Professional Documents
Culture Documents
Researchgate TSSA2017 HaryadiIndex
Researchgate TSSA2017 HaryadiIndex
net/publication/320728316
CITATIONS READS
11 119
2 authors:
1 PUBLICATION 11 CITATIONS
Bandung Institute of Technology
118 PUBLICATIONS 297 CITATIONS
SEE PROFILE
SEE PROFILE
All content following this page was uploaded by Sigit Haryadi on 31 October 2017.
Abstract— This research is focused in a formula called Haryadi Index, which implemented in the
assessment of telco industry. This formula will be contested with the other existing formula to gauge
competition level, equality level, fairness level, and correlation level. The Herfindahl-Hirschman Index will
be compared against Haryadi Index for competition level, Gini Index will be compared for equality level,
Jain Index for fairness level, and Pearson Correlation for the correlation level calculation.
The research produces the following : (1) Haryadi Index can provide better number for an industry with
a different level of competition, whereas the Herfindahl-Hirschman Index do worse; (2) Haryadi Index has
more precise equality level coefficient than the Gini Index; (3) Haryadi Index can give more appropriate
indexes for a variety levels of fairness, while Jain Index gives a relatively bigger indexes for a variety levels
of fairness; (4) Haryadi Index gives a new alternative for correlation counting which refer to the linear
equation, while more research will be needed to find the better situation to use the Haryadi Index or the
Pearson Correlation appropriately.
Keywords— Statistics, Haryadi Index, Herfindahl-Hirschman Index, Gini Index, Jain Index, Pearson
Correlation.
I. INTRODUCTION
In April 2016, Sigit Haryadi, Institut Teknologi Bandung, found a formula which he claimed to be better and
more precise than other existing method. The formula, which called Haryadi Index, is used for statistical
analysis such as competition, equality, fairness, and correlation level analysis.
In telecommunication industry, there are many assessments used for analyses. Some of these assessments
were developed by experts from law, sociology, economics, and statistics. These analyses is very important
not only for companies, but also for governments. They usually use these assessments for making policies or
products according to the assessment results. This is a problem because different analysis usually need
different formula. Haryadi Index presents a new solution for tackling these problems using only 1 formula.
So, this paper will present some comparison results between Haryadi Index and the existing methods of
some analyses such as Herfindahl-Hirschman Index, Gini Index, Jain Fairness Index, and Pearson Correlation.
1
II. THEORIES
A. Competition Level
Competition Level is seen as a level which sees the health of the industry. A Perfect Competition Level is
marked with many companies with identical products, while they also cannot control the market price. The
other way, an Unhealthy or Highly Concentrated Competition Level means there is a monopoly in the market.
An established way to calculate the Competition level is using the Herfindahl-Hirschman Index (HHI). HHI
was found by Orris Herfindahl and Albert Hirschmann. The HHI is defined as,
𝐻 = ∑𝑁 2
𝑖=1 𝑆𝑖 (1)
where N is the number of the companies in the market, while Si is the market share of the i-th company. HHI
1
is valued from 𝑁 until 1.
TABLE I HHI COMPETITION LEVEL
HHI Competition Level
H ≤ 0.01 Highly Competitive
H ≤ 0.15 Unconcentrated
0.15 < H ≤ 0.25 Moderate Concentration
H > 0.25 High Concentration
Table I shows the level of the competition defined by the HHI.
B. Equality Level
Equality Level is usually used as a statistical measure to represent the income or wealth distribution of a
nation’s residents. The commonly used formula to measure the equality level is the Gini Index.
Gini Index (GI) or also called as Gini coefficient was introduced by Corrado Gini in 1909 [10]. GI original
formula is defined as
∑𝑁 𝑁
𝑖=1 ∑𝑗=1|𝑥𝑖 −𝑥𝑗 |
𝐺(𝑁) = (2)
2 ∑𝑁 𝑁
𝑖=1 ∑𝑗=1 𝑥𝑗
where xi and xj is the wealth or income of person i, and there are N persons.
