Resource Curse and Power Balance: Evidence from Oil-Rich Countries

Transcription

1 World Development Vol. 40, No. 7, pp , 2012 Ó 2012 Elsevier Ltd. All rights reserved X/$ - see front matter Resource Curse and Power Balance: Evidence from Oil-Rich Countries KJETIL BJORVATN NHH Norwegian School of Economics, Bergen, Norway MOHAMMAD REZA FARZANEGAN Philipps-University Marburg (CNMS), ZEW Mannheim & TU Dresden, Germany and FRIEDRICH SCHNEIDER * Johannes Kepler University of Linz, Austria Summary. We examine the role of political fractionalization in understanding the resource curse. Using panel data for 30 oil-rich countries, we find that the income effect of resource rents is moderated by the political power balance. With a strong government, resource wealth can generate growth even in an environment of poorly developed institutions, while adding oil revenues to a weak government may have damaging effects on the economy. These results have important implications for the economic prospects of the oil-rich countries in the Middle East, which are currently undergoing profound political changes. Ó 2012 Elsevier Ltd. All rights reserved. Key words oil rents, balance of power, resource curse 1. INTRODUCTION The literature on the resource curse shows that resource wealth may inhibit economic growth (for an excellent review of literature see Frankel, 2010). In particular, the curse applies to point source natural resources like oil (Boschini, Pettersson, & Roine, 2007). Other studies have established that fundamentals such as institutional quality (Mehlum, Moene, & Torvik, 2006) are important conditioning factors for the resource curse. Digging deeper, Hodler (2006) argues that institutional quality is shaped by ethnic fractionalization through a process of rent-seeking, thereby pointing to ethnicity as a fundamental source of the curse. 1 Since both ethnic composition and institutional quality are relatively constant over time, these studies typically amount to a comparison of resource wealth and income across countries. In contrast, the present paper focuses on within country effects. Using panel data for 30 oil-rich countries from 1992 to 2005, we identify political power balance as an important determinant of the efficient use of resource rents. This result holds when we control for the effects of other determinants of income, time varying common shocks, and country fixed effects. It is also robust to various alternative measures of resource abundance and inclusion of the quality of democratic institutions, as well as to the instrumental variable method of estimation (system and differenced GMM (Generalized Method of Moments)) and across different samples. The optimistic message from our study is that oil rich, less developed countries, which typically also have less developed institutions, are not destined to be development disasters. Indeed, with a strong government, resource wealth is likely to be growth enhancing, even when institutions are relatively weak. However, there is another side to this coin, with a more pessimistic message: Even in an environment of sound institutions, adding oil revenues to a fractionalized government can 1308 be like adding fuel to the fire, and cause slower economic growth. Our analysis provides insights which are timely, given the recent political turmoil in many resource-rich countries in the Middle East and North Africa. The literature on the resource curse focuses on rent-seeking as a key mechanism linking high resource wealth to low economic performance. 2 For instance, Torvik (2002) suggests that natural resource rents divert entrepreneurial talent from productive activities to low-productive rent-seeking activities. He concludes that the fall of income due to this re-allocation of entrepreneurs may outweigh the benefits of natural resource rents. Hodler, 2006 also builds a model of rent-seeking, where ethnic groups compete for a share of the resource wealth. Bjorvatn and Selvik (2008) explicitly focus on the power balance of different political factions in society, using Iran as a case in point. Their theoretical model shows that in a situation of high resource rents, political power balance may lead to extensive rent dissipation as this intensifies the competition for power. In contrast, when resource rents are low, a political power balance may be beneficial, since it stimulates investment by the different factions of the élite. Bjorvatn, Farzanegan, and Schneider (2012) lend empirical support to these mechanisms, analyzing factionalism, oil rents, and economic growth in Iran. The present paper builds on these two latter contributions, broadening the perspective from a case study of Iran to a study involving 30 oil rich countries. * We thank Michael Alexeev, Jessica Dewald, Anne-Kathrin Koch, Marcel Thum, the referee as well as participants in the Brown Bag seminars at the TU Dresden and ZEW Mannheim and the 18th Annual Conference of the European Association of Environmental and Resource Economists for helpful comments. Mohammad Reza Farzanegan gratefully acknowledges the Georg Forster Postdoctoral grant of Alexander von Humboldt Foundation. Final revision accepted: March 7, 2012.

