Redistribution, Trade and Corruption: An Empirical Assessment

Redistribution, Trade and Corruption: An Empirical Assessment Antonia Reinecke Hans-Jörg Schmerer Abstract This paper explores the role of institutional quality in the trade and inequality nexus. Does corruption shape the relationship between trade and inequality through its impact on redistribution? Our answer to this question builds on the hypothesis that trade rises inequality through higher per capita income at the top of the income distribution. Motivated by recent theoretical evidence, we argue that governments may intervene through appropriate redistribution schemes that aim at taxing the gains from trade in a way that reduces inequality. Corruption and bad institutions may induce distortions that neutralize those positive effects: inequality rises due to trade liberalization, if bad institutions prevent redistribution schemes by the government. We find that trade increases inequality but the effect hinges on the level of institutional quality. Quite to the contrary to common wisdom, we even find an income inequality reducing effect of trade in countries with high institutional standards as low corruption or a high level of regulatory quality. FernUniversität in Hagen. E-mail: antonia.reinecke@fernuni-hagen.de FernUniversität in Hagen, CESifo and IAB. E-Mail: hans-joerg.schmerer@fernuni-hagen.de 1

I. Introduction There is an emerging consensus in the academic debate that trade causes inequality. Explanations for this result either focus on fair wage considerations as in Egger and Kreickemeier (2009a) or search and matching with screening costs as in Helpman, Itskhoki, and Redding (2010) and Helpman, Itskhoki, Muendler, and Redding (2012). All of those studies are able to associate rising wages at the top and stagnant or even declining wages at the bottom of the income distribution with globalization. Moreover, those more recent models have in common that they build on the seminal heterogeneous firm framework with sorting of heterogeneous firms into export. The presence of labor market frictions from different proveniences can explain why trade increases within-group inequality due to some sort of rent sharing. An otherwise identical worker earns a higher wage, if she is employed at a more productive exporting firm. Going from autarky to free trade still increases per capita income, albeit those gains are usually accompanied by a more dispersed income distribution. Thus, the gains from trade and the type of inequality discussed in the literature mentioned in the last paragraph are different to the welfare gains known from the more canonical trade models. Firstly, countries that open up its economy to international trade benefit from an exit of the least productive firms as domestic competition becomes fiercer due to entry of foreign exporters and the strengthening of domestic exporters. 1 Secondly, more productive firms are more efficient, generate higher profits and pay higher wages due to rent-sharing between workers and firms. This adjustment process results in soaring inequality among unobservable attributes, e.g. firm productivity or worker ability, rather than observable differences in 1 Foreign and domestic establishments compete over market shares within a particular country. 2

skills. This result is in line with recent empirical evidence. The increased availability of rich firm-level data that comprise information on both firms and workers enabled researchers to show that exporting firms pay wage premia to their employees. The seminal papers by Bernard, Jensen, and Lawrence (1995), Bernard and Jensen (1999) and Bernard, Eaton, Jensen, and Kortum (2003) motivated a large and growing literature that sheds light on the determinants for this exporter wage premium. Schank, Schnabel, and Wagner (2010) argue that part of the premium can be explained by unobserved worker heterogeneity. 2 It is difficult to gauge the importance of inequality in the public debate but there seems to be a widespread consensus that governments should intervene using appropriate redistribution schemes. Especially the unexplained differences in income are less easy to defend in the public debate. Is there a channel through which governments can offset the negative labor market effects of globalization through redistributing the gains from trade? Appropriate policy instruments include lump sum tariffs that are redistributed in form of unemployment benefits (this link is explained in De Pinto (2015a) and De Pinto (2015b)) or lump sum tariff payments that are redistributed equally among the whole population as Egger and Kreickemeier (2009b) propose in their model. The latter model goes one step further by investigating the possibility that governments can reduce the inequality promoting effects of trade liberalization by setting a tax that is high enough to reduce overall inequality without purging the entire gains from trade. Thus, their paper can rationalize that governments are indeed able to offset the negative effects of trade on inequality by applying appropriate tax schemes that reduce inequality. Moreover, Gozgur and Ranjan (2015) report some evidence that trade liber- 2 See also Hauptmann and Schmerer (2013) and Felbermayr, Hauptmann, and Schmerer (2014) for more recent empirical evidence on the exporter wage premium. 3

alization fosters redistribution. Combining the arguments proposed by Egger and Kreickemeier (2009a), Gozgur and Ranjan (2015) and Egger and Kreickemeier (2009b) one may have some doubts about the unambiguity of the long run effects of trade on inequality. We argue that those considerations call for a full-fledged empirical analysis that aims at shedding light on the role of the government in the debate on trade and inequality. Using macroeconomic data on openness and income inequality we estimate the relationship empirically. Our main contribution is to take the role of institutions into consideration when estimating the relationship between trade and inequality. Redistribution may reduce the surge of inequality induced by globalization but this positive effect may be distorted in countries with bad institutions. In compliance with Gupta, Davoodi, and Alonso-Terme (2002) the quality of institutions has an influence on income inequality through different channels, i.e. economic growth, biased tax systems, less targeted redistributional policies as well as educational inequality. Corruption is one attribute of low institutional quality. Mauro (1998) as well as Tanzi and Davoodi (2000) identify a negative relation between the level of corruption and public spendings on education, health care as well as social insurance and welfare payments, while they found a positive relation between corruption and public military spending. These findings suggest that public spendings are less equally distributed among citizens. Additionally, Tanzi and Davoodi (2000) descry corruption affects tax revenues and their productivity negatively, which is associated with less progressivity. Corruption or weak enforcement power of a government may lead to a situation, in which the gains from trade are not redistributed efficiently, that magnifies inequality relative to the effect in countries characterized by high 4

