Trade flows of rtas signed between developing economies

Abstract

This paper studies the effect on trade flows of RTAs signed between developing economies. It uses a variation of the gravity model of trade to asses five RTAs: Mercosur, The Andean Community, SICA, the EU, Chile-China.

Contents

Abstract iii

List of Figures vi

List of Tables vi

List of Formulas vi

1. Introduction viii

1.1Background viii

1.2 Problem definition x

1.3 Research Objective x

1.3.1 Major research question x

1.3.2 Minor research question xi

1.4 Theoretical Framework xi

1.4.1 The Gravity model of trade xi

1.4.2 Research Methodology and Design xii

1.4.3 Research Assumptions xii

1.4.4 Research Limitations xii

1.5 Thesis Structure xiii

2. Literature Review xiii

2.1 Trade Creation and Trade Diversion xiv

2.1.1 Trade Creation xiv

2.1.2 Trade Diversion xvii

2.1.3 Gross Trade Creation xviii

2.2 Empirical Evidence from SS RTAs xx

3.Theoretical Framework and Research Methodology xxi

3.1 Theoretical Framework xxi

3.1.1 Multiple Regression Analysis and Model Building xxi

3.1.2 Regression Model Diagnosis xxii

3.1.3 The Gravity Model of Trade xxiii

3.1.4 Research Assumptions xxvii

3.1.5 Research Limitations xxvii

3.2 Research Methodology xxvii

3.2.1 Research Type and Approach xxvii

3.2.2 Data Collection xxx

4. Findings and Results xxxi

4.1 The effect of RTAs xxxi

5. Conclusions xxxiii

6. Appendix xxxiv

7. References xxxvii

List of Figures

Figure 1 – Trade Creation………………………………………………………………….

Figure 2 – Trade Diversion

Figure 3 – Trade Creation Proper vrs. Gross Trade Creation

Figure 4 – Multiple regression hyperplane

List of Tables

Table 1 – Dummy Variable Interpretation…………………………………………………..

Table 2 – RTAs assessed and Members

Table 3 – Regression results of individual years

Table 4 – Regression results of PCS

List of Formulas

Formula 1 – Gravity model equation………………………………………………………

Formula 2 – Log linear form of the gravity model…………………………………………

Formula 3 – Current gravity specifications………………………………………………..

Abbreviations

CGE: Computable General Equilibrium

COMESA: Common Market for Eastern and Southern Africa

FTA: Free Trade Agreement

GATT: General Agreement on Tariffs and Trade

GDP: Gross Domestic Product

MERCOSUR: Mercado Común del Sur – RTA signed between Brazil, Argentina, Uruguay and Paraguay

NAFTA: North American Free Trade Agreement

OLS: Ordinary Least Squares

PCS: Pooled Cross-Section

PTA: Preferential Trade Agreement

RIA: Regional Integration Agreement

RTA: Regional Trade Agreement

SICA: Sistema de Integración Centro Americana – RTA between Honduras, Costa Rica,

El Salvador, Guatemala, Nicaragua Panama and Belize

SS: South-South

UNCTAD: United Nations Conference on Trade and Development

WB: World Bank

WITS: World Integrated Trade Solution

WTO: World Trade Organization

1. Introduction

Background

Four hundred and sixty two RTAs have been notified to the WTO up to February 2010 (WTO,2010). From 1948-1994 the GATT received one hundred and twenty four notifications of RTAs, and since its creation in 1995, the WTO has received over 300 RTA notifications, (WTO,2010). This trend of forming trading blocs is likely to become stronger as more RTAs are currently under negotiation.

Of particular interest to economists, and the focus of this paper, are South-South RTAs, that is, RTAs signed between countries of low income levels. There are reasons to believe that SS RTAs may not only fail to stimulate economic growth among member countries, but also hinder growth for these countries.

In their book Regional Integration and Development, Winters and Schiffer (2003) state that “there is some evidence that North-South RTAs stimulate economic growth in the southern partner, little evidence that North-North RTAs stimulate growth and NO evidence that South-South RTAs do so.? Specifically they argue that SS RTAs do not provide partners with access to technology or knowledge that is characteristic of rich countries; SS RTAs are unlikely to add credibility to government policies and may even hinder investment if not accompanied by liberalization of trade with the rest of the world; and, SS RTAs are “likely to generate only trade diversion and no trade creation?

Mayda and Steinberg (2006) argue that SS RTAs are unlikely to provide the positive effects of competition and economies of scale because partner countries are both small and poor. In addition, the loss of fiscal revenues harms the member country economies’ and finally, SS RTAs are more likely to divert trade rather than create trade. Willmore (1976) and Nicholls (1998) make similar points using the Central American Common Market as an example.