2 RESOURCE CURSE AND POWER BALANCE: EVIDENCE FROM OIL-RICH COUNTRIES 1309 The remainder of the paper is structured as follows. Section 2 presents our empirical strategy and the data. Results are presented and discussed in Section 3. Section 4 concludes the paper. 2. EMPIRICAL RESEARCH DESIGN (a) Data, specification, and empirical strategy To estimate whether the relationship between oil rents and GDP (Gross Domestic Product) per capita varies systematically with the balance of political power, we use the following model: income it ¼ cons þ b 1 oil it þ b 2 power it þ b 3 ðoil it power it Þþb 4 Z it þ l i þ d t þ e it with country i (1,.., 30), and time t ( ). 3 Income is the log of real GDP per capita, oil is the log of oil revenues (as a share of total government revenues), power is a measure of political fractionalization, oilpower is the interaction of oil revenues and power fractionalization, and Z stands for the control variables. All explanatory variables are lagged one year to avoid possible endogeneity problems (see Mehran and Peristiani (2009) for a similar approach). 4 At the margin, the total effect of increasing oil revenues can be calculated by examining the partial derivatives of real income per capita with respect to the oil rent it Þ ¼ it Þ 1 þ b 3 ðpower it Þ ð2þ Based on the theoretical predictions by Bjorvatn and Selvik (2008) and Bjorvatn et al. (2012), we expect the sign of b 1 to be positive and the sign of b 3 to be negative. This means that increasing oil rents in a situation of weak governments lead to a destructive competition, with detrimental effects on income. Of course, there are other factors such as climate, culture, geography, and other unobserved time-invariant factors which are country specific and may correlate with income as well. If such country specific or time specific factors are correlated with oil rents or balance of power, then both pooled cross section and random effects estimations may lead to biased and inconsistent results. 5 We allow for country (l i ) and time (d t ) specific effects, controlling for the unobservable time-invariant country characteristics and shocks which are common to all countries. Furthermore, we can address the spurious business cycle effects by including time fixed effects (Keller, 2004). (b) Dependent and independent variables Following Alexeev and Conrad (2009), Hall and Jones (1999), Easterly and Levine (2003), and Rodrik, Subramanian, and Trebbi (2004) we use the level of real GDP per capita (in log) as our dependent variable, taken from the World Bank (2008). 6 Our main independent variables are oil revenues and balance of power. The most relevant proxy for the oil rent is the share of oil revenues in the government budget. The data are taken from Bornhorst, Gupta, and Thornton (2009) 7. The main proxy for the degree of political power balance is fractionalization of governing parties, or political fractionalization for short, and is taken from Beck, Clarke, Groff, Keefer, and Walsh (2001). The power balance index, which goes from 0 to 1, is defined as the probability that two members of parliament picked at random from governing parties belong ð1þ to different parties. A high fractionalization index indicates that the government consists of a large number of small parties, which we shall think of as a weak government, while a low fractionalization index indicates that the government consists of a small number of strong parties, possibly only one party, and signals a strong government. Our focus on the role of party systems in economic performance is supported by Riker (1964), who emphasizes that strong political parties are a major cause of high public goods provision and economic growth. A large number of independent candidates and candidates from recently formed new parties in presidential or parliament elections indicate a higher degree of fractionalization of the political system and is not conducive to growth. This is confirmed empirically by Enikolopov and Zhuravskaya (2007) who show that political fractionalization is bad for public goods provision. Similarly, Poteete (2009) argues that one of the main causes of the successful development path in the resource-rich country of Botswana was the strong and stable political coalition during the first decades of independence. The control variables include investment as a ratio of real GDP, inflation rate (as a measure of macroeconomic instability), real government consumption as a ratio of real GDP (a proxy for size of government distortions in the economy), trade openness, financial development, and age dependency (to control for the structure of population). Appendix A presents the countries in the sample. The data description and sources are presented in Appendix B. Appendix C reports the descriptive statistics of the major variables. 3. EMPIRICAL RESULTS The results of pooled, fixed, and random effects of panel regressions are presented in Table 1. All specifications in Table 1 show a negative and highly significant effect of the interaction term. This suggests that the political power balance plays an important role in determining the income effect of resource rents. In particular, resource rents are less likely to have a positive effect on income when governments are highly fractionalized, and therefore weak. Note also that the direct effect of oil revenues on real GDP per capita is positive and statistically significant at the 1% level in all specifications. For example, in the specification with country and period fixed effects (model 1.3); a 1% increase in the size of oil revenues leads to a 1.35% increase in levels of real GDP per capita. This direct positive effect of oil rents on economic development is in line with the findings of Alexeev and Conrad (2009). We also observe that power balance as such does not negatively affect income, as indicated by the fact that the estimated coefficient of govfrac is positive and statistically significant. This is in line with the model presented in Bjorvatn and Selvik (2008), which shows that a balanced power structure (which, ceteris paribus, implies a higher level of fractionalization) stimulates investment. In a resource- poor country, this effect may dominate the negative effect of rent-seeking, leading to higher income. In all specifications, we have controlled for other important determinants of income such as trade openness, government consumption, financial development, inflation, investment, and age structure of population. We also control for the common time trend in random effect regressions. 8 Including time trend controls for other factors such as technological progress which may affect the economic development of countries in our sample. The effects of our control variables on income are as expected in theory.

