CORRUPTION AND INTERNATIONAL TRADE: A COMPREHENSIVE ANALYSIS WITH GRAVITY

Transcription

1 CORRUPTION AND INTERNATIONAL TRADE: A COMPREHENSIVE ANALYSIS WITH GRAVITY Salvador Gil-Pareja * Rafael Llorca-Vivero José Antonio Martínez-Serrano University of Valencia Abstract The aim of this paper is to analyze the impact of corruption on trade. To this end, we estimate gravity equations using a wide sample of countries for a long period of time and three different measures of corruption: Corruption Perception Index (CPI), Control of Corruption Index (CCI) and a Structural Corruption Index (SCI). Our results show that perception-based indexes (CPI and CCI) give quite different results with respect to the Structural index (SCI) and that the detected impacts are somewhat sensitive to the formation of regional trade agreements or the level of per-capita income of the trade partners. In particular, when using the CCI index and there is at least one middle or low income country in the pair a positive effect of corruption on trade appears. Moreover, the negative impact of differences in the level of corruption detected with the SCI index appears to be much more relevant when the two countries are high income. Keywords: Corruption, Gravity equation, Income level, International trade, Regional Trade Agreements. JEL code: F14, F15. Salvador Gil-Pareja, Rafael Llorca-Vivero and José Antonio Martínez-Serrano are Full Professors in Applied Economics at the Departamento de Estructura Económica, Facultad de Economía, Av. de los Naranjos s/n, C.P , Valencia, Spain. s: Salvador.Gil- Pareja@uv.es, Rafael.llorca@uv.es (corresponding author), Jams@uv.es Tel Fax This study is part of a research project financed by Ministerio de Economía y Competitividad (ECO R, partially funded by the European Regional Development Fund -ERDF) and Generalitat Valenciana (PROMETEO II/ ).

2 1.- Introduction Corruption in the public sector is probably one of the main elements of failures in governance (Acemoglu and Verdier, 2000) which directly affects institutional quality and, thereby, the well-being of the society. The World Bank and the International Monetary Fund share the opinion that corruption, as one of the determinants of the poor functioning of institutions, is a great obstacle to economic and social development. Diverse authors show how a higher degree of corruption negatively affects economic growth (Mauro, 1995), slows down the flow of foreign direct investment (Wei, 2000) or distorts tax revenues (Mauro, 1998). The uncertainty in the decision making by economic agents, the poor law enforcement affecting business agreements because of the insecurity around the legal solution to disputes or the inefficient allocation of resources are the most common arguments that are behind this adverse impact. However, other authors present corruption as a factor that, under certain circumstances, may facilitate economic exchanges thus improving efficiency (see, for example, Egger and Winner, 2005, for foreign direct investment or Dreher and Gassebner, 2013, for firm entry). The reason is that, in economic systems with extensive and complex regulations as well as weak institutions, bribes can help companies to avoid formal regulatory barriers. That is, corruption would serve as a mechanism for deregulation. It is the greasing the wheels hypothesis. Therefore, in general terms, there are reasons to believe that corruption can act in one or another sense and the real impact will depend on the conditions under which it operates, although on the set of empirical contributions the negative impact usually predominates over the positive one. In this topic, the relationship between corruption and international trade is a less explored area. The motives for expecting that the effect of 1

3 corruption on international trade flows is favorable or detrimental are similar to those outlined above. On the one hand, the improper functioning of the legal framework can hinder the effectiveness of the contracts, a fact that discourages international transactions by increasing the cost of exporting (Anderson and Marcouiller, 2002). For this reason, corruption would have an impact that is similar to the establishment of a tariff. On the other hand, if tariff and/or non-tariff barriers (for instance, complex administrative formalities) are high, bribes could facilitate trade since they would help to overcome these obstacles. In this sense, Dutt and Traca (2010) show that the positive effect of corruption on trade (evasion) will surpass the negative effect (extortion) in those markets in which tariffs are relatively high (over 25%). A reasonable corollary would be that corruption favors international trade in those cases in which destination markets present barriers that exceed a certain threshold. In general, the various existing empirical researches find a negative effect of corruption on international trade. This is the case of Levchenko (2007), focusing on institutional quality, Pomfret and Sourdin (2010), showing that direct trade costs increase with corruption, Zelekha and Sharabi (2012), discussing the case of Israel and Masila and Sigue (2010), on a sample of 47 African countries. On the other hand, Voraveeravong (2013) estimates a positive effect for ASEAN countries whereas Akbarian and Shirazi (2012) find an inverted U relationship between corruption and trade for a sample of countries of the Middle East and Latin America. This implies that, for such countries, an increase of corruption from relatively low levels would increase trade while the same phenomenon from relatively high levels would reduce it. Additionally, the phenomenon of corruption presents different aspects that can influence the degree in which it will affect international trade. This fact is highlighted by Thede and Gustafson (2009). These factors come from the seriousness of such conduct 2

