Institutional heterogeneity in GATT interventions: a panel data analysis. Suryadipta Roy. (Work in progress; do not cite)

Transcription

1 Institutional heterogeneity in GATT interventions: a panel data analysis By Suryadipta Roy (Work in progress; do not cite) JEL Classification: F10, F13, C23. Keywords: Gravity; corruption; interaction; count data, GATT/WTO. 1

2 1. Introduction What is the role of domestic institutions for the success of trade liberalization programs? Are strong institutions important for the General Agreement of Tariffs and Trade (GATT) Treaty to be effective in fostering trade between countries? Do differences in the quality of institutions among the member countries explain the differences in the trade performance of these countries upon joining the GATT? This paper focuses on the above questions. In particular, we investigate the effect of the interaction between the GATT-WTO membership and a measure of the quality of domestic institutions in a country- the level of corruption perceptions, on the volume of bilateral imports. The objective of this paper is twofold: (i) To understand how the treatment effect of the GATT-WTO depends on the level of domestic corruption; and (ii) the other is to examine whether countries upon joining the GATT, either in the capacity of an importer or an exporter, imported less from or exported less to countries with higher levels of corruption. We explore the above questions using panel data on bilateral imports 1 for *** importing and exporting countries obtained from the International Monetary Fund- Direction of Trade Statistics (IMF-DOT), and a measure of corruption perceptions between 1984 and 2010 from the International Country Risk Guide (ICRG) ratings. (Results from Poisson pseudo-maximum-likelihood (PPML) estimator indicate heterogeneity in the impact of corruption on the GATT. High corruption exporters are able to generate smaller volumes of exports compared to the low corruption exporters 1 Following Subramanian and Wei (2007), we focus on imports by an importer i from exporter j, and use this as the dependent variable in our regressions. This is because trade effects from the GATT and the WTO really stem from the reduction of trade barriers by the importer country that should impact the amount of imports by importer i (or exports from exporter j). 2

3 after joining the GATT. However, we do not find any statistically significant impact of corruption in the partner countries on the trade performance of countries once they join the GATT. Our results suggest that strong institutions are necessary for successful trade reform, and there exists substantial heterogeneity in the relationship between institutions and trade liberalization across countries.) Arguably, one of the most comprehensive and over-arching interventions for multilateral trade liberalization till date has the GATT and its successor, the WTO. In one of the most heavily cited papers to study the effect of multilateral trade liberalization, Rose (2004) found that membership in the GATT or the WTO did not have any significant impact on bilateral trade among the member countries. This counter-intuitive result has spawned a huge literature that has examined the effectiveness of the GATT and the WTO in promoting trade from different angles. For example, Subramanian and Wei (2007) suggested various sources of asymmetry in the GATT/WTO system, and found associated patterns of bilateral trade in the data that supported their hypotheses. In particular, their paper suggests an asymmetry between the industrially developed countries and the low income developing countries in the implementation of the multilateral trade liberalization program that explained the lack of success of the GATT in promoting trade among the latter group of countries. Our paper focuses on a different source of asymmetry- the variation in the quality of institutions among the member countries, and the resulting impact of this difference on bilateral trade. In general, countries with stronger institutions (e.g. reflected in lower levels of corruption) are likely to be more active in reciprocal trade liberalization, and implement the tariff cutting measures (as well as drawing down of non-tariff barriers) more seriously than the ones 3

4 with poorer institutions (and higher corruption levels). Thus the GATT-WTO is expected to be more effective in the former group than in the latter group of countries. Similar to Subramanian and Wei (2007), our conjecture is that in viewing the GATT-WTO as a treatment for promoting global trade, it is important to take the heterogeneity among the countries into consideration and that this heterogeneity is responsible for the lack of an average treatment effect of the GATT as was the case in Rose (2004). To this end, we focus on the role of institutional heterogeneity to explain the difference in the impact of GATT in promoting trade between countries. Adjusted Predictions of importer_gatt#honest_imp Adjusted Predictions of exporter_gatt#honest_exp 0 1 Importer joining GATT-WTO 0 1 Exporter joining GATT-WTO honest_imp=0 honest_imp=1 honest_exp=0 honest_exp=1 Figure 1(i) Figure 1(ii) 4

