This paper analyses and compares the economic effects of ASEAN Free Trade Area (AFTA) on the trade flows of Vietnam and Singapore. Using the gravity model, the study shows a number of robust empirical findings. First, on the multilateral trade flows of Vietnam and Singapore, the model reveals no trade diversion following integration. Second, trade flows are not significantly affected immediately following the signing of the AFTA agreement. Third, trade distance still remains a hindrance to trade, indicating that globalization and integration have not mitigated the relevance of physical distance even though technological innovations continue to reduce transportation costs. Fourth, cultural familiarity, as a proxy for linguistic affinity and colonial relationship, continues to be an important determinant of bilateral trade flows. Finally, differences in per capita income among trading partners continue to have negative impacts on bilateral trade. Efforts to narrow the GDP gaps among members, improving social infrastructure, and continued domestic reforms are suggested as remedies for the obstacles to freer flows of trade in the region.
Regionalism has become one of the most salient trends in the global economy. Even after the launch of the WTO multilateral trading system, the number of regional trade agreements (RTAs) has increased rapidly, expanding their scope and geographical reach across continents. In spite of such incremental growth in RTAs, no consensus for the merits of RTAs has been confirmed yet. Instead, these initiatives have led to ongoing debate between economists and politicians over whether RTAs represent ‘building blocs’ or ‘stumbling blocs’. Economists advocating RTAs have heralded such agreements as stepping stones towards worldwide free trade, which benefits individual countries because of the cost reduction resulting from
intensified competition, economies of scale, and diversified consumers’ choices. In contrast, opponents have argued that these initiatives will be stumbling blocs, acting primarily to divert trade from other countries to those countries receiving preferential treatment.
The ‘second-best’ nature of liberalizations under RTAs makes it difficult to determine whether the welfare effects from RTAs will be positive, even for the members of the arrangements. In addition, empirical research has not succeeded in reaching firm conclusions on whether trade creation outweighs trade diversion. For example, Karacaovali and Limao (2005) and Limao (2006) found stumbling bloc effects for the cases of EU and U.S. respectively. In contrast, Miljkovic and Paul (2003) found that trade creation occurs in US agricultural exports to Canada because of Canada - USA Free Trade Agreement. Therefore, the effects of forming a regional integration agreement on the welfare of member countries and non-member countries will depend on individual circumstances and can only be assessed on a case-by-case basis.
Founded in 1992, ASEAN Free Trade Area (AFTA) aims primarily at increasing ASEAN’s competitive edge as the production base for the world market. To achieve this goal, the plan involved a lowering of intra-regional tariffs, through the Common Effective Preferential Tariff (CEPT) scheme, to 0-5 percent within a period of 15 years commencing in 1993. Whether the formation of AFTA has increased intra-regional trade in the South East Asian region still remains a point of controversy in the literature. Some studies (e.g., Elliott and Ikemoto, 2004) have shown that intra-regional trade in ASEAN was strengthened in the 1990s. In contrast, other studies (e.g., Endoh, 2000) have produced opposite results.
This paper centers on the following research objectives: First, using a set of models with fixed and random effects considered, it evaluates the impacts of accession to AFTA on the multilateral trade flows of Vietnam and Singapore as a result of trade creation and trade diversion among member and non-member countries. Second, it investigates the issue of whether or not RTAs can bring about regional economic opportunities and how these opportunities can be fully exploited. Third, due to the varying levels of development among member countries the analysis in this paper aims to offer practical insight into whether or not high and low-income members are equally affected. Fourth, it examines other key determinants of the trade flows of Vietnam and Singapore and finds the potential obstacles therein. Finally, the paper derives policy implications based upon the empirical findings.
This study differs from the previous studies in several aspects. First, instead of pooling data across all countries, we estimate a single equation for Vietnam and Singapore1. This provides us with a better understanding of the impacts of AFTA on the individual countries. Second, this study presents a set of models: fixed effects model and random effects model as discussed in section 2. Third, we decided to choose Vietnam and Singapore as case studies because Vietnam and Singapore are in different stages of development. In addition, although there are a number of studies on the effects of regional trade integration on its members (Athukorala and Menon, 1997; Brada and Mendez, 1983; Clausing, 2001; Fukase and Winters, 2003; Heng and Gayathri, 2004; Tang 2003), little research has focused on comparing the effects of AFTA on Vietnam and Singapore. Another point of interest is that Vietnam’s tariff rates before joining AFTA were relatively high, while those of Singapore were close to zero. Therefore, it would be interesting to see how AFTA affects Vietnam, with its initially higher tariff levels, and Singapore, with its lower tariff levels.
