MSCINTERNATIONALECONOMICS ANDBUSINESS
MASTERTHESIS
SINO-AFRICANTRADE: THEGRAVITY OFINSTITUTIONS
STACEYLEIGHPECK S215 9708 s.l.peck@student.rug.nl
JULY2013
SUPERVISOR: PROF. DR. S. BRAKMAN
CO-ASSESSOR: PROF. DR. J. OOSTERHAVEN
METHODOLOGYSUPERVISOR: DR. P. RAO-SAHIB
TABLE OFCONTENTS
ACKNOWLEDGEMENTS...3
ABSTRACT...4
I. INTRODUCTION...6
II. RELATEDLITERATURE...8
III. METHOD ANDDATA...14
i. Econometric Specification Considerations ...14
Problems Regarding the Log-Normal Specification of the Gravity Equation ...15
From a Log-Normal to a Poisson Specification...16
ii. Model...17
Assessing the Effects of Institutional Quality on Trade Flows...18
Model Analysis Considerations...20
iii. Data...21
iv. Robustness ...22
v. The Endogeneity Problem ...23
IV. RESULTS...27
PARTA: THEEFFECT OFINSTITUTIONALQUALITY ONTRADE...27
i. Baseline Results...27
ii. The Role of Institutions: A Composite View...27
iii. The Role of Institutions: A Composite View – Robustness Check...29
PARTB: THEEFFECT OFINSTITUTIONALHOMOGENEITY ONTRADE...31
i. The Role of Institutional Homogeneity ...31
ii. The Role of Institutional Homogeneity – A Sensitivity Analysis...33
PARTC: ADDRESSINGTHEENDOGENEITYPROBLEM: ANINSTRUMENTALVARIABLEAPPROACH...35
V. DISCUSSION ANDCONCLUSION...36
VI. REFERENCES...40
VII. APPENDIXA ...42
ACKNOWLEDGEMENTS
I would like to start by expressing my sincere gratitude to my supervisor Prof. Steven Brakman for his guidance, patience and motivation throughout the thesis process. His persistent encouragement and constructive comments have made this paper what it is today and I consider having had the opportunity to work together with him an immense privilege.
I would also like to express my appreciation for the invaluable guidance, support and assistance which I received from Dr. Tristan Kohl. His enthusiasm and passion for his work, coupled with his willingness and incredible ability to teach is truly inspiring. It has been an honour to watch and learn from him during my time at The University of Groningen and without his assistance I would not have been able to complete this thesis.
I would also like to thank my research methodology supervisor Dr. Padma Rao-Sahib, as well as Dr.
Maria Grydaki for their valuable contributions to my analytical skills, econometric analysis capabilities and assistance in using Stata. Additionally, I would like to thank my co-assessor, Prof. Jan Oosterhaven for providing me with much food for thought, regarding my thesis, during my defence.
I would like to thank the friends that I have made in Groningen who have become my family away from home. The love, laughter and memories which we have shared have made the experience of studying abroad all the more worthwhile and if it hadn’t been for you I think I may not have made it.
I particularly want to thank my family for their undying love and support. To my parents, it is your incredibly wonderful guidance and unfailing wisdom which has made me the person that I am today and everything I have achieved I owe to you. My sister (and best friend) – thank you! You have been a shining light in my darkest hour, a listening ear when I needed it most and you have given me hope when I least thought it was possible.
Last but not least, I would like to thank my truly special boyfriend Rohan for keeping me going and being beside every step of the way. Your love, support and endless patience has meant more to me than I can ever express in words and I am so blessed to have shared this experience with someone as wonderful as you. Thank you for pushing me to challenge myself with the promise to catch me if I should fall – Baie dankie en ek is regtig lief vir jou!
ABSTRACT
This paper studies the extent to which institutions play a role in determining the size and direction of international bilateral trade flows. Moreover, the same effect is analysed for the case of trade between Sub-Saharan Africa and China; where despite a lower prevailing level of governance trade flows are shown to be high and persistently rising. To this extent, it is further investigated whether institutional homogeneity could be salient in explaining the recent surge in Sino-African trade. The empirical analysis is conducted using a gravity model approach estimated by modified Poisson estimation techniques. Panel data for 144 countries between 1996 and 2007 are considered. Findings indicate that both institutional quality and institutional homogeneity matter in determining bilateral trade flows; and that the effect of exporter institutions are less pertinent in the case of Sino-African trade.
JEL Codes: F14
Keywords: Bilateral trade flows, Gravity Model, Poisson Estimation Techniques, Institutional Quality, Institutional Homogeneity
“The world and all of us are defined by the divide between rich and poor, the haves and the have-nots, the developed and the underdeveloped… It constitutes the difference between the countries of the North and those of the South… Together with China, we are commonly defined by our situation as belonging to the South.”
South African President: Thabo Mbeki, 2001
“Everything is related to everything else, but near things are more related than distant things.”
Tobler, 1970
I. INTRODUCTION
The extent to which institutions play a role in shaping a country’s economic activity has been studied extensively in recent years. Institutions can be thought of as humanly devised behavioural constraints that govern and shape human interactions (North, 1990). These formal and informal ‘rules of the game’
have long been recognized as an important component of a well-functioning market; where institutions exist to reduce the uncertainties that arise from incomplete information and reduce transaction costs.