5 Adjusted Predictions of exporter_gatt#honest_imp Adjusted Predictions of importer_gatt#honest_exp 0 1 Exporter joining GATT-WTO 0 1 Importer joining GATT-WTO honest_imp=0 honest_imp=1 honest_exp=0 honest_exp=1 Figure 1(iii) Figure 1(iv) The importance of corruption perceptions in determining the effect of the GATT- WTO on bilateral trade can be gleaned from Figures 1(i) - 1(iv), which show the association between logarithm of bilateral imports and an interaction term between the corruption measure used in the study and the incidence of an importer or an exporter country joining the GATT. Countries have been classified into two categories- those with poor institutional quality measured by above-average corruption levels, and the ones with high quality institutions as reflected in the below-average corruption levels. The results are based on predicted values of bilateral imports of a pooled OLS regression of the logarithm of bilateral imports on a dummy variable when a country joins the GATT either as an importer ( importer _ gatt 1) or as an exporter (exp orter _ gatt 1) ; a dummy variable that takes the value of one if a country has lower-than-average corruption level- ( honest _ imp 1) if importer, and ( honest _ exp 1) if exporter, and 5

6 an interaction term between these categorical variables. 2 According to Figs. 1(i) and 1(ii), the direct effect of the quality of domestic institutions is clearly important for the success of the GATT in the member countries, either as an importer or as an exporter. Countries with below-average corruption levels (i.e. corruption levels below the sample mean) have experienced significantly greater bilateral trade compared to those with above-average corruption levels (i.e. those with corruption levels above the sample mean), with the latter group experiencing a reduction in their bilateral trade after joining the GATT. On the other hand, based on Figs. 1(iii) and 1(iv), the treatment effect of a country joining the GATT does not seem to depend significantly on the corruption level of its trading partner. There seems The DID treatment effect of joining the GATT for the above-average corrupt partner country appears to be similar to that of a below-average corrupt partner (although there is a difference in the levels of bilateral imports between the two groups). In the paper, we undertake a more rigorous analysis of the interaction effects using a gravity framework. The remainder of the paper is organized as follows. Section 2 provides the theoretical motivation for the study; Section 3 provides a brief literature review; Section 4 lays out the estimation procedure; Section 5 describes the data and the major variables of interest, Section 6 discusses the results and the major findings of the paper; and Section 7 concludes. 2 Note that the approach is conceptually similar to a difference-in-difference (DID) estimation strategy where we estimate the average imports for two groups of countries, i.e. whether the country is a GATT member as an importer or exporter (the treatment variable), and compare that for countries with above-average and below-average corruption levels. As suggested by Puhani (2012), the coefficient of the interaction term in the above regression can be interpreted as the treatment effect of a country joining the GATT. 6

7 2. Theoretical motivation It is widely agreed that the quality of domestic institutions matter greatly for international trade 3. Strong and supportive institutions are supposed to reduce the uncertainty involved in international transactions, and thereby reduce the transactions cost of trade. On the contrary, high levels of corruption increase the uncertainty about the size of expected gains from carrying out an international transaction, and thereby deter international trade. Anderson and Marcoullier (2002) viewed corruption and lack of contract enforcement as components of transactions costs that raised insecurity in international transactions, and resulted in lower trade between developed and developing countries. In the same lines, de Groot et al. (2004) reported that lower levels of corruption increased bilateral trade. Dutt and Traca (2010) contrasted between a trade-taxing extortion effect and a trade-enhancing evasion effect associated with corruption, and found that while corruption impedes trade in an environment of low tariffs, it may actually be have a trade-creating effect when nominal tariffs are in excess of 25% (less than 15% of their observations). Thus, given the widespread evidence of the adverse impact of corruption on trade and the perceived role of the GATT as a tool for promoting global trade, it is reasonable to ask if it has been more effective in countries with lower corruption levels compared to those with high corruption levels. In order to answer to this question, we motivate the theoretical foundations of the multilateral trading system as has been developed by Bagwell and Staiger (1999, 2002, 2011). According to their viewpoint, multilateral trade liberalization pursued under the aegis of the GATT/WTO allow governments to escape from a terms-of-trade-driven prisoner s dilemma situation 3 See World Trade Report (2004) for a survey on the topic. 7

8 that develops when countries attempt to set domestic welfare maximizing tariffs in a unilateral manner. The rationale behind their theory of trade liberalization under GATT and the WTO is therefore referred to as the terms-of-trade theory of trade agreements. By jointly liberalizing trade according to the principle of reciprocal reduction of tariffs, governments are able to achieve lower tariffs and higher trade volumes without requiring any country to experience a loss in its terms-of trade and domestic welfare. This reciprocity principle when accompanied by the Most Favored Nation (MFN) principle of nondiscrimination among the member countries of the GATT, help to preserve the interests of the early trade partners in the negotiation process, and promote trade liberalization across partners over time. Thus, principles of reciprocity and nondiscrimination have been the cornerstone behind multilateral trade liberalization pursued under the GATT. Let us now attempt to provide an intuitive explanation of the impact of institutional asymmetry and its interaction with the membership in the GATT-WTO. We begin by discussing the direct effect of domestic corruption on an importer or an exporter joining the GATT. Suppose there are two importing countries with different levels of corruption that have joined the global trade organization. Since the purpose of the GATT is to promote trade between countries by reducing trade barriers, most notably tariff protection, both the signatory countries should record higher level of imports. However, if corruption exerts a debilitating effect on trade (e.g. due to bribes collected by the customs officials at the border that increases the c.i.f. prices of the imported commodity and reduces the demand for imports), then trade liberalization is likely to generate greater imports in the less corrupt country compared to the more corrupt country. On the other 8

