Income Inequality Revisited:

(1)

"

!

Government Decentralization and

Income Inequality Revisited:

An Inequality Decomposition Analysis

MSc Thesis International Economics & Business

by

S.E.M. Weersink

University of Groningen Faculty of Economics & Business

!

! !

!

E-mail: stephanieweersink@gmail.com Student number: s2081407 Date: January 5^th 2016

Supervisor: Dr. R.K.J. Maseland

Co-assessor: Prof. Dr. D.M. Swagerman

! !

(2)

(3)

Abstract

Recent scholarly work has looked into the impact of government decentralization on income inequality, without considering one important aspect: its composition. Gaining understanding of the composition of inequality is of major importance for policy makers to be able to formulate inequality-reducing policies targeted at the core problem. This thesis covers this gap by decomposing the inequality index and examining the impact of both fiscal and political decentralization on the absolute level and relative share of interregional disparities in total inequality. Evidence indicates that political decentralization within OECD countries is associated with a rise in the level of interregional inequality, both in absolute and relative terms. The impact of fiscal decentralization is also found to be positive, however, the results lack robustness across the different specifications considered. Limited data availability does not allow for drawing any firm conclusions. Still, the results underline the topic’s policy relevance and further research is desired when better data becomes available.

Key words: fiscal decentralization, political decentralization, income inequality, regional

disparities, inequality decomposition.

! !

(4)

(5)

1.!Introduction

Income inequality within most industrialized countries has reached historical highs (OECD, 2015). The richest ten percent of the total population of the 18 industrialized OECD countries earns more than ten times the income of the poorest ten percent. This widespread increase in income inequality has been accompanied by growing public concern and discussion over the impact of the high and often rising gap between rich and poor on society (OECD, 2015).

Whereas disparities between countries have been diminishing in the past couple of years, those within countries have not declined (OECD, 2009). Income inequality is thought to have a negative effect on economic growth and to undermine the social structure of countries. It is therefore no surprise that debates on the issue of income inequality have moved to the top of the policy agenda in many countries (OECD, 2015).

Alongside this trend towards rising income inequalities, there has been an equally prevalent tendency for countries to devolve authority and resources from central to subnational government tiers (Rodriguez-Poze & Gill, 2004). A large number of both developed and developing countries, across unitary and federal states and in rich and poor nations, have either embarked upon, or stated their intention to embark upon, some form of government decentralization (Martinez-Vazquez & McNab, 2003). There are multiple reasons that might explain this increasing interest in decentralization. Often, the primary stated policy objective is to increase the efficiency of public expenditures and to foster economic growth and development (Oates, 1993). Decentralization is seen as a way to break with large centralized bureaucracies and to increase social capital. According to Tselios (2012), the initiatives have often been built on both efficiency and equity grounds.

To date, we have limited understanding of the actual impact of government decentralization on the distribution of income (Sepulveda & Martinez-Vazquez, 2011). However, as many countries are simultaneously implementing policies directed at reducing income inequality, as well as decentralization initiatives, it is important to define the extent to which these policy strategies interact with each other (Sacchi & Salotti, 2014). Most of the existing literature has focused on the impact of fiscal decentralization on interregional inequality (Shankar & Shah, 2003; Rodríguez-Pose & Gill, 2004; Ezcurra & Pascual, 2008), whereas research on within- regional inequality (Morelli & Seaman, 2007; Tselios et al., 2012; Sacchi & Salotti, 2014) and overall income inequality (Sepúlveda & Martinez-Vazquez, 2011) has been more limited.

(8)

What becomes clear, the results of the existing literature are inconclusive. Differing samples, measurement and definition issues, and potential endogeneity are possible explanations for this lack of consensus.

The purpose of this paper is to shed new light on the interactions between government decentralization and income inequality. From the existing literature it seems that one aspect of income inequality has been overlooked: its composition. National averages of the Gini index hide important intra-country variance in the distribution of incomes (Piacentini, 2014), considering overall income inequality encompasses both within-groups and between-groups inequality. A recently published report from the OECD, written by Piacentini (2014), confirmed that there are significant variations in income inequality levels within countries, and that regional breakdowns are valuable for comprehending sources and patterns of income disparities within and between regions. According to Kanbur (2006), inequality decompositions are an effective tool for analyzing inequality and “allow useful depictions of patterns that can be a first step in identifying the proximate causes of inequality” (p. 368).

These insights are highly relevant for monitoring and policy decisions (Piacentini, 2014).

When interregional inequality relatively increases as a consequence of government decentralization, the solidarity and sense of unity between regions may be undermined. High regional disparities can impair national economic growth, since lagging regions have a negative influence on the efficiency at the national level (Tselios, 2012). Moreover, regional disparities may lead to inequality of opportunities and the creation of social tensions, which in turn may cause these disparities to sustain over time (Piacentini, 2014). In an extreme case, it may even create the possibility of disunity (Tselios, 2012). It is therefore important for policy makers to understand the roots of income inequality to be able to formulate policies targeted at the core problem. At given levels of total income inequality, a country where low incomes are highly concentrated in a few areas faces different challenges from a country where low incomes are equally spread across space. Consequently, awareness of the structure of income inequality is important in the formulation of policy and its expected effect (Heshmati, 2004).

