REGIONAL INEQUALITIES, FISCAL DECENTRALIZATION AND GOVERNMENT QUALITY: EMPIRICAL EVIDENCE FROM SIMULTANEOUS EQUATIONS

Transcription

1 GEN - Governance and Economics research Network GEN Working Paper A webs.uvigo.es/infogen/wp February 2015 REGIONAL INEQUALITIES, FISCAL DECENTRALIZATION AND GOVERNMENT QUALITY: EMPIRICAL EVIDENCE FROM SIMULTANEOUS EQUATIONS Andreas P. Kyriacou* Leonel Muinelo-Gallo** Oriol Roca-Sagalés*** * Universitat de Girona and GEN ** Universidad de la República ***Universitat Autònoma de Barcelona and GEN

2 Regional Inequalities, Fiscal Decentralization and Government Quality: Empirical Evidence from Simultaneous Equations Andreas P. Kyriacou 1, Leonel Muinelo-Gallo 2, Oriol Roca-Sagalés 3 1 Departament d Economia, Universitat de Girona, Campus de Montilivi, 17071, Girona, Spain ( andreas.kyriacou@udg.edu) 2 Instituto de Economía de la Universidad de la República, Joaquín Requena 1375, Código Postal 11200, Montevideo, Uruguay ( lmuinelo@iecon.ccee.edu.uy) 3 Departament d Economia Aplicada, Universitat Autònoma de Barcelona, Edifici B, Bellaterra 08193, Barcelona, Spain ( oriol.roca@uab.cat) Abstract There are theoretical arguments supporting the view that regional income inequalities, the degree of fiscal decentralization and the quality of government are simultaneously determined. To the extent that previous work has considered feedback effects between them, it has done so through instrumental variable estimates based on instruments whose validity can be questioned. Moreover, most existing work has estimated the relationship between any two of these variables in the absence of the third. Our empirical evidence, drawn from a sample of 23 OECD countries and based on a simultaneous equation model which accounts for the joint determination of these three variables, suggests that a process of fiscal decentralization, accompanied by measures to improve the quality of government, would be an effective strategy for reducing regional inequalities. Key words: regional inequalities, fiscal decentralization, governance, reverse causality, panel data, simultaneous equations models JEL codes: C33, D73, H71, H73 Acknowledgements The authors would like to acknowledge financial assistance from the Instituto de Estudios Fiscales (Madrid). 1

3 INTRODUCTION In recent years, a good deal of theoretical and empirical work has been devoted towards a better understanding of the interplay between fiscal decentralization, regional inequalities and government quality. What emerges from this work is the appreciation that anyone of these variables is partly determined by the other two or, in other words that these three variables are simultaneously determined. Thus, the empirical evidence to date reviewed in detail in section 2 of this article supports the idea that fiscal decentralization, by empowering poorer regions and fomenting pro-growth policies, reduces regional inequalities. Alternatively, several researchers have reported evidence which indicates that the degree of fiscal decentralization may, itself, depend on the regional inequalities in place because of the conflicting interests of relatively wealthy and relatively poor regions. Moreover, scholars have found that fiscal decentralization has the potential to improve governance across countries, perhaps because of the disciplining effect of inter-jurisdictional competition. It has also been suggested, but not empirically pursued, that corrupt central government officials may resist fiscal decentralization as this would attenuate their possibility to extract rents. Finally, empirical evidence has also emerged indicating that regional inequalities may lead to redistributive conflicts over the territorial distribution of resources, and this may crowd-out efforts towards public sector reforms. Alternatively, it has also been argued that government quality may contribute towards lower regional income differences because more efficient governments have a stronger capacity to address territorial inequalities. Most of the existing empirical studies consider the possibility of simultaneous causality confounding OLS estimates of the hypothesized relationships and have employed 2

4 instrumental variable techniques like 2SLS in an effort to account for this. While cross country work has employed a variety of time constant instruments to deal with the issue, most panel analysis has resorted to lagged values of the explanatory variables as instruments, given the difficulty of finding suitable time variable instruments for the potentially endogenous key variables. Unfortunately, the persistence over time of fiscal decentralization and regional inequalities means that there are serious doubts about the extent to which lagged values represent exogenous instruments. While the same problem does not emerge in cross-section work, it can be argued that some of the instruments employed do not satisfy the exclusion restriction and moreover, there are concerns about sample size, given the limited availability of cross-country data for the variables subject to the analysis. In response to these shortcomings, we propose to reconsider the complex relationship between regional inequalities, fiscal decentralization and the quality of government based on a panel of twenty-three OECD countries over the period 1984 to 2005 and by way of estimation techniques which deal head on with the potential interdependencies. Specifically, we consider a simultaneous equation model which accounts for the joint determination of the relevant endogenous variables. This approach allows us to obtain more appropriate estimates of the relevant parameters and, ultimately, helps us identify the causal relationships between the three variables that we focus on. Based on our analysis, we find strong evidence of a positive impact of fiscal decentralization on governance at the same time as we find that poorly governed countries will tend to be more centralized. Moreover, we find robust evidence for the claim that worse governed countries find it more difficult to reduce regional inequalities as well as the idea that countries with large inequalities will tend to be less well governed. Finally, our findings support the expectation that fiscal decentralization will 3