This subject has been particularly prominent within economic growth and development literature1, where economic scholars have recognized that lower levels of governance produce negative externalities for private transactions, increasing transaction costs and inevitably stifling economic growth.2 Several studies have emerged in recent years, in which a similar argument is proposed for the impact of governance on international trade (Wei, 2000); and as such signifies that the nature of institutions could be vital in shaping both the magnitude and direction of international trade flows.
Globalization has significant effects for global trading patterns. The rapid rise of China as an economic superpower is unprecedented within the contemporary world (van Dijk, 2008), and more specifically, its strengthening relationship with Africa is topical. Trade between SSA3 and China increased from US$7 billion in 2000 to US$59 billion in 2007, making it Africa’s third largest trading partner after The European Union (EU) and Unites States (USA) (Besada et al, 2008).
This improved trading relationship seems somewhat surprising, considering the notion that ‘institutions matter’, because the proposition that African states display some of the worst institutional reforms in the world is a highly feasible one. The continent is characterised by ethnic defragmentation, on-going civil disputes, high-levels of corruption, low regard for human rights, diminished levels of accountability and poor effective governance. These, coupled with poorly developed infrastructure, relatively small domestic markets, largely inadequate education systems and limited proficient commercial experience, make doing business in and with Africa challenging and moreover, a costly venture.
The fundamental question which this paper aims to answer is: does higher institutional quality enhance international trade, and if so, does this same effect hold in the case of trade between China and Africa?
Additionally, the paper assesses whether, in fact, institutional similarity could play a role in explaining the recent surge in Sino-African trade.
The notion that China’s foreign policy is being driven to a large degree by its domestic development strategy and enhanced need for resources (Zweig & Jianhai, 2005) offers vast opportunities for SSA in meeting these demands. China’s progress in the evolution of the world’s economic order could generate
1See Mauro (1995); Knack & Keefer (1997); Hall & Jones (1999); Acemoglu et al (2001, 2002); and Easterly & Levine (2003), La Porta et al (1997, 1998) and Rodrik et al (2004).
2Definitions of governance vary across studies. There seems to be a broad consensus that ‘good governance’ means that the government concerned is accountable, transparent, responsible, effective, and efficient; and follows rule of law, thereby assuring corruption is minimized. (United Nations, 2013)
3SSA and Africa are used interchangeably hereafter. Both terms refer to Sub-Saharan Africa, unless otherwise indicated.
positive gains for the growth performance of these African economies; and as such warrants further investigation. To date, no known studies have considered the effects of institutions on Sino-African trade and as such, this paper aims fill this void. Additionally, this paper strives to add value to the existing literature on institutional quality and trade by both circumventing the problem of endogeneity prevalent in studies of this nature; as well as by addressing several issues with respect to empirical application of the well known gravity model which have recently been brought to light.
The remainder of this paper is structured as follows: Section II provides an overview of related literature while section III provides an in-depth analysis of the econometric specification of the gravity model and discussion regarding the model estimation choice. This is followed by the details of the data employed for the empirical analysis. Section IV offers a fully-fledged examination of the effect of institutional quality and homogeneity on trade carried out with the empirical tools called for in this study. Finally, section V discusses the main findings in greater detail and concludes the investigation.
II. RELATEDLITERATURE
Traditionally international trade theory regards trade as being driven by endowments, technology, preferences and the nature of competition in the market. Consequently, it displays little regard for the notion that institutions play a hand in governing trade levels.
While trade as a percentage of world GDP has increased from around 25% in 1960 to about 58% in 2010 (UNCTAD, 2012a), not all countries have experienced the same level of trade growth. In fact, empirical literature has highlighted the ‘mystery of missing trade’ (see Trefler, 1995) which indicates that trade flows between ‘rich’ and ‘poor’ countries are far lower than levels predicted by traditional sources.
Furthermore, scholars have identified that much of international trade, is among similar countries (Baldwin 1970) and not between countries with disparate factor endowments as suggested by the well- known Heckscher-Ohlin trade theory. Increasingly, scholars have considered whether explanations for these apparent puzzles could be related to the institutional differences between countries. Moreover, institutional factors such as law enforcement, the securing of private property rights and informal institutions may play a significant role in determining trade costs – which in turn are an integral part of trade flow analysis. (Anderson and van Wincoop, 2004; Den Butter and Mosch, 2003).
Because international trade involves ‘business transactions’ between parties operating in different jurisdictions, against different institutional backdrops, involving different currencies and often speaking different languages, the institutions which govern exchange transactions between country pairs are likely to matter significantly more than in the case of domestic exchanges (Wei, 2000). Furthermore, the availability of reliable information and effective risk assessment are concerns for foreigners engaging in international transactions. Even in the case that a country lowers its formal barriers to trade, trade relations may be deterred, and trade volumes reduced, if outsiders do not believe that opportunism and rent-seeking has been mitigated by means of effective institutions. It is, therefore, important to understand to what extent institutions affect international trade flows between countries.
By comparison to the literature on institutions and growth, the impact of institutions on trade flows has received relatively less attention. Anderson and Marcoullier (2002) are amongst the main contributors.
