Content uploaded by Carlos Mendez

Author content

All content in this area was uploaded by Carlos Mendez on Nov 20, 2020

Content may be subject to copyright.

Available via license: CC BY 4.0

Content may be subject to copyright.

Regional Income Disparities and

Convergence Clubs in Indonesia:

New District-Level Evidence

Harry Aginta ∗

Nagoya University

Anang Gunawan

Nagoya University

Carlos Mendez

Nagoya University

November 20, 2020

Abstract

This paper aims to re-examine the regional convergence hypothesis

on income in Indonesia over the 2000-2017 period. By applying a

non-linear dynamic factor model, this paper tests the club

convergence hypothesis using a novel dataset of income at the district

level. The results show signiﬁcant ﬁve convergence clubs in

Indonesian districts’ income dynamics, implying the persistence of

income disparity problems across districts even after implementing

the decentralization policy. The subsequent analysis reveals two

appealing features regarding the convergence clubs. First, districts

belonging to the same province tend to be in the same club, and

second, districts with speciﬁc characteristics (i.e., big cities or natural

resources-rich regions) dominate the highest income club. Overall, our

ﬁndings suggest some insightful policy implications, including the

importance of differentiated development policies across convergence

clubs and inter-provincial development strategies.

Keywords: Regional inequality, Convergence, Indonesia

JEL Codes: O40, O47, R10, R11

∗Corresponding author: aginta.harry@c.mbox.nagoya-u.ac.jp

2REGIONAL INCOME DISPARITIES AND CONVERGENCE CLUBS IN INDONESIA

1. Introduction

There is a growing recognition that reducing income inequality fosters

sustainable development. Speciﬁcally, lower levels of income inequality not

only prevent social conﬂict, but they are also a prerequisite for achieving

social justice. On the other aspects, some studies also ﬁnd that income

inequality negatively affects economic growth (Barro,2008;Ostry et al.,

2014;van der Weide and Milanovic,2018). Moreover, widening income

inequality gives signiﬁcant implications for economic growth as well as

macroeconomic stability (Dabla-Norris et al.,2015).

As one of the most heterogeneous countries in the world, Indonesia

consists of hundreds of ethnic groups with many different cultures and

religious beliefs spreading throughout the world’s largest archipelago. In

economic terms, the country has been experiencing high regional

inequality that persist since its independence. The inequality is shown by

the most developed region, particularly in Java and Sumatra islands with

capital intensive processing industries and the isolated regions that are

barely connected to few regions (Hill,1991). Therefore, one of the main

challenges in Indonesia’s development context is how to reduce regional

inequality and foster regional convergence (Mishra,2009).

The concern about regional inequality became greater when the

decentralization policy was implemented in 2000 (Nasution,2016;Mahi,

2016). The worry seems to be reasonable since the decentralization process

in Indonesia was implemented without much preparation, in the sense that

it was not accompanied by adequate institutional capacity or skilled

ofﬁcials at the local level (Brodjonegoro,2004;Nasution,2016). Some

studies even argue that decentralization contributes to the negative growth

of investment (Brodjonegoro,2004;Tijaja and Faisal,2014) owing to

increased policy inconsistency and business uncertainty at the regional

NEW DISTRICT-LEVEL EVIDENCE 3

level. Motivated by the limited research and an inconclusive answer about

the effect of decentralization on the regional income disparity dynamics,

this paper studies the evolution of regional income disparities and

prospects for convergence across 514 districts of Indonesia over the

2000-2017 period.

This study focuses on the period after the year 2000 because it

corresponds to the beginning of Indonesia’s decentralization era. In this

era, the state budget is allocated to regions, both to provincial and

municipal governments. However, the expected outcome from the

decentralizing policy in reducing regional inequality has yet to be seen

clearly, partly due to the diverse growth barriers and economic

preconditions in each region. Hence, identifying groups of regions facing

similar challenges is of particular relevance concerning the formulation of

policies aiming to reduce regional disparities.

In brief, the results of this paper show that Indonesian districts form ﬁve

convergence clubs, implying that the growth of income per capita in 514

districts can be clustered into ﬁve common trends. This study also ﬁnds a

”catching-up effect” within each club where the initially poor districts tend

to grow faster than the initially rich districts. Further analysis reveals two

appealing features about the convergence clubs. First, districts belonging to

the same province tend to be in the same convergence club. Second, the

highest income club is dominated by districts with speciﬁc characteristics

such as big capital cities or resources-rich regions. Furthermore, the

implementation of classical convergence tests provides supplementary

evidence about convergence speed within each club.

This paper contributes to the regional convergence literature in three

following ways. First, it is based on a convergence framework that

emphasizes the role of regional heterogeneity and the potential existence of

multiple convergence clubs. The results of club convergence analysis are

4REGIONAL INCOME DISPARITIES AND CONVERGENCE CLUBS IN INDONESIA

complemented by two classical tests of convergence, which are

implemented for each identiﬁed club. Second, the analysis is based on a

newly constructed dataset of per-capita income that covers 514 Indonesian

districts over the 2000-2017 period. This granular perspective opens the

possibility of identifying new patterns, which may remain hidden when

using province-level data. Finally, we extend our analysis by applying a

classical convergence test for all convergence clubs to inform the

convergence patterns.

The remainder of this paper is organized as follows. Section 2 discusses

the related literature about regional disparities and convergence in

Indonesia. Section 3 explains our research methodology, while section 4

presents the data and some stylized facts. The results of the formation of

convergence clubs are presented in Section 5. Section 6 discusses the

results in the context of classical convergence indicators, the spatial

distribution of the clubs, and policy issues. Finally, Section 7 closes the

paper with concluding remarks.

2. Related literature

2.1 Regional income disparities in Indonesia

A large body of research has been conducted to analyze economic disparity

among regions in Indonesia. Most of the studies argue that the large

socio-economic disparities among regions in Indonesia are due to larger

unequal economic activities, public infrastructure availability as well as

resource endowment (Esmara,1975;Akita,1988;Garcia and

Soelistianingsih,1998;Hill et al.,2008). One of the pioneers of regional

income disparities analysis in Indonesia at the provincial level was

conducted by Esmara (1975). He studies the inequality of Indonesian

NEW DISTRICT-LEVEL EVIDENCE 5

economic development during 1970’s by analyzing per capita Gross

Domestic Product (GDP) without mining sector. The study ﬁnds that

non-mining per capita income differed by a factor of 12 between the

highest and the lowest income region.

Using Williamson Index, Akita and Lukman (1995) argue that regional

inequality at the provincial level, measured by GDP per capita, had

decreased during 1975-1992. However, using non-mining GDP per capita,

the regional inequality in the same period remained relatively stagnant.

Using a more extended period, Akita et al. (2011) examine the

inter-provincial regional income disparities over the 1983-2004 period. The

study indicates a large regional gap among the main islands in Indonesia.

In addition, they also ﬁnd large disparities across districts within provinces

in the islands.

Moving to the different levels of observation, Tadjoeddin et al. (2001)

analyze the regional inequality at the district level by examining the Theil

and Gini coefﬁcients of GDP per capita from 1993 to 1998. The study shows

that regional disparities in Indonesia are stable at the district level during

the period of analysis. Similarly, the study of Hill et al. (2008) ﬁnds that

regional disparities remained relatively unchanged during 1993-1998.

However, when oil and gas GDP per capita is excluded from the analysis,

the regional inequality kept increasing slightly until 1998.

In the context of the post-Asian Financial Crisis (AFC) 1997/98 era,

Aritenang (2014) argues that the implementation of the decentralization

policy has been considered to increase regional disparities. According to

the decentralization law, districts with abundant natural resources earn a

higher share of revenue than their provincial government and their peers in

the same province. Therefore, a natural resources-rich district receives a

higher revenue share and tends to grow faster. In the end, this might

increase disparities among districts.

6REGIONAL INCOME DISPARITIES AND CONVERGENCE CLUBS IN INDONESIA

2.2 Regional income convergence

A central prediction of the standard neoclassical growth model is that

under common preferences and technologies, economies would tend to

converge to a common long-run equilibrium (Barro and Sala-I-Martin,

1992;Mankiw et al.,1992;Islam,2003;Barro and Sala-I-Martin,2004).1

There is a large literature that aims to test this prediction both across and

within countries. Compared to national economies, the administrative

regions within a country are more likely to share common preferences,

technologies, and institutions. Thus, empirically testing for income

convergence across regions within a country has become a central topic in

the regional growth literature.

The seminal contributions of Barro (1991) and Barro and Sala-I-Martin

(1992) document regional income convergence across the states in the

United States, the prefectures of Japan, and the subnational regions of

Europe. Interestingly, in all these cases, regions appear to be converging at

a similar speed: 2% per year. These convergence dynamics imply that

regional differences within each country would be halved in about 35 years

(Abreu et al.,2005). These results have triggered a large empirical literature

that aims to test the regional convergence hypothesis (Sala-I-Martin,1996a;

Magrini,2004,2009). From a methodological perspective, most papers

evaluate convergence using two complementary analyses. On the one

hand, an analysis of sigma (σ) convergence evaluates whether the income

dispersion decreases over time. On the other, an analysis of beta (β)

1Although the original conception of the Solow growth model aims to explain the

evolution of a single economy over time, its convergence prediction has been empirically

tested across multiple countries, regions, industries, and ﬁrms. As a result, the convergence

hypothesis has been studied from multiple perspectives. The recent work of Johnson and

Papageorgiou (2020) provides a survey of the cross-country convergence literature. Magrini

(2004) provides a survey of the regional convergence literature. The work of Rodrik (2013) is

one of the most inﬂuential papers in the industrial convergence literature, and the work of

Bahar (2018) evaluates convergence across ﬁrms.

NEW DISTRICT-LEVEL EVIDENCE 7

convergence evaluates whether initially poor economies grow faster than

initially rich regions. These two analyses are also related in the sense that

beta convergence is necessary but not sufﬁcient for sigma convergence

(Sala-I-Martin,1996a).

