The effect of regional trade agreement to trade flow evidence of trade creation and trade diversion of asean japan free trade area

UNIVERSITY OF ECONOMICS ERASMUS UNVERSITY ROTTERDAM HO CHI MINH CITY INSTITUTE OF SOCIAL STUDIES VIETNAM THE NETHERLANDS VIETNAM –THE NETHERLANDS PROGRAMME FOR M.A IN DEVELOPMENT ECONOM

UNIVERSITY OF ECONOMICS ERASMUS UNVERSITY ROTTERDAM

HO CHI MINH CITY INSTITUTE OF SOCIAL STUDIES VIETNAM THE NETHERLANDS

VIETNAM –THE NETHERLANDS PROGRAMME FOR M.A IN DEVELOPMENT ECONOMICS

THE EFFECT OF REGIONAL TRADE AGREEMENT TO TRADE FLOW: EVIDENCE OF TRADE CREATION AND TRADE DIVERSION OF ASEAN –

JAPAN FREE TRADE AGREEMENT

BY

PHAM THI HIEN

MASTER OF ARTS IN DEVELOPMENT ECONOMICS

HO CHI MINH CITY, December 2016

UNIVERSITY OF ECONOMICS INSTITUTE OF SOCIAL STUDIES

HO CHI MINH CITY THE HAGUE VIETNAM THE NETHERLANDS

VIETNAM - NETHERLANDS PROGRAMME FOR M.A IN DEVELOPMENT ECONOMICS

THE EFFECT OF REGIONAL TRADE AGREEMENT TO TRADE FLOW: EVIDENCE OF TRADE CREATION AND TRADE DIVERSION OF ASEAN –

JAPAN FREE TRADE AGREEMENT

A thesis submitted in partial fulfillment of the requirements for the degree of

MASTER OF ARTS IN DEVELOPMENT ECONOMICS

By

PHAM THI HIEN

Academic Supervisor:

Prof Dr Nguyen Trong Hoai

HO CHI MINH CITY, December 2016

This is to certify that that this thesis entitled “The effect of regional trade agreement to trade flow: Evidence of trade creation and trade diversion of ASEAN – Japan free trade agreement”, which is submitted in fulfillment of the requirements for the degree of Master of Art in Development Economics to Viet Nam – The Netherlands Program (VNP) The author hereby declares that she edit this thesis individually, using only stated resources and literatures To the best of my knowledge, my thesis does not violate anyone’s copyright as well as any proprietary rights which are fully acknowledge in accordance with the standard referencing practices

HCMC, December 15th, 2016

Pham Thi Hien

he has open the door to welcome me and sent to me the prompt advice

I also would like to thank my co-supervisor Dr Truong Dang Thuy for his enthusiastic support and precious suggestion, which help me overcome the challenges and difficulties in doing regression model, take me in the right direction

I would like to express my gratitude to all lecturers of the Vietnam- Netherlands Program who have provided the interesting lessons to build my economic knowledge during this program In addition, I would like to express my appreciation to the VNP academic staffs for their feedback, cooperation during a long-period time I have learned here

Besides, completing this work would be very difficult without the support from my best friends I

am indebted to them for their help Moreover, I wish to thank all my fellow master students in VNP 21 class who share with me unforgettable memories in this program

Last but not the least, there are also words of deep gratitude for my family who support spiritually and encourage continuously during my thesis writing and my life in general

Table of Contents

CHAPTER I: INTRODUCTION 1

1.1 Problem statement 1

1.2 Research objectives 2

1.3 Research questions 2

1.4 Research scope 3

1.5 Thesis structure 3

CHAPTER 2 LITERATURE REVIEW 5

2.1 Trade theories 5

2.2 Trade creation and trade diversion 5

2.2.1 Trade creation 6

2.2.2 Trade diversion 6

2.3 The gravity model in international trade 8

2.3.1 The origin of gravity model 8

2.4 Theoretical framework 9

2.4.1 Theoretical support and theoretical equation 9

2.5 Empirical support for effect of FTA to ASEAN 12

2.5.1 Empirical support for effect of AFTA to intra-bloc trade flow 12

2.5.2 Empirical support of effect of ASEAN + 1 FTAs 13

2.6 Zero trade data problem 15

2.7 Chapter remark 17

CHAPTER 3: RESEARCH METHODOLOGY 18

3.1 Model specification and validity testing 18

3.1.1 Model specification 18

3.1.2 Model validity testing 22

3.2 Data and data sources 23

CHAPTER 4: RESEARCH FINDINGS AND DISCUSSION 25

4.1 Descriptive statistics of variables 25

4.2 Testing multicollinearity 28

4.3 Regression result 30

4.3.1 Comparison of estimator properties 30

4.3.2 Regression results 31

Chapter 5: Conclusion and policy recommendation 44

5.1 Conclusion 44

5.2 Policy implication 45

5.3 Limitations of the study 46

Reference 47

CHAPTER I: INTRODUCTION 1.1 Problem statement

It is no doubt to saying that recently, regional trade agreements (RTA) have become a popular widespread trend in the international economic system, especially after Doha round of GATT/ WTO According to the definition of WTO, regional trade agreement, included free trade agreements (FTAs) and customs unions (CUs), are the negotiations of two or more parties, in which these participants agree to reduce their current custom barriers, such as tariffs, quotas Since early of the 1990s, RTAs have increased widespread According to reports of World Trade Organization (WTO), until February 2016, there are 625 notifications of RTAs and 419 in which were in force Regarding the Association of Southeast Asian Nations (ASEAN) is considered as a successful model of regionalism and the community is step by step greatly co-operating and integrating to the world economy In addition, Japan, an economy was growing rapidly, involving 17% to world economic in 2005 but reduced to only 6% in 2015 (IMF, 2015) However, her economic performance has a massive influence on the economy of the entire region For evidence, Japan is one of top three trading partners of ASEAN economies, especially Indonesia and the Philippines