institutional quality. 3 But not only the redistribution system itself is prone to corruption. Lambsdorff (1998) shows that a country s level of corruption also has a significant influence on its trade volumes. Disadvantages can emerge through declining shares of more corrupt countries trade in the world market. Moreover, De Groot, Linders, Rietvield, and Sumbramanian (2004) argue that low institutional quality increases transaction cost of trade by boosting insecurity and diminishing the investors trust in business procedures. The authors state that homogeneity of institutional quality between two observed trading partners is an additional source of lower transaction cost due to reduced burdens on market entry and more familiarity regarding business proceedings. The empirical results of De Groot, Linders, Rietvield, and Sumbramanian (2004) support the hypotheses highlighted above: The estimated influence of perceived institutional quality on bilateral trade is highly significant and positive, whereas the divergence of government effectiveness between two countries have a negative impact on bilateral trade. 4 First glimpse at the data. Figure 1 compares the evolution of inequality, trade and a proxy for redistribution over time. The graph shows averages across OECD countries going back to the year 1970. The preferred measure of inequality is the net-gini coefficient. The market Gini has little explantory power in our application because it does not account for transfer payments by the respective government. We are instead interested in the purchasing power of individual 3 One has to admit that the suggested mechanism opposes the results in Itskhoki (2008) that there exists no optimal redistribution strategy that allows tackling income inequality induced by trade without offsetting aggregated welfare gains. 4 The effect of institutional quality and the homogeneity of institutional quality on bilateral trade is estimated by a gravity equation. The World Bank Government indicators are used as proxy for institutional quality. Homogeneity of institutional quality is measured by a dummy variable that takes the value one, if the difference of a respective institutional quality measure between two countries exceeds a previously defined share of the sample standard deviation. Those countries are defined as heterogeneous. 5

households. Figure 1: Openness and inequality The stylized facts in the left panel of Figure 1 reveal a clear trend towards higher globalization but the growth rates are declining in more recent years. The right panel of Figure 1 allows to asses inequality and its evolution over time. Omitting redistribution gives a positive correlation between trade liberalization and soaring inequality: the market Gini grows from 0.4 shortly before 1980 to more than 0.45 around 2005. However, we get a different picture once we look at the net-gini, which fluctuates around the level 0.3. Thus, the overall level of inequality in our sample becomes much lower once transfer payments are accounted for and the positive trend is not as obvious anymore. This first glance at the data can be considered as evidence that points towards a high relevancy of redistribution. Figure 2 illustrates the evolution of corruption and different countries degree of redistribution. Thereby, it confronts corruption (CPI - Corruption Perception Index) as our preferred measure of institutional quality and the degree of redistribution, that is approximated by the difference between market and net-gini. We argue that countries with higher corruption tend to redistribute less. Corruption may very well magnify the negative effects of trade on inequality by hindering governmental redistribution schemes. We observe a clear pattern 6

Figure 2: Openness and redistribution between corruption and redistribution. Countries with low CPI are more corrupt and redistribute less. Our empirical analysis tries to jointly estimate the relationship between the three variables of interest. We expect some interrelationship between corruption, trade and redistribution tested in the following empirical analysis. II. Empirical Analysis We test the hypothesized relationship derived from our theoretical considerations using a dataset that covers 47 countries 5. Inclusion of both developed and emerging economies for a long period that covers the years 1995-2011 gives us enough variation in trade and institutions so that we are able to disentangle the total effect into its between and within components. Our preferred specification 5 The countries included comprise the major developing and developed countries, namely: Argentina, Australia, Austria, Belgium, Bolivia, Bulgaria, Canada, Chile, China, Colombia, Croatia, Czech Republic, Denmark, Ecuador, El Salvador, Estonia, Finland, France, Germany, Greece, Hungary, India, Indonesia, Ireland, Israel, Italy, Japan, Latvia, Lithuania, Mexico, Moldova, Netherlands, New Zealand, Norway, Poland, Portugal, Romania, Russia, Slovakia, Slovenia, South Africa, Spain, Sweden, Thailand, Turkey, United Kingdom, United States. 7

includes country and time fixed effects. The model estimated in our study reads IE it = α + β 1 (G) it + β 2 (Q) it + β 3 (G Q) it + β n (CV) + u it. The dependent variable, IE, is the Net Gini coefficient in country i at time t. The Gini coefficient is calculated on the basis of disposable income, which is total income net of tax payments and other transfer payments received by the workers in the respective country. Variable G describes the economy s level of globalisation. Our preferred proxy for openness is the KOF globalization index, which comprises information on imports and exports as well as other indicators of globalisation. The variable Q stands for the different institutional quality measures considered in this study. To shed light on the role of institutions for the link between trade and inequality we also include the interaction between both measures. We expect that bad institutions magnify the effects of trade on inequality towards more inequality. However, the effect of openness is expected to be ambiguous and may be positive or negative depending on the degree of redistribution. A negative impact of openness on income inequality may be attributed to country specific characteristics, in particular the level of the respective country s development. Emerging economies are usually characterized by a high level of agricultural production, low productivity rates, lower developed infrastructure as well as less targeted social redistribution systems, which leads to higher levels of income inequality. In contrast, highly industrialized countries are characterized by high productivity rates, a high level of industrial production, and hence a distinctive export sector, as well as well developed redistribution systems, which would lead to lower income inequality levels. This is supported by the empirical results of Gozgur and Ranjan (2015) that identify a positive effect of international trade on the level of public redistribu- 8

