue data, reporting data, examination
data and many others.
To overcome this problems, an initiative emerged from West Jakarta Medium Tax
Office to create a web-based Client-Server application that contains the data needed to
create taxpayer profiles classified by month, type of tax, target-realization and so on. The
ease of access and completeness of this new application feature make this profiling
application acceptable not only at tax office in West Jakarta Regional Office but also at
other tax office. In the next period, this application was named by Approweb which stands
for Web-Based Profile Tax Application.
June 2009, DGT issued a Decree of the Director General of Taxes Number KEP-71
/ PJ / 2009 which contained the formation of a team to build a team for the design and
system of Mapping, Profiling, Benchmarking and Data Exchange Applications. This
decree appointed all parties who involved in the development of Approweb in a taxpayer
profiling team which the main task of continuing the development of Approweb.
August 2009, Approweb began to be tested in Regional Offices outside the West
Jakarta Regional Tax Office such as the South Jakarta Regional Tax Office, East Jakarta
Regional Tax Office, West Java Regional Tax Office I, Central Java Regional Tax Office
I, Central Java Regional Tax Office II, and DIY Regional Tax Office.
March 2010, the DGT issued a Letter of the Director General of Taxes Number SE-
41 / PJ / 2010, 2010 which an order to use Approweb for all tax office and regional tax
36
office as tools of monitoring taxpayers’ compliance. By the issuance of this letter,
Approweb was implemented nationwide.
Over time, the weaknesses of Approweb continue to be evaluated and corrected
periodically to support efforts to explore the potential of taxpayers. Weaknesses arise
mainly in terms of modules – application modules, networking, data availability, access
time and other weaknesses. To overcome these weaknesses, in 2011 an Approweb
improvement team was formed to maximize Approweb performance (DGT, 2013).
In 2012, improvements were made to Approweb including server centralization,
single login using SIKKA users and passwords, profile module improvements, data &
analysis module development, supervision modules, and reception modules, adding
comment notifications, revamping the AR activity monitoring menu, revamping access
rights and fixing other bugs (DGT, 2013). In addition, in 2012 a Letter of the Director
General of Taxes Number SE-01 / PJ / 2012 was also issued concerning the
Improvement of the Approweb Application as a Means of Making and Updating Taxpayer
Profiles.
In 2013 the process of developing and refining Approweb was carried out by
developing a download list menu and timeout messages, developing a verification menu,
developing an attribute mapping menu, developing a potential monitoring menu,
improving and developing an AR activity supervision menu, updating data feeding,
transferring data from Apportal to Approweb and synchronizing Approweb with other
application. The development of this potential surveillance menu has then evolved into
a trigger data surveillance module until now.
Supervision through this trigger data is then standardized with the Letter of the
Director General of Taxes Number SE-49 / PJ / 2016 concerning Taxpayer Supervision
through an Information System which contains the duties and functions of each tax office,
Regional Office and DGT Head Office in carrying out taxpayer supervision.
According to Diamendia and Setiowati (2021), approweb was built based on internal
and external input data of DGT. Internal input data includes data contained in annual tax
returns, period tax returns, financial statements submitted by tax payers which are then
processed and juxtaposed with external data such as banking data, insurance and so on.
4.2. Trigger Data

Based on the Letter of the Director General of Taxes Number SE-49 / PJ / 2016, it
is regulated that the Trigger Data is an overview of the overall results of data pairing that
indicates the non-compliance of taxpayers both formally and materially. When it was first
launched in 2016, trigger data was only limited to matching asset data, liability data,
capital data, income or business circulation data and tax credit data contained in the
DGT information system.

The survey on the use of Approweb was conducted by DGT from October 1 to 4,
2020. The survey request email was sent to 8,696 respondents and the number of
incoming responses was 4,917 responses. In other words, the incoming response was
56.54% of the total emails sent.
In terms of work units, this survey was sent to respondents who spread across 352
Tax Offices both strategic and regional taxpayers. Meanwhile, based on the percentage
of respondents who filled out the survey compared to the number of respondent in each
regional tax office, it was found that the average survey level was above 50% (DGT,
2020).
37
5.1. Perceptions of User Satisfaction Regarding the Benefits, Ease and
Accuracy of Trigger Data
Based on the survey, it is known that in general, respondents are satisfied with the
aspects of benefits, convenience and accuracy of trigger data. However, on the aspect
of accuracy, more respondents expressed neutrality than satisfaction with the accuracy
of the trigger data. This facts show that the accuracy aspect needs serious attention if
you want to increase the use of Approweb.
In addition, on the benefit aspect, more respondents expressed extreme
dissatisfaction than respondents expressed great satisfaction. This indicates that many
respondents still use applications other than Approweb in order to extract potential trigger
data.
Based on the perception of the level of importance, it is known that in general,
respondents argue that trigger data is important in terms of benefits, convenience and
accuracy. In fact, some respondents argue that these three aspects are very important
in the trigger data. There are only less than 2% of respondents who state that the aspects
of benefit, convenience and accuracy are not important and less than 1% of respondents
state that these three aspects are very unimportant (DGT, 2020).
Based on the perception of the level of importance, it is known that in general,
respondents argue that trigger data is important in terms of benefits, convenience and
accuracy. In fact, some respondents think that these three aspects are very important in
the trigger data. There were only less than 2% of respondents who stated that the
aspects of benefit, convenience and accuracy were not important and less than 1% of
respondents stated that the three aspect were very unimportant (DGT, 2020).
5.2. 2020 Trigger Data Overview

Based on Approweb data, it is known that the number of trigger data in the period
2016 to 2020 was 19 million data with a value of 7,249 trillion rupiah. Furthermore, it is
known that the largest amount of trigger data is in 2019 with a total of 7 million trigger
data and with a value of 1,965 trillion rupiah.
Based on the data followed-up, the trigger data can be dividen into three parts,
namely: the trigger data has not been followed up, the trigger data is followed up and the
trigger data is followed up by the taxpayer. Based on the numbers data followed-up, in
the period from 2016 to 2020, the number of trigger data that has been followed up is 1.9
million data from a total of 19.44 million trigger data.
In addition, based on Approweb data, it is also known that during the period 2016 to
2020, the number of trigger data followed up by taxpayers was 2.16 million data and
there were still 15.34 million data that had not been followed up with the following details:
If we compare the value of the followed-up trigger data with the total trigger data, the
value of the trigger data that has been followed-up on each year is less than 16%. In fact,
the value of trigger data that not followed-up by taxpayers averages less than 3% as
illustrated in the following chart:
38
Graph 5: Follow-up Trigger Data
100%
80%
60%
40%
20%
0%
2016 2017 2018 2019 2020
Belum ditindaklanjuti Ditindaklanjuti Ditindaklanjuti oleh WP
Source: DJP, 2020.
5.3. Relationship between Trigger Data and Respondent Perceptions

5.3.1 Variables used
As mentioned before, there’re six variables employed in this research. Since the six
variables cannot be observed directly, the authors use the Satisfaction and Importance
variables as the proxy for the six independent variables mentioned earlier. In addition,
the writing team also used the average satisfaction and importance of each category in
each regional office to calculate the relationship between variables.
On the other hand, the dependent variables used in this study are the 2020 follow-
up variables which are the percentage of follow-up levels of trigger data by each Regional
Office in 2020. In this study, the writing team used 34 populations which represented the
level of satisfaction and importance as well as the percentage of trigger data followed up
by 34 DGT Regional Offices. The test summary is as follows:
Tabel 5: Statistik Variabel Dependent And Independent
Source: Author-Processed (2021)
Based on the table above, we can see that each variable has a population of 34 with
the mean, median and percentiles as above. Thus, when described with a histogram, the
population of each variable is as follows:
39
Graph 6: Average Variable Distribution of Satisfaction – Benefits
Chart 7. Average Variable Distribution – Average Satisfaction - Ease
Graph 8: Average Variable Distribution of Satisfaction – Accuracy
Based on the histogram above we can see that the satisfaction variable has a mean
between 0,6 to 0,7. This fact shows that respondent in each Regional Office has a neutral
perception of the benefits, convenience and accuracy of Approweb trigger data.
On the other hand, based on the results of observations, it is known that respondent
in each Regional Office perceive that the trigger data contained in Approweb is important
40
both in terms of benefits, convenience and accuracy. This is evidenced by the mean
value which is in the range of 0,80 to 0,82 with the following distribution:
Graph 9: Average Variable Distribution of Importance – Benefits
Graph 10: Average Variable Distribution of Importance - Ease
Graph 11: Average Variable Distribution of Importance – Accuracy
The description of the histogram of the distribution of variable dependents in this

study, namely variable followed up in 2020, is as follows:
41
Graph 12: 2020 Actionable Variable Distribution
Based on the histogram above we can see that in 2020, the followed-up trigger data
variable has a mean between 0,06. This fact shows that the realization of follow-up
trigger data is still quite low. Therefore, it takes DGT's efforts consistently to improve
Approweb trigger data in terms of the benefits, convenience and accuracy of DGT in
order to increase the success rate of disbursement of trigger data.
5.3.2 Relationship of AR Satisfaction to Data Pemicu Approweb With Prosentase

Penyelesaian Data trigger
Based on the testing using the SEM (Structural Equation Modelling) approach, it is
known that the level of AR satisfaction with the Approweb application has a positive
impact on the value of variabel data that was followed up in 2020. In detail, if the
satisfaction rate rises by 1 point, then the follow-up data variable will increase by 0,977
points (P value = 0,08).
This finding is in line with research conducted by Davis, Bagozzi, & Warshaw (1989)
which states that perceived usefulness and perceived ease of use in information system
applications have a significant impact on the interest in using these applications. The
test details are as follows:
Table 2: Variable Test Results

Estimate S.E C.R PLabel
AverageofSatisfaction_Usefulness  Satisfaction 1
AverageofSatisfaction_Easyofuse  Satisfaction 0,12 0,11 9,956 ***
AverageofSatisfaction_Accuracy  Satisfaction 0,982 0,12 8,418 ***
AverageofImportance_Usefulness  Importance 0,844 0,09 9,102 ***
AverageofImportance_Easyofuse  Importance 0,844 0,08 11,32 ***
AverageofImportance_Accuracy  Importance 1
Successrate2020  Satisfaction 0,977 0,55 1,77 0,08
Successrate2020  Importance -0,573 0,69 -0,834 0,4
Based on the above table we may see that the satisfaction variable has a p-value of
0,08 and a coefficient of 0,977. This figure shows that the latent variable of "Satisfaction"
statistically had a significant influence on the trigger data variables that were followed up
in 2020.
This finding is in line with research conducted by Venkatesh and Davis (2000) which
states that both the influence of social processes and instrumental cognitive processes
such as perceived ease of use significantly affects user acceptance in an information
system application.
42
On the other hand, the latent variable of "Importance" has a p-value of 0,4 and a
coefficient of -0,573. This fact shows that statistically the importance perception of user
on trigger data (importance variable) has no effect on the follow-up rate of followed-up
data (2020 Follow-up variable).
Figure 1: Variable Test Results
Based on the diagram above, it is known that the variables satisfaction-benefit,

satisfaction-ease and satisfaction-accuracy have almost uniform coefficients (between
0,98 to 1,12). This finding shows that simultaneously the satisfaction aspect of the
benefits, convenience and accuracy of Approweb trigger data has a significant effect on
the realization of trigger data followed up by AR.
This finding is also in line with research conducted by Alharbi and Drew (2014) which
states that perceived ease of use and perceived usefulness of application users have an
effect both directly and indirectly on the use of Learning Management System (LMS).
Another study conducted by Venkatesh and Morris (2000) also revealed that
perceived ease of use has a positive and significant effect on perceived usefulness.
Therefore, if DGT wants to increase the realization of follow-up trigger data, DGT must
consistently improve Approweb trigger data in terms of benefits, convenience and
accuracy.
6. CONCLUSIONS AND SUGGESTIONS

6.1 Conclusion
Based on the description above, the conclusion that can be drawn is the aspects of
satisfaction to the benefits, convenience and accuracy of trigger data have a significant
effect on the realization of trigger data followed up by user.
In addition, based on our testing, it was also concluded that the increase in user’s
satisfaction on the Approweb application was positively in line with the data that was
followed up. An increase in satisfaction of 1 point will be responded to by an increase in
data followed up by 0,977 points. Meanwhile, variable importance has no effect on the
follow-up rate of trigger data.
43
6.2 Suggestion
To increase the value and amount of follow-up trigger data, this study recommends
that steps be taken to improve AR satisfaction in using trigger data, for example by
improving accuracy by conducting a more thorough cleansing and testing process of
data before the trigger data is derived.
Furthermore, DGT can also simplify the way trigger data is used, among others, by
grouping similar trigger data. It groups trigger data by cost, turnover, and more. Trigger
data based on turnover that can be grouped include: Turnover Based on PEB, Turnover
vs DPP DN Tax Credit, Equalization of Income Tax Turnover vs Submission in the VAT
Period Tax Return, and Bonded Area WP Turnover vs Data of the Directorate General
of Customs and Excise.

This study attempts to scientifically explain the relationship between the correlation
of AR satisfaction levels to the success rate of Approweb trigger data disbursement.
However, along with the development of taxpayer compliance supervision activities
through the supervisory priority list (DPP) in early 2022, it is necessary to conduct further
research related to the use of trigger data in the DPP.
Daftar Bibliography
Alharbi, S., & Drew, S. (2014). Using the Technology Acceptance Model in
Understanding Academics’ Behavioural Intention to Use Learning Management
Systems. International Journal of Advanced Computer Science and Applications
(IJACSA), 5(1), 143 - 155.
Davis, F. D., Bagozzi, R. P., & Warshaw, P. R. (1989). User acceptance of computer
technology: A comparison of two theoretical models. Management Science,
35(8), 982-1003.
Diamendia, T., & Setyowati, M. S. (2021). Machine Learning-Based Compliance Risk
Management Policy Analysis at the Directorate General of Taxes. Indonesian
Treasury Review, 6(3), 289-298.
DGT. (2013). Approweb Data Module-Based Tax Potential Excavation Guide and
Approweb Analysis. Jakarta: Directorate General of Taxes.
Hox, J., & Bechger, T. (1999). An Introduction to Structural Equation Modeling. Family
Science Review, 11, 354-373.
Lawata, R. S., Nasaruddin, F., & Sari, R. (2020). Analysis of Perceived Usability, Ease
and Complexity of Trigger Data in Approweb at the Primary Tax Service Office.
Journal of Accounting and Finance, 56-69.
doi:https://doi.org/10.52103/jaf.v1i1.85
Santoso, W. N., & Setiawan, D. (2017). Factors affecting the intention to use approweb
by the account representative of the Directorate General of Taxes. Journal of
Accounting, 21(2), 232-252.
Venkatesh, V., & Davis, F. D. (2000). A Theoretical Extension of the Technology
Acceptance Model: Four Longitudinal Field Studies. Management Science,
46(2), 186-204.
Venkatesh, V., & Morris, M. G. (2000). Why Don't Men Ever Stop to Ask for Directions?
Gender, Social Influence, and Their Role in Technology Acceptance and Usage
Behavior. MIS Quarterly, 24(1), 115-139.
44
A SUSTAINABLE APPROACH ON TAX AUDIT
INSPECTION: HOW TAX AGGRESSIVE STRATEGIES
CAN BE DETECTED FROM FEE OF INDEPENDENT
AUDITOR
Rachmad Miftachul Arifuddin

a Sekretariat Jenderal, Jakarta, Indonesia Email: rachmad.miftachul@kemenkeu.go.id
ABSTRACT
Problem with self-reporting data is that taxpayer can be mindful in declaring their tax return. In some
case they may intentionally select information on financial statement for taxation purpose. Tax audit
is an important tool for this problem, but the coverage ratio on corporate enterprise remains low.
Increasing the coverage, might not the only solution to enforce the compliance, since tax audit require
lots of resources. DGT should consider that some variable may influence taxpayer’s behaviour, then
use it on tax audit program. This paper will discuss that fee of independent auditor might affect the
tendency in performing tax aggressive strategies. The analysis observes 22.519 data of audited
financial reports from enterprises in Indonesia. The paper will use Instrument Variable to come up with
endogeneity problem in the dataset. The finding indicates that audit fee has negative impact on tax
aggressive strategies. Thus, data about audit fee could indicate the behaviour of an audited enterprise.
The result from this IV estimator can be used as an insight by DGT in determining their tax audit
strategy.
Keywords: tax audits, tax avoidance, tax compliance, tax enforcement, tax risk
Masalah dengan self-reporting data bahwa wajib pajak (WP) dapat lebih selektif dalam melaporkan
informasi dalam SPT. Pada beberapa kasus WP mungkin sengaja memilah-milah informasi apa yang
akan diberikan, terkait kepentingan perpajakan. Pemeriksaan pajak merupakan salah satu solusi
untuk masalah ini, namun rasio cakupan pemeriksaan pajak pada WP perusahaan masih rendah.
Meningkatkan cakupan pemeriksaan pajak, mungkin bukanlah solusi yang effektif dalam mengatasi
masalah laporan SPT ini, karena pemeriksaan pajak membutuhkan banyak sumber daya. DJP dapat
mempertimbangkan adanya perilaku-perilaku wajib pajak, yang dapat diukur melalui beberapa
variable tertentu. Kemudian, DPJ dapat memanfaatkan fakta-fakta tersebut dalam menentukan
bagaimana proses pemeriksaan pajak. Paper ini akan membahas bahwa fee yang dibayarkan untuk
jasa audit independent, mungkin mempengaruhi kecenderungan perusahaan dalam melakukan
strategi perpajakan yang lebih agresif. Paper ini menggunakan 22.519 data laporan keuangan
auditan dari perusahaan-perusahaan di Indonesia. Dalam penelitian ini, kami menggunakan teknik
Variabel Instrumen (IV) untuk mengatasi masalah endogenitas yang mungkin ada pada model. Hasil
analisis menunjukkan bahwa biaya audit memiliki dampak negatif terhadap strategi perpajakan yang
lebih agresif. Dengan demikian, data tentang audit fee dapat mengindikasikan perilaku perusahaan
yang menjadi sample pemeriksaan pajak. Sehingga, hasil estimator IV ini, dapat digunakan sebagai
masukan bagi DJP dalam menentukan strategi yang lebih efektif dalam melakukan pemeriksaan
pajaknya.
Kata kunci: audit perpajakan, penghindaran pajak, kepatuhan pajak, penegakan hukum pajak, risiko perpajakan
1. INTRODUCTION
1.1. Tax Collection
Improvement in tax collection has become interesting discussion on taxation topic
in Indonesia. The self-assessment system allows taxpayers to independently calculates
and report their tax obligation. Although there has been penalty for un-declared tax return,
in 2020, there was only 77.63% of taxpayer who reported their tax return. Moreover, the
compliance among enterprise entities, was just over 60% (DGT, 2021). In terms of tax
45
revenue to GDP ratio, Indonesian score is relatively low within Asian region. Figure 1
shows the ratio of tax revenue to GDP per capita, according to world bank data (2022).
In 2020, Indonesian score was around 8.3% in 2020, which is lower than the average
score among Asia-Pacific nation.
As a response, DGT has tried to do tax audit even though the coverage remains
relatively low. According to the 2020 report, DGT said that audit coverage ratio is around
2.14% for the corporate enterprise, while about 1.11% for the individual. Although there
has been an increase in audit coverage percentage since 2017, but the increases were
insignificant. Therefore, instead of increasing the coverage, the audit program can also
be strengthened to increase its impact on tax compliance.
By using data of independent public accountant (PA), this paper aims to give
contribution for the discussion on effective tax audit. DGT should consider performing a
risk profiling for the taxpayers. Then, DGT may design audit strategies for each group on
those risk profiles. This strategy can be implemented more effectively, if the program has
considered some variable that can be used to detect risk of taxpayer's compliance.
Indonesian Tax Revenue on GDP Percapita (PPP)

40
DNK
30
Tax Revenue (%ofGDP)
NZL SWE
GRC ITA GBR
FRA LUX
AUT
NLD
ISR AUS
ISL MAC
BEL
NOR
20
FIN
TUR
CHL SMR IRL
THA
MEX KOR
CAN SGP
MYS DEU NAC
10
USA CHE
IDN
0
20000 40000 60000 80000 100000 120000

GDP PPP
Figure 2 Indonesian Tax Revenue on GDP Per capita (PPP) in 2020

Result from this paper might be used by authority to add a variable to enhance their risk
profiling model. This paper tries to provide evidence that taxpayer’s decision in choosing
PA firm, could be used as indicator to detect tendency in performing tax eviction.
1.2. Problem with self-reporting data

Some economic literatures look from taxpayer’s perspective, in determining an
optimum combination between cost and benefit from tax compliance. Under classical
assumption, Andreoni, Erard, and Feinstein (1998) argues that taxpayer will not pay
taxes if the costs of compliance exceed the net benefits gained from aggressive tax
strategies. Discussion in this topic has become interesting when Alm, Bahl, and Murray
(as cited in Vellutini, 20111) use game theory to describe interaction between taxpayer
and regulator. A significant result, brought by these models, is that the probability of
regulator to find tax avoidance practice is not exogenous. Or in another words, there are
variable/parameters that influence the compliance, and can be used by regulator to
detect taxpayer’s compliances.
Interestingly, Vellutini (2011) has found that the probability of detection is
depends on by taxpayer’s information when filing the tax returns. Although he does not
mention specific parameter, his study inspires many taxpayers to be more mindful in
filling their tax return (SPT). It is no doubt that Information from SPT, remains important
for taxpayer’s risk profile. However, DGT may need additional data/resources other than
information, declared by taxpayers. By considering more various variable, the risk
46
profiling model could become more robust, so it can be used by DGT to optimally allocate
their limited resources in performing tax audit.
A risk-scoring model, that use variable other than SPT, may reduce taxpayer’s
ability to consistently underreport their income and avoid audits. Thus, our analysis tries
to use an independent data from auditor about their audit fee. This risk-profiling-audit
strategy is not likely to consume more resources, than other approaches, such as
random audits (Alm, Blackwell, and McKee 2004).
2. LITERATURE AND FORMULATION

Some studies have tried to look audit quality as a variable in determining tax
behaviour. The idea is that the presence of independent auditor could detect the
misstatement on financial report, caused by agency problem. The better audit quality,
the audit process could be more robust. Thus, the presence of qualified auditor my
reduce taxpayer’s tendency to manipulate financial report for their tax purpose.
Jihene and Moez (2019), found evidence that audit quality reduces tax avoidance.
They argue that managers are less motivated to engage in aggressive tax when the
independent auditor is presence. The reasoning is that performing aggressive tax
strategies could bear damaging consequences if auditor detects it. By using a sample of
Norwegian firms from 2000 to 2014, Langli and Willekens (2017) has also found a similar
result. They conclude that high audit quality improves the credibility of financial
information. Company would maintain this credibility for financial purpose. Therefore,
performing aggressive tax strategies could potentially damage this reputation, then,
offsetting the economic benefit of tax avoidance. Kanagaretnam et al. (2016) also
investigated this topic, they found that this negative correlation would be more prevalent
in countries that has stronger investor protection, higher litigation risk, and better audit
environment.
On another hand, Pratiwi, et al (2019) found that audit quality has no impact on tax
aggressiveness. She suspects that all practitioners in assurance industry have pass
competency test, made by association. It means that although the audit quality may differ
in each firm, but they have met a standardised qualification. Some research on specified
industry has also found a similar result. Subagiastra, et al (2017), studied Indonesian
manufacture companies, listed on stock exchange during 2011-2014. Their statistical
evidence says that there is no significant correlation between audit quality and tax
compliance. A similar result has also found in Indonesian companies, listed on
Indonesian SRI index.
There is an ongoing discussion about the link between tax compliance and audit
quality. Some argue that external audit improves the quality of information in financial
statement. The better audit quality means that the information in financial report is more
reliable for investor. Therefore, when a qualified external auditor is presence, performing
aggressive tax strategy could harm company’s reputation in disclosing its financial
performance. Meanwhile, others argue that measuring audit quality on assurance
industry is not relevant, because all practitioners have met a standardised qualification.
Therefore, some research argues that there is no significant impact of audit quality on
tax compliance behaviour.
This paper aims to contribute evidence in this ongoing discussion. Most of research
that doubt the audit quality impact, use a similar approach to proximate audit quality.
Their models define audit quality as a dummy variable of financial reports audited by the
biggest 4 of Indonesian auditor office, and otherwise. This paper will choose Indonesian
company as an observation but will use another proxy in representing audit quality.
In addition, although all Indonesian auditor office have met a standardised
qualification, I expect that qualification is not only factor in determining audit quality.
Infrastructure has also determined audit quality. For example, international bank usually
47
uses automatic system to operate most of their transaction. Auditor will use IT to perform
independent audit for this client. When an advanced technology is used, the audit fee
will be more expensive. Or in another word, the more expensive audit fee, the better its
audit quality, the lower opportunity to manipulate the financial report. Therefore, this
paper expects that audit quality may reduce tax avoidance.
3. DATA SOURCE AND MODEL CONSTRUCTION

This paper use information, reported by audit firms to Ministry of Finance. Under
ministry of Act no 5, year 2011, PA firm are required submitting its financial to the Ministry.
By using this database, we could analyse the financial performance of each auditor office,
as well as how much the audit fee, obtained from each client. This paper uses PA firm’s
reports for the financial year 2021. Our analysis observes 22.519 data of audited financial
reports from enterprises in Indonesia. A company may use more than one audit services
in one financial year, due to their necessity of interim audit report, restatement, or other
purpose.
By using that information, this paper use IV technique (2SLS) and defining model into
the structural equation and the reduced from.
𝐸𝑇𝑅𝑖 = 𝛼 + 𝛽𝜏𝑖 + 𝛾𝑪𝒊 + 𝜀𝑖 ; 𝐸(𝜏𝑖 , 𝑪𝒊 |𝜀𝑖 ) = 𝟎

𝜏𝑖 = 𝛿 + 𝛾𝑍𝑖 + 𝑣𝑖 ; 𝐸(𝑍𝑖 |𝑣𝑖 ) = 𝟎
The 2SLS estimator, implemented in this model, is referred to Gujarati (2003, pp. 679)
and Greene (2012, pp. 271).
a. Effective earning rate (𝐸𝑇𝑅𝑖 )

This paper analyses whether the audit quality has an impact on tax compliance
behaviour. By following Hanlon and Heitzman (2009), this analysis uses 𝐸𝑇𝑅𝑖 as
parameter/metrics to represent tax compliance behaviour. Effective tax rate (ETR) is
measured by dividing tax liability and earning before tax. The rate will capture how much
the average tax payable per dollar of income. If the company use a more aggressive tax
strategies, the rate would be lower.
Meanwhile, they also argue that the effective interest rate (ETR) would not capture
the avoidance. Especially when company do not have a financial disclosure constraint.
For example, a small family enterprise that do not require to declare its financial
performance to third party. In this case, they could lower the taxable income, without
concerning “outside” factor. In this paper, we use audited financial report. It means that
a company do not simply lower their income, as it might negatively impact their
performance to the stakeholders.
In the model, we transform the ETR data into natural logarithmic function. This
transformation is caused by the data, as it has many small scores, but it also has data
with very big score.
b. Audit Fee (𝜏𝑖 )

In the structural equation model, 𝜏𝑖 represent the audit fee. As an economic agent, we
assume that PA main purpose is for economic incentive. PA will always optimise their
resources, subjected to budget constraint. Their procedures will be determined by their
fee. If the fee is acceptable, their service will comply with the standards. But when the
fee is satisfied, they may provide service beyond standards. Due to data availability, this
paper will use audit fee in 2021, as parameter audit quality. Similar to (𝐸𝑇𝑅𝑖 ), we also
transform the fee into natural logarithmic function.
c. Variable 𝑪𝒊 in the structural model, represents matrix of covariate variables in the

model. The covariate is used to reduce the standards errors associated with the
48
treatment effects. In this paper, the covariates are, such as: return on asset, asset to
equity ratio, and revenue-to asset ratio. Information about firm’s efficiency is collected
from audited financial reports. The 𝜺𝒊 and 𝑣𝑖 , represent error in the main model and
reduced form, respectively.
The 2SLS technique, requires instrumental variable ( 𝑧𝑖 ) to influence the tax
compliance metric (𝐸𝑇𝑅𝑖 ), only via audit quality variable 𝜏𝑖 . Then, the instrument model
(reduced from) will be:
𝜏𝑖 = 𝛿 + 𝛾𝑍𝑖 + 𝑣𝑖 ; 𝐸(𝑍𝑖 |𝑣𝑖 ) = 𝟎
This paper use PA firm’s Income as instrument variable (𝑧𝑖 ). The reasoning is that
audit quality is not only determined by PA’s competency, but also by their resources in
performing auditing. For instance, some big audit firm has IT and economic division to
help the auditor, dealing with IT usage or statistics. Auditor them self, may not be familiar
with IT, thus, the help from expertise would be essential to ensure the high-quality service.
Hiring this audit expert is not cheap, only PA Firm that has a high capital can perform
it. The high capital also means high maintenance, thus big PA’s firm need a higher
income to maintain this service. Therefore, this paper will look at PA’s resources (income),
as a proxy for the instrument. The more resources that auditor will use, the higher fee
that they will ask for. More detail explanation about the instrument, can be found in the
section of Exclusion restriction, bellow.
Instrumental Variables (IV) relaxes the un-confoundedness assumption, caused by
problems of heterogeneity such as selection bias, omission, and unobservable (latent)
variable. Let assume that the original model is
𝐸𝑇𝑅𝑖 = 𝛼 + 𝛽𝜏𝑖 + 𝛾𝑪𝒊 + 𝜀𝑖 ; 𝐸(𝜏𝑖 , 𝑪𝒊 |𝜀𝑖 ) = 𝟎
We have variable 𝜇𝑖 (𝐸(𝜇𝑖 |𝜏𝑖 ) ≠ 𝟎) that do not include to the original model. As (𝜇𝑖 )
is not included to the original model, the error 𝜀𝑖 in the equation will contain (𝜇𝑖 ), thus
𝜀𝑖 = 𝜇𝑖 + 𝑢𝑖 , where 𝑢𝑖 is the error term without 𝜇𝑖 . Therefore, the original model would
be
𝐸𝑇𝑅𝑖 = 𝛼 + 𝛽𝜏𝑖 + 𝛾𝑪𝒊 + 𝜇𝑖 + 𝑢𝑖
In this elaborated model, the required condition that 𝐸(𝜏𝑖 , 𝑪𝒊 |𝜀𝑖 ) = 0, can no longer be
hold because the presence of this latent variable, makes 𝐸(𝜇𝑖 |𝜏𝑖 ) ≠ 𝟎. Or in another
words, the correlation between latent variable and variable of interest (in error term),
violates the condition requiring no correlation between error and variable of interest.
Therefore, ordinary linier regression would produce not un-biased estimator.
By using the data, this paper also shows the evidence that OLS estimator is different
with the literature review. To anticipate it, we use IV instead of ordinary linier regression.
Greene (2012, pp. 265) provides mathematical explanation on how IV technique (2SLS)
could provide an unbiased estimation, even if the heterogeneity is presence.
4. Robustness
However, Instrument Variable technique requires two main conditions: the strong
instrument and the exclusion restriction. Although the former can be tested by statistical
tools, the latter might be fundamentally untestable. Therefore, for condition of exclusion
restriction, this paper will try to explain it, narratively.
a. Instrument validity
This condition requires that our instrument need to be correlated with the variable of
interest, where 𝐸(𝜏𝑖 |𝑧𝑖 ) ≠ 𝟎. To test whether the instrument is strong or weak, we use (1)
F-test of excluded instrument on the first stage regression, and (2) Endogeneity test of
endogenous regressors. Figure 2 shows F statistic resulted from our model:
49
Estimates efficient for homoskedasticity only
lnfee Coef. Std. Err.
Statistics robust to heteroskedasticity
t P>|t| [95% Conf. Interval]
lnfinc .2814515 .0036512 77.08 0.000

Number of obs = .2742945
11686 .2886084
revenuetasset .0000852 .000502 0.17
F( 5,0.865
11680) =-.0008987
4.72 .0010692
roa -.0000248 .0026835 Prob 0.993
-0.01 > F =-.0052849
0.0003 .0052353
Total (centered) SS
lntaset =.2247264
24696.41232
.0053312 Centered
42.15 R2
0.000 = .2142764
-0.0101 .2351764
Total (uncentered) SS
assettoequityratio =5.37e-13
63832.12696
4.07e-14 Uncentered
13.18 0.000 R2 = 4.57e-13
0.6092 6.16e-13
Residual SS = 24945.39997 Root MSE = 1.461
_cons 5.387179 .1122432 48.00 0.000 5.167164 5.607195
F test of excluded instruments:

Robust
F( 1, 11680) = 5942.05
lnetr Coef. Std. Err. z P>|z| [95% Conf. Interval]
Prob > F = 0.0000
Sanderson-Windmeijer
lnfee multivariate
-.0794991 F test of-3.46
.0229605 excluded instruments:
0.001 -.1245009 -.0344974
F( revenuetasset
1, 11680) = 5942.05
.0011072 .0007373 1.50 0.133 -.0003379 .0025524
Prob > F roa
= -.0058114
0.0000 .0039415 -1.47 0.140 -.0135366 .0019137
lntaset .0524627 .0115237 4.55 0.000 .0298767 .0750488
assettoequityratio 3.98e-12 5.84e-14 68.21 0.000 3.87e-12 4.10e-12
Figure 3- Ftest of Excluded Instrument
_cons -1.747488 .2176977 -8.03 0.000 -2.174168 -1.320809
Summary results for first-stage regressions

Underidentification test (Kleibergen-Paap rk LM statistic): 2388.118
To indicate that our instrument is strong, weChi-sq(1) need F statistic
P-val = (from the first stage
0.0000
regression) that greater than 10. The regression result

(Underid)
Weak identification test (Cragg-Donald Wald F statistic):
shows that F
(Weakstatistic
id)
7091.857
in our model
is 5942, meaning
Variablethat our
| F( instrument
1, 11680) is strong.
P-val | SW
(Kleibergen-Paap rk Chi-sq( 1) P-val | SW F(
Wald F statistic): 1, 11680)
5942.051
Stock-Yogo weak
lnfee ID 5942.05
test critical values: 10% maximal
5945.10IV size 16.38
Another test is| endogeneity 0.0000
test.| Endogeneity 0.0000 |
15% maximal IV sizelooks whether
5942.05
8.96 the variable of
interest is endogenous
NB: first-stageto thestatistics
test instrument. The null IVhypothesis
20% maximal size
heteroskedasticity-robust is that
6.66 the treatment is
25% maximal IV size 5.53
exogenous, and weStock-Yogo
Source: need to(2005).
be able to reject
Reproduced it.
by permission.
Stock-Yogo weak ID F test critical values for single endogenous regressor:
NB: Critical values are for Cragg-Donald F statistic and i.i.d. errors.
Hansen J statistic (overidentification15%
testmaximal
of all IV size
instruments): 8.96
0.000
20% maximal(equation
IV size exactly identified)
6.66
-endog- option: 25% maximal IV size 5.53
Source: Stock-Yogo
Endogeneity test of (2005).
endogenousReproduced
regressors: by permission. 94.761
NB: Critical values are for i.i.d. errors only. Chi-sq(1) P-val = 0.0000
Regressors tested: lnfee
Underidentification test
Instrumented: lnfee
Ho: matrix of reduced form coefficients has rank=K1-1 (underidentified)
Included instruments: revenuetasset roa lntaset assettoequityratio
Ha: matrix has rank=K1 (identified)
Excluded instruments: lnfinc
Kleibergen-Paap rk LM statistic Chi-sq(1)=2388.12 P-val=0.0000
Figure 4- Endogeneity Test

Weak identification test
The result of this test show that we can reject this null hypothesis (as indicated by p-
Ho: equation is weakly identified
Cragg-Donald Wald F statistic 7091.86
value is less than 5%). Therefore, regarding to F-statistics and endogeneity
Kleibergen-Paap Wald rk F statistic 5942.05
test, we can
conclude that the instrument is not weak.
Stock-Yogo weak ID test critical values for K1=1 and L1=1:
b. Exclusion restriction 15% maximal IV size 8.96
Exclusion means that 𝑧 should not be included in the model of interest, and it
𝑖 20% maximal IV size 6.66
only appear in the instrument model. 𝑧 should influence 𝐸𝑇𝑅 only through 𝜏𝑖 . In another
𝑖
Source: Stock-Yogo (2005). Reproduced by permission. 𝑖
word, our instrument variable should not directly influence the dependent variable, the
NB: Critical values are for Cragg-Donald F statistic and i.i.d. errors.
influence must only through the variable of interest. Some studies use Sargan/Hansen
Weak-instrument-robust inference
test to examine
Teststhe exclusion
of joint restriction,
significance if there
of endogenous are more
regressors instruments
B1 in main equationthan endogenous
variables (Gujarati, 2003
Ho: B1=0 and pp. 713).conditions
orthogonality However, aresince
valid this paper only uses one instrument,
Anderson-Rubin Wald test F(1,11680)=
the Hansen test is not applicable. Thus, our
Anderson-Rubin Wald test
exclusion12.14
Chi-sq(1)= 12.15
P-val=0.0005
restriction only relies on qualitative
P-val=0.0005
explanation. Stock-Wright LM S statistic Chi-sq(1)= 13.35 P-val=0.0003
The exclusion restriction means that AP Firm’s income does not directly influence
NB: Underidentification, weak identification and weak-identification-robust
tax aggressive teststrategy (ETR).
statistics The audit firm income influence how good the audit
heteroskedasticity-robust
service. The high income could indicate that audit firm has more advance tools and
Number of observations N = 11686
resources, toNumber
perform audit service. In addition,
of regressors K =
higher-income-audit-firms
6
also maintain
their reputation;
Numberwhen audit firm
of endogenous found material
regressors K1 = misstatement,
1 they would hesitance to
give “unmodified opinion”.
Number of instruments
Number of excluded instruments
L =
L1 =
6
1
If taxpayer decide to acquire an independent audit from the higher-income-audit-
firm, taxpayers know that the audit firm will perform “high quality audit”. In this case, the
taxpayers would be less likely to perform tax aggressive strategy (by lowering their
income), because the “high quality audit” would expose/found any material misstatement
on financial report. In this case, modifying financial report is risky, as the AP would give
“bad” opinion on their audited financial statement.
To sump up, audit firm’s characteristic influences the taxpayer’s tendency, only
through the audit quality services, given by taxpayers. The tendency of taxpayers to
perform tax aggressive strategy, is influenced by the quality of audit, that is influenced
by the PA firm’s resources (income).
50
This paper has explained how ordinary linier regression may not provide unbiased
estimation when the endogeneity problem is presence.
lnetr lnetr
lnfee 0.097** -0.079**

(0.012) (0.023)
revenuetasset 0.000 0.001
(0.000) (0.001)
roa 0.000** -0.006
(0.000) (0.004)
lntaset -0.005 0.052**
(0.007) (0.012)
assettoequityratio 0.000** 0.000**
(0.000) (0.000)
_cons -3.417** -1.747**
(0.159) (0.218)
R2 0.01 -0.01
N 15,955 11,686
* p<0.05; ** p<0.01
Figure 5- Result from OLS (left) and 2SLS (right)

The estimation, derived from OLS, show positive (and significant) correlation between
audit fee and taxpayer’s tendency to perform aggressive strategy. If the audit fee in
increase by 1% percentage point, the taxpayer’s tendency would also increase by 0.097%
percentage point. However, this result is not supported by most literatures that discussed
in this topic. The literatures argues that the correlation between them should be
significantly negative (Jihene and Moiz, 2019; Langli and Willekens, 2017; and
Kanagaretnam et al., 2016) or it should be non-significantly correlated (Pratiwi, et al,
2019; Subagiastra, et al, 2016). Therefore, we suspect that this difference is caused by
un-confoundedness in the model.
Meanwhile when we apply instrument variable, the impact turns to be negative,
although it derives to a larger standards error. The more qualified audit service, taken by
taxpayers, they tend to be less likely to perform aggressive tax behaviour. This statistical
result is in-line with the literatures in the discussion.
In term of interpreting the result, IV estimator has a slightly different narrative. While
OLS estimator can be interpreted as average treatment effect (ATE)/Average Treatment
Effect on the Treated (ATT), the IV is only interpreted as local average treatment effect
(LATE).
Here, the explanation of ATT from OLS estimation. Let assume that there are two
perfectly identical companies PT A and PT B. PT A decide to acquire the qualified audit
service 𝐸(𝑌𝒂 |𝜏𝑖 = 1) while PT B is not 𝐸(𝑌𝒃 |𝜏𝑖 = 0). By applying OLS, the estimator
would be calculated as:
𝛽 = 𝐸(𝑌𝑎 |𝜏𝑖 = 1) − 𝐸(𝑌𝑏 |𝜏𝑖 = 0)
= 𝐸(𝑌𝑎 |𝜏𝑖 = 1) − 𝐸(𝑌𝑏 |𝜏𝑖 = 1) + 𝐸(𝑌𝑏 |𝜏𝑖 = 1) − 𝐸(𝑌𝑏 |𝜏𝑖 = 0)
= 𝐸(𝑌𝑎 − 𝑌𝑏 |𝜏𝑖 = 1) + 𝐸(𝑌𝑏 |𝜏𝑖 = 1) − 𝐸(𝑌𝑏 |𝜏𝑖 = 0)
From the calculation above, the 𝐸(𝑌𝑎 − 𝑌𝑏 |𝜏𝑖 = 1), can be interpreted as Average
Treatment Effect on the Treated (ATT). Meanwhile, 𝐸(𝑌𝑏 |𝜏𝑖 = 1) − 𝐸(𝑌𝑏 |𝜏𝑖 = 0) reflects
type 1 selection bias, which can be assumed to be 0 as our data use the whole population
in 2021. Thus, in our model, OLS estimator can be interpreted as: if the audit fee in
51
increase by 1% percentage point, the taxpayer’s tendency would also increase by 0.097%
percentage point. This impact is exclusively presence on taxpayers that is randomly
selected from population that were audited in 2021.
As we explained above, OLS estimator might contain unobservable variable, that
affects the estimation result. The unobservable makes OLS estimation could not be
unbiased. Therefore, this paper will use IV to anticipate this heterogeneity problem. The
presence of unobservable, make IV estimation do not produce ATT, but at least it
remains representative on a localise/specific type of taxpayers. Thus, the interpretation
from IV estimation would be if the audit fee in increase by 1% percentage point, the
taxpayer’s tendency would also decrease by 0.079% percentage point. This impact is
exclusively presence on taxpayers that determine audit-firm’s size as a factor in selecting
independent auditor.
The result from this IV estimator can be used as an insight by DGT (directorate
general of taxes) in determining which taxpayers that may have a higher possibility to
perform an aggressive tax strategy. If ETR is determined as a taxpayer’s risk factor, then,
the impact of audit fee on the risk factor (for taxpayers that have audited financial report)
would be ETR*0.079AuditFee. By having the risk profiling, DGT could determine the best
approach of tax investigation program. Thus, the investigation can provide more efficient,
yet provide more benefit for DGT interest.
6. LIMITATION DAN FUTURE RESEARCH

Although IV estimator solve the endogeneity problem, but it has a higher standard of
error. There are treatments that can be used to minimise it. Research may combine IV
with matching estimators. Matching approach such as psmatch (propensity-score
matching) can be used to help in balancing the covariates (𝐶𝑖 ). Combination between
Instrument variable and matching estimator can help researcher to decrease the
standard error.
IV technique is applicable on ex-post data, covering both participant and non-
participant. This technique is suitable to be used in this paper, because it only needs one
year data/information about audited financial report. In this research there has a time-
constraint, that makes collecting and preparing the multiyear dataset become
inapplicable. Future research in this topic could also use the multi-years dataset; and
performing fixed effect to produce a more robust estimator. Both matching and fixed
effect can also be combined with IV to increase the robustness.
7. REFERENCES
Alm, J., Blackwell, C., & McKee, M. (2004). Audit Selection and Firm Compliance with
a Broad-Based Sales Tax, National Tax Journal, 57(2), 209-227, DOI:
10.17310/ntj.2004.2.04
Andreoni, J., Erard, B., & Feinstein, J. (1998). Tax Compliance. Journal of Economic
Literature, 36(2), 818–860. http://www.jstor.org/stable/2565123
Direktorat Jenderal Pajak. (2021). Laporan Tahunan 2020,
https://www.pajak.go.id/sites/default/files/2021-
10/Laporan%20Tahunan%20DJP%202020%20-%20Bahasa.pdf
Greene, W., H. (2012), Econometric Analysis Seventh Edition, Pearson Education
Limited
Gujarati D., N. (2003), Basic Econometric Fourth Edition, McGraw-Hill Higher
Education
Hanlon, M., Heitzman, S. (2009), A Review of Tax Research, in 2009 Journal of
Accounting and Economics Conference.
http://dx.doi.org/10.2139/ssrn.1476561
52
Jihene, F., & Moez, D. (2019). The Moderating Effect of Audit Quality on CEO
Compensation and Tax Avoidance: Evidence from Tunisian
Context. International Journal of Economics and Financial Issues, 9(1), 131–
139. Retrieved from
https://www.econjournals.com/index.php/ijefi/article/view/7355
Kanagaretnam, K., Lee, J., Lim, C., Y., Lobo, G. (2016), Relation between Auditor
Quality and Corporate Tax Aggressiveness: Implications of Cross-Country
Institutional Differences, Auditing A Journal of Practice & Theory, 35(4), 105–
135, https://doi.org/10.2308/ajpt-51417
Langli, J. C., Willekens, M. (2017), Tax Avoidance, Horizontal Agency Conflicts and
High-Quality Auditing in Private Firms, in 2017 Scandinavian Accounting
Research Conference. BI Norwegian Business School
OECD. (2022). Revenue Statistics in Asia and the Pacific 2021, Emerging
Challenges for the Asia-Pacific Region in the COVID-19 Era,
https://www.oecd.org/tax/tax-policy/revenue-statistics-in-asia-and-the-pacific-
5902c320-en.htm
Pratiwi, N., P., S., D., R., Subekti, I., Rahman, A., F. (2019), The Effect of Corporate
Governance and Audit Quality on Tax Aggressiveness with Family Ownership
as The Moderating Variable, International Journal of Business, Economics and
Law, 19 (5).
Subagiastra, K., Arizona, I., P., E., Mahaputra, I. N., Y., K., A. (2017). Pengaruh
Profitabilitas, Kepemilikan Keluarga, Dan Good Corporate Governance
Terhadap Penghindaran Pajak (Studi Pada Perusahaan Manufaktur Di Bursa
Efek Indonesia). Jurnal Ilmiah Akuntansi, 1 (2),
https://doi.org/10.23887/jia.v1i2.9994
Vellutini, C. (2011), Key Principles of Risk-Based Audits. In Khwaja, M., S., Awasthi,
R., Loeprick, J. (Eds.), Risk-Based Tax Audits: Approaches and Country
Experience, ( pp. 13-21). The World Bank
World Bank (2022), Tax Revenue (% of GDP),
https://data.worldbank.org/indicator/GC.TAX.TOTL.GD.ZS
World Bank (2022), “GDP per capita, PPP (current international $)”,
https://data.worldbank.org/indicator/NY.GDP.PCAP.PP.CD
53
EXIT TAX ADOPTION TO PROTECT INDONESIA’S TAX
BASE: ARE WE READY?
Septian Fachrizala, Iva Unnaiza Hanumb
a Directorate General of Taxes, Maastricht, The Netherlands.

Email: f.fachrizal@student.maastrichtuniversity.nl
b Directorate General of Taxes, South Tangerang, Indonesia.
Email: ivaluvu@gmail.com
ABSTRACT
Transfer of residencies, assets, and business arrangements out of jurisdiction borders have spiked
throughout the times, which allegedly could be driven by both sound economic rationale and tax-
saving motives. For the purpose of protecting the tax base without excessively imposing an
administrative burden on the national revenue system, numerous jurisdictions have adopted exit
charge/exit tax in the event of renouncement of tax residency and business restructuring. This paper
discusses the idea of exit charge adoption in Indonesia through a qualitative analysis by reference
to a comparative study of tax law in various jurisdictions. The study shows that the exit charge
adoption is feasible to be undertaken, taking into account the existing legal system, Indonesia’s tax
regulation, and the simplicity aspect within the taxation system, which Indonesia aims to enhance.
Keywords: exit charge, exit tax, change of tax residency, business restructuring
Pengalihan status wajib pajak, aset, dan pengaturan bisnis ke luar jurisdiksi semakin meningkat
seiring berjalannya waktu, yang diperkirakan didorong oleh pertimbangan praktek bisnis yang sehat
maupun oleh motif penghematan pajak. Untuk melindungi basis pajak tanpa secara eksesif
membenambah beban administratif bagi penerimaan negara, beberapa jurisdiksi telah mengadopsi
exit charge/exit tax atas perubahan status residensi perpajakan dan restrukturisasi bisnis. Penelitian
ini membahas gagasan adopsi exit charge/exit tax di Indonesia melalui analisis kualitatif
menggunakan studi komparatif atas hukum perpajakan di beberapa negara. Studi menunjukkan
bahwa adopsi exit charge mungkin untuk diakukan, dengan mempertimbangkan sistem hukum yang
telah ada, peraturan perjakan Indonesia, dan aspek kesederhanaan dalam sistem perpajakan yang
Indonesia tengah berusaha tingkatan.
Kata kunci: exit charge, exit tax, perubahan status residensi perpajakan, restrukturisasi usaha
1. INTRODUCTION
The world economy has come to an era where companies and individuals can
move across jurisdiction borders and seek new advantageous opportunities and
business options in other countries, resulting in the change of residency and assets
shifting through a business restructuring. The adverse of such migration is tax revenue
loss of capital gains that occur when the shifted assets are sold in the country to which
the residence is changed (Beer et al., 2018).
Although, from the economic point of view, business restructuring does not
necessarily mean a tax avoidance strategy, the tax-driven motive of a business
restructuring cannot be completely erased from the objectives of such an event since tax
liabilities are an element of considerable costs borne by the business. The global
business restructuring includes the transfer of significant function and risk of a business
or highly profitable intangible assets overseas. From a tax perspective, it causes income
transfer overseas and a mismatch between the place of value creation and the taxed
54
place. As a result, it is possible for multinational companies to reduce the total amount
of corporate tax burden after the restructuring (Tsuji, 2020).
In responding to the threat of revenue forgone caused by tax avoidance, many
countries have adopted exit tax or exit charge, which is an income tax paid by an
individual or business company when the person shifts his residence from one State to
another or when a cross-border transfer of an entity’s assets occurred (Kubicova, 2016).
The exit tax is commonly levied on the value of unrealized capital gain on the disposal
of assets, which is generally deemed to happen after the residence status is renounced
or after the transaction of transferring a business or asset takes place.
The policy rationale of charging an exit tax as an anti-avoidance measure has
been supported by numerous studies on the effectiveness of existing anti-avoidance
measures and the drawback of tax avoidance on tax revenue collection. For the most
recent years, a one percentage point lower corporate tax rate compared to other
countries will expand before-tax income by 1,5 per cent (Beer,et.al, 2018). The result is
higher than the previous study conducted by Heckemeyer, et.al, in 2017, which
suggested semi-elasticity of as much as 0,8 per cent. Such a study indicates that the
existing anti-avoidance measures has not yet been sufficient to address the tax
avoidance issue.
Despite the debatable view towards its fairness and complication issues, exit tax
has been adopted by many countries. The primary purpose of the adoption is to prevent
speculative, tax-motivated transfers of tax residency. Its proponents consider that the
application of exit tax is of the horizontal tax fairness between residents who stay and
those who leave their home country for good since both will pay capital gain tax
eventually and that the jurisdictions of which the tax residence is renounced substantially
owns the nexus for taxing the capital gain arising from asset disposal of the taxpayer.
On the contrary, some oppose the idea of exit charge due to the double taxation issue
that may arise.
This paper aims to present a qualitative study concerning exit tax/exit charge
implementation by adopting comparative studies of the tax law of jurisdictions which have
adopted exit tax within their taxation landscape, to finally proposes policy
recommendations for the adoption of exit tax on the changes of tax residency changes
and on the business restructuring. The term “exit tax” used in this paper is
interchangeable with the term “exit charge” considering jurisdictions’ best practices,
scholars and the applicability in the Indonesia tax system.
2. THEORETICAL FRAMEWORK
Following De Man, et.al (2011), it is important to make a recourse to the scheme
by Carvalho (2008) that is commonly used to demonstrate the features of particular
taxes and to illustrate the nexus between the facts and their legal consequences when
a tax is constructed.
Figure 1. The Notion of Tax Scheme
55
Source: Carvalho (2008) in De Man, et.al (2011)
 is the mathematical sign of equivalence;

 is the mathematical vector to ascertain that if Fs happens, then Tc should also happen;
 is the mathematical vector that represents that someone has the legal obligation to
pay a certain amount while another has the legal right to demand the payment;
Tn = Tax norm
Fs = Fattispecie, description of a fact
Mc = Material criterion,
Sc = Spatial criterion
Tp = Time period
v = Verb
c = Complement
Tc = Tax consequence
Pc = Personal criteria
As = Active subject, who can demand payment
Ps = Passive subject, who is obliged to pay the due amount
Qc = Quantifying criteria
Tb = Tax base
Tr = Tax rate
The tax norm is composed of a Fattispecie (Fc), description of a fact, which will
generate legal, tax consequences (Tc). A fact is described by material facts (Mc)
reflected by the verb (v) and the complement (c) made in a particular place (Sc) at a
certain time (Tp). A legal consequence (the Tc) will arise as a result of the fusion between
material, spatial, and time variables previously regulated, where the State (As) demands
payment from a person (Ps) a certain sum of money established according to the tax
base (Tb) and the tax rate (Tr), previously set down by the legislator.
With regard to the exit tax, according to the IBFD International Tax Glossary, the
term exit taxes presuppose a cross-border element as the emigration of companies or
individuals or a cross-border transaction (Hug, 2015). States often impose exit taxes in
order to ensure that an amount of previously untaxed income, such as unrealized gains,
which is attributable to their jurisdictions, is taxed before the taxation right is restricted or
lost due to international tax law (Schuch,et.al 2014). In the context of the business
restructuring of a multinational enterprise, exit charge generally refers to a payment
made to compensate for the removal of an asset belonging to an entity whose activity in
the business is being simplified or reduced (Henshall, et.al., 2012).
This contribution is conducted under the framework of legal research. A
comparative study of tax law in various jurisdictions is presented through qualitative
analysis to construct a design of law regime to apply in Indonesia. According to
Prof.Wiman, in Jiang,et.al (2014) mentioned potential advantages of comparative tax law
research: (1) to provide comparable solutions for legislators to make new rules, (2) to
interpret existing rules, (3) to harmonize tax law, (4) to enrich legal education. Wilson
(2017) mentioned the purpose of the study of comparative law is to make a practical
contribution to the local national system.
In this regard, after identifying the research problem and question, the authors
will first provide a description of the exit tax features in various jurisdictions that have exit
taxes regime in their tax systems. The exit tax profiles of countries are obtained by
56
performing a literature review from either relevant countries’ legislations or law scholar
articles. International documents provided by The Organisation for Economic Co-
operation and Development (OECD), such as OECD Model Tax Convention (MTC) and
OECD Transfer Pricing Guidelines (TPG), also play a role as an international reference
in this study. Henceforth, an in-depth comparative analysis is conducted in order to
answer the research question.
4. RESEARCH FINDINGS AND DISCUSSION

4.1. Exit Tax on the Change of Tax Residency
Indonesia’s income tax system prescribes the taxation of a person, comprising
individual and business entity, based on their residency status. As most countries around
the globe do, Indonesia applies worldwide taxation to its resident taxpayers. In essence,
the migration of an Indonesian tax resident would give two tax consequences for either
departure or immigration states. While the tax base of the latter would broaden, the
former would lose its right to tax on the income earned by its taxpayer. In that sense,
when a taxpayer resident of Indonesia moved to another country, he/she would give up
their Indonesian tax residency; thus, Indonesia would not have nexus to levy taxes after
becoming a resident of another jurisdiction.
Further, tax motive is sometimes embedded within the cross-border movement
of residency. Even worse, there is also tax avoidance risk associated with this event.
One can migrate from a high-tax jurisdiction to a lower-tax jurisdiction and subsequently
change his/her residency for tax purposes. OECD is in the view that exit or departure tax
rules may prevent the avoidance of capital gains tax through a change of residence
before the realization of a treaty-exempt capital gain (Paragraph 69 Commentary on
Article 1 OECD Model Tax Convention 2017).
The nature of exit tax on the change of tax residency could be drawn with a notion:
Figure 2. The Notion of Exit Tax on the Change of Residency
Source: De Man, et.al (2011)
According to De Man, et al (2011), the material fact (Mc) of the exit tax is
described by the verb (v) to leave and the complement (c) the State. The spatial criterion
(Sc) is the jurisdiction of the emigration State, and then the Fiscal year is the time period
(Tp). Regarding the tax consequence (Tc), personal criteria (Pc) consists of the
emigration state (As) as an active subject imposing a tax on the passive subject (Ps), the
emigrating person. The tax base (Tb) of the exit tax on migration is the unrealized capital
gain.
Taking consideration best practices from other countries, the authors extracted
several vital elements in crafting the exit charge regime within income tax regulation in
Indonesia:
57
4.1.1 Person Covered
A person, both individual and company, might be subject to the exit charge only
for certain conditions that are attached to them. This means that limitations are needed
to impose exit charge due to the significance purpose. Canadian departure tax, for
instance, limits the imposition to a resident based on the length of the residency period.
Subparagraph 128.1(4)(b)(iv) Canadian ITA stipulates that it excludes a short-term
residency; not more than 5 years residency during the 10-year period preceding
emigration (Yager, et.al. 2002). This residency period is also applied by Australia
pertaining to Capital Gain Tax 25 for an individual migration (Burns, 2002). The U.S
expatriation tax regime also excludes short-term residency, which applies to more than
eight-year permanent residency of the 15 taxable years. Additionally, one of three factors
should also be met by an individual expatriating: (1) Annual income tax, (2) net worth,
and (3) legal compliance (Kwong, 2009).
Taking into account best practices by countries, to the authors’ mind, the
approach taken by the U.S covering both material and formal compliance is essential to
adopt in determining the personal scope of the Indonesian exit charge. Thus, the authors
propose the personal scope of exit charge to be established by Compliance Risk
Management (CRM) assistance. The risk-based assessment provided by CRM will
reflect the taxpayer's formal and material compliance. Moreover, in case where more
considerations are needed, based on broader economic or policy purposes, for instance,
CRM can facilitate this purpose by articulating its compliance variables.
4.1.2 Material Scope

In essence, countries imposed the exit charge for all assets owned by a covered
person with certain exceptions. Canada and the U.S exclude assets that continue to
remain in the states’ tax base (Chand, 2013). In addition, the U.S exit tax includes the
exemption amount of the gains amounted to $600,000. Meanwhile, the Australian exit
tax covers capital gain from all assets with three exceptions: (1) assets that have the
inextricable link with the Australian tax base, (2) assets of a short-term resident, (3)
assets subject to Capital Gain Tax, i.e. non-inventory assets (Burns, 2002). However,
this approach subjects to debate because certain businesses, such as share traders,
could dodge the exit tax because share is regarded as a business inventory. Another
jurisdiction, the Netherlands, has an exit tax that is only applicable to substantial
shareholdings.
To the extent of Indonesia's exit tax, the authors propose the exclusion of assets
that remain under the Indonesian tax base. This will include immovable property and/or
assets recorded in the Permanent Establishment’s statement of the financial position,
given that the gain that will be derived by the assets in question is taxable in Indonesia.
Concerning the former, income arising from an immovable property situated in Indonesia,
even though it is owned by a non-resident taxpayer, will still be taxable in Indonesia
under Article 26 Income Tax Act (ITA), as the taxing right is allocated by Article 6 both
UN and OECD MTC as well as adopted by Indonesian tax treaties with other States.
4.1.3 Mark-to-Market Regime

The exit charge works with the principle of deemed realization on the gain derived
by assets at the moment when the taxpayer is about to cease their tax residency. This
means that the taxable amount of the gain is determined on an unrealized basis. Similar
25 Hereinafter referred as Australian Exit Tax.
58
to mark-to-market tax in the U.S, this regime allows Indonesia to levy an exit charge that
would have been due had its former residents alienated their assets.
In that regard, a follow-up question is: how do we determine the fair market value
of the assets at that point in time? A consideration to referring to a term used in transfer
pricing, “arm’s length principle”, might assist in answering this riddle. In fact, this principle
is widely used by OECD and EU countries (Pellecchia, 2018).26 The point of time of the
assets value would be reasonable if set up at the last day of his Indonesian residency.
From the authors’ view, DGT has already had resources to assess the assets valuation:
The Appraiser Officer. The valuation for tax purposes has been implemented since 2015
with DGT Circular Letter No: SE-61/PJ/2015, which later is revised by SE-05/PJ/2020.
In the later decree, The Appraiser Officer is also given a role in the assessment of the
fair market price of asset transactions, the fair market value of share, and calculating the
value of net assets in tax amnesty. To that end, the authors propose the extension of the
role of The Appraiser Officer in exit taxation purpose by performing an assessment on
the fair market value of the exit tax assets.
4.1.4 Type and Scheme of Exit Tax

One could argue that the exit tax cannot be imposed on Indonesian citizens just
because they have become non-residents and they do not get any income originating
from Indonesia. However, this argument is completely rejected because the exit tax
liability is established at the point of time when a taxpayer is still an Indonesian resident
on an income in the form of unrealized gain arises in Indonesia. The capital gains are
deemed to have been made while he was a resident in Indonesia and not after he
became a resident of the immigration State.
It is important to note that the material notion of exit tax depicts a taxable event
when a person leaves the emigration State. Hence, the authors propose to opt-in for an
immediate exit tax type because the notion means that the immediate exit tax type is
considered the general exit tax. An option of deferral payment until the realization of the
sale of an asset should be made available for a migrated taxpayer as most exit tax
regimes typically incorporate this feature. However, any changes to value that occur,
either gain or loss, in the post-emigration period should not be taken into account
because otherwise, it would not fall under the exit tax scope (De Man,et.al,2011).
The exit charge scheme must reflect the fair market value of the asset at the
moment of emigration; thus, the deferral option should only play a role as a payment
postponer of the established tax liability. This ‘frozen’ moment of exit tax 27 could be
convenient for the taxpayer’s cash flow because it would be levied when the taxpayer
sold the assets later. On the other hand, Indonesia can maintain its taxing right for the
portion of gain accrued therein. There could also be a discussion on whether taxing
unrealized gains in immediate exit charge contradicts the realization principle of the
income tax adopted by the Indonesian tax system. From the authors’ perspective, the
justification for the use of unrealized gains is that Indonesia has already applied a similar
regime in taxing gain on the revaluation of an asset stipulated in Article 4(1) ITA. On top
of that, Indonesia’s income tax adopts the Schanz-Haig-Simons concept of income,
which defines income as an increase in wealth plus economic consumption (Holmes,
2001); thus, unrealized capital gains will be viewed as income (De Man,et.al, 2011).
In combination with the gain on deemed disposal of an asset, the authors propose
the inclusion of gain of exit charge as a part of annual income tax since the capital gain
is part of income subject to tax under Article 4(1) ITA. To calculate the income tax due,
ITA prescribes the capital gains are considered as ordinary income and subject to
26 See Recital no. 10 of the preamble to the ATAD (2016/1164)

27 This term is used by De Man, et.al (2011) to describe the deferred exit tax.
59
income tax and in general, all expenses incurred in obtaining, billing or maintaining
capital gains are deductible (Kristanto, 2022). Henceforth, the immediate exit tax could
be calculated together with the total annual income tax during the fiscal year. The
question is, why do the authors not propose a final tax scheme? Justification of the
inclusion of the exit tax in an ordinary annual income tax would also relate to the
elimination of double taxation, which will be discussed further in section 4.4.
4.2 Exit Charge on Business Restructuring

Corporate restructuring or a business restructuring is a set of discrete decisive
measures to increase the competitiveness of a business entity and enhance its value.
The corporate restructuring also entails an improvement in operational or financing
structure to transform a firm into one that is of higher value or to survive when a
corporate’s business structure is indicated to be dysfunctional (Crum,et.al, 1998).
The idea of applying exit tax on business restructuring has been under the
spotlight since years ago, since anti-avoidance motives is predicted to be embedded
within a restructuring event of the business. Business restructuring could cause income
transfer overseas and a mismatch between the place of value creation and the taxed
place, resulting in a substantial reduction of the total amount of corporate tax burden
after the restructuring (Tsuji, 2020).
The authors prescribe the notion of exit tax on business restructuring using the
notion of tax scheme set out by Carvalho (2008) in De Man, et.al (2011).
Figure 3. Notion of Exit Tax on Business Restructuring
Source: the authors
Exit tax on business restructuring, in essence, describes the fattispecie by the

material criterion (Mc) containing restructure (v) of business (c). The spatial and time
period criterion are in the jurisdiction of the foregoing entity in business restructuring (Sc)
and fiscal year, respectively. The tax consequence will depend on the triggering events.
On the basis of triggering events which do not trigger the change of tax residency 28, the
personal criteria (Pc) is derived by the State of the foregoing entity, as the active subject
imposes a tax on the passive subject, the foregoing entity (Ps) in accordance with the
quantifying criteria (Qc). The tax base (Tb) is the fair market value of disposal of the
assets due to a business restructuring. The notion was not the case while the triggering
event resulted in the change of tax residency, since the whole notion would be referred
to as the notion as described in Figure 2.
In practice, the operation of exit tax on business restructuring is undertaken
through the application of transfer pricing principles and rules, as it involves the analysis
of restructuring from independent entities’ viewpoint – which would enter into transaction
according to sane business reasonings and not tax avoidance arrangement - to
determine whether there are causes that could trigger the exit tax provision to legally
apply.
28 See Triggering events proposed by the authors in Section 4.2.3.3
60
4.2.1 Business Restructuring in Transfer Pricing’s Perspective
Assessing how independent parties would undertake business restructuring is
necessary, since in the transaction between unrelated parties, the transfers that occurred
might lead to a payment in recompense, which in substance is similar to an exit charge.
There are at least three aspects to analyze whether compensation would have
been expected in transactions between independent parties, which are reallocation of
profit potential – also called the “expected future profit”, transfer of something of value,
and termination of existing agreements (OECD, 2022a). If one of these involved in a
business restructuring, an exit charge would generally be triggered.
First, if profit potential is reallocated, compensation to the entity forgoing such
potential in the form of transferred functions or risks might be expected in the context of
sane business logic. Such reallocation often occurs when the functional-assets-and-risks
profile of an entity is altered. As an example, in the event, a fully-fledged distributor is
restructured to be a limited-risk distributor or a sales agent of an overseas principal.
Nevertheless, the mere transformation of an entity’s risk profile does not
necessarily mean a warrant for a compensation payment by itself. Rather, the tax
authority should analyze and compare the historical performance of pre-restructuring and
the projected business performance post-restructuring. A compensation payment would
only be considered at arm’s length if the aforesaid analysis, according to valid evidence,
facts, and circumstances, resulted in a conclusion that actual potential profit has been
reallocated. Besides, tax authorities must include an assessment called the “options
realistically available” factor. The restructuring would only be considered at arm’s length
if no commercially more beneficial opportunities were available to meet the entity’s
objectives. Secondly, another component in a business restructuring event upon
which an exit charge might be a consequence is a transfer of “something of value”, which
includes tangible assets, intangible assets or the rights to such assets, and business
activities as well, known as “ongoing concern”. Transfer of an ongoing concern herein
constitutes a transfer of functioning, economically integrated business unit. According to
the OECD, this means the transfer of assets, bundled with the ability to perform certain
functions and assume certain risks (OECD, 2022b), which might consist of tangible and
intangible property, liabilities associated with holding certain assets and performing
certain functions such as R&D or manufacturing.
Then the third is the termination of existing agreements. If agreements are
terminated or renegotiated to the detriment of the restructured party, it has to be
assessed whether an indemnification needs to be paid to ensure arm’s length conditions
(Gubelmann, 2021). In the case of terminations or renegotiations of arrangements,
changes in the risk and functional profiles of the parties are generally occurred, which
brings consequences for the allocation of profit potential between parties. In addition, the
termination or renegotiation of contractual relationships in the context of a business
restructuring might cause the restructured entity to suffer detriments such as
restructuring costs (e.g. write-off of assets, termination of employment contracts), re-
conversion costs (e.g. in order to adapt its existing operation to other customer needs),
and/or a loss of profit potential (OECD, 2022c). In these situations, the question of
whether indemnification should be paid to the restructured entity (and, if so, how to
determine such an indemnification) arose.
One notable thing is that there should be no presumption that all contract
terminations or substantial renegotiations should give a right to indemnification at arm’s
length. Rather, this will depend on the facts and circumstances of each case. In order to
determine whether, at arm’s length, the restructuring itself would give rise to a form of
61
compensation, it is essential to understand the restructuring, including the changes that
have taken place, how they have affected the functional
analysis of the parties, what the business reasons for and the anticipated benefits
from the restructuring were, and what options would have been realistically available to
the parties.
4.2.2 Triggers of an Exit Tax on Business Restructuring

In general, regulations concerning exit tax on business or corporate restructuring
adopts the transfer pricing principles. As elaborated in the sub-chapter 4.2.1, the trigger
of exit tax implication involves the presence of these three aspects within a business
restructuring: expected future profit, transfer of something of value, and termination of
the existing arrangement. Hence the triggers of an exit tax in numerous jurisdictions are
designed by reference to those aspects taking into account jurisdiction-specific
considerations and interests.
The triggering event of an exit tax on a business restructuring is essential to draw,
as it would limit the application of exit tax which provides certainty on the coverage of the
provision under a transaction-by-transaction basis. This means that only business
restructurings which involve one of the triggers will be imposed by an exit charge.
According to Chua (2022), the general triggers of exit tax on business restructuring are:
1. a transfer of the whole business from one jurisdiction to another; relocation of
employees across the organization;
2. relocation of an asset, such as tangible assets, IP, agreements or clientele;
3. transfer of activity, through termination or substantial renegotiation of existing
arrangements.
In addition, besides the general triggers above, jurisdictions also set out events
that could trigger exit tax within their taxations landscape, which takes into account their
jurisdiction-specific policy rationale. Some of the jurisdictions’ practices in terms of exit
tax triggers are shown below.
Table 1. Comparison of Triggers of Exit Tax on Business Restructuring in the

European Union
Jurisdiction-Specific Provisions
No Jurisdictions concerning Exit Tax Trigger
Elaboration
- Transfers of assets: only applied A company will not be subject to

where the assets remain within the exit taxation if the transfer of
1. Denmark assets is of a temporary nature.
same company and not where
assets are transferred to another Hence, assets that are set to
company. revert to Denmark within a
period of 12 months are not
- Transfers of tax residence. comprised by the new rules on
- Transfers of the business are exit taxation.
carried on by permanent
establishment from an EU
member state to another member
state or a third country.
Transfer of self-developed intangible Conversion from a distributor to

asset a commission agent constitutes
2. France a formal transfer of clientele,
which then triggers a potential
62
capital gain and the applicable
tax to the value of the clientele
transferred.
the change of business form before There should be compensation
and after restructuring. upon the termination of a
3. Germany commercial agent,
commissionaires and buy-sell
distributors (if certain
requirements are met).
- A shift of the place of effective - If the tax basis is
management of a Swiss company to maintained in Switzerland,
4. Switzerland abroad. no exit tax will apply.
- Transfer of assets from a permanent - If the registered office of an
establishment to a company tax entity remains in
resident in another country or to a Switzerland and no tax
non-Swiss permanent treaty applies covering the
establishment. residency, the effective
management office will be
deemed to stay in
Switzerland.
- Transfer of company tax residence Exemption and relief on exit tax

status out of the jurisdiction. are available under participation
5. United - Transfer of assets out of the exemption regime.
Kingdom jurisdiction.
- The cessation of a trade carried out
in the United Kingdom through a
permanent establishment.
Source: Resumed by Authors
4.2.3 Envisioning Exit Charge on Business Restructuring Adoption in

Indonesia
This paper has discussed the fundamental principles of tax implications on
business restructuring and international best practices of how jurisdictions set out their
exit tax regime. In this subchapter, the authors will boil down the ideas of applying an
exit carge regime and feasible policy recommendations on business restructuring to
implement within Indonesia’s taxation landscape.
4.2.3.1 Legal Provision
First, since exit charge implication on business restructuring, in general, heavily
adopts transfer pricing principles, authors propose that the triggering event of exit charge
on business restructuring is conducted under the existing law in Indonesia; thus, no
amendment of the law would be needed.
The obligation to comply with arm’s length principles (i.e. the principles of
independent parties which would drive their behaviour and decision pursuant to a
transaction with an unrelated party) is stipulated in Article 18(3) of the ITL and Ministry
of Finance Regulation Number 22 of 2020, with the related party definition is stipulated
in Article 18(4) of the ITL.
The implementation of exit charge under the existing regulation is of the
consideration that the existing regulation has already embodied the legal authority for
63
imposing a tax on the compensation that would arise when a business restructuring takes
place in accordance with sound business rationale, that in the result is aligned with anti-
avoidance purpose.
4.2.3.2 Entities Covered

Exit charge regimes provision vary across jurisdictions, which contains not only
transfer pricing principles on business restructuring but also jurisdictions-specific
considerations, such as the relationship between commercial law and tax law, the
functional investment and business vehicles available therein, and the existing
international agreements agreed among jurisdictions in a certain geographic region.
Although there are several business forms In Indonesia, such as limited company,
limited partnership, and firm, the authors proposed the covered entity of the exit tax
regime concerning business restructuring is a limited company and its permanent
establishment.
Firstly, business restructuring involves the transfer of activities, assets, and valuable
arrangements, which are likely to occur within resource-intensive entities or a function-
diversified company. As such, the type of entity which is highly possible to restructure is
limited company.
Secondly, one of the main objectives to implement an exit tax regime is the cross-
border anti-avoidance purpose, which involves global investment and business activities
in and out of Indonesia’s jurisdiction. While foreign investment in Indonesia is one of the
essential factors to boost the economy, it can only be carried out through shares
ownership in a limited company established in Indonesia, as stipulated in Law Number
25 of 2007 on Foreign Investment juncto Presidential Regulation Number 10 of 2021 on
Business Investment Activity.
4.2.3.3 The Triggering Events

Policy design regarding business restructuring is heavily intersecting with transfer
pricing principles. Accordingly, the authors suggest the events that could trigger exit
charge to be imposed on business restructuring begin with related party fulfilment as
stipulated in article 18(4) of ITL and its implementing regulation, of which the proxies are
shares ownership, controlling power, and cognation or marital relationship. This means
that in the case of business restructuring, exit charge will not be applied if the parties do
not satisfy the related party definition.
Secondly, the authors proposed to apply exit charge once one of the following
triggers is present in a business restructuring:
1. Transfer of the whole business function from Indonesia to another jurisdiction;
2. Transfer of tangible assets and intangible assets of an entity or a permanent
establishment to another jurisdiction;
3. Transfer of an activity, through termination or substantial renegotiation of an
existing arrangement, of a resident entity of Indonesia to another jurisdiction.
In the event a business restructuring occurred in which one of the triggers is
involved, exit charge should apply. Conversely, no exit charge should apply if the
restructuring does not satisfy the related party transaction definition nor involve the
triggering events.
For example, A co, an Indonesian taxpayer, has been carrying out manufacturing,
warehousing, distribution, and marketing functions in Indonesia. After a group
restructuring, A co will no longer actuate those mentioned functions in Indonesia, as the
manufacturing function will be transferred to its sister company in B jurisdiction. The
restructuring satisfies the related party transaction definition and the triggering events,
which are the transfer of an activity, through the termination of a resident entity of
Indonesia to another jurisdiction. Therefore, an exit charge should be imposed.
64
The application of exit tax on business restructuring is illustrated in the following
diagram.
Figure 4. Illustration of Indonesia’s Exit Charge Application
Source: resumed by the authors
4.3 Interplay with Tax Treaties

There could be a discussion whether the exit tax interacts with the tax treaty when
it deals with cross-border transfer of residency. The proposed exit tax on unrealized gain
on migration can relate to the distributive rule of capital gain in tax treaties laid down in
Article 13 OECD MTC and UN MTC (De Man, et.al.,2011 and Chand, 2013).
While the wordings of Article 13 distribute taxing rights by the words “alienation”,
the exit tax is imposed on the unrealized gain. Accordingly, one of the debates in this
extent is whether exit tax on change of residency to another jurisdiction, is compatible
with article 13(5) OECD MTC which allocates the taxing rights of capital gains to the
residence state of the alienator with regard to ‘other assets’ in Article 13.
The wordings “.the words “alienation of property” are used to cover in particular
capital gains..’ in Paragraph 5 Commentary on Article 13 OECD MTC imply that the term
alienation is not limited to the sale or exchange of property (Moser, 2019). In fact, in the
next paragraph, Paragraph 6 Commentary, OECD enunciates that whether or not there
is a realization has to be determined according to the applicable domestic tax law. It is
65
apparent that the coverage of Article 13 OECD MTC is not only for realized gains but
also unrealized gains with a recourse to domestic law.
Taken into account the interpretation of the Commentaries, the authors follow the
conclusion of some scholars that come up with the broad meaning of “alienation” and
conclude that Article 13(5) of the OECD MTC does not prevent the emigration state from
levying unrealized capital gain tax to its resident taxpayers.29 In addition to that, it is the
sovereignty of a State to apply domestic law to its residents. Tax treaties do not prevent
the application of domestic tax rules according to which a person immediately before
ceasing to be a resident, is considered to have alienated property for capital gain tax
purposes (OECD, 2015).
4.4 Double Taxation Issue

Another issue that has become subject to debate is that the potential conflict of
double taxation associated with the imposition of exit tax on change of tax residency by
the State of emigration. This could be the case when the hosting country taxes the
respective capital gains at the taxable amount established from the difference between
sale price and historical book value when they are realized. This means that the
immigration State does not take into account the market value of the assets at the
moment when the taxpayer changes his residency, while on the other hand, the
emigration State had taxed the unrealized gains which are calculated as the difference
between fair market value at the point of emigration and the historical book value.
To illustrate, suppose Mr.A an Indonesian resident is emigrating to Singapore in
31 July 2020 and thus subject to exit tax on the unrealized gains at market value at the
point of time prior to the departure. Exit tax is calculated at fair market value 31 July 2020
minus book value of the asset bought in 2015. Mr.A is deemed to have sold the assets
before he emigrated. Five years later, in 2025 Mr.A’s asset is sold while he is a
Singaporean resident. In that case, if capital gain tax in Singapore is calculated as selling
price minus historical book value, the portion of gains unrealized in Indonesia would be
taxed twice. A simple formula below could be drawn for a better understanding:
TTb = Te + Ti
TTb = %Te(FMv – Hbv) + %Ti(Sv – Hbv)
TTb = Total Tax burden

Te = Tax in emigration State
Ti = Tax in immigration State
% = Tax rate
FMv = Fair market value of an asset at the point of emigration
Hbv = Historical book value of an asset
Sv = Sales value of an asset
From the formula above, it could be noted that without any relief, double taxation
can arise from both countries tax the difference between the value of their tax bases and
the asset’s historical book value. Scholars have constructed two solutions based on the
practices taken by countries levying exit tax. Firstly, granting the step-up value on the
capital gain in the immigration State. Stepped-up basis refers to a tax policy that looks
at the market value of assets at the time a person inherits them instead of the value when
the prior owner purchased the assets.30 In the aforesaid illustration, double taxation could
29This reason supports the views by Chand (2013) and Moser (2019)
30The definition by Wex Definition Team, Cornell Law School, Legal Information Institute can be
accessed through https://www.law.cornell.edu/wex/stepped-up_basis. (accessed 28 July 2022)
66
be avoided by ensuring that Mr.A’s assets to be recognized at market value in Singapore
immediately after Mr.A becomes a resident therein (step-up). In that way, when the gain
is realized, the Singaporean capital gain tax will be imposed at the amount of the
difference between the sale price and the fair market value at the time when the assets
were first recorded in Singapore. Applying the step-up to the formula, it can be seen that
each country would have a different portion of the cake of taxation on the asset in
question. While Indonesia only taxes the difference between the fair market value at the
point of emigration in 2020 and the historical book value in 2015, Singapore taxes the
difference between sales value in 2025 and fair market value in 2020. The formula would
be as follows:
TTb = %Te(FMv – Hbv) + %Ti(Sv – Fmv)
In the step-up context, Indonesia has to ensure that the emigrating individual is
provided with a step-up in the immigration State. Hence, bilateral negotiation between
Indonesia and the hosting jurisdictions includes a step-up clause in the tax treaties.
Some States include step-up clauses on exit tax in their tax treaties, for instance,
Canadian tax treaties, U.S tax treaties, Denmark-South Africa DTC and Germany-South
Africa DTC. In the authors’ opinion, it is likely to be a long way for Indonesia to implement
this solution considering the need to ratify its tax treaties.
Another approach is that a reverse-credit method can be applied to relief double
taxation. Countries like Canada and the Netherlands for instance, provides a reverse
credit under domestic law to foreign tax paid pertaining to the pre-emigration portion of
the gains (Chand, 2013). An illustration of the reverse credit can be made from the
previous example. To the case at hand, since Singapore imposes a tax on the part of the
gain subject to the Indonesian exit tax, the tax base is the difference between the fair
market value at the emigration date in 2020 and the historical book value in 2015, a tax
credit for the tax levied in Singapore in respect of the part of the gain can be provided by
Indonesia. Adding variable of Foreign Tax Credit (Ftr) to the formula could be drawn as
follows:
TTb = Te + Ti - Ttr
TTb = %Te(Fmv – Hbv) + %Ti(Sv – Hbv) - Ftr
TTb = %Te(Fmv – Hbv) + %Ti(Sv – Hbv) - %Ti(Fmv – Hbv)
TTb = %Te(Fmv – Hbv) + %Ti(Sv – Fmv)
From the formula above, it can be seen that with the reverse tax credit, the total
tax burden of the tax payer will be levied only for the difference between sales value and
historical book value; thus, no double taxation occurs. In the authors’ opinion, following
Chand (2013), the reverse credit method is suitable to adopt in Indonesia to relief double
taxation arising from the proposed immediate exit tax. To that end, Indonesia could
unilaterally include the reverse credit in Article 24 ITA scheme. This will answer the
question of why the ordinary annual income tax scheme is opted in by the authors instead
of the final tax. If final tax is in place, there will be challenging to provide this reverse
foreign tax credit to relief double taxation.
Another alternative, as suggested by the OECD, is bilateral negotiation through
Mutual Agreement Procedure (MAP), that each State should provide relief as regards
the exit tax that was levied by the other State on the part of the income that accrued
while the person was a resident of that other State (OECD, 2015) could be taken into
account.
67
5. CONCLUSION
As the transfer of residency and entities’ valuable assets, functions, and
arrangements through business restructuring is increasing from time to time, while the
effectiveness of the existing anti-avoidance measures, in general, is indicated to be
decreasing, an adoption of exit charge in the renouncement of tax residency and
business restructuring is considered a workable mechanism to protect Indonesia’s tax
base. The proposed exit tax regime is set out under existing Indonesia’s tax regulation,
which allows the efficient implementation without excessive administrative burden as well
as such a lengthy legislative process. The exit charge constitutes a spectrum of income
tax, which applies to the change of residency under the capital gain tax provision and the
specific business restructuring under the transfer pricing provision.
6. IMPLICATION AND LIMITATION

Although exit tax regime has been common in some jurisdictions, research and
discussion concerning exit tax regime in Indonesia are relatively scarce. The policy
recommendation in this paper is expected to provide a new perspective in the effort of
Indonesia to strengthen its taxation system through a presentation of a comparative
study on international best practices, taking into account Indonesia’s specific variables
comprising of existing legal basis and tax system. The sequence of research on exit tax
adoption is encouraged and is suggested to include statistical data regarding the positive
impact of exit tax regimes to jurisdictions’ revenue collection and the adverse impact on
foreign investment and the economy, as well as to offer a more comprehensive evidence-
based study.
BIBLIOGRAPHY
[1] Burns, Lee. (2002). Australia - The tax treatment of transfer of residence by
individuals. IFA Cahiers 2002 - Volume 87b. IBFD
[2] Carvalho, Paulo de Barros. (2010). Curso de Direito Tributário. 22nd ed.
[3] Chand,Vikram. (2013). Exit Charges for Migrating Individuals and Companies:
Comparative and Tax Treaty Analysis. 67 Bull. Intl. Taxn. 4/5. Journals IBFD
[4] Chua, Caroline. (2022). Managing an Exit Tax. KPMG Blog.
https://home.kpmg/ch/de/blogs/home/posts/2022/02/managing-exit-tax.html
[5] Crum, Roy L.,& Goldberg, Itzhak. (1998). Restructuring and Managing the
Enterprise in Transition. Washington: World Bank
[6] De Man, Fernando., & Albin,Tiiu. (2011). Contradicting Views of Exit Taxation
under OECD MC and TFEU: Are Exit Taxes Still Allowed in Europe?. Intertax:Issue
12.pp. 613-625.
[7] Gubelmann, Beat (2021) Transfer Pricing Aspect of Business Restructuring.
KPMG Switzerland Blog.
https://home.kpmg/ch/en/blogs/home/posts/2021/03/transfer-pricing-aspects-of-
business-restructuring.html
[8] Heckemeyer, Jost H., & Overesch, Michael. (2017). Multinationals profit response
to tax differentials: Effect size and shifting channels. Canadian Journal of
Economics, 2017, vol. 50, issue 4, 965-994
[9] Henshall, John., & Achim, Roeder. (2012). Business restructuring: Exit Charges
for Restructuring in Europe. International Tax Review.
https://www.internationaltaxreview.com/article/2a69ixqe8r3jlftbzy8sg/business-
restructuring-exit-charges-for-restructurings-in-europe
[10] Holmes, Kevin. (2001). The Concept of Income:A Multi-Disciplinary Analysis. IBFD.
68
[11] Hug, Johanna Clara. (2015). The Taxation of Gains derived from the Alienation of
Shares in Real Estate Companies Under the OECD Model Tax Convention.
Maastricht University Faculty of Law.
[12] Jiang,Quanfang & Yuan,Yifan. (2014). Legal Research in International and EU Tax
Law. 54 Eur. Taxn. 10. IBFD
[13] Kristanto, Henrietta. (Last Reviewed: 30 April 2022).Indonesia - Corporate
Taxation sec. 1. Country Tax Guides IBFD.
[14] Kubicova, Jana (2016). Exit Tax in the World of International Migration of
Companies and Individuals. 16th International Scientific Conference of
Globalization and Its Socio-Economic Consequences, Slovac Republic, p.1101
[15] Kwong, Yu Hang Sunny. (2009). Catch me if you can: Relinquishing Citizenship
for taxation purposes after the 2008 Heart Act. 9. Houston Bus. & and Tax J. 3, p.
428
[16] McConville, Mike, & Chui, Wing Hong. (2017). Research Methods for Law: Vol.
Second edition. Edinburg University Pers.
[17] Moser, Karen. (2019). Recent Developments and Challenges Regarding Exit
Taxes in the Context of Tax Treaties: Article 13 of the OECD Model and Change
of Residence. 73 Bull. Intl. Taxn. 10. IBFD.
[18] OECD. (2015). Action 6 Final Report 2015 – Preventing the Granting of Treaty
Benefits in Inappropriate Circumstances. OECD Publishing. Paris.
https://www.oecd-ilibrary.org/docserver/9789264241695
en.pdf?expires=1660534059&id=id&accname=guest&checksum=932B3F74DFC
3DDE02E2EB3A3B7AE8F1A
[19] OECD. (2022a). OECD Transfer Pricing Guidelines for Multinational Enterprises
and Tax Administrations 2022. OECD Publishing:Paris.
https://doi.org/10.1787/0e655865-en.
[20] OECD. (2022b). OECD Transfer Pricing Guidelines for Multinational Enterprises
and Tax Administrations 2022. OECD Publishing:Paris. p.381, para 9.68.
https://doi.org/10.1787/0e655865-en.
[21] OECD. (2022c). OECD Transfer Pricing Guidelines for Multinational Enterprises
and Tax Administrations 2022. OECD Publishing:Paris. p.384, para 9.76.
https://doi.org/10.1787/0e655865-en.
[22] Pellecchia, Massimo. (2018). Exit Taxation and Restructuring Operations: A
Comparison of EU and Italian Tax Law. 58 Eur. Taxn. 8. Journal Articles & Opinion
Pieces IBFD.
[23] Schuch, Josef, & Pinetz, Erik in Lang, et. al. (2014). The OECD MC and its Update
2014.
[24] Sebastian Beer, Ruud de Mooij, and Li Liu. (2018). International Corporate Tax
Avoidance: A Review of the Channels, Effect Sizes, and Blind Spots, IMF Working
Paper WP/18/168, p.7
[25] Tsuji, Mie. (2020). Global Business Restructuring of Insurance Industry and
Taxation, Conference Proceeding. 1st Croatian – Japanese Conference June 16
2020, p.22
[26] Yager, James.,& Dyer,Geoffrey J.R. (2002). The tax treatment of transfer of
residence by individuals. IFA Cahiers 2002 - Volume 87b. IBFD
69
MACHINE LEARNING: CLASSIFIYING TAXPAYER’S
SUPERVISING ZONE BASED ON THE STREET
ADDRESS USING NATURAL LANGUAGE PROCESSING
ALGORITHM
Reno Iqbalsah
KPP Pratama Jakarta Koja – Direktorat Jenderal Pajak Email: reno.iqbalsah@pajak.go.id
ABSTRACT
Machine Learnig: Classifiying Taxpayer’s Supervising Zone Based on the Street Address Using
Natural Language Processing Algorithm. Assigning Taxpayers into the respective Account
Representatives is a crucial step to optimize Taxpayers supervision. But, the large amount of
registered Taxpayers and missing data hase been a great challenge. A lot of Taxpayers only include
their street addresses and no additional information such as RT, RW, etc. This will result that each
Taxpayers address has to be manually searched in internet and manually assigned, this process is
not a really efficient and takes a lot of time. Thus, using only street address names I will try to solve
this problem using a machine learning model. There are a lot of machine learning model that are
present on this modern day. But, since the street addresses are mostly text data, I will use the
common text processing machine learning, the Natural Language Processing (NLP) algorithm. Using
the ‘Bag of Words’ model, this model will parse through a set of text data or usually called document,
then classify them based on the similarity of the features given using cosine similarity.
Keywords: Machine Learning, Natural Language Processing, Supervision Zones, Cosine Similarity
Machine Learning: Pengelompokan Wajib Pajak dalam Zona Pengawasan Berdasarkan Alamat dan
Nama Jalan menggunakan Algoritma Natural Language Processing. Melakukan assign wajib pajak
kepada Account Representative merupakan Langkah krusial dalam meningkatkan pengawasan
Wajib Pajak. Namun, banyaknya jumlah wajib pajak dan kurangnya data yang memadai menjadikan
proses ini cukup sulit. Banyak Wajib Pajak yang tidak mencantumkan alamatnya secara lengkap,
dan hanya mencantumkan nama jalan, tanpa mencantumkan RT, RW, dsb. Hal ini masih mungkin
dilakukan dengan melakukan assign manual satu per-satu dengan mencari alamat wajib pajak di
internet, tapi metode ini sangat tidak efisien dan memakan waktu yang lama, Dengan menggunakan
Machine Learning, saya akan berusaha memecahkan masalah tersebut. Salah satu Machine
Learning yang biasa digunakan dalam pemprosesan data teks adalah Natural Language Processing.
Dengan beberapa sampel alamat yang jelas dan sudah ter-assign secara baik, dapat dilakukan
training terhadap model teresebut, dan kemudian digunakan untuk memprediksi zona pengawasan
yang tepat untuk wajib pajak lainnya berdasarkan kemiripan pada features dalam data tersebut.
Algoritma ini menggunakan model ‘Bag of Words’ untuk menentukan klasifikasi menggunakan cosine
similarity.
Kata kunci: Machine Learning, Natural Language Processing, Zona Pengawasan, Cosine Similarity
1. Introduction
Regarding to Nota Dinas Direktur Ekstensifikasi dan Penilaian No. ND-
56/PJ.06/2022, KPP Pratama Jakarta Koja’s supervising zone has been divided into 55
zones and supervised by Account Representatives from 5 Supervising Division (Seksi
Pengawasan). Every Taxpayers has to be assigned the respective Account
Representatives of each zone using the Mapping application. This app automatically
assigns Taxpayers 1 of 55 supervising zones based on their address’ geotagging.
Unfortunately, not all Taxpayers has their addresses geotagged, and the remainder of
the untagged Taxpayers’ addresses has to be manually assigned.
70
However, in KPP Pratama Jakarta Koja’s case, these remainders are in an
extremely large amount. With 344.229 Taxpayers registered (June 18th, 2022), only
3.222 Taxpayers are assigned automatically using point of interest (POI). The problem
is that the addresses are not all standardized and there are a lot of missing information.
We could actually search each street names and assign it one by one to the respective
supervising zones. But it would take a lot of time to assign more than 300.000 Taxpayers.
Thus, useing Machine Learning algorithm I will try to classify all these remainders and
assign them into their respective zones based on their street addresses.
2. Theorethical framework
This algorithm will use the Bag of Words model. Bag of Words turns a set of text or
document, into a vector containing numbers representing the count of each word in the
document. This step is called vectorization. Since it is pretty hard to compute the
similarity of text data quantitively, turning the text into vector makes it easier for the
computer to process the data and compare the similarity of one document to the other.
Table 0.1 Example of Vectorization

Address JL MANGGA DURIAN MERAH MANIS KP TANAH
JL MANGGA
MERAH, KP → 1 2 0 1 0 1 0
MANGGA
JL DURIAN MANIS,
→ 1 0 1 1 1 0 1
TANAH MERAH
Based on the vectorization illustrated in the Table 2.1, the vectorized values can be
saved into vector A and vector B. With the values of vector A = [1,2,0,1,0,1,0] and B =
[1,0,1,1,1,0,1]. We could measure the similarity of those vectors using Cosine Similarity.
This formula is based on Euclidean Dot Product formula where the dot product of two
non-zero vectors can be calculated using the following equation:
𝐴 ∙ 𝐵 =∥ 𝐴 ∥ . ∥ 𝐵 ∥ . cos(𝜃)
or
𝑛
𝐴 ∙ 𝐵 = ∑ 𝐴𝑖 . 𝐵𝑖
𝑖=1
Then, we could define the cosine similarity as:
𝐴∙𝐵
𝑐𝑜𝑠𝑖𝑛𝑒 𝑠𝑖𝑚𝑖𝑙𝑎𝑟𝑖𝑡𝑦 = 𝑆𝐶 (𝐴, 𝐵) ∶= cos(𝜃) =
∥ 𝐴 ∥ .∥ 𝐵 ∥
↓↓
∑𝑛𝑖=1 𝐴𝑖 . 𝐵𝑖
cos(𝜃) =
√∑𝑛𝑖=1 𝐴𝑖 2 . √∑𝑛𝑖=1 𝐵𝑖 2
Using these equations, we could calculate the Cosine Similarity of vector A and B is
equal to 0,41. This is the basic of the Bag of Words model. Then, I will try to apply this
71
into a larger dataset and try some different methods to get the best model that suits our
need to classify Taxpayers’ supervising zones.
3. Research methods
The methods used in this research is mainly data processing, data cleansing, and
fitting the dataset into a machine learning model. I will use a dataset of correctly assigned
Taxpayers’ addresses to build an algorithm based on the addresses and supervising
zones. And then, predict the other supervising zones using said model.
3.1. Data Collection
The data used in this paper is a sample data containing Street Addresses from the
first five supervising zone in KPP Pratama Jakarta Koja. The data has been labelled as
Zone 1 to 5. The data collection method used is using Python selenium web driver to
scrape random addresses from Google Maps. This method used to minimize the use of
actual Taxpayer’s addresses, to prevent any privacy issues. For actual use in your own
office if you wish to replicate this research, you can use actual Taxpayer’s address for a
better accuracy.
As a control variable, the data will be split into training set and test set. The test set
will be 20% with a random state of 101 in Scikit-Learn Train/Test Split function. This
being said, reproducing the research with different parameters, might result in a slightly
different performance of the model.
3.2. Tools Used

The research will be done in Python environment using some of built-in functions
and additional libraries for data processing and machine learning building.
3.2.1. Hardware
The hardware used is computer with the following specifications:
1. AMD Ryzen 7 3750H 64-bit.
2. 8 GB of DDR4 ram.
3. NVIDIA GeForce GTX 1660Ti, Max-Q Design, with 6 GB display memory.
3.2.2. Software
The operating system used in the research is Windows 10 Home 64-bit. The main
data processing tool is Python version 3.9.7 with Anaconda distribution. In addition to the
built-in functions in Python, these following libraries would be installed:
1. Data processing libraries:
- Numpy (1.20.3), array processing library for numbers, strings, records, and
objects;
- Pandas (1.3.4), data structures and data processing library, mainly for data
frames.
2. Web Scraping libraries:
- Selenium Web Driver,
- Parsel.
3. Data visualization libraries:
- Matplotlib (3.4.3), basic data visualization library;
- Seaborn (0.11.2), advanced data visualization library based on Matplotlib;
- Geopandas (0.10.2), Pandas’ geographical extension for processing geo
data frames;
- Folium (0.12.1), interactive geographical plotting library
4. Scikit-Learn (0.24.2), machine learning modules for Python.
72
4. RESULTS
After using the web scraping method to collect data, about 573 random street
addresses are collected. These street addresses are from Zone 1 to 5 in KPP Pratama
Jakarta Koja’s supervising zone, and all of them are from the Kelurahan Sukapura.
These are some examples of the sample addresses:
Table 0.2 Sample Addresses Scraped from Google Maps

Address Label
Jl. Mahoni II 1-17, RT.12/RW.5, Sukapura, Kec. Cilincing, Kota Jkt
ZP 1
Utara, Daerah Khusus Ibukota Jakarta 14140
Jl. Panda Lestari II 1-3, RT.8/RW.9, Sukapura, Kec. Cilincing, Kota
ZP 1
Jkt Utara, Daerah Khusus Ibukota Jakarta 14140
Jakarta Utara, RT.5/RW.5, Sukapura, Kec. Cilincing, Kota Jkt Utara,
ZP 1
Daerah Khusus Ibukota Jakarta 14140
Jl. Anoa Lestari II 3-8, RT.3/RW.9, Sukapura, Kec. Cilincing, Kota
ZP 1
Jakarta Utara, RT.1/RW.5, Sukapura, Kec. Cilincing, Kota Jkt Utara,
ZP 1
Daerah Khusus Ibukota Jakarta 14140
Jl. Panda Lestari I 10-16, RT.8/RW.9, Sukapura, Kec. Cilincing, Kota
ZP 1
Jakarta Utara, Sukapura, Kec. Cilincing, Kota Jkt Utara, Daerah
ZP 1
Khusus Ibukota Jakarta
Jl. Anoa Lestari I 18, RT.2/RW.9, Sukapura, Kec. Cilincing, Kota Jkt
ZP 1
Jl. Sukapura No.55, RT.3/RW.5, Sukapura, Kec. Cilincing, Kota Jkt
ZP 1
JL. Elang, Sukapura, Cilincing, Komplek Walikota Blok A 3/10,
Jakarta, RT.12/RW.5, Sukapura, Cilincing, North Jakarta City, ZP 1
Jakarta 14110
Jl. Tipar Cakung No.55, RT.2/RW.5, Sukapura, Kec. Cilincing, Kota
ZP 1
Jl. Taman Orchard, RT.8/RW.9, Sukapura, Kec. Cilincing, Kota Jkt
ZP 1
These are not the full addresses used in the research. For convenience, the full data and
source codes will be presented in separate files. After collecting the data, next step is to
process the data using Python to create a Machine Learning model based on the bag of
words theory.
4.1. Creating Basic Model
73
The basic model consists of three steps, Tokenization, Vectorization, and model
training. After those three steps are done, we will put the model into a data pipeline for
an easier reusability and model adjustment.
4.1.1. Tokenization
Tokenization is a step of dividing text or document into each word, this will make the
text easier to process in the next step, the vectorization. To do the tokenization, I created
a Python function, this function will mostly do the following step:
- Remove all punctuations from the text,
- Split the text using into each word by removing the whitespaces in the text
and return it as a list.
And I’m going to call this function as text_process function.
Here are some examples when the function is being invoked to the sample address
text data:
Table 0.3 Example of Tokenization

Address Processed Address
Jl. Mahoni II 1-17, RT.12/RW.5, → [“Jl”, “Mahoni”, “II”, “117”,
Sukapura “RT12RW5”, “Sukapura”]
Jl. Panda Lestari II 1-3, → [“Jl”, “Panda”, “Lestari”, “II”, “13”,
RT.8/RW.9, Sukapura “RT8RW9”, “Sukapura”]
Jl. Anoa Lestari II 3-8, RT.3/RW.9, → [“Jl”, “Anoa”, “Lestari”, “II”, “38”,
Sukapura “RT3RW9”, “Sukapura”]
This function will not be directly invoked to the data, rather, this function will be passed
as analyzer in the vectorization process.
4.1.2. Vectorization
Vectorization is a step of converting the text into vectors. This is done to make the
computer machine easier to process the data, since text data are harder to compute. In
this step, I’m utilizing Scikit-Learn’s built-in text processor called CountVectorizer and
use my self-defined text_process function as the analyzer. The CountVectorizer will then
convert my text data into a sparse matrix containing the count of each word. This is what
called as the bag of words data.
Table 0.4 CountVectorizer's Sparse Matrix Output

Address 1 Address 2 … Address N
Word 1 Count 0 0 … 0
Word 2 Count 2 1 … 1
… … … … …
Word N Count 1 0 … 0
This sparse matrix will be passed as features in the machine learning training step.
Meanwhile the already classified supervising zone will be passed as the target label. I
assigned the sparse matrix from the vectorization process as a variable called bow, a
short for bag of words.
4.1.3. Model Training

In the model training, I will use one of the most popular classifier for natural language
processing (NLP), the Naïve Bayes classifier. The Naïve Bayes classifier is based on
74
Bayes’ Theorem developed by Thomas Bayes to describe the probability of an event
based on prior knowledge of the condition that might affect the outcome of an event.
Bayes’ Theorem stated as mathematical equation as follows:
𝑃(𝑦)𝑃(𝑥1 , … , 𝑥𝑛 ∣ 𝑦)
𝑃(𝑦 ∣ 𝑥1 , … , 𝑥𝑛 ) =
𝑃(𝑥1 , … , 𝑥𝑛 )
Where y and x are events and P(x) is not 0.
 𝑃(𝑦 ∣ 𝑥1 , … , 𝑥𝑛 ) is the probability of event y, given the set of x occur;
 𝑃(𝑥1 , … , 𝑥𝑛 ∣ 𝑦) is the probability of the set of x, given the y occurs; and
 𝑃(𝑦) and 𝑃(𝑥1 , … , 𝑥𝑛 ) is the probability of y and x occurs in any circumstances.
Using the naïve assumption that 𝑃(𝑦 ∣ 𝑥1 , … , 𝑥𝑛 ) is equal to 𝑃(𝑦 ∣ 𝑥𝑖 ), the equation is
simplified into:
𝑛
𝑃(𝑦) ∏𝑖=1 𝑃(𝑥𝑖 ∣ 𝑦)
𝑃(𝑦 ∣ 𝑥1 , … , 𝑥𝑛 ) =
𝑃(𝑥1 , … , 𝑥𝑛 )
𝑛
Where the 𝑃(𝑦) ∏𝑖=1 𝑃(𝑥𝑖 ∣ 𝑦) is the product of all 𝑃( 𝑥1 ∣ 𝑦 ) ∗ 𝑃( 𝑥2 ∣ 𝑦 ) ∗ … ∗ 𝑃( 𝑥𝑛 ∣ 𝑦 ).
Given the input is saved in the model as the training set, the 𝑃(𝑥1 , … , 𝑥𝑛 ) is constant, we
can use the following classification rule:
𝑛
𝑃(𝑦 ∣ 𝑥1 , … , 𝑥𝑛 ) ∝ 𝑃(𝑦) ∏ 𝑃(𝑥𝑖 ∣ 𝑦)

𝑖=1
⇓
𝑛
^
𝑦 = arg 𝑚𝑎𝑥 𝑃(𝑦) ∏ 𝑃(𝑥𝑖 ∣ 𝑦)
𝑦
𝑖=1
To train the model, I created a machine learning object called zp_model and assign
the Scikit-Learn’s MultinomialNB (Multinomial Naïve Bayes) as the value. Then, fit the
bow object as the feature (X), and the supervising zone as the labels (y).
4.1.4. Creating Data Pipeline

The bag of words algorithm contains a lot of process, to make the program more
reusable, Scikit-Learn has a built-in function called Pipeline to summarize some machine
learning steps and contain it in a single object, transforming the three steps process into
just one process. By using pipeline, it’s easier for the model to be reused and adjusted,
just by modifying the parameters or the steps used. For the rest of the research, I will be
using this pipeline to adjust the model and try to find best parameter that suits the data.
I created a pipeline object called model_pipeline and used the fit method to train
the model. Then, I use the predict method on the object to return the prediction done by
the model. This prediction will be used as a comparation for evaluation purpose.
4.1.5. Model Evaluation

As being said above, the data used in training the model is split into training set
and testing set using Scikit-Learn’s TrainTestSplit function, with 80% data used as
75
training set and 20% as testing set. Using Scikit-Learn’s classification_report function, I
printed the model’s accuracy by comparing the predicted value of the model and the true
value of the labels. With this basic model, I got the model accuracy as follows:
Table 0.5 Basic Model Accuracy

Precision Recall F1-score
ZP 1 90% 97% 93%
ZP 2 76% 68% 72%
ZP 3 70% 78% 74%
ZP 4 78% 81% 79%
ZP 5 82% 78% 80%
accuracy 81%
Precision: Number of true positive compared to number of predicted positive

Recall: Number of true positive compared to actual positive value
F1-Score: Comparation of precision and recall calculated as:
𝑝𝑟𝑒𝑐𝑖𝑠𝑖𝑜𝑛 ∙ 𝑟𝑒𝑐𝑎𝑙𝑙
𝐹1 = 2 ∙
𝑝𝑟𝑒𝑐𝑖𝑠𝑖𝑜𝑛 + 𝑟𝑒𝑐𝑎𝑙𝑙
From the table 4.4 we can see that the model has the accuracy of 81%. Which is
good enough. But we will try to improve the model to be better.
4.2. Adjusting the Model

To improve my model, I tried some different methods and functions that is relevant
to the data I’m using.
4.2.1. Using TF-IDF

Term Frequency – Inversed Document Frequency is a statistical method to weight
the text based on the occurrence of the term t. Term Frequency is the number of the
word appears on a single text/document compared to the total number of words in the
text, denoted as follows:
𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑎 𝑠𝑝𝑒𝑐𝑖𝑓𝑖𝑐 𝑤𝑜𝑟𝑑 𝑎𝑝𝑝𝑒𝑎𝑟𝑠 𝑜𝑛 𝑎 𝑡𝑒𝑥𝑡/𝑑𝑜𝑐𝑢𝑚𝑒𝑛𝑡

𝑇𝐹(𝑡) =
𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑤𝑜𝑟𝑑𝑠 𝑖𝑛 𝑎 𝑡𝑒𝑥𝑡/𝑑𝑜𝑐𝑢𝑚𝑒𝑛𝑡
Meanwhile Inverse Document Frequency is the logarithm of total number of the

document, divided by the number of documents in which a term t appears.
𝑇𝑜𝑡𝑎𝑙 𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝐷𝑜𝑐𝑢𝑚𝑒𝑛𝑡𝑠

𝐼𝐷𝐹(𝑡) = log10
𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝐷𝑜𝑐𝑢𝑚𝑒𝑛𝑡 𝑖𝑛 𝑤ℎ𝑖𝑐ℎ 𝑎 𝑠𝑝𝑒𝑐𝑖𝑓𝑖𝑐 𝑤𝑜𝑟𝑑 𝑎𝑝𝑝𝑒𝑎𝑟𝑠
By doing so, a word that appears in a text a lot of time will be weighted higher because
it is considered important, but if a word appears in a lot of other text/document, it will be
weighted less since it is considered as a common word.
Using Scikit-Learn’s Pipeline, I created a machine learning object that do the
following functions:
76
- Tokenize and vectorize the text using text_process function as analyzer
- Weight the vector using the TF-IDF function from Scikit-Learn
- Train the data using the Multinomial Naïve Bayes from Scikit-Learn
With this model, I got the following performance:
Table 0.6 TF-IDF Model Accuracy

ZP 1 90% 97% 93%
ZP 2 78% 75% 76%
ZP 3 100% 78% 88%
ZP 4 82% 88% 85%
ZP 5 82% 78% 80%
accuracy 84%
The model’s accuracy has been slightly improved by using the TF-IDF from 81% to 84%.
4.2.2. Adjusting the Text Process Function

In the table 4.2, we can see something odd on the table. Where the number in the
street address stated as number 1-17, but after being tokenized, the number merged into
117. It also happens with the RT and RW in the address. The RT.12/RW.5 is merged
into RT12RW5. This happens because we plainly remove any punctuations from the text,
so that after a punctuation is removed, any remaining words around the punctuation are
merged.
I adjusted the text_process function and created text_process1 function as
replacement. This function does anything the old function does, but given some
additional steps by splitting the street number e.g. “1-17” into “1”,”7” and splitting the
“RT.X/RW.Y” into “RTX”,”RWY”. Keep in mind that we don’t want to split the RT and RW
into “RT”,”X”,”RW”,”Y”. Since, it will result in weird behavior where “RT 3 RW 4” will have
50% similarity to “RT 7 RW 8”, since both has the word RT and RW. By dividing it into
“RTX”,”RWY” each RT and RW is counted as one part with the number instead of being
divided. In addition to that, I also added some step to remove some common words like
“Jakarta Utara”, “Jalan”, “Jl” since all the document has that word in every of them.
Table 0.7 Adjusted Text Process Function

Address Processed Address
Jl. Mahoni II 1-17, RT.12/RW.5, → [“Jl”, “Mahoni”, “II”, “1”, 17”, “RT12”,
Sukapura, Jakarta Utara “RW5”, “Sukapura”]
Jl. Panda Lestari II 1-3, → [“Jl”, “Panda”, “Lestari”, “II”, “1”, 3”,
RT.8/RW.9, Sukapura “RT8”, “RW9”, “Sukapura”]
Jl. Anoa Lestari II 3-8, RT.3/RW.9, → [“Jl”, “Anoa”, “Lestari”, “II”, “3”, “8”,
Sukapura, Jakarta Utara “RT3”, ”RW9”, “Sukapura”]
Using text_process1 function as replacement of the text_process, I got the
following performance:
77
Table 0.8 Model Performance Using Adjusted Text Process Function
ZP 1 94% 100% 97%
ZP 2 93% 89% 91%
ZP 3 100% 78% 88%
ZP 4 93% 96% 94%
ZP 5 91% 91% 91%
Accuracy 93%
The accuracy has been greatly increased from 84% to 93% using the adjusted
text_process1 function.
4.3. Testing Using Different Classifier

In this paper, in addition to Multinomial Naïve Bayes classifier, I also used some
other classifier to test which classifier has the highest performance. The classifier I used
in this research are Random Forest classifier and k-Nearest Neighbor classifier. The
other parameter will be using our latest model, thus utilizing the new text_process1
function and TF-IDF.
4.3.1. Using Random Forest Classifier

Random Forest is a classifier based on the Decision Tree classifier. Decision Tree
is a set of Boolean value of True or False representing presence of particular feature of
a classification.
Figure 0.1 Illustration of Decision Tree
Source: Wikipedia
However, Random Forest uses a set of multiple decision tree to make the model more
flexible and minimize the error.
78
Figure 0.2 Random Forest Illustration
Source: Wikipedia
Using the Random Forest classifier, the model has accuracy of 88%, slightly lower
than our latest model using Multinomial Naïve Bayes. Illustrated in the classification
report below:
Table 0.9 Random Forest Accuracy

ZP 1 94% 100% 97%
ZP 2 82% 82% 82%
ZP 3 71% 56% 63%
ZP 4 92% 88% 90%
ZP 5 88% 91% 89%
accuracy 88%
4.3.2. Using k-Nearest Neighbor Classifier

K-Nearest Neighbor (KNN) is a classifier used in statistical learning to determine a
class of a data based on the “nearest” class in a diagram. The distance is measured
using actual distance of each data in the diagram. KNN chooses k number of neighbor
and determine which class is present more in the dataset. For a better understanding,
here is how KNN works in illustrated diagram:
79
Figure 0.3 KNN Illustration
Source: Wikipedia
In the figure 4.3, if we try to determine the circle’s class, we can choose the number of
k-Nearest Neighbor to classify it. If we use number of k = 3, 3 nearest neighbors are 1
square and 2 triangles, making the square is classified as triangle. in the other hand, if
we choose the number of k = 5, the nearest neighbors are 3 squares and 2 triangles,
making it classified as square.
Choosing the number of k used in the KNN is the most crucial thing. Defining the
number of k too low will make the model too flexible and a little change will affect the
model in a large way, and vice versa. That’s why, I tested the number of k between 1-40,
collect all list of the error and used the elbow method to determine the right amount of k
to use.
Figure 0.4 Elbow Method to Determine the Number of K
80
In the Figure 4.4 we can see the error rate drops from high to low and then constantly
having a considerably low error rate, making shape of a folded elbow. The elbow method
encourages us to pick a number which is closest to the elbow, in this case the number
is 10. Picking the number too far to the left will result in the low model accuracy but
picking number too far to the right will most probably overfit the model. Thus, I took the
number 10 as k value which results in 91% of accuracy, very close to the Naïve Bayes
model I used earlier.
Table 0.10 KNN Model Performance

F1-
Precision Recall score
ZP 1 93% 97% 95%
ZP 2 89% 89% 89%
ZP 3 100% 78% 88%
ZP 4 89% 92% 91%
ZP 5 91% 91% 91%
Accuracy 91%
5. Conclusions
After completing this research, I can say that we could create a model to classify
Taxpayers’ supervising zone using Natural Language Processing with a considerably
high accuracy. The most effective model is the one using TF-IDF, adjusted text
processing function, and using Naïve Bayes classifier. It’s fairly predicted since
Multinomial Naïve Bayes is the most popular classifier used in NLP. Also stated in the
documentation page of Naïve Bayes on Scikit-Learn’s website, Multinomial Naïve Bayes
works best when faced with count vectorized data or TF-IDF data.
The most crucial thing in creating the model is processing the text into an acceptable
feature to be fitted in the model. This is proven that when I adjusted the text processing
function, the performance jumped from 84% to 93% which is very impressive. However,
my method of text processing can not be directly applied to differently structured text
data. Thus, a high understanding on the data structure, how changes on the feature
affects the model, and a clear vision on what the purpose of the model is a really
important key on producing a high-quality machine learning model. Reciting a saying in
the machine learning community, “Trash in, Trash out”, which means whatever data we
put in the model is the most determining thing on the model quality. That’s why, most
process of machine learning is usually spent on data preprocessing.
With high accuracy in this model, this model can be implemented in the real life when
assigning Taxpayers to the respective Account Representatives in KPP Pratama Jakarta
Koja. However, this should be furthermore discussed with the higher authority and
decision makers such as Office’s Executives. Since, reassigning and rearranging the
Taxpayers will results in change in the revenue collection policy.
6. Implications and Limitations

The model I created is limited to used in KPP Pratama Jakarta Koja, since the sample
data used are addresses around KPP Pratama Jakarta Koja’s supervising area. This
81
model is also limited to the data which is divided into each Kelurahan. Using this model
to predict all possible supervising zone in a KPP will result in different (most probably
lower) performance and accuracy.
To counter this limitation, users should be dividing the data into each Kelurahan
before using the model. Or, if some Kelurahan is unknown, we could create a model to
classify street addresses into each Kelurahan first. Using a multi-layered model is way
better than using a single model to do all the classification.
Appreciations
All praise to our lord Allah SWT and all his blessing that I could finish this paper on time. I hope
that this paper will bring benefits to other practitioners and can be replicated to help you all in
office works. Special thanks to all that participated and help me to finish this paper, especially to:
1. My beloved wife Iffah Tsurayya and son Shankara Alfariq Lakusae, who always give me
support to finish this paper.
2. My parents, Moh. Abrori (Alm) and Ibadah Yuliati Masitoh (Almh), even though both of
you are gone now, I will never forget anything you taught me.
3. Jose Portilla, author of Udemy’s course titled “Python for Data Science and Machine
Learning Bootcamp” which is my first step into data science and statistical learning.
4. My coworkers at KPP Pratama Jakarta Koja, especially my division coworkers
And all other parties who contributes to this paper creation.
REFERENCES
Bag of Words Model. Wikipedia. https://en.wikipedia.org/wiki/Bag-of-words_model
F-Score. Wikipedia. https://en.wikipedia.org/wiki/F-score
James, Gareth M. 2013. An Introduction to Statistical Learning with Applications in R.
New York: Springer.
K-Nearest Neighbor. Wikipedia. https://en.wikipedia.org/wiki/K-
nearest_neighbors_algorithm
Naïve Bayes. Scikit-Learn. https://scikit-learn.org/stable/modules/naive_bayes.html
Precision and Recall. Wikipedia. https://en.wikipedia.org/wiki/Precision_and_recall
Random Forest. Wikipedia. https://en.wikipedia.org/wiki/Random_forest
TF-IDF. Wikipedia. https://en.wikipedia.org/wiki/Tf%E2%80%93idf
82
ATTACHMENTS
All the source code and data will be available on my Github repository in this following link:
https://github.com/renoob21/NLP-DJP-Call-For-Paper
Note:
I’m using Jupyter Notebook files in the project (*.ipynb), for better experience in reading the
source code, I advise to use Jupyter Notebook. But, it’s also accessible through any Python IDE
by opening the python files (*.py).
EVALUATION OF CORPORATE INCOME TAX RATES
REDUCTION POLICY APPLIED TO THE GENERAL
EQUILIBRIUM APPROACH AND THE DEGREE OF SELF-
FINANCING METHOD: CASE STUDY IN INDONESIA
[Ryan Mohammad]a, [Helmi Zus Rizal]b

a [Directorate General of Taxes, Jakarta, Indonesia] Email: [ryanmohammad @pajak.go.id]
b [Directorate General of Taxes, Gianyar, Indonesia] Email: [helmi.zus@kemenkeu.go.id]
ABSTRACT
The reduction of the corporate income tax rate from 25% to 22% in 2020 is one of the Indonesian
government's initiatives to hasten the country's economic recovery in response to the COVID-19
pandemic. According to our estimation using the General Equilibrium approach and the Degree of
Self Financing (DSF) method, the measurement results demonstrate that nearly a quarter of 93,8%
from the total lost tax revenue resulting from the policy of lowering the corporate income tax rate will
be made up for by an increase in other tax revenues in the form of: 17.55 percent of individual
income tax revenue, 5.94 percent of VAT, and 0.000197 percent of Tax on Deposits. However,
given that only a quarter of 93,8% from the lost corporate income tax revenue will be recovered, the
government's decision to keep the corporate income tax rate at 22 percent in 2022 and beyond is
still reasonable.
Keywords: degree of self financing, general equilibrium, tax rate reduction policy
Salah satu upaya pemerintah Indonesia dalam mempercepat pemulihan ekonomi karena pandemi
COVID-19 adalah menurunkan tarif PPh Badan dari 25% menjadi 22% di tahun 2020. Berdasarkan
estimasi kami melalui pendekatan General Equilibrium dan metode Degree of Self Financing (DSF),
hasil pengukuran menunjukkan bahwa hampir seperempat dari 93,8% kerugian penerimaan pajak
akibat kebijakan penurunan tarif PPh Badan akan kembali pulih dari peningkatan penerimaan pajak
lainnya berupa: 17,55% dari penerimaan PPh Orang Pribadi, 5,94% dari PPN, dan 0,000197% dari
PPh Final atas bunga simpanan karena kegiatan ekonomi memberikan respons positif terhadap
kebijakan tersebut. Namun, dengan mempertimbangkan hanya seperempat dari 93,8% kerugian
penerimaan PPh Badan yang akan pulih, keputusan pemerintah untuk mempertahankan tarif PPh
Badan sebesar 22% di tahun 2022 dan seterusnya sudah tepat untuk menjaga porsi penerimaan
PPh Badan di APBN tetap signifikan tanpa menimbulkan distorsi yang berlebihan.
Kata kunci: degree of self financing, general equilibrium, kebijakan penurunan tarif pajak
1. INTRODUCTION
1.1. Research Background
In accordance with the Law on Harmonization of Tax Regulations, one of the
Indonesian government's efforts to support economic recovery efforts due to the COVID-
19 pandemic is a fiscal policy that reduces the corporate income tax rate from 25% to 22%,
effective in the 2020 tax year, and so on (HPP). This fiscal policy was first stated in the
Job Creation Law as a form of government assistance to ease the burden on corporations
and reduce the threat of layoffs (PHK) by employers amid the economic slowdown caused
by the pandemic. The Government of Indonesia's efforts to implement a policy to reduce
corporate income tax rates are part of the tax reform pillar, following the strategy of OECD
member countries in the form of a broad base and low-rate tax, which means lowering
rates followed by broadening the tax base. (OECD Tax Policy Reform, 2017).
The policies implemented have consequences. The main effect of the policy of
lowering corporate income tax rates is a loss of revenue. Efforts to relax a tax object, on
the other hand, will have an effect on increasing revenue from other tax objects (Gale and
Andrew, 2016). The authors intend to use the General Equilibrium approach and the
Degree of self-financing (DSF) model in this paper to assess the impact of lowering
corporate income tax rates in Indonesia on corporate income tax revenues in 2020.
2. THEORETICAL FRAMEWORK AND THE DEVELOPMENT OF

HYPOTHESIS
2.1. The General Equilibrium Approach to measure Tax Policy Distortions
Tax policies in general will cause distortions in the form of inefficiencies in market price
formation. The imposition of a high tax rate in a country imposes an additional burden on
both the producer and the consumer. This additional burden is known colloquially as
Deadweight Loss. (Hines dan James R. Jr, 1999).
Harberger (1964) employs the General Equilibrium approach in measuring
Deadweight Loss, which states that Deadweight Loss from imposing taxes on goods in a
market can be measured by how much the tax affects the price of the goods and added
by the interaction of reducing or increasing distortions due to taxes on substitutes that are
not taxed in other markets because consumers are more likely to switch to goods that are
less taxed or whose prices are not affected by the tax. The Harberger triangle, which is
formed by the area of the tax revenue box, the demand curve, and the normal supply curve
without tax, is used to calculate the total deadweight loss from taxation.
Chart 1: Harberger Triangle– Deadweight Loss

Source: Harberger (1964)
According to Goulder and Williams (2003), calculating Deadweight Loss in the General
Equilibrium approach can provide a more accurate description of the excess burden
caused by price increases due to higher taxes than the Harberger Triangle formula
(excess-burden triangle formula). However, the results will be less valid if the taxable
goods used as a measure are important or cannot be easily replaced by economic
activities based on experience or entertainment (leisure), as well as if the goods are
complementary goods or substitutes for other highly taxed goods. Consider commodity
goods. Furthermore, changes or irregularities that arise as a result of tax fluctuations that
affect the price of goods will have an impact on the active labor market.
2.2. Degree of Self Financing (DSF)

According to Srensen (2014), the formula for calculating his Deadweight Loss
considers the types of taxes that generate the most revenue in Sweden, such as Individual
Income Tax, Corporate Income Tax, VAT, and Tax on Deposit Interest. The formula is
intended to empirically measure the Deadweight Loss on the imposition of various types
of taxes and their implications for capital, labor, and taxable goods/services. Sørensen
(2014) explains that, in addition to measuring Deadweight Loss, the formula can also be
used to calculate the Degree of Self Financing (DSF), which is related to the effects of
lower tax rates. DSF is a portion of the lost tax revenue as a result of a reduction in the
rate of one type of tax, which will be recovered later through receiving of another type of
tax because economic conditions respond positively to the tax rate reduction policy. DSF
can be used as an indicator to calculate tax revenue generated by the policy of lowering
tax rates.
2.3. Research Question