3 Table 1. Oil, fractionalization of government parties and income (panel regressions) Variables Dependent variable: log (rgdp p.c.), Pooled OLS, 1 lag of IVs (1.1) Oilrev 3.01 *** (5.25) c. fe, 1 lag of IVs (1.2) 1.22 *** (5.54) of IVs (1.3) 1.35 *** (5.64) of IVs (dropping Govex) (1.4) 1.96 *** (5.97) of IVs (excluding Norway) (1.5) 1.35 *** (5.46) c.re,1 lag of IVs (1.6) 1.29 *** (6.15) cre, 1 lag of IVs excluding Norway (1.7) 1.34 *** (6.28) c & tfe, using 2 lags of IVs (1.8) 1.05 *** (4.61) c & tfe, using 5 lags of IVs (1.9) Govfrac 2.13 ** (2.28) 0.38 *** (2.53) 0.33 ** (2.13) 0.38 ** (2.15) 0.37 * (1.80) 0.42 ** (2.44) 0.44 ** (2.22) 0.13 (0.82) 0.17 (1.28) Oilrev * Govfrac 6.23 *** ( 3.03) 2.08 *** ( 3.53) 1.86 *** ( 3.14) 2.27 *** ( 2.99) 1.94 *** ( 2.91) 2.11 *** ( 3.71) 2.14 *** ( 3.52) 1.31 ** ( 2.25) 1.06 ** ( 2.28) Trade 0.33 ** (2.47) (0.78) (1.13) 0.12 (1.27) (1.13) (1.06) (1.04) 0.21 *** (3.04) 0.26 *** (3.52) Govex 0.33 (1.56) 0.45 *** ( 6.91) 0.45 *** ( 7.35) 0.45 *** ( 7.37) 0.44 *** ( 6.75) 0.43 *** ( 6.66) 0.42 *** ( 7.35) 0.20 ** ( 2.40) Credit 0.34 *** (3.16) 0.01 (0.38) (0.02) 0.02 ( 0.24) (0.03) 0.03 (0.86) 0.03 (0.77) 0.03 ( 0.91) 0.00 ( 0.00) Inflation * ( 1.86) ( 3.67) ( 4.20) 3.95E 05 * ( 1.79) ( 4.23) ( 3.74) ( 3.74) ( 4.04) *** ( 2.68) Invest ( 0.48) 0.07 (1.25) 0.05 (0.88) ( 0.10) (0.88) (1.28) 0.07 (1.26) 0.04 (0.57) 0.11 ** (2.19) Age dep 2.26 *** ( 6.40) 0.11 (0.61) (0.33) 0.38 (1.51) 0.05 (0.22) 0.07 ( 0.39) 0.13 ( 0.63) 0.01 () 0.53 ** ( 1.96) Country fixed effects No Yes Yes Yes Yes Yes (RE) Yes (RE) Yes Yes Time trend Yes Yes No No No Yes Yes No No Time fixed effects No No Yes Yes Yes No No Yes Yes Obs. (countries) 248 (25) 248 (25) 248 (25) 257 (26) 236 (24) 248 (25) 236 (24) 226 (25) 154 (24) R Note: t-statistics are in parentheses, basing on heteroskedasticity-robust standard errors (White diagonal s.e. and covariance; df correction); constant is not shown. c & tfeand re refer to country and time fixed and random effects. All variables except for inflation are in log. Results are robust using higher lags of explanatory variables. * Significant at 10% level. ** Significant at 5% level. *** Significant at 1% level *** (2.80) 1310 WORLD DEVELOPMENT

4 Table 2. Oil, fractionalization of government parties, income, and political institutions (panel regressions) Variables Dependent variable: log (rgdp p.c), of IVs (2.1) of IVs, dropping Govex (2.2) c & t,1 lag of IVs ex.norway (2.3) of IVs, using oil rents per capita(2.4) of IVs, using oil rents per capita), excluding Norway (2.5) One step SYS. GMM, using Oilrev (2.6) One step Diff. GMM, using Oilrev (2.7) Oilrev 1.39 *** (4.13) 2.07 *** (7.51) 1.42 *** (4.21) 0.13 *** (3.42) 0.13 *** (3.30) 0.46 ** (2.26) 0.45 ** (2.19) Govfrac 0.28 * (1.80) 0.31 (1.62) 0.29 (1.32) 0.60 ** (2.07) 1.07 ** (2.46) 0.33 (1.54) 0.30 (1.25) Oilrev * Govfraco 1.75 *** ( 2.99) 2.11 *** ( 3.51) 1.77 *** ( 2.56) 0.10 ** ( 2.23) 0.19 ** ( 2.47) 1.25 ** ( 2.33) 1.24 ** ( 2.00) Trade 0.07 (0.96) 0.11 (1.58) 0.07 (0.95) (0.72) (0.75) ( 0.02) ( 0.00) Govex 0.45 *** ( 7.32) 0.45 *** ( 7.35) 0.44 *** ( 8.24) 0.44 *** ( 8.09) 0.26 * ( 1.74) 0.25 * ( 1.81) Credit 0.00 (0.22) 0.01 ( 0.34) 0.00 (0.14) 0.03 (1.04) 0.31 (0.88) 0.01 (0.22) 0.00 (0.16) Inflation ( 4.15) 4.17E 05 ( 0.97) ( 4.17) ** ( 2.44) ** ( 2.29) * ( 1.79) * ( 1.86) Invest (0.95) 0.00 ( 0.00) (0.96) 0.05 (0.88) 0.05 (0.90) 0.13 *** (2.89) 0.12 ** (2.47) Age dep 0.12 (0.64) 0.45 * (1.96) (0.36) 0.11 (0.57) 0.07 (0.37) 0.48 ( 0.87) 0.34 ( 0.60) Polity 0.21 (0.92) 0.32 (1.18) 0.23 (0.89) 1.15 *** ( 2.89) 1.36 *** ( 3.02) 0.12 (0.62) (0.49) Polity * Oilrev 0.15 ( 0.28) 0.33 ( 0.48) 0.24 ( 0.37) 0.24 *** (3.38) 0.27 *** (3.42) Country fixed effects Yes Yes Yes Yes Yes Yes Yes Time fixed effects Yes Yes Yes Yes Yes Time trend Time trend Lagged dependent variable 0.95 *** (17.39) 0.90 *** (12.27) Obs. (countries) 248 (25) 257 (26) 236 (24) 258 (25) 246 (24) 253 (25) 226 (25) R Sargan (p-value) 54.3 (0.49) 57.3 (0.38) AR(1)-p-value AR(2)-p-value Note: t-statistics are in parentheses, basing on heteroskedasticity-robust standard errors (White diagonal s.e. and covariance; df correction); constant is not shown. c & tfeand re refer to country and time fixed and random effects. All variables except for inflation are in log. The values reported for the Sargan test are the p-values for the null hypothesis of instrument validity. The values reported for AR (1) and AR (2) are the p-values for first and second order auto correlated disturbances in the GMM models. * Significant at 10% level. ** Significant at 5% level. *** Significant at 1% level. RESOURCE CURSE AND POWER BALANCE: EVIDENCE FROM OIL-RICH COUNTRIES 1311