More recently, the convergence literature has been shifting its focus

from the classical emphasis on average behavior and common long-run

equilibrium to a new emphasis on heterogeneous behavior and multiple

equilibria (Apergis et al.,2010;Bartkowska and Riedl,2005;Zhang et al.,

2019). This approach emphasizes the notion that, even within countries,

there could be persistent differences in endowments, preferences, and

technologies. As such, regional economies may not smoothly converge to a

unique long-run equilibrium, but instead, multiple convergence clubs may

characterize the regional economic system.

2.3 Regional income convergence in Indonesia

Many scholars have conducted studies on inequality at the regional level in

Indonesia using various convergence frameworks (see Table 1). Garcia and

Soelistianingsih (1998) employ beta convergence method to investigate the

existence of convergence in income per capita across provinces during

1975-1993. The study shows that regional income disparities tend to

converge, and it may take between thirty to forty years to reduce income

differences by half. However, Hill et al. (2008) argue that the results of

convergence by Garcia and Soelistianingsih (1998) are sensitive to the

period analyzed and the unstable performance of the oil and gas sector.

Also, the study of Hill et al. (2008) shows that during the ﬁnancial crisis and

its aftermath, that is from 1997 to 2002, there was no signiﬁcant

convergence at the regional level. Similarly, using conventional estimation

of sigma and beta convergence, Tirtosuharto (2013) does not ﬁnd regional

8REGIONAL INCOME DISPARITIES AND CONVERGENCE CLUBS IN INDONESIA

convergence during the Asian ﬁnancial crisis, the recovery period, and the

beginning of the decentralization era.

At the district level, the study by Akita (2002) shows a similar conclusion

that regional income inequality increased during 1993-1997. This result

does not contradict other studies at the provincial level since it shows that

inequality increased among certain districts within some provinces. In

addition, Akita et al. (2011) show that when the Asian ﬁnancial crisis hit in

1997, regional inequality has declined since some big cities were hit harder

than less developed districts. However, during the recovery period, regional

inequality increased again until 2004 and remained uncertain afterwards.

Analysis at the district level has also been conducted by Aritenang (2014)

using exploratory analysis, Spatial Autoregressive (SAR) Lag Model, and

Spatial Error Model (SEM) to capture the spatial effects. By considering the

role of the neighborhood, the study ﬁnds that the convergence rate is

higher throughout the decentralization era. A similar approach was

conducted by Vidyattama (2013). However, the result of the study indicates

inconclusive ﬁnding on convergence. The Williamson index measurement

shows slight increases, although insigniﬁcant, while the beta convergence

estimates reveal convergence at both the district and the provincial levels

during 1999-2008. In addition, the study needs a longer period of

observation since the overall trend of convergence is still very weak.

Another recent study was conducted by Kurniawan et al. (2019) by

applying club convergence analysis on provincial dataset from 1969 to

2012. In their study, some missing data at the provincial level are

interpolated in order to build a balanced panel dataset. The results show

two convergence clubs in terms of all investigated variables.2Using a

similar method, Mendez and Kataoka (2020) examine the disparities in

2The study examines the dynamics of four socio-economic indicators: per capita gross

regional product, the Gini coefﬁcient, the school enrolment rate, and the fertility rate.

NEW DISTRICT-LEVEL EVIDENCE 9

Table 1: Studies on per capita income convergence in Indonesia

Author(s) Observation Methods Findings

Garcia and

Soelistianingsih

(1998)

(26 provinces)

1975-1993

1980-1993

1983-1993

Absolute and

conditional beta

convergence

Results: Absolute convergence;

conditional convergence

increases convergence speed.

Akita (2002)

(27 provinces

and

303 districts)

1975-1983

1983-1993

1993-1999

Williamson’s

weighted coefﬁcient of

variation (CVw)

and Theil index

Results: Convergence at province

and no convergence at district level.

The mining sector matters and

income inequality at the district

level increases in1993-1997.

Hill et al.

(2008)

(26 provinces)

1975-2004

Unconditional

and conditional

beta convergence

Results: Convergence before the

Asian Financial Crisis (AFC) 1997

and no convergence after the AFC.

The speed of convergence declines

along with the decreasing in the

mining sector.

Akita et al.

(2011)

(26 provinces)

1983-2004

Bi-dimensional

decomposition

Results: Convergence before

the AFC 1997 and no convergence

after the AFC. The convergence

before the AFC is due to poorer

performance of the resource-rich

provinces.

Tirtosuharto

(2013)

(26 provinces)

1997-2000

2001-2012

Sigma convergence

and unconditional

beta convergence

Results: No sigma convergence,

beta convergence in 1997-2000 and

no beta convergence in 2001-2012.

Vidyattama

(2013)

(26 provinces

and

294 districts)

1999-2008

Unconditional beta

convergence, Spatial

Autoregressive Lag,

and Spatial Error Model

Results: Insigniﬁcant convergence in

income per capita and signiﬁcant

convergence in HDI at both province

and district levels.

Aritenang

(2014)

(292 districts)

1994-2004

1994-2000

2001-2004

Spatial autocorrelation,

Spatial Error Model, and

Spatial Autoregressive

Lag Model

Results: Strong evidence of spatial

autocorrelation, the convergence rate

is higher during decentralization.

Kurniawan et al.

(2019)

(33 provinces)

1969-2012 Club convergence Results: Two convergence clubs in

four socio-economic indicators.

Mendez and

Kataoka

(2020)

(26 provinces)

1999-2010 Club convergence

Results: Two convergence clubs in

labor productivity, four clubs in

physical capital, two clubs in human

capital, and unique convergence club

in efﬁciency.

Source: Authors’ documentation from many sources.

10 REGIONAL INCOME DISPARITIES AND CONVERGENCE CLUBS IN INDONESIA

labor productivity, capital accumulation, and efﬁciency across 26 provinces

from 1990 to 2010. The study ﬁnds that labor productivity, physical capital,

and human capital are characterized by two, four, and two convergence

clubs, respectively. Meanwhile, a unique convergence club is found to be

related to the efﬁciency variable. The study suggests the importance of

capital accumulation and efﬁciency improvements in promoting

productivity growth as well as reducing the disparities among regions in

Indonesia.

3. Methodology

3.1 Classical beta and sigma convergence

The most common method of convergence analysis is mainly based on

classical models such as namely sigma convergence and beta convergence

(Bernard and Durlauf,1995;Hobijn and Franses,2000;Phillips and Sul,

2007). Sigma convergence refers to the decreasing in growth dispersion (in

most cases, the growth of income per capita) across countries or regions

over time. Differently, beta convergence is seen in negative correlation

between the initial level of income capita and its growth. Implicitly, this

means that low-income countries tend to grow relatively faster than

high-income countries and thus are able to catch up (Barro,1991;Barro

and Sala-I-Martin,1992).

The concept of beta convergence can be differentiated into absolute and

conditional convergence (Islam,1995,2003;Mankiw et al.,1992;

Sala-I-Martin,1996b). On one side, absolute beta convergence assumes

that countries will approach a particular common steady-state growth path

over time, given the variability in the initial condition of each country. On

the other side, the notion of conditional beta convergence implies

NEW DISTRICT-LEVEL EVIDENCE 11

convergence occurs towards different paths of steady-state growth given

the assumption that countries have distinctive characteristics, such as

accumulation in human and physical capital, institution, economic and

political system, and other factors affecting economic growth. Many

researchers ﬁnd that the dispersion of income per capita across economies

follows the patterns of clusters rather than the direction of a common

growth path (Quah,1996;Phillips and Sul,2009;Basile,2009). This is not

only true for largely diversiﬁed cases such as cross country analysis, but this

trend has also been observed in more integrated economies like those in

Western Europe (Corrado et al.,2005).

Some studies also started to ﬁnd convergence patterns across countries,

regions, industries, etc., when analyzing socio-economic variables (Barro,

1991;Barro and Sala-I-Martin,1992). According to Barro and Sala-I-Martin

(1992), this convergence pattern can be generalized as follows:

(1/T )·log yiT

yi0

=α−[1 −e−βT ]

T·log(yi0) + wi,0T(1)

where yis the analyzed variable, irepresents a region, 0and Tare the initial

and ﬁnal times, βis known as the speed of convergence, αincludes

unobserved parameters including the steady-state and wi,0Tis the error

term. Referring to equation (1), if there are robust signs of beta

convergence, then a different parameter known as the ”half-life” can be

deﬁned as follows:

half ·lif e =log2

β(2)

This parameter indicates the time required for the average region to reduce

the gap between its initial and the ﬁnal equilibrium state by half.

12 REGIONAL INCOME DISPARITIES AND CONVERGENCE CLUBS IN INDONESIA

3.2 Relative convergence test

Phillips and Sul (2007) develop log tconvergence test, an innovative

method to investigate the existence of multiple convergence clubs based on

a clustering algorithm. This method is favorable because of its superiority

in the sense that it allows the time series not to be co-integrated, thus

allowing individual observation to be transitionally divergent (Bartkowska

and Riedl,2005). The method also concludes that the absence of

co-integration in respective time series does not necessarily deny the

existence of convergence (Phillips and Sul,2007). Due to its advantages,

numerous researchers have utilized this approach with applications in

convergence analysis on various economic indicators such as per capita

income, ﬁnancial development, and energy.

The relative convergence test suggested by Phillips and Sul (2007,2009)

is based on the decomposition of the panel-data variable of interest in the

following way:

yit =git +ait (3)

where git is a systematic component and ait is a transitory component.

To separate common from idiosyncratic components, equation 3can be

transformed with a time-varying factor as follows:

yit =git +ait

µtµt=δitµt(4)

where δit contains error term and unit-speciﬁc component and thus

represents an idiosyncratic element that varies over time, and µtis a

common component.

To be more speciﬁc, the transition path of an observed economy

towards its own equilibrium growth path is explained by δit, while µtdepicts

a hypothesized equilibrium growth path that is common to all economies.