5 tend to reduce regional inequalities but not the claim that regional income differences have an incidence of the extent of fiscal decentralization. Our analysis also indicates that previous work examining the determinants of regional inequalities and fiscal decentralization may suffer a degree of omitted variable bias because it fails to control for the confounding influence of government quality. The article is structured as follows. In the next section, we review the theoretical and empirical findings which have illuminated the complex relationship between the three variables of interest. Having done so, in the following two sections, we describe the data and empirical methodology that we employ in the article. After that we report and discuss our empirical findings before concluding. REVIEW OF THE LITERATURE: THE CAUSAL CHANNELS In an effort to elucidate the causal channels linking the degree of fiscal decentralization, the size of regional inequalities and the quality of government consider figure 1 which we expand upon in a series of sub-sections. <<Figure 1 about here>> Fiscal decentralization and regional inequalities From a theoretical perspective, fiscal decentralization may either narrow or widen regional inequalities (arrow 1). It can reduce regional inequalities because it implies better informed and more specific policies (Oates 1972, 1993) or because it can set the stage for inter-jurisdictional competition for fiscal resources something which should restrain inefficient local government (Brennan and Buchanan 1980; Weingast 1995; McKinnon 1997; Qian and Weingast 1997). On the other hand, regional inequalities may widen to the extent that wealthier regions with a stronger fiscal capacity can 4

6 outcompete poorer ones (Prud homme 1995; Keen and Marchand 1997; Oates 1999) and because fiscal decentralization implies a reduction in the redistributive capacity of central government (Prud homme 1995; Rodríguez-Pose and Gill 2004) which moreover may be swayed by wealthier regions opposed to redistribution (Rodríguez- Pose and Ezcurra 2010). But differences in incomes across regions can also have a bearing on the degree of fiscal decentralization (arrow 2). Lessmann (2012) has explained that in the face of growing regional inequalities, the central government may move to centralize or decentralize fiscal resources depending on whether it thinks that centralization or decentralization will contribute towards fiscal convergence. Beramendi (2007) describes how the degree of fiscal decentralization may emerge as a result of the conflicting interests of poor and wealthy regions: the former prefer more centralization since this increases the resources available for territorial redistribution while the latter oppose centralization for the same reason. The empirical evidence is generally supportive of the argument that fiscal decentralization will tend to reduce inequalities (Costa-Font, 2004; Gil et al. 2004; Ezcurra and Pascual, 2008; Lessmann 2009) although this result seems to be contingent on the level of economic development (Rodríguez-Pose and Ezcurra 2010; Lessmann 2012) and the quality of government (Kyriacou et al. 2014). Insofar as the effect of inequalities on the degree of fiscal decentralization is concerned, Stegarescu (2009) and Saachi and Salotti (2014a, b) report evidence indicating that greater income differences across regions are associated with less fiscal decentralization while Beramendi (2007) finds the opposite or that higher inequalities are related to more decentralization. 5

7 To the extent that previous empirical work has dealt with a two way relationship between fiscal decentralization and inequalities, it has mostly resorted to lagged values of the fiscal decentralization indicators either directly as explanatory variables (Ezcurra and Pascual 2008; Saachi and Salotti 2014a, b) or indirectly as instrumental variables (Lessmann 2009; Rodríguez-Pose and Ezcurra 2010; Kyriacou et al. 2014). However, and as pointed out by Lessmann (2012), the relative persistence of the key variables over time compromises this empirical strategy. Instead this author exploits the cross-section variation of a sample of both developed and developing countries and proposes a variety of instruments for fiscal decentralization including the share of the population using a European language, the extent of democracy, legal origin and the prevalence of Protestantism, Catholicism or Islam. The validity of these instruments is, itself, compromised however most obviously because several authors have explained how many of these variables are in fact associated with government quality (see for example, Hall and Jones 1999 and La Porta et al 1999), something for which Lessmann (2012) does not control for in his regressions. Given the possibility that government quality can directly affect regional inequalities (as more fully explained below), the extent to which the exclusion restriction is satisfied in Lessmann s (2012) estimates is not clear. 1 Fiscal decentralization and government quality Consider next the relationship between fiscal decentralization and the quality of government. Government quality has been broadly defined to include the extent to which the state protects property rights, is not plagued by corruption, and provides public goods in a cost-effective way (La Porta et al. 1999). Several theoretical arguments have been advanced to explain why fiscal decentralization may either reduce or increase government quality (arrow 3). Fiscal decentralization can improve 6

