Natural Resources & Income Inequality: The Role of Ethnic Divisions

Transcription

1 DEPARTMENT OF ECONOMICS OxCarre (Oxford Centre for the Analysis of Resource Rich Economies) Manor Road Building, Manor Road, Oxford OX1 3UQ Tel: +44(0) Fax: +44(0) _ OxCarre Research Paper 23 Natural Resources & Income Inequality: The Role of Ethnic Divisions Ruikang Marcus Fum (University of Melbourne) Roland Hodler (University of Melbourne & OxCarre) Direct tel: +44(0) celia.kingham@economics.ox.ac.uk

2 Natural Resources and Income Inequality: The Role of Ethnic Divisions Ruikang Marcus FUM and Roland HODLER May 18, 2009 Abstract We hypothesize that natural resources raise income inequality in ethnically polarized societies, but reduce income inequality in ethnically homogenous societies; and we present empirical evidence in support of this hypothesis. JEL classification: D7, O1 Key words: Natural resources; inequality; ethnic polarization; resource curse Department of Economics, University of Melbourne. Department of Economics, University of Melbourne; and OxCarre, University of Oxford. (corresponding author) 1

3 1 Introduction Ross (2007, p. 238) notices that surprisingly little is known about the relationship between natural resources and income inequality, but that resource rich countries appear to be neither more nor less unequal, on average. Inspired by Hodler (2006) and Brunnschweiler and Bulte (2009), who show that natural resources curse ethnically divided societies by fueling rent seeking contests and civil conflicts, we hypothesize that ethnic divisions could affect this relationship. 1 In resource-rich societies with a few large ethnic groups, each group may channel time and effort away from productive activities to rent seeking or fighting activities, while only one group gets the resource rent. This group ends up richer than it would be in the absence of this rent while the other groups end up poorer. Resource rents may thus increase income inequality in divided societies. In contrast, resource rents may reduce income inequality in homogenous societies in which they may be shared more or less equally, or even used to support the poor. This would be consistent with the observation that redistribution tends to be more generous in ethnically homogenous societies (Alesina and Glaeser, 2004); and also with the idea that excluding parts of the population from getting a share of the resource rent is more difficult in the absence of easily observable differences in physical appearance (Caselli and Coleman, 2006). Finally, if there are many small ethnic groups, no single group may be strong enough to capture the entire resource rent. We thus expect the resource rent to be shared among various ethnic groups - either peacefully or during long civil conflicts - and not to have any strong positive or negative effects on income inequality. In the remainder of this paper, we empirically test the hypothesis that natural resources raise income inequality in ethnically polarized societies, but reduce income inequality in 1 Leamer et al. (1999), Gylfason and Zoega (2002), and Goderis and Malone (2008) discuss alternative channels through which natural resources may affect income inequality. Van der Ploeg (2008) reviews the literature on the so-called resource curse. 2

4 ethnically homogenous societies. 2 Data and methodology For this purpose, we need measures of ethnic polarization, resource abundance, and income inequality. We use Montalvo and Reynal-Querol s (2005) index of ethnic polarization (EP). This index ranges from zero to one. It is close to one in polarized countries with two large ethnic groups capturing roughly half the population each, and close to zero in homogenous countries in which almost everybody belongs to the same ethnic group. EP is also low in countries with many small ethnic groups. Our resource measure is based on the World Bank s (1997, 2005) estimates of natural capital, including fuel and non-fuel subsoil wealth as well as land wealth from cropland, pastureland, forests and protected areas. We use log average natural capital per capita in 1994 and 2000 (NC). This variable is widely used in the literature and arguably a more accurate measure of resource abundance than variables with denominators measuring economic performance (e.g., Hodler, 2006; Brunnschweiler and Bulte, 2008, 2009). We use Gini coefficients from the World Income Inequality Database (WIID) to measure income inequality. As it is well known, Gini coefficients are reported differently across countries. We derive our measure, GINI, as follows: To get a relatively large sample, we rely on observations based on household level income data rather than individual income data. We combine observations based on post-tax income and expenditure/consumption data. We however exclude observations based on pre-tax income data because we care about inequality after the political processes of rent seeking and redistribution, and also because these observations are less comparable with observations based on other types of data. Due to the varying levels of quality and gaps in the data, we further exclude observations based on data of the lowest quality (quality 4), and we average the Gini 3