Before integrating into ASEAN regional economies, Japan was playing an important role in the regional development In the 1970s, 25% per total import and export values of ASEAN were doing with Japan Moreover, with lower cost in materials and labors, ASEAN markets were attractive destinations of capital investment flow from Japanese companies It generated work jobs and increased working wages, especially, with high technologies and high-trained employees, they provided a valuable opportunity for learning and transferring in this area during the 1980s to 1990s The increasingly integrated business need a major opportunity to strengthen linkages between ASEAN and Japan That is the reason for raising a needful talk about a regional agreement Since 2003, the government of Japan and the 10 countries of ASEAN completely signed the general framework of bilateral free trade agreement named ASEAN-Japan FTA (officially a comprehensive economic partnership), hereinafter referred as AJCEP At the end of December

2008, the last official round was finalized, an agreement signed among Asian countries, included: Brunei Darussalam, Cambodia, Indonesia, Laos PRD, Malaysia, Myanmar, Philippines,

Singapore, Thailand, Vietnam and Japan has been forced, support multilateral trading by reducing the tariff The origin objectives of this FTA are to encourage free trade across the border in intra-bloc ASEAN and Japan, strengthen Asian countries, Japan economic integration, enhance their economic in the world market, are transparent in trading procedure and maintain sustainability in the economic area It seems a major opportunity for high-tech and modern industries of Japan such

as automobile, electronic, etc to enter ASEAN markets as well as encourage assembly line in regions for Japanese firms

Statically, after trade agreement in force, in 2013, two-way trading volume obtained $229 billion compared with $128 billion in 2000 In this year, Japan reported 14% and 15% for import and export value to ASEAN, Thai Land ($22.5 billion), Indonesia ($ 32.2 billion) and Malaysia ($29.6 billion) are top three Asian biggest exporters to Japan (ASEAN Statistics, 2014) The notable products mainly exported from ASEAN to Japan are foods, manufactured goods, textiles, crude material Conversely, machinery and equipment transportation to gather with chemical and advanced technology manufacturing products are important to major export from Japan to ASEAN countries For example, according to Japan automobile Manufacturer Association statistics in

2014, about 47% Japanese cars, 80% truck vehicles and 85% buses were consumed as final products in ASEAN markets

1.3 Research questions

According to numerous studies before, the effect of RTAs has no guarantee positive effect to help its member countries integrating with the global market In many cases, RTAs actually caused some negative effects Therefore, this study aims to find the answers to these questions following:

- How the trade creation and trade diversion in general total export have been caused by the free trade agreement which was signed by AJCEP to ASEAN member countries?

- How the trade creation and trade diversion have been affected by the free trade agreement which is by AJCEP to ASEAN member countries in the five sub-catalogues: food products, agricultural products, manufactured products, Machinery and equipment of transportation and clothing and accessories and textile, fabric?

1.4 Research scope

To estimate the effect of AJCEP, we employ a panel data set will be collected with period from

2000 – 2015 with total 5,920 observations with included 09 ASEAN countries: Brunei Darussalam, Cambodia, Indonesia, Lao PDR, Malaysia, Myanmar, Philippines, Singapore, Thailand, Vietnam and 15 biggest trading partners of Japan 2015 include: The United State, China, South Korea (Korea Rep.), Hong Kong SAR China, Australia, Saudi Arabia, The United Arab Emirates, Russian Federation, Switzerland, New Zealand, United Kingdom, Germany, Mexico, Netherland and Japan

To our knowledge, there is the rapid development in investing effect of RTAs in theoretical as well

as empirical accesses However, most of them are usually focus on general questions: whether or not RTAs have affected to trade flow or created trade creation, trade diversion There are two main problems that many previous studies had

The first problem is estimation challenges of the gravity model which solve around the heteroscedasticity and the frequency of zero trade observations These problems cause challenges

in concerning the most suitable estimation technique to avoid biased and un-misinterpreted result The second advantage is we do estimate regression model by using two sets of trade flow data The first data set is aggregated data is used to examine for bilateral total export flow The second dataset is disaggregated data is optimized to estimate the AJCEP affect to five separate sub-categories: agriculture, manufacturing, chemical industry, machinery, transportation industry and clothing and accessories and textile, fabric By two different approaches, we can analyze impacts

of AJCEP in general and in the specific commodity in particular as well

1.5 Thesis structure

After finishing introduction chapter, the rest of this paper is arranged as follow Chapter 2 presents the literature review in trade theories in international trade flow, theoretical support of gravity model in international trade, empirical support in order so to see the development of contribution

studies of AJCEP effect to ASEAN member as well as Japan In addition, this chapter reviews empirical support of the methods to solve the popular issue of frequency of zero trade data Chapter

3 states methodology, model construction, model estimation methods and data scope that used in the study Chapter 4 interprets the result and findings from the regression model Chapter 5 summaries the thesis result and recommendation suggestion as well as limitation of the study

CHAPTER 2 LITERATURE REVIEW

In this chapter, we will summary some related trade theories which are popularly used in international trade Then, a review of theoretical and empirical support for gravity model on trade are added In addition, we consider about some literature reviews about zero trade data and the developing of estimation techniques which some previous studies used

2.2 Trade creation and trade diversion

Before Viner (1950), most of the studies assumed that tariffs between countries caused reducing welfare, therefore a customs union or free trade agreement would improve welfare He drew the distinction between trade creation and trade diversion effects of an FTA According to his study,

an FTA does not completely improve welfare The positive or negative effects of an FTA depend

on the comparison of the magnitude of trade creation and trade diversion If an FTA causes trade creation more than trade diversion, it implies that this FTA has raised welfare

2.2.1 Trade creation

When an FTA is in force, in general, we expect that with an elimination of tariff as well as trade incentive policies, FTA will encourage trade flow that would not have existed before It allows member countries to concentrate and trade with their comparative advantages and get the benefits from economic of scale All of the trade creation cases will increase country’s national welfare

𝑆𝑀 and 𝑆𝑁 are represented for exporting supply of product 𝑋 from intra-bloc member countries and extra-blocs non-member countries