Trade creation and trade diversion are concepts that were introduced by Jacob Viner in 1950. Both terms refer to the redirection of trade flows as a consequence of an RTA. In trade creation, goods that were previously produced by a local economy are instead imported from more efficient producers in countries within the RTA. Trade diversion refers to the redirecting of trade from the more efficient producer to a less efficient producer within the RTA. In both cases, trade creation and trade diversion, the trade flows are affected by the reduction of tariffs to member countries typical of RTAs. Trade creation and trade diversion are explained with more detail in section 2.1 of this paper.

A number of studies have been conducted to assess the effects of SS RTAs in partner countries –most of them attempt to determine if the RTAs were trade creating or trade diverting – e.g. Evans (1998), Lewis et al. (1999), Flores (1997), Cernat (2001,2003)), Subramanian and Tamirisa (2001), Mayda and Steinberg (2006). Different methods have been used and the results are mixed. As a reference, this paper focuses on the results of Cernat (2001, 2003), Flores (1997), and Mayda and Steinberg (2006). Different methods were used in these studies and the results were mixed.

Cernat (2001) used the log-linear form of the gravity equation to assess nine SS RTAs. He finds evidence that suggests that SS RTAs are “less trade diverting than theoretically predicted?. Cernat (2001) findings suggest that Mercosur and the Andean Community were overall, trade diverting. On the other hand Flores (1997), using a CGE analysis, concluded that Mercosur was trade creating.

Mayda and Steinberg (2006) use a difference-in-difference estimation strategy at commodity level to assess the impact of COMESA on Ugandan imports. They present evidence that “South-South trade agreements create positive but little economic gains, through changes in trade patterns, for their members.? This is different from Cernat (2001) results, which indicate that “imports into COMESA members from third countries were on average 30 per cent higher than those predicted without the trade diversion dummy variable?. Mayda and Steinberg (2006) find evidence that no trade diversion takes place in COMESA.

The mixed results from these studies, the increasing number of SS RTAs underway and the high number of countries wanting to join completely or in part in these RTAs poses the following questions: Why do policy makers from these countries advocate in favor of these RTAs? Should these RTA’s be pursued?, and the still not categorically answered question: Are South-South Regional Trade Agreements trade creating or trade diverting? Using the gravity model, this paper aims to get evidence from SS RTAs from the Americas.

1.2 Problem definition

Do South-South Regional Trade Agreements create trade or divert trade? The literature on this topic is vast and contradictory. Everybody thinks that SS-RTAs are trade diverting. Some papers present evidence of this. Other present evidence that they are actually trade creating. Finally others find evidence of very little trade creation and no significant evidence of trade diversion.

With so many RTAS in place and many others underway, it is important to understand the effects of creating these trade blocs.

Should poor countries pursue RTAs with poor countries? Are SS RTAs building blocks or stumbling stones towards the world liberalization of trade?

1.3 Research Objective

The main objective of this paper is to determine if MERCOSUR, Andean Community, and SICA were trade creating or trade diverting in the years 1995, 1998, 1999, 2003, 2007.

1.3.1 Major research question

Is there significant evidence of trade creation or trade diversion on the years 1995,1998,1999,2003,2007 for Mercosur, Andean Community and SICA?

1.3.2 Minor research question

Is there significant evidence that suggests that RTA members of the above mentioned RTAs increased trade between them and their partners?

Is there significant evidence that suggests that members of the above mentioned RTAs increased trade between them and third countries?

Is there significant evidence that suggests that the increase in trade between RTA partners of the above mentioned RTAs is higher than the decrease in trade between RTA members and third countries?

1.4 Theoretical Framework

1.4.1 The Gravity model of trade

The gravity model uses Newtonian gravity principles to study human behavior. It is widely used by economists and social scientists to predict flows of trade, people, goods, money, and other variables as an effect of changes in economic policies, fiscal policies, new laws, bans and other distortions to the flow of a given variable.

The original gravity model of trade assumes that two countries will trade more or less depending on the sizes of their economies and the distance between their economic centers. It was created independently by Tinbergen (1962) and Pöyhönen (1963) and augmented in later years to include other independent variables that may cause a change in trade flows. These augmented versions of the basic gravity model may include: population of the two countries, presence of common borders, same language, common colonizer, and others that the researcher regards as relevant.

The gravity model specifications used in this paper are similar to those of Cernat(2001) and Cheng & Hall (2003). These specifications are used to run OLS regressions on trade data of 1995, 1998, 1998, 2003 and 2007. One set of pooled data including the years mentioned is analyzed using the same gravity specifications.