Policy actors need to be aware not only of the impact of government decentralization on the level of overall or interregional income inequality, but also on the inherent composition of inequality. In this line of reasoning, the key hypothesis driving this thesis concerns the impact of government decentralization on the composition of income inequality and the relative change of this composition over time. Using a sample of 22 OECD countries, the paper

(9)

provides an empirical test of the influence of decentralization on the composition of inequality by looking into regional inequalities within OECD countries on NUTS 2 level, and examining the extent to which these explain total inequality – as indicated by the Gini index – over time.

The influence of government decentralization on this ratio is subsequently investigated. In this respect, this thesis differs from previous studies as its main focus is not on the impact of government decentralization on the absolute level of overall, intraregional or interregional income inequality, but instead on the composition of inequality. In addition, this thesis dives into the causality issue, since some literature has indicated decentralization might be endogenous to income inequality (Beramendi, 2003; 2007).

The remainder of this thesis is organized as follows. In the next section, government decentralization will be defined, after which the existing literature in the field of decentralization, intraregional, interregional, and overall income inequality will be reviewed.

In section III the methodology and data are outlined. The empirical results follow in section IV and section V concludes.

" !

(10)

2.!Literature!review!

2.1! Government!decentralization!

Government decentralization is defined as the devolution of power from central to subnational governments and is characterized in the literature as a multi-dimensional process (Ezcurra &

Rodríguez-Pose, 2013). Recent scholarly work on the interactions between government decentralization and income inequality has typically resorted to one of these dimensions, being fiscal decentralization. Fiscal decentralization relates to the devolution of resources from national to subnational authorities. A dimension often disregarded in the literature is political decentralization, characterized by the devolution of power or autonomy. It is important to note that these two dimensions do not always match (Ezcurra & Rodríguez-Pose, 2013). A subnational government may have a large part of the public budget at its disposal, however, it may lack the decision-making authority on how to spend it (Gil Canaleta et al., 2004). Looking into both of these dimensions of government decentralization – fiscal as well as political – will therefore provide a more complete picture when analyzing the interactions between government decentralization and the composition of income inequality.

2.2! Decentralization!and!interregional!inequality!!

Regional disparities represent an ever-present development challenge (Shankar & Shah, 2003). Whereas disparities between countries have been diminishing over the past couple of years, those within countries have not declined (OECD, 2009). Persistent interregional disparities may impair national economic growth and can lead to unequal opportunities and the creation of social tensions, which in turn may cause these disparities to sustain over time (Piacentini, 2014). As Tselios (2012, p. 1279) argues, “the underutilization and underperformance of workers and productive capacity in lagging regions lowers overall national wealth.” Moreover, large territorial disparities can represent a serious threat by causing frictions, us-versus-them thinking and even creating the potential for disunity.

Decentralization initiatives have often been built on efficiency and equity grounds (Tselios, 2012). It is believed that fiscal decentralization will contribute to economic growth and development (Oates, 1993) and, ultimately, to the reduction of regional inequalities (Qian &

Weingast, 1997). However, whether the last objective is the actual result remains a topic of debate in the literature and no clear conclusion can be drawn. In development economics there is a presumption that decentralized fiscal constitutions would lead to ever-widening interregional disparities (Shankar & Shah, 2003). According to Prud’homme (1995), fiscal

(11)

decentralization tends to weaken the equalization role of the government performed through social and territorial transfers, with spatially regressive effects as a result. Whereas centralized public sectors will attempt to create a more balanced distribution by channeling resources from richer to poorer areas, this is less likely within a decentralized government function. The influence of poorer areas over the allocation of monetary resources and transfers across the country may be reduced. As Rondinelli (1990, p. 492) argues: “Even the strongest advocates of decentralization recognize that central governments often enact and implement policies that lead to greater territorial justice or the redistribution of wealth more effectively than do territorial units.” The transfer of power and resources to subnational governments is thought to disproportionally benefit those regions with a greater capacity to actually fulfill allocative and productive efficiency, which naturally are the most prosperous regions with better socio- economic endowments and higher quality institutions (Rodriguez-Pose & Ezcurra, 2009). As a consequence, decentralization initiatives may further enhance the differences in institutional capacities and socio-economic endowments across regions.

Although increasing regional inequalities are driven by many forces, there is a clear congruency between the timing of decentralization initiatives and increasing regional disparities (Rodríguez-Pose & Gill, 2004). Starting from the 1980s many developed countries have increased their degree of fiscal decentralization. For example, Belgium became a federal state in the early 1990s and Italy is also moving in that direction (Ezcurra & Pascual, 2008).

Similarly, in Portugal and Spain fiscal decentralization is in its course of development.

Alongside this trend towards decentralization, income inequalities in most OECD countries have steadily increased. Rodriguez-Pose and Gill (2004) argue that decentralization has increased, rather than decreased, regional inequalities.