Sørensen (2014) used the General Equilibrium approach to simulate DSF calculations
for Sweden. The calculation results show that the DSF level of Swedish corporate income
tax is 38.5%, which means that the reduced corporate income tax revenue as a result of
the tax rate reduction policy will recover from the increase in individual income tax, VAT,
and tax on deposit interest with a total of 38.5%.
According to the Job Creation Law, the Indonesian government reduced the corporate
income tax rate from 25% to 22% in the year 2020. In accordance with the Law on the
Harmonization of Tax Regulations, the plan to reduce the corporate income tax rate
gradually from 22% in the 2020 and 2021 tax years to 20% in the 2022 tax year was
canceled, so the policy of reducing the corporate income tax rate remains at 22%
beginning with the 2020 tax year. The policy of reducing corporate income tax rates will
undoubtedly reduce corporate income tax revenue, but the economic climate will respond
positively to this policy, resulting in an increase in tax revenues from other types of taxes
(Gale and Andrew, 2016). So that it can be interpreted that the decrease in revenue from
corporate income tax will be offset by an increase in revenue from other types of taxes,
such as individual income tax, tax on deposit interest, and value-added tax. No one has,
however, measured the level of DSF for the Corporate Income Tax rate reduction policy
using the General Equilibrium approach prior to this point.
Based on the research background and theoretical basis, the authors formulate the
problem as follows:
1. What is the DSF level of corporate income tax as a result of Indonesia's 2020
policy of lowering corporate income tax rates using the General Equilibrium
approach??
3. Research Methodology
3.1. Method of Selection and Data Collection
The author uses secondary data as follows:
1. The 2020 Indonesia’s Inflation Rate Data based on report from Bank Indonesia;
2. The 2020 BI Interest Rate Data based on report from Bank Indonesia;
3. BI’s total liabilities data for the implementation of monetary policy in 2020 based
on report by Bank Indonesia;
4. The total realization of domestic and foreign investment for 2020 data based on
the report from The Central Bureau of Statistics;
5. The data of Gross Domestic Fixed Capital Formation in 2020 based on the report
from the Central Bureau of Statistics.
6. Information on the total number of registered taxpayers for the year 2020 based
on the DGT Performance Report for 2021
7. Percentage-based data for the Indonesian consumer price index in 2020 based on
the Central Bureau of Statistics Report
8. Data on Indonesia's per capita income for 2020 is based on the Central Bureau of
Statistics Report
9. Data on profit returns on corporate bonds or stocks for 2020 based on the Pefindo
Report
3.2. Variable Operational Definitions
To answer the research question, the authors use the DSF calculation formula for
corporate income tax for the policy of lowering corporate income tax rates using the
General Equilibrium approach, as follows:
Sørensen (2014) described each variable in the formula as follows:

𝜀𝐿𝑊 wage elasticity in relation to the number of active workers
Tw the effective marginal rate of individual income tax with a deduction for social
insurance purposes
Tc the effective marginal rate of the VAT rate
Mk the effective marginal rate of the corporate income tax rate
𝜀𝐾𝑃 the level of capital to capital user cost elasticity
𝜌 the capital usage cost, consist of the sum of Indonesia's rate of return on capital,
asset depreciation, and corporate income tax, all measured per unit of capital.
𝛿 the level of asset depreciation (using average method)
𝜃𝑘 proportion of investment to salary wages
𝑠
𝜃 proportion of savings to wages and salaries
𝜀𝑊
𝑆 level of wage elasticity of savings
Tr the effective marginal rate of the tax rate on deposit interest
3.3. Data Analysis Method

Sørensen (2014) quantifies each variable in the previously described formula before
calculating the DSF. The obtained secondary data is pre-processed in order to determine
the effective marginal rate of personal income tax rates, the effective marginal rate of VAT
rates, the effective marginal rate of corporate income tax rates, the marginal rate of tax
rates on effective deposit interest, wage elasticity on the number of active workers, wage
elasticity to savings, and the elasticity of the user's cost of capital to capital. Following
quantification of these variables, they are summed according to the formula to determine
the DSF level of personal income tax, corporate income tax, tax on deposit interest, and
VAT due to the corporate income tax rate reduction policy.
4. Result and Discussion

4.1. Determination of the Effective Marginal Rate of Individual Income Tax Rates
The following formula is used by the author to calculate the effective marginal rate of
personal income tax rates:
According to Sørensen (2014), “s” is the total deduction allowed in calculating taxable
income related to employees' social security needs when they retire because the formula
considers the cycle of individual taxpayers when they are actively working and when they
are entering retirement. Assuming that individual employees and individual entrepreneurs
will set aside around 5% of their gross income for retirement purposes, the author uses
the maximum total pension contributions per year as the value of s, which is 5% of gross
income in one year. twp is an individual progressive tax rate with a weighted average of
18.75%. Based on the quantification results, the effective marginal rate of individual
income tax is:
tw = 0,05 + (1 – 0,05) 0,1875

= 0,05 + 0,178125
= 0,228125
4.2. Determination of Corporate Income Tax Rates' Effective Marginal Rate

The following formula is used by the author to calculate the effective marginal rate of
corporate income tax rates:
𝜌 is the average value of 𝜌d and 𝜌e. 𝜌d represents the value of the user cost of capital
arising from debt, whereas 𝜌e represents the value of the user cost of capital arising from
investment. Using depreciation data (𝛿 ) of 0.106, data on the percentage difference
between fiscal depreciation and gross domestic fixed capital formation growth (a) of
6.244%, data on corporate income tax rates (𝜏) of 16.5%, data on Indonesia's inflation in
2020 (𝜋) of 2.03%, and data on Indonesia's net return on corporate capital (r) of 6.78%,
the calculations for 𝜌, 𝜌d, and 𝜌e are as follows:
𝜌d = (0,0678 - (0, 165/(1 - 0, 165)) (0,0203 + (0,0678)(0,0624)) + 𝛿

= 0,0678 – (0, 165/0,835) (0,0226) + 𝛿
= 0,0678 – 0,0044 + 𝛿
= 0,0634 + 0,106
= 0,169
𝜌e = (0,0678/0,835) (1 – ((0, 165) (0,0624))) + 0,106

= ((0,081) (0,989)) + 0,106
= 0,186
𝜌 = (𝜌d + 𝜌e)/2
= (0,169 + 0,186)/2
= 0,177
After 𝜌 determined, the effective marginal rate of corporate income tax rates (mk) can be
determined as follows:
mk = (0,177 - 0,106 - 0,0678) / (0, 177 - 0,106)
= 0,0037/0,071
= 0,052
4.3. Determination of the Effective Marginal Rate of VAT Rates

The following formula is used by the author to calculate the effective marginal rate of VAT
rates:
𝑡ℎ𝑐 adalah tarif PPN efektif atas Barang Kena Pajak atau Jasa Kena Pajak (BKP/JKP) yang
berhubungan dengan sewa atau jual-beli rumah sejumlah 10%, sedangkan 𝑡0𝑐 adalah tarif
PPN efektif atas BKP/JKP barang selain yang berhubungan dengan sewa atau jual-beli
rumah. Tingkat distorsi bukan berasal dari pengenaan pajak yang mempengaruhi harga
BKP/JKP (𝛽ℎ) diasumsikan sebesar 0 karena pengenaan PPN di Indonesia dilakukan
secara menyeluruh dengan tarif tunggal sehingga distorsi yang timbul hanya dari
pengenaan pajak saja. Penghitungan tingkat marginal efektif tarif PPN adalah sebagai
berikut:
𝑡ℎ𝑐 is the effective VAT rate on Taxable Goods or Taxable Services (BKP/JKP) related to
renting or buying and selling houses in the amount of 10%, while 𝑡0𝑐 is the effective VAT
rate on BKP/JKP goods other than those related to renting or selling houses. The level of
distortion does not arise from the imposition of taxes affecting the price of BKP/JKP (𝛽ℎ)
assumed to be 0 because VAT is imposed in Indonesia at a single rate across the board,
resulting in distortions arising solely from the imposition of taxes. The effective marginal
rate of VAT rates is calculated as follows:
tc = (1) (10%) + (1 – 1) 10%

= 10% = 0,1
4.4. Determination of the Effective Marginal Rate of Income Tax Rate on Savings
Interest
The following formula is used by the author to calculate the Effective Marginal Rate of
Income Tax Rate on Savings Interest:
Before calculating the effective marginal rate of the Income Tax rate on deposit interest
(tr), the author first calculates the personal deposit rate (tF) and the institutional deposit
rate (tt) with a 2020 inflation rate (𝜋) is 2.03%, the 2020 interest rate of The Central Bank
of Indonesia (r) is 3.75%, the deposit interest tax rate (tIs) is 20%, and the income tax rate
on dividends (tFs) is 10% using the formula explained by Sørensen (2014) as follows:
tt = (0,2 (0,0375 + 0,0203)) / 0,0375
= 0,3082
tF = (0,1 (0,0375 + 0,0203)) / 0,0375
= 0,1541
After determining (tF) and (tt), the author calculates the effective marginal rate of Income
Tax on deposit interest with an elasticity level of personal and institutional savings (∅) of
1, the total realization of foreign and domestic investment (SF) in 2020 is 826.3 trillion
rupiah, and Central Bank total liabilities for monetary policy implementation (SI) in 2020 is
1,330.3 trillion rupiah using the formula explained by Sørensen (2014) as follows:
sI = 1/1 + (1-0,1541/1-0,3082) (826,3/1.330,3)

= 1/1 + (0,8459/0,6918) (0,6211)
= 1/1,759
= 0,568
𝛾 = 0,568 (0,6918) –2/0,568 (0,6918) –2 + (0,432) (0,8459) –2
= (0,568/0,69182) / (0,568/0,69182) + (0,432)/(0,8459) 2
= (0,568/0,478) / (0,568/0,478) + (0,432)/(0,715)
= 1,188 / (1,188 + 0,604)
= 1,188 / 1,792
= 0,662
1 - tr = [(0,662)(0,478) + (0,338)(0,715)] 0,5
= [0,316 + 0,241] 0,5
= 0,746
tr = 0,254
4.5. Determination of Wage Elasticity of Labor

The following formula is used by the author to calculate the Wage Elasticity of Labor:
Previously, the author calculated the value of labor elasticity (w) using the consumer price
index in percentage (P) for 2020 of 1.56, per capita income (W) for 2020 of 54,580,000,
and the number of registered active workers with Taxpayer Identification Number (NPWP)
(L) in 2020 of 45.43 million using the formula explained by Sørensen (2014) as follows:
w = (54.580.000) (1 - 0,228))/1,56
= 27.010.102
𝐿
After determining (w), then wage elasticity of labor (𝜀𝑊 ) is as follows:
𝐿
𝜀𝑊 = (27.010.102) (1) / (1) (45.430.000)
= 0,594
4.6. Determination of Wage Elasticity of Savings

The following formula is used by the author to calculate the Wage Elasticity of Savings:
𝑆
The calculation of wage elasticity on savings ((𝜀𝑊 ) is as follows, using a known (w) value
of 27,010,102 and a (S) value of 1,330,363,586,000,000:
𝑆
𝜀𝑊 = (1) (27.010.102)/ (1.330.363.586.000.000) (1)
= 0,0000002
4.7. Determination of Capital Demand Elasticity on Capital User Costs

The author calculates the wage elasticity of capital demand on the cost of using capital
(𝜀𝑃𝐾 ) using the assumptions used by Auerbach and Kotlikoff (1987), with a value of 𝜀𝑃𝐾 of
1 because this value is consistent with empirical observations that conclude that the
overall profit percentage of GDP is constant in the long run.
4.8. Determination of the Savings Portion to Salary Wages

The following formula is used by the author to calculate the Savings Portion to Salary
Wages:
Previously, the authors calculated the value of future consumption prices (p) in advance
using the following calculations.
p = 1 / 1 + 0,0375 (1 - 0,254)
= 1 / 1 + 0,0279
= 0,972
Once the p-value is determined, the percentage of savings to wages (𝜃 𝑠 ) is calculated as
follows:
𝜃 𝑠 = (0,972) (0,0375) (1,565) (1.330.363.586.000.000) / (45.430.000)(54.580.000)

= 756.471.342.271.320 / 2.479.569.400.000.000
= 0,30
4.9. Determination of Investment Portion to Salary Wages

The following formula is used by the author to calculate the Investment Portion to Salary
Wages:
𝜃 𝑘 = (0, 144 - 0,106) (4.897.050.000.000.000) / (45.430.000) (54.580.000)

= (186.087.888.220.000) / (2.479.569.400.000.000)
= 0,075
4.10. Determination of DSF

Setelah seluruh variabel yang diperlukan dalam menentukan DSF terkuantifikasi, penulis
melakukan penghitungan DSF dari tiap objek pajak sesuai dengan formula yang
dijelaskan Sørensen (2014) sebagai berikut:
After quantifying all of the variables required, the authors calculate the DSF for each tax
object using the formula described by Sørensen (2014) as follows:
DSF Individual Income Tax Revenue

𝐿
(𝜀𝑊 𝑡𝑤 )
={ }
1−𝑡𝑤
= (0,228125) (0,594) / (1 – 0,228125)

= 0,13550625 / 0,771875
= 0,1755
= 17,55%
DSF Value Added Tax Revenue

𝐿
(𝜀𝑊 𝑡𝑐 ( 1−𝑡𝑤) )
={ 1−𝑡𝑤
}
= (0,1) (1 - 0,228125) (0,594) / (1 - 0,228125)

= 0,0594
= 5,94%
DSF Corporate Income Tax Revenue

𝐾 𝜌− 𝛿 𝐿
(𝑚𝑘 ( 𝜀𝑃 ( )+ 𝜃𝑘 𝜀𝑊 )
𝜌
={ 1−𝑡𝑤
}
= ((0,052) (1) ((0,177 - 0,10625)/ 0,177) + (0,075) (0,594)) / (1 - 0,228125)

= (0,0036) + (0,04455) /0,771
= 0,062
= 6,2%
The DSF from Corporate Income Tax revenue is 6.2% shows that the total loss of
Corporate Income Tax revenue due to a decrease in the Corporate Income Tax rate is
93.8% because a decrease in tax rates has an effect on expanding the tax base (Gale dan
Andrew, 2016).
DSF from Income Tax Revenue on Deposit Interest

𝑆
(𝑡𝑟 𝜃𝑠 𝜀𝑊 )
={ 1−𝑡𝑤
}
= (0,254) (0,30) (0,0000002) / (1 - 0,228125)

= 0,00000197
= 0,000197%
The total DSF for the policy of reducing corporate income tax rates in 2020 is as follows:
The calculation results show that individual income tax revenue with a DSF rate of 17.55%,
VAT revenue with a DSF rate of 5.94%, and revenue from Income Tax on deposit interest
with a DSF rate of 0.000197% will recover 100% of the total decrease in corporate income
tax revenue reduced by 6.2% DSF from corporate income tax itself.
5. Conclusion
According to our calculations, using Sørensen formula (2014), the total DSF of
Corporate Income Tax, Individual Income Tax, VAT, and Income Tax on deposit interest
for the policy of reducing Corporate Income Tax rates in 2020 is 29.690197 percent. These
calculations indicate that 93.8% of the decrease in corporate income tax revenue caused
by the policy of reducing corporate income tax rates will be offset by an increase in revenue
from other types of taxes, namely 17.55 % from individual income tax, 5.94 % from VAT,
and 0.000197% of Income Tax on deposit interest.
6. IMPLICATION AND BOUNDARIES

The results of the DSF calculation of corporate income tax have implications for the
policy of reducing corporate income tax rates in 2020, as the percentage of tax revenue
returns from personal income tax ranks highest among the other variables. According to
the mandate of the Job Creation Act, this indicates that the policy of reducing the
Corporate Income Tax rate stimulates economic activity through an increase in active
employment. The government's decision to maintain the Corporate Income Tax rate at a
fixed level of 22% and not reduce it to 20% in 2022 in accordance with the Law on the
Harmonization of Tax Regulations is deemed appropriate because reducing the Corporate
Income Tax rate to 20% carries the risk of decreasing the proportion of Corporate Income
Tax revenue that is greater in the state budget posture, taking into account that only a
quarter of the 93.8% lost corporate income tax revenue can be replaced by other sources.
Due to limited resources, time, and data, the author focuses solely on DSF
measurements of the corporate income tax rate reduction policy in this study. As
suggestions for future research, the authors suggest measuring Deadweight Loss for the
policy of increasing VAT rates in 2022 using the General Equilibrium approach as soon as
the necessary data becomes available.
Appreciation
We appreciate Indonesia Endowment Fund for Education or Lembaga Pengelola Dana Pendidikan
(LPDP) assistance with the research for this article.
BIBLIOGRAPHY
Auerbach, A. J. dan L. Kotlikoff (1987). Dynamic Fiscal Policy. Harvard University Press,
Cambridge, Massachusetts.
Gale G. William dan Andrew A. Samwick (2016). Effects of Income Tax Changes on
Economic Growth. Economic Studies at Brookings September 2014.
Goulder, L.H. dan R.C. Williams (2003). The Substantial Bias from Ignoring General
Equilibrium Effects in Estimating Excess Burden. Journal of Political Economy 111,
898-927
Harberger, A.C. (1964). The Measurement of Waste. American Economic Review 54, No.
3, 58-76
Hines, James R., Jr. (1999). "Three Sides of Harberger Triangles" (PDF). Journal of
Economic Perspectives. 13 (2): 167–188. doi:10.1257/jep.13.2.167
Sørensen, P. B. (2004). Measuring Taxes on Capital and Labor: An Overview of
Methods and Issues. Chapter 1 in P.B. Sørensen (ed.): Measuring the Tax
Burden on Capital and Labor. Cambridge, MA: MIT Press
Sørensen, P. B. (2010a). Swedish Tax Policy: Recent Trends and Future Challenges.
Report to the Expert Group on Public Economics (ESO) 2010:4.
Regeringskansliet, Finansdepartementet, Stockholm
Sørensen, P. B. (2011). Calibrating the life cycle model. Technical appendix to paper on
Measuring the Deadweight Loss from Taxation in a Small Open Economy,
Department of Economics, University of Copenhagen, October 2011.
Sørensen, P. B. (2014). Calculating the Deadweight Loss from Taxation in a Small Open
Economy. Report to the Expert Group on Public Economics (ESO) 2010:4
Supplement. Regeringskansliet, Finansdepartementet, Stockholm.
Organisation for Economic Co-Operation and Development (2017): Tax Policy Reforms
2017 OECD and Selected Partners Economies. OECD Publishing
https://doi.org/10.1787/9789264279919-en
DETERMINANTS OF TAX MORALE (EMPIRICAL STUDY
ON MSME IN JAVA)
Rama Daneshwaraa, Riko Riandokob

aDirectorate General Of Taxes Email: rama.daneshwara@pajak.go.id
b Department of Accounting, State Finance Polytechnic STAN Email: riandoko@stan.ac.id
ABSTRACT
One of the reasons Indonesia's tax ratio has not reached the target is the informal sector and tax
avoidance. The individual's desire to pay or avoid taxes is related to tax morals. This study aims to
determine the effect of trust in the government, trust in tax administration, national pride, the
likelihood of being caught in tax evasion efforts, and the level of punishment on tax morale. The
research object is the owner of the micro, small and medium enterprises (MSMEs) in the Java Island
region. The research approach used is quantitative research. The data used are primary data
consisting of 395 respondents' answers collected. Data analysis used the Ordinary Least Squares
(OLS) method. The test results state that trust in the government, trust in tax administration, national
pride, likelihood of being caught in tax evasion efforts, and the level of punishment simultaneously
and partially have a significant positive effect on tax morale.
Keywords: Tax morale, Trust, National pride, Likelihood of being caught, Severity of punishment.
Salah satu penyebab tax ratio Indonesia belum mencapai target adalah sektor informal dan
penghindaran pajak. Keinginan individu untuk memilih membayar atau menghindari pajak berkaitan
dengan moral pajak. Penelitian ini bertujuan untuk mengetahui pengaruh kepercayaan terhadap
pemerintah, kepercayaan terhadap administrasi pajak, kebanggaan nasional, peluang tertangkap
dalam upaya penggelapan pajak, dan tingkat hukuman terhadap moral pajak. Objek penelitian
adalah pemilik usaha mikro, kecil, dan menengah (UMKM) di wilayah Pulau Jawa. Pendekatan
penelitian yang dilakukan adalah penelitian kuantitatif. Data yang digunakan merupakan data primer
yang terdiri dari 395 jawaban responden yang berhasil dikumpulkan. Analisis data menggunakan
metode Ordinary Least Squares (OLS). Hasil pengujian menyatakan bahwa kepercayaan terhadap
pemerintah, kepercayaan terhadap administrasi pajak, kebanggaan nasional, peluang tertangkap
dalam upaya penggelapan pajak, dan tingkat hukuman secara simultan dan parsial memiliki
pengaruh positif signifikan terhadap moral pajak.
Kata kunci: Moral pajak, Kepercayaan, Kebanggaan nasional, Peluang tertangkap, Tingkat hukuman.
1. INTRODUCTION
Indonesia's tax ratio has not been able to reach the target that has been set. The
Ministry of Finance of the Republic of Indonesia, in the 2019 APBN outlook, has set
Indonesia's tax ratio target of 11.1%. Meanwhile, according to the International Monetary
Fund (IMF), a developing country like Indonesia should have a tax ratio of 15% in order to
be able to finance its expenses (Gaspar et al., 2019). However, the realization of
Indonesia's tax ratio only reached 10.7% in 2019 (Directorate General of Taxes, 2019).
One of the causes of Indonesia's low tax ratio, mentioned by the OECD in the 2019
Revenue Statistics in Asia and Pacific Economies report, is the informal sector and tax
avoidance. Some large-scale companies legally enforce tax avoidance to avoid paying
large amounts of tax. Small-scale companies also apply this practice. However,
Mickiewicz et al. (2019) stated that companies with a smaller scale tend to reduce their
tax obligations with a different strategy, namely committing tax evasion by engaging in the
informal sector economy or not reporting all of their activities to the tax authorities.
In Indonesia, the participation of MSME actors in the tax amnesty program indicates
that there are still business activities or income that has not been reported to the
Directorate General of Taxes. According to the 2017 annual report of the Directorate
General of Taxes, the tax amnesty program was attended by 973,426 taxpayers, with
details of 111,415 MSME Corporate Taxpayers (11.45%) and 322,189 MSME Individual
Taxpayers (33.10%). Comparing the number of MSMEs participating in the tax amnesty
with the total MSMEs registered in Indonesia shows that there is still potential for tax
revenue from MSME activities.
Tax evasion is one of the challenges in the tax system adopted by Indonesia. The tax
reform in 1983 introduced the tax system used in Indonesia, the self-assessment system.
This system requires business actors to play an active role in fulfilling their tax obligations.
According to Alm & McClellan (2012), the decision of a business to choose to pay taxes
according to applicable regulations or evade taxes depends on the role and intentions of
the business owner. That decision is also related to the motivation of the business owner.
Individual intrinsic motivation to comply with and pay taxes is defined by Torgler &
Schneider (2005) as tax morale. Lago-Peñas & Lago-Peñas (2010) revealed that business
actors have lower tax morale than those who work as employees. In contrast, research by
Blanthorne & Kaplan (2008) states that individuals with higher moral levels are less likely
to evade taxes.
Many factors influence individual tax morale, but this study adopted the neo-
institutional framework by Scott (2014), according to Mickiewicz et al. (2019), to explain
how the environment affects the tax morale of business actors. This framework is still
rarely used in tax moral research in Indonesia. The neo-institutional framework also
provides a broad perspective on how institutions influence individual behavior. As stated
by the pillars of regulation, institutions influence individual behavior through sanctions and
influence normative judgments about how something should be done. Furthermore,
institutions provide a cultural-cognitive pillar that assumes individuals adopt specific social
identities. This identity defines their role in a group, limiting their behavior.
The normative pillar relates to the growing sense of trust resulting from the
community's assessment or evaluation of an institution carrying out its duties and
responsibilities. The cultural-cognitive pillar is related to the social identity held by the
individual, which will later affect the individual's behavior. Some examples of social identity
are nationalism and patriotism. Furthermore, the regulatory pillar is concerned with
monitoring, sanctioning, and making rules that can influence an individual's decisions or
behavior.
Most of the previous studies only discussed tax morale as an unknown cause of the
emergence of tax evasion without considering how tax morale arose or was maintained
(Feld and Frey, 2002). Based on this, this study aims to determine the effect of trust in the
government and tax administration (normative pillar), national pride (cultural-cognitive
pillar), and the likelihood of being caught as well as the level of punishment (regulative
pillar) on the tax morale of business actors.
One of the social obligations to the state is paying taxes. Business owners pay taxes
to support the government and, at the same time, expect the benefits. Therefore, the
government is expected to act reasonably and reliably in managing taxes (Frey & Torgler,
2007). According to Hug & Spörri (2011), positive government performance evaluation
results will encourage business owners to be more cooperative with tax policies set by the
government. When business owners believe that the government and tax authorities have
worked well, a sense of trust will grow.
Furthermore, the slippery slope theory (Kirchler et al., 2008) states that high individual
trust can foster the individual's compliance to pay taxes. However, if business owners
perceive the government and tax authorities as untrustworthy and dishonest, this can
undermine positive attitudes toward tax compliance (Murphy, 2004). Government
authorities will be damaged when business owners feel that their contributions and
benefits received from taxes are not balanced (Alon & Hageman, 2013). In this situation,
business actors will tend to migrate to the informal sector or commit tax evasion by not
reporting their business activities.
The research of Mickiewicz et al. (2019) has found a significant positive relationship
between trust in the government and tax morale. This study emphasizes the need to
distinguish between trust in government and the specific impact of confidence in the tax
system itself. This study argues that trust in government and, more specifically, in those
aspects of government that is most directly related to taxes, such as the tax office and the
taxation system, will affect the tax morale of business owners. This opinion is in line with
research conducted by Panjaitan (2019), Basri & Al Azhar (2017), Khaerunnisa & Wiratno
(2014), and Lago-Peñas & Lago-Peñas (2010), which state that trust has a significant
positive effect on tax morale.
H1: Business owners who express trust in the government have higher tax morale.
H2: Business owners who trust tax administration have higher tax morale.
Individuals adopt a social identity that defines their role, limiting how they behave.
Social identity theory (Tajfel, 1972) assumes that what an individual does is a
manifestation of collective values usually preserved in a social group in which the
individual is a part. Hogg et al. (1995) also stated that social identity implies that individuals
see themselves as members of a social group and that individual behaviour adopts the
perspective of that group. Some forms of cooperative behaviour become taken for granted
and are no longer questioned at a deeper level of identification. On the other hand, loose
identification leads to the possibility of non-compliance (Scott 2014).
The sense of belonging and pride in being part of the Indonesian state shapes what
should be done as an Indonesian citizen. Studies on tax morale have found that indicators
of social identities, such as patriotism and nationalism, are positively related to tax morale
(Heinemann, 2011). Research by Panjaitan (2019) and Torgler & Schneider (2005)
concluded that the greater the pride of national identity, the higher the individual's intrinsic
motivation to pay taxes. Khaerunnisa & Wiratno's (2014) research also mentions that a
person's feelings of love for his country will foster a sense of pride, motivating a person to
do good things for his country. Based on the results of the empirical research, the following
hypothesis is formulated.
H3: Business owners with tremendous pride in their country have higher tax morale.
According to Scott (2014), the regulatory process is the formulation of rules,
monitoring, and sanctioning activities. The cause of a person being disobedient or
choosing to evade taxes, according to Allingham & Sandmo (1972), is a trade-off between
the profits obtained with the possibility of being investigated and fined. Rational choice
theory (Becker, 1968) also assumes that individuals have preferences among several
choices that allow the individual to state the desired option by considering available
information, the possibility of events, and the potential costs and benefits of the available
alternatives. This is reinforced by the research of Heinemann (2011), which argues that
tax morale is influenced by an egocentric bias which states that a person will adjust his
behaviour based on their interests. Furthermore, Blanthorne & Kaplan (2008) suggest that
taxpayers have lower tax morale in a job with a more significant opportunity to avoid taxes.
In slippery slope theory, prevention factors are also part of the authorities' authority,
affecting individual compliance in paying taxes. Therefore, the prevention factor affects a
person's tax morale. Sanctions are also considerations for someone to pay obediently or
evade taxes. The research of Mickiewicz et al. (2019) concludes that someone who thinks
the chance of being caught is greater when committing tax evasion and the severity of the
punishment received has a significant positive effect on tax morale. Therefore, this study
formulates the following hypothesis.
H4: Business owners who perceive a greater likelihood of being caught in tax evasion tend
to have higher tax morale.
H5: Business owners who perceive tax evaders to be punished more severely when
caught tend to have higher tax morale.
3. METHOD, DATA, AND ANALYSIS
3.1. Sample, Selection, and Data Sources
This study wanted to observe the effect of several factors based on Mickiewicz et al.
(2019) research on the tax morale of micro, small and medium enterprises in Java. MSME
owners were chosen as the population because, according to Mickiewicz et al. (2019),
small-scale companies tend to practice tax evasion by not reporting all their business
activities compared to large companies that avoid taxes legally to minimize their tax
obligations. According to data from the Ministry of Cooperatives and SMEs, MSMEs also
significantly contribute to Indonesia's GDP, reaching 60.51% and absorbing 96.92% of the
workforce in 2019. The composition of micro, small and medium enterprises also makes
up the majority of business units in Indonesia. There are 65.46 million units in Indonesia
or the equivalent of 99%. MSMEs are one sector that it is difficult to monitor the fulfillment
of their tax obligations. This difficulty is because most MSMEs are still engaged in the
informal sector (Ministry of Finance, 2020).
Several reasons are underlying the research locations throughout the province of Java
Island. First, several island areas are met by people from various ethnic groups, customs,
religions, and different backgrounds. In the first quarter of 2020, Indonesia's economic
structure was partially dominated by groups of provinces on the island of Java, with a
contribution of 59.14% to Indonesia's GDP (Central Bureau of Statistics, 2020).
Furthermore, based on the 2014 report on the competitiveness of MSMEs in Indonesia by
Bappenas, it is explained that the productivity index is relatively high for each province in
the Java Island region, including DKI Jakarta (100), West Java (27.09), Banten (24.56). ),
East Java (14.71), Central Java (13.32), and DI Yogyakarta (12.12).
This research is quantitative research using primary data obtained through
questionnaires. The sample selection technique uses a non-probability sampling
approach, namely accidental sampling. Accidental sampling can be interpreted as a
sampling technique based on coincidence. Anyone willing to fill out a research
questionnaire suitable as a data source can be used as a sample (Sugiyono, 2018). A
research sample is some MSME owners who are domiciled in a province located on the
island of Java and are willing to fill out a questionnaire via a google form. The number of
sample members is often expressed by the sample size and is determined by the Krejcie
and Morgan formula in this study. According to this formula, for a population of more than
1 million, 384 samples are needed.
3.2. Variables
The variables used in this study adopted Mickiewicz et al. (2019) research are as
follows.
Table 1 Variable
No. Variable Name Description
Dependent Variable
1 Tax morale Is it justifiable to do tax fraud when the opportunity arises?
Independent Variable
1 Trust in Government How much do you trust the government?

1. Tax authorities are corrupt (reverse coded).
Trust in Tax
2 2. The tax authorities manage taxation fairly.
Administration
3. Tax authorities behave honestly.
3 Nationalism Being an Indonesian citizen is important to me
1. What are the chances of someone being caught for underreporting

operating income?
Likelihood of being 2. What is the probability that someone is caught for underreporting the
4
caught number of employees?
3. What are the chances of someone being caught for underreporting

an employee's salary?
Severity of What is the impact on companies that receive penalties for reporting
5
punishment less income than they should?
Control Variable
1 Female
2 Education
3 Age
4 Employees
5 Sector
6 Performance
Source: Adapted from Mickiewicz et al. (2019)
3.3. Data Analysis

The research method used is a quantitative method with the following basic
equations.
Tax moralei,t = α + β1Trust in governmenti,t + β2Trust in tax administrationi,t + β3
Nationalismi,t+ β4Likelihood of being caughti,t + β5Severity of
punishmenti,t + Female + Education + Age + Employees + Sector +
Performance + ε
Hypothesis testing in this study uses Ordinary Least Squares (OLS), which begins with
testing the research instrument and must meet Gaus Markov's assumptions. The research
instrument testing conducted on the data from the questionnaire included validity and
reliability tests. Research data must also be BLUE (Best Linear Unbiased Estimator).
Furthermore, hypothesis testing uses the coefficient of determination test, simultaneous
test (f-test), and partial test (t-test).

4.1. Result
Research data has been collected through the distribution of questionnaires as many
as 395 respondents' answers. The results of the questionnaire collection showed the
profile of the respondents consisting of gender, age, education, business sector, and
business domicile. Respondents of female sex dominated as many as 246 respondents
and male respondents as many as 149 respondents.
Based on age, respondents were grouped into three categories. In the first category,
with an age range of 18 – 31 years, there are 170 respondents. In the age range of 32-44
years, there are 115 respondents. Respondents over 45 years of age are in the minority,
with 110 respondents.
Based on education level, research data is dominated by respondents with the latest
education S1/Diploma IV with 180 respondents. Furthermore, the education level is
Masters with 64 respondents (16.2%), Diploma with 59 respondents (14.9%),
SMP/Equivalent with five respondents (1.3%), and SD/Equivalent with one respondent or
0.3 % of the total respondents.
Based on the business sector, respondents were dominated by other sectors, with 194
respondents or 49.1%. The second-largest business sector is retail trade, with 122
respondents or 30.9%. Respondents with the service business sector are 51 (12.9%),
wholesale trade is 22 (5.6%), and manufacturing is 6 or 1.5% of the total respondents.
Based on the business domicile, most respondents were domiciled in the DKI Jakarta
area, namely 213 or 53.9%. Furthermore, the area with the second-largest respondent is
West Java, with 43 respondents or 15.7%. The distribution of data for other regions in
Central Java with 43 respondents (10.9%), East Java with 32 respondents (8.1%), Banten
with 27 respondents (6.8%), and DI Yogyakarta with the least respondents with 18
respondents or 4.6% of the total respondents.
4.2. Regression Analysis

Table 2 Regression
Variabel Unstandardized Coef. T Sig.
B Std. Error Two-tailed One-tailed
(Constant) 1.853 0.953 1.945 0.053 0.026
Trust in government 0.524 0.162 3.242 0.001 0.000
Trust in tax administration 0.504 0.149 3.380 0.001 0.000
Nationalism 0.256 0.143 1.794 0.074 0.037
Likelihood of being caught 0.351 0.154 2.272 0.024 0.012
Severity of punishment 0.373 0.137 2.728 0.007 0.003
Adj r2 0.155
F (sig) .000b
Source: Processed from SPSS 25
Based on Table 2, the value of the coefficient of determination (R2) is 0.155. This
coefficient means that the independent variables and control variables in this research
model can explain the variation of the dependent variable by 15.5%. The significance
value (F sig) is 0.000 < 0.05, so H0 is rejected, and H1 is accepted so that it can be
concluded that the independent variables have a simultaneous effect on tax morale.
Based on the results of the t-test in Table 2, the significance value is less than alpha 0.05,
so H0 is rejected, and H1 is accepted. Therefore, all research hypotheses were accepted.
4.3. Discussion
4.3.1. The effect of trust in government on tax morale
Business owners pays taxes as a form of obligation to the government so that later
they can reap the benefits of their contributions. This study found that a business owner
who believed in the government tended to have higher tax morale. The results of this study
are supported by the slippery slope theory (Kirchler et al., 2008), which assumes that trust
in the government will make business owners obey existing regulations, including
compliance with tax obligations because they believe that the provisions made by the
government will have a good impact. The research of Mickiewicz et al. (2019) states that
tax morale can be improved by having confidence in how the tax is collected and spent as
government expenditure has gone well and correctly.
A positive assessment of the evaluation of the implementation of taxes makes
business owners cooperative with government policies (Marien & Hooghe, 2011). Based
on the questionnaire data, it is known that 49.4% of business owners expressed
confidence and 7.1% of business owners expressed greater confidence in the
government's performance. Business owners who carry out tax obligations because they
believe in the government indicate the emergence of social legitimacy. Social legitimacy
is seen as quite important, Mickiewicz et al. (2019) add that the lack of social legitimacy
will cause business owners to choose not to pay taxes or evade their taxes by not reporting
all of their business activities.
Khaerunnisa & Wiratno's (2014) research finds that when the level of trust in the
government is at the highest position, citizens will be motivated to carry out their tax
obligations. This is due to their confidence in the government that the income from this tax
will help in development. Other studies, Basri & Al Azhar (2017), Hug & Spörri (2011), and
Lago-Peñas & Lago-Peñas (2010), have similar results, trust in the government has a
positive influence, which means the more significant the trust, the tax morale will increase.
4.3.2. The effect of trust in tax administration on tax morale

When a business owner states that he believes the tax administration has been
running fairly, he tends to have higher tax morale. The conclusions resulting from this
study are in line with Mickiewicz et al. (2019), which concludes that trust in tax
administration positively affects tax morale. As has been explained in the previous
hypothesis regarding the trust in the government variable, the results of the evaluation of
the implementation of tax administration also affect the behavior of business owners.
According to Murphy (2004), a person's tendency to choose to cooperate and comply with
predetermined behavior is influenced by a sense of obligation. This feeling can increase
or decrease depending on the evaluation of the relevant institution. According to
Mickiewicz et al. (2019), perceptions of justice are essential and can be achieved if there
is no arbitrariness and favoritism in tax services.
Poor administrative practices can lead to corruption and favoritism. If business owners
think that the tax authorities are dishonest, corrupt and do not believe the rules used in
collecting tax revenue and spending it are fair, it can weaken the positive attitude of paying
taxes and justify tax evasion. This is confirmed by the findings of Torgler et al. (2008), who
found empirical evidence regarding the assessment of the tax collection process that is
honest, not corrupt, and fair associated with high tax morale. Mickiewicz et al. (2019) also
state that if business owners think that the benefits they receive are not balanced with
their contribution, it will encourage business owners to migrate to the informal sector.
Based on the questionnaire data, it can be concluded that most respondents think that
the implementation of tax administration in Indonesia has been running honestly and fairly.
In line with the slippery slope theory, which argues that the emergence of trust will
encourage business owners to obey the rules. Therefore, the tax authority should improve
its image by providing information regarding how the tax is collected and spent effectively
and efficiently.
4.3.3. The effect of nationalism on tax morale

When a business owner expresses an intense pride in his country, he tends to have
higher tax morale. Social identity theory (Tajfel, 1972) assumes that what an individual
does is a manifestation of collective values usually preserved in a social group in which
the individual is a part. This theory is supported by Hogg et al. (1995), which states that
social identity implies that individuals see themselves as members of a social group and
that individual behavior adopts the perspective of that group. Nationalism is seen as the
self-perception of individuals who feel they belong to a particular group of countries and
believe they must contribute to that group. Mickiewicz et al. (2019) also state that external
cultural systems shape individuals' interpretations of their environment and guide how they
should act.
A sense of pride in being a citizen increases the motivation of business owners to
contribute to the country by being obedient in paying taxes. This conclusion is supported
by research by Khaerunnisa & Wiratno (2014), which states that one's love for Indonesia
will create a feeling of pride in being a citizen so that this factor can motivate a person to
do good things for his country to ensure that his country can run well.
Based on the questionnaire data, it is known that most of the respondents stated that
being an Indonesian citizen is essential for them. The results of the descriptive analysis
also explain that the respondents have relatively high nationalism. It implies that the
people are proud to be Indonesian citizens. Feelings of pride make people behave
cooperatively consciously without any doubt and further increase the intrinsic motivation
to pay taxes, as research has been done by Torgler & Schneider (2005) in Austria.
4.3.4. The effect of likelihood of being caught on tax morale

When business owners state that tax evaders' opportunity to be caught is more
significant, they have higher tax morale. The study results align with the conclusions of
Mickiewicz et al. (2019), which states that the likelihood of being caught has a positive
effect on tax morale. According to Scott (2014), the third pillar of the institution is regulation
relating to the activities of setting rules, monitoring, and imposing sanctions. Rules are set
to provide limits for community behavior, and monitoring is carried out as a form of
evaluation of whether the rules have been adhered to properly. Scott (2014) adds that
individuals comply with laws and regulations because they seek the accompanying
rewards or avoid existing sanctions. Rational choice theory (Becker, 1968) also assumes
that individuals have preferences among several choices that allow the individual to state
the desired choice by considering available information, the possibility of events, and the
potential costs and benefits of the available choices, and acting consistently in choosing
the best course of action.
Concerning tax morals, according to Allingham & Sandmo (1972), the cause of
individuals to disobey or choose to evade taxes by not reporting or reporting income but
less than it should be is a trade-off between the profits earned and the probability of
detection and fines (penalties). In other words, if the possibility of someone being caught
through a rigorous inspection mechanism due to tax evasion is greater, the business
owner will prefer to comply with his tax obligations. This is because business owners
assume that the costs to be incurred will be greater than the profits obtained. This
statement is also reinforced by research by Heinemann (2011), which argues that tax
morale is influenced by an egocentric bias which states that someone will adjust their
behavior based on their interests.
Based on the slippery slope theory, one of the reasons people comply with tax
obligations is the authority of the authorities. The authority of the authority relates to the
ability of the tax authority to detect and punish tax violators. The results of the
questionnaire data show that the perception of the opportunity for business owners to be
caught for committing tax violations only ranges from 25-50%. The Directorate General of
Taxes' performance report also explained that the audit coverage ratio in 2019 only
reached 1.58%, lower than the IMF benchmark, which was 2%. It can illustrate how the
ability of the DGT to detect tax violations. Therefore, the implementation of the results of
this research needs to be carried out by the DGT by increasing the scope and quality of
tax audits so that the possibility of tax evasion being detected will be even more significant
and will increase the tax morale of business owners.
4.3.5. The effect of severity of punishment on tax morale

According to Scott (2014), one of the pillars of regulation is the provision of sanctions.
Giving sanctions/punishments is a deterrence variable according to slippery slope theory.
The analysis between the benefits and costs of this deterrence variable can affect tax
morale because individuals will usually adjust their norms or attitudes according to their
interests (Mickiewicz et al., 2019). Rational choice theory (Becker, 1968) states that
individuals have preferences among several choices by considering the potential costs
and benefits of the available options. Based on this theory, the business owner's decision
to comply with paying taxes or embezzling taxes considers the benefits derived from
reduced taxes paid and the number of sanctions received later if caught violating tax rules.
The imposition of sanctions is a stimulus in growing tax morale for business owners.
It is also in line with the research results of Mickiewicz et al. (2019), which concludes that
the level of punishment received by perpetrators of tax evasion has a significant positive
effect on tax morale. The majority of respondents in this study argue that the sentence
received by the perpetrators of violations, namely the business owner, will bear a fine.
With sanctions, business owners will be motivated to pay taxes according to applicable
regulations because they find it difficult to pay tax penalties.
The authorities in slippery slope theory include monitoring activities and providing
sanctions. Based on the regression results from the research model, it is known that the
severity of punishment has less significance than the likelihood of being caught and has a
more significant coefficient. The severity of punishment can be concluded to have a more
substantial influence on tax morale than the likelihood of being caught. Therefore, the
imposition of sanctions/penalties should be adjusted to the applicable rules and
regulations. The size of the sanction/punishment is also adapted to the level of the
violation committed. It is done so that the tax authorities still guarantee the creation of
justice.
5. CONCLUSION
This study aims to find the relationship between trust in government, trust in tax
administration, nationalism, the likelihood of being caught, the severity of punishment and
tax morale. From the result of data processing, conclusions can be drawn as well as
answering the formulation of the problem in this study that trust in government, trust in tax
administration, nationalism, the likelihood of being caught, and severity of punishment
have a positive and significant effect on the tax morale of MSME in Java.
6. LIMITATION AND SUGGESTIONS

This research is still far from perfect, considering that there are still many limitations
that can affect the final result of the research. These limitations and constraints include:
first, the research object is only limited to the owners of MSMEs domiciled in Java. Second,
the research time is short, so the authors use non-probability sampling techniques. Third,
dissemination of questionnaires through google form media so that respondents fill in
without the assistance of the author results in differences in respondents' understanding
of the contents of the questionnaire.
Further researchers can complete this research by expanding the research object,
using other sampling methods, multiplying sources and literature relevant to the study,
and using other variables not used in this study.
REFERENCE
A. Book and Articles
Allingham, M. G., & Sandmo, A. (1972). Income tax evasion: A theoretical analysis.
Taxation: critical perspectives on the world economy, 3, 323-338.
Alm, J., & McClellan, C. (2012). Tax morale and tax compliance from the firm’s perspective.
Kyklos, 65(1), 1–17. https://doi.org/10.1111/j.1467-6435.2011.00524.x
Alm, J., & Torgler, B. (2006). Culture differences and tax morale in the United States and
in Europe. Journal of Economic Psychology, 27(2), 224–246.
https://doi.org/10.1016/j.joep.2005.09.002
Alon, A., & Hageman, A. M. (2013). The Impact of Corruption on Firm Tax Compliance in
Transition Economies: Whom Do You Trust? Journal of Business Ethics, 116(3),
479–494. https://doi.org/10.1007/s10551-012-1457-5
Basri, Y. M., & Al Azhar, A. A. (2017). ANTESEDEN DAN KONSEKUENSI MORAL PAJAK
(Studi Pada Wajib Pajak Orang Pribadi di Kota Pekanbaru). JIAFE (Jurnal Ilmiah
Akuntansi Fakultas Ekonomi), 3(2), 61–75. https://doi.org/10.34204/jiafe.v3i2.760
Becker, G. S. (1968). Crime and punishment: An economic approach. In The economic
dimensions of crime (pp. 13-68). Palgrave Macmillan, London.
Blanthorne, C., & Kaplan, S. (2008). An egocentric model of the relations among the
opportunity to underreport, social norms, ethical beliefs, and underreporting behavior.
Accounting, Organizations and Society, 33(7–8), 684–703.
https://doi.org/10.1016/j.aos.2008.02.001
Cahyonowati, N. (2011). Model moral dan kepatuhan perpajakan: wajib pajak orang
pribadi. Jurnal Akuntansi Dan Auditing Indonesia, 15(2), 161-177.
Frey, B. S. (1997). Not just for the money: an economic theory of personal motivation.
Choice Reviews Online, 35(02), 35-1009-35–1009.
https://doi.org/10.5860/choice.35-1009
Frey, B. S., & Torgler, B. (2007). Tax morale and conditional cooperation. Journal of
Comparative Economics, 35(1), 136–159. https://doi.org/10.1016/j.jce.2006.10.006
Gaspar, V., David Amaglobeli, Garcia-Escribano, M., Prady, D., & Soto, M. (2019). Fiscal
Policy and Development: Human, Social, and Physical Investment for the SDGs. IMF
Staff Discussion Note SDN/19/3. IMF Staff Discussion Notes.
Ghozali, I. (2018). Aplikasi Analisis Multivariate dengan Program IBM SPSS 25.
Semarang: Badan Penerbit Universitas Diponegoro.
Gujarati, D. N., & Porter, D. C. (2009). Basic econometrics (international edition). New
York: McGraw-Hills Inc.
Heinemann, F. (2011). Economic crisis and morale. European Journal of Law and
Economics, 32(1), 35–49. https://doi.org/10.1007/s10657-010-9165-z
Hogg, M. A., Terry, D. J., & White, K. M. (1995). A Tale of Two Theories: A Critical
Comparison of Identity Theory with Social identity theory. Social Psychology
Quarterly, 58(4), 255. https://doi.org/10.2307/2787127
Hug, S., & Spörri, F. (2011). Referendums, trust, and tax evasion. European Journal of
Political Economy, 27(1), 120–131. https://doi.org/10.1016/j.ejpoleco.2010.06.005
Khaerunnisa, I., & Wiratno, A. (2014). Pengaruh Moralitas Pajak, Budaya Pajak, Dan
Good Governance Terhadap Kepatuhan Wajib Pajak. Jurnal Riset Akuntansi &
Perpajakan (JRAP), 1(02), 200–210. https://doi.org/10.35838/jrap.v1i02.80
Kirchler, E., Hoelzl, E., & Wahl, I. (2008). Enforced versus voluntary tax compliance: The
“slippery slope” framework. Journal of Economic Psychology, 29(2), 210–225.
Lago-Peñas, I., & Lago-Peñas, S. (2010). The determinants of tax morale in comparative
perspective: Evidence from European countries. European Journal of Political
Economy, 26(4), 441–453. https://doi.org/10.1016/j.ejpoleco.2010.06.003
Maria, D. (2014). Pengaruh Modernisasi Sistem Administrasi Perpajakan Terhadap
Tingkat Kepatuhan Pengusaha Kena Pajak di Kantor Pelayanan Pajak (KPP)
Pratama Bandar Lampung. Jurnal Bisnis Darmajaya, 1(1), 38-54.
Marien, S., & Hooghe, M. (2011). Does political trust matter? An empirical investigation
into the relation between political trust and support for law compliance. European
Journal of Political Research, 50(2), 267–291. https://doi.org/10.1111/j.1475-
6765.2010.01930.x
Mickiewicz, T., Rebmann, A., & Sauka, A. (2019). To Pay or Not to Pay? Business Owners’
Tax morale: Testing a Neo-Institutional Framework in a Transition Environment.
Journal of Business Ethics, 157(1), 75–93. https://doi.org/10.1007/s10551-017-3623-
2
Murphy, K. (2004). The role of trust in nurturing compliance: A study of accused tax
avoiders. Law and Human Behavior, 28(2), 187–209.
https://doi.org/10.1023/B:LAHU.0000022322.94776.ca
Pandiangan, L. (2008). Modernisasi & reformasi pelayanan perpajakan: berdasarkan UU
terbaru. Elex Media Komputindo.
Panjaitan, I., & Ekonomi, F. (2019). Kepatuhan Pajak Pada Dimensi Moralitas UMKM. 3(2),
30–40.
Rahayu, Siti Kurnia. 2010. Perpajakan Indonesia: Konsep dan Aspek Formal. Bandung:
Graha Ilmu
Scott, R. (2014). Institutions and organizations (4th ed.). Thousand Oaks: Sage.
Sekaran, U., & Bougie, R. (2016). Research Methods for Business "A SkillBuilding
Approach" 7th ed. Chichester: John WIley & Sons Ltd
Sophar, L. (1997). Ensiklopedi Perpajakan. Penerbit Erlangga, Jakarta.
Sugiyono. (2018). Metodologi Penelitian Kombinasi (Mixed Methods). Bandung: CV
Alfabeta.
Syarif, F. (2019). Integrasi Nilai Kearifan Lokal Terhadap Penanamanan Nasionalisme
Bagi Siswa Sekolah Dasar. Al Amin: Jurnal Kajian Ilmu Dan Budaya Islam, 2(02),
187–195. https://doi.org/10.36670/alamin.v2i02.26
Tajfel, H. 1972. Social categorization (English translation of "La cat6gorisation sociale").
In S. Moscovici (Ed.), Introduction a la psychologie sociale, vol. 1: 272-302. Paris:
Larousse.
Torgler, B., Demir, I. C., Macintyre, A., & Schaffner, M. (2008). Causes and Consequences
of Tax morale: An Empirical Investigation. Economic Analysis and Policy, 38(2), 313–
339. https://doi.org/10.1016/S0313-5926(08)50023-3
Torgler, B., & Schneider, F. (2005). Attitudes towards paying taxes in Austria: An empirical
analysis. Empirica, 32(2), 231–250. https://doi.org/10.1007/s10663-004-8328-y
B. Regulations
Direktorat Jenderal Pajak. 2020. Laporan Tahunan Direktorat Jenderal Pajak Tahun 2019
Direktorat Jenderal Pajak. 2018. Laporan Tahunan Direktorat Jenderal Pajak Tahun 2017
Bappenas. 2014. Laporan Analisis Daya Saing UMKM di Indonesia
C. Website
Badan Pusat Statistik. (2020). Pertumbuhan Ekonomi Indonesia Triwulan I-2020 No.
39/05/Th. XXIII, 5 Mei 2020. Bertita Resmi Statistik: https://www.bps.go.id/Brs
OECD. (2020). Revenue Statistics in Asia and the Pacific 2020 ─ Indonesia. Diambil
kembali dari oecd.org: https://www.oecd.org/tax/tax-policy/revenue-statistics-asia-
and-pacific-indonesia.pdf
Suwiknyo. (2019, 7 27). Ini Penyebab Rendahnya Rasio Pajak Indonesia. Dipetik 7 30,
2021, dari Bisnis.com: https://ekonomi.bisnis.com/read/20190727/259/1129457/ini-
penyebab-rendahnya-rasio-pajak-indonesia
INDONESIA’S TAX REVENUE:
IS IT SUSTAINABLE AND A GOOD AUTOMATIC
STABILIZER?
Delecia Brilliantyas Bungabangsaa, Benny Gunawan Ardiansyahb
aDirectorateGeneral of Taxes, Ministry of Finance
Jalan Jend Gatot Subroto Kav 40-42, Jakarta Selatan, Indonesia
Email: 1401190219_delecia@pknstan.ac.id
bPolytechnic State of Finance STAN, Ministry of Finance

Ja;an Bintaro Utama V, Tangerang Selatan, Indonesia
Email: bennygunawan.ardiansyah@pknstan.ac.id
ABSTRACT
This study analyzes Indonesia's long-run and short-run tax buoyancy. Buoyancy measures the
response of tax revenues to changes in economic growth. Buoyancy indicates the level of
responsiveness of tax revenue when business profit, people’s income and consumption increase. The
object of this research is total tax revenue, corporate income tax (CIT), personal income tax (PIT),
value added tax (VAT), and excise tax in Indonesia. The research was conducted using the Fully
Modified Least Square (FMOLS) method to determine the long-run tax buoyancy coefficient and the
Autoregressive Distributed Lag Model (ARDL) for the short-run tax buoyancy. This study uses
secondary data in the form of time series data with an observation period of 1967-2020. The research
was conducted to see whether Indonesia's tax revenues are sustainable in the long-run and become
an automatic stabilizer. The results show that the long-run tax buoyancy of Indonesian total tax
revenue, PIT, VAT, and excise tax exceed 1, which means that the tax revenue can guarantee
Indonesia's fiscal sustainability, but the CIT revenue shows otherwise. Tax buoyancy of Indonesia's
total tax revenue, PIT, VAT, and excise are less than 1 which means it cannot be an automatic
stabilizer, but the CIT shows the opposite.
Keywords : tax buoyancy, economic growth, tax revenue
Kajian ini mencoaba melakukan analisis daya pungut Indonesia dalam jangka panjang dan jangka
pendek. Daya pungut pajak mengukur respon penerimaan pajak terhadap pertumbuhan ekonomi.
Daya pungut pajak menunjukkan tingkat responsivitas penerimaan pajak ketika laba perusahaan,
pendapatan penduduk, dan konsumsi masyarakat mengalami perubahan, biasanya merupakan suatu
peningkatan. Objek penelitian ini adalah jumlah penerimaan pajak, pajak penghasilan badan (PPh
Badan), pajak penghasilan pribadi (PPh OP), pajak pertambahan nilai (PPN), dan cukai di Indonesia.
Penelitian ini dilakukan dengan menggunakan metode Fully Modified Least Square (FMOLS) untuk
menentukan koefisien daya pungut pajak jangka panjang dan Autoregressive Distributed Lag Model
(ARDL) untuk daya pungut pajak jangka pendek. Data yang digunakan berupa data sekunder berupa
data time series dengan periode pengamatan 1967-2020, yang bertujuan untuk melihat apakah
penerimaan pajak Indonesia berkelanjutan dalam jangka panjang dan menjadi penstabil otomatis.
Hasil penelitian menunjukkan bahwa daya apung pajak Indonesia dalam jangka panjang dari total
penerimaan pajak Indonesia, PPh OP, PPN, dan cukai melebihi 1, yang berarti penerimaan pajak
dapat menjamin kesinambungan fiskal. Sedangkan, penerimaan PPh Badan menunjukkan sebaliknya.
Daya pungut pajak dari total penerimaan pajak, PPh OP, PPN, dan cukai Indonesia kurang dari 1 yang
artinya tidak bisa menjadi penstabil otomatis, tetapi PPh Badan menunjukkan sebaliknya.
Kata kunci: daya pungut pajak, pertumbuhan ekonomi, penerimaan pajak
1. INTRODUCTION
Indonesia have ratified the Bill on Harmonization of Tax Regulations into a law. This
rule is considered essential, considering that tax revenue itself has always been the
backbone of government financing in Indonesia. Even though it plays a big role, the
potential for tax revenue so far can and needs to be increased. However, Indonesia’s tax
ratio is still relatively low, where in 2022, Indonesia’s tax-to-GDP ratio is only around 8,37%
to 8.42%. Actually, it had been decreased to the 8,33%and 8,2% for the year of 2020 and
2021 respectively. The pandemic caused a contraction to economic growth, which also
affected the government's fiscal stability (Modjo, 2020). Reforms are deemed necessary
to fix the budget deficit and increase the tax ratio. Fiscal consolidation is carried out by
implementing policies oriented towards increasing voluntary compliance, increasing the
tax base, improving tax system that prioritizes the principles of fairness and legal certainty,
increasing tax revenue performance, and reforming tax administration.
In principle, taxation may be one of the strongest policies or factors influencing
economic activity and growth (Yanikkaya and Turan, 2017). In recent years, tax growth in
Indonesia has always been below economic growth. It is also indicated by the relatively
low tax-to-GDP ratio in Indonesia. Quoted from the OECD Report (Revenue Statistics in
Asian and Pacific Economies 2020 Indonesia), the tax ratio compared to Gross Domestic
Product (GDP) in Indonesia in 2018 was 11.9%. This figure is lower than the average for
OECD member countries and several Southeast Asia countries such as the Philippines,
Singapore, Malaysia, and Thailand. The declining tax ratio shows that the Indonesian
economy has succeeded in growing. However, there are still many potentials that escape
from the taxation. It causes tax revenues not to significantly increase compared to GDP
which tends to increase (Pink, 2021).
According to the OECD (2020), tax contribution in Indonesia is dominated by
income tax in the form of corporate income tax at 34%, followed by VAT with a contribution
of 29%. Currently, there is a shift in the tax structure in Indonesia in line with the increase
in tax revenue on consumption. The dependence of state revenues on taxes on
consumption also occurs in Bulgaria (Tanchev and Todoror, 2019). The tax reforms
carried out in Bulgaria to reduce the rates of several types of taxes are intended to
stimulate economic growth to increase income from direct and indirect taxes. However,
after a recession after 2008, the decline in consumption resulted in a decrease in indirect
tax revenues. The same thing happened in Indonesia. In 2009, there was a decrease in
tax revenue caused by the decrease of value-added tax. The government needs the
proper tax structure in responding to economic, social, and environmental developments
to maintain the sustainability of tax revenue growth (Schratzenstaller, 2015). In
determining the optimal tax structure, tax buoyancy testing can serve as an indicator for
the formulation and design of tax policies (Dudine and Jalles, 2017). Dudine and Jalles
(2017) add that tax buoyancy illustrates the role of income policy in ensuring fiscal
sustainability in the long term and an automatic stabilizer of the economy during the
business cycle in the short run. Based on the Macroeconomic Framework and the
Fundamentals of Fiscal Policy for 2022, Indonesia's average tax buoyancy from 2016 to
2019 has a value below 1, indicating a reduced tax base. It happens because the tax
revenue cannot fully capture real sector economic activities such as the basis of the
informal sector and digital economy. Darussalam, quoted from Santoso (2020), said there
is a tax gap in sectors with a GDP value that grows significantly. Yustinus was quoted by
Hendartyo (2020) as saying the same thing and told that extensification and expansion of
the tax base needed to increase tax growth.
Previous research has shown a long-run positive relationship between VAT
buoyancy and corporate income tax with economic growth in Bulgaria (Tanchev and
Todorov, 2019). In addition, previous research related to the estimation of tax buoyancy
in Indonesia was carried out by Nurhidayati (2016) to determine the tax buoyancy
coefficient and the elasticity of VAT. Similar studies were conducted in other countries to
assess tax buoyancy, including Dudine and Jalles (2017). They analyzed tax buoyancy in
developed countries, emerging market economies, and low-income countries. Upender
(2008) conducted a study to measure the degree of tax buoyancy in India, Belinga (2014)
tested tax buoyancy in OECD countries, and Ashraf and Sarwar (2016) identified the
determinants of tax buoyancy in developing countries. Analysis using long-run and short-
run tax buoyancy can be an additional option to help determine the design of a tax
structure that can support tax revenue and its relation to economic development. In
carrying out tax reform, the government is expected to determine the types of taxes that
are the primary support for sustainable tax revenues in the long term and become
automatic stabilizers in the short run. In addition, the government is also expected to
formulate policies that can stimulate the growth of tax revenues from the suitable types of
taxes.
According to Hubbard (2017), long-run economic growth increases productivity as the
average standard of living increases. The standard of living is generally measured based
on real GDP per capita, which is the sum of the production levels in the economy adjusted
for changes in the price level. The growth rate is influenced by macroeconomic policies,
such as taxation, consumption, and investment (Egbunike et al., 2018). Fluctuating GDP
growth indicates a movement in the country's economy.
The theory related to the economic growth model that is still often used today is
the aggregate demand model or economic growth proposed by Keynes (Soewardi et al.,
2018). Two variables affect the Keynesian consumption equation, namely taxes and
income, or what is commonly called a function of disposable income. Disposable income
is individual income after deducting tax obligations or, in other words, income that is ready
to be spent. Keynesian theory of growth explains that both taxes and government
spending in fiscal policy impact aggregate output.
Taxes indirectly impact aggregate output, while government spending has a direct
impact. The increase in disposable income will increase purchasing power so that the
marginal propensity to consume (MPC) of the community will also increase. The increase
in disposable income can be caused by cuts in government revenues (from taxes). In the
Keynesian equation, it is stated that if there is an increase in the consumption function,
there will also be an increase in aggregate output, and vice versa (Froyen 2002).
In addition, changes in tax rates will have an economic impact under the Laffer
curve theory, which specifically discusses the effect of tariffs on tax revenues. The tax rate
has a parabolic correlation curve with one maximum point in this theory. This shows that
there will always be two levels of tariffs with the same tax revenue (Laffer, 2004). Changes
in tax rates have an arithmetic impact and an economic impact (Laffer, 2004). The
arithmetic impact occurs with a decrease in tax rates, which will reduce tax revenues and
vice versa. This happens because tax revenue results from the multiplication between the
tax rate and the tax base. On the other hand, the economic impact pays attention to the
positive effect of a low tax rate on the desire to work, output, and employment opportunities.
This is because low tax rates provide incentives for these activities. On the other hand,
increasing tax rates will penalize these activities.
Taxes are one of the main political tools used by the state to achieve economic,
social, and political goals. Taxation as a fiscal policy tool enables countries to do effective
practices in achieving macroeconomic goals (Aytac, 2018). Tax revenue is considered a
relatively stable form of fiscal financing (Egbunike et al., 2018). Tax revenue is the primary
key for the government in achieving the government's fiscal policy goals in the form of
economic growth (Gurdal et al., 2020). Hubbard (2011) explained that the government
could make changes to government spending and taxes in carrying out fiscal policy.
Changes in taxation can change income, impact consumption, and have an indirect effect
on aggregate demand.
Tax policies are formulated to influence the country's economy. Taxes also play a
role in regulating the economy to increase state revenues. The purpose of tax policy is to
collect revenue, encourage investment, and create equity. Therefore, the approaches
taken by the government must be able to balance these three objectives optimally.
Tax is a fiscal policy tool that functions as a source of state revenue and as a
regulator of the economy (Schiller, 2005). In addition to financing government spending
(budgetary function), taxes also function as a tool to control the economy (regulatory
function). According to Hubbard (2017), in carrying out fiscal policy, both through
expansionary and contractionary fiscal policies, taxes are one of the instruments used.
For example, when the economy experiences a recession marked by a decline in real
GDP, the government can intervene through expansionary fiscal policies. On the other
hand, a contractionary fiscal policy is used when the government wants to slow down the
real GDP growth rate (Sjafri, 2006). Hubbard (2017) added that in the expansionary policy,
the policy taken by the government is to lower taxes followed by increasing government
spending, which is expected to restore long-term equilibrium. In contrast, taxes are raised
in a contractionary fiscal policy, and government spending is lowered.
According to Pistone (2019), the government must carry several limitations in
collecting taxes. The first principle in tax collection is that taxes should not erode capital.
Capital erosion can lead to reduced future tax revenues, which threaten the survival of a
productive economy. Furthermore, the taxes collected must not exceed the government's
needs in funding the needs of the community. Finally, the cost of collecting taxes must not
exceed the amount of tax collected.
In addition, the government needs to consider the main principles in deciding the
best solution to increase tax revenues (Stiglitz, 2015). There are two concepts in the
principle of fairness, namely horizontal equity and vertical equity (Stiglitz, 2015). Horizontal
equity focuses on identical individuals or, in basically similar economic circumstances,
should be treated the same and pay the same taxes. In vertical equity, individuals who
can pay or receive more significant benefits from government services must pay higher.
One of the relationships between tax revenue and economic growth can be seen
through tax buoyancy. Tax buoyancy measures the responsiveness of tax revenue to
economic growth (Leuthold and N'Guessan, 1986). A tax is buoyant if the increase in tax
revenue increases by more than one percent from the one percent increase in national
income or output. Buoyancy reflects discretionary change and automatic income growth
(Mansfield, 1972). Tanchev and Todorov (2019) stated that tax buoyancy is a term to
measure and show the level of tax responsiveness along with an increase in GDP, so it
can be said that an increase in tax revenue and collection is the result of an increase in
national income. If the level of tax responsiveness shows a positive value, the applicable
tax system can be said to be buoyant and efficient (Ashraf and Sarwar, 2016). Belinga et
al. (2014) and Deli et al. (2018) state that if the tax buoyancy exceeds 1, the increase in
tax revenue will exceed the GDP growth, which will affect a higher tax-to-GDP ratio. This
can reduce the deficit ratio so that financing will remain sustainable.
Furthermore, Belinga et al. (2014) stated that tax buoyancy might experience
differences in the short-run and long-run. Short-run buoyancy is related to the stabilization
function of fiscal revenues. Automatic stabilizers refer to income and expenses that can
automatically adjust up or down following economic cycle changes (Baunsgaard et al,
2009 and Hubbard, 2017). Burnside (2005) stated that automatic stabilizers are
components in the budget that can respond automatically to the business cycle without
direct government intervention.
Burnside (2005) stated that taxes could act as a natural stabilizer in the business
cycle. Tax revenue and marginal tax rates tend to rise following an increasing cycle so
that the fiscal surplus moves along with the business cycle. From the perspective of
Keynesian theory, these factors cause the fiscal policy to have a natural and automatic
tendency to dampen fluctuations. If the short-term buoyancy is more than 1, the increase
in tax revenue exceeds the rise in GDP, which indicates that the tax system is a good
automatic stabilizer. On the other hand, the buoyancy coefficient showing a value of less
than one indicates that the tax has not become an automatic stabilizer.
Long-run buoyancy plays a vital role in linking economic growth with long-term
fiscal sustainability. According to Hubbard (2017), the long-run period can vary and
includes periods that can cause all aspects to change. Long-run buoyancy that shows a
value of more than 1 in a ceteris paribus condition results in higher growth to improve the
fiscal balance on the income side of the budget and vice versa. Deli et al. (2018) explain
that the estimation of long-run tax buoyancy measures the reaction of tax revenues to
long-term structural changes and can be used to measure the impact of growth on long-
term fiscal sustainability.
Fiscal sustainability can be demonstrated by a relatively manageable budget deficit
and a declining debt-to-GDP ratio. This situation can illustrate the realization of healthy
fiscal conditions ongoing (Fiscal Policy Agency, 2019b). Burnside (2005) relates fiscal
sustainability to the government's ability to maintain its fiscal policy continuingly to
maintain its financial position in a solvent state. Fiscal sustainability is also defined as the
state's ability to finance spending in the long term while maintaining the budget function
within safe limits, maintaining a positive primary balance, and increasing tax ratios. Fiscal
sustainability reflects the government's ability to consistently meet its long-term financial
responsibilities, characterized by ensuring sufficient income available to provide services
and a sustainable level of capital needed by the public (Chapman, 2008).
3. RESEARCH METHOD
This study aimed to determine the value of long-run and short-run tax buoyancy
and its function to fiscal sustainability and an automatic stabilizer in Indonesia. The
dependent variable used is five variables in the form of tax revenue (TR), corporate income
tax revenue (CIT), personal tax revenue (PIT), value-added tax (VAT), and excise tax
(EXCISE) in Indonesia. The independent variable used is economic growth as measured
by GDP in Indonesia.
The study used secondary data in the form of time-series data. The object of this
research is Indonesia's tax revenue and GDP in 1967-2020. Tax revenue was chosen
because it is the largest source of the component of state revenue in Indonesia. Corporate
income tax, personal income tax, value-added tax, and excise tax are the most prominent
types of tax revenue in Indonesia.
In determining the long-run and short-run tax buoyancy coefficients, the data will
be analyzed using the Fully Modified Least Squares (FMOLS) approach and Auto-
Regressive Distributed Lag Model (ARDL). The models are used to examine the effect of
changes in GDP on tax revenues, both total tax revenue and tax revenue per type of tax.
Based on Belinga et al (2014), tax buoyancy is generally measured by regression
from the logarithm of tax revenue to the logarithm of GDP. Traditionally, in estimating the
value of tax buoyancy, Leuthold and N'Guessan (1986) used the following model:
Log Tk = a0 + a1 log GDP + ek,

T is tax revenue, GDP is economic growth, e is the stochastic disturbance, a1 is
estimated tax buoyancy because a1 measures the percentage of response variables on
the left side for every one percent change on the right side.
In estimating a variable and its effect value, to determine the relationship between
long-run and short-run tax buoyancy, the ARDL equation is used. The equations used in
the estimation method with cointegration analysis are as follows.
𝑝 𝑝
∆ 𝑙𝑛 𝑙𝑛 𝑥𝑡 = 𝛽 + 𝛽1 ∑ ∆𝑙𝑛𝑥𝑡−𝑖 + 𝛽2 ∑ ∆𝑙𝑛𝐺𝐷𝑃𝑡−𝑖 + 𝛽3 𝑙𝑛𝑥𝑡−1 + 𝛽4 𝑙𝑛𝐺𝐷𝑃𝑡−1 + 𝜀𝑡

𝑡=1 𝑡=1
x = dependent variable in the form of tax revenue, corporate income tax, personal
income tax, value added tax, which will be tested for each variable
GDP = economic growth variable
β1, β2 = coefficient of short-run dynamics
β3, β4 = long-term relationship coefficient
Δ = difference between two variable values in successive time periods
ε = normally distributed error
The unit root test is carried out using the intercept and trend and intercept models. At
the first difference level, the output of the test results shows the ADF statistical value that
exceeds the critical value for all variables. Changes in the significance level of the
probability value at α = 5% indicate that H0 is rejected and accepts the alternative
hypothesis. With these outputs, it can be concluded that the LNTR, LNCIT, LNPIT, LNVAT,
LNEXCISE, and LNGDP variables no longer have a unit root problem. These variables
are considered stationary at the first difference level.
Johansen's cointegration test can be done by comparing the trace statistic or the max-
eigenvalue statistic with the critical value to determine the existence of cointegration. The
trace statistic value or the max eigenvalue statistic that exceeds the critical value indicates
cointegration or long-run equilibrium. Trace statistics and max eigenvalue statistics show
a value greater than the critical value at the 5% level. This shows a cointegration that
shows a long-term relationship between variables.
Diagnostic tests on classical assumptions were carried out through normality,
autocorrelation, and heteroscedasticity tests. Normality test performed using the Jarque-
Bera method. The test results of all models of the dependent variable with the economic
growth variable show a probability value greater than 0.05. If the probability value is more
than 0.05, the residue can be said to be normally distributed. The following classic
assumption test is the autocorrelation test using the Breusch-Godfrey Serial Correlation
LM Test. The test results on all dependent variables with GDP show a probability value of
F that exceeds 0.05, so it can be said that the model is free from autocorrelation. A
heteroscedasticity test was performed using the Breusch-Pagan-Godfrey method. In
determining the presence or absence of heteroscedasticity, the Chi-Square probability
value greater than 0.05 indicates no heteroscedasticity in the regression residue. The
results of the heteroscedasticity test show that all models of the dependent variable with
the variable of economic growth are free from heteroscedasticity because they have a
Prob Chi-Square value is more than 0.05.

4.1 Indonesia Long-run Tax Buoyancy
From the results of tests carried out using the FMOLS model, the long-run tax buoyancy
coefficient can be determined. The results of the FMOLS test and its relation to long-run
tax buoyancy is presented in Table 3.
Table 3. Long-run Tax buoyancy Coefficient based on FMOLS

Variable Long-run
Coefficient Prob. R-squared
Dependent Independent Buoyancy
LNTR LNGDP 1.070300 0.0000 0.996629 K>1
LNCIT LNGDP 0.888752 0.0000 0.954975 K<1
LNPIT LNGDP 1.134667 0.0000 0.991664 K>1
LNVAT LNGDP 1.159032 0.0000 0.989221 K>1
LNEXCISE LNGDP 1.037177 0.0000 0.996143 K>1
Source: Data, processed
The value of long-run tax buoyancy is determined based on the coefficient of the
LNGDP variable. The interpretation of long-run tax buoyancy based on Table 3 is as
follows.
a. The long-run tax buoyancy value for Indonesia's total revenue is 1.07. It shows that
an increase in the rate of economic growth by 1% will be followed by an increase in
tax revenues of 1.07%. This value shows the sensitivity of Indonesia's total tax
revenue to economic activity and demonstrates that tax performance is more than
economic performance.
b. The value of long-run tax buoyancy for corporate income tax revenue is 0.89. The
regression coefficient of 0.89% shows that for every 1% increase in the rate of
economic growth, corporate income tax revenues will only increase by 0.89%. This
value reflects the ability of corporate income tax, which cannot keep pace with
economic performance.
c. The long-run tax buoyancy value for personal income tax is 1.13. This value shows
that the personal income tax is sensitive to changes in economic growth of 1.13 times.
A 1% change in economic growth will be followed by a 1.13% increase in personal
income tax.
d. The value of long-run tax buoyancy for value added tax is 1.15. The value of 1.15
shows the response of VAT receipts to GDP. Changes in the rate of economic growth
of 1% will be followed by a response to VAT of 1.15%.
e. The value of long-run tax buoyancy on excise is 1.03. This value shows that 1% GDP
growth will increase excise revenue by 1.03%
The previous section described the value of each long-term tax buoyancy
coefficient for total tax revenue and per type of tax. The value of the tax buoyancy
coefficient in the long run shows the ability of taxes to support the revenue side. The
results show that tax revenue can maintain fiscal sustainability from the revenue side,
except for corporate income tax. In the long term, Indonesia's tax revenues and all types
of taxes except corporate income tax have good responsiveness to changes in economic
growth. In accordance with Dudine and Jalles (2017), tax buoyancy illustrates the role of
revenue policy in achieving fiscal sustainability in the long run. Tax revenues that move in
line with economic output can help support the sustainability of fiscal policy in the long run.
A tax buoyancy coefficient of more than 1 implies that changes in tax policy are
positively correlated and have a good impact on tax revenue. Sjafri (2006) supports it,
which states that the increase in VAT and income tax, which is the most significant tax
contributor in Indonesia, provides a positive value in total tax revenue. It is due to the
positive results of changes in the tax structure by the government. The research results in
this paper indicate that personal income tax and value-added tax have a positive role in
total tax revenue. However, this is not the case with corporate income tax.
Personal income tax is one tax type with a long-run tax buoyancy coefficient of
more than 1. This value is in line with Dudine and Jalles's (2017) research, which states
that in emerging market economies, long-run tax buoyancy exceeds the coefficient value
of 1. Labor growth has contributed to the increase in personal income tax. Based on
Central Bureau of Statistics’ data (n.d) the trend of the working population has increased
from 1986-2020. This indicates a growing personal income tax base. An increase in the
value of the personal income tax will move in line with an increase in wages and a
decrease in unemployment. The growth of personal income tax revenue is also seen due
to government regulations related to MSME taxes. Tax compliance increases along with
the ease of tax compliance schemes offered by the government. The number of taxpayers
and tax receipts has constantly increased since this tax was implemented in 2013 (Ministry
of Finance, 2019). However, currently, the personal income tax is still unable to contribute
to the total tax revenue significantly. Corporate income tax and VAT are still the main
contributors to tax revenue. This could be due to the low compliance of taxpayers in the
informal sector.
Indonesia's tax revenue is influenced by the contribution of corporate tax revenue,
value-added tax revenue, personal income tax, and excise tax, respectively. However,
based on testing, corporate tax revenue is a type of tax that is less responsive to economic
growth. The government is trying to formulate a corporate tax policy that supports
increased investment (Foreign Direct Investment) to boost growth in industrial sectors. To
attract FDI, the government provides incentives in tax allowances and tax holidays.
However, unfortunately, tax holidays and tax allowances are considered less attractive to
investors. Reporting from Santoso (2021), the Minister of Finance stated that the
realization of investment from investors who received these incentives only reached 1.96%
of the total investment plan for 2018-2021. Lewis (2019) stated that tax holidays are
inefficient because these incentives cannot capture other factors outside of taxes which
sometimes have a more critical role in investment decisions. In addition, Arnold (2012)
stated that exceptions can cause unnecessary distortions, including distortions in
corporate tax revenues and the erosion of the tax revenue base. This causes the growth
of long-term corporate income tax revenues due to FDI to be unattainable. In addition to
inefficient tax incentives, factors other than taxes are less supportive in increasing FDI in
Indonesia. These factors include inadequate infrastructure, labor, and geographical
conditions that do not support the investment climate.
Another factor that causes long-term corporate income tax buoyancy not to reach
a value of 1 is that corporate income tax still relies on natural resource revenues. One of
them was caused by the realization of income tax revenues from the mining sector, which
grew by -23.25% due to the downward trend in mining commodity prices (Ministry of
Finance, 2020). In addition, the development of digital business presents challenges for
the Indonesian tax system. Challenges arise because the applicable regulations have not
captured taxes from several new business models. It can lead to tax evasion from
loopholes in the regulations.
In the long run, taxes on consumption are one of the most reliable contributors to
total tax revenue as a source of revenue. In general, the value-added tax is considered a
more "friendly" type of tax on economic growth (Johansson et al., 2008). As output
increases in line with Indonesia's GDP growth trend, total consumption in Indonesia will
increase. The increase in Non-Taxable Income is a trade-off from individual income tax
receipts. Still, it provides compensation to the community to spend their money so that it
indirectly increases VAT. This is in line with Keynes' theory which states that when the tax
burden is reduced, disposable income will increase so that it can increase people's
purchasing power.
In addition to Personal Income Tax and VAT, excise has a long-run tax buoyancy
coefficient more than 1. This value shows that excise revenue is sensitive to changes in
economic growth. Excise has the property of controlling the consumption of certain goods
to reduce the negative externalities caused. With these characteristics, excise rates tend
to increase to maintain the level of consumption of certain goods. The increase in excise
rates will directly impact increasing tax revenues. Currently, the government is still making
policy updates related to excise rates and expanding the objects and subjects of excise
so that excise revenues will tend to increase in the long run. Lewis (2019) states that the
relatively inelastic excise tax base causes the tax imposed to be efficient.
To increase sustainable tax revenue, the government needs to determine a
combination of taxes that can optimize the contribution of these taxes. An increase in tax
revenue cannot only be done by increasing tax rates because it will cause distortion and
inequality (Arnold, 2012). This is under the theory of the Laffer curve, which is that there
is an optimum limit between tax revenues and tax rates. Therefore, to determine a tax
structure that can respond to changes in state revenues, the government should ensure
that each type of tax is well designed so that the policy direction taken by the government
does not distort long-run growth. Efficient tax reform by considering an appropriate and
balanced tax structure is expected to make tax revenue the main supporter on the revenue
side and reduce deficit pressure.
4.2 Indonesia Short-run Tax buoyancy

From the results of tests carried out using the ARDL model, the short-run tax buoyancy
coefficient can be determined. The ARDL test results for the relationship between the
independent variable and the short-run dependent variable and its relation to the short-
run tax buoyancy are presented in Table 4.
Table 4. Short-run Tax buoyancy Coefficient Based on ARDL

Variable Short-run
Coefficient Prob. R-squared
Dependent Independent Buoyancy
LNTR LNGDP 0.819652 0.0000 0.998917 K<1
LNCIT LNGDP 1.440105 0.0000 0.994775 K>1

LNPIT LNGDP 0.629734 0.0022 0.997586 K<1
LNVAT LNGDP 0.184484 0.0208 0.997222 K<1
LNEXCISE LNGDP 0.445028 0.0045 0.999173 K<1
Source: Data, processed
The value of the short-run tax buoyancy can be determined by the short-run GDP
coefficient based on the ARDL test. Each model of the dependent and independent
variables has a different coefficient value at the 95% confidence level. The value of short-
run tax buoyancy for each dependent variable based on Table 4 is as follows.
a. The value of short-run tax buoyancy on the variable total tax revenue (total revenue)
is 0.82. It shows that a 1% rise in GDP growth will increase the value of total tax
revenue by 0.82% in the short term. This value indicates that tax revenue performance
cannot match economic growth performance.
b. The value of the short-run tax buoyancy of the corporate income tax variable is 1.44.
Corporate income tax is the only type of tax with a short-run tax buoyancy value
exceeding 1. It shows that only corporate income tax is sensitive to economic growth
performance in the short run. If economic growth changes by 1%, corporate income
tax will follow a 1.44% rise.
c. The value of short-run tax buoyancy on the individual income tax variable is 0.63. The
personal income tax will respond by 0.63% to a 1% change in economic growth. It
shows that the personal income tax is not sensitive to changes in economic growth in
the short run.
d. The value of the short-run tax buoyancy of the value-added tax is 0.18. It shows that
value-added tax growth only responds by 0.18% if GDP grows by 1%. The low value
of short-run tax buoyancy indicates the performance of the value-added tax cannot
keep up with the economic growth.
e. The value of the short-run tax buoyancy of excise is 0.45. In the short term, excise tax
performance can respond to 0.45% to a 1% change in economic growth. The low
value of short-run tax buoyancy indicates that excise taxes cannot keep pace with
economic growth performance.
Short-run tax buoyancy has a relationship with tax revenue that is responsive to
GDP in its function as a good automatic stabilizer. Short-run tax buoyancy provides an
overview of how state revenues react to changes in GDP in the business cycle (Dudine &
Jalles, 2017). Dudine and Jalles (2017) added this picture to be one of the essential things
from the government's point of view in assessing current and future budgetary constraints.
The short-run tax buoyancy of Indonesia's tax revenue is at a coefficient value
below 1. This value is in line with Dudine and Jalles' research (2017) results, which states
that Indonesia's short-run tax buoyancy is less than 1. This is due to the taxes compiling
state revenue having a low tax buoyancy value.
In the short run, only corporate income tax can be relied on for its responsiveness
to GDP output. The tax buoyancy coefficient for Indonesian corporate income tax is in line
with the short-run tax buoyancy coefficient for emerging market economies, which is more
than 1 (Dudine and Jalles, 2017). Although the government no longer applies progressive
rates in the corporate income tax scheme, the high value of corporate income tax
buoyancy can be caused by the decline in the corporate income tax rate that has been
carried out several times. The company's tax burden will be reduced and will provide a
larger cash flow to trigger companies to be able to carry out investment planning. It is in
line with Dobbins and Jacob (2016). They state that a reduction in the corporate income
tax rate will impact the company's investment strategy and corporate funding decisions.
In the case of personal income tax, the short-run tax buoyancy coefficient of less
than 1 indicates that this type of tax is less sensitive to changes in economic growth. This
could be due to the individual structure in Indonesia. However, there is an increasing trend
in the working population. Based on the Central Bureau of Statistics’ publication (2020),
most workers in Indonesia are not permanent wage earners. This condition causes a high
dependence on compliance with the fulfillment of each individual's tax obligations. Lewis
(2019), in his publication, mentions that reducing informality and increasing reporting
requires increased monitoring of non-compliance to stimulate voluntary compliance. The
increase in the Non-Taxable Income limit is another factor that causes the tax base to
decrease. Higher Non-Taxable Income causes reduced amount subjected to tax.
The value of the value-added tax buoyancy in the short run is the lowest. This could
be due to the many exceptions of value-added tax and the complex design of the VAT
administration. Arnold (2012) states that the efficiency level of value-added tax depends
on how broad the tax base and efficient administration are.
The exemption to the VAT tax object is relatively broad, with various types of
facilities provided by the government. Exemptions to VAT objects can impact the
narrowing of the tax base and a decrease in the level of tax compliance. In addition, the
exclusion of VAT objects will have a distorting impact. One of them causes the seller's
cash flow to be disrupted because some exemptions cause the input tax obtained by the
seller to be unable to be credited. This is in line with Crawford et al. (2010), which explains
that the distortion caused by the VAT exemption will break the chain of output tax and
crediting of input taxes, thereby indirectly creating an element of taxation on production
activities. The tax gap in this type of tax is described in the Tax Expenditure Report. Based
on the 2019 Tax Expenditure Report data, the estimated VAT and sales tax on luxury
goods lost due to exceptions reached 166,920 billion Rupiah or equivalent to 1.05% of
GDP.
Excise is also a component of tax revenue which has a short-run tax buoyancy
value of less than 1. As a regulatory function, the government expects to increase the
excise rate, which almost always occurs to control the consumption of certain goods. For
example, an increase in excise tax rates on tobacco is expected to influence consumer
behavior to reduce health risks. Lower consumption resulting from higher excise rates will
result in lower revenues (Patrick and Janos, 2016). In addition, the complexity of the
mechanism for imposing excise duty with tiered excise rates can cause a gap for
companies to avoid tax by avoiding the highest tariffs (WHO, 2020). This condition is
compounded by several modus operandi in counterfeiting excise stamps for cigarettes,
ethyl alcohol, and beverages containing ethyl alcohol.
Based on the short-run tax buoyancy value obtained, Indonesia's tax revenue as
a whole has not been able to carry out its role as an automatic stabilizer in the business
cycle. This is in accordance with Belinga et al. (2014), which states that tax revenue does
not function as an automatic stabilizer if the short-run tax buoyancy is less than 1.
Automatic stabilizers reflect revenues that can automatically adjust to changes in the
economic cycle that directly impact business and household income (Baunsgaard et al,
2009). The low value of the tax buoyancy of Indonesian tax revenues can be caused by
the tax component that makes up state revenue has a short-term tax buoyancy of less
than 1. A low value indicates that tax growth is below its tax base. In the short run and
relation to the response to the business cycle, the government can rely on corporate
income tax revenues.
5. CONCLUSION AND SUGGESTION

The results show that Indonesia's tax revenue, seen from the long-term tax buoyancy
coefficient, can maintain fiscal sustainability. The tax-buoyancy value of Indonesian tax
revenues is 1.07. A value that exceeds 1 indicates that Indonesia's tax revenue is
responsive to changes in GDP growth so that it can be relied upon to support state
revenues. With the movement of tax revenues in line with economic growth in the long run,
it is hoped that the government can reduce the value of the budget deficit. Not all type of
Indonesian tax has a value of tax buoyancy of more than 1. Corporate income tax has a
coefficient of 0.89 due to the distorting effect on growth. Other types of taxes have a long-
run tax buoyancy coefficient of more than 1, from the highest to the lowest, namely value
added tax, personal income tax, and excise. These types of taxes can be optimized to
support tax revenues so that Indonesian tax revenues can ensure long-run fiscal
sustainability.
Meanwhile, Indonesia's tax revenue seen from the short-run tax buoyancy
coefficient has not been able to become an automatic stabilizer. The short-run tax
buoyancy value based on the ARDL test results is 0.82. This means that in the short run
tax revenues are less sensitive to changes in economic growth. Tax revenues have not
been able to control output in changing economic conditions that fluctuate. The value of
tax buoyancy for almost all types of taxes in Indonesia in the short run is relatively low.
The tax buoyancy coefficient of personal income tax, value added tax and excise tax has
not yet reached 1. Only corporate income tax has a short-run tax buoyancy coefficient of
more than 1. In short-run conditions and its relation to the automatic stabilization function,
the government can rely on tax revenues corporate income.
6. IMPLICATION AND SCOPE OF THE RESEARCH

In carrying out this research, the authors encountered several limitations due to the time
and availability of data. First, this study uses secondary data from 1967-2020. The
extended data period has caused tax regulations to undergo many reforms, resulting in
changes to the tax terminology, system, and mechanism. This causes the research to be
constrained by the difficulty of grouping data for each type of tax. In addition, this study
only examines four variables of the kinds of tax revenues that have the most significant
value in Indonesian tax revenues. As a result, this study has not been able to explain the
responsiveness of other types of taxes to changes in economic growth other than the four
types of taxes. Third, this study uses the FMOLS and ARDL methods in finding the value
of tax buoyancy according to the research of Tanchev and Todorov (2019). This causes
this research not to explain the value of tax buoyancy calculated by methods outside the
study.
Further research can be directed to use a complete variable of tax revenue by
considering other types of taxes to provide a comprehensive picture of buoyant taxes. A
more thorough analysis can be carried out in determining the tax structure that can support
Indonesian tax revenues. Research can also be developed by conducting a tax buoyancy
analysis on a sectoral basis. In addition, further research can consider the study by adding
an indicator of tax elasticity.
For the government as a policymaker, the government needs to consider
increasing the optimal potential for each type of tax that supports long-run tax revenues.
To achieve the goal of long-run fiscal sustainability, the government can strengthen the
tax structure on value-added tax, personal income tax, and excise. In the long run, the
government may reconsider using corporate income tax as a basis for state revenue and
optimizing other types of taxes. Meanwhile, the government can optimize the corporate
income tax in the short run, which functions as an automatic stabilizer. In increasing tax
revenue, the government can expand the tax base by removing exemptions to the object
of value-added tax and extensification on personal income tax. Besides that, the
government can improve the administration system and tax information system to
minimize tax avoidance, improve supervision and compliance, and improve the tax
structure to achieve an optimal tax system.
REFERENCES
Arnold, J. M., Brys, B., Heady, C., Johansson, Å., Schwellnus, C., & Vartia, L. (2011). Tax
Policy for Economic Recovery and Growth. The Economic Journal, 121(550), F59–F80.
doi:10.1111/j.1468-0297.2010.02415.x
Arnold, Jens Matthias. (2012). Improving the Tax System in Indonesia. OECD Economics
Department Working Papers No. 998.
Ashraf, M., & Sarwar, S. (2016). Institutional Determinants of Tax buoyancy in Developing
Nations. Journal of Emerging Economies and Islamic Research 2016, Vol .4, No.1.
Auerbach, A.J.& Feenberg, D. (2000). The Significance of Federal Taxes as Automatic
Stabilizers.
Aytaç, D. (2018). Optimal Taxation of Consumption in the Scope of Changing. Taxes and
Taxation Trends. doi:10.5772/intechopen.74378.
Barro R. J. (1990). Government Spending in a Simple Model of Endogenous Growth.
Journal of Political Economy, 98(2):103–125.
Baunsgaard, T., Symansky, S.A., & Cottarelli, C. (2009). Automatic Fiscal Stabilizers.
International Monetary Fund Volume 2009: Issue 023.
Belinga, V., Benedek, D., Mooij, R., & Norregaard, J. (2014). Tax buoyancy in OECD
Countries. IMF Working Paper.
Blanchard, O. (2000). Commentary, FRBNY Economic Policy Review. April 2000, 69-73.
Burnside, C. (2005). Fiscal Sustainability in Theory and Practice. The World Bank,
Washington DC.
Chapman, J. I. (2008). State and Local Fiscal Sustainability: The Challenges. Public
Administration Review, 68, S115–S131. doi:10.1111/j.1540- 6210.2008.00983.x
Colciago, A., Ropele, T., Muscatelli, V.A., and Patrizi, T. (2008). The Role of Fiscal Policy
in a Monetary Union: Are National Automatic Stabilizers Effectives? Review of
International Economics.16:591-610.86
Crawford, I., Keen, M., and Smith, S. (2010). Value Added Taxes and Excises. in James
Mirrlees and others (eds), Dimensions of Tax Design: The Mirrlees Review, Oxford:
Oxford University Press for Institute for Fiscal Studies.
Deli, Y., Rodriguez, A. G., Kostarakos, I., & Varthalitis, P. (2018). Dynamic tax revenue
buoyancy estimates for a panel. ESRI Working Paper, No. 592.
Dobbins, L. & Jacob, M. (2016) Do corporate tax cuts increase investments? Accounting
and Business Research, 46:7, 731-759, DOI: 10.1080/00014788.2016.1192985
Dudine, P. & Jalles, T., (2017). How Buoyant is the Tax System? New Evidence from a
Large Heterogeneous Panel. IMF Working Papers.
Dynan, K. & Sheiner, L. (2018). GDP as a Measure of Economic. Hutchins CenterWorking
Paper #43.
Egbunike, C., Emudainohwo, O. & Gunardi, A. (2018). Tax Revenue and Economic
Growth: A Study of Nigeria and Ghana. Signifikan: Jurnal Ilmu Ekonomi. 7.
Froyen, Richard T., (2002). Macroeconomics: Theories and Policies, Seventh Edition.
Upper Saddle River, New Jersey: Prentice Hall.
Fullerton, Don. (1980). Can Tax Revenues Go Up When Tax Rates Go Down? OTA Paper,
41.
Ghozali, I. (2018). Aplikasi Analisis Multivariate dengan Program SPSS 25 Edisi 9. Badan
Penerbit UNDIP.
Gurdal, T., Aydin, M., dan Inal, V. (2020). The relationship between tax revenue,
government expenditure, and economic growth in G7 countries: new evidence from
time and frequency domain approaches. Economic Change and Restructuring.
Hansen, B. & Phillips, P. (1990). Statistical Inference in Instrumental Variable Regression
with I(I) Processes. Review of Economic Studies.
Hidayat, A, & Amachi, TB. (2013). Anslisis Penerimaan Pajak dengan Pendekatan Produk
Domestik Bruto Sektoral. Universitas Indonesia.
Hubbard, G. & O’Brien, A. P. (2017). Macroeconomics, 6th Edition. Pearson.87
Johansson, A., Heady, C. Arnold, J., Brys, B., & Vartia, L. (2008). Tax and Economic
Growth. Economic Department Working Paper No. 620.
Laffer, A. B., (2004). The Laffer Curve: Past, Present, And Future.
Laffer, A. B., Moore, S., & Tanous, P. J. (2008). The End of Prosperity: How Higher Taxes
Will Doom the Economy-If We Let It Happen. New York/London/Toronto/Sydney:
Threshold Editions.
Lee, Y., & Gordon, R. H. (2005). Tax structure and economic growth. Journal of Public
Economics, 89(5-6), 1027–1043. doi:10.1016/j.jpubeco.2004.07.002
Leuthold, J., & N'Guessan, T. (1986). Tax Buoyancy vs Elasticity in a Developing
Economy. BEBR faculty working paper no. 1272.
Lewis, C. (2019). Raising more public revenue in Indonesia in a growth - and equity-
friendly way. OECD Economics Department Working Papers, No. 1534. OECD
Publishing, Paris. https://doi.org/10.1787/a487771f-en.
Mansfield, CY. (1972). Elasticity and Buoyancy of a Tax System: A Method Applied to
Paraguay. International Monetary Fund Staff Papers.
Mawia, M., & Nzomoi, J. (2013). An empirical investigation of tax buoyancy in Kenya.
African Journal of Business Management Vol. 7(40).
Modjo, Mohamad Iksan. (2020). Memetakan Jalan Penguatan Ekonomi Pasca Pandemi.
Jurnal Perencanaan Pembangunan Vol. 4 No. 2: Edisi Khusus Covid-19.
Nurhidayati. (2016). Estimasi Elastisitas dan Buoyancy Pajak Pertambahan Nilai (PPN).
OECD. (2006). International VAT/GST Guidelines.
OECD. (2009). National Accounts at a Glance 2009.
Patrick, J. & Janos, N. (2016). How to design and enforce tobacco excises? International
Monetary Fund.
Pedroni, P. (2000). Fully modified OLS for heterogeneous cointegrated panels.
Nonstationary Panels, Panel Cointegration, and Dynamic Panels, 93–130.
doi:10.1016/s0731-9053(00)15004-288
Pesaran, M.H. & Shin, Y. (1999) An Autoregressive Distributed Lag Modelling Approach
to Cointegration Analysis. In: Strom, S., Ed., Chapter 11 in
Econometrics and Economic Theory in the 20th Century the Ragnar Frisch Centennial
Symposium, Cambridge University Press, Cambridge, 371-413.
Phillips, P. C. B.. (1995). Fully Modified Least Squares and Vector Autoregression.
Econometrica, 63(5), 1023-1078. doi:10.2307/2171721
Pistone, P., Roeleveld, J.J., Hattingh, J., Nogueira, J., West, C. (2019). Fundamentals of
Taxation: An Introduction to Tax Policy, TaxLaw and Tax Administration. Amsterdam :
IBFD.
Schiller, B. R. (2005). Essential of Economics. Columbus :Mcgraw-Hill,Inc
Schratzenstaller, M. (2015). Sustainable tax policy. Revue de l'OFCE.
Setyoningrum, D., & Purwanti, E.Y. (2020). Analisis Tax buoyancy pada Asean-5 Tahun
2002-2016. Indonesian Journal of Development Economics.
https://doi.org/10.15294/efficient.v3i2.39295.
Sjafri, Rika Sari. (2006). Analisis Tentang Penerimaan Pajak Sebagai Fungsi dari Produk
Domestik Bruto Kaitannya dengan Tax buoyancy dan Elastisitas Pajak di Indonesia.
Tesis Universitas Indonesia.
Soewardi, Tiara J., Anandab, Candra F., & Erlandoc, A. (2018). Analisis Hubungan
Kebijakan Fiskal Dan Makroekonomi Daerah Kabupaten/Kota Di Jawa Timur.
Simposium Nasional Keuangan Negara.
Stiglitz, J.E. & Rosengard, J.K. (2015). Economics of the Public Sector Fourth Edition. W.
W. Norton & Company, Inc.
Tanchev, S. dan I. Todorov. (2019). Tax buoyancy and Economic Growth: Empirical
Evidence of Bulgaria. Journal of Tax Reform. 2019;5(3):236–248. DOI
10.15826/jtr.2019.5.3.070
Walewski, Mateusz. (2001). Searching for the Laffer Curve in Transition Economies.
10.1007/978-1-4615-1709-2_6.
Wanniski, Jude.(1978). Taxes Revenues And The Laffer Curve. National Affair.50: 3-16.
World Health Organization. (2020). Raise Tobaco Taxes and Prices for a Healthy and
Prosperous Indonesia.
Yanikkaya, H. & Turan, T. (2017). Tax Structure and Economic Growth: Do Differences in
Income Level and Government Effectiveness Matter?. GTU Working Paper No: 2017-
04
TOWARD SEAMLESS TAXATION
THROUGH A SPLIT PAYMENT MECHANISM
ON THE MARKETPLACE AND QUICK RESPONSE
(QR) PAYMENT SYSTEM
Handhung Dwi Nugroho

Directorate General of Taxes, Indonesia Email: handhungdn@gmail.com
ABSTRACT
The rapid development of the digital economy in Indonesia has presented both challenges and
opportunities for the Directorate General of Taxes (DGT) in collecting taxes. Although digital
economy transactions are difficult to monitor, electronic data from digital economy transactions can
be easily assembled. The author uses the desk review approach in conveying ideas about
alternative governments' efforts to increase tax revenue by utilizing the development of the digital
economy. Based on the studies, the authors propose implementing a split payment mechanism,
especially in trading transactions through the marketplace or using the Quick Response (QR)
Payment System. The implementation of the split payment mechanism can provide benefits to
Taxpayers and the DGT. The split payment mechanism's performance will assist taxpayers in
carrying out their tax obligations, namely calculating, paying, and reporting taxes. On the other hand,
implementing the split payment mechanism can increase tax revenues with low supervision costs
and provide better government cash flow.
Keywords: split payment, e-commerce, marketplace, fintech, QR payment
Perkembangan ekonomi digital yang sangat pesat di Indonesia telah menghadirkan tantangan
sekaligus peluang bagi Direktorat Jenderal Pajak (DJP) dalam mengumpulkan penerimaan negara.
Meskipun transaksi ekonomi digital sulit untuk diawasi, data elektronik dari transaksi ekonomi digital
dapat dengan mudah dikumpulkan. Penulis menggunakan pendekatan metode deskreview dalam
menyampaikan gagasan mengenai upaya yang dapat ditempuh pemerintah untuk meningkatkan
penerimaan pajak dengan memanfaatkan perkembangan ekonomi digital. Berdasarkan kajian yang
telah dilakukan, penulis mengusulkan untuk menerapkan mekanisme split payment khususnya
pada transaksi perdagangan melalui marketplace dan transaksi dengan menggunakan Quick
Response (QR) Payment System. Impelementasi mekanisme split payment dapat memberikan
manfaat kepada Wajib Pajak dan DJP. Implementasi mekanisme split payment akan membantu
Wajib Pajak dalam melaksanakan kewajiban perpajakannya yaitu penghitungan dan penyetoran
pajak serta mendapatkan pencatatan yang tertib guna memudahkan pelaporan. Di lain pihak,
implementasi mekanisme split payment dapat meningkatkan penerimaan pajak dan memperlancar
cash flow pemerintah dengan biaya pengawasan yang rendah.
Kata kunci: split payment, e-commerce, marketplace, fintech, QR payment
1. INTRODUCTION
1.1. Background
The development of information and technology in recent years has changed how
people do their activities and how entrepreneurs run their businesses. With a
smartphone's help, everyone can shop for various needs and make multiple payments
through bank transfers or other payment applications very quickly. Likewise,
entrepreneurs no longer need to depend on a network of distributors or large showrooms
to be able to sell their products. The development of e-commerce and payments through
financial technology (fintech) has made all of this possible. Based on data presented by
katadata (2021), the growth of e-commerce in Indonesia has grown almost fourfold in the
last three years. In addition, the use of electronic money in Indonesia also increased by
344% in the same period (Bank Indonesia, 2022).
This rapid growth of the digital economy must receive serious attention from the
government, especially the Directorate General of Taxes (DGT), as the tax authority in
Indonesia. With such rapid growth, DGT must be able to ensure that these digital economy
actors have fulfilled their tax obligations correctly. Since the repeal of Minister of Finance
Regulation Number 210/PMK.010/2018 concerning Tax Treatment of Trade Transactions
Through Electronic Systems (e-commerce), which had not even been implemented in
2019, there has not been a significant breakthrough made by DGT to optimize tax revenue
from perpetrators business through electronic systems.
Several countries in the world have made various breakthroughs so that no tax
potential from their countries is lost due to the development of the digital economy. The
European Union has introduced a Mini One-Stop Shop (MOSS) scheme. Australia has
required online trading companies (including marketplaces) with a particular income value
to register with the Australian Tax Office. In addition, several countries, such as Italy,
Poland, and Romania, have implemented a split payment mechanism to ensure that VAT
payments are carried out correctly. The split payment mechanism is a tax collection
method that separates payments made by customers into the taxpayer's account in the
amount of income/transaction value and the state account in the amount of the tax payable.
Apart from presenting challenges in collecting taxes, the growth of the digital economy,
especially e-commerce and fintech, also present opportunities that can be exploited to
increase tax revenues. Opportunities that arise from this digital economy include the
existence of electronic data collected from various existing transactions. This data can
assist taxpayers in carrying out their tax obligations or for the DGT monitoring taxpayer
compliance. With the available information, taxpayers can fulfill their tax obligations
without feeling burdened or even without feeling that they have implemented it
(seamlessly), and the tax authorities can achieve optimal compliance without much
intervention. Therefore, the development of existing technology and information is
expected to not only be able to change the way a person does business or activities but
also change the way taxpayers exercise their tax rights and obligations. In the end, this
convenience is expected to increase taxpayer compliance and tax revenues in Indonesia.
1.2. Problem Statement

The digital economy has proliferated in recent years. In Southeast Asia, digital
economic value (Gross Merchandise Value/GMV) in 2015 was recorded at US$ 32 billion
(1.5% of GDP). This value has almost doubled in three years, so it was worth US$ 72
billion (2.8% of GDP) in 2018 and is expected to continue to increase to US$ 240 billion
(8% of GDP) in 2025 (katadata, 2018). The development of the digital economy has
resulted in a new way of payment through fintech and one of the new business models,
including e-commerce.
In line with the growth of the digital economy in the Southeast Asia region, the use of
electronic money in Indonesia has grown by 344% in the last three years. It has increased
by more than 1,100% in the previous five years, as shown in the Graph of the Number of
Electronic Money Instruments. There are two types of electronic money in Indonesia: e-
money and e-wallet. Electronic money in the form of e-money refers to electronic money
in the form of cards. Examples of e-money are e-toll cards, flazz cards, and others that
can be used for offline payments, such as TransJakarta or KRL payments. Meanwhile, an
e-wallet is an electronic money in the form of an application. Gopay and OVO are some
examples.
700,00
Number of Instrumens (milion)
600,00
500,00
400,00
300,00
200,00
100,00
0,00
2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021
Graph of The Number of Electronic Money Instruments

Source: Bank Indonesia
Electronic money is snowballing because it provides convenience and simplicity in

doing transactions. Everyone who is going to make a transaction no longer needs to carry
physical money (paper or coins) to make payments. They only need to bring e-money or
a smartphone to buy the goods or use the services they want. Sellers or service providers
are also beneficial because they don't have to bother preparing money back in
transactions. Many MSMEs or places of business have accepted payments using
electronic money. This convenience in transactions encourages electronic money to be
able to quickly enter the informal economy, which is one of the sectors that are difficult to
reach by the tax authorities in Indonesia.
Among the various new business models that are developing, electronic commerce
(e-commerce) is one that we can easily find, and its development is massive. OECD (2002,
p.89) defines e-commerce as buying and selling goods or services through computer
networks with explicit methods for ordering/buying or receiving orders/purchases. People
can order goods or services via computer, but payment and delivery can be made offline.
E-commerce transactions can be carried out between companies, households, individuals,
governments, or others. Interaction between sellers and buyers in e-commerce
transactions can be carried out through websites owned by sellers or third parties that
provide online markets (marketplaces). An example of websites sellers directly own is
www.samsung.com, and examples of marketplaces are Tokopedia and Bukalapak. Based
on the 2021 Bank Indonesia Annual Meeting Report as quoted in Katadata (2021), among
other things, it is stated that e-commerce transactions in Indonesia in 2021 are predicted
to grow by 51.6% and are projected to continue to increase in 2022 with a value reaching
IDR 530 trillion or growing by 31.4%.
600
530
Transaction value (Rp triliun)
500
403
400
300 266
206
200
106
100
0 *) projected number
2018 2019 2020 2021* 2022*
Graph of E-Commerce Transactions in Indonesia

Source: Katadata (2021)
The rapid development of the digital economy, primarily e-commerce, has at least
three direct impacts on tax revenues. First, e-commerce can cut several distribution chains
from producers to consumers. With online trading, producers can directly sell their
products to users. Therefore, this has an impact on the income that distributors receive.
Second, online trading is just a change in the way of doing transactions that don't actually
create a new market. In other words, e-commerce consumers are mainstream-market
consumers who move to online markets (Simkin, Bartlett, & Shim, 2011). Therefore,
income from taxpayers routinely supervised by the tax authorities shifts to taxpayers who
may need to be entirely handled properly. Third, online trading poses a challenge to the
tax apparatus in carrying out supervision since, generally, their business is not physically
visible. Thus, possession of sufficient data by the tax authorities is an absolute
requirement to support optimal supervision.
The inaccessibility of the informal and digital economy in Indonesia is one of the issues
that the Indonesian government must appropriately handle to increase the tax ratio.
Indonesia has a reasonably low tax ratio when compared to other countries in the world.
According to the International Monetary Fund (2011, p.60), the average tax ratio for lower-
middle-income countries, which includes Indonesia, is 16.5%. Even Indonesia's tax ratio
is lower than the average tax ratio of low-income countries (lower income), which equals
13.9%. In continuing efforts to increase the tax ratio, Indonesia is faced with the fact that
there has been a decrease in the tax ratio in the last ten years.
With the industrial revolution 4.0 taking place, where the internet has become an
inseparable part of the life of the world community, the government must start thinking
about turning all the challenges resulting from technological developments into
opportunities to increase tax revenues. By using the desk review approach based on
secondary data sourced from journals, government publications, and other relevant
sources, the author will convey ideas about the efforts that the government can take to
increase tax revenues by utilizing digital economic developments.
2. ANALYSIS
The Indonesian government remains active in facing the rapid development of the
current digital economy. The government has issued Presidential Regulation Number 74
of 2017 concerning the 2017-2019 Road Map for the National Electronic Trading System
(E-Commerce Road Map) as the primary guide for all elements of the government in
formulating policies related to e-commerce. Furthermore, the Minister of Finance
stipulated Minister of Finance Regulation Number 210/PMK.010/2018, which regulates
the provisions for fulfilling tax obligations for e-commerce. The Minister of Finance
Regulation was drafted to guarantee equal treatment between trading through electronic
systems and conventional trading. Regarding the taxation aspect, there is nothing new
regulated in PMK 210/PMK.010/2018. One mechanism regulated explicitly in this
regulation is the obligation for marketplace platform providers to report the recapitulation
of trading transactions carried out by traders or service providers to the DGT every month.
However, in the end, PMK 210/PMK.010/2018 was revoked before it took effect, which
was planned to start on April 1, 2019.
In order to deal with economic challenges due to the threat of the Corona Virus
Disease 2019 (Covid-19) pandemic, the government issued a Government Regulation in
Lieu of Law Number 1 of 2020 regarding state financial policy and financial system stability,
which was subsequently stipulated in Law Number 2 of 2020. Through these provisions,
the government stipulates policies in the field of taxation which, among other things,
regulate the imposition of Value Added Tax (VAT) on the utilization of Intangible Taxable
Goods or Taxable Services from outside the customs area within the customs area as well
as the imposition of income tax on Transaction Through the Electronic System (TTES).
This is a breakthrough in handling digital economy taxation. However, even though Law
Number 2 of 2020 provides a legal basis for the imposition of VAT and income tax on
TTES activities, so far, it has only been implemented regarding VAT collection. Thus, the
government still needs specific steps to implement the income tax on TTES.
The presence of the government in regulating tax treatment for e-commerce is
essential. Forst (1999, p.711) emphasized that provisions regarding e-commerce taxation
are needed to prevent taxpayers from trying to avoid paying taxes. Furthermore, Simkin,
Bartlett, & Shim (2011, p.68) also conveyed the importance of taxation for e-commerce to
ensure equality of treatment with conventional business, the e-commerce business is a
large business with significant revenue potential, and in general, Internet users are part of
society who have the ability to pay taxes.
Several countries in the world have even made various breakthroughs so that no tax
potential from their countries is lost due to the development of the digital economy. Since
2015, the European Union has introduced a Mini One-Stop Shop (MOSS) scheme. MOSS
is an online service created to make it easier for companies to manage tax collection from
consumers and to help companies comply with tax regulations, so they don't have to get
caught up in complicated administrative matters. Furthermore, since July 1, 2018, the
Australian government has obliged online trading companies (including marketplaces)
with a particular income value to register with the Australian Tax Office and are required
to collect tax from consumers following Australian tax regulations. Table 1 shows a
comparison of e-commerce taxation in the two countries.
Table 1. Comparison of E-Commerce Taxation in the European Union and Australia

No Criteria Uni Eropa Australia
1 Tax type VAT Goods and Services Tax
2 Tariff According to the buyer's 10%

country
3 Threshold According to the buyer's AUD 10.000

country
4 Deposit € 8.000 for each country -
5 Taxpayer Yes Yes, with omzet more than AUD

registration 75.000 per year
6 Mechanism MOSS Collected by marketplace
Efforts to tax the digital economy are still being studied and discussed by practitioners
and academics. Various countries are still trying to find the best alternative for their
respective countries not to lose their tax revenues. Through HM Revenue & Customs
(HMRC), the British government is also conducting a study to find the best alternative for
collecting taxes, especially regarding VAT collection in the digital economy. HMRC (2018)
believes that a long-term tax collection solution must rely on more than just the
marketplace. Tax collection must pay attention to the future economy, where people can
buy goods with various channels and various alternative ways of payment. Therefore,
HMRC developed a tax collection mechanism by considering the payment cycle. This
mechanism is known as the split payment mechanism, where banks are systemically
expected to be able to separate the portion of tax payable that must be deposited to the
state when buyers make payments for the transactions they make. With the cooperation
between HMRC and the banks that serve as payment channels for commerce, taxes can
be collected without depending on the presence of taxpayers who may be outside the
territory of the UK in connection with transactions that occur via the internet.
Even though the UK is still in the stage of studying the implementation of the split
payment mechanism, several countries have implemented this mechanism. Countries
implementing a split payment mechanism include Italy, Poland, and Romania. Italy has
implemented this mechanism since January 1, 2015, although it is limited to business-to-
government (B2G) transactions. Suppliers will charge VAT on goods or services delivered
to specific agencies in this mechanism. Furthermore, the customer will divide the payment
of bills received from suppliers in the amount of the price of the goods and deposit the
VAT payable to the state account. This mechanism is similar to the VAT collection carried
out by the Treasurer of the Government or SOE in Indonesia.
The split payment mechanism in Poland has been implemented since July 1, 2018,
with a broader range of transactions compared to the implementation in Italy. The split
payment mechanism in Poland does not require customers to separate payments. Poland
mandates banks to separate payments for transactions into business bank accounts and
VAT special accounts (VAT bank accounts) belonging to providers of goods or services.
In Romania, the split payment mechanism has been implemented since January 1,
2018, for taxpayers with specific VAT payable limits. Companies that are obliged to make
split payments are required to open a separate account (VAT bank account) for the
collection and payment of VAT. The taxpayer can only use the VAT bank account for input
and output VAT payments. In the split payment mechanism in Romania, the goods and
services provider charges VAT, and then the customer transfers the VAT directly to the
VAT bank account belonging to the goods and services provider.
3. POLICY RECOMMENDATION
“Insanity is doing the same thing over and over and expecting different results" (Albert
Einstein). This well-known phrase from Einstein can be used as a philosophy for DGT to
increase Indonesia's tax ratio. DGT can't just rely on old efforts to improve the declining
tax ratio. DGT must take extraordinary steps that have never been taken before. Especially
with the fast-changing times, there is no reason for DGT not to innovate.
It is common practice to secure tax revenue from the digital economy by only expecting
data to be submitted to the DGT as previously stipulated in PMK 210/PMK.010/2018. DGT,
in recent years, has received a flood of data from various parties. This data is vital for
DGT, but converting existing data into tax revenue is complex. It requires considerable
effort and costs for the DGT to carry out data analysis, appeals to taxpayers, audits, and
billing.
The challenge of supervising tax compliance in Indonesia is also closely related to the
current self-assessment tax collection system. With a self-assessment system, taxpayers
are entrusted with fulfilling their tax obligations, starting from registering and calculating
the tax payable as well as depositing and reporting the tax. Therefore, tax officers
supervise by comparing the data owned by the DGT and the data that the taxpayer has
reported. In addition, often, the self-assessment mechanism cannot work correctly
because the taxpayer needs to fulfill his obligations properly. Taxpayers need to
understand their obligations, taxpayers feel bothered and lazy to carry out their
responsibilities, or even taxpayers deliberately neglect their responsibilities are some of
the causes of self-assessment not going well. Thus, aspects of ease of fulfilling obligations
and monitoring compliance must be a concern for the government in formulating policies.
The development of tax services in Indonesia, especially digital-based ones, has been
widely utilized by taxpayers, for example, e-filing, e-invoices, e-forms, and so on. The
government developed these services intending to facilitate taxpayers in carrying out their
tax obligations, which in turn still rely on voluntary compliance. By depending on tax
supervision on voluntary compliance, in the end, Taxpayers can make choices regarding
reporting, calculating, and paying taxes. Options that taxpayers can consider are not only
to comply or not comply but also include efforts made to become compliant such as
recording transactions, compiling financial reports, filling out tax returns, meeting
reporting, and payment deadlines, or seeking information to understand tax provisions
(OECD, 2020, p.11). Fulfilling tax obligations that place too much emphasis on voluntary
compliance can trigger injustice between compliant and non-compliant taxpayers (OECD,
2022, p.12).
As mentioned in the OECD Report (2020), two of the six essential elements that the
tax authorities must pay attention to in developing Tax Administration 3.0 to deal with
digital transformation are “embedded within taxpayer natural systems” and “part of a
resilient system of systems" (p.12). Based on these two aspects, tax collection should be
part of the Taxpayer's business process so that it is carried out seamlessly by the
Taxpayer. Collaboration between tax institutions and taxpayers is needed in developing
tax services so that they can reduce administrative burden, increase data security, and be
transparent and reliable. This step can also reduce the dependence of tax collection on
voluntary compliance due to the integrated system design in the taxpayer's business
processes. In addition to the inclusion of tax administration in the business processes of
taxpayers, the taxation system owned by the taxation institution is no longer the only point
for processing taxation data. Similar to the role of Taxable Entrepreneurs in collecting VAT
and employers deducting Article 21 Income Tax, digital platforms can also become
"agents" of tax institutions that carry out tax administration processes in their systems.
Concerning the various problems, challenges, and opportunities faced by Indonesia in
increasing the tax ratio as described above, extraordinary solutions are needed by utilizing
existing technology and information developments. With an increasing number of people
making payments through fintech and shopping through marketplaces, the authors
propose that the Indonesian government can implement a split payment mechanism for at
least these two digital services. The split payment mechanism in Indonesia can be
implemented by giving a role to fintech and marketplace service providers to make split
payments between the taxpayer's income and the tax obligations they bear. Thus,
taxpayers who earn income through these two instruments will no longer be preoccupied
with the duty to calculate and deposit taxes (seamless). On the other hand, the
government can also quickly and accurately collect taxes without relying on voluntary
compliance.
Split payment mechanisms through fintech and marketplaces can be implemented for
calculating and paying the final 0.5% Income Tax for MSMEs or VAT for Taxable
Entrepreneurs. In general, fintech or marketplace will calculate and deposit taxes owed by
sellers or service providers who receive payments through fintech or marketplaces. Thus,
DGT can monitor tax payments in real time, and taxpayers will receive a recapitulation of
tax payments that they can use for reporting through a tax return. This mechanism will
also help taxpayers who find carrying out their tax obligations challenging.
3.1. Implementation of Split Payment Mechanism on Marketplace

One of the e-commerce trading media is the marketplace. Through the marketplace,
everyone can easily sell the goods or services they offer to potential buyers via the internet.
The seller, buyer, and all transaction data are entirely recorded in the marketplace
database. In addition, the marketplace also provides peace of mind to buyers with the joint
account system offered, where the seller only processes sales after the marketplace
receives payment from the buyer. Then the money is only transferred to the seller after
the buyer states that he has received the goods according to what he wants.
The transaction process through the marketplace is relatively simple and easy. Before
carrying out the bidding process, sellers are required to register into the marketplace
application system by submitting some information. After registration is complete, new
sellers can place offers of goods or services to be sold. Furthermore, prospective buyers
can choose various products in the marketplace. If the buyer has made a payment, the
seller sends the goods ordered to the buyer. After the goods are declared received by the
buyer, the payment money is transferred to the seller's account.
Based on the description above, DGT can implement a split payment mechanism by
intercepting the process in the marketplace. Armed with the seller's data owned by the
marketplace, the marketplace platform provider can separate the seller's tax (income tax
article 4 (2) or VAT) and the seller's transaction value during the transfer process to the
seller. So, the marketplace can systematically generate billing codes and simultaneously
deposit taxes into the state treasury for each transaction. By utilizing existing records on
the marketplace, taxpayers can easily monitor their transactions and use them in preparing
periodic or annual tax returns. In addition, tax officers can also easily monitor if there are
sellers who should have been confirmed as Taxable Entrepreneurs based on tax payment
data that has been made.
Flowchart of Split Payment Mechanism on Marketplace
3.2. Implementation of Split Payment Mechanism on Fintech

The types of financial technology (fintech) services currently developing in Indonesia
are diverse. An e-wallet service that can be used for digital payments via a quick response
code (QR code) is growing rapidly. We can see this with the rise of the financial services
industry and conventional banks currently developing digital payment technology. With
QR payment in Indonesia having become an open system where the QR code at a
merchant is universal and can be accessed by all e-wallet service providers, the
development of cashless transactions will be faster and more widespread.
With QR payments, payments for transactions at outlets (stores) directly or e-
commerce (online) can be made via a smartphone by scanning the QR code. As the name
implies, payments via QR code are made quickly and easily, no longer needing a card to
be swiped at the EDC machine. To get a QR code as an identity for payment at the outlet,
merchants/service providers (taxpayers) must register with e-wallet service providers.
Furthermore, the QR code that has been obtained is installed at the outlet or website
where the Taxpayer sells his goods/services. Generally, e-wallet service providers make
transfers (settlements) of all payments received by taxpayers via QR payments into their
accounts every day (accumulation in a day) or based on a disbursement request by the
account owner.
Based on the description above, DGT can implement a split payment mechanism for
fintech service providers. Fintech service providers can be obliged to separate taxpayers'
tax and income when transferring payments via QR payments that have been collected.
With the data held during registration, the fintech service provider creates a billing code
on behalf of the Taxpayer and, at the same time, deposits taxes into the state treasury for
each transfer (settlement) made. As with the split payment mechanism in the marketplace,
taxpayers and DGT can monitor tax payments created.
Flowchart of Split Payment Mechanism on Fintech
3.3. Optimization of Split Payment Mechanism

Implementing the split payment mechanism is expected to reduce the tax gap between
the existing potential and the collected tax revenue, which can increase the tax ratio. The
split payment mechanism has several advantages, including the following:
a. Increased tax revenue because it will be increasingly difficult for taxpayers to do tax
evasion;
b. Tax revenues are quickly collected to the state treasury so that they can help the
government's cash flow;
c. Low monitoring costs because DGT does not only receive transaction data from the
marketplace but also in the form of tax revenue received;
d. Taxpayers are no longer bothered with the obligation to calculate and deposit taxes;
e. Taxpayers obtain complete transaction data recording of the transactions they make;
f. DGT can focus on exploring potential tax revenue by supervising based on consumer
transaction data.
Apart from the advantages described above, several potential problems can arise from
implementing the split payment mechanism. By understanding the benefits and challenges
of implementing the split payment mechanism, it is hoped that the government will also
prepare other steps that must be taken to optimize the results achieved and reduce the
negative impacts that arise. The following are potential problems and efforts the
government must prepare regarding implementing the split payment mechanism.
a. Burden taxpayers’ cash flow considering that taxpayers can calculate the tax owed at
the end of the month, but the split payment mechanism requires them to deposit taxes
every time a transaction occurs.
Burden taxpayers’ cash flow will be widely felt in implementing split payments for VAT
payments. In the general practice of calculating and paying VAT, taxpayers can deduct
all input taxes they receive from the output taxes that must be paid so that taxpayers
only pay the difference. In the split payment mechanism, all output taxes must be paid
in advance when the transaction occurs without being able to deduct the input tax
taxpayers have. Thus it has the potential to burden the cash flow of taxpayers.
As a solution to these problems, DGT must guarantee a quick and easy tax refund
mechanism. With a reliable tax return, taxpayers have guaranteed certainty in doing
business.
b. Differences in treatment between taxpayers who operate conventionally (not included
in the split payment mechanism) and those who are included in the split payment
mechanism.
The implementation of the split payment mechanism, as previously explained, still
needs to reach all existing taxpayers. For taxpayers who operate
manually/conventionally, they still fulfill their tax obligations in accordance with the
once generally accepted tax provisions. Thus, more effort is needed to ensure tax
payments are made following the requirements. With the various advantages of the
split payment mechanism, the government will likely encourage taxpayers to enter the
existing split payment scheme voluntarily.
The government's efforts to encourage taxpayers to no longer transact manually are
also a financial inclusion program promoted since 2013. The formation of a more
inclusive financial ecosystem can provide many benefits to various parties. Therefore,
implementing the split payment mechanism in the tax sector can also be a driving force
for accelerating financial inclusion in Indonesia.
The government can provide tax facilities to encourage taxpayers further to enter into
the split payment mechanism and offer incentives to taxpayers already in the split
payment scheme. For example, tax facilities that can be provided are a tax deduction
of 5% - 10% of the tax payable, which is paid through a split payment mechanism.
c. The emergence of tax collection costs must be borne by the marketplace and fintech
providers.
In carrying out the role as a party in charge of depositing tax payments from taxpayers,
marketplace and fintech providers act as Collecting Agents as stipulated in Minister of
Finance Regulation Number 32/PMK.05/2014 concerning Electronic State Revenue
Systems as amended by Minister of Finance Regulation Number 202/PMK.05/2018.
Based on these provisions, Collecting Agents may be given compensation or state
revenue services for each successful transaction of Billing Codes. Thus, this clause
can be used as a reward for marketplace and fintech providers so that their business
will continue to grow and they are expected to become reliable government partners.
4. CONCLUSION AND RECOMMENDATION

The tax ratio is a measurement commonly used to measure the general description of
a country's tax conditions. Indonesia's tax ratio is relatively low compared to several
countries in the world. When Indonesia is eager to increase its tax ratio, data for 2013-
2020 shows that Indonesia's tax ratio has decreased. According to the International
Monetary Fund (2011, p.60), the average tax ratio for lower-middle-income countries,
which includes Indonesia, is 16.5%. To be able to increase the tax ratio, DGT is required
to be able to follow various economic developments and optimize available technological
developments.
The rapid growth of the digital economy in Indonesia in recent years can be made it a
priority for the government to continue to increase the tax ratio. Apart from the increasing
proportion of the digital economy in Indonesia, digital economic growth also promises data
integration and ease of transaction processes. With electronic data that is increasingly
easy to obtain, a collaboration between tax institutions and taxpayers will be able to reduce
administrative burdens and increase data security, transparency, and reliability. Tax
administration is expected to become part of the business processes of taxpayers, and
tax institutions are no longer the only point for processing tax data. Thus taxpayers can
fulfill their tax obligations without feeling burdened or even without feeling that they have
implemented it (seamlessly), and the tax authorities can achieve optimal compliance
without much intervention.
E-commerce and fintech are digital economy sectors that are proliferating in Indonesia
but have yet to be entirely handled by the tax authorities in Indonesia. By utilizing e-
commerce that operates on marketplaces and fintech that develops QR payment systems,
DGT can implement a split payment mechanism to collect taxes and ensure that taxes are
calculated and paid correctly. The split payment mechanism is a tax collection method that
separates payments made by customers into the taxpayer's account in the amount of
income/transaction value and the state account in the amount of the tax payable. The split
payment mechanism can be implemented in collecting income tax article 4 (2) and VAT
by involving third parties, namely marketplace providers and fintech providers, to split the
proportion of the tax and the proportion of the transaction value the right of the taxpayer.
Implementation of the split payment mechanism can provide benefits to taxpayers and
the Directorate General of Taxes. From the taxpayer's side, the performance of the split
payment mechanism will assist them in carrying out a self-assessment system, namely
calculating and depositing taxes and obtaining orderly records to facilitate reporting. With
the split payment mechanism that has become part of the Taxpayer's business process,
the Taxpayer can become compliant without having to be burdened with various
administrative obligations. On the other hand, implementing the split payment mechanism
can increase tax revenues and facilitate government cash flow with low monitoring costs.
This mechanism simultaneously creates structured compliance by reducing excess
dependence on voluntary compliance. However, the government also needs to be aware
of the potential burden on taxpayers’ cash flow, costs incurred for marketplace and fintech
providers, and the potential for unequal treatment between taxpayers.
The split payment mechanism that the author describes in this article is different from
the one already implemented in Italy, Poland, and Romania or planned to be implemented
in the UK. The government cannot simply imitate or refer entirely to these countries but
must prepare the necessary steps before implementing this split payment mechanism.
The author is also fully aware that this paper has many limitations. Therefore, further
studies and the government's active role are needed to realize the implementation of the
split payment mechanism. Some aspects that need to be considered by the government
include the following:
a. Prepare regulations as the basis for implementing the split payment mechanism.
Tax collection by the government must be based on applicable legal provisions.
Likewise, the implementation of the split payment mechanism must be based on the
requirements that cover it. It is time for the government to reformulate the provisions
governing taxation in the digital economy sector as a replacement for PMK
210/PMK.010/2018, which was repealed prior to implementation. Learning from
experience, implementing a split payment mechanism will be very sensitive to the
business world, so it needs adequate communication and public involvement.
b. Amend Minister of Finance Regulation Number 184/PMK.03/2007
Article 4 of the Minister of Finance Regulation Number 184/PMK.03/2007 concerning
the Determination of Due Dates for Payment and Deposit of Taxes, Determination of
Places for Payment of Taxes, and Procedures for Payment, Deposit, and Reporting of
Taxes, as well as Procedures for Installment and Postponement of Tax Payments as
amended by the Minister of Finance Regulation Number 80/PMK.03/2010, stipulates
that tax payments and deposits are made at the post office or bank appointed by the
Minister of Finance. With regard to these provisions, it is necessary to amend these
provisions by including other places of payment (other than postal or bank) to be used
as places for payment and deposit of taxes.
c. Requiring marketplace and fintech providers who provide QR payments to become
Collecting Agents
Minister of Finance Regulation Number 32/PMK.05/2014 concerning the Electronic
State Revenue System as amended by the Minister of Finance Regulation Number
202/PMK.05/2018 regulates, among other things, the mechanism for appointing Other
Collecting Agents, implementing UAT, administering state revenues, etc. Since this is
a prerequisite for marketplaces and fintech to receive tax payments, all marketplaces
and fintech must carry out all the regulated procedures. To further facilitate and
expedite this program, the government can review these provisions in the hope that
they will be simplified further.
d. Prepare a study on the provision of tax facilities to create a digital financial ecosystem.
In line with the financial inclusion program and increasing tax revenue, the government
is expected to provide incentives for taxpayers who have entered the split payment
scheme. The form and size of the facilities that can be delivered to taxpayers must be
studied in depth to obtain them optimally. Therefore, further studies are needed
regarding the provision of this facility.
BIBLIOGRAPHY
European Commission (2017) Analysis of the Impact of the Split Payment Mechanism as
an Alternative VAT Collection Method
Forst, D.L. (1999) Old and New Issues in the Taxation on Electronic Commerce. Berkeley
Technology Law Journal, 14(2), 711-719
HM Revenue & Customs (2018) Alternative Method of VAT Collection – Split Payment
International Monetary Fund (2011) Revenue Mobilization in Developing Countries
Katadata (2018) Ekonomi Internet Indonesia Baru Mencapai 2,9% Terhadap PDB
(https://databoks.katadata.co.id/datapublish/2018/11/30/ekonomi-internet-
indonesia-baru-mencapai-29-terhadap-pdb)
Katadata (2021) Transaksi E-Commerce Indonesia Diproyeksikan Capai Rp 403 Triliun
pada 2021 (https://databoks.katadata.co.id/datapublish/2021/11/25/transaksi-e-
commerce-indonesia-diproyeksikan-capai-rp-403-triliun-pada-2021)
OECD (2002) Measuring the Information Economy. OECD Publishing
OECD (2020) Tax Administration 3.0: The Digital Transformation of Tax Administration,
OECD Forum on Tax Administration, OECD Publishing
OECD (2022) Toward Seamless Taxation: Supporting SMEs to Get Tax Right, OECD
Forum on Tax Administration, OECD Publishing
Simkin, M.G., Bartlett, G.W. & Shim, J.P. (2011) Pros and Cons of E-Commerce Taxation.
International Business & Economics Research Journal 1(2), 61-71
DOES GIVING CARROT WORK EFFECTIVELY? A STUDY
OF TAX RELAXATION IN INDONESIA
[Christine Novita Dewi]a [Steffy Madelen Priskila Pasaribu]b
[Dielanova Wynni Yuanita]c
a[Universitas Kristen Duta Wacana Jl. dr. Wahidin Sudirohusodo No. 5-25 Yogyakarta,
Indonesia] Email: [christine_n_dewi@staff.ukdw.ac.id]
b[Universitas Kristen Duta Wacana Jl. dr. Wahidin Sudirohusodo No. 5-25 Yogyakarta,
Indonesia] Email:[steffymdlnpriskila@gmail.com]
c[Universitas Kristen Duta Wacana Jl. dr. Wahidin Sudirohusodo No. 5-25
Yogyakarta, Indonesia] Email:[dielanova@staff.ukdw.ac.id]
ABSTRACT
This study aims to empirically prove the effect of tax understanding, tax awareness and tax
relaxation on the taxpayers’ compliance. The research sample was 65 respondents using a
purposive sampling method. The writer used descriptive statistical analysis and multiple linear
regression analysis as the analysis methods. The results show that tax understanding has a
positive effect on tax compliance, while the tax awareness and the tax relaxation do not
significantly affect taxpayer compliance during the Covid-19 pandemic.
Keywords: tax compliance, understanding, awareness and relaxation
Penelitian ini bertujuan untuk membuktikan secara empiris pengaruh pemahaman pajak,
kesadaran pajak dan relaksasi pajak terhadap kepatuhan wajib pajak. Sampel penelitian sebanyak
65 responden menggunakan metode purposive sampling. Teknik analisis data yang digunakan
yaitu uji statistik deskriptif dan analisis regresi linear berganda. Hasil penelitian menunjukkan
bahwa pemahaman pajak berpengaruh positif dan signifikan terhadap kepatuhan pajak,
sedangkan kesadaran pajak dan relaksasi pajak tidak terlalu berpengaruh terhadap kepatuhan
wajib pajak di masa pandemi covid-19.
Kata kunci: kepatuhan, pemahaman, kesadaran & relaksasi pajak
1. INTRODUCTION
Research Background
Tax is the largest source of revenue for the Indonesian government. It is because the
government has a policy that people's income is given to the government in accordance
with the tax rate as a support for the government. It is known that the taxpayer’s
compliance ratio in 2020 is 78%, even though the target ratio is 80%. One of the
requirements as a Taxpayer is having a Taxpayer Identification Number. After becoming
a taxpayer, the person has tax rights and obligations. The taxpayers who is able to fulfill
their tax rights and obligations is called compliant taxpayers. By registering themselves
as a taxpayer, the taxpayer should understand the tax regulations. The taxpayers will
think that the function and purpose of paying taxes will not have much impact on them
and feel unsure about complying with tax regulations if they do not understand the
taxation (Ruky, et al. 2018).
The Indonesian government has three tax collection systems; an official assessment
system assisted by the tax authorities, a self-assessment system where the taxpayers
calculate their own tax payable, and a withholding system assisted by related agencies.
However, the taxpayers’ compliance still relies on taxpayers' awareness even though the
collection system has been assisted by the tax authorities or other agencies. According
to Megawangi and Setiawan (2007) tax awareness is when people voluntarily do their
tax obligations and this is a form of responsibility from the people to the government. The
global situation has been in turmoil since the beginning of 2020 because of the
Coronavirus Disease 2019 (Covid-19) emergence. Indonesia is one of the countries
affected by the virus, therefore the government decided to tighten the activities of its
people, which caused a contraction in the economy. The government issued a tax
relaxation policy to help people affected by the pandemic and revive the country's
economy.
2. THEORETICAL FRAMEWORK AND HYPHOTHESIS

DEVELOPMENT
This study implements the attribution theory which focuses on the behavior of a
person to another person, observes people’s behavior, and uses the theory of planned
behavior to predict and understand about people’s behavior.
2.1. Taxpayer compliance

Based on Nurkhin et al. (2018), taxation compliance is a condition when the taxpayers
do their obligations and get their rights related to the taxation. According to Rahayu
(2010), the taxpayers’ compliance is classified as formal and material compliance.
2.2. Tax understanding

Tax understanding is a condition when the taxpayers know and understand about the
taxation. A taxpayer is said to understand taxation when the taxpayer uses the self-
assessment system, they understand the steps for paying and reporting taxes. If they
have low tax understanding, the taxpayer will face difficulty doing their taxation
responsibility. It means the taxpayer’s compliance will decrease. Therefore, when the
taxpayer has a good tax understanding, then the level of tax compliance will increase.
The previous study finds that tax understanding is proven to affect the taxpayer’s
compliance (Nurkhin et al., 20`8; Agustiningsih, 2006; Ruky et al., 2018).
H1: tax understanding has a positive effect to taxpayer’s compliance.
2.3. Tax awareness

According to Tanilasari and Gunarso (2017), tax awareness is a condition of the
taxpayers who have the desire and awareness to comply with applicable tax regulations
in order to meet their tax obligations. Based on Theory of Planned Behavior (TPB), tax
awareness can be linked with the taxpayers believing them to pay the tax. The taxpayers’
low awareness of paying and reporting their tax results in lower tax compliance.
Otherwise, if the taxpayers’ awareness is high, then the tax compliance is also higher.
The previous studies show that tax awareness affects the taxpayers’ compliance
(Nurkhin et al., 2018; Agustiningsih, 2016; Ruky et al., 2018; Suriambawa & Setiawan,
2018).
H2: tax awareness has a positive effect on taxpayers’ compliance.
2.4. Tax Relaxation

The government issued a tax relaxation policy to help people in facing the pandemic
and hoping to revive the country’s economy. The policy benefits the taxpayer by getting
flexibility to report their tax and get tax reduction. The taxpayers can neglect their tax
obligations if they do not know and do not experience the tax relaxation. Otherwise, the
tax relaxation policy can boost the taxpayers’ compliance who experience the policy to
report and pay their tax. The previous study states that the tax relaxation policy affects
the taxpayer’s compliance (Dewi et al., 2018; Alfina et al., 2021).
H3: tax relaxation has a positive effect on taxpayers’ compliance

3. RESEARCH METHODS
This study uses the quantitative method. The target population of this study is the
individual taxpayers in Yogyakarta Primary Tax Office. There were 65 respondents in
the research sample. Questionnaires are used as a data collection technique. The
questionnaires are distributed directly or via online using a purposive sampling method,
a method to decide the sample through some considerations (Sugiyono, 2014). This
study uses the Likert Scale as a data measurement technique with 1-4 intervals in order
to sort more accurate data without losing a lot of data. The descriptive statistical
analysis, classical assumption test, and multiple linear regression analysis are used to
analyze the data.

