The Analysis of Potential FTA Effect on Trade Flows
between Ukraine and its Trading Partners as Regards to its
Vector of Economic Integration: A Gravity Model
Approach.
Master's Thesis, academic year 2013-2014 Faculty of Economics and Business Universiteit van Amsterdam Kateryna Soloviova, 10605134 Supervisors: dr. W. E. Romp dr. C.A. Stoltenberg July 2013
Abstract
The thesis investigates the effect of potential trade liberalization on bilateral trade flows for Ukraine depending on the vector of its economic integration: a customs union with Eurasec (Eastern vector) or a free trade area with the European Union (Western vector). While conducting the analysis the augmented gravity model was used. The effect of FTA establishment was computed taking a distance, a common language and the difference in corruption levels into account. The results demonstrate that the Eastern scenario will lead to more intense trade.
Table of contents 1. Introduction 4 2. Literature Overview 7 2.1. Technical Literature 7 2.2. Empirical Literature 10 3. Methodology 18
3.1. The Gravity Model Specification 18
3.2. Potential FTA effect on Ukraine-Eurasec
and Ukraine-EU trade flows 19
4. Data 21
5. Estimated Results 27
5.1. ‘What-If’ Analysis for Ukraine, Eurasec and the EU 27
5.2. Sensitivity analysis 30
6. Conclusions 35
References 36
1. Introduction
In the modern world economy more and more countries become involved in integration processes. Trade volumes increase substantially over time (see picture 1.1) and in order to lower operation costs and mitigate barriers to trade many free trade agreements (hereinafter referred to as ‘FTA’) have been established since early 1990s, such as the EU, NAFTA, ASEAN, MERCOSUR and SICA and the vast amount of bilateral trade agreements, e. g. China and various European countries or free trade agreements between former Soviet republics. Some economic integration unions are currently at the stage of elaboration. For example, Australia holds talks on FTA establishment with Saudi Arabia and other GCC countries; South Korea and Canada have initiated their bilateral free trade agreement in June, 2014; EU is constantly liberalizing trade with Asian partners and strengthening the ties with the East. Therefore, the effect of foreign trade areas on trade flow dynamics should be quantitatively estimated.
Picture 1.1. Gross Volumes of Trade for OECD Countries in 1980-2013 (billion USD)
Source: OECD.
Many researchers see e.g. Baier and Bergstrand (2007), DeRosa (2008) and Urata and Okabe (2007), conducted a thorough analysis of such an effect. As a rule they use gravity models augmented by various parameters that can affect trade. Gravity models
0 200 400 600 800 1000 1200 1400 1600 1800 2000
are proven empirically to be a reliable technique for trade estimations. Even though FTA impact on trade usually gets positive and significant values, its magnitude varies considerably. For example, an increase in trade can be relatively modest if exports and imports between particular countries have been historically intense (as in case of the EU, see Bun & Klaassen (2006)) or the countries’ economies are small (as in case of intra-Maghreb trade, see DeRosa (2008)). The inclusion of additional control variables can also lower the effect of concern. Moreover, sometimes the estimated results appear to be counterlogical. The studies described below show that a common language could have negative impact on the effect of FTA on trade flows, whereas the estimated parameter for distance can become positive.
The thesis also uses the gravity model of trade but contrary to existing literature it focuses mainly on levels of corruption of countries under study. This paper assumes corruption levels to be fixed over time as the fluctuations of corruption perception indices within the observed period can be neglected. Two different approaches were used to compute aggregated corruption perception indices, however, the estimated results suggest that differently computed differences in levels of corruption did not substantially change both the value of parameters and their significance.
The analysis uses GDP, real effective exchange rates, the existence of a free trade area between two countries, distance, language and difference in corruption levels as explanatory variables. The latter three parameters are included in the regression both directly and indirectly (the effect of distance, language and corruption is corrected for FTA in the model).
Favourable geographic location of Ukraine and other political and economic reasons stipulated for two different integration scenarios for the country. One is focused on the Eastern direction. In particular, it implied the entrance of Eurasec customs union, which currently includes Belarus, Kazakhstan and Russia. Ukraine became an observer-country of the union in 2002, but then the integration process was suspended. The other scenario is concentrated on the Western vector. It assumed an establishment of a free trade area between Ukraine on one side and the European Union on the other side within 10 years from the date of signature of the association agreement (the date of signature was set on December, 2013, however, the talks were not successful). Adoption of one of the strategies, either ‘Eastern’ or ‘Western’ one
will have large-scale consequences both for Ukraine and the world economy. This thesis estimates the effect of foreign trade area foundation on trade flows as regards to each of the aforesaid potential scenarios.
The results from the gravity model regressions are used to estimate the effect of a FTA establishment between Ukraine on one side and Eurasec and the EU on the other.
The results suggest that due to similar levels of corruption and a common language the FTA effect on Ukraine-Eurasec trade flows is more than ten times higher than for Ukraine-EU trade. In particular, the establishment of the customs union with Eurasec member countries will lead to at least 193.6% increase in trade, whereas the free trade area with EU members will intensify trade with Ukraine by 9.7%
The thesis is structured as follows: the next section provides the summary of technical literature needed to ‘build-up’ an appropriate modification of the gravity model empirical literature that used the gravity model approach to assess FTA impact on bilateral trade flows. Section three explains the construction of the gravity model variation and the estimation of FTA impact in case Ukraine enters one of the economic integration unions; section four focuses on the dataset and the construction of aggregate real effective exchange rate, corruption level and aggregate corruption level explanatory variables. The following section describes results obtained in the gravity model analysis. The final section summarizes the main conclusions of the thesis.
2. Literature Overview
Many researchers (see e. g. Coulibaly & Fontagné (2014), Martinez-Zarzoso & Nowak-Lehmann (2003), Zhu (2013)) tried to estimate the effect of free trade agreements using the gravity model of trade and variations on this model. The gravity equation is one of the most well known tools to distinguish ex-post effects of international economic integration unions (FTAs, custom unions etc.)2.
2.1. Technical Literature
Head & Mayer (2013) provides a theoretical framework for constructing and applying gravity models of trade. It shows the empirical evidence of proportionality of economy size (GDP) to exports and imports and inverse proportionality of distance to trade. However, the effect of distance was ignored by many economists for a long time; instead of that the concepts of ‘borderless world’, ‘the death of distance’ and ‘world is flat' possessed the central position in economic theory. Head & Mayer (2013) not only demonstrates the importance and benefits of such an approach, but also contains some criticisms. For example, conventionally, the error term is expressed as the difference between data and predictor. After taking logarithms the error term is presented by:
ln (𝑢!) = ln 𝑋!" − 𝑙𝑛(𝑋!")
where 𝑋!" is the prediction of 𝑋!", which depends on the vector of observables and the vector of coefficients for these variables.
