INTRODUCTION
Background
The Association of Southeast Asian Nations (ASEAN) was established in 1967 with 5 original members, namely Indonesia, Malaysia, the Philippines, Singapore and Thailand
Currently, the association consists of 10 member states, as successively Brunei Darussalam, Vietnam, Laos, Myanmar (Burma) and Cambodia In October 2003, ASEAN leaders decided to create a single market for a huge economy of US$722 billion and nearly 550 million people (2003) The building of the ASEAN Economic Community (AEC) by the end of 2015 is a major milestone of integration effort attempted in the developing world, offering variety of opportunities for each member states to attract inward Foreign Direct Investment (FDI) and increase trade flows AEC has been considered as successor following the completion of the ASEAN Free Trade Area (AFTA), which was signed in 1992 in Singapore by 6 original ASEAN countries
At that time, the AFTA agreement aimed to eliminate the tariff on a wide range of products between ASEAN members
The establishment of AFTA has attracted the remaining members, Vietnam has been joined since 1995, Laos and Myanmar (1997), Cambodia (1999) Under the Common Effective Preferential Tariff (CEPT) Scheme, the products are firstly classified into two different groups depending on the willingness of each member - Inclusion List (IL) and Exclusion List (EL) Those product under IL were subject to removal of tariff to between 0% and 5% by 2008 for original AFTA members Later on, the tariff reduction schedule was revised several times Finally, the end year of IL was moved to 2002 The EL was later devided into a Temporary Exclusion List (TL) and Sensitive List (SL) Those products in TL would be shifted to IL in the future Whereas, additional time for both
IL and TL were given to the newer member countries, namely Cambodia, Laos, Myanmar and Vietnam (CLMV) to implement their private routine of tariff reduction for products which originate from within AFTA members In addition, non-tariff measures (NTMs) or non-tariff barriers (NTBs) were excluded under the CEPT scheme
However, since 2010 the ASEAN Trade in Goods Agreement (ATIGA) has come into effect to enhance and supersede the CEPT/AFTA with greater goal ambiguity covering both tariff and non-tariff elimination
The creation of AFTA (or ATIGA) has generated a bigger internal market that each member is able to entry into lower trade cost market and attract more investment into their economy that allow producing on a large scale
1.1.2 ASEAN economics and trade overview
Nowadays, the ASEAN region has become well known in the world due to its high real Gross Domestic Product (GDP) growth in a long period of time
Figure 1.1: Real GDP growth (annual %) of ASEAN region, 1980-2020
Firgue 1.1 illustrates the real GDP growth rate of the ASEAN region compared to the group of advanced economies and the average across the whole world from 1980 to
2020 (estimated data) ASEAN’s real GDP growth have increased remarkably, except during two crisis periods: the Asian financial crisis in 1998 and the Global crisis in 2008
The average growth rate of GDP in the whole period 1980-2018 of three groups ASEAN, advanced economies and the average of the world is 5.3%, 2.2% and 3.3% respectively
Followed by that data, the average value of this important indicator reveals that ASEAN is one of the fastest growing economies in the world
On the other hand, trade to GDP ratios measure the relative importance of international trade in each ASEAN economy Table 1.1 below shows the trade to GDP ratios or openness indicator of 10 ASEAN countries In general, almost ASEAN countries open its economy to the world, partially due to its economic integration efforts The high trade openness which trade to GDP ratios is greater than 100 percent bringing many economic benefits, from increasing technology transfer, supplying more job, enhancing the total factor productivity for the economic development It seems that the smaller economies in land area and GDP than trade more and vice versa That can be seen from the table 1.1, for example Singapore is the most dynamic economy in ASEAN in comparison with Indonesia which is the largest country in both land area and GDP in ASEAN region
Table 1.1: Share of total trade on GDP (%) of ASEAN countries, 1993-2018
Source: Author’s calculations based on WITS country profile data
From the calculation formula of GDP by expenditure approach, the rapid of expansion in GDP of ASEAN region has been definitely contributed by export-import component 1 The entire value of trade in goods and services in this region has been playing an important role in GDP measurement under expenditure approach
By looking at the figure 1.2, the total merchandise trade of ASEAN region has sharply increased from 2000 to 2017 equivalent to an increase of nearly 3.5 times (ASEAN Secretariat, 2018)
1 The formula for GDP by expenditure approach: GDP = C + I + G + (X-M)
Figure 1.2: Total value of exports and imports of goods and services in ASEAN,
According to report of ASEAN Secretariat (2014) in the special edition “the removal of trade barriers has been centered on the removal of intra-ASEAN tariffs through commitment CEPT under the AFTA” Although, figures in trade indicates that “ASEAN depended a great deal on the consumption and production needs of the rest of the world”
(Mahfuz Kabir et al., 2014), but intra-ASEAN trade still has been increased dramatically, even though faster than overall ASEAN trade and extra-ASEAN trade From 1993 to
2013, the intra-ASEAN trade has surged by more than 7 times, while extra-ASEAN trade has increased more than 5 times
Figure 1.3: Trend of ASEAN total trade, extra-ASEAN trade and intra-ASEAN trade,
Regarding to a very fast increasing in intra-regional trade and the success of trade surplus in ASEAN region, from 2000 to 2017 intra-ASEAN collectively is the largest market
Figure 1.4: Share of export values of goods by intra-ASEAN and its major trading partners (%), 2000-2017
Overall, there is a significant gap in trade by export and import value among 10 ASEAN countries Singapore is the largest exporter, followed by Thailand, Malaysia, Viet Nam, Indonesia and Philippines Meanwhile, Brunei Darussalam, Laos, Cambodia and Myanmar are among trade less countries
Figure 1.5: Shares of export and import value of goods by ASEAN country, 2000-
Specially, according to ASEAN Community in Figure reported in 2014, ASEAN records a surplus in its trade of agro-based products, rubber-based products, wood-based products, fisheries products, textile products, electronic products and automotive products with its major trading partners, although the surplus was relatively small, in general According to the report, in the case of automotive products, including both final and intermediate products, by 2013 ASEAN achieved a small surplus with Australia and New Zealand, and USA Meanwhile, trade deficit was large with Japan, small in the case of China, Republic of Korea, and EU-28
Figure 1.6: Trend of intra- and extra-ASEAN export of priority integration sector of automotive products, 1993-2013