9 hand, for two exporting countries with varying corruption levels that have joined GATT, reciprocal trade liberalization and the MFN member status that comes from the membership is likely to open up foreign markets for exports. However, the more corrupt exporter country is likely to face higher transactions costs (see Musila and Sigue (2010) for a discussion on the effect of corruption on exports in general and in case of Africa in particular) that increases the f.o.b. prices of its exports and makes it less competitive in the foreign market. Thus, we anticipate that exporters with lower corruption levels are likely to benefit more from joining the GATT, and be able to generate higher export levels compared to those with higher corruption levels. Does an importer or an exporter joining the GATT/WTO affects its pattern of trade with a high corruption trading partner relative to a low corruption trading partner? While the GATT/WTO does not have any explicit anti-corruption proposal of its own, it is an open question if joining the trade organization has any indirect impact on the domestic institutions of the member countries that either facilitate or impede trade with partners with high corruption levels. Thus, if trade liberalization under the GATT is accompanied by greater transparency, e.g. due to streamlining of customs procedures among the GATT members, then firms from both the highly-corrupt and the less-corrupt countries will find it easier to undertake business with the GATT members, either as an exporter or an importer. In such a scenario, there will be no difference in trade performance between the two groups of countries when they engage in trade with a GATT member country (as suggested in Figs. 1(iii) and 1(iv)). On the other hand, if the GATT/WTO has a built-in anti-corruption agenda, then the GATT members might be less inclined to engage in trading relationships with high corruption countries. In such situations, the GATT will lead to the member countries substituting 9

10 their trade away from the highly corrupt countries towards low corruption countries. In the following sections, we undertake rigorous empirical analysis to uncover these effects. 3. Literature review As discussed in the previous section, our paper contributes to the vast literature on the effectiveness of the GATT-WTO in promoting bilateral trade. However, we focus on an aspect of multilateral trade liberalization that has been relatively less studied in the literature- the institutional heterogeneity among the GATT members as an explanation for the diverse performance of the GATT in promoting global trade. D Souza (2012) has investigated the effect of the 1997 OECD Anti-bribery Convention on bilateral exports. Given that the Convention has an explicit agenda against corruption and attempts to reduce the incidence of bribery in international trade, it is expected to increase the transactions costs of doing business between the firms in the OECD countries and firms from high corruption countries. The results indicate that countries implementing the program were able to export less to importers with higher corruption levels. In our paper, we seek to find out if there has been a similar indirect effect of an importer or exporter joining the GATT-WTO on bilateral imports. Jansen and Nordas (2004) have studied the issue of the importance of domestic institutions for successful trade liberalization. They found that domestic tariff reduction led to greater bilateral trade only when it is complemented with strong institutions. On a similar note, Cheptea (2007) found that improvement in institutions in the Central & Eastern European (CEE) countries generated as much bilateral trade between these countries and the existing members of the EU as in case of removal of tariff and non-tariff barriers. In comparison to these studies, we 10

11 undertake an impact evaluation of the GATT on bilateral trade, and try to understand situations in which the GATT has been more or less effective due to the pernicious effect of corruption on bilateral trade. 4. Estimation procedure In order to assess the impact of corruption on the performance of the GATT, we employ the workhorse of empirical international trade, namely the gravity model. Our approach takes into account some of the recent developments in the literature as they relate to various aspects of the model and the estimation procedure. First, we use countrypair fixed effects in our gravity specifications on bilateral trade. This allows us to control for multilateral resistance terms in the gravity equations as discussed in Anderson and van Wincoop (2003). Second, the analysis uses bilateral trade data over time. This has the advantage of mitigating the bias generated by heterogeneity across countries, since it allows us to control for country-pair heterogeneity by using the pair fixed effects. Third, we deviate from the standard way of estimating the log-linear version of the gravity model and instead estimate the model in levels. This is motivated by the presence of a large number of zero trade flows between countries in bilateral trade data series as is evident from Figure 2 which plots a histogram of the volume of bilateral imports (of less than $10,000). Given that by taking the logarithm of zero trade flows, these observations get dropped from the analysis thereby leading to the loss of potentially useful information on the effect of trade costs on bilateral trade (viz. if the lack of trade relationship is due to prohibitive trade costs, or if the countries decide not to participate in trade due to high levels of fixed costs of entering into trade relationships). Hence we implement the 11

