A Deeper Insight into the Effects of the ASEAN-China Free Trade Agreement: Mutual Differences, Trade Creation and Trade Diversion.

(1)

A Deeper Insight into the Effects of the ASEAN-China Free Trade

Agreement: Mutual Differences, Trade Creation and Trade Diversion.

Author: Jordy Jansema Student Number: 3539946

Supervisor: Prof. Dr. J.H. Garretsen Co-assessor: Prof. Dr. B. Los University of Groningen

(2)

Abstract

The aim of this research is twofold. First, this paper will examine different ASEAN-China Free Trade Agreement (ACFTA) effects between two groups of ASEAN member countries. On the other hand, it will examine the impact of the ACFTA on export flows. Here, it will focus on the trade creation and diversion effects, where in both cases the gravity model of trade is used. There was no concluding evidence found for mutual differences between the ASEAN member countries. At the same time, evidence does suggest that trade levels between intra-ACFTA member countries are lower than normal. However, a robustness test shows that this result should be interpreted with care.

(3)

In November 2000, the leaders of the Association of Southeast Asian Nations (ASEAN) and China agreed to look into the possibility of creating a free trade area. Willing to establish more integration and economic cooperation between the China and the ASEAN member countries. The ASEAN-China Free Trade Agreement (henceforth ACFTA1) was created, covering an economic region with 1.7 billion consumers with trade worth 39.5 billion US$ occurring between ASEAN and China in 2000.2

Twenty years later was the first moment in time when ASEAN overtook the European Union (EU) as China’s largest trading partner. By surpassing both the EU and the United States, it became China’s largest trading partner in the first three months of 2020. This happened after already becoming China’s second largest trading partner in 2019 with trade valued at 644 billion US$.3 A trend of growth that can also be found in figure 1, showing the trade volume between China and ASEAN from 2011 till 2017.

Figure 1 ASEAN-China trade volume in USD from 2011-2017.4

What caused this significant growth in trade volume throughout the years? Can it be contributed to the existence of the ACFTA, or are there other important factors that made this growth in bilateral trade possible?

An interesting detail of the ACFTA is the fact that the timeline for the reduction and elimination of tariff lines is not equal to everyone.5_{Table 7 in appendix I gives a schematic timeline}

1_{To avoid confusion while referring to the ASEAN-China free trade agreement, for example in regard to the}

Central American Free Trade Agreement (CAFTA), the acronym “ACFTA” will be used throughout this paper, as is often found in the literature.

2_{Forging closer ASEAN-China economic relations in the twenty –first century. October 2001. A report submitted}

by the ASEAN-China expert group on economic cooperation.

3_{Source: ASEAN Briefing, Top trading partner in Q1 2020. Retrieved on 14-11-2020 from:}

https://www.aseanbriefing.com/news/asean-overtakes-eu-become-chinas-top-trading-partner-q1-2020/

4_{Source: CGTN, China-ASEAN in numbers: Trade ties. retrieved on 14-11-2020 from:}

https://news.cgtn.com/news/3d3d414e3145544d7a457a6333566d54/share_p.html

5_{The full timeline for tariff reduction and elimination for tariff lines placed in the normal- and sensitive track can}

(5)

2 overview of the differences in the reduction and elimination of tariff lines between the two groups of countries.6 The first difference can already be found in the Early Harvest Program (EHP). The first group, further mentioned in this paper as “ASEAN-6”, refers to the countries of Brunei Darussalam, Malaysia, Singapore, Indonesia, Thailand and the Philippines. Where the second group, referred to as “the newer ASEAN member countries”, are Lao PDR, Cambodia, Viet Nam and Myanmar.

The differences in the reduction and elimination of tariff lines are reason to state the first question of this research: does the ACFTA affect the ASEAN-6 and the newer ASEAN countries

differently? This question must be asked to exam if this could create unequal effects on bilateral

trade flows between the two groups of ASEAN countries and China. To the author’s best knowledge, this will be the first attempt in the literature that evaluates possible different trade creation effects between the two groups of ASEAN member countries. An important question influencing future free trade agreement (FTA) negotiations. The answer will show whether or not being a first implementer of a FTA could be more beneficial than being a later implementer. The second objective of this paper is to: re-examine the trade creation and diversion effects

resulting from the ACFTA. Using the most up-to-date available data as possible, this

re-examination will be a useful contribution to the existing literature by showing the true effects of the ACFTA after 16 years of implementation. This will expand upon the depth of the previous work in terms of timespan and countries examined.

To study the ex post effects of free trade agreements on bilateral trade flows, the “gravity equation” has emerged as the empirical workhorse in international trade throughout the years. The gravity equation is well known for making it possible to explain variation in country pairs trade flows. To conduct this research, this paper will follow the methodology proposed by recent literature on this topic

The aim of this research is twofold. First, this paper will examine different ACFTA effects between two groups of ASEAN member countries. On the other hand, it will examine the impact of the ACFTA on export flows. Here, it will focus on the trade creation and diversion effects, where in both cases the gravity model of trade is used. There was no concluding evidence found for mutual differences between the ASEAN member countries. At the same time, evidence does suggest that trade levels between intra-ACFTA member countries are lower than normal. However, a robustness test shows that this result should be interpreted with care. The setup of this paper is as follows. Chapter 2 of this paper will describe the existing literature about FTAs and the ACFTA. Here, the main questions of this research are further introduced. In chapter 3, there are descriptions of the method and model which will be used throughout this research. This chapter goes more in depth about the gravity equation and the main variables are interest are examined. Chapter 4 covers the data itself and chapter 5 includes a discussion about the results of the use of the gravity equation on this data. Chapter 6 contains a robustness check. Chapter 7 brings about the final conclusions of this work and chapter 8 discusses the applicable limitations which apply to these conclusions. Finally, chapter 9 includes an additional analysis drawn from including other FTAs from around the world.

6_{Take the information in table 6 in appendix I with notion where the table is possibly outdated. However for a}

(6)

3

2. Literature review and hypothesis.

2.1 FTAs

Trade agreements (TAs) have been an important topic in the international trade studies. Especially after the seminal contribution by Rose (2004) who questioned the assumed positive impact of the Wold Trade Organization (WTO) on international trade. As a result, research looking into the effect of TAs started to increase. Research looking into both the effect of individual FTAs and FTAs in general has been done. This paper takes a look at previous research studying the effects of different FTAs from around the world to gain a better understanding of possible effects of FTAs, before looking into the existing literature about the ACFTA.

Starting with the fact that after 40 years of gravity equation estimates of the effect of FTAs on trade flows without having clear and convincing empirical evidence, the authors Baier and Bergstrand (2007) answered ‘yes!’ to the question of whether or not FTAs actually increase members’ international trade using the gravity model. This is done by treating FTAs as endogenous variables subject to interaction effects.

While TAs are mostly used as a binary variable in the literature. Kohl et al. (2013) state that trade agreements have many differences between them. Therefore, looking into how heterogeneity within 296 trade agreements, using a gravity model, can affect international trade differently. Showing that it can have both, a positive and negative effect.