Their seminal paper develops a model of import demand within an insecure world and demonstrates that the quality of institutions, like those which promote sound contract enforcement and secure private property rights, are highly relevant for shaping the volume of trade; in that corruption and imperfect contract enforcement act as a ‘hidden tax or tariff’ which increases transaction costs. Using a gravity model approach they are able to show that insecurity, as a result of low quality institutions, constrain trade by increasing the price of traded goods. These findings are confirmed by studies such as that of Anderson (2000), Anderson and Young (2000), Rodrik et al (2002) and De Groot et al (2004) who empirically show that there exists a significantly positive correlation between institutional quality and trade flows. A comparably smaller, yet growing body of literature stresses the dynamic nature of
institutions and assesses the effects of institutional reforms on bilateral trade. Studies indicate that improvements in domestic institutions are of key importance in promoting trade flows. In line with the existing literature, it seems fair to assert that institutions are expected to have a positive effect on economic activity in general and on trade in particular (Jansen & Nordås, 2004).
Various authors have considered that the potential mechanism through which institutional quality affects trade is that of comparative advantage; and have demonstrated this through studies focusing specifically on sectors or the production of goods that are more sensitive to contract enforcement and property rights.
Rajan and Lee (2003) show that bilateral trade volumes are more affected by institutional quality in sectors which are classified as being institutionally intensive. Berkowitz et al (2003) study the role of imperfect contract enforcement in international trade and the extent to which it affects comparative advantage. More specifically, they indicates that the quality of institutions in an exporting country are most important for enhancing trade, as a result of the difficulty of detailing the nature of complex products in complete ex ante contracts. Costinot (2004) further affirm that better institutions are a complementary source of comparative advantage in more complex industries. Nunn (2007) computes the contract dependence of final goods sectors and combines this with data on trade flows and the quality of judicial institutions in a country. He finds that countries with sound contract enforcement institutions specialise in the production of goods where relationship-specific investments are important. His findings advocate the notion that institutional frameworks matter more significantly in explaining countries’
patterns of trade than their endowments of capital and labour. Nunn and Trefler (2012) further indicate that institutions are ‘statistically and economically important determinants of comparative advantage even after controlling for endowments and technology’.
Building on this stream of literature, Levchenko (2007) proposes that institutional differences could very well be a source of comparative advantage in North-South trade; but due to the fact that the gains from trade, when institutions are the source of comparative advantage, are different across regions, the incentive to enhance domestic institutional quality may vary. More specifically, drawing on a broad series of literature which demonstrates that developed countries (“the North”) have better institutions than developing countries (“the South”)4, he shows that when institutions are the source of comparative advantage, the North gains most from trade while the South experiences conflicting effects. Furthermore, in an additional study (dated 2004) he provides empirical evidence to suggest that while institutional quality may be a source of comparative advantage between trading partners in equilibrium (where countries increasingly adopt the best attainable level of institutional quality to promote trade); if one partner has a sufficiently strong technological comparative advantage in the production of traded goods, institutions will not improve in either country. Hence, in cases where ‘other’ sources of comparative
4 La Porta et al (1997, 1998) and Acemoglu et al (2001, 2002) are examples from the literature.
advantage play a major role in driving trade flows, institutional quality may be less important for trade;
and it is this insight that may be particularly salient in the discussion of Sino-African trade.
While traditionally world trade has been dominated, to a large extent, by exchanges between countries of the North, as well as between the North and the South – South-South exports reached US$3.5 billion in 2010, accounting for 23% of world trade (UNCTAD, 2012b). The recent surge in international business activity between China and SSA is a striking characteristic of this new trend in South-South trade and investment.
Figure 1 displays the spectacular growth in Sino-African trade between 1996 and 2007 measured in millions of US dollars. Moreover, while Africa exhibited a trade deficit with respect to China for much of the last decade, from around mid-2000s, for the first time, Africa reached a trade surplus vis-à-vis China (De Grauwe et al, 2012). Figure 2 depicts the variation in aggregate African governance over the same period. Institutional quality within this graph has been measured using a database constructed by Kauffman et al (2002): commonly known as The World Bank World Governance Indicators (WBWGI);
which contains six indicators argued to proxy various aspects of governance5. The indicators have been scaled (between -2.5 and 2.5) such that lower values reflect weaker institutional quality and vice versa.
These figures indicate the extent to which Sino-African trade has increased, despite only marginal changes in prevailing African governance.
The growing significance of the relationship between China and Africa has spurred an increasing interest in its interactions and a burgeoning literature in both academia and the public and business press6. An apparent attribute of the China-Africa relationship, highlighted by the literature, is that it is evolving along a number of vectors, the most important of which are seen to be trade flows, investment and aid (Ajakaiye, 2009). Unfortunately, most of the work conducted has focused on the relationship in general with little focus on one single aspect such as that of Sino-African trade flows. Moreover, the limited research which has been done has to a large extent focused on the implications of the increasing trade levels between China and Africa and whether they will in fact promote growth or exploitation in various African states. The subject of governance on trade has yet to be explored in the case of Sino-African trade (Large, 2008) and to this extent, little clear insight exists with respect to whether institutional quality plays a role in these high and rising trade flows.