4 (if the corruption is severe it can limit or even prevent transactions), its prevalence (which increases the costs of looking for an honest partner), its function (obstruction of market competition) or its predictability (the more predictable is the lower costs implies). Other authors also highlight several factors that it is necessary to take into account. For instance, De Jong and Bogmans (2011) outline the importance to determine if countries act as exporters or importers. Horsewood and Voicu (2012) argue that differences in the ethical standards of countries would also negatively affect trade and Knack and Azfar (2003) pose the problem of sample selection and bias resulting from those databases that incorporate a small number of countries. Our aim is to shed some additional light on this issue. To this end, we have made use of three different measures of corruption. Two of them belong to the so called perception-based indexes. The other is the result of a structural modelling which takes into account the causes and consequences of corruption across countries. The results obtained differ depending of the type of index used, the income group of the trading partners and the fact that the pair of countries eventually belong to a regional trade agreement. The rest of the paper is as follows. In the next section, we describe the methodology used. Section 3 offers the data sources. Section 4 presents the results and section 5 concludes. 2.-Methodology The vast majority of the aforementioned papers analyzing the impact of corruption on trade make use of the well know gravity model, which name proceeds from its analogy to the Newtonian physics. Gravity equations relate bilateral trade flows to the economic size of countries, the distance between them and other factors affecting trade 3

5 barriers as, for instance, the fact that the partners share a common language, are part of the same regional trade agreement, etc. 1 Since the appearance along the 2000s of diverse influent papers, there are some relevant points that it is necessary to take into account in the estimation of gravity equations in order to avoid misspecification issues. In particular, Anderson and van Wincoop (2003) outline the necessity to account for relative trade barriers by including in the estimation multilateral resistance terms, which may vary over time in panel data models (Klein and Shambaugh, 2006 or Baier and Bergstrand, 2007). Additionally, specific factors in the pair of countries (unobserved heterogeneity) influencing their mutual trade but which are invariant over time have to be also controlled for (Egger, 2002). The existence of a high volume of zero trade flows in extent samples 2 (Helpman, Melitz and Rubinstein, 2008) as well as econometric problems resulting from heteroscedastic residuals (Santos Silva and Tenreyro 2006 and 2010) are the further elements to be considered. With the above comments in mind and following the proposal of Santos Silva and Tenreyro (2006, and 2010) to control for zeroes in the dependent variable and heteroskedastic residuals, we run a non-linear Poisson estimator (Pseudo-Maximum Likelihood Estimation PPML) to the following gravity equation: XX iiiiii = ββ 0 + ββ 1 llllyy iiii YY jjjj + ββ 2 RRRRRR iiiiii + ββ 3 CCCC iiiiii +ββ 4 CCCCCCCCCCCC iiii + ββ 5 CCCCCCCCCCCC jjjj + λλ tt + μμ iiii + uu iiiiii (1) 1 The first authors to apply gravity equations to trade flows were Tinbergen (1962) and Pöyhönen (1963), although lacking a theoretical basis. Anderson (1979) first provided the theoretical background. Subsequent developments came from Bergstrand (1985 and 1989), Deardoff (1998), Evenett and Keller (2002), Anderson and van Wincoop (2003) or Helpman, Melitz and Rubinstein (2008), among others. 2 This is because the existence of non-observable firm heterogeneity and fixed trade costs in line with the so-called new-new trade theory (Melitz, 2003). 4

6 where i and j denote trading partners, t is time, and the variables are defined as follows: Xij are the bilateral export flows from i to j, Y denotes Gross Domestic Product (GDP), RTA is a binary variable which is unity if i and j belong to the same Regional Trade Agreement, CU is a binary variable which is unity if i and j belong to the same Currency Union, Corrup is an index of corruption (our variable of interest), λt are time dummies, μμ iiii are country-pair fixed effects and uijt is the standard classical error term. 3 Ideally, our PPML estimations should incorporate into the regressions countryyear fixed effects (accounting for variations in multilateral resistance over time) and country-pair fixed effects (controlling for unobserved heterogeneity). However, two serious problems appear. Firstly, country-specific corruption variables are perfectly collinear with country-year fixed effects. Secondly, and more important, the PPML estimation with country-year and country-pair fixed effects does not reach convergence in any of the cases. Therefore, we have to make a choice. Our decision is to estimate by including country-pair fixed effects. As stated by Baltagi et al. (2015, pp. 624 and 630): The lion s share of the variance in log bilateral trade flows xijs is contributed by the exporter-importer (time-invariant) component and Empirically, country-pair effects explain much more of the variation in bilateral exports or imports, foreign direct investment, or migration than country-time effects.. Moreover, country-pair fixed effects adequately control for the main sources of endogeneity (see Baier and Berstrand, 2007 or Gil-Pareja et al., 2016). The cost of this decision is to obviate the fact that multilateral resistance may change over time, which probably involves some bias. However, from the above analysis it seems that this is not really an important shortcoming. 3 Obviously, the effect of country-pair dyadic variables (distance, common language, etc.) are controlled for bilateral fixed effects. 5