The results of these regressions provide evidence of gross trade creation and diversion as specified by Balassa (1967)

1.4.2 Research Methodology and Design

The paper uses standard OLS analysis, with bilateral imports as a dependent variable and 17 independent variables: GDP of the importing country, GDP of the exporting country, Population of the importing country and population of the exporting country, distance between the capital cities of each country pair, Intra_x dummy variable for each RTA, Extra_x dummy variable for each RTA. The values of GDPs, distance and populations are used in their logarithmic form.

GDPs and population data was collected from the WB databank. Trade data was collected from UNCTAD’s database using the WB bank’s WITS application.

1.4.3 Research Assumptions

Costs of transportation are proportional to the great circle distance between economic centers of countries studied

All countries have one economic center, namely their capital cities.

The error coefficient of the log-linear gravity model used in this paper is normally distributed with a mean of zero and constant variance for all observations. It is also assumed that error pairs are uncorrelated.

GDPs, population, and trade data collected belongs to the population

1.4.4 Research Limitations

1.5 Thesis Structure

The remainder of this paper is organized as follows: Chapter 2 presents a literature review that explains trade creation and trade diversion, the effect of both and findings of previous papers that assess RTAs. Chapter 3 explains the gravity model used on the paper, how data was collected and organized, and the considerations in analyzing data. Chapter 4 summarizes the findings and Chapter 5 concludes.

2. Literature Review

There is extensive literature on RTAs. This literature either predicts the effects of a RTAs using a computable-general equilibrium analysis or they measure the effects of an FTA using aggregate data or commodity level data.

The concern of most authors, and the reason why they conduct their research, is that FTAs and specially SS FTAs may divert trade rather than create it. In the former case, purchases from an efficient producing country are replaced by purchases of a less efficient FTA partner.

This section serves three purposes: 1. It explains trade creation and trade diversion to the reader so she can better understand the methodology used to assess the selected RTAs. 2. It presents the reader with the results of previous findings so that the reader can compare the results of this paper with previous results of other authors. 3. It gross trade creation and diversion so that the reader can understand the results of the research.

2.1 Trade Creation and Trade Diversion

Trade creation and trade diversion as defined by Viner (1950), refer to changes in flow of trade between nations. Trade creation happens when trade is switched from less efficient producers of one country to more efficient producers in another country – a better allocation of resources. In trade diversion trade is shifted from more efficient producers in one country to less efficient producers in another country –a worsening in the allocation of resources.

2.1.1 Trade Creation

Trade creation can be defined as the net welfare gain that results from the initiation of an RTA, both on the production and on the consumption side. Some economists though, think that it is more precise to think of trade creation only as the increase in welfare from the production side (Senior-Nello S, 2010). In this paper the former definition of welfare is considered.

To understand trade creation, imagine the following scenario (Figure 1): The country in question, Country X, say Honduras, imports product Q from country M (United States) at price Pw+t, which includes an ad valorem tax and is the same price offered by other nations in the world, including country E (El Salvador). At this price, Honduras imports 20 units and consumes 60. The remaining 40 units are imported from the US. This is illustrated by the Honduran supply and demand lines in Figure 1 and the perfectly elastic supply curve with free trade of El Salvador. It is understood that a change in Honduran imports of product Q cannot affect the world price of product Q.

Figure 1. Trade Creation

If Honduras signed an RTA with El Salvador and the price of product Q from El

Salvador dropped to PE, Honduras would now produce 10 units of product Q, consume 70, and import the difference of 60 units. Because El Salvador now offers a lower price for product Q, Honduras now imports this product from El Salvador and not from the US.

The consumer surplus gains of this RTA are represented by areas a+b+c+d. The loss in producer surplus is indicated by area a. The loss of tariff revenue for Honduras is area c. Therefore the net welfare increase of this RTA between El Salvador and Honduras is indicated by triangles b and d.

Triangle b represents the amount of production that was shifted from less efficient producers in Honduras to more efficient producers in El Salvador – a better allocation of resources. Triangle d represents the increase in consumption of product Q.

2.1.2 Trade Diversion

Trade Diversion is illustrated in figure 2. Again the supply and demand lines are those of Honduras for product Q. Line S1 and S2 are the perfectly elastic supply curves of USA and El Salvador respectively, and lines S1+t and S2+t are the tax inclusive supply curves of the same two countries.

Figure 2. Trade Diversion

Honduras imports product Q from the US at tax inclusive price Pw+t. El Salvador offers product Q at price PE+t and thus does not benefit from Honduran purchases.

At price Pw+t Honduras produces 20 units, consumes 60, and imports 40 from the US. If Honduras and El Salvador now form an RTA and do not include the US, tariffs will be removed on imports from El Salvador but not from imports from the US. After forming the RTA Honduras would produce 10 million units, consume 80 million and import 60 million units of product Q from El Salvador at price PE.