There are, nevertheless, reasons to assume that government decentralization contributes to a reduction in regional disparities. A study conducted by Shankar & Shah (2003), investigating a sample of 8 industrial and 17 developing countries, resulted in a significant negative effect of decentralization on regional disparities. The empirical model used, however, did not consider the level of decentralization in the various countries considered. Countries were labeled either unitary or federal as if indicating centralized versus decentralized, which can be considered a very crude measure of government decentralization. Consequently, their results are not entirely convincing. Ezcurra & Pascual (2008) investigated the link between fiscal decentralization and regional disparities in a set of European Union countries, and also found

(12)

that decentralization forms an incentive for regional competition, as the ability of subnational governments to stay in power depends on their performance in obtaining a level of economic growth and development similar to that enjoyed by the rest of the country (Ezcurra & Pascual, 2008). Government decentralization is also likely to create competition for the scarce resources originating from the national government. The playing field is, however, not level with richer and larger regions likely to have a greater say in central decision-making (Rodríguez-Pose & Ezcurra, 2009). This may imply that poorer regions have less influence over the allocation of financial resources and become less well protected by the central government as decentralization evolves, while richer and more powerful regions might well benefit from the process (Rodríguez-Pose & Gill, 2004).

According to Rodríguez-Pose & Ezcurra (2009), the economic environment in which decentralization occurs might influence its effect on interregional inequality. A rich and developed country with high-quality institutions might experience different consequences from decentralization than a poor developing country with low institutional quality. This aspect has often been overlooked in the literature (Rodríguez-Pose & Ezcurra, 2009). Hence, the factors that are expected to push government decentralization towards greater regional disparities might have less of an impact within the OECD countries considered in this thesis.

Redistribution regimes tend to be more present in these countries and the negative effect of decentralization might therefore be mitigated (Rodríguez-Pose & Ezcurra, 2009).

As mentioned before, the empirical evidence regarding the interactions between government decentralization and the evolution of interregional disparities is scarce and inconclusive. The majority of the studies focusing on developing countries tend to find that decentralization causes regional disparities to rise, while the evidence from studies covering developed countries is more mixed (Rodríguez-Pose & Ezcurra, 2009). The prevailing view remains that decentralization initiatives tend to increase regional inequalities, by disproportionally benefiting those regions that have a greater capacity to realize efficiency gains (Rodríguez- Pose & Ezcurra, 2009). Economic forces towards greater concentration of economic activity in prosperous and influential regions seem to be both stimulated and facilitated by decentralization initiatives (Rodríguez-Pose & Gill, 2004). Breaking with this regional concentration of power and resources may be at the cost of national growth (Shankar & Shah, 2003). Altogether, the above considerations lead to the following hypothesis:

Hypothesis 1: Government decentralization leads to more interregional inequality

(13)

2.3! Decentralization!and!intraregional!inequality!

In comparison to the literature on the impact of government decentralization on interregional disparities, the literature concerning the impact on intraregional disparities is more limited (for exceptions see: Morelli & Seaman, 2007; Tselios et al., 2012; Sacchi & Salotti, 2014). As Tselios et al. (2012) mention: “our knowledge about the influence, if at all, of decentralization processes on interpersonal inequalities is extremely limited and patchy” (p. 1279).

Morelli & Seaman (2007) assess the impact of decentralization within the United Kingdom on household income equality. Their focus is on intraregional inequality, corresponding to within-regional or interpersonal inequality. It is argued that decentralization will improve the distribution of income when resources are utilized more efficiently at the local level. The findings show that intraregional convergence is apparent. However, the role of decentralization in this trend appears to be limited.

Tselios et al. (2012) look at the impact of fiscal, as well as political decentralization on within-regional income inequality in the European Union. Through the use of regionally aggregated micro-economic data for more than 100,000 individuals, it is shown that the impact of fiscal decentralization on intraregional income inequality is negative and much stronger than expected. The effect of political decentralization is less clear. One of the arguments, related to the main fiscal federalism theory, is that subnational governments may have an information advantage over national governments when it comes to responding to the needs of local citizens. Decentralization is considered to encourage greater voice and participation, empowering the under-represented groups in society. Transferring authority and resources to subnational governments can therefore promote greater harmony between public policies and local needs (Tselios et al., 2012). When interpersonal inequalities are considered to be an important issue at the local level, it can be expected that local policies will counter these inequalities more effectively than policies implemented by bureaucrats in distant central governments (Tselios et al., 2012). Regions characterized by high interpersonal inequality may choose a more progressive tax-transfer scheme compared to regions with low preference for redistribution. In contrary to the view that lagging regions would be disadvantaged because of capacity and funding constraints, which could hinder them from effectively implementing policies aimed at reducing intraregional inequality, Tselios et al. (2012) find that it is exactly these less well-off regions which seem to profit the most from the inequality- reducing effects of fiscal decentralization.

(14)

On the contrary, Sacchi & Salotti (2014) find that a higher level of fiscal decentralization is associated with a more unequal interpersonal income distribution. However, the authors use the national Gini index as a measure for inequality, which incorporates both inequality within and between regions. The possibility exists that fiscal decentralization’s negative impact on the distribution of income is a consequence of increased regional inequalities, not increased interpersonal inequalities. Hence, the study is not entirely convincing and the conclusions drawn need to be considered with caution.