7 Apart from country-specific corruption variables we have incorporated into the regression two additional pair-specific variables. The first one is the summation of the values of the respective corruption variable for the exporter and the importer (we let S- Corrup). It would reflect something like the whole corruption in the pair. Obviously, in this case we are not able to distinguish trade flows between, for instance, two middle corrupt countries and a high corrupt country and a low one. The second index correct this shortcoming. It is the square of the difference between the corruption indices for the exporter and the importer (let D-Corrup). Knack and Azfar (2003) have found that trade is reduced as far as the differences in this institutional characteristic widens. As noted before, some papers highlight the fact that corruption have a positive effect in countries with high barriers to trade because the evasion effect exceeds the extortion effect. One of the most striking events in the last two decades at international scale is the massive proliferation of regional trade agreements all over the world, regardless of their degree of development. This fact provides valuable information. The hypothesis to test is whether corruption affects more those countries that form trade agreements among them, since presumably only the negative aspects (misallocation of resources, obstacles to competition, etc.) are prevalent, than those countries which hold high regulations among themselves. If this is the case, a jointly priority would be liberalization and eradication of corruption. Therefore, a second equation to estimate is the following: XX iiiiii = ββ 0 + ββ 1 llllyy iiii YY jjjj + ββ 2 RRRRRR iiiiii + ββ 3 CCCC iiiiii +ββ 4 CCCCCCCCCCCC iiiiii + ββ 5 CCCCCCCCCCCC iiiiii RRRRRR iiiiii + λλ tt + μμ iiii + uu iiiiii (2) 6

8 where the interaction of the pair-specific corruption variable with the dummy for RTAs will capture the impact of liberalization on the relationship between corruption and trade. 3.- Data The export flows from country i to country j come from the Direction of Trade (DoT) dataset, which is constructed by the International Monetary Fund (IMF). The data comprise bilateral merchandise trade between 139 countries and territories (see Table A1) over the period GDP data are from the World Development Indicators (World Bank). The indicators of Regional Trade Agreements (RTA) have been built using data from the World Trade Organization, the Preferential Trade Agreements Database (The Faculty of Law at McGill University) and the website The indicators of currency unions are taken from Reinhart and Rogoff (2002), CIA's World Factbook, and Masson and Pattillo (2005). The sample includes around 200 regional trade agreements (plurilateral and bilateral) and 17 currency unions. In order to construct our variable of interest we have resorted to three different sources. The first one is the well known Corruption Perception Index (CPI) from Transparency International. This index is built since 1995 and goes from 0 (a highly corrupt country) to 10 (a very clean country) although the scale have changed in last years from 0 to As defined in the website, countries score depending on how corrupt their public sectors are seen to be. That is, it is based on values given by people with founded opinions as analyst, businesspeople or experts in the area. Nowadays the index 4 Obviously, we have used the same scale for all the period. 7

9 is available for the majority of countries in the world but at the outset there were only 41 countries ranked. The second measure of corruption is the Control of Corruption Index (CCI) which is elaborated in the context of the project Worldwide Governance Indicators from the World Bank. It combines data from a set of other sources (for instance CPI) and, as such, it is based on the views of informed people and public and private organizations. The range of values goes from -2.5 (very bad governance) to +2.5 (very good governance) for around 200 countries and territories and for the years 1996, 1998, 2000 and 2002 to Finally, we resort to the (what we call) Structural Corruption Index (SCI) by Dreher et al. (2007). Contrary to perception-based indexes, which according to diverse authors can be misleading, this index considers corruption as a latent variable connected to its causes and consequences. Therefore, the respective values are obtained basing on much more objective factors than just opinions which, additionally, are probably slow moving over time even in the event of important social changes at the respective countries. The index is constructed for around 100 countries for the periods , , and In our panel data sample we have decided to assign the respective value to the year that is in the middle of each period, that is, we have data for the years 1978, 1983, 1988 and Contrary to that occurs with the other two indices (CPI and CCI) in this case higher values imply worse performance. For instance, for the period the least corrupt country (Switzerland) has a value of whereas the most corrupt country (Guinea-Bissau) presents a value of Results 8