NEW DISTRICT-LEVEL EVIDENCE 13

Equation 4is therefore a dynamic factor model containing a factor loading

coefﬁcient δit that represents the idiosyncratic distance between a common

trending behavior, µt, and the dependent variable, yit. Furthermore, to

characterize the dynamics of the idiosyncratic component, δit,Phillips and

Sul (2007) propose the following semi-parametric speciﬁcation:

δit =δi+σiξit

log (t)tα(5)

where δirepresents the heterogeneity of each economy but constant over

time, ξit is a weakly time-dependent process with mean 0 and variance 1

across economies. Under the condition given in equation 5, convergence

occurs when all economies move to the same transition path as such,

lim

t→∞

δit =δand α≥0(6)

In order to estimate the transition coefﬁcient δit,Phillips and Sul (2007)

construct a relative transition parameter, hit, as

hit =yit

1

NPN

i=1 yit

=δit

1

NPN

i=1 δit

(7)

where the common component, µtin equation 4is eliminated by dividing

the independent variable, yit, with the panel average. Thus, hit represents

the transition path of economy iagainst the level of cross-sectional average,

implying the calculation of individual economic behaviors relative to other

economies. Then, hit converges to unity, that is hit →1, when t→ ∞.

Later, the notion of convergence can be transformed into the following

equation that describes the cross-sectional variance of hit,

Ht=1

N

N

X

i=1

(hit −1)2→0(8)

14 REGIONAL INCOME DISPARITIES AND CONVERGENCE CLUBS IN INDONESIA

where the cross-sectional variance converges to zero, Ht→0.

The null hypothesis in equation 6is veriﬁed in counter to the alternative

hypothesis HA:δi6=δfor all ior α≥0. Finally, Phillips and Sul (2007)

empirically evaluate this null hypothesis by using the following log t

regression model:

log H1

Ht−2 log{log(t)}=a+blog(t) + εt

for t= [rT ],[rT ]+1, . . . , T with r > 0

(9)

where rT is the initial observation in the regression, which implies that the

ﬁrst fraction of the data (that is, r) is discarded.

Based on Monte Carlo experiments, Phillips and Sul (2007) suggest

applying r= 0.3when the sample is small or moderate T≤50. A fairly

conventional inferential procedure is also suggested for equation 9. To be

more speciﬁc, a one-sided ttest with heteroskedasticity-autocorrelation

consistent (HAC) standard errors is used. In this setting, the null hypothesis

of convergence is rejected when the t-statistics (tˆ

b) is smaller than -1.65.

3.3 Clustering algorithm

Even though the null hypothesis of overall convergence in the full sample is

rejected, it does not necessarily mean that the convergence in the

subsample of the panel is not present. Therefore, following Phillips and Sul

(2007), we exploit the feature of the model in equation 7to reveal the

presence of multiple convergence clubs in subsample. For that purpose, we

use an innovative data-driven algorithm developed by Phillips and Sul

(2009), which can be summarized in the following four steps:

1. Ordering: Sample units (districts) are arranged in a decreasing order

according to their observation in the last period. In this paper, the

NEW DISTRICT-LEVEL EVIDENCE 15

ordering is conducted using the average of the last 1

3.

2. Constructing the core group: A core group of sample units (districts) is

identiﬁed based on the ﬁrst kunit of the panel data set (2 ≤k≤N).

If the tˆ

bof the kunit is larger than -1.65, the core group formation is

established. If the tˆ

bin the ﬁrst kunit is smaller than -1.65, the ﬁrst

unit is dropped, and then the log ttest for the next units is conducted.

The step is continued until the tˆ

bof the pair units is larger than -1.65.

If there are no pairs of units showing tˆ

blarger than -1.65 in the entire

sample, it can be concluded that there are no convergence clubs in the

panel.

3. Deciding club membership: Sample units (districts) not belonging to

the core group are re-evaluated once at a time with log tregression.

A new group is formed when the tˆ

bis larger than -1.65. Otherwise, if

the additional units give a result that tˆ

bis smaller than -1.65, then the

convergence club only consists of the core group.

4. Iteration and stopping rule: The log tregression is applied for the

remaining sample units (districts). If the process shows the rejection

of the null hypothesis of convergence, steps 1 to 3 are performed

again. The remaining sample units (districts) are labeled as divergent

if no core group is found, and the algorithm stops.

The representation of the relative transition curve for different

economies in the club convergence framework can be illustrated in Figure

1. The ﬁgure clearly shows that the transition curves for different regions

form a funnel. The four regions, which are region A, B, C, and D, differ in

their initial conditions as well as in their transition paths. However, region A

and region B’s relative transition curve converge into the same value, which

is Club 1. In comparison, region C and region D are characterized by

16 REGIONAL INCOME DISPARITIES AND CONVERGENCE CLUBS IN INDONESIA

time

1 t

Club 1h1

h2

Club 2

Region D

Region C

Region A

Region B Cross-sectional average

Figure 1: An illustration of transition paths and convergence clubs

medium and low initial conditions, consecutively reﬂecting a typical

developing region with a slow growth rate and a poor region that grows

rapidly. With time, the transition path of both regions converge into Club 2.

4. Data and stylized facts

4.1 Data construction

This study uses annual GDP per capita at the district level from 2000 to

2017. However, not all of the data are available for every year in each

district. In addition, there are some missing observations caused by the

splitting up of new districts during the decentralization period. Since the

club convergence test of Phillips and Sul (2007) requires balanced panel

data, we constructed a balanced panel dataset of 514 districts by solving the

NEW DISTRICT-LEVEL EVIDENCE 17

missing observations through interpolation/imputation.3Similar with the

study of Kurniawan et al. (2019), the imputation process in our study was

conducted using a linear regression method with the year and reference

districts as candidates of regressors. Hence, since we only predicted the

missing values from its trend, it would not signiﬁcantly alter the

convergence results.4Table 2shows the descriptive statistics of the dataset.

Table 2: Descriptive statistics

Mean Standard deviation Max/Min

2000 2017 2000/2017 2000 2017 2000/2017 2000 2017

GDP per capita (in thousand IDR) 25,032 36,041 0.69 62,172 42,823 1.45 465.40 105.05

Log of GDP per capita 9.57 10.20 0.94 0.81 0.67 1.20 1.81 1.56

Trend log of GDP per capita 9.57 10.20 0.94 0.81 0.68 1.20 1.81 1.56

Relative trend of log GDP per capita 1.00 1.00 1.00 0.08 0.07 1.28 1.81 1.55

As mentioned in the methodology, we transformed the GDP per capita

data into a log form. Like common macroeconomic data, the GDP per

capita has two prominent features, which are long-run growth trends in

aggregate and a cyclical component that represents ﬂuctuations in the

shorter periods, known as the business cycles. When analyzing business

cycles in observed data by regression or ﬁltering, it is necessary to isolate

the cyclical component from the trend (Phillips and Shi,2019). Therefore,

following Uhlig and Ravn (2002), we ﬁltered the GDP per capita series using

Hodrick-Prescott (HP) ﬁlter technique with smoothing parameter (λ)

equals to 6.25.

3We combined actual data of the new districts and the reference districts and compared

them to ensure that measurement error caused by the interpolation is minimum. Details of

this interpolation process are provided in Appendix A.

4For the robustness check of our interpolation results, we implemented sigma and

beta convergence tests using the number of districts in the year 2000 (342 districts). As

reported in Appendix E, we found no signiﬁcant difference in sigma and beta convergence

coefﬁcients between the full sample of 514 districts and the smaller number of districts.

18 REGIONAL INCOME DISPARITIES AND CONVERGENCE CLUBS IN INDONESIA

4.2 Stylized facts on regional disparities in Indonesia

This subsection illustrates how income disparities across districts in

Indonesia have evolved over time. Figure 2measures regional disparities as

the standard deviation of the log of GDP per capita. This measurement

approach is commonly used in the regional growth literature and is

generally referred to as the study of sigma convergence (Barro and

Sala-I-Martin,1992;Magrini,2004;Sala-I-Martin,1996b). A process of

sigma convergence occurs when regional disparities decrease over time.

Figure 2highlights this process by pointing out that the standard deviation

of the log of GDP per capita has been systematically decreasing over the

2000-2017 period.

.65

.7

.75

.8

SD of Log GDP per capita

2000 2005 2010 2015 2020

Figure 2: Evolution of regional disparities: Sigma convergence approach

Notes: GDP refers to the district-level gross domestic product, and it is measured based

on constant prices of 2010. The source of the data is the Central Bureau of Statistics of

Indonesia and interpolation results.

We also show the process where the initially poor regions are catching

up and growing faster than initially rich regions. This catching-up process is

largely documented in the economic growth literature and is referred to as

NEW DISTRICT-LEVEL EVIDENCE 19

the study of beta convergence (Barro and Sala-I-Martin,1992;Magrini,2004;

Sala-I-Martin,1996b). Figure 3highlights this process by pointing out that

regions with a low GDP per capita in 2000 have grown faster than initially

rich regions over the 2000-2017 period. Interestingly, the richest regions in

1990 experienced large negative growth rates in subsequent years. Thus, the

(beta) convergence process is arising not only because of the faster growth of

the poorest regions but also because of a systematic reduction in the income

of the richest regions.

-2

-1

0

1

2

Growth Rate 2000-2017

810 12 14

Log GDP per capita in 2000

Figure 3: Evolution of regional disparities: Beta convergence approach

Notes: GDP refers to the district-level gross domestic product, and it is measured based

on constant prices of 2010. The source of the data is the Central Bureau of Statistics of

Indonesia and interpolation results.

By construction, the summary statistics of sigma and beta convergence

only describe the behavior of an average or representative economy

(Magrini,2004). They fail to describe more complex convergence dynamics

that could occur beyond the mean of the income distribution. They also fail

to accommodate the notions of multiple equilibria and convergence clubs,

which could arise when the regions’ performance is highly heterogeneous.

20 REGIONAL INCOME DISPARITIES AND CONVERGENCE CLUBS IN INDONESIA

In Indonesia’s context, a high degree of regional heterogeneity has been

previously documented using province-level data. Moreover, it has been

argued that only focusing on average patterns is likely to be incomplete at

best or misleading at worst (Mendez and Kataoka,2020). In an attempt to

start documenting the degree of regional heterogeneity using district-level

data of recent years, Figures 4and 5show the evolution of regional

disparities beyond the scope of the average or median district.