The RTA has diverted trade from more efficient producers in the US to less efficient producers in El Salvador, so there is a worsening in the allocation of resources. On the other hand 10 million units are now imported from El Salvador instead of being produced at home in Honduras. At the same time 40 million units that were previously imported from the US are now being imported from El Salvador.

The welfare loss from trade diversion is reflected rectangle f. The 40 million units that were imported from more efficient producers in the US whose free trade price is $1.00 are now imported from El Salvador at $2.00. The welfare loss is $40 million.

The welfare gain from the customs union is calculated as the areas of triangles b and d. Triangle b is the welfare gain in the production side: $5 million. Triangle d is the welfare gain in the consumption side: $10 million.

The total impact on welfare as a result of the RTA is given by the sum of the areas of triangles b and d minus the area of rectangle f (b+d-f): welfare gain minus welfare loss. In this case the RTA generated a welfare loss of $25 million.

Figure 2 illustrates that the idea of trade creation and trade diversion can be misleading. If, for example, the sum of areas of triangles b and d would be greater than the area of rectangle f, the RTA would cause a net welfare gain. In this scenario, although trade has been diverted from more efficient producers in one country to less efficient producers in another, the RTA increased welfare for the RTA signing country.

2.1.3 Gross Trade Creation

Following the lead of Jacob Viner, Balassa (1967) evaluated the effects of the European Common Market with reference to its trade creating and trade diverting effect using Tinbergen (1962) and Pöyhönen (1963) model –the gravity model. In his work he developed model that captured substitution of less efficient domestic and foreign suppliers for more efficient foreign suppliers – gross trade creation; which is different than Viner’s definition of trade creation according to which trade is created only at the expense of local producers.

To illustrate the difference gross trade creation and trade creation proper as defined by Viner (1950), consider three trading partners of one particular product – countries A, B, and C, product Q (See Figure 3). Before signing a RTA with country B, Country A imports product Q from both, Country B and Country C in equal amounts and has 4 local producers of the same product (Figure 3a).

In the case of trade creation proper (Figure 3b), after signing a RTA with country B, Country A continues to import equal amounts of product Q from countries B and C but has reduced the number of local producers of the same product. More efficient producers in Country B have absorbed market share from local producers in Country A – trade creation proper.

Gross trade creation on the other hand (Figure 3c), considers that trade is created not only when local producers are substituted, but also when producers in third countries are substituted. In this case, after signing a RTA with country B, Country A decreases its imports of product Q from Country C and increases imports of the same product from Country B while keeping the same number of local producers. It is important to note that gross trade creation assumes that substituted producers in Country C were less efficient than producers in country B; the contrary would constitute trade diversion.

Figure 3. Trade Creation Proper vrs Gross Trade Creation

Like in Cernat (2001), this paper evaluates the gross trade creating effects of the assessed RTAs.

In his paper, Balassa (1967) provides evidence of trade creation in the European Common Market during six years since the Market’s establishment. Again, trade creation applies to the substitution of any less efficient producer for a more efficient one, independent of the producer’s base country. The why of the expected differences between the results of developed country RTAs and SS RTAs is explained in the next section.

2.2 Empirical Evidence from SS RTAs

A number of studies have been conducted to assess the effects of SS RTAs in partner countries –most of them attempt to determine if the RTAs were trade creating or trade diverting – e.g. Evans (1998), Lewis et al. (1999), Flores (1997), Cernat (2001), Subramanian and Tamirisa (2001), Cernat (2003), Mayda and Steinberg (2006). Different methods have been used and the results are mixed. This paper uses methods similar to Cernat (2001) and Cheng & Wall (2003).

In his paper, Cernat(2001) used the log-linear form of the gravity equation to asses nine SS RTAs. He finds evidence that suggests that SS RTAs are “less trade diverting than theoretically predicted?. Cernat’s(2001) findings suggest that Mercosur and the Andean Community were overall, trade diverting.

Mayda and Steinberg(2006) use a difference-in-difference estimation strategy at commodity level to assess the impact of COMESA on Ugandan imports. They present evidence that “South-South trade agreements create positive but little economic gains, through changes in trade patterns, for their members? (Mayda and Steinberg, 2003). This is different from Cernat’s(2001) results, which indicate that “imports into COMESA members from third countries were on average 30 per cent higher than those predicted without the trade diversion dummy variable?. Mayda and Steinberg (2006) find evidence that no trade diversion takes place in COMESA.

The mixed results from these studies, the increasing number of SS RTAs underway and the high number of countries wanting to join completely or in part in these RTAs poses the following questions: Why do policy makers from these countries advocate in favor of these RTAs? Should these RTA’s be pursued?, and the still not categorically answered question: Are South-South Regional Trade Agreements trade creating or trade diverting? Using the gravity model, this paper aims to get evidence from SS RTAs from the Americas.