10 The results for the estimations of equation (1) for the three measures of corruption considered are in Table 1. First of all, given that higher values for the CPI and CCI indexes imply better performance and the contrary occurs with the SCI index, in our regressions we have decided to change the sign of the two former variables for a clearer understanding of the parameters estimated. In this sense, a negative value of the parameter for the corruption variable in all the three cases implies an adverse impact of corruption on trade flows and the contrary occurs with a positive value. As usual, the gravity equation works well: trade flows increase with countries incomes and the fact that these countries eventually share a regional trade agreement or a currency union. 5 When using the CPI index (column 1) we observe that for both, the exporter and the importer, higher levels of perceived corruption are related with lower volumes of trade flows. The respective parameters are negative and significant at the 1% level. In order to have an idea of the magnitude of this effect we consider the values for the best performance 6 (-10) and the worse one (-0.4). Doing this, our calculations reveal that the potential loss (ceteris paribus) for being a totally corrupt country with respect to being a totally clean country is around 36% for the exporter and 63% for the importer. Things change a little when considering the CCI index. In this case, only the parameter for the importing country is significant. Acting in the same way as before (in this case the best performer shows a value of and the worse one a value of 2.06) the estimated loss in trade for the importer is around 77%, close to the impact observed with CPI. However, things change a lot in the estimations with the structural index (SCI). As it is observed, now trade flows increase with corruption for exporter and importer countries, although in this last case the parameter is only significant at the 10% level. Doing the same 5 The differences observed between the regressions with CPI and CCI with that with SCI are dependent on important differences in sample size, especially with the time period considered. 6 Remember that lower values of the corruption variables means lesser corruption in our regressions. 9

11 calculations as before, we obtain that the most corrupt country would export (ceteris paribus) 65% more than the less corrupt one and this figure is going to be 33% for importer countries. Although the number of observations for the regression with SCI is much smaller than in the other two cases, this result seems strange and needs for a deeper analysis. Before doing this, we are going to analyze the results obtained with the pairspecific variables. The corresponding parameters estimated are in Table 3. With respect to the summation of the values of the index for the exporter and the importer (S-Corrup) the pattern is the same as before: the higher the level of corruption in the pair the lower the trade flows between partners when measuring corruption by the CPI and the CCI indexes. The contrary occurs when we use the SCI index. However, when considering differences in the level of corruption we detect no impact using as regressors the perception-based indexes whereas a negative (and significant) effect is observed in the case of the structural index. That is, to the extent that the two countries in the pair differ in their level of corruption their mutual trade is lower. In particular, for the biggest difference in the sample between a pair of countries trade decreases by 35%. The next step is to check the premise about the influence of regional trade agreements on the relationship between corruption and trade. The estimations of equation (2) for the summation (S-Corrup) and the difference (D-Corrup) measures are in Table 3. Again, the results obtained with the CPI and CCI measures clearly differ from those with the SCI index. With respect to the first measure, the estimations show that the negative effect of corruption on trade is lower for the pair of countries belonging to a regional trade agreement, contrary to our expectations. That is, the elimination of barriers between a pair of countries also induces the drastic reduction of corruption as a barrier to trade. However, no impact of regional trade agreements in the relationship of interest is 10

12 detected for the SCI index. Relating to our second corruption pair-specific measure (D- Corrup) regional trade agreements only affect the impact of corruption on trade with the SCI variable (column 6) and in a very strong positive way. One interesting point is whether the described pattern depends on the level of income of countries. To this end, we have used interactive dummies considering, on the one hand, only those pairs formed exclusively by trade flows among high income countries (high-income) and, on the other hand, those formed by the rest of possible combinations (rest of combinations). That is, the pairs among each high income and the rest (middle and low income countries) and the pairs among middle and low income countries. In order to group countries by level of income we have resorted to the World Bank data source. 7 The list of countries is in Table A1 in the Appendix. The first point to consider is whether there are relevant differences in the level and variability of the corruption variables between both samples. The descriptive statistics for the three corruption measures are in Table 4. As observed, the sample containing all the combinations of countries income except that between high-income countries is much larger. As expected, the means of the respective variables reflect that developed countries are less corrupt on average, a fact that is also reflected in the range of the maximum and minimum values. However, the standard deviations are similar in all the cases implying that differences in countries corruption across pairs are also relevant even between highincome countries. Therefore, there is not a priori reasons to expect a different impact of corruption on trade across the two considered samples unless this variable would impact in a different way the cost of trading between countries depending on the level of income. 7 See 11