Looking at individual effects of FTAs can yield similar results. The effect of one of the world’s biggest FTAs, in the form of the North American Free Trade Agreement (NAFTA), is be found positive and significant by Gould (1998), at least partially. Where there is a positive effect on trade flows between the US and Mexico. But, the NAFTA is not significantly impacting the trade flows between the US and Canada or Canada and Mexico. This is explained by the author by the fact that a negotiated FTA between the US and Canada was signed five years before the implementation of the NAFTA. This FTA, called the Canada-United States Free Trade Agreement (CUSFTA), had a substantial trade creation effect while there was only little evidence of trade diversion, according to Clausing (2001). At the same time, evidence exists showing that FTAs affect trade positively outside of North America. Abedini and Peridy (2008) show, using the gravity model, that for the Greater Arab Free Trade Area (GAFTA), regional trade has increased by 20% since the GAFTA had been implemented in 1997. Where Peridy (2005a) looked into trade effects of the Euro-Mediterranean Free Trade Agreement (EMFTA). Finding, also using the gravity model, that the EMFTA increased Mediterranean countries exports to the European Union by 20-27%.

Common results in the research mentioned above are the facts that the gravity model is used and secondly, that the implemented FTA seems to positively influence trade, with a few exceptions. More research regarding the effects of FTAs will be presented in the next paragraph when looking at the existing ACFTA literature.7

(7)

4

2.2 The ACFTA

Research with regard to the effects of the ACFTA is not only done after the FTA has been put into force. Several researches tried to show ex ante how the FTA could possibly influence the members of ASEAN and China.

Chirathivat (2002) shows, by using a Computable General Equilibrium (CGE) model, that both ASEAN and China could expect an increase in Gross Domestic Product (GDP) growth. Where ASEAN members could be seen as a possible alternative source of inputs for China regarding natural resources and intermediate products. Generating trade gains for both sides. Park et al. (2008) also apply the CGE model to determine whether or not the ACFTA would be beneficial for both parties. Anticipated results differed between ASEAN members. The FTA was expected to benefit the richer countries such as Singapore and Malaysia more than the poorer ones like Lao PDR, Cambodia and Myanmar.

According to Qiu et al. (2007), using the Global Trade Analysis Project (GTAP) model, the ACFTA would promote bilateral trade in the agricultural sector, generating positive effects on the economic development of both sides. Accelerating exports of the agricultural commodities in which China has an comparative advantage. However, regional agricultural development differences were expected. Namely, agriculture in southern China would suffer and other regions of China are expected to benefit from the ACFTA. Also Lee and Mensbrugghe (2007) used the CGE model to examine alternative FTA scenarios and the effects in East Asia on patterns of trade and sectoral adjustments. Using a dynamic global CGE model finding that any FTA with Japan, China or Korean would bring relatively large welfare gains to all member countries.

The gravity model is another commonly used method which tries to predict the effects of FTAs. Sheng et al. (2012) looked into how trade in parts and components will be influenced by the ACFTA using an extended gravity model. They argued that existing studies generally assumed that trade flows are mainly in final goods, underestimating the impact of a FTA if the nature of trade among the members is highly influenced by a large and growing proportion of component trade. Results of that paper stated that component trade will have a substantially larger impact on the trade flows between China and ASEAN than is predicted by existing literature. The larger trade flows, mainly in parts and components, between the two regions would also create more integrated industries. Finally, they argued that this will also generate positive spill-over to the rest of the world.

(8)

5 expected export performance in the sample data period shows a decrease of competitiveness. Resulting in a decline in the share of exports in commodities.

A general positive effect following from the ACFTA was predicted by the aforementioned research although differences between countries and industries might exists. However, three different papers which examined the effect of ACFTA for Indonesia draw different conclusions. Specifically negatively effects upon the maize industry and the agricultural trade flows, while at the same time a decline in the share of exports in commodities will hit the country.

After the implementation of the ACFTA, several papers have tried to show already how it has influenced different countries and its industries.

The next four researches that look at the effects of reducing and removing tariff barriers ex post use the gravity model which found the following results: Yang and Martinez-Zarzoso (2014) found that the ACFTA promotes total trade volume, showing that trade did not only increase among intra-bloc member countries, but also between extra-bloc and intra-bloc countries. The biggest positive impact in terms of exports being in manufactured goods and chemical products. An important finding by Wang (2018) is the fact that the ACFTA created an increase in the bilateral trade volume between the China ASEAN members. At the same time there has been a trade transfer in the share of China’s biggest import and export partners. It can be seen that the EU, Japan, the United States and South Korea in general have shown a market decrease trend when it comes to the proportion of imports. Taguchi et al. (2016) find by reviewing different ASEAN FTAs that the biggest trade creation effect was due to ACFTA. The trade diversion effect was negative. Arguing that the larger trade creation effect in ACFTA might be due to fact that it is filling a wider gap between the general tariff rate and the preferential rate for ASEAN in China. Finally, Alleyne et al. (2020) looked into the impact of export efficiency flow from ASEAN into China. Finding that the majority of ASEAN members export efficiency to China improved. Also stating that the long-run relationship between ASEAN members and China will increase their collective bargaining power on the global market.

(9)

6 Although this research found positive effects resulting from the implementation of the ACFTA, one of the first works looking into the effect of ACFTA on trade flows in East-Asia found no conclusive result regarding the impact on intra-regional trade. Again using the gravity model, they showed that although ACFTA stimulates intra-regional trade and impacts trade with the rest of the world, positively when it comes to exports and negatively in the case of imports. They argued that both ASEAN trade agreements with China and South Korea are not yet found to be favourable for intra-regional trade, stating that it is not impacting East-Asian trade flows. A conclusion can be drawn that a general consensus seemed to be that total trade volume has increased after the implementation of the ACFTA. Additionally, bilateral trade between ACFTA members and China also increased. Improving export efficiency of ASEAN members and benefitting non SOE most. Nevertheless, not all research is showing positive effects. Interestingly, Guilhot (2010) show that AFCTA is not impacting East-Asian trade flows.

2.3 This research

There are clear differences in the timeline for eliminating tariffs between the ASEAN-6 and the newer ASEAN countries. This research will make use of the gravity model which makes it possible to research the effect of a FTA ex-ante implementation. Using the newest available data to look at the fact whether or not the ACFTA is affecting the ASEAN-6 and newer ASEAN countries differently. The second objective of this paper is to re-examine the trade creation and trade diversion effect of the ACFTA.

“Examining the ASEAN-China Free trade agreement; does the ACFTA affect the ASEAN-6 and the newer ASEAN countries differently?”

In the first place, showing whether or not the liberalization of trade between ASEAN and China actually led to the growth of trade in the region is something worth investigating, simply for the fact that the ASEAN-China agreement is being one of the world’s largest free trade zones. Besides that, there are also big differences between ASEAN countries when it comes to their level of development and economic structures. This could indicate possible differences for (groups of) countries having a different timetable of their implementation of the same FTA. The results of this could change future FTA negotiations. Further,more could be learned about whether or not being a first implementer of a FTA would be more beneficial. Finally, it aims to show ASEAN member governments the impact of the ACFTA which could influence future governance. This is, to the best of the author’s knowledge, the first attempt in the literature that looks for different effects between member countries within the ACFTA.

“Re-examine the ASEAN-China free trade agreement; an analysis of trade creation and diversion effects.”