tion by estimating an econometric model. 6 The results in their study reveal a positive and highly significant effect of trade on the degree of redistribution in an economy. Consistently, high developed economies associated with a high level of trade openness tend to be more engaged in redistribution than emerging economies. With respect to the coefficients obtained from regressions that focus on within income inequality, the expectations regarding the estimated coefficients are different. In compliance with Egger and Kreickemeier (2009a), trade liberalization leads to firm selection and higher inequality but the impact of the quality of institutions is expected to be negative. Thus, we expect a positive coefficient of trade but a negative of the interaction term between openness and institutional quality. The gains from trade are expected to be distributed more equally in economies with better institutions. Consequently, we expect that the potential positive effect of openness on income inequality can be attenuated, or even overcompensated by strong institutions, which ensure that not only one specific part of income distribution benefits from trade, but redistribute welfare gains from trade among the whole distribution. All regressions include per capita GDP, the age dependency ratio 7 and population (in millionen inhabitants) as further control variables. IV strategy Potential endogeneity of trade and corruption likely biases the benchmark OLS estimates. Controlling for country fixed effects already deals with that issue but we go one step further using an instrumental variable (IV) approach. Our main concern is that corruption in the underlying regression analysis is endogenous due to potential reverse causality: While high corruption is expected to distort redistribution schemes, and thereby fosters inequality, it 6 The dependent variable in their study is specified by the difference between market Gini and the net Gini. This difference is positively correlated with openness. 7 "Age dependency ratio is the ratio of dependents people younger than 15 or older than 64 to the working-age population those ages 15-64. Data are shown as the proportion of dependents per 100 working-age population.", World Bank, 2016. 9

can be hypothesized that high inequality create an incentive for individuals to become corrupt in order to receive personal advantages. Economic openness and corruption as well as their interaction are instrumented by the interaction between the Frankel and Romer trade share and the indicator of Government Effectiveness (GE) as well as Regulatory Quality (RQ). Furthermore, we introduce both indexes, GE and RQ, separately as instruments. Frankel and Romer (1999) use a gravity approach to generate an instrument for trade openness. In contrast to classical gravity estimates, they exclusively introduce geographical characteristics determining bilateral trade as size measured as population and area, distance between two countries, common boarder and landlockedness in their regressions. Based on this first step they estimate a country-individual geographic component of overall trade. They argue that geographic characteristics of an economy are not determined by income or other aspects affecting income (such as political measures). However, the Frankel and Romer (1999) trade share is time invariant and thus not appropriate in a time-series panel analysis. We tackle this problem by generating various interactions with other exogenous variables. We are mainly interested in variables that are not affected by the distribution of income across individuals, at least not through variables that are not controlled for in the regressions. A government s efficiency may be influenced by inequality only if a crucial number of voters are unsatisfied with the situation and try to put pressure on the government in order to change the situation. Likely that voters are mainly concerned about the fact that government inefficiency spurs corruption and thus inequality. A second channel may go through redistribution. The same line of arguments holds for regulatory quality. In order to assess the validity of our instruments we report all relevant test statistics at the bottom of each regression table. The Sargan test allows to test the exclusion restriction. The first stage F-statistics jointly tests whether 10

the excluded instruments are significant. The rule of thumb is fulfilled as all F-statistics are greater than 10. Moreover, Shea s partial R-squared statistics are also sufficiently high, which supports the validity of our instruments. III. Data Table 1 reports the first and second moments as well as minimum and maximum values of all variables used in our study. The net Gini stems from The Standardized World Income Inequality Data (SWIID) by Solt (2014), who proposes a way to make the Gini coefficients from different countries comparable. 8 We use two different globalization measures in our study. The Penn World Table provides openness meassured as the sum of imports and exports relative to the respective country s level of GDP 9. Dreher (2006) argues that globalization also depends on other indicators not included in im- or exports. The KOF Globalisation Index introduced by Dreher (2006) is constructed out of three components: Firstly, economic globalisation depending on trade in goods and capital plus international money transactions. Secondly, social globalisation is included through a country s share of foreign population, tourism and Internet users. Thirdly, political globalisation comprises association with membership in international organizations, international treaties and participation in U. N. security councils. The focus of our analysis is on the Economic Globalisation Index and the overall Globalisation Index. All variables are normalized so that their values range from zero to one. A lower number indicates that the economy is less globalised, whereas the value one corresponds to the maximum degree of globalisation. As proxy for institutional quality we use the Corruption Perception Index (CPI) developed 8 Solt (2014) uses a costumised algorithm that calculates missing values under usage of data from national statistical offices, regional data collection as well as academic studies. 9 They use real GDP in million USD. 11

Table 1: Summary Statistics Variable Obs Mean St. Dev. Min Max Net Gini 855.347.089.203.657 KOF I 846.717.132.369.924 KOF II 846.690.148.260.970 (Imp+Exp)/GDP 799.8659 1.754.0741 36.368 Imp/GDP 799.470 1.1483.039 24.316 Exp/GDP 799.395.618.034 12.052 CPI 756 5.661 2.361 1.7 10 RQ 705.628.226 1.76e-08 1 GE 705.540.255 8.63e-09 1 Ln(GDP/POP) 799 9.545.954 5.753 11.066 Dependency 893.511.074.345.808 Population 799 87.716 239.449 1.340 1324.353 by Transparency International. The observations take values between zero and ten. The lowest value zero is associated with the highest level of perceived corruption, whereas the highest value indicates a zero perceived corruption within the respective country. Furthermore, indicators from The Worldwide Governance Indicators data set, developed by Kaufmann, Kraay, and Mastruzzi (2010), allow to approximate other dimensions of institutional quality. The World Bank dataset includes six dimensions that assist evaluating the state of governance. Our study makes use of the variable Regulatory Quality (RQ). All indicators in the original data set range from -2.5 to 2.5. We normalize the measure in order to guarantee comparability. Thus, our measure takes values from zero to one, with zero describing a weak level of institutional quality. Economies that are characterized by high institutional quality take a value equal to one. As explained in the introductory part, the channels corruption affects inequality are diverse: by distorting the tax collection process, by influencing public spending decisions, and thereby, effectiveness of redistribution, as well 12