TABLE II GI EQUALITY LEVEL
GI Equality Level
G < 0.3 Low Inequality
0.3 ≤ G ≤ 0.5 Moderate Inequality
G > 0.5 High Inequality
Table II shows the level equality defined by the Gini Index [10]. Gini Index is valued from 0 to 1. The lower
the GI is, the more equal the income or wealth a person is.
C. Fairness Measure
Fairness Measure is a method used in network engineering to determine if the users receive the fair share
of the throughput.
Fairness Measure is usually calculated using Fairness Index or also known as Jain Fairness Index (JI). JI was
found by Rajendra Jain in his paper on September 1984. JI is defined by [1] as,
2
2
(∑𝑛
𝑖=1 𝑥𝑖 )
𝑓(𝑛) = ∑𝑛 2 , 𝑥𝑖 ≥ 0 (3)
𝑖=1 𝑥𝑖
1
where xi represent the throughput for the ith user connection and n is the users. JI ranges from 𝑁 until 1.
The higher JI, the more allocation the users receive.
D.Correlation Coefficient
In Statistics, correlation coefficient is a number which shows relation of two random variables.
Correlation can also predict that there is a causal relationship from some event.
There are several correlation coefficient which is used, but one of the most common is Pearson
Correlation Coefficient (PCC). PCC was developed by Karl Pearson in 1880. This method is sensitive only to a
linear relationship between two variables. PCC is defined as,
∑𝒏
𝒊=𝟏(𝒙𝒊 −𝒙
̅)(𝒚𝒊 −𝒚
̅)
𝒓= (4)
√∑𝒏 ̅)𝟐 √∑𝒏
𝒊=𝟏(𝒙𝒊 −𝒙 ̅)𝟐
𝒊=𝟏(𝒚𝒊 −𝒚
where xi and yi is the sample of X and Y series, for i=1,2,3,..,n. While 𝑥̅ and 𝑦̅ is the sample mean value of X
and Y, the PCC is defined as r.
PCC ranges from -1 to +1. Value of -1 means the samples have a strong negative correlation, 0 value
means no correlation, and +1 means a strong positive correlation.
E.Haryadi Index
Haryadi Index (HI) was found by Sigit Haryadi on April 2016. This statistical formula offers a versatility
which other formula cannot provide. [4, 6, 8, 11]
Haryadi Index has 2 important factor in the formula:
1. Element Harmony
This element calculates the harmony degree of a system. If this elements is wiped, the formula
function to measure the competition or fairness level fails to do its work properly.
𝑬𝒍𝒆𝒎𝒆𝒏𝒕 𝑯𝒂𝒓𝒎𝒐𝒏𝒚 = ∑𝑵 𝟐
𝒊=𝟏 𝑺𝒊 (5)
Si is the share of the ith sample, while N is the total sample population.
2. Element Gradation
This element functions as a consistent gradation for if the data contains diverse elements. If this
element is wiped it will affect the gradation of the formula to be inconsistent.
𝟐
𝑬𝑮(𝑵) = ∑𝑵 𝑵
𝒊=𝟏 ∑𝒋=𝟏(𝑺𝒊 − 𝑺𝒋 ) (6)
Si and Sj is the share of the ith and jth sample, while N is the total sample population.
The complete HI formula is defined as,
𝟏
𝑯𝑰(𝑵) = 𝟐 (7)
𝑵{∑𝑵 𝟐 𝑵 𝑵
𝒊=𝟏 𝑺𝒊 +∑𝒊=𝟏 ∑𝒋=𝟏(𝑺𝒊 −𝑺𝒋 ) }
where Si and Sj is the share of the ith and jth sample, while N is the total sample population.
3
TABLE III HI COMPETITION LEVEL
HARYADI INDEX COMPETITION
HI = 1.00 Perfect
LEVEL
0.75 < HI < 1.00 Fair
0.50 < HI ≤ 0.75 Unbalanced
HI ≤ 0.50 Unfair
HI ≤ {(N-1)/2N} Concentrated
4
TABLE VII REVENUE DATA OF THE FIRST SCENARIO
Revenue (Billion Rupiahs)
Year
A B C Total
2013 60,031 23,855.27 22,876.18 106,762
2014 66,252 24,085.10 23,460.02 113,797
2015 76,055 26,768.53 22,876.06 125,700
5
Table X shows the result of the second scenario. After the merger of B and C, we can see that the revenue
of the companies are not too far away in Table VIII. But, the interesting thing is the competition level which
is stated by the HHI and HI. HHI still assumed that the competition level is “High Concentration” while HI is
“Fair”. Looking at the condition, this proves that HI is better than HHI because HI can produce Index number
which is closer to the real condition.