Looking specifically at Africa, it is a well-known that the continent displays underdeveloped market institutions, constraints on business competition and weak governance (Haber et al 2003); and much literature has proposed that it is this ‘crisis of governance’ which underlies its vast development problems (World Bank, 1989). In their paper, which focuses, specifically, on drivers of growth, Rodrik et al (2002) show that institutional quality has a positive effect on a country’s trade shares.
5 The disaggregated indicators of governance are further detailed in Table 1 contained in Section III.
6See Kaplinsky et al (2007); van Dijk (2008); Rotberg (2008); Zafar (2008) and Besada et al (2008)
FIGURE1: SINO-AFRICANTRADE FIGURE2: VARIATIONAFRICAN GOVERNANCE
Source: IMF, direction of trade and statistics Source: World Bank World Governance Indicators
However, in another paper in which the trade performance of SSA is assessed, without accounting for variances in institutional quality, Rodrik (1998) asserts that that trade liberalization has the same effect for trade in SSA as it does for other countries. These results seem contradictory and thus the question arises whether the positive correlation between institutional quality and trade, found within the afore- mentioned literature, is also relevant for Africa.
In line with the previously discussed notion of institutional quality affecting trade through comparative advantage, Bigsten et al (2000) demonstrate that the absence of formal efficient legal and judiciary systems, which promote political stability, ensure contract enforcement and secure private property rights, hinders interaction between manufacturing firms based in African countries and potential foreign partners. More specifically, the authors examine contracts and contract ‘flexibility’ using survey data for six SSA countries7. Their findings show that contracts are often dishonoured as a result of unpredictable behaviour and suggest that this ‘inefficiency’ may explain why foreign firms find it difficult to deal with African partners. Additionally, within his 2003 book “In Search of Prosperity”, Rodrik shows the importance of good institutions in promoting trade and growth performance of individual countries such as China, Botswana and Mauritius. Hence, these studies as well as those previously discussed, suggest that African trade should be stifled by its weak prevailing institutional quality. However, as shown above, this is not the current case for trade between China and Africa.
A common justification for this somewhat paradoxical yet maturing exchange relationship, within the literature, is that it is facilitated by Sino-African diplomacy which constitutes policies of respect for African state sovereignty, together with a guarantee from China that it in no ways seeks to propagate economic or political hegemony over any African state or their internal affairs. To this extent, African states are offered the opportunity of enhancing their international trade levels without having to meet
‘conditional’ trading requirements as imposed by traditional western trading partners (Renard, 2011).
Moreover, China is considered to demonstrate little regard for transparency and other aspects of good
7The countries considered for the study are Burundi, Cameroon, Cote d’Ivoire, Kenya, Zambia and Zimbabwe.
0 5000 10000 15000 20000 25000 30000
1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007
Exports from SSA to China Exports from China to SSA
-0.5 -0.45 -0.4
1996 1998 2000 2002 2004 2006
Composite Measure
governance within its own business dealings (Rotberg, 2008) which implies that it is uniquely positioned to take advantage of a structural alignment with SSA states. This may be further enhanced by a
‘similarity’ of disregard for high institutional quality.
This notion that trade is to some degree fostered between countries that are not “too” different is not a new one and could imply that it is not only good institutions which matter for trade but also, similarity in the institutions of trading partners. Dixit (2003) considers, following the work of Li (2003) which documents that formal rule-based governance has high fixed costs but low marginal costs, that informal institutions should be used in instances where the volume and scope of trade is limited; while a mix of formal and informal governance might be an optimum otherwise. He acknowledges that ‘different governance institutions are optimal for different societies, for different kinds of economic activity and at different times’. Rauch and Trindade (2002) study the effect of ethnic Chinese networks on trade and indicate that the presence of such networks has an important positive impact for bilateral trade which is larger for differentiated goods than homogenous products.
Literature which addresses institutional similarity and its effects on trade (and FDI) is scarce8; however, it could be particularly salient for the discussion on increased trade levels between China and Africa.
De Groot et al (2004) argue that similarity in perceived institutional quality may allow for similar norms or behaviours as well a mutual trust to develop in doing business. Furthermore, it can result in familiarity in informal business practise, which in turn reduce transaction costs and promote bilateral trade. Traders with similar institutional frameworks are better equipped to operate in each other’s environments which create a higher level of trading compatibility. The authors explicitly examine the effects of institutional indicators on bilateral trade flows, when the institutions are ‘controlled for’ as trade barriers within a standard gravity model. They make use of bilateral trade data in 1998 for a set of 175 countries in their investigations and capture the quality of institutions using the afore-mentioned WBWGI database. Their findings indicate that having a similar institutional framework promotes bilateral trade volumes even when controlling for institutional quality. Although intriguing, their study does suffer from empirical shortcomings, such as the use of cross sectional data, which limits the scope for inference.
Consequently, this paper aims to provide a more precise and reliable analysis of the effects of both institutional quality and homogeneity on bilateral trade flows. In line with the literature explored above, this paper investigates the hypothesis that institutional quality matters for shaping the direction and volume of international bilateral trade flows. Additionally, the extent to which this same relationship holds in the case of trade between China and Africa is explored. Lastly, the paper investigates whether institutional similarity could explain the recent surge in Sino-African trade relations.