(10)

7

3. Method and model

3.1 The gravity model

It has been shown that the gravity equation is a well-known used method to analyse the size of bilateral trade flows between different geographical entities. Based on Newton’s law of universal gravitation, Tinbergen (1962) was one of the first founders of the concept applied in international trade flows. Tinbergen argued that there is a clear similarity between the mutual attraction between planets proportional to their size and proximity, and the amount of trade between countries with respect to their economic size (often in GDP or GNP) and proximity. A basic gravity model following this theory is written as:

(1) Tradeij

= α

𝑌_𝑖𝛿1𝑌_𝑗𝛿2

𝐷𝑖𝑠_𝑖𝑗𝜃

Where TRADEij is the amount of bilateral trade flows, which depends on the gross domestic

product of importer Yi and exporter Yj being divided by the geographical distance, Disij, between

the two countries. The amount of trade being positively related to the size of the economies of the participating countries and the distance affecting it in an inverse manner. Although it had a big impact on international trade literature, it was lacking a theoretical foundation. By adding a constant elasticity of substitution and a Cobb-Douglas production function, Anderson (1979) made an important contribution to the gravity equation literature. Assuming product differentiation throughout the following years, several empirical applications in the literature contributed to the improvement of the gravity equation and its performance.

A generalized gravity model of trade followed from Martinez-Zarzoso & Nowak-Lehmann (2003) looks as follows:

(2) Tradeij = β0 Y_iδ1Y_jδ2 N_iδ3N_jδ4Dis_ijδ5F_ijδ6 uij

Where the same basic assumption holds; the volume of trade between a pair of countries (Tradeij) is a function of GDP values for the two countries (Y_iδ1Y_jδ2), the countries’ population

(N_iδ3N_jδ4), the geographical distance between them (Dis_ijδ5), how trade may be affected by a set of binary variables (F_ijδ6 ) representing other factors which either enhance or prevent trade between countries and uij being the error term. Where subscripts i and jindicate the role of

exporter and importer respectively.

(11)

8 Below is an overview of the dummy control variables that are used:

- Language being 1 if countries i and j have a common language and 0 otherwise. - Contiguity being 1 if countries i and j share a land border and 0 otherwise.

- Colony45 being 1 if countries i and j have colonial links after 1945 and 0 otherwise.

3.2 The first model specifications

To answer the main questions, two sets of binary variables will be created. These parameters will be of the most interest in the end, answering the main questions of this research. Throughout this paper, there will be a distinction made between the two sets of binary variables of interest and their results. For both sets of binary variables,2004 is the first year that the ACFTA is implemented even though the official date that the agreement went into force is 1 July 2003. 8 There is no agreement at play for the first six months of 2003. Therefore 2004 was chosen to be the first year to capture the full effect of the ACFTA.

The first set of binary dummy variables consists of three variables. Answering the question: “does the ACFTA effect the ASEAN-6 and the newer ASEAN countries differently”

- ACFTA_6ijt, which takes a value of 1 if both countries i and j belong to the ACFTA in year t

and 0 otherwise. This is only including the ASEAN-6 countries and China.

- ACFTA_NEWijt , which takes a value of 1 if both countries i and j belong to the ACFTA in

year t and 0 otherwise. This is only including the newer ASEAN member countries and China.

- ACFTA_ALLijt taking a value of 1 if both countries i and j belong to the ACFTA in year t and

0 otherwise.

Where parameter ACFTA_6ijt measures the effect on bilateral trade from the moment the

ACFTA between the ASEAN-6 and China is in use and tariffs are reduced mutually. The second variable, AFCTA_NEWijt, measures the same effect of tariff reduction from the moment ACFTA

is actively in use for the newer ASEAN countries and China. ACFTA_ALLijt captures the overall

effect of the ACFTA on trade creation. The ACFTA_ALLijt dummy variable should only be

picking up relative differences in trade between ACFTA members relative to non-ACFTA members. If trade is created at a moment when the ACFTA is in use then at least one of the parameters should be positive. This would mean then that the ACFTA promotes intra-regional trade, where trade levels are higher than normal. Aiming to show possible differences in the effect of the agreement between the two groups.

A second set of binary variables makes it possible to evaluate the trade creation and trade diversion effects of the ACFTA. Notice that the variable, ACFTA_ALLijt , is copied from the

first set of binary dummy variables as was mentioned above. To make a distinction between trade creation and trade diversion, the overall effect of the ACFTA is needed. There is no reason to make a new variable if the results will be the same.

8_{According to the Framework Agreement on Comprehensive Economic Co-Operation Between ASEAN and the}

People’s Republic of China the EHP started on the 1st_{of July 2003. Retrieved on 07-11-2020 from:}

(12)

9 - ACFTA_ALLijt taking a value of 1 if both countries i and j belong to the ACFTA in year t and

0 otherwise.

- ACFTA_EXPijt, taking a value of 1 if the exporter i is a member of the ACFTA in year t while

destination country j is not and 0 otherwise.

- ACFTA_IMPijt, taking a value of 1 if exporter i is a non-ACFTA member in year t while

destination country j is a ACFTA member and 0 otherwise.

Following the methods of Carrère (2006), Martinez-Zarzoso et al. (2009) and Yang and Martinez-Zarzoso (2014) for the additional explanation of trade creation and trade diversion effects. A positive and statistically significant, ACFTA_ALLijt, coefficient indicates trade

creation effects where the ACFTA promotes intra-regional trade and where trade levels are higher than normal. A positive and statistically significant coefficient for ACFTA_EXPijt,

indicates a trade creation effect in terms of exports where there is more (regional) integration which leads to an increase of export activities with ACFTA members exporting more to the rest of the world. On the other hand, a statistically significant negative coefficient indicates an export diversion effect. This would be meaning a decrease in exports from ACFTA members to the rest of the world. The final variable is ACFTA_IMPijt. Trade creation effects in terms of

imports occur if a statistically significant coefficient is positive with the amount of imports increasing from the rest of the world to ACFTA countries. A statistically negative coefficient indicates thus a trade diversion effect in terms of imports.

According to Carrère (2006) and Martinez-Zarzoso et al. (2009), one binary variable which only captures the intra-bloc trade is insufficient to determine whether or not there is a net trade creation due to the ACFTA. Table 1 shows that it is possible that trade creation and diversion effects may offset each other. If, for example, intra-bloc exports increase (β1 > 0), but at the

same time this effect is offset by a reduction in imports from extra-bloc countries (β3 < 0), there

is no overall trade creation effect to be found. If both β1 and β2 are > 0, there is pure trade creation

where exports from intra-bloc countries to extra-bloc countries increased. If β1 is positive while

β2 is not, trade creation effects and exports diversion effects are at play at the same time and the

higher magnitude decides whether or not trade creation prevails.