5 1312 WORLD DEVELOPMENT On average, 50% of total government revenues in the countries of our sample from 1995 to 2002 depend on oil revenues. The size of government spending in oil based economies is relatively high and is usually financed through oil rents. The bivariate correlation between oil revenues and government spending in our sample is not too high (0.42). Even extreme collinearity (as long as it is not perfect) does not violate OLS (Ordinary Least Squares) assumptions. OLS estimates are still unbiased and BLUE (Best, Linear, Unbiased Estimator). 9 However, in order to examine the sensitivity and robustness of our main results, we drop the government expenditure variables from general specification. Our main results as presented in Model 1.4 are quite robust without government expenditure as well. In our sample, Norway has an established democracy, ranking as one of the least corrupt countries in the past years. As a robustness test, we have removed Norway from panel estimations and results are presented in Models 1.5 and 1.7 in Table 1. The main results remain robust without Norway. The random effects estimation presented in Models do not differ from fixed effects estimations qualitatively. Our main variables of interest such as oil revenues and the interaction term remain statistically significant with the expected sign in the random effects models. 10 In Models 1.8 and 1.9, we also re-estimate the same specification using 2 and 5 year lags of explanatory variables instead of 1 year lags. Using higher order lags of independent variables reduces the risk of the endogeneity problem within our estimations. We notice that our main results are robust in models with a higher number of right hand side variables lags. The structure of government and institution in countries of our sample varies to different degrees. For example, Iran is a highly factionalized semi-democratic country while Saudi Arabia is an absolute monarchy, Kuwait is a constitutional monarchy, and Norway is an established democracy. It is therefore important to control for the quality of political institutions in our analysis. In Table 2, we have controlled for the role of political institutions using the Polity2 index from the Polity IV dataset (Marshall & Jaggers, 2009). Polity2 scores are between 10 and +10. A +10 refers to a strongly democratic state and 10 to strongly autocratic. We have rescaled the Polity2 index from 0 to 1. The higher the rescaled Polity index, the higher the quality of democratic institutions. We aim to examine whether including this variable affects the intermediary role of fractionalization in government parties in the oil rents-income nexus. In other words, we want to make sure that the effect of fractionalization of government parties does not reflect the quality of political institutions. We notice that even by controlling for the Polity index and its interaction with oil revenues, our main variables of interest remain robust with expected sign and are highly significant (see Model 2.1). As in Table 1, we drop the government expenditure variable in Model 2.2 for the robustness analysis. We notice that this does not affect our main finding. Excluding Norway also does not change our main results (Model 2.3). Furthermore, we present the results using an alternative proxy for the oil resources. We use per capita oil rents instead of the share of oil revenues in total government revenues. Alexeev and Conrad (2009) argue that the best measures of the role of natural resources in the long run growth are per capita measures. Using oil wealth as a measure of GDP or export in growth regressions may bias estimates in favor of the resource curse hypothesis. There can be factors unrelated to natural resources which affect the structure of the economy