The variance of the error term depends on the aforesaid variables, therefore, conditional mean zero assumption for the error term is not satisfied, so the estimates are biased and inconsistent when using OLS method. Secondly, the model assumes trade flows to be strictly positive, however, a real data set can include many zero flows. Head and Mayer (2013) suggest several ways to eliminate the problem. One of the most commonly used methods is adding 1 to a trade flow value. In such a case a zero trade flow will remain zero after taking logs instead being an unobserved component. However, the authors mention that the results might become biased. As various simulations show such arithmetical changes can not only change the magnitude of the estimated coefficient, but also affect its sign.
To estimate the effect of free trade agreements on trade I will use an augmented gravity model approach. The canonical gravity model (see e. g. Shepherd, (2013)) is19:
𝑋!" = к!!
! ! !!
!!"! exp (𝜀!"),
where 𝑋!" represents bilateral trade flows (e. g. the sum of exports and imports), 𝑌! 𝑎𝑛𝑑 𝑌! are GDPs of countries i and j respectively, 𝐷!" is distance between countries, 𝜀!" is the error term.
The log-linearized version of the model is:
log 𝑋!" = 𝑙𝑜𝑔к + 𝛼𝑙𝑜𝑔𝑌! + 𝛽𝑙𝑜𝑔𝑌!− 𝛾𝑙𝑜𝑔𝐷!" + 𝜀!"
where 𝜀!" is an error term.
Several papers that use gravity model approach as a major tool in their research and also provide a thorough description of their methodology will be discussed in the following paragraph. For example, Bun & Klaassen (2002) aims at assessing the impact of the euro on trade within the monetary union. The effect is expected to be positive, as a common currency eliminates exchange rate risks and minimizes transaction costs. It explains carefully the importance of fixed effects introduction, which can assist in correction for various invariant trade determinants: either time or country unchanged. The authors also argue the possibility of correlation between the error term and other parameters. As the data sample for this research consists mostly of the states within the EU, trade policies and, as a result, cultural and political issues, which might have a vast effect within a monetary union, can influence trade itself. However, this potential correlation is neglected in the study. Unlike other augmented gravity models discussed below, the regression in this paper is an augmented distributed lag type of model, i. e. it also includes first and second lagged values of exports for estimating the current one. Furthermore, it takes into account real exchange rates, real exchange rate volatility (is encompassed to measure an accurate direct effect of Euro, because, as stated before, a common currency reduces real exchange rate volatility), a monetary union dummy variable and FTA dummies for two different areas: Europe and America. The model also includes first and second
lags of the aforesaid indicators. As the research suggests, exchange rate volatility appeared to be insignificant for exports dynamics. Euro has increased exports by 3,9% in 1999, and the effect became larger in subsequent years. Bun & Klaassen (2002) argues that the long-run impact will amount to 37,8%, with a half-life reached by 2006.
Bun & Klaassen (2006) studies the effect of the euro on trade. They show the difference in estimations after correcting for explanatory variables that enable to assess the direct effect of a common currency. The specification of the regression hardly differs from a standard gravity model, including only FTA and euro dummy variables and country-pair specific time trends together with standard variables. After estimating the regression without time trends the authors state, that there is an upward bias in the impact of euro introduction. As bilateral trade flows between countries within data sample were growing constantly throughout the whole estimated period and a common currency was established only at the end of this sample, euro dummy variable captures also the effect of independent increase in trade, getting bigger than it actually is. After the inclusion of country-pair specific time trends in the regression Euro appeared to account for 3% increase in bilateral trade. However, the estimate lies at the edge of significance/insignificance. To sum up, the research argues that a common currency effect is exposed to upward bias and is not as substantial as it is believed to be.
Anderson and Wincoop (2003) provides a thorough analysis of the impact of geographical factors (distance and border) on trade between the states of the USA and the provinces of Canada and explains the distance puzzle. The paper includes the dummy variable, which is equal to 1 for inter-provincial trade and 0 for state-province trade flows. When constructing the gravity model the authors take into account that small economies are more sensitive to trade barriers than big ones. The paper also includes the unobservable cost factor equal to the loglinear function of bilateral distance and the border parameter (it is equal to 1 when the regions are located in the same country and equal to 1 plus the cross-border tariff, otherwise). The regression contains the ratio of trade to the product of two countries’ GDP as the dependent variable and distance, a common border and price indices as explanatory variables. The estimated results fully confirm the assumption of the magnitude of trade barriers depending on the size of the economy. In particular, the resistance is higher for
Canadian provinces than for US states. When replacing resistance terms with region dummies the effect of distance becomes more negative. The paper provides an explanation for this phenomenon, stating that fixed or neglected transportation costs make the impact of distance more substantial.
The key outlines for the build-up of an augmented gravity model are presented in the table 2.1.1 hereinafter.
Table 2.1.1. Technical literature summary
Name Main concepts
Head & Mayer (2013)
Empirical evidence on positive impact of GDP and negative impact of distance on trade; criticism on particular crucial assumptions of gravity models and solutions of their elimination.
Shepherd (2013) Gravity modelling outlines for economic researchers; the basic technique explanation.
Bun & Klaassen (2002)
Fixed effects introduction; usage of augmented distributed lag type of model for the gravity equation.
Bun & Klaassen (2006)
Introduction of the idea of direct effect; comparison of the results for the total effect and the direct effect corrected by explanatory variables.
Anderson and Wincoop (2003)
Explanation of high coefficients for a common border variable; explanation of the sign and magnitude of the effect of distance on trade flows.
2.2. Empirical Literature
Baier & Bergstrand (2007) uses an augmented gravity model, which includes free trade agreement, language and adjacency (same border) binary variables together with standard ones. This study aimed at providing a thorough empirical analysis and in order to conduct it the data starting from 1960 to 2000 for 96 potential trading
partners was used with zero trade flows being excluded. The paper gives a grounded confirmation of a predictable answer. Baier & Bergstrand (2007) uses country fixed effects to account for multilateral price terms, going in line with Anderson and Wincoop (2003) methodology. The paper includes fixed effects rather than random effects for two reasons. Firstly, the endogeneity bias can occur mostly due to time invariant parameters that are likely to influence both FTA existence and trade volumes. Thus, controlling for them in the regression makes the results consistent. Secondly, empirical literature analysing fixed and random effects suggests that the former are proven to be more reliable. Having conducted regression analysis the researchers argue that the unbiased average treatment effect varies from 0.61 to 0.76. Stated differently, an FTA will increase two member countries’ trade about 100% after 10 years. However, the study contains several gaps. For example, it does not provide any information on the impact of an FTA on trade between non-member country pairs. Furthermore, it does not take into account that FTAs themselves may differ. Finally, the drawback I consider to be the most substantial is the fact the direct effect of a free trade area can be influenced by other parameters, such as distance or economic size.