8 governance for the same reasons it can promote regional convergence: because it empowers better informed local officials and because the resultant interjurisdictional competition can put a break on bad government. Alternatively, decentralization may lead to fiscal or regulatory `overgrazing' in the context of multilevel government (Shleifer and Vishny 1993), can lead to the capture of local officials by economic interest groups (Prud'homme 1995; Bardhan and Mookherjee 2000; Bardhan 2002) and can mean that government in some uncompetitive regions either compete for capital by helping agents to cheat the central government in ways that reduce the latter's capacity to enforce regulations and collect taxes (Cai and Treisman 2004) or give up on business friendly policies and dedicate themselves to predation instead (Rodden and Rose- Ackerman 1997; Cai and Treisman 2005). Both cross section and panel based evidence has been provided to explain the impact of fiscal decentralization on government quality. Employing a cross section of countries, Fisman and Gatti (2002) find that more fiscally decentralized countries tend to be less corrupt and this result has been replicated by scholars based on both cross section and panel evidence (Enikolopov and Zhuravskaya 2007; Fan et al. 2009; Kyriacou and Roca-Sagalés 2011a and 2011b). All these authors recognize the possibility that officials in less well governed countries may resist fiscal decentralization so as to maintain their access to public resources (arrow 4). And all, except Fan et al (2009), attempt to deal with this issue. Fisman and Gatti (2002) employ legal origin as an instrument for fiscal decentralization to deal with reverse causality in their estimates, a strategy which they recognize is weakened by the fact that legal origins tends to affect government quality through other channels unaccounted for (as explained by La Porta et al. 1999). Enikolopov and Zhuravskaya (2007) and Kyriacou and Roca-Sagalés (2011a) employ country area as an instrument in their cross-section regressions but do not 7

9 contemplate the possibility that country area may affect government quality through regional inequalities, a factor they fail to control for. In their panel estimates, lagged values of fiscal decentralization are used as instruments something which raises the previously referred to limitations in view of the persistence of fiscal decentralization over time. Regional inequalities and government quality Consider finally the relationship between government quality and regional inequalities. With respect to the impact of regional inequalities or government quality (arrow 5), Kyriacou (2012) shows that regional diversity, which may be described by either territorially segregated ethnic groups (Alesina and Zhuravskaya 2011) or regional inequalities, may reduce government quality because it undermines generalized trust. To deal with reverse causality this author employs lagged values of regional inequalities in cross-section instrumental variable regressions. Kyriacou and Roca-Sagalés (2014) argue, and provide empirical evidence to the effect that regional income differences may lead to redistributive conflicts over the territorial distribution of resources something which will tend to crowd out government efforts towards public administration reform and improving government efficiency overall. The relative stability of regional inequalities over the ten year period analyzed, leads them to propose geographic area as an instrument for the former all the while providing evidence of the strength of the instrument and the satisfaction of the exclusion restriction. On the other hand, the use of geographic area as an instrument is less justified in panels which extend over a longer time period which would have the obvious advantage of increasing the number of observations. 8

10 Reversing the direction of causality, it has also been suggested that government quality may influence the evolution of regional inequalities (arrow 6). Ezcurra and Rodríguez- Pose (2014) argue that both the effectiveness of regional development strategies and the capacity of regions to attract foreign direct investment (FDI) are likely to be negatively affected by poor governance. Their empirical evidence documents the positive impact of government quality on regional convergence. In another article, Kyriacou et al. (2014) consider both the direct impact of government quality on regional inequalities and the mediating effect of the former on the impact of fiscal decentralization on regional inequalities. They find that better governance tends to reduce regional inequalities directly, while the mediating effect of governance indicates that fiscal decentralization increases regional inequalities in low government quality settings and reduces them in high government quality ones. 2 To control for reverse causality in their regressions, Kyriacou et al. (2014) employ lagged values of government quality as instruments and strive to reinforce the exogeneity of the instruments by taking lags of several lengths. Ezcurra and Rodríguez-Pose (2014) take a different approach and instrument government quality with an index of press freedom which they show is correlated with government quality while, they argue, it is exogenous to regional inequalities. Omitted variable bias The overwhelming majority of the empirical work that has centered on the relationship among the three key variables we describe in this article has tended to ignore the confounding influence of one of the three. As can be appreciated in table A1 in the Appendix, almost all previous studies calibrating the impact of fiscal decentralization on regional inequalities have ignored the role of government quality. Similarly, those studies which have focused on the impact of regional inequalities on fiscal decentralization do not consider the impact of governance in their estimates. 3 Moreover, 9

11 that empirical work which has considered the impact of fiscal decentralization on government quality has failed to control for the confounding influence of regional inequalities. Consequently, previous work focusing on the impact of one of the key variables on another may suffer from omitted variable bias. In sum, a case can be made to the effect that fiscal decentralization, regional inequalities and government quality are simultaneously determined. And previous studies have centered on the impact of one of these variables on another (without controlling for the influence of the third) while attempting, through instrumental variables methods, to control for the acknowledged feedback effects. Arguably therefore, there is a need to empirically reconsider the relationship between these three variables through a technique which deals head-on with the full set of possible feed-back effects. The rest of the article will be devoted to just this purpose. DATA In this section we review the indicators employed to measure fiscal decentralization, regional inequalities and government quality. Selecting countries and time period on the basis of the availability, frequency and quality of the data available to measure these variables gives rise to an unbalanced panel of twenty-three OECD countries over the period 1984 to 2005 (see, respectively, tables A2 and A3 in the Appendix, for the list of countries included and for sources and definitions of all the variables reported in this article). Turning first to our chosen measure of regional inequalities within countries, we use the population-weighted coefficient of variation (PW-CV), a standard measure that is independent of the scale, population size and number of regions considered, and moreover satisfies the Pigou-Dalton principle (Cowell 1995). 4 The PW-CV is the 10