According to Figure 1, before integrating a free trade agreement, 𝑆𝑀+ 𝑡 and 𝑆𝑁+ 𝑡 are denoted for supply curves from intra-bloc and extra-bloc respectively Assuming that non-member countries provide product 𝑋 at a lower price than member countries do, 𝑆𝑀 lies under 𝑆𝑁 or 𝑆𝑀 +

𝑡 lies under 𝑆𝑁+ 𝑡 graphically The difference between 𝐷0− 𝑆0 is country import demand from non-members

After free trade agreement was formed, the supply curve 𝑆𝑁+ 𝑡 is unchanged because tariff is still applied to non-member countries Meanwhile, the tariff is no longer counted to supply source from 𝑆𝑀 In this case, the equilibrium price of product 𝑋 in the country will be 𝑃1 and the difference between 𝐷1− 𝑆1 will be import demand from member countries instead of non-members countries

as before Considering the domestic consumer surplus is the sum of area 𝑎, 𝑏, 𝑐, 𝑑 while 𝑎 is the surplus of domestics manufacturers falls Regarding government, when a free trade agreement has been in forced, government is no longer collected tax revenue because currently all importing values comes from member countries, denoted by sum of area 𝑐 Therefore, the total effect of trade creation caused by free trade agreement is the sum of areas 𝑏 and 𝑑

Regarding trade diversion, the switching to the higher-cost manufacturers in intra-bloc members instead of lower-cost from extra-bloc members is denoted for trade diversion denoted by 𝑒 area The total effects of free trade agreement in overall will be determined by comparing the magnitudes of trade creation and trade diversion effects If trade creation exceeds trade diversion effect, welfare is enhanced due to free trade agreement Oppositely, trade diversion effect exceeds trade creation, it means that country welfare is decreased due to the free trade agreement

Figure 1: Trade creation and trade diversion

2.3 The gravity model in international trade

2.3.1 The origin of gravity model

Gravity model has been used as a workhorse tools to analyze the international trade flow It was developed from the law of universal gravitation found by Isaac Newton in 1967 The law state that every two points attract another one with a force that is in direct proportion to the product of their masses and inverse proportion to the distance between them in square

𝐹𝑖𝑗 = 𝐺𝑀𝑖𝑀𝑗

𝑑𝑖𝑗2 (1) Where:

𝐹𝑖𝑗 is the gravity force between two of masses

𝑀𝑖 , 𝑀𝑗 are the masses of the first and second point respectively

𝑑𝑖𝑗2 is the distance from fist point center to the second point center in square

𝐺 is the gravitational constant with determined value equal 6.674 x 10-11 N.(m/kg)2

The first study using gravity model which is derived from Newton’s law of gravitation to analyze international trade flows by Tinbergen in 1962 For trade model, the bilateral trade volume between two countries 𝑖, 𝑗 has been used to replace for the force of gravity and economic sizes

𝑌𝑖, 𝑌𝑗have been used to replace for the masses of 𝑀𝑖, 𝑀𝑗 respectively Generally, the gravity formulation has been established in the following form:

𝑋𝑖𝑗 = 𝐴𝑌𝑖

𝛼 𝑌𝑗𝛽

𝑑𝑖𝑗𝛾 (2) Where 𝛼, 𝛽, 𝛾 may take the value different to 1 They depend on the elasticity of economic sizes

of exporting country, importing country and distance respectively In case, 𝛼 = 𝛽 = 1 and 𝛾 = 2,

it has the same formulation of Newton’s equation Usually, economic sizes are defined as GDP, GNP, real GDP, real GNP, income per capita or population These essential variables represent for supply and demand force of each country that determine country’s trade volume

Regarding distance variable, it is defined by geography distance between two economics hubs or capitals counted in land miles Tinbergen stated that distance is not only a proxy represent for real distance but also may stand for many other market factors which influence to trade volume such

as transportation cost, transit cost, communication exchange cost or even culture cost

Usually, in economic regression, the simplest gravity model is estimated under OLS by taking logarithm equation (2) and adding error term 𝜀𝑖𝑗 The coefficient result obtained will be interpreted

as elasticity because the regression took the double log form:

𝑙𝑜𝑔 𝑋𝑖𝑗 = 𝑙𝑜𝑔 𝐴 + 𝛼𝑙𝑜𝑔 𝑌𝑖+ 𝛽 𝑙𝑜𝑔 𝑌𝑗− 𝛾𝑙𝑜𝑔 𝐷𝑖𝑗 + 𝜀𝑖𝑗 (3)

According to the explanation above, the coefficients will be interpreted as follow:

If the economy of country 𝑖/𝑗 increases by one percent, the trade volume between two countries will increase 𝛼/𝛽 percent respectively while other factors are held constantly Similarly, trade volume will reduce 𝛾 percent if the distance between two countries increases by one percent All

in the cases, error terms 𝜀𝑖𝑗 is supposed that it is independent and normal distribution

2.4 Theoretical framework

2.4.1 Theoretical support and theoretical equation

The first noble work in applying gravity model to international trade by Tinbergen, 1962 However, it was still missing powerful theoretical application basic and stood outside of mainstream due to the persistent perception of a physical gravity model more than an economic model The first important contribution we have to mention is the work of Anderson (1979) He built gravity model based on Cobb-Douglas or CSE preference function under assumptions following: each country specialize in trade completely, i.e., goods are differentiated by the origin

of a country (named Armington assumption), the preferences of consumers are homothetic and alike across regions, no transport cost, tariff and barrier in trade Consistent with idea that gravity model depends on the share of expenditure of national income spent for international trade, therefore, it can be estimated from a function of population and income Overcome the assumption Armington of Anderson (1979), Bergstrand (1985 and 1989) built a gravity model with monopolistic competition created by Paul Krugman (1980) It implies that countries have specialization in production and customer have a variety of preference, therefore, they will trade a

different kind of commodities from the identical country Deardorff (1998) and Krugman (1985) contributed the new theory to gravity model by applied literature comparative advantage of Heckscher-Ohlin theory Eaton and Kortum (2002) derived gravity model by using Ricardian model, Helpman et al (2008) added firm heterogeneity to obtain the model

