However, this is not always the case because the institutional context matters. The assumption that corruption rather ‘greases the wheel’ of trade in countries with poor institutional quality has a long history going back to Bhagwati (1982). Following this, I hypothesize that less-developed economies with high regulatory burdens (i.e. developing countries) corruption, unlike other malfunctioning institutions, can be beneficial for trade. The empirical analysis using a Zero-Inflated Negative Binomial and Poisson Pseudo-Maximum Likelihood estimation technique shows that in general corruption ‘sands the wheel’ for developed and developing countries. Interestingly, when institutional quality is controlled for using the ZINB model, the results suggest that in an environment with weak institutional quality, distortions create a context in which corruption has a trade enhancing effect.

Keywords: International trade, corruption, gravity model

(3)

2

The “mystery of missing trade” states that the actual volume of international trade is less compared to the prediction by economic theory (Trefler, 1995). It is argued that if trade were frictionless, then the observed trade volumes would be almost five times larger (Eaton & Kortum, 2002). This is of particular interest considering that international trade is widely accepted as an important element for fostering economic growth, particularly in emerging markets. Thus, barriers which hamper trade volumes, such as the quality of institutions in both importing and exporting countries are considered to be highly important. Overcoming such barriers may enhance trade flows and essentially solve the mystery of missing trade. The aim of this paper is to focus on the effects on trade flows of one particular aspect of institutional quality, namely corruption. Corruption is defined as the use of public office for private gain and is a form of rent-seeking (Thede & Gustafson, 2012). Hence, corruption can take the form of demands for special payments and bribes linked with import and export licenses, exchange controls, tax assessments, police protection, or loans (Charron, Lapuente, & Rothstein, 2010). Thus, corruption is considered to have impact on international trade.

The topic of corruption is particularly interesting because even though corruption is one of the main impediments for conducting business in the world, only limited empirical testing has been done with respect to the effects of corruption on trade (Anderson & Marcouiller, 2002, Zelekha & Sharabi, 2012). Moreover, corruption is different compared to other ‘bad’ institutions (i.e. political and market instability, red tape, inefficient contract enforcement and weak property rights) because unlike other malfunctioning institutions, corruption can be beneficial for trade in countries with an already distorted economy and where regulatory burdens are high (Bhagwati, 1982). Thus, the impact of corruption on trade flows is not straight forward, resulting in two opposing viewpoints which makes this topic especially appealing. On the one hand it is argued, that corruption has a diminishing effect on international trade volumes in environments characterized by high institutional quality (i.e.

effective rule of law, high property right enforcement and low tariffs) because it can increase insecurity, transaction costs and distort investments, i.e. ‘sands the wheels’ of trade (De Groot, 2004, Jansen & Nordas, 2004 and Shepherd, 2009). On the other hand it is argued, that in poor institutional environments characterized by defective governance practices, high tariff levels and extensive bureaucratic ‘red tape’, corruption can actually offset the effects of such distortions and thus lead to a trade enhancing effect (Parayno, 1999, Dutt & Traca, 2010 and de Jong & Bogmans, 2011). Bhagwati (1982) proposes that corruption includes a subset of rent-seeking activities which are directly unproductive, profit-seeking (DUP) activities with a respect to economic activity. While these activities are considered to be harmful in a strong institutional setting, Bhagwati (1982) argues that DUP activities may actually be welfare enhancing in weak institutional environments. More precisely

(5)

4 these activities ‘present ways of making a profit by undertaking activities which are directly unproductive’ (Bhagwati, 1982, p.989). Even though these theories do not directly analyze the impact of corruption on international trade flows, they show that corruption is a mean of promoting of trade, i.e. ‘greases the wheels’. Thus, rent-seeking may behave as a countervailing force to already existing distortions and indirectly as second-best practices contribute to an increase in international trade flows.

This paper aims to answer the question whether corruption ‘sands or greases the wheel’ of international trade. Furthermore, it is considered whether there is a difference in this relationship between developed and developing countries as well as between environments characterized by weak and strong institutional quality. With respect to existing research on causes and effects between trade and institutions, the causality problem remains a key issue. On the one hand, the ‘conventional’ view states that institutions matter for international trade and as such higher levels of corruption may affect the level of trade flows between country-pairs i.e. enhancing or diminishing the level of imports (Anderson & Marcoullier, 2002). On the other hand, a further approach indicates that higher volumes of trade may play a role in determining the level of corruption prevalent within an economy i.e. with an increase in imports, officials have a greater opportunity to demand more bribes which results in more corruption in a given country (Ades & di Tella, 1999). Moreover, it is also likely that the level of corruption as well as international trade flows respond simultaneously to an omitted factor for instance the legal framework or the historical evolution of a nation. This paper addresses the endogeneity problem by using an instrumental variable (IV) regression approach which allows for one-way causality to be established and accounts for omitted factors.

According to Zelekha & Sharabi (2012) there is only a limited amount of studies within the economic literature that examine the effects of corruption on world trade. Furthermore, most of the studies use the corruption index from Transparency International or the Worldbank. The problem with these indexes is that despite using fixed effects, their inconsistency over time could lead to biased estimates in case of a panel data study. Thus, this paper contributes to the field by studying the relationship between corruption and trade by making use of an enhanced corruption index from the Political Risk Service Group and addressing the issue of endogeneity.

The remainder of the paper is structured as follows: Section 2 provides a literature review on previously conducted studies; by first focusing on institutions and then narrowing down to corruption.

Next Hypotheses are formulated for further investigation. In Section 3 the method and data used in the empirical study are explained. Section 4 provides the main results and discussion of the empirical analysis. A conclusion of the paper and recommendations for further research is given in Section 5.

(6)

5 2 Literature Review

The role of institutions - the efficiency of judicial system in order to enforce contracts, security of property rights, investor protection, and the like is a widely discussed topic among researches.

Regarding the empirical evidence of Mauro (1995), Shleifer and Vishny (1997) and Acemoglu, Johnson and Robinson (2001) two important facts of institutions are pointed out. First of all, institutions matter significantly for the economic performance of a nation. According to North, institutions are “the underlying determinant of the long-run performance of economies” (North, 1990, p.107). Good quality institutions are fundamental factors for political and market stability, can reduce information asymmetry and decrease the risk regarding property rights as well as contract enforcement and thus have an important impact on economic activity. Secondly, institutions in developed countries (i.e. the North) are much better than in developing countries (i.e. the South) (Levchenko, 2007).