Coulibaly & Fontagné (2014) offers a detailed analysis of impact of infrastructure besides the one of trade liberalization. The authors collected data on paved road coverage between trading partners being members of Economic Community of West African States. They included common border, common language, number of transit borders, infrastructure and FTA dummy as explanatory variables in their gravity model. All the estimation results are comparable to previous findings. Furthermore, the percentage of paved roads coverage between countries appeared to have a broad positive effect (a 10% increase in the parameter results in 11% increase in trade flows). However, the authors also mention some criticism to the obtained results. As the import flows specified by the category of the product for West African countries are very low, the estimates might not be robust. Moreover, the region of concern is not fully explored; therefore, several observations are missing.
In my opinion, one of the most thorough analyses was conducted in Urata & Okabe (2007). Before estimating the effect of free trade agreements on trade they observe two indicators of intra-FTA interdependence: a relative share of intra-FTA members trade (a ratio between intra-region trade and the trade between the region and the rest
of the world) and a trade intensity index (a ratio between a relative share of intra-FTA members trade and the trade between the region and the rest of the world divided by world trade). These indicators demonstrated a substantial increase after FTA establishment in all the regions except the EU (which is in line with my own estimates). Such kind of an observation illustrates the necessity of actual estimation of a preferential trade agreement formation. The specification of the augmented gravity model is comparatively simple and straightforward. Besides ‘standard’ variables and FTA dummies it also includes income, adjacency and language. The sample consists of 63 countries being representatives of different regions and different integration unions. As expected, the establishment of a preferential trade agreement positively influences the volume of trade both for aggregated data and regional analyses. Furthermore, the regional evaluation showed that MERCOSUR is a very closed economic integration unit while the EU and NAFTA are more closed then AFTA. The authors also mention several criticisms to their research. For example, the study does not take into account the fact that in some FTAs the tariffs have been removed gradually, though the dummy variable could not capture that.
There is also a vast variety of regional studies that try to embrace the impact of different factors on bilateral trade flows using the gravity model approach. For example, Martinez-Zarzoso & Nowak-Lehmann (2003) concentrates on Latin American region in particular, thoroughly investigating the historical perspective of future MERCOSUR member states. Their augmented gravity model includes the following components: bilateral trade flows, GDP, distance, population, preferential trade agreement dummy variable, infrastructure, real exchange rate and a difference in per capita incomes in a country pair. One of the main drawbacks in the paper is the data sample, which is relatively out-dated (1988-1996). The researchers run the regression also controlling for fixed and random effects and obtained the following results: the estimate for preferential trade agreement dummy was positive and significant for intra-bloc trade, thus, entering a free trade agreement and subsequent trade liberalization fosters international trade.
The research of Rahman, Bin Shadat & Das (2006) is focused on the South-Asian region. The paper states that the empirical literature published before this paper and observing the same free trade area has provided different results not only for trade creation and diversion effects, but also for the significance of parameters. A specific
feature of this augmented gravity model specification is the inclusion of import-to-GDP ratio, which aims at capturing a country’s openness. In authors’ opinion, it should be significantly positive as since economic openness encourages trade. Evidently, the paper proves a strong positive impact of a free trade agreement on trade for aggregated data. However, the effect is not equal for different member states, i. e. several countries have been negatively affected by the creation of SAFTA. It could happen due to negative export performance and/or structural limitation. Though, more developed countries of the region, such as India, considerably benefited from economic integration.
Another paper, which goes in line with the aforesaid studies concentrates on South Africa. To conduct ex-post analysis research the Kanda & Jordaan (2010) also use a gravity model, augmented with real effective exchange rate, population and preferential trade agreement dummy variables. For estimating trade creation and trade diversion effects for South Africa, the paper observes 39 major trading partners of the country. Kanda & Jordaan (2010) provides a thorough explanation for the signs of the model parameters. For example, population demonstrated a strong negative effect on trade. The authors state that growth in population (taking into consideration specific feature of South Africa) leads to higher crime rates and illiteracy, which curb trade. Contrary to papers mentioned above this estimation shows that a preferential trade agreement appeared to have a negative effect on trade. The authors provide a reason for that, mentioning that SADC agreement has not fully taken effect by the time the study was published, i. e. not all trading partners that aimed at joining the free trade area actually did it by 2008. Moreover, the tariffs even between the parties to the agreement have been completely liberalized.
The following paper contains not only the analysis of FTAs, which were already established, but also projections of creating such an area in the region of concern. In the first section Zhu (2013) estimates the effect of an FTA on trade for various economic integration unions. The model used is augmented by the following indicators: population, language, belonging to the same continent, a country’s area and colonial relations between countries. Zhu (2013) comes up with a statement that on average two countries within the same FTA have 144% higher trade than two countries otherwise. In order to make projections for trade intensity between China, Japan and Korea the author compares the median FTA effect with the one for this
region. As the latter indicator exceeds the former he infers that the abovementioned region has ‘above average success’ in increasing trade by FTA establishment. Further the paper discusses projections under four different scenarios: a baseline scenario (where GDP and population growth assumptions are based on data from the World Bank and Blue Chip Economic Indicators, respectively): scenario 1, which assumes low, medium and high rates of growth and low growth rates for Japan in particular, scenarios 2-4 based on medium growth case, whereas scenario 2 implies RMB appreciation and Yen depreciation, scenario 3 assumes tariff elimination and scenario 4 accounts for non-tariff barriers. All the scenarios demonstrated that FTA between China, Japan and Korea would increase merchandise trade by 21%-46% and services trade by 49%-79%.
Brun, Carre`re, Guillaumont and Melo (2005) provides the analysis of so-called ‘distance paradox’. Sometimes distance shows positive effect on trade in gravity models. The paper includes population and a common language, a common land border, a common colonizer, the condition of being landlocked, a free trade area and a common currency dummies as control variables in the regression. When using such gravity model variation Brun, Carre`re, Guillaumont and Melo (2005) reports that the effect of distance is positive. The puzzle remains even after augmenting the model with transportation costs. The positive effect of distance on trade was significantly reduced when remoteness was included in the regression. However, the ‘distance paradox’ has different magnitude for different countries. In particular, high-income countries experience the puzzle only for standard gravity model, but it disappears when the aforesaid controls are included. The positive effect always remains for low-income countries.