Table 1 FTA Trade effect outcomes

Export effects Import effects

β2 > 0 β2 < 0 β3 > 0 β3 < 0 β1> 0 Pure TC (X) TC + XD (β1> β2 ) or XD (β1< β2) Pure TC (M) TC + MD (β1> β3) or MD (β1 < β3) β1 < 0 XE XD + XC ME MD + MC

Note: β1 is the coefficient of FTA_ALL denoting exporters among ACFTA member countries. β2 is the coefficient of FTA_EXP

denoting exports from ACFTA member countries to non-member countries. Finally, β3 is the coefficient of FTA_IMP denoting

(13)

10

3.3 The final model setup

To test whether the implementation of the ACFTA has created more bilateral trade among the member countries, and to see whether there are different effects for groups of countries over time, the two sets of different ACFTA dummy variables will be tested through a gravity equation. As described before, the inclusion of a FTA dummy variable in a gravity equation can be problematic. Yang and Martinez-Zarzoso (2014) stated that in order to obtain unbiased estimates for a FTA dummy variable, while at the same time overcoming the problem of country-specific heterogeneity that will be ignored if all countries in the FTA are treated as a homogeneous group, a panel data with a fixed effects model should be used. This way time-invariant factors that vary bilaterally will be accounted for. Further explaining that the panel data approach will be controlling for both time-invariant country-pair and country-and-time unobserved heterogeneity. However, as a panel data set is used, including repeated observations of pairs of countries over time, both observable and unobservable effects may arise. Explanatory variables control for most important bilateral time-varying effects, such as TAs. Whereas bilateral time-invariant effects are taking into account with fixed effects. In the end it is still possible that a correlation pattern between pairs of countries over time are present in the error term. To capture any intra-cluster correlation of the trading pair, Bertrand et al. (2004) introduced clustering the errors over country-pairs and provided evidence that this procedure does quite well in general. Throughout this work, standard errors will therefore be clustered by country-pair.

A 10 year overview by Kepaptsoglou et al. (2010) found that most empirical gravity studies tend to find better results while using the fixed effects model rather than using the random effects model.9

The extended gravity model will look as follows, where variables are in natural logarithms: (3) ln(Xijt) = β0 + β1ln(GDPit) + β2ln(GDPjt) + β3ln(Popit) + β4ln(Popjt) + β5ln(Distanceij) + β6ln(Contiguityij) + β7ln(Languageij) + β8ln(col45it) + β9(ACFTAijt) + εijt

First, equation (3) will be estimated using the pooled OLS technique. Both sets of binary ACFTA dummy variables will have their own table with results. Despite the fact that coefficients will be possibly biased and inconsistent, this equation is used to provide a benchmark for models that will follow. Results regarding the different effects between ASEAN countries are shown in table 2. Results for the trade creation and diversion dummies are shown in table 3. The expected effect for GDP is positively associated with a country’s level of trade as a higher amount of GDP would attract more trade. The expected effect of a country’s population on bilateral trade is ambiguous. Yang and Martinez-Zarzoso (2013) argue that larger populations imply larger domestic markets and therefore being less dependent on international trade. However, Dinh et al. (2011) argue that a bigger population can be seen as a bigger market and the larger the market is, the more it will trade, indicating a positive population sign. The sign for distance, a proxy for trade costs, is expected to be negative. The remaining control binary variables; contiguity, shared language and colonial links after 1945, are described to be trade promoting and therefore the expected signs are positive.

(14)

11 The two sets of ACFTA dummy variables are of primary interest. A positive sign would indicate that after the implementation of the ACFTA, more bilateral trade between the participating members of the FTA has occurred and the ACFTA would be trade promoting. Determining whether or not the net trade creation effect is also positive will be done with the results following from the final model.

The reason why a country would join a specific FTA is hard to identify and often correlated with the level of trade. For example, due to the fact that countries in a FTA are likely to form a FTA with partners with which they already trade more in general. There, gravity equations often encounter the problem of endogeneity bias due to omitted variables.

Anderson and van Wincoop (2003) made an important contribution to the existing gravity literature showing that there are border effects in trade using a Non-linear Least Squares (NLS) model. Introducing the concept of multilateral resistance terms, indicating that the costs of bilateral trade between two countries are not only affected by bilateral trade costs but also by the relative weight of trade costs in comparison to trading partner countries in the rest of the world. To avoid having biased estimates, multilateral resistance terms should be taken into account for the model parameters. To tackle the problem of multilateral resistance terms, this paper will follow the method of Baldwin and Taglioni (2006). Therefore, equation 4 includes individual country dummies. These country dummies aim to capture time-invariant individual country-specific characteristics that are omitted out the model. Furthermore, to control for unobserved time-varying phenomena, yearly fixed effects are added to the equation.

(4) ln(Xijt) = β0 + β1ln(GDPit) + β2ln(GDPjt) + β3ln(Popit) + β4ln(Popjt) + β5ln(Distanceij) + β6ln(Contiguityij) + β7ln(Languageij) + β8ln(col45ij) + β9(ACFTAijt) + θt + Ii + Ij+ εijt

Where the time-invariant effects is denoted by θt. Both Ii and Ij are country-specific effects that

account for the multilateral resistance terms.

Baier and Bergstrand (2007) make use of the methodologies following from Anderson and van Wincoop (2003), and further extend the data series from a cross-section to a panel dataset. They argued that country-pair fixed effects should be added in order to eliminate the endogeneity bias coming from using a FTA dummy variable. Therefore, in equation 5 are country-pair fixed effects also added to the equation. Country-pair fixed effects make it possible to control for country-pair heterogeneity. Note that due to this process, the fixed effect model is not able to estimate the impact of time-invariant bilateral determinants, the control variables; ln(Distanceij), ln(Contiguityij), ln(Languageij) and ln(col45ij) drop out of the equation due to

perfect collinearity.

(5) ln(Xijt) = β0 + β1ln(GDPit) + β2ln(GDPjt) + β3ln(Popit) + β4ln(Popjt) + β9(ACFTAijt) + θt +

Ii + I+ Iij + εijt

Where country-pair fixed effects are denoted as Iij.

(15)

12 Finally, the most important finding from Baier and Bergstrand (2007) is the fact that multilateral resistance terms should not only be country specific but also be time-varying in a panel setting. Therefore, country-time varying fixed effects are added to the equation.

(6) ln(Xijt) = β0 + β9(ACFTAijt) + θt + Iij + Iit + Ijt + εijt

Where country-and-time effects terms Iit and Ijt are introduced to control for time-varying

multilateral resistance. Note that these fixed effects make it impossible to estimate the country-specific variables; GDP and POP. Perfect collinearity, again, removes them from the equation.

(16)

13

4. Data

Making use of a panel data set consisting of 163 different countries, including the ASEAN countries and China as the effect of the ACFTA should not only be estimated only on member countries but on all countries10. Since China and most ASEAN members are global traders and are part of a global integrated trading world. The data covers the time period from 2000 until 2018. In trying to obtain the most consistent estimates as possible, several years of data before the first implementation of the Early Harvest Program are included as well. The year 2000 is chosen because of possible bias from earlier years following from the Asian crisis starting in 1997 and the following early years of recovery. The year 2018 is the most recent data that is available. In total, the data consists out of 339.609 observations.

Data for exports is taken from the IMF’s DOTS database and shown in million US$. Exports being the dependent variable, a measure of the volume of bilateral trade. It should be recognized that the DOTS dataset does not include zero trade values for data that is missing. The dataset provides only results where data is available for the years and countries that do have an observation. Data for a country’s GDP and population are obtained from the World Bank’s World Development Indicators (WDI). GDP data is measured in current US$, to decrease the chance of the use of incorrectly deflated data. GDP and Population are used as representations of market size. Data for the other control variables: distance, common colonial links after 1945, common language and contiguity are obtained from the Centre d’Etudes Prospectives et d’informations Internationales (CEPII)11_.