as by manipulating the patterns of trade. The measures of institutional quality applied in this empirical analysis are aggregated indices, and hence, capture institutional quality as a whole and not one specific channel. Consistently, we estimate the overall effect of institutional quality. The data of the control variable dependency ratio originates from the World Bank. The per capita GDP is calculated based on data of the Penn World Table (PWT) data, version 8.1. III. Empirical Results In the following section, we present the benchmark regression outcomes for our different specifications. The results document a fairly robust finding: The effect of trade on inequality hinges on the quality of institutions. The overall effect of openness turns out to be positive in regressions that were purged of the between variation of the data but the interaction indicates that the marginal effect turns negative in countries with good institutions. Quite to the contrary of most of the existing studies we find some evidence that trade can reduce inequality in the long run. The role of corruption for trade and inequality Our preferred globalization measure is the KOF index on globalization. The coefficients associated with this variable are reported in the first row of Table 2. The third row reports the preferred proxy for corruption. The fifth row contains the estimates for the interaction of both variables of interest. We always include a bunch of controls and different combinations of fixed effects. In the first column we estimate Ordinary Least Squares (OLS) without controlling for any unobserved heterogeneity other then the time trend. Column (2) includes country-level random effects and column (3) puts in country fixed effects. At 13

first glance, our results seem to support the predicted positive coefficient of openness, if unobserved heterogeneity is controlled for: trade can be associated with more inequality. However, the marginal effect has to be interpreted together with the coefficient of the interaction term, which is negative. Thus, the positive effect is mitigated by the degree of corruption. A lower level of corruption reduces the magnitude of the effect of openness on inequality. We evaluate the marginal effect using the summary statistics reported in the previous section. The marginal effect in column (1) is negative, which implies that more open economies tend to have a lower level of inequality. Remember that the first model is always OLS, which bases the point estimates upon both between and within variation of the data. The between variation likely stems from the fact that more open economies are located in more developed countries with better welfare systems and lower overall inequality. Therefore, it is not surprising that the coefficient likely reflects the impact of the between variation. The interaction reveals that this effect becomes even stronger in less corrupt economies. Comparing two countries with the same level of openness, we find that the more corrupt economies are also having a more dispersed income distribution. This low precision of the OLS estimates motivate the regressions in column (2) and (3). The inclusion of fixed- and random-effects on the country level absorbs some or all of the between variation of the data. Fixed-effect estimates identify the coefficients solely based on the within-variation over time. The sign of the coefficient changes from negative to positive but the coefficient of the interaction term remains negative. Both are highly significant. Evaluating the marginal effect using the summary statistics reported above, we find that the marginal effects in column (2) and (3) turn from positive to negative at a level of perceived corruption equal to CPI RE = 0.236/0.047 = 5.02 and 14

CPI FE = 0.115/0.017 = 6.7610. Those numbers are close to the mean value of perceived corruption in the sample, which is around 5.66. Quality of governance and regulations Corruption is one source of distortion of the gains from trade. We argue that corrupt societies distort the rent-sharing mechanism between firms and workers through lowering the rents that can be redistributed between firms and workers. The institutional quality of governance and regulations is a much broader concept of measuring corruption. Countries with good regulatory institutions may be less plagued by corruption so that redistributing the gains from trade is likely more efficient. Coefficients related to the KOF globalization indicator are negative in column (4) but positive in (5) and (6). All are highly significant. The coefficient of the interaction term is always highly significant and negative. The results are identical to the ones discussed in the last paragraph. Countries with better regulatory quality tend to benefit from globalization in terms of a lower inequality. Only in countries with low regulatory quality trade causes more inequality. The cutoff for which the marginal effects turns from positive into negative is around 0.469-0.570. Alternative measure of the KOF globalization index As a first robustness check of our results we estimate our model with a narrower concept of international trade and include exclusively the economic component of the KOF globalisation Index. The Economic Globalisation Index represents 10 All countries characterized by a CPI above this threshold are members of the OECD. Additionally, these countries belong to the high-income countries according to the World Bank classification. These results support the hypothesis that especially industrialized countries are characterized by an efficient redistributing system that is not distorted by bad institutions. 15

Table 2: Benchmark results: Inequality, trade and institutions (1) (2) (3) (4) (5) (6) Net Gini Net Gini Net Gini Net Gini Net Gini Net Gini b/se b/se b/se b/se b/se b/se KOF I -0.136** 0.236** 0.115* -0.215*** 0.237** 0.147** (0.06) (0.11) (0.06) (0.08) (0.12) (0.07) CPI 0.032*** 0.027* 0.007 (0.01) (0.01) (0.01) KOF CPI -0.044*** -0.047** -0.017* (0.01) (0.02) (0.01) RQ 0.253*** 0.367** 0.213*** (0.08) (0.15) (0.07) KOF RQ -0.327*** -0.505** -0.258** (0.11) (0.22) (0.10) GDP/POP -0.004 0.028 0.083*** -0.005 0.019 0.069*** (0.00) (0.03) (0.01) (0.00) (0.03) (0.02) Dependency 0.405*** 0.337*** 0.386*** 0.360*** 0.301*** 0.359*** (0.04) (0.10) (0.04) (0.04) (0.09) (0.05) Population 0.000*** 0.000*** 0.000*** 0.000*** 0.000*** 0.000** (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) Constant 0.261*** -0.233-0.716*** 0.314*** -0.172-0.640*** (0.06) (0.28) (0.14) (0.07) (0.27) (0.17) Time FE x x x x x x Country RE x x Country FE x x Obs. 754 754 754 609 609 609 R-sq within 0.603 0.961 0.596 0.963 adj. R-sq 0.591 0.958 0.584 0.958 Standard errors in parentheses. Coefficients are significant at the 10 percent (* p<0.10), 5 percent (** p<0.05) or 1 percent (*** p<0.010) level. The dependent variable is the net-gini coefficient. KOF I is our preferred economic globalization measure. CPI denotes Corruption Perception Index, which is our preferred measure of institutional quality. RQ denotes Regulation Quality. GDP/POP is a control for per capita GDP and Dependency stands for Dependency Ratio. a more classical approximation of economic openness, but includes more dimensions than only imports and exports. It consists of actual flows 11 and trade 11 Trade including imports and exports, Foreign Direct Investment, Portfolio Investment and income payments to foreign national as percent of GDP, respectively 16