13 As observed, the consideration of the two alternative samples have some consequences on the estimated parameters affecting our variable of interest. On the one hand, when considering the CPI measure, the negative impact appears for corrupt imported countries in the high income sample but for corrupt exported countries in the rest of combinations sample. However, the summation (S-Corrup) measure gives similar results in both samples. Another interesting difference is that, in the case of high income countries, the existence of a regional trade agreement between the pair of countries has a negative impact on mutual trade if the partners differ in their level of corruption. That is, it seems that trade liberalization eliminates the evasion effect. The most radical changes occur when considering the CCI measure, for which the results obtained for the high income case are qualitatively identical to that of the CPI measure. As observed, a positive and significant impact of corruption in the exporting country on trade is detected in the estimations with the rest of combinations interactive dummy. That is, in this case the greasing wheels hypothesis is accomplished. This positive effect is also reflected in the summation pair-specific dummy. Finally, we obtain interesting results when using the SCI variable. As observed, the positive effect of corruption on trade occurs in both types of combinations of countries. For the high income case this impact takes place when countries act as exporters and importers but the significance is only at 10% level. For the rest of combinations case the impact takes place only for countries acting as exporters and the estimated coefficient almost doubles that for high income and is significant at 5%. The summation dummy is positive in both cases but the contrary occurs: the level of significance is higher for the high income case than for the rest of combinations case, reflecting in this last case the non-significance for importers. 12

14 In both cases ( high income and rest of combinations ) differences in the level of corruption in the pair of countries have a negative impact on bilateral trade flows. However, the magnitude is much bigger for high income countries implying that the loss of trade for the larger difference in corruption across these partners in the sample is around 75% whereas the value drops to around 25% for the alternative group considered. This negative effect vanishes when countries form RTAs, especially for the rest of combinations case for which it seems that differences in the level of corruption favors trade if countries belong to a regional trade agreement. Apart from the time period considered (we can do nothing) one of the important differences between the estimations with the CPI and CCI indexes with respect to the SCI index lies in the number of countries considered in the sample. In the two first cases we have (depending on the year) till 139 countries, in the last case (also depending on the year) just 85 countries, which are those with at least two years observations (in bold in Table A1). Therefore, as an additional robustness check we have repeated the estimations with the CPI and CCI indexes but the same sample of countries as in the case of the SCI index. As before, the first estimations are done with dummies that consider all the possible combinations of countries. The latest estimations divide these dummies considering, one the one hand, only high income countries and, on the other hand, the rest of combinations. As it can be seen, things do not change in a relevant manner and some of the most interesting conclusions are reinforced. First, when considering the impact of corruption on international trade for pairs of only high income countries the results are qualitatively (and almost quantitatively) the same with the sole exception of the loss of significance of the parameter which corresponds to the interactive dummy of differences in the level of corruption with RTAs in the CPI case. Second, in the rest of combinations case there is also one exception which corresponds to the negative and significant impact of 13

15 differences in corruption levels across countries with the CPI measure, in line with the result obtained with the SCI measure. More interesting, when considering the CCI index the detected positive impact of corruption on trade for exporter countries is now strongest and with a high level of significance as before (also positive for the importer country but non-significant). This is also the case for the corresponding pair-specific corruption dummy. The above comments are reflected in the estimations aggregating all the possible combinations. The two main differences with respect to consider the whole sample of 139 countries are the following: For the CPI case, differences in levels of corruption between countries negatively impact trade, for the CCI case, the pair-specific variable for summation of levels of corruption in the pair, although with the same negative sign, losses its significance level. 5.- Conclusions The effects of corruption on international trade flows are ambiguous. On the one hand, corruption may act as a barrier to trade increasing the cost of doing international business. On the other hand, corrupt agents may facilitate trade in countries with strict regulations (greasing the wheels hypothesis). To determine which factor dominates is an empirical matter. In this paper, we have used three different measures of corruption in order to analyze this point. Two of them, the Corruption Perception Index (CPI) and the Control of Corruption index (CCI), belong to the known as perception-based indexes. The third (SCI), constructed by Dreher et al. (2007), comes from a structural modelling which takes into account causes and consequences of corruption. 14

16 The results show that differences appear in the estimated impact depending on the measure of corruption used. In general, we observe that corruption negatively impacts trade when perception-based indexes are used whereas the contrary occurs when we alternatively include into the regression the structural index. Additionally, in this last case, differences in corruption levels in the pair of trading partners negatively affect their trade flows. These results have to be qualified when we take into account the level of percapita income of countries or the fact that they form part of a regional trade agreement. In particular, when using the CCI index and there is at least one middle or low income country in the pair a positive effect of corruption on trade appears. Moreover, the negative impact of differences in the level of corruption detected with the SCI index appears to be much more relevant when the two countries are high income. In general terms, regional trade agreements counterbalance the negative effect of corruption on trade flows but there are some exceptions. References Acemoglu, A. and Verdier, T. (2000): The choice between market failures and corruption, American Economic Review 90, pp Akbarian, R. and Shirazi, H. (2012): The effect of corruption on trade volumes of seleted countries in the Middle East and Latin America ( ), Quarterly Journal of Quantitative Economics 8, pp Anderson, J.E. (1979): A theoretical foundation to the gravity equation, American Economic Review 69, pp Anderson, J. and Marcouiller, D, (2002): Insecurity and the pattern of trade: An empirical investigation, Reiew of Economics and Statistics 84, pp