8 See Lavallee (2005), Islam & Reshef (2006), Cuervo-Cazurra (2006) and Darby et al (2010).
In order to conduct the investigation, this paper employs a standard tool for modelling international trade flows; namely the gravity model. Recently, international trade literature has shown a renewed interest in the theoretical foundations of the gravity model; which has revealed that several issues with respect to its empirical applications remain unresolved. Problems of unobserved heterogeneity, heteroskedasticity and the incidence of zero-valued trade flows pose challenges to the conventional log-normal specification and the use of ordinary least squares (OLS) estimation techniques. In light of this, recent literature has suggested the use of various modified Poisson fixed effects estimations as viable alternatives.
In the face of on-going empirical investigation and debate regarding the ‘best’ alternative for gravity estimation, this paper considers the use of a zero-inflated negative binomial ZI(NB) maximum likelihood estimator (MLE) technique to explore the effects of institutional quality and homogeneity on trade flows of 144 country-pairs between 1996 and 2007. Moreover, the effects of both institutional quality and institutional homogeneity are investigated with relevance to both margins of trade; namely the number of trading-pairs (the extensive margin) and the volume of trade (the intensive margin).
Results obtained are compared to those estimated using a Poisson Pseudo-Maximum Likelihood (PPML) estimation technique to test for robustness. Endogeneity is discussed and addressed making use of instrumental variables.
III. METHOD ANDDATA
i. Econometric Specification Considerations
A standard tool for modelling international trade flows is the gravity equation, which is inspired by Newton’s law of universal gravitation. In general, the equation is based on the idea that trade between country pairs is an increasing function of their economic mass and a decreasing function of the geographical distance between them.
The model provides a direct link between trade flows and trade barriers i.e. those linked to countries’
and trading-pairs’ characteristics. It has become a workhorse for the study of bilateral trade flows, can be derived from a variety of models and most importantly for this study, allow for the explicit examination of trade costs. As previously discussed, trade costs are an integral part of trade flow analysis, and it is argued that besides distance, institutional factors such as law enforcement, the securing of private property rights and informal institutions play a direct role in determining both the frequency and magnitude of trade costs. Given that the aim of this study is to explore the effect of institutional qualities prevailing across different countries, the gravity model is assumed to be an obvious choice in that is can easily be augmented in order to substantiate the effect of intangible trade barriers which have implications for the costs of trading. Additionally, by including a multiplicative interaction term to the model (further discussed in the model specification to follow), we can unequivocally identify the extent to which the effects of ‘institutional distance’ matter for trade between Africa and China.
Most notably, the equation is considered to be a simple, yet versatile tool for modelling the determinants of trade which is sensible and intuitive while demonstrating substantial explanatory power (Deardoff, 1998). The model has proven to be empirically stable over time and across different samples of countries and methodologies.9
A basic form of the gravity equation is given by:
= (1.1)
Where is the bilateral aggregate import by importer i from exporter j in year t. is a proxy for the representative country’s economic mass (measured by GDP) and represents the physical distance between countries i and j. represents a variety of variables which control for a countries’ land size, the number of landlocked countries or islands in dyad and whether countries share a common language, border or colonial history. Traditionally, the model is transformed into a linear stochastic form and is estimated using OLS regression techniques. The log-linear format is specified as follows:
Ln( ) = + ln ( ) + ( ) + − ln ( ) + (1.2)
9 For a comprehensive review of the model see van Bergeijk & Brakman (2010)
where represents the constant term and is the error term, assumed to be independent and identically distributed across countries.
A shortcoming of the specification given in (1.2) is that it suffers from omitted variable bias, due to the fact that it does not account for unobserved price indexes. More specifically, as shown by Anderson and Van Wincoop (2003), trade intensity depends not only on bilateral trade expenses, but also on GDP- share average weighted multilateral trade costs. To this extent there is a need to include ‘multilateral resistance terms’ (MRT) to account for the ‘N-body properties’ of the trade system. While Anderson van Wincoop make use of cross sectional data within their study, Feenstra (2004) illustrates than within panel data studies, MRTs can be accounted for by including region specific fixed effects.
This yields a log-normal fixed-effect specification of the gravity equation which is commonly employed within the literature. The specification is as follows:
ln( ) = + ( ) + ( ) + − ( ) + + + + (1.3)
Where and represent fixed effects for importing countries i and j; and is included to correct for common shocks and trends across time periods considered.10
Problems Regarding the Log-Normal Specification of the Gravity Equation
Recent literature has highlighted two main problems with the log-normal specification of the gravity equation which should be considered for empirical analysis of trade data. These include dealing with zero-valued trade flows and the presence of heteroskedasticity. These problems are discussed below, followed by a description of a viable alternative which allows for their circumvention, namely the use of Poisson and modified Poisson estimation techniques.
Zero-Valued Trade Flows
A particular challenge regarding gravity model specification is dealing with zero-valued trade flows.11A typical trade dataset is likely to contain a large number of missing or zero-valued trade flows. The incidence of zero trade flows represents either that there is no trade between a specific country pair or that the trade level is insignificantly small and thus rounded down to zero. In contrast, the case of missing values points to information within the dataset which have gone unreported and as such, have the potential to be both positive or zero. A log-normal model specification is problematic when considering the incidence of zeros since the natural logarithm of zero is undefined. Various strategies to circumvent the ‘zero problem’ have been adopted in the literature12, however these often result in a substantial portion of valuable information about the patterns and characteristics of trade being excluded from the model or biased estimation results. The balanced panel dataset considered for this study covers
10Baier & Bergstrand (2007) indicate that fixed effects used to account for MRT should be time-varying to avoid issues pertaining to endogeneity bias, however, the inclusion of a large number of dyad dummy variables and valuable time invariant data with respects to institutional quality, time-invariant fixed effects have been included.