Figure 4shows how the quantiles of the distribution have evolved over

time. Panel (a) indicates that when we evaluate the regional dynamics of

GDP per capita, within any logarithm transformation, regional disparities

have been increasing over time. Increasing disparities are evident not only

when we measure the gap between the quantile 95 and quantile 5, but also

when we measure the gap between the quantile 75 and 25. In the statistics

literature, this latter gap is referred to as the interquartile range (IQR) and is

commonly used as a dispersion statistic that is robust to extreme values.5

Panels (b) and (c) indicate that the logarithm version of GDP per capita

shows less diverging dynamics. Despite of this transformation, the IQR,

which encompasses half of the entire distribution, shows very little signs of

regional convergence. Panel (d) normalizes the trend of log GDP by the

cross-sectional mean of each year. This transformation helps to remove the

common increasing trends observed in panels (b) and (c). Based on this

transformation, regional disparities have been evolving differently within

the income distribution. Most of the reduction in the disparities arises from

the tails of this distribution, particularly the upper tail. In contrast,

disparities around the center of the distribution show little change. The size

of the IQR has been almost constant over the entire 2000-2017 period.

Figure 5provides a more detailed perspective on the dynamics reported

5In the convergence literature, the IQR is also used to study sigma convergence (Mendez-

Guerra,2018).

NEW DISTRICT-LEVEL EVIDENCE 21

25000

50000

75000

2000 2005 2010 2015

GDP per capita

Quantile

q95

q75

q50

q25

q05

(a) GDP per capita

8.5

9.0

9.5

10.0

10.5

11.0

11.5

2000 2005 2010 2015

Log of GDP per capita

Quantile

q95

q75

q50

q25

q05

(b) Log GDP per capita

8.5

9.0

9.5

10.0

10.5

11.0

11.5

2000 2005 2010 2015

Trend Log of GDP per capita

Quantile

q95

q75

q50

q25

q05

(c) Trend log GDP per capita

0.90

0.95

1.00

1.05

1.10

1.15

2000 2005 2010 2015

Relative Trend Log of GDP per capita

Quantile

q95

q75

q50

q25

q05

(d) Relative trend log GDP per capita

Figure 4: Evolution of regional disparities: Distributional quantile approach

Notes: GDP refers to the district-level gross domestic product, and it is measured based

on constant prices of 2010. The source of the data is the Central Bureau of Statistics of

Indonesia and interpolation results.

22 REGIONAL INCOME DISPARITIES AND CONVERGENCE CLUBS IN INDONESIA

.8

1

1.2

1.4

Relative Trend Log GDP per capita

2000 2005 2010 2015 2020

Figure 5: Evolution of regional disparities: Relative convergence approach

Notes: GDP refers to the district-level gross domestic product, and it is measured based

on constant prices of 2010. The source of the data is the Central Bureau of Statistics of

Indonesia and interpolation results.

NEW DISTRICT-LEVEL EVIDENCE 23

in Figure 4d. Instead of just displaying ﬁve hypothetical regions, Figure 5

shows the actual dynamics of all regions.6Its main ﬁnding is consistent

with that of Figure 4; that is, the process of regional convergence is not

homogeneous across the distribution of districts. Regions at the top of the

distribution are converging faster than those at the bottom. Regions around

the middle of the distribution show little progress in the reduction of

regional disparities.

5. Results

5.1 Relative convergence test

After applying the log ttest to the income per capita data across 514

Indonesian districts over the 2000-2017 period, we were able to reject the

null hypothesis of overall convergence at the 5% signiﬁcant level, where ˆ

bis

signiﬁcantly <0 and tˆ

bis −22.28 (see Table 3). This implies that

convergence for all districts is not present, indicating that the income

growth process of 514 Indonesian districts from 2000 to 2017 does not show

a single equilibrium steady-state. This ﬁnding is consistent with the

empirical evidence documented in the study of Hill et al. (2008), where no

signiﬁcant convergence was observed after the 1997/98 Asian ﬁnancial

crisis period, as opposed to signiﬁcant convergence before the crisis.7Our

ﬁnding also supports the study of Tirtosuharto (2013), who concludes the

lack of regional convergence for the period from 2003 to 2012 following

economic recovery from the Asian ﬁnancial crisis and the beginning of the

6As regions can change their ranking in the income distribution, the quantiles of Figure

4do not necessarily track the performance of a unique region over time.

7The study of Hill et al. (2008) shows the variability in the pace of beta convergence

across subperiod during 1975-2002; 2% during the oil boom (1975-81), 2.8% in the era of

major policy reforms (1981-86), 1.7% for the period 1986-92 as the export-oriented reforms

took place, 1% during the 1990s, and no convergence in the crisis and post-crisis period.

24 REGIONAL INCOME DISPARITIES AND CONVERGENCE CLUBS IN INDONESIA

decentralization era.

Table 3: Global convergence test

Coefﬁcient Standard Error t-statistic

log(t) -0.52 0.02 -22.28

Note: The null hypothesis of convergence is rejected when

the t-statistic is less than -1.65.

This lack of overall convergence also entails policy implications related

to the implementation of the decentralization policy in Indonesia after the

crisis. As pointed out by Azis (2008) and Nasution (2016), by and large, the

performance of regional growth after the decentralization has been

unsatisfying. One of the major causes of this unexpected outcome is the

variability in the capacity of regional institutions and local leaders to

leverage local resources, in addition to counterproductive policies issued by

local governments (e.g., imposing hidden fees, allocating funds to

unnecessary projects), and inﬂexible and over-regulated national policies.

Therefore, to avoid potential detrimental effect of decentralization on

regional disparities, in particular developing countries like Indonesia, the

regional government is required to improve the quality of institutional

factors such as accountability, people’s empowerment as well as

redistribution capacity (Azis,2008;Rodriguez-Pose and Ezcurra,2010).

5.2 Clustering algorithm and convergence clubs

Although overall convergence across Indonesian districts does not prevail,

the log ttest brings the possibility to observe the existence of several

convergence clusters, as explained in Section 3.3. Therefore, we applied the

NEW DISTRICT-LEVEL EVIDENCE 25

test procedure to investigate convergence clubs. As shown in Table 4, we

found ﬁve signiﬁcant initial clubs.8The ﬁrst convergence club consists of 6

districts; the second club consists of 126 districts; the third club consists of

178 districts; the fourth club contains 181 districts, and the ﬁfth club

consists of 23 districts. The rows correspond to the ﬁtted coefﬁcients and

t-statistic in each club.

The order of the convergence clubs is sorted from the districts with the

highest to the lowest GDP per capita, that is, Club 1 refers to the highest

GDP per capita group and Club 5 displays the lowest GDP per capita group.

The result of this club convergence test implies that the development of

income per capita in 514 Indonesian districts can be grouped into ﬁve

common trends during 2000-2017.

Table 4: Local convergence test

Club1 Club2 Club3 Club4 Club5

Coefﬁcient 0.42 -0.08 0.37 -0.04 0.49

t-statistic 4.97 -1.52 5.26 -1.60 6.55

N. of regions 6 126 178 181 23

Note: The null hypothesis of convergence is rejected when the t-

statistic is less than -1.65.

Then, following Phillips and Sul (2009), we checked the possibility of

whether any of those identiﬁed clubs can be merged to form larger

convergence clubs.9As shown in Table 5, the club merging test result

suggests that the convergence hypothesis is rejected (ˆ

bis signiﬁcantly <0

and tˆ

bis smaller than -1.65). Hence, the initial ﬁve clubs are conﬁrmed as

8See Appendix B for complete members of each club.

9The clubs merging steps are outlined in Appendix C.

26 REGIONAL INCOME DISPARITIES AND CONVERGENCE CLUBS IN INDONESIA

the ﬁnal convergence clubs.

Table 5: Clubs merging test

Club1+2 Club2+3 Club3+4 Club4+5

Coefﬁcient -0.14 -0.27 -0.30 -0.20

t-statistic -3.08 -6.99 -15.44 -9.52

Note: The null hypothesis of convergence is rejected when the t-

statistic is less than -1.65.

Furthermore, measuring the gap between clubs is also useful to

understand the income disparities among convergence clubs. For this

purpose, we show the mean per capita income of each club in the second

column of Table 6. The statistics suggest that the gap of income per capita

between clubs is arguably large, particularly between Club 1 and Club 2,

where the average income per capita of districts in Club 1 is IDR 231

million, about four times larger than that in Club 2. This implies that Club 2

has very little progress in catching up with Club 1. Table 6also reﬂects

severe income inequality problems among districts in Indonesia, where the

average income per capita in the last club is only about 3% of the that in the

ﬁrst club.

NEW DISTRICT-LEVEL EVIDENCE 27

Table 6: Characteristics of the clubs 2000-2017

Mean Std. Deviation Min Max

Club 1 231,289 196,580 7,058 932,664

Club 2 56,961 58,557 7,718 658,303

Club 3 20,090 12,178 3,402 304,400

Club 4 13,469 7.952 4,005 194,717

Club 5 7,549 5.959 2,004 59,292

Note: The income per capita data is in a thousand IDR.

Figure 6shows the transition paths of members in each club by

comparing income per capita of each district (in log form) relative to clubs’

average. All ﬁve clubs exhibit different convergence behaviors and

transition paths within the club, depending on each district’s initial

conditions and development process. We also capture one asymmetric

transition pattern within the club. On the one hand, some districts with a

higher level of income at the initial period experience a sufﬁciently large

income reduction at the ﬁnal period and move downwards to the club’s

average level. Most of these districts are those relying on natural resources

(e.g., mining and natural gas processing) such as Mimika (Club 1), Bontang

(Club 2), Lhokseumawe (Club 3), Aceh Utara (Club 4), and Aceh Timur

(Club 5). The last three districts also suffer from prolonged security issues

that led to the Martial Law enactment in 2003, followed by the Tsunami

disaster in 2004. On the other hand, none of the districts with a lower

income level at the initial period record signiﬁcant improvement. This

asymmetrical pattern implies that the convergence process within clubs

(particularly Club 4 and 5) is inﬂuenced by the depleting income in

28 REGIONAL INCOME DISPARITIES AND CONVERGENCE CLUBS IN INDONESIA

wealthier districts.