12 population-weighted standard deviation of Gross Domestic Product (GDP) per capita level within a country, divided by the country s GDP per capita level. Specifically, PW-CV = 1 y n i 1 p ( y y ) i 1/ 2 2 i where y is the average country GDP per capita, y i and p i are, respectively, the GDP per capita and population share of region i and n is the number of regions. All European regions are defined at the NUTS 2 level, except for Belgium, Germany, the Netherlands and the UK which are so at the NUTS 1 level. Australian regions are defined as states and territories; USA and Mexican regions are states and in Canada they are provinces and territories. The PW-CV reflects inequalities between a country`s regions, taking into consideration their relative population weights. It ranges between 0 (equality) and 1 (maximum inequalities). According to this indicator, regional inequalities are higher in Belgium, Slovakia and, especially, Mexico and lowest in the Netherlands and Australia. As can be appreciated from the summary statistics in table A4 in the Appendix, the limited within-country variation (standard deviation of ) is eclipsed by crosscountry differences in regional inequalities (standard deviation of ). In order to measure fiscal decentralization, the inter-jurisdictional competition literature emphasizes the importance of tying local expenditures to local revenues for the proper functioning of competition since vertical transfers may create incentives for local officials to ignore competitive pressures for better management (Qian and Weingast 1997; Qian and Rolands 1998; Oates 1999; Zhuravskaya 2000; Jin et al. 2005). With this in mind, we employ a measure of fiscal decentralization based on the OECD General Government Accounts and defined as subnational revenue share of consolidated general government revenue, after subtracting from state and local revenues, grants from other governments (Gemmel et al. 2013). 11

13 From the summary statistics in table A4 we can observe that the mean value of decentralization is This mean belies a considerable cross-sectional variability in our sample as attested by a between standard deviation of On average, in the period 1984 to 2005, revenue decentralization is highest in the United States, Switzerland and Canada and lowest in Portugal, Slovakia and Greece. As was the case for our regional inequalities measure, the degree of within country variation is, alternatively, relatively limited pointing to the relative stability of this indicator over time (within standard deviation of ). As previously explained, this compromises the effectiveness of employing lagged values of this indicator as instrumental variables in an effort to deal with the possibility of reverse causality. To measure government quality we rely on the International Country Risk Guide (ICRG) database as developed by the Political Risk Services Group to assess the political, economic and financial risks across countries. The ICRG data have been widely used in work exploring the determinants of government quality and the impact of the latter on economic development. 5 It is based on the perceptions of a worldwide network of experts on a range of governance dimensions. These perceptions extend across state institutions and jurisdictions whether at the local, regional or central levels and as such they capture general or country-wide perceptions of government quality. The governance dimensions include corruption, law and order and bureaucratic quality. Corruption refers to the demand for bribes by political and administrative bodies as well as patronage, nepotism, job reservation, favors-for-favors, etc. The variable law and order assesses the strength and impartiality of the legal system as well as popular observance of the law. Bureaucratic quality refers to the strength, expertise, autonomy and recruitment and training mechanism of the civil service. 12

14 Because the first two dimensions are measured on a scale from 0 to 6 while the last one does so from 0 to 4, we normalize each dimension between 0 and 1. The quality of government indicator used here is obtained by summing up these normalized values and thus, ranges from 0 to 3 where a higher number implies higher government quality. Using an aggregate indicator, accounts for the possibility that each individual index may suffer a degree of measurement error (Mauro 1995). In our sample, government quality is low in Mexico, Southern and Eastern Europe and high in the remaining countries. While the gap between the within-country and between-country variations is not as large as in the case of our regional inequalities and fiscal decentralization measures, the variability of government quality across countries and over time is still relatively limited (within and between standard deviations of and respectively compared to a mean value of ). <<Figures 2 to 4 about here>> Figures 2 to 4, plot our indicators of fiscal (revenue) decentralization, regional inequalities and government quality against each-other using average values for each country over the whole sample period ( ). A look at the figures indicates that greater regional inequalities are associated with lower government quality and less fiscal decentralization while more fiscally decentralized states tend to have higher government quality. This is very much in line with the empirical work reviewed in the previous section, but of course, it says nothing about the direction of causality nor does it take into account the confounding influence of omitted variables. In the next section we turn to our empirical methodology which is designed to deal with these important issues. 13

