Tax Evasion Influence in Public Debts

You might also like

Download as docx, pdf, or txt
Download as docx, pdf, or txt
You are on page 1of 12

IS NOT clearly written in a concise grammatical correct English manner

Tax Evasion effects on public debt in EU countries

ABSTRACT

In this article we study the influence of tax evasion phenomena on public debt in EU
countries. For this purpose, we chose the PMG method over the period 2016-2021.
Our study shows that tax evasion has a positive influence on public debt in the long-
run dynamics, while this influence is negative in the short-run dynamics.

Keywords: tax evasion, public debt, EU zone, PMG

Introduction

The effects of tax evasion on public debt are not known for sure. However, almost all
literature on our topic agrees that tax evasion is detrimental to the well-being of any economy.
In fact, tax evasion is an obstacle to growth, highlighting at least poverty. Also, tax evasion is
reducing capital productivity, and the state disposable income (Aghion et al, 2016).

To our knowledge, however, there is no previous study providing a precise negative effect of
tax evasion on public debt. Investigating the impact of tax evasion on public debt in a sample
of 126 countries over the period 1996 to 2012 by simultaneously applying the OLS, fixed
effects, GMM and instrumental variables methods, (Cooray et al, 2017) shows that an
increase in tax evasion leads to an increase in public debt. By applying the GMM method on a
large sample of countries over the period 1995-2015, (Benfratello et al, 2017) come to the
conclusion that tax evasion leads to an increase in public debt.

Tax evasion and corruption were found to be an important explanatory determinant of public
debt in Spanish autonomous groups (González-Fernández et al, 2014). Similar results have
been found in the case of OECD countries (Grechyna, 2012).

Our work aims to contribute to the literature that exists around our topic of study by making
an empirical study of how public debt is affected by tax evasion in the case of EU zone
countries. In this work, we seek to find out how tax evasion affects public debt in EU zone
countries over the period 2016 to 2021.
In order to answer this question, we use the technique of robust cointegration with inter-
individual dependence, which is particularly important in the context of our study as our
country sample is made up of a set of countries that belong to the same zone and are therefore
likely to present certain similarities. We also use a set of unit root tests on panel data. The
rest of the paper is organised as follows: in section 2, we present the data and the
methodology. In section 3, we present and interpret the estimation results and conclude in
section 4.

Description of data and research methodology

For this research, we use a sample of 11 countries 1 in the European zone over the period from
2016 to 2020. The availability of data justifies the choice of this study period. Our dependent
variable is the ratio of public debt to GDP, measured as the ratio of general government gross
debt to GDP.
Our data are taken from various databases including : WEO (World Economic Outlook), IMF
(International Monetary Fund). We use the corruption index of the International Country Risk
Guide (ICRG) to capture the corruption indicator that constitutes our independent variable of
interest. The choice of this index is justified by two reasons, considered here as advantages of
using this index. First, this corruption index covers a longer period of study, since 1984 while
taking into account a larger number of countries (36 countries), and second, this index is not a
composite index, so its annual comparison is more reliable compared to some other indices
such as Transparency International and World Governance Indicators (Hessami, 2014). It is
important to note that the ICRG corruption index has values ranging from 0 (highly corrupt
countries) to 6 (low corrupt countries).
In order to better interpret our results, this indicator has been normalised and re-parameterised
according to an intuitive logic, so that a high score refers to a high risk of corruption and a
low score refers to a low risk of corruption. Thus, we define the corruption variable Corrup, as
follows:
Corrupt = 1 - (ICRGt. score / ICRGmax) equivalent to corrt = 1- (ICRGt/6 score).

1
Burkina Faso, Côte d'Ivoire, Mali, Niger, Senegal, Togo, Cameroon, Republic of Congo, Gabon, Equatorial
Guinea and Guinea-Bissau. Benin, Chad and the Central African Republic are not included in the panel due to
the unavailability of data for certain variables in our model.
Statistical analysis

Variable Obs Mean Std. Dev. Min Max sources


Corrup 176 1.881866 .5671236 1 3.75 ICRG
Dettepub 176 58.39709 48.29697 .585 228.89 WEO
Depubl 176 78.53851 21.60199 16.71296 111.9283 WDI
Stabilipo 176 .8423295 .7057994 0 2 ICRG
Com 176 82.76464 48.07443 30.73252 351.1057 WDI
Table 1: Descriptive statistics table

Sources: WEO: World Economic Outlook; WDI: World Development Indicators; ICRG: International, author's
calculations.