The two sets of ACFTA dummy variables are created by the author. The same rule of thumb holds for both sets of variables, where 2004 is the first moment the ACFTA is implemented. The ACFTA_NEWijt takes a value of 1 if both the newer ASEAN countries and China belong

to the ACFTA in year t. ACFTA_6ijt takes a value of 1 if both the ASEAN-6 and China belong

to the ACFTA in year t. ACFTA_ALLijt takes a value of 1 for all ASEAN members and China

in year t that the ACFTA is implemented. When one or more of these coefficients is positive and significant the conclusion can be drawn that the ACFTA has a positive trade creation effect and intra-regional trade levels are higher than normal trade levels.

Three more ACFTA dummy variables were created, making it possible to measure the specific trade effects of the ACFTA. In the end, a distinction can therefore be made between trade creation and trade diversion. ACFTA_ALLijt which takes a value of 1 if both countries i and j

belong to the ACFTA in year t and zero otherwise. ACFTA_EXPijt, which takes a value of 1 if

the exporter i is a member of the ACFTA in year t while destination country j is not. The final variable is ACFTA_IMPijt, which takes a value of 1 if exporter i is a non-ACFTA member in

year t while destination country j is a ACFTA member and zero otherwise.

10_{See appendix III for an overview of all included countries in the sample data.}

11_{Data coming from the Geodist CEPII dataset from:}

(17)

14

5. Results and discussion

The main results are presented in Table 2 and Table 3. These are differentiated by the two main questions in the same way as was done in the setup of the two models. Table 2 shows the results regarding the differences between the ASEAN new countries and the ASEAN-6. Table 3 shows the results regarding the trade creation and diversion effects.

5.1 ASEAN_new VS ASEAN-6

Column (1) of table 2 follows equation (3) under the pooled OLS technique. No further fixed effects are added. Though ignoring multilateral resistance terms and time-invariant observed heterogeneity make it likely that the ACFTA coefficients are biased, these results are only a first benchmark for the following models. ACFTA_NEWijt is statistically significant only at the 10%

level, ACFTA_6ijt is statistically significant at the 5% level and ACFTA_ALLijt is statistically

significant at the 1% level. The primary representations are all statistical significant, including having a positive sign, as was anticipated. In column (2), the yearly fixed effects were added. Following Wincoop (2003), country fixed effects were added to control for multilateral resistance terms. The main ACFTA variables change drastically where no direct differences can be found in the main proxies. The ACFTA4 and ACFTA6 both were found to be insignificant while the ACFTA10 variable becomes highly negative but staying statistically significant. Column (3) presents the results when including single country-pair fixed effects. Adding country-pair fixed effects make it impossible to estimate the impact of time-invariant determinants. Therefore, distance, contiguity, common language and being colonized after 1945 fall out of the model. All variables of interest become statistically insignificant here. In column (4) are both multilateral resistance terms introduced and country-pair fixed effects. The main variables of interest are again insignificant. The final results are presented in column (5), including time-varying multilateral resistance terms and country-pair fixed effects as described by Baier and Bergstrand (2007). Controlling for all possible determinants, the results in column (5) provide unbiased estimates of the ACFTA variables. Also in the final model, they were found to be insignificant.

(18)

15

Table 2 Gravity estimates dividing New ASEAN countries and ASEAN-6

Note: Robust and clustered standard errors by country-pair are used to compute the t-values. t-Values are reported

below each coefficient. θt: denotes time effects. Ii(j): country fixed effects. Iit(jt): country time-varying fixed effects.

Iij: country pair fixed effects. The estimates for the fixed effects are omitted due to space considerations.

ACFTA_NEWijt takes a value of 1 if both the newer ASEAN countries and China belong to the ACFTA.

ACFTA_6ijt takes a value of 1 if both the ASEAN-6 and China belong to the ACFTA. ACFTA_ALLijt takes a

(19)

16

5.2 Trade creation and trade diversion

Table 3 presents the results regarding the trade creation and diversion effects of the ACFTA. Again, column (1) used a pooled OLS technique where no further fixed effects were added. Here all variables were statistically significant at the 1% level. The main proxies show the expected trend with all three ACFTA variables being positive. This gives a first indication of trade creation effects of the ACFTA. In column (2) were yearly and country fixed effects added. The ACFTA_ALLijt variable becomes negative but insignificant. While the ACFTA_EXPijt and

ACFTA_IMPijt stay significant and positive. In column (3) were only country pair fixed effects

included. Compared to column (2) are there no significant changes to be found. Where only the magnitudes of the ACFTA_EXPijt and ACFTA_IMPijt became less. In column (4) were

yearly, country-pair and country fixed effects altogether included. Again there are not many differences compared to the previous column. Unbiased estimates should come from column (5) where time-varying multilateral resistance terms and country-pair fixed effects are included in the model. This equation controlled for all determinants that vary with it and jt, while also controlling for all time-invariant effects. Therefore all variables, except the ACFTA, drop out of the equation. The variable of interest ACFTA_ALLijt is significant again. However, this time

the sign is negative which contradicts intra-bloc trade creation effects. Notice that the two variables, ACFTA_EXPijt and ACFTA_IMPijt dropped out of the results. This is because of

collinearity. In the current set-up of the variable, ACFTA_EXPijt is 1 for all exports from an

ACFTA country to a non-member country. The sum of ACFTA_ALLijt + ACFTA_EXPijt will

therefore always be equal to 1 for any exporter that belongs to the ACFTA at time t. Therefore collinearity can be observed with exporter-time fixed effect. The same story is true for the import side, ACFTA_IMPijt, so both variables drop out of the model.The missing values make

it impossible to calculate the net trade creation effect.

Interestingly, the ACFTA_ALLijt variable is negative in column (5). The average effect is

22.1%12_{lower than expected from normal levels of trade. A negative effect, contrary with the}

aim of introducing the FTA.

It might still be possible that net trade creation effects are positive. Since the exporter and importer diversion effects are missing in the final model it is not directly possible to show this. However, looking at column (3) and (4), both ACFTA_EXPijt and ACFTA_IMPijt were positive

and significant. The estimates of column (4) show that there are welfare gain effects for countries outside the trade bloc (positive export diversion effect or export expansion), respectively by 24.36%. There is a positive trend in the exports from non-member countries to ACFTA countries, indicating import expansion, by 25.36%.

12_{The average effect is calculated as [exp(-0.250)-1)*100]. The effects of ACFTA_EXP and ACFTA_IMP are}

(20)

17 Net trade creation effects are calculated following table 1 and shown in table 3. The first row of the table combines the coefficients of column (4) and (5) from table 4. Since coefficients

ACFTA_EXP and ACFTA_IMP drop out, for the final model the coefficients of column (4) are

used. To show that even though the intra-bloc trade has decreased, net trade creation effects might still be positive. The second row shows the results of the final model, column (5). There are no values for ACFTA_EXP and ACFTA_IMP. The ACFTA_ALL coefficient is negative. Therefore, the overall net trade creation effect is negative, indicating that the ACFTA did not bring any welfare gains to the member countries.