Recently, many researchers do the theoretical contribution in gravity model by importantly concerning the usage of variables and specification In this section, say thanks to the contribution

of Anderson and van Wincoop (2003) who developed monopolistic competition framework based

on the Armington assumption and constant elasticity of substitution (CES) Assume that customer utility among countries are identical and homothetic, trade gravity equation was specified as below:

𝑉𝑖𝑗 = 𝑌𝑖𝑌𝑗

𝑌 𝑤 ( 𝑡𝑖𝑗

𝑃𝑖𝑃𝑗)1−𝜎 (4) Where 𝑉𝑖𝑗 is bilateral trade volume, 𝑌𝑖, 𝑌𝑖, 𝑌𝑤 is income of country 𝑖, 𝑗 and global income respectively, 𝑡𝑖𝑗, 𝑡𝑗𝑖 are bilateral trade barrier between countries 𝑖, 𝑗, denotes all bilateral trade resistance and assumed equally They include distance and some binary variables such as common border, colony, trade agreement etc 𝜎 is the elasticity of substitution,𝑃𝑖𝑃𝑗, is multilateral trade resistance and 𝑃𝑖, 𝑃𝑗 are consumer price index of country 𝑖, 𝑗 respectively and have a function as below:

Equation (1) and (2) show clearly that any changing in bilateral trade resistance 𝑡𝑖𝑗 in the numerator

will impact to multilateral trade resistance in the denominator and the ratio 𝑡𝑖𝑗

𝑃𝑖𝑃𝑗 will impact to bilateral trade function

To estimate border effect on international trade, they used the Non-linear Least Square (NLS) technique Even though they show a consistently and efficiently result that the bilateral trade cost

is depended on distance, locating landlocked, sharing the border and common language, this study has not overcome several mistakes on its three assumptions The first assumption is the two-way systematization trade cost between two countries It violates if the existence of bilateral or multilateral trade agreement The second violation is differences in the variety of customer preference when they assumed that there is trade volume balance between two countries The last problem is on assumption about the only period of data while they missed time-varying estimator

on the regression model

Following the methodologies of Anderson and van Wincoop (2003), Baier and Bergstrand (2007) developed the model by extension to panel data and used time-varying fixed effect to eliminate bias estimation caused by time-varying trade cost variables Baldwin and Taglioni (2006) regressed the model with the same method when choosing county-pair fixed effect to reduce endogeneity bias caused by FTA dummy variable

From this study, many authors use it in the different type of economic issues as the workhorse due

to its ability to correctly estimate bilateral relationship, for example, immigrant, foreign direct investment as well as trade flow This model has first theoretical clarification presented by Anderson (1979) and theoretical basic later proved by Helpman and Krugman (1985), Bergstrand (1989), and Deardorff (1998) In additional, this model was applied to many studies, can be referred to the collective paper by Sen and Smith's (1995)

In particular, there are various empirical studies when applied this model to international trade flow Soloaga and Winters (2001), Antonucci and Manzocchi (2006) used this model to examine the influence of country’s characteristics such as border, geography, distance combined with trade agreement to trade flow

2.5 Empirical support for effect of FTA to ASEAN

2.5.1 Empirical support for effect of AFTA to intra-bloc trade flow

The first FTA signed between ASEAN countries is AFTA in 1992 The origin members include six of ten ASEAN countries: Brunei Darussalam, Malaysia, Philippines, Singapore, and Thailand The rest of four countries have joined in Vietnam (1995), Lao PRD (1997), Myanmar (1997) and Cambodia (1999) Under this agreement, the tariff rate was reduced up to 99 percent for six origin countries and 95 percent for rest of four countries by 2010 At the present, elimination of tariff under AFTA has been completed

At the first stage of implementation of AFTA, there are many studies predicted that effect of AFTA

to trade creation would be small According to DeRosa (1995) estimated the effect of AFTA to intra-bloc by using CGE, he found that effect of Most Favored Nation (MFN) has created free trade liberalization more than effect caused by AFTA Alternatively, Frankel and Wei (1995) used gravity model as an ex-ante analyst to test this effect He concluded that trade flow between ASEAN countries was still affected by other outside important factors than ASEAN relation Increasingly, Endoh (1999) was as the first author introduced and applied two new definitions trade creation and trade diversion to analysis effect of an FTA According to his result, during sample period 1960-1994, ASEAN countries has not absorbed the effect of AFTA in encouraging trading flow within members He assumed that the result implied that the trade proportion between ASEAN countries was still small

Since the 2000s, a development of the methodology to estimate gravity model has been raised For evidence, Soloaga and Winter (2001) has applied Tobit model to evaluate the effect of some major PTAs on bilateral trade to ASEAN countries According to his result, the coefficient of trade intra-bloc was insignificant and negative However, outside ASEAN trade were significantly encouraged Following next movement on methodology, Carrère (2006) applied Hausman and Taylor method by using instrument variable and panel data, she showed a positive trade flow within ASEAN and reducing import from rest of the world

With the development of methodology as well as the interest on growing of FTA of ASEAN, a number of studies focused on their impact on ASEAN countries economies have been grown moderately Elliot and Ikemoto (2004) applied gravity model to examine the effect of AFTA on

trade creation and trade diversion Matching with some previous studies, they found that ASEAN countries have been gained benefit from AFTA both create on and diversion when coefficients effect were positive and significant With the same result, Bunn et al (2009) employed two kinds

of FTA dummies, first is FTA dummies which take value one if two member countries of AFTA from 1992 and 0 otherwise The second dummy was AFTA dummies multiplied to time trend to capture the possessive tariff elimination under AFTA He found that AFTA has positively affected

to trade throughout collected data and proposed that the including of unobserved explanation variables to estimation model carefully to absorbed trend in trade was needful