15 EMPIRICAL METHOD The empirical strategy Given the potential interdependence between the key variables, it is necessary to apply an empirical method that considers their mutual influence in order to avoid severe specification errors. Consequently we propose to estimate a full system for the joint determination of the relevant endogenous variables. Specifically, we propose to estimate a simultaneous equation model (SEM) which consists of a series of three equations describing regional inequalities, fiscal decentralization and government quality. The proposed model takes the following form: RI it 0 1FD1, it 2GQ1, it 3X1, it u1, it (1) FD it 0 1RI 2, it 2GQ2, it 3Y 2, it u2, it (2) GQ it 0 1RI 3, it 2FD3, it 3Z3, it u3, it (3) where RI it, FD it, and GQ it refer to our dependent variables of regional inequalities, fiscal decentralization and government quality, respectively. In turn, i 1,..., N refer to countries, t 1,..., T to years, ( 0, 0, 0 ) are constant terms, ( 1, 1, 1 ) and ( 2, 2, 2 ) are the coefficients associated to the endogenous variables, ( 3, 3, 3 ) are the coefficient vectors associated to the explanatory variables X 1,it, Y 2,it and Z 3,it, respectively while u n, it (n = 1, 2, 3) refers to the error terms for each of the three equations of the SEM model. This model thus deals explicitly with the a-priori interdependence between government quality, fiscal decentralization and regional inequalities and, as such, allows us to obtain 14

16 more appropriate estimates of the relevant parameters. Moreover, we assume that the disturbances from the different regression equations at a given point in time are correlated, because of common unobservable factors. In order to exploit efficiency gains from the correlation of error terms across equations and the mutual influence of the relevant endogenous variables, the full set of three equations of the SEM will be jointly estimated using system instrumental variables methods (SIV). 6 Because the modern approach to SIV methods is based on the principle of the generalized method of moments (GMM) 7, the SEM model is estimated by way of two GMM estimators. On the one hand, we consider a (two-step) three-stage least squares GMM estimator (3SLS- GMM) with an unadjusted weight matrix 8, and secondly, we will use a GMM heteroskedastic and autocorrelation consistent estimator (GMM-HAC) that, as the name implies, accounts for any type heteroskedasticity and contemporaneous correlation of the errors across equations. 9 It is important to mention that an important feature of GMM estimation is that by selecting different weight matrices, we can obtain estimators that can tolerate heteroskedasticity, autocorrelation, and other features of the error terms. Compared to a single-equation approach, these system estimation methods are able to spell out feed-back simultaneities among the key endogenous variables, and as a result obtain consistent and more efficient estimates. Moreover, these two estimation methods are appropriate when some of the control variables (in vectors X, Y and Z) are endogenous (see Wooldridge 2010). In this case, and in the presence of significant within-country variability, we employ lags of the endogenous control variables as instruments (see, most notably Barro 2000). 10 In every equation, the number of exclusions is sufficient for the order condition of the identification to be satisfied. More specifically, each equation of the system is identified 15

17 given that at least two control variables actually appear in the other two equations but not in the reference equation (see Wooldridge 2010, Theorem 9.1). In turn, the rank condition of every equation of the system (1) to (3) can be safely assumed to hold in a model of this size. 11 The selection of the control variables included in each regression of the system is guided by the need to reduce omitted variable bias. Thus, there is the a- priori expectation that the control variables may be related to the three key variables we scrutinize in this article. The control variables We always control for GDP per capita, country openness to international trade and the extent to which a country is characterized by the presence of ethnically heterogeneous and territorially segregated ethnic groups. Wealthier countries can apply their greater resources to reduce regional inequalities (Lessmann 2009). Moreover, they may be more decentralized if decentralization is desirable but expensive (Wheare 1951; Panizza, 1999) or less decentralized if rising income leads to a stronger demand for centrally provided redistributive, social and infrastructural policies (Wallis and Oates 1988; Letelier 2005). Lastly, richer countries may enjoy better governance because of income s positive effect on education, literacy and depersonalized relationships (Treisman 2000). Insofar as country openness is concerned, increasing trade or economic integration may contribute towards widening regional inequalities (Giannetti 2002; Petrakos et al. 2005; Rodríguez-Pose 2012). Relatedly, economic integration may increase the probability of asymmetric shocks across regions, pointing towards the suitability of centralized redistribution to account for these, thus reducing the scope of fiscal decentralization (Stegarescu 2009). Alternatively, more open countries because they face stronger competitive pressures may have better governments (Ades and Di Tella 1999; Ezcurra 2012). 16

18 We control for geographically segregated ethnic groups since the inherent heterogeneity may make it more difficult to agree on the inter-territorial distribution of resources thereby undermining the capacity to promote regional convergence (Kyriacou and Roca-Sagalés 2012; Kyriacou et al. 2014). Moreover, geographically concentrated ethnic groups may call for greater autonomy thereby increasing the degree of fiscal decentralization (for example, Panizza 1999 and Alesina and Spolaore 2003), but it could also be the case that in the presence of such groups, central government may resist decentralization in the fear of precipitating the break-up of the state (Arzaghi and Henderson 2005; Tanzi 2000). Finally, ethnic segregation may be inimical to government quality because it reduces generalized trust, it increases the tendency of citizens to vote for political candidates on the basis of ethnicity rather than competency, and it facilitates secessionism leading central government to employ resources to deal with such movements resources which could have been employed to improve governance (Alesina and Zhuravskya 2011). Our sample includes several countries which were previously members of the Soviet Union. In equations 1 and 2 we control for the transition from a socialist to a capitalist economic system. It has been argued that the transition process may have increased regional inequalities because it seems to have benefited capital cities and major urban areas while harming agricultural and manufacturing regions (Petrakos 2001). Moreover, transition implied moving from highly centralized planned economy to a decentralized market economy (Bardhan 2002). In equations 1 and 3 we control the size of the public sector since countries with larger state sectors may have a greater capacity to address regional inequalities (Rodríguez- Pose and Ezcurra 2010) but also because bigger public sectors increase the scope for rent-seeking thus worsening government quality (Tanzi 2008) or can be associated with 17