Table 3 Net trade creation effect

ACFTA_ALL ACFTA_EXP ACFTA_IMP Net effect Net TC% Column 4&5 (combined) -.260*** 0.218*** 0.226*** TD + XE + ME = 0.184 20.2% Column 5 -.260*** -- -- TD = -2.60 -22.1%

(21)

18

Table 4 Gravity estimates for trade creation and trade diversion effects

Iij: country pair fixed effects. The estimates for the fixed effects are omitted due to space considerations.

ACFTA_ALLijt taking a value of 1 if both countries i and j belong to the ACFTA in year t and zero otherwise.

Following, ACFTA_EXPijt, taking a value of 1 if the exporter i is a member of the ACFTA in year t while

destination country j is not. The final variable is ACFTA_IMPijt, taking a value of 1 if exporter i is a non-ACFTA

member in year t while destination country j is a ACFTA member and zero otherwise The symbols *,** and *** denote statistical significance at the 10%, 5% and 1% levels respectively.

(22)

19

5.3 Discussion ASEAN_new vs ASEAN-6

The insignificant results of the ACFTA variables make it impossible to state final conclusions about possible different effects of the same FTA for the two groups of ASEAN countries. This was confirmed by table 4 where the robustness check was done using the PPML model. A possible explanation for this, as described by Yang and Martinez-Zarzoso (2014), might be due to the fact that different results by models that try to estimate the effects of FTAs on trade flows depend highly on how researchers control for unobserved country heterogeneity. This implies that the results are highly reliant on the correct model specifications. In Head and Mayer (2014), it was shown how many different possibilities and models there are available for using the gravity equation. It is possibly that not enough tests were done to determine whether or not this was the best fitting model out there to answer the main question of the paper.

Another explanation might be that there are simply no direct differences between the two groups of ASEAN countries. The differences between the ASEAN countries might be too small to be found. One reason for this might be that the time period covered in this dataset, for which the ACFTA is at play, is more than 14 years. It is not unreasonable to think that the countries integrated their trading system more and more over time, making differences between them smaller. The literature has suggested that not all findings favoured trade creation due to the ACFTA in the first place [see Guilhot (2010)]., where differences per country could be expected as well [see Dianniar (2013); and Ferrianta et al. (2012)]. Categorizing the two groups of ASEAN countries might not be sufficient enough to expose differences between them.

5.4 Discussion trade creation and trade diversion

Column (5) of table 4 shows only a coefficient for the ACFTA_ALLijt variable. As explained

earlier, the other two ACFTA variables of interest dropped out of the equation. ACFTA_EXPijt,

is 1 for all exports from an ACFTA country to a non-member country. The sum of

ACFTA_ALLijt + ACFTA_EXPijt will therefore always be equal to 1 for any exporter that

belongs to the ACFTA at time t. The fixed effects control for the total change in exports for a specific country in a given year. Multicollinearity exists. Therefore, is it not possible to determine the net trade creation effects using the final model. This problem can be avoided by, for example, taking the estimates in column (4) as final. However, the possible bias in estimated coefficients in column (4) for not including exporter and importer yearly fixed effects cannot be justified. The unobservable changes in country size and multilateral resistance terms that could be correlated with a countries belonging to an FTA should be controlled for by these fixed effects.

(23)

20 - FTAijt being 1 for pairs ij that are in the same FTA in period t and 0 otherwise, excluding the

ACFTA and ACFTA member countries other FTAs. This way is controlled for all other FTAs actively in the world.

- ACFTA_creationijt being 1 for pairs ij that are in ACFTA in period t and 0 otherwise. ACFTA

country-pairs are here being taken into account.

- ACFTA_diversionijt being 1 for pairs ij that have an FTA with an ACFTA-partner in period t

and 0 otherwise. This variable take pairs of non-ACFTA countries and ACFTA countries into account.

(24)

21

6. Robustness sensitivity check

As discussed previously, a frequently encountered issue while using a gravity model is the presence of zero trade flows. The high frequency of zeros normally requires an adjustment of the gravity model being used to find the best estimates. Several methods that deal with zero trade flaws have been proposed in the literature. The PPML model proposed by Santos Silva and Tenreyo (2006) is an often used model to encounter this phenomena. Although the IMF DOTS database did not include the years and countries for which bilateral trade data is missing, the PPML model can still be used as a robustness check. In Santos Silva and Tenreyo (2006), it was shown that even though omitting the zero values is not ideal, only small consequences are found in the paper. The most important factor to use the PPML is that it is less affected by heteroscedasticity and allows for more flexibility in the error term. The model has gained popularity in the literature in the last few years.

Table 7 and 8 can be found in appendix II, presenting the results of the PPML model. The dependent variable, exports, is not written as a logarithm in the PPML model, based on the idea that taking the natural logarithm of the value of trade flows may lead to biased estimates in the presence of heteroscedasticity. Notice that in both tables the population variable becomes insignificant in almost all the performed models. There is no direct explanation for why this is happening.

Table 7 shows the results with the ACFTA_NEW and ACFTA_6 variables being of a primary interest. In the final model, column (5), the ACFTA_6 has become statistically significant at the 5 percent level and the coefficient is positive. Indicating that trade levels have been higher than normal for the ACFTA_6 countries. The other two variables of interest are still insignificant. Therefore, the conclusion stays the same. There is no evidence of different ACFTA effects between the two ASEAN member country groups.

The PPML model looking at the trade creation and diversion effects is presented in table 8, where there are no real differences for the ACFTA_EXP and ACFTA_IMP variables to be found. However, as can be seen in column (5),insignificant results for the ACFTA_ALL variable were found in the final model. This is contrary to the main findings in table 4.

To check the adequacy of the estimated model, the Ramsey heteroscedasticity-robust RESET test was performed, following Silva and Tenreyro (2006). The test indicates signs of misspecification. The RESET test was done solely for the trade creation and diversion model. Using the model specifications of column (1) and (4) as in table 8, where the RESET test cannot be rejected for the model specification as in column (1), finding a p-value of 0.000. The result was rejected in column (4) where the p-value of the test is 0.156. This indicates that after the inclusion of the fixed effects found in column (4), the model shows no signs of misspecification. It is expected that the same result will hold for the estimates in column (5) under the right variable specifications.

(25)

22

7. Conclusion

The aim of this research was twofold. First, is recognized the fact that it is possible to distinguish two groups of ASEAN member countries, the newer ASEAN member countries and the ASEAN-6. For whom a clear difference in the timeline of the reduction and elimination of tariff lines can be found. Raising the question whether or not this could result in different effects following from the implementation of the ACFTA between them. Secondly, this paper analyzed the impact of the ACFTA on export flows focusing on the trade creation and diversion effects. Using the gravity model with an aggregated dataset of export flows for 163 countries covering the period dating from 2000 to 2018 to obtain unbiased and consistent estimates. Time-varying multilateral resistance terms and country-pair fixed effects were used to deal with endogeneity bias following from omitted variables.

There was no direct evidence found of different ACFTA effects for the two groups of ASEAN member countries. A possible explanation might be that the ACFTA does not affect the member countries at all or differences decreased over time and are minimal. However, a definite answer to the fact why there are no different effects within the same group of ACFTA member countries is outside the scope of this paper. For now, the conclusion is that there are no statistically significant differences.