B. Equality Level
To prove that HI can be used in many cases there are 3 scenarios used for this.
First scenario, we used data of distributed income in the city, countryside, and both from [5]. Then, using
the data we calculate the equality level using HI and compare it with the existing method, Gini Index.
TABLE XI INCOME DISTRIBUTION DATA OF THE FIRST SCENARIO
40% 40% 20%
Year
Low Moderate High
2010 17.57 36.99 45.44
2011 16.10 34.79 49.11
2012 16.00 34.53 49.48
City
2013 15.40 34.83 49.77
2014 15.62 34.89 49.49
2015 15.83 34.60 49.57
2010 20.98 38.78 40.24
2011 19.96 37.46 42.58
Country 2012 20.60 37.57 41.82
side 2013 21.03 37.96 41.00
2014 20.94 38.40 40.65
2015 20.42 37.53 42.05
2010 18.05 36.48 45.47
2011 16.85 34.73 48.42
2012 16.98 34.41 48.61
Both
2013 16.87 34.09 49.04
2014 17.12 34.60 48.27
2015 17.10 34.65 48.25
Because the GI had already provided in the data, we only calculate HI using (7).
6
TABLE XII RESULT OF THE FIRST SCENARIO
40% 40% 20%
Year GI HI
Low Moderate High
2010 17.57 36.99 45.44 0.38 0.67
2011 16.10 34.79 49.11 0.42 0.60
2012 16.00 34.53 49.48 0.42 0.60
City
2013 15.40 34.83 49.77 0.43 0.58
2014 15.62 34.89 49.49 0.43 0.59
2015 15.83 34.60 49.57 0.43 0.59
2010 20.98 38.78 40.24 0.32 0.78
2011 19.96 37.46 42.58 0.34 0.75
Country 2012 20.60 37.57 41.82 0.33 0.77
side 2013 21.03 37.96 41.00 0.32 0.78
2014 20.94 38.40 40.65 0.32 0.78
2015 20.42 37.53 42.05 0.33 0.76
2010 18.05 36.48 45.47 0.38 0.68
2011 16.85 34.73 48.42 0.41 0.62
2012 16.98 34.41 48.61 0.41 0.62
Both
2013 16.87 34.09 49.04 0.41 0.62
2014 17.12 34.60 48.27 0.41 0.63
2015 17.10 34.65 48.25 0.41 0.63
The result of the first scenario is presented in Table XII. The important thing is GI actually shows result of
inequality and HI shows equality. We use table II and IV as a reference for GI and HI to define the equality
level.
According to GI, the equality level for income distribution in the city, countryside, and both are all
moderately unequal. For HI, the equality level for income distribution in the countryside is fair, while in the
city and both are all unbalanced.
But, if we calculate GI using (2) and the data in table XI the result differs with the one provided in the data.
This could mean that the GI provided was calculated using different modification of GI formula, or GI is
calculated using the real income and then combined with the income distribution data, or it can also be said
that GI proved to be inconsistent when calculated using different ways.
It can be concluded from the result that HI proved to be better than GI, because GI cannot produce
consistent result if there are slight changes.
But, the equality level result of HI from Table XII cannot be used as a reference because HI works by dividing
the share equally. In this case it should be ⅓ for each share of income distribution, not 40%:40%:20%.
The second scenario, using regional gross domestic product per capita (PDRB) from 33 provinces in
Indonesia according to [2] we calculate the equality level using HI only.