While there is a wide consensus regarding the effects of institutions on economic performance, the impact of institutions on international trade has only received limited attention (Méon, & Sekkat, 2006). Only recently, researchers started to emphasize that institutional quality plays a key role in determining international trade and production patterns (Gustafson & Thede, 2012). This view has been supported by Anderson and Marcouiller (2002) and Levchenko (2007) who empirically analyze that malfunctioning economic institutions reduce international trade.

Traditionally, economic theory has identified endowments, technology, consumer ‘love of variety’ as well as market competition as drivers of international trade. However, the “mystery of missing trade” states that the actual volume of international trade is less compared to the prediction by economic theory (Trefler, 1995). It is argued that if trade were frictionless, then the observed trade volumes would be almost five times larger (Eaton & Kortum, 2002). Thus, barriers which hamper trade volumes, such as the quality of legal, financial and other institutions in both importing and exporting countries are considered to be highly important (Nunn & Trefler, 2013).

Overcoming such barriers may enhance trade flows and essentially solve the mystery of missing trade. In fact, Nunn and Trefler (2013) show that domestic institutions are statistically and economically important determinants of comparative advantage even after controlling for traditional sources of comparative advantage (i.e. endowments and technology). In comparison to the traditional causes institutions run through different direct and indirect channels. Contracting institutions such as laws on the books, contractual flexibility (La Porta et al., 2008) and those institutions which are related to property rights, the labor market and financial development are direct sources of a countries comparative advantage. Considering the aforementioned fact, that institutions in developed countries

(7)

6 are better than those in developing countries, it could be said that relative to developing countries, developed countries have a comparative advantage concerning institutions.

The institutional context of a country is a tight net of formal institutions, informal institutions as well other distortions such as corruption. Corruption is different compared to other distortions and thus needs special attention. Following empirical and theoretical literature in line with Bhagwati (1982) there are two opposing views an how corruption essentially effects international trade, i.e. it has the ability to either ‘sand’ or ‘grease’ the wheels of trade which makes this topic especially appealing.

The effect of corruption on international trade

The rent-seeking theory was one of the early economic instruments established to model corruption in the public sector (Lambsdorf, 2006). The term “corruption” is often derived from a principal-agent theory while in this paper corruption is defined as the use of public office for private gain (Thede &

Gustafson, 2012). There has been an extensive debate on the role of corruption within the economy.

On the one hand, existing literature argues that global corruption negatively effects the volume and pattern of trade due to the reduction of the security of property rights and misallocation of resources i.e. it ‘it sands the wheels’ of trade. On the other hand, economic scholars continue to debate whether corruption may be beneficial in already distorted environments due to its availability to avoid democratic delays and overcome other aspects of defective governance (Bhagwati, 1982) i.e. “greases the wheels". Thus, corruption can facilitate imports and exports in the case when the quality of customs is low and the tariff structure is complicated (Lavallée, 2005).

Corruption "sands the wheels”

In general the social as well as economic effects of corruption for an economy are considered to be directly negative. “The [World] Bank has identified corruption as the single greatest obstacle to economic and social development. It undermines development by distorting the rule of law and weakening the institutional foundation on which economic growth depends” (The World Bank, 2009). During the Doha round of the World Trade Organization’s (WTO) negotiations, discussions focusing on trade facilitation were undertaken in order to promote transparency, decrease red tape and reduce corruption in customs. Considering these statements, there may be causes to expect that high corruption levels reduce trade flows between countries. As such the focus of the WTO to reduce corruption (i.e. in customs) can cause an increase in trade flows between countries (Dutt & Traca, 2010).

One of the earliest works illustrating these effects is that of Anderson and Marcouiller (2002) who indicate that corruption enhances the insecurity in international trade causing an increase in the

(8)

7 transaction cost which in turn leads to a trade decrease. The authors indicate that corruption generates a price mark-up which behaves as “hidden tax” increasing transaction costs and substantially distorting investments and reducing trade flows. Likewise De Groot et al. (2004) show that transaction costs increases in a more corrupt environment and hence the amount of trade declines.

Jansen and Nordas (2004) investigate the effect of changes in institutions on trade flows and conclude that a lower level of corruption causes a higher integration in the world economy as well as that countries import less from more corrupt countries. In corrupt environments where bureaucrats can successfully derive bribes from importers, bribes may act as a tax and thus decrease trade levels (Dutt

& Traca, 2010). Shepherd (2009) studies the influence of bribing at customs and the impact of the level of corruption on international trade. Corrupt civil servants who are trying to take every opportunity to extract a bribe may cause delays in the transaction process that would have not appeared otherwise. To this extend and in line with existing literature it is generally expected that corruption reduces international trade volumes and negatively effects trade patterns. Therefore I test the following hypothesis

Hypotheses 1: Higher levels of corruption have a negative impact on import volumes (i.e. intensive margin) in the sample as a whole

Corruption “greases the wheels"

In the case of developed countries, there is a high regard for rule of law, higher levels of governance and effective institutions as well as greater contract enforcement. It is expected that corruption would harm the economy and thus ‘sands the wheel’ on international trade in a strong institutional environment. Despite the conventional belief that corruption has a negative impact on trade flows, numerous authors have also considered that corruption may indirectly result in positive externalities for developing countries which suffer from a lack of resources, lower levels of efficiency, low contract enforcement, bureaucratic delays and red tape as well as an inefficient ‘rule of law’. The

“greasing the wheel” for international trade assumption has a long history going back to Bhagwati (1982) and is still present in modern theory.

In the case of developing countries suffering from distortions, Bhagwati (1982) suggests that the diversion of resources from directly productive to directly unproductive activities is fundamentally different from such diversions occurring in distortion-free environments. Within his argument, context is critical, such that rent-seeking (i.e. DUP activities) may exert a positive impact on a country’s output and welfare.