2.1. Analytical framework
To identify the effects of the FTA, it is important to disentangle the effects of regional integration from other changes in the economy. A standard way to control for those effects is to run a gravity model2, and see whether the estimated relationships change as a consequence of implementing the FTA (e.g., Brada and Mendez, 1983 and Carrère, 2006). For estimation purpose, the final regression equation is expressed in log-linear form as follows:
Where: Tijt is the total trade volume between country i and country j at the time t3
GDPitis Gross Domestic Product of country i at the time t.
GDPjtis Gross Domestic Product of country j at the time t.
POPit is the population of country i at the time t.
PGDP_DIFijt is the difference in per capita GDP between country i and country j at the time t (in absolute value), measured as |PGDPit - PGDPjt|.
POPjtis the population of country j at the time t.
DISTij: Distance between the capital city of country i and the capital city of country j.
BORDij is a dummy variable that equals 1 if both country i and country j have a common border, and zero otherwise.
LANGij is a dummy variable that equals 1 if both country i and country j have the same language, and zero otherwise.
EX_COLij is a dummy variable that equals 1 if country i ever has been colonized by country j or vice versa, and zero otherwise.
AFTAijt is a dummy variable that equals 1 if both country i and country j belong to AFTA at the time t, and zero otherwise.
The inclusion of supply factor of the exporting country (GDPit) and demand factor of the importing country (GDPjt) is justified on the ground that higher level of exporting country’s GDP indicates higher level of production for exports, while higher level of importing country’s GDP suggests higher level of demand for imports. Therefore, it is expected that trade increases with the country size, as measured by GDP (Chionis and Liargovas, 2002; See Frankel, 1993), with other factors kept constant.
Elliott and Ikemoto (2004), Tang (2003), and Roberts (2004) incorporated per capita GDP difference variable, log (|GDPCi - GDPCj|), in order to test for the Linder Hypothesis. According to Linder Hypothesis, countries with similar level of per capita income tend to trade more with each other because they have the most similar demand patterns, and produce similar but differentiated products (Markusen et al., 1995). The positive sign fits standard H-O-S framework (e.g., Peridy, 2005), whereas a negative sign supports the Linder hypothesis (e.g., Tang, 2003).
The theoretical justification for population variables (POPit and POPjt) is somewhat imprecise. On the one hand, large population could promote a division of labor and allow more industries to reach efficient economies of scale. Thus, opportunities for trade with foreign partners in a wide variety of goods will increase, suggesting a positive impact of population on bilateral trade (See Brada and Mendez, 1983; Oguledo and Macphee, 1994). On the other hand, populous countries are assumed to be larger in area and thus endowed with a greater quantity and variety of natural resources. The bigger absorption effect of this domestic market causes less reliance on international trade transactions, indicating a negative impact of population on bilateral trade (See Aitken, 1973; Bikker, 1987; Endoh, 1999; Endoh, 2000; Linnemann, 1966; Martinez-Zarzoso and Nowak-Lehmann, 2003; Sapir, 1981). Therefore, the coefficients for population could be positive or negative, depending on which effect is dominant, an absorption effect or economies of scale effect.
Distance between trading partners (DISTij) is used as a proxy for several distance-related variables such transport cost, cost of time, “psychic distance”4 or “cultural cost”, and access to relevant market information (See Linenman, 1966). All of these factors reflect the cost of international transactions of goods and services and are expected to affect trade negatively (Al-Mawali, 2005; Clarete et al., 2003; Martinez-Zarzoso, 2003; and Bougheas et al., 1999). Therefore, we expect that the sign of the coefficient for DISTij variable is negative.
Since linguistic affinity, ex-colony and commonly shared borders tend to reduce cultural distance and therefore encourage bilateral trade, it is expected that the coefficients for these three dummy variables are positive (Clarete et al., 2003; Endoh, 1999; Geraci and Prewo, 1977; Martinez-Zarzoso, 2003; Peridy, 2005).
Finally, a dummy variable is included to capture the integration effect of the FTA. The coefficient on FTA could be negative or positive depending on a case-by-case basis (see Brada and Mendez, 1983; Baier and Bergstrand, Ghosh and Yamarik, 2004; 2007; Cyrus, 2004; Yu and Zietlow, 1995). A positive and significant coefficient on the FTA dummy could imply that its members have traded with each other more than the hypothetical level predicted by basic explanatory variables.