Among studies about AFTA effect, some of them focused on tariff reduction progress under an agreement on the common effective preferential tariff scheme (CEPT) Pelkmans-Balaoing (2007) used a short-time sample data from 2001-2003 in aggregated and disaggregated trade data flow to estimate AFTA effect to ASEAN member countries Although limited on data, they focused on the effectiveness of preference margin on trade carefully The results showed that AFTA has no or slight effect to intra-bloc trade, in essence, meanwhile creating a positive effect on some range of products in which preference margin is more than 25 percent Moreover, his result implied that the cost of applying AFTA would be higher than benefit obtaining from AFTA when the difference

of tariff of Most Favored Nation (MFN) and AFTA is slight Okabe and Urata (2013) investigated the effect of tariff elimination under CEPT through 52 products of ASEAN member countries in the period 1980-2010 They found that AFTA has created trade creation effect However, the magnitude of effect to new members such as Cambodia, Lao, Vietnam was pure small This result could be explained by the small share of new members as well as a subsequent schedule of tariff reduction Moreover, the impact to trade flow is not extremely strong From these arguments can conclude that tariff reduction is not an essential measurement to promote regional trade flow To promote ASEAN trade flow as well as increase welfare for all member countries, other factors such as trade facilitation, eliminating non-tariff barriers (NTMs), equalizing rules of origin (RoO)

as well as enhancing AFTA utilization should be considered

2.5.2 Empirical support of effect of ASEAN + 1 FTAs

A definition has been raised in many recent studies about free trade agreements (FTAs) and regional free trade agreements (RTAs) are “noodle bowl effect: stumbling or building block?” This definition is completely correct with ASEAN’s trade agreements status The reason for that

is to gather with General Agreement on Tariff and Trade (GATT), World Trade Organization (WTO), ASEAN has multilateral FTAs with six major economy countries named: Australia, New Zealand, China, Japan, Korea and India since the middle of the 2000s In addition, a remarkable number of bilateral FTAs has been signed and effected between these countries and ASEAN members, such as Japan has bilateral FTAs with Indonesia, Malaysia, Philippine, Singapore, Thailand, Vietnam; India has bilateral FTAs with Malaysia, Singapore; China has formed bilateral FTAs with Singapore, Thailand…

Following integration trend in international trade when FTAs related to ASEAN in force, a number

of ex-ante and studies has been made to estimate the effect of these FTAs caused to ASEAN Estrada et al (2011) used simulation analyst CGE model to compare the effect to welfare of member countries caused by ASEAN + China, ASEAN + Japan and ASEAN + Korea with current ASEAN + 1 FTAs He found that these FTAs has caused positive expectation and fascination to members China, Korea and Japan Using same as econometric technique CGE model to predict the impact of ASEAN + 1 FTAs, Sheng et al (2012) estimated gravity model period from 1980-

2008, he predicted that ASEAN-China (ACFTA) has caused massive impact to trading flow Especially, it affected to adjacent production linkage internationally and positive impact spreading among ASEAN countries Bano et al (2013) analyzed trade effect caused by ASEAN-Australia-New Zealand (AANZFTA) with data after 1980 with positive effect to trade across ASEAN countries and New Zealand, Australia was proposed Chandran (2012) worked with India-ASEAN FTA (AIFTA) focusing on fishery division, he quoted that the FTA has improved trade by tariff reduction, particularly low-developed countries Okabe (2015) used CGE model to forecast the effect of current ASEAN + 1 FTAs include ACFTA, AKFTA and AJCEP She found that trade creation has been caused by ACFTA, AKFTA in two sub-categories: industrial supplies and capital goods Misa (2015) used the sample data from 2002 to 2012 and gravity model to estimate the effect of ASEAN +1 FTAs They found that ASEAN-China FTA and ASEAN-Korea FTA has cause positive trade creation in industrial supplies and capital goods among member countries In addition, ASEAN-China also facilitates consumption goods In contrary, ASEAN-Japan has not revealed the impact on many cases

In general, the impact of AJCEP appeared limitedly at the moment of ex-post analysis Meanwhile, most of the studies about AJCEP show a negative or unclear effects The reason could be used to

explain this result was tariff reduction schedule and RoO certification while other former FTAs have been implicated in the longer time That is one of the reason encourage us re-estimate the effect of AJCEP to ASEAN members by using by the ex-post analyst

2.6 Zero trade data problem

Zero trade flow between a given pair of countries is a problem has been widely discussed Because the traditional way usually used to estimate gravity model is taking logarithms leading to drop out zero value to the data set However, zero data is not completely mean non-trade between two countries It may imply the trade volume with very small flows or even missing or loss when reporting process in many cases

In reality, several alternative zero approaches have been discussed The first method is truncate zeros data and still estimate log-linear by OLS By this method, zero trade data will be completely deleted from the matrix The second solution is the censoring method by substituting a small constant volume, for example, one dollar before taking logarithms trade value when estimating gravity model in level However, these methods can lead to inconsistent estimates and distort the significant result Burger et al., 2009, Gomez-Herrera, 2013 Moreover, according to Flowerdew and Aitkin (1982) indicate that the result is sensitive when replacing ad hoc zero data for small value is not guaranteed for an expected regression result, can-not be avoided inconsistent estimation Eichengreen and Irwin (1998), Heckman (1979), Helpman et al., (2008) debated that

in case zero data do not follow the random distribution, deleting zero data from trade matrix can lead to loss valuable information and create bias results

With the same result, Linder and Groot posit that applying truncation or censor method when dealing with zero data may lead to misunderstanding bilateral trade patterns Because maybe due

to far distance, low level in GDP or non-linkage in culture or historical, non-profitable in trade, firms make decision reducing trade or non-trading, then we eliminate zero data will cause underestimating coefficients Therefore, there is an attention on finding an appropriate technique

to deal with zero data recently

Some early studies have used Tobit model to deal with it, for example, Rose (2004), Andersen and Marcouiller (2002) With Tobit model, data will be fitted when data is observed in some range, involving rounding part of the observation to zero and rounding up the zero trade flow which below

positive value However, Linder and Groot, 2006 debated the appropriateness when using Tobit model to fit data when zero trade According to their argument, zero trade flow can-not be censored

at zero because desired bilateral trade can-not be negative except one or both countries in a pair have GDP equal 0, however, it is not real In addition, they indicated that censoring at positive value is also not appropriated, even 1$ only Because in UNCOMTRADE trade flow database, sometimes zero trade are caused by actual economic decision situation In this case, taking care of zero trade is needless Therefore, subject to measurement errors, this method will high influence

to regression results, Frankel (1997)