17 Anderson, J.E. and van Wincoop, E. (2003): Gravity with gravitas: A solution to the border puzzle, American Economic Review 93, pp Baier, S. L. and Berstrand, J. H. (2007): Do free trade agreements actually increase members international trade?, Journal of International Economics 71, pp Baltagi, B. H., Egger, P. and Pfaffermayr, M. (2015): Panel data gravity models of international trade in Badi H. Baltagi (Ed.) The Oxford Handbook of Panel Data, Oxford University Press. Bergstrand, J.H. (1985): The gravity equation in international trade: some microeconomic foundations and empirical evidence, Review of Economics and Statistics 67: Bergstrand, J.H. (1989): The generalised gravity equation, monopolistic competition, and the factor proportions theory in international trade, Review of Economics and Statistics 71: Deardoff, A.V. (1998): Determinants of bilateral trade: does gravity work in a neoclassic world?, in: Frankel, J. (Ed.), The Regionalization of the World Economy, University of Chicago Press, Chicago. De Jong, E y Bogmans, C. (2011): Does corruption discourage international trade?, European Journal of Political Economy 27, pp Dreher, A. and Gassebner, M. (2013): Greasing the wheels? The impact of regulations and corruption on firm entry, Public Choice 155, pp Dreher, A., Kotsogiannis, C. and McCorriston S. (2007): Corruption around the world: Evidence from a structural model, Journal of Comparative Economics 35, pp Dutt, P. and Traça, D. (2010): Corruption and bilateral trade flows: Extortion or Evasion?, Review of Economics and Statistics 92, pp

18 Egger, P. (2002): An econometric view on the estimation of gravity models and the calculation of trade potentials, The World Economy 25, pp Egger, P. and Winner, H. (2005): Evidence on corruption as an incentive for foreign direct investment, European Journal of Political Economy 21, pp Evenett, S.J. and Keller, W. (2002): On theories explaining the success of the gravity equation, Journal of Political Economy 110: Gil-Pareja, S; Llorca-Vivero, R. and Martínez-Serrano, J. A. (2016): The effect of nonreciprocal preferential trade agreements on benefactors exports, Empirical Economics, DOI /s y. Helpman, E., Melitz, M. and Rubinstein, Y. (2008): Estimating trade flows: trade partners and trade volumes, Quarterly Journal of Economics 123, pp Horsewood, N. and Voicu, A. M. (2012): Does corruption hinder trade for the new EU members, Economics 6, pp Klein, M.W., Shambaugh, J. C. (2006): Fixed exchange rates and trade, Journal of International Economics 70, pp Knack, S. and Azfar, O. (2003): Trade intensity, country size and corruption, Economics of Governance 4, pp Levchenko, A. (2007): Institutional quality and international trade, Review of Economic Studies 74, pp Masila, J. W. and Sigue, S. P. (2010): Corruption and Internationa Trade: An empirical investigation of African countries, The World Economy 33, pp Mauro, P. (1995): Corruption and growth, Quarterly Journal of Economics 110, pp Mauro, P. (1998): Corruption and the composition of government expendture, Journal of Public Economics 69, pp

19 Melitz, M.J. (2003): The impact of trade on intra-industry reallocations and aggregate industry productivity, Econometrica 71, pp Pöyhönen P (1963): A tentative model for the volume of trade between countries. Weltwirtschaftliches Arhiv 90, pp Pomfret, R. and Sourdin, P. (2010): Why do trade costs vary?, Review of World Economics 146, pp Santos Silva, J. M. C. and S. Tenreyro, (2006): The log of gravity, The Review of Economics and Statistics 88, pp Santos Silva, J. M. C. and S. Tenreyro, (2010): On the existence of the maximum likelihood estimates in Poisson regression, Economics Letters 107, pp Thede, S and Gustafson, N. (2009): The multifaceted impact of corruption on international trade, International Trade Working Papers, Lund University. Tinbergen J (1962): Shaping the World Economy. Twentieth Century Fund, New York. Wooldridge J (2010): Econometric analysis of cross section and panel data. The MIT Press, Cambridge, Massachusetts, 2nd edn. Wei, S. (2000): How taxing is corruption on inernational investors, Review of Economics and Statistics 82, pp Voraveeravong, P. (2013): Corruption impacts on bilateral trade between ASEAN countries during 2006 to 2011: Gravity model Approach, World Journal of Social Sciences 3, pp Zelekha, Y. and Sharabi, E. (2012): Corruptions, institutions and trade, Economics of Governance 13, pp