restrictions 12 weighted by 50%, respectively. The estimation results including the KOF Economic Globalisation Index instead of the total Globalisation Index can be found in Table (3). Our estimation results are robust against the change of the globalisation measure and the benchmark regression results can be restored: Controlling for nothing else than the time trend, the effect of globalisation is negative and significant, while the sign of the estimated coefficient changes controlling for between variation. The coefficient of the interaction term is invariably negative and highly significant. But, quite to our surprise, the cutoff is even higher for the alternative KOF globaliation measure. Only the least corrupt countries with CPI higher than 6.65 tend to benefit from trade liberalization, if inequality is the main concern. For regulatory quality the marginal effect changes at the critical value of RQ = 0.758 IV regression Table 4 and 5 show of IV regression results, thereby the regression in Table 4 includes exclusively time fixed effects, while regression estimates in Table 5 include both time and country fixed effects. As a further control variable we include redistribution 13 in column (3) and (4), respectively. Our results still exhibit the expected signs and are highly significant. Economic globalisation fosters income inequality, while the interaction between high quality institutions and globalisation has an income inequality reducing effect, indicating that strong institutions are decisive for the effectiveness of the redistribution of gains from trade. The high values of first stage F-statistics as well as of the partial 12 Concealed import barriers, mean tariff rate, taxes on international trade in percent of current values and capital account restrictions 13 Redistribution is defined as the difference between Market Gini and Net Gini. 17

Table 3: Benchmark results: Inequality, trade and institutions (1) (2) (3) (4) (5) (6) Net Gini Net Gini Net Gini Net Gini Net Gini Net Gini b/se b/se b/se b/se b/se b/se KOF II -0.022 0.206*** 0.154*** -0.023 0.213*** 0.156*** (0.05) (0.07) (0.04) (0.07) (0.08) (0.05) CPI 0.004 0.015* 0.011** (0.01) (0.01) (0.01) KOF CPI -0.017** -0.031** -0.023*** (0.01) (0.01) (0.01) RQ 0.056 0.190** 0.156*** (0.06) (0.08) (0.05) KOF RQ -0.159* -0.281** -0.203*** (0.09) (0.13) (0.07) GDP/POP -0.017*** 0.038 0.083*** -0.022*** 0.019 0.068*** (0.00) (0.03) (0.01) (0.00) (0.03) (0.02) Dependency 0.459*** 0.313*** 0.378*** 0.388*** 0.275*** 0.339*** (0.04) (0.10) (0.04) (0.04) (0.09) (0.05) Population 0.000*** 0.000*** 0.000*** 0.000*** 0.000*** 0.000** (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) Constant 0.319*** -0.289-0.739*** 0.380*** -0.136-0.612*** (0.06) (0.30) (0.14) (0.08) (0.26) (0.16) Time FE x x x x x x Country RE x x Country FE x x Obs. 754 754 754 609 609 609 R-sq within 0.542 0.962 0.525 0.963 adj. R-sq 0.528 0.959 0.511 0.959 Standard errors in parentheses. Coefficients are significant at the 10 percent (* p<0.10), 5 percent (** p<0.05) or 1 percent (*** p<0.010) level. The dependent variable is the net-gini coefficient. KOF I is our preferred economic globalization measure. CPI denotes Corruption Perception Index, which is our preferred measure of institutional quality. RQ denotes Regulation Quality. GDP/POP is a control for per capita GDP and Dependency stands for Dependency Ratio. R 2 suggest relevance 14 of the used instruments. As a further test for weak instruments we analyses the correlation between endogeneous variables and instruments as well as between instruments. The correlation is high, which 14 An instrumental variable must fulfill two conditions: Cov(z, u) = 0 and Cov(z, x) = 0. 18

indicates strength of used instruments. 15 Due to the fact that more instruments than potential endogeneous variables find application, the regression model is over identified. To test, if the overidentiying restriction is valid, we conduct a Sargan test. The p-values always exceed the critical value of p 0.05, consistently we can conclude that the overidentifying restrictions are valid. 15 Results of the correlation analysis can be found in Appendix I. 19

Table 4: IV Regressions (1) (2) (3) (4) Dependent Variable Net Gini Net Gini Net Gini Net Gini Estimator IV-REG IV-REG IV-REG IV-REG Sample period (all years) (years>2000) (all years) (years>2000) KOF 0.598*** 0.538*** 0.791*** 0.781*** (0.18) (0.18) (0.16) (0.17) CPI 0.125*** 0.137*** 0.103*** 0.107*** (0.02) (0.03) (0.01) (0.02) KOF CPI -0.176*** -0.185*** -0.148*** -0.152*** (0.03) (0.03) (0.02) (0.02) Dependency 0.435*** 0.319*** 0.473*** 0.418*** (0.08) (0.08) (0.06) (0.07) GDP/POP -0.020** -0.025** 0.006 0.008 (0.01) (0.01) (0.01) (0.01) Population 0.000** 0.000 0.000*** 0.000*** (0.00) (0.00) (0.00) (0.00) Redistribution -0.820*** -0.841*** (0.08) (0.09) Constant -0.073 0.050-0.362** -0.352** (0.17) (0.17) (0.15) (0.16) Time FE x x x x Country FE x Sargan p-value 0.6090 0.9266 0.4626 0.5025 F-stat 1st stage: KOF 136.227 117.626 87.3945 71.1582 F-stat 1st stage: CPI 380.899 371.212 319.686 309.404 F-stat 1st stage: KOF CPI 342.206 279.595 283.544 236.44 Partial R-sq: KOF 0.5295 0.5292 0.4298 0.4221 Partial R-sq: CPI 0.7886 0.8048 0.7667 0.7898 Partial R-sq: KOF CPI 0.7679 0.7713 0.7332 0.7401 Number of obs. 523.000 417.000 523.000 417.000 R-sq within 0.223 0.112 0.584 0.569 adj. R-sq 0.195 0.079 0.568 0.552 Standard errors in parentheses. Coefficients are significant at the 10 percent (* p<0.10), 5 percent (** p<0.05) or 1 percent (*** p<0.010) level. The dependent variable is the net-gini coefficient. KOF I is our preferred economic globalization measure. CPI denotes Corruption Perception Index, which is our preferred measure of institutional quality. RQ denotes Regulation Quality. GDP/POP is a control for per capita GDP and Dependency stands for Dependency Ratio. We use Government Efficiency, Quality of Regulations and the interaction with both a Frankel and Romer trade share as instruments for the KOF, CPI and the interaction between KOF and CPI. 20