4.1. Questionnaires Distribution and Respondents’ Demography
There are 126 questionnaires distributed for this study. 26 of them are not returned
and 100 of them are returned, but 35 of the questionnaires are not suitable for this
study. The total of the questionnaires to be analyzed is 65 questionnaires. There is the
respondent’s demography below:
Tabel 4.1 Respondents’ Demography

Number
No Characteristic %
of People
1 Male 34 52,3%
2 Female 31 47,7%
3 < 25 years old 13 20,0%
4 26 - 35 years old 30 46,2%
5 36 - 55 years old 17 26,2%
6 > 55 years old 5 7,7%
7 High School graduate 14 21,5%
8 Diploma graduate 12 18,5%
9 Bachelor graduate 35 53,8%
10 Postgraduate graduate 4 6,2%
11 Civil Servant 4 6,2%
12 Employee 37 56,9%
13 Entrepreneur 21 32,3%
14 Others 3 4,6%
Table 4.1 shows 34 of the respondents are male and 31 of the respondents are
females. Based on the age characteristic, 13 of the respondents are < 25 years old, 30 of
them are 26-35 years old, 17 of them are 36-55 years old, and 5 of them are 55 years old.
Based on the educational background, 14 of the respondents are high school graduates,
12 of them are diploma graduates, 35 of them are bachelor graduates, and 4 of them are
postgraduate graduates. Based on their occupation, 4 of them are civil servants, 37 of
them are employees, and 3 of them are entrepreneurs.
4.2. Research interval analysis
Table 4.2 Research Interval

No Interval Category
Totally disagree
1 1.00 ≥ X ≤ 1.75
disagree
2 1.75 ≥ X ≤ 2.50
agree
3 2.50 ≥ X ≤ 3.25
Totally agree
4 3.25 ≥ X ≤ 4.00
Table 4.2 shows that the totally disagree category has an interval range
between 1.00 and 1.75, the disagree category has an interval range between 1.75
and 2.50, the agree category has an interval range between 2.50 and 3.25, and the
totally agree category has interval range between 3.25 to 4.00.
4.3. Validity and Reliability Test Result

4.3.1. Validity test
This test is used to determine the accuracy of each indicator. If the result of the
test is valid and reliable, it can proceed to the next stage of analysis. The
questionnaires are valid if the value of rcount > rtable.
Table 4.3.1.1 Variable Y Validity Test Result

No rcount Sig. rtable descripti
on
Y_1 0.257 0.05 0.24 Valid
Y_2 0.563 0.05 0.24 Valid
Y_3 0.608 0.05 0.24 Valid
Y_4 0.491 0.05 0.24 Valid
Y_5 0.471 0.05 0.24 Valid
Y_6 0.639 0.05 0.24 Valid
Y_7 0.605 0.05 0.24 Valid
Y_8 0.210 0.05 0.24 Not Valid

Y_9 0.604 0.05 0.24 Valid
The validity result from the taxpayers’ compliance variable test shows that the
8th statement is not valid because the value of the rcount is smaller compared with
the rtable. It means that the 8th statement is not included in the next analysis.
Otherwise, the other statement is included in the next analysis.
Table 4.3.1.2 Variable X1 Validity Test Result
No Rcount Sig. rtable Description
X1_1 0.440 0.05 0.24 Valid
X1_2 0.519 0.05 0.24 Valid
X1_3 0.437 0.05 0.24 Valid
X1_4 0.472 0.05 0.24 Valid
X1_5 0.490 0.05 0.24 Valid
X1_6 0.665 0.05 0.24 Valid
X1_7 0.392 0.05 0.24 Valid
X1_8 0.510 0.05 0.24 Valid
X1_9 0.587 0.05 0.24 Valid
X1_10 0.557 0.05 0.24 Valid
The variable validity test result table of the tax understanding shows that all of
the variables are valid. It means all the variables are included in the next analysis.
Table 4.3.1.3 Variable X2 Validity Test Result

No rhitung Sig. rtabel Description
X2_1 0.211 0.05 0.24 Not Valid
X2_2 0.360 0.05 0.24 Valid

X2_3 0.484 0.05 0.24 Valid
X2_4 0.452 0.05 0.24 Valid
X2_5 0.542 0.05 0.24 Valid
X2_6 0.409 0.05 0.24 Valid
The result of the test shown on the 4.3.1.3 table reveals that the first statement
is not valid. The rcount value is smaller than the rtable value. It means the first
statement is not included in the next analysis.
Table 4.3.1.4 X3 Variable Validity Test Result

No Rcount Sig. rtable Description
X3_1 0.240 0.05 0.24 Valid
X3_2 0.630 0.05 0.24 Valid
X3_3 0.568 0.05 0.24 Valid
X3_4 0.548 0.05 0.24 Valid
X3_5 0.463 0.05 0.24 Valid
X3_6 0.684 0.05 0.24 Valid
X3_7 0.651 0.05 0.24 Valid
X3_8 0.490 0.05 0.24 Valid
X3_9 0.609 0.05 0.24 Valid
The table 4.3.1.4 shows that all the variables are valid. It means all the
variables are included in the next analysis.
4.3.2 Reliability Test

According to Ghozali (2018) the critical r value is 0.7. In order to get a reliable
result, the value of ralpha > rcritical.
Table 4.3.2 Reliability Test Result

No rcount ralpha rCritical Description
Reliable
1 Taxpayers’ compliance 0.813 0.7
2 Tax understanding 0.831 0.7 Reliable
3 Tax awareness 0.704 0.7 Reliable
4 Tax Relaxation 0.843 0.7 Reliable
The reliability test result shows that the taxpayers’ compliance, tax
understanding, tax awareness, and tax relaxation variables are reliable because
the ralpha > rcritical. It means the variables can be used for the data processing.
4.4. Descriptive Statistics Analysis

Descriptive statistics are useful in depicting data in the form of tables.
Below is the result of descriptive statistical analysis.
Table 4.4 Descriptive Statistics Analysis Result

Min Mean
N Range Max Std.Dev Variance
Statistic Std.Error
Y 65 14 18 32 24.66 0.401 3.237 10.477

X1 65 18 21 39 30.40 0.442 3.561 12.681
X2 65 11 9 20 15.45 0.250 2.016 4.063
X3 65 25 25 36 25.40 0.449 3.622 13.119
There are descriptions of the descriptive statistical analysis as is shown in Table 4.4:
1. The taxpayers’ compliance variable (Y) has 65 data with 18 as the
minimum score. The maximum score is 32. The score difference
between the maximum and the minimum score is 14. The average score
is 24.66, the standard deviation is 3.237 and the variance is 10.477.
2. The tax understanding variable (X1) has 65 data. The minimum score is
21 and the maximum score is 39. The score difference between the
maximum and minimum score is 30. The average score is 30.40, the
standard deviation is 3.561, and the variance is 12.681.
3. The tax awareness variable (X2) has 65 data. The minimum score is 9
and the maximum score is 20. The score difference between the
standard deviation is 2.016, and the variance is 4.063.
4. The tax relaxation variable (X3) has 65 data. The minimum score is 25
and the maximum score is 36. The score difference between the
standard deviation is 3.622 and the variance is 13.119.
4.5. Normality test

This test uses One Sample Kolmogorov Smirnov test method by looking at the
significance probability. The significance probability value can be stated as
normally distributed, if its value is bigger than the alpha value, which is 0.05.
Table 4.5 Normality Test Result

Unstandardized
Residual
N 65
Mean 0.0000000
Normal Parametersa,b
Std. Deviation 2.20552635
Absolute 0.079
Most Extreme Differences Positive 0.062
Negative -0.079
Kolmogorov-Smirnov Z 0.640
Asymp. Sig. (2-tailed) 0.808
a. Test distribution is
Normal.
b. Calculated from data.
In conclusion, the normality test result above shows that the data is normally
distributed because the significance value is bigger than 0.05.
4.6. Multicolinearity Test

The study used a multicollinearity test to test the correlation between the
independent variables in a regression. The existence of the multicollinearity can be
determined by looking at the analysis of the Tolerance value section which is higher
than 0.10 and the Variance Inflation Factor (VIF) value does not exceed 10.
Table 4.6 Multicollinearity Test Result

Collinearity Statistics
Tolerance VIF
(Constant)
Tax Understanding 0.448 2.233
(X1)
Tax awareness (X2) 0.653 1.531
Tax Relaxation (X3) 0.619 1.614
a. Dependent Variable: Taxpayers’ Compliance (Y)
The output above shows the tolerance value of the tax understanding variable is
0.488, the tax understanding is 0.653, and the tax relaxation is 0.619, which is
greater than 0.10. The VIF value of the tax understanding is 2.233, the tax
awareness is 1.531, and the tax relaxation is 1.614 which is smaller than 10. Those
numbers mean there is no multicollinearity in the regression model.
4.7. Heteroscedasticity Test using The Glejser Test

The heteroscedasticity test aims to test the regression model experiencing
different variances from the residuals of one observation to another.
Heteroscedasticity can be seen from the significant variable value. There is no
heteroscedasticity in the regression model if the significance value is bigger than
0.05.
Table 4.7 Heteroscedasicity test result

Model Unstandardized t Sig.
Coefficients
B Std. Error
(Constant) 0.689 1.488 0.463 0.645

Tax understanding (X1) 0.017 0.065 0.267 0.790
1
Tax awareness (X2) 0.017 0.096 0.176 0.861
Tax relaxation (X3) 0.013 0.055 0.247 0.806
The output above proves the significance of the tax understanding, the tax
awareness, and the tax relaxation variables are stated as homoscedastic
because the value is bigger than 0.05.
4.8. Determination Coefficient Test (R2)

The R2 test aims to measure the variance of independent variables to the
dependent variable. The R2 value can be considered as influential if the value is
more than 0.5.
Tabel 4.8 Determination Coefficient Test Result

Model Summary
Mod R R Adjusted Std. Error of

el Square the
R
Estimate
Square
1 0.732a 0.536 0.513 2.259
a. Predictors: (Constant), Tax Relaxation (X3), Tax Awareness
(X2), Tax Understanding (X1)
The output above shows the R2 value is 0.513, which means the independent
variables affect the dependent variable. Otherwise, other variables affect the
taxpayers’ compliance by 48.7%.
4.9. Simultaneous Test (F Test)

The study conducted this test to understand the effect of the independent
variables on the dependent variables. The hypothesis is accepted if the significance
value does not exceed 0.05.
Table 4.9 F Test Result

ANOVAa
Model Sum of df Mean F Sig.
Squares Square
Regressio 359.236 3 119.745 23.463 0.000
b
n
1 Residual 311.318 61 5.104
Total 670.554 64
a. Dependent Variable: Taxpayer compliace (Y)

b. Predictors: (Constant), Tax Relaxation (X3), Tax awareness (X2), Tax Understading
(X1)
The table 4.9 shows its significance value is 0.000, which means it is smaller
than 0.05. It concludes that the independent variables: the tax understanding, the
tax awareness, and the tax relaxation; collectively affect the taxpayers’ compliance.
4.10. Multiple Linear Regression Analysis

Regression that has one dependent variable and has two or more independent
variables is called multiple linear regression. The use of this analysis is to
understand how big the effect of the independent variables has on the dependent
variable.
Table 4.10 Regression Analysis Result

Model Unstandardized Standardize t Sig.
d
Coefficients
Coefficients
B Std. Error Beta
(Constant) 3.769 2.699 1.39 0.168

6
1 X1 0.608 0.118 0.669 5.12 0.000
9
X2 0.219 0.173 0.137 1.26 0.211
5
X -0.038 0.099 -0.043 - 0.701
3 0.385
The following equations are compiled from the table above:
𝑌 = 𝛼 + 𝛽𝑋1 + 𝛽𝑋2 + 𝛽𝑋3 + 𝑒

𝑌 = 3.769 + 0.608 𝑋1 + 0.219 𝑋2 + (−0.038 𝑋3) + 𝑒
a. The constant value of 3.769 states the value of the taxpayer’s

compliance is 3.769 if the independent variables: the tax understanding,
the tax awareness, and the tax relaxation; in this study do not change.
b. The tax understanding variable (X1) regression coefficient is 0.608. This
figure shows tax understanding has a positive influence on the taxpayer
compliance. If the tax understanding increases by 1 point, it can increase
the taxpayers’ compliance by 0.608.
c. The tax awareness (X2) regression coefficient is 0.219. This figure
shows tax awareness has a positive influence on the taxpayers’
compliance. If the tax awareness increases by 1 point, it can increase
the taxpayers’ compliance by 0.219.
d. The tax relaxation (X3) regression coefficient is -0.038. This figure shows
tax relaxation has a negative influence on the taxpayers’ compliance. If
the tax relaxation increases by 1 point, it can decrease the taxpayer’s
compliance by 0.038.
4.11. T test
The t test in multiple linear regression is applied to test whether the
independent variable has a significant effect on the dependent variable. It is known
that ttable value is 1.99.
Table 4.11 T Test Result
Model Unstandardized Standardized t Sig.

Coefficients Coefficients
B Std. Error Beta
(Constant) 3.769 2.699 1.396 0.168
1 X1 0.608 0.118 0.669 5.129 0.000
X2 0.219 0.173 0.137 1.265 0.211
X3 -0.038 0.099 -0.043 -0.385 0.701

a. Dependent Variable: taxpayers’ compliance (Y)
Table 4.10 shows the tax understanding variable affects the taxpayer
compliance because the tcount value is 5.129 > ttable 1.99 and sig 0.000 <0.05.
The variable X2, the tax awareness, tcount value is 1.265 and sig 0.211, which
means the tax awareness does not really affect taxpayer compliance during this
pandemic. The X3 variable, the tax relaxation, does not significantly affect taxpayer
compliance because its significant value is greater than 0.05.
4.12. Discussion
4.12.1. The Effect of The Tax Understanding on The Taxpayers’
Compliance
The results of the tests show the tcount of the tax understanding variable value
is 5.129 and the significant value is 0.000. It means that the tax understanding
variable partially has a positive and significant effect on taxpayer compliance. The
taxpayer’s good understanding of the taxation can increase the taxpayers’
compliance. This study’s findings are in line with the study conducted by Purnaditya
and Rohman (2015), Agustiningsih (2016), Ruky et al. (2018).
4.12.2. The Effect of The Tax Awareness on The Taxpayers’ Compliance

The result of the tests show the tcount of the tax awareness variable value is
1.265 and the significant value is 0.211. It means that the tax awareness variable
partially does not have any influence on the taxpayers’ compliance during Covid-
19 pandemic. It happens because of the economic instability and the taxpayers do
not have any encouragement and motivation to fulfill their obligations.
4.12.3. The effect of The Tax Relaxation on The Taxpayers’ Compliance

The result of the tests shows that the tcount of the tax relaxation variable value
is -0.385 and the significant value is 0.701. It means the tax relaxation variable
partially does not affect the taxpayer compliance. In conclusion, the tax relaxation
does not affect the individual taxpayers, but it is focused on saving business entities.
This study in line with the study conducted by Dewi et al. (2020), Widiiswa et al.
(2021).
5. Conclusion
In accordance with the conducted test results, this study presents some main
conclusions. First, the higher and stronger the level of the taxpayers’ understanding of
the taxation rights and obligations can increase taxpayers’ compliance. The taxpayers
with a good taxation understanding are aware of the administrative sanctions as a
consequence of non-compliance with tax regulations, so that taxpayers who have a
high taxation understanding will be more compliant with the tax regulations as a way
of avoiding tax administrative sanctions.
Second, the taxpayers’ tax awareness represents the taxpayers’ urge to pay and
report their taxes in accordance with the regulations. The test results show that the
taxpayers’ awareness does not show any effect on taxpayers’ compliance. Third, the
test results show that the tax relaxation provided by the government has not been able
to influence and have an impact on the taxpayers’ compliance. Tax relaxation policies
in Indonesia are mostly aimed at the corporate taxpayers, while the tax relaxation for
individual taxpayers is not as massive as the tax relaxation for the corporate taxpayers.
In short, the tax relaxation is not able to increase taxpayers’ compliance.

6.1. Study Implication
The study is conducted for the government or policy makers. The first implication,
the writer can conduct a study and analyze several variables that can be a factor
influencing the taxpayers’ compliance. The variables are the tax understanding, the
tax awareness, and the tax relaxation. The aims are to help the government or the
policy makers to know things that should developed or improved in the future.
The second implication of this study is for tax policies. The government tax policies
and tax relaxation should pay more attention to the taxation aspects of the individual
taxpayers. The third contribution, the government or policy makers can find out the
response of the taxpayers to tax relaxation during the pandemic.
6.2. Limitation
The researchers conducted study from start to finish well. However, this study had
several limitations. This study only examines several variables that can have an impact
on taxpayers’ compliance. The variables are tax understanding, tax awareness and
tax relaxation. This study implements various variables that may affect the taxpayers’
compliance. Another limitation in this study was the limited sample due to the specific
criteria and the distributed questionnaire during the Covid-19 pandemic.
6.3. Suggestions
1. The next researcher is expected to add the corporate taxpayers as the
object of the study.
2. Add others variables to the next study, such as the tax sanctions and
other variables that are not applied in this study.
Bibliography
Agustiningsih, W. (2016). Pengaruh Penerapan E-filing, Tingkat Pemahaman
Perpajakan, dan Kesadaran Wajib Pajak Terhadap Kepatuhan Wajib
Pajak di KPP Pratama Yogyakarta. Jurnal Nominal, Vol. V, No. 2.
Alfina, Z. & Diana, N. (2021). Pengaruh Insentif Perpajakan Akibat Covid-19,
Pemahaman Perpajakan, Kesadaran Wajib Pajak Terhadap Kepatuhan
Wajib Pajak dalam Menyampaikan Surat Pemberitahuan Tahunan (Studi
Pada Wajib Pajak yang Terdaftar di KPP Pratama Malang Utara). E-JRA,
Vol. 10, No. 04., 47-57
Dewi, S., Widyasari & Nataherwin. (2020). Pengaruh Insentif Pajak, Tarif Pajak,
Sanksi Pajak dan Pelayanan Pajak Terhadap Kepatuhan Wajib Pajak
Selama Masa Pandemi Covid-19. Jurnal Ekonomika dan Manajemen. Vol.
9, No. 2. Hlm. 108-124.
Febriani, Kusmuriyanto. 2015. Analisis Faktor-Faktor yang Mempengaruhi
Kepatuhan Wajib Pajak. Accounting Analysis Journal
http://journal.unnes.ac.id/sju/index.php/aaj
Ghozali, I. (2018). Aplikasi Analisis Multivariate dengan Program SPSS (Edisi
Sembilan). Badan Penerbit Universitas Diponegoro.
Mardiasmo. (2016). Perpajakan. Yogyakarta. Andi.
Nurkhin, A., Novanty I., Muhsin & Sumiadji. (2018). The Influence of Tax
Understanding, Tax Awareness and Tax Amnesty toward Taxpayer
Compliance. Jurnal Keuangan dan Perbankan. 22(2), 240-255.
http://jurnal.unmer.ac.id/index.php/jkdp
Purnaditya, R.R., & Rohman, A. (2015). Pengaruh Pemahaman Pajak, Kualitas
Pelayanan dan Sanksi Pajak Terhadap Kepatuhan Wajib Pajak (Studi
Empiris Pada WP OP yang Melakukan Kegiatan Usaha di KPP Pratama
Semarang Candisari). Diponegoro Journal of Accounting. Vol. 4, No. 4, 1-
11. http://ejournal-s1.undip.ac.id/index.php/accounting
Rahayu, Siti Kurnia. 2010. “Perpajakan Indonesia: Konsep dan Aspek Formal”.
Yogyakarta: Graha Ilmu.
Ruky, N. E., Putra, W.E., Mansur, F. (2018). Pengaruh Pemahaman Peraturan
Perpajakan, Kesadaran Wajib Pajak, Akuntabilitas Pelayanan Publik dan
Kewajiban Moral Terhadap Kepatuhan Wajib Pajak (Studi Empiris pada
Kantor Bersama SAMSAT Kota Jambi). Jurnal Riset Akuntansi &
Keuangan. 6 (3), 405-418. http://ejournal.upi.edu/index.php/JRAK
Saputro, R. & Meivira, F. (2020). Pengaruh Tingkat Pendidikan Pemilik, Praktika
Akuntansi dan Persepsi Atas Insentif Pajak Terhadap Kepatuhan Pajak
UMKM. Jurnal EMBA. Vol. 8, No. 4 (1059-1068).
Sugiyono. (2014). Metode Penelitian Kuantitatif, Kualitatif, dan R&D. Bandung.
Alfabeta.
Suriambawa, A. & Setiawan, P.E. (2018). Sosialisasi Perpajakan Memoderasi
Pengaruh Kesadaran Wajib Pajak dan Sanksi Perpajakan Pada
Kepatuhan WPOP. E-Jurnal Akuntansi Universitas Udayana. Vol. 25.3.
DOI: : https://doi.org/10.24843/EJA.2018.v25.i03.p21
Tanilasari, Y., & Gunarso, P. (2017). Pengaruh Kesadaran Wajb Pajak dan
Kualitas Pelayanan Fiskus Terhadap Kepatuhan Wajib Pajak Orang
Pribadi Pada Kantor Pelayanan Pajak Pratama Malang Selatan. Jurnal
Akuntansi Perpajakan, Vol. 3, No. 1.
RECONSTRUCTION CONCEPT OF THE MEANING OF
PERMANENT ESTABLISHMENT PHYSICAL
PRESENCE
I Nyoman Darmayasa*
I Dewa Made Partika
Politeknik Negeri Bali, Badung-Bali, Indonesia. Email: nyomandarmayasa@pnb.ac.id
ABSTRACT
This study aims to reconstruct the concept of the meaning of physical presence in the criteria for
determining foreign individuals and foreign entities as permanent establishments. This research
is conceptual research as a reconstruction process. Reconstruction uses the terminology of
physical presence which is adjusted to the presence of a new post-pandemic order, namely
maintaining distance in certain situations. The terminology of maintaining distance is translated
as the distance between foreign individuals and foreign entities and service users. This study
proposes a reconstruction of the concept of the meaning of physical presence, permanent
establishments subject, and permanent establishments object. The concept of Significant
Economic Presence is relevant to be applied to the fulfilment of three aspects of the criteria,
revenue aspect, digital aspect, and user aspect. Reconstruction of the permanent establishments
subject is by determining digital aspect and user aspect. Reconstruction of permanent
establishments object is by determining digital aspect of income. This study provides a theoretical
contribution to the meaning of Significant Economic Presence which replaces the meaning of the
physical presence of a permanent establishment. Thus, the potential for permanent
establishments taxation is not limited to the potential value added tax but can also be on the
potential income tax.
Keywords: Permanent Establishment, Physical Presence, Reconstruction, Significant Economic Presence
Penelitian ini bertujuan untuk merekonstruksi konsep makna kehadiran fisik pada kriteria
penentuan orang pribadi asing dan badan asing sebagai bentuk usaha tetap (BUT). Penelitian ini
merupakan penelitian konseptual dengan pendekatan rekonstruksi makna kata. Rekonstruksi
menggunakan terminologi kehadiran fisik yang disesuaikan dengan hadirnya tatanan baru pasca
pandemi yaitu menjaga jarak. Terminologi menjaga jarak diterjemahkan dengan jarak antara
orang pribadi asing dan badan asing dengan pengguna jasa. Penelitian ini mengusulkan
rekonstruksi konsep makna kehadiran fisik BUT, subjek BUT, dan objek BUT. Konsep Significant
Economic Presence relevan diterapkan dengan pemenuhan tiga aspek kriteria, aspek
pendapatan, aspek digital, dan aspek pengguna. Rekonstruksi subjek BUT dengan penentuan
aspek digital dan aspek pengguna. Rekonstruksi objek BUT dengan penentuan aspek
pendapatan. Penelitian ini memberikan kontribusi teoritis pada makna significant economic
presence sebagai makna baru kehadiran fisik suatu BUT. Kontribusi praktis pada peningkatan
potensi pemajakan BUT tidak terbatas pada potensi pajak pertambahan nilai, namun bisa pada
potensi pajak penghasilan.
Kata kunci: Bentuk Usaha Tetap, Kehadiran Fisik, Rekonstruksi, Significant Economic Presence
1. INTRODUCTION
1.1. Background
The rapid development of technology propagates in various sectors starting from
the information sector to the economic sector. The digital economy continues to grow
exponentially, especially because of technological advances and the availability of
global internet connectivity. It can be observed that technological developments go
hand in hand with changes in people's consumption behavior (Appel et al., 2020;
Grewal et al., 2020). This has stimulated the growth of businesses that take advantage
of technological developments by building businesses on marketplaces, for example,
Shopee, Spotify, and so on (Akter et al., 2020; Ihsan, 2020; Nurhayati & Wolff, 2021).
Like the swift flow of water, the presence of these businesses is gradually felt to be
very strong in the community because all needs, such as entertainment, shopping or
any kind of transaction can be fulfilled through a marketplace that can be accessed
from digital media with smartphones (Belanche et al., 2020; Sharmin et al., 2021).
In the development of the digital economy, by foreign trading through electronic
systems (PMSE) business actors can receive profits from e-commerce transactions in
the destination country and transfer these profits to the jurisdiction where the foreign
PMSE business actors are domiciled. Even though it seems unfair, this concept is a
legal practice of international taxes contained in an international tax framework known
as anti-erosion of the tax base and profit-shifting or BEPS (Rebecca & Grade, 2021).
Taxation of digital transactions still using the concept of physical presence as a
criterion for allocating tax rights to source countries has an impact on increasing
significant base erosion practices (Suwardi et al., 2020). BEPS is the subject of reform
from the Organization for Economic Co-Operation and Development (OECD). The
debates and problems that take place in the OECD related to this matter are how to
determine the relative distribution of company costs and income to be distributed
between countries of origin and destination countries.
The group of countries that are members of the OECD recommend the application
of a significant economic presence (SEP) to determine the existence of a permanent
establishment (PE) for digital-based multinational companies in economic activity in a
country. Along with technological and business developments that are taking place,
more and more foreign tax subjects are receiving large incomes from e-commerce
transactions that are utilized by consumers from Indonesia. Thus, a tax policy is
needed that regulates this matter clearly and unequivocally in order to optimize tax
revenue for e-commerce transactions by foreign individuals and foreign entities. One
of the efforts of the tax authorities is to broaden the definition of PE through the SEP
concept.
1.2. Research Question

The phenomenon of the development of the digital economy encourages the
urgency of reconstructing the concept of the meaning of the physical presence of PE.
Fulfillment of physical presence criteria becomes irrelevant, because of the increasing
cross-jurisdictional cross-border business models involving foreign individuals and
foreign entities. It is not enough to review the criteria for determining PE with unilateral
measures or only revising domestic regulations (Kurniawan, 2020; Wijaya &
Utamawati, 2018). The enactment of Law No. 2 of 2020 in Article 6 paragraph 6 in the
form of setting provisions that use the criteria for the significant economic presence of
a PE which is still limited to Trade Operators Through Electronic Systems (PPMSE).
Thus, it is necessary to review the criteria for determining PE at the global consensus
level. In Hikmah (2020), using the regulatory impact analysis method in determining a
PE, his proposal cannot only change one article but several regulations at once.
Countries connected to world trade are bound by multilateral and/or bilateral
agreements, so it is deemed necessary to review the criteria for determining PE in tax
treaties.
The terminology of physical presence as a criterion for determining the subject of
a PE needs to be reconstructed, so that the scope can be wider according to the
context of space and time. The spatial context is related to the space for cross-
jurisdictional business growth in response to state policies in dealing with a pandemic.
The context of time is related to the current reality as the momentum for post-pandemic
economic recovery. The research question is how to reconstruct the concept of the
meaning of the physical presence of PE. The purpose of this study is to reconstruct
the criteria for determining PE from the terminology of physical presence,
reconstruction of PE subjects, and reconstruction of PE objects. Achievement of
research objectives are through the formulation of a reconstruction concept that uses
the SEP concept from the OECD by considering the hierarchy of legal principles in
Indonesia. In the following description, the researcher presents a review of the
literature that supports the concept of reconstruction.
2.1. Significant Economic Presence
Each country that has proposed and implemented the SEP concept has a different
format and form of implementation; therefore, there is no official international definition
that explains the meaning of SEP or significant economic presence. Referring to the
OECD consultation documents, SEP is an approach that states the presence of a tax
in a country's territory when a foreign company earns significant income under certain
conditions (OECD, 2019b). The certain conditions in question are conditions that can
ensure that there is planned or continuous interaction between a company and a
region of the country through the digital economy or other means. According to Breza
(2018), SEP emphasizes that a permanent establishment or physical office is not the
main requirement for taxing foreign companies. If the e-commerce company fulfills
certain conditions and earns profits from the country concerned, it is considered to
have fulfilled the physical presence and the tax authority of that country has the right
to tax the company.
When a jurisdiction wants to apply the SEP concept according to the task force on
digital economy, there are three factors that must be observed. These are certain
factors that can be used as a basis for testing SEP (OECD, 2015). The second factor
is the digital factor which includes local domain names, local digital platforms, and
local payment options. The third factor is the user-based factor. This factor is based
on the number of users such as the number of final online contracts, the level of
monthly active users, and the volume of digital content acquired through digital
platforms.
The SEP provisions are considered fulfilled if the foreign tax subject has fulfilled
the requirements for interacting through digital technology in that country. This is
regulated in the publication addressing the tax challenges of the digitalization of the
economy in 2019. These requirements include the volume of digital content, user base,
website maintenance in local languages, responsibility for final delivery of goods by
the provider company, payment in the form of local currency, and ongoing sales
promotion or marketing activities (OECD, 2019a). Adopting the SEP concept from the
OECD, the researchers paired with the terminology of keeping a distance in the
description of physical distancing as a new order behavior.
2.2. Physical Distancing as a New Order Behavior

The term keeping distance or physical distancing has adorned our lives for the last
two years. Unwittingly, the social distancing policy has become a new way of life for
society and even worldwide. The real manifestation of keeping a distance is in the
form of restrictions in the form of travel bans implemented worldwide which affect more
than 90% of the world's population (Gössling et al., 2020). Initially in the form of a
country's policy to respond to a pandemic, it eventually became a new order. The
order of keeping distance becomes a good habit in situations of protecting yourself
from the surrounding environment. In terms of terminology, referring to the Big
Indonesian Dictionary, the term keeping your distance comes from two basic words,
namely keeping and distance.
The word guard means to be safe and or not to be disturbed. The word distance
means the space between and indicates the length or distance between two objects
or places. These two words, when strung together, become keeping a distance, which
means space between objects so that there are no distractions. If this word is
associated with physical presence as a criterion for determining foreign individuals
and foreign entities, it becomes relevant not from the semantic meaning of the word,
but from the activities that arise from the word. Changes in the operating procedures
of companies that adapt to the context of space and time are closely related to
government policies that are the new order of today's society.
The new order that was born from government policies targeted various
community activities and changed people's consumption patterns. In the context of
space, there is a new space or the growth of new businesses that optimize
technological developments in response to the new order of social distancing. In the
context of time, there is no time limit that binds business activities to grow in
accordance with the development of the social structure. However, the taxation policy
on transactions that grow in the context of space and time cannot be adjusted
automatically. The concept of taxation still adheres to the legal hierarchy (Krisnapati,
2019; Rinaningsih, 2020), especially regarding transactions across jurisdictional
borders involving foreign individuals and foreign entities. The following description is
related to the hierarchy and principles of tax legislation in Indonesia.
2.3. Lex Specialist derogate Lex Generalist

The principle of lex specialist derogate lex generalist is one of the basic postulates
in legal science which literally means special law sets aside general law (Hiariej, 2021).
It is recorded in the history of classical law that this principle was the brainchild of
Aemilius Papinianus, a legal expert known for his constructive and critical thinking in
the formation and development of law during the Roman Empire. Specificity takes
precedence over general matters because it cannot be denied that special things are
more important. The rationality of prioritizing special laws is carried out because
special laws are made to regulate more specific needs that have not been regulated
by general law, and special laws are also made with the intention of completing or
even making improvements or corrections to general law. This statement is in line with
the thoughts of Jeremy Betham, a legal philosopher, who stated that special law is
considered a correction to general law because it is prepared with deeper and more
precise views or thoughts on the subject of the law (Irfani, 2020).
Applying the lex specialist principle is not easy to do, bearing in mind that there
are no absolute rules governing a special law against another general law. In his book
entitled "Positive Indonesian Law", Prof. Bagir Manan suggests there are several
things that can be used as guidelines in implementing lex specialist derogate lex
generalist. The guidelines that can be used include other provisions stipulated in these
general provisions which remain in force except for provisions already regulated in
special provisions; special provisions must be equal to general rules (for example laws
and laws), and must be in the same legal environment as general provisions (Manan,
2004).
The problem of conflicting special law and common law is also an interesting study
in international law. International law is dynamic in line with current developments of
globalization which have caused relations between countries to become more intense.
International law is law between countries (Hasim, 2019). The complexity of
international interactions that occur has caused the understanding of international law
to also become broader to regulate the structure and activities or behavior of an entity
or company on an international scale (Arista, 2020). International law is complicated
because it involves relations between countries which have their own legal sovereignty
and are of course oriented toward the national interests of each country. In contrast to
national law which is structured in a hierarchical legal system, international law is more
vulnerable to issues of conflicting norms because it is only manifested through
international agreements. Therefore, the principle of lex specialist derogate lex
generalist is a way that can be applied in overcoming the problem of conflicting norms.
The principle of lex specialist derogate lex generalist is considered as a concept to
reconstruct the meaning of physical presence in determining PE.
3. RECONSTRUCTION METHODOLOGY
3.1. Research Paradigm
Paradigm is a researcher's perspective on a reality (Kamayanti, 2016; Triyuwono,
2013); in this study, it is a researcher's perspective on a reality that needs to be
reconstructed. Conceptual research consists of theory synthesis, theory adoption, and
proposed perspectives (Jaakkola, 2020). The proposed perspective bridges the
theoretical framework with propositional proposals (Cornelissen, 2016). The reality in
this study is the reality of determining PE using physical presence criteria. This
research is conceptual research with a qualitative approach. Qualitative research uses
the main data in the form of qualitative data in the form of text and the aim is not to be
generalized (Darmayasa & Aneswari, 2015). The reality of the text is in the form of
laws and regulations and their derivatives related to the determination of PE criteria.
3.2. Reconstruction Stages
The meaning of the term physical presence in determining foreign individuals and
foreign entities meeting the criteria for a permanent establishment is irrelevant if it is
related to business development during and after the pandemic. The growth of various
cross-jurisdictional businesses involving foreign individual and foreign entity tax
subjects in accordance with the spatial context without distance and time context
during and after the pandemic has become a reconstruction concept to produce new
meanings. The principle of lex specialist derogate lex generalist considered as a
principle for ratifying tax treaties is a special provision and the Income Tax Law is a
general provision. The special provisions of the tax treaty agreement are bilateral in
nature between the government and the treaty partner countries.
Regarding consideration of legal aspects in the preparation of tax treaties, one of
which is the provisions issued by the OECD. The SEP concept is currently being
discussed by the OECD. The SEP concept does not require physical presence, which
is the main requirement for imposing taxes involving foreign individuals and foreign
entities (Breza, 2018). The three factors contained in the SEP concept are income-
based factors, digital factors, and user-based factors (OECD, 2015). The three factors
in the SEP concept are used as part of the reconstruction of the meaning of the
physical presence of PE. Referring to the SEP concept, this considers the income
aspect, digital aspect, and the user aspect associated with the terminology of keeping
distance as a new order and considering the lex specialist derogate lex generalist
principle as the basis for preparing the concept of reconstruction, which is presented
in Chart 1.
Significant Economic Presence (SEP)

Foreign Lex Specialist
Individuals Derogate
Lex
Tax
Generalist
Treaty Income-Based Digital User-Based
Factors Factors Factors
Foreign PE Physical
Entities Presence
PE Tax Subject
Reconstruction
PE Tax Object
Chart 1. Stages of Reconstruction of the Meaning of

PE's Physical Presence
Source: (Researcher, 2022)
Based on Chart 1 regarding the stages of reconstruction of the meaning of the

physical presence of PE, there are three stages of reconstruction. The first stage is
the reconstruction of the concept of the meaning of the physical presence of PE. This
reconstruction emphasizes the determination of foreign individuals and foreign entities
using the SEP concept from the OECD. The second stage is the reconstruction of the
subject of PE which emphasizes the subject of foreign individuals and foreign entities
as PE in the digital and user aspects. The third stage, the reconstruction of permanent
establishments emphasizes the income earned by foreign individuals and foreign
entities as permanent establishments. In the following description, the researcher
presents the reconstruction process which begins with the reconstruction of the
concept of the meaning of the physical presence of PE.
4. RECONSTRUCTION
4.1. Reconstruction of the Concept of the Meaning of Physical Presence PE
The first stage of reconstruction begins with the reconstruction of the concept of
the meaning of the physical presence of PE. Provisions for the physical presence of
PE are contained in the physical determination of PE in the Income Tax Law. The
determination that arises from the context of taxation when viewed from the challenges
in terms of determining the tax subject and tax object, where things that allow the PE
itself to be biased and insignificant are related to the aspects of place of business,
location, and permanent, thereby this causes challenges to determine tax rights for
source country (Wijaya & Utamawati, 2018). Furthermore, Article 6 paragraph (6) of
Law No. 2 of 2020 states that foreign traders, foreign service providers, and/or foreign
PPMSE can be treated as PE and subject to income tax if they meet the provisions of
significant economic presence (Law No. 2 Year, 2020). The taxes authority stipulates
three provisions for significant economic presence, namely, the consolidated gross
turnover of a business group up to a certain amount; sales in Indonesia up to a certain
amount; and/or active users of digital media in Indonesia up to a certain number.
Article 6 paragraph 6 of Law No. 2 of 2020 has been able to capture digital transactions
in Indonesia. According to Kurniawan (2020), the implementation of Law No. 2 of 2020
are unilateral measures not yet on a global consensus. Ratification is required at the
tax treaty level through the reconstruction of the meaning of the physical presence of
PE.
At the reconstruction stage, the concept of the meaning of the physical presence
of PE is illustrated by an Indonesian Tax Treaty with Singapore. The ratification of
Indonesia's Tax Treaty with Singapore was carried out on 4 February, 2020, while it
became effective on 23 July, 2021. There are several new agreements contained in
the revised tax treaty. In terms of PE, there has been no new agreement regarding
the determination of PE using the SEP concept from OCED. The new agreement is
limited to changes in dividend, interest, royalty and branch profit tax rates, other
agreements are more about rearranging the exchange of information and anti-tax
evasion (DJP, 2020b; Republik Indonesia, 2020). This means, there is no change in
Article 5 paragraph 1, so that PE means a permanent place of business where all or
part of a company's business is carried out. The emphasis in paragraph 1 is on the
word place (Republik Indonesia, 2020). It is necessary to expand the meaning of the
word place in accordance with the development of businesses involving foreign
individuals and foreign entities in accordance with the current context of space and
time. The next reconstruction is in the form of reconstruction of the concept of PE tax
subjects.
4.2. Reconstruction of PE Tax Subject Concept
The tax authorities are taking advantage of the momentum of accelerating
business growth that utilizes the digital economy as a source of tax revenue. This is
done by the tax authorities to maintain financial system stability. Regarding taxation of
Google Asia Pacific Pte. Ltd. whose place of business is in Singapore, in Press
Release No. SP-29 PMSE batch 1, Google Asia Pacific Pte. Ltd. has been appointed
as PMSE VAT collector as of 1 August, 2020 (DJP, 2020a). Tax is imposed on Google
because it is deemed to have complied with existing regulations both in terms of
transaction value and/or the amount of traffic (access) within 12 months. The system
is that Google will charge 11% VAT (previously 10%) to the Google accounts of
customers who buy digital products or services in Indonesia (Google Workspace
edition). Buyers can provide Google with a taxpayer identification number (TIN) for
VAT purposes to be printed on the buyer's invoice.
Although Google Asia Pacific Pte. Ltd is indicated to have an SEP presence it
cannot be subject to income tax because there are no rules on SEP criteria
(Darmayasa & Kumontoy, 2022). This is because there are two pillars of discussion of
digital economy taxation which are agreed upon in the inclusive framework. One thing
that is easily visible in pillar one is the right to taxation itself. Taxation of electronic
transactions does not meet the principles of fairness and equity because they are
considered to cause double taxation problems. Unilateral measures used by various
countries in collecting state revenues also have implications for double taxation, the
potential for a trade war that will affect the collection of the tax itself.
The high rate of electronic transactions encourages the government to tax digital
transactions. This is because PMSE taxation is one of the efforts to increase the
source of state tax revenue. However, PMSE taxation can be carried out when the
government is ready for the possibility of changing transactions and of course
maintaining the double tax avoidance agreement or tax treaty. In the end, this issue is
something that needs to be negotiated by all the countries involved to determine the
international SEP. In the following description, the researcher presents the
reconstruction of the concept of PE tax objects.
4.3. Reconstruction of PE Tax Object Concept

Taxation of PE is regulated by PMK-35/PMK.03/2019 concerning the
determination of PE (Kementerian Keuangan RI, 2019). PE has the same tax
obligations as corporate tax subjects in Indonesia which have been regulated in Article
2 paragraph (1a) (DGT, 2021). Based on Article 2 paragraph (5) of the Income Tax
Law, PE is a form of business that is used by individuals who do not reside in Indonesia,
individuals who are in Indonesia for no more than 183 days within a 12 month period,
and entities that are not established and are not domiciled in Indonesia to run a
business or carry out activities in Indonesia, for example, building branch companies,
factories, workshops, etc., which of course have facilities in the form of land and
buildings and include equipment, machines, and warehouses owned or used by
electronic transaction organizers to carry out business activities through the
marketplace and are permanent.
Basically, economic activity in the form of trading in goods or services carried out
online via the internet, or what is often called the digital economy, is a phenomenon
that changes almost all aspects of life, even though, from a political point of view, each
country still has territorial boundaries, but digital economic activity can be done
anywhere and anytime. This is considered effective, considering that users do not
need to spend a lot of time and effort in obtaining the desired goods or services.
Producers, consumers and distributors do not need to be at one time and one place
to carry out these buying and selling transactions. Therefore, it can be said that the
flow of money in and out of a country is heavy. In fact, the outflow of money was very
heavy to countries that ruled the digital economy, at a time when many countries were
positioned as consumer countries. This problem certainly creates difficulties for the
state in determining the imposition of income tax, because, to apply it, a business or
activity must be determined as a PE first, whereas in determining to become a PE it is
necessary if it meets the criteria, in the form of physical presence, except for things
currently agreed upon by the international community, such as assembly, installation
and construction projects, the provision of services for more than sixty days out of
twelve months, and the existence of an agent with a non-free position.
Governments in each country are currently fighting for a company or digital
economic activity to be designated as a permanent establishment and carry out its tax
obligations in consumer countries. Currently, the issue of digital economy taxes is
being intensively discussed, bearing in mind that Indonesia is one of the countries in
Southeast Asia with the fastest and very rapid growth in the digital economy, so that
the government can attract the attention of entrepreneurs or digital economy rulers in
seeking profits. In the emergence of this fact, the government has issued several
regulations regarding tax regulations on digital economic activities with the aim of
obtaining benefits from tax revenues. One of them is the plan to use the SEP concept
with a specific form of implementation and format, although on the other hand the
government feels the need to wait for a decision and agreement from the OECD.
Each country that has implemented this concept has a different format and form
of implementation, due to the absence of an official international definition. The
government stipulates the provisions of the SEP concept through several specified
criteria, including based on the consolidated turnover or income of business groups,
the number of sales in Indonesia to the number of active users in digital media
(internet). If the determination of PE cannot be imposed on foreign tax subjects due to
a double tax avoidance agreement, then foreign tax subjects who meet the SEP
criteria will be subject to electronic transaction tax. With this regulation, it is hoped that
foreign companies that do not have a physical presence in Indonesia but are included
in the SEP category, such as Google, Amazon, etc., are taxed. In the midst of
preparations for tax imposition for trade organizers through foreign electronic systems
that meet SEP requirements, an issue arose to postpone the implementation of digital
taxes, on the grounds that PE was deemed irrelevant considering that each country
has different format criteria with different implementation; apart from that Indonesia
recently established Electronic Transaction Tax (PTE) in Law No. 2 of 2020, which
can refer to domestic laws made, so that they are not included in the tax treaty and
result in value added tax and income tax that cannot be imposed in that country.
4.4. Proposed SEP Concept as Criteria for Determining
In this section, the researcher presents a series of three stages of reconstruction,
starting from the reconstruction of the concept of the meaning of physical presence
using the new meaning of SEP, then the reconstruction of the subject of PE and the
reconstruction of the object of PE. The SEP concept proposal as a criterion for
determining PE contains three aspects. The aspect of physical presence uses a new
meaning with the criteria for determining whether there is income flowing from the user
country. This meaning means that Article 5 paragraph 1 of the tax treaty according to
the illustration between Indonesia and Singapore needs to be reviewed by adding an
explanation that physical presence is not the main requirement which is interpreted by
the existence of a significant economic presence in the form of a stream of income.
The reconstruction of the PE subject refers to the SEP concept from the OECD on
digital administration aspects. Digital aspects in the form of local domain names, local
digital platforms, and local payment options are aspects of determining the subject of
PE. Reconstruction of PE objects is in accordance with the SEP concept from the
OECD on the income aspect. The existence of revenue streams from user countries
is the criterion for determining PE objects.
Reconsideration of the criteria for determining PE in Law No. 2 of 2020 needs to
be expanded in scope not only limited to PPMSE. Improvements apart from unilateral
limited global consensus level are needed. It can be started by reviewing the tax treaty
with agreement partner countries that have potential taxes on digital transactions. The
review is an effort to create legal certainty, considering that legal certainty is one of
the factors considered by investors and factors that can encourage tax compliance
(Darmayasa, Arsana, & Putrayasa, 2022; Darmayasaet al., 2022; Keen & Slemrod,
2017; Pickhardt & Prinz, 2014; Pui Yee et al., 2017; Slemrod, 2019). The proposed
reconstruction of the meaning of PE's physical presence using the new meaning of
SEP from the OECD is presented in Chart 2.
Significant Economic
Foreign Presence (SEP)
T
Individuals A
X
Law Digital - Tax Subject
No. 2 Reconstruction
T
2020 R
Foreign E
Entities A
T Income - Tax Object
Y
Unilateral Measures Lex Specialist Derogate Lex Generalist
Chart 2. Reconstruction of the Meaning of PE's Physical Presence

Source: (Researcher, 2022)
5. CONCLUSIONS AND RECOMMENDATION
Referring to the research objective, which is to reconstruct the concept of the
meaning of the physical presence of PE, this research leads to a conclusion of a new
meaning resulting from the reconstruction. The development of the digital economy,
especially with the impact of Covid-19, forced the acceleration of the use of technology
by reducing physical presence to become the new culture of society. Of course, this
new culture is not yet relevant to the meaning of the physical presence of PE as stated
in the provisions of tax regulations, especially in tax treaties.
Along with the accelerated development of digital technology, it leads to the
development of digital economic transactions to become a new tax base. The basis of
taxation is not limited to the physical existence of an entity to meet the criteria for a
PE, but with the proposal of SEP it becomes relevant. This proposal is important in
the effort to foster the principle of fair and equitable tax collection from the perspective
of an entity's source of income.
This research contributes theoretically to the concept of PE taxation. The
reconstruction of the meaning of PE's physical presence after the two-year pandemic
has brought about a new order of activities. Maintaining physical distance becomes a
choice and a habit when you feel unsafe and comfortable. This concept should also
be relevant to the reconstruction of the meaning of the physical presence of PE
taxation. The proposed concept of this new meaning is relevant to the current
development of digital economic transactions. The concept of taxation with a new
meaning of physical presence is defined as the presence of economic activities
influencing various PE taxation policies.
The results of the reconstruction of the concept of the meaning of PE's physical
presence reinforce the view that the SEP concept can be an alternative for determining
PE criteria. The concept of SEP is relevant in the context of the growing space for
digital-based businesses involving foreign individuals and foreign entities as a
response to the new order of post-pandemic physical distancing behavior. It is hoped
that the tax authorities will continue to push for tax treaty ratification with treaty partner
countries according to priority in countries with high income tax potential. This
research is still only conceptual, further research is needed that is more specific to
certain subjects, objects, and partner countries. The researcher presents this idea, as
an effort to strengthen the view and increase self-confidence that Pajak Kuat
Indonesia Hebat.
6. REFERENCES
Akter, S., Motamarri, S., Hani, U., Shams, R., Fernando, M., Mohiuddin Babu, M., &
Ning Shen, K. (2020). Building dynamic service analytics capabilities for the
digital marketplace. Journal of Business Research, 118(June), 177–188.
https://doi.org/10.1016/j.jbusres.2020.06.016
Appel, G., Grewal, L., Hadi, R., & Stephen, A. T. (2020). The future of social media in
marketing. Journal of the Academy of Marketing Science, 48(1), 79–95.
https://doi.org/10.1007/s11747-019-00695-1
Belanche, D., Flavián, M., & Pérez-Rueda, A. (2020). Mobile Apps Use and WOM in
the Food Delivery Sector: The Role of Planned Behavior, Perceived Security and
Customer Lifestyle Compatibility. Sustainability, 12(10), 1–21.
www.mdpi.com/journal/sustainability
Breza, M. (2018). What is Economic Presence Nexus and How Does it Affect You?
http://www.otcpas.com/advisor-blog/what-economic-presence-nexus/
Cornelissen, J. (2016). Developing Propositions, A Process Model or A Typology?
Addressing The Challenges of Writing Theory Without a Boilerplate. Academy of
Management Review, 42(1).
Darmayasa, I. N., & Aneswari, Y. R. (2015). Paradigma Interpretif Pada Penelitian
Akuntansi Indonesia. Jurnal Akuntansi Multiparadigma, 6(5), 350–361.
Darmayasa, I. N., Arsana, I. M. M., & Putrayasa, I. M. A. (2022). Reconstruction of
The Slippery Slope Framework Tax Compliance Model. ACRN Journal of
Finance and Risk Perspectives, 11, 19–32.
https://doi.org/https://doi.org/10.35944/jofrp.2022.11.1.002
Darmayasa, I. N., Arsana, I. M. M., Putrayasa, I. M. A., & Larasati, P. I. (2022).
Reconstruction of Compliance Risk Management Towards a Humanist Approach.
International Conference on Applied Science and Technology on Social Science
2021 (ICAST-SS 2021), 647, 492–498.
https://doi.org/https://doi.org/10.2991/assehr.k.220301.081
Darmayasa, I. N., & Kumontoy, M. (2022). Optimizing Income Tax and Value Added
Tax on E-Commerce Transaction. Proceedings of the First Lekantara Annual
Conference on Public Administration, Literature, Social Sciences, Humanities,
and Education, LePALISSHE 2021, August 3, 2021.
https://doi.org/10.4108/eai.3-8-2021.2315147
DGT. (2021). Susunan dalam Satu Naskah Undang-Undang Perpajakan Edisi 2021
dengan Perubahan UU CIPTAKER. Direktorat P2Humas DJP.
DJP. (2020a). Gelombang Pertama, Enam Perusahaan Ditunjuk Sebagai Pemungut
PPN Produk Digital Luar Negeri. In Siaran Pers Nomor: SP-29/2020, Jakarta, 7
Juli 2020. https://pajak.go.id/sites/default/files/2020-07/SP 29 Daftar PMSE
Gelombang 1.pdf
DJP. (2020b). Siaran Pers SP-03/2020 tentang Indonesia Dan Singapura Perbarui
Perjanjian Pajak (p. SP-03/2020).
Gössling, S., Scott, D., & Hall, C. M. (2020). Pandemics, tourism and global change:
a rapid assessment of COVID-19. Journal of Sustainable Tourism, 7(3), 1–20.
https://doi.org/10.1080/09669582.2020.1758708
Grewal, D., Hulland, J., Kopalle, P. K., & Karahanna, E. (2020). The future of
technology and marketing: a multidisciplinary perspective. Journal of the
Academy of Marketing Science, 48(1), 1–8. https://doi.org/10.1007/s11747-019-
00711-4
Hasim, H. (2019). Hubungan Hukum Internasional Dan Hukum Nasional Perspektif
Teori Monisme Dan Teori Dualisme. Mazahibuna Jurnal Perbandingan Mazhab,
1(2), 166–179.
Hiariej, E. O. S. (2021). Asas Lex Specialis Systematis dan Hukum Pidana Pajak.
Jurnal Penelitian Hukum De Jure, 21(1), 1.
https://doi.org/10.30641/dejure.2021.v21.1-12
Hikmah, M. (2020). Penerapan Regulatory Impact Analysis Dalam Penetapan BUT Di
Indonesia Di Era Ekonomi Digital. Simposium Nasional Keuangan Negara 2020,
699–718.
Ihsan, D. M. (2020). Similar. Trenasia; TrenAsia. https://www.trenasia.com/melihat-
10-e-commerce-paling-hits-di-indonesia/
Irfani, N. (2020). Asas Lex Superior, Lex Specialis, dan Lex Posterior: Pemaknaan,
Problematika, dan Penggunaannya dalam Penalaran dan Argumentasi Hukum.
Legislasi Indonesia, 16, No. 3, 305–325.
Jaakkola, E. (2020). Designing conceptual articles: four approaches. AMS Review,
10(1–2), 18–26. https://doi.org/10.1007/s13162-020-00161-0
Kamayanti, A. (2016). Metodologi Penelitian Kualitatif Akuntansi: Pengantar
Religiositas Keilmuan. In Yayasan Rumah Peneleh.
Keen, M., & Slemrod, J. (2017). Optimal tax administration. Journal of Public
Economics, 152, 133–142. https://doi.org/10.1016/j.jpubeco.2017.04.006
Kementrian Keuangan RI. (2019). PMK No. 35/PMK.03/2019 Tentang Penentuan
Bentuk Usaha Tetap. Kementrian Keuangan RI.
Krisnapati, B. (2019). Direktorat Jenderal Pajak Sebagai Penegak Hukum
Transformatif Di Indonesia. Scientax, 1(1), 41–57.
https://doi.org/10.52869/st.v1i1.7
Kurniawan, A. M. (2020). Pemajakan Perdagangan Melalui Sistem Elektronik (PMSE)
Lintas Negara (Unilateral Measures Vs Global Consesnsus). Simposium
Nasional Keuangan Negara, 315–334.
Manan, B. (2004). Hukum positif Indonesia: satu kajian teoritik (Cetakan 1). UII Press.
Miftakhul Nur Arista, A. F. F. (2020). Hubungan Hukum Internasional Dan Hukum
Nasional. Ma’Mal: Jurnal Laboratorium Syari’Ah Dan Hukum, 1(4), 365–376.
Nurhayati, H., & Wolff. (2021). E-commerce in Indonesia - Statistics & Facts. Statista.
OECD. (2015). Addressing the Tax Challenges of the Digital Economy, Action 1 2015
Final Report. OECD Publishing. https://www.oecd-
ilibrary.org/docserver/9789264241046-
en.pdf?expires=1660437229&id=id&accname=guest&checksum=13EFE4F4E5
67B010FDF6C46ED98D4C44
OECD. (2019a). Addressing The Tax Challenges of The Digitalisation ff The Economy.
In Base Erosion and Profit Shifting Project Public.
https://www.oecd.org/tax/beps/public-consultation-document-addressing-the-
tax-challenges-of-the-digitalisation-of-the-economy.pdf
OECD. (2019b). Development Co‑operation Report 2019: A Fairer, Greener, Safer
Tomorrow. OECD Publishing. https://doi.org/10.1787/9a58c83f-en
Pickhardt, M., & Prinz, A. (2014). Behavioral dynamics of tax evasion - A Survey.
Journal of Economic Psychology, 40, 1–19.
Pui Yee, C., Moorthy, K., & Choo Keng Soon, W. (2017). Taxpayers’ perceptions on
tax evasion behaviour: an empirical study in Malaysia. International Journal of
Law and Management. https://doi.org/10.1108/IJLMA-02-2016-0022
Rebecca, & Grade, A. (2021). Digital Taxation in Indonesia Center for Indonesian
Policy Studies. 1–12.
Republik Indonesia. (2020). Tax Treaty Indonesia - Singapura.
Rinaningsih. (2020). Asas “Dapat Dilaksanakan” Dalam Peraturan Perpajakan.
Scientax, 1(2), 184–202. https://doi.org/10.52869/st.v1i2.30
Sharmin, F., Tipu Sultan, M., Badulescu, D., Badulescu, A., Borma, A., & Li, B. (2021).
Sustainable Destination Marketing Ecosystem through Smartphone-Based
Social Media: The Consumers’ Acceptance Perspective. Sustainability
(Switzerland), 13(4), 1–24. https://doi.org/10.3390/su13042308
Slemrod, J. (2019). Tax compliance and enforcement. Journal of Economic Literature,
57(4), 904–954. https://doi.org/10.1257/jel.20181437
Suwardi, Budiandri, A., S., C., & A., G. N. (2020). Memajaki Transaksi Ekonomi Digital:
Studi Kasus Di India, Perancis, Dan Australia. Jurnal Pajak Dan Keuangan
Negara (PKN), 2(1), 1–12. https://doi.org/10.31092/jpkn.v2i1.971
Triyuwono, I. (2013). [Makrifat] Metode Penelitian Kualitatif [Dan Kuantitatif] Untuk
Pengembangan Disiplin Akuntansi. Makalah Ini Diseminarkan Dalam Acara
Simposium Nasional Akuntansi Ke-16 Di Manado Pada Tanggal 25-27
September 2013.
UU No. 2 Tahun 2020. (2020). Penetapan Peraturan Pemerintah Pengganti Undang-
Undang Nomor 1 Tahun 2020 Tentang Kebijakan Keuangan Negara dan
Stabilitas Sistem Keuangan Untuk Penanganan Pandemi Corona Virus Disease
2019 (COVID-19) dan/atau Dalam Rangka Menghadapi Ancaman yang (p. 53).
Wijaya, S., & Utamawati, H. (2018). Pajak Penghasilan dari Ekonomi Digital atas
Cross-Boarder Transaction. Jurnal Online Insan Akuntan, 3(2), 135–148.
THE TAXATION ASPECT REVIEW OF CASH POOLING
BASED ON INDONESIAN REGULATIONS
Benny Oktis Yanurwendaa, Rindah Febriana Suryawatib

a Direktorat Jenderal Pajak, Jakarta, Indonesia Email: yanurwenda@gmail.com
bDepartement of Business, Faculty of Vocational Studies, Universitas Airlangga Email:
rindah.febriana.s@vokasi.unair.ac.id
ABSTRACT
Review of Taxation Aspect of Cash Pooling Based on Indonesian Regulations. Cash pooling is
formed by companies that are in one business group. Therefore, most of the transactions are
affiliated transactions that must meet the Arm’s Length Principles. The study aimed to review the
implementation of the Arm’s Length Principles in cash pooling. This research utilized a qualitative
approach by reviewing cash pooling transactions with regulations in Indonesia and best practices in
transfer pricing. Based on the role of the leader, affiliated transactions in cash pooling are: payment
of loan interest and payment of interest on deposits in scheme of leader as an in-house bank and
payment of fees in scheme of leader as a service provider.The transactions then should be
examined about their compliance to the Arm’s Length Principles, including evaluating of price
indicators and selecting tested party of each transaction.
Keywords: affiliated transaction, Arm’s Length Principles, cash pooling, Income Taxes
Tinjauan Aspek Perpajakan atas Cash Pooling Berdasarkan Regulasi di Indonesia. Cash pooling
dibentuk oleh perusahaan-perusahaan yang berada dalam satu grup usaha. Oleh karena itu,
sebagian besar transaksinya merupakan transaksi yang memiliki hubungan istimewa yang harus
memenuhi Prinsip Kewajaran dan Kelaziman Usaha. Penelitian ini bertujuan untuk melakukan
tinjauan penerapan Prinsip Kewajaran dan Kelaziman Usaha pada cash pooling. Penelitian ini
menggunakan pendekatan kualitatif dengan melakukan penelaahan transaksi cash pooling dengan
regulasi di Indonesia dan best practice dalam transfer pricing. Berdasarkan peran leader, transaksi
terafiliasi pada cash pooling adalah: pembayaran bunga pinjaman dan pembayaran bunga
simpanan pada skema leader sebagai in-house bank serta pembayaran fee pada skema leader
sebagai penyedia jasa. Transaksi-transaksi tersebut kemudian dilakukan pengujian penerapan
PKKU di antaranya dengan meneliti indikator harga dan tested party atas masing-masing transaksi.
Kata kunci: cash pooling, Pajak Penghasilan, Prinsip Kewajaran dan Kelaziman Usaha, transaksi terafiliasi
1. INTRODUCTION
1.1. Background
Each company will strive to maximize the utilization of its assets in order to provide
better results. This can be accomplished through a variety of approaches in
management planning. For example, if a company owns vacant land that is not used
in its business operations, the company can rent it out to generate regular rental
income without spending too much money. The company can also invite other
companies that specialize in specific industries to work on this vacant land, and the
land owner will receive a profit share.
Cash is a critical asset for any business that must be properly managed. Because
modern financial management believes that "cash is king," the availability of cash is
crucial for the business. Cash must be available in sufficient quantities for the
company's operations, but it should not be idle in excessive amounts, as this reduces
the company's potential income.
Good cash management will become increasingly important for a business group
with multiple companies. Each company will have different cash requirements. Some
businesses may have excess cash, whereas others may require funding for operations
or business expansion. When confronted with this situation, a business group typically
develops a strategy to ensure that the overall use of its cash is properly managed to
provide maximum benefits.
A cash pooling scheme is one strategy that a business group can implement. Cash
pooling is a centralized cash management strategy for all subsidiaries that allows the
company to optimize the debit and credit positions of cash accounts to maximize the
value of these positions. It is embodied as bank service that allows corporations to
externalize intra-group cash management, allowing them to manage their global
liquidity more effectively and affordably (Colangelo, 2016). Cash owned by all
business groups spread across each company can be managed optimally using this
scheme. Research supports the fact that it provide some benefits: reduction of foreign
exchange risk, possibility to receive and present the information on bank accounts of
a given unit, and no need to make decisions concerning borrowing or placement of
liquid assets (Grzywacz, 2008).
1.2. Research Problem

Based on previous concern, research into the tax implications of cash pooling is
critical. A cash pooling incorporates companies in a business group where the majority
of cash pooling transactions are influenced by affiliated company. Transactions
impacted by such special relationships must adhere to the Arm’s Length Principles
(ALP) or Prinsip Kewajaran dan Kelaziman Usaha (PKKU). Failure to comply may lead
to a taxpayer being accused of profit shifting, resulting in a new tax bill. Hence,
research on the cash pooling business process and its relationship to related tax
provisions is critical in order to mitigate the risks associated with this practice.
The purpose of this research is to examine the impact of tax provisions in
Indonesia on the practice of cash pooling by business groups. The study will
concentrate on the application of the Arm’s Length Principles (ALP) practices to
transactions impacted by related party in cash pooling. Using qualitative method, this
study compared cash pooling business practices with related tax provisions. The
following topics will be covered:
a. identification of transactions influenced by related party;
b. price indicators; and
c. selection of tested party.
2. THEORETICAL FRAMEWORK AND PROBLEM STATEMENT

DEVELPMENT
2.1. Cash Pooling Business Process
According to Ciezki and Drozdz, a business group's ability in liquidity management
is critical. The ability of a company to meet its short-term obligations in a timely manner
is referred to as liquidity management. Failure to handle this properly can have serious
consequences for the company, even leading to bankruptcy. As a result, in a business
group, the parent company typically plays a role in implementing liquidity management
for the entire business group.
In order to manage liquidity, the parent company will analyze the entire company's
cash flow structure, including both internal and external party generated cash flows.
The two main issues in cash management, according to Sierpiska and Wdzki (2002),
are cash flow velocity and efficiency.
Implementing liquidity management will add another financial cost to a business
group. Banking institutions provide cash management services in the form of cash
pooling to streamline these activities. Cash pooling, according to Grabowska (2012),
is the process of transferring cash between the accounts of the entities involved and
the entity group's main account.
Several accounts can be managed collectively in cash pooling. Furthermore, this
entails coordinating liquidity and related risks. According to the OECD, it is expected
that this scheme will reduce the financial costs of the entire business group by
reducing the spread between loan interest and deposit interest as well as transaction
costs. In general, the party involved in organizing cash pooling are as follows:
1. Independent Bank
Essentially, the bank still plays the role of a third-party financial management
institution in cash pooling, albeit with a process that simultaneously incorporates
multiple enterprises. For all participants in the cash pool, the bank serves as the cash
manager. In addition to performing other pertinent tasks, the bank will keep participant
funds, transmit money between participants, charge management fees, provide
interest on cash pooling depositors' deposits, and so forth.
From the banking perspective, the cash pooling system uses two strategies:
notional pooling and cash concentration. While the bank will aggregate the balances
of many firm accounts in notional pooling, the money will still be in each cash pooling
member's individual accounts. On the other hand, in a cash concentration mechanism,
money belonging to all cash pool participants is physically transferred to one combined
account.
2. One particular company as the cash pool leader
In cash pooling, the cash pool leader must be one particular company. Basically,
the role of the cash pool leader is to oversee how the available funds are used for the
benefit of all cash pool participants. The position of the cash pool leader might be
played in one of two ways: as a in-house bank or as a service provider.
The cash pool leader performs the role of an independent bank when acting as
the company's in-house bank. The role of the cash pool leader is to take on members'
surplus funds and distribute them to those who ask for funding. As a result, the cash
pool leader can act as an intermediary for providing internal capital needs in the
business group, allowing capital to be available more quickly and at a lower cost.
In the meantime, in order to maximize benefits for the whole business group, the
cash pool leader serves as a fund manager for the group as a whole. The needs of
every member will determine how these benefits are provided. The benefit is improved
deposit rates for members with extra funds. The upside is a lower cost of capital for
members who require money, though.
No matter what capacity the leader plays, it will mostly be responsible for
communicating with the bank. With a consolidated fund capacity, the leader can
bargain with the bank on many issues for the advantage of the entire business group.
The leader might, for instance, bargain over deposit interest rates, loan interest rates,
and export privileges.
3. Other companies as cash pooling member
Companies participated in cash pooling scheme of a business groups, which are
not as the leader are referred to as cash pooling members. As a member, their role in
the cash pool relies on whether the company has an excess or not sufficient cash. If
the company has extra funds, cash pooling will serve as a source of funding for other
members. While this is going on, if a member is experiencing a cash deficit, they will
start using the cash pooling's available funds.
2.2. Bank as the Business Unit

The Decree Number 7 of 1992 about Banking and its Amendments emphasizes
that the primary role of banking is to collect and disperse public fund. Banks take
government funds in the form of savings and then disperse them via loans or other
means. This bank serves a very vital role in economic growth since it directs cash from
party with excess funds to commercial operations that can stimulate the economy.
According to Levin, banks will be able to stimulate a nation's economic growth if they
can effectively perform this role of financial intermediary.
Banks offer incentives in the form of deposit interest to encourage the collection of
public funds. The bank will charge the debtor loan interest in order to cover this deposit
interest. The Net Interest Margin is the difference between the interest on a loan and
the interest on a deposit. Claeys and Vennet argue that the bank's location's market
conditions are reflected in the amount of net interest margin. A high Net Interest Margin
is a sign that banks are operating inefficiently and that there is less competition in the
banking industry. A high Net Interest Margin also suggests that banking regulations
are inadequate and that there is a high level of asymmetric information.
2.3. Accounting for Cash Pooling Activities

The existence of cash pooling will result in cash transfers from one company to
another.Cash will flow from cash pool members to leaders at the time the cash pool is
formed. Cash will flow in the opposite direction when it is used for business or other
reasons, from the leader to the other participants in the cash pool. Three different
classifications of cash-related activities identified by PSAK 2 are as follows:
1. the company's operating activities, which are mostly derived from its primary
revenue-generating activities.
2. Investing activities, particularly those involving assets meant to provide income
and cash flow in the future.
3. financial activities, including cash flows from capital providers to the business.
Some transactions pertaining to the creation and use of cash pooling funds must
be recorded, depending on the type of cash pooling used. The vast majority of cash
transfers between companies in the context of establishing or withdrawing cash
pooling funds do not involve sales of goods or services that can generate revenue. As
a result of these transactions, each company involved now has debts, liabilities, and
receivables.
Based on PSAK 1, liabilities can be divided into two categories: short-term and
long-term liabilities. Short-term liabilities must meet one of the following criteria: (1)
expect to settle these liabilities in one normal cycle; (2) have the liability for trading
purposes; (3) the liability is due to be settled within 12 months after the reporting
period; or (4) do not possess the unconditional right to defer settlement of the liability
for at least 12 months after the reporting period. Liabilities are categorized as long-
term liabilities if they do not meet certain requirements.
2.4. Best Practice of Tax Applications on Affiliated Transactions Related to

Cash Pooling
According to the OECD, a corporate group often has a treasury department that
manage its financial operations. To guarantee that there is enough cash available at
the appropriate time and location, the person in charge of Treasury must optimize the
liquidity circumstances across the whole business group. Treasury responsibilities
include the management of cash and liquidity, as well as corporate financial
management.
The availability of cash in the company group's ongoing operations is a key
concern for the cash and liquidity management’ person in charge. Meanwhile, the
person in charge must act with a long-term perspective when performing the role of
corporate financial management. Business groups must be aware of any risks that
may arise while maintaining business continuity in order to make financial decisions
that optimize the cost of capital, such as issuing bonds, bank loans, and investment
options.
The treasury's function might also be carried out by creating a cash pool. When
compared to situations without cash pooling, generally speaking, the presence of cash
pooling should offer better conditions for all involved party. This viewpoint affects the
distribution of reward among all cash pooling party, including cash pool leaders and
cash pool members.
When a related party relationship has an impact on a transaction, the decision to
proceed with the transaction may be influenced by an interest to reduce the total
amount of tax that must be paid. As a result, the transaction's fairness might not be in
line with the Arm’s Length Transaction (ALP). According to Article 18 paragraph (4) of
Income Tax Law Number 8 of 1983 and its amendments, a special relationship (related
party relationship) exists if:
a. Taxpayers have a direct or indirect capital participation of at least 25% (twenty
five percent) in other taxpayers; a relationship between taxpayers where there
is a participation of at least 25% (twenty five percent) in two or more taxpayers;
b. The taxpayer controls other taxpayers; two or more taxpayers are under the
same control, either directly or indirectly; or
c. There is a family relationship, whether it be blood or adoption.
Cash Pooling is used by companies in the same business group to conduct
transactions. As a result, these transactions are influenced by a particular relationship.
It must be tested to determine whether or not the transaction complies with ALP.
Cash pooling transactions can take the form of providing cash management services
or savings and loan transactions, depending on the cash pooling scheme used. The
Minister of Finance Regulation Number 22/PMK.03/2020 states in Article 14
paragraphs (1) and (2) that transactions in the form of service transactions and
transactions related to loan costs must be completed in preliminary stages before ALP
testing. The following are the preliminary steps that must be carried out:
1) in the case of service transactions, documentation demonstrating that the
services:
1) were genuinely rendered by the service provider and accepted by the
service receiver;
2) were necessary for the service recipients; and
3) brought about financial gains for the service recipients;
4) is not a shareholder activity;
5) is not a passive association activity that benefits a party only because
the party is a member of a business group;
6) is not a repetition of activities already carried out by the taxpayer;
7) is not a service that offers incidental benefits; and
8) in the case of on-call services, it is not a service that can be immediately
obtained from an independent party without prior arrangement.
2) for loan interest-related transactions, evidence of:
1) conformity with the substance and actual circumstances;
2) borrower need; and
3) use to obtain, retain, and collect income in compliance with the rules of
laws and regulations governing income tax;
4) satisfy the loan characteristics, which include:
a) the creditor recognizes the loan economically and legally;
b) there is a loan maturity date;
c) there is a requirement to settle the loan principal;
d) there are payments according to a predetermined payment schedule
for both the loan principal and the yield; and
e) at the time the loan is obtained, the borrower has the capacity to obtain
a loan from an independent creditor and repay the loan principal and
yield.
f) based on a loan agreement created in accordance with the relevant
laws and regulations;
g) legal repercussions may occur if the borrower fails to repay the loan's
principal and/or yield;
h) the lender has the right to pursue collection as an independent
creditor; and
5) offer borrowers some financial advantages.
3. RESEARCH METHOD
This is a qualitative study, which examines the current taxation related to cash
pooling practices. This study further explores how ALP -related tax regulations are
applied to cash pooling-related transactions. Therefore, the base of analysis of this
research are:
a) Decree Number 7 year 1983 Concerning Income Tax and its Amendments,
and
b) Regulation of the Minister of Finance Number 22/PMK.03/2020 Concerning
Procedures for Implementing Transfer Pricing Agreements serve as the
foundation for testing in this study (PMK-22),
as well as best practices for implementing transfer pricing in accordance with the UN
Practical Manual on Transfer Pricing for Developing Countries and OECD Transfer
Pricing Guidelines for Multinational Enterprises and Tax Administrations.
The steps performed in conducting research to achieve the expected results are as
follows:
a. Identifying related transactions that need to comply with ALP;
b. Research on price indicators; and
c. Choosing tested party.

4.1. Transactions among Cash Pooling Members
Haller and Chand proposed that the financial arrangement of the cash pooling
must be in compliance with domestic regulations governing the funds transfer. For
instance, financial agreements reached through cash pooling are regarded as loans
in Germany. As a result, the cash pooling recipient will be considered to have received
the loan, and the sender of the funds will be the owner of the receivables resulting
from the loan.
Generally, transactions can be divided into two categories: balance sheet
transactions and income statement transactions. Transactions involving the transfer
of assets from one party to another are known as balance sheet transactions. Income
statement transactions, on the other hand, are transactions that involve income
statement accounts, such as income or expenses, in order to have an impact on the
transacting entity's equity position. This income statements’ transaction relates to the
determination of the tax payable.
Cash pooling involves three partners, as was previously mentioned. Therefore,
transactions involving cash pooling may proceed between these three parties.
Transactions between independent banks and leaders or members are considered
independent when examined from the perspective of the relationship.
Meanwhile, affiliated transactions take place whenever a transaction is made
between member companies in a cash pool, particularly between leaders and
members. Therefore, affiliated transactions must be subject to the implementation of
ALP. It is required to identify the associated transactions in order to test the
implementation of ALP. The leader's role in the cash pooling process will have an
impact on the type of transactions.
a. leader as the in-house bank
The leader performs as a bank for the business group is referred to as the "in-
house bank." The leader of the cash pooling group controls all cash flows. Members
with extra cash appear to be conserving it for the leader. In the opposite situation, if a
member needs money, the leader will lend them money from the cash pooling funds.
Several types of transactions occur between related parties in this scheme. Cash
transfers from members to leaders will occur as funds are accumulated in cash
pooling. The funds are deposited into the leader's account at the cash pooling
organizing bank. During cash pooling operations, members can only withdraw funds
when making payments for operational transactions that must be paid or depositing
funds into the cash pool if there is payment from the buyer for the sales he makes.
Meanwhile, if any members require loans, the leader, acting as corporate treasury, will
conduct corporate financial management activities to determine the viability of
providing loans. If it is possible, the leader will make loans to these members.
Furthermore, the leader will receive loan interest from members who borrow and
deposit interest from members who have positive cash balances in cash pooling.
Figure 1 describes the transaction relationship of each party.
Independent bank
Deposit’s
Saving interest
Grup Leader/in-
house bank Loan
Saving
Debt’s
Deposit’s
interest
interest
Group member/ Group member/

Cash surplus Cash deficit
Figure 1. Cash Pooling Transaction Flow Using the Leader

Scheme as the In-House Bank.
The leader's source of income is impacted by the mechanism. As the in-house

bank, the leader earns the Net Interest Margin/spread, which is the return generated
from cash pooling fund invested less interest payments. The return on this investment
may come from loan interest or from interest paid by the bank due to a positive balance
in the overall cash pooling fund. Meanwhile, interest obligations raise from both
interest liability to banks and interest that is a right of cash pooling members, whether
paid or not paid.
Figure 1 shows how transactions with independent banks, such as cash deposit
transactions, which are balance sheet transactions, and transactions interest earned
on bank deposits, which are income statement transactions, will proceed through the
cash pooling leader. Since earning interest on deposits from the bank is a profit and
loss statement activity, there is typically no special relationship between the bank and
the leader, making this transaction an independent one. As a result, this transaction
can be seen as satisfying the ALP.
Several transactions are between leaders and members. The existence of a
special relationship between the party involved has an impact on the transaction, thus
it must satisfy the ALP. The affiliated transactions include the following:
1) cash transfers for the purpose of creating a cash pooling and depositing
sales/other revenues from members to leaders;
2) withdrawal of funds from the leader to the cash pooling members;
3) granting loans to members;
4) interest payments on deposits from leaders to members; and
5) loan interest payments from members to leaders.
Among the abovementioned transactions, those for cash transfer, receiving
money, and making loans transactions appear on the balance sheet. While loan
interest payments from members to leaders and deposits from leaders to members
are income statement transactions, respectively. Therefore, ALP compliance is
required for transactions involving the payment of interest on deposits and loans.
Nevertheless, because the balance sheet transaction will serve as the foundation for
the calculation, this income statement transaction is still inextricably linked to the one
that occurs.
b. leader as a service provider

When the leader assumes the role of a service provider, the funds accumulated in
the cash pool are truly under the control of each member. In terms of financial
management, the leader serves as a service provider, meeting the demands of all
members. For instance, the leader bargains for the best interest rate and identifies
more affordable and long-lasting loan sources.
Members appear to be in direct contact with independent banks in this scheme,
both in the form of physical cash pooling and notional cash pooling. When payments
are made for sales performed by the members, they transfer cash into the bank to
build a cash pool. The bank will reimburse each member for the interest on the deposit
from the remaining balance of each members’ fund balance. In the meanwhile, it
appears as though the group's leader is paid by the members for providing cash
pooling services. In practice, a member who requires a loan may apply for one from
the organizing bank. However, in this study, this was not considered because the
transaction was actually outside the scope of cash pooling. Figure 2 depicts a
transaction illustration.
Independent bank
Saving
Deposit’s negotiation
interest
Group member Group Leader

fee
Figure 2: Cash Pooling Transaction Flow with the Leader

Scheme Serving as a Service Provider
The figure above clearly illustrates that members who conduct business with
independent banks directly in the form of deposits and withdrawals, which are balance
sheet transactions and interest receipts, which are income statements transactions.
Banks and members commonly do not have any affiliated relationship, hence, this
transaction is regarded as being in compliance with ALP. The leader's receipt of fees,
which is a transaction that appears on the income statement and is affected by an
affiliation. Therefore, the ALP implementation for this fee payment transaction needs
to be tested.
4.2. Tax Implementation of Cash Pooling

a. Identification of Affiliated Cash Pooling Transactions
Transactions impacted by a special connection must be checked for ALP
compliance, as was previously stated. Because income statement transactions
directly affect the amount of income tax liability, these transactions are the ones that
this test focuses on the most. The following cash pooling transactions need to be
tested based on the previous identification:
1) In the leader scheme as an internal bank:
a. Payment of loan interest from members to leaders; and
b. Payment of interest on deposits from leaders to members; and
2) In the leader scheme as a service provider: Payment of fees from
members to leaders.
Basically, as a research of Article 9 Paragraph 1 of PMK-22, After the introductory
steps, as previously explained, ALP research is conducted for cash pooling
transactions. In line with Article 8 Paragraph 2, the execution of this ALP is essentially
carried out by comparing the terms and price indicators of transactions that are
impacted by a special relationship with the terms and price indicators of comparable
independent transactions. As a result, for the aforementioned cash pooling
transactions, it is required to determine the right price indications.
For loan interest and deposit interest in the leader as in-house bank scheme, it is
found some value relevant to transactions, which is the loan value/deposit value, the
amount of interest paid, and the interest rate. The interest rate should be the indicator
of the transaction for paying interest on loans/deposits depending on the features of
the transaction, out of the three indicators. The ALP test will therefore compare the
interest rates charged on transactions to independent interest rates for similar
scenarios.
It is required to first research the nature of the transaction before paying fees to
the leader scheme as a service provider. There are certain transactional
characteristics that are categorized as low value-adding intra-group services
transactions in the OECD Transfer Pricing Guidelines. These characteristics include:
1) being supportive;
2) not being a part of the group's core activities in terms of profit-earning;
3) not requiring unique and valuable intangibles and not participating in the
creation of such intangibles; and
4) not being involve in significant risk control or causing significant risks.
The services offered by the leader in the leader plan fit the aforementioned criteria
when referring to such characteristics. The leader's services fall within the category of
low value adding services as a result. The OECD advises using the simplified
technique with a maximum profit level of 5% of the total relevant costs for low value
adding services. While all members benefit from cash pooling lead to all members are
required to pay for these services. As a result, the following procedure is used to
establish price indications for paying this leader for cash pooling services:
1) Calculation of all leaders' expenses related to cash pooling operations;
2) Establishment of the appropriate profit level;
3) Distribution of fees to all members.
ALP implementation necessitates identifying tested party after identifying pricing
indications. A price indicator's "tested party" is the party that executes the transaction.
According to the previously completed functional analysis, the choice of the tested
party must be consistent.
The selection of tested party is typically based on circumstances in which transfer
pricing methods are the most accurate to use and accurate comparison data are
available. In terms of the transfer pricing method, the party that has a most simple
business process typically being chosen as the tested party. In terms of the availability
of data to comparative analysis, tested party were chosen based on which party
naturally engages in transactions with independent party. The tested party for related
transactions in cash pooling are chosen based on the following criteria:
1) Repayment of interest on loans
Members as income generator and the leader-in-house bank as income recipients
are the people involved in paying loan interest. Between the two, by nature member
as borrower has choices to propose loans to independent parties. However, because
these investments are made to cash pool interest rather than serve as loans in the
context of business operations, they cannot be considered as loans. Leaders do
allocate funds in separate banks in the meantime. As a result, from the leader's
perspective, it is typically impossible to find reliable comparison data because the
leader can typically only lend money to other cash pooling members. Members should
therefore be chosen as the tested party for loan interest payment transactions.
2) Payment of interest on deposits
The leader-in-house bank pays deposit interest to members who are income
recipients. From the leader's perspective, the sole sources of deposits are the
members' cash pools. Members can principally choose where to invest their money in
other financial organizations or a cash pool. Members are therefore become a more
reliable choice to become tested party in this transaction.
3) Payment of fees
As previously discussed, in a cash pooling arrangement where the leader serves
as a service provider, the leader's work is classified as low-value intra-group services.
For transactions that fall under this category, the price indicator often consists of a
specified percentage profit over the service provider's overall related costs. The fact
that the person delivering the service is the leader indicates that in this transaction,
the leader is the appropriate party to be chosen as a testing party.
Table 1 below provides a summary of the full above description.
Table 1. The Overview of ALP Implementation on Affiliated Transactions in Cash

Pooling
In-house Bank Scheme Service Provider

Scheme
Loan Interest Deposit Interest Service Fee
Income Generator Member Leader Member
Income Receipient Leader Member Leader
Price Indicators Loan interest rate Deposit interest rate Profit level based on
low value-adding
services
Price Allocation Based on loan Based on deposit Cost allocations to all

value value members
Tested party member Member Leader
Reason for selecting the Data for Data for comparison Low value-adding
test party comparison is is more reliable services
more reliable
5. CONCLUSION
Cash pooling is a method for a business group to manage its cash. Cash pooling
allows business groups to optimize the use of their internal funds, such as the use of
group cash by one of the companies experiencing a cash shortage, reducing the use
of external funding sources. A business group with a large accumulation of funds may
also have better financial opportunities, such as negotiating a better rate of return or
seeking cheaper sources of financing.
Three groups are often involved in cash pooling: independent banks, cash pooling
leaders, and cash pooling members. Cash pooling can be classified into two
categories in terms of the leader's role: the leader as an in-house bank and the leader
as a service provider. The transactions made in the cash pooling are impacted by
these variation in the leader's function.
Companies in the same industry group participate in cash pooling, resulting in
some of the transactions being influenced by affiliated relationship transaction. Tax
implications arise from this situation. Those transaction should conform the Arm’s
Length Principles (ALP) or Prinsip Kewajaran dan Kelaziman Usaha (PKKU). Hence,
it is crucial to conduct research on how to implement ALP to cash pooling.
Finding the relevant associated transactions is a required preliminary step in the
early stages. The leader's role in the cash pooling process affects these transactions.
Loan interest payments and deposit interest payments are related transactions that
must satisfy the ALP in the leader scheme as an in-house bank. Meanwhile, the
transactions identified in the leader scheme as a service provider, are considered as
fee payment transactions.
Basically, two steps of research are required to conduct related transactions in
cash pooling. According to PMK-22, it is necessary to make loan interest and service
charge payments in the beginning to verify their existence. In addition, research on
the implementation of ALP for all associated transactions is required, including study
of pricing indicators and tested party.
The loan interest rate serves as the price indicator for interest payments on loans.
The member who acquired the loan is the tested party in this transaction since the
comparison data is more reliable. Therefore, the interest rate applied to cash pooling
will be compared with the independent loan interest rate that can be obtained by
members as debtors in order to assess the implementation of ALP loan interest
payments in a cash pooling.
The deposit interest rate is used as a pricing indication for payments of deposit
interest. The chosen tested party is a member in order to have an accurate comparison
data. As a result, in payment of deposit interest transactions, the applied deposit
interest rate is considered to comply with ALP if it is comparable to the independent
deposit interest rate that the member may earn as the deposit's owner.
Under the leader as a service provider scheme, services that the leader offers are
classified as low value adding services. The OECD recommends a simplified
approach when implementing ALP for services that fit into this category. The whole
relevant costs are used to determine the service delivery value, and then a maximum
of 5% profit may applied. As a result, the tested party is as the service provider in this
transaction.

This study investigates how the Arm’s Length Principles (ALP) or Prinsip
Kewajaran dan Kelaziman Usaha (PKKU) might be applied to related cash pooling
transactions. This research offers recommendations for ALP implementation in cash
pooling. Taxpayers may follow some procedures that diverge from the findings of this
study. This is not necessarily a deviant application and requires more in-depth
research.
This study includes a major study of literature cause a limitation to the results.
Hence, this study's ability to provide additional empirical data is limited. Conducting a
more depth analysis of cash pooling activities in the practices is necessary in order to
capture specifics phenomena of cash pooling and its taxation aspects.
REFFERENCES
CIęŻKI, D., & Drożdż, W. (2019). Using multicurrency cash pooling in the liquidity
management of a capital group. Polityka Energetyczna, 22.
Colangelo, Antonio (2016) : The statistical classification of cash pooling activities, ECB
Statistics Paper, No. 16, ISBN 978-92-899-2417-7, European Central Bank (ECB), Frankfurt a.
M., https://doi.org/10.2866/6893
Haller, A. (2019). Application of the arm’s length principle to physical cash pooling
arrangements in light of the OECD discussion draft on financial
transactions. Intertax, 47(4).
Hutagalung, E. N., & Ratnawati, K. (2013). Analisa Rasio Keuangan terhadap Kinerja
Bank Umum di Indonesia. Jurnal Aplikasi Manajemen, 11(1), 122-130.
Oecd Publishing. (2017). OECD Transfer Pricing Guidelines for Multinational
Enterprises and Tax Administrations 2017. OECD.
Raharjo, P. G., Hakim, D. B., Manurung, A. H., & Maulana, T. N. (2014). The
determinant of commercial banks’ interest margin in Indonesia: An analysis of fixed
effect panel regression. International Journal of Economics and Financial Issues, 4(2),
295-308.
VAN HERKSEN, MONIQUE, JIE-Ajoen, CLIVE, & BAI, F. (2020). INSIGHT: New
OECD Guidance on Transfer Pricing Aspects of Financial Transactions—Part 2. Tax
Management Transfer Pricing Report, 2020(February), 1-11.
Vasco, D. C. (2021). Transfer pricing guidance on financial transactions. Inclusive
framework on BEPS: actions 4, 8-10 de OECD. Crónica tributaria, (180), 175-178.
ASSESSING TAXPAYERS’ ABILITY TO PAY: A
MACHINE LEARNING APPROACH31
Sukaryo a, Adi Marhadi b

a Direktorat Jenderal Pajak, Jakarta, Indonesia. Email: sukaryo1@gmail.com
b Direktorat Jenderal Pajak, Manokwari, Indonesia. Email: a.marhadi@gmail.com
ABSTRACT
Tax revenue remains one of the challenging fiscal issues in Indonesia. Improving tax collection
capability through comprehensive reform has been an influential agenda, especially for the
Directorate General of Taxes. One of the critical improvement areas is the utilization of
information technology in tax assessment and audit functions. This study explores the taxpayers’
ability concept to complement the existing taxpayer’s monitoring module, specifically the case
selection and targeting function. The 5Cs of credit analysis (Character, Capacity, Capital,
Condition, and Collateral) are used to proxy the taxpayers’ ability to pay. This study uses several
machine learning algorithms to explore the ability to pay concept for corporate taxpayers, limited
to those administered in large and medium tax offices. The machine learning approaches
provide a more robust classification output than the non-machine learning model, i.e., logistic
regression, based on the quantitative and qualitative performance comparison. However,
several challenges need to be addressed to improve the model implementation.
Keywords: machine learning, ability to pay, taxpayers, tax compliance.
Penerimaan pajak masih menjadi salah satu tantangan kebijakan fiskal di Indonesia. Upaya
peningkatan kemampuan pengumpulan pajak melalui reformasi pajak merupakan salah satu
tujuan kebijakan utama, khususnya bagi Direktorat Jenderal Pajak. Salah satu fokus perbaikan
utama adalah pemanfaatan teknologi informasi untuk menunjang fungsi pengawasan dan
pemeriksaan pajak. Penelitian ini menelaah konsep kemampuan bayar (ability to pay) untuk
mendukung fungsi pengawasan wajib pajak, khususnya terkait fungsi penentuan sasaran
pengawasan dan pemeriksaan. Konsep 5Cs dalam analisis kredit (Character, Capacity, Capital,
Condition, dan Collateral) digunakan untuk menjembatani konsep kemampuan bayar wajib
pajak. Penelitian ini menggunakan beberapa algoritma machine learning (pembelajaran mesin)
untuk memetakan konsep kemampuan bayar bagi wajib pajak badan, khususnya yang
diadministrasikan pada kantor pajak besar dan madya. Secara umum, pendekatan machine
learning menghasilkan keluaran yang dapat diandalkan, jika dibandingkan dengan model non-
machine learning, yaitu regresi logistik, berdasarkan penilaian kinerja berbasis kuantitatif dan
kualitatif. Namun, terdapat beberapa tantangan yang harus diatasi untuk meningkatkan
kemampuan model.
Kata kunci: machine learning, kemampuan bayar, wajib pajak, kepatuhan pajak.
1. INTRODUCTION
Tax revenue is a daunting challenge in Indonesia, with a stagnated revenue to
GDP ratio of roughly 11 per cent or considerably lower than the regional and income-
31
The Ability-to-pay module has been officially launched and deployed on 14 July 2021. This paper
only explores the background study and initial development of the module.
group average (de Mooij et al., 2018). In general, the low revenue ratio suggests an
inefficiency in tax revenue collection (IMF, 2019). Consequently, there has been
mounting pressure to improve tax collection through various policy and administrative
measures. Furthermore, from 2017 to 2020, the government, through the Directorate
General of Taxes (DGT, Indonesia’s tax authority), carried out the latest
comprehensive tax reform, which encompasses policy and administrative pillars
(DGT, 2020).
However, even though the reform generally aligned with the Medium-Term
Revenue Strategy (MTRS) framework, there is still some potential for improvement,
for example, the administrative reform (de Mooij et al., 2018). Arguably, the tax
administration is one of the ever-growing challenges as the increasing taxpayers’
baseline is followed by limited filing compliance and suboptimum administrative
capability.
In 2008, the DGT administered around 10.68 million taxpayers, and the number
quadrupled in 2019, reaching approximately 45.95 million taxpayers. Indeed, the
lodgment rate improved considerably, from 30.96 per cent in 2008 to 73.06 per cent
in 2019. However, the filing rate is lower than the 85 per cent on-time filing rates of
OECD and selected economies (OECD, 2019).
Meanwhile, the growth of human resources lagged, from 31.8 thousand in 2009 to
46.6 thousand in 2019. Compared to regional peers, Indonesia’s labour force to tax
officer ratio of 2952 is higher than China (1979), Singapore (1739), Malaysia (1142),
and Australia (721). At the same time, the DGT suffers from productivity complications:
higher staff allocation for routine and supporting activities and audit resources
misallocation (de Mooij et al., 2018). Therefore, the DGT could harness the
Information Technology (IT) potential to improve risk management and case selection
functions for targeting high non-compliance risks (de Mooij et al., 2018).
The use of technology to promote tax compliance has earned a more pivotal role,
especially in tax audit and assessment functions (OECD, 2016). In particular, the
emerging implementation of the advanced analytics approaches improves audit case
selection (OECD, 2016). By large, an analytics-driven approach could promote a more
efficient process (Davenport and Harris, 2007).
In recent years, DGT has introduced and incorporated technology initiatives,
including data analytics, as one of its leading strategies (DGT, 2016; DGT, 2017; DGT,
2020). One of the highlights of data analytics implementation in the DGT is the
Compliance Risk Management (CRM) program (de Mooij et al., 2018; DGT, 2017). In
short, the CRM clusters the taxpayers based on their non-compliance risk and then
map them according to probability and consequence level. As a result, the tax
verification staff could prioritize assessment activity for the highest risk group.
However, as a relatively new case selection tool implemented in DGT, the CRM
function has yet to account for the taxpayers’ future outlook, i.e., the taxpayers’
financial capability to pay the tax due if the DGT conducted the audit. One of the
commonly used methodologies to assess audit performance is the outcome approach,
based on the cost-benefit principle (OECD, 2006). In other words, an audit would be
generally favourable if it could generate considerable future revenue. Also, as the
OECD (2006) suggested, the primary aspect of a successful audit is related to the
case selection approach. Therefore, it would be beneficial for DGT to improve its case
selection capability to detect taxpayers’ collectability or future ability to pay.
Based on micro-level tax return data, this paper will explore the ability to pay (ATP)
concept in Indonesia. Particularly, this research tries to follow the relevant ATP
concept that would be applicable and build a relevant machine-learning model that
could predict future taxpayers’ ATP. This paper adopts the machine-learning approach
to utilize the DGT’s data analytics infrastructures and address the pattern-recognition
challenges with the micro-level dataset (Khandani et al., 2010). Furthermore, several
studies showcased that the machine-learning techniques offer a powerful prediction
power when used in financial sectors, e.g. to model credit risk (van Liebergen, 2017;
Khandani et al., 2010; Provenzano et al., 2020; Bellotti and Crook, 2009). Thus, we
seek to reproduce the findings in tax administration settings.
This paper aims twofold: to provide an introductory study on the ATP concept and
complement the CRM implementation as case selection tools for tax verification
function in Indonesia. The ATP module would provide valuable information for account
officers and auditors before initiating tax assessments or at least prioritizing the CRM
execution based on taxpayers’ predicted financial capability. The rest of this paper will
discuss the relevant ability-to-pay concept, limitation of the study, methodology, result,
and conclusion.
Equity has been the heart of taxation principle, and one of the primary examples
is Adam Smith’s argument that “[t]he subjects of every state ought to contribute
towards the support of the government, as nearly as possible, in proportion to their
respective abilities” (Smith, 1776)32. In particular, Musgrave (1996) argued that the
interpretation of fairness in taxation is primarily aligned with the ability to pay concept.
Furthermore, Musgrave (1996) added that the equity principle in taxation consists of
two concepts, horizontal equity (“[p]eople with equal ability should pay the same”) and
vertical equity (“those with higher abilities should pay more”).
The quantification of the ability to pay notion is arguably somewhat limited,
especially when applying it to the tax administration domain. From a policy
perspective, we generally could determine the taxpayers’ ability to pay based on their
realized income or wealth, e.g., one of the bases for progressive income tax 33 .
However, under the administrative perspective, e.g., tax audit, the tax authority should
assess both present and future taxpayers’ financial capability as the audit examines
past compliance resulting in future tax liability34. Therefore, this study will adopt the
32
Musgrave (1996) argued that Smith’s assertion combined two principles of equity, benefit and ability
to pay.
33
For example, Slemrod (1996) suggested that we could interpret the ability to pay principle as
“equalizing the sacrifice due to tax”; thus, “tax should rise with income” even though the implementation
might not as straightforward as the concept proposed. Similarly, Gruber (2011) use the Haig-Simons
comprehensive income definition that define “taxable resources as an individual’s ability to pay taxes” or
“individual’s potential annual consumption, potential annual consumption, plus any increase in his or her
stock of wealth”.
34
One of the relevant concept in tax audit is the collectibility or a taxpayers’ ability to pay the future tax
assessment as a basis to conduct tax audit (OECD, 2006)
ability to pay concept as taxpayers’ financial capability based on an accounting
perspective and credit analysis.
From an accounting perspective, assets are a relevant proxy of one’s financial
capability. International Accounting Standard (IAS) states that an asset is “a present
economic resource controlled by the entity as a result of past events. ([a]n economic
resource is a right that has the potential to produce economic benefits)” (Alibhai et al.,
2020). Specifically, IAS classifies current assets as “cash, cash equivalents and other
assets that are expected to be realized in cash, or sold or consumed during one normal
operating cycle of the business” (Alibhai et al., 2020). As a comparison, IAS defines
income as “[i]ncreases in assets or decreases in liabilities, that result in increases in
equity, other than those relating to contributions from holders of equity claims” (Alibhai
et al., 2020). Therefore, it would be possible to capture taxpayers’ financial capability
based on their assets, specifically the current assets.
Furthermore, credit analysis will be a relevant concept in assessing taxpayers’
financial capability. Following its description, credit analysis is a “systematic and
comprehensive assessment of a firm’s capacity and willingness to pay its financial
obligations” (Ganguin and Bilardello, 2005). A systematic credit analysis approach
under “the 5 Cs of credit” could provide a comprehensive explanation for the ability to
pay concept: Character, Capacity, Capital, Conditions, and Collateral (Ganguin and
Bilardello, 2005; Golin and Delhaise, 2013).
According to Ganguin and Bilardello (2005), Character is related to a firm’s
reputation, strategy, and leadership. The Character factor in credit analysis is primarily
related to a firm’s management, corporate governance, and financial policy. In
general, companies with aggressive growth strategies create higher-risk credit
profiles. The second C, Capacity, highlights the firm’s ability to generate cash or
income. Meanwhile, Capital is related to the firm’s capitalization or investment. In other
words, the Capital in credit analysis focuses on the amount of the owner’s investment
in the firm. Conditions portray the firm’s competitive environment and how well-
positioned the firm is in the market. In measuring firms’ competitiveness, it is essential
to incorporate various underlying factors that could determine firms’ market position,
e.g., the cost factor for price competition. Lastly, Collateral highlights the alternative
source of funding to repay the obligations.
Among the five aspects, evaluating one’s capability through past and present
financial statement data is the “core of credit analysis” (Golin and Delhaise, 2013).
Therefore, Capacity, or measuring the firms’ cash flow and financial risks, remains the
primary factor in credit analysis. However, quantifying and capturing the 5 Cs might
be challenging given the specific nature of the business and information availability
(Golin and Delhaise, 2013). In particular, Golin and Delhaise (2013) illustrated the
sectoral-based credit analysis arrangement and different analytics approach for
various business scale.
3. METHODOLOGY
3.1. Methodology and Data
This study’s primary approach centred around the supervised learning
classification method, using taxpayers’ ability to pay as the target variable. In
particular, this study explores several machine learning algorithms to ‘label’ taxpayers’
ability to pay for the following year (t+1) based on the current year’s predictors (t+0).
As a target variable, this research uses current assets as a proxy for taxpayers’ ability
to pay, following the accounting and credit analysis principle.
As the machine learning algorithms, this paper explores several classification
models to get an optimum approach based on accuracy and resource efficiency:
Support Vector Matrix (SVM), Naïve Bayesian, tree-based Gradient Boosting, and
Random Forest classifier methodology. The tree-based approach is highly relevant in
this study as it could capture the non-linearity trend and multiple categoric variables,
optimized for a relatively smaller dataset compared to the neural network algorithm.
The selection for the SVM algorithm is based on its capability in generalization and
pattern recognition (Burges, 1998). Meanwhile, the Naïve Bayes approach is one of
the most competitive classifications of machine learning techniques, especially in
independent and dependent features (Rish, 2001). The tree-based Gradient Boosting
algorithms are generally more favourable due to their efficiency, accuracy, and
interpretability (Ke et al., 2017). Lastly, this paper also compares the Random Forest
classification result as one of the most versatile algorithms in machine learning (Biau
& Scornet, 2016). This research utilizes logistic regression as a non-machine learning
approach for comparison.
When evaluating and comparing the models’ output, this study uses 75% of the
sample dataset as the “train data” to build the machine learning model. The remaining
25% of the dataset will serve as “test data” for model evaluation. This paper focuses
on five parameters to evaluate the model’s performance: accuracy, the area under the
curve (AUC), recall, precision, and execution time. The accuracy measures the
“correct” predictions based on the total number of samples. Meanwhile, precision and
recall emphasize the proportion of the “correct” identifications, with precision focusing
on “false positive” and recall focusing on “false negative”. The AUC parameter
measures the performance of the classification model, focusing on the true positive
rate (TPR) and false positive rate (FPR). The four parameters will take a value of 0 to
1, where 1 represents a model with 100% “correct” prediction. Lastly, this study
incorporates execution time, measured in seconds, as a proxy for model efficiency,
i.e., an efficient model will require a somewhat lower run time. Arguably, the lower run
time will be one of the significant parameters during the scaling-up and deployment
stage, especially applying the model with an extensive, high dimensionality dataset.
In order to forecast the ability to pay, there are two possible outputs: continuous
or categorical. Under the supervised learning methodology, both are plausible under
the “regression problem” for continuous values and the “classification problem” for
discrete values. However, to make the model easier to understand, explain, and
possibly faster, this study adopts the latter approach by constructing the taxpayers’
ability to pay as discrete value (Liu et al., 2002). This study categorized the taxpayers’
ability to pay, i.e., current assets, into five classes: “very low”, “low”, “moderate”, “high”,
and “very high”. Admittedly, the data distribution highly stretched from both tails,
making the grouping less representative. Therefore, this research further classified the
corporate taxpayers based on their administering tax offices. Accordingly, the cutoff
points for the five classes for taxpayers’ ability to pay vary across the tax office groups.
This research utilizes administrative and external data for around 60 thousand
corporate taxpayers, summarized as follows:
Table 1. Data Distribution, Train and Test Set
Source: Authors’ Calculation
Tax Office Total Data Train Data (75%) Test Data (25%)
Madya 54,878 41,158 13,720
Khusus 7,097 5,322 1,775
LTO 962 721 241
Total 62,937 47,201 15,736
This paper finds that the “Madya” and “Khusus” tax offices group are comparable
based on the data distribution. Thus, the corporate taxpayers will further be classified
into two tax office groups: Large and Medium. Accordingly, the modelling will be built
separately for the two classes.
3.2. Limitation
This study, however, is subject to data availability limitations. First, the data
limitation arises from asymmetric information from tax return forms and third-party data
exchange, especially for individual taxpayers who submit the 1770S or 1770SS forms.
At the same time, the attempt to construct taxpayers’ features under the 5 Cs of credit
analysis approach faces – at least – data representativeness challenges, e.g. relevant
information and data matching problems. In other words, increasing the sample size
would result in reduced dimensionality. Also, due to data availability issues, this study
limits the observations for corporate taxpayers administered in the medium and large
taxpayer offices.
Another data availability challenge is regarding the external data. While it is
possible to construct tax return data using several fiscal years’ information, the case
of third-party external data is contrasting: the data is limited to recent years. Therefore,
this study incorporates data from the 2019 and 2020 fiscal years to address the data
availability issues.
The last identified challenge is obtaining a representative dataset to measure the
5 Cs. Based on the administrative and external data, this study managed to proxied
the 4 Cs of credit – except Collateral. Arguably, the Capital and Collateral principles
are highly correlated: the total asset figure represents the amount of capital investment
and part of the firm’s collateral ability. Therefore, this study employs the 4 Cs principle:
Capacity, Capital, Character, and Condition.
Table 2. Independent Variables Mapping and Classification

Variable 4 Cs Type Remarks
Turnover Capacity Numeric

Commercial Nett Profit Capacity Numeric
Operating Cashflow Capacity Numeric
Working Capital Capacity Numeric
Nett Revenue Capacity Numeric
Gross Revenue Capacity Numeric
Cash Equivalents Capital Numeric
Nett Assets Capital Numeric
Total of Assets Capital Numeric
Current Liabilities Capital Numeric
Total Liabilities Capital Numeric
Domestic account Capital Numeric To complement the Cash

balance/Exchange of Equivalents variable
Information Data
Outstanding Credit Capital Numeric To complement the Current

Liabilities and Total Liabilities
Credit Collectibility Character Discrete 1. Current ;
2. Special Mention;
3. Substandard;
4. Doubtful;
5. Bad.
Status of PKP (Value- Character Discrete 1. Non-PKP;

added tax subject)
2. PKP.
Tax return filing status Character Discrete 1. Non-filers;
2. Normal;
3. Filer, correction.
Firm’s maturity Character Discrete 1. 0 – 5 Year ;
2. 5 – 10 Year;
3. 10 – 15 Year;
4. more than 15 Years.