Even though the evidence is inconclusive and limited, theoretical contributions predict that decentralization may reduce interpersonal inequality. The line of argumentation is that higher responsibilities of subnational governments bring local policymakers closer to their citizens, favoring policies more sensitive to the existence of interpersonal disparities and poverty (Le Galès, 2002; Brenner, 2004). This is in line with the findings by Tselios et al. (2012) and leads to the formulation of the following hypothesis:

Hypothesis 2: Government decentralization leads to less intraregional inequality

On grounds of limited data availability, this hypothesis cannot be tested within this thesis. In the methodology and data section the underlying reason will be further explained.

2.4! Decentralization!and!overall!inequality!

Apart from the effect of fiscal decentralization on interregional and intraregional inequality, some studies have taken a more general approach by looking at the impact on total income inequality. It seems that there is possibly some confusion around the inequality concept, since the definitions of interpersonal, intra-regional and overall income inequality are not always aligned. “Interpersonal” is in some studies referring to overall income inequality (see: Sacchi

& Salotti, 2014), while in other cases it refers to within-regional inequality (see: Tselios, 2012). The study by Tselios et al. (2012) clearly examines inequality within regions and claims that our knowledge about interpersonal inequality is still ‘very limited and patchy’.

Sacchi & Salotti (2014) build their research on this statement, however, they use the national Gini index as a measure for inequality and therefore the focus is seemingly not on within- regional inequality, but on overall income inequality instead. As follows, the results of Sacchi

& Salotti (2014) can actually be placed under this heading and point in the direction of fiscal decentralization having a negative impact on the overall income distribution.

(15)

Taking the definition issues into account, mixed results arise from analyzing the interactions between fiscal decentralization and total income inequality. Sepulveda & Martinez-Vazquez (2011) have scrutinized the consequences of fiscal decentralization on a country’s income distribution on the basis of households. Using a large international panel dataset of 34 countries covering the past three decades, it appeared that fiscal decentralization tends to improve the distribution of income as long as the general government still represents a significant share of the economy (20% or more). In an earlier paper, Sacchi & Salotti (2011) also explored the effects of fiscal decentralization on overall income inequality. Fiscal decentralization was characterized by three tax revenue decentralization indices and four expenditure decentralization indices to measure the concept more thoroughly. The results indicated that expenditure decentralization is not associated with significant effects on income inequality, whereas tax revenue decentralization does lead to a more unequal distribution across households (Sacchi & Salotti, 2011). However, their measure of income inequality is based on gross household income, while post-tax disposable income is far more appropriate when looking into the effects of fiscal decentralization.

Again, no well-grounded conclusion about the sign of the relationship can be drawn from the existing literature. It is important to be aware, however, that even when decentralization would cause total inequality to decrease, interregional inequality may still increase and vice versa. Therefore, the impact of fiscal decentralization on the absolute level of inequality is not the topic of interest in this thesis. Instead, the focus will be on its composition and the contribution of interregional income variation to overall inequality.

Decomposing inequality is important for the formulation of policies, their anticipated effect and in evaluation of the impacts of inequality and redistribution on the welfare among regions (Heshmati, 2004). Accordingly, this thesis will decompose the Gini index and investigate the composition of income inequality over time, and the impact decentralization may have on changes within this composition. As will be explained in the methodology section, due to limited data availability and the fact that the Gini index is not decomposable into purely a between- and within-component, the specific focus will be on the proportional contribution of between-regional income variation to total inequality over time. The following hypothesis is therefore the main hypothesis of interest in this thesis:

Hypothesis 3: Government decentralization makes overall income inequality relatively more interregional

(16)

Endogeneity*issue*

The dominant assumption in the literature is that decentralization affects inequality, eliminating the possibility of effects working in the opposite direction. It is conventionally thought that decentralization lowers the likelihood of redistribution between regions, increasing regional economic inequality (Sacchi & Salotti, 2011). However, some authors claim inequality to be a determinant of decentralization (Bolton & Roland 1997; Beramendi 2003, 2007; Bodman & Hodge, 2010). Following Bodman & Hodge (2010), a more heterogeneous population increases the ‘ideological’ distance from the median voter, implying that voters tend to favor a lower share of national government expenditure.

Population heterogeneity may, besides factors such as differences in ethnicity and language, be caused by income inequality (Bodman & Hodge, 2010). In this line of argument, Bolton &

Roland (1997) argue that high territorial economic disparities may increase the desire for greater regional autonomy and government decentralization. Beramendi (2003) empirically examined whether the degree of regional inequality in a country affects the incentives for fiscal decentralization and argued that “whatever the impact of decentralization on the distribution of income may be, it is to a large extent a function of the internal structures of inequality” (p. 3). Differences in the demand for redistribution associated with interregional income inequalities lead to different preferences for decentralization. In another study by Beramendi (2007), the between-group share of inequality revealed a positive relation with the degree of decentralization in OECD countries. In case of considerable regional disparities, rich regions might want to protect themselves against undesired interregional redistributive policies by striving for fiscal decentralization. Moreover, prosperous regions might want to become more autonomous since they feel lagging regions keep them back in economic growth and lower national wealth (Tselios, 2012), which creates friction and us-versus-them thinking.