Theoretical Framework and Research Methodology

***Intro***

Problem Definition

Research Objective

Research Questions

3.1 Theoretical Framework

3.1.1 Multiple Regression Analysis and Model Building

Figure 4. Regression Hyperplane

Multiple regression analysis is a method of inferential statistics that measures the relationship between two or more independent variables and one dependent variable. The multiple regression model is given by:

Where:

y = dependent variable

= regression constant of the population

= regression coefficient for each variable xj=1,2,…k

k = number of independent variables

= error of the model

Different from a simple regression equation –which forms a straight line in a two-dimensional space to represent the linear relationship between two variables – the multiple regression model forms a hyperplane in a multidimensional space (Figure 4). This hyperplane represents the relationship between the dependent variable and k independent variables.

To build a multiple regression model, that is, to construct a mathematical equation that represents the relationship between independent and dependent variables, a researcher must decide:

The question that needs to be answered

The potential independent variables

What is a representative sample of the population – should be at least four times the number of independent variables (Groebner, et al, 2008)

The model used in this paper is well known and widely used by social scientists to measure the flow of various types of variables. This model is explained in section 3.1.3.

3.1.2 Regression Model Diagnosis

To ensure the significance of an OLS regression analysis results, the following evaluation criteria are usually used (Groebner, et al, 2008):

The coefficient of determination (R2 and R2 adjusted)

Significance of the overall model (F-test)

Significance of individual variables (t-tests)

Size of the standard deviation of the model

Multicollinearity of variables

The coefficient of determination measures the proportion of variation in the dependent variable that can be explained with the independent variables used by the model. The value of R2 may range from 0-1, with 1 representing a perfect linear relationship between dependent and independent variables. Higher values of R2 are preferred as they would indicate that the chosen independent variables explain better the variations in the dependent variables.

A derivate indicator, called adjusted R2, takes into account the number of independent variables in the model, and their contribution the variations in the dependent variable. Because R2 increases when independent variables are added to the model, even if the new variables have no relationship with the dependent variable, adjusted R2 evaluates the model more precisely.

The Significance of the overall model can be determined by comparing the Significance F value given in the regression output of a statistical software application, and the critical value for a given alpha level.

The critical value for a given alpha level is determined using t-tables and statistical procedures explained in Groebner (2008).

The Significance of individual variables is determined by comparing their calculated t-values with the critical t-value of the model. If their calculated t-values are greater than their critical t-values the variable is considered significant. To determine the critical t-values of independent variables, degrees of freedom need to be calculated and interpolated with the desired level of significance in a t-table. For detailed explanations see Groebner (2008).

The size of the standard deviation of the model measures the dispersion of observed values of the dependent variable, and the predicted values for the same variable. It is up to the researcher to determine an acceptable range for the standard error estimation.

Multicollinearity occurs when two variables provide overlapping information to explain the variation in the dependent variable. To measure multicollinearity the researcher can use the VIF as an indicator. Generally, if the VIF < 5 for a particular independent variable, multicollinearity is not considered a problem (Groebner, 2008).

3.1.3 The Gravity Model of Trade

Following Isaac Newton’s principle of gravity, according to which two bodies will attract each other more when their sizes are increased and the distance between them is shortened; the gravity model explains trade flow between two countries based on the size of their economies and the distance between their economic centers.

The equation representation of the gravity model of trade is:

(Formula 1)

Where Fg represents trade flow, G is the constant, m1 and m2 are the economic dimensions of the two countries in question, and d is the distance between the two countries. In its basic log-linear form, the gravity equation is as follows:

(Formula2)

Where is the bilateral trade flow between countries i and j at time t, α is the constant, is the natural logarithm of the GDP of country i, is the natural logarithm of the GDP of country j, is the natural logarithm of the distance between country i and country j, and ε is the normally distributed error.

This basic gravity model is usually augmented by including other variables like adjacency, common language, colonial links, common currency, and RTA membership among others. Different authors have suggested many different specifications for the gravity model of trade [1] , however there is no consensus about which model specification is more accurate and serves best in assessing RTAs. Moreover other authors have suggested that the gravity model is biased due to endogeneity and reverse causality (Magee, 2003) and have led others to use entirely different methods to asses RTAs (Mayda & Steinberg (2006).

This paper uses a gravity model specification that is similar to Cernat (2001) but considers Cheng & Walls (2003) suggestions of eliminating dummy variables that might capture unintended trade distorting variables.

To assess trade creation and trade diversion in nine RTAs, Cernat(2001) adds two dummy variables to an already augmented specification of the model: Intra_RTA and Extra_RTA. The Intra_RTA dummy becomes a 1 when both, the importing and the exporting countries, are partners in the RTA being assessed by the two dummies. The Extra_RTA dummy becomes one when the importing country is part of the assessed RTA but the exporter is a third country.