From the table above we can see that not all standard deviations are strong, which
means that the variances are minimal between these variable values. It is therefore not
appropriate to log-transform them as is often the case in order to normalise the series.
Although our control variables that we use in this work are not identical, we have selected
them by focusing on the previous literature (Cooray et al, 2017; Benfratello et al, 2017).
Among these control variables, we were able to select the corruption variable, public
expenditure as a percentage of GDP, political stability and trade openness. In Table (1) above,
we present the descriptive statistics and data sources for our different variables.
Dettepub Corrup Depubl Stabil~o Com
Dettepub 1.0000
Corrup -0.1984 1.0000
Depubl 0.3697 0.1290 1.0000
Stabilipo -0.2520 0.1427 -0.2946 1.0000
Com -0.1553 0.0598 -0.5343 0.0731 1.0000
Table 2: Correlation matrix table
Sources: WEO: World Economic Outlook; WDI: World Development Indicators; ICRG: International Country
Risk Guide, author's calculations.

It can be seen from table (2) above that the correlations between our different explanatory
variables are not strong enough to cause problems of muticollinearity between the explanatory
variables.
4
3
2
1

0 50 100 150 200 250


Dettepub
Fitted values Tax evasion

Figure 1: Scatterplot between corruption and public debt

Sources: WEO: World Economic Outlook; ICRG: International Country Risk Guide, author's calculations.

Figure 1 above presents the correlation between the corruption variable as measured by the
ICRG index and the variable representing public debt as measured by the public debt
indicator in the WEO database. The scatterplot shows a negative relationship between
corruption and public expenditure. Econometric methods are needed to explore the
longitudinal nature of our data and to properly condition the link between public debt and
corruption.

Model specification
The main objective of this study is to analyse the effect of corruption on public debt in the Cfa
franc zone countries. To do so, we use a specification that is very close to other forms of
specifications often encountered in previous work, for example (Kaufman, 2010; Cooray et al,
2017). Taking into account the variables regularly encountered in the literature, we can
formulate our econometric model as follows:
Dettepubi ,t =α i+ β1 Corrupi ,t + β 2 X i , t +ε i ,t

where i is the country index (i =1 to n), t is the time index (t =1 to T), Debtr is the variable
representing public debt, Corrup is the corruption index of the ICRG, X is the column vector
of other control variables. The coefficients β i correspond to the parameters to be estimated
and α i being the vector of individual (country) fixed effects, ε corresponds to the error term.
In compact form, the model is written as follows:
Dettepub i ,t =α i+ β, x i , t +ε i , t

Where β represents the column vector of the parameters to be estimated β i and x i ,t is the
column vector of the explanatory variables of our model, the prime here represents the
transpose.

With the data thus declined, the theoretical model and its variables specified, and the
econometric methodology known, we can now proceed to the different estimations and the
interpretation of the results that will follow.

Unit root tests

We use two generations of unit root tests. We justify this choice by two reasons. The
countries of the CFA franc zone are not immune to heterogeneity linked to their different
economic structures (first generation tests) or even to inter-individual dependence linked to
the management and sharing of a common currency (second generation tests).

First Test in level First difference test Conclusion


generation Levinet alii) Maddala et Levinet alii Maddala et Wu I(1)
tests Wu (2002) (1999) (2002) (1999)

Corrup 15.8049 1.0780 -1.3e+04 4.4070 yes


(1.0000) (0.1405) (0.0000) (0.0000)
Dettepub -0.8346 1.3863 -8.9902 3.6781 yes
(0.2020) (0.0828) (0.0000) (0.0001)
Depulb -0.2786 -0.8513 -11.4222 6.6336 yes
(0.3903) (0.8027) (0.0000) (0.0000)
Stabilipo 2.6883 -1.7965 -1.9355 7.8627 yes
(0.9964) (0.9638) (0.0265) (0.0000)
Com -1.1001 1.0987 -10.5280 8.4246 yes
(0.1356) (0.1359) (0.0000) (0.0000)
Table 3: Panel unit root test results table
Second generation Test in level First difference Conclusion
tests test I(1)
Corrup -0.374 (-2.26) -2.370 (-2.28) yes
Dettepub -1.755 (-2.26) -3.632 (-2.28) yes
Depulb -2.079 (-2.26) -4.331 (-2.28) yes
Stabilipo 1.381 (-2.26) 0.585 (-2.28) not
Com -1.762 (-2.26) -3.543 (-2.28) yes
Table 4: Table of results of Pesaran panel unit root tests (2007)