Similar to Figure 4d, in Figure 7we plot the transition paths of clubs over

time. However, instead of using the absolute income per capita (in log

form) on Y axes, in Figure 7we compare the transition of clubs relative to

the cross-sectional average of all clubs. The parallel pattern of the clubs’

transition path indicates that the clubs do not converge over time. Even

though Club 3 appears to slightly close its gap to Club 2, the transition path

of the other clubs reﬂects prolonged and stable dispersion between clubs,

where Club 4 and 5 are systematically below the average, while Club 1 and 2

are consistently above the average.

NEW DISTRICT-LEVEL EVIDENCE 29

.8

1

1.2

1.4

Relative Trend

Log GDP per capita

2000 2005 2010 2015 2020

Club 1

.8

1

1.2

1.4

Relative Trend

Log GDP per capita

2000 2005 2010 2015 2020

Club 2

.8

1

1.2

1.4

Relative Trend

Log GDP per capita

2000 2005 2010 2015 2020

Club 3

.8

1

1.2

1.4

Relative Trend

Log GDP per capita

2000 2005 2010 2015 2020

Club 4

.8

1

1.2

1.4

Relative Trend

Log GDP per capita

2000 2005 2010 2015 2020

Club 5

Figure 6: Convergence clubs and transition paths

Notes: GDP refers to the district-level gross domestic product, and it is measured based

on constant prices of 2010. The source of the data is the Central Bureau of Statistics of

Indonesia and interpolation results.

30 REGIONAL INCOME DISPARITIES AND CONVERGENCE CLUBS IN INDONESIA

.9

1

1.1

1.2

Relative Trend Log GDP per capita

2000 2005 2010 2015 2020

Club = 1

Club = 2

Club = 3

Club = 4

Club = 5

Figure 7: Convergence clubs trends

Notes: GDP refers to the district-level gross domestic product, and it is measured based

on constant prices of 2010. The source of the data is the Central Bureau of Statistics of

Indonesia and interpolation results.

5.3 Sensitivity to the trend estimation

We also implemented the log ttest by using the smoothing parameter (λ) of

Hodrick-Prescott (HP) ﬁlter equals to 400, which is used in the study of

Phillips and Sul (2007). Similar to the results discussed in the previous

section, the existence of global convergence is rejected at the 5% signiﬁcant

level (ˆ

bis signiﬁcantly <0 and tˆ

bis −23.02). Then we proceeded with the

club convergence test by following the same procedures discussed in

Section 3.3. We found twelve convergence clubs initially. Next, we applied

the merging test procedure to investigate whether any of those initial

subgroups can be merged to form convergence clubs with a larger number

of members.

As a result, we also found ﬁve ﬁnal signiﬁcant convergence clubs. The

NEW DISTRICT-LEVEL EVIDENCE 31

Table 7: Characteristics of the clubs, 2000-2017 (Trend parameter 400)

Mean Std. Deviation Min Max

Club 1 198,036 166,863 7,058 932,664

Club 2 50.595 39,595 3,531 417,149

Club 3 20,534 12,502 2,102 304,400

Club 4 11,825 3,505 4,292 59,292

Club 5 5,809 1,746 2,004 10,376

Note: The income per capita data is in a thousand IDR.

ﬁrst and second initial clubs merge into ﬁrst convergence club with 14

members while the third initial club becomes the second club with 106

members. Next, the third new club is formed by the fourth, ﬁfth, and sixth

initial clubs with 240 members and becomes the largest ﬁnal club (47% of

the total number of districts). Next, the fourth ﬁnal club is constructed by

clubs 7, 8, 9, 10, and 11 of the initial clubs with 132 members. Finally, the

twelfth (the last) initial club stays unmerged with 22 members (details of

results are presented in Appendix D). Consistent with the previous analysis,

the statistics shown in Table 7also imply a huge income gap among clubs.

Referring to the average of per capita income in all clubs, one may quickly

capture that the biggest income gap lies between Club 1 and Club 2, while

the income gap between districts in lower clubs (Club 4 and Club 5) is

much smaller.

32 REGIONAL INCOME DISPARITIES AND CONVERGENCE CLUBS IN INDONESIA

6. Discussion

6.1 Convergence within clubs

This section evaluates the convergence patterns within each club using the

classical frameworks of sigma and beta convergence. Figure 8shows the

evolution of the standard deviation of the log of GDP per capita for each

club. All sub-ﬁgures share the same axes in order to facilitate comparability

between clubs. Consistent with Figure 6, there is a stronger process of

convergence within clubs than between clubs. In particular, the districts of

Club 1 (see Appendix B for a detailed list) show the largest reduction in

regional disparities. Although the other clubs start from lower levels of

disparities, they show relatively less progress over time.

Figure 9shows the negative relationship between the initial level of

income and its subsequent growth rate. Within each club, initially poor

regions are growing faster than initially rich ones. Thus, a process of beta

convergence is also taking place within each club. Compared to the global

convergence process suggested by Figure 3, the slope of each convergence

club is steeper. This difference suggests that the (local) speed of

convergence within each club is faster than that of the global process.

Table 8provides further details about beta convergence within each

club. Again, relative to the global ﬁt (Figure 3), the ﬁt of the local models is

higher. The R-squared ranges from 0.64 in Club 5 to 0.86 in Club 1. The

districts belong to Club 3 converge at the highest speed (5.3% per year).

Thus, it is expected that disparities within Club 3 would be reduced by half

in just under 13 years. This fast local convergence contrasts with the global

model, which predicts that disparities would be halved in 42 years.

NEW DISTRICT-LEVEL EVIDENCE 33

0

.5

1

1.5

2

SD Log GDP pc

2000 2005 2010 2015 2020

Club 1

0

.5

1

1.5

2

SD Log GDP pc

2000 2005 2010 2015 2020

Club 2

0

.5

1

1.5

2

SD Log GDP pc

2000 2005 2010 2015 2020

Club 3

0

.5

1

1.5

2

SD Log GDP pc

2000 2005 2010 2015 2020

Club 4

0

.5

1

1.5

2

SD Log GDP pc

2000 2005 2010 2015 2020

Club 5

Figure 8: Evolution of disparities within clubs: Sigma convergence approach

Notes: GDP refers to the district-level gross domestic product, and it is measured based

on constant prices of 2010. The source of the data is the Central Bureau of Statistics of

Indonesia and interpolation results.

34 REGIONAL INCOME DISPARITIES AND CONVERGENCE CLUBS IN INDONESIA

-2

-1

0

1

2

Growth 2000-2017

810 12 14

Log GDP pc in 2000

Club 1

-2

-1

0

1

2

Growth 2000-2017

810 12 14

Log GDP pc in 2000

Club 2

-2

-1

0

1

2

Growth 2000-2017

810 12 14

Log GDP pc in 2000

Club 3

-2

-1

0

1

2

Growth 2000-2017

810 12 14

Log GDP pc in 2000

Club 4

-2

-1

0

1

2

Growth 2000-2017

810 12 14

Log GDP pc in 2000

Club 5

Figure 9: Evolution of disparities within clubs: Beta convergence approach

Notes: GDP refers to the district-level gross domestic product, and it is measured based

on constant prices of 2010. The source of the data is the Central Bureau of Statistics of

Indonesia and interpolation results.

NEW DISTRICT-LEVEL EVIDENCE 35

Table 8: Evolution of disparities within clubs: Beta convergence approach

Beta Convergence Half-life R-square

coefﬁcient speed in years

Club 1 -0,54∗∗∗ 0,046 15,09 0,86

Club 2 -0,43∗∗∗ 0,032 21,14 0,74

Club 3 -0,59∗∗∗ 0,053 12,98 0,83

Club 4 -0,51∗∗∗ 0,042 16,36 0,69

Club 5 -0,52∗∗∗ 0,043 15,05 0,64

6.2 Geographical distribution of the convergence clubs

Now, we provide a geographical view of club membership as seen in Figure

10. A few regularities are visible from the map. First, the province effect is

notably obvious; districts belonging to the same province tend to be in the

same club (Barro,1991;Quah,1996). This pattern applies almost to all

clubs. For example, districts in provinces of East Kalimantan and Riau tend

to be grouped in Club 2. Similarly, Aceh, West Sumatra, West Kalimantan,

and Central Kalimantan also show comparable pattern where most of the

districts in these provinces are clustered in Club 3, and mostly the districts

in Maluku and Nusa Tenggara provinces dominate Club 4 and Club 5. More

surprisingly, districts belonging to the same club also tend to be

geographically close. To put it another way, the clubs seem to be spatially

concentrated. This could indicate some spatial agglomeration effects

(Martin and Ottaviano,2001) driven by factors like spatial externalities or

spillovers (Quah,1996).

Second, the distribution of clubs is also related to the spatial

distribution, implying the prolonged existence of classical regional

36 REGIONAL INCOME DISPARITIES AND CONVERGENCE CLUBS IN INDONESIA

Figure 10: Spatial distribution of the convergence clubs

development problem in Indonesia; that is, the eastern regions of the

archipelago are still lagged in development. It can be seen from the

membership of the ﬁfth club where out of 23 members, 21 districts are

located in the eastern provinces of Indonesia, i.e., South Sulawesi, Nusa

Tenggara, and Papua.

6.3 Policy implications

The difference in the progress of inter-regional development is natural. It is

related to the variation in potential that each region has, both natural

resources and geographical location. In addition, variation in the regional

ability to manage their resources and potential are also factors that

differentiate the success rate of development in each region. Despite the

Indonesian economy’s ability to maintain robust economic growth after the

Asian ﬁnancial crisis in 1997/98, the persistent income gap between regions

still becomes one of major problems that could potentially be a source of a

worse complication in the future. Not only could trigger social dispute

NEW DISTRICT-LEVEL EVIDENCE 37

stemmed from the perception of injustice among fellow communities,

regional income inequality could also pose downside risks to the national

economic growth.

To reduce regional income inequality, the Indonesian government needs

to have a clear and accurate picture of regional imbalances among regions.