Regarding to the trade value of automotive industry in ASEAN, this region has remained a net importer despite its export value is closely catching up with import value Remarkably, the value of extra-ASEAN export in the case of automotive was 2 times that of intra- ASEAN export compared to other industries those was 7 to 15 times (ASEAN Secretariat,
2014) This figures imply that intra-regional trade flows of automobile are increasingly significant important However, ASEAN still import mainly from neighboring ASEAN countries such as China, Japan, South Korea, etc
Further, according to the ASEAN Investment Report (ASEAN Secretariat, 2020), steady economic growth, increasing in middle-class consumer, deepening regional integration ASEAN has attracted more the Multinational Enterprises (MNEs) increasing their value chain participation in automotive industry
Therefore, AFTA was expected to accelerate easier obtaining and more efficient intra- regional and intra-industry trading, creating an environment that is advantageous to achieve economies of scale in automobiles and its components
Table 1.2: ASEAN-5 motor vehicle production (in units), 2006-2018
Note: Data for other countries are not available
According to figure collected from the ASEAN Automotive Federation, five largest economies in producing automobile are Indonesia, Malaysia, Philippines, Thailand, and Viet Nam produced totally over 44 million units of motor vehicles from 2006 to 2018
Definition
Intra-regional trade defined in this study as the average of intra-AFTA exports and imports, which focuses on trade exchange between ASEAN member countries
This study uses the definition of the automobile industry in global value chain (GVC) data The automobile industry is broadly defined as an industry of motorized vehicles consisting of four wheels and powered by internal engines, while narrowly defined as the car industry or motor vehicle industry (Dowlah, 2018) In general, the automobile industry is understood as covering a wide range of motor vehicles, including automobiles (cars), buses, motorcycles, off-road vehicles, trucks However, this study focuses on the car industry and its parts and components industry, therefore the narrow definition is taken into account The UN Comtrade Database classifies its traded products according to the Harmonized Commodity Description and Coding Systems, or
Harmonized System (HS) The automobile industry’s imports and exports are classified mainly under sub-headers at HS-4 digit level and HS-6 digit level 3
Trade creation and Trade diversion
Trade creation and trade diversion is an economic term as Jacob Viner (1950) firstly pointed out that related to economic efficiency and welfare-effect definitions However, in this study trade creation and trade diversion are different The study follows the definitions that were firstly used by Endoh (1999) and applied by Sattayanuwat and Nantarat (2017) and Wong, C K et al (2017): (1) Trade creation refers to trade more with non-AFTA member; (2) Trade diversion refers to trade less with non-AFTA member.
Research objectives and research questions
In regard to notable achievements in trade liberalization under AFTA and ATIGA that tariff and non-tariff are eliminated, which have officially been in effect since 1993, there was a few existing empirical studies conducting research on the impacts of AFTA on intra-regional trade flows at aggregated data level (Elliott and Ikemoto, 2004; Indira M
Hapsari and Carlos Mangunsong, 2006; Trung Kien, 2009; Ismail and Wong, C K.,
2013) Thus, one of disadvantages in using pooled data to investigate the effects of free trade policy such as AFTA is not enable to reveal the detailed changes in trade volume of a specific industry Therefore, it is necessary to apply the gravity model in sector analysis, especially in the case of the automotive industry in the context of ASEAN countries’ clear target of becoming industrialized
Regarding disaggregated level, only Pakasa Bary (2015) and Wong, C K et al (2017) used trade data in manufacturing for their analysis On the other hand, Herath et al
(2014) implemented a thorough investigation on the ASEAN’s eight agrifood commodities by applying the gravity model Meanwhile, Okabe and Urata (2014) conducted a large-scale examination on almost all ASEAN industries focusing on investigating the trade creation effect due to tariff reduction Those studies have focused on the trade creation and trade diversion effects of AFTA without, however, using an appropriate estimator as explained later On the other hand, as mentioned above the
ASEAN’s automobile industry has increasingly affirmed its position in the world and contributed significantly to the intra-regional trade value
Hence, the purpose of this study is to empirically investigate the effects of ASEAN Free Trade Area (AFTA) on intra-regional trade flows of automotive industry for the period of 1993-2018 Moreover, in the context of “openness” of ASEAN countries, the study also considers the effect of AFTA on non-members In addition, compared with previous studies, this study is the only one applying the gravity model and the recent appropriate techniques to avoid biased estimates in the case of automotive industry in the context of growing strong regional integrations in ASEAN region
Based on the main research objective, the principal research questions that the study attempts to answer are:
1) Whether AFTA has trade creation and trade diversion effects in the case of automotive industry?
2) Are there any difference in the elasticity of tariff reduction between original and new AFTA members?
Hypotheses
Based on the estimated results from existing papers, AFTA has significantly enhanced the trade flows between members since 1993 Almost the studies showed a trade creation effect among them For the case of automotive industry, the study initially believes that AFTA also has resulted a trade creation effect among AFTA members With regard to the increasing trend in trade volumes in ASEAN region, a decreasing export flows would be occurred from ASEAN automotive exporter to non-ASEAN importer in the world to take the advantage of the removal of tariff and non-tariff barriers inside the region In addition, the significant increasing trend of FDI into ASEAN in recent years contributed to the export trade value of ASEAN countries and domestic market as well
The trade surplus recorded in the figure 1.2 suppose that the overall export trade value is greater than import trade value in a long period of time That means ASEAN countries have gradually less depended on import from outside On the other hand, the level of economic development and the automotive industry are very different between two groups of countries – original and new AFTA members – as discussed above (except for
Brunei), so the impact of tariff reduction on automotive trade flows could be different in these two groups
Therefore, the main hypothesis of this study for AFTA dummies as follows:
1) AFTA has resulted trade creation effects among members, trade diversion effect on export flows with non-ASEAN countries and trade creation effect on import flows with non-ASEAN countries
2) The elasticity of tariff elimination on automotive trade flows is significantly different between original and new AFTA members.