DeRosa (2008) estimates the impact of FTA establishment on bilateral trade flows under different integration scenarios for Maghreb countries. The augmented gravity model used in the study contains distance, joint GDP, joint GDP per capita, language, a common border, the possibility of being landlocked, the possibility of being an island, land area, a common colonizer, the current status of the colony, the possibility of ever being a colony, preferential trade agreement and joint FDI stocks. The first integration scenario takes only AMU area into account. The next option assumes EU and US bilateral FTA establishment with leading Maghreb countries: Algeria, Morocco and Tunisia. The last scenario implies the creation of the same free trade
area including all Maghreb countries. The results show that the FTA effect under the first scenario is positive but not large due to small size of Maghreb members’ economies. The second scenario showed substantial enhancement of trade compared to the previous option. The third variant, that assumed FTA extension to all Maghreb countries demonstrated only a modest improvement in free trade agreements as regards to the second scenario. Moreover, the coefficients for EU-AMU free trade area were smaller than those for US-AMU, because currently the trade between the European Union and Maghreb is more extensive.
Fidrmuc and Fidrmuc (2009) explores the effect of language on trade. It analyzes bilateral trade flows among 29 countries that are current or potential EU members. The languages taken into account are English, French, German and Russian. The gravity model is augmented by adjacency, the possibility of belonging to the same federation in the past (applicable in Eastern Europe) and language control variables. Moreover, language parameters cover both native languages and communicative probabilities, i. e. the probability that two randomly chosen inhabitants of the countries in a country pair can communicate in a particular language. The study proves languages to have a strong effect on trade; however, the effect can be ambiguous. In some cases a common language had negative impact on trade flows. The authors state that it could happen because the countries do not use fully their language opportunities or would rather prefer English, which is the most widely spoken in the countries from the dataset. The other reason mentioned in the study is historical animosity that can create barriers to trade.
The main results of the abovementioned papers are listed in the table 2.2.1. Table 2.2.1. Key results from the empirical literature
Name Control variables included Results (FTA effect)
Baier &
Bergstrand (2007)
Free trade agreement, language and adjacency
Average treatment effect of a free trade area varies from 0.61
to 0.76. Coulibaly &
Fontagné (2014)
Common border, common language, number of transit
Positive and significant impact of FTA on trade; broad positive
borders, infrastructure and FTA dummy variable.
effect of paved roads on trade.
Urata & Okabe (2007)
Income, adjacency and language.
Positive influence of preferential trade agreement on trade both for aggregate data and
region-specific analyses. Martinez-Zarzoso
& Nowak-Lehmann (2003)
Population, preferential trade agreement dummy variable, infrastructure, real exchange rate and a
difference in per capita incomes in a country pair.
Positive and significant estimate for the FTA effect on trade.
Rahman, Bin Shadat & Das (2006)
Population, common border, common language and import-to-GDP ratio.
A strong positive impact of the free trade area establishment on trade, which varies depending on the member state.
Kanda & Jordaan (2010)
Real effective exchange rate, population and
preferential trade agreement dummy variables
Negative effect of preferential trade agreement on trade.
Zhu (2013) Population, language, belonging to the same continent, a country’s area and colonial relations between countries.
Two countries within a FTA have on average 144% more intense trade than two countries otherwise.
Brun, Carre`re, Guillaumont and Melo (2005)
Population, language, a common land border, a common colonizer, the condition of being landlocked, a free trade
Positive impact of distance on bilateral trade flows contrary to the expected results. The impact is bigger for trade flows between low-income countries.
area, a common currency, transportation costs, remoteness
DeRosa (2008) Joint GDP per capita,
language, a common border, the possibility of being landlocked, the possibility of being an island, land area, a common colonizer, the current status of the colony, the possibility of ever being a colony, preferential trade
agreement, joint FDI stocks
Positive and modest impact of FTA on trade within AMU area and when establishing EU-AMU and US-AMU free trade areas; positive and substantial impact of FTA on trade when
establishing bilateral free trade agreements between Us and EU, on one side, and leading
Maghreb countries on the other side.
Fidrmuc and Fidrmuc (2009)
Adjacency, the possibility of belonging to the same federation in the past, language
A positive effect of common languages on trade for most of bilateral flows. A negative effect appeared for specific country groups.
3. Methodology
3.1. The Gravity Model Specification
The gravity model described in previous chapter will be augmented using factors that can enhance or discourage international trade. One of the most deteriorating factors in international trade is corruption. Many countries, which have the status of strategic world trade partners, are fully exposed to it. Therefore, I will include variable 𝛥𝐶𝑂𝑅𝑅𝑈𝑃𝑇!" that reflects the difference in corruption perception indices between two countries within a country pair subject to Transparency International.
Further, it would be reasonable to include real effective exchange rate, 𝑅𝐸𝐸𝑅!"#
in
the gravity model. For example, Guechari (2012) argues that real effective exchange rate has significant positive impact on the trade balance of a particular country.Moreover, several dummy variables will also be included in the regression. First of all, I want to capture the existence of free trade agreements in regions of concern, as it is the major parameter of the model. Thus,
𝐹𝑇𝐴!"# = 1 𝑖𝑓 𝑏𝑜𝑡ℎ 𝑟𝑒𝑔𝑖𝑜𝑛𝑠 𝑎𝑟𝑒 𝑝𝑎𝑟𝑡𝑖𝑒𝑠 𝑡𝑜 𝑠𝑎𝑚𝑒 𝑓𝑟𝑒𝑒 𝑡𝑟𝑎𝑑𝑒 𝑎𝑔𝑟𝑒𝑒𝑚𝑒𝑛𝑡
0 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒
Another special feature that can significantly affect the volumes of trade is a common language. As Egger & Lassmann (2011) argues, that a common language has a significant positive effect on trade, which increases over time. Moreover, the effect is even higher for emerging markets. Therefore, I am going to include another dummy, which stands for the regions speaking the same language:
𝐿𝐴𝑁𝐺!" = 1 𝑖𝑓 𝑏𝑜𝑡ℎ 𝑟𝑒𝑔𝑖𝑜𝑛𝑠 𝑠𝑝𝑒𝑎𝑘 𝑡ℎ𝑒 𝑠𝑎𝑚𝑒 𝑙𝑎𝑛𝑔𝑢𝑎𝑔𝑒
0 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒
Typically, free trade agreements occur regionally. One can think of the following examples: Greater Arab Free Trade Area, ASEAN Free Trade Area, Common Market for Eastern and Southern Africa etc. These multilateral trade agreements usually imply shorter distances and sometimes common languages between members. Furthermore, high levels of corruption can prevent the establishment of a free trade area between trade partners. Therefore, parameters 𝐿𝐴𝑁𝐺!", 𝛥𝐶𝑂𝑅𝑅𝑈𝑃𝑇!" and 𝐷!"
will include the direct effect of the free trade dummy variable as well as their effect corrected by distance, level of corruption and a common language.