especially in the export sector. Thus, a resource rich country which suffers from these factors may have a high oil/gdp or oil/export ratio, presenting a spurious evidence for the resource curse hypothesis in regressions. Oil rents (in US dollars) are calculated as the production volume times the difference between the international market unit price and the average production cost. We have divided the total oil rents by the population of each country to calculate the per capita oil rents. Models 2.4 and 2.5 show the results, using per capita oil rents instead of oil revenues in total revenues. The direct effect of per capita oil rents on income is positive and significant at the 1% level. However, the magnitude of the effect is now smaller. A 1% increase in the size of per capita oil rents raises per capita income by 0.13%. The moderating effect of fractionalization of government parties remains statistically significant at the 5% level in models with per capita oil rents. Excluding Norway, as in Model 2.4, does not change our results. As we can see from Models 2.4 and 2.5, the interaction of the Polity index and oil rents is positive and statistically significant. Higher quality of political institutions can amplify the positive income effects of oil rents. This is in line with findings of previous literature such as Mehlum et al. (2006). The direct effect of Polity on income is negative and significant in Models 2.4 and 2.5. This means that, given a negligible share of oil rents, higher political openness may reduce the speed of economic growth. This can be due to the increased political instability in poor resource economies by opening the political system. Previous studies have found an inverted U-shaped relationship between democracy and conflict (Hegre, Ellingsen, Gates, & Gleditsch, 2001; Muller & Weede, 1990). Full autocracy and full democracy should go along with a low risk of conflict (and higher income) since highly autocratic regimes can repress dissent and thus avoid civil violence, while highly democratic societies can solve problems peacefully. For middle values of the Polity index, we can thus make no clear predictions. Furthermore, as per capita income tends to be persistent over time, we estimate a dynamic specification (difference GMM) using the Arellano and Bond (1991) estimator, which allows the specification of a common lagged effect (see Model 2.7 in Table 2). We use three lags of potentially endogenous variables as instruments. The Sargan test validates the adequacy of the instruments, and the failure to reject the null hypothesis of the validity of the instruments indicates that the specification is correct. In addition, the other diagnostics are also satisfactory. The absence of first order serial correlation is rejected and the absence of second order serial correlation is not rejected. The dynamic GMM differences the model to remove country specific effects or any time-invariant country specific variable, eliminating any endogeneity because of the correlation of these country specific effects and the right hand side regressors. It also addresses the possible non-stationarity of explanatory variables (Baltagi, Demetriades, & Law, 2009). The results in Model 2.7 provide further support for our basic models regarding a direct positive effect of oil revenues on income and a moderating effect of fractionalization of government parties. We also report the results of system GMM introduced by Blundell and Bond (1998) in Model 2.6. The System GMM estimator combines the first differenced equation and levels equation to estimate the model, employing both lagged levels and differences as instruments. The model is estimated using a first difference transformation to remove the individual country effect. In addition, to control for the country specific effects, the system GMM preserves the cross country dimension of the data which is lost when we use first differenced GMM (Castelló-Climent, 2008). Again, our results provide empirical supports for our main hypotheses. The diagnostic statistics are also satisfactory. The Sargan test does not reject the over-identification restriction, indicating that instruments are not correlated with the error term and, thus, are valid.