7
TABLE XIII DATA OF THE SECOND SCENARIO
PDRB (Thousand rupiahs)
Province
2011 2012 2013
Aceh 6,652.08 6,912.07 7,137.52
Sumatera Utara 9,515.62 9,971.72 10,431.66
Sumatera Barat 8,370.65 8,784.84 9,205.66
Riau 9,154.37 9,617.51 9,945.97
Jambi 5,292.70 5,646.58 6,021.42
Sumatera Selatan 7,157.47 7,609.60 8,049.08
Bengkulu 5,065.06 5,306.22 5,540.42
Lampung 5,230.60 5,502.79 5,755.31
Kep. Bangka Belitung 9,076.26 9,393.92 9,676.76
Kepulauan Riau 24,057.57 24,909.74 25,665.05
DKI Jakarta 43,195.94 45,509.95 47,774.70
Jawa Barat 7,613.76 7,988.92 8,355.29
Jawa Tengah 5,721.70 6,055.96 6,382.36
DI Yogyakarta 6,305.35 6,561.24 6,834.07
Jawa Timur 9,649.71 10,283.36 10,885.77
Banten 8,607.46 8,929.01 9,243.06
Bali 7,771.90 8,186.29 8,576.19
Nusa Tenggara Barat 4,263.21 4,157.39 4,334.10
Nusa Tenggara Timur 2,767.46 2,867.82 2,976.62
Kalimantan Barat 7,160.20 7,448.63 7,772.47
Kalimantan Tengah 8,825.07 9,194.19 9,644.55
Kalimantan Selatan 8,642.48 8,978.04 9,282.23
Kalimantan Timur 20,447.80 22,147.59 22,698.16
Sulawesi Utara 8,542.91 9,103.16 9,671.14
Sulawesi Tengah 7,027.34 7,559.01 8,156.16
Sulawesi Selatan 6,740.78 7,225.27 7,692.69
Sulawesi Tenggara 5,534.42 5,977.64 6,275.62
Gorontalo 2,956.50 3,132.33 3,321.11
Sulawesi Barat 4,406.87 4,711.52 4,952.51
Maluku 2,861.72 3,029.70 3,128.99
Maluku Utara 3,026.62 3,157.90 3,279.50
Papua Barat 8,280.36 8,668.87 9,106.88
Papua 7,274.75 7,208.25 8,117.64
Table XIII shows the data of regional gross domestic product per capita from 33 provinces in Indonesia.
8
Third scenario, using internet speed data of 5 cities in Indonesia from [9] we calculate the equality level
using HI only.
TABLE XIV RESULT OF THE SECOND SCENARIO
Year 2011 2012 2013
HI 3.88% 3.85% 3.89%
Using (7) with N=33, we calculated the HI and is presented in Table XIV. Province with the highest PDRB,
DKI Jakarta, compared to the lowest, East Nusa Tenggara, is almost 16 times bigger. It can be concluded that
the equality level of PDRB in Indonesia is not equal at all.
TABLE XV INTERNET SPEED DATA OF THE THIRD SCENARIO
Internet Speed
City
(Mbps)
Jakarta 7.8
Tangerang 7.25
Bekasi 4.29
Surabaya 4.17
Bandung 3.9
After calculating the internet speed share for each city and using (7), the result of the third scenario is HI
= 0.639071489. Refer to Table IV, this means that the equality level of internet speed in Indonesia is still
unbalanced. But, this equality cannot be used as a reference for internet speed research because the sample
used in this scenario is only 5 cities.
C. Fairness Level
In purpose of comparing HI and the existing method to calculate Fairness Level, Jain Index, one scenario is
hatched.
The scenario is planned using internet price of two telecommunication operators [3] and regional gross
domestic product per capita (PDRB) from 33 provinces in Indonesia [7] within 4 years from 2013. The internet
price will be divided with the PDRB of the same year, then the share will be used to measure the fairness level
using (3) for Jain Index and (7) for HI.