Finally, in order to mitigate the bias in the estimates a country pair fixed effect and a time fixed effect will be incorporated in the gravity model. The time fixed effect will be captured by a set of dummy variables, each of which represents a particular year within the period observed.
Taking into account all the abovementioned factors the gravity model will look the following way: 𝑥!"! = 𝛼!" + 𝛽!𝑦!" + 𝛽!𝑦!"+ 𝛽!𝐹𝑇𝐴!"#+𝛽!𝑑!"𝐹𝑇𝐴!"#+ 𝛽!𝛥𝑐𝑜𝑟𝑟𝑢𝑝𝑡!"𝐹𝑇𝐴!"#+ 𝛽!𝑟𝑒𝑒𝑟!" + 𝛽!𝑟𝑒𝑒𝑟!" +𝛽!𝐿𝐴𝑁𝐺!"𝐹𝑇𝐴!"# + 𝛾! !"#! !!!""# 𝑇𝐹𝐸!"+ 𝑢!"#
where lower case letters express the logarithms of the upper case notations, 𝛼!" is a country-pair fixed effect, 𝑇𝐹𝐸!" is a set of year dummies and 𝑢!" is an error term.
3.2. Potential FTA effect on Ukraine-Eurasec and Ukraine-EU trade flows
The potential effect of trade agreements between Ukraine and former Soviet republics and Ukraine and EU will be estimated. Recently Ukraine was standing on the brink of strategic changes. It had two scenarios of developing its economic integration: an Eastern and a Western one. The Eastern scenario implied entering Eurasec, i. e. a custom union initiated by Russia. Currently Ukraine has a status of an observer country. The Western option suggested an association agreement with the European Union, which was signed in March (political provisions) and June (economic provisions). One of key outlines of the agreement is a build-up of a free trade area in a ten-year period, starting from the day of signing it.
Therefore, I will estimate the effect of a free trade area on bilateral trade flows for each of the vectors of economic integration by including the aforesaid regions in the country-pair data set. It is possible to conduct such an analysis with the existing data and get reliable estimates, because the average treatment effect in this case does not depend on time:
𝐸𝑓𝑓𝑒𝑐𝑡!"#$%
= 𝐸𝑓𝑓𝑒𝑐𝑡!"#$%!!− 𝐸𝑓𝑓𝑒𝑐𝑡!"#$%!!= 𝛽! ∗ 1 − 𝛽! ∗ 0 + 𝛽!𝑑!" ∗ 1 − 𝛽!𝑑!" ∗ 0
+ 𝛽!𝛥𝑐𝑜𝑟𝑟𝑢𝑝𝑡!"∗ 1 − 𝛽!𝛥𝑐𝑜𝑟𝑟𝑢𝑝𝑡!" ∗ 0 + 𝛽!𝐿𝐴𝑁𝐺!" ∗ 1 − 𝛽!𝐿𝐴𝑁𝐺!"∗ 0 =𝛽!𝑑!"+ 𝛽!𝛥𝑐𝑜𝑟𝑟𝑢𝑝𝑡!" + 𝛽!+ 𝛽!!𝐿𝐴𝑁𝐺!"
4. Data
The data sample is comprised of 15 country pairs being representatives of three different regional free trade agreements. The data goes back to 1990 up to 2013 in order to have the observations for both conditions: before and after the establishment of the free trade areas described below.
The first region is represented by NAFTA members, i. e. the USA, Canada and Mexico, where the former two share the same language and demonstrate similar levels of corruption, whereas the latter differs from the US in both of the parameters. Second region consists of the MERCOSUR member states: Argentina, Brazil and Paraguay. Argentina and Brazil have close corruption perception indices, but speak different languages, with the situation between Argentina and Paraguay being opposite. Finally, the Netherlands, Belgium, Portugal, Estonia, Poland, Romania, Germany – all being EU-members – and Moldova, which has close economic and political ties with the EU without possessing its membership, make up the third region. The European region is presented more largely than the former two in order to conduct further analysis. The Netherlands shares the same language and has relatively close level of corruption with Belgium; however, it does not have any of the indicators in common with Portugal. The following six country pairs demonstrate trade relations between Estonia, Romania and Moldova on one side and their geographically and economically close and distant partners on the other: Netherlands and Poland, Belgium and Poland, Romania and Germany, respectively. Estonia and Romania are recent EU members; the size of their economies and their locations are comparable to the parameters of Ukraine. Republic of Moldova is currently been strongly integrated in the economy of the European Union, which has much in common with ‘Ukraine-EU’ economic relations scenario. Such a selection assists in capturing the specific features of trade dynamics between Ukraine and the union to assess the potential impact of a FTA on trade flows between them.
Picture 4.1. Bilateral Trade Flow Dynamics 1990-2013 (million of Euro)
Source: United Nations Trade Statistics Division.
0 100000 200000 300000 400000 500000 600000 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013
USA -‐ Mexico
0 500 1000 1500 2000 2500 3000 3500 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013Argentina -‐ Paraguay
0 1000 2000 3000 4000 5000 6000 7000 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013Netherlands -‐ Portugal
Trade volumes in the regions of concern showed positive dynamics both after signing but not yet establishing and after establishing free trade agreements. The most indicative in this case are bilateral trade flows between Argentina and Paraguay, which increased roughly by 50% when MERCOSUR became effective (picture 4.1). EU member country pair did not demonstrate a substantial increase in trade after the effective date of Maastricht Treaty. The reasons for that are historically intensive intra-region trade and integration process that dates back to Schuman Plan of 1950. All the variables are expressed in nominal terms, as in accordance with Ben Shepherd (2013) real indicators deflated by a particular factor do not adequately capture the unobserved multilateral resistance terms, and could produce biased results19.
The abovementioned corruption perception index will be included as a fixed effect due to the following reasons. As the research of Dutt & Traca (2010) suggests high levels of corruption usually discourage trade10. However, the existence of very high tariffs together with the disposition towards corruption can lead to the opposite results. In order to avoid huge tariff payments people resort to bribery spending less, thus, the expenditures of trade agents decrease and this fact facilitates. To sum up, the abovementioned paper proved the significant effect of corruption on international trade. As the data from Transparency International is available only from 1995, the index will be included as a fixed effect in the regression. But prior to doing that I am going to examine corruption index dynamics of the countries of concern (picture 4.2). One can see that for the most of the countries levels of corruption remained relatively constant. Nevertheless, there are two ‘outliers’ in the sample. First of all, Argentinian index plummeted dramatically in 1995, this trend faded away in 1998. This spike has occurred due to the severe economic crisis, which started in 1995 because of Mexico’s currency devaluation (tequila crisis) that triggered the fears of same processes in Argentina. The interest rates increased significantly in 1995, and that symbolized the beginning of a sharp recession18.