(9)

8 In other words, in a weak institutional context with market distortions some mechanism (i.e. DUP activities) which would be otherwise rather harmful for strong economies, could actually improve economic outcomes (Shleifer & Vishny, 1994). In this context Bhagwati (1982) proposes that a particular subset of DUP activities with respect to economic activity, is corruption.

Figure 1 reiterates the basic figure shown in Bhagwati (1980, p.358). It depicts the production frontier of a small open economy which trades on international markets before and after the introduction of a i.e. a tariff of a medium and large size (indices t and tt). Costs are measured as ‘output loss of good X’. The diagram indicates that in the case of no distortions the production and consumption points lie at p*, where the production frontier (PPF) and the highest indifference curve are tangent to the world price ratio, respectively. Exogenously introducing a tariff alters the domestic price ratio to P^t, where the economy remains producing at the initial PPF which results in a drop of national output. Under DUP circumstances it is indicated that rent-seeking ‘consumes’ resources which are used in the production of goods X or Y. Thus, the PPF shifts down. As it can be seen in the case of a higher distortion, DUP activities increase output as compared to a situation where the same distortion was introduced without rent-seeking (Xttr

> Xtt).

Consequently, corruption may result in diminishing rent-seeking distortions in a second-best world caused by poorly functioning institutions (Leff, 1964 and Huntington, 1686). In the context of international trade, corruption is similar to others DUP activities such as tariff evasion or smuggling.

If these activities take place in initially distorted situations then the DUP activities (and so corruption) could have output and welfare enhancing effects. Even though Bhagwati’s theory does not directly investigate the effect of corruption on the volume of trade flows, it presents corruption as a mean of greasing the wheels of commerce (Lavallée, 2005). Thus, distorted economies conducting rent- seeking may have a comparative advantage over distorted economies who do not undertake any DUP activities.

With a weak government, rent-seeking activities are common in a setting with asymmetric information between the State and its agents generating sufficient room for opportunistic behavior.

Government involvement and restrictions in private markets is seen as one source of corruption. In the case of ‘bad institutions’ such as poor quality customs and restrictive regulations, corruption can decrease regulatory restrictions and consequently ease international trade flows (De Jong &

Bogmans, 2011). The malfunctioning of bureaucracy such as slowness, tedious regulations and low

Figure 1 Bhagwati DUP activities, Source: Fischer, 2006

(10)

9 quality of civil servants are seen to be the most impediments that corruption could grease. For example, bribes can give incentives to bureaucratic to speed up the process and therefore corruption can reduce the time queuing (Lui, 1985). Aurioal and Benaim (2000) presented the consequences of public sector corruption in a growth model. They analyzed that an equilibrium with corruption may be favored over one where corruption is absent because corruption mechanisms bypass bureaucratic red tape. Thus, they were able demonstrate the positive effects of rent-seeking. Additionally, Coppier and Michetti (2006) showed when the State fails to ensure proper monitoring mechanism, higher corruption in turn can be associated with enhanced production. For instance trade restrictions such as quantitative import limits for certain goods or protection of a certain home industry from foreign competition through tariffs will result in bribing officials to obtain import licenses or to keep on maintaining monopoly status. Hence, corruption can take the form of demands for special payments and bribes linked with import and export licenses, exchange controls, tax assessments, police protection, or loans (Charron, Lapuente, & Rothstein, 2010). Furthermore, country studies show a trade enhancing effect from corruption (Dutt & Traça, 2010). Parayno (1999) indicates that businesses conducted in the Philippines have become accustomed to paying bribes for customs services. In order to avoid misclassification and undervaluation in the declaration process, bribes are common ways for enterprises to avoid the official trade barriers. Arduz (2000) shows that in Bolivia, most goods have to go through a system of parallel customs. Instead of the official trade taxes, in this system corrupt customs officers charge their own taxes.

Similar results are found by De Jong and Udo (2005) who show that in countries with low custom quality, corruption (bribing) increases international trade. Channels through which corruption can enhance trade are for example ‘grease money’ where a corrupt official receives money in order to speed up the time of an economic activity with complex and burdensome rules and regulations (U Myint, 2000). Furthermore, customs officials take bribes and in turn allow importers to avoid the enforcement of trade regulations (Thede & Gustafson, 2012). Thus, corruption can be seen as a way to bypass rigidities imposed by governments (Lavallée, 2005). Dutt and Traca (2010) use the International Country Risk Guide and analyses a sample of 84-90 countries over the time period 1982-2000. They show that the direction of relationship between corruption and trade flow is influenced by the level of tariffs. The authors further emphasize that the relationship between corruption and international trade is non-linear. In the case of severe regulations (e.g. high tariffs), corruption has a positive effect on trade flows. More precisely when the level of tariffs is high then the impact of corruption presents an inverted U-shape curve.

In light of this it seems fitting that different economic settings might also influence the relationship between corruption and trade. Especially between developed and developing countries

(11)

10 the effects of corruption on trade might be different. Very few studies have considered whether differences exist and therefore this paper aims to fill this gap. The countries are additionally grouped according to their development status. In line with the discussed literature it is expected that in developing countries where the initial economic situation is already distorted corruption is likely to have a positive impact on trade. Thus the following hypothesis is investigated

Hypotheses 2: In developing countries corruption enhances the volume of imports

While hypothesis 1 and 2 investigate the effect on corruption on an already existing trade relationship (intensive margin) the paper further analyses if corruption is a reason that countries have not traded with each other at all or if corruption increases the probability for a future trade relationship (extensive margin). It is expected that corruption increases the probability for future trade for developing countries and decreases it for developed countries. In that line following hypothesis is formulated

Hypotheses 3: Corruption affects the extensive margin of international trade

Studies which have considered the effects of corruption on trade suffer from several shortcomings which this paper attempts to overcome. While many studies employ the gravity model for testing the effects of corruption on trade, recent interest in the theoretical foundation of the gravity model have indicated that common errors with respect to its empirical specification exist. More specifically, this paper addresses issues with regards to the misspecification of the dependent variable, the inclusion of a multilateral resistance term (MRT), the inappropriate deflation of trade flows by common price indexes¹, the prevalence of unobserved heteroskedascity and the incidence of zero trade flows^.. Additionally, two estimation techniques, a Zero-Inflated Negative Binomial and a Poisson pseudo- maximum likelihood model are used to overcome several statistical problems regarding gravity equations estimations modeling bilateral trade. Moreover, this paper overcomes the shortcoming of inconsistent corruption index use, by employing an enhanced corruption index. A specific obstacle faced by the literature is the one of reserve-causality. While corruption is shown to exert a strong impact on international trade, it could be that international trade flows affect the level of corruption prevalent within countries. Lastly, this paper addresses this problem of endogeneity by establishing a one way causality using instrumental variables.