In such a situation, decentralization might actually be necessary to induce various regions to remain part of a federation (Tanzi, 1995).

Political actors, when deciding on the level of centralization or decentralization, are thought to be aware of the structure of inequality in the territory under their jurisdiction. Beramendi (2003, 2007) argues therefore that decentralization is endogenous to income inequality. In most of the existing literature, the possibility of bi-directional causality has been overlooked.

This thesis aims to gain insight into whether the relative share of interregional inequality within a country can be considered a determinant of the level of decentralization. This approach allows for the causality issue to be taken into account.

(17)

3.!Methodology!and!data!

3.1! Method!for!decomposing!income!inequality!!

A decomposition analysis will be performed in order to assess the significance of the spatial component in the aggregate level of income inequality. After having decomposed the commonly used Gini index, the subsequent step will be analyzing the possible influence of both fiscal and political decentralization on this composition.

The national Gini coefficient measures total inequality and is, in contrast to generalized entropy measures, not perfectly decomposable into the sum of within- and between-groups components. Instead, decomposing the Gini index results in a within-groups component, a between-groups component and a residual or interaction term. Let G be the Gini coefficient and let the subgroups, or in this case regions, be indexed by k = 1, 2, … , n. The decomposition can then be written as (Lambert & Aronson, 1993):

! = !_#+ %_&'!_&+ (

where !_# represents the between-groups Gini coefficient, which would be obtained if all incomes in every region were to be replaced by the relevant regional mean, %_& is the product of population share and income share going to region k, and !_& is the Gini coefficient for income within region k. The final term, R, is a residual term or ‘interaction effect’, which evaporates when the regional income distributions do not overlap (Lambert & Aronson, 1993). It is, however, hard to assume that this residual is zero. Instead, the interaction effect can be very significant and for some parts of the world the interaction term even accounts for the highest contribution to overall inequality (Lambert & Decoster, 2005). The residual is thought to be a between-groups and within-groups effect at the same time: “it measures a between-groups phenomenon, overlapping, that is generated by inequality within-groups”

(Lambert & Aronson, 1993: p. 1224). Due to limited data availability, this thesis can only compute the between-group component !_# and can therefore not make a distinction between within-regional inequality and the interaction term. There is no data on the variances and degrees of asymmetry within regions, but only on regional income means with which !_# can be determined. Regional means may however become less relevant when asymmetry within regions is high. To illustrate, when regional means come close and the variances and degree of asymmetry within these regions is relatively high, the overlapping component will become

(18)

more important. According to Milanovic (2002), the more important the overlapping component, the less one’s income depends on where he or she lives.

In spite of the inability to decompose the Gini index into its three components – which can be considered a limitation of this thesis – the topic of interest can still be analyzed. The ratio !_# versus total inequality ! reveals the relative importance of ‘net’ inequality between regions in explaining total inequality. Therefore, and in the absence of any superior alternative, this thesis proceeds on the assumption that the between-regional inequality expressed as a proportion of the overall Gini coefficient (!_#/!) captures the importance of the contribution of interregional income disparities to total inequality.

In order to measure the between-group component of the Gini coefficient, the method by Shorrocks & Wan (2005) is followed, according to which the between-group component !_# can be defined as:

!_# = 2

,^-. /(._&− .)

234₅ 6

&78

= 9_&:_& :_;− :_;

6

;7&

&

;78 6

&78

where countries are partitioned into < regions (k = 1, 2, …, m) with mean income ._& and population size ,_&. Regions are positioned in an order of increasing mean income. Vector 9_&

(._&/.) captures the differences in relative mean income between regions and :_& (,_&/,) denotes the region’s population share. In case population share would not be accounted for, the outcome of !_#'would have very little to tell about income inequality ! among the country’s citizens. This is basically because regions are of unequal population size and thus, a rapid increase in the income of a poor and small region will not have the same effect on total inequality as the same per capita increase in a poor and populous region (Milanovic, 2006).

3.2! Data!on!variables!

3.2.1! Gini!index!

The Gini index is a widely-used measure for the overall income inequality within a society.

Measurement problems and limited data availability are a common problem in the literature and it is hard to find series with consistent definitions across a large number of countries.

Most countries do not publish national Gini indices on a year-to-year base, making panel data research difficult to effectuate. The World Bank and the OECD provide reliable measures of

(19)

the Gini index which allow for easy comparison, unfortunately however these databases only have a very limited amount of country-year observations before the year 2005.

In order to cope with the problem of missing income inequality data, several other databases are considered which have alternative measurement methods, combine various databases or fill in data gaps in a systematic way. An example of a database using an alternative method is the “Estimated Household Inequality Data Set (EHII)” by the University of Texas Inequality Project (UTIP). The EHII contains estimated Gini coefficients for gross household income, derived from measures of cross-sector industrial pay inequality, which are computed using the between-groups component of Theil's T-statistic. The database provides reasonable estimates of the Gini coefficient for a large number of countries and years. A database combining various datasets to fill in the gaps is the UNU-WIDER World Income Inequality Database (WIID), which combines different data sources like household surveys and official Gini indices from national statistics. The most extensive database is the Standardized World Income Inequality Database (SWIID), offering the greatest coverage of Gini indices by using multiple imputations to fill in data gaps (Solt, 2014).