12 (pseudo) Poisson maximum likelihood estimator (PPML) estimator that can be applied to the levels of trade and thereby avoid dropping of zero trade observations. As argued by Santos Silva and Tenreryo (2006), in the presence of heteroskedasticty, the PPML is a robust estimator. Another attractive approach of the PPML is that the estimator will remain consistent under the assumption that the conditional mean distribution is correctly specified, although the data need not be distributed as Poisson (Cameron and Trivedi, 2005). One shortcoming of using the Poisson distribution is the restrictive assumption of equal dispersion, i.e. the mean is equal to the variance, which is often violated with most datasets (as in our case). However, given that the violation of the assumption of equal variance does not affect the estimates of the coefficients but only causes the estimated standard errors of the coefficients to be downward biased, we employ a variance estimator that is robust to correlation within country pairs Value of imports in current US dollars Logarithm of Bilateral Imports Figure 2 The main econometric specification implements a Poisson pseudo-maximumlikelihood (PPML) estimator on the baseline specification: Bilateral t t t Im portij aij at ijzij 1 Im porter _ GATT * Corruption _ importer t i 12

13 2 t Im porter _ GATT * Corruption _ exp orter j t 3 Exporter _ GATT * Corruption _ importer i t t + 4Exporter _ GATT * Corruption _ exp orter j + ij (1) where, t Bilateral Im port ij is the volume of bilateral imports into importer i from exporter j in a particular year t; t Z ij are the vector of control variables that explain bilateral trade in a gravity framework. These include (logarithms of) real GDP, population, real exchange rate for the importer and the exporter countries; (logarithm of) distance between the two countries; and a set of dummy variables that reflect the country characteristics, e.g. common legal origin, common currency, if the two countries share a common border, if there was a colonial relationship between the two countries, if the two countries share any regional trading agreement (as articulated under Sec. XXIV of the GATT), and if a developed country has accorded a GSP treatment to a developing country trade partner. ij denotes the country-pair fixed effects that control for multilateral trade resistance in our gravity regressions. This addresses what Baldwin and Taglioni (2006) refer to as the gold medal mistake in the gravity regressions which fail to account for the multilateral trade resistance terms and thereby suffer from omitted variable bias in that the omitted terms might be correlated with the trade costs. Finally, the year fixed effects in the regressions ( ) corrects for the problem of incorrectly deflating nominal trade data and t the GDP data by an aggregate price index (e.g. US dollars). It is to be noted that the specification employed in the current study substantially addresses the major problems in the gravity models as has been discussed in Baldwin and Taglioni (2006). According to these authors, studies that estimate the gravity model without including the multilateral trade resistance terms discussed in Anderson and van Wincoop (2003) suffer from 13

14 omitted variable bias since the omitted terms might be correlated with trade costs. Moreover, given that the gravity equation is theoretically derived from a modified expenditure function, the value of bilateral trade refers to the value of spending by one country on the goods produced by another country. Studies that estimate gravity models by averaging the trade flows between two partners instead of treating uni-directional trade flows thus are subject to measurement error in the dependent variable. Finally, deflating the value of trade flows by an aggregate price index (e.g. US dollars) creates further bias in estimation, viz. if the price deflator includes non-traded goods prices. Our study addresses all the above estimation issues. First, we account for the multilateral resistance terms by including country-pair fixed effects in the regressions. Second, we use the value of bilateral imports as the dependent variable. Third, the inclusion of the year fixed effects addresses the problem of inappropriate deflation of trade flows. 5. Data sources and summary statistics Descriptive statistics for the major variables of interest are reported in Table 1. The observations are based on 76 importing and exporting countries that have been included in the final regressions, i.e. we have non-missing data on all the explanatory variables for these countries. The choice of countries was driven by data availability for the major variable of interest in the study, i.e. corruption. Our measure of corruption ( Corr it ) is from the International Country Risk Guide (ICRG) published by the Political Risk Services (PRS) Group. This is an assessment of corruption within the political system reflecting demand for special payments and bribes to obtain regular official services (financial corruption) as well as other distortionary practices like government 14

15 patronage or special favors for select business interests (political corruption). Both forms of corruption distort economic and financial incentives, reduce efficiency, and introduce political instability. The index ranges from 0-6 with higher values denoting very low corruption; for interpretation purposes, this was rescaled such that higher values denote greater corruption. Corruption data are available for 174 countries during which is broader than the other commonly used sources of corruption data like the Transparency International or the World Bank s Worldwide Governance Indicators; however the data coverage is not uniform for all countries. In order to measure the heterogeneity of the effect of the GATT/WTO on an importer or an exporter joining the GATT/WTO, we use two indicator variables (Importer-GATT/ Exporter-GATT) that takes the value of one once the importing or the exporting country has joined the GATT/WTO in any year t, and a value of zero before it has joined the GATT/WTO. For example, given that Albania joined the GATT/WTO for the first time on 8 September, 2000, the above two dummy variables takes the value of zero for Albania during , and of one during On the other hand, since the United States (US) has been a member of the GATT since 1 January, 1948, the dummy variables take the value of one for the US throughout the sample period. The dummy variables suggest substantial participation in the GATT. For example, in our sample, around 88% of the countries have become members of the GATT over the years (until 2010). The dependent variable in the regressions is the nominal value of bilateral imports (in current US dollars) for any year available from the IMF Direction of Trade Statistics. As discussed in the previous section, the import data are strongly skewed to the right 15