Recently, Head and Mayer (2013) has proposed new approach when dealing with a set of data with

25 percent by gravity model, based on Eaton and Kortum (2001), named EK Tobit model By this method, they will replace all zero trade data from country 𝑖 to all destination country 𝑗by minimum level of trade data recorded This method has two advantages First, without any criteria, we will easily collect minimum trade value which used to replace Second, easily estimate the model by using command 𝑖𝑛𝑡𝑟𝑒𝑔 in Stata

Another attention new method was developed by Santo Silva and Tenreyro (2006; 2011) is Poison Pseudo Maximum Likelihood (PPML) to deal with logarithm transformation and zero trade flow data They deal with a set of data with a share of zero trade at 62 percent According to their argument, PPML estimation will fix the problem by regressing model in the level instead of taking logarithms By this method, it will address observed heterogeneity, provide the instinctive way to estimate the model with zero trade because we no need to do log-linearized and lowest bias among other estimators However, Martin and Pham (2008), Burger et al., 2009, argued controversially that PPML method has some limitation, especially do not take account unobserved heterogeneity More recently, Head and Mayer (2014) proposed new method name Multinomial Pseudo Maximum Likelihood (MPML) In this method, dependent variable will be 𝑌𝑖𝑗

𝑌𝑗 as a market share variable and estimate by Stata command 𝑝𝑜𝑖𝑠𝑜𝑛 along with fixed effect

Another method proposed by Burger et al., (2009) to take care unobserved heterogeneity are Negative Binomial Pseudo Maximum Likelihood (NBPML) and Zero-inflated Pseudo Maximum Likelihood (ZIPML) However, they posit that this method is not well-suited for the cases a

number of zero trade data predicted by the model is lower than a number of zero trade data flow observed

In sum, according to listed review, each method has own pros and cons and the best method has not yet to be defined, remain in debate and are not an unclear decision However, in this paper, data is updated for time period recently from 2000-2015, the trade zero data were recorded at 15 percentage and concentrate on data of Cambodia and Brunei In 2015, the total export value of Cambodia and Brunei to the rest of the world were reported at 0.42 percent and 0.34 percent per total export value of the countries in this sample Therefore, we chose the simplest estimation method in case of present zero trade is dropping them out the data

CHAPTER 3: RESEARCH METHODOLOGY

In this chapter, we will diagnose the gravity model applied to international trade from multiplicative to logarithmic form We did it by adding variables which effect to bilateral trade to gravity model to investigate the impact of AJCEP to ASEAN as well as Japan Then, we process

by using a variety of estimation techniques such as OLS, Fixed effect model (FEM), Random effect model (REM), Hausman-Taylor estimator In addition, an overview of data scope and data sources are also mentioned in this Chapter

3.1 Model specification and validity testing

3.1.1 Model specification

Starting with the multiplicative equation of gravity model in international trade:

𝑋𝑖𝑗 = 𝛽0𝐺𝐷𝑃𝑖𝛽1𝐺𝐷𝑃𝑗𝛽2𝑃𝑂𝑃𝑖𝛽3𝑃𝑂𝑃𝑗𝛽4𝐷𝐼𝑆𝑇𝑖𝑗𝛽5𝐹𝑖𝑗𝛽6𝑢𝑖𝑗𝑡 (7) Following to the gravity model of international trade, the function of the total bilateral volume of export of a couple of countries 𝑋𝑖𝑗 is included their GDPs, population, distance and 𝐹𝑖𝑗 denotes a set of dummy elements which encouraging or discouraging bilateral trade flow included Border, Language, Colony, Land-lockedness, Free trade agreement

For estimation target as well as time dimension plus, we change model (7) into log-liner format which is given as below:

GPD at current US$ is the first independent variables which collected from the database of World trade organization WTO This variable denoted for total value within a certain time 𝑡at current US$ of all final goods and services produced in a country In gravity model, it includes 𝐺𝐷𝑃𝑖𝑡and 𝐺𝐷𝑃𝑗𝑡 which consider 𝑖 is exporting country and 𝑗 is importing country According to utility theory, when the income and output of a country increase, it will increase consumer demand for goods and service, leading to increasing production and export Nellis and Parker (2004), 𝐺𝐷𝑃 presents for country’s income and purchasing power, therefore, GDP will have the positive sign with total import plus export volume However, Basat (2002) indicated that positive relation only with middle development countries, there is no evidence for low and high development countries

 Weighted distance DIST (𝐷𝐼𝑆𝑇𝑖𝑗)

Weighted distance is the third independent variable is calculated by Mayer and Zignago (2005), based on inspired idea of Head and Mayer (2002) is calculating geographic distance between two countries 𝑖 and 𝑗 by biggest cities distance, inner cities distance being weighted by population ratio

of the city to the total country’s population

The reason for using this method instead of simple geodesic distance which calculated by using longitudes and latitudes is to avoid over or underestimate the effect of the border Taking an example of trading volume between Vietnam and China, one of the factor include to trade volume

is the comparison of domestics transportation cost inside China internally and international transportation cost between China-Vietnam and population in these cities The general function was developed by Head and Mayer (2002) to calculate the weighted-distance from country 𝑖 to country 𝑗 is:

𝑑𝑖𝑗 is the weighted-distance between two countries 𝑖 and 𝑗

𝑝𝑜𝑝𝑘, 𝑝𝑜𝑝𝑙 are population agglomeration of cities 𝑘, 𝑙 belongs to country 𝑖 and 𝑗 respectively

𝑝𝑜𝑝𝑖, 𝑝𝑜𝑝𝑗 are the total population of country 𝑖 and 𝑗 respectively

𝑑𝑘𝑙 is the geographic bilateral distance between two city 𝑘, 𝑙

According to Bakman, Garretsen and Marrewijk (2001), the further distance, the higher in the cost of transportation, culture difference, the higher cost of trade In other words, the relation between distance and trade flow are a negative correlation