In this context, the results of this study suggest that the growth path of

income per capita among 514 Indonesian districts during the period of

2000-2017 does not converge to the same steady-state level. Similar to

Kurniawan et al. (2019), this ﬁnding implies the absence of global

convergence of income per capita among Indonesian regions. Instead, the

growth process of Indonesian districts constitutes ﬁve local convergence

clubs.

Interestingly, there is distinct characteristic across clubs, in particular

between the highest income club (Club 1) and the lowest one (Club 5). At

one end, Club 1 is dominated by regions with typical characteristics, i.e., big

cities or natural resources-rich regions like Central Jakarta (the central

district of the nation’s capital city), Kediri (the largest national tobacco

producer), Morowali (the location of recently developing nickel-based

industrial park), Membramo Raya, Mimika and Teluk Bintuni (the natural

resources-rich districts in the coastal area of Papua island, respectively).

While at the other end, Club 5 predominantly consists of districts that have

long been struggling with poverty issues. In addition, the income gap

among these ﬁve clubs is also considerably large, suggesting that the

potential regional development policies might be different across clubs. For

example, the development policies for districts in Club 1 might be directed

to seeking new sources of growth to avoid income stagnation. Meanwhile,

the majority of districts in Club 2, 3, and 4 could focus their program on

developing the middle-sized cities and more programs on improving

connectivity. Differently, policies on basic infrastructures and public

38 REGIONAL INCOME DISPARITIES AND CONVERGENCE CLUBS IN INDONESIA

services provision should be implemented in districts of Club 5.

Furthermore, the spatial distribution of clubs can provide non-trivial

information for inter-provincial policymaking to reduce income inequality.

For example, Figure 10 shows that some districts in Club 5 share the border

with districts in Club 1, implying the potential to further strengthen positive

spillover from the rich districts to their poor neighboring districts. In Papua

province, for instance, Pegunungan Arfak (Club 5) is the direct neighbor of

Teluk Bintuni (Club 1); Deiyai, Puncak, and Nduga (Club 5) share the border

with Mimika (Club 1); Puncak Jaya and Tolikara (Club 5) are the neighbors

of Membramo Raya (Club 1). Among others, inter-provincial policies such

as strengthening connectivity and promoting trade between these regions

are highly favorable. The same fashion can also be applied in some poor

districts of Club 5 such as Aceh Timur in Aceh province, Blora in Central

Java province, and Jeneponto in South Sulawesi province.

Lastly, given the persistence of the west-east development gap observed

in this study, the central government policies to support the development

of physical infrastructures and basic public services provision in the eastern

parts of the archipelago are strongly suggested to reduce regional income

inequality.

7. Concluding remarks

A development process is sustainable when it favors social inclusion and

the reduction of regional disparities. In this context, our study documents

the evolution of disparities in income per capita among Indonesian

districts after the implementation of decentralization policy. We use a novel

district-level dataset that covers 514 Indonesian districts over the 2000-2017

period. From a methodological perspective, the convergence club test

proposed by Phillips and Sul (2007) is applied to evaluate whether all

NEW DISTRICT-LEVEL EVIDENCE 39

districts converge to a common steady-state growth path.

The main ﬁndings are as follows. First, there is no overall convergence in

income per capita among Indonesian districts after decentralizing. Instead,

we ﬁnd ﬁve convergence clubs that describe the evolution of income

disparities across Indonesian districts. Consistent with previous literature,

our results imply that income disparity across Indonesian districts remains

a major problem even after implementing decentralization policies in the

early 2000’s. Second, we observe large and persistent differences between

clubs, where the catching-up effects seem to exist only within clubs, but not

between them. This pattern calls for differentiated development policies

based on the composition of the clubs. Third, although districts belonging

to the same province tend to converge to the same club, there is evidence

that some provinces are composed of largely different districts, which

belong to different clubs. Finally, the spatial distribution of convergence

clubs clearly shows persistence in the east-west regional divide. In this

context, the central government should coordinate regional policies to

support the development in the eastern parts of Indonesia.

Based on the data construction and research methods of this paper,

there are at least three avenues for further research. First, the comparability

of regional data is crucial in the context of Indonesia, where the number of

districts has largely increased over time. In this paper, we use a simple

time-series interpolation method to construct a balanced panel dataset. As

there is no unique and optimal interpolation method, further studies could

apply other methods to re-evaluate the convergence clubs’ composition.

Second, alternative convergence analyses can be used to evaluate the

composition and dynamics of the convergence clubs. In particular,

distributional convergence methods could complement the club

convergence approach used in this paper. Finally, recent studies about

regional convergence emphasize the role of spatial spillovers in

40 REGIONAL INCOME DISPARITIES AND CONVERGENCE CLUBS IN INDONESIA

accelerating the convergence speed. Hence, formally integrating spatial

spillovers into a club convergence framework is a promising direction for

further research.

References

Abreu, M., De Groot, H. L. F., and Florax, R. J. G. M. (2005). A meta-analysis of Beta-

convergence: The legendary 2%. Journal of Economic Surveys, 19(3):389–420.

Akita, T. (1988). Regional Development and Income Disparities in Indonesia. Asian

Economic Journal, 2(2):165–191.

Akita, T. (2002). Income inequality in Indonesia. International University of Japan,

Economic and Management Series, 2002-04-01.

Akita, T., Kurniawan, P. A., and Miyata, S. (2011). Structural Changes and Regional

Income Inequality in Indonesia: A Bidimensional Decomposition Analysis. Asian

Economic Journal, 25(1):55–77.

Akita, T. and Lukman, R. A. (1995). Interregional Inequalities in Indonesia: A

Sectoral Decomposition Analysis for 197592. Bulletin of Indonesian Economic

Studies, 31(2):61–81.

Apergis, N., Panopoulou, E., and Tsoumas, C. (2010). Old wine in a new bottle:

Growth convergence dynamics in the EU. Atlantic Economic Journal, 38(2):169–

181.

Aritenang, A. F. (2014). The spatial effect of ﬁscal decentralisation on regional

disparities: the case from Indonesia. Indonesian Journal of Geography, 46(1):1–

11.

Azis, I. J. (2008). Institutional constraints and multiple equilibria in

decentralization. In Review of Urban and Regional Development Studies:

NEW DISTRICT-LEVEL EVIDENCE 41

Journal of the Applied Regional Science Conference, volume 20, pages 22–33.

Wiley Online Library.

Bahar, D. (2018). The middle productivity trap: Dynamics of productivity

dispersion. Economics Letters, 167:60–66.

Barro, R. and Sala-I-Martin, X. (1992). Convergence. Journal of Political Economy,

100(2):223–251.

Barro, R. and Sala-I-Martin, X. (2004). Economic Growth. MIT Press, Cambridge,

Mass.

Barro, R. J. (1991). Economic growth in a cross section of countries. Quarterly

Journal of Economics, 106(2):407–443.

Barro, R. J. (2008). Inequality and Growth Revisited. Working Paper Series on

Regional Economic Integration, 11.

Bartkowska, M. and Riedl, A. (2005). Regional convergence clubs in Europe:

Identiﬁcation and conditioning factors. Economic Modelling, 29(1):22–31.

Basile, R. (2009). Productivity polarization across regions in Europe: The role of

nonlinearities and spatial dependence. International Regional Science Review,

32(1):92–115.

Bernard, A. B. and Durlauf, S. N. (1995). Convergence in international output.

Journal of Applied Econometrics, 10(2):97–108.

Brodjonegoro, B. (2004). The Effect of Decentralization on Business in Indonesia. In

Basri, M. C. and van der Eng, P., editors, Business in Indonesia : New Challenges,

Old Problem. Singapore: Institute of Southeast Asian Studies.

Corrado, L., Martin, R., and Weeks, M. (2005). Identifying and interpreting regional

convergence clusters across Europe. The Economic Journal, 115(502):C133–C160.

42 REGIONAL INCOME DISPARITIES AND CONVERGENCE CLUBS IN INDONESIA

Dabla-Norris, E., Kochhar, K., Ricka, F., Suphaphiphat, N., and Tsounta, E. (2015).

Causes and Consequences of Income Inequality: A Global Perspective. IMF Staff

Discussion Notes, SDN/15/13.

Esmara, H. (1975). Regional income disparities. Bulletin of Indonesian Economic

Studies, 11(1):41–57.

Garcia, J. G. and Soelistianingsih, L. (1998). Why do differences in provincial

incomes persist in Indonesia? Bulletin of Indonesian Economic Studies, 34(1):95–

120.

Hill, H. (1991). Unity and Diversity: Regional Economic Development in Indonesia

since 1970. Oxford University Press, London.

Hill, H., Resosudarmo, B., and Vidyattama, Y. (2008). Indonesia’s changing

economic geography. Bulletin of Indonesian Economic Studies, 44(3):407–435.

Hobijn, B. and Franses, P. H. (2000). Asymptotically perfect and relative convergence

of productivity. Journal of Applied Econometrics, 15(1):59–81.

Islam, N. (1995). Growth empirics: A panel data approach. Quarterly Journal of

Economics, 110(4):1127–1170.

Islam, N. (2003). What have we learnt from the convergence debate? Journal of

Economic Surveys, 17(3):309–362.

Johnson, P. and Papageorgiou, C. (2020). What Remains of Cross-Country

Convergence? Journal of Economic Literature, 58(1):129–175.

Kurniawan, H., de Groot, H. L. F., and Mulder, P. (2019). Are poor provinces catching-

up the rich provinces in Indonesia? Regional Science Policy & Practice, 11(1):89–

108.

Magrini, S. (2004). Regional (Di) Convergence. In Henderson, V. and Thisse, J.,

editors, Handbook of Regional and Urban Economics, chapter 62, pages 2741—-

2796. Elsevier, Amsterdam.

NEW DISTRICT-LEVEL EVIDENCE 43

Magrini, S. (2009). Why Should We Analyse Convergence Using the Distribution

Dynamics Approach? Regional Science (Scienze Regionali), 8(1):5–34.