16 (about 16% of the observations involve zero trade flows between countries). Table 1 also suggests that the distribution of bilateral imports displays over-dispersion since the variance of imports is much greater than the mean, i.e. greater variance than might be expected in a Poisson distribution. Data on real Gross Domestic Product (in constant 2000 US dollars), country population, and the real exchange rates have been obtained from the World Bank World Development Indicators for different years. The distance measure, obtained from the CEPII, is the measure of distance between most populated cities. All the control variables have been found to be important determinants of bilateral trade in previous studies, e.g. Rose (2004), Subramanian and Wei (2007), D Souza (2012) among others. These include a binary variable if the importer and exporter country use a common currency at time t; a dummy variable that indicates if the trading partners share a common border; if the countries have ever been involved in a colonial relationship; a binary dummy variable that indicates that the trading partners belong to the same regional trade agreement at time period t; a dummy variable indicating if the exporter country has been a GSP beneficiary or vice versa at time period t; and a dummy variable indicating if the countries have a common official language. As mentioned previously, we focus on the interaction between the dummy variable indicating the incidence of a country joining the GATT-WTO, and the country-level measure of corruption perceptions to understand the interplay between governance and institutions. 16

17 6. Results 6.1 Log-linear model We begin by estimating the traditional log-linear version of the specification laid out in equation (1). In each specification, we interact the GATT-WTO treatment dummy variable with the measure of corruption perceptions to investigate the impact of corruption on trade, and to understand if this effect is dependent on countries being members of the GATT-WTO. A significant negative (/ positive) interaction term between the GATT-WTO treatment dummy for a country (either as exporter or importer) and its domestic corruption level (where higher values of the corruption index denote greater corruption) would suggest that compared to the non-gatt-member countries, domestic corruption levels had a greater (/ lesser) negative impact on the ability of the country to generate bilateral exports or allow imports from the partner country. On the other hand, a significant negative interaction term between the GATT-WTO indicator variable for a member country and the corruption level in its trading partner would suggest that the impact of the GATT-WTO on bilateral trade is dependent on partner countries corruption levels, i.e. a GATT importer (/ exporter) might import less from (/ export less to) countries with higher corruption levels. Based on the results for the panel fixed effects regression for the log-linear model in column (1), we find that for the reference category i.e. the non-gatt countries (Importer-GATT = 0, Exporter-GATT = 0), one standard deviation increase in domestic corruption levels in the importing country is associated with approx. 13% (=exp(-0.14)-1) reduction in bilateral imports while a similar increase in corruption in the exporting country increases bilateral imports by about 13% (=exp(0.12)-1). However, compared to 17

18 the non-gatt countries, the GATT importers import about 13% more (=exp(0.12)-1) for every one standard deviation increase in domestic corruption. The GATT-WTO treatment dummy is found to be positive and highly significant in case of both the importers and the exporters- being in the GATT is associated with 16% greater import volume and about 31% greater export volume for the member countries. Turning to the specification in column (2) that tests for the dependence of the impact of GATT on partner s corruption levels, we observe that for every standard deviation increase in the exporters corruption levels, bilateral imports for the GATT member countries is found to be approx. 17% greater compared to that for the non-gatt importers. The interaction term between the Exporter-GATT dummy variable and the importer country corruption level is found to be not significant. Moreover, both the GATT-WTO treatment variables remain positive and highly significant. Moving on to the comprehensive specification in column (3), the results for the major variables of interest, i.e. the GATT treatment dummy variables, the importer- and exporter-level corruption indexes, and their interaction terms remain broadly unchanged from the stand-alone specifications in columns (1) and (2), except that exporter-level corruption is found to be not significant. In order to better interpret the results in the presence of the different interaction terms, Fig. 3 plots the predicted values of bilateral imports from the comprehensive specification in col. (3). The graph at the top left-hand corner reveals a divergence between the GATT-member countries and the non- GATT countries in the impact of domestic corruption on imports- while corruption, in general, had a negative impact on trade, this detrimental effect has been larger in the non- GATT countries compared to the countries in the GATT (steeper slope for the non- GATT importing countries). The graph at the bottom left-hand corner suggests an even 18