Operating Cashflow Condition Discrete 1. Negative;

Status
2. Stagnant;
3. Positive
Status of receiving a tax Condition Discrete 1. No;

refund in the last six
months 2. Yes
Conglomerate group Condition Discrete 1. No;
2. Yes

Table 3 compares the classification result based on the prediction performance for
every algorithm on the test set. The model output suggests that the machine learning
approach performs better than the non-machine learning model, i.e., the logistic
regression. The machine learning models average around 0.7 of accuracy, precision,
and recall, while the run time varies across algorithms. Among the machine learning
algorithms, the Gradient Boosting groups (“gbc”, “lightgbm”, “xgboost”, and “catboost”)
emerge with the highest performance and efficient running time – for both Large and
Medium taxpayer models. For example, the non-Gradient Boosting model
performance for accuracy is around 0.49, while the Gradient Boosting approach
reaches 0.73. However, the Gradient Boosting algorithms are relatively more
exhaustive, with an average run time of around 30 seconds or 30 times slower than
the non-Gradient Boosting model (approximately 0.88 seconds).
Table 3. Model Performance Comparison

Acc Prec Time

Model AUC Recall
uracy ision (secs.)
Large Offices
Non-machine learning
- Logistic regression 0.2500 0.5508 0.2169 0.1307 0.1180
Machine learning
- SVM 0.1718 0.0000 0.2286 0.1893 0.0560
- Naïve Bayes (“nb”) 0.5261 0.8403 0.5204 0.5240 0.0160

Acc Prec Time
Model AUC Recall
uracy ision (secs.)
- Gradient Boosting Classifier (“gbc”) 0.6954 0.9104 0.6880 0.7072 0.5360
- Light Gradient Boosting Machine

0.6926 0.9152 0.6880 0.7072 0.2020
(“lightgbm”)
- Extreme Gradient Boosting (“xgboost”) 0.6770 0.9141 0.6674 0.6915 24.2280
- Catboost Classifier (“catboost”) 0.7082 0.9194 0.7053 0.7185 8.2400
- Random Forest Classifier (“rf”) 0.6718 0.9158 0.6649 0.6822 0.4140
- Decision Tree Classifier (“dt”) 0.6224 0.7600 0.6212 0.6257 0.0200
Medium Offices
Non-machine learning
- Logistic regression 0.1184 0.7836 0.2076 0.2671 9.5620
Machine learning
- SVM 0.1357 0.0000 0.2196 0.3107 0.6180
- Naïve Bayes 0.3904 0.7114 0.3807 0.3748 0.4420
- Gradient Boosting Classifier 0.7557 0.9322 0.7672 0.7576 40.6560
- Light Gradient Boosting Machine 0.7607 0.9342 0.7690 0.7623 2.5680
- Extreme Gradient Boosting 0.7572 0.9324 0.7655 0.7586 132.3740
- Catboost Classifier 0.7612 0.9344 0.7688 0.7629 33.2160
- Random Forest Classifier 0.7552 0.9318 0.7662 0.7573 4.7200
- Decision Tree Classifier 0.6565 0.7793 0.6663 0.6568 0.7940
Selecting the model: Performance and Efficiency. Following the result from Table
3, this study compares the five performance parameters to select the optimum model
for predicting the taxpayers’ ability to pay. In order to create comparable figures for
the five parameters, this study calculates the z-score for all parameters. Based on the
standardized scores for the eight machine learning algorithms, the “lightgbm” (Light
Gradient Boosting Machine) and “rf” (Random Forest Classifier) offer more balanced
performance and efficiency. While the “catboost” (Catboost Classifier) performs
relatively better than the “lightgbm” and “rf”, the run time is significantly higher: 8.24
seconds for Large Offices and 33.22 seconds for Medium Offices. As a comparison,
the “lightgbm” requires 0.20 (Large Offices) and 2.57 seconds (Medium Offices),
whereas the “rf” doubles the “lightgbm” run time to 0.41 seconds and 4.72 seconds,
respectively. Figure 3 compares the performance score for all algorithms.
The efficiency aspect, or running time, is highly significant during the scaling up
and deployment stage, especially when there are around 2 million corporate taxpayers
in the tax administration data. To illustrate, increasing the sample size from 241 (test
set, Large Offices) to 15495 (test set, Medium Offices), or around 64 times, requires
more than ten times the initial run time (11.4 times for “lightgbm” and 12.7 times for
“rf”). Using the back-of-the-envelope approach, scaling up the dataset to around two
million taxpayers would result in more than 300 times the execution time for “lightgbm”
and “rf” models.
Figure 1. Machine Learning-based Model Performance Comparison

Aside from performance measurement using the test and train dataset, this
research validates 120 sample taxpayers with the operational unit, i.e., the Account
Representative. On average, around 65% of the samples are correctly classified: 70.6%
for Medium Offices and 61.2% for Large Offices. Meanwhile, according to the Account
Representative, the survey results suggest that 38.8% of the corporate taxpayers in
Large Offices are classified as lower than the perceived ability to pay. The number is
relatively lower for Medium Offices. The under-classified taxpayers are around 17.6%
for Medium Offices, and about 11.8% were over-classified. In general, the survey-
based model performance, around 70% of correct classification, echoes the
quantitative performance. Based on further enquiries, the survey respondents
incorporated taxpayers’ willingness to pay when assessing the over and under-
classification.
5. CONCLUSION AND RECOMMENDATION

This study explores the taxpayers’ ability to pay concept using the existing 5 Cs
principle of credit analysis: Character, Capacity, Capital, Condition, and Collateral.
Several machine learning models were developed, using the taxpayers’ current
assets as proxies for the ability to pay – limited to the large and medium-taxpayers
group. The machine learning approaches provide robust performance compared to
the logistic regression. Among the eight machine learning algorithms, the Light
Gradient Boosting Machine and Random Forest produce a relatively balanced
performance and efficiency. Table 4 summarizes the taxpayers’ ability to pay for the
Large Offices and Medium Offices group.
Table 4. Taxpayers’ Ability to Pay Classification Result (Light Gradient Boosting

Machine Model)
Ability to Pay Large Offices Medium Offices
Very Low 283 (29.42%) 16,285 (26.28%)
Low 203 (21.1%) 13,597 (21.94%)
Moderate 255 (26.51%) 14,355 (23.16%)
High 111 (11.54%) 9,572 (15.44%)
Very High 110 (11.43%) 8,166 (13.18%)
Nevertheless, as an initial study assessing the taxpayers’ ability to pay, there are
several challenges in the model development and output. First, increasing the
dimensionality or variables from the administrative and external data sources will be
favorable. For example, the survey result introduced the qualitative variable of
taxpayers’ willingness to pay, which will be a relevant variable for taxpayers’
Character. One possible proxy for the variable is the taxpayers’ historical payment for
tax assessments or taxpayers’ delinquency records. Also, it is critical to include the
Collateral principle in assessing the taxpayers’ ability to pay, mainly using external
data sources rather than the tax return information.
Lastly, improved dimensionality and sample are instrumental in improving the
model’s performance. With around 0.7 accuracies, it would be plausible to fine-tune
the models to achieve higher classification performance. Besides, it is crucial to test
the model with the taxpayers administered in the small tax offices, which will bring
more challenges of more observations and limited data availability. Future research
on taxpayers’ ability to pay needs to incorporate more comprehensive observation and
historical data, allowing the machine algorithms to “learn” distinct features related to
the taxpayers’ ability to pay. Also, if data and information are available, it is relevant
to introduce the ability to pay concept for individual taxpayers.
Moreover, exploring the possibility of employing an “ensemble” approach,
combining several machine learning algorithms, is essential to create a more robust
prediction model – while accounting for the efficiency aspect. Arguably, the taxpayers’
ability to pay could be a critical complementary feature in the DGT taxpayers’
assessment and supervisory functions. However, it would require extensive model
development and training to unlock its optimum capability.
REFERENCES
Alibhai, S., Bakker, E., Balasubramanian, T. V., Bharadva, K., Chaudhry, A., Coetsee,
D., . . . van der Merwe, M. (2020). Interpretation and Application for IFRS
Standards. Somerset: John Wiley & Sons, Inc.
Bellotti, A., & Crook, J. (2009). Support Vector Machines for Credit Scoring and
Discovery for Significant Features. Expert System with Applications 36, 3302-
3308.
Biau, G., & Scornet, E. (2016). A random forest guided tour. Test, 25(2), 197-227.
Blaise, G., & Bilardello, J. (2005). Fundamentals of Corporate Credit Analysis. New
York: McGraw-Hill.
Burges, C.J. A Tutorial on Support Vector Machines for Pattern Recognition. Data
Mining and Knowledge Discovery 2, 121–167 (1998).
https://doi.org/10.1023/A:1009715923555
Davenport, T. H., & Harris, J. G. (2007). Competing on Analytics: The New Science of
Winning. Cambridge, MA: Harvard Business Press.
de Mooij, R., Nazara, S., & Toro, J. (2018). Implementing a Medium-Term Revenue
Strategy. In L. E. Breuer, J. Guajardo, & T. Kinda (Eds.), Realizing Indonesia’s
Economic Potential (pp. 109-140). Washington, DC: International Monetary
Fund.
Directorate General of Taxes. (2016). Annual Report 2015. Jakarta: Directorate
General of Taxes.
General of Taxes.
General of Taxes.
Golin, J., & Delhaise, P. (2013). The Bank Credit Analysis Handbook: A Guide for
Analysts, Bankers and Investors. Singapore: John Wiley & Sons Singapore Pte.
Ltd.
Gruber, J. (2011). Public Finance and Public Policy Third Edition. New York: Worth
Publishers.
International Monetary Fund. (2019). Indonesia Selected Issues: IMF Country Report
No. 19/251. Washington, D.C.: International Monetary Fund.
Ke, G., Meng, Q., Finley, T., Wang, T., Chen, W., Ma, W., ... & Liu, T. Y. (2017).
Lightgbm: A highly efficient Gradient boosting decision tree. Advances in
neural information processing systems, 30.
Khandani, A. E., Kim, A. J., & Lo, A. W. (2010). Consumer credit-risk models via
machine-learning algorithms. Journal of Banking & Finance 34(11), 2767-2787.
Liu, H., Hussain, F., Tan, C. L., & Dash, M. (2002). Discretization: An Enabling
Technique. Data Mining and Knowledge Discovery 6, 393–423.
Musgrave, R. A. (1996). Progressive Taxation, Equity, and Tax Design. In J. Slemrod
(Ed.), Tax Progressivity and Income Inequality (pp. 341-356). Cambridge:
Cambridge University Press.
OECD. (2006). Strengthening Tax Audit Capabilities: General Principles and
Approaches. Paris: OECD Publishing.
OECD. (2016). Advanced Analytics for Better Tax Administration: Putting Data to
Work. Paris: OECD Publishing.
OECD. (2016). Technologies for Better Tax Administration: A Practical Guide for
Revenue Bodies. Paris: OECD Publishing.
OECD. (2019). Tax Administration 2019: Comparative Information on OECD and other
Advanced and Emerging Economies. Paris: OECD Publishing.
Provenzano, A. R., Trifirò, D., Datteo, A., Giada, L., Jean, N., Riciputi, A., ... & Nordio,
C. (2020). Machine learning approach for credit scoring. arXiv preprint
arXiv:2008.01687.
Rish, I. (2001, August). An empirical study of the naive Bayes classifier. In IJCAI 2001
workshop on empirical methods in artificial intelligence (Vol. 3, No. 22, pp. 41-
46).
Slemrod, J. (1996). Tax Progressivity and Income Inequality. (J. Slemrod, Ed.)
Cambridge: Cambridge University Press.
Smith, A. (1776). An Inquiry Into the Nature and Causes of the Wealth of Nations
(2008 edition). London: Oxford University Press.
van Liebergen, B. (2017). Machine Learning: A Revolution in Risk Management and
Compliance? Journal of Financial Transformation 45, 60-67.
FINAL TAX REGRESSION
Muhamad Indrawan Yudha Prawiraa, Leo Fransb, Rezki Destianac

a Directorate General of Taxes, Jakarta, Indonesia Email: muhamad.prawira@pajak.go.id
b Directorate General of Taxes, Jakarta, Indonesia Email: leo.frans@pajak.go.id
c Directorate General of Taxes, Jakarta, Indonesia Email: rezki.destiana@pajak.go.id
ABSTRACT
The aim of this research was to determine the progressivity pattern of final income tax and its impact
on the personal income tax system which is currently on progressive form. The result shows that
final income tax distracts the progressivity pattern of income tax, in fact, it tends to be regressive for
some groups of taxpayers.
Keywords: final income tax, progressivity, regressive, income distribution, inequality, tax rates
Penelitian ini bertujuan untuk mengetahui bagaimana pola progresivitas pajak penghasilan final dan
dampaknya terhadap sistem pemajakan penghasilan orang pribadi yang saat ini bersifat progresif.
Hasil analisis menunjukkan bahwa pajak penghasilan final ini membuat distraksi terhadap pola
progresivitas pajak penghasilan, bahkan cenderung regresif bagi beberapa golongan wajib pajak
tertentu.
Kata kunci: PPh final, progresivitas, regresif, distribusi penghasilan
1. INTRODUCTION
The Income Tax Law (UU PPh) emphasizes that the tax system for personal
income tax (PPh OP) in Indonesia is progressive tax system. It is shown by
incremental tax rates ranging from 5 to 30 percent. Even in the Law on Harmonization
of Tax Regulations (UU HPP), the highest rate was increased to 35 percent.
For most people, this progressive tax system is claimed to be the fairest system
(compared to regressive, flat, and proportional systems) because it is considered
capable of easing the tax burden of those with low income. Wealthy people with higher
incomes are considered to have the extra financial power to pay higher taxes.
Whereas for those with mediocre income, most of the income received is generally
spent on basic needs and survival.
Even so, some think that the progressive tax system is unfair because it treats the
rich and the poor differently. Moreover, this system is also proven to influence tax
evasion35 behavior in Indonesia as evidenced in the research of Estri & Djamaluddin
(2019).
Aside from these pros and cons, the authors want to examine the facts regarding
the PPh OP tax system practice in Indonesia, which should be progressive instead of
proportional, or even regressive. It can be seen by examining the individual taxpayers’
reported income structure and the amount of tax paid on their Annual Income Tax
35
Tax evasion is an effort to reduce taxes by violating tax regulations (Suandy, 2014)
Return (SPT Tahunan). The initial assumption is that those income and taxes have
been fully reported.
A survey conducted by Forbes magazine in 2021 (table attached) ranks the 50
richest people in Indonesia. The survey results revealed that the 50 wealthiest people
have net worth or net assets of around IDR 9.45 trillion to IDR 610.89 trillion (Forbes,
2021). Being taxpayers with top income among the population of Individual Taxpayers
(WP OP), the assumption that they are also supposed to pay taxes with the highest
rate needs further analysis. Apart from that, their income structure also needs to be
analyzed to see the alignment between the taxes paid and the level of income earned
for both income subject to final and non-final income tax.
Kristiaji and Awwaliatul (2020) in one of their writings have stated that the
application of final income tax is one of the solutions for applying presumptive tax36. In
addition, the tax mechanism is also considered simple and easy to implement by
taxpayers and tax officers. In his writings, James Alm (2004) stated that this
presumptive tax is generally applied so that the tax authorities can deal with hard-to-
tax problems.
Ahmad and Stern (1991) also support the argument that the tax base used in
presumptive tax is obtained from indicators that are simpler than the tax base in
general. The term presumptive itself indicates a legal assumption that the tax base is
taxed not lower than the assumed amount (Bongwa, 2009).
On the other hand, the IMF (1996), in one of its studies, revealed that presumptive
taxation was applied for several reasons, including relieving the burden of compliance
and administration for taxpayers with relatively low turnover. In addition, the IMF also
said that tax avoidance or evasion could be overcome by presumptive taxes. Lastly,
the system is believed to promote equity in the distribution of the tax burden when
conventional approaches are unreliable due to corruption in taxpayer compliance and
administration.
However, despite the series of benefits stated by the IMF, there are potential
problems arising. One of them is related to the administrative burden, which is when
the taxpayer has high turnover, is this benefit still relevant and fair if the tax rate
charged is flat rate? Furthermore, regarding the second benefit, when the effective tax
ratio for taxpayers subject to Final Income Tax is lower than non-final income tax, will
this provide incentives for taxpayers to commit tax avoidance by opting for income
subject to Final Income Tax? According to Allingham-Sandmo-Model (1972), the tax
rate is one of the most influencing factors for non-compliant taxpayer behavior.
It should be remembered that the spirit of the PPh OP tax system is related to the
budgetary function to finance national development and the redistribution of income,
the ultimate goal of which is to reduce social inequality. Therefore, Article 17 of UU
PPh states that the PPh OP rate was set to be progressive. Moreover, the UU HPP
emphasizes the progressive nature of PPh OP taxation by expanding the coverage of
36
presumptive tax is a tax calculation method using the indirect estimation method to calculate the amount of the taxpayer's
tax liability (IMF, 1996)
the lowest bracket to IDR 60 million and increasing the top marginal tax rate by 5
percent to 35 percent.
Meanwhile, the application of the final tax rate is regulated and reported separately
from the taxpayer’s income, which is subject to the general rate. For example,
dividends received/earned by individual domestic taxpayers are subject to income tax
of 10 percent and are final. This rate does not differentiate the income earners, nor
does it consider the size of the dividends received. For example, the Final Income Tax
rate for renting land and/or buildings, for both individuals and corporate taxpayers, is
set at the same rate, 10 percent of the gross amount of the rental value. Since the
range of the rates is small or tends to be fixed, it is causing the application of the final
rate to somewhat deviate from the spirit of progressive PPh OP taxation.
It is also suspected to cause the gap in socio-economic inequality in society to
widen. The rich can use this regulation as a loophole to reduce their tax burden by
increasing income proportion subject to the final rate. Thus, they can avoid the
maximum progressive rate of 35%. Meanwhile, those who only earn income from their
employers are busy being taxed with progressive income rates.
In testing the structure of income and taxes paid by individual taxpayers, we
examine the population of the 2018-2020 SPT Tahunan PPh OP data filed by the
taxpayers. The data is then processed using the STATA application to test the WP
OP's ETR (Effective Tax Rate) value of all income levels, final and non-final. It is done
to see the income distribution and to test the taxpayers' strategy in "allocating their tax
burden."
According to our hypothesis, the effective tax burden of "super-rich" WP OPs could
be slightly under the "rich" WP OPs in percentage level or even far lower. This may
happen because of the indiscriminate nature of Final Income Tax and is a distraction.
If proven, it may be necessary to revisit the final tax formulation to make it more
appropriate.
Based on the 2020 SPT Tahunan database alone, for example, there are
12,193,963 WP OPs who have filed their SPT Tahunan. From this population data,
19,052 were excluded because their ETRs were below 0 and/or above 1. Then, the
authors used the 3x Standard Deviation method to detect the presence of other data
outliers. That is, if there is WP OP in the population reporting total income of more
than the standard deviation value multiplied by 3, then the data is considered outliers.
With this method, as many as 33 data have been removed from the population
because they have an annual income of over IDR 7.8 trillion.
It should also be conveyed that there is a potential for miscalculation of taxes
reported by taxpayers. 57,223 WP OP reported final taxes with rates above 20%, and
151 WP reported non-final taxes with rates above 30%. Those two final and non-final
tax rates are the highest, so when there is a calculation result that exceeds these two
rates, there is a potential for miscalculation by the taxpayers.

4.1. Income Layer Distribution
Table 1. The Layer of Income based on the Income Tax Law
Layer # Range of Taxable Income Rates

1 0 s.d. Rp50 million 5%
2 >Rp50 million s.d. Rp250 million 15%
3 >Rp250 million s.d. Rp500 million 25%
4 >Rp500 million 30%
Source: Article 17 UU PPh, processed
As long as the effective date of the UU HPP has yet to come into effect, the
provisions as stipulated in the UU PPh above are still effective in reporting 2021 SPT
Tahunan. In the UU PPh, there are four layers of PPh OP’s tax rate which we can see
in the table above, ranging from 5 to 30 percent. Of course, this layer only applies to
non-final income received by taxpayers, not including final income. The population
distribution of taxpayers based on the provisions in force above is as follows.
Graph 1. Distribution and Contribution of Non-Final Income Tax

Source: 2020 SPT Tahunan WP OP, processed
Looking at the distribution of the number of WP OP reported in the 2020 SPT

Tahunan, the population based on the PPh OP’s tax rate layers tends to be skewed
to the right. It can happen if the WP's average income is greater than the median and
the median is greater than the modes gathered at the first layer or the taxable income
layer of up to IDR 50 million.
The amount of income tax payable indicates the opposite phenomenon and the
pareto principle applied. WP OP contribution by income layer tends to be skewed to
the left. It means that the smallest number of taxpayers gives the contribution of
income tax owed in the population on the right of the graph. However, remember that
income tax is still tax payable by taxpayers from non-final sources. To see the shape
of the distribution as a whole will be discussed in the next section.
Table 2. Types of Final Income Tax Rates based on the Income Tax Law
Rates Final Income
0-1% MSMEs, Stock Transactions
>1 s.d. 5% Derivative Transactions, Construction, Income Tax on the Transfer of

Rights on Land and Buildings
>5 s.d. 10% Rentals, Construction, Dividends, Savings, Cooperative Interest
>10 s.d. 20%37 Deposits, SBI Discounts, Current Account Services, Deposit Interest
Source: PPh Law, processed
To provide an overview of the income layers subject to Final Income Tax rates, a
rate classification is carried out based on four layers, namely 0 – 1% (layer
1), >1% – 5% (layer 2), >5% – 10% (layer 3). ), >10% – 20% (layer 4). Based on this
classification, an analysis of the distribution of taxpayers who have income subject to
final income tax is carried out. The results of the analysis can be seen in the graph
below.
Graph 2. Distribution and Contribution of Final Income Tax

In line with the pattern of non-final income, a similar thing also occurs in the
distribution of income subject to Final Income Tax with a certain layer. The most
striking difference between the contribution of non-final PPh and Final PPh is that the
contribution of tax revenue at the highest tariff layer Non-final PPh is the income layer
with a contribution value of around Rp. rates > 10% – 20%. It can indicate that in the
upper-income layer, there is regressiveness of tax imposition on the upper percentile
layer when the taxpayer has both non-final and final income.
37
There is also a Final Income Tax rate that is more than 20%, namely final Income Tax on income originating from prizes or
sweepstakes. The rate is charged at 25%. However, it is not included in the analysis considering it is a casuistic income.
4.2. ETR by Income Percentile
In the figure below, the average ETR per percentile of non-final income increases
as the total income of the taxpayer increases. The ETR itself is obtained by dividing
the tax that has been owed by the income received by each taxpayer for a year. For
the below case, ETR is obtained by dividing non-final tax by non-final income. From
the graph, for the non-final tax itself, our tax system is indeed progressive, provided
that the taxpayer has no other income besides the non-final income itself.
Graph 3. Average of Non-Final ETR based on Non-Final Income Percentile

Furthermore, Graph 4 displays the average ETR for each percentile of taxpayers
who have income subject to Final Income Tax. The graph shows that as much as 80%
of taxpayers have no income subject to final income tax. In addition, it can also be
concluded that there is no ETR progressivity toward higher-income taxpayers. Thus,
when taxpayers have income subject to final income tax, they tend to benefit from the
final income tax rate compared to taxpayers who do not have final income.
Graph 4. Average Final Effective Tax Rate by Final Income Percentile
However, it should be borne in mind that the income structure of taxpayers comes
not only from labor income38 typically taxed or capital income39 generally subject to a
final tax system but a combination of the two. Therefore, the income percentile must
be based on the total income received by each taxpayer during the year, regardless
of whether it comes from labor or capital income. Thus, this time the ETR is obtained
by dividing all taxes payable (final and non-final) by all income received (labor and
capital).
38
Revenue from work
39
Passive income that comes from investment or placement of other funds outside of employment
Graph 5. Average Non-Final Effective Tax Rate based on Total Income Percentile
So, after conducting a two-sided test (final plus non-final), this analysis results
show that Indonesia's taxation system is not fully progressive. For the lower percentile,
the ETR is higher than some of the percentiles above it. On the other hand, the top-
three percentile has a lower ETR than the several percentiles below them. The
question is, why is the percentile distribution this time different from the previous
percentile? It can happen because, in total, the percentile group members may differ
due to differences in income structure. Thus, taxpayers can move with an income
structure with a more significant proportion of their final income than their non-final
income to move up to the percentile class.
Likewise, from the calculation of the average rate below, it can be seen that the
average ETR of final income is greater than the ETR of non-final income. This can be
interpreted that in general, taxpayers who have income subject to Final Income Tax
have a higher ETR when compared to taxpayers who are subject to non-final Income
Tax. However, from the previous graph and the Final Income Tax distribution graph,
it is known that there is no progression in the Final Income Tax rate. Thus, when
taxpayers with very large or super-rich incomes have income that is dominated by
income subject to Final Income Tax, they will benefit because their ETR will tend to
be lower.
Table 3. Mean Differences of Final and Non-Final Income

Graph 6. WP Income Structure based on Total Income Percentile
Let us look at the income composition in graph 6 and compare it to graph 7.

Starting from the 79th percentile and above, final income holdings are steadily
increasing. It even outnumbered his non-final earnings in the 99th and 100th
percentile. However, compared to the ETR in that group, it is declining. Its role is fragile
to be able to increase the total ETR amount. This pattern has been consistent for the
last three years (2018-2020). Based on this insight, the nature of final taxation is
regressive for the upper class (especially the highest three percentiles), and it is not
even an exaggeration to say that it can create a super-rich class of society.
Graph 7. Mean of Effective Tax Rate based on Total Income Percentile
The next question is, what kind of capital income makes the top-ten percentile very
rich? In the 100th percentile, the income from the sale of shares is the primary source
of their wealth. However, what is intriguing is the composition of the other final gross
income, which is consistently in all percentile layers. It was initially used for MSME
reporting. Therefore, in this regard, further research is needed to determine whether
formally the taxpayer has been correct and/or honest in his reporting.
Graph 8. Composition of Final Income in the 10 Highest Percentiles

4.3. Tax Administration for High-Wealth Taxpayers

Currently, WP OPs with certain incomes are monitored by three KPP types: KPP
HWI, KPP Middle throughout Indonesia, and KPP Entities and Foreign Individuals.
Based on the aggregate data for the three types of KPP in 2020, apart from Badora
KPP, the income structure of WP OP is generally dominated by income subject to final
rates. In contrast, the PPh WP OP structure at the KPP is generally supported by non-
final PPh. It is in line with the research results in the previous subheading on the WP
OP population throughout Indonesia.
Graph 9. Income structure of WP OP in the selected KPP

Graph 10. PPh OP structure in the selected KPP
The table below shows that the WP OP ETR at KPP Badora, HWI, and Madya are
24.8%, 11%, and 9.5%. This ETR is much lower than the highest Article 17 Income
Tax Rate in 2020, which is 30%. It is due to the low final PPh rate, especially the Final
PPh rate other than deposits.
Table 4. Mean Effective Tax Rate of PPh OP at Selected KPP
Compared to the average National Non-Final ETR of 4.03%, the Non-Final ETR in
this selected KPP is much higher. Likewise, with the final taxation, the ETR of this
selected KPP is still much higher than the average National Final ETR of 5.2%. It may
be because high-wealth individuals from all over Indonesia are gathered in this KPP
category. However, this also means that the existence of the HWI, Badora, and Middle
KPP types is administratively appropriate to handle wealthy WP OPs in Indonesia.
Table 5. Mean Difference between Final and Non-Final Income

4.4. Final Tax Treatment

In the United States, the tax rate paid on capital income or realized capital gains
depends on the taxpayer's total income and the time the taxpayer holds the asset
before selling it. If he sells the asset for less than one year of ownership, the profit will
be calculated as ordinary income and taxed based on a progressive income tax group.
Gains earned on assets sold after a more extended period of ownership are
considered long-term capital gains and are taxed separately at a lower rate.
For cases in Indonesia, Final Income Tax is not only subject to capital income
which tends to be passive. Many Final Income Taxes are subject to active income,
such as construction services or MSME Income Tax. So, it is appropriate that they are
subject to non-final taxation because the taxation is no longer an incentive for them to
invest.
The conditions put forward by the IMF for final taxation as mentioned above,
namely easing the burden of compliance and administration of taxpayers who have
low turnover, only apply to Micro Enterprises, excluding Small and Medium
Enterprises, so that the maximum limit for taxpayers who can enjoy MSME is
appropriate. To be taken down.
Furthermore, the reason for final taxation to combat tax avoidance or tax evasion
is irrelevant to business conditions in Indonesia. Final taxation understates the
meaning of operating profits and expenses. Taxpayers that are included in the final
regime, both WP OPs and agencies generally have similar advantages, so they no
longer need to evasion as long as they can control the reported turnover value.
As for the final condition of final taxation to promote fairness in the distribution of
the tax burden, when the general approach is unreliable because there is corruption
in taxpayer compliance and administration, it is likely to occur in Indonesia. However,
it can only be applied to some types of final income. Of the various types of income
subject to final tax, only construction services partially meet this requirement. Why
partly because it is still unfair for entrepreneurs in other sectors.
The classic reason raised for final taxation on construction services should be
corrected by eradicating corruption or by using the services of an independent third
party, for example, to assess the operating margin of this business activity. In this
modern and open era, it is fitting for the public to see how the actual condition of
construction service business kitchens is operating in Indonesia.
At least the authors formulate two kinds of approaches that can be taken to
gradually remove the final taxation system, especially for WP OPs. The first is to
equate the Final Income Tax rate with the ETR value of the taxpayer that has been
reported in the SPT Tahunan for the previous Tax Year (n-1). For example, in 2020, a
person reports his income with a ratio of "PPh: Net Income" alias ETR of 17%. Every
final income he received in 2021 is subject to Final Income Tax, equivalent to his non-
final income, which is 17%. This technique can eliminate final taxation distortions so
that individual income tax remains progressive. In addition, the state will undoubtedly
enjoy positive externalities in the form of increased tax revenues.
The second approach that can be taken if the previous approach has worked well
is annualizing every income received, both final and non-final. This method will tax
each person according to his income during the year. There is no longer a gap
between passive income and active income. To realize this method requires
technological readiness and transparent law enforcement. It is because there is a lag
between the income received and when he reports his income. The number of stock
transactions, for example, must be able to be recorded neatly and in detail.
Furthermore, the AR function, as stipulated in PMK 45/2021 in carrying out
supervision of taxpayer compliance, is very vital. He must be able to oversee WP as
a whole and detect transactions that are not only carried out physically or in
cyberspace.
By changing the final taxation system approach, the authors hope to restore the
OP PPh taxation system to maintain its progress and be fair for all parties.
5. CONCLUSION
Based on the analysis that has been done before, based on testing of the top
percentile of taxpayers with the highest income, the results show a regressiveness of
the ETR of all taxpayers' income. One of the contributors to this regression is the
income component subject to Final Income Tax. This conclusion is supported by
income data for HWI taxpayers, whose income is dominated by income subject to
Final Income Tax. However, researchers see that the risk of filling in the SPT Tahunan
by taxpayers is still inherent.
From the conclusions above, it is necessary to consider the imposition of income
rates subject to Final Income Tax that are dynamic or progressive following the level
of income received so that there is an increase in ETR comparable to Taxpayers
whose income is subject to non-final Income Tax. Apart from that, DGT also needs to
develop a prepopulated filling system for income subject to Final Income Tax to reduce
filling errors.
All information and results of this analysis can be disseminated widely to obtain
other academic and regulatory views, which play a role in improving the taxation
system in Indonesia, primarily related to PPh OP.

Any views, ideas, and/or ideas contained in this paper are not representative of
the policies issued by the author's place of work - the Head Office of the Directorate
General of Taxes. However, they are solely the professional responsibility of the
author.
In conducting this research, the authors have various limitations, including that
limited data access causes the SPT data used in the research to only come from a
few tax years, namely 2018, 2019, and 2020 tax years, years in which the HPP Law
has not been effectively applied. What is more, the community's economic condition
was unstable at this time due to the pandemic. However, the tax year used is the most
recent in the DGT tax database. In addition, the authors have made efforts so that
data consistency is not an issue in this study.
Furthermore, the author also needs to remind us that the data used in this paper
is the WP OP's self-assessment version outlined in the SPT Tahunan, so the tax office
has done no research or inspection. It means there may be typos, wrong numbers,
wrong columns, or wrong application of tax law. Originality is key to source data and
can have positive implications for research.
Finally, in formulating two approaches that can be taken to eliminate the final
taxation system, the authors have yet to conduct research or simulations due to time
and space limitations in this study. For this reason, the authors will conduct further
research separately in less time since this research was completed.
Acknowledgment
We thank Mr. Surya Adi Setiawan, S.S.T., Ak., M.S.E., MPP, as superiors who always support
us in carrying out analyzes and providing freedom of expression, as well as Himawan Saputro,
Ph.D., for his guidance in conducting data analysis.
BIBLIOGRAPHY
Ahmad, E., & Stern, N. (1991). The Theory and Practice of Tax Reform in Developing
Countries. Cambridge University Press.
https://EconPapers.repec.org/RePEc:cup:cbooks:9780521265638
Alm, J., Martinez-Vazquez, J., & Schneider, F. (2004). ‘Sizing’the Problem of the Hard-
to-Tax. Contributions to Economic Analysis, 268, 11-75.
Allingham, M. G., & Sandmo, A. (1972). Income tax evasion: A Theoritical analysis.
Journal of Modern Accounting and Auditing, 1, 323–338.
Bongwa, A. (2009). Managing Ethiopian Cities II: Informality in Ethiopia: Taxing the
Hard to Tax (No. IHS WP 22). Institute for Housing and Urban Development
Studies (IHS). http://hdl.handle.net/1765/32181
Dauchy, Navarro-Sanchez, Seegert (2021) Taxation and Inequality: Active and
Passive Channels. Review of Economic Dynamics (Volume 42). 2021
Estri & Djamaluddin (2019). Does the Progressive Personal Income Tax Drives Tax
Evasion in Indonesia? The 3rd International Conference on Accounting, Business,
& Economics (UII-ICABE 2019).
Forbes. (2021). Indonesia’s 50 Richest. https://www.forbes.com/indonesia-
billionaires/list/
IMF (1996). Tax Law Design and Drafting. DDTC Working Paper 2220.
https://www.imf.org/external/pubs/nft/1998/tlaw/eng/ch12.pdf
Kristiaji, B. B., & Mukarromah, A. (2020). Meninjau Konsep dan Relevansi PPh Final
di Indonesia. DDTC Working Paper.
https://ddtc.co.id/research/publications/working-paper/meninjau-konsep-dan-
relevansi-pph-final-di-indonesia/
IBFD (2022). OECD Glossary of Tax Terms.
https://www.oecd.org/ctp/glossaryoftaxterms.htm
Undang-Undang Republik Indonesia Nomor 36 Tahun 2008 tentang Perubahan
Keempat atas Undang-Undang Nomor 7 Tahun 1983 Tentang Pajak Penghasilan
Undang-Undang Republik Indonesia Nomor 7 Tahun 2021 tentang Harmonisasi
Peraturan Perpajakan
Peraturan Pemerintah Nomor 51 Tahun 2008 tentang Pajak Penghasilan dari
Penghasilan Usaha Jasa Konstruksi
Peraturan Kementerian Keuangan 45/PMK.03/2021 tentang Account Representative
pada Kantor Pelayanan Pajak.
Suandy, Erly (2008). Perencanaan Pajak (ed. 4) HVS. Jakarta: Penerbit Salemba.
Thuronyi, Victor (1996). Tax Law Design and Drafting (Volume 1; International
Monetary Fund: 1996)
ATTACHMENT 1
List of the 50 Richest People in Indonesia in 2021
Source: Forbes, processed

ATTACHMENT 2
Do-File
/*GENERATE VARIABEL BARU: TOTAL PENGHASILAN DAN TOTAL PPH
(FINAL DAN NON-FINAL)*/
gen totalpengh = jml_ph_neto + jml_dpp_ph_final
gen totalpph = jml_pph_terutang + jml_pph_final
gen etr = totalpph / totalpengh
gen inc_mil = totalpengh / 1000000
replace jml_ph_neto = 0 if jml_ph_neto ==.
replace jml_dpp_ph_final = 0 if jml_dpp_ph_final ==.
replace etr=0 if etr==.
drop if jml_ph_neto < 0
drop if jml_dpp_ph_final < 0
gen etrnonfinal = jml_pph_terutang / jml_ph_neto
gen etrfinal = jml_pph_final / jml_dpp_ph_final
drop if etrnonfinal >=1
drop if etrfinal >=1
drop if etr >=1
/*DISTRIBUSI PENGHASILAN BY LAYER UU PPH*/

gen layer = 1 if jml_pkp <= 50000000
replace layer = 2 if jml_pkp > 50000000
tab layer
total jml_pph_terutang, over (layer)
/*DISTRIBUSI XTILE*/
gen income_class=.
replace income_class= 20 if jml_ph_neto > 5000000000
replace income_class= 19 if jml_ph_neto <= 5000000000
total jml_pph_final, over (income_class)
xtile decile_pengh = totalpengh, nq(10)
xtile pctile_pengh = totalpengh, nq(100)
replace pctile_pengh=1 if pctile_pengh==.
xtile pctile_pengh_nonfinal = jml_ph_neto, nq(100)
replace pctile_pengh_nonfinal=1 if pctile_pengh_nonfinal==.
/*sebaran ETR per lapisan penghasilan maupun percentile penghasilan*/

tabstat inc_mil, by(pctile_pengh) s(n mean sd min max)
tabstat etr, by(pctile_pengh) s(n mean sd min max)
*ETR lapisan non-Final*
tabstat etrnonfinal, by( pctile_pengh_nonfinal ) s(n mean sd min max)
tabstat etrnonfinal, by( pctile_pengh ) s(n mean sd min max)
*ETR lapisan Final*
tabstat etrfinal, by( pctile_pengh ) s(n mean sd min max)
*ETR all lapisan*
tabstat etr, by( pctile_pengh ) s(n mean sd min max)
/*buat diagram ETR per percentile*/

preserve
collapse (mean) etr, by (pctile_pengh)
graph twoway (connected etr pctile_pengh if etr<=1 & etr>=0, lwidth (thin) msymbol
(smcircle)), graphregion(fcolor(white))
restore
/*data buat diagram stacking final vs non-final per percentile*/

total jml_ph_neto jml_dpp_ph_final, over (pctile_pengh)
/*data buat diagram jenis final top percentile*/

total jml_ph_bruto_diskonto_sbi jml_ph_bruto_obligasi
jml_ph_bruto_penjualan_saham jml_ph_bruto_honorarium_apbn jml_ph_bruto_phtb
jml_ph_bruto_bangun_guna_serah jml_ph_bruto_sewa_tanah_bangunan
jml_ph_bruto_usaha_jaskon jml_ph_bruto_transaksi_derivatif jml_ph_bruto_dividen
jml_ph_bruto_lainnya if pctile_pengh >89, over ( pctile_pengh)
SIMILARITY INDEX FOR ESTIMATING TAX REFORM
PROGRAM AND ITS IMPACT ON TA REVENUE:
INDONESIA CASE STUDY
[Gede Satria Pujanggo PG.]a, [Dewa Made Cakrabuana Aristokra]b, [Ryan Mohammad]c
a Directorate General Taxation, Tabanan, Indonesia Email: gede.satriapg@gmail.com
b Directorate General Taxation, Tabanan, Indonesia Email: aristokra@gmail.com
c Directorate General Taxation, Jakarta, Indonesia Email: ryanmohammad1990@gmail.com
ABSTRACT
This study estimates the significance of tax reform on tax revenues, both in the short and in the long run. We
find it’s dificult to estimate tax reform due to the lack of related studies. We offer estimator of tax reform using
Similarity Index, a modification Bray-Curtis Dissimilarity Index’s. Similarity Index estimates how far the
coverage of Indonesia's tax structure compared to average developed countries. Furthermore, we measure
the impact of tax reform on tax revenues using modified Vector Autoregressive. We compute data of
Indonesian tax revenues, GDP, GDP per capita, inflation, and Indonesia's international trade within 40 years,
results show that tax reform variable has positif effect, both short and long term, on tax revenue.
Keywords: tax reform, tax revenue, Similarity Index, Vector Autoregressive
Penelitian ini mengestimasi signifikansi reformasi perpajakan terhadap penerimaan pajak. Penulis
mengalami kesulitan untuk mengestimasi materi tersebut karena kurangnya studi terkait. Untuk mengatasi
kesulitan menentukan proksi sebagai indikator yang mampu mendefinsikan elemen-elemen reformasi
perpajakan Indonesia secara holistik, penulis menawarkan estimator reformasi perpajakan menggunakan
Similarity Index, modifikasi sederhana dari Bray-Curtis Dissimilarity Index. Estimator ini mengestimasi sejauh
mana cakupan struktur pajak Indonesia dibanding rata-rata cakupan struktur pajak negara-negara maju.
Selanjutnya untuk mengukur dampak reformasi perpajakan pada penerimaan pajak, penulis menggunakan
metode Vector Autoregresive. Dengan menghimpun data penerimaan pajak Indonesia dan enam belas
negara maju, PDB, PDB per kapita, inflasi hingga ekspor dan impor Indonesia dalam jangka waktu 39 tahun,
hasil penelitian menunjukkan bahwa variabel reformasi perpajakan memiliki hubungan kasualitas, baik itu
jangka pendek maupun jangka panjang, terhadap variabel penerimaan pajak. Penerimaan pajak dapat
menikmati dampak positif dari reformasi perpajakan untuk kurun waktu yang cukup lama.
Kata kunci: reformasi perpajakan, penerimaan pajak, Similarity Index, Vektor Auto Regresif
1. INTRODUCTION
1.1. Background
The rise of uncertainty in global economy caused by COVID-19 pandemic have
affecting tax revenue a lot of countries, which is an important concern for the
government to keep it stable in the middle of economic activities decline. Even though
government have alternative nation income, such as treasury bonds, tax revenue still
nation main income for government in the long term.
However, government attempt for reaching optimal tax revenue is difficult for
developing countries, Indonesia included. Therefore, Indonesia Directorate General
Taxation (DGT) has implemented tax reform program for raising the effectivenes of
tax revenue colleting that has been conducted for over past 40 year until now.
Currently, Indonesia DGT developing their core system tax administration for
establishing comprehensive reform program so taxation system become easier,
integrated, reliable, accurate, and certain. Based on presidential regulation number 40
year 2018, those renewal tax core system will redesign a significant amoung of tax
administration business process through Commercial Off-the-Shelf (COTS) and tax
data improvement.
On renewal tax core system opening (Jakarta, 11/18/2020), Indonesia Ministry of
Finance, Sri Mulyani, state the main goal of tax reform is the significant growth of tax
revenue and tax ratio with GDP. Sri Mulyani even point out the minimum of growth
atleast double from current tax revenue and tax revenue with a measurable indicator.
Silvani and Baer (1997) explain that tax administration reform is a part of fiscal
reform for aiming macroecnomic stability and restructuring taxation so it become more
efficient, not distorted easily by market, and easier for administration. Indonesia has
implemented tax reform program from 1983, but is it affecting positively for tax revenue?
Eka (2019) concluded that tax reform administration in Indonesia affecting tax revenue
negatively because after tax reform program have been implemented, the amount tax
revenue potential is not affecting tax revenue collected, which indicate there is
problem in DGT productivity.
We have not found empirical studies explaining the corelation and impact of tax
reform program to tax revenue. We also have not yet found study that explaining the
reliable estimator for tax reform program in Indonesia. In this paper, we attempt a
model for estimating tax reform so we could evaluate the impact of tax reform program
to tax revenue in Indonesia.
2. THEORITICAL FRAMEWORK
Here are several definition of tax reform program:
- Government Law Number 7 year 2021 regarding Tax Regulation
Harmonization explain that for increasing sustainable economic growth and
accelerating economy recovery, government need fiscal consolidation strategy
that focus on improving nation budget deficit and tax ratio. Those strategy
including: the implementation of: tax collecting performance, tax reform
program, tax base expansion, designing taxation system that putting first on
equality and law certainty, and increasing taxpayers voluntary compliance. In
attempt of improving tax ratio, the Tax Regulation Harmonization elaborate that
government need conduct several effort, such as tax reform program for
organization, human resource, technology information based data, business
process, and tax regulation.
- OECD (2010) define tax reform as program for change one or more taxation
or taxation system, so the function could be improved, such as increasing tax
revenue, expanding tax base, better taxpayer compliance, and it could
contributed to fiscal consolidation for better equality on taxation system.
- Satya (2017) explain that tax reform program is changing taxation system
significantly and comprehensively for improving tax administration, tax
regulation and tax base so tax revenue will growth stronger. One of major aim
for tax reform in Indonesia is reaching tax ratio to 14% in 2020.
- Sinaga (2017) explain tax reform is conducted for elevating taxation system to
become more equal, comprehensive and valid data for expanding tax base,
and integrating short term and long term tax revenue in more sustainable way.
Tax reform is crucial because stronger, reliable, and accountable tax institution
is needed. Therefore, tax reform should restructure human resource, budget,
business process, information system, and infrastructure of tax institution and
tax regulation so DGT have capability for detecting tax potential and
manifesting it to become tax revenue efficiently and effectively.
Based on those definitions, we found there are four main element in tax reform
program, which are: (1) significant change on taxation structure or system, (2) proper
tax regulation, (3) improvement on tax administration, system, and procedure, and (4)
the main goal is to rise tax revenue and tax ratio.
Generally, DGT as Indonesia’s tax institution had been conducted four period of
tax reform program:
- In 1983, Indonesia change their taxation system from official assesment into
self assesment
- In 2002 to 2008, DGT’s tax reform focuses on improving human resource,
organization and business process. It resulted on modernization institutional
unit based on taxpayer segmentation which consist of Regional Tax Office,
Large Tax Office, Medium Tax Office, and Tax Office.
- In 2009 to 2016, DGT’s tax reform focuses on ease of business for facing
economic decline after global finance crisis.
- Lastly, from 2016 to this day, DGT’s tax reform focuses on five main pillar in
tax administration, which are organization reinforcement, increasing human
resource quality, improving business process, renewal information system and
database, and regulation refinement.
We found several studies attempt for learning the relation between tax reform
program and tax revenue. Kusi et al (1998) describe that government need to
implement taxation system that is responsive with economic growth because it has
high potential growing tax revenue without adding unpopulist fiscal policy such as
increasing tax rate in careless way. Taxation system also need to shift from income
tax, but also consumption tax. Based on Government act on Tax Harmonization, we
can see Indonesia has their concern on consumption tax by increasing Value Added
Tax (VAT) rate from 10% to 11%, which eventually rise to 12% approximately in 2025
and at the same time decreasing income tax rate. Gnagnon and Brun (2019)
concluded that tax reform program affecting positively to tax revenue. However, Eka
(2019) concluded that tax reform on tax administration affecting tax revenue negatively.
Based on those studies founding and theoritical framework, we hypothetizing that tax
reform program have positive impact on improving tax revenue in long term.
3.1. Data Selection and Collection
Our paper use two set of variabel. First, we estimate the value of tax reform
program in Indonesia. Second, we measure tax reform program impact on
Indonesia’s tax revenue. Lastly, we conduct coverage quantification.
We found several paper that have estimated tax reform program performance.
Desai and Hines (2003) attempt to evaluate United States international tax reform
program proposal with Capital Ownership Neutrality (CON) and National
Ownership Neutrality (NON) approach as benchmark. CON will be achievable if
all country not imposing tax for foreign income, so capital owners will not consider
the difference of tax rate in foreign and they will be encouraged to join the market
for alocating their capital to the place with highest productivity. Nation welfare
could be in highest level if there is no tax on domestic company foreign income
(NON) and global welfare could be maximal if foreign income tax of capital owner
around countries is aligned each other.
Rosen (1976) suggesting the way of evaluation taxation system changes in
United States based on how it impacted social welfare. Chan et al (1999) and
Sajadifar et al (2012) use General Equilibrium approach for evaluating tax reform
performance, spesifically on value added tax changes and its social welfare impact
in Iran and Vietnam.
For estimating the range of tax reform program in Indonesia, we attempt to
quantify it by compare Indonesia tax structure with developed countries tax
structure using Similarity Index. Similarity Index is developed from Bray-Curtis
Dissimilarity Index, a method used in ecology study for quantifyin sample
difference (Bray, Curtis, & Roger, 1957; Greenacre, 2018). Bray Curtis
Dissimilarity Index equation is presented as follow:
𝛴𝑗 |𝑥𝑗 − 𝑦𝑗 |
𝑑𝑥𝑦 =
𝛴𝑗 |𝑥𝑗 + 𝑦𝑗 |
(1)
X and Y are two different country sample and j is tax structure compenent from
those countries. Gnangnon (2019) mention that tax structure component could be
proxied using direct tax to GDP ratio, indirect tax to GDP ratio, and international
tax to GDP ratio, so we use those variabel for measuring tax structure compenents
on country sample. Gnangnon and Brun (2019) excluding tax revenue from nature
resource because it could be misleading indicator for policy perspective.
Therefore, Disimilarity Index of Indonesia’s tax reform in year t, 𝑑𝑖𝑡 , to
developed countries equation is presented as follow:
|𝐷𝑖𝑟𝑒𝑐𝑡 𝑇𝑎𝑥𝑖𝑡 −𝐷𝑖𝑟𝑒𝑐𝑡 𝑇𝑎𝑥𝑎𝑡 |+ |𝐼𝑛𝑑𝑖𝑟𝑒𝑐𝑡 𝑇𝑎𝑥𝑖𝑡 −𝐼𝑛𝑑𝑖𝑟𝑒𝑐𝑡 𝑇𝑎𝑥𝑎𝑡 |+ |𝐼𝑛𝑡 𝑇𝑟𝑑 𝑇𝑎𝑥𝑖𝑡 −𝐼𝑛𝑡 𝑇𝑟𝑑 𝑇𝑎𝑥𝑎𝑡 |
𝑑𝑖𝑡 = (𝐷𝑖𝑟𝑒𝑐𝑡 𝑇𝑎𝑥𝑖𝑡 −𝐷𝑖𝑟𝑒𝑐𝑡 𝑇𝑎𝑥𝑎𝑡 )+ (𝐼𝑛𝑑𝑖𝑟𝑒𝑐𝑡 𝑇𝑎𝑥𝑖𝑡 +𝐼𝑛𝑑𝑖𝑟𝑒𝑐𝑡 𝑇𝑎𝑥𝑎𝑡 )+(𝐼𝑛𝑡 𝑇𝑟𝑑 𝑇𝑎𝑥𝑖𝑡 +𝐼𝑛𝑡 𝑇𝑟𝑑 𝑇𝑎𝑥𝑎𝑡 )
(2)
In this equation 𝐷𝑖𝑟𝑒𝑐𝑡 𝑇𝑎𝑥𝑖𝑡 , 𝐼𝑛𝑑𝑖𝑟𝑒𝑐𝑡 𝑇𝑎𝑥𝑖𝑡 , and 𝐼𝑛𝑡 𝑇𝑟𝑑 𝑇𝑎𝑥𝑖𝑡 are
Indonesia’s tax structure components in year t. 𝐷𝑖𝑟𝑒𝑐𝑡 𝑇𝑎𝑥𝑎𝑡 , 𝐼𝑛𝑑𝑖𝑟𝑒𝑐𝑡 𝑇𝑎𝑥𝑎𝑡 , and
𝐼𝑛𝑡 𝑇𝑟𝑑 𝑇𝑎𝑥𝑎𝑡 are average of developed countries tax structure component in year
t. We choose Australia, Austria, Belgia, Canda, Denmar, French, German, Japan,
Dutch, New Zealand, Protugal, Swedish, Switzerland, English, and United states
as developed countries.
Zero value in Bray-Curtis Dissimilarity Index means that the sample is identical
and one value means that the sample is totally difference, so this index is
interpreted as percentage (Greenacre, 2018). For measuring the similarity index
of tax structure Iindonesia with developing countries, we use equation as follow:
Similarity Index for estimating tax reform is the result of substracting 1 with 𝑑𝑖𝑡 .
This equation indicate that the higher 𝑇𝑎𝑥 𝑅𝑒𝑓𝑜𝑟𝑚 value, the higher similarity rate
of Indonesia tax structure with developed countries tax structure. In the other side,
the lower 𝑇𝑎𝑥 𝑅𝑒𝑓𝑜𝑟𝑚 value show high distinctiveness of Indonesia tax structure
to developed countries tax structure (Bray et al., 1957; Gnangnon & Brun, 2019;
Greenacre, 2018).
After determining similarity index, we measure the impact of tax reform
program to tax revenue. There is limitation on calculating tax reform index because
the index depend on tax revenue variable. The tax revenue variabel also depen
on several factor. We collecting those factors to Vector Autoregressive (VAR)
equation. VAR is statistic model for analyzing macroeconomic data. VAR model
has advantage on analyzing multivariate data arranged in time series (Warsono et
al, 2019). VAR equation is presented as follow:
𝑌𝑡 = 𝛽10 + 𝛽11 𝑌𝑡−1 + ⋯ + 𝛽1𝑝 𝑌𝑡−𝑝 + 𝛾11 𝑋𝑡−1 + ⋯ + 𝛾1𝑝 𝑋𝑡−𝑝 + 𝑢1t
𝑋𝑡 = 𝛽20 + 𝛽21 𝑌𝑡−1 + ⋯ + 𝛽2𝑝 𝑌𝑡−𝑝 + 𝛾21 𝑋𝑡−1 + ⋯ + 𝛾2𝑝 𝑋𝑡−𝑝 + 𝑢2t
(4)
In VAR Model, time series variable vector will be regressed to lag vector of
those variable vector. Lag vector symbolized with 𝑝 so VAR(𝑝) model consist of
two variable as showin in equation number (4). Generally, VAR model compesed
with two or more endogenous variable. However, in this paper, we use exogenous
variable in VAR model so the model become VAR with exogenous variable (VARX),
which is known as dynamic model. VARX equation as follow:
In this equation, Γ is endogenous variable, 𝑝 is lag, Ψ is exogenous variable,

and q equal to zero (Warsono et al, 2019).
We conducted two test for achieving best interpretation from VAR model,
which are testing causality between endogenous variable use Granger Causality
and long term cointegration between variable use Johansen Cointegration Test.
We also testing data stasionery use Augmented Dickey-Fuller, optimal lag
determination use Final Preiction Error (FPE) criteria, Akaike Information Criterion
(AIC), Schwarz Information Criteria (SC), and Hannan-Quinn Information Criteria
(HQ), and VAR test stabilization, Autocorellation test, and Normality test to residual.
3.1. Data Selection and Collection

We use secondary data as follow:
1. Indonesia tax revenue and developed countries tax revenue from: Australia,
Austria, Belgia, Canada, Denmark, French, German, Japan, Netherland, New
Zealand, Portugal, Sweden, Switzerland, England, and United States, year
1980 to 2019 based on World Development Indicators bank data consist of
a. Central tax revenue
b. Tax revenue from income, profit, and capital gain
c. Tax revenue from good and service transaction
d. Tax revenue from International trade
2. Indonesia GDP year 1980 to 2019 based on World Development Indicators
bank data
3. GDP per capita year 1980 to 2019 based on World Development Indicators
bank data
4. Indonesia inflation rate year 1980 to 2019 based on World Development
Indicators bank data
5. Export and import Indonesia year 1980 to 2019 based on World Development
Indicators bank data
3.2. Operational Variable Definition

For searching correlation of tax reform and tax revenue, we use previous
studies about structural determinant of public revenue. Variables we use in this
research as endogeneous variable are: Tax Revenue (TAXREV), Tax Reform
(TAXREF), and GDP per Capita (GDPC), and international transaction openness
(TRD_OPEN) as exogeneous variable.
We use tax revenue ratio to GDP as a proxy for measuring tax revenue variable.
For tax reform variable, we use estimation result based on equation number (3).
International transaction openness generally measured by trading opennes
indicator, which calculated by sum of export and import of good and service to
GDP. For testing equation model number (5), we transform variable data into log
(Baunsgaard & Keen, 2005; Desai & Hines, 2003; Gnangnon & Brun, 2019).
4.1. Statistic Data Variable Description
We use secondary data from World Development Index bank data for
testing our model. We convert our data into log form because it is useful for
changing data skewness, so it become near to normality. For data with value
less than one, we use transformation as follow:
Data variable description presented as table 1.
Table 1. Statistic Description

Source: Processed by us using eviews application
Variable
Explanation
TAXREV TAXREF GDPC INF TRD_OPEN
Mean 0.056864 0.224604 6.694934 0.037671 0.183395
Median 0.056035 0.224272 6.769772 0.0321 0.182046
Maximum 0.086169 0.266414 7.767182 0.199895 0.292668
Minimum 0.03479 0.163187 5.488854 0.012966 0.138054
Std. Dev. 0.013286 0.02305 0.746402 0.030207 0.028272
4.2. Unit Root Test

We conduct unit root test to every variable data separately and
simultaneously with Augmented Dickey-Fuller test. We found the test result is
stationery good in point level, inclunding point of 1st difference as presented in
table 2
Table 2. Unit Root Test Result on Every Variable

No. Variable Unit Root Test (Augmented Dickey-Fuller)
1
Probablity in level Probablity in 1st
point difference
1 TAXREV 0.2157 0.00000
2 TAXREF 0.5365 0.00000
3 GDPC 0.5887 0.00002
4 INF 0.0003 0.00000
5 TRD_OPEN 0.0734 0.00000
If unit root test result shows with α value 5%, it means Null Hypothesis
(H0), which lead to not stationery data. Based on our testing, inflation variable
(INF) is stationery in level point, but all variable except INF are stationery in
1st difference point. Unit root test also conducted to all variable data
simultaneously with resulted variable data is stasionery in point of level.
Table 3. Unit Root Test Result to Each of Variable

Cross-
Statistic Prob.** sections Obs
Null: Unit root (assumes common unit root process)
Levin, Lin & Chu t* -1.680507 0.0464293 5 193
Null: Unit root (assumes individual unit root process)
Im, Pesaran and Shin W-stat -2.559629 0.0052391 5 193
ADF - Fisher Chi-square 27.027194 0.0025787 5 193
PP - Fisher Chi-square 29.001312 0.0012454 5 195
** Probabilities for Fisher tests are computed using an asymptotic Chi
-square distribution. All other tests assume asymptotic normality.
Since all variable have passed unit root test, VARX test could be
conducted.
4.3. Optimal Lag Determination

Lag is range of time needed for one independent variable for giving
respond to dependent variable. Optimal lag determination important because
unoptimal lag have a risk of multi cholinerity or lost degree of freedom that
could lead to misintrepretation on result. For knowing optimal lag, we use Final
Prediction Error (FPE), AKaika Information Criterion (AIC), Schwarz
Information Criteria (SC), and Hannan-Quin Information Criteria (HQ) criteria.
Based on FPE, AIC, SC, and HQ test result, the optimal lag for our
research is four. Those results shown in table 4 shown as follow:
Table 4. VAR Lag Order Selection Criteria

Lag LogL LR FPE AIC SC HQ
0 412.6704 NA 1.07e-15 -23.12403 -22.76852 -23.00130
1 443.2360 50.65148 4.70e-16 -23.95634 -22.88982 -23.58818
2 472.6842 42.06886 2.29e-16 -24.72481 -22.94727* -24.11120
3 485.7246 15.64845 3.08e-16 -24.55569 -22.06713 -23.69664
4 513.2793 26.76744* 2.03e-16* -25.21596* -22.01639 -24.11147*
Schwarz Information Criteria recomending 2 lag, while other criteria

recommending 4 lag. We choose 4 lag because we folowing recommendation
majority of criteria beside Schwarz Information Criteria and considering
residual data condition from normality and autocorellation tes result.
4.4. Stability Model Test and Residual Diagnostic

Stability model test used for testing model we use with spesific criteria lag
is on stasionery. Unstable model VAR/VARX could make interpretation of test
result become invalid. The stability of model showed if inverse roots of
charateristic polynomial have modulus under one or inside circle unit. Stability
model test result show inverse roots of charateristic polynomial have modulus
under one, which concluded the VAR/VARX model is stable, presented in table
5 as follow:
Table 5. Roots of Characteristic Polynomial

Root Modulus
0.146458 - 0.866777i 0.879064
0.146458 + 0.866777i 0.879064
-0.830931 - 0.239473i 0.864750
-0.830931 + 0.239473i 0.864750
0.510286 - 0.591157i 0.780934
0.510286 + 0.591157i 0.780934
0.740453 - 0.199648i 0.766897
0.740453 + 0.199648i 0.766897
-0.281465 - 0.663160i 0.720419
-0.281465 + 0.663160i 0.720419
-0.559317 - 0.309112i 0.639051
-0.559317 + 0.309112i 0.639051
0.055351 - 0.546825i 0.549619
0.055351 + 0.546825i 0.549619
-0.054610 - 0.323481i 0.328059
-0.054610 + 0.323481i 0.328059
No root lies outside the unit circle.
VAR satisfies the stability condition.
For testing the residual diagnostic, we use normality and autocorellation

test. Normality test in lag 4 for residual model show normal result in α value
5%, Null Hypothesis: Residual multivariat normal as shown in table 6.
Table 6. Probability Normality Residual Test

Variable
Explanation Joint
TAXREV TAXREF GDPC INF
Skewness 0.331 0.5276 0.871 0.6632 0.816
Kurtosis 0.4143 0.5324 0.1091 0.5393 0.406
Jarque-Bera 0.4468 0.6741 0.2734 0.7533 0.6963
For autocorellation show no problem of autocorellation in lag 4 with α value

5%, Null hypothesis: no serial correlation to lag h, as shown in table 7. In
conclusion, the model could give a proper estimation so hypothesis test could
be proceed.
.
Table 7. Autocorrelation Residual Test

Sumber: Processed by us using eviews application
Lag LRE* stat Df Prob. Rao F-stat Df Prob.
1 23.87276 16 0.0923 1.692929 (16, 31.2) 0.1017
2 14.35653 16 0.5722 0.889569 (16, 31.2) 0.5861
3 12.74512 16 0.6913 0.772273 (16, 31.2) 0.7026
4 15.03164 16 0.5223 0.940202 (16, 31.2) 0.537
4.5. Causality Tax Reform to Tax Revenue

We use Granger Causality test for understanting the capability of tax reform
casuality to tax revenue. Granger Causality Test is statistic hypothesis test for
determining time series data is reliable for predicting other time series data. If the
changes of variable could predict other variable value in the future, it means that
the variable is granger-cause other variable.
Based on Granger Causality test for model, tax reform, GDP per Capita, and
inflation Granger-cause variable of tax revenue with probability of overall in 0,0000
(α value 5%). Our Granger Causality test not estimating casuality international
trade variable, exogenous variable.
Table 8. Granger Causality Tests Model Result

Sumber: Processed by us using eviews application
Dependent variable: D(TAXREV)

Variable Chi-sq Df Prob.
D(TAXREF) 16.38428 4 0.0025
D(GDPC) 17.99296 4 0.0012
D(INF) 16.3272 4 0.0026

All 67.40859 12 0.0000
We conducted cointegration test with Johansen Cointegration Test for finding

long term cointegration of variable. Cointegration test result identified there are two
cointegration vector, so we use vector error corection model (VECM). The Granger
Causality test result with VECM model show that tax reform, GDP per Capita, and
Inflation variable Granger-Cause to tax revenue variable with overall probability in
0,0000 (α value 5%). These result inline with casuality test with VARX model, which
concluded that tax reform, GDP per capita, and inflation as independent variable
have Granger-Cause relation with tax revenue as dependent variable in short term
and long term.
Table 9. Johansen Cointegration Test

Hypothesized Trace 0.05

No. of CE(s) Eigenvalue Statistic Critical Value Prob.**
None * 0.799259 91.44256 47.85613 0.0000

At most 1 * 0.571269 36.84747 29.79707 0.0065
At most 2 0.171116 8.051976 15.49471 0.4599
At most 3 0.04796 1.671024 3.841466 0.1961
Trace test indicates 2 cointegrating eqn(s) at the 0.05 level
* denotes rejection of the hypothesis at the 0.05 level
**MacKinnon-Haug-Michelis (1999) p-values
Tabel 10. Granger Causality Tests VECM Result

Dependent variable: D(TAXREV)
Variabel Chi-sq Df Prob.
D(TAXREF) 16.96300 4 0.0020
D(GDPC) 18.39659 4 0.0010
D(INF) 12.77498 4 0.0124
All 62.36539 12 0.0000
We conduct Impulse Response Function (IRF) test to found the range of

dependent variable react to the change of variable in one standard deviation or
standard-error-shock to other variable. Based on IRF analysis result, one standard
deviation shock of tax reform varibale have affective positive impact to tax revenue
significant in first three period, then decrease in positive region until eight period
until down to negative region in ninth period. However, the impact return to positive
after tenth period. In conclusion, our hypothesis that tax reform have positive impat
to tax revenue in long term is accepted.
Graph 1 Impulse Response Tax Reform to Tax Revenue in 10 periode Graphs

5. CONCLUSION
Tax reform policy is well known and used for elevating countries budget. For
example, East Europe in transition economic period, tax reform has conducted
deeply and sometime done briefly. In other side, tax reform conducted in long
period for other country in Europe, Japan, and any Asia country, which will lead to
redesign their taxation system substantially (Brys, Matthews, & Owens, 2011).
We have not yet found adequate tax reform empirical study. Eka (2019)
conducted study of Indonesia’s tax reform on administration, which concluded that
tax administration reform to tax revenue but increasing taxpayer compliance.
Gnagnon and Brun (2019) found that tax reform has positive impact to tax revenue
in developing countries. We understand the lack of tax reform empirical study
because it is difficult for choose proper proxy for estimating elements of Indonesia’s
tax reform comprehensively. In this study, we feature Similarity Index, the result of
simple modification of Bray-Curtis Dissimilarity index, for estimating Indonesia’s
tax reform. Similarity Index estimate Indonesia tax structure compared with
developed countries tax structure, which if the index near to one, then the similarity
rate is high.
We use the result of similarity index as a result as a proxy for tax reform variable
so we could conduct test for finding the relation of tax reform variable with tax
revenue variable. Based on our VARX and VECM model result, we concluded that
tax reform variable has causality in short and long term with tax revenue variable.
Our analysis results in line with Gnangnon and Brun (2019), but opposite with Eka
(2019) founding.
Furthermore, our IRF test result to VECM found that tax reform will give positive
impact to tax revenue approximately in long term, even though it is predicted will
have negative impact until ninth period, it will generate positive impact after tenth
period. We suggest our result could be reference for Indonesia’s tax institution for
reviewing or changing tax policy regarding tax reform aim at the right time.
Our research has estimated Indonesia’s tax reform elements using

appropriate index for testing its impact on tax revenue. However, we understand
we our limited resource, time, and research data that make us have not yet capture
some of tax reform element such as good corporate governance. We suggest for
the next researcher for developing those element for improving tax reform program
estimation.
Appreciation
We appreciate Indonesia Endowment Fund for Education or Lembaga Pengelola Dana
Pendidikan (LPDP) assistance with the research for this article.
BIBLIOGRAPHY
Baunsgaard, T., & Keen, M. (2005). Tax Revenue and ( or ?) Trade Liberalization.
Bray, J. R., Curtis, J. T., & Roger, J. (1957). This content downloaded from
147.8.31.43 on Mon. Source: Ecological Monographs, 27(4), 325–349.
Brys, B., Matthews, S., & Owens, J. (2011). Tax reform trends in OECD Countries.
OECD Taxation Working Papers No. 1, (1).
Chan, N., Ghosh, M., & Whalley, J. (1999). Evaluating Tax Reform in Vietnam
Using General Equilibrium Methods. University of Western Ontario,
Department of Economics, UWO Department of Economics Working
Papers: 9906, 1999. Retrieved from
http://search.proquest.com/docview/56349646?accountid=17248
Desai, M. A., & Hines, J. R. (2003). Evaluating international tax reform. National
Tax Journal, 56(3), 487–502. https://doi.org/10.17310/ntj.2003.3.03
Eka, I. W. A. (2019). Dampak reformasi administrasi perpajakan indonesia kepada

kepatuhan dan penerimaan pajak 1. Jurnal Ekonomi Dan Pembangunan,
27(1).
Gnangnon, S. K., & Brun, J. F. (2019). Trade openness, tax reform and tax revenue
in developing countries. World Economy, 42(12), 3515–3536.
https://doi.org/10.1111/twec.12858
Greenacre, M. (2018). Chapter 5 Measures of distance between samples.

Correspondence Analysis and Related Methods, 1–10. Retrieved from
http://www.econ.upf.edu/~michael/stanford/maeb5.pdf%0Ahttp://www.econ
.upf.edu/~michael/stanford/
Kusi, N. K., Economic, A., Pour, C., Recherche, L. A., & En, E. (1998). TAX
REFORM AND REVENUE Tax reform and revenue productivity in Ghana.
Rosen, H. S. (1976). A METHODOLOGY FOR EVALUATING TAX REFORM

PROPOSALS. Journal of Public Economics, 6(1–2), 105–121.
Sajadifar, S. H., Khiabani, N., & Arakelyan, A. (2012). A computable general

equilibrium model for evaluating the effects of value-added tax reform in
Iran. World Applied Sciences Journal, 18(7), 918–924.
https://doi.org/10.5829/idosi.wasj.2012.18.07.1772
Satya, Venti Eka (Juni 2017). "Optimalisasi Penerimaan Pajak melalui Reformasi
Pajak" (PDF). Majalah Info Singkat Ekonomi dan Kebijakan Publik. Vol. IX
no. 12. Jakarta: Pusat Penelitian Badan Keahlian DPR RI. hlm. 13–
16. ISSN 2088-2351
Silvani, C., Baer, K. (1997). Designing a Tax Administration Reform Strategy:

Experiences and Guidelines. INTERNATIONAL MONETARY FUND
Warsono, Russel, E., Wamiliana, Widiarti, & Usman, M. (2019). Vector

autoregressive with exogenous variable model and its application in
modeling and forecasting energy data: Case study of PTBA and HRUM
energy. International Journal of Energy Economics and Policy, 9(2), 390–
398. https://doi.org/10.32479/ijeep.7223
APPENDIX
Table 1. Variable Definition
Variable Symbol Variable Definitions Sumber
TAXREV Tax Revenue to GDP World Development

(based on Local Currency Unit) Indicators Data
TAXREF Tax reform indicator
GDPC GDP per Capita World Development

Indicators Data
(based on Local Currency Unit)
INF Inflation World Development

Indicators Data
TRD_OPEN International transaction openness, World Development

measured by sum of international Indicators Data
transaction to GDP ratio
(based on Local Currency Unit)
THE IMPACT OF THE ALTERNATIVE TAX BASE
MEASUREMENT POLICY ON THE VAT REVENUE
PERFORMANCE IN THE INDONESIAN
AGRICULTURAL SECTOR
[Singgih Priyogo]a, [Rus’an Nasrudin]b

a [Directorate General of Taxation, Jakarta, Indonesia] Email: [singgih.priyogo@gmail.com]
b
[University of Indonesia, Jakarta, Indonesia] Email: [rusan.nasrudin@gmail.com]
ABSTRACT
Informality issues and government distributional objectives cause the need for VAT special
treatment for the agricultural sector. The treatment forms and resulting impacts vary, depending on
each country's conditions and necessities. This study aims to estimate the impact of the alternative
tax base measurement policies on the VAT revenue performance in the Indonesian agricultural
sector. Using input-output table modeling, the results show that the policy has a positive impact on
the compliance level and VAT revenue in the agricultural sector, but reduces the aggregate national
VAT revenues, and increases the VAT burden that must be borne by the agricultural sector
entrepreneurs. The normal VAT mechanism is a preferred choice for the long-term goal, but the
alternative policy is still needed during the transition period.
Keywords: VAT Revenue Performance, VAT Compliance Level, Agricultural Sector, Input-Output Model
Dampak Kebijakan Alternatif Pengukuran Basis Pajak Terhadap Kinerja Penerimaan PPN Sektor
Pertanian Indonesia. Isu informalitas dan tujuan distribusi pemerintah menyebabkan perlakuan
khusus PPN atas sektor pertanian masih diperlukan. Bentuk perlakuan dan dampak yang
ditimbulkan beragam, tergantung kepada kondisi dan kebutuhan di masing-masing negara. Studi
ini bertujuan untuk mengestimasi dampak kebijakan pengukuran alternatif basis pajak terhadap
kinerja penerimaan PPN sektor pertanian di Indonesia. Menggunakan permodelan table input-otput,
hasil yang diperoleh menunjukkan bahwa kebijakan tersebut berdampak positif terhadap tingkat
kepatuhan dan kinerja penerimaan PPN sektor pertanian, namun menurunkan agregat penerimaan
PPN nasional, dan meningkatkan beban PPN yang harus ditanggung oleh pengusaha sektor
pertanian. Mekanisme PPN normal adalah pilihan yang lebih baik dalam tujuan jangka panjang, tapi
kebijakan alternatif masih diperlukan selama masa transisi.
Kata kunci: Kinerja Penerimaan PPN, Tingkat Kepatuhan PPN, Sektor Pertanian, Model Input-Output
1. INTRODUCTION
VAT imposition on the agricultural sector is often considered difficult due to two
main factors, informality and the government distribution objectives. The
agricultural sector is generally dominated by small farmers who do not keep
records of their business activities, and only a small proportion of large farmers do
it properly. This condition creates administrative problems that cause normal VAT
provisions cannot be fully applied. The agricultural sector is also a source of
livelihood for most low-income groups and its production is needed to fulfill public
consumption, production inputs for other business sectors, as well as important
commodities for export purposes. The government for distribution purposes and
other political considerations can stipulate a special VAT treatment for this
important sector.
Most countries broadly exclude the agricultural sector from the VAT imposition,
and only a few countries have fully taxed this sector. Developments in the countries
that have fully imposed VAT on the agricultural sector, such as New Zealand and
Denmark, are not entirely encouraging, that the special VAT treatment of
agriculture is not inevitable (Ebril et al., 2001). The special VAT treatment practice
in the agricultural sector varies among countries, adjusting to the conditions and
necessities of the country. The question of what form of VAT treatment is proper
for the agricultural sector and how effective its impacts are to optimize VAT
revenue is still not fully answered. Although the normal treatment of VAT equates
to other business sectors in the long term has become a clear goal. However,
during the transition period, the high costs of administration and collection,
distribution objectives, and political considerations are still a strong argument for
giving special treatment to this important sector.
This study will estimate the impact of VAT policies on the Indonesian
agricultural sector, which provides an alternative basis for imposing VAT on certain
agricultural commodities. This provision is regulated in the Minister of Finance
Regulation number 89 of 2020 which has been in effect since July 2020. This
treatment is a different policy design from VAT regulations which are usually
mandatory. This policy gives taxable entrepreneurs options to choose the method
of calculating the tax imposition basis on the delivery of certain agricultural
products, whether to continue using the selling price or switch to using another
value of 10% of the selling price but lose the right to credit input tax. This change
in VAT policy has the potential to influence the economic decisions of the affected
producers and consumers, change the basis of VAT, and ultimately have an impact
on the VAT revenue that can be collected by the government.
This policy is intended as a solution to the problem of imposing VAT on the
Indonesian agricultural sector. The application of the normal VAT mechanism, after
the Supreme Court granted the judicial review of the VAT exemption facility for
agricultural products in 2013, creates difficulty in fulfilling VAT obligations,
especially for farmers/taxable entrepreneurs who have administrative issues and
limited understanding of tax provisions. They can choose to use an alternative
method that is simpler, no longer required to account for input VAT, with a lower
effective rate. Meanwhile, the normal mechanism for crediting input VAT can still
be used by farmers/taxable entrepreneurs who need it. This policy is expected to
accommodate the needs of all taxable entrepreneurs in the agricultural sector,
thereby increasing compliance and VAT revenue in the agricultural sector, which
so far have not been optimal.
Based on previous studies, various factors were found that affect the
performance of VAT revenue in a country. The interaction between tax structure
(tariff, tax base, and threshold), taxable economic activity, and compliance with tax
regulations play an essential role in determining the performance of VAT revenue
(Ebril et al., 2001). Tax rate, tax knowledge, and perceived fairness are several
factors, among others, that affect the level of compliance according to a review by
Kirchler et all. (2008) on the existing literature. The marginal tax rate has a mixed
empirical impact on compliance (see Alm et al., 1992; Baldry, 1987; Park & Hyun,
2003). Reduces complexity and increases tax knowledge have a positive impact
on compliance (Clotfelter, 1983; Kirchler & Maciejovsky, 2001). Perceived as a fair
tax system will increase trust and boost voluntary tax compliance (Braithwaite,
2003b; Wenzel & Thielmann, 2006).
Single tariff and broad consumption base are characteristics of a good structure
of the VAT system (James, 2015). This can lower efficiency costs and simplify
administrations because the tax authority does not have to allocate its limited
resources to identify different economic activities with different treatment (Bird &
Gendron, 2007). The complex tax system and narrow tax base will result in low
VAT collection efficiency (Bogetic and Hassan, 1993). The higher VAT rates could
be associated with lower compliance, and the use of multiple rates has an adverse
effect on compliance (Agha and Haughton, 1996). Furthermore, political and
economic structural factors (such as the level of urbanization, the share of the
agriculture sector, and the degree of economic openness) have impacts on the
performance of VAT revenue in a country (Aizenman & Jinjarak, 2008).
The study conducted by Iswahyudi (2018) shows that the decline in the VAT
revenue performance in Indonesia was mainly influenced by VAT expenditure
policies and taxpayers' non-compliance. Reducing the scope of VAT exemptions
and simplifying the VAT system with the application of a single VAT rate can have
a positive impact on VAT collection. Sugana & Hidayat (2014) in a study of the
potential and tax gaps in VAT revenue show that increasing the level of compliance
has a higher positive impact on VAT revenue than increasing the VAT rate. These
two studies show that a simple VAT system and increased compliance levels play
an important role in improving VAT revenue performance in Indonesia.
The special treatment of VAT in the Indonesian agricultural sector allows two
methods of imposing VAT on the delivery of a certain agricultural product, namely
normal method and other value method. Taxable Entrepreneurs have a right to
decide what method they will use. The other value methods offer lower effective
rates with the consequence that the right to credit input VAT is forfeited. This
makes the alternative method simpler because the administration of input VAT,
which is considered onerous and complicated for some Taxable Entrepreneurs in
the agricultural sector, is no longer needed. The existence of method options
offered makes this policy fairer because the Taxable Entrepreneur can assign a
method that best suits their individual needs and tax administration capacities.
However, this method options also yield effective VAT rate differences on selling
the same agricultural product.
The impact of the special treatment on the VAT revenue performance cannot
be concluded using the previous studies finding. This happens because of contrary
impacts interaction caused by the policy change. Simplified VAT administration,
lower effective rates and a fairer VAT policy for all agricultural sector entrepreneurs
can have positive impacts on VAT compliance and revenue performance. On the
other hand, the tariff differences for the same agricultural products can distort the
decisions of economic agents, increasing the complexity of fulfilling obligations and
administering taxes, which have negative impacts on the level of VAT compliance
and revenue performance. The option methods for imposing VAT on certain
agricultural products allow for mixed-effects on VAT revenue in the agricultural
sector.
The alternative method of VAT imposition is a deviation from the normal VAT
provisions. The smaller tax base due to the use of alternative methods potentially
reduces VAT revenues. However, VAT on production inputs that cannot be
credited will be an additional government revenue from production activities, which
should not be intended to be subject to VAT as a consumption tax. This non-
creditable input VAT also has the potential to increase the tax base if it is added
as a component of the selling price and has a cascading effect at the production,
distribution, and final consumption stages. The impact of this policy on VAT
revenues is the accounting net effect of the increases and decreases in the
revenue generated.
This study aims to estimate the impact of alternative tax base measurement on
VAT revenue, and assess the magnitude tax gap between VAT theoretical
collections and VAT actual collections in the agricultural sector. The VAT treatment
on the Indonesian agricultural sector is one of the various forms of special VAT
treatment in the agricultural sector, and the impact on the VAT revenue
performance cannot be concluded using the previous studies finding. The study
results are expected to provide a basis for evaluating and formulating VAT policies
in the agricultural sector that require special treatment depending on their
characteristics to optimize the potential VAT revenues as a source of financing for
development and prosperity programs.
2. THEORETICAL FRAMEWORK AND DEVELOPMENT OF

RESEARCH MODEL
2.1. Theoretical Framework
Taxation has an important role for the government in managing the country's
economy through fiscal policy. Fiscal policy is the government's effort to use its
revenues and expenditures to influence the country's macroeconomic conditions.
Even more broadly, according to Keynesians, fiscal policy can affect aggregate
demand and the level of economic activity. Taxation is an essential factor from the
revenue side in fiscal policy. Changes in tax policy adopted by the government
may lead to changes in aggregate demand and the level of economic activity in the
affected sectors, including other related sectors, thereby resulting in aggregate
changes at the national level. When the government raises/lowers taxes, it will
impact decreasing/increasing the aggregate level of consumption, production, and
investment nationally.
Taxation has a vital function as revenue adequacy to ensure that government
revenues are appropriate and sufficient to finance the government expenditure
needs over time. If these conditions are not achieved, the government must resort
to borrowing, printing money, selling assets, or slowing down the implementation
of development programs. These actions will ultimately harm the people, especially
the poor. In addition, taxation also plays a role in revenue stability to maintain the
continuity of the government's fiscal policies. So, it is important to have a tax
system that is effective in collecting revenue and efficient because it does not
distort economic activity to ensure a healthy budget policy in financing
development and welfare programs over time.
The performance of tax revenues in a country is strongly influenced by the
scope of the tax base, tax rates, and the level of tax compliance. Changes in tax
policy will impact the level of tax revenue that the government will collect. For
example, when the government increases/decreases VAT rates, the direct impact
is an increase/decrease in tax revenue on the same tax basis. While the indirect
effect is due to an increase/decrease in prices due to an increase/decrease in tax
rates, which will reduce/increase aggregate consumption and production, which
means a decrease/increase in the tax base. Thus, how much tax revenue is
generated from a change in tax policy depends on comparing the direct and indirect
impacts. A good tax policy can create opportunities/distortions for taxpayers,
leading to more productive efforts, not just tax planning to reduce the tax burden.
The new policy possibly impacts aggregate demand and supply of agricultural
commodities through the price changes mechanism due to changes in the effective
tax rate. The provision of an alternative measurement of the tax base will impact
the size of the VAT base. With a single statutory VAT rate of 10%, the change in
the tax base from the selling price to another value (10% of the selling price)
causes a change in the effective tax rate to 1% of the selling price. Because this is
optional, there will be differences in effective VAT rates on certain agricultural
commodities, which will cause price differences that can distort economic
decisions from both the producer and consumer sides and will change aggregate
demand and supply nationally.
The new policy will also cause differences in the value of VAT paid by taxable
entrepreneurs on the sale of certain agricultural products, which will impact the
VAT revenues that the government will collect. The VAT that must be paid using
the normal mechanism is (10% x selling price) - input tax on production inputs.
Meanwhile, VAT that must be paid using another value mechanism is (10% x (10%
x selling price)) or 1% of the selling price without considering the input tax. The
burden of the non-creditable input tax can be transferred to the consumer through
the selling price component or will be borne by the producer as a profit deduction.
These two conditions will increase VAT revenue because the input tax will become
the basis for further VAT taxation (cascading effect), or there will be taxation at the
producer level, which is not statutorily the subject who should bear the VAT burden.
The options offered by the VAT policy can be considered in tax planning for
both producers and consumers in maximizing their respective utilities. Producers,
to maximize profit, will choose a VAT collection method to minimize the tax and
administrative burden they have to bear. Meanwhile, to maximize utility,
consumers will choose products that provide the same level of benefits with
minimum sacrifice (lower price plus VAT). The choice of VAT collection method
taken by producers will impact the choice of products that consumers will buy. The
interaction of these choices in the aggregate will impact the basis and revenue of
VAT from the agricultural sector that the government will collect.
The VAT revenue that the government will collect due to changes in VAT policy
in the agricultural sector will technically be significantly influenced by the large
proportion of Taxable Entrepreneurs who switch to using alternative mechanisms.
The greater the proportion of using other values as the basis for imposing VAT, the
greater the potential for a decrease in VAT revenue due to a lower effective VAT
rate. But on the other hand, there will be additional government revenue from VAT
on production inputs that cannot be credited by the Taxable Entrepreneur. This
non-creditable input VAT also can increase the VAT base when the Taxable
Entrepreneur can transfer the burden of input VAT to the buyer by adding it to the
selling price component. The more inelastic the demand for price changes, the
greater the opportunity to transfer the burden of input VAT to consumers. The
addition of VAT revenue from the cascading of tax on input will be even greater
when the production and distribution chain takes longer to deliver to the final
consumer (Ebril et al., 2001). The impact of this policy on VAT revenue is the net
value of the increase and decrease in revenue that occurs.
2.2. Research Model Development
VAT applicable in Indonesia has several main characteristics, namely: VAT is
a tax on consumption that is imposed in stages on each production and distribution
line until final consumption. To avoid the effect of double taxation, VAT applies a
crediting mechanism. As an indirect tax, the seller acts as the person responsible
for the tax, but the consumer is the party who bears the tax burden. The VAT
negative list principle defines that all goods and services are goods and services
subject to VAT at a single rate unless otherwise stipulated by law. The destination
principle is applied by imposing VAT on domestic consumption and imports and
applying zero tariffs to exports. In addition, VAT also provides some VAT
exemptions and facilities for particular commodities or sectors.
Input-Output table has a data structure that can be used to describe the VAT
collection mechanism. The I-O table presents information about transactions of
goods and services that occur among economic sectors in a statistical matrix in an
economy and a certain period of time. The contents along the rows of the I-O table
show the allocation of output produced by a sector to meet intermediate and final
demand. Meanwhile, the columns show the input structure used by each sector in
the production process. The I-O table contains data on final demand and detailed
information on sectoral inputs and outputs that are able to describe the inter-
sectoral linkages in economic activity. This data structure draws the mechanism
for collecting VAT as a tax on consumption imposed on each production and
distribution chain. So, the I-O table can be used to build VAT modeling.
There are three approaches commonly used in VAT modeling, namely
aggregate national account approach, sectoral national account approach, and
input-output table approach (Jenkins et all., 2000). The three different approaches
should yield consistent results in projected VAT revenues and compliance rates.
However, since I-O tables provide the most detailed accounting data for supply
and use, and final consumption in an economy, I-O VAT modeling can be
considered the best approach for estimating the disaggregated impact on VAT
collection and compliance at the commodity level due to policy changes (Le, 2007).
Many studies use I-O modeling for various purposes, among others, Jenkins
et. al (2000) develop an I-O model to estimate the tax base and simulated revenue
based on data and the case in Canada which carried out tax reform from sales tax
to VAT. Le (2007) develops an IO model for estimating the VAT basis taking into
account the impact of small businesses' outputs and inputs with turnover below a
predetermined VAT threshold for Romania. Manasan (2002) formulates an I-O
model to estimate the performance indicators of VAT revenue for each industrial
sector in the Philippines. Warwick et al. (2022) use input-output tables to estimate
the impact of VAT exemptions on consumer prices because VAT is attached to the
supply chain. Then it is incorporated with a microsimulation model to estimate the
impact of VAT policies on tax revenues, inequality, and poverty.
Mark (2003) used the I-O modeling approach to calculate potential VAT
revenue in Indonesia. He uses the 1995 I-O table to estimate the impact of the VAT
exemption policies on certain goods and services that are widely practiced in
Indonesia on the performance of VAT revenue. The result is that the potential VAT
revenue in Indonesia is 5.83% of GDP, but the actual revenue that can be collected
is only 3.2%. Eliminating VAT exemptions in some sectors did not generate much
additional revenue, but instead reduced revenues in some sectors. Sugana &
Hidayat (2014) estimates the level of revenue and compliance of VAT using the
2008 I-O table for validating the compliance rate in the study conducted by Marks
(2003). They make adjustments in determining the proportion of taxables, the level
of compliance, and the tax year studied. The result is that the VAT revenue
estimated in the 2013 fiscal year is Rp. 381 trillion which is only less than 1% of
actual revenue, and a compliance rate of 53%.
This study extends the model built by Sugana & Hidayat (2014) because the
model is considered the most proper and allows for some adjustments to
accommodate the objectives of this study. In this study, the use of alternative
methods proportion is added to estimate the impact of the alternative policy and
simulate cascading effects when taxable entrepreneurs can transfer the burden of
uncreditable input VAT to buyers. In addition, adjustments were made in
determining the taxable proportion and below threshold proportion, using the latest
input-output tables and utilizing available DGT administrative data.
This study tries to adopt and use the principles of the I-O model formation
based on the study of Sugana & Hidayat (2014), by incorporating the alternative
treatment of VAT in the agricultural sector and using the latest I-O table data in
2016. This study also utilizes administrative data from the DGT in estimating the
proportion of taxable entrepreneurs who switch to using alternative methods to
analyze the impact of policies on VAT revenue and compliance. The research
period is the 2021 tax year because it is the period after the policy was issued,
including the first month of a tax year when the transition to the use of the
alternative method is allowed, and the transition period from the effects of the
COVID-19 pandemic.
This study uses the 2016 I-O data table which contains the input and output of
185 business sectors, which are then projected to get a value that can reflect the
conditions of the 2021 projection year. Gross up factor is needed to project the
value of final household expenditure, gross fixed capital formation and intermediate
inputs in 2021 based on I-O table data in 2016. Gross up factor is obtained by
comparing the GDP value of each business sector in the projection year with the
base year using the formula as follows:
𝐺𝐷𝑃𝑖,𝑡 𝐺𝑖 : Gross up factor of commodity i
𝐺𝑖 = … (1)
𝐺𝐷𝑃𝑖,0 𝐺𝐷𝑃𝑖,𝑡 : GDP of commodity i in the projected year
𝐺𝐷𝑃𝑖,0 : GDP of commodity i in the base year
The gross up factor then used to project the value of final consumption,
intermediate inputs, and fixed capital formation as follows:
𝐶𝑖,𝑡 = 𝐺𝑖 × 𝐶𝑖,0 …(2) 𝐶𝑖,𝑡 : Final consumption in the projected year
𝐶𝑖,0 : Final consumption in the base year
𝐼𝑛𝑡𝑖,𝑡 = 𝐺𝑖 × 𝐼𝑛𝑡𝑖,0 …(3) 𝐼𝑛𝑡𝑖,𝑡 : Intermediates input in the projected year
𝐼𝑛𝑡𝑖,0 : Intermediates input in the base year
𝐼𝑖,𝑡 = 𝐺𝑖 × 𝐼𝑖,0 …(4) 𝐼𝑖,𝑡 : Capital formation in the projected year
𝐼𝑖,0 : Capital formation in the base year
The first step that needs to be done is to estimate the effective VAT base from
the three main components of the I-O table, namely the supply matrix, use matrix,
and final demand matrix. The final demand data needs to be adjusted to exclude
the proportion of goods and services that are excluded, subject to zero tariffs,
exempted, and borne by the government. Furthermore, the proportion of final
demand sold by non taxable entrepreneurs or entrepreneurs with a turnover below
the threshold is also taken into account. So that the effective VAT base is obtained
from the final demand data that will be subject to VAT as follows:
𝑩𝑪𝒊 = 𝑪𝒊 ∗ 𝝆𝒊 ∗ (𝟏 − 𝝅𝒊 ) ∗ ((𝟏 − 𝐀𝒊 )(𝟏 − 𝑫𝑷𝑷𝒊 ) + 𝑫𝑷𝑷𝒊 ) . . . (𝟓)
𝐵𝐶𝑖 : Taxable final demand base of sector i
𝐶𝑖 : Total final demand of sector i
𝜌𝑖 : Taxable proportion of sector i
𝜋𝑖 : Non taxable entrepreneur proportion of sector i
𝐴𝑖 : proportion of using the alternative mechanism of sector i
𝐷𝑃𝑃𝑖 : Proportion of alternative tax imposition basis of sector i
The taxable proportion (𝜌𝑖 ) will be 1 if all commodities in a sector are subject to
VAT, and 0 if otherwise. If only a part of the commodities are subject to or excluded,
then it can be contains value between 0 to 1. The non taxable entrepreneur
proportion 𝜋𝑖 is based on the proportion of value added of micro and small
businesses in each business sector, assuming that small entrepreneurs with
turnover below the threshold are not taxable entrepreneur. The proportion of using
the alternative mechanism (𝐴𝑖 ) is estimated based on the applicable VAT provisions and
available DGT administrative data. 𝐴𝑖 will be worth 0 if there is no alternative mechanism
based on applicable regulations, and 1 if otherwise. The existence of a basis option for
imposing VAT on certain commodities in the agricultural sector allows 𝐴𝑖 to be worth
between 0 to 1. The proportion of the alternative tax base imposition (𝐷𝑃𝑃𝑖 ) is determined
based on the applicable VAT regulations.
The taxable proportion of the supply and use matrix between business sectors
consists of taxable inputs used by non taxable entrepreneur and taxable
entrepreneur who produce commodities that are excluded and exempt from VAT
because they cannot credit the input tax, so they are treated as final consumption.
It is also necessary to consider the VAT regime that applies to intermediate inputs
and who sold the intermediate inputs. Consideration of the VAT mechanism used
will also have an impact on how large the VAT basis from intermediate inputs must
be taken into account in the model. Intermediate inputs used by taxable
entrepreneur that use other value mechanisms will increase the taxable
intermediate input base. Meanwhile, intermediate inputs purchased from taxable
entrepreneur using other value mechanisms will reduce the taxable intermediate
input base. The taxable basis of intermediate inputs can be formulated as follows:
𝑩𝑰𝒏𝒑𝒋 = (Max((𝟏 − 𝝆𝒋 ); 𝝅𝒋 ) + 𝐴𝒋 ) ∗
𝑛 𝑰𝒏𝒑𝑗,𝑖
∑ (𝝆𝒊 ∗ (1 − 𝝅𝒊 ) ∗ ((1 − 𝐴𝒊 )(𝟏 − 𝑫𝑷𝑷𝒊 ) + 𝑫𝑷𝑷𝒊 ) ∗ … (6)
𝑖 𝑰𝒏𝒑𝑗
BInpj : Taxable intermediate input base of sector j
Inp𝑗 : Total intermediate input of sector j
Inp𝑗,𝑖 : Total intermediate input of sector j from sector i
ρi : Taxable proportion of intermediate input sector i
πi : Non taxable entrepreneur proportion of intermediate input sector i
𝐴i : Proportion of using the alternative mechanism of sector i
DPPi : Proportion of alternative tax imposition basis of sector i
ρj : Proportion of taxable output generated by sector j
πj : Proportion of output sold by non taxable entrepreneur in sector j
𝐴j : Proportion of using the alternative mechanism of sector j
The fixed capital formation and investment is basically not a VAT basis,
because taxable entrepreneurs can credit input tax on the transaction. The fixed
capital formation includes construction of business places, purchase of machinery
and equipment, purchase of livestock seeds and so on. However, if it is carried out
by non taxable entrepreneurs, the fixed capital formatin will be treated as an object
of VAT, so it will be treated the same as intermediate inputs. The purchase of new
housing by households is an object of VAT and is included in the component of
fixed capital formation in the I-O table, so it must be added as a tax base by making
adjustments to the land value that has not been taken into account in the I-O table.
The above steps will result in an effective VAT base of final demand,
intermediate inputs, and fixed capital formation. The effective basis of VAT when
multiplied by the applicable VAT rate, the potential VAT revenue will be obtained
at a full compliance rate of 100%.
𝑉𝐴𝑇𝑖 = 𝐵𝑖 × 𝑡 . . . (7)
The difference between the actual administrative data of VAT revenue and the
potential VAT revenue from the estimated model is a tax gap that shows the level
of taxable entrepreneur compliance in carrying out VAT obligations correctly.
𝑅𝑖𝐴 𝐾𝑖 : Compliance rate of commodity i

𝐾𝑖 = . . . (8) 𝑅𝑖𝐴 : Actual VAT revenue of commodity i
𝑅𝑖𝑃
𝑅𝑖𝑃 : Potential VAT revenue of commodity i
The impact of the alternative policy of measurement VAT imposition basis on
VAT revenue in the agricultural sector is carried out by comparing the results of
the estimation of potential VAT revenue at the full compliance rate between
conditions that exist in 2021 with simulation results when there is no policy and
when the policy is fully mandatory, not as an option for taxable entrepreneur.
Meanwhile, the gap between theoretical and actual revenue that represent the rate
of compliance of the agricultural sector is measured by comparing the level of
compliance in 2021 with the level of compliance in the previous year where this
alternative policy has not been implemented.

4.1. Data Preparation
The gross-up factor is obtained by comparing the value of sectoral GDP in 2021
with 2016 as the base year. This method does not consider changes in the
technical coefficients which indicate changes in technology, prices, and
classifications used. However, the projected value is assumed to represent the
conditions of the projection year, so that it can be used to measure the impact of
alternative policies on revenue and compliance levels. The gross-up factor is
needed to estimate the value of household final consumption, gross fixed capital
formation, and intermediate inputs at the aggregate level of each sector in the
projected year. The estimated final consumption and gross fixed capital formation
values then adjusted based on the aggregate expenditure data reported by Central
Bureau of Statistics.
The taxable proportion is determined based on the technical, applicable VAT
regulations, and administrative issues. There are 42 sectors in which the taxable
proportion is 0 because all of the outputs are exempt from VAT, and 22 sectors in
which only part of the output is subject to/excluded from VAT so that the taxable
proportion is between 0 and 1. The last discussed taxable proportion is calculated
using DGT administrative data derived from the VAT period report. This proportion
is obtained by dividing the total sales that are excluded, exempted, and not
collected from VAT by the total amount of sales reported by taxable entrepreneur
for each Classification of Business Fields (KLU). It should be considered that the
data used only covers sales made by registered entrepreneurs, and there is no
incentive for taxable entrepreneurs to report sales that are excluded from VAT. So
that the proportion obtained will tend to be larger than if the data on sales made by
non-PKP are available and included in the calculation. However, the resulting value
is assumed to be sufficient to represent the actual condition.
The proportion of entrepreneurs below the VAT threshold is obtained using the
2016 economic census data by comparing the income data of Micro and Small
Enterprises (MSEs) with the total income of MSEs and Large Medium Enterprises
(LMEs) per business sector. Meanwhile, the proportions for the agriculture,
forestry, and fishery sectors were obtained using available data from the
agricultural census. The proportion obtained is considered to reflect the portion of
entrepreneurs who have a business turnover in one year less than the VAT
threshold, which is 4.8 billion. It is assumed that small entrepreneurs with a
turnover below the threshold are not taxable entrepreneurs, although based on the
applicable VAT provisions, they have the right to voluntarily register as taxable
entrepreneurs.
The proportion of alternative method is obtained based on DGT administrative
data from tax invoices issued using code 04 which indicates that the VAT on the
sale is calculated using other values. Based on the minister of finance regulation
number 89 of 2020, there are 27 sectors of the I-O table that are entitled to use
other values as the basis for imposing VAT. There are other provisions that
regulate the use of other values, but because the proportion is not significant, all
sales in tax invoices with code 04 from the 27 sectors are assumed to be sales as
regulated in the minister of finance regulation number 89 of 2020, which is then
divided by the total taxable delivery of each sector.
4.2. Policy Impact on VAT Revenue
The projected final household expenditure, gross fixed capital formation, and
intermediate input in 2021 and the previously obtained proportions are then
inserted into the I-O model that has been formed before. This step will obtain an
estimate of potential VAT revenue at the level of 100% compliance when
alternative policies are applied. The next step is to conduct a simulation to see the
VAT revenue potential if the policy is not applied and when the policy is fully applied
by certain agricultural sectors. It is also significant to conduct a simulation by
adding non-creditable VAT input to the selling price component to estimate the
impact of the cascading effect. The impact of alternative policies on VAT revenues
can be obtained by comparing the VAT revenue potential from those conditions as
shown in table 4.1.
The VAT revenue potential for the agricultural sector when the alternative
method applies in 2021 is 0.35% greater than the VAT under normal conditions
applied. These conditions indicate that the policy can increase VAT revenues in
the agricultural sector. But cumulatively, the VAT revenue decreases 0.01% due
to the decrease in the revenue potential from the other business sectors, except
the trading sector. The policy will increase the national revenue potential by 0.79%
when the policy is fully implemented (no longer as an option for taxable
entrepreneur) in certain agricultural sectors. The two sectors experiencing an
increase in revenue potential are business sectors that could apply the other value
bases, while the others, experiencing a decrease in VAT revenue potential, are
sectors outside the PMK-89 policy but have links to the agricultural sector and trade
in certain agricultural products.
The sectors for the supply of accommodation, food and drink, processing
industry, and construction are the three sectors that have the greatest linkages to
the agricultural sector, as seen from the increase in the proportion of other uses of
value in the agricultural sector, causing a relatively large decrease in the potential
for VAT revenue compared to other business sectors. Meanwhile, other business
sectors tend to be unaffected (relatively less than 1%) by changes in VAT policy in
the agricultural sector, particularly the information and communications sector, and
real estate. This can be a consideration for the government in setting VAT policies
in the agricultural sector, it is also necessary to consider the impact it will have on
other related sectors.
Tabel 4.1. Full Compliance VAT Revenue Potential (Billion Rp)
No Condition Full % No Condition % Full %

% Impact
Sector Alternative 2021 Alternative Impact Alternative* 2021* Impact* Alternative* Impact*
(a) (b) (b-a)/a (c) (c-a)/a (d) (e) (e-d)/d (f) (f-d)/d
Agriculture, Forestry and Fisheries 43.251 43.404 0,35% 61.400 41,96% 43.801 43.957 0,36% 62.203 42,01%
Mining and excavation 51.234 51.204 -0,06% 51.201 -0,06% 51.889 51.860 -0,06% 51.872 -0,03%
Processing industry 257.252 257.015 -0,09% 248.685 -3,33% 260.498 260.265 -0,09% 251.903 -3,30%
Electricity and Gas Supply 17.749 17.743 -0,04% 17.743 -0,04% 17.978 17.972 -0,03% 17.977 0,00%
Water Supply, Waste Management, Waste 1.542 1.541 -0,07% 1.541 -0,09% 1.562 1.561 -0,07% 1.561 -0,06%
and Recycling
Construction 67.171 67.043 -0,19% 65.741 -2,13% 68.018 67.890 -0,19% 66.590 -2,10%
Wholesale and Retail Trade; Car and 108.813 109.149 0,31% 109.120 0,28% 110.190 110.534 0,31% 110.536 0,31%
Motorcycle Repair
Transportation and Warehousing 56.583 56.576 -0,01% 56.576 -0,01% 57.296 57.291 -0,01% 57.307 0,02%
Provision of Accommodation and Food and 20.815 20.748 -0,32% 19.922 -4,29% 21.079 21.011 -0,32% 20.181 -4,26%
Drink
Information and Communication 46.311 46.309 0,00% 46.309 0,00% 46.894 46.894 0,00% 46.907 0,03%
Financial Services and Insurance 16.789 16.782 -0,04% 16.780 -0,05% 16.958 16.952 -0,04% 16.954 -0,02%
Real Estate 71.638 71.638 0,00% 71.638 0,00% 72.540 72.541 0,00% 72.562 0,03%
Company Services 4.876 4.875 -0,03% 4.875 -0,03% 4.938 4.937 -0,03% 4.938 0,00%
Government Administration, Defense and 26.543 26.521 -0,08% 26.414 -0,48% 26.878 26.857 -0,08% 26.756 -0,45%
Mandatory Social Security
Education Services 17.519 17.504 -0,09% 17.497 -0,13% 17.741 17.726 -0,08% 17.724 -0,09%
Health Services and Social Activities 15.625 15.608 -0,11% 15.585 -0,26% 15.825 15.807 -0,11% 15.788 -0,23%
Other Services 90.779 90.770 -0,01% 90.697 -0,09% 91.923 91.917 -0,01% 91.870 -0,06%
Total 914.491 914.430 -0,01% 921.723 0,79% 926.006 925.971 -0,00% 933.631 0,82%
*Taxable entrepreneurs can add input VAT burdens to product prices

When taxable entrepreneurs can transfer the input VAT burden to buyers
through the price component, it will increase the VAT base, which means it will
increase the potential for government revenue. The revenue potential of all
business sectors increases compared to the previous condition, this indicates a
cascading effect from unrecovered VAT input. The impact of alternative policies on
revenues is also higher, the increase in VAT revenue from the agricultural sector
being 0.36%, higher than the condition when the VAT input burden was borne by
producers. However, the cumulative policy impact is still negative, which means
that the increase in revenue due to the cascading effect is still lower than the
decrease in revenue due to the use of an alternative VAT basis.
Table 4.2. shows that the decrease of VAT revenue potential comes from
consumption activities due to a decrease in the base of VAT imposition, which
other values imposition base is smaller than the normal VAT provisions. This
means that consumers will enjoy a decrease in the effective VAT rate so that the
tax burden that must be borne by consumers is relatively lower in consuming a
commodity. Meanwhile, the increase of potential VAT revenue occurred in
production activities due to the addition of the effective VAT basis from non-
creditable input VAT. This additional VAT burden will be borne by producers as a
reduction in profits or will be transferred as an increase in selling prices which will
be borne by producers in the next stage to the final consumer.
When producers can transfer the input VAT burden to buyers, cascading effect
increases the potential state revenue. This increase comes from all economic
agents. However, the impact of alternative policies in the agricultural sector is still
negative on final consumption, while the business sectors and government
expenditure experience an increase in revenue potential. Consumers benefit from
this alternative policy, while producers have to bear more of the tax burden. This
condition worsens the VAT regressivity for taxable entrepreneurs, mostly those
with lower profits.
In the agricultural sector, the sub-sectors that are directly affected by
alternative policies have decreased potential VAT revenues from consumption
activities and increased VAT revenues from production activities, namely the food
crops, horticulture, plantation, and forestry. Meanwhile, the livestock, fishery, and
agricultural services which not covered by the alternative policies, the potential
VAT revenue from consumption activities is not affected, and the potential from
production activities decreases slightly, which is shown in table 4.3.
This alternative policy has the potential to reduce VAT base and revenue from
consumption activities and increase VAT base and revenue from production
activities for the business sectors covered by the policy. Business sectors outside
the scope of the policy will experience a decrease in potential VAT revenues from
production activities, while potential revenues from consumption activities will not
be affected. The greater the linkage of the business sector outside the scope of
the policy to the business sectors listed in the policy, the greater the negative
impact of decreasing potential VAT revenue from production activities.
This condition clarifies the negative effect of alternative policies on the agricultural
sector. Farmers/entrepreneurs in the agricultural sector must bear a greater tax
burden if they switch to using other value mechanisms. Meanwhile, entrepreneurs
in other business sectors who use certain agricultural products as production
inputs will benefit more because of the lower VAT burden they have to bear. This
condition increases the unfair VAT treatment for entrepreneurs in the agricultural
sector compared to entrepreneurs in other business sectors. Based on these
considerations, the general VAT provisions remain a better choice in creating a fair
and efficient VAT policy with minimum possible economic distortion.
Tabel 4.2. Policy Impact on Economic Agent (Billion Rp)
No Condition Full No Condition % Full %

Agent % Impact % Impact
Alternative 2021 Alternative Alternative* 2021* Impact* Alternative* Impact*
(a) (b) (b-a)/a (c) (c-a)/a (d) (e) (e-d)/d (f) (f-d)/d
Consumer 461.098 460.180 -0,20% 458.492 -0,57% 466.842 465.925 -0,20% 464.351 -0,53%
Producer 383.889 384.747 0,22% 393.727 2,56% 388.786 389.666 0,23% 398.879 2,60%
Government 69.503 69.503 0,00% 69.503 0,00% 70.378 70.380 0,00% 70.400 0,03%
Total 914.491 914.430 -0,01% 921.723 0,79% 926.006 925.971 -0,00% 933.631 0,82%
*Taxable entrepreneurs can add input VAT burdens to product prices
Tabel 4.3. Policy Impact on VAT Revenue Agriculture Sub-Sector
% Impact (VAT Rev. Condition 2021 vs No % Impact (VAT Rev. Full Alternative vs No
Agriculture Sub-Sector Alternative) Alternative)
Consumer Business Total Consumer Business Total
Crops -0,21% -0,01% -0,06% -89,44% 48,32% 11,07%

Horticultural Plants -5,84% 0,14% -0,03% -90,00% 5,08% 2,29%
Plantation crops -1,03% 1,80% 1,70% -90,00% 170,24% 160,58%
Forestry and Logging -0,97% 3,37% 2,83% -90,00% 174,34% 141,37%
Farm 0,00% -0,19% -0,19% 0,00% -1,15% -1,13%
Fishery N/A -0,17% -0,17% N/A -0,30% -0,30%
Agricultural and Hunting Services 0,00% -0,09% -0,09% 0,00% -0,20% -0,18%
Total -0,79% 0,42% 0,35% -75% 48% 42%

However, under certain circumstances, this alternative policy is needed and
provides benefits for farmers/small entrepreneurs, due to limited administrative
capabilities and understanding of tax provisions, unable to implement the normal
provisions of VAT properly. When the benefits of crediting input VAT are less than
the administrative and compliance costs that must be borne, then simpler
provisions are the worthy choice.
4.3. Policy Impact on Compliance Rate

The gap between actual revenue and potential revenue indicates a compliance
rate. The compliance rate is determined by complex interaction factors such as tax
policy regime, general tax culture, the effectiveness of tax administration, and
mainly, tax evasion and avoidance tendencies (Le, 2007). Potential revenue is
estimated using taxable proportion, proportion below VAT threshold, and
proportion using alternative methods determined based on data and applicable
VAT policies. Meanwhile, not all Taxable Entrepreneurs comply in paying and
reporting all taxable transactions that occur. So it is possible that the realization of
VAT payments differs from the potential revenue that should be collected. The
imbalance condition from both value reflects the level of taxpayer compliance.
The compliance rate is calculated by dividing actual revenue realization with
estimated potential revenue in a full compliance rate (Sugana & Hidayat, 2014).
Although the potential revenue is estimated using a static model and does not take
into account changes in taxpayer behavior, but the actual revenue realization will
show the taxpayer's response to VAT policy changes, ceteris paribus. The actual
revenue from each sector is obtained based on the VAT payments of each KLU
sourced from DGT administrative data. The first step is to match 1626 sectors
based on KLU with 185 sectors in the I-O table. This data is the realization of gross
VAT revenue because it has not been deducted by the VAT overpayment that must
be returned to the taxpayer. So that this data can reflect the rate of compliance of
each business sector. Furthermore, it then is grouped into 17 sectors as presented
in table 4.4.
The policy impact on the compliance rate can be seen by comparing the results
of the compliance rate calculation of each sector among tax years. The year in
which the latest I-O tables exist (2016) is suggested as the base year (Le, 2007).
The compliance rate in 2020 is also used as a comparison because it is the closest
year when alternative policies have not been fully put into practice, and 2019 before
the impact of the COVID-19 pandemic in Indonesia. The compliance rate of the
agricultural sector in 2021 is 35.5%, higher than the compliance rate in previous
years. This shows that alternative policies that are fully implemented in 2021 have
a positive impact on increasing the compliance rate in the agricultural sector,
ceteris paribus.
There was a decrease in the compliance rate of the agricultural sector in 2020 and
2019 compared to the base year 2016, which was influenced by the outbreak of
COVID-19 at the end of 2019 in several countries and began to enter Indonesia in
early 2020 and getting worse globally. Due to the interconnectedness of global
economic chains, economic changes due to a pandemic in a country will affect
domestic economic conditions. For example, a decrease in demand for agricultural
commodities in the importing country will have an impact on the supply and prices
of domestic products. Thus, changes to the VAT base will affect potential VAT
revenues. In addition, the uncertainty of the pandemic impact can affect changes
in taxpayer behavior, which affects the actual VAT payment.
Tabel 4.4. Compliance Rate
Sector 2021 2016 2019 2020
Agriculture, Forestry and Fisheries 35,46% 27,72% 23,73% 19,75%

Crops 1,98% 0,96% 1,08% 1,09%
Horticultural Plants 3,36% 1,90% 2,32% 1,91%
Plantation crops 84,39% 61,20% 50,75% 46,02%
Forestry and Logging 113,83% 154,72% 120,94% 89,52%
Farm 1,16% 2,09% 2,60% 0,63%
Fishery 11,48% 13,02% 17,97% 11,10%
Agricultural and Hunting Services 63,72% 53,42% 61,08% 57,14%
Mining and excavation 162,20% 114,17% 113,00% 96,01%
Processing industry 71,58% 72,30% 71,09% 62,22%
Electricity and Gas Supply 28,69% 52,63% 48,44% 31,00%
Water Supply, Waste Management, Waste and Recycling 81,20% 66,17% 76,53% 63,51%
Construction 88,53% 121,01% 111,44% 54,67%
Wholesale and Retail Trade; Car and Motorcycle Repair 156,54% 141,12% 144,82% 119,34%
Transportation and Warehousing 28,36% 22,76% 22,39% 24,78%

Provision of Accommodation and Food and Drink N/A N/A N/A N/A
Information and Communication 39,28% 45,35% 36,30% 32,38%
Financial Services and Insurance 42,61% 52,15% 41,36% 37,19%
Real Estate 13,40% 17,19% 15,17% 11,46%
Company Services 55,23% 69,99% 64,68% 51,33%
Government Administration, Defense and Mandatory Social Security 3632,95% 4280,12% 4361,33% 1933,15%
Education Services N/A N/A N/A N/A
Health Services and Social Activities N/A N/A N/A N/A
Other Services 2,77% 3,66% 3,79% 2,85%
Total 65,59% 66,13% 64,44% 53,57%

The compliance rate in total for 2021 is 65.6% shows that there is still a lot of
potential VAT revenue that can be increased. This figure is lower than the
compliance rate for the base year 2016 (66.1%). This condition can be interpreted
that the additional complexity of the VAT system administration due to alternative
policy has a negative impact on the total compliance rate. In many cases, the
compliance rate is substantially below 100%, but for some sectors, the compliance
rate is above 100%. This may occur due to the intensification programs of tax
collection or an increasing number of taxpayers. In addition, the I-O table only
describes formal economic activities, while the potential VAT from informal
activities is difficult to estimate. So that the actual revenue of a sector can exceed
the estimated revenue potential. Besides, the weakness of the I-O model which
assumes a sector only produces one type of homogeneous output, when in fact it
is possible to have more than one business activity with different VAT treatment.
For example, the accommodation provision sector does not have the potential VAT
revenue because the hotel services delivered are exempt from VAT, but it is
possible to have other businesses, such as renting space for ATMs, which are
subject to VAT.
The plantation crops experienced the greatest increase in compliance rate and
had the greatest potential for VAT revenue compared to other agricultural sub-
sectors. The existence of superior commodities such as palm oil and rubber, as
well as the substantial number of large and medium companies entering this
sector, support the higher rate of compliance and VAT revenues. Meanwhile, the
livestock, food crops, and horticulture have a low level of compliance, which can
be caused by the high proportion of informal entrepreneurs that limits DGT to carry
out supervision due to data limitations. In addition, these sectors also received
many VAT exemptions which caused the revenue potential to decrease.
5. CONCLUTIONS
Based on the results and previous analysis, several things can be concluded
that the alternative tax base measurement policy has a positive impact on
increasing VAT revenue in the agricultural sector, which is 0.35% in 2021, and
increase compliance level of the agricultural sector which reach 35.5% in 2021. An
easier and fairer VAT policy can increase voluntary compliance in the agricultural
sector so that more entrepreneurs will be obedient in fulfilling their tax obligations.
But the policy reduces the potential for national VAT revenue by -0.01% due to a
decrease in the potential VAT revenue that occurs in other business sectors. The
additional administration complexity of the VAT system due to the alternative policy
has a negative impact on the aggregate compliance rate.
The increase in revenue from business activities indicates an additional VAT
burden for producers which reduces disposable profit. It will exacerbate the
regressivity of VAT mostly for lower profit producers. Although the input VAT
burden can be transferred to the buyer through the selling price of the product,
there is an unfair treatment of VAT on agricultural products compared to other
sector products or imports that are free from VAT on production inputs. That will
have an impact on the level of productivity and competitiveness of the agricultural
sector compared to other business sectors.