Mahi, B. R. (2016). Indonesia Decentralization: Evaluation, Recent Movement and

Future Perspectives. Journal of Indonesian Economy and Business, 31(1):119–133.

Mankiw, N. G., Romer, D., and Weil, D. N. (1992). A contribution to the empirics of

economic growth. Quarterly Journal of Economics, 107(2):407–437.

Martin, P. and Ottaviano, G. P. I. (2001). Growth and agglomeration. International

Economic Review, 42(4):947–968.

Mendez, C. and Kataoka, M. (2020). Disparities in regional productivity, capital

accumulation, and efﬁciency across Indonesia: A club convergence approach.

Review of Development Economics.

Mendez-Guerra, C. (2018). Beta, Sigma and Distributional Convergence in Human

Development? Evidence from the Metropolitan Regions of Bolivia. Latin

American Journal of Economic Development, 30(Nov):87–115.

Mishra, S. C. (2009). Economic inequality in Indonesia: Trends, causes and policy

response. UNDP Report for Strategic Asia, pages 139–206.

Nasution, A. (2016). Government Decentralization Program in Indonesia. Asian

Development Bank Institute Working Paper Series, 601.

Ostry, J. D., Berg, A., and Tsangarides, C. (2014). Redistribution, Inequality, and

Growth. IMF Staff Discussion Note, 14/02.

Phillips, P. C. B. and Shi, Z. (2019). Boosting: Why you can use the HP ﬁlter. Cowles

Foundation Discussion Paper No.2212, (2212).

Phillips, P. C. B. and Sul, D. (2007). Transition modeling and econometric

convergence tests. Econometrica, 75(6):1771–1855.

Phillips, P. C. B. and Sul, D. (2009). Economic transition and growth. Journal of

Applied Econometrics, 24(7):1153–1185.

44 REGIONAL INCOME DISPARITIES AND CONVERGENCE CLUBS IN INDONESIA

Quah, D. (1996). Empirics for economic growth and convergence. European

Economic Review, 40(6):1353–1375.

Rodriguez-Pose, A. and Ezcurra, R. (2010). Does decentralization matter for

regional disparities? (A) cross-country analysis. Journal of Economic Geography,

10(5):619–644.

Rodrik, D. (2013). Unconditional convergence in manufacturing. Quarterly Journal

of Economics, 128(1):165–204.

Sala-I-Martin, X. (1996a). Regional cohesion: Evidence and theories of regional

growth and convergence. European Economic Review, 40(6):1325–1352.

Sala-I-Martin, X. (1996b). The classical approach to convergence analysis. The

Economic Journal, 106(July):1019–1036.

Tadjoeddin, M. Z., Suharyo, W. I., and Mishra, S. (2001). Regional disparity and

vertical conﬂict in Indonesia. Journal of the Asia Paciﬁc Economy, 6(3):283–304.

Tijaja, J. and Faisal, M. (2014). Industrial Policy in Indonesia: A Global Value Chain

Perspective. ADB Economics Working Paper Series, 411.

Tirtosuharto, D. (2013). Regional inequality in Indonesia: Did convergence Occur

Following the 1997 Financial Crisis? Proceeding of The Paciﬁc Conference of

Regional Science Association International.

Uhlig, H. and Ravn, M. (2002). On adjusting the Hodrick-Prescott ﬁlter for the

frequency of observations. Review of economics and statistics, 84(2):371–376.

van der Weide, R. and Milanovic, B. (2018). Inequality is Bad for Growth of the Poor

(but Not for That of the Rich). The World Bank Economic Review, 32(3):507–530.

Vidyattama, Y. (2013). Regional convergence and the role of the neighbourhood

effect in decentralised Indonesia. Bulletin of Indonesian Economic Studies,

49(2):193–211.

NEW DISTRICT-LEVEL EVIDENCE 45

Zhang, W., Xu, W., and Wang, X. (2019). Regional convergence clubs in China:

Identiﬁcation and conditioning factors. Annals of Regional Science, 62(2):327–

350.

46 REGIONAL INCOME DISPARITIES AND CONVERGENCE CLUBS IN INDONESIA

Appendix A: Interpolating new districts data

During the decentralization, the number of districts in Indonesia increased

signiﬁcantly from 342 districts in 2000 to 514 districts in 2017. Due to this

administrative proliferation, the sequential time series of GDP per capita at

the constant price at the district level are difﬁcult to obtain. Therefore, in

this paper, we construct the balanced panel dataset for 514 districts level

deﬂated by the constant price of 2010 from the Central Bureau of Statistics

and INDODAPOER-World Bank. Speciﬁcally, we interpolated all missing

values in both the regional GDP at constant price and the number of

population. Also, we adjusted the historical data in some reference districts

to avoid the structural break, i.e., the time series of reference districts before

and after split-up.

Similar to Kurniawan et al. (2019), this paper uses a linear regression

method with the reference district and (or) year as regressor(s) for the

interpolation. We assumed that the new district is having co-movement

with its reference district. The new districts’ trend refers to their actual

available data and follows its reference district (or year) when they are

interpolated. For example, in Figure 11 we illustrate the comparison of GDP

per capita of one of the proliferated districts, the Sungai Penuh City that

became a new district in 2009 after separated from its origin district, the

Regency of Kerinci.

In brief, our imputation/interpolation steps can be summarized as

follows:

1. Constructing the reference district: The historical data of original

district before split-up and the sum of the composite of new district(s)

data after the proliferation are used as reference district.

2. Imputation and data adjustment: The time-series data of the original

NEW DISTRICT-LEVEL EVIDENCE 47

district are adjusted by subtracting the reference district’s time series

data with the new district’s imputed data before the proliferation year.

3. Constructing population data: The missing data of each district’s total

population are imputed using linear calculation of the population’s

share to the total province population.

4. Calculating regional GDP per capita: The per capita GDP at constant

price for districts level is obtained by dividing GDP by each district’s

total population.

Figure 11: Comparison of imputed data and its reference

48 REGIONAL INCOME DISPARITIES AND CONVERGENCE CLUBS IN INDONESIA

Appendix B: Club membership

Club 1 [6]: Jakarta Pusat, Kediri, Mamberamo Raya, Mimika, Morowali, Teluk

Bintuni.

Club 2 [126]: Badung, Balangan, Balikpapan, Banda Aceh, Bandung,

Banggai, Bangka Barat, Banyuwangi, Barito Utara, Batam, Batang Hari, Batu

Bara, Bau-Bau, Bekasi, Bengkalis, Berau, Bima, Bintan, Bitung, Bojonegoro,

Bontang, Boven Digoel, Bukittinggi, Bulungan, Buton Utara, Cilacap,

Cilegon, Cirebon, Dumai, Fakfak, Gianyar, Gresik, Indragiri Hilir, Indragiri

Hulu, Jakarta Barat, Jakarta Selatan, Jakarta Timur, Jakarta Utara, Jayapura,

Jayapura, Kampar, Karawang, Karimun, Keerom, Kendari, Kepulauan

Anambas, Kepulauan Meranti, Kepulauan Seribu, Kolaka, Kolaka Utara,

Konawe Utara, Kota Baru, Kota Batu, Kuantan Singingi, Kudus, Kutai Barat,

Kutai Kartanegara, Kutai Timur, Labuhan Batu, Labuhan Batu Selatan,

Labuhan Batu Utara, Lamandau, Madiun, Magelang, Mahakam Ulu,

Makassar, Malang, Malinau, Mamuju Utara, Manado, Manokwari, Maros,

Medan, Mesuji, Minahasa Utara, Mojokerto, Mojokerto, Morowali Utara,

Muara Enim, Murung Raya, Musi Banyuasin, Nabire, Natuna, Nunukan,

Padang, Padang Panjang, Palembang, Palu, Pangkajene Kepulauan,

Parepare, Pariaman, Pasir, Pasuruan, Pekanbaru, Pelalawan, Pematang

Siantar, Pinrang, Purwakarta, Rokan Hilir, Salatiga, Samarinda, Sarmi,

Sawahlunto, Semarang, Siak, Sibolga, Sidoarjo, Sorong, Sorong, Sumbawa

Barat, Sungai Penuh, Surabaya, Surakarta, Tabalong, Tana Tidung,

Tangerang, Tanjung Jabung Barat, Tanjung Jabung Timur, Tanjung Pinang,

Tapanuli Selatan, Tarakan, Tegal, Wajo, Wakatobi, Waropen, Yogyakarta.