19 better working conditions for public servants thus potential improving governance (Van Rijckeghem and Weder 2001). Our measure of public sector size is made up of two important components of public spending namely, current public spending and public investment (both as a share of GDP). In equation 1 we employ these sub-components separately while in equation 3 we add them together and employ an aggregate measure. When explaining regional inequalities (equation 1) we moreover control for human capital, private investment and the number of regions. Previous work has identified the growth promoting effects of human capital (Mankiw et al. 1992; Barro 2001) and private investment (Voitchovsky 2005 and Lin et al. 2009). In the absence of indicators capturing regional differences in human capital endowments and private investment, we turn to an indicator of the former defined as the average years of schooling of the population aged 25 and over from Barro and Lee (2001) and we measure private investment as total investment minus public investment as a percentage of GDP. We also control for the number of regions in a country since the territorial level used to calculate our measure of regional inequalities is not always the same (see Lessmann 2012). In our fiscal decentralization equation (equation 2) we, moreover, control for the geographic size of countries and whether or not a country can be classified as federal in nature because of the expectation that larger and/or federal countries, be more fiscally decentralized (see, for example, Panizza 1999 and Rodden 2004). Finally, in the equation explaining government quality (equation 3) we control for the degree of inter-personal income inequality since it has been argued that in unequal societies, relatively poor individuals may see petty corruption as a legitimate way of obtaining basic public services while relatively wealthy people may buy out public 18

20 sector officials to maintain their privileges (You and Khagram 2005; Glaeser et al. 2003). We also control for the percentage of protestants in the population and a range of legal traditions by way of an indicator which takes the value of 1 if a country has had a Soviet legal system in the past, 2 if it has had a civil law system (French, German or Scandinavian) and 3 if it has a tradition of British common law (see also Seldayo et al. 2010). The Protestant reformation may have put a check on the development of an overbearing and interventionist state while a country s legal tradition gives some indication of the power of the state relative to property owners, with socialist law systems favoring the former and common law system favoring the latter to the benefit of good economic government (La Porta et al. 1999). EMPIRICAL RESULTS We report our empirical findings in tables 1 and 2. The former employs annual data while the latter uses four year averages. The large difference in the number of observations in each case means that table 2 can be seen as a robustness check of the results based on annual data. Moreover, using four year averages helps us to focus on the structural relationships since it tends to reduce short-run fluctuations due to the business cycle (see also, Lessmann 2009 and 2012; Kyriacou et al. 2014). <<Tables 1 and 2 about here>> To facilitate the interpretation of the results we will discuss them in the context of figure 1. To avoid repetition because of space constraints, we will refrain here from citing previous theoretical and empirical work in relation to all the relationships captured by this figure and direct the reader to our extensive discussion of this work and corresponding citations above. Beginning with the relationship between fiscal decentralization and regional inequalities, our estimates support the idea that 19

21 decentralization tends to reduce inequalities (arrow 1) but not the argument that greater regional inequalities have an incidence on the degree of fiscal decentralization (arrow 2). Depending on whether we employ annual data or four year averages, we find either a negative or positive impact of inequalities on fiscal decentralization. More importantly, regardless of the type of data used and the estimation method, the estimated impact of regional inequalities on fiscal decentralization is never statistically significant. Turning next to the relationship between fiscal decentralization and government quality, we find robust evidence that these two dimensions are simultaneously determined. In particular, our estimates support the expectation that fiscal decentralization will tend to improve the quality of government (arrow 3), a result which has received wide support by previous empirical work. But we also find evidence for the expectation that countries with better quality government tend to be more fiscally decentralized (arrow 4) thus supporting the, until now, untested claim that corrupt and or inefficient officials may resist the fiscal decentralization of the country since this would reduce their capacity to access public resources. The two-way relationship between the degree of fiscal decentralization and the quality of government is robust to the type of data (annual data versus four year averages) and estimation method employed (3SLS-GMM or GMM- HAC). Figure 1 also captures the possibility of a two-way relationship between government quality and regional inequalities: the presence of regional inequalities may undermine or crowd-out efforts towards better governance (arrow 5) at the same time as government quality may help reduce inequalities perhaps because it implies more effective regional development strategies and a stronger capacity of regions to attract FDI (arrow 6). Our empirical results provide considerable support for the existence of both these causal relationships. We always find regional inequalities to have a negative and statistically 20