1 Common errors regarding gravity model specification are further discussed in section 3.

(12)

11 3 Method and Data

In order to establish what effect corruption has on international trade flows the paper makes use of a standard gravity model with standard controls. This model is used because empirical evidence for the gravity equation in international trade is strong. It has become the workhorse for modeling bilateral trade flows within the literature and has been shown to be empirically sound across various data, time periods and methodologies. It has evolved from a simple from of gravity equation to a log-normal fixed effect specification (Anderson & van Winccop, 2003). However, recent studies on gravity model have shown that traditional gravity models suffer from several statistical problems.

3.1.Statistical Problems

There are several factors which may yield to biased and inconsistent coefficient estimates which have to be taken into account first. I discuss each of them in the following section.

a) Gold, silver and bronze medal errors

Baldwin and Taglioni (2006) identify three common errors in the literature using the gravity model called the gold, silver and bronze medal errors.

The gold medal mistake refers to an omitted variable problem which could result in heteroskedasticity in the model. If this is not controlled, the estimates will be severely upward biased and it is not possible to achieve accurately estimate the effects on trade (Baldwin and Taglioni, 2007).

While omitted variable bias is a general problem within regression analysis, Anderson and van Wincoop (2001) highlight that in the gravity model unobserved multilateral price indices could act as trade barriers with respect to trade flows between importing and exporting countries. Therefore, unobserved variables (i.e. distances, borders etc.) should be included in the model to account for unobserved trade cost. Thus, they demonstrate that it is necessary to include what they refer to as a multilateral trade resistance term (MRT) (Cecilio, Escalona, & Gómez, 2011). In order to solve this problem I include in line with Anderson and van Wincoop (2001) and Feenstra (2004) country (for exporter and importers) fixed effects which proxy for the existence of unobserved trade barriers.

Silver medal mistake takes place when the bilateral trade flows are averaged and when the log of the sum instead of the sum of the logs are used (Baldwin & Taglioni, 2006). The aim of using a gravity model is to explain uni-directional trade. However, when bilateral trade flows (imports and exports) are averaged then essentially the equation depicts the averages of two-way exports and thus do not reflect uni-directional trade. While Baldwin and Taglonini (2006) point out that there is nothing intrinsically wrong with this, because it is done without reference to theory, most researchers go on to confuse the estimation of the log of the averages with the averages of the log. According to Anderson and van Winccop (2001) and Feenstra’s (2004) gravity estimation of the dependent

(13)

12 variable must be unilateral and not be aggregated. This allows to identify for asymmetries between importers and exporters. To encounter for this mistake I only focus on imports and represent each country pair twice.

The Bronze medal mistake arises by inappropriately deflating the trade flows by a common price index. This results in biases via spurious correlation that is not because of causality, but due to the dependence of the two variables on another unobserved factors. More specifically, in the case that trade data is deflated using the US Consumer Price Index for a specific year (as an example) essentially inflation standards in the US during that year are being applied to all countries within the sample. This is unrealistic because not all countries have the same inflation. By including time fixed effect this problem is eliminated (i.e. variation over time is controlled for). Thus by taking care of the gold medal mistake I do not face any difficulties regarding the bronze medal mistake and deflating trade by the US Consumer Price Index (Baldwin & Taglioni, 2006).

b) Heteroscedasticty

An important key assumption of regression is that the variance of the errors is constant across observations i.e. the errors are homoscedastic. When this assumption is violated and the errors are heteroscedastic or have non-constant variance, then the standard errors computed are incorrect and the test statistics as well as confidence intervals might be misleading. According to Santos Silva and Tenreyro (2006) results from a log-linear estimations can be biased in the presence of heteroskedasticity due to the fact that Jensen’s inequality implies that E(ln y) ≠ ln E(y) (i.e. the expected value of the logarithm of a random variable is unequal to the logarithm of its expected value). It is crucial to point out that the parameter estimates as well as the standard errors are affected and thus a completely different estimation methodology than Ordinary least square (OLS) is required to solve for this problem. In order to remedy for heteroskedasticity I first make use of a Zero-Inflated Negative Binomial (ZINB) estimation technique and then of a Poisson Pseudo-Maximum Likelihood technique (PPML) which will be further discussed below.

c) Zero Trade flows

Most central bureaus of statistics only report trade figures exceeding a certain threshold. Thus, if trade turns out to be zero or negligible this will lead to problems in analyzing a log-liner form of the equation used in the paper. The balanced panel of this study covers 120 countries over 10 years. For total imports 20 percent of the observations are equal to zero, reflecting that either these countries do not trade with each other at all or the trade volumes are really small and got rounded to zero. An additional 10 percent of the observations consist of missing values. To this extent using a traditional

(14)

13 log-linear form of the equation has the drawback that it prohibits the examination of both the intensive and extensive margin of trade. Because the absence of trade is also essential information that needs to be taken into account in the estimation in order to prevent sample selection bias, the observations cannot be simply dropped. Therefore this study applies a ZINB and a PPML model to take zero and missing trade values into account, while the ZINB model further accounts for the intensive and extensive margin.

3.2.Econometric Specification

Because econometric estimation of the gravity model has revealed that typically the model suffers significantly from unobserved heterogeneity, a more favorable model is a Poisson model (PM). In general Possion models are making use of count data, but according to Wooldridge (2002) these models can also be used with non-negative variables. According to Santos Silva and Tenreyoro (2006) by applying a multiplicative, non-linear specification it solves the zero trade problem and strict assumptions of the OLS (i.e. homoscedasticity and normality). However, this model only accounts for observed heterogeneity, where different values of the predictor variables lead to a different conditional mean value. Thus, it suffers from the presence of over-dispersion (i.e. the conditional variance of the data exceeds the conditional mean) resulting in a biased model fit and in deflated standard errors of the estimates. According to Burger, van Oort and Linders (2009) a modified Poisson model, the Negative Binomial Model (NB) can relax the assumption of equi- dispersion of the Poisson model by explicitly modeling between-subject heterogeneity. Thus a likelihood ratio test of the over-dispersion parameter α is employed. When α is larger than zero, it is an indication for overdispersion and concluded that a Negative Binominal model is preferred over a Poisson distribution (Cameron and Trivedi, 1986).