6.1. Implications and Recommendations
Based on the conclusions above, it can be seen that the agricultural sector's
compliance rate of 35.5% indicates that this sector still has a lot of potential for
intensification and extensification to increase VAT revenues. Some policies can be
applied, among others, by providing simple and fair VAT administration, increasing
tax knowledge of taxpayers, and focusing on monitoring sectors with high revenue
potential and low levels of compliance.
Returning to the normal VAT mechanism is a preferred choice because of
these strong considerations: the small economy of scale and uneven alternative
policy impact on VAT revenue; the addition of VAT regressivity and unfair VAT
treatment for producers in the agricultural sector; and an increasingly complex VAT
system due to differences in VAT treatment. The normal VAT mechanism,
theoretically, can generate greater aggregate VAT revenue with minimal economic
activity distortion.
However, the VAT best practice as a stable and efficient source of revenue
through applying a single tariff and a broad tax base requires ideal conditions with
as few administrative limitations as possible. Meanwhile, the agricultural sector, at
least for now, has not been able to fulfill these ideal conditions due to the problem
of informality. Thus, alternative policies are still needed during the transition period
to accommodate farmers/agricultural entrepreneurs constrained by administrative
issues and limited understanding of tax provisions. The simple alternative method
can increase the number of farmers/entrepreneurs in the agricultural sector
involved in the VAT system. Thus monitoring and increasing tax knowledge of
taxpayers become more possible to do. In this case, the alternative policy is
regulated as a temporary policy with specific criteria and a limited time allowed to
use the alternative method.
6.2. Limitations
The model used is static and does not explicitly take into account changes in
behavior in response to policy changes. This limitation arises from the assumptions
used in preparing the input-output table, which makes the input coefficients or
technical coefficients assumed to be constant during the analysis or projection
period. The model cannot fully describe the actual conditions, so it is required to
set several assumptions to accommodate these limitations. The process of
generating the Input-Output table is assumed to be the same as the process of
imposing VAT. The criteria for determining the classification of business sectors in
the input-output table are assumed to be the same as the criteria for Classification
of Business Fields (KLU) determined by the DGT. The sample of DGT
administration data used in the model is assumed to represent actual conditions at
the population level, although there may be overstated/understated actual
conditions.
Each business sector in the input-output table and KLU only produces one type
of homogeneous output, although in actual conditions it possibly produces other
outputs outside the main business. This assumption causes differences in the
results of the estimated revenue potential based on the model with the actual
revenue based on administrative data. An example is the hotel sector, which main
output is providing accommodation, which is exempt from VAT because it is
subjected to local tax, the model will estimate zero potential VAT revenue.
Meanwhile, revenue based on actual data from the DGT administration allows for
VAT revenue from activities outside the main business, such as renting space for
ATMs or shops.
REFERENCES
Agha, A., & Haughton, J. (1996). Designing VAT Systems: Some Efficiency
Considerations. The Review of Economics and Statistics, 303-308.
Aizenman, J., & Jinjarak, Y. (2008). The collection efficiency of the Value Added Tax:
Theory and international evidence. The Journal of International Trade &
Economic Development, 17(3), 391-410.
Alm, J., Jackson, B. R., & McKee, M. (1992). Estimating the determinants of taxpayer
compliance with experimental data. National Tax Journal, 107-114.
Baldry, J. C. (1987). Income tax evasion and the tax schedule: Some experimental
results. Public Finance= Finances publiques, 42(3), 357-383.
Bird, R., & Gendron, P.-P. (2006). Is VAT the best way to impose a general
consumption tax in developing countries? Bulletin for International
Taxation, 287.
Bogetic, Z., & Hassan, F. M. (1993). Determinants of Value-Added Tax Revenue: A
Cross-section Analysis. World Bank Publications.
Braithwaite, V. (2003a). Dancing with Tax Authorities: Motivational Postures and
Non-Compliance. Taxing Democracy: Understanding Tax Avoidance and Tax
Evasion, 1-11.
Clotfelter, C. T. (1983). Tax evasion and tax rates: An analysis of individual returns.
The Review of Economics and Statistics, 363-373.
Ebrill, L., Keen, M., Bodin, J. P., & Summers, V. (2001). The Modern VAT. Washington,
D.C.: International Monetary Fund.
Iswahyudi, H. (2018). Where has the money gone?: The case of Value Added Tax
revenue performance in Indonesia. Munich Personal RePEc Archive, 1-29.
James, K. (2015). The Rise of the Value Added Tax. New York: Cambridge University
Press.
Jenkins, G. P., Kuo, C.-Y., & Shukla, G. (2000). Tax Analysis and Revenue Forcasting.
Cambridge, Massachusetts: Harvard Institute for International
Development.
Kirchler, E., & Maciejovsky, B. (2001). Tax compliance within the context of gain and
loss situations, expected and current asset position, and profession. Journal
of Economic Psychology, 22(2), 173-194.
Kirchler, E., Hoelzl, E., & Wahl, L. (2008). Enforced versus voluntary tax compliance:
The “slippery slope” framework. Journal of Economic Psychology, 29(2), 210-
225.
Le, T. M. (2007). Estimating the VAT base: Method and application. Tax Note
International, 46(2), 203.
Manasan, R. G. (2003). Estimating industry benchmark for the value added tax.
Philippine Journal of Development, 71.
Marks, S. V. (2003). The value added tax in Indonesia: The impact of sectoral
exemptions on revenue potential and effective tax rates. Jakarta:
Nathan/Checchi Joint Venture Partnership for Economic Growth (PEG)
Project∗Under USAID Contract #497-C-00-98-00045-00 (Project #497-
0357).
Park, C. G., & Hyun, J. K. (2003). Examining the determinants of tax compliance by
experimental data: A case of Korea. Journal of Policy Modeling, 25(8), 673-
684.
Sugana, R., & Hidayat, A. (2014). Analisis Potensi dan Kesenjangan Penerimaan
Pajak Pertambahan Nilai di Indonesia Tahun 2013. Jurnal Ekonomi dan
Pembangunan Indonesia, 1-40.
Warwick, R., Harris, T., Phillips, D., Goldman, M., Jellema, J., Inchauste , G., & Goraus-
Tan´ ska, K. (2022). The redistributive power of cash transfers vs VAT
exemptions: A multi-country study. World Development, Volume 151,
March 2022, 105742.
Wenzel, M., & Thielmann, I. (2006). Why we punish in the name of justice: Just
desert versus value restoration and the role of social identity. Social justice
research, 19(4), 450-470.
FORECASTING VALUE-ADDED TAX (VAT) REVENUE
USING AUTOREGRESSIVE INTEGRATED MOVING
AVERAGE (ARIMA) BOX-JENKINS METHOD
Muchamad Irham Fathonia, Akbar Saputrab

a Directorate General of Taxes, Indonesia. Email: irham.fathoni@kemenkeu.go.id
b Directorate General of Taxes, Indonesia. Email: akbar.saputra@kemenkeu.go.id
ABSTRACT
We propose a method to forecast Value-Added Tax (VAT) revenue for Indonesia government using
Autoregressive Integrated Moving Average (ARIMA) Box-Jenkins method. ARIMA Box-Jenkins
method, introduced by statisticians George E. P. Box and Gwilyam M. Jenkins, is a statistical
method that works as time series forecasting model based on the behavior of the observed variable
data. We experimented the ARIMA Box-Jenkins method using time-series analysis of VAT revenue
data of two Tax Offices—Surabaya Karangpilang Tax Office and Jakarta Setiabudi Dua Tax Office—
of Directorate General of Taxes (DGT) from the year 2017 until 2021. The result shows that
forecasted VAT revenue resembles the real VAT revenue more closely than when compared to the
actual VAT target by Indonesia government. We then argue that this result may be used as a fail-
safe tax revenue target, that can work as a tool to better measure DGT performance and a starting
point for tax intensification strategy.
Keywords: ARIMA, Box-Jenkins, value-added tax, tax revenue forecast
Kami mengajukan suatu cara untuk meramalkan penerimaan Pajak Pertambahan Nilai (PPN) bagi
pemerintah Indonesia, menggunakan metode Autoregressive Integrated Moving Average (ARIMA)
Box-Jenkins. Metode ARIMA Box-Jenkins, diperkenalkan oleh ahli statistik George E. P. Box dan
Gwilyam M. Jenkins, adalah suatu metode statistik yang berguna sebagai model peramalan
berbasis waktu berdasarkan perilaku data yang diobservasi. Kami bereksperimen dengan metode
ARIMA Box-Jenkins menggunakan analisis time-series atas data penerimaan PPN dari dua Kantor
Pelayanan Pajak—Kantor Pelayanan Pajak Pratama Surabaya Karangpilang dan Kantor Pelayanan
Pajak Pratama Jakarta Setiabudi Dua—pada Direktorat Jenderal Pajak (DJP) dari tahun 2017
sampai dengan 2021. Hasilnya menunjukkan bahwa nilai ramalan penerimaan PPN dari metode
tersebut lebih mendekati nilai penerimaan PPN sesungguhnya, daripada ketika dibandingkan
dengan target penerimaan PPN aktual yang ditetapkan pemerintah. Selanjutnya, kami berpendapat
bahwa hasil ramalan ini dapat digunakan sebagai target fail-safe atas penerimaan PPN, yang dapat
berfungsi sebagai alat untuk mengukur kinerja DJP dengan lebih baik dan sebagai titik permulaan
penentuan strategi intensifikasi pajak.
Kata kunci: ARIMA, Box-Jenkins, pajak pertambahan nilai, ramalan penerimaan pajak
1. INTRODUCTION
In the first semester of 2022, Directorate General of Taxes (DGT) of the Ministry of
Finance of Indonesia recorded 55.7% growth of tax revenue, compared to that of the
same period in the previous year (Kurniati, 2022). This translates to 58.5% of tax revenue
target for 2022 has been collected. This seems like good news, complementary to a
record that DGT finally broke in 2021 for achieving 103.9% of tax revenue targeted in
the National Budget—a record that has not been broken for the last twelve years.
1
However, for the second semester of 2022, with the enactment of President Decree
No. 98 of 2022, tax revenue target is revised to grow to IDR 1,783 billion—18.1% higher
than the initial target, which was IDR 1,510 billion (Wildan, 2022). Despite some factors
that may still accelerate tax revenue collected (i.e., Income Tax revenue collected in tax
return period ended in March to April, DGT’s Voluntary Disclosure Program that ran from
January to June, and global commodity price boom), there is still a long way for DGT to
achieve this designated target.
In the realm of tax administration in Indonesia or any countries in the world, national
budget revision—let alone tax revenue target revision—is not new. Even in Indonesia,
tax revenue target has been regularly revised in every budget year due to its dynamic
nature and its proximity to how the economy changes. VAT, as the second biggest
component of tax revenue after Income Tax, is one of the most dynamically changed
type of tax because it highly reflects the economy and national goods and services
transactional volumes. While no research found negative effects of tax revenue target
revision, for Indonesia (and especially DGT) this might highlight a structural issue.
Currently, breaking down national tax revenue target per month and for each vertical
unit in DGT is still done conventionally, meaning that there is no specific method used to
allocate national tax revenue target per month for each Tax Offices in DGT. This may
result in biases, because after certain period, even tax offices in the same region may
show significant differences in terms of tax revenue collected. Regarding performance
measurements, this also addresses a new issue: after the enactment of President
Decree No. 96 of 2017, performance benefits are given to DGT employee based on
organizational performance achievement, which factors the percentage of tax collected
compared to target. Thus, if tax revenue target is revised and broken down without
specific scientific calculation, Tax Offices may find it hard to achieve tax revenue target
allocated to them and eventually fail to achieve their targeted organizational
performance—resulting in decrease in performance benefits received by employees of
said Tax Offices.
More than biases affecting tax revenue target breakdown and performance
measurement, a solid tax revenue target is crucial to implement an appropriate and
measurable intensification strategy. With proper forecasting to determine accurate tax
revenue target, it will be helpful for DGT’s vertical units to determine the direction of tax
supervision strategies and potential revenue measurements in each tax period. If in one
period there is a shortfall of tax revenue collected from the predetermined target, DGT
may re-evaluate its tax intensification and supervision strategy to achieve said target in
the remaining months of the year.
With that said, we proposed a forecasting method called Autoregressive Integrated
Moving Average (ARIMA) to determine tax revenue target for DGT. Forecasting is
basically a conjecture or estimate of future events, and forecasting is an important tool
in effective and efficient planning (Makridakis et al., 1999). ARIMA Box-Jenkins
method—named after statisticians George Box and Gwilym Jenkins—is a time series
forecasting technique that is only based on the behavior of observed variable data (Box
& Jenkins, 1970). The ARIMA model completely ignores the independent variables
because this model uses the present value and past values of the dependent variable to
produce accurate short-term forecasts (Cryer & Chan, 2008).
2
In this research, we use ARIMA Box-Jenkins method to specifically determine VAT
revenue target for DGT. VAT is chosen because it contributes up to 33% of total tax
revenue collected every year. As we said in previous paragraphs, VAT revenue target
changes dynamically following the economic situation and the goods and services
transactional activities. However, in this case, we propose ARIMA Box-Jenkins method
to see how VAT revenue can be projected in ceteris paribus. From the result of our
forecast, we further elaborate its effectiveness and accuracy, as well as its usefulness
as a fail-safe VAT revenue target that goes side-by-side with the official VAT revenue
target specified in the National Budget.
2.1. Tax Collection and Revenue Target in DGT
Generally, tax revenue target is calculated based on the tax buoyancy or tax elasticity
of economic growth and tax revenue growth (Direktorat Jenderal Pajak, 2021).
Calculating tax revenue target this way is relatively simple because it only uses historical
economic growth and elasticity data, as well as assumptions of economic growth in the
next fiscal year, as main indicators. Tax revenue target calculation can also be compiled
from a combination of various methods on a micro scale. Potential tax revenue, on the
other hand, is measured based on the expansion of the tax base and the dynamics of
certain economic sectors, types of taxes, and regions. Using this micro-scale approach,
the methods and indicators used to calculate tax revenue targets can vary widely.
Currently, allocating tax revenue targets to each of DGT’s vertical units tends to be
based on historical data adjusted with regional micro-scale parameters such as new
taxation bases, business growth, and regional economic projected growth. That said,
there is no quantitative calculations or methods used to allocate tax revenue target per
unit or per region more comprehensively and more contextually. It is considered
necessary to develop a tax revenue target based on a quantitative calculation using
econometric analysis in the form of time series data. By looking at the pattern of how
various factors affect tax revenues in certain regions, tax revenue allocated for certain
DGT vertical units can be projected. The variables used may also vary, such as economic
growth, income per capita, inflation, commodity prices, business growth, and so on.
These methods can be referred to as top-down macroeconomic data-based methods.
However, it is also necessary to set tax revenue targets based on conditions that are
not affected by any variables (ceteris paribus). This is to show the pattern that will be
projected based on historical data and can be used as a fail-safe tax revenue target,
aside from the official tax revenue target.
2.2. Forecasting Methods
Forecasting is crucial in policy formulation, due to the time lag between when the
policy is formulated and when it will be implemented. Forecasting must be performed
with certain principles, i.e., forecasting involves errors, forecasting should use a
benchmark for forecasting errors, and short-term forecasting is more accurate than long-
term. There are also steps to follow while forecasting, i.e., defining the purpose of
forecasting, creating a data plot diagram, selecting the right forecasting models,
calculating forecasting errors, and choosing the best forecasting method with the
smallest error.
3
There are two forecasting methods: qualitative and quantitative. Qualitative method
is used where there is no mathematical model, usually because the existing data is not
representative enough to predict the future (long term forecasting). Quantitative method
is based on the availability of raw data, accompanied by a series of mathematical rules
to predict future results. Quantitative methods are divided into 3 types:
a. Model Time Series Analysis (Time Series)
b. Regression Models
c. Econometric Model
One method that can be used to forecast tax revenue targets is Time Series Analysis
introduced in 1970 by George E. P. Box and Gwilyam M. Jenkins through their book
Time Series Analysis: Forecasting and Control. Time series can be interpreted as a
series of data obtained based on observations of an event in the order in which it
occurred. The time of the incident can be a period in units of seconds, minutes, hours,
days, months, years, and other time periods, all of which are a series of observational
data based on the time of the incident with a certain time interval which is better known
as a time series (Cryer & Chan, 2008).
The rationale for the Time Series is that the current observation (Zt) depends on one
or more previous observations (Zt-k). In other words, time series is made because
statistically there is a correlation (dependency) between series of observations. To see
the dependencies between observations, we can perform a correlation test between
observations which is often known as the autocorrelation function (ACF). In ACF, each
observation is expressed as a random variable Zt obtained on a certain time index (ti) as
a sequence of observations, so that the time series data is written Zt1, Zt2, Zt3, …, Ztn. In
the time series method, there are several things that need to be considered such as the
stationarity of the data, the autocorrelation function, and the partial autocorrelation
function.
2.3. ARIMA
ARIMA is one of time-series models. This model consists of AR (Autoregressive),
MA (Moving Average), or ARMA (Autoregressive Moving Average) components, and if
the data is not stationary in its mean, a differencing process is carried out. ARIMA Box-
Jenkins model is one of the time series model forecasting techniques that is only based
on the behavior of the observed variable data.
The condition that must be met to make a forecasting model is that the data is
stationary. AR model will be used to find any relations between current value and
previous value by adding arbitrary values, while MA model will be used to find any
relations between current value and previous residual value (Wei, 2006). The
autoregressive process, as the name implies, is a self-regressive process. The general
form of a p or AR(p) level autoregressive process is
𝑍𝑡 = 𝜙1 𝑍𝑡−1 + 𝜙2 𝑍𝑡−2 + ⋯ + 𝜙𝑝 𝑍𝑡−𝑝 + 𝑎𝑡
which shows that the present value of a process can be expressed as the weighted sum
of past p values plus one current random error. In this case, it is assumed that at is
independent of Zt-1, Zt-2, etc. So, it can be seen that Zt is regressed on the past p value
of Z.
4
The identification of ARIMA Box-Jenkins model can be used as a step in identifying
the non-stationary model. If the data is not stationary in its mean, then it should be
differenced and if it is not stationary in its variance then it can be transformed by Box-
Cox. After the data is stationary in both its mean and variance, the next step is to plot
the Autocorrelation Function (ACF) and Partial Auto Correlation (PACF) which are used
to identify the initial ARIMA model.
3.1. Research Methods, Variables, and Formulas
This research does not try to test any hypotheses, so it uses a qualitative approach
to find its conclusion. However, in the process, we use quantitative formulas using
numerical data analysis and statistical models. Box-Jenkins method is used to define
suitable ARIMA model for our time-series data (Daniel, 1989), and it requires data to be
stationary before getting processed. With AR and MA model integrated in our ARIMA
model, we utilize degree of AR (p), degree of differences (d), and degree of MA (q) to
create a function ARIMA (p,d,q) as follows:
𝑌𝑡 − 𝑌𝑡−1 = 𝜑0 + 𝜑1(𝑌𝑡−1−𝑌𝑡−2) + ⋯ + 𝜑𝑝(𝑌𝑡−𝑝−𝑌𝑡−𝑝−1) + 𝜀𝑡 − 𝜔1𝜀𝑡−1 − 𝜔2𝜀𝑡−2 − ⋯ − 𝜔𝑞𝜀𝑡−𝑞
where
𝑌𝑡 = variable of time t,
𝜑0 = constant value,
𝜑𝑝 = p-th AR parameter,
𝜔𝑞 = q-th MA parameter,
𝜀𝑡 = error value of time t, and
𝜀𝑡−1,−2,…,𝜀𝑡−𝑞 = previous error values on related time series.
The initial step of our analysis is determining a forecasting model suitable for the VAT
revenue data, using time-series method with the following steps:
a. creating time series plot on the VAT revenue data, to check if there are any seasonal
patterns;
b. performing Box-Cox method to identify whether the data has stationarity problems in
its variance and/or mean—if the data is not stationary in its mean, then differencing
is performed, but if it is not stationary in variance, then data transformation is
performed;
c. identifying data stationarity through Autocorrelation Function (ACF) and Partial
Autocorrelation Function (PACF) plots, which are used to check whether the data is
finally stationary after transformation and differencing, as well as to find suitable
models;
d. performing initial ARIMA (p,d,q) model estimation, which consists of three stages—
identifying model, assessment and testing, and running the model—and results in a
temporary model used for later estimation and diagnostic checking;
e. performing parameter estimation based on the temporary model;
f. performing diagnostic checking;
g. selecting the best model.
5
The accuracy and effectiveness of the forecasting method can be seen from the
difference between the forecast value and the actual value. Two methods are used to
measure forecast accuracy:
a. MAPE (Mean Absolute Percentage Error) measures the accuracy of the estimated
model value expressed in the form of the average absolute percentage error and can
be expressed as follows:
1 𝑛 |𝑌 − 𝑌
𝑡
̂𝑡 |
𝑀𝐴𝑃𝐸 = ∑
𝑛 𝑡=1 𝑌𝑡
where
𝑌𝑡 = actual value on year t
𝑌̂𝑡 = forecast value on year t

𝑛 = number of data
b. MSE (Mean Squared Error) measures the accuracy of the estimated value of the
model expressed in the average square of the error and can be expressed as follows:
1 𝑛 2
𝑀𝑆𝐸 = ∑ (𝑌𝑡 − 𝑌̂𝑡 )
𝑛 𝑡=1
where
𝑌𝑡 = actual value on year t
𝑌̂𝑡 = forecast value on year t

𝑛 = number of data
In addition to MAPE and MSE, we also used Root-Mean-Square Error (RMSE) and
Ljung-Box test to measure accuracy. RMSE, as the name suggests, is the root of MSE,
and often preferred to MSE as it is on the same scale as the data (Hyndman & Koehler,
2006). Ljung-Box test is a statistical measure to test if there are some autocorrelation
groups in a time-series data that value different from zero (Ljung & Box, 1978).
3.2. Research Scope
Due to limitation and unavailability of broader data, we use VAT revenue data from
only two Primary Tax Offices in DGT, which are Surabaya Karangpilang Tax Office and
Jakarta Setiabudi Dua Tax Office. These two tax offices are located in two big cities in
Indonesia, which are Surabaya and Jakarta, respectively. These two tax offices run the
same responsibilities, meaning that there are no special responsibilities or functions
given to these tax offices that may make them different from other tax offices in DGT.
We take monthly VAT revenue data of 2017 to 2022 from both offices. To better
represent the true VAT revenue, we only take regular VAT payments in consideration
and omit VAT payments due to fines or sanctions, or VAT payments withheld by the
government, treasurers, and other VAT collectors.
6
In this study, forecasting will be carried out using VAT revenue data from Surabaya
Karangpilang Tax Office and Jakarta Setiabudi Dua Tax Office. To assist us, we utilized
Minitab for data visualization, as well as EViews and SPSS for data processing.
4.1. Forecasting VAT Revenue in Surabaya Karangpilang Tax Office
We observed Surabaya Karangpilang Tax Office’s VAT revenue for the period
January 2017 to December 2021, consisting of 60 observations. The first step of this
research is to determine the forecasting model in accordance with the data on the
amount of Value Added Tax receipts analyzed using the time series method. Value
Added Tax revenue data is included in time series data that can be predicted using the
ARIMA method with the following stages of analysis.
a. Identifying Models
1) Data Plotting
We plotted VAT revenue data from Surabaya Karangpilang Tax Office, as shown on
Figure 5.
Figure 5: Surabaya Karangpilang Tax Office VAT Revenue Data Plot
Figure 5 shows that the data was not stationary, because a trend appears in the plot.
Non-stationary data in variance can be seen from series plot, where points are not
spread out evenly because they increase or decrease with time. We have to transform
non-stationary data to be stationary. Diagnostic checking using Box-Cox method showed
a rounded value of 0.50 and data interval between -0.46 and 1.29, or less than 1. We
transformed this non-stationary data until we got a rounded value of 1.
7
Figure 6: Box-Cox Plot Before and After Transformation
2) Data Stationarity and Differentiation

We then performed stationarity test in mean for this data using data plotting. Again,
we found trend in the data, so we have to transform it using differentiation. Figure 7
shows time-series data plot after transformation and differentiation. From here, we found
that the data has already been stationary.
Figure 7: Time-series Data Plot After Differentiation 1
3) ACF and PACF plot

We continue our testing by further performing stationarity test in mean using ACF
and PACF plot. The result is shown in Figure 8. From that figure, we can visually analyze
that the data is not stationary in mean, because we found that some ACF plot appeared
to die down. We can also tell from Figure 8 that the plot does not fluctuate constantly
around parallel lines. Thus, it is necessary for us to perform differentiation. The result of
the data after differentiation is shown in Figure 9.
8
Figure 8: ACF and PACF Plot
Figure 9: ACF and PACF Plot After Data Differentiation
Figure 9 shows that data plot is now fluctuating constantly along the mean line. Thus,
we conclude that the data is now stationary in mean. We can move forward by finding
the best and most feasible model to do forecasting. In Figure 9, we see that ACF bar
crosses the line on lag 1 and PACF bars show sinusoidal pattern but crosses the line on
lag 1. From this, we find several possible models, i.e., ARIMA (1,1,1), ARI (1,1) and IMA
(1,1). However, considering the time-series nature of VAT revenue data, we predict that
the more feasible models are ARIMA (0,1,1), ARIMA (1,1,1), and ARIMA (1,1,0). From
here, we can continue to estimate the parameters.
b. Testing Model Feasibility

After predicting three feasible forecasting models, we need to calculate the criteria of
the best model among those three. The output of this calculation is shown in Table 11.
Table 11: Comparison of Statistical Value of RMSE, MAPE, MAE, and Ljung-Box
of Three ARIMA Models (row in bold shows the best model)
Model RMSE MAPE MAE Ljung-Box
ARIMA (0,1,1) 8,403 x 109 18,178 6,409 x 109 54,736
9 9
ARIMA (1,1,1) 8,359 x 10 17,755 6,248 x 10 46,467
ARIMA (1,1,0) 8,750 x 109 18,265 6,623 x 109 74,236
9
From three outputs, we can find the best model by comparing the statistical value of
RMSE, MAPE, MAE, and Ljung-Box of three feasible ARIMA models. The smaller the
statistical value, the more feasible and suitable the model with our observation data. In
Table 11, we can see that the smallest statistical value came from ARIMA (1,1,1). Thus,
ARIMA (1,1,1) is considered the best model for our time-series data.
Further feasibility testing on ARIMA (1,1,1) shows MAPE of 17.755, which is sufficient
considering it is still on the range between 10 to 20. We then estimate ARIMA (1,1,1)
mode parameters, as shown in Table 12
Table 12: ARIMA (1,1,1) Model Parameters
Estimate SE t Sig.
VAT- No Trans- Constant -52,302,636.085 283,614,207.478 -0.184 0.854
Model_ formation AR Lag 1 0.308 0.182 1.690 0.097
1 Difference 1.000
MA Lag 1 0.835 0.110 7.604 0.000
Using the following time-series equation of ARIMA (1,1,1):

𝑍𝑡 = 𝜇 + 𝜑1 𝑍𝑡−1 + ⋯ + 𝜑𝑝 𝑍𝑡−𝑝 + 𝑎𝑡 − 𝜓1 𝑎𝑡−1 − ⋯ − 𝜓𝑞 𝑎𝑡−𝑞
we can substitute estimate value from Table 12, with 𝜑𝑝 as coefficient of AR(p) and 𝜓𝑞
as coefficient of MA(q), as follows
𝒁𝒕 = −𝟓𝟐. 𝟑(𝟏𝟎𝟔 ) + 𝟎. 𝟑𝟎𝟖𝒁𝒕−𝟏 + 𝒂𝒕 − 𝟎. 𝟖𝟑𝟓𝒂𝒕−𝟏
However, because the model is differentiated to lag 1, we expand 𝑍𝑡 = 𝑌𝑡 − 𝑌𝑡−1 , as

follows:
𝑌𝑡 − 𝑌𝑡−1 = −52.3(106 ) + 0.308(𝑌𝑡−1 − 𝑌𝑡−2 ) + 𝑎𝑡 − 0.835𝑎𝑡−1
𝒀𝒕 = −𝟓𝟐. 𝟑(𝟏𝟎𝟔 ) + 𝟏. 𝟑𝟎𝟖 𝒀𝒕−𝟏 − 𝟎. 𝟑𝟎𝟖 𝒀𝒕−𝟐 + 𝒂𝒕 − 𝟎. 𝟖𝟑𝟓𝒂𝒕−𝟏
Thus, our ARIMA (1,1,1) model equation.
c. Forecasting
Using our ARIMA (1,1,1) equation, we can now start to forecast VAT revenue of
Surabaya Karangpilang Tax Office throughout 2022. The result of our forecast can be
seen in Table 13.
Table 13: VAT Forecasted Revenue of Surabaya Karangpilang Tax Office
Year Month VAT Forecasted Revenue
2022 1 Rp 32,749,473,639
2022 2 Rp 31,859,834,811
2022 3 Rp 31,593,374,692
2022 4 Rp 31,513,565,906
2022 5 Rp 31,489,661,983
2022 6 Rp 31,482,502,401
2022 7 Rp 31,480,357,999
10
4.2. Forecasting 2022 8 Rp 31,479,715,718 VAT Revenue in
Jakarta 2022 9 Rp 31,479,523,346 Setiabudi Dua
Tax Office 2022 10 Rp 31,479,465,727
We can 2022 11 Rp 31,479,448,470 repeat the same
process to 2022 12 Rp 31,479,443,301 forecast VAT
revenue in Total Rp 379,566,367,993 Jakarta
Setiabudi Dua Tax Office.
Observation data still ranges for the same period, which is January 2017 until December
2021, resulting in 60 observations.
a. Identifying Models
1) Data Plotting and Stationarity
We plotted VAT revenue data from Jakarta Setiabudi Dua Tax Office to detect its
stationarity, as shown on Figure 10.
Figure 10: Jakarta Setiabudi Dua Tax Office VAT Revenue Data Plot
From Figure 10, we can tell that—different from Surabaya Karangpilang VAT revenue
data—Jakarta Setiabudi Dua data is relatively more stationary in variance. Thus, we
skipped differentiation and move forward to ACF and PACF plotting.
2) ACF and PACF Plot

Figure 11 shows ACF and PACF plot for our data. We can see that dying down
occurs, so our data tends to be non-stationary in mean. We then perform order-1
differentiation on the data, resulting in a more significant down value that indicates
stationarity (see Figure 12).
11
Figure 11: ACF and PACF Plot Before Differentiation
Figure 12: ACF and PACF Plot After Differentiation
From Figure 12, we can see that ACF shows sinusoidal pattern and crosses the
confidence limit line on lag 1, while PACF crosses the line on lag 2. Thus, we can predict
possible ARIMA models, which are ARIMA (1,1,2), ARI (1,1), and IMA (1,2).
b. Testing Model Feasibility

We then compare statistical value of the three possible ARIMA models. The
comparison is shown in Table 11.
Table 14: Comparison of Statistical Value of RMSE, MAPE, MAE, and Ljung-Box
(row in bold shows the best model)
Model RMSE MAPE MAE Ljung-Box
10 10
ARIMA (1,1,2) 2.765 x 10 24.358 1.909 x 10 10.071
ARIMA (1,1,0) 3.410 x 1010 24.962 2.085 x 1010 19.816
10 10
ARIMA (0,1,2) 2.850 x 10 25.211 1.974 x 10 13.192
12
The smallest statistical value of RMSE, MAPE, MAE, and Ljung-Box is found on
ARIMA (1,1,2). Thus, ARIMA (1,1,2) is the best model to forecast our data. Model
estimation of ARIMA (1,1,2) is shown in Table 15.
Table 15: ARIMA (1,1,2) Model Parameters

Estimate SE t Sig.
VAT- No Trans- Constant 1,477,258,589.000 893,046,818.543 1.654 0.104
Model_ formation AR Lag 1 -0.818 0.121 -6.770 0.000
1 Difference 1.000
MA Lag 1 -0.117 10.438 -0.011 0.991
Lag 2 0.883 9.204 0.096 0.924
Using the following time-series equation of ARIMA (1,1,2):
𝑍𝑡 = 𝜇 + 𝜑1 𝑍𝑡−1 + ⋯ + 𝜑𝑝 𝑍𝑡−𝑝 + 𝑎𝑡 − 𝜓1 𝑎𝑡−1 − ⋯ − 𝜓𝑞 𝑎𝑡−𝑞
we can substitute estimate value from our model parameters, with 𝜑𝑝 as coefficient of
AR(p) and 𝜓𝑞 as coefficient of MA(q), as follows:
𝒁𝒕 = 𝟏. 𝟒𝟕𝟕(𝟏𝟎𝟗 ) − 𝟎. 𝟖𝟏𝟖 𝒁𝒕−𝟏 + 𝒂𝒕 − 𝟎. 𝟏𝟏𝟕 𝒂𝒕−𝟏 + 𝟎. 𝟖𝟖𝟑 𝒂𝒕−𝟐
However, because the model is differentiated to lag 1, we expand 𝑍𝑡 = 𝑌𝑡 − 𝑌𝑡−1 , to

create our ARIMA (1,1,2) equation as follows:
𝑌𝑡 − 𝑌𝑡−1 = 1.477(109 ) − 0.818 (𝑌𝑡−1 − 𝑌𝑡−2 ) + 𝑎𝑡 − 0.117 𝑎𝑡−1 + 0.883 𝑎𝑡−2

𝒀𝒕 = 𝟏. 𝟒𝟕𝟕(𝟏𝟎𝟗 ) + 𝟎. 𝟏𝟖𝟐 𝒀𝒕−𝟏 + 𝟎. 𝟖𝟏𝟖 𝒀𝒕−𝟐 + 𝒂𝒕 − 𝟎. 𝟏𝟏𝟕 𝒂𝒕−𝟏 + 𝟎. 𝟖𝟖𝟑 𝒂𝒕−𝟐
c. Forecasting
The final result of Jakarta Setiabudi Dua Tax Office forecasted VAT revenue for the
year 2022 is shown in Table 16.
Table 16: VAT Forecasted Revenue of Jakarta Setiabudi Dua Tax Office
Year Month VAT Forecasted Revenue
2022 1 Rp 95,494,519,785
2022 2 Rp 98,079,903,747
2022 3 Rp 95,964,946,856
2022 4 Rp 97,695,073,876
13
2022 5 Rp 96,279,754,482
2022 6 Rp 97,437,547,502
2022 7 Rp 96,490,422,293
2022 8 Rp 97,265,212,041
2022 9 Rp 96,631,400,220
4.3. Discussion
2022 10 Rp 97,149,885,945
From the forecasting
2022 11 Rp 96,725,741,999
results of both Surabaya
2022 12 Rp 97,072,710,254
Karangpilang and Jakarta
Total Rp 1,162,287,119,000
Setiabudi Dua VAT revenue, we
found that forecasted revenue tends to be lower than official VAT target. For example,
Surabaya Karangpilang’s official VAT revenue target for the year 2022 is Rp
403,781,162,000, but its forecasted VAT revenue is Rp 379,566,367,993—which is Rp
24,214,794,007 or 5.99% lower. This is also the case for Jakarta Setiabudi Dua’s official
and forecasted VAT revenue: the official VAT revenue target of Jakarta Setiabudi Dua
Tax Office is Rp 1,349,070,583,000, but the forecasted VAT revenue for the same year
is Rp 1,162,287,119,000—which is Rp 186,783,464,000 or 13.84% lower than the official
target. Official VAT targets in both tax offices refer to the revised tax revenue target after
the enactment of President Decree No. 98 of 2022.
When compared to both Tax Offices’ net VAT revenue (at least in the first semester
of 2022), their deviations from their forecasted VAT revenues are smaller than those
from their official target. Surabaya Karangpilang’s net VAT revenue for the first semester
of 2022, for instance, is Rp 186,415,317,564. This is roughly 92.33% of half of Surabaya
Karangpilang’s official VAT revenue target, but 97.76% compared to the forecast result
of January to June 2022. The same thing happens for Jakarta Setiabudi Dua’s data: VAT
revenue of Jakarta Setiabudi Dua in January to June 2022 is Rp 829,734,161,466,
translating to 123.01% of half of its official VAT revenue target or 109.30% of forecasted
VAT revenue of January to June 2022.
14
Figure 13: Comparison of Net VAT Revenue, Official VAT Target, and Forecasted
VAT Revenue of Both Tax Offices, for first half of 2022.
Seeing that differences of net VAT revenue and forecasted revenue is closer to zero
than those of net VAT revenue and official VAT target, one should not necessarily think
that forecasted revenue describes real VAT revenue pretty well. This is because the
official target might have taken other external factors, while forecasted revenue as the
result of ARIMA method only uses historical data. That said, we can think of forecasted
revenue as tax revenue that the office should normally achieve when everything goes
just the same as it did in previous years. This works as some sorts of fail-safe target: a
target that some tax offices should normally achieve, if external factors are omitted and
taxpayers pay taxes with the same abilities. Therefore, if some tax offices achieve less
than their fail-safe target, it is an indication that these offices perform less than they
historically did. On the other hand, if some tax offices achieve more than their fail-safe
target, it is indicated that they perform better than they normally did. In this case,
Surabaya Karangpilang’s VAT revenue in the first half of 2022 still falls short below its
fail-safe target—since, historically, it can achieve at least 5 billions rupiahs more than
what it achieved until June 2022. Likewise, Jakarta Setiabudi Dua goes stronger than it
normally did, because it achieves around 71 billions rupiahs higher than what ARIMA
predicted it can.
This fail-safe target can further be used to better measure DGT’s performance in
terms of the amount of tax it collected. Looking back at Figure 13, if for instance Surabaya
Karangpilang achieves about Rp 190 billions of VAT until June 2022, it is a little unfair to
say that this tax office still falls away from its designated target. This is because when
compared to what it historically can achieve, it is only Rp 1 billion shy. The same logic
goes for Jakarta Setiabudi Dua’s performance: if, for instance, it achieves Rp 700 billions
of VAT in the first semester of 2022, we cannot say that it has achieved more than its
target. This is because, historically speaking, Jakarta Setiabudi Dua still can achieve Rp
59 billion more.
This fail-safe target can also work as a starting point for decision making. Heads of
Tax Offices of DGT can use this fail-safe target to see how much they fall below their
normal tax collection performances and what strategies they have to do to fulfill the gap.
15
This should be their first priority. Meanwhile, the differences between the official tax
revenue target and the forecasted tax revenue can be seen as the impact of external
factor—thus, a second priority. In case of VAT, if this difference is positive, it means that
current economic situation is better than how it was in previous years. If the difference is
negative, it means that previous years’ situation was worse.
5. CONCLUSION
We forecast VAT revenue of two Tax Offices—Surabaya Karangpilang Tax Office
and Jakarta Setiabudi Dua Tax Office—of DGT using ARIMA Box-Jenkins method. Using
historical VAT revenue of 2017 until 2021, after several tests we found that ARIMA
(1,1,1) is the best model for Surabaya Karangpilang, while ARIMA (1,1,2) is the best
model for Jakarta Setiabudi Dua. The results show that forecasted VAT revenue
resemble closely with VAT revenue of both offices, meaning that differences of
forecasted VAT revenue and real net VAT revenue is closer than zero, than differences
of forecasted revenue and official VAT revenue target defined by Indonesia government.
We further argue that these results can be used as a fail-safe target to better measure
DGT’s performance. Since until now there is no specific method to calculate and allocate
tax revenue target, the result of our ARIMA Box-Jenkins forecasting can be used as an
initial tool to find if certain tax offices achieve more or less than what they historically can.
Especially for VAT which, by nature, highly relies on the economic situation, this fail-safe
target can work as a first-priority target that should be the foundation of tax intensification
strategies for tax offices in DGT.
6. LIMITATION
This research is, in many extents, limited due to unavailability of comprehensive tax
revenue data, both in terms of the scope and the details of the data. Further research is
eagerly welcomed to see how ARIMA can forecast nationwide tax revenue target, for all
types of taxes—not just VAT.
Acknowledgements
The authors wish to thank Mr. Eko Radnadi Susetio, the Head of Surabaya Karangpilang Tax
Office, and Mrs. Rina Lisnawati, the Head of Jakarta Setiabudi Dua Tax Office, who have granted
support and data availability for this research.
REFERENCES
Box, G., & Jenkins, G. (1970). Time Series Analysis: Forecasting and Control. Holden-
Day.
Cryer, J., & Chan, K. (2008). Time Series Analysis. Springer.
Daniel. (1989). Statistik Nonparametrik Terapan. Penerbit Gramedia.
Direktorat Jenderal Pajak. (2021). Laporan Tahunan 2020.
Hyndman, R. J., & Koehler, A. B. (2006). Another look at measures of forecast accuracy.
International Journal of Forecasting, 22(4).
Kurniati, D. (2022, July 27). Penerimaan Pajak Tumbuh 55,7%, Sri Mulyani: Kenaikan
16
Luar Biasa Kuat. DDTC News. https://news.ddtc.co.id/penerimaan-pajak-tumbuh-
557-sri-mulyani-kenaikan-luar-biasa-kuat-40839
Ljung, G. M., & Box, G. (1978). On a measure of lack of fit in time series models.
Biometrika, 65(2).
Makridakis, S., Wheelwright, S. C., & McGee, V. E. (1999). Metode dan Aplikasi
Peramalan Jilid 1. Penerbit Erlangga.
President Decree No. 96 of 2017 concerning Amendment to President Decree No. 37 of
2015 concerning Employee Performance Benefits in the Directorate General of Taxes,
(2017) (testimony of President of the Republic of Indonesia).
President Decree No. 98 of 2022 concerning Amendment to President Decree No. 104
of 2021 concerning Details on the State Budget for Budget Year of 2022, (2022)
(testimony of President of the Republic of Indonesia).
Wei, W. W. S. (2006). Time Series Analysis Univariate and Mutivariate Methods.
Pearson.
Wildan, M. (2022, June 30). Perpres Revisi APBN 2022 Terbit, Target Perpajakan Resmi
Naik. DDTC News. Perpres Revisi APBN 2022 Terbit, Target Perpajakan Resmi Naik
17
USING DATA MINING TO DETECT FRAUD PATTERNS
FROM TAXPAYER’S FINANCIAL STATEMENT
Achmad Ginanjar, Agung Septia Wibowo

Direktorat Jenderal Pajak, Jakarta, Indonesia Email: aginoffice@gmail.com Direktorat Jenderal
Pajak, Jakarta, Indonesia Email: agung.septia@gmail.com
ABSTRACT
The usage of machine learning in analysing financial statement is one stream that less explored compared
to data mining mainstream such as Natural Language Processing (NLP) or image explorations, yet it is a
promising area to explore. This study explores state of the art linear regression theorem to analyse
detectable pattern in taxpayer’s financial statement, utilizing method that conceptually adopting the basic
concept of both vertical and horizontal approaches in financial statement analysis. Using data of financial
statements reported to Indonesian Tax administration and historical records of the taxation audits, the study
shows that there are detectable patterns generated from the experiment. This study uses linear regression’s
of financial statement account values that represent degree of changes over the studied year. Furthermore,
this study uses yearly value of each financial statement account of an entity to build a unique point that
represent itself among other points. a clustering method then done to finally group sets that are having similar
pattern.
This study shows that the method used in this experiment can be used to analyse how an entity reports its
financial statement over the years and clustering it based on the possibility of committing fraud derived from
historical audit record of financial statement pattern. These pattern then being validated by the occurrence
of underpayment/ overpayment of corporate income from tax audit results. By examining the cluster results,
this study shows that some clusters identify labelled pattern quite well where 2 out of 3 labels can be identified
accurately. The comparation resultsbetween unsupervised clustered method versus labeled criteria show a
significant probability of fitness. This study, however, did not evaluate the feature importance that might show
the reason why the clusters are formed.
Keywords: accounting, machine learning, clustering, regression, horizontal analysis, vertical analysis
Penggunaan pembelajaran mesin dalam analisis laporan keuangan adalah salah satu bidang yang belum
banyak dieksplorasi dibandingkan dengan proses data mining yang umum seperti Natural Language
Processing (NLP) atau Computer Vision (CV), padahal area ini cukup menjanjikan untuk dianalisis. Studi ini
mengeksplorasi teorema regresi linier tingkat lanjut untuk menganalisis pola yang dapat dideteksi dalam
laporan keuangan wajib pajak, dan menggunakan metode yang secara konseptual mengadopsi pendekatan
analisis vertikal dan horisontal dalam analisis laporan keuangan. Dengan menggunakan data laporan
keuangan Wajib Pajak yang dilaporkan kepada administrasi perpajakan Indonesia dan catatan pemeriksaan
perpajakan atas pajak penghasilan terkait, penelitian ini menunjukkan bahwa ada pola yang dapat dideteksi
yang dihasilkan dari eksperimen yang dilakukan. Penelitian ini menggunakan regresi linier dari nilai akun-
akun laporan keuangan yang mewakili tingkat perubahan nilai dalam jangka waktu sesuai tahun yang diteliti.
Selanjutnya, penelitian ini menggunakan nilai akun per tahun dari setiap akun laporan keuangan untuk
membangun titik unik yang merepresentasikan setiap pola laporan keuangan Wajib Pajak pada rentang
periode tertentu. Metode unsupervised clustering kemudian dilakukan untuk mengelompokkan laporan
keuangan Wajib Pajak yang memiliki pola yang sama.
Studi ini menunjukkan bahwa metode yang digunakan dalam eksperimen ini dapat digunakan untuk
menganalisis bagaimana suatu entitas melaporkan laporan keuangannya selama bertahun-tahun dan
mengelompokkannya berdasarkan kemungkinan melakukan kecurangan yang berasal dari pola laporan
keuangan. Pola ini kemudian divalidasi dengan data kurang bayar/lebih bayar penghasilan badan dari hasil
pemeriksaan pajak. Dengan melihat hasil klaster, penelitian ini menunjukkan bahwa proses machine
learning dapat mengidentifikasi pola label dengan cukup baik dimana 2 dari 3 label dapat diidentifikasi
18
secara akurat. Hasil perbandingan antara metode unsupervised clustering dengan label sebenarnya dari
data historis menunjukkan probability of fitness yang signifikan. Studi ini, terbatas dalam hal tidak
mengevaluasi tingkat kepentingan fitur-fitur yang mungkin menunjukkan alasan mengapa klister dapat
terbentuk.
Keywords: akuntansi, pembelajaran mesin, klastering, regresi, analisis horisontal, analisis vertical
1. INTRODUCTION
Machine learning (ML) has introduced new approaches to understand and solve
problems in many areas. Supported by the advancement in technology, mathematics
and statistics, researchers now able to develop advanced machine learning models that
never be thought before. Multidimensional visualization for example, with the new
formula like TSNE, now can be done with a consumer processing unit like laptop. The
advancement of technology also creates a new way on how data created, stored, and
accessed. The easiness of data creation moves ourworld to the era of big data. In the
big data era, data are too big to be processed. Hence, more machine learning
approaches are needed to help human in decision making based on abundantavailable
data.
Finance and accounting are areas that have a bright future for the implementation
of machine learning. Big companies like EY, Pricewaterhouse Coopers (PwC), Deloitte,
and KPMG have developed their own method to approach ML in pursue of a better
understanding from their data.Many of the method available are used to understand the
data as a single data, for example frauddetection in a financial report, or determine group
of business base on a financial report. Furthermore, ML approaches can also be used to
determine the practice of tax fraud. Vasco et al(2018) has conducted a study related to
Taxpayer Segmentation for Personal Income Tax fraud with data mining techniques with
internal tax administration data. The study shows that by examining the characterization
of potential fraud taxpayers through probability distribution of variables, the profile of
potential fraud taxpayers can be clustered into group which showing the degree of
probability in committing tax fraud.
Adopting those approaches, this paper would employ machine learning techniques
and analysison the financial statement reported to tax administration agency to examine
patterns of taxpayersposting behaviors. It will focus on the degree of change of each
account of a company from its historical financial statement of an entity then cluster
them based on the reporting patterns. Furthermore, clusters generated from the
processes then validated with tax audit reports that showing whether taxpayers from
each cluster are fraud or not fraud based on tax assessments issued. Based on that,
this research is expected to be able to dig deeper into the potential to builda model of
data mining techniques combined with a financial statement analysis approach to be
used to detect corporate income tax fraud from financial statement patterns.
Prior studies have shown the use of data mining techniques can help the process of
data analysisin accounting and taxation areas. Experiment was done by Cai, et al (2016)
and Becirovic, et al (2020) observed the usage of machine learning theory in accounting
scopes. Both works use thesimilar method in terms of accessing the accounting report
values and the experiments accessedthe account value as a single value to solve their
19
problem respectively. The result was the clusterof each report that also represent the
entity itself. This founding provide opportunity for researchers to explore more relevant
implementation of machine learning methods for accountingresearch.
Furthermore, the use of machine learning tends to be useful to help researcher to
gain insight from untouched accounting data such as use it in detecting the probability
of a company committing fraud. Many studies focus on how this method can provide
efficient and effective approach to capture the pattern of how fraudulent companies
report their financial statements. One example of the use of machine learning to detect
financial fraud has been proposed by Alvarez et, al. (2017) in their study of money
laundering investigation of more than 600 Spanish companies, they combine the use of
Benford's law as a detection tool for anomalies in books andaccounting records, then
applying four machine learning models of pattern recognition (logistic regression, neural
network, decision trees and random forests) in detecting the presence of financial fraud
of money laundering. The study found that random forest algorithms with SMOTE
transformation obtains quite confident result with 96.15% of true negative results and
94,98% of true positive results.
The use of machine learning approach also inspires researchers in taxation area to
conduct studywith similar approach for examining taxation data. Study conducted by Dias
et al., (2016) providean analytical framework that combines dimensionality reduction and
data mining techniques to obtain sample segmentation according to potential fraud
possibility in personal income tax. Furthermore, Dias et al., (2016) classify individual
taxpayers based on their tax evasion risk usingcluster analysis methodology to organize
observations into homogeneous groups which would enable them to identify risky firms
in a more effective manner.
However, we could not find any works that studies the financial account entries
changes as a whole. In general, most studies employ classical accounting techniques
such as ratio analysis orvertical and horizontal to capture partial pattern of a financial
statement. Therefore, as an effort to fill this gap, this paper would like to observe the
possibility of study on overtime account changes as a whole with the help of machine
learning methodology. For addition, this study aims to detect specific pattern on the
changes of financial statement with the possibility of tax fraud occurred. For example,
we want to confirm as if this approach can provide evident that there is aspecific pattern
in which fraudulent taxpayers behave in their tax assessment. This study, however,
would like to focus more on the evaluation of the value changes in financial statement
accounts from the historical values. The main idea of this study is to detect the behavior
or pattern formed from an entity’s financial statement over the years. This approach
adopts the horizontal analysis which compare values of financial statement account
throughout the years and detect anomaly that may rise indicating abnormal pattern or
behavior in financial statement, while at thesame time, this study also adopts vertical
analysis by using the whole account as features.
The study is mixed of the machine learning approaches as we both employed
supervised and unsupervised methods. The linear regression is the part where the
supervised learning used. Onthe other hand, the proof-of-concept section that clustering
the data is done with unsupervised learning. The study then matches the cluster
produced with real world label produced by a complex process business from the Tax
Office. The label is a permanent legal force; therefore, the trueness is valid. The
experiment, unlike others, focus on how posting behavior changes overtime. This is
achieved by exploiting the B1 value of the regression formula.
𝑦 = 𝐵0 + 𝐵1 𝑥 + 𝑒 , {𝑖 = 1, … , 𝑛}
Equation 1 Linear Regression
20
The B1 in above formula express how big the change of x changes the value of y. The
study takes10 years data of each account and calculate the linear regression formula.
3. RESEARCH METHOD
3.1. Analysis Methods
3.1.1 Horizontal Analysis and Vertical Analysis
Financial analysis techniques can be used by investigators to find anomalies and
examine indication of impropriety in financial information. Unexpected lapses in values
are likely to indicate wrongdoing, but may also indicate illegal or fraudulent actions
(Albrecht, et al., 2015). The process of investigating financial statement fraud should
begin by determining the areasof operation in identifying potential fraud schemes, noting
the red flags associated with the identified schemes, establishing effective audit
measures to look for indicators and conducting further investigations to validate the
detection or suspicion of red flags (PWC, 2004). Financialanalysis can also be employed
in examining if there are tax frauds as Taxpayers are obligedto attach financial statement
in their tax reports.
Horizontal Analysis and Vertical Analysis are sample forms of classical financial
analysis (Albrecht, et al., 2015). Those financial analysis are proficient techniques that
can be used fordetecting fraud as well as spotlighting areas of concern (Edelman, 2008).
Horizontal Analysis basically analyzes the changes in the value of an account in the
financial statements from oneperiod to other periods, while Vertical Analysis analyzes
the value proportions of accounts ina period of reporting. In some cases, there are frauds
that normal audit techniques can miss,but they can be effectively detected using
horizontal and vertical analysis (Albrecht, et al., 2015).
However, performing Horizontal and Vertical Analysis in large sets of data needs
enormous effort and may not be an efficient tool if there are many entities to be analyzed.
Therefore, forthis study, we use regression slope to proxy the value changes of financial
report accounts asperformed in horizontal analysis. Furthermore, we use account value
and period as features used in our clustering method so that the proportion of account
value can be captured as in Vertical Analysis usually done.
3.1.2 Linear Regression Analysis

Regression is a mathematical model where the result is a number (float or integer)
that can take any values (continuous). Linear means that the ideal problem-solving
result is following a straight line either decreasing or increasing in a consistent value.
The linear regression is formulated as:
𝑦 = 𝐵0 + 𝐵1 𝑥 + 𝑒 , {𝑖 = 1, … , 𝑛}
Equation 2 Linear Regression
y = is the model result or the predicted value. In this experiment, this value is the value ofan
account given specific year x.
x = is the feature value or the known variable. In this experiment, this value is the value ofthe year
of y
B0 = is the intercept value. It represents the value of y when x is equal to 0
B1 = is the dimensional parameter value. Or it can be understandable as how big the
changes in y for each change in x. Also known as slope.
21
e = error noise. Usually present in the form of normal distribution random value.
3.1. Data Processing

This study uses ten years of accounting’s report data. the studied year was 2011-
2020. For each entity being observed, there might be several accounts consists of ten
years data. All ten years data must be existing regardless the account name, which mean
that an entity havesets of their own unique account name. In this study this condition will
be called the total yearand data availability limitations. The ilustration of the data structure
can be seen as followed:
NPWP AKUN Tahun Nilai

(Taxpayer ID) (Account Name) (Year) (Value in Rp)
ABC cash 2011 200000
… … … …
ABC cash 2020 700000
ABC Short-term Invesment 2011 1200000
… … … …
ABC Short-term Invesment 2020 5700000
XYZ Trade Receivable from 2011 4000000000

Related Parties
… … … …
XYZ Trade Receivable from 2020 5000000000

Related Parties
In the table above, entity ABC have 2 account name that are being used for
calculation. The two account names are ‘cash’ and ‘Short-term Invesment’. Both account
names have 10 yearsof data from the studied years. Similarly, entity XYZ has ‘Trade
Receivable from Related Parties’ that consist of the data from 2011 to 2020.
3.2.1 Data Overview

This study is using more than 900 account names available in the original datasets.
However,due to variability of the data not all 900 names are employed. Some account
names might notbe used at all due to the total year and data availability limitations. Below
are some samples of the account names used. The original data cannot be shared due
to the law restriction.
22
ID Account Name
1 Cash and cash equivalents
2 Short-term Invesment
3 Trade Receivable from Third Parties
4 Trade Receivable from Related Parties
5 Other Receivable from Third Parties
6 Other Receivable from Related Parties
7 Allowance for Uncollectible Accounts
8 Inventory
9 Prepaid Expense
10 Purchase Down Payment
11 Other Current Assets
12 Long-term Receivable
13 Land and Building
14 Other Fixed Assets
15 Subtract: Accumulated Depreciation
16 Investment in Associated Companies
17 Other Long-term Investment
... ...
... ...
... ...
617 Intangible Assets

Figure 1 Account ID table
The total of 617 account names are used for the experiment. This study does not
evaluate thedetails or type of the account names. For each entity, all 617 account names
within 10 yearsof observed data, create a unique dimension for respected entity.
3.2.2 Applying regression

Unlike traditional horizontal and vertical analysis, this study uses a different strategy
to analyse the data. In a horisontal analysis, the difference of an account values for some
periodare being calculated, while in vertical analysis, the percentages of an account in
relative to overall account values are used (Albrecht, et.al., 2015). On the other hand,
this study evaluates both vertical and horizontal values around it’s algorithm. The vertical
values that are an account’s yearly values, are the base value for regression calculation.
Linear regression formula is used to extract 10 years of “degree of changes” from each
account. Generally, ‘degree of changes’ is also known as slope. Other researcher shows
how to process single record as a feature for clustering method (researcher, year).
However, this study focusses on the “degree of change” in each account as a feature.
This degree of changes is collected by using B1 value from above linear regression
formula. The B1 value represent the relation between x and y. The B1 can be seen as
how change of x affect y directly. Therefore, this value also can be interpreted as the
“degree of changes” values.
3.2.3 Dimensional reduction
23
This section works on the dimensional reduction process for the easiness of
clustering and visualization. As proposed by Belman (2003) about the course of
dimensionality, when processing a high dimension data, various phenomena might
occur that do not occur in low dimensional data analysis. To reduce the possibility of
this issue, TSNE algorithm (van der Maaten & Hinton, 2008) and PCA algorithm are
applied to the data.
In general, TSNE algorithm results can be seen as an unsupervised clustering
model. It is work by pulling together data point that, mathematically, have similar
features. This study is applying TSNE with n = 2. Which means that the result of the
model has 2 dimensions / features.
On the other hand, (PCA) works by determining the most important value. The idea
of PCA issimple, reduce dimensionality while preserving the statistical information as
much as possible(Jollife & Cadima, 2016). PCA work by extracting set of values that
represent the distributionof the original dataset. The set of values usually marked as:
𝑃𝐶𝐴𝑛, {𝑛 = 1, … , 𝑥}
PCA1 is considered more significant to the dataset compared to PCA2, and so on.
This studyapplies PCA to the extent of PCA1 and PCA2.

4.1 Data with Regression
The result from applying linear regression to the data is data that consist of
‘id_object’, id_akun, slope and intercept. The ‘id_object’ represent a unique entity. the
‘id_account’ is the representation of account’s name in an accounting journal like ‘cash’,
‘debt’,etc. The ‘slope’ is theslope value, previously called ‘degree of change’, from the liner
regression of the correspondenceaccount. The ‘intercept’ is the intercept value from the
liner regression of the correspondence account. All data that showed in this paper has
been masked to only represent visual concept. The table mined from linear regression
can be seen below.
Id_object Id_account slope intercept

1 0 -3,15E+16 6,35E+19
1 1 -6,81E+15 1,38E+19
1 2 1,44E+17 -2,90E+20
1 3 -3,15E+16 6,35E+19
... ... ...
400000 1 1,75E+15 -3,53E+18
400000 2 -2,80E+15 5,67E+18
400000 3 2,10E+15 -4,20E+18
400000 4 2,41E+15 -4,82E+18
400000 5 1,24E+15 -2,52E+18
Figure 2 Regression Result
24
The data consist of 40.000 rows from 11102 entity. The unique number of account
names are 617. As a common practice, the account name might have variation and
typo within. However, the account names are not treated with any treatment such as
correction either manual or stateof the art NLP algorithm.
After previous step, a table was built. The table has entity as index and account name
as columnname. The values of the table are the B1 values referring to the index and the
column name. this process was done by pivoting the data.
Figure 3 Pivoted Data
4.2 Visualized Data

4.2.1 TSNE
The result of applying TSNE algorithm to the pivoted data with n=2 is two-
dimensional data. The two-dimensional data are mean for easiness of data visualization.
In this experiment the TSNE result has “Feat_1” and “Feat_2” as features. “Feat_1” is
used as “X” point and “Feat_2” is used as “Y” point.
Index Feat_1 Feat_2

0 2.292.285 6.388.725.586
1 358.360.596 663.375.610
2 290.112.335 424.451.019
3 34.406.998 952.894.653
4 34.406.998 952.894.653
5 34.406.998 952.894.653
6 28.423.534 650.764.404
7 34.406.998 952.894.653
8 34.406.998 952.894.653
9 17.357.542 693.593.445
Figure 4 TSNE Result
The result then plotted for visualization. At this level, the visualization does not
explain a lot. However, several conclusions can be drawn by examining the plot. From
the TSNE Visualization figure below. First, we can clearly spot several individual clusters.
TSNE works by pulling together points that are similar. The individual clusters spotted
25
on the figure can be seen as points that are similar. In other words, there are patterns in
the datapoints.
Figure 5 TSNE Visualization
4.2.2 PCA
The principal component analysis (PCA) algorithm works quite differently. PCA is
lookingto substitute data that has similar variance to the original data. The PCA result of
this studycan be seen below. This study tries to evaluate the value of PCA1, PCA2,
PCA3 and PCA4. As shown in the figure there is not any pattern that clearly shown to
the data.
Figure 6 PCA Visualization
4.3 K-Means Clustering

The TSNE data then clustered with K-Means algorithm. For this study we use n=20.
This also means that the experiment tries to predict 20 cluster within the data. At this
point, the usage of K-Means is solely to cluster data to become 20 regions. The result
will be used tocompare the region from K-Means cluster with real label data.
26
Figure 7 K-Means Plot on TSNE
a. Proof of Concept with Labels

4.4.1 Business Categories
Business categories as identified by Klasifikasi Lapangan Usaha (KLU) is the first
categoryto be tested on. The label comes from 11 general category of the tax payer
business. The 11 categories are label that has been assigned to the respected tax payer
during registrationand is a label in the experiment data. The result of the plotted data can
be seen in the figurebelow.
Figure 8 Business Area Plot on TSNE
The figures show that some clusters have some dominant colors while the others
are not.Some business categories can be well separated by TSNE cluster. In the other
words, it can be said that for this kind of business areas there are patterns on how
accountant log their report over the studied years. In contrast, for some business
categories, the algorithmcompletely cannot detect any similarities in pattern. This just
simply means that the algorithm is not fit for the data or the data labeling process itself
has human error in it.
Further evaluation to the experiment result confirms the suggested conclusion
above. Confusion matrix table between K-Means cluster and Business Category on
TSNE cluster shows that category 4 (Construction, Wholesale and Retail Trade; Car
and Motorcycle Repair and Maintenance) dominate almost every cluster. Category 2
(Processing industry) dominate cluster 0 and 2 with 0.67 and 0.5 respectively. The
numbers are not significant as it close to 0.5 or random probability.
27
Business
0 1 2 3 4 5 6 7 8 9 E
category
Cluster
0
0.01 0.23 0.67 0.04 0.01 0.00 0.03 - - - 0.00
1 0.08 0.03 0.03 0.03 0.45 0.13 0.12 0.07 0.05 0.02 -
2
0.04 0.36 0.50 0.06 0.03 0.00 0.01 0.00 0.00 - -
3
0.07 0.05 0.02 0.01 0.56 0.10 0.06 0.10 0.01 0.01 0.00
4 0.01 0.01 0.01 0.01 0.82 0.05 0.03 0.06 0.01 0.00 -
5
0.13 0.03 0.03 0.05 0.25 0.17 0.19 0.08 0.06 0.02 -
6
0.04 0.07 0.04 0.01 0.72 0.03 0.05 0.03 0.00 0.01 -
7 0.01 0.04 0.01 0.01 0.79 0.05 0.04 0.03 0.02 0.01 0.01
8
0.01 0.02 0.01 0.00 0.78 0.04 0.02 0.10 0.02 0.00 0.00
9
0.03 0.10 0.05 0.01 0.56 0.06 0.09 0.03 0.03 0.01 0.02
10
0.04 0.43 0.39 0.06 0.07 0.00 0.01 0.00 0.00 0.00 0.00
11
0.01 0.03 0.01 - 0.83 0.02 0.03 0.07 0.01 0.01 0.00
12
0.03 0.06 0.03 0.01 0.77 0.02 0.03 0.03 0.01 0.00 0.00
13 0.04 0.38 0.47 0.05 0.04 0.01 0.02 0.01 - - -
14
0.02 0.03 0.01 0.00 0.72 0.05 0.08 0.06 0.02 0.01 0.00
15
0.04 0.04 0.05 0.01 0.76 0.02 0.04 0.02 0.00 0.01 0.00
16 0.03 0.31 0.57 0.05 0.02 0.01 0.00 0.00 - - -
17
0.03 0.03 0.02 0.01 0.69 0.07 0.03 0.10 0.02 0.00 -
18
0.03 0.04 0.02 0.00 0.67 0.06 0.08 0.07 0.01 0.01 -
19
0.31 0.03 0.07 0.04 0.16 0.03 0.27 0.07 0.01 0.01 0.01
Figure 9 K-Means vs Business Group
Digit Category
0 Mining
Agriculture, Forestry and Fisheries
1 Processing industry
Water Supply, Waste Management and Recycling, Disposal and Cleaning of Waste
and Garbage
Procurement of Electricity, Gas, Steam/Hot Water and Cold Air
4 Construction
28
Wholesale And Retail Trade; Car and Motorcycle Repair and Maintenance
Transportation And Warehousing

5 Information and Communication
Provision of Accommodation and Provision of Food and Drink
Transportation And Warehousing
6 Information and Communication
Financial Services and Insurance
Professional, Scientific and Technical Services
Real Estate
7 Rental Services, Employment, Travel Agencies and Other Business Support
Professional, Scientific and Technical Services

8 Government Administration and Mandatory Social Security
Health Services and Social Activities
Education Services
Rental Services, Employment, Travel Agencies and Other Business Support
9 Individual Services Serving Households; Activities that produce goods and services
by households that are used alone to meet needs
Culture, Entertainment and Recreation
Activities of International Agencies and Other Extra International Agencies
Other Service Activities
Figure 10 Categories Table
i. Fraud Group
a. Labeled Criteria 1
In this section label criteria 1 is applied. The label criteria 1 is about the
occurrence ofa tax payer to apply for overpayment on their past tax payment.
When a taxpayer applies for overpayment for any value, in this study, the tax
payer is assigned with 1 or True or yellow color. Value 0 or False or blue color
is given for tax payer data that not apply any overpayment.
Figure below shows that there are clear patterns of the labeled case. Most
cluster arecompletely covered with value 0 or blue color. This means that most
tax payer observed did not apply for overpayment. The blue color is concentrated
in the center.For the cluster in the middle the color is completely blue. Yellow or
true can be seen in the outer area of TSNE cluster. Several clusters in the bottom
are dominated by yellow.
29
Figure 11 Fraud Criteria 1 plot on TSNE
Observation to the confusion matrix table between K-Means and Fraud

Criteria Labelon TSNE cluster confirm above result. 10 out of 20 K-Means cluster
are dominated with not Fraud label with probability above 0.7. Cluster 3,6, 14 and
19 have probabilities above 0.7 for fraud label. The rest 7 of K-Means clusters have
probabilitybelow 0.7 that is not considered as significant.
Criteria 1 Fraud Criteria

K-means
Cluster Not Fraud Fraud
0 0.382550 0.617450
1 0.834739 0.165261
2 0.736527 0.263473
3 0.233390 0.766610
4 0.852126 0.147874
5 0.819149 0.180851
6 0.146771 0.853229
7 0.337950 0.662050
8 0.918330 0.081670
9 0.920181 0.079819
10 0.516579 0.483421
11 0.937402 0.062598
12 0.616123 0.383877
13 0.912621 0.087379
14 0.073298 0.926702
15 0.559271 0.440729
16 0.730882 0.269118
17 0.598485 0.401515
18 0.870073 0.129927
19 0.025455 0.974545
Figure 12 K-Means vs Fraud Criteria 1
b. Labeled Criteria 2
30
The label criteria 2 is about the occurrence of a tax payer to be assigned as
underpayment of their past tax payment. When a tax payer is assigned as
underpayment for any value, in this study, the tax payer is labeled with 1 or True
or yellow color. Value 0 or False or blue color is given for tax payer data that
did not show any sign of underpayment.
Similarly with labeled criteria 1, there are clear pattern of the labeled criteria
2. The observation to the figure show that the yellow versus blue color
composition is quite fair. The center TSNE custer are dominant to value 0 or bkue
color. Some cluster on the right and bottom part of the figures are completely
yellow or labelled as true. For the bigger cluster the area that are close to the
center are blue or labeled as false. Themore distance to the center the probability
of yellow occurrences is bigger.
Figure 13 Fraud Criteria 2 plot on TSNE
The confusion matrix between K-Means cluster vsrsus labeled criteria 1 agree
with theprevious figure. 10 K-Means clusters out of 20 are labeled with True with
probability more that 0.7 or 70%. The rest of the cluster have no significant
probability as the probability for both label near 0.5 or 50% of chance.
Criteria 2 Fraud Criteria

K-means
Cluster Not Fraud Fraud
0 0.001736 0.998264
1 0.556180 0.443820
2 0.137821 0.862179
3 0.514979 0.485021
4 0.036125 0.963875
5 0.642741 0.357259
6 0.397408 0.602592
7 0.556593 0.443407
8 0.054662 0.945338
9 0.038270 0.961730
10 0.613953 0.386047
11 0.630691 0.369309
12 0.258232 0.741768
13 0.114413 0.885587
31
14 0.668975 0.331025
15 0.017713 0.982287
16 0.220848 0.779152
17 0.202797 0.797203
18 0.569069 0.430931
19 0.436077 0.563923
Figure 14 K-Means vs Fraud Criteria 2
5. CONCLUSION
The K-Means clusters cover 20 clusters. As The nature of K-Means clustering is
unsupervised learning, there is not any best size number to be calculated. There are
severalways to conduct a good size of clusters, however this study is not about K-Means
clustering. This study focusses on detecting pattern using linear regression on yearly
accounting report.In addition, this study focusses on using TSNE to cluster the data to
clearly show the visualization of the data.
The data clustered in the 20 clusters are labeled with historical data. The first label
is about business area label that assigned during tax payer registration. From the data
available, it isclear that the label does not match with the clustered data except only for
business criteria
4. This means that this method might be only useful to detect an entity pattern that
has business criteria 4. The second and third labels have a similar theme. The label
shows that whether there has been an inspection to certain entity that resulted in a
determination of underpayment, overpayment or nothing at all. More than 50% of the
data can be clustered well for both labels with accuracy more than 70%.
The way the algorithm in this study works makes this algorithm is a distinct method
compared to classical vertical and horizontal analysis. This study explores ‘degree of
change’ (B1) overyearly single account data. In other words, it also can be concluded that
several entities mayhave similar value of B1 even though the original data is not the
same. Also, as shown in thefigure of ‘Pivoted Data’, the B1 values that entitled to a single
entity build a unique dimensionfor the respected entity.
Based on the initial findings of the preliminary research, it can be seen that there is a
potentialfor using linear regression over yearly accounting report to detect patterns. More
study can be done in the future to study other labels or exploring using polynomial
algorithm.

This study shows how to exploit slope value of yearly account reports and detect
pattern within. This study proof that linear regression can be used as an option to detect
pattern on how accountants log their bookkeeping and present their financial statement.
The study alsosuggest that the pattern detected may be show particular behavior for
certain labels or purposes, where in this case is grouping fraudulent taxpayer in
corporate income tax. However, this study still have limitations in some areas that need
to be considered in future study. Firstly, this study does not consider the magnitude of
fraud that can be reflected by the amount of underpayment or overpayment. Considering
the magnitude might add furtherinsight of the patterns detected on the different fraud
magnitude. Secondly, this study basedon the assumtion that all the data can be reflected
as a linear function. Due to the size of thedata, the linearity characteristic of the data was
not tested. Future study might also consider other algorithms such as nonlinear. Lastly,
32
the exploration of this study uses TSNE algorithm as a clustering method. TSNE was
invented not to server this purpose, however, the natureof the algorithm allows such
interpretation. Other unsupervised clustering method might be better fit with the data but
the visualization will need more effort to understand. Future studycan be developed from
the minor point of this study. Better pre-processing data and different(nonlinear) algorithm
might be used to do similar experiments.
7. REFERENCE
Albrecht, W. S., Albrecht, C. O., Albrecht, C. C., & Zimbelman, M. F. (2015). Fraud
examination. Cengage Learning.
Becirovic, S., Zunic, E., & Donko, D. (2020). A Case Study of Cluster-based and
Histogram-based Multivariate Anomaly Detection Approach in General Ledgers.
19th International Symposium INFOTEH-JAHORINA.
Bellman, R. E. (2003). Dynamic Programming. Courier Dover Publications.
Cai, F., Le-Khac, N.-A., & Kechadi, M.-T. (2016). Clustering Approaches for Financial
Data Analysis: a Survey. Advances in Quantitative Analysis of Finance and.
Accounting, 199-211.
De Roux, D., Pérez, B., Moreno, A., Villamil, M.P. and Figueroa, C. (2018), “Tax Fraud
Detection for Under-Reporting Declarations Using an Unsupervised Machine
Learning Approach”, Proceedings of the 24th ACM SIGKDD International
Conference on Knowledge Discovery & Data Mining.
Devereux, Michael & Freedman, Judith & Vella, John. (2012). Tax Avoidance. SSRN
Electronic Journal. 10.2139/ssrn.3754562.
Dias, A., Pinto, C., Batista, J. and Neves, M.E, (2016), “Signaling Tax Evasion, Financial
Ratios and Cluster Analysis”, Working Paper 51, OBEGEF, Observatorio de
Economia y Gestáo de Fraud, Coimbra.
Fox, W. F., Luna, L., and Schaur, G. (2014), “Destination taxation and evasion: Evidence
from US
inter-state commodity flows”, Journal of Accounting and Economics, 57(1), 43-57.
González, P. C., and Velásquez, J. D. (2013), “Characterization and detection of
taxpayers with false invoices using data mining techniques”, Expert Systems with
Applications, 40(5), 1427-1436.
Matos T., de Macebo J.A. and Monteiro J.M. (2014), “An empirical method for discovering
tax fraudsters: a real case study of Brazilian fiscal evasion”. Proceedings of the
19th International Database Engineering & Applications Symposium, New York.
PriceWaterhouseCoopers. (2004). The Emerging Role of Internal Audit in Mitigating
Fraud and Reputation Risks. Internal Audit Services. PriceWaterhouseCoopers.
Wu, R. S., Ou, C. S., Lin, H. Y., Chang, S. I., and Yen, D. C. (2012), “Using data mining
technique to enhance tax evasion detection performance”, Expert Systems with
Applications, 39(10), 8769-8777.
van der Maaten, L., & Hinton, G. (2008). Visualizing Data using t-SNE. Journal of
Machine Learning Research, 2579–2605.
Vasco, C. G., Rodríguez, M. J. D., de Lucas Santos, S., & de Madrid, U. A. (2018).
Characterization and detection of potential fraud taxpayers in Personal Income Tax
using data mining techniques.
33
APPLICATION OF DATA MINING TO TAXPAYERS
ISSUING FICTITIOUS TAX INVOICE USING
CLASSIFICATION TECHNIQUES
Yusrifaizal Gumilar Winata a, Fauziah Noorb, Muhammad Futhra Baharc, Aris Budi
Santosod, Eddy Sukarnoe
aDirektorat Jenderal Pajak, Indonesia Email: yusrifaizal.gumilar@kemenkeu.go.id
bDirektorat Jenderal Pajak, Indonesia Email: noor.fauziah@pajak.go.id
cDirektorat Jenderal Pajak, Indonesia Email: muhammad.futhra@pajak.go.id
dDirektorat Jenderal Pajak, Indonesia Email: aris.budi@pajak.go.id
eDirektorat Jenderal Pajak, Indonesia Email: eddy.sukarno@pajak.go.id
ABSTRACT
Application of Data Mining to Taxpayers Issuing Illegal Tax Invoices Using Classification Techniques.
Taxes are the backbone of state revenue, but tax revenues are not received optimally because of tax
crimes. 44% of all tax crimes are derived from the issuance of Tax Invoices that are not based on actual
transactions. This reduces income from the VAT sector and can even take state money using VAT
refunds. To deal with this, DGT can use data mining to conduct audits more effectively and efficiently.
This research is mixed methods, both qualitative and quantitative. The Cross Industries Standard
Process for Data Mining (CRISP-DM) also used. Data processing and machine learning conducted
with Python programming language. The dataset consisted of 1.071 tax payers that issue illegal tax
invoices and 2.142 non-issuer taxpayers. The research provided a machine learning model that has
Prediction Efficiency of 83.56%, reduction in Examination Effort of 69.31%, and Strike Rate of 90.77%.
The model can then be used with Streamlit and predicts 8 issuer tax payers with a 75% probability from
deployment data that has 1.000 rows.
Keywords: data mining,crisp dm, value added tax, tax invoice
Penerapan Data Mining Terhadap Wajib Pajak Penerbit Faktur Pajak Tidak Sah Menggunakan Teknik
Classification. Pajak adalah tulang punggung penerimaan negara, namun penerimaan pajak tidak
berjalan optimal karena adanya tindak pidana perpajakan. 44 % dari keseluruhan tindak pidana pajak
berasal dari penerbitan Faktur Pajak Tidak berdasarkan transaksi sebenarnya. Hal tersebut
mengurangi pendapatan dari sektor PPN bahkan dapat mengambil uang negara menggunakan
restitusi PPN. Untuk menghadapi hal tersebut DJP dapat menggunakan data mining untuk melakukan
pemeriksaan kriteria seleksi yang lebih efektif dan efisien. Penelitian menggunakan metodologi
campuran antara kualitatif dan kuantitatif dengan mengacu pada Cross Industries Standard Process
for Data Mining (CRISP-DM). Pengolahan data dan proses machine learning dilakukan dengan
pemrograman Python. Dataset yang digunakan untuk melatih dan evaluasi model terdiri dari 1.071
data wajib pajak penerbit dan 2.142 wajib pajak bukan penerbit. Penelitian ini menghasilkan model
machine learning untuk klasifikasi penerbit faktur fiktif dengan Prediction Efficiency sebesar 83,56%,
reduction in Examination Effort sebesar 69,31%, dan Strike Rate sebesar 90,77%. Model tersebut
kemudian dapat digunakan dengan Streamlit dan memprediksi 8 WP penerbit dengan probability 75%
dari data deployment yang terdiri dari 1000 baris data.
Kata kunci: data mining, crisp dm, value added tax, tax invoice
34
1. BACKGROUND
Taxes are the backbone of state revenues, but the Directorate General of Taxes
(DGT) is currently experiencing serious challenges as indicated by the declining tax ratio
up until now it is below 10%. Some of these challenges also come in the form of
increasingly uncertain economic conditions due to the pandemic and limited resources
within the Directorate General of Taxes. This can hinder the Directorate General of Taxes
in fulfilling its mandate, which is optimal tax revenue.
One of the things that causes tax revenue to be not optimal comes from tax crimes.
The issue of tax crime is an inherent risk derived of government tax policies and needs
to be mitigated and managed comprehensively. Law enforcement business processes
at the Directorate General of Taxation must be able to carry out early warning detection
of indications of tax crime and be able to make effective law enforcement activities as
effective and focused deterrent effect.
Looking at the 2020 DGT Annual Report, the most common tax crime handled by
tax investigators is Issuance of Tax Invoices Not Based on Actual Transactions (NBAT)
or Fictitious Tax Invoice with a total of 44% of all cases handled by tax investigators
which can be seen in the figure following.
Figure 1 Tax Crime Cases handled by Tax Investigators

Source : Annual Report of the Directorate General of Taxes for 2020
Fictitious Tax Invoice Crimes are dynamic and adapt quickly to take advantage of
new opportunities for financial gain, often outpacing legislative changes and controls on
e-faktur apps which was designed to prevent them. Facing this, DGT must adapt quickly
to avoid loss of state revenue due to the issuance of fictitious tax invoices.
Tax Invoice NBAT is also referred to as Fictitious Tax Invoice which in simple terms
is an invalid tax invoice because the identity of the issuer does not match the actual
situation or there is no material delivery of goods and/or services even though formally
fulfilling the provisions of the Value Added Tax (VAT) Law. The issuance of fictitious tax
invoices is carried out by taxpayers because of the ease of VAT refunds and the results
of tax evasion will be greater than the costs incurred (Yamin & Putrantri, 2009). Of course,
if it is not handled, then tax revenue from the VAT sector will decrease and harm the
state.
Given the difficulty of supervising all 46,380,119 taxpayers in Indonesia, the DGT
can utilize data mining to analyze and make decisions based on data. Data Mining is the
process of collecting important information which can be in the form of correlations,
patterns, and trends from large data using artificial intelligence, statistical techniques,
mathematics, machine learning, and so on (Larose, 2005).
Looking at previous research, Wu et al.(2012) have implemented data mining on
VAT reporting compliance in Taiwan to conduct screening and audits more efficiently.
35
This research contributes that the use of data mining can support screening activities for
inappropriate VAT Reports in a more scientific way compared to relying on audits based
on manual methods and just personal judgments. The results of the data mining are
expected to be combined with the personal experience of the tax auditor to obtain more
effective and efficient results. Naturally this will help DGT in auditing considering that the
audit coverage ratio is still low at 1.54%.
In order to determine which is the best model produced by data mining activities,
Gupta dan Nagadevara (2007) measure model performance using Strike Rate (SR),
Prediction Efficiency (PE), and Reduction in examination effort (EF). This research
prioritizes precision compared to recall with the reason that audits are carried out
accurately and produce a deterrent effect on other taxpayers.
This study intends to use data mining as a tool for screening taxpayers who issue
fictitious tax invoices. It is hoped that the results of this study can increase the
effectiveness and efficiency of tax audit activities by the Directorate General of Taxes,
especially in the VAT sector.
2. THEORITICAL FRAMEWORK
2.1 Theory of Planned Behavior
Theory of Planned Behavior is explained by Ajzen (1991) as a person's intention to
perform a certain behavior. A person's intention can be seen from attitudes toward
behavior (attitude), subjective norms related to behavior (subjective norms), and control
over perceived behavior (perceived behavioral control). The measure of effort expended
to perform a particular behavior reflects the level of a person's intention. TPB also argues
that a person will perform certain behaviors depending on the benefits and costs of an
activity.
Simply put, a certain behavior can describe a person's intentions. In connection with
this study, the intention of the taxpayer should be seen from his behavior which will then
be analyzed using data mining techniques.
2.2 Theory of Tax Evasion

The difference between tax avoidance and tax evasion lies in the legality of the
actions taken by the taxpayer. Sandmo (2021) explains tax avoidance as an act of legal
by taking advantage of legal loopholes to reduce the tax that must be paid. Taxpayers
do not need to worry if tax avoidance activities are detected by the tax authorities
because their actions do not violate the law. Meanwhile, tax evasion is explained as
reducing tax through violation of the law by not reporting income that should be taxed or
carrying out other illegal activities that make him responsible for administrative actions
from the tax authorities.
Research on tax evasion was conducted by Chen et al. (2010). In this study it was
found that family companies tend to have a lower level of tax aggressiveness than other
companies. This happens because the family company really cares about the
sustainability of the company and avoids conflicts with the tax authorities so as not to
tarnish the family name. From this study it can be concluded that company owners can
also influence the level of tax aggressiveness.
In this case, the taxpayer will be worried if the tax evasion activity is detected by the
tax authorities because their actions violate the law. Related to TPB, if the taxpayer has
the intention to carry out tax evasion, then the taxpayer will show a certain behavior.
2.3 Economic Deterrence Model

This model incorporates the concept of an economically rational taxpayer who will
avoid taxes as long as the results of illegal tax evasion are greater than the penalties
36
received when caught by the tax authorities (Hasseldine & Bebbington, 1991). Using this
model, taxpayers should not commit illegal tax evasion as long as the penalty received
is greater than the profit earned. In this regard, of course taxpayers who commit illegal
tax evasion need to be detected by the tax authorities so that they can be punished.
Tax authorities should focus on preventing tax evasion through tax audits and
penalties for non-compliance (Carvalho & Pacheco, 2014). Supporting this,
Wu et al (2012) have implemented data mining in auditing SPT Period VAT in Taiwan
and showed that data mining can screen non-compliant taxpayers with a more scientific
method than just conducting random audits or experience. auditors only. With the help
of data mining, it is hoped that it can cause a deterrence effect on other taxpayers and
reduce the tax evasion that is carried out.
2.4 Value Added Tax

Value Added Tax is a tax imposed on the delivery of taxable goods (BKP) and/or
taxable services (JKP) within the customs area (territory of the Republic of Indonesia) as
regulated in Undang-undang Nomor 8 Tahun 1983 tentang Pajak Pertambahan Nilai dan
Pajak Penjualan atas Barang Mewah. Value Added Tax is objective, not cumulative, and
is an indirect tax with tax subjects consisting of Taxable Entrepreneurs and non-PKP.
In practice, the tax subject does not deposit his tax directly into the state treasury,
but deposits it through the party that deducts the VAT. In its calculation, Value Added
Tax has Input Tax which can be credited and Output Tax which becomes tax debt. The
amount of this Input Tax directly affects the amount of VAT that needs to be deposited
to the state because it reduces the tax debt from the Output Tax. If the Input Tax has a
larger amount than the Output Tax, then the taxpayer can get a refund.
2.5 Fictitious Tax Invoice

Each Input Tax (Pajak Masukan) and Output Tax (Pajak Keluaran) that is reported
in the Taxpayer Periodic VAT SPT is recorded based on the Tax Invoice received and/or
issued by the Taxpayer himself. VAT-related tax crimes often occur because tax invoices
are not issued based on actual transactions so that taxpayers receive input tax that is
greater than they should be and reduce the amount of VAT that must be deposited to
the state.
According to Peraturan Direktur Jenderal Pajak Nomor PER-19/PJ/2017 Tentang
Perlakuan Terhadap Penerbitan Dan/Atau Penggunaan Faktur Pajak Tidak Sah Oleh
Wajib Pajak , Fictitious Tax Invoices are Tax Invoices issued not based on actual
transactions and/or Tax Invoices issued by Entrepreneurs who have not been confirmed
as Taxable Entrepreneurs.
For this crime, the Directorate General of Taxes imposed a penalty based on
Undang-undang Nomor 6 Tahun 1983 Tentang Ketentuan Umum Dan Tata Cara
Perpajakan (UU KUP) Sebagaimana Telah Beberapa Kali Diubah Terakhir Dengan
Undang-undang Nomor 7 Tahun 2021, namely imprisonment for a minimum of 2 years
and a maximum of 6 years and a fine of at least 2 times the amount of tax on the tax
invoice, proof of tax collection, proof of tax withholding, and/or proof of tax payment and
a maximum of 6 times the amount of tax on the tax invoice, proof of tax collection, proof
of tax withholding , and/or proof of tax payment.
3. Research Method
This study aims to build a model for the detection of tax invoice issuers that are not
based on actual transactions. This research was conducted using mixed methods,
involving both qualitative and quantitative analysis. These two methods are used in the
development of machine learning models with reference to the data mining process. The
research flow will follow the Cross Industry Standard Process for Data Mining (CRISP-
37
DM), as revealed by Brown (2014) as a step-by-step data mining process created by
data miners for data miners. The data mining process is divided into six cycle stages,
which are:
a) Business Understanding
This stage is carried out with a qualitative approach to obtain information about
business problems in the research object which is the case study site.
b) Data Understanding
After information about business problems is learned, the next step is to collect data
that is relevant to the research being conducted. Data acquisition was carried out by
observing DGT's Data Warehouse, conducting profiling to find out the amount, size
and type of data for each data element, and extracting structured data for use in this
study.
c) Data Preparation
After the data is obtained, then the initial processing is carried out to form a labeled
dataset. This stage includes the extraction of independent variables/features that
are used as predictors and the activity of labeling the class that is the target of the
prediction.
d) Modeling
This stage includes activities to train machine learning models with either a single
algorithm or an ensemble algorithm. Examples of single algorithms include K
Nearest Neighbor, Decision Tree, and Logistic Regression, while examples of
ensemble algorithms include Random Forest and AdaBoost. This activity will
produce many machine learning models to compare their performance at the
evaluation stage.
e) Evaluation
This stage includes activities to compare model performance and select the machine
learning model that has the best performance. The performance measure of the
model used is the F1-Score which is the harmonic mean of precision and recall.
According to Gupta & Nagadevara (2007), the model can be further evaluated for
tax audit purposes as follows:
 The model's ability to predict the Issuing Taxpayer is measured using the
Prediction Efficiency (PE) with the formula PE = TP/ (TP+FN).
 Model's ability to reduce the effort required in an audit by reducing in
Examination effort (EF) with the formula EF = 1-(FP+TN)/Total Cases.
 The model's ability to obtain Issuer Taxpayers if each taxpayer predicted to be
an issuer is audited can be measured using the Strike Rate (SR) with the
formula SR=TP/(TP+FP).
f) Deployment
After the model with the best performance measurement is obtained, then the model
is applied to an application so that it can be used to detect invoice issuers that are
not based on actual transactions..
4. RESULT DAN DISCUSSION

The research was conducted using Jupyter-Notebook which is supported by other
libraries available in the Package Installer for Python (PIP) which is commonly used for
data mining activities. The research was conducted by following the six stages of the
CRISP-DM model with the following explanation.
38
4.1 Business Understanding
One of the threats to tax revenues is tax evasion in withholding taxes, such as Value
Added Tax (VAT). The embezzlement scheme used is in the form of issuing and using
tax invoices that are not based on actual (fictitious) transactions.
VAT-Registered businesses collect VAT from sales transactions without issuing a
tax invoice, then issue a fictitious tax invoice to other parties (with a smaller value than
the actual transaction for which a tax invoice was not issued), the tax invoice is used as
a tax credit by other parties, thereby reducing the amount of tax to be paid by the party.
Puppet company networks were created to disguise this embezzlement scheme with a
series of fictitious transactions, so that this scenario is difficult to detect by tax examiners
(Mehta et al, 2009).
4.2 Data Understanding

This study aims to build a detection model for fictitious invoice issuers using taxpayer
characteristic data as a predictor. Characteristics of taxpayers can be obtained from data
on submission of tax returns (SPT Masa PPN), both annual tax returns and periodic VAT
returns.
The Directorate General of Taxes has implemented a data warehouse. All data from
various business processes have been collected in DGT's Data Warehouse. The
required data is obtained by querying the data warehouse.
Predictive target data for the machine learning model to be built are obtained from
law enforcement activities, namely preliminary evidence checks and/or year
investigations with a time span of 2018-2020, as well as eFaktur databases and
Taxpayer Master Files with a timeframe of 2014-2020.
Data is received in the form of comma separated values (.csv) from different tables
and databases so that it is prone to getting null values when merging is performed. The
file received has the smallest size having a size of 59 Kilo Bytes containing approximately
1500 lines of data and the file with the largest size having a size of 600 Mega Bytes
containing approximately 10,000,000 lines of data.
4.3 Data Preparation

Data diolah menggunakan dengan Pandas yaitu library Python yang digunakan
untuk mengolah panel data. Data-data dari tabel-tabel tersebut dilakukan proses
merging hingga membentuk satu tabel data dengan variabel-variabel berikut.
Table 1 Variables used in Data Mining

Kode. Variabel Tipe Data
X1 Ratio of Underpayment to Output VAT Float
Age of VAT-Registered Business (categories divided into
X2 int
0-3, 4-6,7-9, and >10)
X3 Number of output VAT sheets (average) int
X4 Number of input VAT sheets (average) int
X5 Number of Consignment in a year (average) Float
X6 Number of employees on SPT PPh Article 21 (TLK) int
X7 SPT PPh 21 Reporting Compliance (average) boolean
X8 VAT retuns compliance (average) int
X9 Value of Input VAT in a year (average) Float
Y Attributes of the Taxpayer issuing the Fictitious Invoice boolean
39
Source : Processed by Authors
The merging process is carried out using the inner join method. For positive label
data, 1,071 rows of data were obtained from the initial data of 1,551 rows of data.
However, there is still a null value in the number of employees variable which is then
filled in using linear regression based on the number of submissions in a year, the Input
VAT value, and the SPT PPh 21 Reporting Compliance based on the highest correlation
value.
Manual labeling is carried out by referring to the report on the results of preliminary
evidence inspection activities. A taxpayer with a positive label is a taxpayer who issues
an invalid tax invoice and a taxpayer with a negative label is a taxpayer who does not
issue an invalid tax invoice. Taxpayers are classified as issuers of invalid tax invoices if
the taxpayer issues Tax Invoices that are issued not based on actual transactions as
stipulated in PER-19/PJ/2018 concerning Treatment of Issuance and/or Use of Invalid
Tax Invoices by Taxes as explained in previous section. Taxpayer data with a positive
label comes from Taxpayer data that has been examined by preliminary evidence and/or
investigations contained in the SIGAKUM database for the period 2018 to 2020. Then
for taxpayer data with a negative label taken by sampling from strategic taxpayers at
Master Taxpayer Files and e-Faktur databases that have never been subject to
preliminary evidence inspection.
The negative label data obtained as many as 57,541 lines of data which were then
taken randomly as many as 2,142 lines of data as training data and 1000 lines of data
as deployment data. Each variable in table 1 is explained as follows.
4.3.1 Attributes of the Taxpayer issuing the Fictitious Invoice

This variable contains positive label data from SIGAKUM database for 2018-2020
which is then combined with negative label data randomly from the eFaktur and Taxpayer
Master File databases as previously described. The training data used has a distribution
of labels which is illustrated by the following graph.
Figure 2 Distribution of Positive (Penerbit) and Negative (Bukan Penerbit) Label Data
Source : Processed by Authors using Jupyter-Notebook
40
As can be seen in the figure, the distribution of labels is divided into one third of
positive label data and two thirds of negative label data to avoid imbalanced data.
4.3.2 Ratio of Underpayment to Output VAT

The variable ratio of underpayment to output VAT is obtained from the amount of
underpayment or overpayment coupled with the output VAT reported by the taxpayer in
the Periodic VAT reports. The value of this variable is taken as the average value per
year.
4.3.3 Age of VAT-Registered Business

The age of VAT-Registered business is taken from the year the taxpayer was
confirmed or established himself as a VAT-Registered business in the Taxpayer Master
File, reduced by the first year's data recorded in the SIGAKUM database for data with a
positive label and the last year's data is in the eFaktur database for data with negative
labels.
The data has a high range of values from each other, so it is simplified into divided
category types as explained in the following table.
Table 2 Age Division into Categories

Categories Age Range
0 0-3 years
1 4-6 years
2 7-9 years
3 >10 years
Source : Processed by Authors
4.3.4 Number of output VAT sheets and Number of input VAT sheets
For the variable Number of Output VAT Sheets and Number of Input VAT Sheets,
the average value of the taxpayer issuing a tax invoice and crediting the tax invoice were
taken from eFaktur database.
4.3.5 Number of Consignment in a year

The Variable Amount of Delivery in a year is the variable the value of the goods
and/or services delivered by the taxpayer to another party, which in general is a sale and
purchase transaction of taxable goods or taxable services. This variable is the basis for
the imposition of output VAT which becomes a tax debt for taxpayers. Data for this
variable is obtained based on the average value in the eFaktur database.
4.3.6 Number of employees on SPT PPh Article 21

The variable number of employees on SPT PPh Article 21 is obtained based on the
last value or Take Last Known (TLK) in the Taxpayer's Master File based on the latest
SPT PPh Article 21 reported by the taxpayer. Naturally, if the taxpayer has high value
and routine submissions, then the taxpayer requires a large number of workers as well.
4.3.7 SPT PPh 21 Reporting Compliance

The SPT PPh 21 Reporting Compliance variable is obtained based on the average
number of taxpayers reporting their SPT PPh 21 Reports per year. Variables have a
minimum value of 0 and a maximum value of 12.
41
4.3.8 VAT reports compliance
VAT reports Compliance variable is obtained based on the average number of
taxpayers reporting VAT reports each months per year as recorded in the eFaktur
database. The authors did not use the Annual Tax Returns variable because from the
data received, every registered taxpayer complied with reporting the Annual SPT, so the
writing team used the VAT reports Compliance variable to test monthly taxpayer
compliance with VAT obligations.
4.3.9 Value of Input VAT in a year

The input VAT value variable in a year is obtained from the average number of Input
VAT credited by the taxpayer during the year. This variable is suspected to have a
relationship with the fictitious tax invoices because input VAT will reduce the amount of
tax debt that must be paid to the state.
4.4 Modeling
After the training data is prepared, the next step is to do the modeling or variable
fitting of the machine learning algorithm for data mining. At this stage, the authors used
libraries as shown in the following figure.
Figure 3 Libraries used for Data Mining
Source : Processed by Authors using Jupyter-Notebook
The machine learning algorithms used in this study are Logistic Regression, decision
tree, random forest, and KNN because they adjust to the capabilities of the computer
equipment that is owned by authors and the level of popularity that is commonly used in
data mining classification techniques (Gong, 2022). Then the AdaBoost algorithm was
also added because the previous four algorithms were bagging methods and one
boosting method was added to compare their performance.
The five algorithms are measured using the F1-Score because even though
precision is the main goal of the writing team, Gupta & Nagadevara (2007) explain that
recall is also needed to prove the model can be used even if only to certain limits.
Considering that precision and recall are needed, the F1-Score is the right measure in
measuring a model's ability (Lanier, 2020).
In calculating the F1-Score, the writing team uses K-Fold Cross Validation with a
number of folds totaling 10. Cross Validation will divide the training data into 10 parts
with 9 parts for training data and 1 part for data testing which is then tested 10 times to
get more objective test results. After fitting is done to each algorithm, the models are
compared using the subplots depicted in the following graph.
42
Figure 4 Each Models compared using Subplot graphic
Source : processed by authors using Jupyter-Notebook
Based on Figure 4 above, the Random Forest is the best model because it has the
top position and has a fairly low standard deviation compared to the other four models.
The mean value of the F1-Score for each model can be seen in the following table.
Table 3 F1-Score comparison of each models

Algorith models F1-Score (mean)
Logistic Regression 26,17%
Decision Tree 84,66%
Random Forest 90,61%
AdaBoost 86,60%
kNN 64,78%
From the Random Forest algorithm model obtained a True Positive Rate of 84%, a
True Negative rate of 96%, a Positive Precision Rate of 91% and a Negative Precision
Rate of 92%. When visualized with the confusion matrix, jupyter-notebook generated the
following image.
43
Figure 5 Confusion Matrix of Random Forest algorithm model
4.5 Model Evaluation

Out of the five previous algorithm models, the Random Forest algorithm model was
chosen as the best algorithm model for the data you have. Then tuning is applied using
the GridSearchCV function to determine the best parameters for using the Random
Forest algorithm model. The parameters for tuning are n_estimator or the number of
trees in the model and max_depth or the size of one tree in the model. The parameter
tuning is done again using Jupyter-Notebook as seen in the following screenshot.
Figure 6 Syntax used for parameter tuning
44
Based on the results in Figure 6 above, the Random Forest algorithm model will
work better if it uses max_depth = 17 and n_estimators = 59 compared to 1392 other
candidates with an F1-score of 90.90%. These parameters are used in the Random
Forest which then produces a confusion matrix as follows.
Figure 7 Confusion Matrix of Random Forest Model After Tuning

The Random Forest model after tuning has a True Positive Rate of 83.56%, a True
Negative Rate of 96%, a Positive Precision Rate of 90.9%, and a Negative Precision
Rate of 93%. Then for further evaluation with audit purposes, the model has the following
capabilities.
a. PE of 895/895+176 = 83,56%;
b. EF of 1-(91+895)/3213 = 69,31%; and
c. SR of 895/(895+91) = 90,77%.
Out of the three evaluations, the strike rate is preferred because the main goal of
modeling is to be able to channel resources more effectively and efficiently, so that when
the auditor conducts an audit of a taxpayer who is suspected of being the issuer, the
audit has a high probability of getting the same results as predicted. Prediction Efficiency
is used to prove that the model does have the ability to predict the issuing taxpayer better
than random audits. Then the model is saved using the pickle library in (.sav) format.
4.6 Deployment
At the deployment stage, the saved model is implemented in a simple application
using Streamlit. Streamlit is a python framework that is used to create web-based
applications that are commonly used for data mining deployments. The application that
has been made has the following display
45
Figure 8 Interface of Application Deployment using Streamlit
Source : processed by authors using python and web browser
This application accepts an input file in the form of comma-separated values (.csv)
with the order of the NPWP (Tax Identification Number) column, the ratio of
under/overpayment of output VAT, consignment in a year, input VAT in a year, total VAT
reports in a year, input tax invoice sheets in a year, output tax invoices sheets in a year,
the number SPT PPh 21 per year, the number of employees, and the age category of
VAT-registered Business. The application also provides a slider to adjust the probability
value provided by the Random Forest so that the application can only display taxpayers
who are predicted to be issuers with a certain level of probability.
Furthermore, the deployment data that has been prepared at the data preparation
stage is uploaded to the Upload list menu of VAT-Registered Business lists and the slider
is shifted to 0.75 for slider testing and simplification of the display of results.
46
Figure 9 Display of taxpayers who are predicted as issuers by applications with a
minimum probability value of 75%
Source : processed by authors using python and web-browser
As can be seen in Figure 9, the application will only display the NPWP of the
taxpayer who is predicted to be the issuer and has a minimum probability of 75%
according to the slider. The results show that there are 8 taxpayers who are suspected
of being issuers with a probability exceeding 75% which are displayed sequentially based
on the highest probability with the aim of the top taxpayers being prioritized for a tax audit.
5. CONCLUSIONS
Based on the research that has been described previously, the conclusion that can
be drawn is that Random Forest is the most suitable algorithm model to use for the data
owned compared to the Logistic Regression, Decision Tree, kNN, and AdaBoost
algorithm models with
F1-score of 90.61%. The Random Forest algorithm model was then parameter tuned
to the n_estimator and max_depth parameters using GridSearchCV which obtained an
F1-Score of 90.90% which indicated that there was no significant difference between the
Random Forest algorithm model without tuning and the model with tuning.
The Random Forest algorithm model is evaluated to be able to correctly predict
invalid invoice issuer taxpayers if each taxpayer is examined with a Strike Rate/SR of
90.77%. Then the model can reduce the number of examinations compared to random
examinations with a reduction in examination effort/EF of 69.31%. The model can also
classify fictitious tax invoice issuer from all actual fictitious tax invoice issuer taxpayers
with a Prediction Efficiency/PE of 83.56%.
47
Furthermore, the model that has been created can be used in simple applications
using Streamlit and can predict that 8 taxpayers are taxpayers who issue invoices with
a probability that exceeds 75% of the 1000 taxpayer data in the deployment data. For
deployment with different data, this can be done as long as the file uploaded is in the
form of comma-separated values (.csv) and has a column for NPWP (Tax Identification
Number) column, the ratio of under/overpayment of output VAT, consignment in a year,
input VAT in a year, total VAT reports in a year, input tax invoice sheets in a year, output
tax invoices sheets in a year, the number SPT PPh 21 per year, the number of
employees, and the age category of VAT-registered Business sequentially.

This study has limitations on positive label data considering that the audit coverage
ratio is still low at 1.54% and limitations on negative label data which are believed to be
truly negative considering that negative label data could be an invalid invoice issuer
taxpayer but has not been examined or has not been proven.
In the preparation of this study there were still some shortcomings and limitations,
which are the limited time of the authors who carried out the research for approximately
two months and limited data due to confidentiality of the position and limited access so
that the authors could not use more data and methods in the research. Research also
limits the use of positive label data samples with a time range of 2018-2020
7. ACKNOWLEDGEMENTS
The team of authors would like to thank our colleague Ajar Parama Adhi, for his
assistance in obtaining the data needed for this study. In addition, the team of authors
would like to thank:
a. Mr Ag. Sigit Adi Satmoko, Head of Data Science Section at the Directorate of Tax
Data and Information, Directorate General of Taxes; and
b. Mr. Leonard Yulianus, Head of General Subdivision at the KPP Pratama Sanggau,
Directorate General of Taxes;
for the guidance and input given to the authors during this research process.
8. REFERENCES
Ajzen, I. (1991). The theory of planned behavior. Organizational Behavior and Human
Decision Processes, 50(2), 179–211. https://doi.org/10.1016/0749-5978(91)90020-
T
Brown, M. S. (2014). Data mining for dummies. John Wiley & Sons, Inc.
Carvalho, D., & Pacheco, L. M. (2014). Tax compliance, corruption and deterrence: An
application of the slippery model.
http://repositorio.uportu.pt:8080/handle/11328/837
Chen, S., Chen, X., Cheng, Q., & Shevlin, T. (2010). Are family firms more tax aggressive
than non-family firms? Journal of Financial Economics, 95(1), 41–61.
https://doi.org/10.1016/J.JFINECO.2009.02.003
Direktorat Jenderal Pajak. (2021). Laporan Tahunan DJP 2020.
Gupta, M., & Nagadevara, V. (2007). Audit selection strategy for improving tax
compliance: application of data mining techniques. Foundations of Risk-Based
Audits. Proceedings of the Eleventh International Conference on e-Governance,
28–30.
48
Hasseldine, D. J., & Bebbington, K. J. (1991). Blending economic deterrence and fiscal
psychology models in the design of responses to tax evasion: The New Zealand
experience. Journal of Economic Psychology, 12(2), 299–324.
https://doi.org/10.1016/0167-4870(91)90018-O
Lanier, S. T. (2020, October 27). Choosing Performance Metrics.
TowardsDataScience.Com. https://towardsdatascience.com/choosing-
performance-metrics-61b40819eae1
Larose, D. T. (2005). An introduction to data mining. Traduction et Adaptation de Thierry
Vallaud.
Undang-undang Nomor 8 Tahun 1983 tentang Undang-undang Nomor 8 Tahun 1983
tentang Pajak Pertambahan Nilai dan Pajak Penjualan atas Barang Mewah
sebagaimana telah beberapa kali diubah terakhir dengan Undang-Undang Nomor
21 Tahun 2021, (2009).
Peraturan Direktur Jenderal Pajak Nomor PER-19/PJ/2017 tentang Perlakuan Terhadap
Penerbitan dan/atau Penggunaan Faktur Pajak Tidak Sah oleh Wajib Pajak, (2017).
Sandmo, A. (2005). The Theory of Tax Evasion: A Retrospective View. National Tax
Journal, 58(4), 643–663. https://doi.org/10.17310/NTJ.2005.4.02
Gong, D. (2022, February 23). Top 6 Machine Learning Algorithms for Classification.
TowardsDataScience.Com. https://towardsdatascience.com/top-machine-learning-
algorithms-for-classification-2197870ff501
Undang-undang Nomor 6 Tahun 1983 tentang Ketentuan Umum dan Tata Cara
Perpajakan (UU KUP) sebagaimana telah beberapa kali diubah terakhir dengan
Undang-undang Nomor 7 Tahun 2021.
Wu, R.-S., Ou, C. S., Lin, H., Chang, S.-I., & Yen, D. C. (2012). Using data mining
technique to enhance tax evasion detection performance. Expert Systems with
Applications, 39(10), 8769–8777.
Yamin, L., & Putrantri, T. M. (2009). Model Penyelewengan Pajak Menggunakan Faktur
Pajak Fiktif. Bisnis & Birokrasi, Jurnal Ilmu Administrasi Dan Organisasi, 16(1), 1–
7.
P. Mehta, J. Mathews, S. Kumar, S. K., C. Babu, and S. Rao, “Big Data Analytics for
Nabbing Fraudulent Transactions in Taxation System,” 2019, pp. 95–109.
49
IMPACT STREET EARNINGS ON TAX AVOIDANCE
Yohana
Pramita Indonesia University, Tangerang, Indonesia
Email : rayakahyan@gmail.com
Abstract
This study discusses the effect of street earnings on tax avoidance for two reasons. First, the potential
for companies that always minimize tax avoidance, especially after the COVID-19 pandemic. Second,
preventive actions from various parties, including independent commissioners, public accounting firms
including securities analysts who carry street earnings related tax avoidance activity. This study was
conducted on manufacturing companies for the 2015 – 2020 period. This study also added an
expansive test without taking two controlling variables. As a result, street earnings have a significant
negative effect on tax avoidance in expansive testing, meaning recommended companies by analysts
potentially do tax avoidance. The practical implication is the role of securities analysts as recommended
parties for company monitoring. The theoretical implication is transitory items excluded increase
earnings relevant in various tests so further studies are needed such as the effect of street earnings
on corporate governance.
Keywords: street earnings, transitory items, tax avoidance
Kajian ini membahas pengaruh nilai street earnings terhadap penghindaran pajak karena dua alasan.
Pertama, potensi perusahaan yang selalu memininalkan beban pajak khususnya pasca pandemic
covid 19. Kedua, perlunya tindak pencegahan dari berbagai pihak antara lain komisaris independen,
kantor akuntan publik termasuk analis sekuritas yang mengusung street earnings. Kajian ini dilakukan
pada perusahaan manufakur periode 2015 – 2020. Kajian ini juga menambahkan pengujian ekspansif
yaitu tanpa memperhitungkan dua variabel pengendali. Hasilnya, street earnings berpengaruh negatif
signifikan terhadap penghindaran pajak pada pengujian ekspansif, artinya perusahaan yang dinilai baik
oleh analis berpotensi melakukan penghindaran pajak. Implikasi praktis dari kajian ini ialah
menegaskan peran analis sekuritas sebagai pihak yang mampu memonitoring perusahaan. Implikasi
teoritisnya yaitu tanpa menyertakan transitory item, earnings juga relevan dalam berbagai pengujian
sehingga dibutuhkan kajian lanjutkan contohnya pengaruh street earnings terhadap tata kelola
perusahaan.
Kata kunci: street earnings, transitory item, penghindaran pajak.
1. INTRODUCTION
The Covid-19 pandemic has created disasters and challenges for many people in
developing countries as well as economic development, especially for countries with the
weak health systems, poor populations, and weak social safety nets. It is impact rising
commodity prices globally. This weakness is worst by the lack of budgets in several
countries, so government action to progress economically is limited (Kamin & Clements,
2021).
In Indonesia, the impact of the Covid-19 pandemic was quite heavy. More than 1.5
million workers were cut off where 90 percent were laid off temporarirly and 10 percent
were laid off permanently. Inflation increased until March 2020 reaching 2.96 which is
contributed by the price of gold jewelry and food commodities. For the manufacturers
industry, performance of processing industries, from production, new orders, and
employment was decreased as measured by Manufacturing PMI numbers (Hanoatubun,
2020).
Companie’s performance had an impact due to the Covid-19 pandemic. Return on
Asset (ROA) of consumer goods sector decline because people's mobility was limited.
Virus fearness causes activities outside reduce and focus on shopping for basic needs
50
(Junaidi & Salim, 2021). In the property sector, the Covid-19 pandemic also had a
significant negative impact on performance. This pandemic has proven to reduce
people's interest to buy property because their funds only focused on daily needs and
kept as a precaution. Investors also refrain from making any investment because they
see an uncertain economy (Lowardi & Abdi, 2021). Hotel, restaurant, tourism sector
(Esomar & Chritianty, 2021) and financing sector (Esomar, 2021) in similar conditions.
The Indonesia government anticipates performance decline by providing various
incentives including ease delivery of tax notification letters, facilitating import activities,
reducing Corporate Income Tax (Tambunan, 2020), releasing incentives for Income Tax
Article 21 DTP, Income Tax Article 22 import DTP, Final Income Tax / UMKM DTP,
reducing installments of PPh article 25 by 50% and preliminary VAT refund (Ermanis et
al., 2021).
All incentives support aimed national economy back to normal, including restore
companies performance, but there is a negative side, namely tax avoidance. This
practice more significant increase during the Covid–19 pandemic than before through
tax incentives (Barid & Wulandari, 2021) although other studies stated that there were
no significant differences in tax avoidance practices before and during the pandemic
(Firmansyah & Ardiansyah, 2021) means this practice will always exist.
For this reason, it is urgently to prevent this practice, for example presence of
independent commissioner. Unfortunately, role of them are not significant for tax
avoidance (Wardana & Wulandari, 2021), and will reduce control quality (Sparta &
Purnama, 2021) even though they have a lot experience in accounting practices
(Gunawan, 2022). Contrary, independent commissioners can reduce tax avoidance
because they can supervise management to comply with laws and regulations (Wijayanti
& Merkusiwati, 2017).
The second preventive action is through external auditors but unfortunately greater
public accountant firm (KAP) facilitates companies doing tax avoidance (Suyadnya &
Supadmi, 2017). Auditor with high competency was needed (Cahyadi et al., 2020) and
highest fee audit to extend audit scope will barrier tax avoidance (Cahyadi et al., 2020;
Suyadnya & Supadmi, 2017)
The third preventive action and will be tested in this study is role of a security analyst.
Analysts are considered more independent because their function is only as intermediary
between investors and companies (Barker & Imam, 2008) and provides value to
companies known as street earnings (Sadique & Sheikh, 2013) or I/B/E/S earnings
(Entwistle et al., 2010) or analyst consensus earnings (Barth et al., 2012). The value
provide by analysts is better because it is more informative (Bhattacharya et al., 2003),
more predictive (Barth et al., 2012) and more persistent (Brown & Sivakumar, 2003).
Other studies show that street earnings is more influence on stock prices (Rachmawati
& Susilawati, 2008), more effect on stock returns (Cohen et al., 2007) and can determine
firm value (Z. ikhsan Pane et al., 2021). Street earnings also used to indicate financial
distress (Z. I. Pane, 2021b) and more consistent with future earnings if there are few
independent directors (Frankel et al., 2011). Street earnings are considered superior
because of some adjustments (Huang & Skantz, 2016), by exclude certain accounts, for
example gains / losses on sales of asset, claims due to legal liability (Heflin et al., 2015)
or known as special items (Christensen et al., 2011), or transitory effects (Heflin et al.,
2015) or non-recurring items (Baik et al., 2009; Bradshaw & Sloan, 2002; Chen, 2010).
From the explanation above, it can be summarized as follows. First, the background
of this study is urgently need for preventive action against tax avoidance because this
practice is always exist when the companies in normal or difficult conditions, for example
due to the Covid-19 pandemic. Beside external auditors and independent commissioners
who have been studies previously, the role of securities analysts needs to be considered
because they have not personal interest and their function is only intermediary between
51
companies and investors. Second, research purposes this study is to examine the effect
of street earnings values on tax avoidance. This study is needed to find out whether
street earnings which initiated by analysts are adequate to determine tax avoidance
practices. Novelty from this study is relation street earnings to other variables besides
financial distress, firms value, stock market prices and other variables that have been
studied previously.
2. THEORITICAL BACKGROUND AND HYPOTESIS

DEVELOPMENT
2.1 Theoritical Background
Signaling theory has 4 components, namely signaler, signal, receiver and feedback.
Anyone can be a signaler as long as they have information as long as not owned by
other people, for example a manager or companes executive. The signaler obtain
information, both positive and negative, for example the initial stages of research on a
product or service. Receivers are from any parties as long has a lack of information
resulting in what is called information asymmetry (Connelly et al., 2011). This theory is
relevant to explain impact street earnings on tax avoidance for three reasons. First, there
is a signaling provider, namely a security analyst who provide street earnings. Second,
receiver are parties related to information on tax avoidance practices, for example
investors and other stakeholders. Third, the signal itself is information if the companies
actually do tax avoidance.
Securities analysts as signalers provide street earnings which calculated without
following any standard rules (Sadique & Sheikh, 2013), by subtracting net income from
financial statements with transitory items / non-recurring items, for example restructuring
costs, acquisition expenses, gains on asset sales , realized investment gains (Gu & Chen,
2004), write-off and revaluation costs, research and development expense, amortization
of goodwill and certain returns from subsidiaries (Bradshaw & Sloan, 2002). In this study,
three accounts are considered as transitory / non-recurring items, namely amortization,
research and development expense and unusual / exceptional items because three item
are considered irrelevant (Z. I. Pane, 2021a). Unusual / exceptional items are also part
of non-recurring items (Howard et al., 2019; Nagar & Sen, 2016).
Tax avoidance is an legal effort for taxpayers because it does not conflict with tax
regulation where the methods and techniques used tend to take advantage of
weaknesses of laws and tax regulations themselves, to minimize the amount of tax owed
(Anggraeni & Octaviani , 2021 cited Barid & Wulandari, 2021). Tax avoidance calculated
by comparing current year tax expense the total income before tax or known as Effective
Tax Rate (ETR). A low ETR value indicate tax avoidance occur because a low ETR value
can be interpreted as a low amount of the tax expense borne in that period (Firmansyah
& Ardiansyah, 2021).
2.2 Hypothesis Development

Increase street earnings indicate companies is valuable by analyst. Four studies
explain it as follows. First, the correlation street earnings with financial distress state
financial distress lower when street earnings increase. This shows companies that
recommended by analysts are low potential for financial distress (Z. I. Pane, 2021b).
Second, the correlation of street earnings with firms value state that increase in street
earnings is in line with firms value (Z. ikhsan Pane et al., 2021). Third, correlation of
street earnings with stock market value stated that increase in street earnings will
increase in line with stock market price (Rachmawati & Susilawati, 2008). The fourth,
correlations street earnings with research and development expense show that the
52
greater research and development expenditure caused higher revision of street earnings
because analysts see uncertainty in the future from this expenditure (Ho et al., 2007).
Meanwhile, tax avoidance practices can be seek from Effective Tax Rate (ETR),
where lower value is interpreted as the lower the tax expense borne by the companies,
means higher the practice of tax avoidance (Firmansyah & Ardiansyah, 2021) by
reducing companies profits or increase corporate debt (Wardana & Wulandari, 2021).
From explanation above, it can be concluded that higher street earnings indicate
companies will recommend by analyst. The higher the value of Effective Tax Rate (ETR)
indicate tax avoidance is lower. This study attempts to link street earnings and tax
avoidance which have never been studied before, so hypothesis proposed as follow:
H1 : street earnings positive significant to tax avoidance.
Independent variable Dependent Variable
STREET EARNINGS Tax Avoidance

(STREET)
(PP)
Control Variable
Debt Ratio (TD)
Firm Size (SIZE)
Figure 1. Conceptual Framework
Control variables used in this study are firm size and total debt / debt ratio in
accordance with three previous studies. First, the bigger companies indicate greater
companies resources to managing their tax burden (Darmawan & Sukartha, 2014 cited
by (Sparta & Purnama, 2021). Second, companies size also has a significant positive
impact to street earnings (Z. ikhsan Pane et al., 2021). Third, the greater leverage in line
with the amount of third party funding then interest expense will be higher. As a result,
companies will use it to reduce corporate tax calculation (Kurniasih & Sari, 2013 cited by
(Sparta & Purnama, 2021).
Regression model of this study as follow :
PP = α + β1 STREET + β2 TA + β3 TD
where,
PP = Tax Avoidance
STREET = Street Earnings
SIZE = Firms size proxied by Total Aset
TD = Debt ratio to total asset.
This study examines impact of street earnings to tax avoidance in consumer goods
companies listed on the Indonesia Stock Exchange from 2015 – 2020. The
manufacturing sector was chosen because it experienced significant declines similar with
property, financing and tourism (Esomar, 2021; Esomar & Chritianty, 2021; Hanoatubun,
53
2020; Lowardi & Abdi, 2021). Companies also have complete financial reports from 2015
– 2020. In addition, the companies has been operating for more than 20 years because
it is assumed to have a better ability for produce output (Kotha et al., 2011).
Companies financial statement are obtained from OSIRIS database. The OSIRIS
database is a fully integrated public companies database and analytical information
solution. The OSIRIS database is produced by Bureau van Dijk Electronic Publishing,
SA. Bureau van Dijk Electronic Publishing, SA (BvDEP) is a company and business
information provider based in Brussels. BvDEP's marketing center is in London and has
branch offices all over the world, such as Amsterdam, Bahrain, Beijing, Bratislava,
Brussels, Chicago, Copenhagen, Edinburgh, Frankfurt, Geneva, Lisbon, London, Madrid,
Manchester, Mexico City, Milan, Moscow, New York , Paris, Rome, San Francisco, Seoul,
Shanghai, Singapore, Stockholm, Sydney, Tokyo, Vienna and Zurich. The OSIRIS
database provides financial report, ownership, news, ranking, earnings, and stock-
quoted data from publicly traded companies around the world, including banking and
insurance companies. The OSIRIS database has information on more than 45,000
companies from 140 countries consisting of 34,000 companies listed on the Stock
Exchange and 11,000 companies that are not listed or are no longer listed on the Stock
Exchange (Wardani & Hermuningsih, 2011).
Financial reports obtain by OSIRIS database are downloaded in excel form with the
World Vest Base (WVB) analyst report format. This format is a financial data report
intended for portfolio managers and researchers (Intelligence, n.d.).
There are 4 variables in this study, one dependent variable, one independent
variable and two control variables. The four measurement variables are described as
follow:
Table 1. Variable Measurement

Variable Measurement
Tax Avoidance (PP) Tax Expense / earnings before tax (EBT)
Street earnings (STREET) Net profit – transitory item / total asset
Firm Size (SIZE) Log Total Asset
Debt ratio (TD) Total debt / Total Asset

4.1 Descriptive test
Manufacturing companies used as samples are 53 companies from 2015 – 2020
with 213 observations exclude outliers. All valid companies are in the food and beverage
sub-sector, basic chemical industry, ceramics and porcelain, automotive and other
sectors. Descriptive statistics explain as follow,
Table 2. Descriptive Statistic

N Minimum Maximum Mean Std. Deviation
Street Earnings (STREET) 213 -.0337 .1913 .080690 .0460089
Debt ratio (TD) 213 .0917 .7715 .369958 .1677579
Firms Size (SIZE) 213 2.9303 3.2804 3.093831 .0781875
Tax Avoidance (PP) 213 .1707 .4937 .332063 .0650326
Valid N (listwise) 213
First, street earnings. This variable has a minimum value of -0.337 and a maximum
of 0.1913. The minimum value provide by loss of PT Beton Jaya Manunggal, Tbk in 2016
Rp. 5,974,737,984 while the maximum provide by profit of PT Selamat Sampurna, Tbk
in 2015 Rp. 427,628,000,000. The average street earnings is 0.080690 (8.0690%) of
54
companies total assets. The standard deviation (0.0460) which is smaller than the
average value (0.0806) indicates there is no significant gap from the lowest street
earnings value. This shows street earnings data as homogeneous.
Second, debt ratio. This variable has a minimum value of 0.0917 and a maximum of
0.7715. The minimum debt value provide by PT Indospring, Tbk in 2020 Rp.
262,519,771,935 while the maximum value provide by PT Indal Aluminum Industry, Tbk
in 2017 Rp. 936,511,874,370. The average debt owned by the companies are 0.369958
or 36% of the total assets. Standard deviation is 0.1677579 which smaller than the
average of 0.369958 indicates there is no sizable gap from the lowest debt value. This
shows debt value data as homogeneous.
Third, firms size. This variable has a minimum value of 2.9303 and a maximum of
3.2804. The minimum total asset provide by PT Alakasa Industrindo, Tbk in 2016 Rp.
136,618,855,000 and the largest total asset provide by PT Astra International, Tbk in
2019 Rp. 351,958,000,000. The average total asset is 3.093831 or Rp. 3,812,052,678.
Standard deviation is 0.0781875, smaller than the average of 3.093831 indicate there is
no sizable gap from the lowest asset value. This shows the asset value data as
homogeneous.
Fourth, tax avoidance. This variable has a minimum value of 0.1707 and a maximum
of 0.4937. The lowest tax avoidance provide by PT Astra International, Tbk in 2020 of
17% of profit before tax and the highest tax avoidance provide by PT Gajah Tunggal,
Tbk in 2020 of 49% of profit before tax. The average value of tax avoidance is 0.332063
or 33% of profit before tax. The standard deviation is 0.0650326, smaller than average
0.332063 indicate there is no significant gap from the lowest tax avoidance. This shows
that tax avoidance data is homogeneous.
4.2 Classic Assumption test

4.2.1 Normality test
Normality test aims to determine the regression model have normal distribution
between independent variable and dependent variable use Kolmogorov-Smirnov. The
results are shown as follow :
Table 3. Kolmogorov-Smirnov test result

N 213
Normal Parametersa,b Mean .0000000
Std. .06347027
Deviation
Most Extreme Differences Absolute .052
Positive .052
Negative -.049
Test Statistic .052
Asymp. Sig. (2-tailed) .200c,d
a. Test distribution is Normal.
b. Calculated from data.
c. Lilliefors Significance Correction.
d. This is a lower bound of the true significance.
Table above shows 0,200 > 0,05, means data was normal distribution.
4.2.2 Multicollinearity test.

Multicollinearity test aims to determine correlation between independent variables
thrugh tolerance and Variance Inflation Factor (VIF) values . The result as follow :
55
Table 4. Multicollienarity test result
Variable Tolerance VIF
Street Earnings 0,812 1,231
(STREET)
Debt ratio (TD) 0,831 1,204
Firms Size (TA) 0,926 1,079
The results above show that tolerance of street earnings, debt ratio and firms size
are greater than 0.1 and the VIF value is less than 10. This indicates no correlation
between variables or no multicollinearity problems
4.2.3 Autocorrelation test

Autocorrelation test aims to determine correlation between the (current) t-period
error and the (previous) t-1 period error in regression model use Durbin Watson. The
results are shown as follow :
Table 5. Autocorrelation test result

dL dU
1,7382 1,7990
Durbin Watson 0,102
Table above shows Durbin Watson is 0.102 < 1.7382, means there is still an
autocorrelation problem but this study still continued because autocorrelation is more
suitable for time series data (Basuki & Prawoto, 2017)
4.2.4 Heteroscedastisity test

Heteroscedastisity test aims to determine inequality of variance from the residual
one observation to another use a scatterplot graph where the points formed must spread
randomly, spread above or below the number 0 on the Y axis. Scatterplot graphic shows
as follow :
Figure 2. Scatterplot graphic
Graphic above show the points on the scatterplot graph spread randomly and are
spread both above and below number 0 on the Y axis, means no heteroscedasticity
problem
56
4.3 Hypothesis test
4.3.1 Simultaneous test (F-test)
The F-test is conducted to determine all independent variables have impact to
dependent variable. F-test can be done by looking at the significance value of F at the
output of the regression results using SPSS with a significance level of 0.05 (α = 5%).
The test results appear as follow :
Table 6. F-test result

ANOVAa
Model Sum of df Mean F Sig.
Squares Square
1 Regression .043 3 .014 3.472 .017b
Residual .854 209 .004
Total .897 212
a. Dependent Variable: Y [TAV]

b. Predictors: (Constant), X3 [ TA ], X2 [ TD ], X1 [ STREET]
From table above, it shows significant value 0.017 < 0.05, means model is fit.
4.3.2 Individual test (T-test)

T-test is used to determine how significant independent variable to dependent
variable partially. T-test do by looking significance level of each variable with significance
<0.05.
The test results appear as follow :
Table 7. T-test result

Coefficientsa
Model Unstandardized Standardize t Sig.
Coefficients d
Coefficients
B Std. Error Beta
1 (Constant) .437 .177 2.468 .014
Street earnings -.130 .106 -.092 -1.224 .222
(SE)
Debt Ratio (TD) .063 .029 .162 2.186 .030
Firms size (TA) -.038 .058 -.046 -.651 .516
a. Dependent Variable: Y [TAV]
Table above shows total debt has a significant positive effect on tax avoidance (Sig
0.030 <0.05)
4.3.3 Determination coefficient (R2)

This test aims to determine how close combination point between dependent
variable looking R square value. The result appear as follow :
Table 8. R2 test result

Model R R Square Adjusted R Std. Error of Durbin-
Square the Estimate Watson
57
1 .218a .047 .034 .0639242 .102
a. Predictors: (Constant), X3 [ TA ], X2 [ TD ], X1 [ STREET]
b. Dependent Variable: Y [TAV]
Table above shows adjusted R – Square 0.034 or 3.4%, means all independent
variables, street earnings, debt ratios and firms size, only explain 3.4% of tax avoidance
while 96.6% is explained by other variables. These results show two indications. First,
street earnings initiated by analysts and two control variables have not adequate to
explain tax avoidance in Indonesia, especially in the manufacturing companies, then that
there are many other variables outside the companies needed to be studied further for
example audit opinion from public accounting firms. Second, exclude transitory items for
street earnings calculation does not adequate indicating tax avoidance.
4.3.4 Hypothesis test

Hypothesis test determine direction and significant impact independent to
dependent variables through multiple regression analysis. The results appear as follow :
Table 9. Hypothesis test result

Variable Koefisien Prediction Significant Result
Constanta 0,437
Street -0,130 (+) 0,222 rejected
Earnings
Debt Ratio 0,063 (-) 0,030** accepted
Firms size -0,038 (+) 0,516 rejected
R square 0,034
F-test F – 0,017
Statistic
Significant 0,00000
Dependent Variable : Tax Avoidance
Significant Level : 1% (*), 5% (**) and 10% (***)
Table above arise three discussion as follow :

a. Street earnings to tax avoidance. The results show street earnings have no
effect to tax avoidance (p-value 0.222 > 0.05), means H1 is rejected. This result
shows value propose by analyst not adequate to determine tax avoidance
practice. Transitory items exclude on calculation of street earnings not enough
to produce reliability like previous studies when determining firms value (Z.
ikhsan Pane et al., 2021), financial distress (Z. I. Pane, 2021b) and stock market
prices (Rachmawati & Susilawati, 2008).
b. Debt ratio.The test results show debt ratio has significant positive effect on tax
avoidance (p-value 0.030 <0.05) means higher debt impact to avoid tax
avoidance. This result is contrary from previous studies which stated that higher
leverage will reduce tax calculations (Kurniasih & Sari, 2013 as cited by Sparta
& Purnama, 2021). On average, companies have 36% debt of total assets
owned, but this debt does not generate tax burdens, for example payables to
third parties and related parties, customer advances and accrued expenses,
namely transportation, promotion, utilities and audit expenses.
c. Firms size. The test results show that firm size has no significant effect on tax
avoidance (significance 0.516 >0.05). This result contrary from previous studies
where the greater companies assets will increase tax expense (Darmawan &
Sukartha, 2014 as quoted by (Sparta & Purnama, 2021). This study uses
58
manufacturing companies where majority assets are trade receivables, land,
machinery, vehicles and office inventory which do not necessarily have a
significant tax expense, while the average significant tax expense result from
net profit multiplied by a certain percentage according to with article 17
paragraph 1 part b of Law number 26 of 2008 concerning PPh
4.4 Expansive test

This test to drive alternative results by eliminating two control variables and confirm
whether street earnings and tax avoidance has no effect as well as previous results. The
result as follows:
Table 10. Expansive test with and without control variable

Variable With control variable Without control variable
Coefficient Significant Coefficient Significant
constanta 0,437 0,350
Street earnings -0,130 0,222 -0,226 0,020 **
Debt ratio 0,063 0,030**
Firms size -0,038 0,516
R square 0,034 0,021

Dependen Variable : Tax Avoidance
Significant level : 1% (*), 5% (**) and 10% (***)
Table above arise three discussion as follow :

a. Adjusted R square. Tests without two control variables show a smaller adjusted
R square value compared with control variable (0.021 < 0.034). It shows control
variable only contributes 0.013 (1.3%) for explanation tax avoidance practice.
These results also show that street earnings still play a greater role in explaining
tax avoidance 0.021 (2.1%) compared with control variables of 0.013 (1.3%).
b. Street earnings. Expansive test proves that without control variables, street
earnings have significant negative effect on tax avoidance, means greater street
earnings will increase tax avoidance. The higher street earnings reflects
companies are profitable and recommended by analysts for invest. As a result,
the company has opportunity to do tax avoidance. The results also show
transitory items exclude in calculation will produce relevant earnings
accordance with previous studies.
c. Control variable. Expansive test proves presence control variable does not add
significance impact to street earnings although it can be useful to add
explanations in research through the adjusted R square value.
5. CONCLUSION.
This study aims to examine effect street earnings on tax avoidance in manufacturing
companies listed on Indonesia Stock Exchange from 2015 – 2020. This research is also
completed with expansive test by removing two control variables for seek role of street
earnings on tax avoidance in-depth. Based on the results, there are two conclusions.
First, street earnings have a significant negative impact on tax avoidance without
two control variables. This shows the earnings proposed by securities analysts can
determine tax avoidance practice.
Second, higher debt ratio reduce tax avoidance. This shows that company debt is
not always related to tax burdens, for example debt to third parties and related parties,
59
customer advances and accrued costs such as freight, promotion, utilities and audit
expenses.
6. IMPLICATION AND RESEARCH LIMITATION

Based on results and conclusions, there are two research implications. First, the
practical implications regarding the role of analysts through street earnings. besides tax
authorities, security analysts also as monitoring partied. Analysts are considered to play
a more independent role because they do not have professional interest with companies,
only as an intermediary between companies and investors then free from companies
pressure to protect them agains tax regulation. Second, the theoretical implications
regarding the value of street earnings. besides able to determine firms value, financial
distress and stock market prices, street earnings are relevant to determine companies
achievement. Then, further research is needed to expand street earnings, for example
correlation street earnings on corporate governance to ensure analyst valuation is fit with
good corporate governance.
Based on conclusions and implications, limitation of this study regarding formulation
of tax avoidance. Comparison of tax expense and income before tax is not full enough,
so additional methods are required for example Cash Effective Tax Rate (CETR) or by
annual report disclosure such as directors ability to understand tax regulations through
number of training and tax seminars followed.
REFERENCE
Baik, B., Farber, D. B., & Petroni, K. (2009). Analysts’ incentives and street earnings.
Journal of Accounting Research, 47(1), 45–69. https://doi.org/10.1111/j.1475-
679X.2008.00311.x
Barid, F. M., & Wulandari, S. (2021). Praktik Penghindaran Pajak Sebelum dan Setelah
Pandemi Covid – 19 di Indonesia. Jurnal Riset Akuntansi & Perpajakan (JRAP),
8(02), 68–74. https://doi.org/10.35838/jrap.2021.008.02.17
Barker, R., & Imam, S. (2008). Analysts’ perceptions of ‘earnings quality.’ Accounting
and Business Research, 38(4), 313–329.
Barth, M. E., Gow, I. D., & Taylor, D. J. (2012). Why do pro forma and Street earnings
not reflect changes in GAAP? Evidence from SFAS 123R. Review of Accounting
Studies, 17(3), 526–562. https://doi.org/10.1007/s11142-012-9192-9
Basuki, A. T., & Prawoto, N. (2017). Analisis Regresi Dalam penelitian Ekonomi dan
Bisnis (dilengkapi Aplikasi SPSS & Eviews). In Rajawali Press.
Bhattacharya, N., Black, E. L., Christensen, T. E., & Larson, C. R. (2003). Assessing
the relative informativeness and permanence of pro forma earnings and GAAP
operating earnings. In Journal of Accounting and Economics (Vol. 36, Issues 1-3
SPEC. ISS.). https://doi.org/10.1016/j.jacceco.2003.06.001
Bradshaw, M. T., & Sloan, R. G. (2002). GAAP versus The Street: An Empirical
Assessment of Two Alternative Definitions of Earnings. Journal of Accounting
Research, 40(1), 41–66. https://doi.org/10.1111/1475-679x.00038
Brown, L. D., & Sivakumar, K. (2003). Comparing the Value Relevance of Two
Operating Income Measures. Review of Accounting Studies, 8(4), 561–572.
Cahyadi, R., Nasaruddin, F., & Lannai, D. (2020). Pengaruh Structure Ownership Dan
Karakteristik Auditor Terhadap Tax Avoidance Pada Perusahaan Dalam Idxv30 Di
Bursa Efek Indonesia. Invoice : Jurnal Ilmu Akuntansi, 2(1), 86–101.
https://doi.org/10.26618/inv.v2i1.3189
Chen, C. Y. (2010). Do analysts and investors fully understand the persistence of the
items excluded from Street earnings? Review of Accounting Studies, 15(1), 32–
60
69. https://doi.org/10.1007/s11142-008-9079-y
Christensen, T. E., Merkley, K. J., Tucker, J. W., & Venkataraman, S. (2011). Do
managers use earnings guidance to influence street earnings exclusions? Review
of Accounting Studies, 16(3), 501–527. https://doi.org/10.1007/s11142-011-9158-
3
Cohen, D. A., Hann, R. N., & Ogneva, M. (2007). Another look at GAAP versus the
Street: An empirical assessment of measurement error bias. Review of
Accounting Studies, 12(2–3), 271–303. https://doi.org/10.1007/s11142-007-9029-
0
Connelly, B. L., Certo, S. T., Ireland, R. D., & Reutzel, C. R. (2011). Signaling theory: A
review and assessment. Journal of Management, 37(1), 39–67.
https://doi.org/10.1177/0149206310388419
Entwistle, G. M., Feltham, G. D., & Mbagwu, C. (2010). The Value Relevance of
Alternative Earnings Measures: A Comparison of Pro Forma, GAAP and Analysts
Actuals*. Journal of Accounting, Auditing and Finance, 519.
https://doi.org/10.2139/ssrn.1014087
Ermanis, Y., Putri, A. A., & Lawita, N. F. (2021). Pengaruh Insentif Pajak Pandemi
Covid-19, Digitalisasi Administrasi Perpajakan dan Omnisbus Law Perpajakan
Terhadap Penerimaan Pajak (Studi Kasus di KPP Pratama Pekanbaru Tampan
Tahun 2020-2021). Bilancia: Jurnal Ilmiah Akuntansi, 5(4), 444–453.
Esomar, M. J. F. (2021). Analisa Dampak Covid-19 terhadap Kinerja Keuangan
Perusahaan Pembiayaan di Indonesia. Jurnal Bisnis, Manajemen Dan Ekonomi,
2(2), 22–29.
Esomar, M. J. F., & Chritianty, R. (2021). Dampak Pandemi Covid 19 terhadap kinerja
keuangan perusahaan sektor jasa di BEI. Jurnal Konsep Bisnis Dan Manajemen,
7(2), 227–233. https://doi.org/10.31289/jkbm.v7i2.5266
Firmansyah, A., & Ardiansyah, R. (2021). Bagaimana Praktik Manajemen Laba Dan
Penghindaran Pajak Sebelum Dan Setelah Pandemi Covid19 Di Indonesia? Bina
Ekonomi, 24(2), 31–51. https://doi.org/10.26593/be.v24i1.5075.87-106
Frankel, R., Mcvay, S., & Soliman, M. (2011). Non-GAAP earnings and board
independence. Review Accounting Studies, 16(July), 719–744.
https://doi.org/10.1007/s11142-011-9166-3
Gu, Z., & Chen, T. (2004). Analysts’ treatment of nonrecurring items in street earnings.
Journal of Accounting and Economics, 38(1-3 SPEC. ISS.), 129–170.
https://doi.org/10.1016/j.jacceco.2004.09.002
Gunawan, A. (2022). Peranan Good Corporate Governance Dalam Meminimalisir
Praktik Penghindaran Pajak Pada Perusahaan Go Publik. Owner : Riset & Jurnal
Akuntansi, 6(1), 379–385. https://doi.org/10.33395/owner.v6i1.618
Hanoatubun, S. (2020). Dampak Covid 19 terhadap perekonomian Indonesia. JOurnal
of Education, Psychology and Counseling, 2(1), 146–153.
Heflin, F., Hsu, C., & Jin, Q. (2015). Accounting conservatism and Street earnings.
Review of Accounting Studies, 20(2), 674–709. https://doi.org/10.1007/s11142-
014-9311-x
Ho, L. C. J., Liu, C. S., & Schaefer, T. F. (2007). Analysts’ forecast revisions and firms’
research and development expenses. Review of Quantitative Finance and
Accounting, 28(3), 307–326. https://doi.org/10.1007/s11156-006-0013-8
Howard, M., Maroun, W., & Garnett, R. (2019). Misuse of non-mandatory earnings
reporting by companies: Evidence from an emerging economy. Meditari
Accountancy Research, 27(1), 125–146. https://doi.org/10.1108/MEDAR-12-2017-
0247
Huang, Q., & Skantz, T. R. (2016). The informativeness of pro forma and street
earnings: an examination of information asymmetry around earnings
61
announcements. In Review of Accounting Studies (Vol. 21, Issue 1). Springer US.
https://doi.org/10.1007/s11142-015-9345-8
Junaidi, M. J., & Salim, S. (2021). Dampak Pandemi COVID-19 terhadap Kinerja
Perusahaan di Moderasi Pendapatan. Jurnal Ekonomi, 1(November 2021), 208–
226.
Kamin, S., & Clements, B. (2021). The cost of a fair and sustainable COVID-19 grant
for the world’s poorest economies. Sustainability: Science, Practice, and Policy,
17(1), 186–195. https://doi.org/10.1080/15487733.2021.1910406
Kotha, R., Zheng, Y., & George, G. (2011). Entry into New Niches : The effect of firm
age and the expansion of technological capabilities on innovative output and
impact. Strategic Management Journal, 32, 1011–1024.
https://doi.org/10.1002/smj
Lowardi, R., & Abdi, M. (2021). Pengaruh Pandemi Covid 19 terhadap kinerja dan
kondisi keuangan perusahaan publik sektor properti. Jurnal Manajerial Dan
Kewirausahaan, III(2), 463–470.
Nagar, N., & Sen, K. (2016). Earnings management in India: Managers’ fixation on
operating profits. In Journal of International Accounting, Auditing and Taxation
(Vol. 26). Elsevier Inc. https://doi.org/10.1016/j.intaccaudtax.2016.02.003
Pane, Z. I. (2021a). Amortization, Research and Development Expense, Unusual /
Exceptional Item Relevance for Street Earnings. Journal of Social Science, 2(6).
Pane, Z. I. (2021b). Impact street earnings on financial distress. International Journal of
Contemporary Accounting, 3(1), 1–10.
Pane, Z. ikhsan, Sembel, R., & Agustine, Y. (2021). Street earnings as a mediator of
the effect of intellectual capital disclosure, customer value, and research
development activities on firm value. Technium Social Sciences Journal,
25(November 2021), 242–259.
Rachmawati, D., & Susilawati, C. E. (2008). Nilai intellectual capital pada perusahaan-
perusahaan di Indonesia. Media Riset Bisnis & Manajemen, 8(1), 59–92.
Sadique, M. S., & Sheikh, M. A. R. (2013). Investor reaction to strategic emphasis on
earnings numbers: An empirical study. Contemporary Economics, 7(3), 51–64.
Sparta, & Purnama, A. G. (2021). Pengaruh tata kelola perusahaan dan intensitas
modal terhadap penghindaran pajak. Jurnal Ilmiah Akuntansi Dan Ekonomi, 6(2),
38–49.
Suyadnya, I. G., & Supadmi, N. L. (2017). Pengaruh Ukuran KAP, Audit Fee, dan Audit
Tenure pada Agresivitas Pajak. E-Jurnal Akuntansi Universitas Udayana, 21(1),
1131–1159.
Tambunan, M. R. U. D. (2020). Kebijakan perpajakan di Indonesia untuk kemudahan
ekonomi saat masa pandemi Covid-19. Jurnal Administrasi Dan Kebijakan Publik,
V(2), 173–192. https://doi.org/10.25077/jakp
Wardana, M. J., & Wulandari, S. (2021). Analisis Determinan Penghindaran Pajak
Pada Perusahaan Manufaktur yang Terdaftar di Bursa Efek Indonesia.
Kompak :Jurnal Ilmiah Komputerisasi Akuntansi, 14(2), 297–307.
https://doi.org/10.51903/kompak.v14i2.541
Wardani, D. K., & Hermuningsih, S. (2011). Pengaruh Struktur Kepemilikan terhadap
Nilai Perusahaan dengan Kinerja Keuangan dan Kebijakan Hutang sebagai
Variabel Intervening. Jurnal Siasat Bisnis, 15(1), 27–36.
https://doi.org/10.20885/jsb.vol15.iss1.art3
Wijayanti, Y. C., & Merkusiwati, N. K. L. A. (2017). Pengaruh Proporsi Komisaris
Independen, Kepemilikan Institusional, Leverage, Dan Ukuran Perusahaan Pada
Penghindaran Pajak. E-Jurnal Akuntansi Universitas Udayana, 20(1), 699–728.
62
SELF-SERVICE ANALYTICS AND DATA PIPELINE IN
TAX ANALYTICS
Agung Daronoa
aTax Education & Training Center, email: agungdarono@kemenkeu.go.id
ABSTRACT
The shift of paradigm in data analytics (including in taxation) from enterprise analytics tendencies to
self-service analytics requires an adjustment of approach in data governance and the implementation
of tax analytics itself, one of which is the adjustment of the pipeline data structure as a data analytics
component that flows data from the source to the users. This shift is mainly driven by the queue of
user needs on the one side and the availability of various data analytics tools that can be obtained
easily and freely. Using an exploratory case study, this research found that the Directorate General
of Taxes (DGT) should consider implementing self-service analytics and adjusting its data pipeline
as part of the deployed data-technology strategy. This research aims to meet the very high dynamics
of the users of analytical models that align with their duties and functions in the organization. The
consequence of this demand is the refinement of the existing end-user computing approach that
should be equipped with an analytics sandbox approach, allowing data scientists and business
lines/end-users to work hand-in-hand to provide analytical models that support the entire
organization's business processes.
Keywords: compliance, diagnostics, governance, predictive, sandbox, scalability
Adanya pergeseran paradigma dalam analitika data, termasuk di bidang perpajakan, dari
kecenderungan enterprise analytics menuju ke arah self-service analytics memerlukan penyesuaian
pendekatan dalam tata kelola data dan implementasi analitika perpajakan itu sendiri, termasuk dalam
hal ini adalah penyesuaian struktur data pipeline sebagai komponen analitika data yang mengalirkan
data dari sumber hingga dapat dimanfaatkan oleh penggunanya. Pergeseran ini terutama didorong
oleh antrian kebutuhan pengguna di satu sisi dan ketersediaan berbagai perangkat analitika data
yang dapat diperoleh secara mudah dan bebas. Kajian ini dengan menggunakan studi kasus
eksploratif menemukan bahwa Direktorat Jenderal Pajak (DJP) sebaiknya mempertimbangkan
implementasi self-service analytics dan penyesuaian data pipeline-nya sebagai bagian dari
keseluruhan strategi data-teknologi yang dijalankan. Hal ini merupakan upaya untuk memenuhi
dinamika yang sangat tinggi dari para pengguna akan model analitika yang sejalan dengan tugas
dan fungsi mereka. Konsekuensi dari tuntutan ini adalah penyempurnaan pendekatan end-user
computing yang selama ini telah berjalan agar dilengkapi pendekatan analytics sandbox yang
memungkinkan data-scientist maupun lini-bisnis/pengguna-akhir bekerja sama dalam menghasilkan
model analitika yang mendukung keseluruhan proses bisnis organisasi.
Kata Kunci: diagnostik, kepatuhan, prediktif, sandbox, skalabilitas, tata Kelola
1. INTRODUCTION
As the tax authority in Indonesia, the Directorate General of Taxes (DGT) has
explicitly established tax analysis as one of its strategic initiatives to achieve its primary
objective, namely the realization of sustainable tax compliance. In various situations,
data analytics has been used to support various operational activities or formulation and
evaluation of tax policies. The goal is to establish a data-based organization capable of
increasing value from data utilization, drawing conclusions, and making policy decisions
using appropriate and sufficient data analysis (DJP 2015; DJP 2022; DIP 2022). Data
analysis is an essential issue in the field of taxation not only from the side of authorities
and taxpayers but also for various research institutions, academics, consulting firms, and
information and communication technology (ICT) vendors (see, for example, IOTA 2016;
OECD 2017; EY 2017; Microsoft dan PwC 2017; CIAT 2020; ADB 2022; Deloitte 2020;
63
Pijnenburg 2020). From the tax authorities' perspective, the implementation of data
analytics is an effort to minimize the tension (gap) between the amount of tax that has
been paid compared to the amount of tax that should be withheld by the taxpayer (Cleary
2011; Martikainen 2012; Wua et al. 2012; Wessels 2014; Gregg 2015; Stankeviciusa
dan Leonas 2015). Tax data analysis is a technical stage for obtaining information in
making operational decisions or formulating policies in various forms, such as: selecting
taxpayers for mandatory audits, managing tax receivables, improving service quality, or
evaluating the impact of various tax policies that have been stipulated (OECD 2016;
Pijnenburg et al. 2017; ADB 2022).
The dynamics of developing and implementing data analytics lead to situations that
encourage organizations to consider self-service analytics. This term refers to the
preference for independently accomplishing analytical needs by business users in each
line of business rather than relying on data professionals while still considering
adherence to established data governance. The main goal is to reduce data
professionals' service burden to meet the entire organization's analytical needs (HBR
and Alteryx 2016; GoodData 2022; Ali-Eldin et al. 2013; Swanson 2016; Chin 2022;
Alexopoulos et al. 2022; Tableau n.d.).
Another issue that later arose with the growing need for self-service analytics is how
to develop a data pipeline that supports the full implementation of analytics at the
corporate level and self-service analytics itself (Harland et al. 2022; Strod 2020). A data
pipeline is an analytical component that flows data from the source to be further
processed and presented as an analysis result. This component is part of the data
infrastructure, whose function is to stream data from the source so that it can be utilized
by its users (Microsoft 2010; Hevo 2022; Kitching et al. 2021; Raj et al. 2020). The
existence of a data pipeline dramatically influences the entire data analysis process. The
following consequence is that the implementation of self-service analytics will also impact
the pipeline data structure that an organization develops and implements. The main
challenge is obtaining a balance between the flexibility of fulfilling data analysis needs
on the one hand and data governance that wants to protect organizational data from
harmful use.
Self-service analytics and data pipeline as its support has not been discussed
explicitly and in-depth to understand how these two aspects stand in the overall roadmap
for the implementation of tax analytics. The situation mentioned above also applies in
Indonesia. The implementation of tax (data) analytics at DGT will be able to handle
challenges related to the need for self-service analytics and data pipeline development,
as described above. However, as far as the author's observation goes, data pipeline has
become a component of data analytics in the context of its implementation within various
tax authorities (see, for example, IOTA 2016; OECD 2017). Discussion and
understanding of these two matters are becoming increasingly important in connection
with several initiatives that require the support of data management technology, such as
the use of eXtensible Business Report Language (XBRL) or GL Tax Mapping as part of
cooperative compliance (DJP 2016; Djuniardi 2016; Sakti 2021; DJP 2022; DIP 2022).
Based on the description above, it is interesting to discuss the position of self-service
analytics and the existence of data pipelines as part of the overall implementation of data
analytics in tax administration in Indonesia in more specific and in-depth research.
Understanding how these two things work, especially in their interactions with various
other technical-organizational components at DGT, will also facilitate understanding the
implementation of tax data analytics as a whole. From the tax administration's point of
view, it is expected that the research will facilitate the implementation of data analytics
to achieve the projected goal of providing insights that are used to minimize the tax
compliance gap, which so far is considered to be quite broad, which is usually seen from
the low tax ratio in Indonesia compared its peer countries (OECD 2022; DJP 2020:188).
64
Therefore, the focus of this research is to understand how aspects of the data pipeline
and self-service analytics become a part that influences the implementation of tax data
analytics at DGT as the tax authority in Indonesia. In line with the focus of this research,
this study uses an interpretive case study research strategy to reveal various
organizational situations that can be used to understand the existence of data pipelines
and self-service analytics as the phenomena under study.
The paper is presented as follows. The first section will describe the research
background. The second section will exhibit a literature review and previous studies.
Next, the third part will explain the research methodology. Then, the fourth section will
present the research findings and discuss them. Finally, the fifth section presents
conclusions and recommendations.
2. TINJAUAN LITERATUR
Organizations need data as a means of supporting decision-making. This data
comes from a data processing system that is transactional to analytical. The transaction
processing system is the primary data source which will be combined with various other
data sources and then processed to support the data analysis function. Data processing
systems for analytical purposes are known by various terms, such as data warehouse,
data mart, business intelligence, data mining, analytics, or even data science. These
various terms are often used interchangeably even though there are differences in
substance, but in common use, they can be interchanged. In such a context, the various
terms explain that the data that has been collected can be analyzed using various
techniques to gain new insights from the results of the analysis (Cuadrado-Gallego dan
Demchenko 2020; Eger dan Smith 2021; Darono 2016). Davenport dan Harris (2007)
define analytics as the extensive use of data, quantitative analysis, predictive and
explanatory, and fact-based models as tools for decision-making and action in
organizations. White and Imhoff (2010) then briefly define analytics as the science of
analysis. Analog to marketing, which refreshes the notion of sales, Davenport (2013)
showed that analytics is the term used to update everything related to data processing
for organizational decision-making. According to Power et al. (2018), analytics is
different, so with analysis. The existence of the ending -ics in analytics aims to indicate
the presence of a body of knowledge about analysis. Meanwhile, analysis can be
interpreted as an analysis activity itself, namely the breakdown of a subject over its
various parts and the study of each individual part themselves and the relationships
between them to obtain a proper and comprehensive understanding
The results of the author's review of previous studies on analytics bring the author
to an important point that can be used to understand the concept of this analytics (Chen
et al. 2012; Richardson et al. 2019; Duan dan Xiong 2015; Schroeck, et al. 2015;
Tableau, n.d.; SAS 2016). First, they enable the author to understand various related
terms regarding the type of analytics. Second, they also give an understanding of how
data flow in analytics. According to its type, analytics can be described as follows:
(1) based on the type of data source from which the data is processed and analyzed,
analytics is known by several terms such as database analytics; network analytics;
web analytics, big data analytics, and so on ;
(2) based on the processing characteristics and the results of the information it presents,
analytics can be divided into four types, namely: descriptive analytics, which
emphasizes descriptions such as largest, smallest, average, etc.; diagnostic
analytics, which emphasizes data related to compliance with business processes
and specific rules; predictive analytics which emphasizes the development of
models (or machine learning) to make predictions with a specific purpose, and
65
prescriptive analytics which focuses on building models so as to produce various
suggestions that can be considered in decision making;
(3) based on the area or business function it supports, it is known by several terms such
as accounting analytics; analytics audits; marketing analytics, human resource
analytics, procurement analytics, and so on;
(4) based on the differentiation of roles between data-technology functions and end-
users as analytic application providers or line-of-business as end-users, it is known
by several terms: enterprise- analytics (also known as corporate-analytics, or IT-
managed analytics) analytics and on the other hand there is also self-service
analytics as an analytical mechanism developed by the line-of-business/end-users
themselves
Meanwhile, when viewed from the flow of the data analytics process, analytics can
be divided into several stages of work, namely: (1) understanding the context of how
data can assist processes in various organizational functions; (2) data integration and
data pre-processing (there are quite a lot of terms for this stage, such as data ingestion,
data wrangling, data cleansing, data transformation, etc.); (3) model building and data
analysis; (4) presentation/reporting of analysis results (IBM 2021; Chapman et al. 2000;
GoodData 2022; SAS 2016; Tableau n.d.). For all stages of work to run, a data pipeline
is needed. As a component in analytics, the data pipeline flows data from data sources
to be presented to its users. While in the data flow, the data will undergo a transformation
process and be formed with various models according to the objectives of the analysis
process itself (Hevo 2022; Raj et al. 2020; Microsoft 2010). Figure 1 illustrates and makes
connections between constructs about analytics as described above.
Figure 1 Various processes and types of analytics.