Belgium also demonstrated a significant decrease in corruption perception index in 1997-1999. In accordance with Transparency International report this downturn was explained by political shifts in the country. The Kingdom was hit by numerous scandals involving political parties, e.g. Dutroux case, which led to several strategic policy changes and hit the stability of Belgium21.
Therefore, one can conclude that there were no significant changes in corruption perception indices and due to the lack of data I will assume them to be fixed in the augmented gravity model.
Picture 4.2. Corruption Perception Index Dynamics 1995-2013
Source: Transparency International.
Furthermore, to estimate the potential effect of Ukraine entering either a free trade area with the European Union or a customs union with Eurasec, these regions of concern will also be included in a pooled regression analysis. Eurasec is comprised of three member states: Russia, Belarus and Kazakhstan. In this study European Union will be represented by first twelve Euro Area members: Austria, Belgium, Finland, France, Germany, Greece, Ireland, Italy, Luxembourg, Netherlands, Portugal, Spain. Such a selection allows to get the values of real effective exchange rate for a longer period.
The data on aggregate real effective exchange rate for the Euro Zone (which will be represented by the aforesaid twelve member states) is available, however, there is no joint REER values for Russia, Belarus and Kazakhstan. In order to compute it for the countries of concern I will use the guidelines of Bruegel (2012). Their working paper defines it as: 0 10 20 30 40 50 60 70 80 90 100 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 CPI USA CPI CAN CPI MEX CPI ARG CPI BRA CPI PAR CPI NDL CPI BEL CPI PRT
𝑅𝐸𝐸𝑅! =𝑁𝐸𝐸𝑅!∗ 𝐶𝑃𝐼! 𝐶𝑃𝐼!(!"#$%&')
where 𝑁𝐸𝐸𝑅! is a nominal effective exchange rate of a particular country, which they identify as:
𝑁𝐸𝐸𝑅! = 𝑆(𝑖)!!
(!)
!
!!!
where 𝑆(𝑖)! is the nominal bilateral exchange rate between a particular country and
its trading partner i,
𝑤(!) is the weight (share) of partner i in the total trade flows of the country under
study.
𝐶𝑃𝐼! is the consumer price index of the country of concern,
𝐶𝑃𝐼!(!"#$%&') is an aggregate consumer price index of the trading partners, which is computed as follows:
𝐶𝑃𝐼!(!"#$%&') = 𝐶𝑃𝐼(𝑖)!!(!)
!
!!!
where 𝐶𝑃𝐼(𝑖)! is the consumer price index of trading partner i, N is the number of trading partners considered.
Such an approach does not exclude intra-region trade, however, as the paper of Bruegel (2012) states the bias, which can occur with respect to this approach, is relatively small and does not impose large impact on the estimate.
In order to construct a corruption perception index for regions, which consist of more than one country, two different approaches will be used. One of the ways to create an aggregated parameter is to compute a weighted average for all the EU members, using the distance from each of the countries to Ukraine as a denominator. As mentioned before, distance is proved to be a significant factor in influencing trade volumes; therefore, Ukraine tends to have more intense import-export flows with its close geographical partners. Another way to pool the data on corruption of many countries
in one index is using trade flows as a denominator, because the amount of trade is a key indicator for my analysis. Therefore, corruption perception indices for Eurasec and Euro Zone will be represented as:
𝐶𝑂𝑅𝑅𝑈𝑃𝑇(!""#$"!%$) = !!!!𝐴𝑉_𝐶𝑂𝑅𝑅! ∗ 𝑇𝑅𝐴𝐷𝐸!
𝑇𝑅𝐴𝐷𝐸!
! !!!
where 𝐴𝑉_𝐶𝑂𝑅𝑅! is the average of corruption perception indices of country i across the whole observed period,
𝑇𝑅𝐴𝐷𝐸! is the sum of total exports and imports between Ukraine and country i across the whole observed period; or:
𝐶𝑂𝑅𝑅𝑈𝑃𝑇(!""#$"!%$) = !!!!𝐴𝑉_𝐶𝑂𝑅𝑅! ∗ 𝐷!
𝐷!
! !!!
where 𝐷! is the distance between Ukraine and country i.
The estimations for each of the parameters will be provided in the subsequent section. The distance between Ukraine and aforesaid regions will also be calculated as a weighted average of distances between the country and the Euro Zone/Eurasec members with bilateral trade flows being a denominator, due to the fact that there is a strongly positive impact of distance on trade, as stated before.
5. Estimated results
5.1. ‘What-If’ Analysis for Ukraine, Eurasec and the EU
The estimated results for the abovementioned country pairs are presented in tables 5.1.1. and 5.1.2. Firstly, the aggregated corruption perception indices for EU and Eurasec were denominated by distance. Both countries’ GDP estimates are positive and significant which goes in line with the basic principle of gravity models. The estimate for distance, when corrected for free trade agreement effect, doesn’t show the negative sign for all model variations. This parameter is significant at 99% level when language corrected for FTA is dropped out of the regression. The so-called ‘distance puzzle’ is a common phenomenon known in the related literature, it has been mentioned above. In the final regression the parameter loses its significance. When including it in the regression the estimate for FTA dummy changes substantially. Firstly, the parameter loses its significance for regressions (4), (5) and (6). Nevertheless, it becomes significant at 95% level for the final regression specification.
The parameters for real effective exchange rates of both countries within the pair appeared to be different. The estimate for the first country is significant at 99.9% level for all model variations. 𝑟𝑒𝑒𝑟!" has positive impact on bilateral trade flows. It
increases trade by 93.3% and this increase hardly varies for regressions (3)-(7). The real effective exchange rate for country i absorbs the impact of 𝑟𝑒𝑒𝑟!"; in particular, this estimate for country j is insignificant.
Corruption is proved to be an influential parameter for bilateral trade flows. The difference in corruption levels within a country pair negatively affected trade for all model variations including this variable.This factor is also proved to be consistent, as it showed significance at 99.9% level.
Table 5.1.1. Regression estimates for corruption indices denominated by distance
Language is the parameter that showed the results opposite to the ones expected and reported in corresponding literature. Trade partners speaking the same language experience a decrease in export and import flows, with the estimate being significant at 95% level. This might be explained by relatively low trade intensity between countries due to the reasons, not captured in the regression (same structure of produced goods and services, size of the economy, historical or political reasons etc.). Moreover, on the one hand, several trade partners do not fully use the advantage of the common language, giving preference to English. On the other hand, some countries which do not share the same language can speak English, as the most widely spoken language worldwide, to conduct their business.