Deciding on which database to use actually entails making a trade-off between coverage and comparability (Solt, 2014). The data of the World Bank or the OECD is the best for comparability because of its uniformity in measurement, however the large amount of missing country-year observations make these databases not useful for panel data analysis. The EHII database is advantageous for both its coverage and comparability, however, its Gini estimates are based on gross household income which is a major drawback, since the effect of taxation policies is excluded in these coefficients. Gini estimates based on post-tax disposable income are needed, since this thesis looks into the effects of fiscal decentralization. Besides, disposable income is more associated with the real welfare of the population. The WIID and the SWIID both combine different data sources, resulting in reduced data comparability. The SWIID does contain Gini indices based on post-tax disposable household income and uses multiple imputations to fill in the missing country-year observations, which is useful for the time-span considered in this thesis where the Gini is often lacking. It does mean however that interpretation of the results needs to be considered with some caution, since multiple imputation reduces data validity.

(20)

3.2.2! Interregional!inequality!!

Examining the degree of interregional inequality is necessary for the computation of !_# – the between-group component of income inequality – for all countries included in the study. A commonly used measure for determining interregional disparities is regional per capita GDP.

For a sample of 22 OECD countries, data on regional GDP per capita is available for the years 1995 onwards. As an alternative to regional GDP per capita, it can be argued that disposable income per capita is actually the preferred measure. Commuting patterns tend to increase GDP per capita in urban regions where people are employed and decrease GDP per capita in regions where commuters reside, whereas earnings are included in the income of the region of origin (Piacentini, 2014). Interregional inequality might therefore be overestimated when computed based on regional GDP per capita, while a more realistic picture might be created when disposable income per capita is used. Unfortunately, the data availability on regional income per capita is very limited for the years prior to 2000 for a large number of NUTS2 regions, resulting in only 12 countries having sufficient observations for the years 1995-2000.

For robustness, however, both measures will be used and compared in this thesis.

The degree of territorial disaggregation considered in this thesis is NUTS2. This level of disaggregation is meaningful from a policy perspective, since the relatively large NUTS2 regions have considerable responsibilities concerning policy implementation. The choice is also dictated by practical considerations, since there is limited data available on NUTS3 regions. There is merely data on regional GDP per capita for a small sample of countries and, besides, commuting patterns are expected to play a significantly larger role at this level of disaggregation. As a consequence, the use of GDP per capita on such a small scale might provide a distorted picture of the level of interregional inequality. The above considerations result in a preference for the NUTS2 regional breakdown, according to which the following two country samples will be analyzed:

Sample Territorial Level Measure interregional inequality =_> Countries

1 NUTS 2 Regional GDP per capita 22

2 NUTS 2 Regional income per capita 12

For an overview of the countries within each sample, see table A1 and A2 in Appendix 1.

The data on regional GDP and income per capita are drawn from the ‘OECD Regional Database’, covering the years 1995-2005. In order to make comparisons over time and across countries, both regional GDP and income per capita are expressed at constant prices (base

(21)

year 2010) and converted into USD purchasing power parities (PPPs) to express all region’s GDP and income per capita into a common currency.

3.2.3! Government!decentralization!

Measuring government decentralization is not easy and common measures have often been criticized in the literature. However, lacking any reliable alternative, this thesis has chosen the subnational share in total government expenditure as a proxy for fiscal decentralization. It is explained as the expenditure of local and regional governments relative to total government spending and is the most utilized measure of decentralization available. The proxy has been criticized however for failing to capture the degree of expenditure autonomy of subnational authorities, for not making a distinction between tax and non-tax revenue sources, and for failing to identify the proportion of intergovernmental transfers that are conditional or discretionary (Rodríguez-Pose & Ezcurra, 2009). In an attempt to partly overcome this point of critique and provide a more comprehensive picture of the actual powers of subnational authorities, an indicator of political decentralization is introduced as an additional proxy.

Political decentralization is related to the actual power and authority of subnational governments to undertake the political functions of governance (Schneider, 2003). Perhaps the most comprehensive attempt to measure political decentralization and subnational autonomy is the regional authority index (RAI) proposed by Hooghe et al. (2008). According to this index, regional authority can be partitioned into two elements that respectively capture the degree of authority exercised by a subnational government over those who live in its territory (self-rule) and in the country as a whole (shared rule). Data on the regional authority index is taken from the database by Hooghe et al. (2008) for the years 1990-2005. Data on subnational expenditure share in total government expenditure is taken from the IMF Government and Finance Statistics and cover the years 1972-2000.

3.2.4! Control!variables!

A series of variables is included to control for other factors of importance. The last decades have been characterized by great advances in international trade, especially for the countries under study. The effects of greater competition and the gains of trade are unevenly spread across space and thus have an impact on regional inequalities (Ezcurra & Pascual, 2008).