Method and methodology
In order to see the impacts of free trade policies on the volume and direction of trade flows between members after the formation of regional trade institutions, a variety of quantitative methods have used Among those methods, the gravity model is a workhorse of international trade analysis (WTO, 2016) Specially, according to Anderson, James E (2011) “gravity model has long been one of the most successful empirical models in economics” and was officially presented in the textbook
“International Economics: Theory and Policy” (Krugman, P R., Melitz, Jacques and Obstfeld, Maurice, 2017) By using this model, main determinants directly affect trade flows as well as the impacts of membership in other multilateral and bilateral trade agreements can be evaluated by using dummy variables
Gravity model has firstly been introduced by Tinbergen, J (1962) under the book named
“Shaping the world economy: Suggestions for an international economic policy” in trying to understand the trade pattern at the international level Tinbergen proposed the
“gravity equation”, which is derived from Newton's theory of gravitation with belief that if planets according to Newton are attracted to each other in proportion to their sizes and proximity, then countries too The trade between any two countries also be mutually attracted in percentage of their economic sizes (proxy as their GDPs) and in proximity (represented as geographical distance of two countries)
The extraordinary power of gravity equation in explaining bilateral trade flows is to consider the significance of the size of an economy that previous influential models in trade such as Ricardian model 4 and H-O model 5 were not regarded
In general, the study firstly conducts a regression analyses by employing a gravity model to bilateral trade flows of automotive industry in both way export flows and import flows in order to examine the impact of joining AFTA AFTA dummies as proxies for trade creation and trade diversion effects are mainly estimated Secondly, the study attempts to find out whether tariff elimination under the CEPT Scheme is a main factor of trade creation effects among AFTA members Finally, whether there are significant differences between the elasticities of tariff reduction by individual occurred among ASEAN member countries in the period 1993-2018
To undertake the above analyses, the study uses econometrics approach by panel data for estimating parameters in gravity equations In addition, the strong and modern estimators which are Poisson pseudo-maximum likelihood (PPML) high-dimensional fixed effect (HDFE) and Regression HDFE models to be requested Among empirical studies mentioned above, no any existing papers use the same approach
The remainder of this study is structured as follows Chapter 2 briefly reviews the theoretical framework of gravity model and relevant empirical literatures on impacts of AFTA on intra-regional trade flows Chapter 3 a thorough applying the gravity equation in regression analysis after detailed discussing on estimation methodology, common issues of gravity equations and appropriate estimation methods Chapter 4 presents the empirical findings in three case analyses AFTA dummies, the average tariff reduction in 10 ASEAN members and a separable analysis by individual in tariff elimination
Chapter 5 yields a conclusion based on estimated results and discusses on the issues of policy implications, limitations and further research
4 Ricardian model has been developed by David Ricardo
5 H-O model developed by Eli Heckscher and Bertil Ohlin.
LITERATURE REVIEW
Theoretical framework of gravity model
The gravity model introduced in the early 1960s by Tinbergen, J (1962) which is a powerful tool for the estimation of trade flows between countries The standard form of it as follows:
Where, 𝑀 𝑖𝑗 denotes bilateral export or import flows from countries 𝑖 to 𝑗 depend only on the gross national income of the exporter 𝑌 𝑖 and importer 𝑌 𝑗 , vice versa and the geographical distance between two countries 𝐷 𝑖𝑗
Discussing on the development of theoretical foundations of the gravity model, Anderson (1979) was the first economist providing a theoretical framework for the gravity model basing on the Armington’s assumption of constant elasticity of substitution (CES) preferences 6 that goods are differentiated by region of origin 7 His gravity equation derived from the properties of expenditure system Anderson’s article used the hypothesis of identical homothetic preferences across regions His study has either advantages or disadvantages in literature that can or cannot to apply for this research Anderson (1979) was not considering the effects of trade taxes across countries, and according him trade cost depends only in distance In the section of conclusion, he suggested that his methodology approach can be use efficiently at the nations where the structure of tradable goods preference is very comparable, and subsidiary, where the structure of trade tax and transport cost structures are similar
Bergstrand (1985) continued to supported the Armington’s assumption and extended Anderson (1979)’s theoretical foundation but more detailed explanation of the supply side including price in equation that according him previous studies did exclude the price and not explain the multiplicative function form However, Bergstrand (1990) provided a foundation with new assumption that goods are differentiated by producers not by country of origin based on monopolistic competition assumption In addition,
6 CES utility functions are a special case of homothetic preferences
Helpman (1987) asserted that products are distinguished by firms, not only by country
His foundation relying on the assumption of increasing returns to scale and firms were monopolistically competitive
Furthermore, Deardorff (1998) applied both CES structure and monopolistic competition (H-O structure) for specialization Melitz (2003) developed a model also based on the assumption of heterogeneous firms to analyze the effects of the intra- industry on international trade regarding their exporting behavior Thereby, he gave the first theoretical framework for the presence of zero in data of trade flows due to the existence of export market entry cost
Evenett and Wolfgang Keller (2002) examined the empirical success applied the most important theories in terms trade, the H-O theory and the theory of increasing returns to scale Their estimates suggested that increasing returns and factor endowments and can explain for the difference in components of the international variation trade patterns
On the other hand, Anderson and Van Wincoop (2003) provided a method which allows to estimate consistently and efficiently a theoretical gravity equation and calculates correctly the comparative statics of trade fictions by using the estimated general- equilibrium gravity model The model also extended to discuss on “multilateral resistance” problem that related to trade costs
Continuously, Anderson and Van Wincoop (2004) built a strong theoretical system for trade costs function from their previous study They extended the existing gravity equations for better treatment of aggregation and endogeneity problems, better estimates of substitution elasticities to improve understanding of trade costs in gravity model
Anderson and Van Wincoop focused on analyzing the impacts of trade cost through: 1) policy barriers (tariff, nontariff barriers – NTM); 2) transport costs; 3) whole sale and retail distribution costs
Later on, Helpman et al (2008) developed a theoretical foundation that predicts the presence of zeros in data as well as positive value Then, they generalized the empirical gravity equation, which could be called HMR model by two-stage procedures that accounts for the self-selection of producers into export markets and their effect on trade flows The first stage consists of using a Probit equation which estimates the extent of firm’s entry by a function of observable variables In the second stage, a gravity equation in log-linear for that at the same time correct for two kinds of biases including a sample selection bias and bias due to asymmetries in trade between countries from the issue of omitting variables