Table 5: IV Regression, contd. (1) (2) (3) (4) Dependent Variable Net Gini Net Gini Net Gini Net Gini Estimator IV-REG IV-REG IV-REG IV-REG Sample period (all years) (years>2000) (all years) (years>2000) KOF 0.729*** 1.362*** 0.771*** 1.259*** (0.23) (0.33) (0.24) (0.32) CPI 0.040 0.142*** 0.053 0.126** (0.04) (0.05) (0.04) (0.06) KOF CPI -0.098* -0.253*** -0.124** -0.232** (0.06) (0.09) (0.06) (0.09) Dependency 0.507*** 0.479*** 0.575*** 0.496*** (0.07) (0.09) (0.07) (0.08) GDP/POP 0.090*** 0.144** 0.103*** 0.132** (0.03) (0.06) (0.03) (0.06) Population 0.000* -0.000 0.000-0.000 (0.00) (0.00) (0.00) (0.00) Redistribution -0.311*** -0.133 (0.06) (0.09) Constant -1.074*** -1.901*** -1.239*** -1.738*** (0.39) (0.63) (0.40) (0.63) Time FE x x x x Country FE x x x x Sargan p-value 0.9421 0.4044 0.7040 0.5860 F-stat 1st stage: KOF 8.651 6.086 8.673 6.1285 F-stat 1st stage: CPI 7.121 8.412 7.930 9.374 F-stat 1st stage: KOF CPI 8.288 3.730 9.299 5.720 Partial R-sq: KOF 0.1010 0.0602 0.1033 0.0601 Partial R-sq: CPI 0.0583 0.0960 0.0609 0.1004 Partial R-sq: KOF CPI 0.0736 0.0528 0.0762 0.0578 Number of obs. 523.000 417.000 523.000 417.000 R-sq within 0.949 0.948 0.950 0.953 adj. R-sq 0.942 0.939 0.943 0.944 Standard errors in parentheses. Coefficients are significant at the 10 percent (* p<0.10), 5 percent (** p<0.05) or 1 percent (*** p<0.010) level. The dependent variable is the net-gini coefficient. KOF I is our preferred economic globalization measure. CPI denotes Corruption Perception Index, which is our preferred measure of institutional quality. RQ denotes Regulation Quality. GDP/POP is a control for per capita GDP and Dependency stands for Dependency Ratio. We use Government Efficiency, Quality of Regulations and the interaction with both a Frankel and Romer trade share as instruments for the KOF, CPI and the interaction between KOF and CPI. 21

The role of imports and exports Dauth, Findeisen, and Suedekum (2012) argue that imports and exports may have different effects on labor market outcomes. Their focus lies on labor demand. Nevertheless, the effects on inequality may also be different for import and export penetration. If a government is ambitious in redistributing gains from trade, one would expect positive effects of soaring exports. Domestic firms may gain from exports. If governments tax part of those new profits from export, inequality may decrease. However, in countries where openness is driven by imports, one may argue that offshoring of low skilled jobs lead to a decline in wages at the bottom of the income distribution. Those losses are hardly compensated, if firms do not export more. However, if trade is balanced one may expect that both effects cancel each other out. Interestingly, we find no significant effects for import and export measures in the specification that include corruption as institutional quality measure. The picture changes in columns (4) to (5). Inclusion of the regulatory quality measures restores the pattern observed in the benchmark regression table. More openness increases inequality but the effect changes to negative in countries with good regulatory quality. However, the reducing effect of inequality shows up for the interaction with export openness only. The interaction term between imports and the institutional variable is insignificant. This result is in line with the intuition described above. Only exports generate additional rents that can be redistributed. Governments in countries with good regulations may tax some of the profits generated through higher export volumes in order to redistribute them to the citizens with low income. 22