The model uses bilateral trade flows as a dependent variable and 18 independent variables: GDP of importing country, GDP of the exporting country, GDP per capita of the importing country, GDP per capita of the exporting country, Population of the importing country, population of the exporting country, distance between the capital cities of both countries, an adjacency dummy variable, a common language dummy variable, nine Intra_RTA dummy variables (one for each RTA assessed), and nine Extra_RTA dummy variables (one for each RTA assessed). All non-dummy variables expressed in their logarithmic form.

In theory, the Intra_RTA dummies will capture the effect that the assessed RTA had on trade between partners of the RTA; and the Extra_RTA dummy captures the effect of the same RTA on trade of RTA members with third countries.

To diagnose a RTA as trade crating or trade diverting, Cernat (2001) designed an Intra-Extra coefficient table (Table# in this paper). According to this table, if a trade agreement increased trade between its partners at the expense of third countries –diverted trade, the Intra_RTA dummy should be positive and the Extra_RTA dummy negative. If the agreement created trade instead, the coefficients of both dummies would be positive.

Coefficient

Extra_RTA

Intra_RTA

Sign

+

-

+

Trade creation and trade expansion

Trade diversion

-

Trade expansion

Trade contraction

Table 1: Dummy Variable Interpretation

Cheng & Wall (2003) use a fixed-effect panel data analysis to measure the effect on trade of RTAs over time. Their proposed model allegedly controls the heterogeneity bias in the gravity model of trade. In it, Cheng & Wall (2003) drop all dummy variables and even drop the distance variable. They argue that these variables bias the gravity model and they motivate their argument in a number of ways. First, they reason that economic distances are too hard to measure with accuracy because big countries have many economic centers, that are thousands of miles apart and that serve as trade centers for different countries. Moreover, in many cases transportations costs are not proportional to distance and costs can differ significantly if goods are transported by land or by sea.

Second – and a point considered in this paper, they argue that the effects that dummy variables try to capture, like that of common language, colonial history, common cultures, etc. are too difficult to observe and to quantify. In addition, the heterogeneity bias that they attempt to reduce is more severe in countries that have little in common, so they don’t think that these variables can eliminate the bias.

Lastly, in the case of the adjacency dummy, Cheng & Wall (2003) consider that even though adjacency is an important factor, much trade occurs from people crossing the border and this trade is unaccounted for. More importantly, they reject that contiguity is equivalent in terms of its effect on trade considering that contiguous countries like Canada/US and Argentina/Chile have very different levels of trade that can be seemingly attributed to adjacency.

3.1.3.1 Expected Results

Each variable in the gravity model has predictable effects (Oguledo & Macphee, 1994): The GDP variables are expected to affect trade flow positively and thus result with positive coefficients in the regression analysis. All else remaining constant, an increase in GDP in the exporting country indicates greater production, that is, more products available for export. An increase in the importing country’s GDP would lead, given a high marginal propensity to import, to an increase in imports.

The population variable affects trade flows indeterminately. “On one hand a large population may indicate large resource endowment, self sufficiency, and less reliance on international trade. On the other hand a large domestic population may promote division of labor and thus opportunities of trade in a larger variety of goods (Oguledo & Macphee, 1994). Coefficient signs of population variables may be positive or negative.

The distance variable is expected to have a negative effect on trade flows and thus a negative coefficient sign in the regression results.

3.1.4 Research Assumptions

Costs of transportation are proportional to the great circle distance between economic centers of countries studied

All countries have one economic center, namely their capital cities.

The error coefficient of the log-linear gravity model used in this paper is normally distributed with a mean of zero and constant variance for all observations. It is also assumed that error pairs are uncorrelated.

3.1.5 Research Limitations

3.2 Research Methodology

3.2.1 Research Type and Approach

The research done for this thesis is qualitative and inductive. The author uses standard OLS multiple regression analysis to study the effect of 7 independent variables on a single dependent variable. The regression model used is a variation of the gravity model of trade similar to Cernat (2001) and it is designed to produce results for six different RTAs (listed on Table 2). To avoid falling in the dummy variable trap, the author included an additional RTA –USA and Israel – for which no findings are reported. An Intra_RTA dummy variable was included in the model for this additional RTA to prevent it from distorting the regression results.

The model used in this paper is:

(3)

Where,

ln= natural logarithm

Y= imports

α = constant

β = coefficient

= GDP of country i in year t

= GDP of country j in year t

= Population of country i in year t

= Population of country j in year t

= Distance between capital cities of country i and j

= normally distributed error

Intra_RTA = dummy variable, takes value of 1 if both countries are members of the same RTA, zero otherwise -

Extra_RTA= dummy variable, takes value of 1 if the importing country is member of the assessed RTA and the exporting country is a non-member, zero otherwise

To obtain statistically significant results, RTAs of developed countries were included in the model; however a greater emphasis is placed on analyzing the results of SS RTAs: SICA, MERCOSUR and The Andean Community. A Total of five years of trading data were analyzed: 1995, 1998, 1999, 2003, and 2004. For each year more than 1,000 observations were analyzed and 6264 observations for the PCS analysis.