Finally, Novy (2013) provided a measurement to estimate trade costs as a function of observable trade data in time series and panel data analysis He derived a micro foundation that indirectly infers trade frictions base on Anderson and Van Wincoop
Effects of AFTA on intra-regional trade flows at aggregated level
At aggregated level, in investing the intra-regional effects of AFTA Elliott and Ikemoto
(2004) examined data cover the periods 1982-1999 before and after the signing AFTA by using a modified gravity equation with the dataset of five founding countries and non-ASEAN members They found that the intra-ASEAN and extra-ASEAN trade flows were insignificantly affected immediately after signing AFTA agreement
Typically, they investigated the degree of the trade creation and trade diversion effects as a result of AFTA by adding more two dummies for four PTAs dummies Although almost results were showed at high level of statistical significance, all regressions are estimated using OLS and Tobit model to account the zero trade values with very similar results These two methods therefore lead to biased in estimated coefficients
Similarly, Indira M Hapsari and Carlos Mangunsong (2006) examined various determinants on both intra- and extra-regional trade flows of between AFTA members by using a dataset cover the year AFTA implemented (1993-2003) for 5 ASEAN countries and 14 other trading partners, which are advanced countries They used OLS estimation, which are evaluated as a simple and biased estimator They found the intersting findings: firstly, the reduction of tariff had resulted a significant effect in increasing the export trade value; secondly, the similarity of the export structure of ASEAN members is one of the important factors affecting the growth of intra-industry trade among them They also suggested that “AFTA may be causing some trade diversion and shifting trade from countries outside the bloc to possibly less efficient countries inside the bloc”
On the other hand Trung Kien (2009) applied a large panel data of thirty-nine countries during the period 1988-2002 that the Hausman-Taylor (HT) to test in FEM and REM in taking into account heterogeneous effects in error component Trung Kien (2009) found that “the formation of AFTA has resulted in significant trade creation among its members” Importantly, a positive estimated coefficient on the import AFTA dummy suggested that there had been no import trade diversion
In addition, Ismail and Wong, C K (2013) reported a conclusion that there is a positive significant coefficient in AFTA dummy in which confirm that free trade agreement encourages intra-ASEAN trade Basically, their findings based on two estimators – Pooled OLS and RE model with panel data for 25 years from 1986 to 2010 including five ASEAN countries and thirty-nine selected trading partners.
Effects of AFTA on intra-regional trade at disaggregated level
In sector level analysis, Herath et al (2014) applied the gravity model to estimate the impacts of AFTA on agrifood trade and eight agrifood commodities individually They stated that “joint membership has significantly enhanced agrifood trade among member countries of AFTA” even though agrifood was often considered as sensivtive good to protect between countries Likewise, coefficients as representative of trade creation and trade diversion effects shown statistically significant and positive values indicated that trade creation in agrifood trade with the formation of AFTA in tersm of agrifood sector
In a study that involved diverse products, Okabe and Urata (2014) applied a similar analysis method with pervious studies – by using gravity model and panel data approach for a wide range of products from 1980 to 2010 However, the number of dummy variables affecting trade costs such as common border, common language, island and landlocked status was not used They focused on investigating the effects of AFTA by sector level and by member country level Futhermore, RE, FE and HT test are estimators that the paper supposed to use Remarkably, they used tariff reduction variable as proxy for trade creation and trade diversion effects under AFTA on intra- ASEAN trade.In paticular, they investigated trade creation effect by tariff elimination by the CEPT Scheme and trade creation effect for individual AFTA member countries
In general, Okabe and Urata (2014) revealed that positive and significant trade creation effects of AFTA in almost sectors for both imports and exports In addition, the coefficient parameters of tariff elimination on imports tend to be greater than that of exports Moreover, old members received more trade creation effects than new members
However, their results of transport equipment industry that related to automotive industrywill be morewell thought-out in the next section
On the other hand, by using manufacturing trade data Pakasa Bary (2015) exposed that AFTA has expanded the trade largely in two way: trade creation effects happened immediately after signing the trade agreement but tending to decrease gradually; contrarily, trade diversion was not significant in the short period of time, but effects tend to rise overtime He investigated the impact of AFTA in a time-varying context by using bilateral trade data of 8 ASEAN countries and 120 partners from 1990 to 2012
Recently, Wong, C K et al (2017) suggested that main factor’s results obtained from
RE model including are significant By estimating AFTA dummies coefficients, they revealed that AFTA has caused trade creation for exports among AFTA members, compared to non-AFTA members and trade diversion for both exports and imports between AFTA members and non-AFTA members in manufactured products
Table 2.1: The estimated coefficients of AFTA dummies from existing studies
(+) Trade creation Hapsari and Carlos
(-) Trade diversion Wong, C K et al
Overall, the estimated coefficients of AFTA dummies 8 from the previous studies in which the trade creation and trade diversion effects of AFTA are identified based on data trade flows of whole tradable goods The table 2.1 summarizes the different results
8 The explanation of AFTA dummies can be seen more details in the section model specification of this study In this sections, the meaning of AFTA dummies from previous studies are understood as AFTA 1 , AFTA 2 , and AFTA 3 in this study AFTA 1 determines the case both exporter and importer are member of AFTA in the pair country AFTA 2 determines the case only exporter is member of AFTA AFTA 3 determines the case only importer is member of AFTA. from them Therefore, the study applies a similar methodology through AFTA dummies for the case of automotive industry to investigate again the impact of AFTA
In conclusion, the study follows previous proposed model specifications that gravity equations are considered effects of standard variables, time-variant variables, time- invariant variables and focus on AFTA dummies Unlike most of the existing empirical studies, the author used both tariff reduction variable in the modified gravity equation to capture the differences of elasticities of tariff elimination under CEPT Scheme and ATIGA between old and new AFTA members Particularly, the Poisson Pseudo- Maximum Likelihood (PPML) and regression by high-dimensional fixed effect estimators are employed to deal with issues common of gravity model using panel data approach at sector level: heteroskedasticity, autocorrelation, heterogeneity and zeroes
Thus, the author’s main contribution in this study is applying a new approach in method for the case of automotive trade flows that could be applied for other sector in ASEAN region.
