Kengoay. Pdf

ij

ijk

k

ijk

k

ijk

k

ij

ij

j

i

ij

j

i

j

i

ij

u

exRTA

imRTA

RTA

ADJ

COM

PGDP

PGDP

D

PGDP

PGDP

GDP

GDP

M

β

β

β

β

β

β

β

β

β

β

β

β

 

 

 

 

 

 

 

 

 

(2) 

where  M

ij

 is the US dollar value of imports of country  i from trade partner  j.

12

  

GDP

i(j)

 is country  i  (j)’s GDP,  PGDP

i(j)

 is country  i  (j)’s per capita GDP,  D

ij

 is the 

distance between capital cities, COM

ij

 is a complementarity index between countries i 

and  j,  ADJ

ij

  is a dummy variable that is 1 is two countries share a common land 

border and 0 otherwise,  RTA

ijk

  is 1 if both countries  i and  j belong to RTA  k and 0 

otherwise, similarly  imRTA is 1 if only the import country  i belo ngs to RTA  k and 0 

otherwise and likewise, exRTA is 1 if only the export country  j belongs to RTA  k and 

0 otherwise.  The RTA’s considered in this study are ASEAN, EU, NAFTA and 

APEC.

13

  Finally, u

ij

 is the log normally distributed error term, where E(log u

ij

)=0. 

 

We estimate several specifications of equation (2).   To  enable us to make 

comparisons before and after the AFTA process started as well as prior to, and 

following the Asian crisis our data cover the period 1982 to 1999.  We provide 

estimations for  six distinct time periods, three five- year periods 1983-1987, 

1988-1992, 1993-1997 as well as 1998-1999 and two summary periods 1983-1997 

and 1993-1999.

14

 

                                                 

12

 The original gravity model has exports as the dependent variable.  Equation (2) was 

estimated for exports, imports and total trade with similar results.  For reasons of data 

accuracy we report the results with imports as our dependant variable. 

13

 Note that some RTA’s increase their membership over time.  In this study however, 

each regional group is defined for a consistent country membership.  See Appendix B for 

a list of countries included in our ASEAN, EU, NAFTA and APEC dummies. 

14

  The pooling  of the data has the effect of smoothing the effects of business cycles, 

economic shocks and trade imbalances that could affect any given year.  All results were 

---

 

12

 

There are two excluded variables that require further discussion.  The first is the real 

exchange rate.  If our regressions were simple yearly cross sections then real 

exchange rates are not relevant as it is not possible to tell whether a currency is over 

or undervalued.  With pooled data however, competitiveness via the real exchange 

rates matters.  If  we are to take account of the effect of the Asian crisis on trade 

relationships it would seem appropriate to have a measure that can pick up the effect 

of changes in the real exchange rates over the period of study.  In this paper we 

experimented with a number of real exchange rate variables although the existing 

gravity literature provides limited guidance.  Our approach was to include a single 

variable where country  i’s real exchange rate relative to country  j  was defined as 

country  i’s local currency value of one unit of country  j’s currency multiplied by  j’s 

GDP deflator and divided by country i’s GDP deflator where i is the importer and j is 

the exporter country.  See Appendix C for a graphical representation.  This is 

similar to Soloaga and Winters (2001) who include two variables, one for each 

country where country  i(j)’s real exchange rate was defined as the local currency 

value of one US$ multiplied by the US GDP deflator and divided by country  i(j)’s 

GDP deflator.  In both cases the means over our periods are set to zero so that 

movements relative to the mean reflect real exchange rate effects.  The inclusion of 

our variable made little difference to overall results while the Soloago and Winters 

(2001) results were often inconsistent.  The mixed evid ence from previous studies 

makes the results using real exchange rates questionable and are thus, not reported in 

this paper.

15

  The other variable that is usually included in gravity equations is a 

                                                                                                                                            

re-estimated using 1993-1996 instead of 1993-1997 but no discernible differences were 

observed. 

15

 Results of equation (2) including our exchange rate variables are available from the 

author upon request.   

---

 

13

common language dummy but as we are primarily interested in a group of countries 

that all have their own distinct language the results are not reported. 

 

The structure of equation (2) differs from the standard gravity model in two main 

ways.  First, we include an index of complementarity.  One of the characteristics of 

the basic gravity equation is that it does not explicitly include a factor endowment 

variable as, although income level differences reflect factor endowment differences, 

they may also explain product differentiation or demand dissimilarity (see e.g.  

Deardorff 1984 and Frankel 1997).  A complementarity index (COM

ij

) based on 

Drysdale (1967) is included to directly capture factor endowment differences and is 

given by; 

 

(

)

(

)

[

]

(

)

{

}

∑

−

−

=