Reading: The results of the tests applied to the variables in level and first difference, using a model with constant
and individual trend, are presented here. We have represented in brackets the p-values that are associated with
the different test statistics that precede them. The conclusions of these tests are as follows: to accept the null
hypothesis of non-stationarity if the critical probability is above the 5% threshold, and to reject this hypothesis in
the opposite case (and vice versa). All these tests are based on the null hypothesis of non-stationarity against an
alternative hypothesis of stationarity. NB: the final conclusion (integrated series of order 1) was obtained after
repeating these same tests on the variables in first differences. The tests then lead to stationarity.
Sources: ICRG, WEO, WDI, author's calculations.

The results of our different unit root tests that we present in tables (3) and (4) above
globally highlight the presence of a unit root in the variables studied for all our first
generation tests (Levin, Lin and Chu (2002) and Maddala and Wu (1999)) at the 5% 2
threshold. The second generation test of Pesaran (2007) for p=3 lags shows the presence of a
unit root in the dynamics of the variables 3, except for the variable Stabilipo. The results of the
tests on the variables in first difference show that they are stationary. We conclude that the
order of integration of our different variables is 1. These different results on the unit root tests
lead us to test if all our variables are linked by a long term relationship.

Cointegration tests

We use two generations of cointegration tests: Westerlund's tests (2007) and Pédroni's tests
(1995; 1997). Four of the seven statistics proposed by Pédroni reveal the presence of a
cointegrating relationship (see Table 5). We give priority to the tests based on the inter-

2
In order to find the optimal number of delays, it is possible to use the information criteria or to be set
arbitrarily by the modeller. In order to find the optimal number of delays for these tests, we use the
information criteria.
3
It should be noted that the robustness of the results depends on the choice of the number of delays. Indeed,
for this test (Pesaran test), the modeller has to arbitrarily set the number of upstream delays. In order to check
the robustness of the results, it is necessary to vary the number of delays and to study its effect on the results.
We varied several delays and our conclusions remained unchanged regardless of the delay tested.
individual dimension because the homogeneity of the panel is unlikely 4, which leads us to
conclude that there is a cointegrating relationship between public debt and its different control
variables. This conclusion is consistent with two of the four statistics provided by the
Westerlund test (see line Pt and pa in Table 5, associated with heterogeneity of the
cointegrating vectors). It is now possible to estimate the parameters of the cointegrating
vector. To do this, we use the PMG method.

Tests de Pedroni (1995, 1997) Tests de Westerlund (2007)


Alternative hypothesis: common autoregressive coefficients Model with trend and constant
(intraindividual dimension
Tests Statistique P-value Tests Statistique Z-Statistique P-value
Vc 5%
Panel : .6337 1.64 Gt 2.518 14.503 1.000
Statistiquev
Statistique rho 2.633 -1.64 Ga 75.612 39.870 1.000
Statistique t -1.423 -1.64 Pt -5.448 -0.068 0.000
Statistique 4.896 -1.64 Pa -3.871 1.018 0.000
ADF
Hypothèse alternative : coefficients autorégressifs individuels
(dimension interindividuelle)
Statistique P-value
Groupe: 3.703 -1.64
Statistique rho
Statistique t -1.804 -1.64
Statistique 5.544 -1.64
ADF
Table 5: Co-integration test results table

Reading: similar to the approach taken in the case of unit root tests, we present here only the results for models
with constant and trend. For the Westerlund tests, the number of lags and leads is determined from the Akaike
information criterion (AIC), by setting a relatively small lag and lead interval given the small number of
observations per series. The width of the Bartlett kernel window that was used in the semi-parametric estimation
of the long-term variances was determined from the formula 4 ( T /100 ) 2 /9 =3, where T is the number of time
series observations. Only the probabilities calculated using the Bootstrap5 technique are reported. The presence
of a possible cointegrating relationship is detected if the value of the P-value is lower than the 5% threshold
retained for the tests. Z-statistic represents the statistic obtained following the Bootstrap procedure allowing for
inter-country dependence.
Sources: ICRG, WEO, WDI, author's calculations.