6 RESOURCE CURSE AND POWER BALANCE: EVIDENCE FROM OIL-RICH COUNTRIES 1313 Figure 1. Marginal effects of oil revenues on income at different levels of fractionalization. Note: the middle line shows the marginal impact of a 1% increase in the size of oil revenues (% of total revenues) on real GDP per capita at different levels of political fractionalization. The upper and lower lines are 90% confidence intervals (CIs). We reject the absence of first order serial correlation while accepting the absence of second order serial correlation. Furthermore, the lagged dependent variable in both Diff GMM and SYS GMM is positive and significant. The relatively high size of lagged dependent variables indicates its persistency, however, it is statistically different from unity in both models. To sum up, our results show that the final effect of oil rents on growth is conditional on the level of fractionalization among government parties. Using Eqn. (1), we calculate the marginal impact of oil rents on growth at different levels of the fractionalization index (mean, minimum and maximum). We use Models 2.1 and 2.4 (country and time fixed effects panel regression) to calculate the marginal effects of a 1% increase in oil revenues [in total revenues and oil rents per capita]. The results are presented in Table 3. At the maximum level of fractionalization, a 1% increase in the share of oil revenues in total government revenues and per capita oil rents increases income by 0.41% and 0.07%, respectively. In the case of dominance of a single political faction (govfrc = 0), the same increase in oil revenues (or oil rent) has a more pronounced positive effect on growth (1.39% and 0.13%, respectively). The overall impact of oil revenues on growth is positive but higher fractionalization of politics moderates this positive effect. We have also illustrated the marginal impacts of oil on income at different values of government fractionalization, reporting 90% confidence intervals around estimated marginal effects. This approach enables us to determine the conditions under which the oil revenues have a statistically significant effect on the GDP per capita. The results are illustrated in Figure 1. Note that we refer to the results estimated on the basis of two way fixed effects in Table 2 (Model 2.1 with robust standard errors). Table 4. Standardized coefficients Dependent variable: log (rgdp p.c), Variables Standardized coefficients Oilrev_S 0.20 *** (5.63) Govfrac_S 0.04 * (1.83) Oilrev * Govfrac_S 0.12 *** ( 3.03) Trade_S 0.03 (1.03) Govex_S 0.19 *** ( 7.35) Credit_S (0.20) Inf_S 0.92 *** ( 4.14) Invest_S 0.02 (0.93) Age_S 0.03 (0.69) Polity_S 0.03 (1.52) Country fixed effects Yes Year fixed effects Yes Obs. 248 R-square 0.99 Note: t-statistics are in parentheses, basing on heteroskedasticity-robust standard errors (White diagonal s.e. and covariance; df correction). Standardized coefficients show increases or decreases in the dependent variable if the explanatory variable increases by one standard deviation. * Significant at 10% level. ** Significant at 5% level. *** Significant at 1% level. Table 3. Marginal effects of oil revenues on growth at different levels of power balance Marginal effects Oil revenues (based on model 2.1 estimations) Oil rents per capita (based on Model 2.4 estimations) Mean of power balance (govfrac) Maximum of power balance (govfrac) Minimum of power balance (govfrac) Power balance is the logarithm of one plus govfrac( 1). Mean, maximum, and minimum levels of power balance are presented in Appendix C. Oil revenues is the logarithm of one plus oilrev( 1). Oil rents per capita is the logarithm of oil rents per capita ( 1).

7 1314 WORLD DEVELOPMENT The marginal effects (the middle solid line) are statistically significant when the 90% confidence interval bands (dashed lines) fall above or under the zero line. The leftmost seven points represent the statistically significant marginal effects of oil revenues on income. The histogram in the background adds interesting information by showing us how the cases are distributed. The statistically significant marginal impacts cover most parts of our observations. At the maximum level of government fractionalization, the marginal effect is at its minimum but insignificant at the 90% level of confidence. As we can see, for the majority of countries in our sample, the marginal effect of oil revenues on income is positive and significant. 11 In order to examine the relative importance of explanatory variables such as oil revenues and fractionalization of government parties in a growth model, we use standardized variables in regression analysis by subtracting their mean and dividing them by their standard deviation. Table 4 presents the standardized coefficients of the main explanatory variables in a country and year fixed effect regression. Standardized variables have the same unit of measurement, making an interpretation of the results easier. The standardized coefficient indicates the impact of the explanatory variable in terms of standard deviation units. It shows us the number of standard deviations that the dependent variable (standardized real GDP per capita) increases or decreases with a one standard deviation increase in the standardized independent variable. In other words, we can evaluate the relative importance of our main independent variables in explaining growth. We find that inflation, oil revenues, government consumption, and interaction of fractionalization and oil revenues have the most important effects on growth. 