5 coefficients over the period from 1990 to We also use Gini coefficients from the World Development Indicators averaged over the same period (GINIWDI). This measure however includes observations based on pre-tax income data. We estimate the following model: GINI = α + β 1 EP + β 2 NC + β 3 (EP NC) + X Λ + ε, (1) where X is a vector of control variables including log GDP per capita in 1990 (GDP), its squared value (GDP 2 ), and log population in 1990 (POP). Table 1 reports descriptive statistics of our main variables, and the Appendix lists definitions and sources of all variables. The effect of NC on GINI is β 2 + β 3 EP. Our hypothesis suggests that β 2 should be negative, and β 3 positive. This would imply that there is a threshold level of EP below which the effect of NC on GINI is negative, and above which this effect is positive. 3 Empirical evidence Table 2 contains our main results. In column 1 we start by looking at the effects of GDP, GDP 2 and POP on GINI. We find evidence for a Kuznets curve and for lower inequality in more populous societies. In columns 2 and 3 we add NC and EP. As Ross (2007), we also notice no significant relationship between NC and GINI on average. In column 4 we present our baseline regression which includes the interaction term EP NC. We notice that the coefficients on NC is significantly negative, and the coefficient on EP NC significantly positive. These results support our hypothesis. In an average country, natural resources feed income inequality if EP is above the threshold level of 0.52, but reduce income inequality otherwise. 2 Resource rents may therefore contribute to the high inequality in 2 Similarly, these results suggest that EP increases GINI if and only if NC exceeds

6 ethnically polarized Bolivia and Mexico (where EP is 0.77 and 0.65, respectively) as well as to the low inequality in ethnically homogenous Norway (where EP is 0.09). In columns 5 and 6 we show that our results remain unchanged when omitting countries from Latin America or Sub-Saharan Africa, where inequality tends to be higher than elsewhere. It is known from the literature that income inequality is negatively associated with education, trade openness and institutional quality, and that landlocked countries tend to be more unequal (Sylwester, 2004). In table 3 we show that our main results remain significant when controlling for these potential covariates of GINI. 3 In table 4 we explore the role of our measures of divisions, natural resources and income inequality. In columns 1 and 2 we show that it is indeed ethnic polarization rather than ethnic fractionalization or religious polarization that determines how NC affects GINI. The former finding suggests that natural resources do not have a strong effect on inequality in societies with many small ethnic groups. In column 3 we divide NC into its two main components, subsoil wealth (SW) and land wealth (LW). We find that it is primarily the effect of SW on GINI that depends on EP. This finding is not surprising given that SW tends to be more locally concentrated than LW and, therefore, more likely to end up in the hands of a single powerful group. Finally, we test whether our results are an artefact of our construction of GINI. In column 4 we account for the fact that we used Gini coefficients based on different types of data sources by adding a dummy that equals one for observations exclusively based on expenditure/consumption data, and another dummy that equals one for observations exclusively based on post-tax income data. 4 In column 5 we replace GINI by GINIWDI. Our main results survive these tests. 3 Our results also survive when we control for the institutional variables and their interactions with NC. 4 Both dummies equal zero for observations based on data from both sources. 5

7 4 Conclusions Our empirical findings support the hypothesis that natural resources raise income inequality in ethnically polarized countries such as Bolivia or Mexico, but reduce income inequality in ethnically homogenous countries such as Norway. Appendix GINI: Our Gini coefficients for the period See section 2 for details. Source: WIID GINIWDI: Gini coefficient, averaged over the period Source: WDI GDP: Log GDP per capita in Source: WDI POP: Log population in Source: WDI EP: Ethnic polarization. Source: Montalvo and Reynal-Querol (2005) EF: Ethnic fractionalization. Source: Montalvo and Reynal-Querol (2005) RP: Religious polarization. Source: Montalvo and Reynal-Querol (2005) NC: Log average natural capital per capita in 1994 and Source: World Bank (1997, 2005) SW: Log average subsoil wealth per capita in 1994 and Source: World Bank (1997, 2005) LW: Log average land wealth per capita in 1994 and Source: World Bank (1997, 2005) Average years of primary/secondary/higher schooling in total population in Source: Barro and Lee (2000) Trade volume as share of GDP in Source: WDI Constraints on executive in Source: Polity IV Polity2 score for Source: Polity IV Dummy equal to one for landlocked countries. Source: Sylwester (2004) 6