DATA AND METHOD
The estimation methodology and model specification
Regarding to the methodology for empirical research applying gravity model, Anderson and Van Wincoop (2003) based on the assumption of constant elasticity of substitution (CES) preferences provided a basic gravity equation for estimating bilateral trade at country level without considering the case of disaggregated trade:
1/(1−𝜎) where 𝑌 𝑖 and 𝑌 𝑗 are levels of GDP, 𝑌 𝑤 is world GDP, 𝜃 𝑖 is the income share of country
𝑖, and 𝜎 is common elasticity among all goods The term 𝑡 𝑖𝑗 is the unobservable border effect and П 𝑖 and 𝑃 𝑗 are the multilateral unobserved prices
After providing an inherent solution to the price indices as a function of all bilateral trade barriers and income shares With symmetry of trade costs 𝑡 𝑖𝑗 = 𝑡 𝑖𝑗 , П 𝑖 = 𝑃 𝑖 , and their final gravity equation as follow:
In the development of theoretical foundation of gravity model, trade costs became the most important factors to be taken into account Trade costs are large and variable that highly linked to economic policy such as tariff, non-tariff barriers associated with the exchange rate But in the standard form of gravity model, trade costs include all transport, border-related In addition, the matter of “trade costs also reflects the local distribution costs from foreign producer to final user in the domestic country” (Anderson and Van Wincoop, 2004)
At product level, Anderson and Van Wincoop (2004) derived a theory-based gravity model, namely “Class of trade separable general equilibrium model” The model of trade separable obtained under the assumption of separable preferences and technology
This assumption allows “two-stage budgeting needed to separate the allocation of expenditure across product classes from the allocation of expenditure within a product class across countries of origin” (Anderson and Van Wincoop, 2004, p.707) Therefore, the demand function at product level is used as the gravity equation for the estimation
𝐸 𝑗 𝑘 where, 𝑋 𝑖𝑗 𝑘 is defined as exports from 𝑖 to 𝑗 in product class 𝑘, 𝜎 𝑘 is the elasticity of substitution among brands, 𝑝 𝑖𝑗 𝑘 is the price charged by 𝑖 for exports to 𝑗, 𝐸 𝑗 𝑘 is expenditure allocations from 𝑗 to 𝑖, and 𝑃 𝑗 𝑘 is the CES price index:
Undergoing multiple transformation steps, that yields the system:
𝑖 where, 𝑌 𝑘 is world output in sector 𝑘 The indices 𝑃 𝑗 𝑘 and П 𝑖 𝑘 solved as a function of trade barriers {𝑡 𝑖𝑗 𝑘 } and the entire set {𝑌 𝑖 𝑘 , 𝐸 𝑖 𝑘 } Therefore, trade flows depend on trade barriers and the set {𝑌 𝑖 𝑘 , 𝐸 𝑖 𝑘 }
While, trade cost function is assumed as a function of observables 𝑧 𝑖𝑗 𝑚 which includes, adjacency, distance, preferential trade agreement, common language and historical relationship
Replace the equation (4) into the equation (1) and the logarithmic form of the empirical gravity equation transformed by above nonlinear function becomes: ln(𝑋 𝑖𝑗 ) = ln(𝑌 𝑖 ) + ln(𝑌 𝑗 ) + ∑ 𝑀 𝑚=1 (1 − 𝜎)𝛾 𝑚 ln (𝑧 𝑖𝑗 𝑚 )− (1 − 𝜎) ln(П 𝑖 ) −
The term 𝜀 𝑖𝑗 is error term
In practice, the gravity equations apply the natural logarithmic transformation This specification allows to easily interpret estimators as elasticities
In general, a number of variables are added to capture the trade costs component since Anderson and Van Wincoop (2004) including geographical distance, land area, and other dummies for island, landlocked status, common borders, common language, colony history, etc (Andrew, 2004) There seems to not any existing papers use both tariff and non-tariff variable under AFTA agreement to capture the separated effects of them However, there is only Okabe and Urata (2014) considered the tariff difference between MFN and CEPT Scheme for a wide range of sectors
Before Anderson and Van Wincoop (2003) there was lack of theoretical foundation of empirical gravity equations results biases due to omitted variables and has problem with conducting comparative statics exercise The most important contribution of their theory is solving the impact of trade barriers on trade flows as “multilateral resistance terms”
(MRTs) Anderson and Van Wincoop (2003) constructed estimates of the price-raising effects of barriers to multilateral trade However, this procedure requires a non-linear least square (NLS) program to estimate To solve that problem, according to Feenstra
(2004) and Richard Baldwin and Daria Taglioni (2006) there are often two alternative methods to use as proxy for price indexes called “remoteness” variable related to distance to all bilateral partners Another simpler is using country fixed effects for importers and exporters as dummy variables In this study MRTs technique is used The methodology of estimating the impacts of distance and other bilateral variables on trade flows supported replacing the MRTs indexes in equation (1) (Anderson and Van Wincoop, 2004)