4. CONCLUSION Our analysis shows that the evolution of political power balances and imbalances over time can have important effects on a country s ability to benefit from its oil revenues. Indeed, when the level of fractionalization is high, indicating a weak government, oil revenues appear to be fully wasted: Above a critical level of fractionalization, there is no significant, positive effect of oil revenues on income. In contrast, when governments are less fractionalized, for instance, consisting of a single party, oil revenues have a pronounced positive effect on income. We believe that these results have important implications for countries in the Middle East and North Africa which are currently undergoing fundamental changes in their political structure. While these changes may hopefully lead to more sound political and economic institutions, they may also open up for more intensified power struggles. This was certainly the case in Iran, where the omnipotent Shah in 1979 was replaced by a system which invited factionalist infighting, leading to a rather disappointing economic performance. There is a risk that intensified rent-seeking will lead to lackluster growth also in the wake of the current revolutions in the Middle East and North Africa, leading to a period of prolonged political instability in the region. NOTES 1. Other empirical studies on institutions and the resource curse include Brunnschweiler and Bulte (2008), Brunnschweiler (2008), Iimi (2007), and Alexeev and Conrad (2010). 2. There is also an older literature focusing on market related mechanisms for the resource curse, linked to the concept of the Dutch disease. In this case, higher oil prices lead to a higher real effective exchange rate and an appreciation of the domestic currency, thus increasing the price of nonoil exports and causing deindustrialization (see Corden & Neary, 1982; Corden, 1984, and van Wijenbergen, 1984). 3. The main criterion for choosing our sample is the availability of data for our measure of resource wealth, which is the share of oil revenues in total government revenues. Rent-seeking is likely to be driven by these more tangible rents rather than by oil reserves or production, measured in barrels of oil (see Hodler, 2006, for a similar view). 4. We have also used lagged 2, 3, 4 and 5 years explanatory variables in estimations to control robustness of our main results in Table 1. The results hold, even using longer lagged explanatory variables, providing more support for estimated coefficients. In addition, our main independent variable, oil revenues, is largely exogenous in income models in each country. For example, OPEC members must follow the determined level of the crude oil production quota by this organization (Farzanegan, 2011). Global oil prices are also mostly exogenous for the countries in our sample. For instance, Farzanegan and Markwardt (2009) argue that demand for Iranian oil largely depends on global economic growth, energy intensities within industrial countries, speculator operation in the global oil markets, expectations of other key oil producers about current and future developments of the market, international oil companies decisions on liquidation of their stocks and finally, the policy of key oil consumers on strategic petroleum reserves. Alexeev and Conrad (2010) argue that natural resource endowments and the output of a country are largely exogenous. 5. Nevertheless, for comparison of our results we present estimations of pooled, random and fixed effects. 6. If we consider a linear regression of Y on X, then the standard interpretation of the coefficient of X is the change in Y in response to the change in X. In our case, Y is the GDP level, then the change of Y is growth and our coefficient would be the effect of the change of X on changes in the level of the GDP (i.e. growth as a result of the change of X). If, however, our Y is the growth rate, then our coefficient is the change in growth rate of the GDP in response to a change in X. 7. We thank Fabian Bornhorst for making this data available to us. As for a robustness test, we also use the per capita oil rents instead of the share of oil revenues in total government revenues (See Table 2). 8. We do not include common time trend in panel regressions with country and time fixed effects because of perfect multicollinearity between time trend and time fixed effects. 9. For more details see Richard Williams s note at: rwilliam/stats2/l11.pdf. 10. We have carried out the Hausman test under the null hypothesis so that the individual effects are uncorrelated with the other regressors in the model (Hausman, 1978). The p-value = 0 rejects the null hypothesis, indicating that fixed effect models are preferred. 11. Using estimations of GMM in Models 2.6 and 2.7 for the calculation of the marginal impact does not change our results in Figure 1.

8 RESOURCE CURSE AND POWER BALANCE: EVIDENCE FROM OIL-RICH COUNTRIES 1315 REFERENCES Alexeev, M., & Conrad, R. (2009). The elusive curse of oil. The Review of Economics and Statistics, 91(3), Alexeev, M., & Conrad, R. (2010). The natural resource curse and economic transition. Economic Systems, 35(4), Arellano, M., & Bond, S. (1991). Some tests of specification for panel data: Monte Carlo evidence and an application to employment equations. Review of Economic Studies, 58(2), Baltagi, B. H., Demetriades, P. O., & Law, S. H. (2009). Financial development and openness: Evidence from panel data. Journal of Development Economics, 89(2), Beck, T., Clarke, G., Groff, A., Keefer, P., & Walsh, P. (2001). New tools in comparative political economy: The database of political institutions. World Bank Economic Review, 15(1), Bjorvatn, K., Farzanegan, M.R., & Schneider, F. (2012). Resource curse and power balance: Evidence from Iran. Mimeo. Available at: