OLS, PPML, xtpoisson, xtqpml

[John Nilsson]

Hi am running a gravity equation with a large number of countries (i.e. 120 reporter and 120 partner). I would like to run an additional gravity model, but by including a country-pair fixed effect. For my OLS model, I ran:

egen pairid = group(ReporterName PartnerName)
tab pairid, gen(pair_)

reg lnexport lndist contig comlang_off comcol fta_wto EUtrade Brexit EUBrexit year_* pair_*, robust cluster(distw)

This, however, does not allow me to run the OLS because of the large number of regressors (120*120), so I have had to drop 40 countries in reporter and 40 countries in partner. Is there another way to do this? and is there a quicker way since this takes a few hours to run? I also have a similar problem with the PPML.

ppml export lndist contig comlang_off comcol fta_wto Brexit EUtrade BrexitEU year_* pair_*, cluster(distw)

I saw somewhere that xtqpml is adviced to run here. I am pretty sure I am not running it correctly though:

xtpqml export lndist contig comlang_off comcol fta_wto Brexit EUtrade BrexitEU year_*, fe i(pairid) cluster(distw)

The problem is that it drops variables of interest. Should I cluster this by pairid instead?

Sum up:
1) how do I run a quicker version of OLS that takes country pair into consideration?
2) How do I run a quicker version of PPML correctly?

Best regards,
John

[Joao Santos Silva]

Dear John Nilsson,

1) you should use xtreg;
2) what you are doing is correct; with pair FE, all the bilateral variables such as distance and contiguity will drop because their effect is absorbed by the FE.

Best wishes,

Joao

[John Nilsson]

Dear Joao Santos Silva,

Thank you so much for all your responses!

You're truly a treasure to this forum!

Sincerely,
John