Therefore, the interest of this research is focusing on estimating trade creation and trade diversion impacts cause by the presence of regional trade agreement Recently, the most complete methodology in modelling dummies as proxy for trade creation and trade diversion effects has presented in Yang and Inmaculada (2014)
Likewise many previous studies, the author apply gravity equations for bilateral export flows and import flows at product level 𝑘 from one ASEAN country 𝑖 to its 39 main trading partner (country) 𝑗 and to the remaining ASEAN countries, vice versa The study follow methodology of Anderson and Van Wincoop (2004)for building trade cost function at product level Therefore, the estimated equation for the case of automotive industry by panel data is expressed as follows:
The gravity equation in the case of export flows as follows: ln(𝑋 𝑖𝑗𝑡 ) = 𝛽 0 + 𝛽 1 ln(𝑌 𝑖𝑡 ) + 𝛽 2 ln(𝑌 𝑗𝑡 ) + 𝛽 3 ln(𝑦 𝑖𝑡 ) + 𝛽 4 ln(𝑦 𝑗𝑡 ) + 𝛽 5 ln(𝐷𝑖𝑠𝑡 𝑖𝑗 )
The gravity equation in the case of export flows as follows: ln(𝑀 𝑖𝑗𝑡 ) = 𝛽 0 + 𝛽 1 ln(𝑌 𝑖𝑡 ) + 𝛽 2 ln(𝑌 𝑗𝑡 ) + 𝛽 3 ln(𝑦 𝑖𝑡 ) + 𝛽 4 ln(𝑦 𝑗𝑡 ) + 𝛽 5 ln(𝐷𝑖𝑠𝑡 𝑖𝑗 )
𝑋 𝑖𝑗𝑡 and 𝑀 𝑖𝑗𝑡 denotes bilateral export/import value of motor vehicle and its parts between country 𝑖 and 𝑗 at year 𝑡,
𝑌 and 𝑦 denote GDP and GDP per capita, respectively 𝑌 and 𝑦 are representative for economic size and income level, respectively,
𝐷𝑖𝑠𝑡 𝑖𝑗 is a measure of the geographical distance between 𝑖 and 𝑗, which is introduced as proxy for transport costs (in kilometers),
𝐴𝑟𝑒𝑎 𝑖𝑗 is the multiplication of land area of country 𝑖 and country 𝑗 (in square kilometers),
𝐶𝑜𝑚𝑏𝑜𝑟 𝑖𝑗 is a dummy variable which takes a value of 1 if 𝑖 and 𝑗 share a common border, and otherwise,
𝐿𝑎𝑛𝑑𝑙 𝑖𝑗 is the number of landlocked countries in the country-pair (0,1 or 2),
𝐼𝑠𝑙𝑎𝑛𝑑 𝑖𝑗 is the number of island countries in the country-pair (0,1 or 2),
𝐶𝑜𝑚𝑙𝑎𝑛𝑔 𝑖𝑗 is a dummy variable which takes a value of 1 if 𝑖 and 𝑗 have a common language, and 0 otherwise,
𝐶𝑜𝑚𝑐𝑜𝑙 𝑖𝑗 is a dummy variable which takes a value of 1 if 𝑖 and 𝑗 were ever colonies with the same colonizer, 0 otherwise,
𝐶𝑜𝑙𝑜𝑛𝑦 𝑖𝑗 is a dummy variable which takes a value of 1 if 𝑖 ever colonized 𝑗 or vice versa, and 0 otherwise,
𝑊𝑇𝑂 𝑖𝑗𝑡 is a dummy variable which takes a value of 1 if both 𝑖 and 𝑗 belong to WTO organization at year 𝑡, and 0 otherwise,
𝐹𝑇𝐴 𝑖𝑗𝑡 is a dummy variable which takes a value of 1 if both 𝑖 and 𝑗 belong to Free Trade Area at year 𝑡, and 0 otherwise,
𝐴𝐹𝑇𝐴𝑝𝑙𝑢𝑠 𝑖𝑗𝑡 is a dummy variable which takes a value of 1 if both 𝑖 and 𝑗 belong to AFTA-plus-Free Trade Area at year 𝑡, and 0 otherwise,
𝐴𝐹𝑇𝐴 1𝑖𝑗𝑡 is a dummy variable which takes a value of 1 if both 𝑖 and 𝑗 belong to AFTA at year 𝑡, and 0 otherwise,
𝐴𝐹𝑇𝐴 2𝑖𝑗𝑡 is a dummy variable which takes a value of 1 if exporting country 𝑖is a member of the AFTA and importing country 𝑗 is not at year 𝑡, and 0 otherwise,
𝐴𝐹𝑇𝐴 3𝑖𝑗𝑡 is a dummy variable which takes a value of 1 if importing country 𝑗is a member of the AFTA and exporting country 𝑖 is not at year 𝑡, and 0 otherwise,
𝑇𝑎𝑟𝑖𝑓𝑓_𝑟𝑒𝑑𝑢𝑐𝑡𝑖𝑜𝑛 𝑖𝑗𝑡 is the difference of MFN tariff rate and the CEPT tariff rate that impose by an AFTA member country 𝑗 to the exporting country 𝑖 at year 𝑡, vice versa This variable takes a positive or zero value depending on whether the pair country is AFTA member or not,
{𝑒𝑥𝑝_𝑑𝑢𝑚𝑚𝑦}and{𝑖𝑚𝑝_𝑑𝑢𝑚𝑚𝑦} is a dummy variables which controls the effects of exporter or importer,
{𝑇 𝑡 } is a time dummy variable which controls the effects of year,
𝜀 𝑖𝑗𝑡 is error term that represents the omitted other influences on dependent variable
Table 3.1: Possible outcomes of trade effects in an AFTA
Expansion of extra-AFTA export
Export diversion + Contraction of intra- AFTA export
Expansion of extra-AFTA import
Import diversion + Contraction of intra- AFTA import
The hypotheses for those explanatory variables that increase (positive) or decrease (negative) bilateral trade are described shortly as following table, while (+) means positive and (-) means negative:
Table 3.2: Hypotheses of the estimated coefficients
In the study, these time-variant variables GDP, GDP per capita, WTO, FTA, AFTAplus,
AFTA 1 , AFTA 2 , AFTA 3 , Tariff reduction are used to control macroeconomic effects, incentive policies and policy barriers Meanwhile, these time-invariant variables dist, area, border, landlocked, island, language, common colonizer and colony capture the effects of other trade cost component.