Source: author analysis
Referring to the results of several studies (see, for example, Tableau n.d.; Chin 2022;
Swanson 2016; Ali-Eldin et al. 2013), there is a shift from enterprise analytics towards
self-service analytics. Initially, analytics was a top-down product developed by the
organization's ICT function. The line of business as a user performs analytics only on
what is provisioned. Up to a particular level, this will encourage non-ICT users or
business units to take the initiative to develop their own analytical activities. This initiative,
referring to Strod (2020), sometimes creates problems for overall data management and
technology, such as uncontrolled data access or using various non-standard analytical
development tools.
66
In the context of tax administration, tax data analytics is a series of techniques to
obtain information to support decision-making or to review tax policies in various forms,
such as: selecting audited taxpayers, managing tax receivables, improving service
quality, or evaluating the impact of various tax policies that have been set. (OECD 2016).
Deloitte (2020) emphasizes the use of the term tax analytics. They define it as a
combination of technical tax knowledge and advanced information technology to identify
patterns and anomalies to support organizations in setting strategies for the impact of
taxes on overall organizational performance. The OECD (2016:16-20) proposes the term
advanced analytics to represent predictive and prescriptive analytics as one of the
processing and analysis of data for the benefit of tax administration. This study
emphasizes that advanced analytics is not fundamentally new at all. Instead, this term is
more directed at reducing human judgment by relying more on the results of data
analysis to determine various decisions in tax administration. The data analysis
techniques used can be in the form of pattern recognition, outlier detection, cluster
analysis, experimental design, network analysis, and also text mining. ANAO (2008:17)
states that data analytics in tax administration includes all computer-based
methodologies that enable tax authorities to handle large amounts of data from various
data sources, make comparisons, present patterns of interrelationships, or construct
either functional or hypothetical relationships between data. Furthermore, it is explained
that data analytics is divided into two parts (ANAO 2008:137), namely (1) basic analytics,
which includes data exploration and aggregation, as well as statistical profiling and
analysis; (2) advanced analytics which includes the use of data mining technology to find
and develop a model and decision making.
The shift in the pattern of implementation of data analytics from enterprise analytics
to self-service analytics has also been experienced in tax administration (Microsoft and
PwC 2017; EY 2017). Therefore, this shift should also be a concern for the tax authorities
to consider the implementation of self-service analytics as a complementary part of the
existing enterprise analytics.
3. METODOLOGI PENELITIAN
Research methodology is the overall framework of understanding related to
including philosophical assumptions, social context, ethical principles, strategies for
answering questions that will be used, including the impact that there may be new
knowledge obtained from research. Meanwhile, the research method is a procedure for
collecting and analyzing data to answer research problems (Neuman 2014:2; Creswell
2014). Furthermore, Creswell (2014) proposed the term research design as the
application of methodology as a series of research strategies and methods specifically
used in one study. The design set for this study is an exploratory-interpretive case study.
According to Yin (2018), case study research is contemporary empirical research in real-
life contexts, not in the form of surveys or experiments, which are in situations where the
boundaries between phenomena and contexts are not very clear. Case study research
terms can be seen as a trilogy, namely as (1) a research strategy (or mode of inquiry);
(2) a method of analysis; and (3) phenomena with certain limits that are being studied.
The research objectives determined the choice of this strategy to explore and understand
the existence of self-service analytics and data pipelines as part of the implementation
of tax data analytics at DGT.
Exploratory (or explorative) case studies were chosen because the characteristics
of the research problem did not yet show a clear and explicit relationship between
variables (Klein dan Myers 1999; Berntsen et al. 2004; Ponelis 2015). Interpretive in this
context means that understanding is obtained and presented by researchers by
disclosing the meaning of the data that has been collected for further interpretation using
67
theory/literature, previous research, and the experience of researchers using an emic
perspective (research-subject point of view) to gain a deep understanding (Bhattacherjee
2012:35-40; Gerring 2007:69, 214; Yin 2018:311). The data collection method was
carried out through documentation studies (Bohnsack 2014; Coffey 2014; Bowen 2009)
and observation (Erich et al. 2017) of the implementation of data analytics by using a
proxy in the form of involvement in the preparation of training materials involving
researchers as managers of training programs (Erickson et al. 2016; Darono 2016).
Documentation includes regulations, standard operating procedures, application
manuals, and various internal and external articles. Observations were made using
training program preparation activities as a proxy for the tax data analysis activity itself.
Data analysis was carried out using interactive data analysis techniques (Miles et al.
2020) and gap analysis (Zheng et al. 2010).
Gap analysis combined with interactive data analysis of qualitative data is carried
out by collection, condensation, display, and verification processes until conclusions are
carried out continuously and iteratively so that valid final results are obtained. Gap
analysis in this context refers to Zheng et al. (2010) as an analysis technique to identify
a discrepancy between user requirements and available application systems. Figure 2
describes interactive data analysis techniques by adding a gap analysis component and
a conceptual framework for tax data analysis as a tool for collecting, summarizing,
presenting data, and drawing conclusions.
Figure 2 data analysis method with interactive data analysis and gap analysis using a
conceptual framework of data analysis
source: modified from Miles et al. (2020)
4. FINDING AND DISCUSSION

The results and discussion section will be divided into three sub-sections. First, a
description of the latest situation of implementing analytics at DGT as a context for
understanding cases. Second, a discussion on the interplay between the fulfillment of
data analytics solutions for users and the provisions regarding data governance. Third,
a discussion on how these analytical needs can be met with an analytics sandbox
mechanism with a containerization approach.
4.1. Implementasi analitika data perpajakan: konteks kasus

DGT started to develop formal data analytics in 2014 with the development of
Compliance Risk Management (CRM). This initiative aims to utilize a risk-based
compliance approach by treating taxpayers according to their compliance risk. CRM
development is one of the ten transformational themes aimed at increasing the tax ratio
listed in the Decree of the Minister of Finance Number 36/KMK.01/2014 (DJP 2022:8).
68
Referring to Djuniardi (2016; 2018), data analytics was developed from the construction
of a data warehouse so that data is available that supports a single source of truth for all
its users and also develops adequate access capacity. The CRM was initially started
from several use cases, such as network analysis for fraud detection in Value Added Tax
(VAT) transactions or diagnostic analysis in data matching between VAT and income tax.
In the author's view, the structure of the implementation of tax analysis at DGT can
be seen from the technology stack perspective. From this point of view, ongoing tax data
analytics tends to use a best-of-breed approach (see, for example, Light et al. 2001;
Munchbach 2020), which emphasizes a combination of various analytical software
available on the market. Then, it determines the level of suitability for the needs
development of the data analytics framework. Figure 3 shows the technology stack data,
the workflow, and the data pipeline formed from a predefined software selection structure.
Figure 3 DGT tax analytics data flow

source: author analysis, modified from DIP(2022); DJP (2022); Sakti (2021)
This implementation situation shows a tendency to emphasize enterprise analytics
by providing facilities to end-users to use data analysis results through standard
applications (Aproweb, Aportal, etc.). This situation shows that self-service analytics is
not yet a priority for using already available data. Even if referring to DGT (2016), it has
been stated that
"… developing CRM based on sectors supported by the use of predictive data
analysis, trends, and taxpayer behavior, developing self-service analytics, and
integrating analytics into business intelligence" (translation by author)
The problem that is also most likely to arise related to the deployment of self-service
analytics is that there are restrictions on access rights per data governance stipulated in
SE-30/PJ/2019 concerning the Governance of DGT Tax Data Access Authorities. In
addition, technical considerations are primarily related to service capacity (e.g.,
managing bandwidth, CPU, or RAM capacity) required to execute an analysis model so
that users can flexibly use the model.
4.2. User needs and data governance

As described in subsection 4.1, the conditions are because SE-30/PJ/2019 is very
conservative in regulating data access. This circular emphasizes the
minimum/necessary principle in granting data access to meet user needs in carrying out
tasks based on (1) duties and functions; (2) position authority; (3) work area. At this point,
69
a gap appears between user needs for flexibility in developing data pipelines that can be
used to perform self-service analytics. On the other hand, there has been a Director
General of Taxes Circular Number SE-08/PJ/2016 concerning Guidelines for
Management of End User Computing (EUC) which allows work units other than the
Directorate of Communication and Information Technology (ICT) to develop data
analysis applications.
The issue of meeting user needs on the one hand and data management on the
other, of course, does not have to be seen as a diametrical conflict but rather as a
balance point. Between these two interests, of course, there is a gap. The challenge is
to make the gap as minimal as possible. In the author's view, this gap can be identified
through training needs, which can be considered individual user needs, and also
business lines of analytical activity that need to be fulfilled to achieve the objectives of
each user himself. Based on this condition, there are several data analysis training
programs that involve the author as a program manager. By conducting a training needs
analysis, the author (see Table 1) can identify a gap indicating a need for this activity
that has not been fulfilled by the analytical application that has been running.
Table 1 Observation objects by proxy in the form of training program preparation

activities
Training ID The observation results are related to the need
for competence in the field of data analytics
Tax data analytics How do tax officers in the Data Quality Assurance
Section establish a data pipeline to provide the
data needed by the sections in their respective
unit (tax office)
Tax supervision of the How can Account Representatives carry out
digital economy further analysis using comparative data that has
been distributed under their supervisory tasks
Computer-assisted audit Every individual tax auditor needs self-service
tools and techniques for tax analytics by using a data pipeline to compare the
auditors. data they get from the taxpayer with the data
provided by the DGT (SIDJP, Aproweb, Apportal,
Derik, etc.) to produce an audit report. Some tax
examiners have developed applications such as
e-Audit Utilities or Apiseta, which already deploy
some diagnostic analytics principles to obtain
audit findings.
source: author anlalysis
In the day-to-day organization's practice, the need for analytical applications is

provided through applications developed by end users such as MPN Info (Ministry of
Finance 2015), e-Audit Utilities (Dit P2 2015), or Apiseta (DDTC News 2020). Some of
these applications were developed and equipped with descriptive and diagnostic
analytical features. Therefore, application development is one approach to optimizing
various analytical techniques and tools to solve the tax administration task.
4.3. Analytics sandbox using containerization approach

Boswell and Courtright (2022) define the analytics sandbox as an area that becomes
a workspace for users who collaborate to develop specific analytical solutions with some
limitations, such as storage capacity or users. An analytics sandbox is an effort to enable
data scientists or certain users who can create an analytical model in a limited
70
environment to gain confidence that the model will optimally run y if executed as a
standard analytical model that applies to all users (Dietrich et al. 2015). Referring to
Djuniardi (2016; 2018), DGT has also introduced the concept of an analytics sandbox.
However, it is still limited to the ICT unit environment and has not involved business lines
or end users. One of the approaches/mechanisms that can be used in developing
sandbox analytics is containerization. The sandboxed containers approach (see, for
example, Redhat 2022) allows testing of data analytics models in a limited area in terms
of hardware, storage, datasets, and users according to the provisions in SE-08/PJ/2016
and SE-30/PJ/2019. It is in this context that by continuing to use the provisions in SE-
08/PJ/2016, sandbox analytics can be developed involving lines of business or end users
by considering the following points:
(1) the use of open-source or freeware available on the market so that it does not require
an additional budget;
(2) a containerization approach so the developed data analytics model can be validated
with more fit in business cases and datasets.
As a general characteristic, an analytical model that has gone through a sandbox
will decide whether the model will continue to the mass deployment stage so that it can
be used by a broader range of users or even deleted because it is no longer suitable for
use. Therefore, if possible, it is advisable that some of the provisions in SE-08/PJ/2016
relating to licensing procedures in EUC development be changed to a validation test
(curation) mechanism through an analytics sandbox mechanism. In this mechanism, it is
possible to form a panel of experts/practitioners from either the ICT unit or user side to
assess the feasibility of the proposed analytical model.
5. CONCLUSIONS AND RECOMMENDATIONS

Tax analytics has become one of DGT's strategic initiatives to achieve sustainable
tax compliance. The change in the implementation of data analysis, from enterprise
analytics to self-service analytics, needs to be anticipated by taxation. Therefore, the
DGT should be more concerned about implementing self-service analytics as a
complementary part of the existing enterprise analytics. This study concludes that the
development of self-service analytics with its data pipeline will be a crucial issue for the
full implementation of tax analytics to support some DGT initiatives, such as cooperative
tax compliance or the use of the eXtensible Business Report Language (XBRL).
This paper proposes recommendations for improving SE-08/PJ/2016 by changing
the licensing approach to a curation approach based on an analytics sandbox. This
approach is considered better since several users (line of business) tried to develop their
analytical solutions, such as the e-Audit Utilities application, Apiseta, MPN Info, or other
applications deployed limited to units where the developer works. It is hoped that the
provisions regarding EUC that include a curation strategy based on an analytics sandbox
will create a situation that, referring to Strod (2020), "governs self-service analytics
without stifling innovation."
The results of this study on a broader scale are expected to provide a conceptual
framework that can be used to understand how self-service analytics and the
accompanying data pipeline play a role in the implementation of data analytics in the
field of tax administration (including local taxes). In fact, on a broader scale, with the
principle of generalization-analysis in the case study research strategy, it is hoped that
the research findings can be utilized to understand the implementation of data analytics
in the environment of public administration/government organizations.
71
DAFTAR PUSTAKA
ADB. 2022. Launching a Digital Tax Administration Transformation - What You Need To
Know. Manila: Asian Development Bank (ADB).
https://www.adb.org/sites/default/files/publication/792586/digital-tax-administration-
transformation.pdf.
Alexopoulos, Andreas, Yolanda Becerra, Omer Boehm, George Bravos, Vasilis
Chatzigiannakis, Cesare Cugnasco, Giorgos Demetriou, et al. 2022. Big Data
Analytics in the Banking Sector: Guidelines and Lessons Learned from the CaixaBank
Case. In , 273–97. https://doi.org/10.1007/978-3-030-78307-5_13.
Ali-Eldin, Amr, Mohamed Zaghloul, and Mofreh Salem. 2013. Towards a Self-Service
Data Analytics Framework. International Journal of Computer Applications 80
(October): 41–48. https://doi.org/10.5120/13893-1840.
Berntsen, Kirsti E., Jennifer Sampson, and Thomas Østerlie. 2004. Interpretive Research
Methods in Computer Science.
Bhattacherjee, Anol. 2012. Social Science Research: Principles, Methods, and
Practices. Tampa: USF Tampa. http://scholarcommons.usf.edu/oa_textbooks/3.
Bohnsack, Ralf. 2014. Documentary Method. In The SAGE Handbook of Qualitative Data
Analysis, edited by Uwe Flicks, 217–32. London: SAGE Publications Ltd.
Booch, Grady, James Rumbaugh, and Ivar Jacobson. 1998. Unified Modeling Language
User Guide. Reading, Massachusetts: Addison Wesley.
Boswell, Mike, and Andrea Courtright. 2022. Data Lake Zones and Containers. Microsoft
Corp. https://docs.microsoft.com/en-us/azure/cloud-adoption-
framework/scenarios/cloud-scale-analytics/best-practices/data-lake-zones.
Bowen, Glenn A. 2009. Document Analysis as a Qualitative Research Method.
Qualitative Research Journal 9: 27–40.
Chapman, Peter, Janet Clinton, Randy Kerber, Tom Khabaza, Thomas P. Reinartz, Colin
Shearer, and Richard Wirth. 2000. CRISP-DM 1.0: Step-by-Step Data Mining Guide.
In .
Chen, Hsinchun, Roger H. L. Chiang, and Veda C. Storey. 2012. Business Intelligence
and Analytics: From Big Data to Big Impact. MIS Quarterly 36 (4): 1165–88.
Chin, Cedric. 2022. Self Service Analytics Is a Business State. 2022.
https://www.holistics.io/blog/self-service-analytics/.
CIAT. 2020. ICT as a Strategic Tool to Leapfrog the Efficiency of Tax Administrations.
Panama City, Panama: Inter-American Center of Tax Administrations – CIAT.
Cleary, Duncan. 2011. Predictive Analytics in the Public Sector: Using Data Mining to
Assist Better Target Selection for Audit. Electronic Journal of E-Government 9 (2):
132–40.
Coffey, Amanda. 2014. Analysing Documents. In The SAGE Handbook of Qualitative
Data Analysis, edited by Uwe Flick, 367–76.
Creswell, John W. 2014. Research Design: Qualitative, Quantitative, and Mixed Methods
Approaches. Thousand Oaks, California: SAGE Publications.
Cuadrado-Gallego, Juan J., and Yuri Demchenko. 2020. Introduction to the Data Science
Framework. In The Data Science Framework - A View from the EDISON Project, 1–
7. Switzerland: Springer Nature.
Darono, Agung. 2016. Pembelajaran Business Analytics Dan Big Data Dalam
Pendidikan Ekonomi Dan Bisnis. In National Seminar on Accounting and Finance
2016, 8–16. Malang: Universitas Negeri Malang.
Davenport, Thomas H. 2013. Introduction: The New World of Enterprise Analytics. In
Enterprise Analytics Optimize Performance, Process, and Decisions Through Big
Data, edited by Thomas H. Davenport. New Jersey: Pearson Education, Inc.
Davenport, Thomas H., and Jeanne G. Harris. 2007. Competing on Analytics: The New
Science of Winning. Harvard Business Review Press.
72
DDTC News. 2020. DERIK, Digitalisasi Proses Pemeriksaan Pajak, 2020.
https://news.ddtc.co.id/derik-digitalisasi-proses-pemeriksaan-pajak-20282.
Deloitte. 2020. My Data Is a Mess. Is Analytics out of Reach? Deloitte Tax LLP.
Dietrich, David, Barry Heller, and Beibei Yang. 2015. Data Science & Big Data Analytics
Discovering, Analyzing, Visualizing and Presenting Data. John Wiley & Sons, Inc.
DIP. 2022. Implementasi Data Analytics Dalam Administrasi Perpajakan. Presented at
the Kuliah Umum Analitika Data Perpajakan - Direktorat Data dan Informasi
Perpajakan (DIP).
Dit P2. 2015. E-Audit Utilities Manual Guide. Direktuorat Pemeriksaan dan Penagihan.
DJP. 2015. Cetak Biru Teknologi Informasi Dan Komunikasi Direktorat Jenderal Pajak
Kementerian Keuangan Republik Indonesia Tahun 2015-2019: Direktorat Jenderal
Pajak (DJP).
———. 2016. Era Baru! DJP Bangun Kepatuhan Melalui CRM. Direktorat Jenderal Pajak
(DJP). http://www.pajak.go.id/news/era-baru-djp-bangun-kepatuhan-melalui-crm.
———. 2020. Laporan Tahunan DJP 2020. Direktorat Jenderal Pajak (DJP).
https://pajak.go.id/sites/default/files/2021-
10/Laporan%20Tahunan%20DJP%202020%20-%20Bahasa.pdf.
———. 2022. CRMBI: Langkah Awal Menuju Data Driven Organization. Jakarta:
Direktorat Jenderal Pajak (DJP).
Djuniardi, Iwan. 2016. Next Generation Data Analysis The Implementationof Big Data in
Directorate General of Taxes Republic of Indonesia. Presented at the 13th ATAIC
Annual Technical Conference, Melaka, Malaysia.
http://ataic13.hasil.gov.my/SLOT3/IndonesiaDGTBigData.pdf.
———. 2018. Journey to Big Data. Indonesian Ministry of Finance.
Eger, III, Robert J., and Christy Smith. 2021. Integrated Business Intelligence and
Analytics: The Case of the Department of the Navy. Journal of Governmental &
Nonprofit Accounting 10 (1): 26–49. https://doi.org/10.2308/JOGNA-17-001.
Erich, F. M. A., C. Amrit, and M. Daneva. 2017. A Qualitative Study of DevOps Usage in
Practice. J. Softw. Evol. Process 29 (6): n/a. https://doi.org/10.1002/smr.1885.
Erickson, Amy, Patricia Noonan, Jennifer Brussow, and Kayla Carter. 2016. Measuring
the Quality of Professional Development Training. Professional Development in
Education 43 (May): 1–4. https://doi.org/10.1080/19415257.2016.1179665.
EY. 2017. The Tax Authority of the Future: How Tax Authorities Are Using Analytics to
Deliver New Levels of Value. UK: Ernst & Young Global Limited.
GoodData. 2022. BI and Analytics — From Terms to Options. GoodData Inc.
https://www.gooddata.com/resources/starter-guide-bi-analytics-terms-to-options/.
Gregg, Richard. 2015. Passion for Change and Good Data: The Right Combination for
the DATA Act. Journal of Government Financial Management 64 (2): 48–52.
Harland, Tobias, Christian Hocken, Tobias Schröer, and Volker Stich. 2022. Towards a
Democratization of Data in the Context of Industry 4.0. Sci 4 (3).
https://doi.org/10.3390/sci4030029.
HBR, and Alteryx. 2016. The Untapped Power of Self-Service Data Analytics,. Harvard
Business Review Analytic Services & Alteryx.
Hevo. 2022. Pipelines. Hevo Data Inc. https://docs.hevodata.com/pipelines/.
IBM. 2021. CRISP-DM Help Overview. IBM Corporation.
https://www.ibm.com/docs/en/spss-modeler/18.2.0?topic=dm-crisp-help-overview.
IOTA. 2016. Applying Data and Analytics in Tax Administration. IOTA Good Practice
Guide. Intra-European Organisation of Tax Administrations (IOTA). https://www.iota-
tax.org/good-practice-guide-applying-data-and-analytics-tax-administrations.
Kemenkeu. 2015. Catatan Inspirasi Transformasi Inovasi Dan Terobosan Untuk Inspirasi
Perubahan. Balai Pustaka.
73
Kitching, Karen A., Eileen Coble, and Alex Phillips. 2021. DATA Act Dashboard: An
Instructional Case Using Data Visualization. Journal of Governmental & Nonprofit
Accounting 10 (1): 157–74. https://doi.org/10.2308/JOGNA-2021-004.
Klein, Heinz K., and Michael D. Myers. 1999. A Set of Principles for Conducting and
Evaluating Interpretive Field Studies in Information. MIS Quarterly 23: 67–93.
Light, Ben, Christopher P. Holland, and Karl Wills. 2001. ERP and Best of Breed: A
Comparative Analysis. Business Process Management Journal 7 (3): 216–24.
https://doi.org/10.1108/14637150110392683.
Martikainen, Jani. 2012. Data Mining in Tax Administration - Using Analytics to Enhance
Tax Compliance. Master Thesis, Helsinki, Finland: Aalto University School of
Business.
Microsoft. 2010. Microsoft EDW Architecture, Guidance and Deployment Best Practices.
Microsoft Corp. https://docs.microsoft.com/en-us/previous-versions/sql/sql-server-
2008-r2/hh146875(v=msdn.10).
Microsoft, and PwC. 2017. Digital Transformation of Tax Administration. Netherland:
Microsoft and PricewaterhouseCoopers Belastingadviseurs N.V.
Miles, Matthew B., A. Michael Huberman, and Johnny Saldaña. 2019. Qualitative Data
Analysis - A Methods Sourcebook 4th Edition. SAGE Publications.
Munchbach, Cory. 2020. All-In-One Vs. Best-Of-Breed: Putting An End To An Endless
Debate. Forbes.Com, 2020.
https://www.forbes.com/sites/forbesbusinesscouncil/2020/08/05/all-in-one-vs-best-
of-breed-putting-an-end-to-an-endless-debate/?sh=2ad972126024.
Neuman, W. Lawrence. 2014. Social Research Methods: Qualitative and Quantitative
Approaches. 7th ed. Essex, England: Pearson Education Limited.
OECD. 2016. Advanced Analytics for Better Tax Administration Putting Data to Work.
Paris: OECD. https://read.oecd-ilibrary.org/taxation/advanced-analytics-for-better-
tax-administration_9789264256453-en#page1.
———. 2017. 'Advanced Analytics' in Tax Administration 2017: Comparative Information
on OECD and Other Advanced and Emerging Economies. Organisation for Economic
Cooperation and Development (OECD).
OECD, Muhamad. 2022. Revenue Statistics in Asia and the Pacific 2022. Organisation
for Economic Co-Operation and Development (OECD).
https://www.oecd.org/dev/revenue-statistics-in-asia-and-the-pacific-5902c320-
en.htm.
Pijnenburg, Mark Gerardus Franciscus. 2020. Data Science for Tax Administration.
Ph.D. thesis, Netherland: Leiden University.
Pijnenburg, Mark, Wojtek Kowalczyk, and van der Hel-van Dijk Lissette. 2017. A
Roadmap for Analytics in Taxpayer Supervision. The Electronic Journal of E-
Government 15 (1): 19–32.
Ponelis, Shana. 2015. Using Interpretive Qualitative Case Studies for Exploratory
Research in Doctoral Studies: A Case of Information Systems Research in Small and
Medium Enterprises. International Journal of Doctoral Studies 10 (December): 535–
50. https://doi.org/10.28945/2339.
Power, D.J., C. Heavin, J. McDermott, and M. Daly. 2018. Defining Business Analytics:
An Empirical Approach. Journal of Business Analytics 1 (1): 40–53.
https://doi.org/10.1080/2573234X.2018.1507605.
Raj, Aiswarya, Jan Bosch, Helena Holmström Olsson, and Tian J. Wang. 2020.
Modelling Data Pipelines. In 2020 46th Euromicro Conference on Software
Engineering and Advanced Applications (SEAA), 13–20.
https://doi.org/10.1109/SEAA51224.2020.00014.
Redhat. 2022. What Are Sandboxed Containers? Redhat.
https://www.redhat.com/en/topics/containers/sandboxed-containers.
74
Richardson, Vernon J., Ryan Teeter, and Katie L. Terrell. 2019. Data Analytics for
Accounting. New York: McGraw-Hill Education.
Sakti, Nufransa Wira. 2021. Implementation of Data Analytics In Directorate General of
Taxes. Presented at the The 2nd Conference of Belt and Road Initiative Tax
Administration Cooperation Forum.
https://www.britacom.org/news/2ndOnlineMeeting/files/22Slides%20of%20Indonesia
.pdf.
SAS. 2016. What Is Analytics? SAS. http://www.sas.com/en_us/insights/analytics/what-
is-analytics.html.
Schroeck, Michael, Rebecca Shockley, Janet Smart, Dolores Romero-Morales, and
Peter Tufano. 2015. Analytics: The Real-World Use of Big Data How Innovative
Enterprises Extract Value from Uncertain Data. IBM Global Business Services
Business Analytics and Optimization Executive Report. IBM Institute for Business
Value and Saïd Business School at the University of Oxford.
Stankeviciusa, Evaldas, and Linas Leonas. 2015. Hybrid Approach Model for Prevention
of Tax Evasion and Fraud. In 20th International Scientific Conference Economics and
Management - 2015 (ICEM-2015). Kaunas, Lithuania.
Strod, Eran. 2020. Govern Self-Service Analytics Without Stifling Innovation. 2020.
https://datakitchen.io/govern-self-service-analytics-without-stifling-innovation/.
Swanson, Sandra. 2016. Self-Service Analytics. Sebastopol, CA: O'Reilly Media, Inc.
Tableau. n.d. Business Intelligence: A Complete Overview. Tableau Software LLC.
https://www.tableau.com/learn/articles/business-intelligence.
———. n.d. Self-Service Business Intelligence (SSBI): Definition and Examples.
Tableau Software LLC. https://www.tableau.com/learn/articles/business-
intelligence/self-service-
bi#:~:text=BI%20is%20the%20strategic%20process,around%20using%20data%20e
very%20day.
Wessels, Eugene. 2014. Towards a Conceptual Decision Support Systems Framework
Aimed at Narrowing the Tax Gap in South Africa: A Narrative Case Study. Ph.D.
thesis, University of South Africa.
White, Colin, and Claudia Imhoff. 2010. Advanced Analytics and Business Intelligence:
Term Abuse? 2010. http://www.b-eye-network.com/view/13797.
Wua, Roung-Shiunn, C.S. Ou, Hui-ying Lin, She-I Chang, and David C. Yen. 2012. Using
Data Mining Technique to Enhance Tax Evasion Detection Performance. Expert
Systems with Applications 2012 (39): 8769–77.
Zheng, Zhixing, Benjamin Yen, and Minyi Huang. 2010. An Exploratory Study of the Gap
between Client Expectations and Client Perceived Performance of the Delivered
Information System. In PACIS 2010 - 14th Pacific Asia Conference on Information
Systems, 16.
75
USABILITY MEASUREMENT OF E-BUPOT INSTANSI
PEMERINTAH APPLICATIONS
(CASE STUDIES IN THE GENERAL ELECTION SUPERVISORY AGENCY)
Dwi Langgeng Santosoa, Mohammed Lintang Theodiktab

a Directorate General of Taxes, Jakarta, Indonesia Email: dwi.langgeng@gmail.com
b Directorate General of Taxes, Jakarta, Indonesia Email: mohammed.lintang@gmail.com
ABSTRACT
Government Institution Taxpayers are obligated to report periodic tax returns. Such reporting must
be done through the e-Bupot Instansi Pemerintah since the September 2021 tax period. To make it
easier for Government Institution Taxpayers to fulfil these obligations, the e-Bupot Instansi
Pemerintah must have good usability. This research measures usability using the USE questionnaire
method with four parameters (Usefulness, Ease of Use, Ease of Learning, and Satisfaction). The
respondents were 35 treasurers of The General Election Supervisory Agency (Bawaslu) in Indonesia.
It was found that the e-Bupot Instansi Pemerintah got a good average value of usability parameters.
However, there is still a chance for improvement in the Ease of Learning parameter. The Directorate
General of Taxes should enhance the usability value by increasing the Ease of Learning parameter
value by providing adequate instructions, guidebooks, or video tutorials so that Government
Institution Taxpayers can learn how to use the application properly.
Keywords: periodic tax return, application, e-Bupot Instansi Pemerintah, usability, USE questionnaire
Wajib Pajak Instansi Pemerintah diwajibkan melakukan pelaporan SPT masa. Pelaporan tersebut
wajib dilakukan melalui aplikasi e-Bupot Instansi Pemerintah sejak masa pajak September 2021.
Untuk dapat memudahkan bendahara Instansi Pemerintah dalam memenuhi kewajiban tersebut,
aplikasi e-Bupot Instansi Pemerintah harus memiliki usability yang baik. Penelitian ini mengukur
usability menggunakan metode USE questionnaire dengan parameter Usefulness, Ease of Use,
Ease of Learning, dan Satisfaction. Responden penelitian ini adalah 35 bendahara Instansi
Pemerintah Badan Pengawas Pemilihan Umum yang berada di seluruh Indonesia. Ditemukan bahwa
aplikasi e-Bupot Instansi Pemerintah mendapatkan nilai rata-rata parameter usability yang baik.
Walaupun demikian masih terbuka ruang perbaikan pada parameter Ease of Learning. Untuk
memperbaiki hal tersebut Direktorat Jenderal Pajak disarankan untuk meningkatkan nilai usability
melalui peningkatan nilai parameter Ease of Learning dengan menyediakan petunjuk, buku panduan,
atau video tutorial yang memadai agar bendahara Instansi Pemerintah dapat mempelajari
penggunaan aplikasi dengan baik.
Kata kunci: pelaporan SPT masa, aplikasi, e-Bupot Instansi Pemerintah, usability, USE questionnaire
1. INTRODUCTION
1.1. Background
At the end of 2019, the Government issued the Minister of Finance Regulation
Number 231/PMK.03/2019 concerning Procedures for Registration and Deletion of
Taxpayer Identification Numbers, Confirmation and Revocation of Taxable
Entrepreneurs, as well as Withholding and/or Collection, Deposit, and Reporting of
Taxes for Government Institution Taxpayers (PMK-231/PMK.03/2019). This regulation is
the basis for the Directorate General of Taxes (DGT) to administer Government
Institution Taxpayers' tax administration. The administration is carried out, among other
things, by regulating procedures for tax deduction/collection and reporting of tax returns
for Government Institution Taxpayers.
76
It is crucial to control tax administration, among other things, to encourage taxpayer
compliance, bearing in mind that Government Institution Taxpayers have a strategic role
as a tax withholder/collector for every government spending activity. The value of
government spending is enormous and always increases from year to year, as explained
in the following table:
Table 1 Realization of Government Spending in 2018 to 2021

Source:www.bps.go.id
Realization of Government Spending

Government Spending (Billion Rupiah)
2018 2019 2020 2021
GOVERNMENT SPENDING 2.213.118 2.309.287 2.739.166 2.750.028
I. Central Government Spending 1.455.325 1.496.314 1.975.240 1.954.549
1. Spending of Ministries and 846.562 873.426 836.380 1.031.961
Agencies
2. Spending of Non-Ministries 608.763 622.888 1.138.860 922.588
and Agencies
II. Regions Government Spending 757.793 812.973 763.926 795.480
1. Transfer to Regions 697.934 743.159 692.736 723.480
a. Balancing Fund 668.643 711.285 653.359 688.677
b. Regional Incentive Fund 8.231 9.694 18.500 13.500
c. Special Autonomy Fund and 21.060 22.180 20.877 21.303
Privileges Fund DIY
2. Village Fund 59.859 69.814 71.190 72.000
III. Suspend - - - -
To support the implementation of PMK-231/PMK.03/2019, the Director General of

Taxes issues several implementing regulations, including the Director General of Taxes
Regulation Number PER-17/PJ/2021 concerning Forms and Procedures for Making
Withholding Tax Slip, as well as Form, Content, Procedure for Filling in, and Submission
of Periodic Tax Returns for Government Institution Taxpayers. Based on PER-
17/PJ/2021, starting from the September 2021 tax period, the Director General of Taxes
requires all Government Institution Taxpayers to carry out their tax reporting obligations
using a new application, namely the e-Bupot Instansi Pemerintah Application.
This effort is in line with Harjowiryono's research (2019) on attribution theory. This
theory explains that several variables influence government treasury tax compliance,
namely the variable level of tax knowledge, attitude towards tax obligations, quality of
tax services, ease of tax application, and the role of the Directorate General of Treasury.
Attribution theory, proposed by Fritz Heider in 1958 and developed by Kelley in 1973,
tries to find things that cause or motivate someone to do something. This theory is
relevant to explain the factors that influence taxpayer compliance in making decisions
(Susherdianto, 2014).
Furthermore, the purpose of making an application is to make it easier for users
(Siregar & Sari, 2028) and to assist users in completing certain activities (Haerulah &
Ismiyatih, 2017). Susanti, Astari, and Thamrin (2016) explain that an application is a
device created to complete daily tasks, such as advertising, trade, public relations,
games, and other human activities.
Arifin (2018) explains that a proper application must consider at least two things: the
user interface and the user experience. According to Guo (2012), at least the user
experience has four components: value, usability, desirability, and adoptability. Among
77
these four components, the usability component is one of the determining factors for the
success or failure of an application. The usability component explains whether the
application can facilitate the user in completing specific tasks. If an application has high
usability, the user will continue to use the application. Meanwhile, if the usability is low,
the user will leave the application (Rahman & Vitalocca, 2018).
1.2. Problem Formulation

This research aims to measure the usability of the e-Bupot application for
Government Agencies as an application used to support Government Institution
Taxpayers' tax administration control policies based on PMK-231/PMK.032019. The
sample used in this research is the General Election Supervisory Agency (Bawaslu)
working units throughout Indonesia, including the central working unit of Bawaslu and
the provincial working units of Bawaslu.
1.3. Research Purpose

The expected objective of this research is to provide input to the Directorate General
of Taxes regarding implementing the e-Bupot Instansi Pemerintah application for
Government Institution Taxpayers in the future.
2. CONCEPTUAL AND THEORETICAL FRAMEWORK

2.1. Tax Obligations of Government Institution Taxpayers
The taxpayer is an individual or entity, including tax payer and tax withholder, who
has tax rights and obligations under the provisions of the tax laws and regulations (UU
KUP Article 1 paragraph (2)). Government Institution Taxpayers are included in the
category of corporate taxpayers. Government Institution Taxpayers are central
government institutions, regional government institutions, and village government
institutions, which carry out government activities and have the authority and
responsibility for using the budget.
In general, there is no difference in tax obligations for individuals, entities or
government institution taxpayers (www.pajak.go.id, 2021). Specifically for Government
Institution Taxpayers, the government issued Minister of Finance Regulation Number
PMK-231/PMK.03/2019. The PMK-231/PMK.03/2019 describes the tax obligations of
Government Institution Taxpayers, including:
a. registration: every Government Institution Taxpayer must register with the Tax Office
or Tax Service, Dissemination, and Consultation Office whose working area covers
the domicile of the Government Institutions Taxpayer according to the actual
situation. (Article 2 paragraph (1);
b. calculation: Government Institution Taxpayers are income tax collectors of
government spending. The income tax consists of Income Tax Article 4 paragraph
(2), Article 15, Article 21, Article 22, Article 23 and Article 26. Government Institution
Taxpayers are also collectors of Value Added Tax (VAT) (Article 16 paragraph (1));
c. payment: Government Institution Taxpayers withheld taxes are required to pay their
taxes into the state treasury;
d. reporting: Government Institution Taxpayers must report withholding and paying
taxes made in one tax period to the Tax Office.
Reporting is an activity to report or submit certain activities to certain parties or
authorities in which there is a record of information related to the specific activities
(Ruung, Ilat, and Vokas, 2017). In PMK-231/PMK.03/2019 Article 25 paragraph (1), it is
stipulated that Government Institution Taxpayers are required to report withholding and
paying of taxes carried out in one tax period to the Tax Office where the Government
Institution Taxpayers registered.
78
2.2. e-Bupot Instansi Pemerintah Application
To complete the Government Institution Taxpayer's tax reporting provisions in PMK-
231/PMK.03/2029, the Directorate General of Taxes issues Director General of Taxes
Regulation Number PER-17/PJ/2021. PER-17/PJ/2021 Article 1 paragraph (26) explains
that the e-Bupot Instansi Pemerintah application is software provided on a website
owned by the Directorate General of Taxes or specific channels determined by the
Directorate General of Taxes which can be used to create income tax Article 21 /26
withholding tax slip and unification withholding tax slip of Government Institution
Taxpayers, as well as filling out and submitting income tax returns Article 21/26 and
unification tax returns of Government Institution Taxpayers.
According to www.pajak.go.id (2021), some of the backgrounds for using e-Bupot
Instansi Pemerintah application are:
a. convenience: provide convenience and services for Government Institution
Taxpayers to report periodical tax returns;
b. legal certainty: provide legal certainty regarding the status and reliability of
withholding tax slips;
c. compliance: improve compliance by making withholding tax slips and submitting tax
returns;
d. accuracy and validation: improve the accuracy and validation of withholding tax slips
issued by Government Institution Taxpayers;
e. one-stop application: with one application, Government Institution Taxpayers can
perform tax calculations, create withholding tax slips, create billings, create and
submit income tax returns.
The Government Institution Taxpayers can access the e-Bupot Instansi Pemerintah
application starting September 1, 2021, via the website address
https://ebupotip.pajak.go.id/ and https://subunitip.pajak.go.id/ (DGT, 2021). The flow of
use is as follows:
1. login: use the username and password of the Government Institution Taxpayer
djponline application to login into the application;
2. withholding tax slip recording: ecord the withholding tax slip can be made in two ways,
namely "key in" and/or "data import". The "key in" method means recording one by
one withholding tax slips. The "data import" method records large quantities with CSV
formatted data;
3. making billing: before making a payment, Government Institution Taxpayers should
make an id billing. e-Bupot Instansi Pemerintah application has given the facility to
make id billing;
4. posting: transferring withholding tax slips and payment data and updating data on
the tax return. We can proceed with the posting process if we have finished recording
the withholding tax slip;
5. recording tax payment slip: payment of tax payable that the withholding tax slips have
been recorded. Recording the tax payment slip can be done on the menu as follows:
unification tax returns/income tax Article 21, periodic tax returns, and recording the
withholding tax slip;
6. submit the tax return: the last step in using the e-Bupot Instansi Pemerintah
application is to send the tax return online.
2.3. Application in Theoretical Framework

The e-Bupot Instansi Pemerintah application is a breakthrough made by DGT.
Government Institution Taxpayers must use this application to create withholding tax
slips and report periodic tax returns. The purpose of making an application is to make it
easier for users (Siregar & Sari, 2018) and to help users complete specific tasks
(Haerulah & Ismiyatih, 2017). According to Susanti, Astari and Thamrin (2016), an
79
application is a device created to complete daily activities or tasks, such as advertising,
trade and other activities.
Making a good application considers two things, namely, the user interface and user
experience (Arifin, 2018). Galitz (2002) explains that the user interface is a tool to
beautify the application's appearance, making the application heard, seen and felt with
a simple display. A good user interface makes interaction and communication between
users and applications run well (Ganggi, 2019).
Meanwhile, user experience is related to application users' perceptions, responses,
and experiences (ISO in Munthe, Brata, & Fanani, 2018). A good user experience aims
to fulfil the desires and needs of users with the principle of simplicity so that users feel
comfortable using the application (Garret, 2011). User experience focuses on users'
perception and their experience before, when, and after using the application. (Bevan,
Carter & Harker, 2015).
There are four main components of user experience, according to Guo (2012):
a. value: does the application provide added value to the user?
b. usability: does this application make it easier for users to complete specific tasks?
c. desirability: do users have an exciting and enjoyable experience using the
application?
d. adoptability: will people start buying, downloading, installing and using this
application?
Meanwhile, according to Faisal et al. (2016) in Aniesiyah (2018), the user experience
components include interaction design, information architecture, visual design, and
usability. From the two explanations regarding the user experience components above,
there is one component that both of them have, namely usability. A simple definition of
usability is that users must easily use applications (Nielsen, 2012).
2.4. Usability
There are several definitions of usability presented by experts. According to
Hedegaard and Simonsen (2013), usability is a way to measure how an application can
help users adequately complete the tasks given. Of course, this still depends on the
conditions of the application, task and user. Rubin and Chisnell (2008) try to interpret
usability from the word's origin. The origin of the word usability is usable, which means it
can be used and put to good use. Therefore, an application can be successful and
efficient when it is valid and satisfies the user. If the application experiences a failure,
the failure can be removed or minimized (Rahadi, 2014).
Meanwhile, according to ISO 9241, usability is a measuring tool to determine whether
an application can meet user goals effectively, efficiently, and satisfactorily (Marthasari
& Hayatin, 2017). From the three definitions above, it can be concluded that usability is
a measuring tool to determine whether an application can complete tasks or meet user
expectations. Experts differ in opinion regarding the components that are measured in
usability measurements. However, there are also some similarities.
According to ISO 9241 in Hadi (2018), according to its definition, the usability
component includes effectiveness, efficiency, and satisfaction. Effectiveness sees
whether the application can complete the user's task. Efficiency describes how easily the
application can be used and operated by users. Satisfaction explains whether the
application users are satisfied.
Nielsen (2012) argues that usability has five main components:
a. learnability: how far the user can learn this product to complete a work assignment;
b. efficiency: know how precisely and quickly the user is using the application;
c. memorability: What the user remembers in using the application;
d. errors: how the user made a mistake and whether the user can overcome the
problem;
80
e. satisfaction: the level of user satisfaction with the application used (Novita & Amelia,
2019).
According to Petrie & Bevan (2009), there are four usability components:
a. flexibility: how far can the application accommodate changes and user desires;
b. learnability: convenience for users to learn as well as practice in operating the
application;
c. memorability: users still remember how to use this application;
d. safety: users feel safe, and there is no danger in using the application.
Usability is one of the determining factors for the success of an application. Usability
component explains whether the application can facilitate the user in completing specific
tasks. If an application has high usability, the user will continue to use the application.
Meanwhile, if the usability is low, the user will leave the application (Rahman & Vitalocca,
2018)
2.5. Usability Measurement

Why does usability need to be measured? Usability needs to be measured to
determine how far the user's ability is using an application. Besides that, it is also to see
user satisfaction with the application (Kusuma, Noviasari, & Marthasari, 2016). Because
as explained above, usability is the main component of a product or application that is
directly related to the user (Marthasari & Hayatin, 2017).
In general, usability measurements measure three main components: effectiveness,
efficiency, and user satisfaction. In simple terms, there are two methods used. The first
is by trusting the assumptions and estimates of the application maker. The second is to
use the metric usability method (Novita & Amelia, 2019).
Usability measurement is different from user testing. The purpose of user testing is
to see whether an application meets the expectations and needs of the user. User testing
is from the point of view of the application developer. Meanwhile, usability measurements
are carried out to determine whether the application can be used and appropriately
operated by users (Kasih & Delianti, 2020).
According to Tullis and Albert in Santoso (2018), measurable usability can be used
for various things, for example:
a. make the assessment more objective because it is not made based on assumptions
but based on suggestions and input sourced from data;
b. as a comparison of the two applications;
c. to group problems and constraints in the use of the application;
d. as a predictor of actual application usage;
e. provide an overview and illustration to the application owner.
Meanwhile, according to Aelani and Falahah (2012), there are four flows to go
through in measuring usability:
a. choose a suitable questionnaire. Generally, each questionnaire has a different
reason, scope, and approach;
b. determine participants who can represent based on aspects of work, age, gender,
and others;
c. choose a sample to be used as an object;
d. the collected data is processed and interpreted according to its characteristics.
Questionnaires are the most frequently used method for measuring usability.
Hariyanto, Triyono & Kohler (2020) convey several forms of questionnaires that can be
used, including:
a. Questionnaire for User Interaction and Satisfaction (QUIS) (Chin, Diehl, & Norman,
1988),
b. the Software Usability Measurement Inventory (SUMI) (Kirakowski & Corbett, 1993),
c. the Computer System Usability Questionnaire (CSUQ) (Lewis, 1995),
81
d. the questionnaire System Usability Score (SUS) (Brooke, 1996),
e. the USE questionnaire (Lund, 2001).
USE questionnaire is a questionnaire that is often used in measuring usability. USE
is an acronym for Usefulness, Satisfaction, and Ease of Use. In general, these three
things are usability components.
3.1. Research Flow
The following is a flow plan for this research:
Picture 1: Research Flow

Source: Researcher
3.2. The Type of Research
This type of research is quantitative research, where data analysis uses statistical
calculations. The data are explained systematically and descriptively.
3.3. Population and Determination of Respondents

The object of this research is users of the e-Bupot Instansi Pemerintah application
from the General Election Supervisory Agency (Bawaslu) working units throughout
Indonesia. There are 35 working units in the Bawaslu, consisting of 1 central working
unit and 34 provincial working units, with details described in Table 2. One respondent
represents one working unit, so there are 35 respondents. Respondents who filled out
the questionnaire were employees given the task of using and reporting periodical tax
returns through the e-Bupot Instansi Pemerintah application.
Table 2 List of Bawaslu Working Unit Respondents
Source: Bawaslu
Number Bawaslu Number Bawaslu

1 Aceh Province 19 Lampung Province
2 Bali Province 20 Maluku Province
3 Bangka Belitung Province 21 North Maluku Province
4 Banten Province 22 West Nusa Tenggara Province
5 Bengkulu Province 23 East Nusa Tenggara Province
6 DI Yogyakarta Province 24 Papua Province
82
Number Bawaslu Number Bawaslu
7 DKI Jakarta Province 25 West Papua Province
8 Gorontalo Province 26 Riau Province
9 Jambi Province 27 West Sulawesi Province
10 West Java Province 28 South Sulawesi Province
11 Central Java Province 29 Central Sulawesi Province
12 East Java Province 30 Southeast Sulawesi Province
13 West Kalimantan Province 31 North Sulawesi Province
14 South Kalimantan Province 32 West Sumatera Province
15 Central Kalimantan Province 33 South Sumatera Province
16 East Kalimantan Province 34 North Sumatera Province
17 North Kalimantan Province 35 Central
18 Riau Islands Province
3.4. The Data Collection Technique

The data collection technique used was a survey method with a questionnaire as an
instrument. The questionnaire consists of several open questions for respondents. The
questionnaire form was distributed online to be filled in by respondents via Google form
with the link address as follows: https://s.id/usability-bawaslu.
3.5. Question Instrument

Measuring usability in this research used the USE questionnaire created by Lund
(2001). According to ISO, there are three main dimensions of usability: Usefulness,
Satisfaction, and Ease of Use. Furthermore, the dimension of Ease of Use is expanded
into two: Ease of Learning and Ease of Use. So this research measures four dimensions
of the USE Questionnaire: Usefulness, Satisfaction, Ease of Learning, and Ease of Use.
Each of the dimensions above is described in several question instruments. The
question instruments used in the questionnaire are as follows:
Number Questions
Uselfulness
1 It helps me be more effective
2 It helps me be more productive
3 It is useful
4 It gives me more control over the activities in my life
5 It makes the things I want to accomplish easier to get done
6 It saves me time when I use it
7 It meets my needs
8 It does everything I would expect it to do
Ease of Use
9 It is easy to use
10 It is simple to use
11 It is user friendly
12 It requires the fewest steps possible to accomplish what I want to do
with it
13 It is flexible
14 Using it is effortless
15 I can use it without written instructions
16 I don't notice any inconsistencies as I use it
83
17 Both occasional and regular users would like it
18 I can recover from mistakes quickly and easily
19 I can use it successfully every time
Ease of Learning
20 I learned to use it quickly
21 I easily remember how to use it
22 It is easy to learn to use it
23 I quickly became skillfull with it
Satisfaction
24 I am satisfied with it
25 I would recommend it to a friend
26 It is fun to use
27 It works the way I want it to work
28 It is wonderful
29 I feel I need to have it
30 It is pleasant to use
The value measurement technique in this research uses a Likert scale. According to
Sugiyono (2013), the Likert scale helps measure the attitudes, perceptions, and opinions
of a person or group about a social phenomenon. The Likert scale is used for measuring
and weighing. This research uses a 5-point scale from the far left, which indicates a
negative answer, to the far right, which indicates a positive answer.
The Likert scale is made so that respondents answer questions or statements in the
questionnaire at various levels. The value categories are as follows:
Table 3 Answer Grade Categorization

QQ SD D N A SA
Value 1 2 3 4 5
Keterangan:
QQ : Questionnaire Questions
SD : Strongly Disagree
D : Disagree
N : Neutral
A : Agree
SA : Strongly Agree
3.6. Data Processing

The data that has been collected is processed and analyzed using descriptive
statistical methods to obtain the usability value of each component or parameter.

Based on the data collected on the questions in the USE Questionnaire, the
researcher analyzes it with simple data processing. Researchers also used the answers
to the open questions asked in the questionnaire to support and strengthen the analysis.
Before filling out the questionnaire, respondents were asked to do a series of tasks.
These tasks are essential to ensure that the respondent has experience using the e-
Bupot Instansi Pemerintah application. The series of tasks include the following activities:
a. login to the e-Bupot Instansi Pemerintah application;
b. record the withholding tax slip by manual mechanism;
c. record the withholding tax slip by data import mechanism;
d. post the withholding tax slip;
84
e. recording tax payments on the withholding tax slip; and
f. report the withholding tax slip in the periodic tax returns.
4.1. Average Value Per Parameter

The table below shows the average values of each parameter, namely Usefulness
(US), Ease of Use (EU), Ease of Learning (EL), and Satisfaction (SC).
Table 4 Average Value Per Parameter

Source: Processing Results
Number Parameter Average Value
1 US 4,0786
2 EU 3,8359
3 EL 3,8214
4 SC 3,8770
The table above shows that the average value of each parameter is close to 4.
Generally, respondents agree that the e-Bupot Instansi Pemerintah application has
fulfilled all parameters: Usefulness, Ease of Use, Ease of Learning, and Satisfaction.
However, considering that the average value is still not close to 5, the usability of the
Government Agency e-Bupot application is still open to improvement.
Of the four parameters, the Usefulness parameter has the highest value. This
condition shows that compared to the other three parameters, respondents considered
that the Usefulness parameter was felt more in their experience while using the
application. This Usefulness parameter relates to the application's ability to:
a. make respondents work more effectively and productively, and save time;
b. make respondents feel useful, achieve desires more efficiently, and do things that
are expected, as well;
c. be an application that suits the needs of respondents.
While of the four parameters, the Ease of Learning parameter has the lowest value.
This condition indicates that compared to the other three parameters, respondents
considered that they felt less experienced using the application. The Ease of Learning
parameters relates to the application's ability to:
a. quickly and easily learned by respondents;
b. easy to remember the steps for its use by respondents; and
c. easily mastered by respondents.
4.2. The Average Value Per Sub-Parameter

In more detail, based on the results of data processing, it was found that the five sub-
parameters with the lowest scores were sub-parameters number 15 (3,343), 23 (3,629),
14 (3,657), 16 (3,657), and 28 (3,657) respectively each refers to a parameter:
a. Ease of Use (EU)
1) 15: "I can use it without written instructions"
2) 14: "Using it is effortless"
3) 16: "I don't notice any inconsistencies as I use it"
b. Ease of Learning (EL)
23: "I quickly became skilled with it"
c. Satisfaction (SC)
28: "It's wonderful"
85
Graph 1 The Average Value Per Sub-Parameter
Source: Processing Results
From the results above, on a more detailed scale, respondents still really need written
instructions to use the application easily. In addition, respondents also stated that the
Government Agency's e-Bupot application is an application that is considered general
(not extraordinary) but not easy to master.
4.3. Respondents' Problems and Open Feedback

In the questionnaire submitted to the respondents, there were also two open
questions related to problems and related input. The answers to these open questions
are shown in the following table:
Table 5 Resumes of Problems from Respondents

Source: Respondents
Number Problem Group Detail
1 Data There are still difficulties regarding the
recording difference between data from
working subunits and working unit
The process of data reconciliation
requires much effort
Checking data that has been recorded
must be done one by one
2 Use Difficulty understanding the menus in
the application
Instructions for using the application are
not available
3 System Billing code generation problems
Problems related to the internet network
Problems related to calculation
systems/formulas in applications that
are not suitable
86
Table 6 Resume of Feedbacks from Respondents
Source: Respondents
Number Feedback Group Detail
1 Application Need to make a tutorial video on
instructions/training Youtube and a complete guide book
Technical guidance is needed
It is necessary to provide instructions for
using the application that is integrated
with the application
2 Data input feature Posting authority should also be
assigned to working subunits
A difference feature is provided to
facilitate searching for incorrect data
input
Provided a feature to calculate the tax
base input
3 Interconnection There needs to be a connection
between the e-Bupot Instansi
Pemerintah application with other
Ministry of Finance applications so the
treasurer does not have to input the
same data multiple times
The problems and input from the respondents above are in line with the results of
previous analysis, which placed the Ease of Learning parameter as the parameter with
the lowest value. From the table above, it is known that respondents need adequate
instructions, guidebooks, or video tutorials in order to learn how to use the application
properly.
5. CONCLUSION
It has been mandatory for Government Institution Taxpayers to use the e-Bupot
Instansi Pemerintah application since the September 2021 tax period. This application
will make it easier for Government Institution Taxpayers to fulfil their tax
deduction/collection obligations and report them. The e-Bupot Instansi Pemerintah
application must have good usability to meet these expectations. In this research, the
researchers used the USE questionnaire method to analyze the usability.
Based on data obtained from working units of Bawaslu throughout Indonesia, the
researcher found that, in general, respondents agreed that the e-Bupot Instansi
Pemerintah application had met all usability parameters: Usefulness, Ease of Use, Ease
of Learning, and Satisfaction. This result means that the e-Bupot Instansi Pemerintah
application has good usability. However, considering that the average value is still not
optimal yet (not close to 5), the usability of the e-Bupot Instansi Pemerintah application
is still open for improvement, especially related to the Ease of Learning parameter. In
addition, based on the answers to the open questions posed in the questionnaire,
respondents also conveyed problems related to input data, systems, and technical use
of the e-Bupot Instansi Pemerintah application.
The Directorate General of Taxes can make several efforts to increase the usability
of the e-Bupot Instansi Pemerintah application by increasing the Ease of Learning
parameter's value. The Efforts are providing adequate instructions, guidebooks, or video
tutorials. It needs to be carried out so Government Institution Taxpayers can learn how
to use the application properly.
87
The implication of the results of this research for the Directorate General of Taxes is
that apart from focusing on technical application creation, it is also necessary to pay
attention to non-technical factors that can make it easier for users to learn how to use
the application.
There are several limitations to this research. The first limitation is the extensive
research population, namely the Government Institution Taxpayers throughout
Indonesia. However, the sampling, namely Bawaslu working units throughout Indonesia,
was considered sufficient by researchers to represent all regions in Indonesia.
The second limitation is that there is no mechanism for supervising respondents to
carry out the stages before filling out the questionnaire (the stage of implementing the
series of tasks). This limitation is because the respondents are in different places beyond
the physical reach of the researcher.
For further research, a more complex analysis can be developed to determine the
compliance of Government Institution Taxpayers in reporting periodic tax returns with the
e-Bupot Instansi Pemerintah application. This suggestion is given because this research
focuses only on the usability of the e-Bupot Government Agency application.
88
REFERENCES
Books / Journals / Scientific Works
Aelani, K & Falahah (2012). Pengukuran Usability Sistem Menggunakan Use
Questionnaire (Studi Kasus Aplikasi Perwalian Online STMIK “Amikbandung”. In
Seminar Nasional Aplikasi Teknologi Informasi (SNATI).
Aniesiyah, A. N. (2018). Perancangan user experience aplikasi pelaporan keluhan
masyarakat menggunakan metode human-centered design (Doctoral
dissertation, Brawijaya University).
Arifin, I. N. (2018). Evaluasi Dan Perancangan User Interface untuk Meningkatkan User
Experience Menggunakan Metode Human Centered Design dan Heuristic
Evaluation pada Aplikasi EzySchool (Doctoral dissertation, Brawijaya University).
Bevan, N., Carter, J., & Harker, S. (2015, August). ISO 9241-11 revised: What have we
learnt about usability since 1998?. In International conference on human-
computer interaction (pp. 143-151). Springer, Cham.
Directorate General of Taxes (DGT). (2021). Panduan Penggunaan Aplikasi e-bupot
Instansi Pemerintah. Jakarta: Directorate of Information and Communication
Technology.
Galitz, Wilbert O. (2002). The Essential Guide to User Interface Design Second Edition.
John Wiley & Sons, Inc. Canada
Ganggi, R. I. P. (2019). Evaluasi Desain User Interface Berdasarkan User Experience
Pada iJateng. Jurnal Ilmu Perpustakaan, 8(4), 11-21.
Garrett, Jesse James. (2011). The Elements of User Experience: User-Centered Design
for the Web and Beyond, Second Edition. Berkeley: New Riders1249 Eighth
Street
Guo, F. Y., & Principal, U. X. (2012). Not just usability–The four elements of user
experience. Taken from https://www.uxmatters.com/mt/archives/2012/04/more-
than-usability-the-four-elements-of-user-experience-part-i.php accessed on 30
July 2022
Hadi, K. R. (2018). Analisis Dan Perbaikan Usability Aplikasi Mobile KAI Access Dengan
Metode Usability Testing Dan Use Questionnaire (Doctoral dissertation,
Brawijaya University).
Haerulah, E., & Ismiyatih, S. (2017). Aplikasi e-commerce penjualan souvenir
pernikahan pada toko “XYZ”. PROSISKO: Jurnal Pengembangan Riset dan
Observasi Sistem Komputer, 4(1).
Hariyanto, D., Triyono, M. B., & Kohler, T. (2020). Usability evaluation of personalized
adaptive e-learning system using USE questionnaire. Knowledge Management
& E-Learning: An International Journal, 12(1), 85-105.
Harjowiryono, M. (2019). Analisis Faktor-Faktor yang Memengaruhi Kepatuhan
Bendahara Pemerintah dalam Penyetoran Pajak. Indonesian Treasury Review:
Jurnal Perbendaharaan, Keuangan Negara Dan Kebijakan Publik, 4(3),195-217.
Hedegaard, S., & Simonsen, J. G. (2013, April). Extracting usability and user experience
information from online user reviews. In Proceedings of the SIGCHI Conference
on Human Factors in Computing Systems (pp. 2089-2098).
Kasih, A., & Delianti, V. I. (2020). Analisis Usability Nagari Mobile Banking Menggunakan
Metode Usability Testing dengan Use Questionnaire. Voteteknika (Vocational
Teknik Elektronika dan Informatika), 8(1), 124-131.
Kusuma, W. A., Noviasari, V., & Marthasari, G. I. (2016). Analisis Usability dalam User
Experience pada sistem KRS online UMM menggunakan USE Questionnaire.
Jurnal Nasional Teknik Elektro dan Teknologi Informasi (JNTETI), 5(4), 294-301.
Lund, A. M. (2001). Measuring usability with the use questionnaire12. Usability
interface, 8(2), 3-6.
89
Marthasari, G. I., & Hayatin, N. (2017, November). Analisis usability terhadap sistem
Lective Gegulang berbasis USE Questionnaire. In Prosiding SENTRA (Seminar
Teknologi dan Rekayasa) (No. 3).
Munthe, R. D., Brata, K. C., & Fanani, L. (2018). Analisis User Experience Aplikasi Mobile
Facebook (Studi Kasus pada Mahasiswa Universitas Brawijaya). Jurnal
Pengembangan Teknologi Informasi dan Ilmu Komputer e-ISSN, 2548, 964X.
Nielsen, J. (2012). Usability 101: Introduction to usability. Taken from
https://www.nngroup.com/articles/usability-101-introduction-to-usability/
accessed on 30 July 2022
Novita, D., & Amelia, L. (2019). Analisis Usability Aplikasi Pengisian KRS Online STMIK
XYZ Palembang Menggunakan Use Questionnaire. Jurnal Informasi Dan
Komputer, 7(1), 17-28.
Petrie, H., & Bevan, N. (2009). The evaluation of accessibility, usability, and user
experience. The universal access handbook, 1, 1-16.
Rahadi, D. R. (2014). Pengukuran usability sistem menggunakan use questionnaire
pada aplikasi android. JSI: Jurnal Sistem Informasi (E-Journal), 6(1).
Rahman, E. S., & Vitalocca, D. (2018). Analisis Usabilitas Menggunakan Use
Questionnaire Pada Sistem Informasi Smk Negeri 3 Makassar. Jurnal
Mekom, 5(1), 16-22.
Ruung, N. V., Ilat, V., & Wokas, H. R. (2017). Analisis Kontribusi dan Pelaporan Pajak
Hotel dan Pajak Air Tanah terhadap Pendapatan Asli Daerah Kota Manado.
Going Concern: Jurnal Riset Akuntansi, 12(2).
Santoso, J. (2018). Usability user interface dan user experience media pembelajaran
kamus Kolok Bengkala Berbasis Android. Jurnal Sistem dan Informatika (JSI),
12(2), 174-181.
Siregar, H. F., & Sari, N. (2018). Rancang Bangun Aplikasi Simpan Pinjam Uang
Mahasiswa Fakultas Teknik Universitas Asahan Berbasis Web. (JurTI) Jurnal
Teknologi Informasi, 2(1), 53-59.
Sugiyono. 2013. Metode Penelitian Bisnis (Pendekatan Kuantitatif, Kualitatif dan R&D).
Bandung : CV. Alfabeta.
Susanty, W., Astari, I. N., & Thamrin, T. (2019). Aplikasi GIS Menggunakan Metode
Location Based Service (LBS) Berbasis Android. Explore: Jurnal Sistem
Informasi dan Telematika (Telekomunikasi, Multimedia dan Informatika), 10(1).
Susherdianto, R. (2014). Analisis faktor-faktor yang mempengaruhi tingkat kepatuhan
wajib pajak (studi WPOP yang memiliki usaha di Kota Kudus). Thesis.
Website:
https://www.pajak.go.id/id/kewajiban-perpajakan accessed on 30 July 2022
https://pajak.go.id/id/artikel/lebih-mudah-dengan-spt-masa-unifikasi-instansi-
pemerintah accessed on 30 july 2022
Statutory regulations
Republic of Indonesia. (2021). Undang-Undang Nomor 7 Tahun 2021 tentang
Harmonisasi Peraturan Perpajakan sebagai prubahan terakhir dari Undang-
Undang Nomor 6 Tahun 1983 tentang Ketentuan Umum dan Tata Cara
Perpajakan. Lembaran Negara Republik Indonesia Tahun 2021 Nomor 246.
Jakarta: Ministry of State Secretariat.
Republic of Indonesia. (2019). Peraturan Menteri Keuangan Nomor 231 Tahun 2019
tentang Tata Cara Pendaftaran dan Penghapusan Nomor Pokok Wajib Pajak,
Pengukuhan dan Pencabutan Pengukuhan Pengusaha Kena Pajak, serta
Pemotongan dan/atau Pemungutan, Penyetoran, dan Pelaporan Pajak bagi
90
Instansi Pemerintah. Berita Negara Republik Indonesia Tahun 2019 Nomor 1746.
Jakarta: Ministry of Finance.
Republic of Indonesia. (2021). Peraturan Direktur Jenderal Nomor Per-17/PJ/2022
tentang Bentuk dan Tata Cara Pembuatan Bukti Pemotongan dan/atau
Pemungutan Pajak, serta Bentuk, Isi, Tata Cara Pengisian, dan Penyampaian
Surat pemberitahuan Masa Bagi Instansi Pemerintah. Jakarta: Directorate
General of Taxes.
91
IMPACT OF AUTOMOTIVE TAX INCENTIVES DURING
COVID-19: EVIDENCE FROM INDONESIA
Sukaryoa, Sigit Hariyantob

aDirectorate General of Taxes, Jakarta, Indonesia. Email: sukaryo1@gmail.com
bDirectorate General of Taxes, Jakarta, Indonesia. Email: sigit.hariyanto10@gmail.com
ABSTRACT
The COVID-19 pandemic created extraordinary health and economic disruption on a global scale, including
in Indonesia, which experienced negative growth in the major economic sectors. In 2021, as part of the PEN
program, the government introduced additional tax incentives to boost durable goods consumption, the
PPnBM incentives for qualified new car purchases. This study aims to explore the PPnBM incentives’ impact
on automotive sectors in 2021, utilizing the ITSA methodology and seasonal forecasting method with Holt-
Winters exponential smoothing and ARIMA. In particular, using the automobile wholesales and tax
administration data, this paper assesses the impact of the PPnBM incentives on car wholesales, participating
firms’ sales, purchases, and wages expenditure. The result suggests that the PPnBM incentives’ had a
modest impact on wholesales recovery and economic activity for participating automotive manufacturers
during the incentives period.
Keywords: Tax Incentives, Automotive Sectors, COVID-19, ITSA
Dampak Insentif Pajak Kendaraan Bermotor Selama Pandemi COVID-19: Kasus di Indonesia. Pandemi
COVID-19 telah menyebabkan krisis global di bidang kesehatan dan ekonomi, tidak terkecuali dengan
Indonesia. Pada tahun 2021, sebagai bagian dari program Pemulihan Ekonomi Nasional (PEN), pemerintah
memberikan insentif pajak khusus sebagai salah satu bentuk kebijakan fiskal untuk mendorong konsumsi
barang dalam bentuk insentif PPnBM untuk pembelian kendaraan bermotor baru tertentu. Penelitian ini
bertujuan untuk mengukur dampak pemberian insentif pajak tersebut pada industri otomotif di tahun 2021,
dengan menggunakan pendekatan ITSA dan prediksi musiman melalui metode Holt-Winters dan ARIMA.
Dengan kombinasi data penjualan kendaraan bermotor dan data administrasi perpajakan, studi ini
menunjukkan pengaruh kebijakan insentif PPnBM terhadap penjualan mobil, penjualan usaha, pembelian,
dan biaya gaji perusahaan otomotif pemanfaat insentif. Secara umum, terdapat perbaikan kinerja penjualan
dan aktivitas ekonomi perusahaan peserta insentif PPnBM selama periode pemberian insentif – tetapi
sifatnya relatif moderat.
Kata kunci: Insentif Pajak, Sektor Otomotif, COVID-19, ITSA
1. INTRODUCTION
The COVID-19 pandemic caused significant global-scale disruption in health and

the economy, especially in 2020 and 2021. In Indonesia, firms and businesses
experienced the severe impact of the pandemic due to restricted social mobilization and
falling demand (for example, World Bank, 2020; BPS [Statistics Indonesia], 2020).
Consequently, the economy contracted during the three quarters of 2020, brandishing
Indonesia with recession status – a first since the 1998 crisis (BPS, 2021).
Apart from a few resilient sectors, most of the industry groups’ growth slashed,
including manufacturing – the largest sector – which recorded a -2.93% growth in 2020
(BPS, 2021). Specifically, among the manufacturing subsectors, the transport equipment
industry experienced the most alarming contraction, -19.86%, as car production
contracted by 46.37% in 2020 (BPS, 2021). In March 2021, along with COVID-19 Tax
Incentives under the PEN (‘Pemulihan Ekonomi Nasional’ or National Economic
Recovery) Program, the government introduced additional tax incentives, providing a tax
cut for new car purchases (‘PPnBM Incentives’) 40.
40 Luxury goods sales tax or PPnBM (‘Pajak Penjualan atas Barang Mewah’)
92
During a recession or economic downturn, the government could intervene in the
output fluctuation through fiscal policy, stimulating the demand through increased
spending and tax relaxation (for example, see Auerbach and Feenberg, 2000; Atkinson,
2008; Elmendorf and Furman, 2008). While there were concerns about the fiscal stimulus’
effectiveness during an economic crisis, some arguments suggested that discretionary
fiscal stimulus would be “helpful if well-crafted” (Elmendorf and Furman, 2008). Lawrence
Summers (2008) proposed three principles for designing fiscal stimulus during a crisis:
Timely, Targeted, and Temporary. In short, Summers (2008) suggested that the
government provide timely provisions, target “those in needs”, and ensure that the
stimulus will not affect the long-run fiscal position (also see Elmendorf and Furman, 2008;
Taylor and Castillo, 2015).
When the pandemic hit, the global governments prioritized measures to contain the
virus outbreak and, expectedly, led to significant economic disruption (OECD, 2020). In
particular, small and medium-sized enterprises (SMEs) and self-employed businesses
faced a more significant constraint during the crisis (OECD 2020). Therefore, through
fiscal policy, many countries implemented policy responses to support impacted
businesses and households, focusing on liquidity and income support. (OECD, 2020).
However, Chetty et al. (2020) suggested that the primary cause of the slashed economic
output was the spending cut by more affluent households, especially during the initial
period of the crisis. From a recovery perspective, the consumers might decide to spend
later, especially for durable goods after a crisis (“pent-up demand”) (Hodbod et al., 2021;
Baraja & Wolf, 2021; Caballero, 1993). Thus, following the “pent-up demand” mechanism,
it would be relevant for the government to encourage durable goods consumption, e.g.
motorcars, through discretionary fiscal incentives – especially when the automobile
market is relatively sensitive toward price change (Copeland, 2014).
In March 2021, the government – through the Ministry of Finance – issued the
Ministry of Finance regulation to relieve the PPnBM levies on eligible vehicles based on
the criteria by the Ministry of Industry. Before the incentives, car sales in Indonesia were
subject to two indirect taxes, namely the VAT (Value-added taxes) and PPnBM (Luxury
Goods Sales Taxes). In short, the PPnBM is a one-off tax payment at the manufacturing
level or during the importation, ranging from 5% to 30% based on car type and engine
displacement size. Under the Ministry of Finance Regulation number 20 of the Year 2021
(subject to three amendments), the government provided a PPnBM relaxation for
qualifying small-medium passenger car models. The Ministry of Industry regulated the
eligible vehicles based on the local contents criteria and listed the models in Regulation
839 of the Year 2021 (subject to further amendment).
While the macroeconomic indicators suggested that the PPnBM incentives boosted
sectoral performance41, the literature on the impact is somewhat limited42. The initial
study from ISI (2021) suggested that the PPnBM Incentives significantly impacted the
automotive sector recovery, increasing: total output, employment, and household
revenue. However, the study found that while motorcar sales improved from March to
May 2021, it has not reached the pre-pandemic level (ISI, 2021). Therefore, this study
attempted to fill the knowledge gap, incorporating micro-level tax administration data to
41
For example, the Ministry of Industry’s press release on the car incentives
(https://kemenperin.go.id/artikel/23126/Insentif-PPnBM-DTP-Terbukti-Dongkrak-Pertumbuhan-
Manufaktur-; accessed 1 July 2022), the Gaikindo (Automotive Indonesia or ‘Gabungan Industri
Kendaraan Bermotor Indonesia’) statement on the PPnBM Incentives (https://www.gaikindo.or.id/gaikindo-
relaksasi-ppnbm-selamatkan-industri-otomotif/; accessed 1 July 2022), and the Ministry of Finance’s
explanation on the incentives (https://www.kemenkeu.go.id/publikasi/berita/kemenkeu-perpanjang-insentif-
ppnbm-kendaraan-bermotor/; accessed 1 July 2022).
42 Based on author’s knowledge, there is one research on the impact of the PPnBM Incentives on car
sales by ISI (Institute for Strategics Inisiative), presented in the Ministry of Industry’s Webinar in 23
September 2021, available online at https://www.youtube.com/watch?v=JOviIVDj6e0
93
provide an alternative assessment. Also, this study will contribute as an extension of the
initial research on the PPnBM incentives’ impact on the automotive industry, focusing on
the sectoral performance following the incentives based on the firm’s sales, purchases,
and employment.

DEVELOPMENT
This data will incorporate two data sources: motorcar wholesales 43 data from
Gaikindo and aggregate firm-level data based on tax administration. Using the
wholesales data, we attempted to measure the impact of the PPnBM Incentives on car
sales in 2021. Based on the brand and model monthly sales volume, we constructed a
dataset based on the incentives’ eligibility (Ministry of Industry Regulation Number 839
of the Year 202144) as the treated group and comparable car models as the control group.
We match the comparable car models based on the Gaikindo classifications (i.e., sedan
and 4X2 type) and the displacement size. Altogether, we collected around 24 thousand
observations consisting of around a thousand car models from 25 manufacturers (brand)
spanning January 2017 to December 2022, summarized as follows.
GAIKINDO dataset summary
Sample Number of Obs.
Description/Category
Period Eligible (1) Non-Eligible (0) Total
Total 9072 13368 22440
Sedan (engine size < 1500 cc) Jan 2017 to Dec 264 588 852
4X2 Small (< 1500 cc) 2021 6900 4920 11820
4X2 Medium (1500 - 2500 cc) (60 months) 1860 6408 8268
4X4 Medium (1500 - 3000 cc) 48 1452 1500
Tax Administration dataset summary

Number of Descriptive Stat. (log)
Description Sample Period
Obs. Min Mean Max Std. Dev
Incentives Participant 6 Firms
Total Sales Jan 2017 - Dec 2021 28.2917 30.3281 30.7147 0.4423
Total Purchases Jan 2017 - Dec 2021 28.8769 30.2159 30.6806 0.3369
Total Salary/Wage Jan 2018 - Dec 2021 26.5913 27.0212 28.0610 0.3501
Non-Participant 6 Firms
Total Sales Jan 2017 - Dec 2021 0.0000 23.7816 27.0464 4.3065
Total Purchases Jan 2017 - Dec 2021 19.0304 24.2499 27.1164 1.8370
Total Salary/Wage Jan 2018 - Dec 2021 19.3885 21.7039 23.8421 0.6618
Figure 1. Gaikindo and Tax Administration dataset Summary45
For the tax administration data, we aggregated the firm-level data to maintain
anonymity and allow for a more straightforward comparable group. As the treated group,
43 Gaikindo offers several dataset from the https://www.gaikindo.or.id/indonesian-automobile-industry-

data/ , namely the production, wholesales, and retail sales. As the government imposed PPnBM on the
transaction from manufacturers to distributors, the wholesales data would be more representative to
measure the incentives impact on the direct output.
44 For the list of eligible vehicles, see Appendix 1.
45 In the November 2021, the Ministry of Industry released a regulation update, increasing the eligible
vehicle list with five car models under the Affordable Energy Savings Cars. We exclude those car types
from our sample as the PPnBM Incentives encompassed all models in this category, i.e., there is no
comparable category (counterfactuals).
94
we pooled the six incentives participants (manufacturer), all classified under the
automotive manufacturing industry 46. To create a comparable group, we aggregated
other firms registered as automotive manufacturers (within the tax administration
database). As indicators of a firm’s economic performance, this study focused on sales,
purchases, and employment based on the reported tax return data. For sales and
purchases, we aggregated the total sales and purchases based on the VAT (Value-
added tax) return.
Meanwhile, for the employment data, we proxied it with the aggregated total wages
based on the ‘PPh 21’47 return. This study incorporates a parameter for employment as
several studies found that employment strategies, e.g., wage cuts and furlough, were
prevalent during the pandemic (for example, see Chetty et al., 2020; World Bank, 2020;
BPS, 2020; Rosid et al., 2022). Thus, recovered employment would be one of the
indicators of recovery.
Empirical and analysis approach. To measure the impact of the PPnBM incentives
on the eligible vehicles and incentives participants, we incorporated the ‘Interrupted
Time-Series Analysis’ (ITSA). In short, the ITSA approach is a quasi-experimental
methodology to measure an outcome variable by comparing pre and post-intervention
periods (Linden, 2015). Using a control group as a comparison, the regression model is
as follows (Linden, 2015).
Y_t=β_0+β_1 T_t+β_2 X_t+β_3 X_t T_t+β_4 Z+β_5 ZT_t+β_6 ZX_t+β_7 ZX_t T_t+ε_t
In the equation, T_t denotes the time since the start of the study, X_t is an indicator
of the intervention period (1), Z is a dummy variable for control group assignment, while
X_t T_t, ZT_t, ZX_t, and ZX_t T_t are interaction terms (Linden, 2015). Notably, we are
interested in the value of β_t which denotes the difference between treatment and control
groups in the trend of dependent variable following the intervention compared to the
preintervention (Linden, 2015). For this study, we incorporated two periods of time
interruption: March 2020 to denote the start of the pandemic in Indonesia and March
2021 to indicate the PPnBM incentives period.
This study also attempted to measure such phenomena by echoing the ISI findings,
which suggested that car sales have not reached the pre-pandemic level. However,
rather than using the pre-pandemic figures as a comparison (also in Chetty et al., 2020),
we forecasted the 2020 and 2021 levels based on the previous periods’ trends – to create
a non-pandemic counterfactual. The extrapolated figures, e.g. wholesales and firm’s
economic activity, will be a reference level to assess the sectoral recovery. An equal or
higher post-incentives trend (from March to December 2021) will suggest a recovery
indicator. To allow for a more straightforward and intuitive analysis, we alternated
between the Holt-Winters seasonal smoothing and ARIMA (Auto Regressive Integrated
Moving Average) based on the monthly figures of 2017 to 201948 to forecast the target
outcomes (i.e., car wholesales and firm’s economic activity).
4. RESEARCH OUTCOME AND ANALYSIS

In this section, we explored the PPnBM Incentives’ impact on wholesales and
participants’ economic performance, namely the sales, purchases, and wage
expenditure. The first part of the analysis for each parameter began with the post-
46 We refer to the tax administration industrial classification (‘Klasifikasi Lapangan Usaha’ or KLU) code.
For the six incentives participants, all firms are registered under the 29100 KLU code for “Manufacture of
motor vehicles”.
47 Income tax on employment income, generally salary and wage; withheld by the employer.
48 Except for the wage and salary data which only available from January 2018
95
incentives assessment, focusing on the trend differences between the eligible group and
the non-eligible. Subsequently, we measured the recovery trend of the actual figures in
2021 compared to the pre-pandemic trajectory.
Impact on automotive wholesales. Figure 2 below plots the aggregate monthly
wholesales for incentives-eligible and non-eligible motorcars from January 2017 to
December 2021. Based on a quick visual interpretation, we could observe that: (1) the
eligible motorcars took up a more significant market share than the non-incentives group;
(2) within three months into the pandemic, i.e., May 2020, both groups experienced rock-
bottom sales level, but gradually recovered; and (3) at the lowest point, the eligible
vehicles experienced the most severe sales contraction, relative to the pre-COVID trend.
Following the rolling out of the PPnBM Incentives in March 2021, the eligible group
demonstrated an expansive trend while the non-incentives vehicles’ trend dwindled. The
difference grew more evidently towards the last quarter of 2021, nearing the concluding
period of PPnBM Incentives.
Figure 2. Monthly Wholesales (log-scale)

For the first empirical analysis, this paper employed the ITSA approach for the
wholesales parameter, including for subgroups, i.e. sedan, 4X2 small, and 4X2 medium.
Figure 3 summarizes the ITSA output for all assessed indicators. As the primary
objective of this study, we estimated the slope difference between eligible cars and the
non-eligible group during the incentives period. The empirical result suggested that
eligible car models experienced more expansive wholesales from March to December
2021 than non-eligible models (positive coefficient, statistically significant at 5%
confidence level – see line 1). However, the result also suggested no significant
difference in incentives-qualifying cars’ wholesales before and after the PPnBM
Incentives. While the negative coefficient would suggest a slower trend, it is not
statistically significant even at a 10% confidence level.
When looking into the car models, we found that only the small-sized 4X2 category
demonstrated a similar result: a higher post-intervention trend than the non-eligible cars.
The output for other car models – sedan and medium-sized 4X2 – indicated a non-
significant slope difference. Meanwhile, all car models’ results implied that the wholesale
trend was relatively unchanged during the pandemic and incentives period.
Even though the primary output confirmed the initial expectation that PPnBM
Incentives boosted eligible car sales, there are several concerns about the interpretation.
First, a non-price factor might affect the qualifying-car sales during the incentives period,
for example, new car models. At the same time, the price effect might motivate the
consumers to decide on the eligible cars or substitute effect. Lastly, the not statistically
96
significant difference between the pandemic and incentives period might have resulted
from the “pent-up demand”. Visually, it is observable that from August 2020, the
wholesale for both eligible and non-eligible cars had recovered – albeit limited. Thus, we
would argue that the consumers resumed their durable-goods consumption, and the
incentives managed to catch the recovery momentum to bolster the demand.
Impact on Wholesales
Total Sedan 4X2 Small 4X2 Medium

(1) Difference (during PPnBM) 0.0850** 0.0630 0.1170** 0.0782
(Treated - Control) (0.0360) (0.1436) (0.0580) (0.0677)
(2) PPnBM x Incentives -0.0260 0.0779 0.0790 -0.1077

(from March 2021) (0.1049) (0.1601) (0.1213) (0.1173)
(3) COVID x Incentives 0.1175 -0.0149 0.0387 0.2021**

(from March 2020) (0.0986) (0.0741) (0.1068) (0.0962)
(4) Treatment 8.5465*** 2.5502*** 8.0389*** 5.9911***

(Incentives) (0.1108) (0.5486) (0.1781) (0.1850)
(5) PPnBM Period 0.0360 -0.4703 -0.0483 0.2125

(0.2434) (0.4716) (0.4166) (0.4502)
(6) COVID Period -0.6985*** -0.1041 -1.0412*** -0.9093***

(0.1720) (0.3326) (0.2922) (0.3089)
Impact on Sales Impact on Purchases Impact on

Total Domestic Export Total Domestic Import Wages
(1) Difference (during PPnBM) 0.0327 0.0528 -0.4206 0.1545** -0.0584 0.4289 0.0285
(Treated - Control) (0.0748) (0.0822) (0.5334) (0.0611) (0.0414) (0.4726) (0.0597)
(2) PPnBM x Incentives -0.0417 -0.0585 -0.1538 -0.1548 -0.1158* 0.1105 0.0200
(from March 2021) (0.1108) (0.1352) (0.6683) (0.0976) (0.0690) (0.5108) (0.0678)
(3) COVID x Incentives 0.2231** 0.2638** -0.2251 0.1932*** 0.0846 0.4813** 0.0092
(from March 2020) (0.0930) (0.1158) (0.4091) (0.0625) (0.0559) (0.2008) (0.0342)
(4) Treatment 9.7396*** 9.7086*** 17.4038*** 6.6998*** 7.7707*** 10.8958*** 5.2797***

(Incentives) (1.1592) (1.1396) (1.5778) (0.3289) (0.2072) (1.3860) (0.1651)
(5) PPnBM Period 0.2304 0.0157 -2.9159 0.8204 -0.0991 0.5175 0.1230
(0.4275) (0.4788) (4.1695) (0.5388) (0.2887) (2.8343) (0.2558)
(6) COVID Period -2.5086*** -2.5365*** -4.5012 -0.4789 -0.9192*** 0.2022 -0.1971
(0.6908) (0.6906) (3.0682) (0.3396) (0.2157) (1.2159) (0.1933)
Standard errors in parentheses; asterisks indicate significance level at 1% (***), 5% (**), and 10% (*)
Figure 3. The Impact of PPnBM Incentives on Observed Indicators
Measuring the wholesales recovery. In order to assess the recovery, we

incorporated the Holt-Winters exponential smoothing approach to extrapolate the
eligible-car sales. Subsequently, we assessed the recovery indication by comparing the
actual sales with the forecasted wholesale for the period starting from March until
December 2021. We expect comparable or higher actual sales than the projected
amount to affirm the recovery notion. Thus, we performed t-tests on the equality of means
for the projected sales compared to the actual values.
The exponential smoothing approach provided a robust forecasting model with more
than 97% accuracy for the testing period. Figure 4 below exhibited the seasonal model,
24 months extrapolation (from January 2020 to December 2021), and actual wholesales.
The t-test result suggested that there was no significant difference between extrapolated
and actual wholesales during the incentives period (March to December 2021). While
the forecasted value had a relatively higher mean (difference of 0.0806), the p-value is
not significant even at a 10% confidence level: Pr(|T| > |t|) = 0.3189 and Pr(T > t) =
97
0.1594. Therefore, this study would argue that during the PPnBM Incentives period, the
eligible-car sales had recovered to the pre-pandemic trend.
Figure 4. Projected vs Actual Wholesales, Eligible Vehicles (log-scale)
Impact on firms’ total sales. The subsequent analysis is on the firms’ total sales
parameter, based on the reported sales from the VAT returns. Figure 5 below plots the
monthly total sales for PPnBM Incentives participants and non-participants from January
2017 to December 2021. The general trend suggested that: (1) similar to the wholesales
data, the participants of the incentive had a more significant total sales compared to non-
participants; (2) both groups experienced severe contraction during the pandemic, but
incentives participants demonstrated a relatively shorter period of slashed sales; (3)
notably, the participants of the incentive were having a “v-shaped” trend while the non-
participants undergo a somewhat “w-shaped” trend. However, during the PPnBM period,
the non-participants began to recover, and the participants maintained their sales
trajectory.
The ITSA output confirms the less striking difference for both groups post-
intervention, the higher coefficient for treated groups was insignificant even at a 10%
confidence level (see Figure 3). In other words, there was no significant difference in
total sales between treated and control groups during the incentives period. For the
incentives participant, the post-intervention slope is not significantly different compared
to the COVID-19 period. One possible explanation is that the total sales for participating
firms recovered before the PPnBM period. The visual representation in Figure 6 indicates
that the total sales for the participating firms began to stabilize around the fourth quarter
98
of 2020. Similarly, in the ITSA output, the treated groups experience a significantly higher
total sales trend than the control group during COVID – at a 5% significance level.
Figure 5. Monthly Total Sales (value, log-scale)
When zooming into sales components, domestic sales and export demonstrated no
slope difference (not statistically significant) between the treated and control firms –
similar to the total sales result. Also, there is no significant difference for all sales
components pre and post-intervention for incentives participants. Based on the export
components, there are no significant differences for all parameters between the
participants and non-participants (aside from the higher export level for treated groups).
The result might be expected as the PPnBM is a domestic-sales levy and the incentives
policy targeted local consumption rather than bolstering the export activities.
Measuring the total sales recovery. The subsequent analysis for the total sales
parameter assesses the recovery trend: by comparing the total sales of participating
firms with the projected sales. The ARIMA with a 12-period seasonality model provided
a robust fit for the trained dataset (see Figure 6). This research continued with the
participants’ total sales projection during the incentives period, which served as a pre-
pandemic baseline. Based on the t-test approach, this study found no significant
difference between the actual total sales and projected values. The extrapolated total
99
sales had a relatively higher mean (difference of 0.0787), the p-value is not significant at
a 10% confidence level: Pr(|T| > |t|) = 0.2050 and Pr(T > t) = 0.1025. Thus, this study
would argue that the incentive participants recovered their total sales to the pre-COVID
level during the PPnBM incentives period.
Figure 6. Projected vs Actual, Total Sales for incentives participants (log-scale)

Impact on firms’ total purchases. Another primary indicator of a firm’s performance
is purchasing activity. In this study, the firms’ total purchases will be based on the
reported input from VAT returns. We could observe a relatively contrasting trend based
on the monthly purchases for treated and non-treated firms from 2017 to 2021 (see
Figure 7). The visual representation suggests that: (1) similar to the previous parameters
(wholesales and total sales), the treated group had a more significant total purchase
compared to non-participants; (2) both groups experienced slashed purchases during
the pandemic, but incentives participants demonstrated a relatively modest contraction;
(3) the treated firms were having a “v-shaped” recovery while the control groups
demonstrated a “w-shaped” purchases during the pandemic. However, both groups
began to expand during the PPnBM period, which will be subject to subsequent empirical
analysis.
Figure 7. Monthly Total Purchases (value, log-scale)
The ITSA result generally confirms the participating firms’ positive trend during the
treatment period, as shown in Figure 3. The treated firms experienced relatively higher
purchases than the control group (significant at 5% confidence level). However, the
difference was only significant for total purchases, while the coefficients were not
statistically significant for local purchases and imports. Also, for the incentive participants,
the post-intervention slope was not significantly higher than the COVID-19 period, which
might indicate an early recovery prior to the PPnBM incentives. The early recovery for
purchase activity could be observed from the positive coefficients for the treated groups
during the pandemic, to which the import activity contributed. At the same time, during
the COVID-19 period, only the local purchases were significantly affected, while total
purchases and imports were relatively stable (coefficients were not statistically
significant). The early recovery trend was apparent in Figure 7, with total purchases for
the treated group experiencing a positive trend beginning in July 2021.
100
Measuring the total purchase recovery. This research assesses the treated firms’
recovery on total purchases using the ARIMA to set a projection model. Figure 8 below
suggests that the model robustly fits the 2017 – 2019 purchase data. Based on the visual
interpretation, the projected purchases during the treatment period were relatively higher
than participating firms’ actual purchases. The t-test result confirms the speculation: the
extrapolated total purchases had a relatively higher mean (difference of 0.1313),
significant at a 5% confidence level, i.e., Pr(|T| > |t|) = 0.0462 and Pr(T > t) = 0.0231.
Thus, this study would argue that the treated firms had yet to recover their total
purchases to the pre-pandemic level during the incentives period.
Figure 8. Projected vs Actual, Total Purchases for incentives participants (log-scale)
Impact on firms’ employment. The last parameter to measure firms’ economic

activity during the treatment period is employment, proxied with the total salary and wage
spending. This study incorporates salary payments based on the monthly payroll tax data
from 2018 to 2021, comparing the treated firms with non-participants. Figure 9 below
plots the monthly data for treated and control groups, which suggests: (1) the participants
of the incentive had a higher wage spending than the non-participants; (2) both groups
reported a relatively modest drop in salary and wages; (3) following the treatment period,
both groups experienced a relatively sustained level of salary spending. Therefore, it
would require empirical analysis to measure the coefficient and significance level.
The ITSA result confirms the less observable difference between the treated and
control groups. The post-intervention coefficient suggests a steeper slope for
participating firms but is not statistically significant at α = 10% (see Figure 3). For all other
parameters, except the treatment dummy, there are no statistically significant coefficients
– which could represent a relatively stable wage spending before and after the
intervention. One possible explanation might be that the employment-related strategies
in the sample firms were less prevalent during the pandemic. Based on the previous
parameters, incentives participants experienced a somewhat “v-shaped” recovery in
sales and purchases in the second semester of 2020. The uptake might contribute to
relatively stable wage spending. However, the interpretation would require further
101
scrutinization as the tax return data represent workers earning more than personal
income tax allowance.
Figure 9. Monthly Total Wage Spending (value, log-scale)
Measuring the total wages recovery. Figure 10 below plots the Holt-Winters
exponential smoothing forecast based on the treated firms’ monthly wages data from
2018 – 2019. A quick visual analysis would suggest that the actual wages were relatively
weaker than the extrapolated figures. However, given the relatively small magnitude, the
difference might not be statistically significant. The t-test result suggests that while the
mean difference is around 0.1787 for projected wages, it is not significant even at 10%
confidence level, i.e., Pr(|T| > |t|) = 0.3061 and Pr(T > t) = 0.1530. Therefore, this study
would argue that the incentive participants reached the pre-pandemic level of total wages
during the treatment period.
Figure 10. Projected vs Actual, Total Wages for incentives participants (log-scale)
5. CONCLUSION
Conclusion. During the pandemic, automobile firms were among the severely
impacted manufacturing sectors. The wholesales and total sales slashed significantly
during the pandemic for both treated and control groups (see Figure 3). While the
102
coefficient for total purchases during the pandemic was somewhat not statistically
significant, the visual representation suggests that both groups hit the lowest point in
2020. Only wage spending reported a relatively comparable level across the observation
period. Thus, following the ‘targeted’ principle, it might be admissible to implement the
PPnBM Incentives.
In 2021, the government implemented the PPnBM Incentives and disbursed around
IDR 4.6 trillion during the intervention period. However, the empirical analysis in this
research indicates a modest impact of the incentives. Based on the post-intervention
level, only on wholesale and total purchases were the treated groups significantly higher
than the control groups. Meanwhile, the recovery towards the pre-pandemic level is
limited, only observable for total sales. Figure 11 below summarizes the selected results
of wholesales, firms’ total sales, purchases, and wage spending.
Figure 11. Summary Result of Observed Parameters

Treated > Control For treated groups,post- For treated groups, post-intervention
Parameters
Post-intervention? intervention > pre? ≥ pre-pandemic level?
Wholesales Yes No No
Total Sales No No Yes
Total Purchases Yes No No
Salary & Wages No No No
Discussion. Based on the result and modest impact of the incentives, two identified
factors might contribute to the somewhat lackluster conclusions: the “pent-up demand”
and consumers’ decisions. Based on the “pent-up demand” hypothesis, people might be
motivated to postpone their decision on durable goods consumption during a crisis. As
the consumers postpone the purchase, consumption is expected to resume during the
economic recovery. In this study, such phenomena were visually apparent, especially for
the treated group. For example, although limited, the wholesales for incentives-eligible
vehicles began to uptake in the second semester of 2020. Arguably, the recovered trend
would distort the pre-intervention analysis to measure the impact of the PPnBM
Incentives. Therefore, it might be relevant to assess the timeliness of the PPnBM
Incentives in future research.
Also, this paper will argue that consumers’ decisions factor during the intervention
period affected the domestic automobile market: new car models and preference shifting
due to price effect. In 2021 and coincidence with the PPnBM Incentives, several
manufacturers introduced new car models, either refreshments (facelifts) or a new line-
up. For example, Toyota and Daihatsu introduced the new Rocky and Raize models in
2021, and both models are eligible for the incentives. Thus, a consumer’s decision to
buy this model might be motivated by personal preference, PPnBM incentives, or both.
The impact of the new model and refreshments might be significant, e.g., the Rocky and
Raize model contributed to around 10% of the wholesale of eligible small-sized 4X2
groups during the intervention period.
Lastly, given the comparability of several car models, the price effect might
motivate prospective consumers to shift their purchases towards eligible vehicles during
the incentives period. As an illustration, this study tracks the market share (based on the
wholesale) for Toyota Fortuner and Mitsubishi Pajero in the medium-sized 4X2 category
– of which the Fortuner models were eligible for incentives whilst the Pajero were
ineligible. On average, the Fortuner models took up around 48% of the medium-sized
4x2 category in 2019. However, from June to December 2021, the average distribution
for the Fortuner model in the vehicle category significantly increased to 66%. As a result,
103
the interpretation of PPnBM’s impact on automobile wholesale might be biased – a
consumption shift might occur for some comparable car models.
Figure 12. Wholesale distribution for Pajero and Fortuner (medium-sized 4X2
group)
6. RESEARCH IMPLICATION AND LIMITATIONS
APPENDIX
No Model No Model
1 Toyota Yaris 19 Honda Brio RS
2 Toyota Vios 20 Honda Mobilio
3 Toyota Sienta 21 Honda BR-V
4 Toyota Innova 2.0 22 Honda CRV 1.5T
5 Toyota Innova 2.4 23 Honda HR-V 1.5L
6 Toyota Fortuner 2.4 4X2 24 Honda HR-V 1.8 L
7 Toyota Fortuner 2.4 4X4 25 Honda CRV 2.0 CVT
8 Daihatsu Xenia 26 Honda City Hatchback
9 Toyota Avanza 27 Suzuki New Ertiga
10 Daihatsu Grand Max 28 Suzuki XL 7
11 Daihatsu Luxio 29 Wuling Confero
12 Daihatsu Terios 30 Wuling Formo
13 Toyota Rush 31 Toyota Veloz
14 Toyota Raize 32 Toyota Agya
15 Daihatsu Rocky 33 Toyota Cayla
16 Mitsubishi Xpander 34 Daihatsu Ayla
17 Mitsubishi Xpander Cross 35 Daihatsu Sigra
18 Nissan Livina 36 Honda Brio Satya
Appendix 1. Eligible Vehicles based on the Ministry of Industry Regulation
104
Predicted Actual Diff. Pr(T < t) Pr(|T| > |t|) Pr(T > t)
Parameters
(mean) (mean) Predicted - Actual diff. < 0 diff. ≠ 0 diff. > 0
Wholesales 10.6600 10.5793 0.0806 0.8406 0.3189 0.1594
Total Sales 30.5228 30.4442 0.0787 0.8975 0.2050 0.1025
Total Purchases 30.4698 30.3385 0.1313 0.9769 0.0462 0.0231
Salary & Wages 27.2417 27.0630 0.1787 0.8470 0.3061 0.1530
Appendix 2. Predicted vs Actual: t-test summary for all parameters
REFERENCES
Atkinson, R. D. (2008). Timely, Targeted, Temporary and Transformative: Crafting an
Innovation-Based Economic Stimulus Package. Washington, DC: The
Information Technology and Innovation Foundation.
Auerbach, A. J. & Feenberg, D. (2000). The Significance of Federal Taxes as Automatic
Stabilizers. Cambridge: National Bureau of Economic Research.
Badan Pusat Statisik (Statistics Indonesia). (2020). Analisis Hasil Survei Dampak
COVID-19 Terhadap Pelaku Usaha (Analysis of the Impact of the COVID-19 on
Businesses). Jakarta: Badan Pusat Statistik.
Badan Pusat Statistik (Statistics Indonesia). (2021). Laporan Perekonomian Indonesia
(Indonesia Economic Report). Jakarta: Badan Pusat Statistik.
Beraja, M. & Wolf, C. K. (2021). Demand Composition and The Strength of Recoveries.
Cambridge: National Bureau of Economic Research.
Caballero, R. J. (1993). Durable goods: An explanation for their slow adjustment. Journal
of Political Economy, 101(2), pp. 351-384.
Chetty, R. et al. (2020). How Did COVID-19 and Stabilization Policies Affect Spending
and Employment? A New Real-Time Economic Tracker Based on Private Data
Source. Cambridge: National Bureau of Economic Research.
Copeland, A. (2014). Intertemporal substitution and new car purchases. The RAND
Journal of Economics, 45(3 (Fall 2014)), pp. 624-644.
Elmendorf, D. W. & Furman, J. (2008). If, When, How: A Primer on Fiscal Stimulus.
Washington,DC: The Brookings Institution.
Hodbod, A., Hommes, C., Huber, S. J. & Salle, I. (2021). The COVID-19 consumption
game-changer: evidence from a large-scale multi-country survey. Frankfurt am
Main: European Central Bank.
Linden, A. (2015). Conducting interrupted time-series analysis for single- and multiple-
group comparisons. The Stata Journal, 15(2), pp. 480-500.
OECD. (2020). Tax and Fiscal Policy in Response to the Coronavirus Crisis:
Strengthening Confidence and Resilience. Paris: OECD.
Rosid, A., Sanjaya, T. B. & Ardin, G. (2022). The Economic Impact of Covid-19 Pandemic
on Businesses: The Case of Indonesia. Jurnal Anggaran dan Keuangan Negara
Indonesia, 4(1), pp. 86-109.
Summers, L. (2008). Fiscal Stimulus Issues Testimony before the Joint Economic
Committee. [Online] Available at: http://larrysummers.com/wp-
content/uploads/2012/10/1-16-08_Fiscal_Stimulus_Issues.pdf [Accessed 8
February 2022].
Taylor, J. E. & Castillo, A. (2015). “Timely, Targeted, and Temporary?” An Analysis of
Government Expansions over the Past Century. Virginia: Mercatus Center at
George Mason University.
The World Bank. (2020). How COVID-19 is affecting firms in Indonesia. Washington D.C:
The World Bank.
105
NATURAL LANGUAGE PROCESSING ANALYSIS OF
APPEAL DECISIONS IN TRANSFER PRICING TAX
DISPUTES IN INDONESIA
Ananda Anggara a, Reza Darmawan b
a
Inspectorate General of Ministry of Finance, Jakarta, Indonesia
Email: ananda.anggara@kemenkeu.go,id
b
Inspectorate General of Ministry of Finance, Jakarta, Indonesia
Email: reza.darmawan@kemenkeu.go,id
ABSTRACT
Due to the time, money, and data-intensive nature of litigation, legal science and practice have long been
interested in predicting court case outcomes. Natural language processing (NLP) has been introduced as
an option for processing human language in legal documents. Our goal is to determine, based on textual
evidence extracted from a case, specific passages from the parties involved focused on income tax and
value-added tax disputes. This paper is the first systematic study to use text mining to predict the outcome
of transfer pricing abuse cases before the Indonesian tax court. During the phase of topic modelling, we
developed several numbers of LDA topics pertaining to tax court judgment documents. The probability
distribution of each topic from each observation produced by LDA will be used as a textual feature to create
two predictive classification models using linear kernel Support Vector Machine (SVM) for value-added tax
and income tax scenarios. Due to the models' high accuracy, we could identify the most predictive subjects
for tax appeal verdicts using linear kernel SVM.
Keywords: transfer pricing, tax court, appeal verdict, Latent Dirichlet Allocation, Support Vector Machine
Sehubungan dengan kebutuhan akan waktu, uang, dan data yang tidak sedikit dalam penyelesaian kasus
di bidang hukum, praktisi maupun akademisi telah lama tertarik untuk memprediksi putusan pengadilan.
Natural language processing (NLP) muncul sebagai salah satu solusi dalam memproses bahasan tekstual
dalam dokumen hukum. Tujuan studi ini adalah untuk memperoleh bukti, berdasarkan sampel dokumen
tekstual, apakah keputusan pengadilan yang berfokus pada sengketa pajak penghasilan dan pajak
pertambahan nilai dapat diprediksi menggunakan teknologi informasi. Tulisan ini merupakan studi sistematis
pertama menggunakan text mining untuk memprediksi putusan kasus transfer pricing abuse di Indonesia.
Selama fase pemodelan topik, kami mengembangkan sejumlah topik LDA yang berkaitan dengan dokumen
putusan pengadilan pajak. Distribusi probabilitas setiap topik dari setiap pengamatan yang dihasilkan oleh
LDA digunakan sebagai fitur tekstual untuk membuat dua model klasifikasi prediktif menggunakan kernel
linear Support Vector Machine (SVM). Dari hasil studi, didapatkan akurasi model yang baik atas beberapa
fitur tekstual dengan unsur prediktif tinggi dalam memperkirakan putusan pengadilan, baik sengketa pajak
penghasilan dan pajak pertambahan nilai.
Kata kunci: transfer pricing, pengadilan pajak, putusan banding, Latent Dirichlet Allocation, Support Vector Machine
1. Introduction
Due to the litigation process's time, money, and data-intensive nature, legal
sciences and practice have long been interested in predicting the outcome or likelihood
of success in court cases. While investigating the potential legal applications of
information technology, Lawlor projected that in the future, Computers might be capable
of assessing and forecasting the results of court decisions (Lawlor, 1963). Artificial
intelligence technology is one of the current research solutions for the issue. Natural
language processing (NLP), a technique utilized in this study, is a form of artificial
intelligence that is particularly adept at processing human language in legal documents.
The automatic analysis of instances involving abuse of transfer pricing before the
Indonesian tax court is the primary topic of this essay. Tax Court is an institution of justice
106
that has the authority to resolve tax disputes for taxpayers or tax insurers. A taxpayer
who disagrees with the DGT Objection ruling may submit an Appeal (Banding) within
three months after receiving the DGT Objection. Usually, the Tax Court must make an
appeal decision within one year. Any tax owed as a consequence of the Court's decision
is subject to a 100 per cent surcharge.
Meanwhile, transfer pricing (TP) is the price set by taxpayers when they sell, buy,
or share resources with affiliates (Barker et al., 2017). From the perspective of the
definition given above, TP is comprehensible, proving that the process is a legitimate
industrial activity. However, because it is currently seen as a swamp for multinational
corporations to utilize tax evasion techniques, transfer pricing has a bad reputation.
Transfer pricing manipulation or abuse is an alleged fraudulent practice. In order to
circumvent government regulation, this manipulation is typically accomplished by
enacting a policy on transfer prices above or below the opportunity cost. This deviation
uses regional legislation variations, mainly cross-border, concerning tax rates. Shortly,
this scam uses markup or markdown to lower the amount of tax that must be paid.
According to the Tax Justice Network (2021), cross-border tax evasion could cost
Indonesia USD 2,274 million (IDR 33.75 trillion). To reduce the risk of transfer pricing
practices causing losses to the state revenue, the Government of Indonesia has
established a tax policy related to transfer pricing practices as regulated in Article 18 of
the Income Tax Law and its implementing provisions, including provisions on transfer
pricing inspections. However, taxpayers frequently challenged the DGT's Tax
Assessment Letters issued due to tax audits involving transfer pricing. According to the
2020 DGT Performance Report, the DGT's loss rate in the 2020 conflict was 56.9 per
cent. Meanwhile, the DGT would have to court over 8,664 tax disputes in 2020. The vast
majority of cases—3,634 in total—are still decided in what is known as a "full grant."
Our goal is to analyze, based on textual evidence taken from a case, including
particular sections of the parties' arguments, if enterprises accused of participating in
transfer pricing abuse have violated a specific article controlling income tax and value-
added tax. Subsequently, This is the first systematic study to predict the result of cases
involving transfer pricing abuse before the Indonesian tax court by mining existing text
data.
Overall, we think constructing a text-based predicting system of court judgments
may provide attorneys and judges with a helpful aiding tool. The technique may detect
situations swiftly and extract patterns that connect with specific results. It may also be
utilized to generate previous indications for diagnosing suspected breaches of particular
articles in submitted applications and, ultimately, to prioritize decision-making in
circumstances where a violation seems rather probable. On the other hand, we also wish
to strike a fair balance between the pursuit of justice and the financial risks associated
with legal proceedings by supporting taxpayers in anticipating the outcome of an appeal.
2. DESIGN AND HYPOTHESIS

2.1. Indonesia Judicial Procedures: Tax Courts and Appeal Verdict
Following the formal requirements, an appeal must be submitted no later than three
(three) months after the taxpayer receives the objection letter. In addition, the Tax Court
will provide a copy of the appeal letter to the Appellant (DGT).
In this instance, the Tax Court requests the DGT to respond as an Appeal
Explanation Letter. A copy will be provided to the taxpayer by the Tax Court. The
taxpayer, as the appellant, can then reply to a letter known as a Rebuttal Letter. In some
instances, the appellant only submits the Appeal Letter to the Tax Court, or the opposing
party does not respond to the Appeal Explanation Letter on the appellant's appeal, so
the above procedure is not always followed. After completion of the process, the trial is
held. Typically, the trial begins with the Court examining the formal requirements for filing
107
an appeal. The appeal dispute material will be examined if the formal provisions have
been met.
Figure 1 Summary of the Appeal Implementation Process

Source: The Tax Disputes and Litigation Review (Whitehead, 2015)
Accordingly, material objects are discussed multiple times. In this trial, tax dispute-
related evidence will be discussed. The Tax Court conducts the hearing for the reading
of the Appeal Verdict in the presence of the Appellant and Appellant party. The Tax Court
will read the appeal verdict at trial, nevertheless, at its discretion.
2.2. Literature Review