11For more details see Santos-Silva & Tenreyro (2006); Linders & De Groot (2006) and Helpman et al (2008).
12See Frankel (1997) for a description of various procedures and Kohl (2012: Chapter 2) for a brief overview of alternatives.
144 countries over 12 years. Overall, around 50% of the dependent variable data consists of missing values while a further 28% represents zero-valued trade flows. This highlights the necessity to employ a model specification which deals with zero-valued trade flows.
The Presence of Heteroskedasticity
A further specification consideration of the log-normal gravity equation is that it implies log-normality of the random error term, . Santos Silva and Tenreyro (2006) (hereafter SST) indicate that traditional gravity equations, including those which account for MRTs, suffer from severe heteroskedasticity. Log- linear transformation in the presence of such heteroskedasticity can lead to inconsistent biases because the expected value of the logarithm of a random variable depends on higher-order moments of its distribution i.e. Jensen’s inequality implies that E(ln ) ≠ ln E( ).
From a Log-Normal to a Poisson Specification
Given these problems with the log-normal specification rendering its use unfeasible, recent attention has been given to the use of Poisson and modified Poisson models in the context of gravity estimation. The application of Poisson family regressors to bilateral trade analysis was pioneered by SST (2006), and is considered to be a plausible approach in that it has a multiplicative non-linear form and as such is not restricted by the assumptions of OLS. Burger et al (2009) indicate that by using a Poisson regression model the estimates are adapted to the actual data which implies that the sum of the predicted values is virtually identical to the sum of the input values and thus is consistent in the presence of heteroskedasticity. Moreover, SST (2006) suggests that this method provides a natural way of dealing with zero values of the dependent variable – this is further discussed below. While, this family of models traditionally derives from count data analysis; Wooldridge (2002) indicates that Poisson estimators can be applied when considering non-negative continuous variables and as such it provides a viable alternative for modelling international trade data.
A particularly restrictive property of the standard Poisson model is that it assumes equality of the conditional mean and variance i.e. it entails an assumption of equidispersion. Empirical datasets are likely to exhibit over-dispersion and so, this model is argued to yield inefficient estimation of the dependent variable which is demonstrated by the spuriously large z-values and small p-values, due to downward biased standard errors (Cameron & Travedi, 1986). To address this problem, we propose using a modified Poisson estimation technique, namely a negative binomial (NB) MLE.
The fact that NB estimation is only considered in the case of over-dispersion implies the necessity to test for its prevalence. This can be done by observing the over-dispersion parameter, α. In the case that α = 0, the mean and variance of the distribution are considered to be identical i.e. implying the use of a Poisson model. In the case that over- or under-dispersion occurs i.e. α > 0 or α < 0, respectively, a modified Poisson model is deemed appropriate.
While the NB MLE accounts for the overdispersion within trade data it may predict fewer zeros for a given mean than the actual number of zeros prevalent within a specific dataset. In this case the dependent variable is said to suffers from an ‘excess zero’ problem, and a zero-inflated model is suggested to be more appropriate (Cameron & Travedi, 2009).
The ‘zero-inflated’ variation of the NB model lets zero counts occur in two ways: as a realization of a binary process and as a realization of a count process when the binary random variable takes on the value of 1. Hence, the ZINB MLE model is a two-part model and demonstrates a density function of:
g( ) = (0) + 1 − (0) (0), = 0,
1 − (0) , ≥ 1. (1.4)
Estimation of bilateral trade flows making use of this model allows for a distinction analogous to that between the ‘extensive’ and ‘intensive’ margins of trade. In the case of the first binary process, (.), trade flows are estimated using a logit model which essentially divides countries into those that never trade i.e. country pairs with exactly zero probability of trade and those who do not currently trade but that exhibit a latent possibility for trade. The second binary process, (.), is a count process and as such is estimated using either Poisson or NB estimation. Following the work of Kohl (2012) the ‘sub-divided’
country-pairs are hereafter referred to as the ‘passive’ and ‘active’ country groups. The set of variables considered for the (.) density need not be the same as those in the (.) density. However, in the case of international trade flows, factors which are considered to inhibit trade entirely can be considered to be the same as those which limit positive trade between country pairs. Hence, the same sets of regressors are considered for the two parts of the model.