Dummy variables

 Common Language, LANG (𝐿𝐴𝑁𝐺𝑖𝑗)

Language is the first dummy variable, used as a measurement tool to compare culture facture differences According to the theoretical support of Linnermann (1966), Hacker and Johansson (2001), the volume of trade between two countries was influenced by the common language as the communication barrier If two countries speak the same language, they can easy to communicate and reduce transaction cost Therefore, there are two binary values of 𝐿𝐴𝑁𝐺𝑖𝑗 variable, equal 1 if two countries have same language and zero if otherwise, the coefficient is expected positive sign

 Common Border, BOR (𝐵𝑂𝑅𝑖,𝑗)

A pair of countries sharing the common border with each other, they will have a lower cost of transportation, leading higher trade flow, Bakman, Garretsen and Marrewijk, (2001) Therefore, this dummy variable takes the binary value, equal 1 if a pair countries sharing a common border and 0 if otherwise The coefficient is expected positive sign

 Colony, COL (𝐶𝑂𝐿𝑖𝑗)

If two countries have been colonies of each other or a common colonizer, this dummy variable takes value 1 if two countries share a common border and 0 if otherwise Thus, apriori, the coefficient is expected positive sign

 Land-lock, LLOCK (𝐿𝐿𝑂𝐶𝐾𝑖𝑗)

If at least one of countries in a pair is the land-locked countries, this dummy variable will take the value at 1 and 0 if otherwise Various studies tried to highlight the cost of a country when being landlocked Mackellar, Worgotter and Worz (2002) indicate that being landlocked connected with increase import price as well as reduce in export revenue due to higher in intermediate export services, Stone (2001) Therefore, the coefficient is expected negative sign

 Free trade agreement, FTA (𝐹𝑇𝐴𝑖𝑗)

(𝐹𝑇𝐴𝑖𝑗𝑡) is a dummy variable are extended three different set of binary variables, included 𝐹𝑇𝐴_3𝑗𝑖𝑡, 𝐹𝑇𝐴_2𝑖𝑗𝑡, 𝐹𝑇𝐴_3𝑖𝑗𝑡 These variables denote the effect of ASEAN – Japan FTA (AJCEP) 𝐹𝑇𝐴_1𝑖𝑗𝑡 = 1 if both countries 𝑖 and 𝑗 are belonged to AJCEP after 2008 and zero if otherwise 𝐹𝑇𝐴_2𝑖𝑗𝑡 = 1 if exporter 𝑖 is the member of AJCEP officially in year t while destination importing country 𝑗 is not belong to AJCEP and equal zero otherwise 𝐹𝑇𝐴_3𝑖𝑗𝑡 = 1 in case exporter 𝑖 is from extra-bloc AJCEP members in year t and exporting to destination intra-bloc AJCEP country 𝑗 and get zero value if otherwise

If in the regression result, the coefficient of 𝐹𝑇𝐴_1𝑖𝑗𝑡 obtains a positive and statistically, it implies that trade creation effect is generated In other words, intra-regional trade flow has been promoted more when the free trade agreement is in force in time t than normal Similarly, a statistically significant and positive coefficient of 𝐹𝑇𝐴_2𝑖𝑗𝑡 is implied that FTA has created trade creation effect in term of exports Expressing in the different way, export activities has been switched from AJCEP member countries to extra-bloc AJCEP countries by regional integration agreement Contrarily, a statistically significant and negative sign of coefficient of 𝐹𝑇𝐴_2𝑖𝑗𝑡 expresses a decrease in exports value from member AJCEP to non-member AJCEP countries and implies an export trade diversion effect A regression result shows a positive sign and statistically significant the coefficient of 𝐹𝑇𝐴_3𝑖𝑗𝑡, it means that a trade creation effect in terms of import from extra-bloc

to intra-bloc AJCEP In this case, we conclude that AJCEP has encouraged export flow from member countries to member countries Contrarily, we will conclude that a trade diversion effect

non-in relation to imports has been created if the coefficient of 𝐹𝑇𝐴_3𝑖𝑗𝑡 is negative and statistically significant The model should be:

𝑙𝑛𝑋𝑖𝑗𝑡 = 𝛽0+ 𝛽1𝑙𝑛𝐺𝐷𝑃𝑖𝑡+ 𝛽2𝑙𝑛𝐺𝐷𝑃𝑗𝑡+ 𝛽3𝑙𝑛𝑃𝑂𝑃𝑖𝑡 + 𝛽4𝑙𝑛𝑃𝑂𝑃𝑗𝑡+ 𝛽5𝑙𝑛𝐷𝐼𝑆𝑇𝑖𝑗

+ 𝛽6𝐿𝐴𝑁𝐺𝑖𝑗+ 𝛽7𝐵𝑂𝑅𝑖𝑗+ 𝛽8𝐿𝐿𝑂𝐶𝐾𝑖𝑗 + 𝛽9𝐶𝑂𝐿𝑖𝑗 + 𝜃1𝐹𝑇𝐴_1𝑖𝑗𝑡+ 𝜃2𝐹𝑇𝐴_2𝑖𝑗𝑡+ 𝜃3𝐹𝑇𝐴_3𝑖𝑗𝑡 + 𝜇𝑖𝑗𝑡 (9)

3.1.2 Model validity testing

There are many techniques to estimate to gravity model which are mentioned If ignoring panel data and assuming that error term 𝜇 not correlated with the dependent variable and identical distribution and constant variance 𝜀 ~ 𝑁(0, 𝜎2), we can regress model by Pool OLS technique However, pool OLS estimator obtained the biased result due to omitted variables

Therefore, a more proper estimation method should be considered to observe export direction effect and time effect is high appreciated Export direction effect allows its unobserved time-invariant variables such as history, culture, diploma which affect to the trend of export These unobserved variables might be correlated with explanation variables In addition, time effect suggests that some factors such as export potential trend or cycle of business may cause the effect

to the direction of export Therefore, we need to add these effect to regression model and control them To take into account these problems, basically we have two models: fixed effect model and random effect model Then, we use Hausman test to examine whether fixed effect model or random effect model is high appreciated