The model specifications used are similar to Cernat (2001) in that it uses Intra/Extra dummy variables to measure the impact of a particular RTA in the trade flow of a country, allows only one intercept, has different coefficients for each variable, and includes the distance variable. It differs in that it does not include adjacency and language dummies to control for heterogeneity bias as suggested by Cheng & Wall (2003). It also does not use GDP per capita to avoid collinearity [2] with GDP and population variables.

Table 2: RTAs Assessed

Andean

Bolivia, Colombia, Ecuador, Peru, Venezuela

European Union

Austria, Belgium, Denmark, Finland, France, Germany, Greece, Ireland, Italy, Luxembourg, Netherlands, Portugal, Spain, Sweden, United Kingdom

MERCOSUR

Argentina, Brazil, Paraguay, Uruguay

NAFTA

Canada, Mexico, United States

SICA

Belize, Costa Rica, El Salvador, Guatemala, Honduras, Nicaragua, Panama

CC

China, Chile

Considering Cheng & Wall (2003) reasoning of the biasing effect of adding independent variables to the gravity model; this paper has dropped all dummy variables except for the Intra_RTA and Extra_RTA, which capture the trade creating and trade diverting effect that the research pretends to measure. The ‘distance’ variable was kept however, for two reasons: first, this variable is widely used and accepted as a significant one in the economics literature [3] . Second, dropping original variables of the gravity model to investigate the effect of new variables makes it more likely that the new variables will prove (falsely) significant (Anderson, Ferrantino &Schaefer, 2004)

3.2.2 Data Collection

GDPs and population totals of all countries used in the model were downloaded from the WB databank. Trade totals between all countries included in the model were downloaded using from UNCTAD using the WB WITS application.

Physical distances between capital cities (measured as great circle distance) are provided at http://privatewww.essex.ac.uk/~ksg/data-5.html. To guarantee the fidelity of this source, distances of 50 random city pairs were selected and compared with data available at http://www.indo.com/distance and Google Earth software. The distances provided by the three sources were comparable and had only minor differences.

The data used is available at: https://docs.google.com/leaf?id=0B1bFYu9Cjb-SMjhhZDM0ZTctNDE1YS00YWQ4LWE3NWMtM2QzYmYwMTMzNGZl&hl=en&authkey=CJ-88uoI

4. Findings and Results

The results of the OLS multiple regression analysis for all sets of data are displayed in Table 4 and Table 5 of the Appendix. Table 4 shows the results of the individual year regressions and Table 5 shows the results of the PCS analysis of all years considered. This section provides a brief description of the results.

At 0.00, the calculated F significance for all regressions was less than the alpha critical value of 0.05 with a 95% confidence level – the model is overall significant. Adjusted R2 ranged from 0.812 to 0.834 – the model explains a high percentage of the variation in the log of imports. Standard errors ranged from 1.314 to 1.487 [4] .

In the individual year regression results, distance coefficients are, as expected, consistently negative over time and highly significant; showing the negative impact of distance on trade. The coefficients of GDP for importing and exporting countries have consistent positive signs and show a downward trend over time. The coefficients of population are inconsistent and show negative signs in 1995 and 1998 and positive signs in 2003 and 2007. In 1999 the population coefficient of importing countries is positive and the one for exporting countries is negative.

In the Pooled Cross-Section data regression results, coefficients for GDPs and populations are positive and significant, except for the coefficient of populations of the exporting countries, which is not significant. Distance was negatively signed and significant.

4.1 The effect of RTAs

Andean Community Coefficients for Intra_Andean are positive and significant across all years, except for 2007, when the coefficient is not significant. The Extra_Andean coefficient is negative but not significant in all years except 1998, when it is positive and not significant. In the pooled cross-section analysis the Intra dummy was positive and significant, and the Extra dummy was negative but not significant. These results suggest that Andean Community was mainly trade.

European Community Intra_EU coefficients are negative and significant in 1995 and 1998 but not significant in the other three years. Extra_EU coefficients were not significant in all five years and had changing signs. In the pooled cross-section the Intra dummy was negative and significant while the Extra dummy was positive but not significant; suggesting that the EU has been overall trade expanding.

Chile-China In 2007, the only year in which this RTA is evaluated, both Intra and Extra dummies are not significant. In the PCS analysis however, the Intra dummy is positive and significant and the Extra is positive but still not significant. This evidence suggests that over time, this RTA is trade creating and trade expanding.