However, these models can underestimate the true number of zeros in the model indicating that the number of zeros is greater than predicted by the Poisson or Negative Binomial distribution.

Because the trade data contains an excess of zeros, Burger et al. (2009) further extends the Poisson estimation by considering, Zero Inflated Poisson (ZIP) and Zero Inflated Binomial (ZIB) models. The starting point of these models is the presence of two latent groups. One group takes on the value zero because trade flows between countries do not exist, for example due to the lack of resources or trade embargos. The other group has zero values due to the fact that either in a cross section at that specific point of time countries were not trading but could potentially trade to another point of time or trade was reported very close to zero. Consequently, the model is expressed in two parts. Firstly, it is a binary process applying a Logit model, expressing the probability p_i of no bilateral trade at all (extensive margin). Secondly, it is a count model applying a Negative Binominal or Poisson model,

(15)

14 conditional on the probability of having non-zero trade 1- pi (intensive margin) where the trade flows are taking on count values.

The Vuong test is employed in order to examine whether a zero-inflated estimation technique is favored above its non-zero inflated counterpart, with positive values favoring the zero-inflated variant. Thus, I am applying a Zero-Inflated Negative Binomial estimation technique (ZINB) which handles the presence of an excess of zeros as well as over-dispersion to estimate the effect of corruption on imports (further justification for the use of the model is provided in section 4).

However, this model has been criticized lately because it is not scale invariant and thus a PPML model would be more appropriated (Head & Mayer, 2013). Hence the PPML technique is explained in the robustness check and next to the ZINB further used throughout the paper.

This paper makes use of a standard gravity equation with standard controls and is based on the approach of Anderson and Marcouiller (2002) and Feenstra (2004). It accounts for unobserved price indexes formalized as the multilateral resistance term (MRT) by including exporter, importer and time fixed effects².

The sample considered is based on data availability and covers 121 countries (listed in Table 2) over the time period of 1998 to 2007 which yields in 121 x 120 x 10 = 14,5200 observations. This time period is selected based on two reasons. 1998 is chosen because prior to the date difficulties will be faced regarding data availability and 2007 is selected to avoid distortions in the results caused by the financial crisis. Furthermore, in 1998 the corruption index and trade data is available for 121 countries. The dataset consists of trade flows between any two trading partners in each year which creates a balanced panel. The model can be specified as follows:

= × × × × × × ×

× × ( ∗ ) × ( ∗ ) × ×

%^&^' ×%⁽⁾ × %^*⁺× ε

Where i (importer) and j (exporter) denotes trading partners, t denotes time.

Of particular interest is the testing of the hypotheses stipulated above.

Hypothesis 1 will be supported, if the corruption variables for the importing and exporting country and exhibit a significant negatively sign. Consequently, high levels of corruption have a trade diminishing effect on imports i.e. “sands the wheels”.

2 See section 3.1. for further explanation of MRT

(16)

15 With respect to hypothesis 2 where DC takes the value 0 for a developing country, it is expected that the marginal effect of ( ∗ ) and ( ∗ ) is positively significant showing that a higher level of corruption has a trade enhancing effect. Furthermore, the countries are grouped according to their development status. The estimation is first run for developed and then for developing countries. Hypothesis 2 is confirmed when the corruption coefficients and

turn out to be positive i.e. ‘greases the wheels’ of trade for developing countries.

Regarding hypothesis 3 it is expected that the extensive margin of the corruption coefficient and is positive and thus increases the likelihood for future trade for developing countries and is negative and therefore decreases it for developed countries.

3.3.Data

The bilateral imports are used as the dependent variable in order to investigate the influence of corruption. Taking the import values is the standard way when applying the gravity model.

Furthermore, each country-pair is included twice, once as ij and once as ji (Kohl, 2012). This also serves as prevention from making the theoretical “silver medal mistake” (Baldwin and Taglione, 2006). In line with Rose (2004) following expectations about the variables in the equation are made.

GDP is a proxy for the economic size of country i and j. It is suggested that trade (imports) increases with the economic size of the country. Additionally, the effects of population size on international trade are tested. It is assumed that the amount of trade decreases with a higher population due to the phenomenon that countries with a larger population tend to be relatively less open to international trade since these are well equipped with resources within their own border.

Transport and transaction costs increase with distance and thus it is expected that trade decreases with distance. Countries with a common primary language are likely to have a better understanding of each other’s business practices and thus trade more than countries with less-similar environments.

F_i and F_j are country-level fixed effects, representing the MRT that control for the differences in the imports prices faced by trade partners regarding to all their other trading partners (Anderson &

Van Wincoop, 2001; Feenstra, 2004). The use of time-invariant fixed effects has been criticized for being a too general approach and not accounting for inter-temporal effects i.e. the problem of this estimation is that it does not allow for the country-level effects to vary over time. However, in the existence of price levels varying over time, this should be ideally the case. Thus, estimation (1) may suffer from an omitted variables bias resulting from ignoring time-varying terms across countries (Baier & Bergstand 2007). Therefore, the use of time-varying importer and exporter effects, F_it and F_jt are suggested. Additionally, by including Fij (dyad effects) unobserved factors influencing imports

(17)

16 which may be correlated with those influencing corruption should be controlled for (Kohl, 2012).

However, due to the use of a country level corruption index, which does not account for country-pair effects it is not possible to include dyad effects and thus time-invariant fixed effects have been used.