Column (5) of table 4 shows that there is a significant negative effect of the ACFTA_ALLijt .

Indicating that due to the implementation of the ACFTA, contrary to the goal of a FTA, trade levels between intra-bloc member countries are lower than normal. Nevertheless, positive net trade creation effects can still exist. However, in the current model specification it is not possible to determine the net trade creation effect of the ACFTA. The two additional variables of interest, ACFTA_EXPijt and ACFTA_IMPijt, drop out in the final model. This because of

perfect collinearity. The ACFTA_ALLijt variable becomes statistically insignificant, as shown

(26)

23

8. Limitations

This paper is not flawless and comes with its limitations. Firstly, in this research only aggregated trade data is used. As can be seen in the timeline of the reductions and elimination of tariffs, the removal is often not only not equal between countries but also between sectors. Different impacts on different sectors could be expected. Using disaggregate trade data would give a better insight into whether or not the trade effects in regions differ by commodity. However, using aggregated trade data gives a first overall insight into the effects of the ACFTA. The results are important for policy implications for the participating countries.

Secondly, there are no zero trade values included in the dataset. The IMF DOTS database automatically excludes zero value data. Due to time limitations, it was not possible to include all the years for the countries for which no trade data was available. Although including zero trade flows comes with its own shortcomings, the literature acknowledge this problem and showed how to properly handle it. Having a more complete dataset could affect the estimated outcomes which also help in comparing results with other ACFTA research.

A third flaw is the fact that in the current setup of the primary variables ACFTA_EXPijt and

ACFTA_IMPijt, drop out of the model while including time-varying exporter and importer fixed

effects. This is due to perfect collinearity. Anderson and van Wincoop (2003) introduce the concept of multilateral resistance terms, indicating that the costs of bilateral trade between two countries are not only affected by bilateral trade costs but also by the relative weight of trade costs in comparison to trading partner countries in the rest of the world. Baier and Bergstrand (2007) argue that multilateral resistance terms should not only be country specific but also be time-varying in a panel setting. To find unbiased estimates and to determine the net trade creation and trade diversion effects, importer and exporter effects should be time-varying. The

ACFTA_EXPijt and ACFTA_IMPijt should therefore be setup differently. A first proposal is

(27)

24

9. Additional analysis

In this chapter, there is a final attempt to determine the trade creation and trade diversion effects following from the introduction of the ACFTA13_{. This is done with the variable FTA_world}

ijt

that captures the effect of FTAs between countries worldwide. The main variables of interest would be as follows:

- FTA_worldijt being 1 for pairs ij that are in the same FTA in period t and 0 otherwise.

Excluding the ACFTA and ACFTA member countries’ other FTAs. This controls for all other FTAs active in the world. Additional data for FTAs active in the world by country pair comes from Kohl (2014). This dataset covers the years from 2000 to 2013. Notice that therefore the used panel data here covers 5 years less than the panel data used in the main text.

- ACFTA_ALLijt being 1 for pairs ij that are in ACFTA in period t and 0 otherwise. ACFTA

country-pairs are here taken into account. This variable does not vary with the variable used in the main text.

- ACFTA_diversionijt being 1 for pairs ij that have an FTA with an ACFTA-partner in period t

and 0 otherwise. This variable takes pairs of non-ACFTA countries and ACFTA countries into account.

Table 5 shows the results, distinguishing the fixed effects and PPML models and presenting two columns per model. In column (1) and (3) are time, country and country pair fixed effects used. Whereas in column (2) and (4) time-varying country and country pair fixed effects are used. Clear differences in results can be found between the fixed effects and the PPML models. In the fixed effect model column (2), using time-varying multilateral resistance terms, all three main variables of interest are positive at the 10% significance level. Where the ACFTA_ALLijt

and FTA_worldijt are also both significant at the 1% level. In the PPML model, looking at both

column (3) and (4), only one variable is found to be significant. Significant at the 5% level. In column (2) are both ACFTA_ALLijt and ACFTA_divijt were found to be negative., indicating

that the ACFTA had not only a negative effect on intra-bloc trade levels, but also on trade levels between intra-bloc and extra-bloc countries. The positive and significant FTA_worldijt variable

shows that FTAs worldwide however do promote trade levels between countries in general. Interestingly, there are clear differences in results while using the two different models and included fixed effects. Column (4) indicates that neither the ACFTA or FTAs in use worldwide have any effect on trade levels between countries.

Making hard claims about the ACFTA is therefore difficult. This additional analysis together with the results found in the main text of this paper show the importance of the right model specification in the FTA and gravity literature. Therefore, readers should be critical of findings in general and in the way they are presented.

13_{While conducting this research I could not find a direct solution in the literature for the perfect collinearity}

(28)

25 Table 5 adjust later

Iij: country pair fixed effects. The estimates for the fixed effects are omitted due to space considerations. ACFTA.

ACFTA_ALLijt takes a value of 1 for all ASEAN members and China from the moment the ACFTA is

(29)

26

References

Abedini, J. & Peridy, N. 2008. The Great Arab Free Trade Area(GAFTA): an Estimation of Its Trade Effects. Journal of economic integration. Vol. 23(4), pp. 848-872.

Anderson, J.E. 1979. A Theoretical Foundation for the Gravity Equation. The American

Economic Review, Vol. 69(1), pp. 106-116.

Anderson, J.E. 2010. The Gravity Model. NBER Working Paper No. w16576.

Anderson, J.E. & Van Wincoop, E. 2003. Gravity with gravitas: A solution to the border puzzle.

The American Economic Review, Vol. 93(1), pp. 170-192.

Alleyne, A., Zhang, Z. & Mu, Y. 2020. Sustaining International Trade with China: Does ACFTA Improve ASEAN export efficiency? Sustainability 2020, Vol. 12(15).

Baier, S.L. & Bergstrand, J.H. 2007. Do free trade agreements increase members’ international trade? Journal of International Economic. Vol. 71, pp. 72-95.

Baier, S.L. & Bergstrand, J.H. 2009. Bonus vetus OLS: A simple method for approximating international trade-cost effects using he gravity equation.

Baldwin, R. & Taglioni, D. 2006. Gravity for dummies and dummies for gravity equations.

National Bureau of Economic Research, Cambridge. Working paper 12516.

Bertrand, M., Duflo, E., & Mullainathan, S. 2004. How much should we trust Differences-In-Differences estimates? The Quarterly Journal of Economics. Vol. 119(1), pp. 249-275.

Binh, D. T. T., Duong, N. V., & Cuong, H. M. 2011. Applying gravity model to analyze trade activities of Vietnam. Forum for Research in Empirical International Trade Working Paper Carrère, C. 2006. Revisiting the effects of regional trade agreements on trade flows with proper specification of the gravity model. European Economic Review. Vol.50, pp. 223-247.

Chaney, T. 2015. The Gravity Equation in International Trade: An Explanation. NBER Working

paper. No. 19285.

Chirathivat, S. 2002. ASEAN-China Free Trade Area: background, implications and future development. Journal of Asian Economics, Vol. 13(5), pp. 671-686.

Clausing, K.A. 2001. Trade creation and trade diversion in the Canada – United States Free Trade Agreement. Canadian Journal of Economics. Vol. 34(3), pp. 677-696.