20 Table 1: Estimations of the impact of corruption on trade PPML-CPFE Variables CPI CCI SCI LnY ity jt (0.02) *** (0.023) *** (0.037) *** RTA ijt (0.047) ** (0.037) ** (0.037) *** CU ijt (0.044) *** (0.045) *** (0.158) Corrup it (0.011) *** (0.032) (0.153) *** Corrup jt (0.012) *** (0.034) *** (0.123) * Observations 176, ,782 16,456 Sample period: for CPI, 1996, 1998, 2000 and 2002 to 2012 for CCI and 1978, 1983, 1988 and 1994 for SCI. Notes: Regressand: log of real bilateral exports. Robust standard errors (clustered by country-pairs) are in parentheses. *significant at 10%; **significant at 5%; ***significant at 1%. CPFE: country pair fixed effects. CPI: Corruption Perception Index. CCI: Control of Corruption Index. SCI: Structural Corruption Index. 19

21 Table 2: Estimations of the impact of corruption on trade. PPML-CPFE Variables CPI CCI SCI CPI CCI SCI LnY ity jt (0.025) *** (0.023) *** (0.036) *** (0.024) *** (0.023) *** (0.033) *** RTA ijt (0.047) ** (0.039) ** (0.036) *** (0.048) ** (0.039) ** (0.051) *** CU ijt (0.044) *** (0.045) *** (0.155) (0.044) *** (0.044) *** (0.167) S-Corrup ijt (0.009) *** (0.025) ** (0.089) *** D-Corrup ijt (0.001) (0.006) (0.113) ** Observations 176, ,782 16, , ,782 16,456 Summation of the indexes of corruption in the pair (S-Corrupt) and Squared difference of the indexes of corruption in the pair (D-Corrup). Sample period: for CPI, 1996, 1998, 2000 and 2002 to 2012 for CCI and 1978, 1983, 1988 and 1994 for SCI. Notes: Regressand: log of real bilateral exports. Robust standard errors (clustered by country-pairs) are in parentheses. * significant at 10%; ** significant at 5%; *** significant at 1%. CPFE: country pair fixed effects. CPI: Corruption Perception Index. CCI: Control of Corruption Index. SCI: Structural Corruption Index. 20

22 Table 3: Estimations of the impact of corruption on trade. PPML-CPFE Variables CPI CCI SCI CPI CCI SCI LnY ity jt (0.026) *** (0.023) *** (0.037) *** (0.024) *** (0.023) *** (0.033) *** RTA ijt (0.099) ** (0.043) *** (0.059) *** (0.052) ** (0.043) ** (0.030) *** CU ijt (0.043) *** (0.046) *** (0.154) (0.044) *** (0.044) *** (0.169) S-Corrup ijt (0.009) *** (0.027) *** (0.089) *** S-Corrup ijt* RTA ijt (0.008) *** (0.022) ** (0.083) D-Corrup ijt (0.002) (0.007) (0.113) ** D-Corrup ijt* RTA ijt (0.002) (0.011) (0.169) *** Observations 176, ,782 16, , ,782 16,456 Summation of the indexes of corruption in the pair (S-Corrupt) and Squared difference of the indexes of corruption in the pair (D-Corrup). Interaction of corruption with Regional Trade Agreements S-Corrup* RTA and D-Corrup* RTA. Sample period: for CPI, 1996, 1998, 2000 and 2002 to 2012 for CCI and 1978, 1983, 1988 and 1994 for SCI. Notes: Regressand: log of real bilateral exports. Robust standard errors (clustered by country-pairs) are in parentheses. * significant at 10%; ** significant at 5%; *** significant at 1%. CPFE: country pair fixed effects. CPI: Corruption Perception Index. CCI: Control of Corruption Index. SCI: Structural Corruption Index. 21

23 Table 4: Summary Statistics for corruption variables Variable Index Sample Obs. Mean Std. Dev. Min. Max. Corrup it Corrup jt Corrup it Corrup jt Corrup it Corrup jt S-Corrup ijt S-Corrup ijt S-Corrup ijt D-Corrup ijt D-Corrup ijt D-Corrup ijt CPI CCI SCI CPI CCI SCI CPI CCI SCI High income Rest High income Rest High income Rest High income Rest High income Rest High income Rest High income Rest High income Rest High income Rest Notes: High income means that both countries in the pair are high-income countries. Rest are the complementary pairs of High-income. CPI: Corruption Perception Index. CCI: Control of Corruption Index. SCI: Structural Corruption Index. 22