In this paper, we employ two techniques, including the fixed effects model and random effects model. The fixed effects model allows for country-pair heterogeneity and gives each country-pair its own intercept. The fixed effects estimates can help us reduce potential specification errors from omitting important variables. One shortcoming of this model, however, is that it does not allow for time-invariant variables to be included5. Therefore, we include the random effects model in order to incorporate differences between cross-sectional entities by allowing the intercept to change, as in the fixed effects model, but the amount of change is random. The advantage of random effects model is that both time-series and cross-sectional variations are used.
Apart from regression analysis, several trade indices are calculated and incorporated into the present research in order to provide complementary examination of intra-ASEAN trade. They include the Trade Intensity Index, Revealed Comparative Advantage and Intra-industry Trade (See Appendix 1 for the computable formulas).
According to Mátyás (1997), the traditional cross-section approach is affected by a severe problem of misspecification. Drawing on his critique, this paper uses the panel data for 23 countries over the period of 16 years, from 1990 through 2005. They include six ASEAN countries6 (Indonesia, Malaysia, Philippines, Singapore, Thailand and Vietnam) and 17 non-ASEAN countries7 (Australia, Canada, China, France, Germany, Hong Kong, India, Italy, Japan, Korea, Netherlands, Norway, Spain, Sweden, Switzerland, United Kingdom, and USA).
Yearly total trade between two countries is obtained from the IMF Direction of Trade Statistics - CD ROM. The values of commodity trade are extracted from TradeMap. Data on GDP, GDP per capita and population are obtained from World Economic Outlook Database – IMF (WEO). Information regarding language and colonial relationship is obtained from the Economist Intelligence Unit. The distance between the two capital cities is available from Indo.com. Finally, data on tariff rates are extracted from Market Access Map.
3. Empirical results
3.1. Regression results
This paper estimates the gravity model for Vietnam and Singapore respectively over the period of 16 years, from 1990 through 2005. The estimations use annual data consisting of 352 country pairs for Vietnam (Vietnam’s trade with 22 countries for the period from 1990 through 2005) and 352 country pairs for Singapore (Singapore’s trade with 22 countries over the period from 1990 through 2005).
Table 2: Regression Result for Vietnam
Fixed Effects Model
Random Effects Model
Number of Observation
** Significant at the 0.01 level
* Significant at the 0.05 level
Gravity model for Vietnam
The estimation results for Vietnam are given in Table 1. In the fixed effects model, the variables log DISTij, BORDij, LANGij and EX_COLij are dropped because these variables are time-invariant. In the random effects model, the variable “LANGij” is dropped because it causes the multicolinearity problem8.
As indicated in Table 1, the gravity model fits the data well, providing explanation for the major variation in bilateral trade. The basic variables of gravity equation behave as the model predicts. All estimated coefficients, except log POPitand log of POPjt, are statistically significant at the 0.01 significance level.
Our main interest is in the impact of AFTA on intra-bloc trade. The estimated coefficient on the AFTA dummy variable is negative and statistically significant. Therefore, membership in AFTA does not seem to be important per se when other relevant variables are controlled. We estimate that joining AFTA would lead to a decline in Vietnam’s trade with ASEAN countries by roughly 18.4percent, with other variables controlled9.
GDPit turns out to be the most important explanatory variable, not unexpectedly. The coefficient of log GDPit is on the higher side, suggesting that GDP growth in Vietnam triggers and accelerates the expansion of trade. The estimated coefficient of GDPjt is also positive and statistically significant, indicating that the larger GDP of the trading partner is correlated with increased trade with Vietnam. In the estimation, when Vietnam has an increase in GDP by 100 percent, Vietnam’s trade will increase by 172 percent, while an increase in the trading partner’s GDP by 100 percent would lead to an increase in trade with Vietnam by 97 percent.
The coefficient on per capita GDP difference is negative and statistically significant. This indicates that countries with similar level of income tend to trade more with each other. It is estimated if the difference in per capita income between Vietnam and its trading partner increases by 100 percent, the bilateral trade between Vietnam and its trading partner would decrease by 46 percent.
Although the estimated coefficient of log POPjt is statistically insignificant, its negative sign could be indicative that the absorption effect is greater than the economies-of-scale effect in the trading partners of Vietnam10. A large population may indicate a large domestic market and a large resource endowment, so that the bigger absorption effect of this domestic market causes less reliance on international trade transactions with Vietnam. In contrast, positive sign of log POPit may imply that the economies-of-scale effect is greater than the absorption effect in Vietnam, which allows the advantages of economies of scale to be fully exploited.