TABLE XVI DATA OF THE FAIRNESS LEVEL SCENARIO
Internet PDRB
Operator Year Price (Million
(Rp/kB) Rupiahs)
2013 80.6 1,222,374.52
2014 50.68 1,261,752.84
A
2015 36.83 1,296,737.25
2016 39.0648 1,334,626.06
2013 115.85 1,222,374.52
2014 52.5 1,261,752.84
B
2015 35.32 1,296,737.25
2016 64.33 1,334,626.06
9
TABLE XVII RESULT OF THE FAIRNESS LEVEL SCENARIO
Jain Fairness
Operator HI
Index
A 0.879487572 0.593426498
B 0.814089994 0.466891092
According to Table V, the fairness level of internet price from operator A is unbalanced and from operator
B is unfair. According to Jain Index, the fairness level of internet price from operator B is 87.9% fair and from
operator C is 81.4% fair.
D. Correlation Level
There is only one scenario used to measure the correlation level. Using data of revenue and total customer
from two operators, the correlation level will be measured using Pearson Correlation and HI.
The scenario is defined using the revenue and the total customer from two telecommunication operators
within 4 years from 2013 [3]. The revenue will be divided with the total customer of the same year, then the
share will be used to measure the correlation level using (4) for Pearson Correlation and (7) for HI.
TABLE XVIII DATA OF THE CORRELATION LEVEL SCENARIO
Revenue = y Customer
Operator Year (Billion =x
Rupiahs) (millions)
2013 22,876.18 60.549
2014 23,460.02 59.643
A
2015 22,876.06 42.1
2016 21,341.00 46
2013 23,855.27 59.6
2014 24,085.10 63.2
B
2015 26,768.53 69.7
2016 29,184.62 79.1
10
From the data, we can get the linear function for each operator. From the linear function we can get R2 =
0.2879 for B and R2 = 0.9971 for C. For HI linear function, the value of R2 is the same as the HI value.
The other clear difference is HI linear function is y = ax, while the usual linear function is y = ax + b.
Looking at the available data, it is yet to be concluded which method is the best. It is caused by limited
data and more case study needed to see which method is better.
CONCLUSION
IV.
From this research, we can see that Haryadi Index is better than some of the existing method. Herfindahl-
Hirschman Index can’t produce accurate result according to the scenario in this research.
Gini Index also can’t produce a consistent gradation changes if there are a slight changes in the data.
Haryadi Index can define more realistic fairness level than Jain Index from the data.
Haryadi Index has a different linear function than the usual linear function. But, it can’t be concluded yet
which method is better for measuring the correlation level from the data available. More case studies needed
to be able to prove which method is better.
REFERENCES
[1] R. K. Jain, D. M. W. Chiu, W. R. Hawe, “A Quantitative Measure of Fairness and Discrimination for Resoure
Allocation in Shared Computer System,“ Eastern Research Lab, 1984.
[2] https://www.bps.go.id/linkTabelStatis/view/id/1628. Accessed on 29th of August 2017.
[3] W. R. Agustin, S. Haryadi, “Sustainability of Indonesian Telecommunication Operators in the Era of Net
Neutrality,” 3rd ICWT, Palembang, 2017.
[4] S. Haryadi, “Indeks Haryadi dan Penerapan di Ilmu Hukum, Sosiologi, Ekonomi, Statistik dan
Telekomunikasi,” Elex Media Komputindo, 2016.
[5] https://www.bps.go.id/linkTabelStatis/view/id/946. Accessed on 4th of September 2017.
[6] S. Haryadi, “Equality Correlation Method,” Institut Teknologi Bandung, Notebook, 2017.
[7] https://www.bps.go.id/linkTableDinamis/view/id/958. Accessed on 4th of September 2017
[8] S. Haryadi, “Haryadi Index untuk Evaluasi Kompetisi, Kesetaraan dan Korelasi,” Lantip Safari Media,
2016.
[9] http://www.wantiknas.go.id/2016/10/07/tahukah-kamu-kota-mana-yang-memiliki-koneksi-internet-
tercepat-di-indonesia/. Accessed on 29th of August 2017.
[10] W. Riani, “Menakar Akurasi Indeks Gini untuk Mengukur Kesetaraan Pendapatan Rakyat,” Fakultas
Ekonomi dan Bisnis Universitas Islam Bandung, 2016.
[11] S. Haryadi. “Calculation of the Mobile Communication Competition using Haryadi Index,” Institut
Teknologi Bandung, Notebook, 2017.
11