As long as there are judges, there will be case law, and as long as there is case
law, legal academics will analyze it. For centuries, legal scholars used doctrinal research
methods, which include not only describing laws, practical problem solving, and adding
interpretative comments to legislation and case law, but also 'innovative theory building
(systematization), with the more simple versions of that research serving as necessary
building blocks for the more sophisticated ones' (Van Hoecke, 2011; Medvedeva et al.,
2019).
Waltl et al. (2019) explain the early method of legal prediction, which is the nearest
neighbour method and the Issue-Based Prediction Model. One of the oldest methods for
making predictions was the nearest neighbour method, in which the examples closest to
a problem are identified using similarity measurements and an outcome is assigned to
most of those cases. Popple, in 1996, using the nearest neighbour algorithm, added
more complexity to the similarity measures by assigning weights to different fact
descriptors. The nearest neighbour method, in our opinion, is restricted by its definition
to the area of distinguishable neighbours and does not permit accurate forecasts outside
of this area. Furthermore, Waltl et al. (2019) also explain The IBP (Issue-Based
Prediction Model), which blends case-based reasoning with a model of abstract legal
concerns. The IBP Domain Model's tight relationship to case law is evident when legal
issues and relationships are "a distillation and interpretation of two authoritative sources
on trade secret misappropriation (a statute and a Restatement provision)." This model's
topic identification is labour- and knowledge-intensive, and it must be done from scratch
for any other area of law, preventing the building of a general prediction model.
Recent research utilized data from the European Court of Human Rights as an
example and examined how natural language processing technology may be used to
analyze court proceedings documents in order to automatically predict (future) judicial
decisions (Aletras et al., 2016; Panagis et al., 2016; Medvedeva et al., 2019). Aletras et
al. (2016) utilized one of the early natural language processing (NLP) algorithms to
108
forecast European Court of Human Rights court rulings. This research attempts to
construct a predictive model to identify a case in the European Court of Human Rights
and to extract patterns that lead to particular decisions. The data consists of cases on
articles 3, 6, and 8 of the convention. The model accurately predicts 79 per cent of court
decisions using the linear kernel SVM classification. Meanwhile, Medvedeva (2019) get
an average accuracy of 75% in predicting the violation of 9 European Convention on
Human Rights articles. Panagis et al. (2016) used topic modelling techniques to
automatically find latent topics in the decisions of the Court of Justice of the European
Union and the ECtHR.
Katz et al. (2017) conducted a study on a generic method of predicting the
behaviour of the United States Supreme Court using a random forest; accuracy values
of 70.2% at the case result level and 71.9% at the fairness vote level were determined.
Furthermore, Gruginskie and Vaccaro (2018) conducted a study predicting the time of
lawsuits. In this study, four main approaches were used for lead time classification,
namely SVM, nave Bayes (NB), random forest (RF), and neural network (NN). The
findings of this investigation indicate that the SVM approach was chosen because of its
accuracy for modelling complex non-linear decision boundaries, reducing the tendency
of overfitting better than other methods, and having the highest accuracy value of 77.58%.
2.3. Hypothesis
The primary basis of our ideas is that the textual content and other case elements
significantly impact the Court's decision. Based on the premise that language taken from
published decisions of the Court contains a significant number of similarities with
applications made to the Court and briefs presented by parties in litigation matters, it may
serve as a (rough) proxy for them.
3. METHODOLOGY
3.1. Approach
This part summarises the actions taken inside our strategy, which adheres to a
traditional machine learning strategy separated between training and testing methods.
In the training process, the training data will go through three processes, namely the
preprocessing process, Latent Dirichlet Allocation (LDA) analysis, and Support Vector
Machine (SVM) classification. The preprocessing process includes Cleansing and
Stopwords Removal. Using LDA, we generated several topics regarding the tax court
decision documents at the topic modelling stage. The probability distribution of every
topic from each observation produced by LDA will serve as a textual feature to develop
a predictive classification model.
Preprocessing LDA Topic Feature Training &

Evaluation
Data Modelling Extraction Testing
Figure 2 Research Step
After feature extraction, an SVM classification model is developed. The model is

created by separating the data into training and testing sets. SVM will use the training
set to construct a model that identifies a separation function (hyperlane) between classes.
The model will next be evaluated through data testing to determine the amount of
accuracy with which it can forecast classes or categories based on unique new data. In
developing this predictive model, relevant features extracted from the tax court textual
data, as a predictive variable, will be used to predict whether an appeal will be successful
or not.
109
According to research, SVM is considered a machine learning approach with
promising results in text classification, particularly when employing small data sets
(Joachims, 2002; Wang & Manning, 2012). We use a linear kernel because it allows us
to discover critical characteristics indicative of each class by reviewing the learned
weights for each feature (Chang & Lin, 2008). In Aletras (2016), the symbol -1 denotes
cases in which the judge grants the taxpayers' appeal. The cases in which the judges
denied the taxpayers' appeal are denoted by +1. Consequently, factors with positive
weights imply that the government succeeded in the appeal proceedings, whereas those
with negative weights indicate that taxpayers prevailed. The models are trained and
evaluated using stratified 5-fold cross-validation, with 30% of the data retained at each
stage to evaluate expected performance. A stratified shuffle split was implemented in
response to the class disparity.
3.2. Data
We compile examples of transfer pricing on income and value-added tax into a
data collection. We focus mainly on these circumstances for two reasons. First of all, we
were able to learn the most from these articles. Second, both comprise the most crucial
issue, accounting for 81 per cent of state revenue from taxes. The sample of 84 consists
of 66 documents regarding value-added tax and 18 documents regarding income tax.
3.3. Data Extraction

We extracted the textual data from the portable document format of the tax court's
ruling using the Python Py2PDF module. The statement of the government, the
statement of the taxpayer, and the statement of the Court are extracted. The government
and taxpayer statements summarize the application and respondent state's key
arguments. This section almost entirely concentrates on the parties' legal arguments
since, in most cases, domestic courts have authoritatively established the relevant facts.
The dispute's subject, though, has not altered. The court statement offers the legal
arguments that are supposed to support the decision made by the Court. Usually, the
Court integrates its conclusions within a larger body of already-established laws,
principles, and doctrines. This paragraph is assumed to contain nearly solely legal
arguments, occasionally interspersed with factual facts from other sections.
4. RESULTS AND ANALYSIS

4.1. Preprocessing Data
Before stepping onto the research analysis, several preprocessing steps must be
conducted. The first preprocessing is to remove stopwords. Stopwords are obtained from
the Indonesian language nltk python library with case-specific customization. The main
customized stopwords are words always present in almost all documents, so their
presence does not add value to the analysis results. The subsequent preprocessing step
removes non-ASCII characters, URLs, punctuation, numbers, and excess white space
using the regular expression library. This preprocessing aims to produce a list of words
to be included in the dictionary.
4.2. LDA Topic Modelling

The next step is to create bigrams and trigrams. Bigrams are two words that often
appear together in documents and are selected for only bigrams that appear ten times
or more. Trigrams are three words that appear frequently. Bigrams and trigrams
produced will also be included in the dictionary. After generating bigrams and trigrams,
we produce a dictionary using a list of text, including the frequency with which each word
appears in the training set. Next, A corpus-a mapping of (word id, word frequency) is
produced by Gensim by assigning a unique id to each word, bigram, and trigram in the
110
text. For each document, gensim doc2bow will create a dictionary that contains how
many words there are and how many times they appear in the document. The coherence
score is used in the topic selection process. A coherence score measures topic modelling,
and a successful model will generate topics with high coherence scores. The coherence
score is calculated using 'c_v' metrics which shows that the greater the coherence score,
the better the interpretation of the topic modelling results.
Table 17 Coherence Score Measurement Output

Corporate Income-Tax Value-Added Tax
Components Number of Coherence Number of Coherence
Topics Score Topics Score
Government Statement 3 0.508467 5 0.367104
Taxpayer Statement 7 0.45852 5 0.480189
Court Statement 3 0.403887 3 0.541248
Based on the dictionary and corpus produced and the number of topics decided,
we generate an LDA model to analyze the topics of the texts to be used as textual
features. Here is the detail of the generated topics and their top relevant keywords.
4.3. Textual Feature Analysis - Income Tax

The following table shows the topic generated from the LDA model regarding tax
court cases in Corporate Income Tax.
Table 18 Topic Generated From LDA Model Regarding Income Tax

Topic Interpretation Relevant Keywords
(Bahasa - English)
Corporate Income Tax – Taxpayer Statement
Topic 0 analisa kewajaran biaya, know, kuartil, rentang, komponen, aktiva, rugi, penerapan,
komponen penghasilan dan penjualan, fee, pengujian, memperhitungkan, dibandingkan, tertimbang,
biaya usaha sewa, scrap, penyesuaianpenyesuaian, pendapatan, kurs, expenses,
(resonableness analysis of dicatat, ncpm, sependapat, prosedur, principle, komersial, enterprises,
the income and operating pembanding, analisa, pemberian
expenses components)
Topic 1 Perjanjian pinjaman antar bunga, pif, pinjaman, saham, bank, biaya, tagihan, putusan, perjanjian,
perusahaan agreement, loan, pkpu, fee, hak, penyaluran, status, anggaran, annual,
(Intercompany Loan kewajiban, utang, dilaporkan, tppi, memperoleh, fakta, pelabuhan,
Agreement) kreditur, erat, sah, debitur, akun
Topic 2 beban komisi penjualan komisi_penjualan, penjualan, biaya, pofsb, komisi, harga, grup,
(sales commission pemasaran, pembeli, sales, pasar, kontrak, dibebankan, sawit, marketing,
expenses) barang, struktur, manajemen, pembayaran, manfaat, dunia, mencari,
pengiriman, pernyataan, menjalankan, operasional, service, cpo,
ekonomis, plus
Topic 3 pembayaran kompensasi merek, denda, pidana, penjara, diperdagangkan, barang, pembayaran,
merek luas, bisnis, membayar, milyar, sengaja, sejenis, referensi, pemakai,
(payment of trademark barangnya, terdaftar, dipidana, diproduksi, kompensasi, konsekuensi,
compensation) konsekuensinya, mencoba, menerima, menggariskan, milik, dunia,
mencari, hak, harga
Topic 4 Jasa asistensi teknis benang, saran, tenaga, ahli, baku, rekomendasi, bahan, mesin, produksi,
(Technical Assistance Fee) pelanggan, jual, pemberian, ekspor, sumber, pengolahan, teknis,
pengembangan, celup, mengidentifikasi, parameter, tender, menyeleksi,
perencanaan, alat, media, berbagi, staf, berhubungan, pintal, fiber
Topic 5 jasa manajemen yang manajemen, lahan, pembebanan, hsf, pupuk, plasma, ctp, mark, marjin,
dihitung berdasar luasan biaya, dihitung, peredaran, pembanding, luasan, fungsional, pembelian,
lahan pembiayaan, harga, pembayaran, documentation, tnmm, royalti, setuju,
(management fee payable inti, timbul, mengabulkan, kebun, bruto, pengurang, berdasar
based on land area)
Topic 6 analisa customization
wilayah, spesifik, langsung, pasar, dipengaruhi, manufaktur, analisa,
produk wilayah pemasaran dihasilkan, membutuhkan, sisi, technology, produknya, pangsa, industries,
(market-area product
exchange, teknikal, global, customization, dibutuhkan, party,
customization analysis ) circumstances, bersaing, berkualitas, berbeda, affiliated, kesesuaian,
economic, jenis, uang, masuk
Corporate Income Tax – Government Statement
111
Topic 0 pengujian pembebanan Penjualan, biaya, harga, manajemen, sales, pembayaran, dibebankan,
royalti manfaat, tertulis, kredit, pembebanan, saham, neto, argumentasi,
(royalty-fee charging test) management, informasi, eksistensi, royalti, pk, lahan, lazim, dibayar,
dividen, dibayarkan, ketetapan, bayar, rugi, fee, analisa, pengujian
Topic 1 pembebanan biaya bunga bunga, biaya, perjanjian, kredit, dibebankan, pk, piutang, diperoleh, uang,
(interest-fee charging test) mata, menagih, memelihara, informasi, pengujian, pembebanan, kondisi,
singapura, total, rupiah, rincian, keuntungan, dibayarkan, saham,
pendukung, bruto, sub, memenuhi, pelaporan, mempertahankan, lazim
Topic 2 Analisis kesebandingan operating, margin, pembanding, documentation, peredaran, rentang, fee,
operating margin tagihan, invoice, tnmm, pembayaran, analisis, mempertahankan, analisa,
(operating margin sales, operasi, sejenis, diserahkan, uraian, diluar, lazim, pelaporan,
comparability analysis) diperoleh, kondisi, penentuan, bruto, dijalankan, setuju, produksi,
pengujian
Corporate Income Tax – Court Statement
Topic 0 Eksistensi royalti dan royalti, pembayaran_royalti, merek_dagang, manfaat_ekonomis,
technical assistance biaya_royalti, technical_assistance, merek, cup, manfaat, dagang, lisensi,
(royalty and technical intangible, trademark, barang, aktiva, produksi, paten, property,
assistance existence) perbandingan_harga, cost_plus, eksistensi, dividen, assistance, ahli,
trade, fee, berwujud, technical, transactional_margin, sales
Topic 1 Tested party analisa operating_margin, profit_indicator, operasi, analisa_kesebandingan,
kesebandingan profit, transactional_margin, tested_party, technology, analisa, expenses,
(comparability analysis of komponen, rentang, information, perbedaan, cost_plus, peredaran, party,
tested party) indicator, exchange, report, documentation, memperhitungkan, diterima,
company, kesebandingan, enterprise, tnmm, pengujian, entitas, non
Topic 2 Pengujian Biaya bunga dan biaya_bunga, bunga, know, saham, pinjaman, tested_party, pembelian,
know how modal, cost_plus, fee, kerugian, rentang, transactional_margin,
(testing of interest and know- perbandingan_harga, cost, bruto, pemberitahuan, bahan_baku, bank,
how fee) corporation, dividen, kuartil, tnmm, analisis, plus, kontrak, pengeluaran,
rugi, neto, persen
4.4. Distribution of Topic with Regards to Court Decision – Income Tax Cases
4,5
4
4
3,5
3 3
3
2,5
2 0
2 1
1,5
1 1 1 1 1 1
1
0,5
0
0 1 2 3 4 5 6
Graphic 1 Distribution of Taxpayer Statement Topic with Regards to Court

Decision in Corporate Income Tax Cases
At a glance at graphic 1, it is visible that in most observation (4 cases) regarding

the Corporate Income Tax cases, the government lose at court appeals when the topic
of taxpayer statement is topic 4, which is labelled as Technical Assistance Fee, i.e.
dispute regarding technical assistance fees for planning related to all appellant activities
without limitation.
112
6
5 5
5
3 3
3 0
1
2
2
0
0 1 2
Graphic 2 Distribution of Government Statement Topic with Regards to Court

Decision in Corporate Income Tax Cases
From graphic 2, It can also be inferred that in observation regarding the Corporate
Income Tax cases, the government lost at court appeals when the topic of government
statement is topics 0 and 2, which is labelled as pengujian pembebanan royalti (royalty-
fee charging test) and analisis kesebandingan operating margin (operating margin
comparability analysis). The example for topic 0 is while the Appellate accuse that the
appellant cannot explain with sufficient and convincing evidence regarding the existence
and benefits received. Meanwhile, the topic 2 example is the dispute of analysis of the
fairness which the appellant has been carried out. It is known that the TNMM of the
Appellant is still far from the lower limit of the reasonable profit range.
8
7
7
5
4 4
4 0
1
3
2
1 1 1
1
0
0 1 2
Graphic 3 Distribution of Court Statement Topic with Regards to Court Decision

in Corporate Income Case
Meanwhile, the takeaway from graphic 3 is topic 2, labelled as pengujian biaya

bunga dan know-how (testing of interest and know-how fee). The example for this case
113
is while the Appellate accuse that the appellant cannot explain with sufficient and
convincing evidence regarding the existence and benefits received.
4.5. Textual Feature Analysis - Value-Added Tax

The following table shows the topic generated from the LDA model regarding tax
court cases in Value-Added Tax.
Table 19 The Topic Generated From LDA Model Regarding Value-Added Tax
(Bahasa - English)
Value-Added Tax – Government Statement
Topic jasa mengorganisir, pembayaran, pemanfaatan, biaya, memenuhi, pelaksanaan, objek, kliennya,
0 mencarikan, mengatur mengorganisir, kiliennya, mencarikan, pembahasan, mewah, dibayar, berbeda,
kontrak klien informasi, pemberian, kontrak, orang, alat, mengatur, minyak, menerbitkan, dijual,
(organizing, risalah, langsung, audit, bangunan, membuktikan, dilaporkan, mengajukan
procurement, and
mangement of client
contract services)
Topic memasukan searching mewah_mempertahankan, penyerahan_nilainya, mewah, dipungut, nilainya,
1 criteria mempertahankan, mewah_mempertahankan_penyerahan_nilainya, jenis, objek,
(searching criteria input) ditolak, searching_criteria, criteria, searching, kompeten, berhubungan, pasar,
setuju, pelaksanaan, pekerjaan, perundangan, pekerjaan_kebun,
berhubungan_pelaksanaan, tanggapan, kliennya, ketetapan, manufacture,
memasukan, wholesale, base, information
Topic Pajak Masukan masukan, tbs, cpo_pk, unit_kebun, bersifat_strategis, dibebaskan, menghasilkan,
2 produksi TBS Sawit, unit_pabrik, bahan, unit, sawit, dikreditkan, kebun, persetujuan_kepala,
CPO dan PK tandan_buah, strategis, bersifat, penyerahannya_dibebaskan, kelapa_sawit,
(Input VAT for Palm FFB bibit_pupuk, cpo, pabrik, pk, persetujuan, bkp, mewah, tbs_sawit, memproduksi,
CPO, and Output VAT) perolehan, pma
Topic Perjanjian/Kontrak pembelian, harga, peredaran, penelitian, dilaporkan, jual, kredit, perjanjian,
3 ekspor distributor, saham_komisaris, pegawai, saham, harga_jual, komisaris,
(export penggantian, peb, mencari, aktivitas, setuju, ekspor, diperoleh, dihitung, tarif,
agreement/contract) perjanjian_kontrak, dipertahankan, kontrak, tahunan, akuntan, impor,
membuktikan
Topic equal treatment Pajak masukan, kelapa_sawit, tbs, penyerahannya_dibebaskan, bkp, dibebaskan,
4 Masukan penyerahan dikreditkan, menghasilkan, kelapa, sawit, penyerahannya, integrated, terpadu,
TBS kategori, bersifat_strategis, jkp, berwujud, pemanfaatan, mempertahankan,
(equal treatment of input strategis, rangka, perlakuan, equal, treatment, terpadu_integrated, non,
VAT regarding Palm dikreditkannya, berdampak, dapatnya, bersifat
FFB transfer)
Value-Added Tax – Taxpayer Statement
Topic Pajak Masukan a quo barang, pembeli, masukan, pemanfaatan, pk, quo, pembayaran, oil_cpo, cpo,
0 penyerahan CPO kredit, dikenakan, setuju, analisis, perkara, pendukung, usahanya,
(Input VAT a quo penyerahannya_dibebaskan, penyerahannya, inti, menghasilkan, cpo_pk,
transfer of CPO) crude_palm_oil_cpo, perolehan, crude, oil, crude_palm, dibebaskan, sawit,
mengikuti, ekspor
Topic titip olah TBS menjadi olah, cpo, distributor, tbs, setuju, titip, titip_olah, masukan, palm, oil_cpo, pk,
1 CPO diolah, melandaskan, bentuk, koreksinya, mengatur, barang, perkara,
((toll manufacturing of crude_palm_oil_cpo, oil, crude, crude_palm, cpo_pk, dibebaskan, buah, tandan,
ffb to cpo)) segar, tandan_buah, baku, bahan
Topic Criteria pembanding biaya, pembanding, tax, ekspor, konsisten, penggantian, information, electronics,
2 (comparation criteria) technology, technology_pembanding, electronics_information, mark, barang,
pembayaran, mengenakan, tarif, kecuali, kecuali_tergantung, tergantung,
penyelesaian, criteria, margin, dibatalkan, pendekatan, menerima, pemilihan,
pembanding_konsisten, operating, rentang, memilih
Topic Pajak masukan Crude masukan, nilainya_dipungut, pembelian, kelapa_sawit, barang, dikreditkan,
3 Palm Oil dan Palm penjualan, masukan_dikreditkan, sawit, nilainya, kelapa, tbs, palm, perkebunan,
Kernel oil_cpo, oil, cpo, dilaporkan, penghitungan, crude_palm, lokal, dibebaskan,
(Input VAT of CPO and pemakaian, biaya, menghasilkan, crude, buku, kernel, palm_kernel, perolehan
Palm Kernel)
Topic Faktur Pajak Masukan masukan, faktur, pembelian, langsung, dikreditkan, konstruksi, industri, bayar,
4 dan Bukti kontrak menerbitkan, tujuan, masukan_dikreditkan, maksud, tagihan, membayar,
(Input VAT invoice andpenjualan, setuju, perolehan, kontrak, komponen, nama, keluaran, pabrik,
contract document) pembayaran, penggunaan, pembahasan, entitas, total, mengikuti, pemanfaatan,
dijual
Value-Added Tax – Court Statement
114
(Bahasa - English)
Topic titip olah TBS menjadi tbs, masukan, cpo_pk, titip_olah, cpo, olah, titip, harga_jual, dikreditkan, pk,
0 CPO pengkreditan_masukan, dibebaskan, penyerahannya_dibebaskan,
(toll manufacturing of ffb tujuan_produktif, pemakaian, cpo_pko, penyerahannya, pemakaian_pemberian,
to cpo) sawit, pabrik, biaya, perolehan, menghasilkan, distributor, palm, perjanjian, jual,
kebun, dibayar_perolehan, dihasilkan
Topic Criteria pembanding pembanding, biaya, debtor_biaya, electronics_information, operating_margin,
1 (comparation criteria) electronics, margin, display_devices, debtor, peredaran, sebanding, operating,
technology, information, crystal, display, devices, electronic,
dijadikan_pembanding, masukan, extreme_results, mne_group, pengujian,
pembanding_memproduksi, components, group, setuju, independent_enterprise,
entitas, pengujian_financial
Topic jasa mengorganisir, penerima_diluar, klien_mencari, diterbitkan_catatan, vertikal,
2 mencarikan, mengatur penggantian_dibayarkan, vat, berkas_dipungut, berwujud_istilah,
kontrak klien khusus_mengatur, agen_perantara, penilaian_objek, jkp_bkp,
(services organizing, memenuhi_persyaratan, jkp, pembuktian_perundangundangan, dikreditkan,
finding, managing client bkp_berwujud, diluar, fakta_terungkap, wilayah_negara, seri_acuan,
contracts) kepastian_keadilan, faktor, pemasok_klien, quo_dikenakan, pemeriksaannya,
satuan_organisasi, perantara, keyakinan_memori, dikonsumsi_dimanfaatkan
4.6. Distribution of Topic with Regards to Court Decision - VAT Cases
35
30
30
25
20
16 0
15 1
10
5
5 4
3 3 3
2
0
0 1 2 3 4
Graphic 4 Distribution of Taxpayer Statement Topic with Regards to Court

Decision in VAT Cases
At a glance from graphic 4, it is visible that in most observation (30 cases)

regarding the Value-Added Tax cases, the government lose at court appeals when the
topic of taxpayer statement is topic 3, which is labelled as pajak masukan palm kernel
dan crude palm oil (input tax of palm kernel and crude palm oil). The example for this
case is the dispute of Input Tax that the appellant has paid for purchasing fertilizer and
maintenance of the garden that produces Fresh Fruit Bunch (FFB). However, the
appellant does not submit the FFB but sells palm kernel and crude palm oil.
115
25
22
20
15
0
10
10 9 9 1
7
5
5 3
1
0
0 1 2 3 4
Graphic 5 Distribution of Government Statement Topic with Regards to Court

Decision in VAT Cases
From graphic 5, It can also be inferred that in most observation (22 cases)
topic of the government statement is topic 1, which is labelled as memasukan searching
criteria (searching criteria input). The example for this case is while appellate accusation
that the appellant does not use consistent searching criteria based on Osiris data.
35
31
30
25
20 19
0
15 1
10
6
5 5
5
0
0 1 2
Graphic 6 Distribution of Court Statement Topic with Regards to Court Decision

in VAT Cases
Meanwhile, in graphic 6, it is also worth noting that in most observation (31 cases)
topic of court statement is topic 2, which is labelled as jasa mengorganisir, mencarikan,
mengatur kontrak klien (services organizing, finding, managing client contracts). The
example for this case is while appellate accusation that the appellant organizes its clients
to find third parties to provide services to its clients and make payments to third parties,
which fall under the vat rules.
116
4.7. Classification Model
The last step of this study is to predict the topic of each observation, including the
government statement, taxpayer statement, and court statement. The probability
distribution of every topic from each observation will serve as a textual feature to develop
a predictive classification model for the tax court decision.
The problem of predicting the tax court decisions related to transfer pricing cases
is categorized as a binary classification task. We develop our model with a single goal to
predict if, in a particular case, the judge will grant the taxpayers' appeal regarding transfer
pricing cases related to corporate income tax and value-added tax. For that purpose, we
employ the topics generated from the government statement, taxpayer appeal statement,
and the court ruling statement as textual features to train Support Vector Machine (SVM)
classifiers (Vapnik, 1998).
An evaluation is then conducted for each model developed. Court case prediction
of the income tax model has 100% and 80% accuracy for the value-added tax model.
However, we might expect a high accuracy in the income tax prediction due to the limited
number of observations.
Table 20 Precision, Recall And F-Score Per Model

Model Accuracy Class Precision Recall F1-Score Support
-1 (Government Lose) 1.00 1.00 1.00 4
Corporate Income Tax 100%
1 (Taxpayer Lose) 1.00 1.00 1.00 2
-1 (Government Lose) 0.93 0.82 0.87 17
Value-Added Tax 80%
1 (Taxpayer Lose) 0.40 0.67 0.50 3
As for the model on value-added tax, we could see that despite the high accuracy,
this model has a low score of precision and recall when it comes to predicting cases in
which the taxpayers lose or has their appeals to the tax court not granted. This issue
might be due to the limited number of cases in which the government win in the tax court.
The problems provide more evidence for this argument and show how practical
NLP is. To identify the most important factors for predicting whether a court judgment
would benefit the government or the taxpayer, the linear kernel of the SVM model may
be used. The six subjects with the highest positive and negative SVM weights for the
corporate income tax and value-added tax are shown in Tables 5 and 6, respectively.
Topics indicate factual patterning that corresponds to general tendencies in court case
law.
Table 21 The Most Predictive Topics For Corporate Income Tax Decisions *)
Topic Label Words W
Cases Where Taxpayers Lose
Government pembebanan biaya bunga, biaya, perjanjian, kredit, dibebankan, pk, 1.68211385
statement topic 1 bunga piutang, diperoleh, uang, mata, menagih, memelihara,
(interest charge) informasi, pengujian, pembebanan, kondisi,
singapura, total, rupiah, rincian, keuntungan,
dibayarkan, saham, pendukung, bruto, sub,
memenuhi, pelaporan, mempertahankan, lazim
Court statement tested party analisa operating_margin, profit_indicator, operasi, 0.80454798
topic 1 kesebandingan analisa_kesebandingan, profit, transactional_margin,
(comparability analysis tested_party, technology, analisa, expenses,
tested party) komponen, rentang, information, perbedaan,
cost_plus, peredaran, party, indicator, exchange,
report, documentation, memperhitungkan, diterima,
company, kesebandingan, enterprise, tnmm,
pengujian, entitas, non
Taxpayer intercompany loan bunga, pif, pinjaman, saham, bank, biaya, tagihan, 0.54614943
statement topic 1 agreement putusan, perjanjian, agreement, loan, pkpu, fee, hak,
penyaluran, status, anggaran, annual, kewajiban,
utang, dilaporkan, tppi, memperoleh, fakta, pelabuhan,
kreditur, erat, sah, debitur, akun
Cases Where Government Lose
117
Topic Label Words W
Government Analisis operating, margin, pembanding, documentation, -0.45490168
statement topic 2 kesebandingan peredaran, rentang, fee, tagihan, invoice, tnmm,
operating margin pembayaran, analisis, mempertahankan, analisa,
(operating margin sales, operasi, sejenis, diserahkan, uraian, diluar,
comparability analysis) lazim, pelaporan, diperoleh, kondisi, penentuan, bruto,
dijalankan, setuju, produksi, pengujian
Taxpayer jasa manajemen manajemen, lahan, pembebanan, hsf, pupuk, plasma, -0.82825774
statement topic 5 berdasar luasan lahan ctp, mark, marjin, biaya, dihitung, peredaran,
(management services pembanding, luasan, fungsional, pembelian,
based on land area) pembiayaan, harga, pembayaran, documentation,
tnmm, royalti, setuju, inti, timbul, mengabulkan, kebun,
bruto, pengurang, berdasar
Court statement pengujian biaya bunga biaya_bunga, bunga, know, saham, pinjaman, -0.96633566
topic 2 dan know how tested_party, pembelian, modal, cost_plus, fee,
(testing of interest and kerugian, rentang, transactional_margin,
know-how fee) perbandingan_harga, cost, bruto, pemberitahuan,
bahan_baku, bank, corporation, dividen, kuartil, tnmm,
analisis, plus, kontrak, pengeluaran, rugi, neto, persen
*) Listed in descending order by their SVM weight are the most predictive subjects for transfer pricing situations involving
the Corporate Income Tax, represented by the most common terms. Topic labels are inserted manually. Positive weights
(w) suggest more predictive themes for when the taxpayer won in Court, whereas negative weights denote more predictive
topics for when the government lost.
Table 22 The Most Predictive Topics For Value-Added Tax Decisions

Topic Label Words W
Cases Where Taxpayers Lose
Government Perjanjian/Kontrak ekspor pembelian, harga, peredaran, penelitian, dilaporkan, 1.23695821
statement topic (export jual, kredit, perjanjian, distributor, saham_komisaris,
3 agreement/contract) pegawai, saham, harga_jual, komisaris, penggantian,
peb, mencari, aktivitas, setuju, ekspor, diperoleh,
dihitung, tarif, perjanjian_kontrak, dipertahankan,
kontrak, tahunan, akuntan, impor, membuktikan
Taxpayer titip olah tbs olah, cpo, distributor, tbs, setuju, titip, titip_olah, 1.23247106
statement topic menjadi cpo masukan, palm, oil_cpo, pk, diolah, melandaskan,
1 (toll manufacturing of ffb bentuk, koreksinya, mengatur, barang, perkara,
to cpo) crude_palm_oil_cpo, oil, crude, crude_palm, cpo_pk,
dibebaskan, buah, tandan, segar, tandan_buah, baku,
bahan
Court criteria pembanding pembanding, biaya, debtor_biaya, 1.12333047
statement topic (comparison criteria) electronics_information, operating_margin,
1 electronics, margin, display_devices, debtor,
peredaran, sebanding, operating, technology,
information, crystal, display, devices, electronic,
dijadikan_pembanding, masukan, extreme_results,
mne_group, pengujian, pembanding_memproduksi,
components, group, setuju, independent_enterprise,
entitas, pengujian_financial
Cases Where Government Lose
Court jasa mengorganisir, penerima_diluar, klien_mencari, diterbitkan_catatan, -0.71664508
statement topic mencarikan, mengatur vertikal, penggantian_dibayarkan, vat,
2 kontrak klien berkas_dipungut, berwujud_istilah, khusus_mengatur,
(services organizing, agen_perantara, penilaian_objek, jkp_bkp,
finding, managing client memenuhi_persyaratan, jkp,
contracts) pembuktian_perundangundangan, dikreditkan,
bkp_berwujud, diluar, fakta_terungkap,
wilayah_negara, seri_acuan, kepastian_keadilan,
faktor, pemasok_klien, quo_dikenakan,
pemeriksaannya, satuan_organisasi, perantara,
keyakinan_memori, dikonsumsi_dimanfaatkan
Taxpayer pajak masukan crude masukan, nilainya_dipungut, pembelian, -0.96469027
statement topic palm oil dan palm kernel kelapa_sawit, barang, dikreditkan, penjualan,
3 (Input VAT of CPO and masukan_dikreditkan, sawit, nilainya, kelapa, tbs,
Palm Kernel) palm, perkebunan, oil_cpo, oil, cpo, dilaporkan,
penghitungan, crude_palm, lokal, dibebaskan,
pemakaian, biaya, menghasilkan, crude, buku, kernel,
palm_kernel, perolehan
Government jasa mengorganisir, pembayaran, pemanfaatan, biaya, memenuhi, -1.54259323
statement topic mencarikan, mengatur pelaksanaan, objek, kliennya, mengorganisir,
0 kontrak klien kiliennya, mencarikan, pembahasan, mewah, dibayar,
118
(services organizing, berbeda, informasi, pemberian, kontrak, orang, alat,
finding, managing client mengatur, minyak, menerbitkan, dijual, risalah,
contracts) langsung, audit, bangunan, membuktikan, dilaporkan,
mengajukan
*) Listed in descending order by their SVM weight are the most predictive subjects for transfer pricing situations involving
the Value-Added Tax, represented by the most common terms. Topic labels are inserted manually. Positive weights (w)
suggest more predictive themes for when the taxpayer won in Court, whereas negative weights denote more predictive
topics for when the government lost.
As for the final takeaway from the classification model of income tax, we could see
how the Government statement regarding the interest charge (topic 1, table 5) and the
testing of the interest and know-how fee are the predominant topics in an income tax
dispute. Even though both look similar, the combination of the words regarding each
topic could be said somewhat differently. The government statement regarding the
interest charge (topic 1, table 5) mainly talks about the contract, which could be referred
to as the evidence of the disputed transactions. Meanwhile, testing the interest and
know-how fee discusses the method to test the arm length principles, particularly the
tnmm method.
Furthermore, there is exciting output while examining value-added tax. In this
regard, we found that the three topics concerning Cases Where Taxpayers Lose that we
produced are equal on the weight list. However, the Government statement regarding
the export contract/agreement (topic 3) has a slight advantage over the Taxpayer
statement on toll manufacturing of FFB to CPO (titip olah tbs menjadi cpo) (topic 1) and
the Court statement regarding comparison criteria (kriteria pembanding) (topic 1). we
could see how the Government statement regarding the interest charge (topic 1, table 5)
and the testing of the interest and know-how fee are the predominant topics in income
tax disputes. Even though both look similar, the combination of the words regarding each
topic could be said somewhat differently. The government statement regarding the
interest charge (topic 1, table 5) mainly talks about the contract, which could be referred
to as the evidence of the transactions in the dispute. Meanwhile, testing the interest and
know-how fee discusses the method to test the arm length principles, particularly the
tnmm method.
5. Conclusion
This article describes the results of an interdisciplinary investigation into the
application of machine learning to predict the outcome of court decisions based on a
corpus of precedent cases. We were exclusively concerned with predicting transfer
pricing dispute decisions within the framework of Indonesian tax law. If the taxpayer
opposes the tax authority's objection decisions, he or she may file an appeal. However,
the appellant (taxpayer) and appellate (tax authority) spend a tremendous deal of time
and money on this process. In addition, 84 texts were analyzed using natural language
processing. We extracted many features for each occurrence. Textual features were
utilized as input for a Support Vector Machine (SVM) classifier. In the end, the model
evaluation demonstrates that the classifier is effective, with an F1-score between 0.50
and 1.00 and an accuracy between 80 and 100 per cent. In addition to the high predictive
accuracy of our statistical NLP framework, we have described a variety of qualitative
trends, namely topic interpretation, that may impact court decisions. In particular, we
emphasized that knowledge of the case's factual background, as provided by the Court
in the relevant section of its judgements, is necessary for achieving the most accurate
prediction of the Court's average decision result.
6. Implications and Limitations

In conclusion, our research paves the way for future research utilizing diverse data
types (e.g., texts of individual applications, briefs provided by parties, or domestic
119
decisions) from various sources. However, difficulties with data access constitute a
considerable obstacle for scientists working with such legal data. Access to more types
of data, mainly submitted applications and briefs, would facilitate additional study at the
interface of legal science and artificial intelligence.
All the data and workpaper related to this research can be accessed at:
https://github.com/mihoku/tax-court-analytics
Author Contributions
• Nikolaos Aletras and Vasileios Lampos conceived and designed the experiments,
performed the experiments, analyzed the data, contributed reagents/materials/analysis
tools, wrote the paper, prepared figures and/or tables, performed the computation work,
reviewed drafts of the paper.
• Dimitrios Tsarapatsanis conceived and designed the experiments, analyzed the
data,
contributed reagents/materials/analysis tools, wrote the paper, prepared figures and/or
tables, reviewed drafts of the paper.
Author Contributions
 Ananda Anggara (1st Author, corresponding author)

Conceptualization, Investigation/Data Collection, Data Curation, Formal Analysis,
Software, Resources, Validation, Methodology, Writing - Original Draft, Writing
– Final Submission
 Reza Darmawan (Co-Author)
Data Curation, Formal Analysis, Software, Validation, Methodology,
Visualization, Writing - Original Draft, Writing – Final Submission
References
Aletras, N., Tsarapatsanis, D., Preoţiuc-Pietro, D., & Lampos, V. (2016). Predicting
judicial decisions of the European Court of Human Rights: A natural language processing
perspective. PeerJ Computer Science, 2, e93.
Barker, J., Asare, K., & Brickman, S. (2017). Transfer pricing as a vehicle in corporate
tax avoidance. Journal of Applied Business Research (JABR), 33(1), 9-16.
Chang, Y. & Lin, C.. (2008). Feature Ranking Using Linear SVM. Proceedings of the
Workshop on the Causation and Prediction Challenge at WCCI 2008, in PMLR 3:53-64
Géron, A. (2019). Hands-on machine learning with Scikit-Learn, Keras, and TensorFlow:
Concepts, tools, and techniques to build intelligent systems. " O'Reilly Media, Inc.".
Gruginskie, L. A. D. S., & Vaccaro, G. L. R. (2018). Lawsuit lead time prediction:
Comparison of data mining techniques based on categorical response variable. PloS
one, 13(6), e0198122.
Joachims, T. (2002). Learning to classify text using support vector machines (Vol. 668).
Springer Science & Business Media.
Katz, D. M., Bommarito, M. J., & Blackman, J. (2017). A general approach for predicting
the behavior of the Supreme Court of the United States. PloS one, 12(4), e0174698.
Lawlor, R. C. (1963). What computers can do: Analysis and prediction of judicial
decisions. American Bar Association Journal, 337-344.
120
Medvedeva, M., Vols, M., & Wieling, M. (2018). Judicial decisions of the European Court
of Human Rights: Looking into the crystal ball. In Proceedings of the conference on
empirical legal studies (p. 24).
Panagis, Y., Christensen, M. L., & Sadl, U. (2016). On top of topics: Leveraging topic
modeling to study the dynamic case-law of international courts. In Jurix (pp. 161–166).
Vapnik, V. (1998). Statistical learning theory new york. NY: Wiley, 1(2), 3.
Waltl, B., Bonczek, G., Scepankova, E., Landthaler, J., Matthes, F. (2017). Predicting
the Outcome of Appeal Decisions in Germany's Tax Law. In: , et al. Electronic
Participation. ePart 2017. Lecture Notes in Computer Science(), vol 10429. Springer,
Cham. https://doi.org/10.1007/978-3-319-64322-9_8
Wang S, Manning CD. 2012. Baselines and bigrams: simple, good sentiment and topic
classification. In: Proceedings of the 50th annual meeting of the Association for
Computational Linguistics: short papers-Volume 2. 90–94.
Whitehead, S. (Ed.). (2015). The Tax Disputes and Litigation Review. Law Business
Research.
Van Hoecke, M. (2011). Foreword in'methodologies of legal research'. European
Academy of Legal Theory Series, I–IX.
121
DO OPERATIONAL RISKS AND CREDIT RISKS MAKE
BANKS MORE AGGRESSIVE TO AVOID TAXES?
EMPIRICAL STUDY OF BANKING COMPANIES ON THE
INDONESIA STOCK EXCHANGE
Muhammad Shohihul Wahyu Muzakkia, Nafis Dwi Kartikob, Lazuardi Widyanto
Pratamac
a Directorate General of Taxes Ministry of Finance of the Republic of Indonesia Email:
wahyu.muzakki@gmail.com
b Directorate General of Taxes Ministry of Finance of the Republic of Indonesia Email:
nafisdwikartiko@gmail.com
c Directorate General of Taxes Ministry of Finance of the Republic of Indonesia Email:
lazuardiwidyanto16@gmail.com
ABSTRACT
This study aims to analyze the relationship between credit risk and operational risk on the
aggressiveness of tax avoidance in banking companies in Indonesia. Based on the purposive
sampling method, this study selected banks listed on the Indonesia Stock Exchange (IDX) from
2004-2021 as samples. Total observations after elimination amounted to 271 pieces. The test
method in this study uses multiple regression analysis with panel data. The best model estimation
is done using the Chow test, Breusch Pagan Lagrange multiple tests, and Hausman test. The best
model estimation test results in the Fixed Effect Model as the best model. The test results show that
greater credit risk does not affect the aggressiveness of tax avoidance. Meanwhile, higher
operational risk affects the aggressiveness of higher tax avoidance.
Keywords: credit risk, operational risk, tax avoidance, banking.
Penelitian ini bertujuan untuk menganalisis hubungan antara risiko kredit dan risiko operasional
terhadap agresivitas penghindaran pajak pada perusahaan perbankan di Indonesia. Didasarkan
pada metode purposive sampling, penelitian ini memilih perbankanyang terdaftar di Bursa Efek
Indonesia (BEI) pada periode 2004-2021 sebagai sampel. Total observasi setelah eliminasi
berjumlah 271 sampel. Metode pengujian dalam penelitian ini menggunakan analisis regresi
berganda dengan data panel. Estimasi model terbaik dilakukan dengan menggunakan uji chow, uji
breusch pagan lagrange multiple, and uji hausman. Hasil uji estimasi model terbaik menghasilkan
Fixed Effect Model sebagai model terbaik. Hasil pengujian menunjukkan bahwa risiko kredit yang
lebih besar tidak berpengaruh terhadap agresivitas penghindaran pajak. Sedangkan risiko
operasional yang lebih tinggi berpengaruh terhadap agresivitas penghindaran pajak yang lebih
tinggi pula.
Kata kunci: risiko kredit, risiko operasional, penghindaran pajak, perbankan.
1. BACKGROUND
The corporation's mission is to maximize shareholder value through maximizing
profits. Profit maximization may be achieved through selling products or services
(Henderson, 2021; Luyckx et al., 2022). However, if a corporation merely sells
commodities, services, or both, it may not earn maximum profit, especially during
globalization when competition intensifies (Komljenovic, 2020). Companies must
innovate or diversify their goods, explore new markets, and minimize workloads, among
other tactics. The company as an entity has a tax burden that must be paid as an
obligation for the existence of the business carried out as regulated in a country's tax
laws. Companies tend to lower their tax payments to enhance their earnings after taxes.
In the context of a public corporation, a reduction in the tax burden is viewed as
advantageous for shareholders (Wang et al., 2020). Consequently, specific regulatory
action may be implemented in the firm's administration to lower the tax burden. This
122
activity has its incentives for management, as management is motivated by shareholder
interests. This regulatory effort is known as aggressive tax activity. Tax aggressiveness
is a company-specific activity involving transactions designed to lower the amount of
corporate tax due via tax planning (Frank et al., 2009). In addition, Slemrod (2004)
asserts that businesses would take advantage of leniency and legal interpretation gaps
to decrease their tax burden through lawful actions due to creative compliance.
There are difficulties in determining the limits of aggressive tax planning and tax
avoidance. Suandy (2011) defines tax avoidance as engineering tax affairs that are still
within the framework of tax provisions. The company carries out tax avoidance because
the company wants a large amount of profit. The existence of tax avoidance can hinder
the government's efforts to optimize tax revenues to finance state revenues.
Panjalusman et al. (2018) stated that tax avoidance is carried out by taking advantage
of gaps and loopholes in tax regulations to significantly reduce the amount of corporate
tax. The practice of tax avoidance has been widely practised in Indonesia and has
caused state financial losses due to tax evasion by companies. It was noted that from
2001-2009, Indonesia suffered a loss of $109 billion. One of the cases of tax evasion
that occurred in Indonesia is PT. Asian Agri operates in agriculture and plantation. PT.
Asian Agri evaded corporate taxes of IDR 2.6 trillion (Bawoleh, 2021; Putri & Mulyani,
2020). Cobham et al., (2020) in The State of Tax Justice 2020: Tax Justice in the time of
COVID-19 revealed that Indonesia is estimated to face a loss of US$ 4.86 billion per year
or equivalent to Rp. 68.7 trillion (exchange rate) rupiah worth Rp. 14,149 per US dollar)
due to tax evasion. The loss came from corporate taxpayers worth US$ 4.78 billion or
equivalent to Rp 67.6 trillion. Meanwhile, the rest came from individual taxpayers,
reaching US$ 78.83 million or equivalent to Rp 1.1 trillion. In addition, multinational
companies transfer income (shifting income) to countries considered tax haven
countries. Thus, companies that practice this practice end up paying less tax than they
should. Then, individual taxpayers belonging to the upper-class community hide assets
and income declared abroad to avoid the law's reach in their country. The existence of
state losses due to tax evasion cannot be justified. There is a need for follow-up from the
Ministry of Finance, in this case, the Directorate General of Taxes, so that tax loss due
to tax avoidance does not occur, considering that tax revenues have a significant
contribution to State revenues listed in the State Revenue and Expenditure Budget.
Previous research has demonstrated that tax avoidance is connected with increased
company risk. According to Rego & Wilson (2012), CEOs of businesses with lower
effective tax rates (ETR) get more substantial risk-taking incentives, which pushes them
to pursue hazardous tax reduction measures. According to Badertscher et al. (2013),
management ownership is positively associated with ETR, which is consistent with the
notion that poorly diversified owner-managers avoid the inherent hazards of tax reduction
techniques. Badertscher et al. (2013) discovered that firms with lower ETRs charged
higher interest rates to secure bank loans. Banking firms represent an industry with
significant business risk. The banking business is high-risk since it includes the
management of public finances and is played out through various activities, including the
provision of credit, the purchase of securities, and the investment of other funds.
Indonesia's banking system has been rife with instability throughout the previous two
decades. Broadly speaking, the risks faced by the banking industry consist of five main
risks, namely (1) credit risk, (2) market risk, (3) liquidity risk, (4) operational risk, and (5)
capital risk (Sintha, 2020). These risks are represented through various financial ratios
that show the management's performance in managing the bank. This study attempts to
determine the link between banking risk and tax avoidance in Indonesian banking
organizations. Specifically, we investigate the impact of operational risk and credit risk
on tax evasion by banking organizations. Therefore, this study aims to determine if the
increased risk in financial institutions would also increase the risk of tax avoidance.
123
2. THEORETICAL FRAMEWORK AND HYPOTHESES
DEVELOPMENT
2.1. Agency Theory
Agency theory illustrates the company as a meeting point between the company's
owner (principal) and management (agent). Jensen and Meckling state that an agency
connection is an agreement between the manager and the company owner
(Rokhlinasari, 2015). The principal is the group that gives accountability to the agent to
act on behalf of the principal. In contrast, the agent is the party the principal provides the
responsibility to run the company. The agent must account for what has been delegated
to him by the principal (Agustin et al., 2020; Narastri, 2022). The agent and principal's
powers and responsibilities are controlled in the employment contract upon mutual
agreement. The employment agreement is a set of rules governing the profit-sharing
mechanism, whether in the form of returns, returns or risks, which the principal and agent
approve. The employment contract will be optimal if the contract has fairness, namely
being able to balance the principal and the agent, which mathematically shows the
optimal implementation of obligations by the agent and the provision of sufficient
incentives/special rewards from the principal to the agent.
According to Eisenhardt & Eisenhardt (2018), agency theory is based on three
premises, namely: (a) assumptions about human nature, (b) assumptions about the
organization, and (c) assumptions about information. Assumptions about human nature
highlight that humans have the nature to be selfish (self-interest), have limited rationality
(bounded rationality), and do not like risk (risk aversion). Organizational assumptions are
conflicts between members of the organization, efficiency as a productivity measure, and
the presence of Asymmetric Information between principals and agents. While the
assumption about information is that information is seen as a commodity that can trade.
Eka (2018) stated that both the principal and agent have bargaining positions. The
principal, the owner of capital, has the right to access the company's internal information.
In contrast, the agent who runs the company's operations has accurate and
comprehensive information about the company's operations and performance. Still, the
agent does not have absolute authority to make strategic, long-term, and extensive
global decisions. Different functions and backgrounds between the principal and the
agent can cause a conflict of interest in the agency relationship. There is a conflict of
interest between the owner and the agent due to the possibility of the agent acting not in
the principal's interests, thus triggering agency costs. A conflict of interest between the
agent and the principal can lead to information asymmetry. This information asymmetry
will result in the meaning of accounting information being biased. Agency theory can
explain potential conflicts of interest between various interested parties in the company.
2.2. The Effect of Operational Risk and Credit Risk in Banking on Effective Tax Rate
Credit risk is the risk encountered by banks because they channel their funds into
loans to the public. Various reasons make the debtor may not fulfil his responsibilities to
the bank (Mulyati, 2018; Sebayang, 2020). Receivables management is essential for
companies whose operations provide credit because of the more significant the
receivables, the greater the risk. A non-Performing Loan (NPL) demonstrates a bank's
capacity to recover the credit granted by the bank until it is repaid. Non-Performing Loan
(NPL) is the proportion of non-performing (substandard, dubious, and loss) loans to the
total loans given by banks (Mulyati, 2018; Sebayang, 2020). In other words, the bigger
a bank's Non-Performing Loan (NPL), the more significant the performance disruption.
The amount of efficiency and the bank's capacity to carry out operational tasks are
measured by operational expenses (Mardiana & Purnamasari, 2018; Sihotang et al.,
2022). Banks incur operational expenses to conduct their core business operations (such
as interest, labor, marketing, and other operating costs). Operating income is a bank's
124
primary source of income, consisting primarily of interest income from the placement of
money in the form of credit and other operating income. BOPO is quantitatively assessed
using the efficiency ratio. This ratio measures whether the bank's management has
appropriately and efficiently utilized all of its production components (Karamoy & Tulung,
2019; Rajindra et al., 2021). The efficiency of a bank's business is measured using the
ratio of operating costs compared to operating income (BOPO). BOPO compares total
operating expenses with the entire operating income.
Tax provisions related to NPLs can be seen in the Minister of Finance Regulation
Number 207/PMK.010/2015 concerning the Second Amendment to the Minister of
Finance Regulation Number 105/PMK.03/2009 concerning Clearly Uncollectible
Receivables That Can Be Deducted From Gross Income. Article 2 of PMK Number
207/PMK.010/2015 defines that “Uncollectible Debts are receivables that arise from
normal business transactions by their line of business, which is uncollectible even though
efforts have been made to maximum or final collection by the Taxpayer." Requirements
for receivables that are uncollectible can be charged as a deduction from gross income,
as long as they meet the following requirements:
a. has been charged as an expense in the commercial income statement;
b. Taxpayers must submit a list of clearly uncollectible receivables to the Directorate
General of Taxes; and
c. The uncollectible receivables have been submitted for collection to the District Court
or government agency that handles state receivables, or there is a written agreement
regarding the write-off of receivables/debt relief between the creditor and the debtor
on the receivables that are uncollectible or have been published in a general or
special publication, or there is an acknowledgment from the debtor that the debt has
been written off for a certain amount of debt.
The greater the NPL value will affect the amount of tax payable. Of course, NPL will
also have an impact on increasing operating expenses. This addition will result in a
higher BOPO value. The greater the operational expenses, the smaller the ETR value
will be. This shows an increase in tax aggressiveness due to the addition of NPL and
BOPO. So that the hypothesis that we build in this study is that Operational Risk and
Credit Risk affect the Effective Tax Rate of banking companies in Indonesia.
3.1. Data and Samples
Sugiyono (2011) revealed that the population is a generalization area consisting of
objects or subjects with specific quantities and characteristics determined by researchers
to be studied and then concluded. The data sources in this study are from the Indonesia
Stock Exchange and the Indonesian Financial Services Authority. The population used
in this study are 24 companies that carry out their banking sector activities listed on the
Indonesia Stock Exchange in the period 2004-2021. This study has not been able to take
all banking companies on the Indonesia Stock Exchange. The number of companies, as
many as 24 banking companies, was chosen by considering the availability of NPL and
BOPO data on the Indonesian Financial Services Authority website.The method used in
collecting samples in this study is purposive sampling. Purposive sampling is a sampling
method based on specific considerations (Campbell et al., 2020). The considerations
mentioned in the statement in this study are (a) banking companies listed on the
Indonesia Stock Exchange from 2004 to 2021, (b) companies whose financial statements
have been audited, (c) using rupiah currency, (d) the sampled companies do not have a
tax avoidance proxy value (ETR) of less than zero or more than one, and (e) companies
that have complete data for all variables used during the study period. So that the total
observations in this study amounted to 271 observations. The data in this study were
125
processed using STATA 15 software in performing the regression model. For calculating
each variable and processing financial report data using Microsoft Excel.
3.2. Research Model

Hypothesis testing in this study uses the results obtained from the regression results.
Regression models that will be used in research that are considered relevant, namely:
𝐸𝑇𝑅1 = 𝛽0 + 𝛽1 𝑁𝑃𝐿𝑖𝑡 + 𝛽2 𝐵𝑂𝑃𝑂𝑖𝑡 + 𝛽3 𝐹𝑆𝐼𝑍𝐸𝑖𝑡 + 𝛽4 𝐹𝐴𝐺𝐸𝑖𝑡 + 𝛽5 𝑃𝐵𝑉𝑖𝑡 + 𝛽6 𝑅𝑂𝐴𝑖𝑡 +
𝛽7 𝑅𝑂𝐸𝑖𝑡 + 𝛽8 𝑆𝐼𝐺𝑖𝑡 + 𝜀𝑖𝑡 .........................................................(1)
𝐸𝑇𝑅2 = 𝛽0 + 𝛽1 𝑁𝑃𝐿𝑖𝑡 + 𝛽2 𝐵𝑂𝑃𝑂𝑖𝑡 + 𝛽3 𝐹𝑆𝐼𝑍𝐸𝑖𝑡 + 𝛽4 𝐹𝐴𝐺𝐸𝑖𝑡 + 𝛽5 𝑃𝐵𝑉𝑖𝑡 + 𝛽6 𝑅𝑂𝐴𝑖𝑡 +
𝛽7 𝑅𝑂𝐸𝑖𝑡 + 𝛽8 𝑆𝐼𝐺𝑖𝑡 + 𝜀𝑖𝑡 .........................................................(2)
Information:
ETR1 = First Proxy Effective Tax Rate
ETR2 = Second Proxy of Effective Tax Rate
NPL = Non Performing Loan
BOPO = Operating Expenses and Operating Income
FSIZE = Firm size
FAGE = Firm age
PBV = Price-to-Book Value Ratio
ROA = Return on Assets
ROE = Return on Equity
SIG = Sales Income Growth Rate
3.3. Selection of Panel Data Regression Estimation

In estimating panel data, this study uses three model parameters, namely Fixed
Effect Model (FEM), Random Effect Model (REM), and Pooled Least Square (PLS),.
Several tests were carried out by estimating the three models: the Chow test, the
Breusch Pagan Lagrange Multiple tests, and the Hausman test. The three tests aim to
estimate which method is the most suitable for use in research. The Chow test
determines whether the research should use the Fixed effect model or the Pooled Least
Square (PLS) model. If F Restricted < 0.05, then the FE model is a suitable model to
use. To determine the correct model between PLS and RE, use the LM test. The results
of the LM test with Prob F < 0.05 indicate that the RE model is suitable. The step to find
the right model between the RE and FE models is to use the Hausman test. Prob F value
< 0.05 indicates if the FE model is suitable.
3.4. Variable Operations

The following is a table of operationalization of variables used in this study:
Table 1 Variable Operations

Variable Operationalization
ETR1 (Worldwide Income Tax Expense)/(Worldwide total pre-tax
accounting income) (Hanlon & Heitzman, 2010)
Source: Indonesia Stock Exchange
ETR2 (Worldwide Income Tax Expense)/(Earning Before Interest Tax,
Depreciation, And Amortization) (Stamatopoulos et al., 2019)
NPL Ratio of non-performing loans (NPLs) to total gross loans (Naili &
Lahrichi, 2022; Zhang et al., 2016)
Source: Indonesian Financial Services Authority
BOPO Operating Expenses divided by Revenue Operational (Al Iqbal &
Budiyanto, 2020; Wahyuni, 2016)
Source: Indonesian Financial Services Authority
126
Variable Operationalization
Control Variables
FSIZE Natural Logarithm of Total Assets (Aduralere Opeyemi, 2019; Ayuba
et al., 2019)
FAGE Natural Logarithm of Subtraction between the year of measurement
and the year the company was founded (Kieschnick & Moussawi,
2018)
PBV Share Price divided by Book Value per Share (Al Iqbal & Budiyanto,
2020; Wahyuni, 2016)
ROA Pre-tax income divided by total assets (Richardson & Lanis, 2007)
ROE Pre-tax income divided by the book value of equity (Moussu & Petit-
Romec, 2014)

4.1. Descriptive Statistics
Descriptive statistics examine data by summarizing or describing the acquired data
without drawing generally accepted inferences or generalizations. In descriptive
statistics, each variable's mean, standard deviation, minimum, and maximum values are
sought. The statistical descriptive of each variable can be seen in Table 2. Table 2 shows
that the ETR1 variable, as one of the dependent variables in this study, has an average
value of 0.35793 with a minimum value range of 0.00000 and a maximum weight of
0.98934. Meanwhile, the ETR2 value is on average 0.08777 with a standard deviation of
0.10034.
Table 2 Descriptive Statistics

Variable Obs Mean Std. dev. Min Max
ETR1 271 0.35793 0.13924 0.00000 0.98934
ETR2 271 0.08777 0.10034 0.00000 0.96587
NPL 271 3.04493 3.02960 0.00000 24.84000
BOPO 271 93.82375 35.49410 0.86000 287.86000
FSIZE 271 24.26979 1.99620 18.31280 28.14870
FAGE 271 3.86848 0.56126 2.83320 4.83630
PBV 271 2.54494 6.35714 0.32670 85.65000
ROA 271 0.01175 0.01480 -0.08700 0.04140
ROE 271 0.10053 0.10091 -0.54700 0.36290
SIG 271 0.21439 0.52118 -0.70620 5.05490
Source: Source: Author's Calculation
The NPL ratio demonstrates the capability of bank management to manage non-
performing loans provided by banks. The higher the NPL ratio, the worse the credit
quality, which causes the number of non-performing loans to rise it can cause a bank to
be in a challenging condition. The average NPL value for 271 observations is 3,04493,
with a maximum weight of 24,84000. BOPO describes the efficiency of a bank in carrying
out its operational activities. The table above shows that the highest value of BOPO is
287.86, with the lowest value being 0.86. A high BOPO value indicates that the bank
cannot carry out its operational activities efficiently, leading to very minimal company
127
profits. The average BOPO value is 93,82375 above the standard set by Bank Indonesia,
which is 90%. The firm size variable as proxied by FSIZE shows the average value and
standard deviation of 24,26979 and 1,99620, respectively. Meanwhile, the firm age
indicated by FAGE shows an average value of 3,86848. The average values for PBV,
ROA and ROE were 2.54494, 0.01175, and 0.10053, respectively. At the same time, the
value of sales growth shows an average value of 0.21439 or 21.439% per year.
4.2. Correlation Analysis

The Pearson correlation test was done to determine the link between variables in the
study model; the association between variables has a value between 0 and 1. A
correlation value greater than 0.8 suggests a significant relationship between variables,
whereas a correlation value of 1 implies a perfect relationship. Table 3 shows the findings
of the correlation analysis.
Table 3 Correlation Table

Variables ETR1 ETR2 NPL BOPO FSIZE FAGE PBV ROA ROE SIG
ETR1 1.000
ETR2 0.119* 1.000
NPL 0.194*** 0.064 1.000
BOPO 0.156** 0.091 0.405*** 1.000
FSIZE -0.178*** 0.055 0.061 -0.258*** 1.000
FAGE -0.052 0.055 0.046 -0.152** 0.748*** 1.000
PBV -0.196*** -0.008 -0.066 0.118* -0.113* -0.115* 1.000
ROA -0.072 -0.264*** -0.048 -0.468*** 0.453*** 0.417*** -0.074 1.000
ROE -0.143** -0.324*** 0.014 -0.323*** 0.510*** 0.494*** -0.050 0.875*** 1.000
SIG 0.007 0.152** -0.125** 0.036 -0.197*** -0.172*** 0.517*** -0.285*** -0.229*** 1.000
*** p<0.01, ** p<0.05, * p<0.1
The test results show that the credit risk and operational risk variables significantly
correlate to ETR 1, while ETR 2 does not. Based on table 3, the results of the Pearson
correlation test show that the variables FSIZE, PBV, and ROE have a significant
correlation with the first proxy of the effective tax rate (ETR1) at 99% and 95% confidence
levels. Meanwhile, ROA and SIG have no significant correlation with the first proxy of the
effective tax rate (ETR1). In the second model, the control variables, namely ROA, ROE,
and GIS, are significantly correlated with the two proxies of the effective tax rate (ETR2).
4.3. First Model Regression

There are three types of research estimates for the panel data model: pooled least
square (PLS), random effect model (REM), and fixed effect model (FEM). Therefore, it
is required to initially evaluate the estimation of the research model using the panel data
model. The Langrage multiplier, Chow, and Hausman tests can be utilized to determine
the selection of the optimal estimating model. The first model of this investigation used
a fixed effect model estimation consistent with the Lagrange multiplier, Chow, and
Hausman test outcomes. Table 4 displays the results of each estimation.
Table 4 First Model Regression

PLS FEM REM
npl 0.00807*** 0.0135**** 0.00722**

(0.006) (0.000) (0.020)
128
bopo 0.000554* -0.00000749 0.000405
(0.058) (0.980) (0.152)
fsize -0.0196*** -0.0272* -0.0334****

(0.002) (0.088) (0.000)
fage 0.0461** -0.0937 0.0734*

(0.033) (0.439) (0.060)
pbv -0.00644**** -0.00823**** -0.00798****

(0.000) (0.000) (0.000)
roa 4.065*** 1.778 3.329***

(0.001) (0.281) (0.010)
roe -0.551*** -0.303 -0.440**

(0.002) (0.179) (0.016)
sig 0.0491*** 0.0613*** 0.0590***

(0.009) (0.001) (0.001)
_cons 0.591**** 1.356**** 0.835****

(0.000) (0.000) (0.000)
N 271 271 271
Prob > F 0.0000 0.0000 0.0000
R2 0.181 0.211
p-values in parentheses
* p < 0.1, ** p < 0.05, *** p < 0.01, **** p < 0.001
According to table 4, the probabilistic F statistical value of the study model is less
than 1%, indicating that the model is statistically significant at the 99% confidence level.
The degree of confidence implies that the dependent variable significantly affects the
independent variable in the research model. The R-Square score of the research model
is 0.211, indicating that the independent factors can account for 21.10% of the variance
in the dependent variable (ETR1). Independent variables are NPL, BOPO, FSIZE, FAGE,
PBV, ROA, ROE, and GIS. The remaining variance of 78.9% is determined by variables
independent of the research.
The coefficient of the NPL variable is 0.0135, and its probability is 0.000. The
coefficient value of 0.0135 reveals a correlation between non-performing loans and tax
evasion—the greater the value of non-performing loans, the greater the company's tax
avoidance. The greater the NPL value, the greater the ETR value. A high proportion of
non-performing loans does not make tax evasion more aggressive. At a confidence level
of 99.99%, a probability value of 0.000 suggests that the non-performing loan variable is
significantly associated with tax evasion. While the variable BOPO's coefficient is -
0.00000749 with a probability of 0.98, The coefficient value of 0.0135 implies that BOPO
and tax evasion have a negative association. The ETR value will decrease when the
BOPO value rises. This indicates that the bigger a company's operational expenditures,
the less tax it pays; hence, the more aggressive it is. The likelihood value of 0.98
suggests that the BOPO variable has no bearing on tax evasion. As for the control
variables, only FSIZE, PBV, and SIG variables are significant on the tax avoidance
proxy.
129
4.4. Second Model Regression
The second model of this study uses a fixed effect model estimation in accordance
with the results of the lagrage multiplier, chow and hausman tests. The results of each
estimate can be presented in Table 5.
Table 5 Second Model Regression

PLS FEM REM
npl 0.00228 0.00106 0.00303

(0.271) (0.594) (0.135)
bopo 0.0000712 -0.000380** 0.0000123

(0.731) (0.021) (0.948)
fsize 0.00999** -0.0399**** -0.00356

(0.024) (0.000) (0.523)
fage 0.0284* -0.0354 0.0555**

(0.064) (0.589) (0.011)
pbv -0.00116 0.00232** -0.00109

(0.268) (0.037) (0.282)
roa 1.223 0.340 0.978

(0.161) (0.702) (0.246)
roe -0.615**** -0.961**** -0.705****

(0.000) (0.000) (0.000)
sig 0.0335** -0.00717 0.0206*

(0.012) (0.485) (0.091)
_cons -0.235*** 1.313**** 0.00389

(0.005) (0.000) (0.969)
N 271 271 271
Prob > F 0.000 0.000 0.000
R2 0.207 0.506
p-values in parentheses
* p < 0.1, ** p < 0.05, *** p < 0.01, **** p < 0.001
Based on table 6, the probabilistic F statistical value of the research model is below
1%; this value indicates the research model is a significant model at the 99% confidence
level. The confidence level means that the independent variable in the research model
is significantly affected by the dependent variable. The R-Square value of the research
model is 0.506, which indicates that the independent variables of 50.60% can explain
the variation in the dependent variable (ETR2). The independent variables are NPL,
BOPO, FSIZE, FAGE, PBV, ROA, ROE, and GIS. 49.40% is another variation
determined by independent variables outside of the study.
The NPL variable has a coefficient of 0.00106 with a probability of 0.594. The
coefficient weight of 0.00106 shows a positive association between non-performing
130
loans and tax avoidance. The more the importance of non-performing loans, the higher
the level of tax avoidance by the company. A high NPL value means that the higher the
ETR value. It can identify that a high level of NPL does not make tax avoidance more
aggressive. The probability value of 0.594 indicates that the non-performing loan variable
is not significant to tax avoidance. At the same time, the coefficient of the BOPO variable
is -0.000380 with a probability of 0.980. The coefficient value of 0.0135 indicates a
negative association between BOPO and tax avoidance. The higher the BOPO value,
the smaller the ETR value will be. This means that the greater the company's operating
expenses, the smaller the tax paid, which can be more aggressive. The probability value
of 0.021 indicates that the BOPO variable is significant to tax avoidance at the 95%
confidence level. As for the control variables, only FSIZE, PBV, and ROE variables are
substantial on the tax avoidance proxy.
5. CONCLUSION
Non-performing loans and Operating Expenses are both a measure of the risk in
banking companies. From every credit the bank gives to the customer, not all of it can
be appropriately returned, not by the promised time. However, in reality, there are some
customers who, for some reason, the financial ratio used as a representative for the
weight of credit risk is a Non-Performing Loan (NPL). This ratio shows the bank
management's capacity to manage non-performing loans provided by banks. NPL
images credit risk; the smaller the Non-Performing Loan (NPL), the smaller the credit risk
accepted by the bank. In both regression equation models, the results show that NPL
has a positive effect on the tax avoidance proxy. An increase in NPL means that it
indicates an increase in the value of the ETR. On the other hand, a decrease in NPL will
result in a smaller ETR value. The aggressiveness of tax avoidance is measured using
the effective tax rate (ETR) calculated by dividing the income tax expense by the total
pre-tax accounting income. The smaller the tax rate indicates the company is getting
more aggressive in tax aggressiveness. These results suggest that the increase in credit
risk does not affect the aggressiveness of tax avoidance.
While the operational risk variable in banking is proxied by BOPO, the ratio of
operating expenses to operating income (BOPO) is often called the efficiency ratio, which
is used to estimate the capability of bank management to control operational costs to
operating income. The less this ratio means, the more efficient the operational expenses
incurred by the bank involved. A bank's success is based on a quantitative assessment
of a bank's profitability that can be measured using the ratio of operating costs to
operating income. This is because any growth in operations will result in a profit before
tax and eventually reduce the profit or profitability of the concerned bank. The test results
in the first and second models conclude that operational risk variables have a negative
effect on tax avoidance. The smaller the operational risk, the greater the ETR value. On
the other hand, the greater the value of operational risk, the smaller the ETR. The smaller
the ETR variable value indicates tax avoidance's aggressiveness. So the conclusion that
can draw shows that the increase in operational risk affects the aggressiveness of tax
avoidance.

This study concludes that credit risk does not affect the aggressiveness of tax
avoidance. At the same time, operational risk affects the aggressiveness of tax
avoidance. The implications of this research can at least provide input to the government,
companies, and further researchers. First, the government, especially the tax authorities
in Indonesia, should consider operational risk as one of the components used in
measuring banking taxpayer risk. Second, our findings are consistent with the concept
131
that a low ETR reflects the amount to which a company's operations allow it to take
benefit of tax-favourable transactions, as resisted to divergences in managers'
propensity to lower corporation tax payments by taking tax-risky positions. Third, future
researchers can explore the relationship of tax avoidance to other banking risks to show
that the increase in other risks also affects the aggressiveness of tax avoidance. In
addition, further researchers can use other tax avoidance proxies to see the consistency
and robustness of the resulting conclusions.
BIBLIOGRAPHY
Aduralere Opeyemi, O. (2019). The impact of Firm Size on Firms Performance in Nigeria:
A comparative study of selected firms in the Building Industry in Nigeria. Asian
Development Policy Review, 7(1), 1–11.
Agustin, H., Abdul Rahman, F., & Jamil, P. C. (2020). A Critical Islamic Perspective
Towards Agency Theory. International Journal of Economics, Business and
Management Research, 4(8), 43–50.
Al Iqbal, M. H., & Budiyanto, I. (2020). Analisis Pengaruh Kewajiban Penyediaan Modal
Minimum (KPMM), Beban Operasional Pendapatan Operasional (BOPO),
Financing To Deposit Ratio (FDR), dan Inflasi Terhadap Return On Asset (ROA)
Pada Bank Umum Syariah di Indonesia Periode 2016-2019. MALIA: Journal of
Islamic Banking and Finance, 4(1), 1–11.
Ayuba, H., Bambale, A. J., Ibrahim, M. A., & Sulaiman, S. A. (2019). Effects of Financial
Performance, Capital Structure and Firm Size on Firms’ Value of Insurance
Companies in Nigeria. Journal of Finance, Accounting & Management, 10(1).
Badertscher, B. A., Katz, S. P., & Rego, S. O. (2013). The separation of ownership and
control and corporate tax avoidance. Journal of Accounting and Economics, 56(2–
3), 228–250.
Bawoleh, J. R. (2021). PENGGELAPAN PAJAK DAN PROSES PENEGAKAN HUKUM
OLEH PENYIDIK POLRI. LEX PRIVATUM, 9(2).
Campbell, S., Greenwood, M., Prior, S., Shearer, T., Walkem, K., Young, S., Bywaters,
D., & Walker, K. (2020). Purposive sampling: complex or simple? Research case
examples. Journal of Research in Nursing, 25(8), 652–661.
Cobham, A., Bernardo, J. G., Palansky, M., & Mansour, M. B. (2020). The State of Tax
Justice 2020 : Tax Justice in the time of COVID-19. Tax Justice Network, November,
1–83.
Eisenhardt, K. M., & Eisenhardt, K. M. (2018). Agency Theory : An Assessment and
Review. Academy of Management, 14(1), 57–74.
Eka, H. (2018). Corporate finance and firm value in the Indonesian manufacturing
companies. Business Studies, 11(2), 113–127.
Frank, M. M., Lynch, L. J., & Rego, S. O. (2009). Tax reporting aggressiveness and its
relation to aggressive financial reporting. The Accounting Review, 84(2), 467–496.
Hanlon, M., & Heitzman, S. (2010). A review of tax research. Journal of Accounting and
Economics, 50(2–3), 127–178.
Henderson, R. M. (2021). Changing the purpose of the corporation to rebalance
capitalism. Oxford Review of Economic Policy, 37(4), 838–850.
Karamoy, H., & Tulung, J. E. (2019). The Effect of Banking Risk on Indonesia’s Regional
Development Banks.
Kieschnick, R., & Moussawi, R. (2018). Firm age, corporate governance, and capital
structure choices. Journal of Corporate Finance, 48, 597–614.
Komljenovic, J. (2020). The future of value in digitalised higher education: why data
privacy should not be our biggest concern. Higher Education, 1–17.
Luyckx, J., Schneider, A., & Kourula, A. (2022). Learning from alternatives: Analyzing
alternative ways of organizing as starting points for improving the corporation. In
132
The corporation: Rethinking the iconic form of business organization. Emerald
Publishing Limited.
Mardiana, M., & Purnamasari, P. E. (2018). The effect of risk management on financial
performance with good corporate governance as a moderation variable.
Management and Economics Journal (MEC-J), 2(3), 257–268.
Moussu, C., & Petit-Romec, A. (2014). RoE in banks: myth and reality. Available at SSRN
2374068.
Mulyati, E. (2018). The Implementation of Prudential Banking Principles to Prevent
Debtor with Bad Faith. Padjadjaran Jurnal Ilmu Hukum (Journal of Law), 5(1), 89–
108.
Naili, M., & Lahrichi, Y. (2022). Banks’ credit risk, systematic determinants and specific
factors: recent evidence from emerging markets. Heliyon, 8(2), e08960.
Narastri, M. (2022). Principles and Agents: The Phenomenon of Agency Theory in the
Business Sector and the Public Sector.
Panjalusman, P. A., Nugraha, E., & Setiawan, A. (2018). Pengaruh Transfer Pricing
Terhadap Penghindaran Pajak. Jurnal Pendidikan Akuntansi & Keuangan, 6(2), 105.
https://doi.org/10.17509/jpak.v6i2.15916
Putri, N., & Mulyani, S. D. (2020). Pengaruh Transfer Pricing dan Kepemilikan Asing
Terhadap Praktik Penghindaran Pajak (Tax Avoidance) Dengan Pengungkapan
Corporate Social Responsibility (CSR) Sebagai Variabel Moderasi. Prosiding
Seminar Nasional Pakar, 2–4.
Rajindra, R., GUASMIN, G., BURHANUDDIN, B., & ANGGRAENI, R. N. (2021). Costs
and Operational Revenue, Loan to Deposit Ratio Against Return on Assets: A Case
Study in Indonesia. The Journal of Asian Finance, Economics and Business, 8(5),
109–115.
Rego, S. O., & Wilson, R. (2012). Equity risk incentives and corporate tax
aggressiveness. Journal of Accounting Research, 50(3), 775–810.
Richardson, G., & Lanis, R. (2007). Determinants of the variability in corporate effective
tax rates and tax reform: Evidence from Australia. Journal of Accounting and Public
Policy, 26(6), 689–704.
Rokhlinasari, S. (2015). Teori-Teori dalam Pengungkapan Informasi Corporate Social
Responsibility Perbankan. Syria Studies, 7(1), 37–72.
Sebayang, P. (2020). The Impact of the Capital Adequacy Ratio, Non-Performing Loan
Against to Return on Equity (Case Study Private Bank in Indonesia). SHS Web of
Conferences, 76, 1035.
Sihotang, M. K., Hasanah, U., & Hayati, I. (2022). Model Of Sharia Bank Profitability
Determination Factors By Measuring Internal And Externals Variables. Indonesian
Interdisciplinary Journal Of Sharia Economics (Iijse), 5(1), 235–251.
Sintha, L. (2020). Pengelolaan Manajemen Risiko pada Industri Perbankan. Jurnal Mitra
Manajemen, 34–42.
Slemrod, J. (2004). The economics of corporate tax selfishness. National Tax Journal,
57(4), 877–899.
Stamatopoulos, I., Hadjidema, S., & Eleftheriou, K. (2019). Explaining corporate effective
tax rates: Evidence from Greece. Economic Analysis and Policy, 62, 236–254.
Suandy, E. (2011). Perencanaan pajak edisi 5. Jakarta: Salemba Empat.
Sugiyono. (2011). Metodologi penelitian kuantitatif kualitatif dan R&D. Alfabeta.
Wahyuni, S. F. (2016). Pengaruh Loan to Deposit Ratio (LDR), Capital Adequacy Ratio
(CAR) dan Rasio Beban Operasional Pendapatan Operasional (BOPO) terhadap
Return on Assets (ROA) pada Perbankan yang terdaftar di Bursa Efek Indonesia.
Jurnal Agrica, 9(1), 29–37.
Wang, F., Xu, S., Sun, J., & Cullinan, C. P. (2020). Corporate tax avoidance: A literature
review and research agenda. Journal of Economic Surveys, 34(4), 793–811.
133
Zhang, D., Cai, J., Dickinson, D. G., & Kutan, A. M. (2016). Non-performing loans, moral
hazard and regulation of the Chinese commercial banking system. Journal of
Banking & Finance, 63, 48–60.
134
EXAMINING FIRMS’ FINANCING OPTIONS: DO THIN
CAPITALISATION RULES MATTER?
Subagio Efendi*, Anis Anjala Widyanti**, Dinar Ayu Adeline**, Pramuji Handra Jadi**
* Corresponding author. Directorate General of Taxes, Ministry of Finance of Indonesia, Jakarta. Faculty
of Economics and Business, University of Indonesia
Email: subagio.effendi@kemenkeu.go.id
** Directorate General of Taxes, Ministry of Finance of Indonesia, Jakarta
Email: anisanjala@kemenkeu.go.id, dinar.adeline@kemenkeu.go.id, pramuji.handra@kemenkeu.go.id
ABSTRACT
This study extends existing research on firms’ financing choices by examining the use of debt financing over
time. Moreover, we investigate the effect of thin capitalisation rules on firms’ financing strategies and identify
alternative tax avoidance techniques used by the sample firms when thin capitalisations are too costly or no
longer available. Using confidential tax return data of 4,336 largest firms (28,952 firm-years) in Indonesia
from 2009 to 2017, our study finds that the use of debt financing tends to decline over time, implying equity
financing is gaining more popularity in the long run. The declining trend is partly explained by the
implementation of thin capitalisation rules, particularly for foreign affiliate firms. Additionally, we find
consistent shifting of the sample firms’ tax avoidance operations from thin capitalisations to non-debt tax
shelters. These findings suggest that to curb tax avoidance practices optimally, thin capitalisation rules need
to be implemented in concomitance with specific regulations and enforcement activities that restrict the use
of various non-debt tax shelters.
Keywords: financing option, thin capitalisation, tax avoidance method, tax return
JEL classifications: M41; M48; G32; G38; H25; H26
Studi ini berkontribusi pada lingkup penelitian tentang pilihan pembiayaan perusahaan dengan menguji
penggunaan pembiayaan utang dalam rentang waktu yang panjang. Lebih lanjut, kami menginvestigasi
dampak dari peraturan thin capitalisations terhadap strategi pembiayaan perusahaan dan mengidentifikasi
teknik penghindaran pajak alternatif yang digunakan oleh perusahaan sampel ketika thin capitalisations
menjadi berbiaya tinggi atau tidak dapat digunakan. Menggunakan data dari Surat Pemberitahuan Tahunan
Pajak Penghasilan 4,336 perusahaan (28,952 perusahaan-tahun) terbesar di Indonesia dari tahun 2009
sampai 2017, studi ini menemukan penggunaan pembiayaan utang cenderung menurun dalam jangka
panjang, yang berarti pembiayaan ekuitas semakin mendapat preferensi. Sebagian tren penurunan ini dapat
dijelaskan oleh implementasi peraturan thin capitalisations terutama untuk perusahaan yang dimiliki oleh
pemegang saham asing. Sebagai tambahan, kami menemukan penggeseran secara konsisten pada
operasi penghindaran pajak perusahaan sampel dari thin capitalisations ke non-debt tax shelters. Implikasi
dari temuan ini, untuk mencegah penghindaran pajak secara optimal, peraturan thin capitalisations harus
diimplementasikan berbarengan dengan kegiatan atau peraturan yang secara spesifik membatasi
penggunaan non-debt tax shelters.
Kata kunci: financing option, thin capitalisation, tax avoidance method, tax return
JEL classifications: M41; M48; G32; G38; H25; H26
1. INTRODUCTION
Tax implications of firms’ financing options, particularly from excessive debt financing,
have attracted substantial attention from policymakers and academia. The G20 and
OECD (2013), for example, have placed limiting interest deductions and other financial
payments as one important objective of their Base erosion and profit shifting (BEPS)
project.49 Moreover, the incoherence treatments of debt and equity financing costs in
49
Action four of the BEPS project aims explicitly to prevent base erosion through excessive interest deductions or financing
of the production of exempted or deferred incomes using, mainly, a fixed ratio between a firm’s net interest deductions to
135
corporate income tax calculations have been argued as a primary cause of the global
financial crisis (Slemrod, 2009). 50 Nevertheless, existing literature on optimal capital
structures, as reviewed by Auerbach (2002), Graham (2003) and Hanlon and Heitzman
(2010), contends that tax incentives only present third-order importance in the hierarchy
of corporate financing decisions. They argued that firms’ characteristics such as
profitability, size, investment in fixed assets, and production technologies are more
decisive in the real financing choices. However, the long-run trend of firms’ financing
choices remains unclear. Furthermore, the earlier assertions have mainly been drawn
from a sample of firms listed on stock exchanges with publicly-available data. Thus, they
may have limited generalisability due to survivorship bias caused by database errors and
omissions (Conover & Nichols, 2000; Kinney & Swanson, 1993; Mills, Newberry, &
Novack, 2003).
In this study, we analyse confidential tax administration data of the largest firms
operating in Indonesia to complement existing evidence on firms’ financing options also
the relative importance of tax incentives in firms’ financing decisions. Additionally, we
verify earlier claims on the substitution effect hypothesis (e.g. DeAngelo & Masulis, 1980;
Dhaliwal, Trezevant, & Wang, 1992; Graham & Tucker, 2006; MacKie‑Mason, 1990) by
examining alternative tax avoidance methods used by the sample firms when thin
capitalisations are restricted by the implementation of a targeted anti-avoidance rule and
the corresponding enforcements, making it too costly to operate. Recall that financing
choice is an essential part of firms’ business strategy; the findings may assist further
research in evaluating the long-term trend of firms’ financing options. Furthermore, the
findings may guide further research investigating the determinants and consequences of
firms’ financing choices. They also inform policymakers and tax authorities in evaluating
the efficacy of a targeted anti-avoidance reform. Additionally, we consider this study as
a modest response to the call by Graham (2003) and Hanlon and Heitzman (2010) for
more research examining the importance of corporate taxes in the firms’ financing
decisions and the role of non-debt tax shields in explaining the lingering ‘underlevered
puzzle’.
Examining a sample of 4,336 unique firms (28,952 firm-year observations) from
2009 to 2017 and multiple tax avoidance techniques employed by these firms, our study
finds that the use of debt financing tends to decline over time, implying equity financing
is gaining more priority by the sample firms in the long-run. The declining trend is partly
explained by the implementation of thin capitalisation rules, particularly for foreign
affiliate firms. These findings are consistent with the conservative use of debt financing
by larger and more profitable firms suggested by earlier studies (e.g. Graham, 2000,
2003; Graham & Tucker, 2006). Additionally, consistent with Graham and Tucker (2006),
we document consistent shifting of the sample firms’ tax avoidance operations from thin
capitalisations to non-debt tax shelters, suggesting the substitution effect of interest
deductions with non-debt tax shields. Thus, to optimally curb tax avoidance practices,
thin capitalisation rules need to be implemented in concomitance with specific
regulations and enforcement activities that restrict the use of various non-debt tax
shelters.51
This study offers at least three important contributions to the capital structure and
corporate tax avoidance literature. Firstly, we present initial cross-sectional
documentation of the long-run trend of firms’ financing strategies. To the best of our
its level of economic activity and other targeted rules. See OECD (2016) for descriptions of BEPS action four and OECD
(2013) for explanations of BEPS overall global actions.
50
In most income tax systems, interest payments are deductible while dividends are not deductible, which facilitates the
creation of a debt-tax shield.
51
Non-debt tax shelters include lease in-lease out, corporate-owned life insurance, contested liability acceleration strategy,
cross-border dividend capture, transfer pricing, offshore intellectual property havens, and deduction acceleration strategy.
See Wilson (2009) for identifications of tax shelter participants.
136
knowledge, ours is the first empirical study that examines firms’ financing choices over
time. Secondly, our study provides an immediate post-implementation review of a
specific anti-avoidance rule that limits interest deductions. While prior studies perform
their reviews using a sample of listed companies with publicly-available data (e.g.
Buettner, Overesch, Schreiber, & Wamser, 2012; Merlo, Riedel, & Wamser, 2020), we
examine a large set of annual tax return data obtained from the tax authority. The data
covers listed companies and private entities, enabling us to capture a complete picture
of firms’ financing options and to perform more internally valid investigations. Therefore,
the findings are more reliable and relevant for tax authorities and policymakers. Lastly,
by identifying a consistent shifting from thin capitalisations to non-debt tax shelters, our
study provides important insights into specific tax avoidance techniques used by the
sample firms. This highlights specific techniques which warrant more scrutiny from the
tax authority.
The following sections are structured as follows. Section two reviews existing
literature and formulates the hypotheses. This is followed by Section three, which
describes our research design. Further, Sections four and five contain sample selection
and empirical results, respectively. Some sensitivity analyses are presented in Section
six. Finally, Section seven provides concluding remarks and suggestions for further
research.
2. THEORY AND HYPOTHESES DEVELOPMENT

2.1 The long-run trend of firms’ financing options
Modigliani and Miller (1958) authored the ground-breaking article on capital structure.
They asserted that a firm’s value is independent of its capital structure, based on several
assumptions, including the absence of corporate and personal taxes, transaction costs,
imperfect market conditions, and asymmetric information. Responding to Modigliani and
Miller’s (1958) unrealistic assumptions, over time, significant theoretical and empirical
contributions have been made to the capital structure literature, as reviewed by Auerbach
(2002), Graham (2003) and Hanlon and Heitzman (2010). The trade-off theory is
currently dominating the firms’ capital structure debates. The theory generally argues
that by balancing the costs and advantages of an additional unit of debt, an optimal
capital structure maximises the firm’s value. Accordingly, firms maximise their value
when the benefits of debt, namely tax incentives, equal the marginal cost of debt,
particularly bankruptcy expenses (Myers, 2001). Several empirical studies, as reviewed
by Graham (2003), have confirmed the importance of various variables utilised by the
trade-off theory to explain the capital structure preference of firms.
The tax advantages of debt play a significant role in corporate financing decisions,
as interest payments are deductible for corporate income tax purposes, whereas equity
returns are not. Therefore, it is evident that a company need to consider the tax benefits
when determining its optimal capital structure. This incoherence of the tax system results
in a tax-induced preference for debt financing. Although existing studies have long
recognised this debt bias, its economic costs may be greater than previously
documented for several reasons. First, debt bias has become more urgent due to the
recent financial crisis (Slemrod, 2009). Second, debt bias erodes corporate tax bases.
This concern has increased due to the emergence of hybrid financial instruments and
the growing importance of multinational corporations engaging in international tax
planning. In its report to the G20 on financial sector taxation, the IMF noted that the
pervasive debt bias in G20 economies may have led to excessive risk-taking by financial
firms and a banking sector that is too large and inefficient (IMF, 2010). These concerns
have placed debt bias on the top of policy reform agendas of numerous nations, mainly
137
in limiting the tax deductibility of interest through thin-capitalisation rules and earnings
stripping provisions.
Nevertheless, multiple studies contend that tax incentives only present third-order
importance in the hierarchy of corporate financing decisions. They argued that firms’
characteristics such as profitability, size, investment in fixed assets, and production
technologies are more decisive in the real financing choices. Further, the existence of
non-debt tax shields reduces marginal tax benefits of interest deductions over time,
which leads to underlevered capital structures (Graham, 2000, 2003). Therefore,
consistent with the substitution effect hypothesis uttered by earlier research (e.g.
DeAngelo & Masulis, 1980; Dhaliwal et al., 1992; Graham & Tucker, 2006; MacKie‑
Mason, 1990), this study predicts that the use of debt financing tends to decline in the
long-run. To formally examine the utilisation of debt financing over time, we propose the
following hypothesis:
Hypothesis 1: The use of debt financing is declining over time.
2.2 The impact of thin capitalisation rules to firms’ financing options

Multinational corporations (MNCs) are incentivised to alter their group financing strategy
to shift profits from entities with high taxes to entities with reduced taxes. In achieving
this objective, the parent entity typically invests equity capital in a low-tax affiliate, which
is subsequently used to provide internal loans to affiliated entities in high-tax jurisdictions.
Because interest payments are deducted from the corporate tax base, the corresponding
revenue is removed from the high-taxed entities and taxed at a low or nil rate at the
entities located in tax-favourable jurisdictions or tax havens. To restrict the adverse
effects that debt financing might pose on tax collection, many countries have imposed
thin capitalisation rules. The previously described debt-shifting incentives can be limited
by establishing a maximum Debt to equity ratio (DER) that restricts a firm’s ability to
deduct interest payments from its income tax base. Tax avoidance through debt shifting
is a common problem that attracts substantial concerns. Studies show that more than 61
countries have adopted the thin capitalisation rules by 2012 (e.g. Merlo et al., 2020).
Multiple empirical studies have documented that thin capitalisation rules have
effectively reduced debt utilisations in the firms’ capital structure (Blouin, Huizinga,
Laeven, & Nicodème, 2014; Buettner et al., 2012; Clemente-Almendros & Sogorb-Mira,
2016). The two most critical features of thin capitalisation rules are DER restrictions for
related-party debt and total debt. Although there is a decline in internal debt utilisations,
implementing these regulations leads to the increasing use of external debt as a
substitute (Blouin et al., 2014; Buettner et al., 2012; Overesch & Wamser, 2010). Thus,
broader definition of debt is advisable to augment the intended effects (Buettner et al.,
2012). However, in the absence of restrictions on using internal debts, firms with access
to financing from related parties (i.e. parent companies or affiliates) can channel external
fundings to related parties and charge a higher interest rate. They can, nevertheless,
adhere to the maximum DER while reaping the tax benefits of internal debts.52
The Indonesian tax authority released Ministry of Finance Regulation number

169/010/2015 as the thin capitalisation rules to prevent base erosion caused by
excessive debt-financing. Indonesia’s thin capitalisation rules limit the number of
deductible borrowing costs (e.g. interest payments, loan arranger fees, lease expenses,
loan premiums, mortgage fees) using a maximum DER of 4:1. Accordingly, if a firm’s
DER surpasses the threshold, any borrowing costs that correspond to the excess ratio
52
Preventing this debt channelling scheme, the G20 and OECD (2016) recommend the implementation of a group-level
fixed ratio complementing the entity-level comparison between net interest deductions and the level of economic activity.
138
are not deductible in the corporate tax calculations. These rules apply to both related
and third-party debts, domestic or foreign. The maximum DER 4:1 threshold under the
thin capitalisation rules applies to all corporations established or domiciled in Indonesia,
except for certain entities subject to specific tax provisions (i.e. financial firms, mineral
and mining firms, oil and gas contractors, firms that subject to final income taxes). The
rules initiate their implementation in the fiscal year 2016. Consistent with the findings of
earlier studies, we predict that implementing thin capitalisation rules reduces the use of
debt financing by the Indonesian firms. Therefore, this study proposes the following
directional hypothesis:
Hypothesis 2: Thin capitalisation rules are negatively associated with debt

financing.
2.3 Substitution of thin capitalisations with other tax avoidance methods

Operating tax minimisation strategies are associated with substantial risks and
uncertainties caused by the opaque nature of tax avoidance, yet they can also provide
financial benefits for the firm and its shareholders. Moreover, to maximise firm value,
managers must be optimally aggressive by reducing income tax payments and the
necessary costs of tax planning (Hanlon & Slemrod, 2009; Scholes, Wolfson, Erickson,
Maydew, & Shevlin, 2014). Therefore, firms need to weigh the relative costs of individual
tax avoidance methods and try to optimise every viable opportunity to lower their tax
burdens.53 Prior studies (e.g. Desai, 2003; Graham & Tucker, 2006; Lisowsky, 2010;
Lisowsky, Robinson, & Schmidt, 2013; Wilson, 2009) suggest tax sheltering is the most
aggressive and sophisticated techniques in the tax aggressiveness continuum, whereas
taking benefits from temporary and permanent non-conformities between financial
accounting standards and tax laws are argued to be the least costly strategies (Desai,
2003). Additionally, due to the collective government initiatives in combating base
erosion and profit shifting, transfer pricing, tax havens, and treaty-shopping are
proclaimed as the most costly methods to implement (Ernst & Young, 2013, 2016).
Existing evidence (e.g. Efendi, Czernkowski, Morton, & Bond, 2021b) shows that firms
are moving toward the most sophisticated and less-costly tax avoidance techniques in
the long-term.
Thin capitalisations are one of the main and less-costly tax avoidance techniques
(Buettner et al., 2012; Efendi et al., 2021b; Merlo et al., 2020; OECD, 2013). Firms
typically alter their financing strategies by using more internal and external debts when
the corresponding interest deductions substantially lower their marginal tax rate. In other
words, firms will continuously add one unit of debt until the marginal benefit (i.e. tax
savings) equals the marginal cost (i.e. bankruptcy expenses). However, the existence of
non-debt tax shields (e.g. investment tax credit, depreciation expense, fiscal loss carry
forward, tax shelters) act as preferred substitutes to thin capitalisations and reduces the
tax incentives of using debt financing (DeAngelo & Masulis, 1980; Dhaliwal et al., 1992;
Graham & Tucker, 2006; MacKie‑Mason, 1990). Furthermore, firms near tax exhaustion
utilise less debt financing since they face an almost zero marginal benefit on interest
deductions (MacKie‑Mason, 1990).
These findings reveal firms’ tendency to employ the most sophisticated and less-
costly tax avoidance methods in their tax planning. Consistent with the substitution effect
hypothesis, this study predicts that when thin capitalisations are restricted or too costly,
firms tend to substitute debt financing with other viable tax avoidance techniques. To
53
Agency, implementation and outcome costs are the essential considerations in selecting a specific tax avoidance
method suggested by Wilde and Wilson (2018). See Wilde and Wilson (2018, p. 64) for a detailed explanation of an
agency-based framework for corporate tax planning decisions.
139
formally examine the changes in firms’ tax planning after the implementation of thin
capitalisation rules, we propose the following hypothesis:
Hypothesis 3: Firms substitute debt financing with other tax avoidance methods
after the implementation of thin capitalisation rules.
3. RESEARCH DESIGN
This section explains our methodological approaches to investigating the outlined
hypotheses. The methods begin with constructing the proxy for the use of debt
financing, LEV, followed by examining its trend over time (𝐻𝑦𝑝𝑜𝑡ℎ𝑒𝑠𝑖𝑠1 ). We then
outline the methods to examine the effect of thin capitalisation rules on debt
financing (𝐻𝑦𝑝𝑜𝑡ℎ𝑒𝑠𝑖𝑠2 ) and the substitution of debt financing with alternative tax
avoidance techniques (𝐻𝑦𝑝𝑜𝑡ℎ𝑒𝑠𝑖𝑠3 ).
3.1 Assessing firms’ financing decisions
Consistent with prior studies (e.g. Buettner et al., 2012; MacKie‑Mason, 1990), this study
infers firms’ financing choices from the utilisation of debt financing. Accordingly, we
construct the measure of a firm’s financial leverage, LEV, by accumulating a firm’s
internal and external long-term debts and dividing them by its total assets as follows:
𝑇𝑜𝑡𝑎𝑙 𝑙𝑜𝑛𝑔𝑡𝑒𝑟𝑚 𝑑𝑒𝑏𝑡𝑖𝑡
𝐿𝐸𝑉𝑖𝑡 = 𝑇𝑜𝑡𝑎𝑙 𝑎𝑠𝑠𝑒𝑡𝑠
(1)
𝑖𝑡
The measure is theoretically describing the portion of long-term debts used
to finance a firm’s assets. Following a standard balance sheet equation, that is,
total assets are equal to total liabilities plus shareholders’ equities, we interpret a
close to one LEV as a firm that excessively uses debt financing in its asset
acquisitions, thus a potentially tax aggressive firm, while a near-zero LEV as a
firm relying on equity financing, a potentially tax compliant firm.
3.2 Firms’ financing options over time

In examining the use of debt financing over time, consistent with earlier studies (e.g.
Dyreng, Hanlon, Maydew, & Thornock, 2017; Efendi et al., 2021b; Thomsen & Watrin,
2018), we evaluate the linear time trend of LEV throughout the sample period by
regressing it to a time trend variable as follows:
𝐿𝐸𝑉𝑖𝑡 = 𝛼 + 𝛽𝑇𝐼𝑀𝐸𝑖𝑡 + Σ𝛿𝑘 𝐶𝑂𝑁𝑇𝑅𝑂𝐿𝑆𝑖𝑡 + Σ𝛿𝑚 𝐼𝑁𝐷𝑈𝑆𝑇𝑅𝑌𝑚𝑡 + 𝜀𝑖𝑡 (2)
(𝑖 = firms 1 − 4,336 and 𝑡 = 2009 − 2017)
The dependent variable is LEV, as previously defined. The explanatory variable of
interest is TIME, calculated as the fiscal year for each observation less the number 2009,
which is the first year in the sample period. Thus, TIME takes a value between 0 to 8,
which corresponds to the observation period from 2009 to 2017. Consistent with the
outlined hypothesis, this study anticipates a declining trend in debt financing over time;
therefore, β needs to be negative and significant in the regression result of estimating
equation (2).
3.3 The effect of thin capitalisation rules to firms’ financing options
To examine the immediate effect of thin capitalisation rules on firms’ financing choices,
our study employs a difference-in-differences estimation by assigning sample firms in
fiscal years after the implementation of thin capitalisation rules in Indonesia (i.e. the fiscal
years 2016 and 2017) as the treatment group and sample firms in fiscal years prior to
the implementation (i.e. the fiscal year 2009 to 2015) as the control group. Thus, we
assign LEV as the dependent variable and an indicator variable, THINCAP, as the
explanatory variable in the following regression equation:
𝐿𝐸𝑉𝑖𝑡 = 𝛼 + 𝛽𝑇𝐻𝐼𝑁𝐶𝐴𝑃𝑖𝑡 + Σ𝛿𝑘 𝐶𝑂𝑁𝑇𝑅𝑂𝐿𝑆𝑖𝑡 + Σ𝜃𝑙 𝐼𝑁𝐷𝑈𝑆𝑇𝑅𝑌𝑙𝑡 + Σ𝛾𝑚 𝑌𝐸𝐴𝑅𝑚𝑡 + 𝜀𝑖𝑡 (3)
140
(𝑖 = firms 1 − 4,336 and 𝑡 = 2009 − 2017)
LEV is a firm’s use of debt financing, as previously defined. The variable of interest
is THINCAP, a dummy variable coded one for sample firms in fiscal years 2016 and 2017
and zero for sample firms in other fiscal years. Consistent with the outlined hypothesis,
we predict that thin capitalisation rules limit the tax deductibility of interest payments and,
thus, lower firms’ incentives to use more debt financing; accordingly, β needs to be
negative and significant in the regression result of estimating Equation (3). Conversely,
if thin capitalisation rules are irrelevant in firms’ financing strategies, the coefficients
should be insignificant.
3.4 Substitution of thin capitalisations with other tax avoidance

methods
Lastly, to examine changes in the sample firms’ tax planning following the
implementation of thin capitalisation rules, we partially replicate estimation
methods employed in Efendi et al. (2021a; 2021b) by regressing individual
alternative tax avoidance methods to LEV in the sample firms from 2015 to 2017
fiscal years using the following equation:54
𝐿𝐸𝑉𝑖𝑡 = 𝛼 + 𝛽1 𝑁𝑂𝑁_𝑅𝑂𝑈𝑇𝐼𝑁𝐸𝑖𝑡 + 𝛽2 𝑆𝐻𝐸𝐿𝑇𝐸𝑅𝑖𝑡 + 𝛽3 𝐿𝑂𝑆𝑆𝑖𝑡 +
𝛽4 𝑇𝑅𝐸𝐴𝑇𝑌𝑖𝑡 + 𝛽5 𝐻𝐴𝑉𝐸𝑁𝑖𝑡 + 𝛽6 𝑃𝐸𝑅𝑀𝐷𝐼𝐹𝐹𝑖𝑡 + 𝛽7 𝑇𝐵𝑇𝐷𝑖𝑡 +
𝛽8 𝐶𝑂𝑁𝐹𝑂𝑅𝑀𝑖𝑡 + Σ𝛿𝑘 𝐶𝑂𝑁𝑇𝑅𝑂𝐿𝑆𝑖𝑡 + Σ𝛿𝑚 𝐼𝑁𝐷𝑈𝑆𝑇𝑅𝑌𝑚𝑡 +
Σ𝜃𝑛 𝑌𝐸𝐴𝑅𝑛𝑡 + 𝜀𝑖𝑡 (4)
(𝑖 = firms 1 − 4,336 and 𝑡 = 2015 − 2017)
LEV is a firm’s use of debt financing and, thus, our proxy for thin capitalisations, as
previously defined. Consistent with Efendi et al. (2021a, 2021b), alternative tax
avoidance methods are NON_ROUTINE, a proxy for aggressive transfer pricing
techniques, calculated as total non-routine intra-group transactions scaled by total net
sales; SHELTER, a quantitative probability of tax sheltering as suggested by Wilson
(2009); LOSS, a dummy variable for loss carryover and shift-to-loss techniques
suggested by De Simone et al. (2017) and Hopland et al. (2018), coded one if a firm
reports fiscal losses and zero otherwise; TREATY, a proxy for treaty-shopping,
measured as total internal transactions with affiliates domiciled in Indonesia’s tax treaty-
partners scaled by total net sales55; HAVEN, a proxy for income shifting to tax havens,
measured as total intra-group transactions with affiliates located in tax havens scaled by
total net sales 56 ; PERMDIFF, permanent difference to proxy for permanent non-
conforming tax avoidance method consistent with Frank et al. (2009), measured as total
permanent fiscal adjustments scaled by total assets; TBTD (temporary portions of a
firm’s book-tax differences to proxy for temporal non-conforming tax avoidance method,
calculated as total temporary fiscal adjustments scaled by total assets; and CONFORM,
a proxy for conforming tax avoidance techniques as suggested by Badertscher et al.
54 The Indonesian thin capitalisation rules were legislated in 2015 and began their implementation in 2016.
Therefore, we include the fiscal year 2015 in the analysis to completely capture the changes in firms' tax
planning.
55
Consistent with Efendi et al. (2021a, 2021b), TREATY is a rough measure of treaty-shopping since it
captures the magnitude of a firm’s intra-group transactions with affiliates domiciled in treaty-partner
countries, yet it fails to segregate the tax-motivated transactions from those that commercially justified.
56
Following Efendi et al. (2021b), we classify a country or jurisdiction as a tax haven if it is listed in the
noncooperative jurisdictions for tax purposes proposed by OECD (2000), Dharmapala and Hines (2009),
the European Union (2021), or the US Senate (2011). See Appendix B of Efendi et al. (2021b) for a
complete list of the tax haven countries.
141
(2019). Accordingly, by estimating Equation (3), our study assumes firms have the
capability to complement excessive interest deductions with other tax avoidance
techniques; therefore, the sign of the coefficients of individual methods in the regression
result of estimating Equation (4) will reflect the complementation or substitution of
interest deductions with other methods in firms’ tax planning.
Our variables of interest are all individual alternative tax avoidance methods:
NON_ROUTINE, SHELTER, LOSS, TREATY, HAVEN, PERMDIFF, TBTD, and
CONFORM. If sample firms have tendencies to substitute interest deductions with
alternative tax avoidance methods, we expect negative and significant coefficients of
these alternative tax avoidance techniques in the result of estimating Equation (4).
However, if these alternative methods are unnecessary in the sample firms’ tax planning,
the coefficients must be insignificant.
3.5 Control variables

Control variables in all regression equations are various internal characteristics
associated with firms’ financing decisions suggested by earlier studies: SIZE to control
for the size-effect of firms financing choices as proposed by Dhaliwal et al. (1992); ROA,
return on assets, to control for firms’ profitability as suggested by Graham (2000, 2003)
and Graham and Tucker (2006); FOREIGN to control for firms’ international operations
as proposed by Rego (2003); FIXED, total fixed assets, to capture moral hazard and
debt securability effect as suggested by MacKie-Mason (1990) and Dhaliwal et al. (1992);
INVEN, inventory intensity, to control for firms’ investment choices as proposed by Gupta
and Newberry (1997); and industry dummies using two-digit of Indonesia’s standard
industry classification to control for production technologies, the degree of asymmetric
information between managers and investors across industries, and other industry-fixed
effects as suggested by MacKie-Mason (1990) and Dhaliwal et al. (1992). Additionally,
Equations (3) and (4) control for year-fixed effects to capture unobserved macro-
economic effects and time-specific enforcement policies that influence firms’ financing
decisions, as proposed by MacKie-Mason (1990). The complete definitions of individual
control variables are presented in Appendix A.
4. SAMPLE SELECTION AND DESCRIPTIVES

This study analyses comprehensive firm-level data exclusively provided by the
Indonesian tax authority, Directorate General of Taxes, for research purposes. The panel
database comprises all firms administered at the two regional offices with the most
significant tax revenue contributions, namely the Large Taxpayers Regional Office and
the Jakarta Special Regional Tax Office, covering the 2009 to 2017 periods. The data is
collected directly from the firms’ annual corporate income tax returns and multiple
attachments. However, due to tax information secrecy provisions, all data are provided
anonymously; therefore, we cannot match specific financial information with the
corresponding firm. This necessary anonymity does not limit our ability to perform the
analysis.
Investigating confidential tax return data provides several advantages. Since the
legislation mandates tax return reporting, these large taxpayers must report their actual
financial performance accompanied by the audited financial statement. Therefore, our
dataset offers high validity and reliability. This favourable characteristic of the dataset
allows us to construct specific variables, run sensitivity analysis, and implement time
trend examination to capture the progression of individual variables over time. Employing
a dataset from only the two largest regional offices, however, has its drawbacks. First, it
only reflects the largest firms in Indonesia and does not represent the entire population
of firms in Indonesia that is affected by the thin capitalisation rules. Second, taxpayers
142
administered within these two offices are dominated by compliant taxpayers due to
intensive supervision and enforcement activities. Thus, there is a possibility of selection
bias within the sample that could affect the generalisability of our empirical results. This
sample bias, however, could be treated by controlling firm-level characteristics and
limiting the empirical result concluded in this paper only to this population.
Table 1 Sample composition
In selecting the sample firms, we begin with 53,573 firm-year observations

covering all firms administered in the two largest regional offices for nine fiscal years
from 2009 to 2017. In order to properly analyse a sample group affected by the thin
capitalisation rules, firms that are exempted from the regulation are excluded from the
sample. The exclusion includes 3,430 small firms that are subject to final tax rates; 1,117
natural extraction firms, including oil, gas, coal, and mineral mining, also geothermal
firms that are subject to specific DER provisions in their contracts with the government;
5,946 firms in the specific industries that subject to final tax rates;57 768 financial services
firms; 13,210 firms that report incorrect balance sheets with long-term liabilities to value
higher than the assets; and 150 firms that apply incorrect statutory corporate tax rates.
Finally, these filterings give us 4,336 firms (28,952 firm-years) that we further classify
into two panels based on industry and ownership classifications. The selected sample
group has a variety of industries dominated by non-automotive wholesale trading,
suggesting a low risk of industry bias.58 Furthermore, 33.47 per cent of the sample are
purely domestic, and the rest are firms with overseas shareholders. Panel A of Table 1
summarises the sample selection criteria, Panel B presents the detailed industries
distribution, and Panel C displays the ownership classification.
Table 1 Sample composition
Variable N Mean Standard Min. p.25 Median p.75 Max.

deviation
Panel A: Descriptive statistics
LEV 28,952 0.1148 0.1944 0.0000 0.0000 0.0207 0.1392 1.0000
THINCAP 28,952 0.2269 0.4188 0.0000 0.0000 0.0000 0.0000 1.0000
SHELTER 25,727 0.0367 0.1399 0.0000 0.0000 0.0000 0.0000 1.0000
SHELTER_RESI
23,606 0.0000 0.7041 –7.0859 –0.3238 0.1239 0.4261 6.4731
D
LOSS 28,952 0.0783 0.2686 0.0000 0.0000 0.0000 0.0000 1.0000
TREATY 26,173 0.0387 0.1439 0.0000 0.0000 0.0000 0.0000 1.0000
HAVEN 26,276 0.0077 0.0662 0.0000 0.0000 0.0000 0.0000 1.0000
PERMDIFF 23,784 –0.0319 0.1432 –1.0000 –0.0511 –0.0043 0.0052 0.9871
TBTD 23,882 –0.0209 0.0938 –0.9972 –0.0201 –0.0000 –0.0000 0.9798
CONFORM 21,466 0.0000 0.0319 –0.0047 –0.0159 –0.0047 0.0085 0.6605
SIZE 28,951 24.8080 2.3271 0.0000 23.3207 24.8076 26.3816 32.7391
ROA 23,874 0.0589 0.1718 –0.9950 0.0000 0.0479 0.1296 0.9990
FIXED 28,792 0.2429 0.2365 0.0000 0.0365 0.1762 0.3866 1.0000
57
See Efendi (2020) for further descriptions of the combination of comprehensive and schedular/final income tax systems
in Indonesia.
58
In mitigating industry bias, we control for industry-fixed effects in all regression estimations.
143
Sample selection criteria Firm-years
(2009-2017)
Panel A: Sample selection summary
All firms administered at the Large Taxpayers Regional 53,573
Office and the Jakarta Special Regional Tax Office
Less:
INVEN Small firms 28,936 0.1556 0.1836 0.0000 0.0000 0.0935 (3,430)
0.2493 1.0000
FOREIGN Coal and mineral
28,935 mining
0.0001firms which sign
0.0019 their contract
–0.0001 0.0000 0.0000 (182)0.0000 0.1318
Financing of works beforeN 2009 Foreign affiliates Domestic firms Differences
options Oil and gas firms which
Foreign sign their
Domestic production
Mean sharing Mean
Median Median (800)Mean Median
contracts before 2009
Panel B: Comparison of foreign affiliates and domestic firms
LEV Geothermal firms 9,691
19,261 0.1223 0.0308 0.0999 0.0000 (135)0.0223* 0.0308*
ROA Firms that
15,928subject to
7,946 final income
0.0606 taxes
0.0487(e.g. 0.0555 0.0469 (5,946)
0.0051* 0.0018*
construction
SHELTER_RESID services,
is the residuals real estates,
of regressing SIZE toshipping,
SHELTERair to lessen multicollinearity problems, while other
transportations)
variables are as defined in Appendix A. TBTD and PERMDIFF are scaled by lagged total assets. ROA, FOREIGN,
TBTD, LEV,Financial servicesare
and PERMDIFF censored to –1 and 1. Similarly, NON_ROUTINE, TREATY,
firms (768)
HAVEN, FIXED, and
INVEN are Firms
censoredwith 0 and 1. Abalance
to incorrect foreign affiliate is a firm where at least one of its shareholders
sheet (LEV>1) (13,210)is not domiciled
in Indonesia. Differences between means are tested using
Firms which apply incorrect statutory income tax rates t-tests, while differences in medians
(150) are tested using
Wilcoxon Final
rank-sum
sample(Mann-Whitney) tests. The asterisk (*) indicates the differences are significantly
28,952different than zero
at five per cent confidence level.
Industry and ownership description Frequency Number of Firm-years
(%) firms (2009-2017)
Panel B: Industry classification
Agriculture 2.44 107 705
Automotive manufacturing 2.81 99 813
Basic chemicals manufacturing 5.69 215 1,647
Broker, dealer, and financial securities 5.03 278 1,455
Clothing and apparels 3.93 162 1,139
Electronic and optical parts manufacturing 2.21 88 640
Foods manufacturing 5.29 214 1,533
Glass, cement, and gips manufacturing 2.05 76 594
Management services 4.79 214 1,388
Metal products manufacturing 2.69 108 780
Non-automotive wholesale trading 17.04 39 4,933
Large transport vehicles manufacturing 2.07 75 599
Rubber and plastic products 4.56 178 1,320
manufacturing
Textile manufacturing 3.26 128 945
Warehouse and transportation services 2.32 101 671
Wooden products manufacturing 2.42 103 702
Other 31.4 2,151 9,088
Total 100 4,336 28,952
Panel C: Ownership classification
Domestic firms 33.47 2,359 9,691
Foreign affiliates 66.53 3,126 19,261
Total 100 5,485 28,952
Industry classification follows standard industry classification developed by Indonesia’s Central
Bureau of Statistics (2015). Other industries are various manufacturing, trading, and services
with less than two per cent frequency. A foreign affiliate is a firm where at least one of its
shareholders is not domiciled in Indonesia. The total number of firms is different between Panel
B and Panel C due to changes in firms’ ownership structure during observation periods.
Our study examines the long-run trend of firms’ financing options and the impacts
of thin capitalisation rules on firms’ financing choices. Panel A of Table 2 presents
descriptive statistics of the use of debt financing, alternative tax avoidance methods, and
control variables employed in the analysis. It shows that, among others, the mean
(median) of LEV is 0.1148 (0.0207), indicating, on average, sample firms are
underlevered comparable with similar conditions reported by earlier studies using
different research settings (e.g. Graham, 2000; Graham & Tucker, 2006). Furthermore,
the mean and median of TBTD and PERMDIFF are negative, suggesting low financial
reporting incentives among sample firms and an escalating risk of using conforming tax
avoidance schemes. Panel B reveals that the mean and median
144
of LEV and ROA significantly differ between foreign affiliates and purely domestic firms
in the sample group, implying different financing strategies and profitability among the
two subsamples. Accordingly, Figure 1 displays visual illustrations of the different
financing choices between subsamples over the examination periods. It also portrays
declining trends in the use of debt financing among the two subsamples after the
implementation of thin capitalisation rules, suggesting an immediate effect of the anti-
avoidance measures. However, the effect seems to last only in the foreign affiliates
subsample.
Table 2 Descriptive statistics
Figure 1 Trends of leverage by ownership classifications
This figure shows the average use of debt financing by foreign affiliates and purely domestic firms
in the sample group over the observation period.
Table 3 provides bivariate correlations between LEV and alternative tax avoidance
methods. The table reveals linearly and monotonically positive correlations
between LEV and other tax avoidance techniques (i.e. NON_ROUTINE, LOSS,
TREATY). LEV maintains a positive monotonic correlation with SHELTER, yet the
correlation becomes negative linearly. Overall, Table 3 indicates that collinearity among
tax avoidance methods employed in the analysis is generally low.
145
Table 3 Bivariate correlations of tax avoidance methods
Variable LEV NON_ROUTINE SHELTER LOSS TREAT HAVEN PERMDIFF TBTD CONFORM
Y
LEV 0.1021* 0.1735* 0.0766* –0.0480* – 0.0048
0.1796* 0.0892*
0.0706*
NON_ROUTINE 0.0487* 0.2345* 0.0285* 0.3282* 0.2221* –0.0809* –0.0141 0.0863*
SHELTER –0.1416* 0.0478* 0.0920* 0.1107* 0.0483* 0.1508* 0.0271* 0.2363*
LOSS 0.1019* 0.0194* 0.0905* 0.0255* 0.0123 0.1119* 0.0990* 0.0660*
TREATY 0.0255* 0.4352* –0.0970* – 0.1102*
0.0316* 0.2716* –0.0116
0.0642*
HAVEN 0.0002 0.3906* –0.0619* – 0.0485*
0.0144 0.1890* –0.0230*
0.0382*
PERMDIFF 0.0751* – – – –0.0323*
–0.0091 –0.0831* 0.2388*
0.0498* 0.0203* 0.3603*
TBTD 0.0734* – – –0.4150* –0.1712*
0.0133 –0.0175* 0.0159
0.0249* 0.0393*
CONFORM –0.0007 0.0498* 0.0057 –0.1347* –
–0.0943* 0.0417* 0.1830*
0.2910*
This table displays correlations between main variables. Spearman correlations are presented above the diagonal and Pearson correlations are presented
below. Asterisk (*) indicates correlation coefficients are statistically different from zero at one per cent confidence level. All variables are as defined in
Appendix A. TBTD and PERMDIFF are scaled by lagged total assets. TBTD and PERMDIFF are censored to –1 and 1. Similarly, NON_ROUTINE, TREATY,
and HAVEN are censored to 0 and 1.
146
RESULTS
The first hypothesis examines the long-run trend of firms’ financing choices. Considering the
less importance of tax incentives in firms financing decisions and the existence of various non-
debt tax shields, we predict that the use of debt financing tends to decline over time. Table 4
presents the results of regressing LEV to a time trend variable, TIME. It reports a downward
trend of LEV during observation periods, as predicted. The slope of TIME is –0.0027 (p<0.01),
indicating a linear decline of debt financing by about 0.0027 percentage points per year.
Accordingly, Figure 1 visualises this finding by plotting the mean of LEV for foreign affiliates
and purely domestic subsamples for each year from 2009 to 2017. The figure clearly shows
downward trends of LEV for both subsamples throughout the observation period.
Table 4 Regression analysis of firms’ financing options over time

Variable Predicted sign Coefficient t-statistic
TIME – –0.0027 –5.11***
SIZE + 0.0133 23.62***
ROA – –0.1710 –21.31***
FIXED + 0.1608 21.53***
INVEN + 0.0177 2.39**
FOREIGN + –0.6177 –2.01**
Constant –0.1652 –9.29***
Industry fixed effects Yes
N 23,736
𝑅2 12.21
This table reports coefficients of the following regression equation (2):
𝐿𝐸𝑉𝑖𝑡 = 𝛼 + 𝛽𝑇𝐼𝑀𝐸𝑖𝑡 + Σ𝛿𝑘 𝐶𝑂𝑁𝑇𝑅𝑂𝐿𝑆𝑖𝑡 + Σ𝛿𝑚 𝐼𝑁𝐷𝑈𝑆𝑇𝑅𝑌𝑚𝑡 + 𝜀𝑖𝑡
The equation includes SIZE, ROA, FIXED, INVEN, and FOREIGN as control variables. In
addition, we control for industry-fixed effects using two-digit of Indonesia’s standard
industry classification. All variables are as defined in Appendix A. ROA is censored to –1
and 1. Similarly, FIXED, INVEN, and FOREIGN are censored to 0 and 1. Untabulated
variance inflation factor (VIF) analysis shows individual VIF of the independent variables
is less than two suggesting low correlations between independent variables and, thus,
minimum risk of multicollinearity. Standard errors are robust and clustered by firm and
year. 𝑅2 is stated in percentage (%). The asterisk (*) indicates the statistical significance
of the coefficients at 1 per cent (***), 5 per cent (**), and 10 per cent (*) significance level,
respectively.
The effects of thin capitalisation rules are examined in the second hypothesis. Prior
studies (e.g. Buettner et al., 2012; Merlo et al., 2020) contend that the implementation of thin
capitalisation rules restricts the use of debt financing. Accordingly, we predict that the sample
firms tend to use less debt financing, thus more equity financing, after implementing this anti-
avoidance provision. Table 5 reports the results of our difference-in-differences estimations
using full sample firms (column (c)), also partitioned by foreign affiliate firms (column (d)) and
purely domestic firms (column (e)). Column (c) reveals a negative and significant (p<0.01)
relation between THINCAP and LEV, suggesting the use of debt financing is significantly
reduced following the implementation of thin capitalisation rules, as expected. The coefficient
of LEV is –0.0209, indicating a reduction of sample firms’ long-term debts by about 0.0209
percentage points after implementing the specific anti-avoidance rules.
Further, examining the effect of thin capitalisation rules across different firms’ ownership
characteristics, Column (d) and Column (e) report similar relations
between THINCAP and LEV as previously identified. However, the relevance and magnitude
are substantially reduced using a purely domestic firms sample, implying the effects of thin
capitalisation rules are more lasting for foreign affiliates firms.
147
Table 5 Difference-in-differences analysis on the effect of thin capitalisation rules
Variable Predicted Sample group
sign Full sample Foreign affiliates Domestic firms
(a) (b) (c) (d) (e)
THINCAP – –0.0209 –0.0209 –0.0154
(–4.53)*** (–3.73)*** (–1.83)*
SIZE + 0.0133 0.0116 0.0146
(12.39)*** (8.80)*** (8.70)***
ROA – –0.1708 –0.1993 –0.1048
(–15.56)*** (–15.26)*** (–5.51)***
FIXED + 0.1609 0.1729 0.1433
(11.93)*** (10.74)*** (6.63)***
INVEN + 0.0178 0.0170 0.0168
(1.39) (1.11) (0.81)
FOREIGN + –0.6188 –0.4065 –1.5699
(–1.31) (–0.76) (–2.20)
Constant –0.1659 –0.0946 –0.2481
(–4.77)*** (–2.14)** (–5.03)***
Industry fixed effects Yes Yes Yes
Year fixed effects Yes Yes Yes
N 23,736 15,848 7,888
𝑅2 12.22 12.69 17.68
𝐿𝐸𝑉𝑖𝑡 = 𝛼 + 𝛽𝑇𝐻𝐼𝑁𝐶𝐴𝑃𝑖𝑡 + Σ𝛿𝑘 𝐶𝑂𝑁𝑇𝑅𝑂𝐿𝑆𝑖𝑡 + Σ𝜃𝑙 𝐼𝑁𝐷𝑈𝑆𝑇𝑅𝑌𝑙𝑡 + Σ𝛾𝑚 𝑌𝐸𝐴𝑅𝑚𝑡 + 𝜀𝑖𝑡
The equation includes SIZE, ROA, FIXED, INVEN, and FOREIGN as control variables. In addition,
we control for year and industry-fixed effects using two-digit of Indonesia’s standard industry
classification. All variables are as defined in Appendix A. ROA is censored to –1 and 1. Similarly,
FIXED, INVEN, and FOREIGN are censored to 0 and 1. Untabulated variance inflation factor (VIF)
analysis shows individual VIF of the independent variables is less than four suggesting moderate
correlations between independent variables and, thus, low risk of multicollinearity. A foreign affiliate
is a firm where at least one of its shareholders is not domiciled in Indonesia. Standard errors are
robust and clustered by firms. 𝑅2 is stated in percentage (%). The asterisk (*) indicates the statistical
significance of the coefficients at 1 per cent (***), 5 per cent (**), and 10 per cent (*) significance level,
respectively.
The last hypothesis examines changes in firms’ tax planning after the implementation of
thin capitalisation rules. Consistent with the substitution effect hypothesis, we anticipate that
when thin capitalisations are restricted, firms tend to substitute debt financing with other viable
tax avoidance techniques. In investigating this hypothesis, following Efendi et al. (2021a,
2021b), we begin with transforming SHELTER into its residual form, SHELTER_RESID, by
regressing it to SIZE, a control variable. This orthogonalised-transformation is necessary to
correct structural multicollinearity problems caused by cross-correlations between these
variables (i.e. consistent with Wilson (2009), SIZE is one important component of SHELTER’s
construction) and improve the model’s precision. Further, we simultaneously regress
SHELTER_RESID and other alternative tax avoidance methods to LEV by estimating
Equation (4).
Table 6 presents the result of this estimation. It shows that LOSS, PERMDIFF,
and TBTD are positively and significantly (p<0.01) associated with LEV suggesting fiscal loss
carry-over, also permanent and temporary differences between accounting standards and tax
laws are complementing the use of interest deductions in minimising firms’ tax burdens.
Therefore, restricting interest deductions will systematically limit the use of these
accompanying tax avoidance techniques. Nevertheless, SHELTER_RESID is negatively
related (p<0.01) with LEV, indicating firms with non-debt tax shelters in operation use less
148
debt financing, consistent with similar shifting reported by Graham and Tucker (2006).59 This
finding supports the substitution effect hypothesis documented by earlier studies (e.g.
DeAngelo & Masulis, 1980; Dhaliwal et al., 1992; Graham & Tucker, 2006; MacKie‑Mason,
1990). However, inconsistent with MacKie-Mason (1990), we find a complementary effect
between fiscal loss carryforwards and the use of debt financing. Additionally, some costly tax
avoidance methods (e.g. NON_ROUTINE, TREATY, HAVEN) are irrelevant in the firms’
financing strategies, indicating firms’ incentive to maximise the net benefits of tax avoidance.
Table 6 Regression analysis of firms’ financing options and alternative tax avoidance
methods
Variable Tax avoidance method Predicted Coefficient t-statistic
sign
NON_ROUTINE Inter-jurisdictional income – 0.0099 0.70
shifting
SHELTER_RESID Tax sheltering – –0.3047 –30.55***
LOSS Loss carry-over – 0.0912 12.51***
TREATY Treaty-shopping – –0.0074 –0.57
HAVEN Income shifting to tax havens – 0.0094 0.33
PERMDIFF Non-conforming permanent tax – 0.6471 13.90***
avoidance schemes
TBTD Non-conforming temporary tax – 0.7725 14.33***
avoidance schemes
CONFORM Conforming tax avoidance + –3.8610 –18.55***
schemes
SIZE + 0.0087 9.29***
ROA – 0.1063 3.06***
FIXED + 0.0839 7.81***
INVEN + 0.0043 0.39
FOREIGN + 0.1663 0.67
Constant –0.0703 –2.48**
Industry fixed effects Yes
Year fixed effects Yes
N 8,300
𝑅2 58.65
𝐿𝐸𝑉𝑖𝑡 = 𝛼 + 𝛽1 𝑁𝑂𝑁_𝑅𝑂𝑈𝑇𝐼𝑁𝐸𝑖𝑡 + 𝛽2 𝑆𝐻𝐸𝐿𝑇𝐸𝑅𝑖𝑡 + 𝛽3 𝐿𝑂𝑆𝑆𝑖𝑡
+ 𝛽4 𝑇𝑅𝐸𝐴𝑇𝑌𝑖𝑡 + 𝛽5 𝐻𝐴𝑉𝐸𝑁𝑖𝑡 + 𝛽6 𝑃𝐸𝑅𝑀𝐷𝐼𝐹𝐹𝑖𝑡 + 𝛽7 𝑇𝐵𝑇𝐷𝑖𝑡 + 𝛽8 𝐶𝑂𝑁𝐹𝑂𝑅𝑀𝑖𝑡
+ Σ𝛿𝑘 𝐶𝑂𝑁𝑇𝑅𝑂𝐿𝑆𝑖𝑡 + Σ𝛿𝑚 𝐼𝑁𝐷𝑈𝑆𝑇𝑅𝑌𝑚𝑡 + Σ𝜃𝑛 𝑌𝐸𝐴𝑅𝑛𝑡 + 𝜀𝑖𝑡
The equation includes SIZE, ROA, FIXED, INVEN, and FOREIGN as control variables. In addition, we
control for year and industry-fixed effects using two-digit of Indonesia’s standard industry classification.
SHELTER_RESID is the residuals (𝜀𝑖𝑡 ) of regressing SIZE to SHELTER to lessen multicollinearity
problems, as follows:
𝑆𝐻𝐸𝐿𝑇𝐸𝑅𝑖𝑡 = 𝛼 + 𝛽1 𝑆𝐼𝑍𝐸𝑖𝑡 + 𝜀𝑖𝑡
All variables are as defined in Appendix A. TBTD, PERMDIFF, and ROA are censored to –1 and 1.
Similarly, NON_ROUTINE, TREATY, HAVEN, FIXED, INVEN, and FOREIGN are censored to 0 and
1. Untabulated variance inflation factor (VIF) analysis shows individual VIF of the independent variables
is less than four suggesting low correlations between independent variables and, thus, minimum risk
of multicollinearity. Standard errors are robust and clustered by firms. 𝑅2 is stated in percentage (%).
The asterisk (*) indicates the statistical significance of the coefficients at 1 per cent (***), 5 per cent (**),
and 10 per cent (*) significance level, respectively.
59
CONFORM is also negatively related to LEV. However, following Badertscher et al. (2019), CONFORM is an inverse
measure of corporate tax avoidance; therefore, the negative relation cannot be interpreted as a substitution effect.
149
Finally, the directions of association of firms’ internal characteristics (i.e. SIZE, ROA,
FIXED, INVEN, FOREIGN), as control variables, with LEV in nearly all regression estimations
are consistent with their theoretical predictions suggested by prior studies, implying that
several characteristics are informative in explaining firms’ financing strategies. Also, it reflects
the high comparability of our sample group with that of similar studies.
SENSITIVITY ANALYSIS
This section presents several additional analyses to evaluate the sensitivity of initial
findings on the effect of thin capitalisation rules on firms’ financing choices and alternative tax
avoidance techniques used when interest deductions are restricted. Firstly, we examine the
consistency of thin capitalisation rules’ effects across LEV’s distribution using quantile
regressions. Secondly, we assess firms’ tax planning changes over time using fixed-effect
panel data estimations. Finally, we evaluate the robustness of our difference-in-differences
estimations using propensity scores matching analysis.
4.1 Quantile regression analysis on the effect of thin capitalisation rules