Club 3 [178]: Aceh Tengah, Agam, Asahan, Bandar Lampung, Banggai

Kepulauan, Banggai Laut, Banjarmasin, Bantaeng, Banyumas, Barito

Selatan, Barito Timur, Barru, Belitung, Belitung Timur, Bengkulu, Bengkulu

Tengah, Binjai, Blitar, Blitar, Bogor, Bolaang Mongondow Selatan, Bolaang

NEW DISTRICT-LEVEL EVIDENCE 49

Mongondow Timur, Bolaang Mongondow Utara, Bombana, Bone, Buleleng,

Bulukumba, Bungo, Buol, Buton, Buton Selatan, Cimahi, Dairi, Deli

Serdang, Denpasar, Dharmasraya, Donggala, Enrekang, Gorontalo,

Gorontalo, Gorontalo Utara, Gunung Mas, Gunung Sitoli, Indramayu, Intan

Jaya, Jambi, Jayawijaya, Jember, Jembrana, Jombang, Kaimana, Kapuas,

Karanganyar, Karangasem, Karo, Katingan, Kendal, Kep.Siau Tagulandang

Biaro, Kepulauan Mentawai, Kepulauan Sangihe, Kepulauan Selayar,

Kerinci, Ketapang, Klungkung, Konawe, Konawe Kepulauan, Konawe

Selatan, Kotawaringin Barat, Kotawaringin Timur, Kubu Raya, Kupang,

Lahat, Lamongan, Lampung Selatan, Lampung Tengah, Lampung Timur,

Lampung Utara, Langkat, Lhokseumawe, Limapuluh Kota, Lingga,

Lumajang, Luwu, Luwu Timur, Luwu Utara, Magetan, Malang, Mamberamo

Tengah, Mamuju, Mandailing Natal, Mappi, Mataram, Merangin, Merauke,

Metro, Minahasa, Minahasa Selatan, Minahasa Tenggara, Muaro Jambi,

Muna, Muna Barat, Musi Rawas, Musi Rawas Utara, Nagan Raya, Pacitan,

Padang Lawas Utara, Padang Pariaman, Palangkaraya, Palopo, Pangkal

Pinang, Parigi Moutong, Pasaman Barat, Pasuruan, Pati, Payakumbuh,

Pegunungan Bintang, Penajam Paser Utara, Pesawaran, Pohuwato, Polewali

Mandar, Pontianak, Poso, Prabumulih, Pringsewu, Probolinggo, Pulang

Pisau, Raja Ampat, Rejang Lebong, Rokan Hulu, Sabang, Sambas, Samosir,

Sarolangun, Semarang, Serang, Serang, Serdang Bedagai, Sidenreng

Rappang, Sigi, Sijunjung, Simalungun, Singkawang, Sinjai, Sleman, Solok,

Solok, Soppeng, Sorong Selatan, Sragen, Sukabumi, Sukamara, Sukoharjo,

Sumbawa, Supiori, Tabanan, Takalar, Tana Toraja, Tanah Bumbu, Tanah

Datar, Tanah Laut, Tangerang Selatan, Tanjung Balai, Tapin, Tasikmalaya,

Tebing Tinggi, Tebo, Teluk Wondama, Ternate, Toba Samosir, Tojo

Una-Una, Toli-Toli, Tomohon, Toraja Utara, Tuban, Tulang Bawang, Tulang

Bawang Barat, Tulungagung, Yalimo.

Club 4 [181]: Aceh Barat, Aceh Barat Daya, Aceh Besar, Aceh Jaya, Aceh

50 REGIONAL INCOME DISPARITIES AND CONVERGENCE CLUBS IN INDONESIA

Selatan, Aceh Singkil, Aceh Tamiang, Aceh Tenggara, Aceh Utara, Alor,

Ambon, Asmat, Bandung, Bandung Barat, Bangka, Bangka Selatan, Bangka

Tengah, Bangkalan, Bangli, Banjar, Banjar, Banjar Baru, Banjarnegara,

Bantul, Banyuasin, Barito Kuala, Batang, Bekasi, Belu, Bener Meriah,

Bengkayang, Bengkulu Selatan, Bengkulu Utara, Biak Numfor, Bireuen,

Boalemo, Bogor, Bolaang Mongondow, Bondowoso, Bone Bolango,

Boyolali, Brebes, Buru Selatan, Buton Tengah, Ciamis, Cianjur, Cirebon,

Demak, Depok, Dogiyai, Dompu, Empat Lawang, Ende, Flores Timur,

Garut, Gayo Lues, Gowa, Grobogan, Gunung Kidul, Halmahera Barat,

Halmahera Selatan, Halmahera Tengah, Halmahera Timur, Halmahera

Utara, Hulu Sungai Selatan, Hulu Sungai Tengah, Hulu Sungai Utara,

Humbang Hasundutan, Jepara, Kapuas Hulu, Kaur, Kayong Utara,

Kebumen, Kediri, Kepahiang, Kepulauan Aru, Kepulauan Sula, Kepulauan

Talaud, Kepulauan Tidore, Kepulauan Yapen, Klaten, Kolaka Timur,

Kotamobagu, Kulon Progo, Kuningan, Lampung Barat, Landak, Langsa,

Lebak, Lebong, Lombok Barat, Lombok Tengah, Lombok Timur, Lombok

Utara, Lubuklinggau, Madiun, Magelang, Majalengka, Majene, Malaka,

Maluku Barat Daya, Maluku Tengah, Maluku Tenggara, Maluku Tenggara

Barat, Mamasa, Mamuju Tengah, Manokwari Selatan, Maybrat, Melawi,

Mempawah, Mukomuko, Ngada, Nganjuk, Ngawi, Nias, Nias Barat, Nias

Selatan, Nias Utara, Ogan Ilir, Ogan Komering Ilir, Ogan Komering Ulu,

Ogan Komering Ulu Selatan, Ogan Komering Ulu Timur, Padang Lawas,

Padang Sidempuan, Pagar Alam, Pakpak Bharat, Pamekasan, Pandeglang,

Pangandaran, Paniai, Pasaman, Pekalongan, Pekalongan, Pemalang,

Penukal Abab Lematang Ilir, Pesisir Barat, Pesisir Selatan, Pidie, Pidie Jaya,

Ponorogo, Probolinggo, Pulau Morotai, Pulau Taliabu, Purbalingga,

Purworejo, Rembang, Sampang, Sanggau, Sekadau, Seluma, Seram Bagian

Barat, Seram Bagian Timur, Seruyan, Sikka, Simeulue, Sintang, Situbondo,

Solok Selatan, Subang, Subulussalam, Sukabumi, Sumba Barat, Sumba

NEW DISTRICT-LEVEL EVIDENCE 51

Timur, Sumedang, Sumenep, Tambrauw, Tangerang, Tanggamus, Tapanuli

Tengah, Tapanuli Utara, Tasikmalaya, Tegal, Temanggung, Timor Tengah

Selatan, Timor Tengah Utara, Trenggalek, Tual, Way Kanan, Wonogiri,

Wonosobo.

Club 5 [23]: Aceh Timur, Bima, Blora, Buru, Deiyai, Jeneponto, Kupang,

Lanny Jaya, Lembata, Manggarai, Manggarai Barat, Manggarai Timur,

Nagekeo, Nduga, Pegunungan Arfak, Puncak, Puncak Jaya, Rote Ndao, Sabu

Raijua, Sumba Barat Daya, Sumba Tengah, Tolikara, Yahukimo.

52 REGIONAL INCOME DISPARITIES AND CONVERGENCE CLUBS IN INDONESIA

Appendix C: Test for clubs merging

Phillips and Sul (2009) propose a merging test to evaluate whether the clubs

identiﬁed according to the clustering algorithm described in Section 3.3

can be merged. Thus, we used a club merging algorithm by Phillips and Sul

(2009) to test for merging between the adjacent clubs. The procedure works

as follows:

1. The log ttest is applied on the ﬁrst two initial groups identiﬁed in the

clustering mechanism described in Section 3.3. If the t-statistic is

larger than -1.65, these two groups together form a new convergence

club;

2. The log ttest is repeated by adding the next club, and the process

continues until the condition of t-statistic is larger than -1.65 is

achieved;

3. If the convergence hypothesis is rejected, that is when t-statistic is

larger than -1.65 does not hold, we assume that all previous groups

converge, except the last added one. Hence, we restart the merging

algorithm from the club for which the hypothesis of convergence does

not hold.

NEW DISTRICT-LEVEL EVIDENCE 53

Appendix D: Convergence test using a trend

parameter of 400

Table 9: Global convergence test (trend parameter of 400)

Coefﬁcient Standard Error t-statistic

log(t) -0.53 0.02 -23.02

Note: The null hypothesis of convergence is rejected when the t-statistic is less than -1.65.

Table 10: Local convergence test (trend parameter of 400)

Club1 Club2 Club3 Club4 Club5 Club6

Coefﬁcient 0.61 0.27 0.02 0.25 0.03 0.67

t-statistic 6.25 3.25 0.42 3.89 1.25 51.39

N. of regions 4 10 106 186 50 4

Club7 Club8 Club9 Club10 Club11 Club12

Coefﬁcient 2.43 1.81 1.33 0.81 2.54 0.00

t-statistic 9.48 8.73 11.84 11.88 8.24 -.07

N. of regions 24 70 10 24 4 22

Note: The null hypothesis of convergence is rejected when the t-statistic is less than -1.65.

54 REGIONAL INCOME DISPARITIES AND CONVERGENCE CLUBS IN INDONESIA

Table 11: Merge test (trend parameter of 400)

Club1+2 Club2+3 Club3+4 Club4+5 Club5+6 Club6+7

Coeff 0.30 -0.07 -0.24 0.02 0.08 2.25

T-stat 3.79 -1.49 -6.39 0.58 3.85 9.86

Club7+8 Club8+9 Club9+10 Club10+11 Club11+12

Coeff 1.08 1.32 0.19 0.27 -0.16

T-stat 8.41 8.56 6.20 6.49 -6.05

Note: The null hypothesis of convergence is rejected when the t-statistic is less than -1.65.

Table 12: After-merge clubs (trend parameter of 400)

Club1 Club2 Club3 Club4 Club5

Coeff 0.30 0.02 0.01 0.03 0.00

T-stat 3.79 0.42 0.41 0.71 -0.07

Size 14 106 240 132 22

Note: The null hypothesis of convergence is rejected when the t-statistic is less than -1.65.

NEW DISTRICT-LEVEL EVIDENCE 55

.8

1

1.2

1.4

GDP per capita

2000 2005 2010 2015 2020

Club 1

.8

1

1.2

1.4

GDP per capita

2000 2005 2010 2015 2020

Club 2

.8

1

1.2

1.4

GDP per capita

2000 2005 2010 2015 2020

Club 3

.8

1

1.2

1.4

GDP per capita

2000 2005 2010 2015 2020

Club 4

.8

1

1.2

1.4

GDP per capita

2000 2005 2010 2015 2020

Club 5

Figure 12: Convergence clubs and transition paths (trend parameter of 400)

56 REGIONAL INCOME DISPARITIES AND CONVERGENCE CLUBS IN INDONESIA

.8

.9

1

1.1

1.2

GDP per capita

2000 2005 2010 2015 2020

Club = 1

Club = 2

Club = 3

Club = 4

Club = 5

Figure 13: Convergence clubs trends (trend parameter of 400)

NEW DISTRICT-LEVEL EVIDENCE 57

Appendix E: Classical convergence analysis for a

sample of 342 districts

.66

.68

.7

.72

.74

.76

SD of Log GDP per capita

2000 2005 2010 2015 2020

Figure 14: Sigma convergence across 342 districts

-2

-1

0

1

2

Growth Rate 2000-2017

810 12 14

Log GDP per capita in 2000

Figure 15: Beta convergence across 342 districts