Data description
The most difficulty in building a dataset for a gravity model is using many explanatory variables The dataset is therefore extensive collected from a variety of different data sources for different kind of variables Then, the dataset need to be merged in a single database That task took too much time in organizing
This study uses a big dataset cover 49 countries (10 ASEAN countries and its 39 main trading partners) in 26 years (from 1993 to 2018) The list of these countries is provided in Appendix 7 For the 49 countries the author constructed a matrix of trade flows represent 870 (10×9 + 39×10 + 10×39) pair countries And the maximum observation is 22620consisting of exports from country 𝑖 to country 𝑗 and vice versa However, about 35% of export flows are zeros Therefore, the author need to choose an appropriate econometrics model to deal with that problem The most advantages of this database is sample size and econometrics estimation would be more confident
Trade data: The bilateral trade flows in value base on disaggregated data with high level of commodity aggregation followed by Harmonized Commodity Description and Coding Systems (HS) at 4-digit level and 6-digit level that collected from United Nation (UN) Comtrade 9 through the World Bank’s WITS portal for almost countries since 1962
All bilateral trade on FOB exports or CIF imports in current US dollars converted from national currencies at nominal exchange rates Therefore, to calculate the effects of USD currency on trade value, the author used the annual trade weighted U.S dollar index (index 19970) which reflects the strength of the dollar relative to the currencies of a broad group of major U.S trading partners accounted for well over 90 percent of total U.S imports and exports in 1997 U.S dollar index published by the Federal Reserve of the United States of America 10 Since the data for 1993 and 1994 are not available, the author US dollar index of 1995 to replace
The data of bilateral trade flows has been separated into two different flows, export flows and import flows Based on the availability of distinguished trade flow, gravity equations are estimated in both sides – export flow and import flow
9 UN Comtrade is the first and foremost database for trade by commodity introduced by WTO’s guide book (World Trade Organization, 2012)
10 https://fred.stlouisfed.org/series/TWEXBANL
Bilateral trade flows related to automotive industry based on all version of 59 HS codes
That will help cover all data for the period 1993-2018 Description of each code followed by HS17 can be seen in Appendix 2 This study will use 59 codes to cover all automotive parts mostly for motor vehicles such as motor vehicles, motor vehicle engines and parts, motor vehicle bodies, automotive electrical and instruments, and other automotive components such as clocks, seats that represent for automotive industry
The most important things that could be seen carefully when collecting bilateral trade data is lack of data from mostly sources For instance, in export flow, there is 8470 zeros accounting for 35% of missing data The lack of data lies mainly in the small economies of ASEAN
Table 3.3: The availability of bilateral trade flows from UN Comtrade
Real GDP and real GDP per capital in constant 2010 US dollars by countries were obtained from WDI
Geographical distance in kilometer collected from the website Distance.1km.net 11 for air between cities This research uses capital cities of 49 countries to calculate the physical distance between start location and destination For the case of Myanmar, nowadays Naypyidaw is the capital and third-largest city of Myanmar However, the old capital city was Yangon, which is the largest city so far 12 Therefore, this research calculated the distance from Myanmar to other countries based on the geographical location of Yangon
11 Distance between two latitude-longitude coordinates are calculated with function based on: https://www.geodatasource.com/
12 https://www.britannica.com/place/Yangon
The author exploited the Central Intelligence Agency (CIA)’s World Factbook for a number of country-specific variables or time-invariant variables These variables include land area (in sq km), physically contiguous neighbors, landlocked and island status, language, colonizers and dates of independence Almost these variables have been controlled as dummy variables (0 or 1)
Other dummies related to impacts of regional and multilateral free trade agreements obtained from WTO website 13 and Market access Map of International Trade Centre 14 , include: WTO, FTAs, AFTA-plus-FTAs, AFTA, see more detail in Appendix 8
Tariff and non-tariff data of ASEAN countries obtained from the WTO Integrated Database, ASEAN Secretariat and WITS – World Integrated Trade Solution (World Bank).
The estimation method and common issues of panel data in gravity model
This section will present the real applications of the estimation of gravity model by panel approach which is helpful in dealing with common problems related to multiple gravity estimations Basically, there are two different methods – linear and non-linear by taking logarithm from the original equations in STATA This study applies both of them due to different set of advantages and disadvantages
Recall from the theory-consistent gravity model by Anderson and Van Wincoop (2004) consists of the multilateral resistance terms as mentioned above And these MRTs do not correspond to any price indices from national statistics therefore it cannot be observable Therefore, it is necessary to find an alternative estimation approach that allows us calculated the effects of inward and outward multilateral resistance to trade between countries, even though these factors do not need to include in the estimators
Baier and Bergstrand (2009) suggested two strategies to solve this problem by applying fixed effects model (FEM) especially to estimate the effects of trade agreement on trade flows However, normally the FEM cannot estimate the effects of time-invariant variables such as distance, a common language, common border, landlocked and island status, etc So random effects model (REM) is called Depending upon the assumptions
14 https://macmap.org/ about the error term in panel data model that whether they are fixed or random, the Hausman-Taylor test is applied for FEM and REM models Most of these existing empirical papers employed these estimation methods (Trung Kien, 2009; Okabe and Urata, 2014; Wong, C K et al., 2017; Ismail et al., 2013)
The advantage of both FE and RE is solving endogenous problem due to explanatory variables correlated to error term However, the disadvantage of two these models is not appropriate in estimating in the case dependent variable is limited or the presence of zero trade flows in data Consequently, the estimation results of FE and RE is biased (Gómez-Herrera, 2013; Santos Silva and Silvana Tenreyro, 2006; Mahfuz Kabir et al.,
Regarding to the advantages of panel data shown in methodology framework that permits considering the change of variables over time and related to specific time or individuals (Andrew and Van Wincoop, 2001; Melitz ,2007)
Nevertheless, according to Gómez-Herrera (2013), there are several problems of gravity equation in empirical studying that required to solve by alternative techniques He argued that the unobserved heterogeneity, the presence of heteroskedasticity in trade data and the presence of zeros are the most important issues surrounding gravity model and panel data He provided a survey of the most recent literature concerning the specification and estimation methods of gravity equation In the end, Gómez-Herrera
(2013) showed that the Heckman sample selection model performs better estimation results On the other hand, Santos Silva and Silvana Tenreyro (2006) offered the most natural procedure thus can be corrected by applying a Poisson pseudo maximum- likelihood (PPML) But each of them has its own set of advantages and disadvantages in dealing common issues of gravity equations For instance, only PPML can deal with heteroskedasticity Meanwhile, “Heckman allows for separate data generating processes for the zero and non-zero observations, whereas Poisson assumes that all observations are drawn from the same distribution” (Shepherd, 2016)
In addition, at disaggregated data level, Helpman et al (2008) extended Heckman sample selection to also take into account the biased due to the heterogeneity of firms