Table 3.3.1 Description of the Data

Variable Description Source

Im_ijt Represents the imports by country i from country j in year t. The data is deflated by the US

Consumer Price Index (All consumer Goods, 1983-84 = 100)

Kohl (2012) obtained from the Bureau of Labor Statistics (2008)

GDP It is the real GDP of country i and j at year t in 1990 international dollars

Kohl (2012) obtained from Maddison (2007)

Pop Number of population to control for the size of country i and j at year t

Kohl (2012) obtained from Maddison (2007)

Dis_ij The Geodesic (great circle) distance between capitals of country i and j

Kohl (2012) obtained from Centre d'Etudes Prospectives et d'Informations Internationales (CEPII)

Lang a dummy with value 1 if i and j have a common primary language

Kohl (2012) obtained from CEPII Cor Annual corruption levels, making use of the

ICRG indexes of corruption

International Country Risk Guide Dataset (ICRG) created by the Political Risk Service Group.

Obtained from Le Van Ha DC*Corruption Interaction term which takes the value 1 when the

trading partner is a developed country Fi a set of importer fixed effects

Fj a set of exporter fixed effects

Ft a set of time fixed effects to correct for common trends and shocks

ε Is the error term

The main interest of this paper is the corruption variable which is taken from the International Country Risk Guide Dataset (ICRG) and is created by the Political Risk Service Group. In order to explain the existence of regional corruption, the corruption index for both the importing and exporting country is included (Treismann, 2000). In many previous studies the corruption index from Transparency international or the Worldbank is applied (Sandholtz & Koetzle, 2000, Treisman, 2000, Lambsdorff, 2006). However, these indexes suffer from inconsistency over time and thus leads to bias in the estimates of panel studies. These measures pool together multiple sources to receive an index of corruption which results in a greater degree of imprecision and inconsistency. The main problem is that various sources are applied for countries in different years and that these sources can alter in any year. Consequently, these measures suffer from a limited amount of internal consistency which leads to problems in year-to-year comparison (Charron, et al., 2010). In order to avoid these drawbacks in this study the ICRG index is used. The Political Risk Service Group shows that

(18)

17

“corruption within the political system, which may be damaging to the economic and financial environment, reduces the efficiency of government and business by enabling people to assume positions of power through patronage rather than ability and introduces an inherent instability in the political system” (Charron, et al., 2010, p.70). An advantage of the ICRG is that the measures focus on how corruption effects international investors and does not emphasis its impact on the lives of a country’s citizens. Furthermore, the index is not constructed on the aggregation from survey data containing various sources but is related to the evaluations made by a panel of experts. The ICRG index is an annual index, covering across counties the time period 1982-2012. It has been used by Knack and Keefer (1995), Dutt and Traça (2010) and among others. The index specifies that higher corruption refers to “high government officials are likely to demand special payments and illegal payments are generally expected throughout lower level of government in the forms of bribes connected with import and export licenses, exchange controls, tax assessment, police protection, or loans“ (Knack & Keefer, 1995, p.225). The index ranges from 0 to 6, where a higher number indicates a lower level of corruption. In order to ease the interpretation of the coefficient this variable will be multiplied by (-1), so that it increases with corruption.

Furthermore, an interaction term (Developing countries × Corruption) is included to distinguish the effect of corruption on imports between developed and developing countries.

Corruption, poor governance and weak institutions seem to be specifically relevant in developing countries. But since the initially economic situation in developing countries is already distorted, the effect of corruption can be different to the one in developed countries (Bhagwati, 1982).

Descriptive statistics of the dataset of this study are provided in Table 3.3.2 while Table 3.3.3 provides an overview of the countries included in the dataset.

Table 3.3.2 Descriptive statistic

Variable Obs. Mean Std.

Dev. Min Max

Imports 131539 369.91 3397.35 0 214440

Ln GDPimporter 127080 10.9 1.87 6.60 16.04

Ln GDPexporter 127080 10.9 1.87 6.60 16.04

Ln Population importer 126000 2.53 1.56 -1.29 4.17

Ln Population exporter 126000 2.53 1.56 -1.29 4.17

Ln Distance 140420 8.74 0.77 2.35 9.89

Common language 140420 0.16 0.36 0 1

Corruption index importer 145200 -2.78 1.67 0.00 40.80

Corruption index exporter 145200 -2.78 1.67 0.00 40.80

Developed Country *Corruption importer 145200 4.34 9.26 0.00 36.00 Developed Country *Corruption exporter 145200 4.34 9.26 0.00 36.00

(19)

18 Table 3.3.3 Countries included in the Dataset

Developed Countries Developing Countries Australia, Austria, Belgium,

Bulgaria, Canada, Cyprus, Denmark, Finland, France, Germany, Greece, Hungary, Iceland, Ireland, Italy, Japan, Luxembourg, Malta, Netherlands, New Zealand, Norway, Poland, Portugal, Romania, South Korea, Spain, Sweden, Switzerland, United Kingdom, United States

Albania, Algeria, Angola, Argentina, Bahamas, Bahrain, Bangladesh, Bolivia, Botswana, Brazil, Brunei, Burkina Faso, Cameroon, Chile, China, Colombia, Congo, Costa Rica, Cuba, Dominican Republic, DR Congo, Ecuador, Egypt, El Salvador, Ethiopia, Gabon, Gambia, Ghana, Guatemala, Guinea, Guinea-Bissau, Guyana, Haiti, Honduras, Hong Kong, India, Indonesia, Iran, Iraq, Israel, Jamaica, Jordan, Kenya, Kuwait, Lebanon, Liberia, Libya, Madagascar, Malawi, Malaysia, Mali, Mexico, Mongolia, Morocco, Mozambique, Myanmar, Namibia, Nicaragua, Niger, Nigeria, North Korea, Oman, Pakistan, Panama, Papua New Guinea, Paraguay, Peru, Philippines, Qatar, Saudi Arabia, Senegal, Sierra Leone, Singapore, Somalia, South Africa, Sri Lanka, Sudan, Suriname, Syria, Tanzania, Thailand, Togo, Trinidad & Tobago, Tunisia, Turkey, Uruguay, Venezuela, Vietnam, Yemen, Zambia, Zimbabwe

To begin the analysis of how corruption affects international trade, I first turn to an investigation of the corruption levels of the countries included in the sample. Table 3.3.4 lists the ten most and least corrupt countries, according to its corruption index in 2007. A brief examination of the rankings in the table shows that the countries which are least corrupt are characterized as being democratic, have high levels of political as well as economic freedom and are incorporated into the world economy (Sandholtz & Koetzle, 2000). Countries belonging to the least corrupt ones are for example the Scandinavian countries such as Denmark, Finland, Norway, and Sweden. In line with the literature review it is expected that for those countries corruption “sands the wheels”.