The establishment of the customs union between Ukraine and Eurasec will increase bilateral trade flows by 253.6%. Similarly, the free trade area between Ukraine and the EU accounts for a 16.9% increase in imports and exports between the regions. Thus, the results show that smaller distance between trade partners and similar levels of corruption given trade liberalization will lead to more intense trade.
F-test showed that the parameters are jointly significant for all gravity model variations presented herein.
Table 5.1.2. uses corruption perception indices denominated by trade for EU and Eurasec. The estimates for separate parameters, such as GDP, real effective exchange rate, corruption corrected by FTA etc do not vary substantially from the ones stated in table 5.1.1. However, the difference between FTA effect on Ukraine – Eurasec and Ukraine – the EU trade flows becomes even bigger. The former generates a 344.6% increase in trade flows between the regions of concern, while the latter accounts for 11.1% more trade. The bigger difference can occur due to the fact that bilateral trade flows (used as a denominator for aggregated corruption indices) between Eurasec members and Ukraine have been more intense than between the EU and Ukraine.
Table 5.1. 2. Regression estimates for corruption indices denominated by trade
When dividing the FTA effect into components one can see the following results: the estimates for 𝛽!𝑑!", 𝛽! and 𝛽!!𝐿𝐴𝑁𝐺!" are identical for panel data with both distance and trade aggregated corruption perception indices. The computed effect of distance
is even higher for Ukraine – the EU pair (0.339) than for the Ukraine – Eurasec pair (0.308). As the obtained estimates for language are negative it adversely affects the trade between Ukraine and Eastern European partners of concern (compared to zero effect for the trade with the European Union). However, the impact of differences in corruption not only absorbs the negative effect of language, but also makes the FTA effect on Ukraine – Eurasec trade flows at least 15 times higher than for Ukraine – the EU. Similar levels of corruption (denominated by distance) generate the impact equal to 22.4% decrease in trade for the former pair and to 82.8% decrease for the latter. When the indices of concern are denominated by trade the effect changes to -2.6% and -83.7%, respectively. Therefore, the difference in corruption is the key factor affecting the overall FTA effect.
5.2. Sensitivity analysis
The results of the sensitivity analysis for different corruption perception indices estimates are presented in tables 5.2.1. and 5.2.2.
Table 5.2.1. Regression estimates for corruption indices denominated by distance
Table 5.2.2. Regression estimates for corruption indices denominated by trade
Regression (1) drops time fixed effects out of the gravity model. In this case the estimates for separate regression parameters are identical in tables 5.2.1. and 5.2.2. However, the coefficients for FTA dummy variable and language corrected by FTA become insignificant. Corruption remains an important factor impacting trade demonstrating 99.9% significance.
The upward bias can be seen in the effect of a free trade area establishment for regression (1). Bilateral trade flows between Ukraine on one side and Eurasec and the EU on the other will be enhanced by 232.3% and 20% for distance denominated indices, respectively, with the latter estimate being more than twice bigger than in the baseline gravity model.
Higher FTA effect values were obtained also for trade denominated indices: 308.4% (40.4% higher than in table 5.1.2) increase for Ukraine – Eurasec trade flows and 14.6% (10.4% higher than in table 5.1.2) increase for Ukraine – the EU trade flows. The bigger difference in FTA effect estimates in table 5.2.2 than in table 5.2.1 can
occur due to the difference in corruption levels which becomes smaller for Ukraine – Eurasec pair and, in turn, bigger for Ukraine – EU when trade flows are used as a denominator instead of distance for aggregated corruption perception indices computation.
Regression (2) differs from the model described in the methodology section because of the omission of country-pair fixed effect (captured by 𝛼!") and the inclusion of direct effect of distance, corruption and language dummy variable. Tables 5.2.1 and 5.2.2. suggest that FTA coefficient becomes more significant (99.9% level). The estimate for this dummy becomes strongly negative, which does not confirm the common hypothesis (trade liberalization leads to trade encouragement). Moreover, in model (2) a common language corrected by FTA loses its significance even at 95% level (compared to the baseline regression). The distance between the countries in a country pair and the common language dummy demonstrate negative and 99% level significant and a huge positive 99.9% level significant effects on trade, respectively, which goes in line with common knowledge and related literature. However, the difference in corruption levels appeared to be insignificant contrary to the expected results. The same parameter corrected for FTA existence is not significant, despite the fact that it was significant at 99.9% level for the baseline regression. The computed effects of Eurasec customs union and the EU foreign trade area establishment became counterlogical and opposite to the ones obtained in various gravity model specifications. Trade liberalization accounts for 25.8% less trade with Eurasec and 199.5% more trade with the EU for distance denominated corruption indices and, respectively, 29.5% less and 209.3% more trade for trade denominated corruption indices. Tables 5.1.1 and 5.1.2. provide opposite estimates for Ukraine – Eurasec bilateral trade flows and much bigger estimates for Ukraine – the EU exports and imports.
Regression (3) uses bilateral real exchange rates instead of real effective exchange rates in the gravity model. In case of Ukraine – Eurasec trade flows the real exchange rate between Ukrainian hryvnia and Russian ruble will be used. Russian Federation possesses the first place in geographical structure of the international trade of Ukraine both in imports (28.7% of the total imports volume in the 1st quarter of 2014) and exports (19.5% of the total exports volume in the 1st quarter of 2014). The real exchange rate for Ukraine – the EU trade flows for the years 1999 - 2013 will be
captured by hryvnia – euro currency pair, whereas the exchange rate for hryvnia – deutsche mark will be taken for the period from 1993 (no earlier data on Ukrainian currency is available due to the transitory period that occurred after obtaining the independence in December, 1991) to 1998. According to the United Nations Trade Statistics Division the trade flows between Ukraine and Germany have always been the most intense for the former compared to the rest of the European countries. Moreover, the volume of imports and exports between the two countries also surpasses the ones between Ukraine and many other trade partners of the world, with only China accounting for bigger share of international trade of Ukraine. This trend hardly varied for the years 1993 – 1999 (see picture 5.2.3.).
Picture 5.2.3. Volumes of trade between Ukraine and its trading partners 1993-1998 (billion USD)
Source: UN COMTRADE
In this case the coefficient for FTA dummy loses again its significance. The bilateral real exchange rate has a modest positive effect on trade flows (a 2.9% increase), which is significant at 95% level. The difference in FTA effects on trade between the regions of concern becomes even bigger than in regression (1) and the baseline regression. Specifically, for distance denominated corruption indices trade liberalization will lead to a 240.4% increase in imports and exports between Ukraine and Eurasec and a 4% increase between Ukraine and the European Union. As table 5.2.2 suggests the same parameters there show that the trade can be increased by
0,00 2,00 4,00 6,00 8,00 10,00 12,00 14,00 16,00 18,00 1993 1994 1995 1996 1997 1998 Russia Eurasec Germany "The EU (12 Jirst Eurozone members)"
330.6% for the former pair and even decreased by 1.4% for the latter. A huge difference in the price levels between advanced European economies and Ukraine lead to trade deterioration between them, while comparable prices within Post Soviet countries make it easier to enhance imports and exports.