8 References Alesina, A., Glaeser, E., Fighting Poverty in the U.S. and Europe: A World of Difference. Oxford University Press, Oxford. Barro, R.J., Lee, J.W., International data on educational attainment: Updates and implications. CID Working Paper 42. Brunnschweiler, C., Bulte, E., The resource curse revisited and revised: A tale of paradoxes and red herrings. Journal of Environmental Economics and Management 55, Brunnschweiler, C., Bulte, E., Fractionalization and the fight over natural resources: Ethnicity, language, religion, and the onset of civil war. OxCarre Research Paper Caselli, F., Coleman II, W.J., On the Theory of Ethnic Conflict. Discussion Paper, LSE. Goderis, B., Malone, S.W., Natural resource booms and inequality: Theory and evidence. OxCarre Research Paper Gylfason, T., Zoega, G., Inequality and economic growth: Do natural resources matter? CESifo Working Paper 712. Hodler, R., The curse of natural resources in fractionalized countries. European Economic Review 50, Leamer, E.E., Maul, H., Rodriguez, S., Schott, P.K., Does natural resource abundance increase Latin American income inequality? Journal of Development Economics 59, Montalvo, J.G., Reynal-Querol, M., Ethnic polarization, potential conflict and civil wars. American Economic Review 95, Ross, M.L., How mineral-rich states can reduce inequality. In: Sachs, J.D., Stiglitz, J.E., Humphreys, M. (eds.), Escaping the Resource Curse, Columbia University Press, New York. Sylwester, K., A note on geography, institutions, and income inequality. Economic Letters 85, van der Ploeg, R., Challenges and Opportunities for Resource Rich Economies. OxCarre Research Paper World Bank, Expanding the measure of wealth. Environmentally Sustainable Development Studies and Monographs 17. World Bank, Where is the Wealth of Nations? Measuring Capital for the XXI Century. IBRD/World Bank, Washington, DC. 7

9 Table 1: Summary Statistics Variable Number of observations Gini coefficient (GINI) 79 Mean 42.6 Standard Deviation 10.8 Minimum 22.9 Maximum 73.9 Ethnic Pol. (EP) Natural Capital (NC) GDP Table 2: Main Results Dependent Variable: GINI (1) (2) (3) (4) (5) (6) GDP 27.31*** 27.29*** 23.07*** 19.45*** 13.31** 21.28*** (5.909) (5.956) (5.487) (5.546) (6.593) (7.905) GDP *** (0.391) *** (0.393) *** (0.362) *** (0.365) ** (0.427) *** (0.495) Population (POP) *** (0.682) *** (0.691) *** (0.610) *** (0.545) *** (0.598) *** (0.679) Ethnic Pol. (EP) 6.824* (3.658) *** (31.54) ** (37.45) *** (27.32) Natural Capital (NC) (1.068) (1.153) *** (2.073) ** (2.557) *** (2.018) EPxNC 11.37*** (3.2) 10.88** (4.462) 12.23*** (3.339) Omitted Observations Latin American countries Sub-Saharan African countries Adjusted R 2 Observations Notes: ***, **, and * indicates significance at the 1%, 5%, and 10% level. Robust standard errors are in parentheses. All regressions include an intercept.

10 Table 3: Additional Controls Dependent Variable: GINI (1) (2) (3) (4) (5) (6) Ethnic Pol. (EP) ** (32.27) ** (32.31) *** (32.11) *** (32.13) ** (30.05) (36.22) Natural Capital (NC) ** (2.314) *** (2.124) *** (2.080) *** (2.116) ** (2.012) * (2.637) EPxNC 10.46*** (3.821) *** (3.839) 12.08*** (3.839) 11.85*** (3.842) 9.790*** (3.637) 8.072* (4.323) Primary Schooling (0.997) (1.047) Secondary Schooling (1.675) (1.594) Higher Schooling (7.153) (7.045) Trade volume as share of GDP (0.024) (0.022) Constraints on Executive (0.491) (1.305) Polity2 score (0.179) (0.472) Dummy for Landlocked Controls Adjusted R 2 Observations Notes: See table GDP, GDP 2, POP *** (2.619) *** (2.925)

11 Table 4: Alternative Measures Dependent Variable: GINI GINIWDI (1) (2) (3) (4) (5) Ethnic Pol. (EP) ** (45.14) *** (29.52) (27.71) ** (29.65) ** (27.35) Ethnic Frac. (EF) (38.57) Religious Pol. (RP) (31.92) Natural Capital (NC) *** (2.147) ** (2.281) ** (2.049) *** (1.9) Subsoil Wealth (SW) Land Assets (LW) *** (0.730) (1.8) EPxNC 14.82*** (5.474) 11.16*** (3.518) 9.484*** (3.555) 8.838*** (3.324) EFxNC (4.661) RP*NC (3.742) EPxSW EPxLW 3.048** (1.452) (4.328) Dummy for expend./cons. data ** (2.602) Dummy for posttax income data Controls Adjusted R 2 Observations Notes: See table GDP, GDP 2, POP (2.427)