In order to determine whether a standard or ZI model should be used to account for an excess of zeros a Vuong (1989) test can be used. The test statistic obtained is standard normally distributed, with positive values favouring the ZI model. A further examination as to whether a ZI model improves the fit of the data can be achieved by using the user-written countfit package (Long & Freece, 2006) in Stata which facilitates a multiple model comparison. It is shown in section iii below that these test criteria strongly favour the use of ZINB MLE over that of NB and Poisson (see p.22)
ii. Model
The gravity model which is used for the analysis of institutional quality on bilateral trade flows has a multiplicative form and is as follows:
= × × × × × × × × × × ( × ) ×
( × ) × × × × × × × ×
where , represents the population size of the importing (exporting) country at time t; , represents a country’s land area; represents a vector of categorical variables which may affect the ease of trading bilaterally, such as sharing a common border, linguistic or colonial links and which
(1.5)
represents the participation of a country in a common trade bloc membership. is a binary variable which controls for the prevalence of high oil or mineral endowment of a respective SSA country. The details of these individual variables are described in Table 10 contained in Appendix A.
Assessing the Effects of Institutional Quality on Trade Flows
The variables of particular interest for this paper are those which account for variations in institutional quality across different countries.
, represents the individual level of governance prevailing in the importing (exporting) country at time t. Although the concept of governance is widely discussed there is no consensus around a single definition of governance. Within this paper, the definition put forward by Kauffman et al (2010) is employed. To this extent, governance reflects ‘the traditions and institutions by which authority in a country is exercised. These include (a) the process by which governments are selected, monitored and replaced; (b) the capacity of the government to effectively formulate and implement sound policies; and (c) the respect of citizens and the state for the institutions that govern economic and social interactions among them.’ In line with this definition, the WBWGI dataset constructed by Kauffman et al (2002) has been employed to proxy for institutional quality. The index consists of six indicators of perceived institutional quality across countries which are in line with the above definition. The indicators scores have been scaled between -2.5 (weakest institutions) and 2.5 (strongest institutions). Table 1 provides an overview of each of these indicators.
In order to identify the overall effect of institutions on bilateral trade flows, a general index of governance, represented by a simple arithmetic average of each of the separate indicators, is considered (hereafter referred to as the composite indicator). In line with the literature discussed in section II, it is hypothesized that quality of institutions and bilateral trade share a significantly positive correlation.
Because governance is measured as an index which varies between -2.5 and 2.5, the coefficients represent semi-elasticities and thus are interpreted with respects to an increase in institutional quality of one standard deviation from the mean. The effect on trade of a one standard deviation from the average institutional quality provides a good indication of the average impact of variation in institutional quality on trade.
By considering the variable , , we are able to identify the effect of a one unit change in the institutional quality( , )of the country of origin and importing country on the bilateral trade flows ( ) between them. To this extent,it is expected that an increase in the institutional quality of a country of one standard deviation from the mean, will result in a higher level of trade as well as a reduced probability that a country pair will fall into the passive trading group i.e. we postulate that and will be significantly positive for active trading group and significantly negative for the passive trading group.
In order to assess whether the same effect holds in the case of China and Africa, this study employs a multiplicative interaction term, namely (SnAf × Inst). SnAf is a binary variable which takes on the value of 1 in the case that a trading partnership comprises of China and one of the SSA countries within the sample. By including this term, we allow for the development of a conditional hypothesis (Brambor et al, 2005). More specifically, we are able to explicitly examine the effects of institutional quality on bilateral trade in the case of China and Africa i.e. the coefficients of this variable reflects the effect of institutions on trade dependent on the condition that these belong to a trading relationship between China and Africa. As a consequence, these coefficients are not interpreted as the average effects of institutions on trade as would be the case in a more linear-additive regression model but rather reflect the conditional marginal effects.
In considering the literature discussed in section II, it is postulated that in the case of China and Africa the effect of institutions on trade will be weaker than the ‘general’ effect. Figures 1 and 2 contained in section II explicitly illustrate that trade levels have risen while the composite effects of institutional quality have remained relatively constant over the period under investigation. This implies that institutions matter to a lesser degree in the case of trade between China and Africa. To this extent, we hypothesize that the and are likely to depict a negative sign for the active trading group. The intuition is then that the original positive effect of institutions on trade volume as measured by , is reduced.Additionally, the effect for the passive trading group is expected to be positive.
In addition to the effects of institutional quality on trade, this paper considers whether institutional similarity has an effect on bilateral trade, with a view to exploring whether this could perhaps be a factor which may explain the recent surge in Sino-African trade. In order to test this, the model incorporates
, as a binary variable used to measure institutional homogeneity; which takes on the value of 1 in the case that countries within the country-pair are considered to be institutionally homogeneous and TABLE1: INSTITUTIONALVARIABLES
VARIABLES DETAILS
(a) The process by which governments are selected, monitored and replaced:
Voice and Accountability (VA) Perceptions of the extent to which a country’s citizens are able to participate in the selection of the government together with freedom of expression, freedom of association and free media.
Political Stability & absence of
Violence/Terrorism (PV) Perception of the likelihood that the government will be destabilized or overthrown through unconstitutional or violent means, including politically motivate violence and/or terrorism.
(b) The capacity of the government to effectively formulate and implement sound decisions:
Government Effectiveness (GE)
Perception of the quality of public services, the quality of civil service and the degree of its independence from political pressures, the quality of formulation and implementation, and the credibility of the government’s commitment to such policies.
Regulatory Quality (RQ) Perception of the ability of the government to formulate and implement sound policies that permits and promotes private sector development.
(c) (c) The respect of citizens and the state for the institutions that govern economic and social interactions among them:
Rule of Law (RL)
Perception of the extent to which agents have confidence in and abide by the rules of society, and in particular the quality of contract enforcement, property rights, the police, and the courts, as well as the likelihood of crime and violence.