Table 6: Econometric results Imports, Exports and Interaction (1) (2) (3) (4) (5) (6) Net Gini Net Gini Net Gini Net Gini Net Gini Net Gini b/se b/se b/se b/se b/se b/se Imp -0.011 0.052 0.027-0.065 0.092** 0.066* (0.03) (0.04) (0.02) (0.05) (0.05) (0.04) Exp 0.074 0.090 0.033-0.043 0.175** 0.116* (0.05) (0.08) (0.04) (0.08) (0.08) (0.06) CPI 0.002-0.005-0.006*** (0.00) (0.00) (0.00) Imp CPI 0.004-0.009-0.003 (0.01) (0.01) (0.00) Exp CPI -0.022*** -0.014-0.001 (0.01) (0.01) (0.01) RQ 0.011 0.071** 0.064*** (0.02) (0.03) (0.02) Imp RQ 0.093-0.145** -0.090 (0.08) (0.07) (0.06) Exp RQ -0.058-0.272** -0.147* (0.11) (0.13) (0.09) GDP/POP -0.026*** 0.038 0.087*** -0.031*** 0.012 0.063*** (0.00) (0.03) (0.01) (0.01) (0.02) (0.02) Dependency 0.498*** 0.284*** 0.347*** 0.425*** 0.255*** 0.324*** (0.03) (0.10) (0.04) (0.04) (0.09) (0.05) Population 0.000*** 0.000*** 0.000*** 0.000*** 0.000*** 0.000*** (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) Constant 0.339*** -0.150-0.661*** 0.421*** 0.036-0.498*** (0.05) (0.27) (0.13) (0.05) (0.22) (0.15) Time FE x x x x x x Country RE x x Country FE x x Obs. 754 754 754 609 609 609 R-sq within 0.591 0.962 0.577 0.963 adj. R-sq 0.577 0.958 0.562 0.958 Standard errors in parentheses. Coefficients are significant at the 10 percent (* p<0.10), 5 percent (** p<0.05) or 1 percent (*** p<0.010) level. The dependent variable is the net-gini coefficient. Imp stands for import openness and Exp stands for export openness. CPI denotes Corruption Perception Index, which is our preferred measure of institutional quality. RQ denotes Regulation Quality. GDP/POP is a control for per capita GDP and Dependency stands for Dependency Ratio. 23

Motivated by this decomposition into the effects of imports and exports we ask, if both work into the same direction by lumping them together into openness defined as imports plus exports. Total trade openness is positively associated with inequality in the OLS regressions exclusively controlling for the time trend. Including random or fixed effects destroys significance of openness in all regressions for both the direct and the interaction effects of openness. The effect of the institutional measures itself is ambiguous. Further robustness checks As a robustness check we include both institutional variables, the CPI as well as RQ in our regression. The number of observations reduces due to some missing values in both variables. The results turn out to be robust against inclusion of additional variables and/or changes in the number of observations. The coefficients mainly replicate the results documented in Table (2). Economic globalization can be associated with lower inequality in regressions that identify the effects on both between and within variation of the data. However, the sign of the coefficients turn from negative to positive in the random and fixed effects regressions. All coefficients are significant. Corruption is significant only in the random and fixed effects specifications. The sign changes depending on the interaction terms included. Specification (1) to (3) include the interaction between KOFI and CPI, whereas specification (4) to (6) include the interaction between KOFI and RQ index. The critical level of the CPI index for which the marginal effect turns from positive to negative lies around 4.7 to 6.8 in regressions (2) and (3). In regressions (4) to (6) we find a critical value of RQ between 0.53 and 0.78. Those estimates are very much in line with our benchmark results. Thus, a further loss in observations does not change the overall picture of our results. 24

Table 7: Econometric results Imports+Exports (1) (2) (3) (4) (5) (6) Net Gini Net Gini Net Gini Net Gini Net Gini Net Gini b/se b/se b/se b/se b/se b/se Trade 0.046*** 0.000-0.009** 0.042*** 0.005-0.002 (0.01) (0.01) (0.00) (0.01) (0.01) (0.00) CPI 0.003-0.006** -0.006*** (0.00) (0.00) (0.00) Trade CPI -0.014*** -0.000 0.002* (0.00) (0.00) (0.00) RQ 0.023 0.049* 0.047** (0.02) (0.03) (0.02) Trade RQ -0.113*** -0.015 0.002 (0.02) (0.01) (0.01) GDP/POP -0.026*** 0.038 0.089*** -0.032*** 0.013 0.067*** (0.01) (0.03) (0.01) (0.01) (0.03) (0.02) Dependency 0.497*** 0.279*** 0.346*** 0.418*** 0.248*** 0.331*** (0.03) (0.10) (0.04) (0.04) (0.08) (0.05) Population 0.000*** 0.000*** 0.000*** 0.000*** 0.000*** 0.000*** (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) Constant 0.340*** -0.139-0.672*** 0.418*** 0.049-0.525*** (0.05) (0.27) (0.13) (0.05) (0.22) (0.15) Time FE x x x x x x Country RE x x Country FE x x Obs. 754 754 754 609 609 609 R-sq within 0.590 0.962 0.572 0.963 adj. R-sq 0.577 0.958 0.559 0.958 Standard errors in parentheses. Coefficients are significant at the 10 percent (* p<0.10), 5 percent (** p<0.05) or 1 percent (*** p<0.010) level. The dependent variable is the net-gini coefficient. Trade denotes import + export openness from the Penn World Table. CPI denotes Corruption Perception Index, which is our preferred measure of institutional quality. RQ denotes Regulation Quality. GDP/POP is a control for per capita GDP and Dependency stands for Dependency Ratio. As a further robustness check we estimate the impact of globalisation and institutional quality on redistribution, directly. In compliance with Gozgur and Ranjan (2015), redistribution is quantified by the difference between market and net-gini coefficient. Results can be found in table (6) and (7). The KOF 25