Econometric estimation and interpretation of results

The PMG (Pooled Mean Group) approach presented by Pesaran et al. Firstly, this approach
allows the heterogeneity of countries to be taken into account, secondly, this approach allows
the dynamics of countries to be taken into account and finally, it takes into account the non-
stationary nature of the variables. This approach makes it possible to include heterogeneity in
the short-term parameters while maintaining homogeneity in the long-term parameters. This is

4
Indeed, although the countries of the CFA franc zone share a common currency, the countries of the CEMAC
zone, which are for the most part oil-producing countries, are characterised by a higher volatility of their
indicators compared to the countries of the WAEMU zone.
5
The Bootstrap is a statistical inference technique based on a succession of resamplings allowing, among other
things, a very fine sensitivity analysis. In the case of this test, the Bootstrap makes it possible to calculate
probabilities that are robust to the presence of common factors in the time series
particularly relevant in the context of our study, since the convergence criteria established for
the Cfa franc zone should eventually make it possible to "erase" the heterogeneity that
currently exists. As a result, the PMG approach makes it possible to determine the speed of
adjustment of the public debt towards the long term equilibrium.

The dynamic panel specification of the model expressed as an error correction model is
given as follows:

p−1 q−1
∆ Debt i , t=v i ( Debt i ,t −1−β , x i ,t )+ ∑ λi , τ ∆ Debt i ,t −τ + ∑ δ ,i , τ ∆ xi , t−τ + μ i+ ε i ,t
τ=1 τ=0

The term in the brackets contains the long-term parameters. The speed of adjustment
towards the long-run equilibrium is represented by the parameter vi . The model does not
admit a long-run relationship for vi ≥0. For vi significantly less than zero with an absolute
value not too large, then there is a long-run relationship. More particular importance is given
to the parameter β containing the parameters of the long-run relationship between the
variables.

variables Coefficients Std. Err. P>|z|


Corrup 16.61849 9.236034 0.042
Depubl 2.191617 .7151138 0.002
Stabilipo 43.02736 32.30425 0.183
Com .7297429 .3471399 0.036
Table 6: Estimated long-term relationships table (PMG method)

variables Coefficients Std. Err. P>|z|


Δ(Corrup) -15.44155 5.816704 0.005
Δ(Depubl) .1852726 .3341729 0.579
Δ(Stabilipo) -11.80545 8.527218 0.166
Δ(Com) -.3061407 .1315878 0.020
ECM(-1) -.1917492 .0686679 0.005
Table 7: Estimated table of short-term relationships (PMG method)

Reading: Δ denotes the first difference operator. In tables (6) and (7) above, we have plotted the estimated values
of the long-run coefficients (table 6) and the simple arithmetic means of the estimated short-run coefficients per
country (table 7). The models are estimated taking into account individual fixed effects.
Sources: ICRG, WEO, WDI, author's calculations.

The coefficient associated with the corruption variable, which is our independent
variable of interest, contributes significantly and positively to the explanation of the variable
Public debt at the 5% threshold in the long-run relationship. This result means that an increase
in the level of corruption leads to an increase in the amount of public debt. The negative
character of corruption on any economy is confirmed by this result. In particular, this result
confirms the negative effect of corruption on the amount of public debt. This result is in line
with the results previously found by (Cooray et al, 2017; Benfratello et al, 2017). According
to these authors, there is a positive relationship between corruption and public expenditure.
Conversely, the effect of corruption on public debt is significantly negative in the short run.
On the other hand, other control variables, namely: public expenditure, political stability and
trade openness have positive and significant effects on public debt in the long run. In the short
run, political stability and trade openness have a negative impact on public expenditure but
this negative impact is only significant for the trade openness variable and not for the political
stability variable. However, the short-run coefficient of the Depubl variable has a positive but
insignificant influence on public expenditure.

The estimation results show that the coefficient associated with the variable measuring
the recall force (ECM (-1)) is significantly negative. This result confirms the existence of a
stable long-run relationship between corruption and the different explanatory variables of our
model.