Following the implementation of thin capitalisation rules which govern the maximum DER
allowed for income tax purposes, we evaluate the consistency of the intended effects of the
regulations across the overall LEV’s distribution, rather than focusing on the conditional mean
effects under conventional OLS estimations, using quantile regressions. Accordingly, LEV is
divided into eight quantiles to test the consistency in each quantile.
Recently, quantile regressions have been used to obtain evidence about points in the
distribution of the dependent variable further than the conditional mean. We utilise quantile
regressions to determine if the association of interest differs between ‘quantiles’ in the
conditional distribution of the dependent variable. Additionally, a quantile regression minimises
the total absolute deviation of the residuals, making it more robust to outliers than a standard
OLS. As a result, it can increase the accuracy of our prior estimates.
Further, we rerun Equation (3) using quantile regression specifications and present the
results of these estimations in Table 7. It reveals how thin capitalisation rules impact firms’
with long-term debts to total assets ratios above quantile (40), as intended. As can be seen in
Table 7, THINCAP is negative and significant (p<0.05) in quantile (50) to (90), consistent with
the regulations as the proportion of DER of more than four is not permitted and is subject to
further adjustment. Contrarily, firms with a lower portion of long-term debts (i.e. towards the
left tail of LEV’s distribution) appear not to be significantly affected by the rules as they satisfy
the statutory DER limitation.
4.2 Firms’ financing options and alternative tax avoidance methods over time
The earlier finding on the changes in firms’ tax planning reported in Table 6 may be
sensitive to any unobservable internal characteristics. To mitigate the potential concern that
control variables included in the previous estimations are not fully capturing the effects of
sample firms’ specific characteristics, we re-estimate regression Equation (4) using a fixed
effects-panel data specification. Furthermore, earlier studies (e.g. Griliches & Hausman, 1986;
Plesko, 2003) have asserted that panel data model, through analysing deviations of individual
means, is capable of correcting estimation bias caused by serial correlations of the
explanatory variables and omitted-individual effects. Nevertheless, the model may exacerbate
the negative correlations between errors in variables’ measurement and the residuals.
Table 8 reports the result of this estimation. It reveals a qualitatively comparable result
with the earlier finding reported in Table 6. Specifically, Table 8 presents a negative and
significant (p<0.01) association of SHELTER_RESID with LEV, confirming that the shifting of
excessive debt financing to non-debt tax shelters is not sensitive to any unobservable internal
characteristics. Similarly, LOSS, TBTD, and PERMDIFF remain positively related
to LEV, suggesting the complementary roles of these tax avoidance techniques to debt
financing persist over time. Additionally, HAVEN becomes positively and significantly (p<0.05)
150
associated with LEV, indicating that firms tend to use intra-group debt financing from entities
situated in tax havens over time.
151
Table 7 Quantile regressions on the effect of thin capitalisation rules
Variable Predicte LEV quantile
d sign (20) (30) (40) (50) (60) (70) (80) (90)
THINCAP – –0.0001 –0.0002 –0.0002 –0.0012 –0.0030 –0.0049 –0.0142 –0.0287
(–0.26) (–1.30) (–1.57) (–2.37)** (–2.30)** (–2.12)** (–2.81)*** (–3.54)***
SIZE + 0.0001 0.0024 0.0047 0.0068 0.0084 0.0109 0.0145 0.0093
(1.75)* (21.64)*** 35.62*** (44.16)*** (38.22)*** (34.03)*** (13.59)*** (5.26)***
ROA – –0.0001 –0.0021 –0.0093 –0.0278 –0.0539 –0.0973 –0.1925 –0.4037
(–0.35) (–4.66)*** (–9.79)*** (–13.34) (–17.43)*** (–18.01)*** (–19.05)*** (–26.97)***
FIXED + 0.0003 0.0093 0.0333 0.0902 0.1844 0.3020 0.4253 0.4821
(1.59) (12.24)*** (14.38)*** (17.04)*** (24.61)*** (32.04)*** (29.12)*** (27.27)***
INVEN + 0.0001 0.0015 0.0016 0.0011 –0.0011 –0.0054 0.0033 0.0258
(0.67) (4.43)*** (2.06)** (0.93) (–0.53) (–0.86) (0.26) (1.00)
FOREIGN + 0.2262 0.1432 0.2045 0.1427 0.0367 –0.1218 –0.1622 –0.5681
(2.10)** (4.04)*** (3.71)*** (0.88) (0.27) (–0.28) (–0.50) (–3.09)***
Constant –0.0021 –0.0562 –0.1086 –0.1542 –0.1811 –0.2216 –0.2521 0.0341
(–1.75)* (–22.17)*** (–36.26)*** (–47.49)*** (–36.39)*** (–32.29)*** (–9.75)*** (0.75)
N 23,736 23,736 23,736 23,736 23,736 23,736 23,736 23,736
Pseudo 𝑅2 0.00 0.70 2.60 4.29 6.27 8.62 10.55 11.54
This table reports coefficients of quantile regression specifications of the following quantile regression equation (3):
𝐿𝐸𝑉𝑖𝑞 = 𝛼 + 𝛽𝑇𝐻𝐼𝑁𝐶𝐴𝑃𝑖𝑡 + Σ𝛿𝑘 𝐶𝑂𝑁𝑇𝑅𝑂𝐿𝑆𝑖𝑡 + Σ𝜃𝑙 𝐼𝑁𝐷𝑈𝑆𝑇𝑅𝑌𝑙𝑡 + Σ𝛾𝑚 𝑌𝐸𝐴𝑅𝑚𝑡 + 𝜀𝑖𝑡
The equation includes SIZE, ROA, FIXED, INVEN, and FOREIGN as control variables. All variables are as defined in Appendix A. ROA is censored to –
1 and 1. Similarly, CAPITAL, INVEN, and FOREIGN are censored to 0 and 1. Standard errors are robust. Pseudo 𝑅2 is stated in percentage (%). t-statistics
are displayed in parentheses. The asterisk (*) indicates the statistical significance of the coefficients at 1 per cent (***), 5 per cent (**), and 10 per cent (*)
significance level, respectively.
152
Table 8 Panel data estimation of firms’ financing options and alternative tax
avoidance methods
Variable Tax avoidance method Predicted Coefficient t-statistic
sign
NON_ROUTINE Inter-jurisdictional income – 0.0121 1.47
shifting
SHELTER_RESID Tax sheltering – –0.1819 –87.20***
LOSS Loss carry-over – 0.0651 19.27***
TREATY Treaty-shopping – 0.0008 0.11
HAVEN Income shifting to tax havens – 0.0333 1.98**
PERMDIFF Non-conforming permanent tax – 0.2072 23.50***
avoidance schemes
TBTD Non-conforming temporary tax – 0.3151 24.43***
avoidance schemes
CONFORM Conforming tax avoidance + –2.1256 –40.14***
schemes
SIZE + 0.0163 10.93***
ROA – 0.1867 22.74***
FIXED + 0.0619 8.15***
INVEN + 0.0031 0.33
FOREIGN + –0.5677 –1.22
Constant –0.3124 –8.36***
N (group) 20,661
(3,711)
𝑅2 42.42
This table reports coefficients of fixed-effect panel data estimation of the following regression equation
(4):
𝐿𝐸𝑉𝑖𝑡 = 𝛼 + 𝛽1 𝑁𝑂𝑁_𝑅𝑂𝑈𝑇𝐼𝑁𝐸𝑖𝑡 + 𝛽2 𝑆𝐻𝐸𝐿𝑇𝐸𝑅𝑖𝑡 + 𝛽3 𝐿𝑂𝑆𝑆𝑖𝑡
+ 𝛽4 𝑇𝑅𝐸𝐴𝑇𝑌𝑖𝑡 + 𝛽5 𝐻𝐴𝑉𝐸𝑁𝑖𝑡 + 𝛽6 𝑃𝐸𝑅𝑀𝐷𝐼𝐹𝐹𝑖𝑡 + 𝛽7 𝑇𝐵𝑇𝐷𝑖𝑡 + 𝛽8 𝐶𝑂𝑁𝐹𝑂𝑅𝑀𝑖𝑡
+ Σ𝛿𝑘 𝐶𝑂𝑁𝑇𝑅𝑂𝐿𝑆𝑖𝑡 + Σ𝛿𝑚 𝐼𝑁𝐷𝑈𝑆𝑇𝑅𝑌𝑚𝑡 + Σ𝜃𝑛 𝑌𝐸𝐴𝑅𝑛𝑡 + 𝜀𝑖𝑡
The equation includes SIZE, ROA, FIXED, INVEN, and FOREIGN as control variables.
SHELTER_RESID is the residuals ( 𝜀𝑖𝑡 ) of regressing SIZE to SHELTER to lessen multicollinearity
problems, as follows:
𝑆𝐻𝐸𝐿𝑇𝐸𝑅𝑖𝑡 = 𝛼 + 𝛽1 𝑆𝐼𝑍𝐸𝑖𝑡 + 𝜀𝑖𝑡
All variables are as defined in Appendix A. TBTD, PERMDIFF, and ROA are censored to –1 and 1.
Similarly, NON_ROUTINE, TREATY, HAVEN, FIXED, INVEN, and FOREIGN are censored to 0 and 1.
𝑅2 is stated in percentage (%). The asterisk (*) indicates the statistical significance of the coefficients at
1 per cent (***), 5 per cent (**), and 10 per cent (*) significance level, respectively.
4.3 Propensity scores matching analysis

The previous finding on the effect of thin capitalisation rules on the use of debt financing
reported in Table 5 may be sensitive to any unobservable difference between pre and
post-implementation periods. To mitigate this potential concern, we re-estimate Equation
(3) using propensity score matching based on SIZE and ROA in multiple specifications
(i.e. nearest neighbour matching, radius matching, kernel matching, stratification
matching). Using propensity scores, we can match treated and non-treated observations
so they are comparable with respect to SIZE and ROA. One advantage of this non-
parametric approach is that it is a very intuitive avenue to estimate treatment effects
because we determine treated observations by a set of observable variables and then
compare them to relevant matches (observations that are not affected but similar in other
1
respects). Eventually, the propensity score can be explained as the probability of
receiving a treatment given pre-treatment characteristics.
Consistent with Wamser (2014), we employ multiple matching methods to identify
the effect of thin capitalisation rules. Nearest neighbour matching matches the control
unit with the closest propensity score to each treated observation. In contrast, radius
matching matches treatment units with control units only if the propensity score falls into
a particular range. The stratification method divides the sample into intervals based on
the value of the propensity score. Within each interval, treated and control units have, on
average, similar propensity scores. Finally, kernel matching does not only use some of
the control observations. The counterfactual result is, instead, created by this matching
estimator using weighted averages of all controls.
When evaluating various matching strategies, the trade-off between quantity and
quality of matches must be considered. Higher matching quality can lower the bias, while
additional information can increase the efficiency of the estimates. When compared to
nearest neighbour matching, radius matching, for instance, tends to steer clear of poor
matches. Since different matching algorithms imply a trade-off between bias and
efficiency, we report the results for the different methods.
Table 9 presents the result of these estimations. It reveals that, except for the
radius matching, the coefficient of THINCAP is negative and significant of thin
capitalisation rules on debt financing is negative and significant (p<0.1), consistent with
earlier findings, suggesting the role of thin capitalisation rules in restricting the use of
debt financing is robust notwithstanding any unobservable difference between pre and
post-implementation periods. Estimates range from –0.005 to –0.013, depending on the
matching method. A coefficient of –0.013 illuminates that treatment by the thin
capitalisation rules and related enforcements decreases firms’ long-term debts to total
assets by 1.3 percentage points.
Table 9 Propensity scores matching analysis

Matching After thin Before thin Average t-statistic
method capitalisation capitalisation treatment
rules rules effect
(N) (N)
Nearest 6,337 4,954 –0.007 –1.777* (0.004)
neighbour
Radius 6,337 17,521 –0.005 –1.620 (0.003)
Kernel 6,337 17,521 –0.005 –2.155** (0.002)
Stratification 6,337 17,521 –0.013 –2.508*** (0.003)
This table reports coefficients of the propensity score matching specifications of the following
regression equation (3):
𝐿𝐸𝑉𝑖𝑡 = 𝛼 + 𝛽𝑇𝐻𝐼𝑁𝐶𝐴𝑃𝑖𝑡 + Σ𝛿𝑘 𝐶𝑂𝑁𝑇𝑅𝑂𝐿𝑆𝑖𝑡 + 𝜀𝑖𝑡
The equation includes SIZE and ROA as control variables. All variables are as defined in
Appendix A. ROA is censored to –1 and 1. Standard errors are in parentheses. The kernel
matching reports bootstrap standard errors. The asterisk (*) indicates the statistical
significance of the coefficients at 1 per cent (***), 5 per cent (**), and 10 per cent (*)
significance level, respectively.
2
CONCLUSION AND FURTHER RESEARCH
This paper examines firms’ financing strategies over time and how they alter their capital
structure after being targeted by a stricter rule on thin capitalisations. We investigate a
set of confidential tax returns data of the largest firms operating in Indonesia from 2009
to 2017 to identify and test the time trend of firms’ financing choices, the effect of thin
capitalisation rules on firms’ financing options, and alternative tax avoidance techniques
used by the sample firms when thin capitalisations are too costly or no longer available.
Applying standard OLS, difference-in-differences, quantile regressions, fixed-effect
panel data estimations, and propensity scores matching, this study finds consistent
evidence of the negative effect of thin capitalisation rules on debt financings, as intended.
The findings also suggest that the effect is varied by ownership structure. Foreign
affiliated firms are mostly affected by these regulations rather than purely domestic firms.
Additionally, the time trend analysis reveals that debt financing generally decreases over
time, implying that firms are shifting their financing options to equity financing. Lastly,
since interest deductions are restricted by the new anti-avoidance rules, consistent with
the substitution effect hypothesis, firms are switching their tax planning using more
sophisticated and less-costly techniques. This study finds consistent shifting of firms’ tax
avoidance strategies to non-debt tax shelters after implementing thin capitalisation rules.
Viewed from a regulatory perspective, these findings suggest that to curb tax avoidance
practices optimally, thin capitalisation rules need to be implemented in concomitance
with specific regulations and enforcement activities that restrict the use of various non-
debt tax shelters. In addition, tax authorities need to target risky foreign affiliates firms
by further ruling the composition of debt financing allowed either using direct or hybrid
loans. Also, they may consider strengthening the deterrence effects of the regulations
by employing a more direct measure of excessive interest deductions (i.e. a fixed ratio
of interest payments to earning before interest, taxes, depreciation, and amortisation
(EBITDA)) as suggested by the G20 and OECD’s BEPS Action 4 (2016).60 Additionally,
to assess the medium and long-term impact on the economy, policymakers need to
consider the effect of thin capitalisation rules on firm’s cost of capital and, hence, on real
investments, in which a non-deductibility of interest expenses is comparable to an
additional tax that would risk crowding-out productive capitals, rather than tax bases.
Albeit investigating more reliable tax return data, our study is not without limitations. The
period of sample observation is relatively short for drawing a solid conclusion about
systematic changes in firms’ financing strategies. Hence, the trend needs to be
interpreted with caution. Further, the thin capitalisation rules post-implementation
periods observed in this research are relatively short. Therefore, the negative coefficients
of THINCAP cannot be interpreted as the true efficacy of the regulations but rather their
immediate effects on firms’ financing choices. Accordingly, further research with longer
post-implementation periods observation can examine the effect more accurately.
60
The proposed fixed ratio of interest payments and EBITDA is 10-30 per cent (2016).
3
Finally, it has not been clear whether the imposition of thin capitalisation rules is generally
beneficial for Indonesia, even though our results indicate that such rules can effectively reduce
tax incentives of debt financing and, thus, curb firms’ aggressive tax planning. Some studies have
elucidated that restricting opportunities for tax planning might result in adverse consequences
for multinationals’ investment and reinforce tax competition (e.g. Janeba & Smart, 2003;
UNCTAD, 2015). Thus, plausible future research may investigate whether imposing thin
capitalisation rules extends adverse effects on real investment in Indonesia. Also, we call for
more research on determining how specific regulations may affect firms’ long-term financing
strategies.
ACKNOWLEDGEMENT
The authors are thankful to both Director of Dissemination, Services, and Public
Relations and Director of Tax Data and Information of the Directorate General of Taxes
for providing confidential corporate income tax return data.
Appendix A Variable definitions

Variable Definition
𝐶𝑂𝑁𝐹𝑂𝑅𝑀𝑖𝑡 = A proxy for firm i’s conforming tax avoidance strategies suggested by
Badertscher et al. (2019), measured as the residuals of the following
regression equation:
𝑇𝐴𝑋𝐸𝑆𝑃𝐴𝐼𝐷_𝑇𝑂_𝐴𝑆𝑆𝐸𝑇𝑆𝑖𝑡 = 𝛼0 + 𝛼1 𝐵𝑇𝐷𝑖𝑡 + 𝛼2 𝑁𝐸𝐺𝑖𝑡 + 𝛼3 𝐵𝑇𝐷𝑖𝑡 X
𝑁𝐸𝐺𝑖𝑡 +𝛼4 𝑁𝑂𝐿𝑖𝑡 +𝛼5 ∆𝑁𝑂𝐿𝑖𝑡 + 𝜀𝑖𝑡
𝐹𝐼𝑋𝐸𝐷𝑖𝑡 = Total non-current assets of firm i scaled by total assets at the end of year t.
𝐹𝑂𝑅𝐸𝐼𝐺𝑁𝑖𝑡 = Total foreign income of firm i scaled by total assets at the end of year t.
𝐻𝐴𝑉𝐸𝑁𝑖𝑡 = Firm i’s total intra-group transactions with affiliates located in tax havens
scaled by total net sales at the end of year t.
𝐼𝑁𝑉𝐸𝑁𝑖𝑡 = Total inventory of firm i scaled by total assets at the end of year t.
4
Variable Definition
𝐿𝐸𝑉𝑖𝑡 = Financial leverage as a proxy for thin capitalisations, measured as total long-
term debts of firm i scaled by total assets at the end of year t.
𝐿𝑂𝑆𝑆𝑖𝑡 = A dummy variable for fiscal losses, coded one if firm i has fiscal loss
compensation at the end of year t and 0 otherwise.
∆𝐿𝑂𝑆𝑆𝑖𝑡 = Changes in affiliate i’s fiscal loss carry-forward of scaled by lagged total
assets at the end of year t.
𝑁𝑂𝑁_𝑅𝑂𝑈𝑇𝐼𝑁𝐸𝑖𝑡 = Total non-routine intra-group transactions (i.e. transfers of non-current and
financial assets; payments of royalties, interests, service fees, and other
expenses) of firm i scaled by total sales at the end of year t.
𝑃𝐸𝑅𝑀𝐷𝐼𝐹𝐹𝑖𝑡 = Permanent difference of firm i, measured as total permanent fiscal
adjustments scaled by total assets at the end of year t as suggested by Frank
et al. (2009).
𝑅𝑂𝐴𝑖𝑡 = Total pre-tax net income of firm i scaled by total assets at the end of year t.
𝑆𝐻𝐸𝐿𝑇𝐸𝑅𝑖𝑡 = A probability of firm i’s involvement in tax sheltering activities at the end of
year t proposed by Wilson (2009) as follows:
𝑆𝐻𝐸𝐿𝑇𝐸𝑅 = −4.30 + 6.63 ∗ 𝐵𝑇𝐷 − 1.72 ∗ 𝐿𝑒𝑣𝑒𝑟𝑎𝑔𝑒 + 0.66 ∗ 𝑆𝑖𝑧𝑒 + 2.26
∗ 𝑅𝑂𝐴 + 1.62 ∗ 𝐹𝑜𝑟𝑒𝑖𝑔𝑛𝐼𝑛𝑐𝑜𝑚𝑒 + 1.56 ∗ 𝑅&𝐷
However, we exclude research and development expenses from the variable
construction since they are not statistically significant in Table 5 of Wilson
(2009 p. 988).
𝑆𝐼𝑍𝐸𝑖𝑡 = Log natural of firm i’s total assets at the end of year t.
𝑇𝐵𝑇𝐷𝑖𝑡 = A temporary component of firm i’s book-tax differences, measured as total
temporary fiscal adjustments scaled by total assets at the end of year t.
𝑇𝐼𝑀𝐸𝑖𝑡 = A time trend variable, calculated by deducting the respective fiscal year with
2009 as the first year of the sample period.
𝑇𝐻𝐼𝑁𝐶𝐴𝑃𝑖𝑡 = A dummy variable for the application of thin capitalisation rules in Indonesia,
coded one if the fiscal year is 2016 or 2017 and 0 if the fiscal year is 2009 up
to 2015.
𝑇𝑅𝐸𝐴𝑇𝑌𝑖𝑡 = Firm i’s total intra-group transactions with affiliates domiciled in Indonesia’s
treaty-partner countries scaled by total net sales at the end of year t.
REFERENCES
Auerbach, A. J. (2002). Taxation and Corporate Financial Policy. Handbook of public
economics, 3, 1251-1292.
Badertscher, B. A., Katz, S. P., Rego, S. O., & Wilson, R. J. (2019). Conforming tax
avoidance and capital market pressure. The accounting review, 94(6), 1-30.
Blouin, J., Huizinga, H., Laeven, M. L., & Nicodème, G. (2014). Thin capitalization rules
and multinational firm capital structure: International Monetary Fund.
Buettner, T., Overesch, M., Schreiber, U., & Wamser, G. (2012). The impact of thin-
capitalization rules on the capital structure of multinational firms. Journal of Public
Economics, 96(11-12), 930-938.
Central Bureau of Statistics of Indonesia. (2015). Indonesia Standard Industrial
Classification. In.
Clemente-Almendros, J. A., & Sogorb-Mira, F. (2016). The effect of taxes on the debt
policy of spanish listed companies. SERIEs, 7(3), 359-391.
Conover, T. L., & Nichols, N. B. (2000). A further examination of income shifting through
transfer pricing considering firm size and/or distress. The International Journal of
Accounting, 35(2), 189-211.
5
De Simone, L., Klassen, K. J., & Seidman, J. K. (2017). Unprofitable affiliates and income
shifting behavior. The accounting review, 92(3), 113-136.
DeAngelo, H., & Masulis, R. W. (1980). Optimal capital structure under corporate and
personal taxation. Journal of Financial Economics, 8(1), 3-29.
Desai, M. A. (2003). The divergence between book income and tax income. Tax policy
and the economy, 17, 169-206.
Dhaliwal, D., Trezevant, R., & Wang, S.-w. (1992). Taxes, investment-related tax shields
and capital structure. Journal of the american taxation Association, 14(1), 1-21.
Dharmapala, D., & Hines Jr, J. R. (2009). Which countries become tax havens? Journal
of Public Economics, 93(9-10), 1058-1068.
Dyreng, S. D., Hanlon, M., Maydew, E. L., & Thornock, J. R. (2017). Changes in
corporate effective tax rates over the past 25 years. Journal of Financial
Economics, 124(3), 441-463.
Efendi, S. (2020). Covid-19 and corporate tax avoidance: Measuring long-run tax
burdens as an alternative bailout test. Kajian Ekonomi & Keuangan, 4(3), 216-
244.
Efendi, S., Czernkowski, R., Morton, E., & Bond, D. (2021a). Cross-jurisdictional tax
arbitrage and the flexibility of transfer pricing aggressiveness: New evidence from
foreign subsidiaries’ tax return data. Working paper. Accounting Discipline
Group, University of Technology Sydney Business School. Retrieved from
http://hdl.handle.net/10453/159781
Efendi, S., Czernkowski, R., Morton, E., & Bond, D. (2021b). Examining long-run
corporate tax avoidance: How firms avoid taxes over time? Working paper.
Accounting Discipline Group, University of Technology Sydney Business School.
Retrieved from http://hdl.handle.net/10453/159781
Ernst & Young. (2013). Global Transfer Pricing Survey. Retrieved from
https://www.ey.com/en_au/transfer-pricing-international-tax-survey
Ernst & Young. (2016). Transfer Pricing Survey Series. Retrieved from
https://www.ey.com/en_au/transfer-pricing-international-tax-survey
European Union. (2021). The EU List of Non-cooperative Jurisdictions for Tax Purposes.
Retrieved from https://eur-lex.europa.eu/legal-
content/en/TXT/PDF/?uri=uriserv:OJ.C_.2021.066.01.0040.01.ENG
Frank, M. M., Lynch, L. J., & Rego, S. O. (2009). Tax reporting aggressiveness and its
relation to aggressive financial reporting. The accounting review, 84(2), 467-496.
Graham, J. R. (2000). How big are the tax benefits of debt? The journal of finance, 55(5),
1901-1941.
Graham, J. R. (2003). Taxes and corporate finance: A review. The Review of Financial
Studies, 16(4), 1075-1129.
Graham, J. R., & Tucker, A. L. (2006). Tax shelters and corporate debt policy. Journal of
Financial Economics, 81(3), 563-594.
Griliches, Z., & Hausman, J. A. (1986). Errors in variables in panel data. Journal of
econometrics, 31(1), 93-118.
Gupta, S., & Newberry, K. (1997). Determinants of the variability in corporate effective
tax rates: Evidence from longitudinal data. Journal of accounting and public
policy, 16(1), 1-34.
Hanlon, M., & Heitzman, S. (2010). A review of tax research. Journal of accounting and
Economics, 50(2-3), 127-178.
Hanlon, M., & Slemrod, J. (2009). What does tax aggressiveness signal? Evidence from
stock price reactions to news about tax shelter involvement. Journal of Public
Economics, 93(1-2), 126-141.
Hopland, A. O., Lisowsky, P., Mardan, M., & Schindler, D. (2018). Flexibility in income
shifting under losses. The accounting review, 93(3), 163-183.
6
International Monetary Fund. (2010). The IMF’s Report to the G-20 and Background
Material: Financial Sector Taxation. In: IMF.
Janeba, E., & Smart, M. (2003). Is targeted tax competition less harmful than its
remedies? International Tax and Public Finance, 10(3), 259-280.
Kinney, M. R., & Swanson, E. P. (1993). The accuracy and adequacy of tax data in
Compustat. The Journal of the American Taxation Association, 15(1), 121.
Lisowsky, P. (2010). Seeking shelter: Empirically modeling tax shelters using financial
statement information. The accounting review, 85(5), 1693-1720.
Lisowsky, P., Robinson, L., & Schmidt, A. (2013). Do publicly disclosed tax reserves tell
us about privately disclosed tax shelter activity? Journal of Accounting Research,
51(3), 583-629.
MacKie‑Mason, J. K. (1990). Do taxes affect corporate financing decisions? The journal
of finance, 45(5), 1471-1493.
Merlo, V., Riedel, N., & Wamser, G. (2020). The impact of thin‑capitalization rules on the
location of multinational firms’ foreign affiliates. Review of International
Economics, 28(1), 35-61.
Mills, L. F., Newberry, K. J., & Novack, G. F. (2003). How well do Compustat NOL data
identify firms with US tax return loss carryovers? Journal of the american taxation
Association, 25(2), 1-17.
Modigliani, F., & Miller, M. H. (1958). The cost of capital, corporation finance and the
theory of investment. The American economic review, 48(3), 261-297.
Myers, S. C. (2001). Capital structure. Journal of Economic perspectives, 15(2), 81-102.
Organisation for Economic Co-operation and Development. (2000). Towards Global Tax
Cooperation: Progress in Identifying and Eliminating Harmful Tax Practices.
Retrieved from Paris, France: https://www.oecd.org/ctp/harmful/2090192.pdf
Organisation for Economic Co-operation and Development. (2013). Action Plan on Base
Erosion and Profit Shifting. Retrieved from
https://www.oecd.org/ctp/BEPSActionPlan.pdf
Organisation for Economic Co-operation and Development. (2016). Limiting Base
Erosion Involving Interest Deductions and Other Financial Payments, Action 4 -
2016 Update.
Overesch, M., & Wamser, G. (2010). Corporate tax planning and thin-capitalization rules:
evidence from a quasi-experiment. Applied Economics, 42(5), 563-573.
Plesko, G. A. (2003). An evaluation of alternative measures of corporate tax rates.
Journal of accounting and Economics, 35(2), 201-226.
Rego, S. O. (2003). Tax‑avoidance activities of US multinational corporations.
Contemporary Accounting Research, 20(4), 805-833.
Scholes, M. S., Wolfson, M. A., Erickson, M., Maydew, E., & Shevlin, T. (2014). Taxes &
business strategy: Prentice Hall Upper Saddle River, NJ.
Slemrod, J. (2009). Lessons for tax policy in the Great Recession. National Tax Journal,
387-397.
Thomsen, M., & Watrin, C. (2018). Tax avoidance over time: A comparison of European
and US firms. Journal of International Accounting, Auditing and Taxation, 33, 40-
63.
U.S. Senate. (2011). Stop Tax Haven Abuse Act. Retrieved from
https://www.congress.gov/bill/112th-congress/senate-bill/1346/related-
bills?r=98&s=1
United Nations Conference on Trade and Development. (2015). World Investment
Report: Reforming International Investment Governance. Retrieved from
https://unctad.org/en/PublicationsLibrary/wir2015_en.pdf
7
Wamser, G. (2014). The impact of thin‑capitalization rules on external debt usage–a
propensity score matching approach. Oxford Bulletin of Economics and
Statistics, 76(5), 764-781.
Wilde, J. H., & Wilson, R. J. (2018). Perspectives on corporate tax planning:
Observations from the past decade. The Journal of the American Taxation
Association, 40(2), 63-81.
Wilson, R. J. (2009). An examination of corporate tax shelter participants. The
accounting review, 84(3), 969-999.
8
TAX AMNESTY AMIDST THE CORONAVIRUS
PANDEMIC IN INDONESIA, MALAYSIA, AND THE
PHILIPPINES: A COMPARATIVE ANALYSIS
Rosa Vanda Melaniaa
a Direktorat Jenderal Pajak, Indonesia. Email: rosa.vanda@pajak.go.id
ABSTRACT
Tax Amnesty Amidst the Coronavirus Pandemic in Indonesia, Malaysia, and the Philippines: A
Comparative Analysis. Amidst the coronavirus pandemic, 3 (three) countries in South East Asia, namely
Indonesia, Malaysia, and the Philippines implemented tax amnesty program. Even though the program is often
seen as a revenue-raising tool, apparently, gaining additional tax revenue is not the only reason behind the
implementation of the program in those countries. By utilizing literature review approach, this research aims to
provide comparative analysis of tax amnesty implementation in those countries, including the objectives, basis of
implementation, the program target, taxes covered, and whether the programs fulfill the principles of voluntary
disclosure outlined by OECD. It is found that the design of tax amnesty in each country is quite distinctive.
However, some similarities on the benefit offered and consequences of non or partial compliance are identified.
In addition, this study found that even though the objectives and terms of amnesty program has been clearly
stated, Malaysia and the Philippines offer leniency by extending the program period. Finally, further research is
required to study the impact of tax amnesty program on tax revenue and taxpayer compliance in each country.
Keywords: tax amnesty, voluntary disclosure, tax compliance, comparative
Analisis Perbandingan Pengampunan Pajak Saat Pandemi Coronavirus di Indonesia, Malaysia, dan Filipina. Di
tengah terjadinya pandemi coronavirus, 3 (tiga) negara di Asia Tenggara, yaitu Indonesia, Malaysia, dan Filipina
menerapkan program pengampunan perpajakan. Meskipun program tersebut seringkali disebut sebagai sarana
untuk meningkatkan penerimaan negara, peningkatan penerimaan pajak tidak menjadi satu-satunya alasan
diterapkannya program pengampunan pajak di negara-negara tersebut. Dengan menggunakan metode kajian
literatur, studi ini bertujuan untuk memaparkan analisis perbandingan mengenai implementasi program
pengampunan pajak di tiap negara tersebut, termasuk dari sisi tujuan, dasar implementasi, target program, ruang
lingkup program, dan apakah program pengampunan pajak yang diterapkan memenuhi prinsip-prinsip yang
dijelaskan oleh OECD. Hasil dari studi ini menunjukkan bahwa desain dari setiap program pengampunan pajak di
masing-masing negara cukup berbeda satu sama lain. Namun demikian, beberapa kesamaan dapat diidentifikasi,
yaitu mengenai insentif yang ditawarkan dan konsekueansi jika wajib pajak tidak sepenuhnya jujur dalam
mengikuti program tersebut. Studi ini juga menunjukkan bahwa meski tujuan dan persyaratan mengikuti program
telah ditetapkan secara jelas, Malaysia dan Filipina menetapkan perpanjangan periode program. Akhir kata, perlu
adanya studi selanjutnya utnuk mempelajari efek program pengampunan pajak terhadap penerimaan dan
kepatuhan pajak di masing-masing negara.
Kata kunci: amnesti pajak, pengungkapan sukarela, kepatuhan perpajakan, perbandingan
1. INTRODUCTION
Many countries have implemented tax amnesty program, or voluntary disclosure
as mentioned by OECD on its guidance, for various reasons. First, governments expect
to gain an immediate, additional tax revenue using this program. Second, governments
may expect to increase future compliance of taxpayers. Third, governments may utilize
this program for certain purposes, for instance, encouraging repatriation of capital and
improving domestic investments (Baer & Le Borgne, 2008). Other reasons to implement
tax amnesty includes the reduction of administrative cost and reduction the authority
burden of tax cases to be investigated or audited (Sabnita, 2020).
9
Due to coronavirus pandemic that started to hit across the world in early 2020,
many countries face economic crisis because the pandemic has caused various activities
to put in a halt. Government spending is increased, mostly on healthcare expenditure
and support relief for affected business and individuals, while tax revenue decreased, as
economic activities were limited due to containment measures and government provided
tax exemption and facilities to ease the burden of taxpayers.
Tax amnesty is sometimes implemented during crisis in order to generate
immediate tax revenue. Even though the pandemic has not ended yet, current situation
has been better and countries start to enter recovery phase. Amidst the coronavirus
pandemic, 3 (three) countries in South East Asia, which are Indonesia, Malaysia, and
the Philippines has implemented tax amnesty program. Even though the program is often
seen as a revenue-raising tool, apparently, gaining additional tax revenue is not the only
reason behind the implementation of the program in those countries.
This paper will discuss the implementation of tax amnesty in 3 (three) countries in
Southeast Asia, which is Indonesia, Malaysian and the Philippines, amidst the
coronavirus pandemic. This research aims to provide comparative analysis of tax
amnesty implementation in those countries, including the objectives, basis of
implementation, the program target, taxes covered, and whether the programs fulfill the
principles of voluntary disclosure outlined by OECD.
2. TAX AMNESTY: DEFINITION, TYPES, AND PRINCIPLES

2.1 Definition
Tax amnesty discussed in this paper includes broad term such as tax pardon, tax
remission, and voluntary disclosure. Baer and Le Borgne (2008) defines tax amnesty as:
a limited-time offer by the government to a specified group of taxpayers to pay a

defined amount, in exchange for forgiveness of a tax liability (including interest and
penalties), relating to a previous tax period(s), as well as freedom from legal
prosecution.
Meanwhile, OECD (2015) uses the term of voluntary disclosure program, defined
as “opportunities offered by tax administrations to allow previously non-compliant
taxpayers to correct their tax affairs under specified terms”.
Tax amnesty programs typically offer sanction dispensation on financial and/or
legal aspect of taxation. In financial aspect, government usually grants reduction of tax
liabilities, which can be in form of decreasing or eliminating interest or penalties, or
reducing the initial tax liabilities. Whereas, in legal aspect, government typically offer a
waiving of civil and criminal penalty (Baer & Le Borgne, 2008). In several countries,
taxpayers who participate in the program will not be subject to tax audit for previous
unreported taxable income or tax returns. Another applied approach is a guarantee that
their tax returns in future periods will not be subject to tax audit, under certain term and
conditions, such as it must be through a standard selection process, be depending on
the capacity of the tax authority to perform tax audit, and tax audit will be performed
nonetheless if serious case like fraud is suspected.
2.2 Design Features of Tax Amnesty

Typical tax amnesty that have been implemented have common design features,
in term of targeted taxpayers, the scope of tax being covered in the program, the amnesty
(benefit for taxpayers), and the duration. Pellechio (1993) outlines the possible design
features, which includes basis of implementation, program target, taxes covered,
program duration/ period, and incentives offered, as explained on the Table below.
10
11
Table 1 Possible Design Features of Tax Amnesty
Source: Adapted from Tax Amnesties (Pellechio, 1993)
2.3 Principles of Successful Tax Amnesty Program
With prudent design, tax amnesty programs could be beneficial, not only for the
participated taxpayers, but also for the government. OECD (2015) outlines several
principles for successful program as following.
1. Be clear about the program objectives and terms.
Tax amnesty should be clear on the program objectives, for instances, addressing
the problem of non-compliance taxpayers, reducing the load of tax audit cases or
tax investigation cases on court for certain period, and providing opportunities for
taxpayers to improve their tax administration.
Terms of a tax amnesty program should be designed as specific as possible. If the
program is time-limited, the duration of the program must be clearly stated. How
taxpayers could participate in the program, the benefits or privilege gained, data
protection, and also any exception of participation need to be clearly explained.
Hence, authority must provide guidelines that are easily accessible and
communicate the program’s terms effectively to the public. In term of the covered
taxes, it is better if tax amnesty program includes all taxes, since the taxpayers’
situation could be fully captured.
2. Deliver evident and cost-effective increases of revenue.
The result of tax amnesty should be evident and cost-effective, meaning that the
result is demonstrable and the administrative or implementation cost is considerable
effective. In measuring the cost, in addition to the program cost, government should
include the opportunity costs of revenue that could be captured in case there is no
tax amnesty program. However, the result or benefit is not only about revenue, but
also the increasing level of compliance, including the compliance of compliant
taxpayers.
3. Be consistent with the generally applicable compliance and enforcement regime
The terms of tax amnesty program should be attractive yet be in line with ongoing
compliance and enforcement arrangement. While authorities may aim to counter
non-compliance using this program, they should also be careful not to damage the
overall voluntary compliance.
4. Help to deter non-compliance.
12
Tax authority needs to show that identification of non-compliance activities and the
resolution are part of the tax amnesty program. Taxpayers who persist in being non-
compliant should be subject to a more serious consequences, such as higher
interests or penalties.
5. Improve levels of compliance among the population eligible for the program.
Tax amnesty program should be designed to give increased risk for those eligible to
participate but choosing not to do so, yet they will be detected anyway, and the
program should give evident incentives to encourage participatory. Tax authority
needs to inform that the consequences is harsher for those who are eligible to
participate but choosing not to do so and for those who participated but only made
partial disclosure. It needs to be easy for taxpayers to come forward and make
disclosure in the program. Nevertheless, a repetitive amnesty program with similar
terms, may harm the rate of compliance.
6. Complement the immediate yield from disclosures with measures that improve
compliance in the longer-term.
The role of tax amnesty as revenue-raising tools is seen as controversial, since the
effect on the future compliance is arguably (Alm, Martinez-Vazquez, & Wallace,
2009). The increasing revenue from the program should not sacrifice long-term
compliance, thus, the program must have credibility that the “special terms” offered
is time-limited and serious consequences may avail if taxpayers missed this
opportunity. Evidence shows that tax amnesty program used to reinforce other
measures of ensuring tax compliance can avoid the risk of negative trend of long-
term tax compliance. For instances, tax amnesty can be used to accompany critical
changes in tax administration, such as the implementation of automatic exchange
of information. In other countries, tax amnesty has been initiated when new
governmental administration took over. Furthermore, many other programs
implement harsher sanctions after the amnesty period.
This paper utilizes a qualitative method using literature review. Secondary data
were obtained from various sources, such as reports, news, articles, and press release
from government websites, and tax regulation databases. Relevant academics papers
were also collected from various scholarly databases, mainly ScienceDirect, SSRN and
Google Scholar.
This paper findings will be comprised of data and information collected from
sources mentioned above. First, the overview the implementation of tax amnesty in
Indonesia, Malaysia, and the Philippines will be described. Next, this paper will discuss
the features of each program based on features covered by Pellechio (1993) and identify
whether the programs satisfy every principle of a successful voluntary disclosure
program explained by OECD.
4. RESULTS/FINDINGS
4.1 Voluntary Disclosure Program in Indonesia
On January 2022, Indonesia implemented Voluntary Disclosure Program (VDP),
which stipulated on chapter 5, article 5, of Law Number 7/2021 on Tax Regulation
Harmonization. Further, the Ministry of Finance (MoF) issued Regulation of the Minister
of Finance (PMK) Number 196/PMK.03/2021 regarding Implementation Procedure of
Voluntary Disclosure Program. The program was held for the period between 1 January
2022 until 30 June 2022.
According to DGT, VDP is a program to increase taxpayers’ voluntary compliance
and being implemented based on the principle of simplicity, legal certainty, and
13
practicality. This program provides an opportunity for taxpayers who participated on Tax
Amnesty 2016-2017 to disclose unreported assets and individual taxpayers to voluntarily
disclose assets that have not been reported in annual income tax returns fiscal year
2016-2020. One factor that becomes the background of this program is the
implementation of automatic exchange of information (AEOI) and tax information from
agencies, institutions, associations, and other parties (ILAP), thus, taxpayers are given
opportunity to come forward and fulfill their tax obligation.
In order to participate, taxpayers need to disclose their unreported net assets (the
value of assets subtracted with debt/liability related to the asset acquisition), then pay
final income tax based on the value of the net assets, on condition that DGT has not
found data/information regarding the assets being disclosed. Taxpayers are considered
participating in the program when they received certificate after completing the procedure.
There are 2 (two) types of policy and several final income tax rates that applied in
this program as described on the table x below.
Table 2 Subject and Final Income Tax Rate of Voluntary Disclosure Program in
Indonesia
Source: Summarized from PMK-196/PMK.03/2021
Policy I Policy II
Subject Individual Taxpayers and Individual Taxpayers
Corporate Taxpayers, who
had participated on Tax
Amnesty 2016-2017
Rate of Final Income Tax 11% for declaration of assets 18% for declaration of assets
located overseas located overseas
8% for repatriation of asset 14% for repatriation of asset
located overseas and located overseas and
declaration of assets in declaration of assets in
Indonesia Indonesia
6% for repatriation of asset 12% for repatriation of asset
located overseas and located overseas and
declaration of assets in declaration of assets in
Indonesia, which will be Indonesia, which will be
invested on: invested on:
 Government Securities; or  Government Securities; or
 business activities in the  business activities in the
natural resource sector or natural resource sector or
renewable energy sector. renewable energy sector.
Assets acquirement that Net assets acquired from Net assets acquired from
can be disclosed in this 1985-2015 which were not 2016-2020 which were not
program disclosed under Tax Amnesty disclosed in Tax Return 2020
2016-2017 and are still owned by
taxpayers at the end of 2020
fiscal year.
Taxpayers participated in Voluntary Disclosure Program would gain several

benefits. First, they will enjoy significant lower rate of income tax compared to the usual
income rate, especially for those who will invest their disclosed assets. For comparison
with the VDP tariff explained above, the current rate of corporate income tax (CIT) is 25%
and the highest rate of personal income tax (PIT) is 30%.
Second, taxpayers who participated on Policy I would not be subject to
administrative sanction of 200% for net assets that were undisclosed during Tax
14
Amnesty 2016-2017. Meanwhile, taxpayers participated on Policy II would be exempted
from any administration sanction and tax audit on any tax obligation for fiscal year of
2016-2020, unless there is new data or information of undisclosed assets. Furthermore,
for both policies, data and information declared in this program cannot be used as a
basis for tax investigation and/or criminal prosecution.
Final income tax rates in VDP program are varied depends on taxpayers’
commitment, they can choose to disclose their net asset, repatriate, or invest it.
Repatriation of assets can be done by transferring their assets through bank(s) in
Indonesia in accordance with the provisions of laws and regulations in banking sector.
The assets must be transferred into Indonesia at the latest September 30, 2022.
Taxpayers should keep the assets in Indonesia for a minimum period of 5 (five) years
from the date of certificate issuance. This condition also applies for taxpayers who
declare their net assets located in Indonesia.
For taxpayers who participated in the program and stated that the net assets will
be invested, the disclosed assets can only be invested through certain schemes, which
are investment through government securities, and/or investment through business
activities in natural resources processing industry or renewable energy sector in
Indonesia. The investment must be done at the latest on September 30, 2023 and must
be kept for a minimum period of 5 (five) years since the time it was invested. However,
taxpayers are allowed to change their investment scheme twice during the period of
investment.
The participated taxpayers are required to report the asset repatriation and/or
investment to the DGT by submitting a realization report electronically. Taxpayers who
failed to repatriate and/or invest their net assets will be imposed an additional final
income tax.
4.2 Voluntary Disclosure and Amnesty Program in Malaysia
The Royal Malaysian Customs Department (RMCD) introduced Special Voluntary

Disclosure and Amnesty Program (VA Program) that provides an opportunity for
taxpayers to voluntarily disclose their indirect tax liabilities, including duty, tax, levy,
penalty and/or surcharge, outstanding on or before 31 October 2021. A guideline of VA
Program for Indirect Taxes was published by RMCD on 31 December 2021. Various
indirect taxes covered in this program are:
1) Import and export duty;
2) Sales tax;
3) Service tax, including that on imported services and digital services;
4) Excise duty;
5) Goods and services tax;
6) Tourism tax; and
7) Departure levy.
Malaysia government implements this program to reduce tax leakages eroding
national revenue, and encourage taxpayers, whether individuals or companies, to
improve their focus on tax governance, by offering various incentives. The incentives
provided in this program will be a one-off offer.
VA Program Participation Letter (VA2) will be issued by RMCD to confirm taxpayer
eligibility to participate. The VA Program comprises of 2 (two) categories, namely
Voluntary Disclosure Program and Amnesty Program.
1) Voluntary Disclosure
This program provides opportunity for taxpayers to report any errors unknown or
undiscovered by RMCD. This program is available for taxpayers that have
underpayment of indirect tax under the relevant indirect tax regulation, but Bill of
15
Demand (BOD) has not been issued yet by the RMCD. Taxpayers who eligible to
participate comprises of:
a) Companies/individuals who have registered and performed the following
activities:
(1) Submitted NIL return incorrectly
(2) Not submitted any return/declaration when required
(3) Under declared and underpaid duty/tax/levy
(4) Charged tax (e.g. sales tax, service tax or GST) on goods/services that are
not subject to tax
b) Companies/individuals who are liable to be registered or licensed but are not
registered/licensed, under the relevant Acts, and have collected tax/levy but not
paid tax/levy to RMCD;
c) Companies/individuals who are liable to be registered or licensed but are not
registered/licensed, under the relevant Acts, and did not collect the tax/levy from
customers
d) Companies/individuals who are not liable to be registered or licensed, under the
relevant Acts, but have collected tax/levy and not paid to RMCD
e) Companies/individuals who acquired imported taxable services under the GST
Act and Services Act but have not accounted for tax payable to RMCD
f) Companies licensed/registered under the Sales Tax Act that declare sales tax
not in accordance with the sales tax assessment, involving sales between
related parties
g) Companies/individuals who obtain duty/tax/levy facilities/exemptions but did not
comply with the conditions
h) Importer/ exporter/ local excise manufacturer/ licensed manufacturing
warehouse/ license warehouse that
(1) Underpaid of duty/tax during import and export declaration;
(2) Used the incorrect tariff code when declaring the goods
(3) Used an exemption facility in error
(4) Made amendments to the cost of goods in determining the excise value
under section 2 of the Excise Act
i) Importers who are connected to Multinational Enterprise (MNE)
j) Companies/individuals who has not been audited but earmarked for an audit in
2022
k) Companies/individuals who being audited
l) Companies/individuals who has been audited
2) Amnesty Program
Amnesty program is targeted for taxpayers whose offenses have been identified by
RMCD and a Bill of Demand (BOD) has been issued. Any company/individual that
has been issued a BOD on or before 31 August 2022 for the liability period on or
before 31 October 2021 are eligible to participate in this program. BOD must be
issued on or before 31 October 2021. The calculation of incentive will be based the
outstanding balance of BOD on the date of the application is received. BOD that has
been registered in court and has obtained a judgement is not eligible for the Amnesty
Program.
VA Program is divided into 2 (two) phases. The period of Phase 1 is from 1 January
2022 until 30 June 2022, while Phase 2 is running from 1 July 2022 until 30 September
2022. The participation in each phase is indicated by the completion of relevant payment.
Incentives offered by the VA program comprises of:
1) Penalty and tax remission incentives will be given on a blanket approval without the
need to submit application
16
2) Audits will not be conducted for the activity and period involved
3) Information related to the disclosure will be kept confidential
4) A minimum compound is imposed for each voluntary disclosure under the same Act
5) Enforcement actions will be stopped once all BODs are fully paid (for Amnesty
program)
Taxpayers who participated in Phase 1 will enjoy more significant potential tax
savings compared to those participated in Phase 2. In Phase 1, tax incentives given is
up to 30%, while in Phase 2, tax incentives given is up to 15%. Furthermore, in Phase 1,
the participated taxpayers are given 90% and 100% of penalty incentives, and in Phase
2, they are given 80% and 50% penalty incentives. However, due to strong interest on
the program, Malaysia government decided to give extension to taxpayers who wish to
participate in Phase 1, so they were allowed to make payment by 14 July 2022, as long
as the application has been submitted by 30 June 2022.
Table 3 Phase and Incentives of VA Program in Malaysia

Source: Summarized from Guideline of VA Program provided by RMCD
Phase 1 Phase 2
Period of VA Program 1 January 2022 – 30 June 2022 1 July 2022 – 30 September
(6 months) 2022
(3 months)
Tax Incentives Up to 30% Up to 15%
Penalty Incentives 90% and 100% 80% and 50%
4.3 Tax Amnesty in The Philippines
The Philippines has implemented multiple tax amnesty over the years. On
February 2019, The Philippines government issued Republic Act (RA) 11213 (Act
Enhancing Revenue Administration and Collection by Granting an Amnesty on All
Unpaid Internal Revenue Taxes Imposed by The National Government for Taxable Year
2017 and Prior Years with Respect to Estate Tax, Other Internal Revenue Taxes, and
Tax on Delinquencies), which became effective on 24 April 2019. This act is issued in
order to protect and enhance revenue administration and collection, and make the
country’s tax system more equitable, by simplifying the tax compliance requirements.
This act regulates 2 (two) programs, namely Estate Tax Amnesty (ETA) and Tax
Amnesty on Delinquencies (TAD).
This act was actually implemented prior to the coronavirus pandemic. However,
the Philippines government extended the program in order to give more opportunity for
taxpayers due to the pandemic.
Furthermore, on September 2020, the Department of Finance (DOF) issued RR
21-2020 that regulates the Voluntary Assessment and Payment Program (VAPP) in
order to collect additional revenue to fund government expenditure during the
coronavirus pandemic. VAPP allows qualified taxpayers to settle unpaid taxes (internal
revenue taxes), including on one-time transactions (ONETT), which could otherwise be
collected through audit or enforcement effort, and would be provided certain privileges.
1) Estate tax amnesty

This program is a one-time opportunity to settle estate tax obligation, that will
give reasonable tax relief to estates with deficiency estate taxes. Initially, estate tax
amnesty will be run for the period of 2 (two) years, which ended on 15 June 2021,
as stated in Section 6 RA 11213. Further, Revenue Regulation (RR) No. 06-2019
regulates the implementation of the provisions of Estate Tax Amnesty. The
Philippines government extends the program by the stipulation of RA 11569 which
17
amends Section 6 of RA 11213. Hence, the deadline of estate tax amnesty program
is currently on 14 June 2023. The extension of estate tax amnesty program is further
regulated on RR No. 17-2021.
This program covers the estate of decedent(s) who died on or before 31
December 2017, whose estate taxes have remained unpaid or have accrued. The
executor/administrator of the estate, or the legal heirs, transferees or beneficiaries
who participated in this program only need to pay six percent (6%) of the total net
taxable estate at the time of death. This rate is applicable at every stage of transfer
of the estate/property without penalty. However, the minimum Estate Tax Amnesty
Tax that should be paid is P5,000 for the transfer of estate of each decedent.
This amnesty provides opportunity for executor/administrator of the estate, or
the legal heirs, transferees or beneficiaries to enjoy the current lower rate of estate
tax that has been stipulated on Republic Act (RA) Number 10963, known as the Tax
Reform for Acceleration and Inclusion Law (TRAIN Law), that became effective on
1 January 2018. The estate tax rate on this act is set on 6% of the net estate value.
Meanwhile, prior to the implementation of TRAIN Law, the estate tax rates were
ranging from 5%-20%, and the estate tax and the return should be paid and filed
within 6 (six) months from the decedent’s death. Late payment of estate tax will be
subject to imposition of 25% to 50% surcharge, 20% interest per year, and a
compromise penalty.
2) Tax Amnesty on Delinquencies

All taxpayers with delinquent internal revenue tax liabilities covering taxable
year 2017 and prior years, on or before 24 April 2019, may avail of this program.
This program allows taxpayers (any person, natural or judicial) who has delinquent
account to take advantage of lower rates to settle their obligations/liabilities. On the
government side, this program loosens the burden of administrative and judicial
dockets of slow-moving cases of tax delinquencies. This offer starts from April 2019
until 23 April 2020 (one year), but the government decided to extend it until 30 June
2021 due to coronavirus pandemic.
Tax Amnesty on Delinquencies covers all national internal revenue taxes such
as income tax, withholding tax, capital gain tax, donor’s tax, value-added tax, other
percentage taxes, excises tax and documentary stamp tax collected by the Bureau
of Internal Revenue (BIR), including VAT dan excise taxes collected by the Bureau
of Customs for taxable year 2017 and prior. Participated taxpayers shall pay the
following tax amnesty rates:
a) 40% of the basic tax assessed for delinquent accounts and assessments which
have become final and executory
b) 50% of the basic tax assessed for tax cases subject of final and executory
judgement by court
c) 60% of the basic tax assessed for pending criminal cases filed with Department
of Justice (DOJ)/Prosecutor’s office or courts for tax evasion and other criminal
offenses under National Internal Revenue Code of 1997
d) 100% of the basic tax assessed for withholding agents who withheld taxes but
failed to remit the same to the BIR (Bureau of Internal Revenue)
3) Voluntary Assessment and Payment Program (VAPP)

The Department of Finance (DOF) and the Bureau of Internal Revenue (BIR)
initiate VAPP by issuing RR 21-2020. The objective of VAPP is to collect additional
revenue to fund government expenditure related to coronavirus pandemic. This
program allows qualified taxpayers to settle unpaid taxes (internal revenue taxes),
18
including on one-time transactions (ONETT), which could otherwise be collected
through audit or enforcement effort, and would be provided certain privileges.
Originally, the application of VAPP is from 21 September 2020 - 31 December
2020. However, the government extends the period by issuing new regulation (RR
33-2020), thus taxpayers are able to participate in this program until 30 June 2021.
The VAPP applies to all internal revenue taxes covering the taxable year
ending 31 December 2018 and fiscal year 2018 ending on 31 July 2018 to June
2019, including taxes on one-time transaction (ONETT), such as estate tax, donor’s
tax, capital gains tax (CGT), as well as ONETT-related creditable withholding tax
(CWT) expanded withholding tax and documentary stamp tax (DST). Any person,
natural or judicial, including estates and trusts, liable to pay for internal revenue
taxes for the above period, who unintentionally paid incorrect internal revenue tax or
failed to file tax returns, except taxpayers who are already issued Final Assessment
Notice that have been final, persons under investigation as a result of verified
information filed by informer, taxpayers with cases involving tax fraud, and taxpayer
with pending tax evasion cases/other criminal offenses of tax code.
Taxpayers who want to participate in this program should submit application
along with the required documents. Then, the Revenue Officer will evaluate the
application. A Certificate of Availment (COA) will be issued if the applications are
approved. Taxpayers who received COA are eligible to participate in VAPP by
paying the liable tax using the stipulated rate which considerably lower than the
common tax rate.
Participated taxpayers with a duly issued of COA shall not be audited for fiscal
year 2018, for the tax covered by the availment. In case the taxpayers’ returns for
the covered period are currently being audited, the audit will be suspended. If the
taxpayers’ availment has been determined to be valid and COA has been issued,
the audit will be cancelled.
19
5. DISCUSSION
Based on the findings explained on the previous section, the table below
summarizes the features of tax amnesty program in Indonesia, Malaysia, and the
Philippines.
Table 4 Summary of Tax Amnesty in Indonesia, Malaysia, and The Philippines

Source: Compiled by Author
Feature Indonesia Malaysia The Philippines
Program Basis Law Number 7/2021 Guidelines of VA Estate Tax Amnesty:
(Regulation) Regulation of MoF Programme on TITLE II of RA11213
(PMK) number Indirect Tax, published RR No. 06-2019
196/PMK.03/2021 by Royal Malaysian Extension:
Customs Department, RA 11569
31 December 2021 RR No. 17-2021
Tax Amnesty on
Delinquencies:
TITLE IV of Republic
Act (RA) 11213
RR No.4-2019
Extension:
RR No. 5-2020
RR No. 7-2020
RR No. 10-2020
RR No. 11-2020
RR No. 15-2020
RR No. 32-2020 (ext)
VAPP:
RR No.21-2020
RR No.33-2020 (ext)
Target of The Policy I: Individual and Individual and Estate Tax Amnesty:
Program Corporate Taxpayers Corporate Taxpayers Executor or
who had participated administrator of the
on Tax Amnesty 2016- estate, legal heir,
2017 transferees or
beneficiaries of
estates.
Policy II: Individual Tax Amnesty on
Taxpayers Delinquencies:
Any person, natural or
judicial
VAPP:
Any person, natural or
judicial
Taxes Covered Declaration of Indirect Tax: Estate Tax Amnesty:
undisclosed assets is Estate Tax
20
considered as 1. Import and export Tax Amnesty on
additional income, duty; Delinquencies:
subject to Final 2. Sales tax; National Internal
Income Tax. 3. Service tax,
Revenue Taxes, such
including that on as income tax,
imported services withholding tax, capital
and digital services;gain tax, donor’s tax,
4. Excise duty; value added tax,
5. Goods and services excise tax, and
tax; documentary stamp
6. Tourism tax; and tax collected by BIR or
7. Departure levy. Bureau of Customs.
VAPP:
all internal revenue
taxes, including taxes
on one-time
transaction (ONETT),
such as estate tax,
donor’s tax, capital
gains tax (CGT).
Benefit for  Lower Income Tax Penalty incentives: Estate Tax Amnesty:
participated rate on the declared 90% and 100% in Lower rate of estate
taxpayers assets. Phase I tax (6%) based on
 Policy I: The 80% and 50% in TRAIN Law.
declared assets Phase II Tax Amnesty on
would not be Tax Incentives: Delinquencies:
subject to penalty of Up to 30% in Phase I Lower rates applied for
200%. Up to 15% in Phase II tax liabilities
 Policy II: free from Exemption from audit Termination of any
administrative criminal case related
sanction and to tax and immune
excluded from tax from suits, including
audit for the fiscal penalties.
year of 2016-2020 VAPP:
Lower rates applied for
tax liabilities
Exemption from audit
Companion  Implementation of - -
policy/ follow-up Tax Regulation
policy Harmonization Law:
administration
reform,
consolidative tax
policy, and
broadened tax
bases.
 Priority Supervision
on taxpayers who
are encouraged by
DGT to participate
on VDP but did not
participate.
21
Duration 6 (six) months 9 months, divided into Estate Tax Amnesty:
1 Jan 2022 – 30 June 2 phases: 2 years (15 June 2019
2022 Phase I: 6 months – 15 June 2021), then
1 January 2022 – 30 extended for 2 more
June 2022 years, until 14 June
Extension of Phase 1 2023
payment: by 14 July Tax Amnesty on
2022, as long as the Delinquencies:
application has been 1 year (15 June 2019 –
submitted by 30 June 15 June 2020), then
2022 extended until 14 June
Phase II: 3 months 2021
1 July 2022 – 30 Sept VAPP:
2022 21 September 2020 -
31 December 2020
Then extended until 30
June 2021
From the above explanation, it is found that Indonesia and the Philippines
stipulated legislative laws followed by various regulations as implementation basis of tax
amnesties program during the coronavirus pandemic, while Malaysia only issued
guideline for the implementation of VA Program. Malaysia government then issued FAQ
to further provide explanation on participation procedure of VA Program.
The 3 (three) countries state clear objectives and goals of tax amnesties
implementation. Malaysia government aimed to reduce tax leakages eroding national
revenue, and encouraged taxpayers to improve their indirect tax administration or
governance. Meanwhile, Indonesia aimed to provide opportunity for taxpayers to
disclosed their unreported asset (as proxy of unreported income) before the tax authority
fully make use of data from of automatic exchange of information (AEOI) and tax
information from agencies, institutions, associations, and other parties (ILAP). For the
Philippine, objectives of the program were varied: estate tax amnesty aimed to provide
reasonable tax relief for the decedents; TAD program aimed to reduce the load of
delinquent tax cases on the BIR and the courts, which could result in minimizing
administrative cost; and VAPP focused one gaining additional revenue to fund
government expense in regards Covid-19 pandemic. Of all three countries, only the
Philippines who explicitly stated enhancing revenue collection as the program objective
on the Republic Act.
Incentives offered by Indonesia, Malaysia and the Philippines are quite attractive.
Indonesia and the Philippines reduce the tax rate significantly and exempt taxpayers
from sanctions/penalties, while Malaysia provide high remission, both for tax liabilities
and penalties. All three countries also exempt participated taxpayers from tax audit and
investigation. However, for Indonesia, Policy II is targeted only for individual taxpayers.
According to the DGT, the compliance rate of individual taxpayers is still need to be
improved compared to corporate taxpayers. Experience from Tax Amnesty 2016-2017
indicates that tax compliance of individual taxpayers who participated on the program
shows better trend compared to non-participant, hence, DGT intends to encourage more
individual to become more compliance through VDP (Ulya, 2021).
Other than tax and penalty reduction, amnesty program in Indonesia, Malaysia and
the Philippines also provide other benefits. Indonesia applies distinctive tax rate for
taxpayers who repatriate and invest their disclosed assets on government bonds or
business activities in natural resources processing industry or renewable energy sector
22
in Indonesia. This condition shows that Indonesia government uses VDP to repatriate
flight capital/asset back to Indonesia and support the development of certain domestic
business sectors. Malaysia also aims to encourage taxpayers to improve their indirect
tax governance, while the Philippines aims to simplify the tax administration procedure.
These designs show that the three countries not only focus on additional revenue and
tax/penalty reduction, but also make use amnesty program to support other initiatives.
Furthermore, despite the exemption of audit offered, all three countries stated that
such privilege can be revoked if the participated taxpayers did not fully comply in the
amnesty program. Indonesia DGT states that if the authority found new information
regarding undisclosed asset, while Malaysia government indicates that audit still can be
performed if fraud is detected. For the Philippines, if the authority finds strong evidence
that participated taxpayers did not fully comply by certain degree, the privilege of audit
and investigation exemption can be revoked. Indonesia would apply additional final
income tax if such cases happen, while Malaysia and the Philippines will cancel all the
benefits given. Such designs would encourage participated taxpayers to comply, as non-
compliance found after the period will face harsher consequences.
It is quite apparent that improving long-term compliance is also being the focus of
tax amnesty in Indonesia and Malaysia. In the process of encouraging participation in
the VDP, the DGT of Indonesia sent encouragement letter that contains detected
undisclosed assets to numerous individual taxpayers. This initiative shows that the DGT
is able to gain such information as a result of automatic exchange of information (AEOI),
thus, taxpayers should make a good use of current opportunity to come forward and
settle any tax liabilities. Meanwhile, Malaysia could expect positive trend of indirect tax
compliance if businesses improved their tax governance. However, in the Philippines,
the impact of the amnesty program, especially TAD and VAPP, on long-term compliance
rate could be arguable and requires further observation, since the program focuses more
on minimizing the administrative cost of the authority rather that encouragement of better
tax governance for taxpayers.
Lastly, Malaysia and the Philippines provide leniency in the implementation of the
tax amnesty program in form of extension. Malaysia extends the payment deadline of
Phase I due to high interest of participation, conditional on taxpayers must have already
submitted application by 30 June 2022. The Philippines extends the TAD program
several times and extends the VAPP program for 6 (six) months. Such extension should
be carefully implemented because it could harm the credibility of tax amnesty program.
Credibility of amnesty program is essential as it could encourage more taxpayers to fully
comply when participating in the program.
6. CONCLUSION AND LIMITATION

6.1 Conclusion
Many countries face heavier fiscal burden due to coronavirus pandemic. Amidst
the pandemic, Indonesia, Malaysia, and the Philippines implemented tax amnesty
program for various backgrounds. Indonesia implements Voluntary Disclosure Program
(VDP), Malaysia implements Voluntary Disclosure and Amnesty Program (VA Program),
and the Philippines implements several programs: Estate Tax Amnesty, Tax Amnesty on
Delinquencies (TAD), and Voluntary Assessment and Payment Program (VAPP).
Apparently, the features of tax amnesty programs in the three countries are significantly
distinctive one from another, that could be identified from the basis of implementation,
program target, taxes covered, implementation period, and incentives offered.
Nevertheless, some similarities also could be identified. A common incentive offered is
the exemption from tax audit and investigation for taxpayers who participated in the
program. Another similarity of those programs is that taxpayers may face heavier
23
consequences and lose the program benefits if tax authorities found that they were not
fully complied when participating in the amnesty programs.
While those countries clearly set the objectives of their tax amnesty program, only
the Philippines outlines the intention of gaining additional revenue using the amnesty
program, in order to fund government expenditure related to the coronavirus pandemic.
The objective is explicitly stated on the Revenue Regulation (RR) No. 21-2020 that
regulates VAPP implementation. Indonesia encourages participants of the program to
repatriate flight capital/asset back to Indonesia and to invest in government bonds or
business activities in natural resources processing industry or renewable energy sector
in Indonesia by providing lower rate for those who chose the options, while Malaysia
focuses on the improvement of taxpayers’ indirect tax governance. Improving long-term
compliance is apparent for Indonesia and Malaysia tax amnesty design, but such goal is
questionable as the Philippines’ design of tax amnesty program focuses more on
minimizing the administrative cost of the authority rather that encouragement of better
tax governance for taxpayers.
Lastly, even though all three countries provided clear terms of the program,
Malaysia and the Philippines provide extension of the program. Especially for the
Philippines, whose program extension lacks in boundary, this may reduce the credibility
of the amnesty program.
6.2 Limitation
This research only conducted literature review in providing comparative analysis

on the implementation of tax amnesty in Indonesia, Malaysia, and the Philippines.
Comparison of tax revenue gained from the programs were not included as amnesty
programs in Malaysia and the Philippines were not yet ended. While this research
included a discussion whether amnesty programs in those countries fulfill the principles
of successful voluntary disclosure program outlined by OECD, it could not provide
analysis of the actual impact of the programs, particularly on tax revenue boost and tax
compliance. It would be interesting for future research to study the implication of the
program, on both short-term and long-term compliance, as well as the cost effectiveness
of the programs, using a quantitative approach.
24
REFERENCES
Alm, J., Martinez-Vazquez, J., & Wallace, S. (2009, September). Do Tax Amnesties Work? The
Revenue Effects of Tax Amnesties During the Transition in the Russian Federation. Economic
Analysis & Policy, 39(2), 235-253.
Baer, K., & Le Borgne, E. (2008). Tax Amnesties: Theory, Trends, and Some Alternatives.
International Monetary Funds, Washington, DC.
Bureau of Internal Revenue. (2022). Tax Amnesty Act. Republic of The Philippines. Retrieved
2022, from https://www.bir.gov.ph/index.php/tax-amnesty-act.html
Gita-Carlos, R. A. (2021, June 30). Estate Tax Amnesty Extended until June 2023. Manila.
Retrieved from https://www.pna.gov.ph/articles/1145527
Ministry of Finance. (2021, December 2021). Regulation of The MInister of Finance (PMK)
Number 196/PMK.03/2021 concerning Implementation Procedure of Voluntary Disclosure
Program. Retrieved from https://jdih.kemenkeu.go.id/in/dokumen/peraturan/bb933c55-e3b9-
4a1c-6e44-08d9cc20e0d9
OECD. (2015, August). Update on Voluntary Disclosure Programmes: A Pathway to Tax

Compliance.
Pellechio, A. (1993). Tax Amnesties. (F. A. Fund, Ed.) Washington: unpublished.
Razon, R. A. (2021, June 15). Green Light for Tax Amnesty Act Extension. Retrieved from
https://home.kpmg/ph/en/home/insights/2021/06/green-light-for-tax-amnesty-act-
extension.html
RMCD. (2022, June 29). Portal Rasmi Jabatan Kastam Diraja Malaysia. Retrieved from
http://www.customs.gov.my/ms/_layouts/15/ApplikasiKastam/DetailsPengumuman.aspx?ID=
469
Royal Malaysia Custom Department. (n.d.). Frequently Asked Questions VA Programme and
Voluntary Disclosure Programme. Malaysia. Retrieved from
https://myva.customs.gov.my/FAQ_BI_VA_30_12_2021.pdf
Royal Malaysia Customs Department. (2021, December 31). Retrieved from

http://www.mysst.customs.gov.my/:
http://www.mysst.customs.gov.my/assets/document/Annoucement/GUIDELINE%20VA%20PR
OGRAM_20211231_merged.pdf
Sabnita, N. (2020, 02). What is Known about Tax Amnesty? A Scoping Review. Indonesian Tax
Journal, 3(2), 36-50.
Ulya, F. N. (2021, 11 05). Program Pengungkapan Sukarela Kebijakan II Khusus Perseorangan, Ini
Alasan Kemenkeu. Jakarta, Indonesia. Retrieved from
https://money.kompas.com/read/2021/11/05/143800726/program-pengungkapan-sukarela-
kebijakan-ii-khusus-perseorangan-ini-alasan?page=all
25
TAXATION, DIGITAL FINANCIAL SERVICES, AND
THEIR IMPACTS ON FINANCIAL INCLUSION
Ariza Ayu Ramadhania, Citra Handayani Nasruddinb
a
[Fiscal Policy Agency, Jakarta, Indonesia] Email: [arizaayu@kemenkeu.go.id]
b
[Fiscal Policy Agency, Jakarta, Indonesia] Email: [citra.handayani@kemenkeu.go.id]
ABSTRACT
In this conceptual article, we discuss the burgeoning Digital Financial Services (DFS) through the lens of
taxation and financial inclusion. First, to provide a clearer insight into the phenomenon, we explain the
theoretical framework and elucidate the concept Digital Financial Services and Financial Inclusion. We
maintain that the rise of DFS affects financial inclusion through digital channels or digital financial inclusion.
We show the benefits of financial inclusion for individuals, families, society, financial institutions, and the
government. We also present criticism on DFS, including biases causing negative relationship between
digital finance and financial inclusion. Second, we investigate the link between taxation and Digital Financial
Inclusion. In doing so, we provide an international perspective of taxing DFS and compare it with policy
actions and practices in Indonesia. We identify DFS tax measures and incentives provided by the
government and conclude that those instruments are potentially to be a barrier for financial inclusion.
Ultimately, we contend that governments roles on financial regulations will include taxation. Therefore, we
recommend in-depth review on the DFS tax implementation in Indonesia.
Keywords: Digital Financial Services, Taxation, Financial Inclusion, Fintech
Melalui artikel konseptual ini, kami membahas Layanan Keuangan Digital (LKD) yang sedang berkembang
melalui lensa perpajakan dan inklusi keuangan. Pertama, untuk memberikan pemahaman yang lebih jelas
tentang fenomena tersebut, kami memaparkan kerangka teori dan menjelaskan konsep Layanan Keuangan
Digital dan Inklusi Keuangan. Kami berpendapat bahwa kenaikan DFS mempengaruhi inklusi keuangan
melalui saluran atau inklusi keuangan digital. Kami menunjukkan manfaat inklusi keuangan bagi individu,
keluarga, masyarakat, lembaga keuangan, dan pemerintah. Kami juga menyampaikan kritik terhadap LKD,
termasuk adanya bias yang menyebabkan hubungan negatif antara keuangan digital dan inklusi keuangan.
Kedua, kami menginvestigasi hubungan antara perpajakan dan Inklusi Keuangan Digital. Untuk itu, kami
memberikan gamabran tentang perspektif internasional mengeai perpajakan DFS dan membandingkannya
dengan praktik kebijakan yang ada di Indonesia. Kami mengidentifikasi langkah dan insentif pajak DFS yang
diberikan oleh pemerintah dan menyimpulkan bahwa instrumen tersebut berpotensi menjadi penghalang
bagi inklusi keuangan. Pada akhirnya, kami berpendapat bahwa peran pemerintah dalam regulasi keuangan
juga perlu mencakup perpajakan pada sektor keuangan. Oleh karena itu, kami merekomendasikan kajian
mendalam atas penerapan pajak LKD di Indonesia.
Kata Kunci: Pelayanan Keuangan Digital, Perpajakan, Inklusi Keuangan, Keuangan Digital
1. INTRODUCTION
After the global financial crisis, the Covid-19 pandemics and the physical distancing
policies have accelerated digital transformation in many sectors, including the
financial industry, through digital financial services (Auer et al., 2020; Didier et al.,
2021). FinTech also refers to Digital financial services (DFS), which are financial
services relying on digital technologies in their provision and use by the consumers.
Well-known Fintech innovations today include crypto assets, blockchains, artificial
intelligence, Robo-advisers, peer-to-peer lending, equity crowdfunding, and mobile
payment systems. Digitalization in the financial industries benefited countries and
26
individuals by providing safe and affordable financial services and increasing
consumption, remittances, and investment (Demirguc-Kunt et al., 2022).
An essential benefit of DFS is improving digital financial inclusion (D'Silva et al., 2019;
Ozili, 2020; Demirguc-Kunt et al., 2022). Based on the Global Findex 2021, for ten
years from 2011 to 2021, there was a 50 percent increase in global financial inclusion
from 51 percent to 76 percent (Demirguc-Kunt et al., 2022). Financial inclusion
provides benefits, such as affecting financial resilience and stability during economic
hardship and emergencies (Moore et al., 2019; Han & Melecky, 2013; Collins,
Morduch, Rutherford, & Ruthven, 2009). In addition, digital financial inclusion benefits
societies by disrupting the financial ecosystem forcing financial institutions to provide
cheaper services with good quality (Ozili, 2018). However, critics of DFS argue that
DFS only benefits the middle-high income population (Ozili, 2018). Furthermore,
Phillippon (2019) argues that to be beneficial for low-income people, the cost of
Fintech must be very low or negligible to bestow widespread advantages. Based on
those arguments, DFS could decrease financial inclusion, especially for the poor and
underprivileged population.
While DFS could also benefit governments by distributing financial assistance,

another aspect regulated by the government is the taxation of digital financial services.
There is limited literature investigating issues at the intersection of digital financial
inclusion and taxation. Most extant studies examine the impacts of financial inclusion
on improved tax revenues or a tax-based (Oz-Yalaman, 2019; Al-Own & Bani-Khalid,
2021; Oleschak, 2021). All of the research in this area suggests that financial inclusion
benefits taxation. However, it is still unclear whether the opposite is true.
The study examines the implication of digital financial services and taxation on
financial inclusion. The study makes two significant contributions. Firstly, to the
theoretical body of knowledge on digital finance, taxation of digital financial services,
and financial inclusion. There is limited literature linking taxation on DFS to financial
inclusion. Secondly, to policy development by exploring the role of DFS on digital
financial inclusion and including possible negative impacts that arise from taxation to
digital financial inclusion. This article would help policymakers or financial regulators
to understand better the current trend in financial innovation and international taxation
practices regarding DFS. By unpacking these issues, policymakers could shift to the
social and behavioral impacts of tax policies like financial inclusion.
The main question we would like to address through this article is the impact of
taxation and DFS on digital financial inclusion. Accordingly, we divide our analysis
into two sections: Digital Financial Services and Financial Inclusion and Taxation and
Digital Financial Inclusion. First, we explain how the DFS thrives and the sector's
importance in social-economic development, including tackling problems arising from
the Covid-19 pandemic. We further discuss the benefits spilled over from individuals,
households, businesses, government, and society as a whole. To put it into
perspective, we also explore the critics and negative consequences of DFS.
Secondly, to understand the possible implications of taxation on digital financial

inclusion, we examine tax provisions in the financial sector, both in the international
and domestic context. Specifically, we investigate standard practices in some
countries to acquire a broad perspective and compare them to the ones implemented
in Indonesia. Drawing on such a finding, we further identify areas in which taxation
can influence customer access to digital financial products and services, thus affecting
27
financial inclusion in Indonesia. In analyzing tax provisions in the digital financial
sector, we focus on three instruments, i.e., payment system, peer to-peer-lending,
and investment products and services. In this conceptual article, we do not, however,
closely examine digital saving products and insurance.
2. ANALYSIS AND DISCUSSION

2.1. Digital Financial Services and Financial Inclusion
2.1.1 The Rise of Digital Financial Services (DFS)
With FinTech taking by storm in recent years after the global financial crisis and during
the Covid-19 pandemics, the result is the transformation in the financial industries that
have been captured in many studies. Bank of International Settlement/BIS (2019)
defined FinTech as innovation from digital technology creating new development or
modifying existing provision of financial services. The term FinTech also refers to
Digital financial services (DFS) which are financial services relying on digital
technologies in its provision and use by the consumers. Innovation in the financial
industry may disrupt current structures, blur boundaries, and as a result democratize
financial access (Feyen et. al., 2021; Philippon, 2016). On the one hand, the
innovation facilitated new players from small start-up companies to big tech firms to
target consumers by providing digital financial services. On the other hand, incumbent
financial players, such as banks, adopted digital technologies in improving their
efficiencies to compete with big techs and FinTech players (BIS, 2019; Frost et. al.,
2019).
After the global financial crisis, the Covid-19 pandemics and the physical distancing
policies urged and accelerated digital transformation in many sectors including in
financial industry. The pandemic speed up the shift to electronic money (Auer et. al,
2020), escalated e-commerce (BIS 2020, Alfonso et al, 2021), and might accelerated
the work on central bank digital currencies (Auer et al 2020). Finally, Didier et al (2021)
also showed that countries with strict physical restrictions during the Covid-19
outbreak experienced a higher increase in financial apps downloaded.
Well-known FinTech innovations today include cryptoassets, blockchains, artificial

intelligence, robo-advisers, peer-to-peer lending, equity crowdfunding, and mobile
payment systems. Digitalization in the financial industries benefited countries as well
as individuals. For individuals, mobile money provides safety and affordability for
users to store money or transfer them which could affect to an increase in remittances,
consumption and investment (Demirguc-Kunt et. al., 2022). The financial services
such as mobile money was crucial especially in the time of pandemics where people
around the world were restricted to do social activities but still needed to pay their bills.
In achieving development goal, studies found that workers who received wages and
social benefit from the government directly to their account had more savings that
those who receive in cash (Blumenstock, Callen, and Ghani, 2018).
Moreover, DFS were crucial for the government to directly transfer social assistance
by using digital technologies. Demirguc-Kunt et. al. (2022) show that 18 percent of
individuals paid utility bills directly from an account and around one third of these
adults did so for the first time after the onset of the COVID-19 pandemic. Furthermore,
the same study also pointed out that digital merchant payment also increased after
the pandemics showing some data from China and India. These figures asserted
28
previous study by The World Bank (2014) that Fintech also beneficial in improving
financial inclusion, developing basic services, and expanding usage from financial
services to non-financial services.
2.1.2 Digital Financial Inclusion

The formal definition of financial inclusion is not yet available. However, in general,
financial inclusion refers to the “universal access to, and use of, a wide range of
reasonably priced financial services” (Fayen et al, 2021). Financial inclusion also
refers to efforts to increase the number of adults in accessing formal financial services
largely through formal bank account ownerships (Ozili, 2018). Through financial
inclusion, people will be able to save, launch business and invest more in education,
which in turn, would contribute to increase economic growth and reduce poverty (Beck
et al., 2007;Bruhn & Love, 2014 in Ozili, 2018).
Similar to the various definitions, the determinants of financial inclusion were varied
as well. Recent literature related to financial inclusion from 2014 to 2020 has been
reviewed by Ozili (2020) showing that there are three school of thoughts about the
determinants of financial inclusion, which are financial literacy, financial innovation,
and other policies or strategies. Financial innovation and technology allowed financial
institutions to reach the underprivileged by breaking the infrastructural and structural
barriers (Ozili, 2020). Hence, digital financial services positively contribute to financial
inclusion. Mobile phone and the internet have been known to be the game changers
in many aspects including in the financial industry. Therefore, the provision of digital
financial infrastructure would definitely improve financial inclusion in a country (D’Silva
et al, 2019). Furthermore, technological innovation that is not directly related to
finance such as government-issued biometrics identification cards were in facts could
lowered cost of access, and promoted account ownership among adults (Demirguc-
Kunt et al, 2018).
The rise of DFS affects financial inclusion through digital channels or digital financial
inclusion. Digital financial inclusion refers to the access and usage of formal financial
services for underserved and excluded individuals with the help of digital technology.
Ozili (2018) suggested that the digital financial inclusion’s process commenced when
the financially underserved or excluded population who have some kind of formal
bank accounts required digital access to basic financial transactions remotely. Rapid
digitalization in the financial industry has resulted in the surge of financial inclusion
between the global financial crisis and the Covid-19 pandemics; hence, creating
digital financial inclusion. Based on the Global Findex 2021, global account ownership
at a bank or other formal institution, such as a credit union, microfinance institution,
or a mobile money provider, reached 76 percents of adults. During 10 years from 2011
to 2021, there were 50 percent increased in global financial inclusion from 51 percent
to 76 percent (Demirguc-Kunt et al, 2022). From the same data, it was suggested that
the large share of the increasing account ownership in Sub-Saharan Africa stems
from the adoption of mobile money. In other words, the account owners have just
included in the financial system through DFS.
In general, financial inclusion brings many benefits especially for the underprivileged
population. Study shows that families and business who are financially included are
more resistant during economic shocks than those who are financially excluded
(Moore et al, 2019). Financial inclusion could provide stability for low income
individuals through the possibility of savings which could secure their personal finance
29
during hard times and emergencies such as illness or unemployment (Han & Melecky,
2013; Collins, Morduch, Rutherford, & Ruthven, 2009). For financial institutions,
financial inclusion especially a substantial rise in the number of small savers, could
reduce procyclicality risk, hence, promoting banking system stability through reducing
banks’ dependence on “non-core” financing, which tend to be more volatile during a
hardship (Khan, 2011). Briefly, previous literatures show that financial inclusion can
be one of the efforts to achieve Sustainable Development Goals through alleviating
poverty and women empowerment or gender equality.
For digital financial inclusion, there are several on-top advantages. On the one hand,
it disrupts the banking system and forces them to provide more affordable services of
good quality in order to reduce the risk of losing consumers to other banks (Ozili,
2018). On the other hand, it helps banks lowering their costs by reducing bank
branches and administrative paperworks (IFC, 2017; Manyika et al., 2016). It could
improve the well-being of individuals and business to digitally acces their fund in their
accounts to perform financial transactions (CGAP, 2015). Digital financial inclusion is
also beneficial for financial regulators for decreasing the amount of physical cash
circulated and crucial to reduce the level of high inflation in poor and developing
countries (GPFI, 2016). However, Ozili (2018) suggested that the expected
advantages of digital financial inclusion could be realised completely if the cost of
performing a digital transaction (such as mobile phones, personal computers and
related devices) is low or negligible, especially for poor individuals. Overall, digital
financial inclusion benefits financial service consumers, DFS providers, government
or regulators and the economy by improving access to finance for poor people,
decreasing cost of financial intermediation for financial institutions and increasing
aggregate government spending (Ozili, 2020).
2.1.3. Critics on DFS and Financial Inclusion

Despite the benefits for the economy, digital financial services and digital financial
inclusion also bring consequences. First, it requires updated policies and regulations
to reveal Fintech’s full potential in providing broadly distributed welfare advantages
(Philippon, 2019). There are huge challenges for regulators to keep up with the speed
of technological progress. Moreover, regulators are required to juggle between
promoting technological innovation and protecting the consumers including their data
privacy. DFS are prone to cyber-attacks which threat customers’ data on fintech
platforms. This could cause distrust in fintech channels or cause people to avoid
digital transactions until the customers get strong consumer protection (Malady, 2016).
Besides, another criticism was suggested by Ozili (2018) showing three biases
causing negative relationship between digital finance and financial inclusion. First,
Fintech players are profit-seeking companies that could discriminate costumers
based on their incomes or other socio-economic data and use different marketing
strategy to approach them. When they use a less-aggresive tactic to convince low-
income customers that may lower financial inclusion for low-income costumers. Ozili
(2018) also suggested that Fintech providers might be bias to provide DFS based on
location or geographical condition. Rural and remote area might be assessed as a
high risk area causing a Fintech firm to discontinue its services, thereby decrease
financial inclusion. Finally, Ozili mentioned educational bias by Fintech providers
which discriminate poor and uneducated individuals because the firms think that those
consumers are not profitable for their companies. From this study, we could
30
summarized that the Fintech biased would harm financial inclusion for low-income
individuals and not for the medium to high income groups.
2.2. Taxation and Digital Financial Inclusion

2.2.1 The Relationship between Taxation and Digital Financial Inclusion
There is limited literature investigating issues at the intersection of digital financial
inclusion and taxation. Most of the extant studies examine the impacts of financial
inclusion on improved tax revenues or tax based. Oz-Yalaman (2019), for instance,
assesses the impact of financial inclusion on tax revenue and finds a significant and
positive relationship between those variables. In the same vein, Al-Own & Bani-Khalid
(2021) show that higher financial inclusion is associated with more tax revenue.
In addition, Oleschak (2021) finds a significant and robust negative relationship between
financial inclusion and inflation, and a positive relationship between financial inclusion
and tax revenue, even after controlling for major macroeconomic variables. Meanwhile,
Mpofu’s study in 2022 reveals that mobile money taxes generated from financial
inclusion, in particular from mobile transactions such as receiving money, making
payments, and withdrawing money, is an opportunity for widening the tax bases. The
measure is considered as an arm to tax the informal sector in its informal status without
formalising it, and thus providing a reprieve for the post-pandemic reconstruction
expenditure.
Furthermore, as digital financial products and services are outcomes of technology

advancements, opportunities and risks associated with this novel innovation become
objects of examination. In a sense, issues arising from digital financial provision, such
as reliable financial and ICT infrastructure and regulatory or legal framework, are
intertwined with tax collection. That way, as suggested by Oleschak’s study (2021),
governments that work closely with the private sector to lay the groundwork for using
digital technology will have a greater chance to increase both financial inclusion and tax
collection. Ultimately, all of the research suggests that financial inclusion indeed benefits
taxation. It is, however, still unclear whether the opposite is true.
2.2.2. Taxation in the Digital Financial Sector: International Practices

Taxing financial institutions is an old concept that goes way back to 1694, when the
London Stock Exchange levied a transaction tax with a value dependent on stock
transfers (Puawska, 2021). It gained more popularity in post crisis time, such as after the
Great Depression and following the collapse of the Bretton Woods system, with
endorsement from prominent economists, not least of whom are Keynes and Tobin, for
the broader utilization of the financial transaction tax (Puawska, 2021).
The global financial crisis in 2008 has further triggered a burgeoning debate on tax
measures for the financial sector (Directorate-General for Taxation and Customs EU,
2010; Shackelford et al., 2012; de Mooij et al., 2013; Puawska, 2021). The arguments
centered around to what extent taxing the financial sector should be expanded to
intervene and combat systemic externalities in the industry. The proponents of higher
levies highlight the role of taxes in improving efficiency and stability, as well as reducing
the volatility and adverse effects of excessive risk-taking in financial markets. There is
also a recognition that the financial industry is the most lucrative sector, and it received
significant support from governments during the crisis. To that end, the advocates
31
suggest that the industry should make a fair contribution to the community through
taxation. Meanwhile, the opponents argue that tax distortions may play a role in
heightening the vulnerability of the financial sector.
Generally, there are three types of taxes proposed for the financial sector, namely
Financial Stability Contribution (FSC), imposed against particular banking institution
balance sheet components, Financial Activities Tax (FAT), levied based on a financial
institution's overall profitability, and Financial transaction tax (FTT), charged on a certain
kind of financial transaction with a certain purpose (Claessens et al., 2010; Cottarelli,
2010; IMF, 2010; Directorate-General for Taxation and Customs EU, 2010; Shackelford
et al., 2012; Puawska, 2021).
Value Added Tax (VAT) or Goods and Services Tax (GST) which serves as the key
source of revenues for many countries in the world (De Mooij & Swistak, 2022), is
commonly exempted on financial services, for instance through an input-tax mechanism
(Baydur, 2021). Such that, the end customers are not charged with the tax but financial
service providers are not able to claim a tax credit for the tax paid on the services or
products they deliver. Countries applying VAT or GST exempt on some of financial
services, among others, are New Zealand, Australia, Canada, Eropean Union, Israel and
Argentina (Chaudhry et al., 2015). In Australia, the Reduced Input Tax Credit is used to
counterbalance the distorsive effects of the input tax, while in South Africa and Singapore,
financial institutions like banks can claim input VAT for certain services (Chaudhry et al.,
2015).
However, in the wake of a flourishing digital financial industry, governments around the
world have taken advantage of this opportunity by enacting specific taxes targeted at
related products and services provisions, including the VAT. For developing countries,
the benefits of digital finance utilization can facilitate greater financial inclusion and boost
gross domestic product by 6% in 2025, according to a McKinsey’s report in 2016
(Manyika et al., 2016). It also holds the prospect of a broader tax base and revenues.
This applies in particular to emerging economies facing challenges to safeguarding their
domestic revenue mobilization, such as in Africa (OECD, 2019), Tanzania (Pilla, 2016),
Ghana (Sarkodie, 2022), Kenya, and Uganda (Anyanzwa, 2021). As projected by
Maherali (2017), $4.1 trillion in tax revenue was generated globally between 2014 and
2020 by virtue of financial inclusion.
Nonetheless, beyond revenues, taxes are also set for other purposes. The policy
objectives achieved through this instrument comprise resource redistribution, economic
growth and stability, equity and equality, and reduced negative externalities (Crumbley,
1973; Mpofu, 2022). Against this backdrop, implementing tax measures should consider
three fundamental components of a good tax, i.e., efficiency and certainty, equity, and
administrability (Munoz et al., 2022). Ultimately, to provide a conducive climate for strong
and inclusive economic growth, the government should strike a balance between these
goals.
It is worth noting that there is little evidence of tax implications on financial inclusion since
it is inherently challenging to isolate the variables from market structure and other factors
(Munoz et al., 2022). Due to the ambiguous link between the tax change and behavioral
effect (Crumbley, 1973; Olivola & Sussman, 2015), achieving public policy objectives
through taxes is also far from straightforward. More evidence from various settings and
specific taxes will be necessary to better understand and estimate the impacts of taxation,
and therefore provide a sound justification for levying DFS taxes.
32
2.2.3. International Consensus in Taxing Digital Financial Sector
Along with its high potential for advancing the economy, the DFS tax has also sparked
concerns about its utilization. Naturally, the tax will drive up the price of digital financial
products and services, adding costs for institutions or end customers. In the long run, it
will discourage the market expansion of the DFS (Munoz et al., 2022). In this view,
instead of creating an enabling environment, the DFS tax will distort the market, thus
curbing the growth of financial inclusion. Some governments, such as the European
Commission, have proposed regulations to ensure fair and growth-friendly DFS taxation
(European Commission, 2018).
As digital finance allows cross-border transactions, DFS is interconnected with

international tax. To date, although no consensus has been reached between countries
around the world, concerted efforts have been made through the OECD/G20 Inclusive
Framework (IF) on BEPS. With an initiative called the Two-Pillar Solution, this
cooperation aims to address the tax challenges arising from the digitalisation of the
economy (OECD, 2022). The solution consists of two approaches, Pillar One and Pillar
Two, with each model rule developed separately by different working groups.
Pillar one is made up of eleven building blocks within four frameworks, Amount A,
Amount B, Tax Certainty, and Implementation and Administration. One of its objectives
is to avoid fragmented unilateral tax measures such as digital tax services and disputes
emerging from such approaches (OECD, 2022). Under Amount A, taxing rights on more
than USD 125 billion of the largest corporations’ profits, including those categorized
under the digital economy, are expected to be reallocated every year to market
jurisdictions where goods or services are supplied or consumers are located (OECD,
2022).
However, regulated financial services are to be excluded from this mechanism. The
types of Regulated Financial Institutions (RFI) covered in the draft rules are depositary
institutions, Investment Institution, Insurance Institution, Asset Manager, a Mixed
Financial Institution, and entities exclusively performing on behalf of a Service Entity.
According to the OECD, such an exclusion is due to the industry’s distinct nature of
regulations that already align between the market and the place of tax profits (OECD,
2022). However, it must be pointed out that this deal is not final, and some members
maintain that reinsurance and asset management should not be ruled out (OECD, 2022).
As a matter of fact, to date, the technical work of some of the Pillar One elements is still
underway. Initially, this agreement was planned to come into effect in 2023. However,
country members were not able to reach consensus on those technical aspects by the
middle of this year. Consequently, the draft rules are yet to be finalised and translated
into a signed and ratified Multilateral Convention (MLC), which is mandatory for the
instrument to enter into force. The OECD has delayed Pillar One’s implementation to
2024 (Giles et al., 2022).
2.2.4. Taxation and Digital Financial Inclusion in the Context of Indonesia

In Indonesia, revenues from the financial sector are substantial. From the beginning until
mid 2022, the sector has generated more than 12% of total tax revenues, with 16.2%
annual growthy (Dian Kurniati, 2022), Historically, financial sector has been one of
Indonesia’s four major contributors to tax revenues, together with the manufacturing,
commerce, and mining sectors (Noverius Laoli, 2019). Although the industry contracted
33
to 14.3% in 2020 because of the Covid-19 crisis, it quickly recovered in 2021 (Avisena,
2022). The turning point is projected to be followed by a rapid growth, notably from digital
finance and economy (Arief Rahman Hakim, 2021).
The thriving digital finance and a greater inclusion in the sector during the Covid-19
pandemic is forecasted to prompt domestic economic growth. The prospect of a rosy
future for DFS is substantiated by projected growth achieved by digital financial
transaction this year. E-commerce transactions, for instance, is predicted to rise by 31%,
reaching Rp 536 trillion. Electronic money transaction is expected to attain Rp 360 trillion
or increasing up to 18% from the previous period. Meanwhile, transaction from digital
banking services are estimated to reach Rp 51,000 trillion (Triyan Pangastuti, 2022).
The large number and value of DFS transaction offer opportunities for the government
to expand the tax base and revenues. Tapping into the potentially desirable tax outcome,
the tax authority has regulated taxation for DFS. The tax measures for this sector is
mainly covered by the regulation of the minister of finance number 69/PMK.03/2022 that
is recently introduced this year. Based on the regulation, such a provision is made to
provide legal certainty and ease of administration for taxpayers in fulfilling their tax
obligations on transactions using financial technology, particularly on payment systems
and peer to peer lending. There are two types of taxes administered through this
arrangement, namely income tax and value added tax.
Other financial assets that have become more accessible for the vast majority of
customers by virtue of technology advancement are investment options. The instruments
traded in the Indonesian capital market are stocks, bonds, mutual funds, exchange
traded funds (ETFs), and derivatives (OJK & Kemenkeu, 2019). Several legal
instruments and regulations have also been available to govern the tax provisions on
these financial assets services, including the Law of Income Tax and Government
Regulations concerning the income taxes regarding investment products.
From the Table 1 below, we can see that VAT is largely imposed on DFS provisions such
as payment system, peer-to-peer lending, and investment services. Conversely,
transaction tax with regards to investment product is solely levied through income tax.
Interestingly, we find a relatively low tax rate for some investment products and services
that are considered sophisticated an are in the late stage of financial inclusion such as
stocks and bonds. Moreover, those products and services are arguably less common to
be accessed by low-income individuals. Therefore, a question could raise, whether this
policy is warranted and in line with the objective of financial inclusion and inclusive
economy development.
34
Table 1. Tax Provision for Digital Financial Sector in Indonesia
Sources: OJK & Kemenkeu, 2019; DJPB, 2021; Ministry of Finance, 2022
Type of Digital
Products/Services Type of Tax Regulation Tax Object Tax Subject Tariff
Payment services provision
through Electronic Money/
Minister of Finance Wallet, Payment Gateway, Business Entity (Personal
Payment system Value added tax
Regulation No.69/2022 Switching Services, Clearing, and Corporate)
11%
Final Settlement, Funds
Transfer
Domestic Lenders and

Lender with Permanent 15%
Establishment (PE)
Loan Interests
Minister of Finance
Peer to Peer Lending Value added tax
Regulation No.69/2022
Non PE / Foreign Tax Payer* 20%
Borrowing and Loan Service

Lending Services 11%
Provider (Domestic entity)
Investment Products
Government Regulation Stock Sale Transaction:

No.14/1997 jo Minister of
Finance Decree
No.282/1997 jo Circular - Non-Founder Share Individual & Corporate 0.1%
Income Tax (final)
Letter of Directorate Business Entity
Stocks General of Taxation
No.15/PJ.42/1997 and
No.06/PJ.4/1997 - Founder Share 0.5%
- Domestic Individual 10%

Income Tax (final) Law No. 36/2008 Dividen
35
- Foreign Individual* 20%
- Business Entity with Tax 15%

Identification Number
(NPWP)
- Business Entity without Tax 30%

Identification Number
(NPWP)
- Share sales transactions

Government Regulation
Income Tax (final) - Transfer of equity Business partner company 0.1%
No.4/1995
participation
Individual & Corporate

Business Entity:
15%
- Domestic tax payer and
Law No. 36/2008 Juncto - Interests
Bond Income Tax (final) Permanent Establishment
PP 16 Tahun 2009 - Discount
(PE)
20%
- Non PE Foreign Tax Payer*
Mutual Fund N/A Law No. 36/2008 N/A N/A 0%
Government Regulation Individual & Corporate

Exchange Traded Fund (ETF) Income Tax (final) Sale transaction Business Entity: 0.10%
No.4/1995
Income received/earned from

Government Regulation Individuals and corporate
Derivatives transactions of futures 2.5%
No.17/2009 business entities
contracts
Investment Services
36
Provision of equity crowd
Minister of Finance funding services through an
Equity crowd funding Value added tax Business entity 11%
Regulation No.69/2022 open electronic system
network.
Provision of integrated
electronic communication
Minister of Finance
Investment settlement Value added tax facilities supporting securities Business entity 11%
transaction settlement
activities
Provision of services to
manage investments using
electronic communication
facilities such as advanced
Minister of Finance algorithms, cloud computing,
Investment Management Value added tax Business entity 11%
Regulation No.69/2022 capabilities sharing, open
source information technology,
automated advice and
management, social trading,
and retail algorithmic trading.
Provision of data comparision
Minister of Finance
Market Support Services Value added tax on financial products and Business entity 11%
services
Provision of financial support
services in the form of eco
crowdfunding, Islamic digital
Digital financial support services
Minister of Finance financing, ewaqf, e-zakat, robo
and other financial service Value added tax Business entity 11%
Regulation No.69/2022 advise and credit scoring, d.
activities trading invoices, vouchers or
tokens, and blockchain
application based products
*Note: subject to relevant tax treaty provisions
37
Nevertheless, from a pragmatical perspective, levying VAT on DFS is deemed justifiable,
at least for three reasons. First, VAT is imposed on most of product and services in
Indonesia (Acclime, 2022), therefore it is reasonable to apply it to DST taxes that hold a
promise as a broad-based tax. Second, the proportion of VAT contribution to tax
revenues is still below best practices (Jeven, 2021). Third, like many developing
countries and emerging economies, domestic revenue mobilization through specific tax
on DST is evidently necessary for sustainable finance and development.
That being said, tax measures should also take into account the good principle of
taxation. On the one hand, since the tax levy will increase the relative price and hence
economic behavior change, the tax design should minimize the distorsive effects
(Bhattacharya & Stotsky, 2022). On the other hand, tax fairness must also be considered.
Depending on the degree to which redistribution is set to be achieved, the tax system
needs to address both vertical and horizontal equity (Bhattacharya & Stotsky, 2022).
Eventually, the government should reconcile the efficiency and equity goals in the tax
design while ensuring political and administrative viability before implementing the policy.
Government endeavors to balance tax policy objectives can be seen, among others,
through tax incentives for DFS. Below is a table showing exemptions for both digital
products and services.
Table 2. Tax Facilities/Incentive for Digital Financial Sector in Indonesia

Sources: OJK & Kemenkeu, 2019; Badan Kebijakan Fiskal, 2021
Type of Digital Products/Services Incentives
Payment
Transfer of funds in the same bank to customers holding demand deposits, VAT exemption
time deposits, certificates of deposit, savings, and/or other equivalent forms,
in accordance with the provisions of laws and regulations
Lending
Fund placement services, lending, or financing by lenders VAT exemption

Investment VAT exemption
Fund placement or financing services to securities issuers through facilities VAT exemption
provided by crowdfunding service providers
Fund placement or financing services by investors/investors to issuers of VAT exemption
Securities or other financial instruments through electronic communication
facilities
Mutual Funds (Profits and Assets) Income tax and VAT
exemption
Insurance
Online insurance services VAT exemption
It is discernible that tax incentives for DFS are limited to a few transactions. Those are
payments within the same financial institution, fund placement services, and online
insurance. This data demonstrates a stark difference with practices found in most
1
countries analyzed in the previous section, in which VAT exemption is commonly applied.
Nevertheless, all transactions and profits gained from mutual funds in Indonesia are fully
exempt.
In addition, although there is scant evidence proving the adverse impacts of taxation on
DFS, little data is also available to assert that the reverse condition applies. Yet, one
undeniable claim is that the tax levy will increase the cost for customers of accessing
digital financial products and services. Thus, taxation has the potential to be a barrier for
community members to accessing those provisions, especially those from low income
households. To that end, countermeasures in the form of tax exemptions or other policy
instruments will be essential.
CONCLUSIONS AND SUGGESTIONS

Financial inclusion, which refers to the access and usage of formal financial products
and services, has grown substantially in the past decade. Evidences show that the rise
of DFS is one of the crucial factors for its advancement. Financial innovation was
flourishing during the Covid-19 pandemics when loads of countries introduced physical
restriction policies. During the outbreak, DFS allowed individuals to perform financial
transactions such as using mobile money and also receiving social assistance program
from the government. This means that DFS was beneficial for both the social assistance
recipients and the government as the provider. As mentioned before, Fintech and
financial inclusion were positively correlated. The newest Global Findex 2021 found that
a wide share of mobile money users in Sub-Sahara Africa were new users. Hence,
Fintech has a positive relationship with the increasing financial inclusion.
Financial inclusion is known to have many advantages not only for individuals but also
for private companies and governments around the world. Many studies believed that
financial inclusion could increase households and business’ resilience during economic
shocks, improve access to credit and for the government, it could increase aggregate
expenditures. On top of that, digital financial inclusion could provide saver and more
affordable financial products and services that is substantial for poor individuals, thereby
improving their well-being. On the other end, digital financial inclusion is beneficial for
financial regulators since it reduces the amount of physical money circulated in a country.
Besides its benefits, DFS and digital financial inclusion also poses challenges for
example is data privacy issue. In term of regulation, governments are required to update
financial regulations to incorporate the growing trend in digital technology, improving
consumer protection while fostering innovation.
Firmly, governments roles on financial regulations will include taxation. In the rise global
awareness on financial inclusion and DFS, there are several studies showing a positive
correlation between financial inclusion and tax revenue. However, there is still scarce
robust evidence on impacts of existing taxes on DFS, and about how the design of those
taxes shape impacts digital financial inclusion. Previous studies show that the difficulties
in finding the evidence on the relationship between taxation and financial inclusion rely
on the characteristics of financial inclusion that is hardly being separated from other
variables such as market structure.
With regards to DFS taxes, we are convinced that those instruments should be designed
to be consistent with the principle of good taxation, and deliberately arranged to achieve
broader economic interests and various policy objectives. As such, the government
2
needs to consider the trade-offs between efficiency and equity purposes of the tax and
look ahead to the long terms goals of development.
Considering the types and degree of DFS taxation implemented in Indonesia, we find
that the tax measures have the potential to be a barrier for community members
accessing DFS provisions, especially those from low income households. To that end,
countermeasures in the form of tax exemptions or other policy instruments will be
essential.
Futhermore, the lack of data and evidence on the effects of DFS taxation on financial
inclusion should be a signal for better policy actions. The in-depth review on the DFS tax
implementation in Indonesia could shed light on the effectiveness of the policy and give
a profound understanding of its impacts on government undertakings to improve financial
inclusion and other policy objectives.
It is imperative for the evaluation to incorporate various lenses and perspectives,

including the supply and demand side of DFT, stages for financial inclusion,
characteristics of users from different socio economic status, ecosystems, and relevant
actors involved, in order to match with the unique context of Indonesia. Future study may
also be crucial to investigate the interaction between current regulation of DSF taxation
and those anticipated in the future, such as the Pillar One rules of the OECD’s two pillar
solution.
Altogether, tax policy should become a part of a comprehensive policy package and
national strategy to improve financial inclusion. That way, taxation can carry out its role
to support greater access to DFS by promoting smooth adoption of the products and
services at every stage for broad community members.
3
REFERENCES
Acclime. (2022). Value added tax (VAT) in Indonesia. Acclime Indonesia.
https://indonesia.acclime.com/guides/value-added-tax/
Al-Own, B., & Bani-Khalid, T. (2021). Financial Inclusion and Tax Revenue:
Evidence From Europe. International Journal of Financial Research, 12(2), 27.
https://doi.org/10.5430/ijfr.v12n2p27
Anyanzwa, J. (2021, January 21). Kenya joins Tanzania and Uganda in taxing digital
transactions. The East African.
https://www.theeastafrican.co.ke/tea/business/kenya-joins-tanzania-and-
uganda-digital-tax-3264326
Arief Rahman Hakim. (2021, December 17). Ekonomi Digital Indonesia Diprediksi
Meningkat Pesat [Liputan 6]. liputan6.com.
https://www.liputan6.com/bisnis/read/4778800/ekonomi-digital-indonesia-
diprediksi-meningkat-pesat
Auer, R. (2019). Embedded supervision: how to build regulation into blockchain
finance. BIS Working Paper, 811.
Auer, R., Cornelli, G., & Frost, J. (2020). Rise of the central bank digital currencies:
drivers, approaches and technologies. BIS Bulletin, 3.
Avisena, M. I. R. (2022, February 8). OJK: Sektor Keuangan Indonesia Telah
Membaik. Media Indonesia. https://mediaindonesia.com/ekonomi/470014/ojk-
sektor-keuangan-indonesia-telah-membaik
Badan Kebijakan Fiskal. (2021). Tax Expenditure Report 2020. Kementerian
Keuangan.
https://fiskal.kemenkeu.go.id/files/ter/file/1640354845_laporan_belanja_perpaj
akan_2020_ebook.pdf
Baydur, I. (2021). VAT Treatment of the Financial Services: Implications for the Real
Economy. Journal of Money, Credit and Banking, 53.
https://doi.org/10.1111/jmcb.12780
Bhattacharya, S., & Stotsky, J. G. (2022). Tax Primer: With Applications to Asia and
Pacific Countries. In Asian Development Outlook 2022: Mobilizing Taxes for
Development (p. 31). Asian Development Bank.
https://www.adb.org/sites/default/files/institutional-
document/782851/ado2022bn-tax-primer-applications-asia-pacific.pdf
Blumenstock, J., Callen, M., & Ghani, T. (2018). Why Do Defaults Affect Behavior?
Experimental Evidence from Afghanistan. American Economic Review, 108(10),
2868–2901. https://doi.org/10.1257/aer.20171676
Chaudhry, S. M., Mullineux, A., & Agarwal, N. (2015). Balancing the regulation and
taxation of banking. International Review of Financial Analysis, 42, 38–52.
https://doi.org/10.1016/j.irfa.2015.01.020
Claessens, S., Keen, M., & Pazarbasioglu, C. (2010). Financial Sector Taxation:
The IMF’s Report to the G-20 and Background Material (p. 197). InternatIonal
Monetary Fund.
Cottarelli, C. (2010, April 25). Fair and Substantial—Taxing the Financial Sector.
IMF Blog. https://blogs.imf.org/2010/04/25/fair-and-substantial-taxing-the-
financial-sector/
Crumbley, D. L. (1973). Behavioral Implications of Taxation. The Accounting Review,
48(4), 759–763.
Demirgüç-Kunt, A., Klapper, L., Singer, D., & Ansar, S. (2022). The Global Findex
Database 2021 Financial Inclusion, Digital Payments, and Resilience in the Age
of COVID-19. https://www.worldbank.org/en/publication/globalfindex
4
de Mooij, R., Keen, M., & Orihara, M. (2013). Taxation, Bank Leverage, and
Financial Crises. IMF, 26.
De Mooij, R., & Swistak, A. (2022, March). Value-Added Tax Continues to Expand.
IMF. https://www.imf.org/en/Publications/fandd/issues/2022/03/b2b-value-
added-tax-continues-to-expand
Dian Kurniati. (2022, July 3). Setoran Pajak Seluruh Sektor Utama Positif, Begini
Perinciannya. DDTC News. https://news.ddtc.co.id/setoran-pajak-seluruh-
sektor-utama-positif-begini-perinciannya-40252
Directorate-General for Taxation and Customs EU. (2010). Taxation papers –
Taxation of Financial Sector. Publications Office of the European Union.
https://data.europa.eu/doi/10.2778/1549
DJPB. (2021, Agustus). Pajak Penghasilan Pasal 4 ayat (2). KPPN Kotabumi.
https://djpb.kemenkeu.go.id/kppn/kotabumi/id/informasi/perpajakan/pph-pasal-
4-ayat-2.html
European Commission. (2018). Fair Taxation of the Digital Economy. Taxation and
Customs Union. https://taxation-customs.ec.europa.eu/fair-taxation-digital-
economy_en
Feyen, E., Frost, J., Gambacorta, L., Natarajan, H., & Saal, M. (2021). Fintech and
the digital transformation of financial services: implications for market structure
and public policy. BIS Papers, No 117.
https://www.bis.org/publ/bppdf/bispap117.htm
Giles, C., Fleming, S., & Agyemang, E. (2022, May 24). Landmark OECD
international tax deal pushed back a year. Financial Times.
https://www.ft.com/content/711d8f97-a2ce-48be-aa14-7f01e70f792f
Han, R., & Melecky, M. (2013). Financial inclusion for financial stability: Access to
bank deposits and the growth of deposits in the global financial crisis. World
bank policy research working paper 6577. World Bank.
IFC. (2017). Digital financial services: challenges and opportunities for emerging
market Banks. EM Compass Report, No 42, August. International Finance
Corporation, World Bank. Available: https://www.ifc.org/
wps/wcm/connect/4e45d83f-e049-41d3-8378-2e388ffc1594/EMCompassþ
Noteþ42þDFSþChallengesþupdated.pdf?MOD¼AJPERES.
IMF. (2010). A Fair and Substantial Contribution by The Financial Sector—Final
Report For The G-20. International Monetary Fund.
https://www.imf.org/external/np/g20/pdf/062710b.pdf
Jeven. (2021, December). PPN Dominasi Penerimaan Negara Maju. Pajakku.
https://www.pajakku.com/read/6188fc6c4c0e791c3760bdd3/PPN-Dominasi-
Penerimaan-Pajak-Negara-Maju
Khan, H. R. (2011). Financial inclusion and financial stability: are they two sides of
the same coin. Speech at BANCON. BIS Working Paper. Available:
http://www.bis.org/review/r111229f.pdf
Lyman, T., & Lauer, K. (2015, March 15). What is Digital Financial Inclusion and
Why Does it Matter? Https://Www.Cgap.Org/. Retrieved August 10, 2022, from
https://www.cgap.org/blog/what-digital-financial-inclusion-and-why-does-it-
matter
Maherali, A. (2017). Financial Inclusion, Digital Payments and Their Impact on
Income and Tax Revenue Around the World.
https://dash.harvard.edu/handle/1/33826588
Manyika, J., Lund, S., Singer, M., White, O., & Berry, C. (2016). Digital Finance For
All: Powering Inclusive Growth in Emerging Economies (p. 115). McKinsey
Global Institute.
5
https://www.mckinsey.com/~/media/mckinsey/featured%20insights/Employme
nt%20and%20Growth/How%20digital%20finance%20could%20boost%20gro
wth%20in%20emerging%20economies/MGI-Digital-Finance-For-All-
Executive-summary-September-2016.ashx
Ministry of Finance. (2022). PMK No 69/PMK.03/2022 Tentang Pajak Penghasilan
Dan Pajak Pertambahan Nila! Atas Penyelenggaraan Teknologi Finansial.
Ministry of Finance. https://jdih.kemenkeu.go.id/download/11108bd4-6b31-
4111-a5a9-972969c7be40/69~PMK.03~2022Per.pdf
Mpofu, F. Y. (2022). Industry 4.0 in Financial Services: Mobile Money Taxes,
Revenue Mobilisation, Financial Inclusion, and the Realisation of Sustainable
Development Goals (SDGs) in Africa. Sustainability, 14(14), 8667.
https://doi.org/10.3390/su14148667
Munoz, L., Mascagni, G., Prichard, W., & Santoro, F. (2022). Should Governments
Tax Digital Financial Services? A Research Agenda to Understand Sector-
Specific Taxes on DFS. Institute of Development Studies, 136.
https://doi.org/10.19088/ICTD.2022.002
Noverius Laoli. (2019, October 29). Pendapatan pajak dari sektor keuangan tumbuh
positif, ini kata pengamat pajak. kontan.co.id.
https://nasional.kontan.co.id/news/pendapatan-pajak-dari-sektor-keuangan-
tumbuh-positif-ini-kata-pengamat-pajak
OECD. (2019, November 19). Africa: Urgent action needed to mobilise domestic
resources as tax revenues plateau - OECD. OECD.
https://www.oecd.org/tax/africa-urgent-action-needed-to-mobilise-domestic-
resources-as-tax-revenues-plateau.htm
OECD. (2020). Tax Challenges Arising from Digitalisation – Economic Impact
Assessment: Inclusive Framework on BEPS. OECD.
https://doi.org/10.1787/0e3cc2d4-en
OECD. (2022a). Public consultation document: Pillar One – Amount A: Regulated
Financial Services Exclusion. The OECD Secretariat.
https://www.oecd.org/tax/beps/public-consultation-document-pillar-one-
amount-a-regulated-financial-services-exclusion.pdf
OECD. (2022b). Two Pillar Solution to Address the Tax Challenges Arising from the
Digitalisation of the Economy: Frequently asked questions. OECD.
https://www.oecd.org/tax/beps/faqs-two-pillar-solution-to-address-the-tax-
challenges-arising-from-the-digitalisation-of-the-economy-july-2022.pdf
OECD and Taxation of the Digital Economy. (n.d.). Bloomberg Tax. Retrieved
August 4, 2022, from https://pro.bloombergtax.com/brief/oecd-taxation-digital-
economy/
OJK & Kemenkeu. (2019). Buku 10—Aspek Perpajakan Sektor Jasa Keuangan.
OJK.
https://sikapiuangmu.ojk.go.id/FrontEnd/LiterasiPerguruanTinggi/assets/pdf/B
uku%2010%20-%20Aspek%20Perpajakan%20Sektor%20Jasa%20Keuangan
.pdf
Olivola, C., & Sussman, A. (2015). Taxes and Consumer Behavior. In The
Cambridge Handbook of Consumer Psychology.
https://doi.org/10.1017/CBO9781107706552.021
Ozili, P. K. (2018). Impact of digital finance on financial inclusion and stability. Borsa
Istanbul Review, 18(4), 329–340. https://doi.org/10.1016/j.bir.2017.12.003
Ozili, P. K. (2020). Financial Inclusion Research Around the World: A Review. SSRN
Electronic Journal. https://doi.org/10.2139/ssrn.3515515
Oz-Yalaman, G. (2019). Financial inclusion and tax revenue. Central Bank Review,
19(3), 107–113. https://doi.org/10.1016/j.cbrev.2019.08.004
6
Philippon, T. (2019). On Fintech And Financial Inclusion. Nber Working Paper
SerieS, Working Paper 26330. http://www.nber.org/papers/w26330
Phillipon, T. (2016). The Fintech Opportunity. NBER Working Paper.
https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2819862
Pilla, R. (2016). Person-to-Government payments: Lessons from Tanzania’s
digitization efforts (Case Study). Better Than Cash Alliance. https://btca-
production-
site.s3.amazonaws.com/documents/237/english_attachments/Tanzania-Case-
Study.pdf?1515010379
Puławska, K. (2021). Taxation of the financial sector: Is a bank levy the answer to
the financial crisis? Journal of Banking Regulation.
https://doi.org/10.1057/s41261-021-00178-w
Reed, L. W. (1982, May 1). The Role of Incentive | Lawrence W. Reed.
https://fee.org/articles/the-role-of-incentive/
Sarkodie, A. O. (2022, February 10). Ghana’s electronic transaction tax: Not a bad
idea, but must be properly designed. The Conversation.
http://theconversation.com/ghanas-electronic-transaction-tax-not-a-bad-idea-
but-must-be-properly-designed-176595
Shackelford, D. A., Shaviro, D. N., & Slemrod, J. (2012). Taxation and the Financial
Sector. Oxford University Press.
https://doi.org/10.1093/acprof:oso/9780199698165.003.0006
The Emergence of TaxTech—A New Era of FinTech. (n.d.). Retrieved August 12,
2022, from
https://www2.deloitte.com/content/dam/Deloitte/global/Documents/Financial-
Services/gx-fsi-tax-tech-report.pdf
Triyan Pangastuti. (2022, July 11). BI Ungkap Sektor Digital Jadi Penyelamat
Ekonomi RI di Masa Pandemi. investor.id.
https://investor.id/business/299348/bi-ungkap-sektor-digital-jadi-penyelamat-
ekonomi-ri-di-masa-pandemi
Twarowska-Ratajczak, M., & Szołno-Koguc, J. (2014). Taxation of the financial
sector in the European Union – an attempt to find the best solution. Journal of
Economics and Management, 18.
World Bank. (2015). Global Findex Database 2014: Measuring Financial Inclusion
around the World.
https://thedocs.worldbank.org/en/doc/681361466184854434-
0050022016/original/2014GlobalFindexReportDKSV.pdf
World Bank. (2018, October 1). UFA2020 Overview: Universal Financial Access by
2020. The World Bank IBRD IDA.
https://www.worldbank.org/en/topic/financialinclusion/brief/achieving-
universal-financial-access-by-2020

Proceeding Book of International Tax Conference 2022 2.0 (2)_compressed (1)

Uploaded by

Copyright:

Available Formats

You might also like

Proceeding Book of International Tax Conference 2022 2.0 (2)_compressed (1)

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Proceeding Book of International Tax Conference 2022 2.0 (2)_compressed (1)

Uploaded by

Copyright:

Available Formats

Proceeding Book of

International Tax Conference

7th December 2022

Advisor Material Division

Vice Team Leader I Publication Division

First Edition, 2023

Jakarta, Februari 2023

Graph 1: Type of income taxes in Indonesia and their contribution to total

2.2. Tax Avoidance and Taxpaying Behaviour

2.3. Artificial Neural Networks (ANN)

Table 1: Overview of benefits and drawbacks of the classification algorithms

(a) ANN architecture (b) ANN activation node

Figure 1: ANN architecture and neural network activation node

𝑂𝑢𝑡𝑝𝑢𝑡 = 𝑓(𝑤1 𝑥1 + 𝑤2 𝑥2 + ⋯ + 𝑤𝑛 𝑥𝑛 + 𝑏𝑖𝑎𝑠) = 𝑓(∑ 𝑥𝑖 𝑤𝑖 + 𝑏𝑖𝑎𝑠)

Table 2: Variables under study

3.2. Analytical Approach

4. RESULTS AND DISCUSSIONS

Table 3: Descriptive Statistics and Relationships between Tested Variables

The Pearson's correlation coefficient was further employed to assess the

Panel A: Business sector (n = 538,254) Panel B: Annual turnover category (n = 538,254)

Table 4: ANOVA Results of Variables Under Study by Annual Turnover

Panel A. Gross profit margin Panel B. Operating profit margin

Graph 3: The Difference of Variable Means Among Four Groups

4.3. Artificial Neural Networks (ANN)

Table 5: Case Processing Summary

Small firms (n = 224,181) Medium firms (n = 166,171)

Graph 4: Structure of Neural Networks for the Model Prediction

Table 7: ANN Model Summary

Table 8: Accuracy of classification

Panel A: ROC Curves

Medium-large firms (n = 61,572) Large firms (n = 86,330)

Panel B: Area Under the Curve

Table 9: Independent Variable Importance

For easier comprehension, the results described in Table 9 are diagrammatically

Small firms (n = 224,181) Medium firms (n = 166,171)

Medium-large firms (n = 61,572) Large firms (n = 86,330)

Graph 6: Independent Variable Importance (IVI) by Annual Turnover Category

Table 10: Summary: Accuracy Rate, Predictors, and Areas of Concern

These findings, while preliminary, suggest that it is possible to minimise the

6. IMPLICATIONS AND LIMITATIONS

Note: A = Agriculture, B = Mining and quarrying, C = Manufacturing, D = Electricity and

Appendix 3: Cumulative gain chart

Medium-large firms (n = 61,572) Large firms (n = 86,330)

Medium-large firms (n = 61,572) Large firms (n = 86,330)

0,00 0,05 0,10 0,15 0,20 0,25 0,30

Graph 1. Ten Countries with the Most Efficient Tax System

Graph 2. Ten Countries with the Most Complex Tax System

1. LITERATURE REVIEW AND HYPOTHESIS

Figure 1. A Simple Moderation Model

In a moderation model, the relationship between X and Y (the effect of X on Y) is

Y = 𝛂 + b1X + b2W + b3XW + 𝛆

Figure 2. A Moderation Model in the Study

TCI = 𝛂 + b1RegQual + b2Govern + b3RegQual*Govern + 𝛆

Subsequently, SPSS version 27 will be used to analyze the datasets. The

Table 1. Descriptive Statistics

Next, before analyzing the correlation, a normality test is conducted to determine

Table 2. Tests of Normality (K-S)

4.1.2 Moderated Regression Analysis

Graph 3. Interaction Plots of Regulatory Quality and Tax Complexity as Moderated by