We have to pay attention to the actuality that the model includes not only time-varying elements such as GDP, Population and set of AJCEP FTA dummies but also time-invariant elements named bilateral distance, language, border, colony, land-locked Econometrically, while Fix effect model (FEM) does not allow us to estimate time-invariant variables, Random effect model (REM) can obtain regression result for time-varying variables as well as time-invariant variables Unobserved individual effects are included into the error terms while REM’s assumption is the error term is not correlated with variables in the model Therefore, in many cases, result of REM is inconsistent due to ignoring the possibility of the correlation between the unobserved effects and explanation variables

The alternative model suggested by Hausman and Taylor (1981) to fix the disadvantages of FEM and REM named Hausman-Taylor method It is a random effect including instrument technique to

eliminate the disadvantage of REM when fixing the correlation between included variables in the model and error term Basically, a hybrid of FEM and REM takes the following equation:

Hausman-Taylor is the method in which using variables from X-matrix correlated with 𝛼 to produce unbiased estimates of 𝛽’s by deviation from individual means as well as instruments for

forZwith correlated with 𝛼 by using individual In addition, according to Baier and Bergstrand (2002), FTA dummy variables perhaps endogenous due to the correlation of unobserved variables

or omitted variables such as culture, politic, history Therefore, it will cause biased estimation Wooldridge (2000) suggested to using FTA dummy variables as instruments to fix endogeneity

To check the consistency of Hausman-Taylor estimator to use Hausman test for identification It will compare the estimator between Hausman-Taylor model and Within (Fixed effect model) An acceptance the null hypothesis means the instruments are reliable or no bias because the specific bilateral effects correlate with explanatory variables

over-To compare the potential efficiency of Hausman-Taylor method, we will regress model in many methods: Pool OLS, FEM, REM and Hausman-Taylor

3.2 Data and data sources

A panel data set will be collected with period from 2000 – 2015 with total 5,920 observations with included 09 ASEAN countries: Brunei Darussalam, Cambodia, Lao PDR, Malaysia, Myanmar, Philippines, Singapore, Thailand, Vietnam and 15 biggest trading partners of Japan 2015 include: The United State, China, South Korea (Korea Rep.), HongKong SAR China, Australia, Saudi Arabia, The United Arab Emirates, Russian Federation, Switzerland, New Zealand, United Kingdom, Germany, Mexico, Netherland and Japan Regarding Indonesia, until 2016 record,

Indonesia has not yet to finalize document to join into AJCEP, therefore, Indonesia is not included

in selected data of ASEAN

Trading export value is obtained from the UN-Comtrade database at aggregated and disaggregated level and are extract on the Standard international trade classification (SITC), revision 3 and stay

on nominal data to eliminate error in measurement (Baldwin & Taglioni, 2006) Disaggregated data will be split into five sub-sections separately which absorb the highest tariff reduction on AJCEP agreement: agricultural products (sum export value of SITC No 0, 1, 2, 4, minus 27, 28), manufactured products (sum of export value of SITC No 5, 6, 7, 8 excluding 667 plus 68), chemical products (SITC), Machinery and equipment of transportation (SITC 7) and clothing and accessories and textile, fabric (SITC 84 plus 65)

Data on GDP and population are obtained from World Bank development indicators database The universe of RTAs is withdrawn from the “Regional Trade Agreement Information System” (RTA-IS) database of the World Trade Organization (WTO) Geographic and cultural data on distance, common language and adjacency are from CEPII database

Table 1: List of countries

14 biggest trading partners of Japan 2015

CHAPTER 4: RESEARCH FINDINGS AND DISCUSSION

This section will be divided into two sub-section The first will be statistics of variables and their economics meanings The second one will be econometrics results and its meaning In the second sub-section will be added some testing to find out the best model estimated and reject the unsuitable models

4.1 Descriptive statistics of variables

Variables summary statistics of panel data are presented in Table 2

Table 2 shows that with the full panel data of 24 countries with 5,920 observations cover the period time from 2000 to 2015 Total export value shares 0.34% in total gross domestic products The minimum value to export value is 0 can come from missing data reported, or resulting from rounding process error or even firms in these countries make the decision do not export to each other In the total export trading value, manufactured goods play a majority proportion, about 77.80% and are divided into two sub-categories as follow: Chemical products 8.64%, machinery and transportation equipment 48.72% respectively Agricultural products obtain 6.94% in total export

Table 2: Descriptive statistics of variables

Machinery and transport equipment

Dummy variable 𝐹𝑇𝐴_1 takes the value 1 if both importing and exporting country are belong to free trade agreement AJCEP from 2008 The effective time of each country will be adjusted according to summary in Table 10 The descriptive statistics of other variables in case 𝐹𝑇𝐴_1 = 1

is shown on Table 3

Table 3: Descriptive statistics of variables if pair of countries belongs to AJCEP from 2008

Gross domestics products (Exporting

Gross domestics products (Importing

Machinery and transport equipment

Source: author calculation

Dummy variable 𝐹𝑇𝐴_2 takes the value 1 if both exporting country is belong to AJCEP and importing country is not belong to free trade agreement AJCEP from 2008 The descriptive statistics of other variables in case 𝐹𝑇𝐴_2 = 1 is shown on Table 4

Table 4: Descriptive statistics of variables if exporting country is belong to AJCEP and

importing countries is not belongs to AJCEP from 2008

Gross domestics products (Exporting

Gross domestics products (Importing

Population (Importing country) 952 1.65E+08 3.38E+08 6972800 1.37E+09

Machinery and transport equipment

Source: author calculation

Dummy variable 𝐹𝑇𝐴_3 takes the value 1 if both exporting country is not belong to AJCEP and importing country is belong to free trade agreement AJCEP from 2008 The descriptive statistics

of other variables in case 𝐹𝑇𝐴_3 = 1 is shown on Table 5

Table 5: Descriptive statistics of variables if exporting country is not belong to AJCEP and importing countries is belongs to AJCEP from 2008

Gross domestics products (Exporting

Gross domestics products (Importing

Machinery and transport equipment

Source: author calculation