19 starker difference between the two groups in terms of the impact of exporter-level corruption on bilateral trade. Greater corruption in the exporting countries has been associated with higher imports for the GATT-member countries and lower levels of imports for the non-gatt importers. These predictions from the log-linear gravity model suggest that, compared to the non-members, the GATT has been relatively successful in promoting greater imports among its members as functions of corruption levels, both domestic as well as in the partner countries over the years. We do not, however, observe any significant difference between the two groups of countries in the role of the GATT as an exporting country. From the top right-hand corner of Fig. 3, higher levels of corruption in the importing countries led to reduction in bilateral exports for both the GATT- and the non-gatt exporters. Moreover, as illustrated in the bottom right-hand corner of the graph, the non-gatt countries have had more success in generating exports compared to the GATT members as functions of their domestic corruption levels. The coefficients of the control variables in the comprehensive specification mostly have the expected signs. Real income for the importing countries is found to be an important predictor of bilateral trade while real income for the exporter is found to be not significant. Being part of a currency union, involvement in regional trading arrangements, and GSP specialized treatment from the industrially advanced economies to the developing countries are found to significantly increase bilateral trade. Real exchange rates for both the importing and the exporting countries are found to be not significant. To examine the effects of other traditional gravity variables which are collinear with the country-pair fixed effects and hence get dropped from the regression (e.g. if the country is landlocked, distance, contiguity, common language, and colonial linkage between the 19

20 trading partners), we implemented a pooled OLS specification with importer, exporter, and year fixed effects. While this specification does not address all the criticisms associated with the estimation of a theoretically-consistent gravity model, it does account for time-invariant country-specific factors and global trends. The results on these pairspecific control variables are mostly statistically significant and are in line with the existing literature. We used the Akaike Information Criterion (AIC) and the Bayesian Information Criterion (BIC) as metrics for model selection (comprehensive vs. standalone models). We find that both criteria select the comprehensive model over the standalone models, and the fixed effects models are always selected over the pooled OLS specification. 6.2 Poisson model Results for the Poisson models are reported in Table 3. Cols. (5) and (6) report the results for the country-pair and year fixed effects regressions for the stand-alone models involving the direct and the indirect interaction terms between the GATT-WTO treatment dummy and the country-level measures of corruption respectively, Col. (7) reports the results for the comprehensive model, and Col. (8) reports the results for the comprehensive model for the pooled Poisson regression. For comparison purposes, we focus on the results for the comprehensive specification using the country-pair and year fixed effects for the two models. We observe significant differences in results between the log-linear version in Table 2 and the levels equation in Table 3. First, the Poisson regressions suggest greater role for the GATT-WTO membership on bilateral trade compared to the log-linear models. For example, while the Importer-GATT dummy variable is associated with 36% greater imports in the Poisson regression (compared to 20

21 23% for the log-linear model), the Exporter-GATT dummy variable is associated with 47% more exports compared to 32% for the log-linear specification. These findings are in lines with Westerlund and Wilhelmsson (2011) who observed substantial downward bias associated with the estimates of the treatment dummy variables in their log-linear specifications (both in the truncated least squares model, as well as with log(1 + Bilateral Imports) as the dependent variable). Second, there are important differences in the signs and the levels of significance for the interaction terms between the two models. For instance, the interaction term between the Exporter-GATT dummy and the corruption level in the exporting country is found to be negative and highly significant in the Poisson model. One standard deviation increase in corruption among the GATTexporting countries is found to reduce bilateral exports to their trading partners by 15%. On the other hand, contrary to the log-linear model, importer-level corruption is not associated with any positive impact on bilateral imports among the GATT-importers in the Poisson model. Similarly, given that the interaction terms between the GATTtreatment dummy and partner-country corruption levels were found to be not significant in the PPML regression, we find no evidence of any positive impact of corruption levels in the partner countries on the bilateral trade levels among the GATT members. The differences in the findings are also reflected in Fig. 4. While the GATT member countries have higher bilateral trade levels compared to their non-gatt counterparts, there does not seem to be any significant difference in the impact of importer-level corruption between the two groups. The major difference between the two groups seems to be reflected by the impact of exporter-level corruption upon the GATT members (relative to the non-gatt member countries). The bottom left-hand corner graph suggests that 21