Conclusion

The main objective of this study was to analyse the effect of corruption on public debt in 11
countries of the Cfa franc zone during the period 2000 to 2015. A set of other variables such
as trade openness, political stability and public expenditure were taken into account in our
analysis. Our results revealed that corruption measured by the ICRG corruption index has a
significant and positive influence on public debt. Based on the above results, we formulate a
policy recommendation that is not without relevance. Many African countries in the franc
zone have been steadily increasing their anti-corruption programmes. However, the expected
results are very unsatisfactory and mixed. The countries of the Cfa franc zone must therefore
step up their anti-corruption measures in order to make their public spending more efficient
and, above all, to reduce their debt. The countries of the Cfa franc zone must therefore
increase their efforts to fight corruption in order to make their public spending more efficient
and, above all, to reduce the volume of their sovereign debt.

REFERENCES
Aghion, P., Akcigit, U., Cage, J. and Kerr, W.R. [2016], « Taxation, Corruption, and Growth
», European Economic Review, 86, 24-51.
Benfratello, L., Del Monte, A. and Pennacchio, L. [2017], «Corruption and public debt: a
cross-country analysis », Applied Economics Letters, 24(5) 340-344.
Cooray, A., Dzhumashev, R. and Schneider, F. [2017], « How Does Corruption Affect Public
Debt? An Empirical Analysis », World Development, Vol. 90, 115–127.
d’Agostino, G., Dunne, J. P. and Pieroni, L. [2016], « Government Spending, Corruption and
Economic Growth », World Development, 84, 190-205.
Egger, P. and Winner, H. [2006], « How Corruption Influences Foreign Direct Investment: A
Panel Data Study », Economic Development and Cultural Change, Vol. 54(2), 459-
486.
González-Fernández and González-Velasco, [2014], « Shadow Economy, Corruption and
Public Debt in Spain. Journal of Policy Modeling, 36 (6), 1101–1117.
Grechyna, D. [2012], « Public corruption and Public Debt: Some Empirical Evidence »,
University of Auckland working paper.
Gupta, S., De Mello, L. and Sharan, R. [2001], « Corruption and military spending »,
European Journal of Political Economy, Vol. 17(4), 749–777.
Hessami, Z. [2014], «Political corruption, public procurement, and budget composition:
Theory and evidence from OECD countries », European Journal of Political Economy,
Vol. 34, 372-389.
Imam, P. and Jacobs, D. [2007], « Effect of Corruption on tax revenues in the Middle East »,
IMF Working Paper, 07/270, International Monetary Funds, Washington, DC.
Justesen, M. K. and Bjørnskov, C. [2014], « Exploiting the Poor: Bureaucratic Corruption and
Poverty in Africa », World Development, Vol. 58, 106-115.
Lambsdorff, J. [2003b]. How Corruption Affects Productivity », Kyklos, Vol. 56, 457-474.
Levin A.T., Lin C.F. and Chu C.-S.J. [2002], « Unit Root Tests in Panel Data: Asymptotic
and Finite-Sample Properties », Journal of Econometrics, vol. 108, n° 1, pp. 1-24.
Maddala G.S. and Wu S. [1999], « A Comparative Study of Unit Root Tests with Panel Data
and a New Simple Test », Oxford Bulletin of Economics and Statistics, vol. 61, n° S1,
pp.631-652.
Mo, P. K. [2001], « Corruption and Economic Growth », Journal of Comparative Economics,
Vol. 29, 66–79.
Pedroni P. [1995], « Panel Cointegration, Asymptotic and Finite Sample Properties of Pooled
Time Series Tests with an Application to the PPP Hypothesis », Working Paper in
Economics n° 92-013, Indiana University, 29 pages.
Pedroni P. [1997], « Panel Cointegration, Asymptotic and Finite Sample Properties of Pooled
Time Series Tests with an Application to the PPP Hypothesis: New results », Working
Paper in Economics, Indiana University, 29 pages.
Pesaran H.M. [2007], « A Simple Panel Unit Root Test in the Presence of Cross Section
Dependence », Journal of Applied Econometrics, vol. 22, n° 2, pp. 265-312.
Transparency international [2016], « Corruption perception index»,
Wei, Shang-Jin, [2000], « How Taxing is Corruption on International Investors? », Review of
Economics and Statistics, Vol. 82, 1-11.
Westerlund J. [2007], « Testing for Error Correction in Panel Data », Oxford Bulletin of
Economics and Statistics, vol. 69, n° 6, pp. 709-748.

You might also like