works.bepress.com/mr_farzanegan/11/. Bjorvatn, K., & Selvik, K. (2008). Destructive competition: Factionalism and rent-seeking in Iran. World Development, 36(11), Blundell, R., & Bond, S. (1998). Initial conditions and moment restrictions in dynamic panel data models. Journal of Econometrics, 87(1), Bornhorst, F., Gupta, S., & Thornton, J. (2009). Natural resource endowments and the domestic revenue effort. European Journal of Political Economy, 25(4), Boschini, A. D., Pettersson, J., & Roine, J. (2007). Resource curse or not: A question of appropriability. Scandinavian Journal of Economics, 109(3), Brunnschweiler, C. N. (2008). Cursing the blessing? Natural resource abundance, institutions and economic growth. World Development, 36(3), Brunnschweiler, C. N., & Bulte, E. H. (2008). The resource curse revisited and revised: A tale of paradoxes and red herrings. Journal of Environmental Economics and Management, 55(3), Castelló-Climent, A. (2008). On the distribution of education and democracy. Journal of Development Economics, 87(2), Corden, M. W. (1984). Booming sector and Dutch disease economics: Survey and consolidation. Oxford Economic Papers, 36(3), Corden, M. W., & Neary, J. P. (1982). Booming sector and deindustrialisation in a small open economy. The Economic Journal, 92(368), Easterly, W., & Levine, R. (2003). Tropics, germs, and crops: How endowments influence economic development. Journal of Monetary Economics, 50(1), Enikolopov, R., & Zhuravskaya, E. (2007). Decentralization and political institutions. Journal of Public Economics, 91(11 12), Farzanegan, M. R. (2011). Oil revenue shocks and government spending behavior in Iran. Energy Economics, 33(6), Farzanegan, M. R., & Markwardt, G. (2009). The effects of oil price shocks on the Iranian economy. Energy Economics, 31(1), Frankel, J.F. (2010). The natural resource curse: A survey. NBER working paper, no Hall, R., & Jones, C. (1999). Why do some countries produce so much more output per worker than others?. Journal of Economics, 114(1), Hausman, J. A. (1978). Specification tests in econometrics. Econometrica, 46(6), Hegre, H., Ellingsen, T., Gates, S., & Gleditsch, N. P. (2001). Toward a democratic civil peace? democracy, political change, and civil war, American Political Science Review, 95(1), Hodler, R. (2006). The curse of natural resources in fractionalized countries. European Economic Review, 50(6), Iimi, A. (2007). Escaping from the resource curse: Evidence from Botswana and the rest of the world. IMF Staff Papers, 54(4), Keller, W. (2004). International technology diffusion. Journal of Economic Literature, 42(3), Marshall, M., & Jaggers, K. (2009). Polity IV project: political regime characteristics and transitions, University of Maryland. Mehlum, H., Moene, K., & Torvik, R. (2006). Institutions and resource curse. Economic Journal, 116(508), Mehran, H., & Peristiani, S. (2009). Financial visibility and the decision to go private. The Review of Financial Studies, 23(2), Muller, E. N., & Weede, E. (1990). Cross-national variation in political violence: A rational action approach. The Journal of Conflict Resolution, 34(4), Poteete, A. R. (2009). Is development path dependent or political? A reinterpretation of mineral- dependent development in Botswana. Journal of Development Studies, 45(4), Riker, W. (1964). Federalism: Origins, operation, significance. Boston, MA: Little, Brown and Co.. Rodrik, D., Subramanian, A., & Trebbi, F. (2004). Institutions rule: The primacy of institutions over geography and integration in economic development. Journal of Economic Growth, 9(2), Torvik, R. (2002). Natural resources, rent-seeking and welfare. Journal of Development Economics, 67(2), van Wijenbergen, S. (1984). Inflation, employment, and the Dutch disease in oil exporting countries: A short-run disequilibrium analysis. Quarterly Journal of Economics, 99(2), World Bank (2008). World development indicators. Washington, D.C.: World Bank. APPENDIX A Countries included in the sample OPEC Algeria Angola Ecuador Iran Kuwait Libya Nigeria Qatar Saudi Arabia United Arab Emirates Venezuela Non-OPEC Azerbaijan Bahrain Brunei Cameroon Chad Congo Equatorial Guinea Gabon Indonesia Kazakhstan Mexico Norway Oman Russia Sudan Syria Trinidad and Tobago Vietnam Yemen

9 1316 WORLD DEVELOPMENT Data description Variable rgdp p.c. Oilrev Govfrac Trade Govex Credit Inflation APPENDIX B Description and Source Logarithm of GDP per capita (constant 2000 USD). Source: World Bank (2008) Logarithm of one plus Oil revenues/total revenues of government. Source: Bornhorst et al. (2009) Logarithm of one plus fractionalization of government parties. The probability that two members of parliament picked at random from among the government parties will be of different parties. Missing if there is no parliament, if there are any government parties where seats are unknown, or if there are no parties in the legislature. Scale from 0 to 1.Source: Beck et al. (2001) Logarithm of sum of imports and exports in GDP. Source: World Bank (2008) Logarithm of government consumption in GDP. Source: World Bank (2008) Logarithm of domestic credit to private sector (% of GDP). Source: World Bank (2008) Inflation, consumer prices (annual%). Source: World Bank (2008) Invest Age_dep Polity2 Summary statistics Variable Logarithm of gross fixed capital formation (% of GDP). Source: World Bank (2008) Logarithm of age dependency ratio (% of working-age population). Source: World Bank (2008) Democracy Index. Scale from 10 (full autocracy) to 10 (full democracy). Rescaled from 0 to 1. Source: Marshall and Jaggers (2009) APPENDIX C Obs. Mean Standard deviation Minimum Maximum Log (rgdp p.c.) Log (govfrac( 1)+1) Log (oilrev( 1)+1) Log (govfrac( 1)+1) Log (oilrev( 1)+1)