Therefore, trade openness (OPEN) is included as a control variable, measured as the ratio between total trade (import and export) and GDP. Government size (GSIZE), measured as the total expenditure of general government as a percentage of GDP, is included to capture the importance of the public sector in the economy. It is a proxy for the governments’ capacity to

(22)

redistribute financial resources across regions, which is likely to affect the level and evolution of regional inequalities within any given country. The variable population (POP) is added to account for demographic composition and country size. The larger the size of a country, the greater the inherent heterogeneity and thus the possibility for subnational governments to implement other policies than in smaller, more homogenous countries (Rodriguez-Pose &

Ezcurra, 2009). Furthermore, new economic geography models have indicated that economic development is often accompanied by an uneven spatial development (Krugman, 1998). Per capita GDP (GDPpc) is therefore also included as a control variable to account for the level of economic development or wealth. The dummy variable LEFT represents the dominance of left-wing parties based on their seat share in parliament. When parliament is balanced or there is right-wing dominance, the dummy takes a value of 0. An overview of all variables with both description and source can be found in table A3 in Appendix 2.

3.3! Econometric!models!

The hypothesis of main interest in this thesis is hypothesis 3 (Government decentralization makes overall income inequality relatively more interregional). In order to provide a more complete picture, hypothesis 1 (Government decentralization leads to more interregional inequality) will also be tested, thereby following most of the existing literature on the topic of decentralization and income inequality. As previously mentioned, limited data availability results in the inability to test hypothesis 2 (Government decentralization leads to less intraregional inequality). The available data only allows for computing between-regional inequality, whereas no distinction can be made between intraregional inequality and the interaction term.

The interactions between decentralization and the share of inequality explained by interregional disparities (!_#/!) will be analyzed by performing a panel data regression analysis using the statistical software package StataSE 14. The variable representing the relative share of interregional inequality is log-transformed for interpretation purposes.

Missing data on fiscal decentralization causes the panel to be unbalanced. Both random and fixed effects are considered for running the model and fixed effects turn out to deliver better results. Performing a Hausman-test also indicates that the random effects assumption does not hold. That is, the random effects model is inconsistent, implying that the fixed effects model is the preferred choice. Fixed effects allow to control for omitted country-specific time- invariant factors that might affect the relationship between decentralization and income inequality. It must be noted, however, that both the dependent and independent variables are

(23)

changing slowly over time and that the time-frame of the panel is limited. Besides, a significant part of the variation is expected to be cross-country while the panel data approach will only pick up the time trend. The panel data approach will therefore be complemented by a cross-section analysis. An F-test does indicate that fixed effects are still preferred over a pooled OLS model. Moreover, a panel allows for analyzing the direction of causality, which is something of major interest in this thesis. Hence, the data will first be described based on a cross-section, whereas the main analysis will be focused on a panel data approach with fixed effects.

Altogether, the preceding considerations result in the following “Baseline model” to test the hypothesis of main interest (H3):

Baseline!model:!!

(!_#/!)_2? = ' @₂ + 'A1^CDE′_2? + 'A2^CGE′_2?+ 'A3^C!EGIJ′_2? + A4^CLGMN′_2?+ 'A5^C!PQRM′_2?+ A6GLG′_2?+ A7^CUMDV′_2?+ W_2?'

In which FD represents fiscal decentralization and PD political decentralization. The first hypothesis (H1) will also be tested according to the Baseline model, however, the dependent variable is in this case not the relative share of interregional inequality (!_#/!) but the absolute level of interregional inequality !_#:

!_#2? = ' @₂ + 'A1^CDE′_2?+ 'A2^CGE′_2?+ 'A3^C!EGIJ′_2? + A4^CLGMN′_2?+ 'A5^C!PQRM′_2?+ A6GLG′_2?+ A7^CUMDV′_2?+ W_2?'

The regression results for this last model, testing hypothesis 1, can be found in table A5 and A6 in Appendix 4. The following section will present and discuss the empirical results of the Baseline model for the main hypothesis of interest.

" !

(24)

4.!Empirical!results!

4.1! Baseline!model!

First of all, a cross-sectional analysis is performed in order to describe and gain understanding of the data at hand. The panel data approach has preference, as was indicated previously.

However, the dependent and independent variables are expected to change slowly over time and an important part of the variation is expected to be cross-country, while a panel with fixed effects will only pick up the time trend. Therefore, before analyzing the data using a panel data approach, a cross-sectional analysis is performed. Average values of all variables in the Baseline model are taken for the years 1995-2000. The results for both country sample 1, using regional GDP per capita as a measure for interregional inequality, and country sample 2, using regional income per capita as a measure, are presented in Table 1 below.

Table!1:!Cross-sectional!analysis!!

Dependent variable: : XYZ(!_#/!)

