Microsoft Word Cheongaftaeffects doc

Results 

 

Table 1 shows the results for a sample of all ASEAN imports with positive preferential 

margins both from ASEAN and non-ASEAN partners. The specifications in Columns 1 

and 2 do not include a variable to capture trade creation. In Column 1, the results from 

the OLS within estimator show significant trade diversion with an elasticity of 0.08. 

However, in Column 2, when the Fixed Effects PQML estimator is used, the sign is 

reversed, implying that higher preferential margins increased trade with non-ASEAN 

countries at the HS 6-digit level. This effect is likely to be the result of increasing 

production fragmentation even within a HS 6-digit category following intra-regional 

trade liberalization. The estimated elasticity of intra-regional trade creation is 0.26. Given 

that the average increase in preference margins during the period was 8%, the average 

preferred HS 6-digit category experienced an increase in intra-regional ASEAN imports 

of 2% and even in extra-regional ASEAN imports of 0.7%. Using the average value of 

about US$395,000 in preferred ASEAN imports for a HS 6-digit category in 2001, this 

would translate to an increase in US$8,000 in intra-regional imports and US$2,500 in 

extra-regional imports at the product level. With over 150,000 HS 6-digit products 

involved, the expansion in trade is around US$1.5 billion. The GDP variables are 

estimated with the wrong sign, but they are not statistically significant at the 1% level.  

 

Table 2 performs the estimations by ASEAN country. The estimated trade diversion 

elasticities are all positive, indicating increased trade with non-ASEAN partners. 

However, they are statistically significant at the 5% level for only Brunei, Cambodia, 

Indonesia, the Philippines, and Thailand and range from 0.17 for Thailand to 2.7 for 

Cambodia. As for the intra-regional trade creation elasticities, these are also positive for 

all countries but significant at the 5% level only for Cambodia, Malaysia, Thailand, and 

Vietnam. These range from 0.22 for Thailand to 4 for Cambodia. This country-level 

evidence may be consistent with a story of intra-regional trade liberalization leading to 

further production fragmentation, finer specialization, and increased intra- and extra-

regional trade.  

 

In Table 3, the estimation results are by the HS 1-digit category. In contrast to the 

previous results, the trade diversion elasticities are negative for half the industry 

categories, and significantly negative at the 5% level for HS1 (Vegetable Products and 

Foodstuffs) and HS4 (Raw Hides, Skin, Leather and Fur, and Wood and Wood Products). 

As HS1 represents industries with relatively little production fragmentation, it is not 

surprising that AFTA had trade-diversion effects in this category. As for the trade-

---

 

 

 

8

 

creation elasticities, again, about half the industry categories show a negative sign, but 

there are statistically significant results only among the positive estimates (i.e. HS4 and 

HS8). As such, the net trade creation effect for HS4 was 0.07 (=0.212-0.141) and for 

HS8, it was 0.213 (=0.283-0.077). The HS8 category includes industries such as Metals, 

Machinery, Electrical Products, and Transportation, which are mainly industries with 

complex and intense production fragmentation. HS8 is also the largest HS 1-digit 

category in terms of import value for ASEAN, accounting for about 45% of total ASEAN 

imports. As such, this result implies that AFTA had a positive and significant welfare 

effect on the bulk of ASEAN trade.   

 

As robustness checks, I first test for the presence of heteroskedasticity by using RESET 

and find that the conditional distribution of the data is indeed heteroskedastic. As such, I 

am justified in not using the OLS Within estimator but the Fixed Effects PQML 

estimator. Next, I use a Park-type test (see Santos-Silva and Tenreyro, 2006, p.646 ) to 

check whether OLS estimation of my empirical model would be consistent. The test 

rejects the use of OLS estimation, implying that the PQML estimator should be used for 

consistency.

x