6. Conclusions
The thesis aimed at analysing the effect of potential trade liberalization between Ukraine and Eurasec and Ukraine and the European Union taking into account the distance between regions of concern, the ability to speak the same language and the difference in the levels of corruption. In order to estimate the aforesaid effect the gravity model of trade was used.
The results suggest that similar corruption levels can mitigate the negative effect of this phenomenon on trade, whereas similar language substantially contributes to trade flows enhancement. Thus, the establishment of the customs union with Eurasec appeared to be more encouraging for bilateral trade flows with Ukraine than the creation of the free trade area with the European Union. Specifically, the former option will increase trade at least by 193.6%, while the latter can enhance imports and exports only by 9.7% under the same conditions.
The sensitivity analysis demonstrated that the results obtained are robust as the estimates do not vary substantially from the baseline regression estimates and confirm the claim that Ukraine – Eurasec trade liberalization can be much more lucrative than Ukraine – the European Union FTA.
The analysis conducted can be used for estimating the effect of potential FTAs establishment for various regions and countries in order to choose the option that will lead to more intense trade flows.
There is also room for deepening the research. The gravity model can be augmented by other parameters affecting trade. The data set can also be expanded to provide more precise results. Moreover, as Ukraine signed the economic part of the association agreement on June 27, 2014, it might be reasonable to analyse the dynamics of the trade flows between the two regions in 10 years’ time (when all barriers to trade will be removed by the both parties to the agreement) and verify if the results obtained herein are reliable predictors of the trade flows development.
References:
1. Anderson, James E., and Eric van Wincoop (2003). Gravity with Gravitas: A Solution to the Border Puzzle. American Economic Review, 93(1): 170-192.
2. Baier, Scott L., Jeffrey H. Bergstrand (2007). Do Free Trade Agreements Actually Increase Members’ International Trade? Journal of International Economics, 71 (2007): 72-95.
3. Brun, Jean-Francois, Ce´line Carre`re, Patrick Guillaumont, Jaime de Melo (2005). Has Distance Died? Evidence from a Panel Gravity Model. The World Bank
Economic Review, World Bank Group, vol. 19(1): 99-120.
4. Bun, Maurice J. G., Franc J. G. M. Klaassen (2002). Has the Euro Increased Trade? Tinbergen Institute Discussion Papers, 02-108/2, Tinbergen Institute. 5. Bun, Maurice J. G., Franc J. G. M. Klaassen (2006). The Euro Effect on Trade is
not as Large as Commonly Thought. Oxford Bulletin of Economics and Statistics, 69(4), 473-496.
6. Coulibaly, Souleymane, Lionel Fontagné (2004). South – South Trade: Geography Matters. CEPII, Working Paper 04(08).
7. Cruces, Guillermo, Quentin Wodon (2003). Argentina’s Crisis and the Poor, 1995-2002. London School of Economics, Discussion Paper № DARP 71.
8. Darvas, Zsolt (2012). Real Effective Exchange Rates for 178 Countries: A New Database. Bruegel Working Paper 716 (2012/06).
9. DeRosa, Dean A. (2008). Gravity Model Analysis. Maghreb Regional and Global Integration: A Dream to Be Fulfilled. Peterson Institute. ISBN 978-0-88132-426-6. 10. Dutt, Pushan, Daniel Traca (2010). Corruption and Bilateral Trade Flows: Extortion or Evasion? The Review of Economics and Statistics, MIT Press, vol. 92(4): 843-860.
11. Egger, Peter H., Andrea Lassmann (2011). The Language Effect in International Trade: A Meta-Analysis. Economics Letters, Elsevier, vol. 116(2): 221-224.
12. Fidrmuc, Jan, Jarko Fidrmuc (2009). Foreign Languages and Trade. CEDI
Discussion Paper Series, 09(03), Centre for Economic Development and
Institutions(CEDI), Brunel.
13. Guechari, Yasmina (2012). An Empirical Study on the Effects of Real Effective Exchange Rate on Algeria’s Trade Balance. International Journal of
14. Head, Keith, Thierry Mayer (2013). Gravity Equations: Workhorse, Toolkit and Cookbook. CEPII Working Papers, 2013(27), CEPII research center.
15. Kanda, Patrick T., Andre C. Jordaan (2010). Trade Diversion and Trade Creation: An Augmented Gravity Model Study for South Africa. TIPS Small Grant
Scheme Research Paper Series, 2010.
16. Martinez-Zarzoso, Inmaculada, Felicitas Nowak-Lehmann (2003). Augmented Gravity Model: An Empirical Application to MERCOSUR – European Union Trade Flows. Journal of Applied Economics, vol. 6(2).
17. Rahman, Mustafizur, Wasel Bin Shadat, Narayan Chandra Das (2006). Trade Potential in SAFTA: An Application of Augmented Gravity Model. CPD Working
Paper, 61, Centre for Policy Dialogue.
18. Saxton Jim (2003). Argentina’s Economic Crisis: Causes and Cures. A Joint
Committee Study, 2003, Joint Economics Committee United States Congress.
19. Shepherd, Ben (2013). The Gravity Model of International Trade: A User Guide. United Nations Publication, ARTNeT Gravity Modeling Initiative.
20. Urata, Shujiro, Misa Okabe (2007). The Impact of Free Trade Agreements on Trade Flows: An Application of the Gravity Model Approach. RIETI Discussion
Papers, 07052, Research Institute of Economy, Trade and Industry.
21. Walgrave, Stefaan, Frédéric Varone, Patrick Dumond (2006). Policy With or Without Parties? A Comparative Analysis of Policy Priorities and Policy Change in Belgium, 1991 to 200. Journal of European Public Policy, 13(7): 365-395. 22. Zhu, Edward (2013). The Case for Free Trade Agreements: Historical
Perspectives and a Projection for China, Japan and Korea. Stanford University.
Appendix. Data Sources
Parameter Source
GDP (1990-2012) The World Bank
GDP (2013) National Bureaus of Statistics Real Effective
Exchange Rates (1990-2013)
Bruegel Working Paper 2012/06
Corruption Perception Indices
Transparency International
Volumes of Trade (1990-2013)
United Nations Trade Statistics Division
Volumes of Trade between Ukraine and Belarus, Kazakhstan and Russia (1992-1995)
Statistical Register of Ukraine for the years 1992, 1993, 1994, 1995.
Bilateral exchange rates (LCU to USD)
The World Bank
Consumer Price Indices