Dianmiar, U. 2013. The impact of Free Trade Agreements on Indonesia’s Agricultural trade flows: An application of the gravity model approach. Unpublished master thesis.

Ferrianta,Y., Hanani,N., Setiawan,B. & Muhaimin, W. 2012. Impact of Trade Liberalization Asean-China Free Trade Area (ACFTA) on The Performance of Indonesia Maize Economy.

Journal of Basic and Applied Scientific Research, Vol. 2(7), pp. 6801–6809.

Gould, D.M. 1998. Has NAFTA changed North American Trade? Economic and Financial

(30)

27 Guilhot, L. (2010) Assessing the impact of the main East-Asian free trade agreements using a gravity model. First results. Economics Bulletin, Vol. 1, pp. 282-291.

Head, K. & Mayer, T. 2014. Gravity Equations: Workhorse, toolkit, and cookbook. In: Gopinath G, Helpman E, Rogoff K. (ed)., Handbook of international economics. Vol. 4. Elsevier; 2014. p. 131–195.

Ibrahim, I., Permata, M. I. & Wibowo, W. A. 2010. The impact of ACFTA implementation on international trade of Indonesia. Buletin Ekonomi Moneter Dan perbankan, Vol. 13(1), pp. 23-56.

Indriyan, I. 2016. The effect of ASEAN-China Free Trade Area (ACFTA) on Indonesia Export.

Etikonomi, Vol. 15(2).

Kepaptsoglou, K. Karlatis, M. & Tsamboulas, D. 2010. The Gravity Model Specification for Modeling International Trade Flows and Free Trade Agreement Effects: A 10-Year Review Of Emperical Studies. The Open Economic Journal. Vol 3, pp. 1-13.

Kohl, T., Brakman, S. & Garretsen, J. 2016. Do Trade Agreements Stimulate International Trade Dfiferently? Evidence from 296 Trade Agreements. The World Economy, Vol. 39(1), pp. 97-131.

Lee, H. & v.d. Mensbrugghe, D. 2007. Regional Integration, Sectoral Adjustments, and Natural Groupings in East Asia. Discussion Paper. 07E008.Osaka: Osaka School of International Public Policy, Osaka University.

Martinez-Zarzoso, I. & Nowak-Lehmann, D.F. 2003. A Gravity Model: An Empirical Application To Mercosur-European Union Trade Flows. Journal Applied Economics. Vol. 6(2), pp. 291-316.

Martinez-Zarzoso, I., Nowak-Lehmann, D.F. & Horsewood, N. 2009 Are region trading agreements beneficial? Static and dynamic panel gravity models. North American Journal of

Economics and Finance. Vol.20(1), pp. 46-65.

Moreira, M,M. (2018) Connecting the Dots: A Road Map for a Better Integration of Latin America and the Caribbean. Inter-American Development bank.

Park, D., Park, I. & Estrada, G.E.B. 2009. Prospects for ASEAN–China Free Trade Area: A Qualitative and Quantitative Analysis. China & World Economy, Vol. 17(4), pp. 104–120. Peridy, N. 2005. The trade effects of the Euro-Mediterranean partnership: what are the lessons for ASEAN countries? Journal of Asian Economics. Vol. 16(1), pp. 125-139.

Peridy, N. 2005. Toward a Pan-Arab free trade area: assessing trade potential effects of the Agadir Agreement. The Developing Economies. Vol. 43(3), pp. 329-345.

Piermartini, R & Teh, R. 2005. Demystifying Modelling Methods for Trade Policy. WTO

Discussion paper. No.10.

(31)

28 Qiu, H., Yang, J., Huang, J. & Chen, R. 2007. Impact of China-ASEAN Free Trade Area on China’s International Agricultural Trade and Its Regional Development. China & World

Economy, Vol 15(4), pp.77 – 90.

Rose, A. 2004. Do we really know that the WTO increases trade? American Economic

Review. Vol. 94, pp 98-114.

Santos-Silva, J.M.C. & Tenreyro, S. 2006. The log of gravity. The Review of Economics and

Statistics. Vol.88, pp. 641-658.

Sheng, Yu, Tang, H.C. & Xu, X. 2012. The impact of ACFTA on People's Republic of China–ASEAN trade: Estimates based on an extended gravity model for component trade. ADB Working Paper Series on Regional Economic Integration, No. 99.

Shepherd, B. 2016. The Gravity Model of International Trade: A User Guide (an updated version). United Nations Publication.

Taguchi, H. & Lee, H. 2016. ASEAN-plus-one Free Trade Agreements and their trade effects.

MPRA Paper 72503.

Tinbergen, J. 1962. Shaping the World Economy; Suggestions for an International Economic Policy. Books (Jan Tinbergen). Twentieth Century Fund, New York.

Tristan, K. 2014. Do we really know that trade agreements increase trade? Review of Wold

Economics. Vol. 150(3), pp 443-469.

Wang, C.Z .2018. China-ASEAN Free Trade Area Development Status Quo and Trade Effect Analysis. Modern Economy, Vol. 9, pp. 719-733.

Viner, J. 1950. Customs union theory. Carnegie Endowment for International Peace, New York. Yang, J. & Chen, C. 2008. Agricultural trade between China and ASEAN: Dynamics and prospects. Book: Agriculture and Food security in China: What effect WTFO accession and

Regional Trade Arrangement? ANU Press. pp. 346-371.

Yang, S. & Martinez-Zarzoso, I. 2014. A panel data analysis of trade creation and trade diversion effects: The case of ASEAN-China Free Trade Area. China economic review, Vol. 29, pp. 138-151.

Zhang, L., Cui, L., Li, S. & Lu, J. 2017. Who rides the tide of regionalization: Examining the effect of the China ASEAN Free Trade Area on the exports of Chinese firms. International

(32)

29

Appendix I Background information ACFTA

Table 6 Summary of key dates ACFTA

(33)

30

Appendix II Robustness test PPML Estimates

Table 7 Gravity estimates using the PPML model dividing new ASEAN countries and ASEAN-6

Iij: country pair fixed effects. The dependent variable here is exports in its natural form. The estimates for the fixed

effects are omitted due to space considerations. ACFTA_NEWijt takes a value of 1 if both the newer ASEAN

countries and China belong to the ACFTA. ACFTA_6ijt takes a value of 1 if both the ASEAN-6 and China belong

to the ACFTA. ACFTA_ALLijt takes a value of 1 for all ASEAN members and China from the moment the ACFTA

(34)

31

Table 8 Gravity estimates using the PPML model for trade creation and trade diversion effects

Iij: country pair fixed effects. The dependent variable is here exports in its natural form. The estimates for the fixed

effects are omitted due to space considerations. ACFTA_ALLijt taking a value of 1 if both countries i and j belong

to the ACFTA in year t and zero otherwise. Following, ACFTA_EXPijt, taking a value of 1 if the exporter i is a

member of the ACFTA in year t while destination country j is not. The final variable is ACFTA_IMPijt, taking a

A Deeper Insight into the Effects of the ASEAN-China Free Trade Agreement: Mutual Differences, Trade Creation and Trade Diversion.

A Deeper Insight into the Effects of the ASEAN-China Free Trade

Agreement: Mutual Differences, Trade Creation and Trade Diversion.

Table of contents

= α