24 Table 5: Estimations of the impact of corruption on trade by the level of development. PPML - CPFE Only High Income countries Rest of combinations Variable CPI CCI SCI CPI CCI SCI Corrup it (0.018) (0.042) (0.175) * (0.010) *** (0.041) ** (0.275) ** Corrup jt (0.017) *** (0.041) *** (0.188) * (0.015) * (0.046) (0.307) S-Corrup ijt (0.013) *** (0.030) *** (0.089) *** (0.010) *** (0.035) * (0.246) * D-Corrup ijt (0.002) (0.013) (0.162) *** (0.002) (0.007) (0.116) * S-Corrup ijt (0.014) *** (0.031) *** (0.087) *** (0.009) *** (0.039) (0.251) * S-Corrup ijt* RTA ijt (0.008) *** (0.021) *** (0.077) (0.012) *** (0.032) (0.269) D-Corrup ijt (0.002) (0.013) (0.154) *** (0.002) (0.008) (0.118) ** D-Corrup ijt* RTA ijt (0.003) *** (0.020) ** (0.211) *** (0.003) (0.012) (0.189) *** Observations 176, ,782 16, , ,782 16,456 Notes: Regressand: log of real bilateral exports. Robust standard errors (clustered by country-pairs) are in parentheses.* significant at 10%; ** significant at 5%; *** significant at 1%. CPFE: country pair fixed effects. CPI: Corruption Perception Index. CCI: Control of Corruption Index. SCI: Structural Corruption Index. 23

25 Table 6: Estimations of the impact of corruption on trade for the CPI and CCI measures using the same sample of countries as for the SCI measure. PPML - CPFE All combinations Only High Income countries Rest of combinations Variables CPI CCI CPI CCI CPI CCI Corrup it (0.012) *** (0.038) (0.019) (0.048) (0.011) *** (0.049) *** Corrup jt (0.013) *** (0.041) *** (0.020) *** (0.049) *** (0.013) *** (0.050) S-Corrup ijt (0.009) *** (0.031) (0.015) *** (0.038) *** (0.010) *** (0.040) *** D-Corrup ijt (0.001) ** (0.007) (0.003) (0.019) (0.001) ** (0.007) S-Corrup ijt (0.011) *** (0.034) (0.017) *** (0.040) *** (0.011) *** (0.046) ** S-Corrup ijt* RTA ijt (0.009) *** (0.025) (0.009) *** (0.023) ** (0.012) ** (0.031) D-Corrup ijt (0.001) ** (0.008) (0.003) (0.022) (0.002) ** (0.009) D-Corrup ijt* RTA ijt (0.002) (0.011) (0.004) (0.029) * (0.003) (0.012) Observations 86,493 90,736 86,493 90,736 86,493 90,736 Notes: Regressand: log of real bilateral exports. Robust standard errors (clustered by country-pairs) are in parentheses. * significant at 10%; ** significant at 5%; *** significant at 1%. CPFE: country pair fixed effects. CPI: Corruption Perception Index. CCI: Control of Corruption Index. SCI: Structural Corruption Index. 24

26 High Income Australia Austria Bahrain Belgium-Luxembourg Canada Chile China Hong Kong China Macao Croatia Cyprus Czech Republic Denmark Equatorial Guinea Estonia Finland France Germany Greece Iceland Ireland Israel Italy Japan Korea Kuwait Latvia Lithuania Malta Netherlands New Zealand Norway Oman Poland Portugal Qatar Russia Saudi Arabia Singapore Slovak Republic Slovenia Spain Sweden Switzerland United Arab Emirates United Kingdom United States of America Uruguay Appendix A Table A1: List of Countries Albania Algeria Angola Argentina Armenia Azerbaijan Bangladesh Belarus Benin Bolivia Bosnia and Herzegovina Brazil Bulgaria Burkina Faso Burundi Cambodia Cameroon Cape Verde Central African Republic Chad China Mainland Colombia Congo, D.R. Congo, Republic of Costa Rica Côte d Ivoire Djibouti c(l) Dominican Republic Ecuador Egypt El Salvador Gambia Georgia Ghana Guinea Guinea Bissau Guatemala Guyana Haiti Honduras Hungary India Indonesia Iran Jamaica Jordan Kazakhstan Middle and Low Income Kenya Kyrgyz Republic Lebanon Macedonia Madagascar Malaysia Mali Mauritania Mexico Mongolia Morocco Mozambique Nepal Nicaragua Niger Nigeria Pakistan Panama Paraguay Peru Philippines Romania Rwanda Senegal Serbia and Montenegro Sierra Leone South Africa Sri Lanka St. Tome and Principe Sudan Swaziland Syria Tajikistan Thailand Togo Tonga Tunisia Turkey Uganda Ukraine Venezuela Vietnam Yemen Zambia Zimbabwe Note: In bold the sample of countries used for estimations with the SCI measure. 25