22 significant impact on the quality of government at the same time as we corroborate the expectation that countries with better governments will tend to have smaller income differences across regions. Further, we explore the possibility that previous work reporting evidence of a robust relationship between two of the key variables may be biased because of the absence of the third one. In particular, we drop our measure of government quality from those regressions explaining regional inequalities and fiscal decentralization (see table 3). When we do this, the estimated economic impact and statistical significance of both fiscal decentralization on regional inequalities and, especially, that of inequalities on fiscal decentralization improves markedly. We take this as an indication that previous empirical work examining the determinants of regional inequalities and fiscal decentralization suffers from a degree of omitted variable bias because it fails to control for the quality of government. <<Table 3 about here>> To further consider the robustness of our results we re-estimate our models using alternative measures of fiscal decentralization and regional inequalities. In particular we measure fiscal decentralization by way of expenditure decentralization defined as subnational expenditure share of consolidated general government expenditure, after subtracting grants from other governments from state and local expenditures (again from Gemmel et al. 2013). We also consider alternative indicators of regional inequalities namely, a population-weighted standard deviation of the logarithm of regional GDP per capita, which is widely used in the convergence literature to capture sigma convergence (Barro and Sala-i-Martin 1995), a measure proposed by Theil (1967) that is typically employed in the inequality literature and a coefficient of variation of the 21

23 regional GDP per capita indicator that does not take into account the distribution of the population across all regions. Regardless of the indicators employed our results are maintained. 12 Before concluding this section a word is in order about the estimated impact of our control variables. Reassuringly, our findings in this respect are generally consistent with previous work. Thus, GDP per capita is negatively associated with regional inequalities and positively associated with both fiscal decentralization and government quality. Moreover, we find that more open countries have higher regional inequalities and better quality governments. Third, our evidence indicates that ethnically segregated countries have higher regional inequalities and are more fiscally centralized. The former is suggestive of possible difficulties on agreeing on territorial redistribution in heterogeneous societies while the latter is consistent with the idea that central governments may oppose decentralization in the fear of fomenting secessionism. This result is robust to additionally controlling for the level of democracy in each country based on the assumption that more democratic countries are likely to be more responsive to calls for greater decentralization coming from ethnically distinct regions. In relation to the controls which only appear in any one of the regressions, we find that public investment reduces regional inequalities while these inequalities are higher in countries with more regions and with more human capital, although this latter result must be interpreted with care since our measure of human capital is a rather imperfect proxy. Moreover, we find larger or federal countries to be more fiscally decentralized. Finally, our results indicate that Protestant countries and ones with a tradition of British common law (as opposed to a soviet legal tradition) have better government quality whereas interpersonal income inequalities tend to be inimical to good governance. All these results are consistent with those reported by previous empirical work. 22

24 CONCLUSION A review of the literature dealing with the determinants of regional inequalities, fiscal decentralization and government quality brings forth the view that these three variables are simultaneously determined. And while previous empirical work has tried to deal with the possibility of feedback effects between them, it has done so through instrumental variable estimates based on instruments of questionable validity. In this article we have applied a simultaneous equation model which accounts directly for the joint determination of these three variables, thus allowing us to deal directly with these potential interdependencies. Our empirical findings based on a sample of twenty-three OECD economies over the period 1984 to 2005, indicate the existence of a two-way relationship between the degree of fiscal decentralization and the quality of government: fiscal decentralization has a positive impact on governance, perhaps because it empowers better informed subnational governments and because it promotes interjurisdictional competition; countries with worse quality governments will tend to be more fiscally centralized, possibly because inefficient or corrupt officials may resist decentralization to maintain their access to public resources. We also find robust evidence of a two way relationship between government quality and regional inequalities. Countries with poorer governance are less able to narrow regional income differences because bad governance acts as a constraint on the adoption of efficient regional development strategies and their capacity to attract FDI. Additionally, we also find that countries with larger regional inequalities will tend to be less well governed, possibly because they are plagued with redistributive conflicts over the territorial distribution of resources conflicts which crowd out efforts towards improvements in governance. 23

25 With regards the relationship between fiscal decentralization and regional inequalities, our results support the view that fiscal decentralization can reduce regional inequalities. Alternatively we find no evidence for the claim that more significant regional inequalities affect the centralization of fiscal resources. In this regard, our findings diverge from those reported by previous work. Interestingly, our evidence suggests one reason for this divergence namely, the likelihood that previous work suffers from a degree of omitted variable bias because it fails to control for the confounding impact of government quality. The evidence provided is valuable from a normative or policy perspective. In particular, it suggests that if the objective is to reduce regional inequalities, a process of fiscal decentralization accompanied by measures to improve the quality of government (for example, reducing corruption and improving bureaucratic efficiency) would be an effective strategy, at least in this sample of OECD countries. Our evidence also raises the possibility of structural difficulties facing those striving to adopt suitable policies: on the one hand, the presence of significant regional inequalities may undermine efforts to improve public sector efficiency and, on the other, corrupt or inefficient officials may resist fiscal decentralization. This is not to say that countries with large regional inequalities and lower government quality cannot make improvements through the right policies but, rather, that desirable policy changes may face additional restrictions or constraints. Before ending, it is important to reiterate that our results emerge from a sample of mostly rich and advanced economies. One potentially important avenue for future research, conditional on data availability, is to explore the relationship between the three variables spotlighted in this article in a wider sample which includes both developed and, especially developing countries. Developing countries tend to have more 24

