Help with PPML Gravity Model with GDP

[Samuel Malkun]

Hi everyone!

I'm currently running my master's dissertation model where I'm trying to study whether Colombian exports to Venezuela would have been higher if Venezuela's GDP didn't collapse starting 2013. For that purpose I'm running a PPML model with the following syntax: ppmlhdfe export_value gdp_o gdp_d distance contiguity common_language common_colonizer common_legal_origin member_wto_joint agree_fta agree_cu, absorb(countrycode1 countrycode2 year) d. I'm running my model for each SITC section only with exports from Latin America. My results are as follows for section 0.

As you can see, the expected symbols of my variables is consistent with literature, but I'm having problems with the small value for the GDP coefficients. My model subestimates the effect of business cycles, and by observing predicted trade flow between both countries and real trade flows I see that my model fails to account for this effect and my predictions would be better if GDP coefficients are higher. Is there any way I can forcefully to do this? I have read some literature about PPML and random intercepts, but I'm not sure whether I should implement this strategy.

Thanks!

[Samuel Malkun]

Dear Joao Santos Silva ,

Could you please give me a hand. Your insights would be very helpful.

Regards,
Samuel M.

[Joao Santos Silva]

Dear Samuel Malkun,

The estimate for gdp_d has the wrong sign, right? So, there may be something peculiar about the sample you are using. Also, what you have in mind is not a standard use of the gravity equation, so I am not sure what to suggest. Theory-consistent gravity equations would include sets of fixed effects that that I believe you are not including, but doing it may not help you much. I suggest you talk to your adviser about this.

Best wishes,

Joao

[Samuel Malkun]

Dear Joao Santos Silva,

Thanks for your response and time, I appreciate your feedback. I'm currently including a fixed effect for exporter, importer, and year, what other fixed effect would you suggest? I'm not able to include time-variant exporter or importer fixed effect, as I'm interested in modifying GDP and that would be collinear. My methodology consists in imputing alternate GDPs after estimating my model and observing differences between observed and predicted trade, but I discover that after alternate GDP scenarios trade doesn't change that much which I think is odd. Also, I'm trying to study if there was any trade diversion afterwards and for that matter I'm estimating my model by sectors, I would then see if any of those sectors recovered exports to any other country. I reduced my sample to Latin American exports to any destination. Any suggestions regarding my strategy would come very in handy.

Regards,
Samuel M.

[Joao Santos Silva]

Dear Samuel Malkun,

Since you cannot estimate the model with time-varying importer and exporter fixed effects, you may try to estimate it with other sets of fixed effects, for example just pair fixed effects and so on. Again, I suggest you discuss this with your advisor.

Best wishes,

Joao

[Samuel Malkun]

Dear Joao Santos Silva ,

Thank you very much for your time and very helpful tips. I have incorporated a single pair fixed effect for Colombia & Venezuela and my predictions improve greatly. The coefficient for importer's GDP remains with a relatively small magnitude, but I guess that is what the model predicts and I'll have to go with it.

Thanks again,
Samuel.