Basically, Helpman et al (2008)’s model or HMR model is a generalized gravity equation that accounts for the self-selection of firms into export markets and their impacts on trade volumes This model is supposed to be helpful in predicting the positive as well as zero trade flows at sector level analysis and does not require firm data level
But, there might issue some problems when applying HMR in STATA for mid-sized datasets In addition, there is very few information be found in applying HMR model
Therefore, HMR is not discussed and applied in this study
This study focus on solving existing common issues when applying the gravity model for trade as mentioned above To solve these issues, Santos Silva and Silvana Tenreyro
(2006) discussed the log-linearization of the gravity equation in OLS estimator changes the property of the error term that lead to heteroskedasticity and produce an inefficient standard error of parameters Therefore, the assumption number 3 of OLS regression is violated [ (𝑉𝑎𝑟(𝑢| 𝑋 2𝑖 , ,𝑋 𝑘𝑖 ) = 𝜎 𝑖 2 ] and OSL’s estimation results are biased Normally, the reasons leading to heteroskedasticity may be due to the characteristics of data or omitted variable or functional form misspecification Also, the autocorrelation problem often appears in gravity model of trade Regarding zero trade flows problem in data which is considered as limited dependent variable in statistics, there are several useful models, namely Logit model, Probit model, Tobit model and Poisson model However, trade data is continuous, not as count data So, PPML with command ppml developed by Santos Silva and Silvana Tenreyro (2006) is more outstanding among them In fact, the dataset in this study has over 35% (8470/ 22620 observations) in term of exports and over 43% (9867/ 22620 observations) in term of imports of automotive trade flows being zeros
Later on, Martijn Burger et al (2009) examined the performance of Poisson model (PPLM) and modified Poisson (negative binomial, zero-inflated) in comparison to solve the serious biased results due to omitted zeros across the importing and exporting countries They showed that “the PPML and the zero-inflated Poisson model (ZIPPML) perform the best when comparing the estimated and observed values of the dependent variable, and even outperform OLS in our example” When considering model fit and the relevance of excess zeros, the ZIPPML on average scores best However, after that only PPML showed the best performance in the case of continuous data in trade, even when the proportion of zeros is very large (Santos Silva and Silvana Tenreyro , 2011a) by using ppml command that automatically uses the robust option for estimation But to estimate by ppml with multiple fixed effects, ppmlhdfe command is used from SSC and vce(cluster(clustvar) to specify that the standard errors allow for intragroup correlation, and relaxing the usual requirement that the observations be independent 15 Sergio Correia et al (2020) suggested that by applying PPMLHDFE as standard approach to model count data with nonnegative dependent variable and robust standard errors by clustering country-pair effect as a main reason of heteroskedaticity Meanwhile, HDFE is a computes the residuals to partial-out variables with respect to multiple levels of fixed effects 16 According to Sergio Correria, PPML HDFE is totally different with PPML, instead of searching for problematic regressors, it looks for problematic observations It also adopts variables which is perfect multicollinearity Also, HDFE technique suggested as for controlling the unobservable effects due to MRTs as mentioned in the section of estimation methodology
Besides, the author also applies a considerable alternative method by linear approach –
Reghdfe Regression HDFE is similarly as an estimation linear regressions that control for multiple fixed effects Moreover reghdfe is command in STATA also developed by Sergio Correia to improve over existing commands
Therefore, this study strictly follows the strongest estimator – PPMLHDFE model and REGHDFE model and to take into account multiple fixed effects in absorb(abvars) 17 , the study separates into five different cases (model) of controlling fixed effects for each method, the following tables show precisely
Table 3.4: Description of multiple fixed effects models
Models Model 1 Model 2 Model 3 Model 4 Model 5 t t,it t,jt t,it,jt t,it,jt,ij
Time- and country- and country pair effects
15 https://www.stata.com/manuals13/xtvce_options.pdf
16 http://scorreia.com/demo/hdfe.html
17 Description of using ppmlhdfe can be seen at http://scorreia.com/help/ppmlhdfe.html
Data processing
The nominal trade value of dependent variables collected from the UN Comtrade are converted to the real trade value by using the U.S dollar index to deflating the time effects of U.S dollar currency over time
Then, to avoid the problem of non-positive value after taking natural logarithm of dependent variable and GDP and GDP per capita variables, the author use bilateral trade value at U.S dollars, GDP at thousand U.S dollars and GDP per capita at U.S dollars
In the case of running REGHDFE model, adding a small value (equal 1) into the dependent variable is necessary to keep the maximum observations and avoid missing data However, this technique creates non-informative zeros in the dataset of bilateral trade value Regarding to this problem, PPML DHFE model is more appropriate Both REGHDFE and PPMLHDFE methods allow to absorb multiple levels of fixed effects including heterogeneous slopes and additional robust standard errors (multiple way clustering, etc.) 18
For each case of analysis, the author considers the better estimated coefficient between REGHDFE and PPMLHDFE and between different set of multiple fixed effects
Basically that process is applied for different estimation of AFTA dummies, tariff reduction and tariff reduction interact with country dummies in two different ways, export flows and import flows
18 For Reghdfe description: http://scorreia.com/help/reghdfe.html
FINDINGS
Estimation results of trade creation and trade diversion effects by AFTA
The thesis employs the gravity model by panel data approach described above to estimate the trade creation and trade diversion effects of the ASEAN Free Trade Area (AFTA) agreement on the bilateral automotive trade flows among 49 countries from
1993 to 2018 The table 4.1 presents the main estimation results of AFTA dummy variables by REGHDFE method after considering the functional misspecification problem in PPMLHDFE method Among five different models of multiple fixed effects, the model 1 using time effect show the full and better results of AFTA dummies, only AFTA1 is positive and insignificant, meanwhile AFTA2 and AFTA3 are negative and high significant level Although the adjusted R-squared is not highest one which means a good fit at 60% for existing data that could be acceptable, the result from Ramsey test suggest that the function of this model is correct The impacts of AFTA on export trade flows is observed in the estimated slope coefficients of AFTA1, AFTA2 and AFTA3
The estimated 𝛾 4 coefficient is 0.669 at insignificant level The estimated 𝛾 5 and
𝛾 6 coefficients are -0.691 and -1.749 at significant level, respectively Very different coefficient of 𝛾 4 , 𝛾 5 and 𝛾 6 can be seen in the four remaining models And a different combination of multiple fixed effects leads these AFTA dummies be omitted in the results
In addition, the estimated coefficients of GDP and GDP per capita is significant positive as strongly increasing in automobile’s trade between larger economic size and higher income For instance, if GDP of exporter increase by 1% then export trade flows will increase by approximately 3% The estimate of distance indicates that the physical distance between two countries is still a very important impediment to export flows The sign of those coefficient once again confirm the direction of relationship between them and the bilateral trade variable in theoretical foundation of gravity model
Table 4.1: The effects of AFTA on export trade flows by REGHDFE
REGHDFE method Model 1 Model 2 Model 3 Model 4 Model 5
Ln(Export flow) t t,it t,jt t,it,jt t,it,jt,ij
Note: Regressed by log real trade Robust standard errors (clustering by country-pairs) in parentheses*** p