22 countries after joining the GATT-WTO as importers registered more imports from partners with higher corruption levels while bilateral imports for the non-gatt countries went down with higher corruption levels in their trade partners. This effect is however found to be statistically not significant. The difference in the impact of domestic corruption levels among the GATT and the non-gatt exporters is, however, found to be highly significant as shown in the lower right-hand corner of Fig. 4. Domestic corruption seems to have a strong pernicious impact on bilateral exports, and thus can be posited to be an important explanation for the relative non-performance among the GATT member countries. Moreover, the results suggest that rather than reducing bilateral imports, corruption seems to work through the export channel by reducing the capability of the countries to engage in trading relationships with other countries. Thus policies that reduce domestic transactions costs of doing businesses through greater transparency and promoting strong governance are likely to be complementary to the role of the GATT as an institution for promoting global trade. Overall, the PPML estimator turns out to reject the null hypothesis less in comparison to the log-linear model- another finding that has been reported by Westerlund and Wilhelmsson (2011). One explanation for the difference in the results between the two models is that by dropping country-pairs with zero trade values between countries, the log-linear model creates a downward bias in the impact of corruption on bilateral trade. As has been discussed in Helpman, Melitz, and Rubinstein (2008), firms face varying costs of entering into export markets, and firms from the low-income and the developing countries face even higher costs due to their low quality institutions and higher corruption levels. Thus it is more likely for these groups of countries to report 22

23 missing trade observations. The traditional gravity models that use log-linear specifications would therefore non-randomly leave out observations that incorporate the hidden costs of poor institutions, e.g. corruption, on the propensity of countries to participate in trade that the Poisson correctly accounts for. Third, the Poisson estimates for the effects of the real GDP for the importing and the exporting countries on bilateral import are substantially different from the predictions of the log-linear model. For example, at the mean value of the regressors, the estimated elasticity of imports with respect to real GDP of the exporters and the importers are 0.51 (standard error = 0.16) and 0.05 (standard error = 0.16) respectively, while the corresponding values in the loglinear model are 0.31 (standard error = 0.21) and 2.15 (standard error = 0.22). That the estimated elasticities of imports are inelastic is in conformity with the stylized fact that trade-gdp ratio decreases with increasing GDP of countries, i.e. smaller countries tend to be more open to trade. The results with regard to the other traditional control variables in the gravity model 23

24 Predictive Margins of Importer-GATT Predictive Margins of Exporter-GATT Corruption-Importer Corruption-Importer Importer-GATT=0 Importer-GATT=1 Exporter-GATT=0 Exporter-GATT=1 Predictive Margins of Importer-GATT Predictive Margins of Exporter-GATT Corruption-Exporter Corruption-Exporter Importer-GATT=0 Importer-GATT=1 Exporter-GATT=0 Exporter-GATT=1 Figure 3 24

25 Predictive Margins of Importer-GATT Predictive Margins of Exporter-GATT Corruption-Importer Corruption-Importer Importer-GATT=0 Importer-GATT=1 Exporter-GATT=0 Exporter-GATT=1 Predictive Margins of Importer-GATT Predictive Margins of Exporter-GATT Corruption-Exporter Corruption-Exporter Importer-GATT=0 Importer-GATT=1 Exporter-GATT=0 Exporter-GATT=1 Figure 4 25

26 Import Log(Import) Importer-GATT Exporter-GATT Corruption-Importer (Standardarized) Corruption-Exporter (Standardarized) Log(RealGDP- Exporter) Log(RealGDPpercapita- Exporter) Log(RealGDP- Importer) Log(RealGDPpercapita- Importer) Log(RealExchange Rate-Exporter) Table 1: Summary statistics Definition N Mean Standard deviation Value of bilateral imports by importer i from exporter i in current US dollars Log of bilateral imports Member country joining GATT- WTO as importer Member country joining GATT- WTO as exporter ICRG corruption levelimporter ICRG corruption levelexporter Log of Real Gross Domestic Productexporter Log of Real Gross Domestic Productimporter Log of Real effective exchange rate index (2005 = 100) - exporter Min. Max e e e

27 Log(RealExchange Rate-Importer) Common currency RTA GSP Log(distance) Common language Contiguous Common colony Log of Real effective exchange rate index (2005 = 100) - importer Common currency for the importer and the exporter country Dummy variable indicating regional trade agreement Special GSP treatment accorded to exporter Distance between most populated cities Common official primary language Dummy variable indicating common border Dummy variable if country pair ever in colonial relationship