Table 8: Econometric results CPI and RQ (1) (2) (3) (4) (5) (6) Net Gini Net Gini Net Gini Net Gini Net Gini Net Gini b/se b/se b/se b/se b/se b/se KOF I -0.318*** 0.274** 0.170* -0.332*** 0.316** 0.211*** (0.08) (0.11) (0.09) (0.09) (0.13) (0.07) CPI 0.015 0.034** 0.010-0.005* -0.008** -0.007*** (0.01) (0.02) (0.01) (0.00) (0.00) (0.00) KOF CPI -0.026* -0.058*** -0.025* (0.02) (0.02) (0.01) RQ 0.068** 0.048* 0.053*** 0.218** 0.444*** 0.259*** (0.03) (0.03) (0.02) (0.09) (0.14) (0.07) KOF RQ -0.207-0.597*** -0.311*** (0.14) (0.22) (0.10) GDP/POP -0.003 0.015 0.074*** -0.002 0.019 0.076*** (0.00) (0.03) (0.02) (0.00) (0.02) (0.02) Dependency 0.337*** 0.335*** 0.402*** 0.335*** 0.311*** 0.391*** (0.04) (0.10) (0.06) (0.04) (0.10) (0.05) Population 0.000*** 0.000*** 0.000*** 0.000*** 0.000*** 0.000** (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) Constant 0.391*** -0.163-0.702*** 0.399*** -0.197-0.736*** (0.07) (0.26) (0.19) (0.07) (0.27) (0.19) Time FE x x x x x x Country RE x x Country FE x x Obs. 591 591 591 591 591 591 R 2 within 0.611 0.964 0.610 0.964 adj. R 2 0.598 0.960 0.597 0.960 Standard errors in parentheses. Coefficients are significant at the 10 percent (* p<0.10), 5 percent (** p<0.05) or 1 percent (*** p<0.010) level. The dependent variable is the net-gini coefficient. KOF I is our preferred economic globalization measure. CPI denotes Corruption Perception Index, which is our preferred measure of institutional quality. RQ denotes Regulation Quality. GDP/POP is a control for per capita GDP and Dependency stands for Dependency Ratio. Globalization Index exhibits a positive significant sign in all regressions. Hence, we can conclude that globalisation fosters redistribution, and thereby, support the empirical results of Gozgur and Ranjan (2015). This holds also controlling for the time trend for KOFI I, the economic globalisation index. The estimated 26

coefficient of the CPI is insignificant using KOF I but highly significant and positive estimating the model including KOF II. In contrast, the index for regulatory quality is highly significant and positive using time fixed effects in column (4) for both globalisation indices: Countries characterized by high regulatory quality are associated with higher levels of redistribution. It is not surprising, that the institutional quality measures are not significant controlling for between variation because the within variation of both indices does not fluctuate at a high magnitude over time. The same holds for the gini coefficients, and consequently, also for redistribution. Therefore, globalisation is the driving component analyzing the interaction between globalisation and institutional quality on within country inequality. The estimated sign of regulatory quality changes from positive to negative including country fixed effects. This result is counterintuitive, and requires further research. One possible explanation could be outliers. 27

Table 9: Regression results with redistribution as dependent variable (1) (2) (3) (4) (5) (6) redistr. redistr. redistr. redistr. redistr. redistr. b/se b/se b/se b/se b/se b/se KOF I 0.354*** 0.156 0.114** 0.318*** 0.159** 0.098* (0.03) (0.10) (0.06) (0.04) (0.07) (0.06) CPI 0.002* 0.002-0.001 (0.00) (0.00) (0.00) RQ 0.047*** -0.044-0.077*** (0.02) (0.03) (0.02) GDP/POP 0.013*** 0.019-0.029* 0.014*** 0.034*** -0.005 (0.00) (0.01) (0.02) (0.00) (0.01) (0.02) Dependency -0.013 0.143* 0.138*** -0.003 0.181** 0.212*** (0.03) (0.08) (0.04) (0.03) (0.08) (0.05) Population -0.000*** -0.000*** -0.000-0.000*** -0.000*** -0.000 (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) Constant -0.245*** -0.242*** 0.252* -0.241*** -0.364*** 0.035 (0.03) (0.09) (0.13) (0.03) (0.10) (0.18) Time FE x x x x x x Country RE x x Country FE x x Number of obs. 754.000 754.000 754.000 609.000 609.000 609.000 R-sq within 0.650 0.914 0.653 0.917 adj. R-sq 0.640 0.906 0.644 0.907 Standard errors in parenthesis. Coefficients are significant at the 10 percent, (* p<0.10) 5 percent (** p<0.05) or 1 percent (*** p<0.010) level. The dependent variable is redistribution, measured as difference between market and net Gini. KOFI is our preferred globalisation measure. CPI denotes Corruption Perception Index, which is aour preferred measure for institutional quality. RQ denotes Regulatory Quality. GDP/POP is a control for per capita GDP and Dependency stands for Dependency Ratio. 28

Table 10: Regression results with redistribution as dependent variable contd. (1) (2) (3) (4) (5) (6) redistr. redistr. redistr. redistr. redistr. redistr. b/se b/se b/se b/se b/se b/se KOF II 0.171*** 0.020 0.008 0.132*** 0.034 0.025 (0.02) (0.04) (0.02) (0.02) (0.04) (0.03) CPI 0.005*** 0.003-0.001 (0.00) (0.00) (0.00) RQ 0.084*** -0.037-0.074*** (0.02) (0.03) (0.02) GDP/POP 0.028*** 0.030*** -0.014 0.027*** 0.043*** 0.003 (0.00) (0.01) (0.01) (0.00) (0.01) (0.02) Dependencyl -0.062** 0.144* 0.145*** -0.030 0.173** 0.205*** (0.03) (0.08) (0.04) (0.03) (0.08) (0.06) Population -0.000*** -0.000*** -0.000-0.000*** -0.000*** -0.000 (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) Constant -0.247*** -0.256*** 0.183-0.253*** -0.370*** 0.016 (0.04) (0.09) (0.12) (0.04) (0.11) (0.17) Time FE x x x x x x Country RE x x Country FE x x Number of obs. 754.000 754.000 754.000 609.000 609.000 609.000 R-sq within 0.610 0.913 0.616 0.916 adj. R-sq 0.599 0.905 0.605 0.907 Standard errors in parenthesis. Coefficients are significant at the 10 percent, (* p<0.10) 5 percent (** p<0.05) or 1 percent (*** p<0.010) level. The dependent variable is redistribution, measured as difference between market and net Gini. KOFI I is the Economic Globalisation Index. CPI denotes Corruption Perception Index, which is aour preferred measure for institutional quality. RQ denotes Regulatory Quality. GDP/POP is a control for per capita GDP and Dependency stands for Dependency Ratio. 29