Table 3.3.4 Corruption Index

The least corrupt countries Index The most corrupt countries Index

Finland 6.00 Zimbabwe 0.00

Denmark 5.50 Kenya 0.50

Iceland 5.50 Haiti 1.00

New Zealand 5.50 Paraguay 1.00

Austria 5.00 Venezuela 1.00

Germany 5.00 Iraq 1.00

Luxembourg 5.00 Lebanon 1.00

Netherlands 5.00 Congo, DR 1.00

Norway 5.00 Gabon 1.00

Sweden 5.00 Somalia 1.00

A score of 6.00 indicates the lowest level of corruption

Whereas countries with higher corruption indexes are seen as having weak institutions (i.e. poor protection for property rights, weak contract enforcement, and weakened legal and social order) are more authoritarian, have little entrepreneurship and are less integrated into the world economy.

Countries labeled as being most corrupt are for instance African countries like Zimbabwe, Kenya,

(20)

19 Gabon and Somalia. According to Bhagwati (1982) and modern literature it is expected that corruption “greases the wheels”.

3.4.Robustness Check

An alternative approach for dealing with zero trade flows is proposed by Santos Silva & Tenreyoro (2006) and indicates the use of a Poisson Pseudo-Maximum Likelihood model (PPML). Recent literature has suggested that the ZINB model suffers from serious problems because it is not scale invariant. Especially in the context of gravity estimations the results will vary between whether the dependent variable is in dollars or millions of dollars.

Besides, it is argued that the Poisson model is consistent as a PPML regardless how the data is disturbed (Head & Mayer, 2013). Therefore, advantages of this model are that it is consistent in the presence of fixed-effects which can simple be entered as dummy variables and because of its multiplicative form it can deal with zero trade value observations. Furthermore, its estimates are consistent in the presence of heteroskedasticity and are especially efficient in large samples (King, 1988). Consequently, to take the criticism of the ZINB model into account the PPML is applied as a robustness check.

3.5.Sensitivity Analysis

The previous estimations simply distinguished between developed and developing countries. It was only assumed that developed countries have stronger developed institutions while developing countries suffer from weak institutional quality. As stated before, it is likely that corruption takes place along with weakly developed legal, political and economic institutions. It is expected that when an economy suffers from low quality institutions (i.e. low governance), in line with Bhagwati (1982) corruption mitigates the effect of such a distortion and international trade will be higher than without it. In order to encounter for this effect, the development specification has to be replaced by an indicator of the quality of institutions. Therefore, an interaction term is included which is the product of corruption and a proxy for other distortions (i.e. quality of institutions) (Meon & Sekkat, 2006).

Kuncic (2013)³ grouped over thirty well-known institutional indicators into three homogenous sets of formal institutions (legal, political and economic) and computed its latent institutional quality for 197 countries over the time period of 1990 - 2010. According to Kuncic (2013) the advantage of this method is that it represents almost the whole formal institutional environment of an economy and thus captures the true latent institutional quality which enables to study the countries form a

3 See Kuncic (2013) for discussion and information about the Dataset

(21)

20 comparative institutional perspective. In order to obtain one indicator for institutional quality for each country the average of the three institutional indicators is computed for each year. The indicator ranges from 0 to 1 with values close to 1 showing high quality institutions. Thus, when institutional quality ( -) is low and corruption (Cor) is high, the marginal effect of the interaction term should exhibit a positive sign indicating a trade enhancing effect. Whereas with a high quality of governance the effect of corruption on trade volumes should turn out be to negative.

Furthermore, the sensitivity analysis includes additional controls that are expected to have an impact on international trade flows. It is expected that variables indicating if countries share a common border, have a regional trade agreement, or are a member of WTO have a trade increasing effect.

Table 3.5.5 Description of the Data included in the sensitivity Analysis

Variable Description Source

Regioij A dummy with value 1 if i and j belong to the same regional trade agreement (RTA) at time t and 0 otherwise

Kohl (2012): RTAs drawn from table 2.3. pg 34.

WTOij A dummy with value 1 if i or j are members of the WTO, 2 if i and j are members and 0 otherwise at time t

Kohl (2012) obtained from Tomz at al. (2007)

Border a dummy with value 1 if the i and j share a common border

Kohl (2012) obtained from CEPII

This estimation is done based on the ZINB as well as the PPML technique and the model is specified as follows.

= × × × × × × ×

. /0 × × 10 × 23 ×

× × ( - ∗ ) × ( - ∗ ) × - × - ⁴

%^&^' ×%⁽⁾ × %^*⁺× ε

3.6.Endogeneity

The paper investigates whether a significant correlation exists between imports and corruption with controlling for other variables influencing international trade. However, in examining the effect of corruption on trade flows, two problems arise. First of all, it is likely that the level of corruption as well as international trade flows respond simultaneously to an omitted factor (i.e. economic growth or rule of law). Because omitted variable bias affects the size and significance of coefficients it is important to encounter for it. The second problem is to determine whether corruption has an effect on imports or if imports influence the level of corruption. In previous studies the phenomenon that

(22)

21 international trade can have an effect on domestic institutions is less recognized (Nunn & Treffler, 2012). Ades and di Tella (1999) complete a study in which they claim that international trade leads to a more competitive economy which results in less corruption. The authors empirically show that the level of corruption is affected by the openness of an economy (measured by the amount of imports).

They argue that trade liberalization increases the efficiency of the market which makes them more competitive. This in turn leads to a reduction of available rents and thus to a decrease in the level of corruption in the economy (Chaudhry, 2005). Contrary, if international trade in a country increases, especially if the volume of imports rises officials have more chances to demand higher bribes which results in more corruption in a given country.

Consequently, the problem is that the estimates for the ß coefficients concerning corruption can be biased due to omitted variables and because of reverse causality from corruption to imports.