Control of Corruption (CC) Perception of the extent to which public power is exercised for private gain, including both petty and grand forms of corruption, as well as ‘capture’ of the state by elites and private interests.
Source: Kaufmann et al (2010)
zero otherwise. The measure is determined following criteria put forward by De Groot et al (2004).
More specifically, homogeneity is determined by the criteria that the difference in institutional effectiveness (taken as an absolute value) between two countries in a country pair is below a specified fraction of the sample standard deviation of a particular indicator of governance. Hence, in the case that the absolute difference exceeds the specified fraction, countries are considered to be institutionally heterogeneous. The use of a binary variable which allows for the measurement of a discrete impact is argued by De Groot et al (2004) to be clear and concise in its interpretation. To account for the sensitivity of the dummy variable specification, this paper further considers several similarity criteria i.e.
one, two and three standard deviations.
In line with the notion that trade may be fostered between countries that share a similar institutional framework, similar norms and behaviours that promote the development of mutual trust and familiar business practices as put forward by De Groot et al (2004), it is hypothesized that institutional homogeneity will have a positive effect on trade volumes such that in the case that two countries are considered to be institutionally similar, there is a positive effect for trade volume. Additionally, similarity in institutional quality is expected to promote the probability of trade between county pairs (i.e. reduce the likelihood of an inflated zero occurring) and to this extent we postulate that will depict negative sign for the passive trading group.
Model Analysis Considerations
Renewed interest in the theoretical foundations of the gravity model, in recent years, has led to the identification of common errors which are made by researchers in empirical analysis. Within their 2006 article, Gravity for Dummies, Baldwin and Taglioni (hereafter B&T) highlight three common errors, which they refer to as Gold, Silver and Bronze medal errors, which should be acknowledged and avoided. The authors argue that these mistakes can lead to severely inconsistent and biased estimation.
The Gold Medal Error
While the Newtonian physics principle of gravity suggests that the gravitational pull between two objects is dependent on distance, mass and a gravitational constant; B&T point out that this is not the same in the case for trade. Following the previously discussed work of Anderson and van Wincoop (2003), it has become well recognised that bilateral trade flows depend not only on distance and mass but also on country-specific trade resistance terms. Hence, the relationship between mass, distance and trade is what has been referred to as a ‘gravitational un-constant’ which depends on the scale of trade resistance that stems from the characteristics of each individual country. Moreover, because these characteristics are not constant over time, they have specific implications for panel data studies. While various remedies have been discussed within the literature, this paper employs country specific fixed- effects as well as a time-fixed effect (as specified in section III) to correct for this mistake.
The Silver Medal Error
Baldwin and Taglioni (2006) further highlight that the gravity model is a modified expenditure function in that it depicts the demand by one country for goods produced in another country. This implies that it specifically explains unidirectional trade flows. However, in practise the average of two-way exports is often employed. The authors highlight that this can lead to misspecification in the instance that the log of the averages is mistaken for the average of the log i.e. the ‘Silver medal’ error. A certain way to avoid making this mistake is to consider only import or export values for the dependent variable, while including each country pair twice in the sample. Anderson and van Wincoop as well as Feenstra’s derivations of the gravity equation indicate the necessity for the dependent variable to be unilateral.
Thus, to avoid this mistake, the panel data study considered for this model is arranged by country-pair, regardless of missing values or zero-valued trade flows. This implies that each of the countries yields two observations i.e. once as an importer and once as an exporter.
The Bronze Medal Error
The final ‘Bronze medal’ error occurs when nominal trade values are inappropriately deflated making use of a common (often US) aggregate price index. This is pointed out to be erroneous in that it may result in spurious correlation since various trends in global inflation rates exist. The authors propose the use of a time fixed effect in order to circumvent this problem in instances where every bilateral trade flow is divided by the same price index. As detailed below, bilateral imports (used for measuring the dependent variable) have been deflated using a common price index, however the inclusion of the time fixed effect (see 1.3) assists in the avoidance of the bronze medal mistake.
iii. Data
The balanced panel dataset considered for this study consists of data for 144 countries between 1996 and 2007. This yields 144×143×12 = 247 104 observations. Table 2 lists the countries included in the dataset. Notably, the sample includes 46 out of 48 potential SSA countries.13 The time period investigated was chosen based on data availability and is assumed to provide fair results based on the dynamic properties of institutional quality over time.
A substantial portion of the data used has been taken from Kohl (2012). Appendix A provides a table of variables and their data sources. Annual bilateral imports (c.i.f. imports and exports f.o.b. in US$
millions) are used as the dependent variable. In the case of missing values (which constitutes around 50% of the dataset), the country’s trade partner’s bilateral exports are used as a proxy for that country’s bilateral imports. A 10% c.i.f. rate is assumed when exports replace missing imports. The trade data has been deflated using the US Consumer Price Index (All Consumer Goods, 1983-4 = 100). As previously discussed, this paper employs a time fixed effects to control for common shocks and trends over time which further prevents any estimation bias from the use of a common price index for deflation of data.
13Due to lack of data availability with respect to intuitional quality, Cape Verde and Somalia have been excluded from the sample.