Mercosur Evidence suggests that this RTA has been trade diverting. Intra coefficients were consistently positive and Extra coefficients consistently negative across all years and in the PCS analysis.

NAFTA The coefficients for both dummies of this RTA are not significant in all individual years and in the PCS analysis with changing signs; suggesting that trade between member countries of this RTA and third countries included in this research has not been significantly affected.

SICA For 1995 and 1998 the Intra dummy are positive and significant while the Extra dummy was negative but not significant. For all other years both coefficients are positive with the Extra dummy remaining not significant. The PCS analysis shows a positive sign for both coefficients and significant results only for the Intra dummy. In this case evidence suggests that the RTA was trade creating.

5. Conclusions

6. Appendix

Table 4: Regression results of individual years

 

1995

1998

1999

2003

2007

 

Coef

t-value

Coef

t-value

Coef

t-value

Coef

t-value

Coef

t-value

Constant

-30.728

-25.75

-31.188

-24.88

-32.982

-26.89

-32.247

-25.82

-31.562

-22.43

Distance

-1.22329

-17.04

-1.23281

-16.74

-1.2172

-17.37

-1.09701

-15.96

-1.20662

-15.89

GDP I

1.00612

16.02

1.01011

14.99

0.98304

14.62

0.84344

11.48

0.73006

8.79

GDP E

1.13967

31

1.22413

31.2

1.16704

31.44

0.98887

28.21

0.96548

24.03

Pop I

-0.04471

-0.73

-0.10361

-1.59

0.0338

0.54

0.22958

3.35

0.34328

4.21

Pop E

-0.08754

-1.92

-0.14715

-3.03

-0.06492

-1.48

0.12563

2.99

0.22845

4.72

Intra_MS

0.7473

1.75

1.0875

2.48

1.2544

2.84

1.8183

4.14

1.098

2.07

Extra_MS

-0.4913

-2.68

-0.4225

-2.26

-0.2488

-1.34

-0.8515

-4.51

-0.8198

-2.68

Intra_Nafta

0.0618

0.11

0.2367

0.4

0.1977

0.33

0.01

0.02

-0.1709

-0.25

Extra_Nafta

-0.3528

-1.69

0.033

0.15

0.034

0.16

-0.0101

-0.05

0.1

0.33

Intra_Andean

1.2305

3.55

1.6644

4.65

1.7798

4.92

1.5363

4.3

0.6243

1.33

Extra_Andean

-0.1899

-1.04

0.0506

0.27

-0.1216

-0.64

-0.2313

-1.22

-0.6169

-1.95

Intra_Sica

1.7336

5.2

2.0986

6.14

2.1202

6.27

2.0059

6.07

1.8532

4.31

Extra_Sica

-0.3447

-1.71

-0.2416

-1.17

0.2008

0.99

0.1906

0.95

0.0805

0.25

Intra_EU

-0.911

-3.74

-0.6941

-2.79

-0.4583

-1.89

-0.1507

-0.61

-0.1917

-0.61

Extra_EU

-0.016

-0.08

0.107

0.55

0.236

1.22

-0.0609

-0.3

0.0828

0.3

Intra_CC

-

-

-

-

-

-

-

-

2.123

1.94

Extra_CC

-

-

-

-

-

-

-

-

-0.1856

-0.55

Observations

1168

1173

1298

1306

1309

Parameters*

16

16

16

16

18

Adjusted R2

0.834

0.827

0.827

0.832

0.812

F Significance

0.000

0.000

0.000

0.000

0.000

Std. Error

1.315

1.355

1.376

1.356

1.487

All non-dummy variables are in log form

Table 5: Regression results of pooled data

 

Pooled 5 years

 

Coef

t-value

Constant

-32.5621

-56.9

Distance

-1.19108

-36.71

GDP importing

0.93588

30.36

GDP exporting

1.08886

64.27

Population im

0.0798

2.71

Population ex

0.02259

1.1

Intra_MS

1.3317

6.6

Extra_MS

-0.47453

-5.39

Intra_Nafta

0.0892

0.33

Extra_Nafta

-0.0158

-0.16

Intra_Andean

1.4615

8.77

Extra_Andean

-0.13324

-1.48

Intra_CC

2.8009

2.84

Extra_CC

0.1894

1.03

Intra_Sica

2.0804

13.38

Extra_Sica

0.09169

0.96

Intra_EU

-0.4565

-4.01

Extra_EU

0.08788

0.96

Y95

0.52427

8.57

Y98

0.44138

7.35

Y99

0.42669

7.26

Y03

0.35962

6.25

Observations

6254

Parameters*

22

Adjusted R2

0.825

F Significance

0.000

Std. Error

1.487

All non-dummy variables are in log form