In the random effects model, the results are relatively similar to those of the fixed effects model. The coefficient on AFTAijt is also statistically significant. Again the negative sign indicates that joining AFTA would lead to a decrease of 21 percent in bilateral trade between Vietnam and ASEAN countries.
We also find the traditional positive signs on GDP and colonial relationship, and negative sign on distance. The estimated coefficient on the log of bilateral distance is negative and statistically significant. We estimate that an increase in bilateral distance by 100 percent leads to a 217 percent decline in bilateral trade. An ex-common colonizer could raise trade 96 percent. In this model, the coefficient on common land border is negative, but statistically insignificant.
Gravity model for Singapore
The estimation results for Singapore are presented in Table 2. In the fixed effects model, the variables log DISTij, BORDij, LANGij and EX_COLij are again dropped because these variables are time-invariant.
Table 2: Regression Result for Singapore
Fixed Effects Model
Random Effects Model
** Significant at the 0.01 level
* Significant at the 0.05 level
In the fixed effects model, a very high degree of explanation is achieved. All estimated coefficients, except log PGDP_DIFit, are statistically significant at the 0.01 significance level. All other coefficients are not statistically significant.
With regard to the impact of AFTA on intra-bloc trade, the result is different from that of Vietnam. The estimated coefficient on the AFTA dummy variable is positive and statistically significant. Therefore, a pair of countries that joins AFTA would likely experience an increase in bilateral trade by a roughly 11 percent, with other variables held constant. A very tentative explanation for this could be that Singapore moved quickly to establish itself as a trade and investment partner of these countries once they were included in the AFTA.
The coefficients of logs GDPitand GDPjt are positive and statistically significant. It is estimated that an increase in Singapore’s and the trade partner’s GDP by 100 percent would lead to an increase in bilateral trade of 57 and 102 percent respectively. The coefficient of log POPjt is positive and statistically significant, indicating a greater economies-of-scale effect compared with the greater absorption effect in Singapore’s trading partners. The coefficient on log POPit is negative and statistically significant. This might indicate that larger population of Singapore could reduce its trade flows with partner countries.
In the random effects model, the results are relatively similar to those of the fixed effects model. Again the coefficient on AFTA is positive and statistically significant. Thus, joining AFTA would increase Singapore’s bilateral trade with ASEAN countries by 17.5 percent. The coefficients on logs of GDPit and GDPjt are positive and statistically significant. Therefore, an increase in the GDP of Singapore by 100 percent would raise the bilateral trade between Singapore and its trading partner by 56 percent, while an increase in the GDP of the trading partners would increase bilateral trade by 93 percent.
We also found the traditional negative sign on distance, and positive signs on common language and ex-colonizer variables. The estimated coefficient on the log of bilateral distance is negative and statistically significant. This indicates that an increase in the distance between Singapore and its trading partner by 100 percent would indicate a decrease in bilateral trade by 185 percent. Being ever colonized colonizer would increase bilateral trade between Singapore and its ex-colonizer by 153 percent. The coefficient on common land border is inconsistent with the traditional sign.
Although the coefficient of log PGDP_DIFijt is not statistically significant, its positive sign might indicate that the Singapore’s pattern of trade follows Heckscher-Ohlin theory. As data indicate, Singapore tends to trade more with ASEAN countries, whose per capita incomes are low. Overtime the bilateral trade between Singapore and ASEAN countries increases while the difference in per capita GDP between Singapore and ASEAN countries increases. That is a potential reason why the sign of the coefficient is in favor of H-O-S framework.
3.2. Complementary results
Trade Intensity Index
Vietnam and Singapore’s propensities to trade with ASEAN, as measured by the trade intensity index, are given in Table 3-A and 3-B. Compared with Vietnam, Singapore exhibits larger values of trade intensity index for virtually all ASEAN countries. The result seems plausible since trade intensity indices are often high between neighboring countries, and because Singapore is playing a role as an entrepot. The dominating impression is in terms of trend. Between 1990 and 2005, Singapore’s trade intensity with ASEAN was on a slight increase, whereas Vietnam’s trade intensity with ASEAN declined. This trend implies that the relative importance of ASEAN as a trading partner to Vietnam has been declining, while ASEAN has become more important to Singapore.