For the purpose of this paper it is important to achieve a one-way causal relationship indicating that corruption has an effect on trade and not vice versa.

Frankel and Romer (1999) suggest the use of instrumental variables (IV) in curing potential bias and solving for endogeneity. This method provides a better evidence of a causal influence from corruption to trade. A good IV is uncorrelated with the error term, highly correlated with the endogenous independent variable but independent from the dependent variable (i.e. having no direct effect on imports) except through the potentially endogenous variable to establish exogeneity (Hill, Griffits, & Lim, 2011). In the case of this study, the IV should be highly correlated with corruption but have no influence on trade. Regarding the research field of corruption and international trade, authors have used ethnolinguistic fractionalization, origin of the country’s legal system and the mortality rates of European settlers at the time of colonization as instrumental variables (Mauro,1995;

Ades & Di Tella 1999; Acemoglu, Johnson, Robinson, 2001; Knack & Azfar 2002; Azfar & Lee,

2007). The paper

applies a two-step estimation to the Possion model, where I instrument the corruption variables based on data availability for the sample by the origin of the ethnolinguistic fractionalization (ELF) and country’s legal system.

The reasoning for ELF as being a good instrumental variable for corruption is that ethnic groups have internal solidarity and because members use their power in favor of those belonging to their group, this behavior would eventually lead to corruption (Mauro, 1995, Easterly & Levine, 2002). In other words, more fractionalized countries may have more dishonest bureaucrats who favor members of their own ethnolinguistic group and hence higher levels of corruption are associated with a higher degree of ELF in a country. It is assumed that the extent to which countries are fractionalized along ethnolinguistic lines is exogenous and unrelated to their import levels other than through its

(23)

22 effects on corruption. Therefore this is considered to be a good instrumental variable. This index is taken from the Atlas Narodov Mira (1964) and ranges from 0 to 1 (0<ELF<1 ), where the higher the ELF index more fragmented is the country.

The reasons for origin of the country’s legal system as being a valid instrument for corruption is because the legal system of the country has an effect on the setting of the property rights and this in turn affects corruption (La Porta et al., 1999; Frediksson & Svensson, 2003). According to La Porta et al. (1999) common law systems, mostly found in Britain and its former colonies are different compared to civil law systems, mostly found in continental Europe it its former colonies. While, common law places emphasized on the private rights of individuals and their property rights (Finer, 1997); civil and socialist laws place particular emphasizes on the State’s ability to maintain power and extract resources without much regard for protecting economic interest or liberties of the people (La Porta et al., 1999). Furthermore Treismann (2000) shows that countries with British legal origins tend to stress procedural fairness. It is argued further that British colonial heritage is associated with significantly lower levels of corruption (La Porta et al, 1999; Treisman 2000, Acemoglu et al. 2001).

Because the legal origin of a country can in no way influence a country’s import level other than through the level of corruption, it is assumed to be a good instrument for the estimation of corruption.

In order to account for legal origin, I use a dummy variable which takes the value of 1 when the legal origin of the company law or commercial law of a country has its origin from the British common law, and 0 otherwise.

The two-step estimation involves a two-step regression methodology. At first in a linear probability regression the independent corruption variables are run on the instrumental variables (ELF and origin of the country’s legal system) to obtain the residuals (i.e. 5 86 7 ). In the second step the Poisson model (ZINB and PPML) is fitted on the regressors that contain the first-step residuals. Due to establishing predictions for the independent variables (corruption) in the second stage through the instruments, I am able to correct for the correlation between the error term and the independent variable.

A drawback of the mentioned instrumental variables is that these are time-invariant. However, due to data availability and the fact that these variables can affect imports indirectly through corruption they are assumed to be exogenous within the study.

4 Empirical Results

In order to confirm my model choice I make use of the Vuong test and the α level. According to the results of the Vuong test, zero-inflated models are favored against the non-inflated versions.

(24)

23 Furthermore, overdispersion is confirmed since α > 0 which favors the NB against the Poisson model and the ZIB against the ZIP model. Consequently, the use of the ZINB is confirmed.

The estimation results of the gravity model spanning 1998-2007 are displayed in table 4.1.2.

The results are subdivided into two columns: the ‘extensive margin’ which if positive (negative) indicates the likelihood that a country pair will (not) trade; the ‘intensive margin’ which shows if positive (negative) the extent to which international trade volumes will be enhanced (diminished).

Column (1) presents the estimation results for the basic gravity equation excluding corruption indexes. In column (2) corruption indexes and in column (3) the interaction terms (i.e. Corruption × Developed Country) are added. In column (4) countries are grouped according to their development status.

Regarding the estimates of the basic gravity equation which exhibit the same signs and significant levels throughout columns (1), (2) and (3) the effect of GDP is highly positive and significant at the 1 percent level confirming the initial expectations that trade increases with the economic size of a country. As expected the sign of distance is negative at the 1 percent level of significance, implying that with increasing distance, ceteris paribus, imports decrease significantly.

Population is negatively associated with imports at the 1 percent level of significance, showing that an increase in the population size reduces the volume of imports which is also in line with expectations. When a country pair has a common language, international trade flows are increased significantly confirming my expectations. However, by adding the interaction terms in column (3), the estimate of the population size for the importing country loses significance. The variable of main interest, the corruption indexes of the importer as well as the exporter are further discussed in more detail.

In column (2) the corruption index for exporting countries is negative but not significant.

While the estimator for importing countries is negative and significant at the 1 percent level. If the corruption index increases by one standard deviation (SD) from the mean, international trade flows will decrease by 17 percent (09 .;;<∗ ;.=>− 1), ceteris paribus. In the extensive margin the positive sign of the corruption coefficients illustrate that if countries are more corrupt in general it increases the probability of belonging to the trading group (intensive margin).

In column (3) while the corruption indexes for importing as well as exporting countries exhibit negative signs, the coefficient only for the importing country is significant. The findings of the interaction terms between developed importing and exporting countries and its corruption indexes are as expected. Both corruption coefficients turn out to be significantly negative and thus the marginal effect indicates a trade diminishing effect. The coefficient of the corruption index for importing countries exhibit a positive sign in the extensive margin. Thus, if developed importing

Table of Contents

Table of Contents