Appendices

The gravity model has long been a workhorse for analyzing the determinants of bilateral tradeflows. It has been extensively applied to analyze effects of policy-related as well as behind-the-border trade barriers, regional trading groups, and economic integration policies. The model also owes its popularity to a large extent due to its predictive power to explain bilateral tradeflows as measured by the goodness of fit. In its most intuitive version, the gravity model postulates that bilateral trade depends on the economic size of the trading partners, which reflects market size and purchasing power, and a measure of economic distance (or proximity) between countries to reflect trade costs. Some of the early theoretical foundations for the gravity model have been provided by Anderson (1979) using the Armington assumption whereby goods are differentiated by their country of origin and where consumers have preferences over all the differentiated products. Subsequently, Bergstrand (1985,1989) have shown that the gravity model can be derived from trade models based on monopolistic compe-tition where identical countries trade differentiated goods because consumers have a taste for variety. Deardorff (1998) has derived gravity-type expressions from the standard Heckscher–Ohlin trade model, and has shown that the gravity equation can be consistent with a large class of trade models. More recently, in a significant development, Anderson and van Wincoop (2003) have shown that controlling for relative trade costs is crucial for well-specified gravity models. Their theoretical results suggest that bilateral trade does not only depend on bilateral trade cost between two countries, but also on the relative weight of these costs with respect to the other trade partners (what they refer to as multilateral resistance). Thus, it is important to account for the multilateral resistance terms in the estimation of gravity models.

Given that we focus on the asymmetric impact of corruption between the high income and the low income countries toward their bilateral exports and imports in different industries, corruption in our model is viewed as a component of trade costs. Thus, we augment the traditional gravity model to include a number of country-specific and dyadic factors as components of trade costs. Besides the exporter-and importer-level corruption terms (Corrit and Corrjt), these include distance between the exporter and the importer country, whether the country pair belongs to a regional trade agreement or the existence of any preferential trading agreement, or whether the trade partners share a common language, border or colonial heritage with its trade partner. Our baseline log-linear gravity specification to be estimated for exports of sector k products from country i into country j in year tðXijt^kÞ is as follows:

ln X_ijt^k ¼ b^k₁ln Y_itþ b^k₂ln Y_jtþ b^k₃j ln Yit ln Yjtj þ d^k₁ln y_itþ d^k₂ln y_jtþ d^k₃j ln yit ln yjtj þ c^k1Corr_itþ c^k2Corr_jtþ c^k12Corr_it ln yitþ c^k21Corr_jt ln yitþ h^0kZ_ijtþ a^kijþ a^kt:

ð1Þ The above empirical specification is a reduced form equation that reflects both increasing returns to scale with monopolistic competition, and the factor proportions theory of trade. Exportflows are expected to be directly related to the market size of the trading partners reflected by the real gross domestic product (GDP) ðYit=YjtÞ. Relative factor endowments measured by real per capita GDP ðyit=yjtÞ are also important determinants of tradeflows. According to Markusen and Venables (2000), a large part of manufacturing trade is in the form of intra-industry trade between multina-tionalfirms from developed countries. More specifically, with positive trade costs, multinationals are more likely to exist when the countries are more similar in both relative and absolute endowments, and thereby substitute bilateral trade by horizontal foreign direct investment (FDI). Hence, we introduce the logarithm of the absolute difference in the GDP of country i and j to reflect the substitution away from bilateral trade toward FDI. We also introduce the logarithm of the absolute difference in the GDP per capita between the trading partners. This variable controls for factor endowment differences between countries, and can be treated as a proxy for wage differences between countries. a_ij-s are fixed effects capturing time-invariant country-pair-specific heterogeneity that control for multilateral resistance in the gravity model, and the at-s are year fixed effects capturing time-varying hetero-geneity. Zijt denotes the vector of pair-specific dummy variables referred to above that affect trade costs between countries. The baseline specification implements a pair (i.e., exporter–importer) fixed effects estimator using panel data for different indus-tries. While this has the advantage of mitigating the bias generated by heterogeneity across countries, the pairfixed effects absorb all heterogeneity that is constant over time, and hence, we cannot recover these estimates included in Zijt (e.g., distance, common language, etc.). In order to address this problem, we also estimate a random effects model proposed by Mundlak (1978). This model includes averages of the time variant variables to control for the remaining bilateral unobserved heterogeneity that is time invariant and could be correlated with the idiosyncratic error term. This model accounts for bilateral time-invariant heterogeneity that is correlated with the regressors and allows us to estimate the coefficients of the time-invariant variables.

The key terms in the above equation are the interaction terms between domestic and partner country corruption and per capita GDP, i.e., Corrit ln yitand Corrjt ln yitrespectively. From Eq.1,

@ ln Xijt^k

@Corrit ¼ c^k1þ c^k12ln yit ð1aÞ

For relatively low income countries, the c^k1 term dominates, and the effect of domestic corruption on bilateral exports will be determined by the sign ofc^k1. On the other hand, the effect of domestic corruption on bilateral exports for the high income countries will be determined byc^k12. Similarly, the effect of partner country corruption on bilateral exports can be obtained as

@ ln X_ijt^k

@Corrjt

¼ c^k₂þ c^k₂₁ln y_it ð1bÞ

Thus, the specification allows us to investigate if the effect of corruption on bilateral exports in different industries depends on the level of economic devel-opment proxied by GDP per capita. Given that advanced economies possess better quality institutions reflected in lower levels of corruption, stronger property rights, and better contract enforcement, they tend to have comparative advantage in products whose transactions costs are more sensitive to the quality of institutions (e.g., Berkowitz et al.2006; Nunn2007, Levchenko 2007; Ranjan and Lee2007).

Empirical evidence based on these papers suggest that the more institutionally intensive industries usually involve higher levels of product differentiation, are more “high-tech” in nature, use relatively higher amounts of capital inputs, and involve greater amount of relationship-specific investment in their production process (Rauch (1999)). Thus greater domestic or partner corruption is more likely to affect exports of institutionally intensive goods from the high income countries compared to the low income countries. Moreover, by controlling for differences in factor endowments between countries, the specification also isolates the indepen-dent effect of institutions on comparative advantage and thereby on bilateral exports. A similar specification is also estimated for imports of sector k products by country i from country j in year tðM_ijt^kÞ to understand the difference in the effect of domestic and partner institutions on the high income and the low income countries:

ln M^k_ijt¼~b^k₁ln Y_itþ ~b^k₂ln Y_jtþ ~b^k₃j ln Yit ln Yjtj þ ~d^k₁ln y_itþ ~d^k₂ln y_jtþ ~d^k₃j ln yit ln yjtj þ ~c^k₁Corr_itþ ~c^k₂Corr_jtþ ~c^k₁₂Corr_it ln yitþ ~c^k₂₁Corr_jt ln yitþ ~h^0kZ_ijþ ~a^k_ijþ ~a^k_t:

ð2Þ Given that the high-income countries tend to have comparative advantage on institutionally intensive industries, it is likely that they will import less institutionally intensive goods and the low income countries will import more institutionally inten-sive products. Hence greater domestic or partner corruption is more likely to affect the imports of institutionally intensive goods by the low income countries compared to the high income countries. However, given that the high income countries enjoy lower corruption compared to the low income countries, these countries are likely to have lower transactions costs and hence experience higher imports. Thus, the net asym-metric effect of corruption on imports between the high income and the low income countries in different sectors is more of an empirical question that we take to the data.