Dear Tram,

It appears that GDP is in real form, and PGDP are in value form. The two measures are colinear, so I would try to use only one of them.

Very interesting model.

Why you wouldn't use other variables like common language.. Membership in WTO

Ideally, we want them to be significant. If they are not significant, which means we can easily drop them from the model, then the model becomes no longer a gravity model.

Even without knowing the exact definition of your data, it is clear that it is a terible specification of the equation. If Lall is correct, that GDP and PGDP are real and nominal Gross Domestic Products, respectively, that he is correct, too, that there are strong multicollinearities involved - leading to spurious coefficients. If PGDP is the GDP price index (deflator), then it is even worse: a price increase in one of the countries would lead to the same effect on exports independet of if the inflation happens in country i or j. In general, one would assume that a price increase in i would lead to a reduction of (more expensive) exports to j.

Another problem is the (cor)relation between Y and GDP and openness. Y is part of GDP, therefore there are two-sided effects: Y==>GDP and GDP==>Y (simultaneous equation bias, which may be considerable for a big neighbour country). As openness is usually defined as (exports+imports)/GDP, there are several correlations within this equation.

If afta is a dummy with 1 if ASEAN and 0 otherwise, the coefficient should not be significantly negative, because one would expect a higher trade between ASEAN countries. The same is true, when border is a measure of a common border (either 1 for a common border or the relative length of the common border).

I think it is necessary to put more effort in a careful specification and to find transformations, which avoid collinearities as far as possible.