28 Table 2: Fixed effects log-linear model Dependent variable: log(bilateral Import) (1) (2) (3) (4) Fixed effects loglinear Fixed effects loglinear Fixed effects loglinear 1.importer_gatt 0.15** 0.27*** 0.21*** [0.05] [0.05] [0.06] [0.06] z_icrgm -0.14*** -0.09* -0.18** [0.04] [0.04] [0.06] [0.07] 1.importer_gatt #c.z_icrgm Pooled OLS loglinear 0.12** 0.10* 0.10* [0.04] [0.04] [0.05] 1.exporter_gatt 0.27*** 0.25*** 0.28*** 0.14* [0.06] [0.06] [0.06] [0.07] z_icrgx 0.12* [0.05] [0.04] [0.06] [0.06] 1.exporter_gatt #c.z_icrgx 1.importer_gatt #c.z_icrgx 1.exporter_gatt #c.z_icrgm [0.05] [0.05] [0.05] 0.16*** 0.15*** [0.04] [0.04] [0.05] [0.04] [0.04] [0.05] ln_realgdp_exp [0.14] [0.14] [0.14] [0.15] ln_realgdp_perc apita_exp 1.24*** 1.28*** 1.28*** 1.78*** [0.15] [0.15] [0.15] [0.16] ln_realgdp_imp 1.44*** 1.41*** 1.43*** 0.66*** [0.15] [0.15] [0.15] [0.15] ln_realgdp_perc ** apita_imp [0.14] [0.14] [0.14] [0.15] o.ln_dist -1.38*** ln_realexrate_i mp ln_realexrate_e xp [0.03] [0.03] [0.03] [0.03] [0.03] ** [0.02] [0.02] [0.02] [0.03] o.comlang_off 0.58*** [0.07] comcur 0.14** 0.12** 0.12** -0.23* [0.04] [0.04] [0.04] [0.12] o.contig [0.16] 28

29 o.colony 0.96*** [0.11] rta 0.43*** 0.42*** 0.42*** 0.18*** [0.04] [0.04] [0.04] [0.05] gsp 0.36** 0.35* 0.35* 0.88*** [0.14] [0.14] [0.14] [0.07] o.landlock -1.51** [0.50] _cons *** *** *** [3.90] [3.88] [3.90] [4.08] Country-pair fixed effects Year fixed effects Yes Yes Yes No Yes Yes Yes Yes Importer fixed No No No Yes effects Exporter fixed No No No Yes effects H0: log(real GDP- Exporter)=1 H0: log(real GDP- Importer)=1 N adj. R pseudo R 2 AIC 343, , , , BIC 343, , , , Robust standard errors adjusted for clustering within country-pair in brackets; + p<0.1, * p<0.05, **p<0.01, *** p<0.001; estimates for year and country-pair/importer/exporter fixed effects not reported. 29

30 Table 3: Dependent variable- Bilateral Imports (5) (6) (7) (8) Conditional Fixed Conditional Fixed Conditional Fixed Pooled Poisson effects Poisson effects Poisson effects Poisson 1.importer_gatt 0.32*** 0.36*** 0.36*** 0.45*** [0.08] [0.07] [0.09] [0.10] z_icrgm [0.03] [0.04] [0.05] [0.08] 1.importer_gatt #c.z_icrgm [0.03] [0.04] [0.04] 1.exporter_gatt 0.49*** 0.33*** 0.47*** 0.51*** [0.07] [0.07] [0.08] [0.09] z_icrgx 0.15*** exporter_gatt #c.z_icrgx 1.importer_gatt #c.z_icrgx 1.exporter_gatt #c.z_icrgm [0.04] [0.05] [0.06] [0.08] -0.16*** -0.15*** -0.17*** [0.04] [0.04] [0.05] [0.05] [0.05] [0.08] [0.04] [0.04] [0.07] ln_realgdp_exp 0.70* 0.72** 0.69* 0.35 [0.27] [0.28] [0.28] [0.30] ln_realgdp_perc apita_exp ** [0.26] [0.26] [0.26] [0.26] ln_realgdp_imp [0.23] [0.23] [0.23] [0.28] ln_realgdp_perc apita_imp ln_realexrate_i mp ln_realexrate_e xp 0.95*** 0.94*** 0.94*** 1.36*** [0.23] [0.23] [0.23] [0.25] 0.39*** 0.39*** 0.39*** 0.37*** [0.07] [0.07] [0.07] [0.07] 0.28*** 0.27*** 0.28*** 0.27*** [0.06] [0.06] [0.07] [0.06] comcur 0.21*** 0.20*** 0.21*** 0.16* [0.05] [0.05] [0.05] [0.06] rta 0.27*** 0.27** 0.27*** 0.71*** [0.08] [0.08] [0.08] [0.08] gsp [0.10] [0.10] [0.10] [0.16] ln_dist -0.57*** [0.04] comlang_off 0.06 [0.08] 30

31 contig 0.50*** [0.08] colony 0.34*** [0.10] landlock -3.35*** [0.83] _cons 1.30 [6.11] Country-pair Yes Yes Yes No fixed effects Year fixed effects Yes Yes Yes Yes Importer fixed No No No Yes effects Exporter fixed No No No Yes effects H0: log(real GDP- Exporter)=1 H0: log(real GDP- Importer)=1 N pseudo R AIC 7.70e e e e+13 BIC 7.70e e e e+13 Robust standard errors adjusted for clustering within country-pair in brackets; + p<0.1, * p<0.05, **p<0.01, *** p<0.001; estimates for year and country-pair/importer/exporter fixed effects not reported. 31