26 significant regional inequalities (Lessmann 2012), have traditionally been more centralized (World Bank 1999/2000) and have lower government quality (La Porta et al. 1999). And this configuration is very much in line with both the theoretical arguments and empirical evidence provided in this article suggesting a negative and two-way association between inequalities and government quality on the one hand, and fiscal centralization and the quality of government on the other. Widening the sample to include less developed countries would serve as a useful check on the robustness of our results, at the same time as it could provide new insights on the complex relationship between regional inequality, fiscal decentralization and the quality of government. REFERENCES Acemoglu, Daron, Simon Johnson and James Robinson (2001). The Colonial Origins of Comparative Development: An Empirical Investigation. American Economic Review, 91: Ades, A. & Di Tella, R. (1999) Rents, competition and corruption, American Economic Review, 89(4), Alesina A, Spolaore E (2003) The size of nations. MIT Press, Cambridge MA, London Alesina A, Zhuravskaya E (2011) Segregation and the quality of government in a cross-section of countries. American Economic Review 101(5): Andrews, D.W.K. (1991) Heteroskedasticity and autocorrelation consistent covariance matrix estimation. Econometrica 59, Arzaghi M, Henderson V (2005) Why countries are fiscally decentralizing. Journal of Public Economics 89: Avery, R. (1977) Error components and seemingly unrelated regressions. Econometrica 45, Baltagi, B.H. (1981) Simultaneous equations with error components. Journal of Econometrics 17, Baltagi, B.H. (2008) Econometric Analysis of Panel Data, fourth ed. Wiley and Sons, Chichester Bardhan, P. (2002) Decentralization of government and development. Journal of Economic Perspectives, 16(4): Bardhan P, Mookherjee D, 2006, ``Decentralization, corruption and government accountability: an overview'', in International Handbook on the Economics of Corruption Ed. S Rose- Ackerman (Edward Elgar, Cheltenham, Glos) pp Barro, R. (2000) Inequality and growth in a panel of countries, Journal of Economic Growth, 5(1), Barro R (2001) Human capital and growth. American Economic Review 91(2):

27 Barro R, Lee J (2001) International data on educational attainment: Updates and implications. Oxford Economic Papers 53(3): Bazzi S, Clemens M (2013) Blunt instruments: avoiding common pitfalls in identifying the causes of economic growth. American Economic Journal: Macroeconomics 5(2): Beramendi, P. (2007). Inequality and the Territorial Fragmentation of Solidarity. International Organization 61, Bjorn, E., Krishnakumar, J. (2008). Measurement errors and simultaneity, in: Mátyás, L., Sevestre, P. (Eds.), The Econometrics of Panel Data. Fundamentals and Recent Developments in Theory and Practice, Chapter 10, Brennan G, Buchanan J (1980) The power to tax. Analytical foundations of a fiscal constitution. Cambridge University Press, Cambridge Cai H, Treisman D, 2004, ``State corroding federalism'' Journal of Public Economics Cai H, Treisman D (2005) Does competition for capital discipline government? Decentralization, globalization and public policy. American Economic Review 95(3): Cameron, A. C., and P. K. Trivedi (2005). Microeconometrics: Methods and Applications. New York: Cambridge University Press Microeconometrics Using Stata. Rev. ed. College Station, TX: Stata Press. Costa-Font, J. (2010). Does devolution lead to regional inequalities in welfare activity? Environment and Planning C: Government and Policy 28(3): Davidson, R., and J. G. MacKinnon Econometric Theory and Methods. New York: Oxford University Press. Elazar D, 1995, ``From statism to federalism: a paradigm shift'' Publius. The Journal of Federalism Enikolopov R., Zhuravskaya E. (2007) Decentralization and Political Institutions. Journal of Public Economics 91(11-12): Ezcurra, R. (2012). Is there a link between globalization and governance? Environment and Planning C. Government and Policy 30(5): Ezcurra R., Pascual P. (2008) Fiscal decentralization and regional disparities: Evidence from several European Union countries. Environment and Planning A 40: Ezcurra R, Rodríguez-Pose, A. (2014). Government quality and spatial inequality: A crosscountry analysis, Environment and Planning A 46(7): Gil C, Pascual P, Rapun M (2004) Regional economic disparities and decentralization. Urban Studies 41(1): Glaeser, Edward, Jose Scheinkman and Andrei Shleifer (2003). The injustice of Inequality. Journal of Monetary Economics, 50: Fan S., Lin C., Treisman D. (2009) Political decentralization and corruption: evidence from around the world. Journal of Public Economics 93: Fisman R., Gatti R. (2002) Decentralization and corruption: Evidence across countries. Journal of Public Economics 83: Gemmell, N., Kneller, R., Sanz, I., Fiscal decentralization and economic growth: Spending versus revenue decentralization. Economic Inquiry 51(4): Giannetti M (2002) The effects of integration on regional disparities: Convergence, divergence or both? European Economic Review 46(3): Gil C., Pascual P., Rapun M. (2004) Regional economic disparities and decentralization. Urban Studies 41(1):