Country sample 1: GDP Country sample 2: Income

Variables: (1) (2) (3) (4)

FD -0.0121** -0.000293

(0.00464) (0.0165)

PD -0.000177 0.00325

(0.00694) (0.0120)

OPEN 0.00559** 0.00719*** -0.0107 -0.0103

(0.00202) (0.00236) (0.00565) (0.00574)

GSIZE 0.0388*** 0.0336*** 0.0268 0.0230

(0.00872) (0.0107) (0.0236) (0.0259)

logPOP 0.0182 0.0449 -0.0476 -0.0437

(0.0524) (0.0712) (0.203) (0.202)

logGDPpc -0.182 -0.410** -0.778 -0.892

(0.160) (0.190) (0.782) (0.730)

LEFT -0.264 -0.0116 -0.0264 0.0228

(0.201) (0.220) (0.736) (0.428)

Constant -1.148 0.372 6.686 7.835

(1.520) (2.332) (7.030) (6.774)

Observations 21 21 12 12

R-squared 0.829 0.746 0.800 0.802

Standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

The results from the cross-section are mainly in contradiction with what was expected after reviewing the literature. Fiscal decentralization is found to be negatively correlated with the

(25)

relative share of regional inequality at a five percent significance level, whereas political decentralization does not seem to have any influence. Interpreting these results means that an increase in fiscal decentralization has a negative effect on the share of interregional inequality, ceteris paribus, which is in line with the findings by Ezcurra & Pascual (2008). It must be noted, however, that in the pooled OLS model country-fixed effects are not controlled for. As a consequence, there may be some form of omitted-variable bias in which case the standard errors cannot be trusted. The β-coefficient could then also be zero or even positive, since the confidence interval may widen. Besides, this thesis is mostly interested in whether decentralization alters the share of interregional inequality within a country over time and in the direction of causality, while the results of a cross-section only provide a snapshot of the situation.

Altogether, the considerations above provide more than enough reason to perform a panel data analysis with fixed effects. This approach is able to pick up the time trend and allows for analyzing the direction of causality. Summary statistics of the main variables of interest are provided in table A4 in Appendix 3 and the regression results are outlined and commented on below. First, country sample 1 will be covered, using regional GDP per capita to calculate the between-regional inequality component. Subsequently, the sample based on regional income per capita will be discussed. In order to allow for comparison of both measures of interregional inequality and the resulting outcomes, country sample 2 is also analyzed on the basis of regional GDP per capita. For an overview, please refer to Table 2 below.

Table!2:!Overview!of!country!samples!and!measures!of!interregional!inequality!for!the!Baseline!model!!

Sample Territorial Level Measure interregional inequality =_> Countries

1 NUTS 2 Regional GDP per capita 22

2 NUTS 2 Regional income per capita 12

3 NUTS 2 Regional GDP per capita 12

4.1.1! Sample!1:!Regional!GDP!per!capita!

The results of the fixed effects model for country sample 1 are displayed in Table 3. As opposed to the results from the cross-section, both fiscal and political decentralization have a significant positive effect on the relative importance of between-regional inequality in total inequality (!_#/!)'. The effect remains positively significant after the inclusion of several control variables, which confirms the robustness of the coefficients and shows that the effects

(26)

of fiscal and political decentralization on interregional inequality are not spurious correlations resulting from the exclusion of relevant variables. From the results it seems that decentralization initiatives lead to relatively more interregional inequality, ceteris paribus. An additional robustness check was performed, leaving out the large federations of the United States, Canada and Australia on grounds of form of government and the magnitude of NUTS2 regions. The results are presented in table A7 of Appendix 5 and provide further evidence for the previously found results.

Concerning the control variables included in the model, the variable ‘OPEN’ exhibits the expected sign and is significant at the one percent level across all specifications (except for model 10). It seems that the effect of greater competition and the gains of trade are unevenly spread across regions and thus have an impact on the relative share of regional inequalities in total inequality. The control variables ‘GSIZE’ and ‘LEFT’ do not seem to have any influence on the outcome, whereas ‘logPOP’ and ‘logGDPpc’ show significance at the five percent level in the last model. Country-wide advances in economic development or wealth seem to be associated with relatively greater spatial inequality, ceteris paribus. This can be related to the new economic geography model in that not all regions evenly benefit from economic development, in which clustering forces generate an uneven distribution of economic activity and income across space. The sign of control variable ‘logPOP’ is contrary to what was expected and shows a negative correlation with the relative share of interregional inequality.

It must be noted, however, that both population size and GDP per capita are not significant in the specifications including fiscal decentralization. Therefore, these coefficients lack robustness and the interpretation becomes more questionable. "

4.1.2! Sample!2:!Regional!income!per!capita!

The results of the fixed effects model for country sample 2 are presented in Table 4. Due to limited data availability on regional income per capita and fiscal decentralization, the amount of observations in the first five columns is very limited. The effect of both fiscal and political decentralization on (!_#/!) is not significant, with the exception of the sixth model in which a significant positive effect for political decentralization is found. Nevertheless, when control variables are added the coefficient loses its significance. From these results it seems there is no real relationship between government decentralization and the relative importance of between-regional inequality in explaining total inequality. However, it is troublesome to draw conclusions with the limited amount of observations at hand. In order to find out whether the

Income Inequality Revisited:

!

Government Decentralization and

Income Inequality Revisited:

An Inequality Decomposition Analysis

MSc Thesis International Economics & Business

! !

! !

Contents

1.!Introduction

" !

2.!Literature!review!

3.!Methodology!and!data!

" !

4.!Empirical!results!