Best Statalisters

I am trying to estimate a gravity model for trade effects of trade preference scheme GSP+ accession for my bachelor's thesis. I am using panel data for 49 exporters and 15 importers.
I want to run regressions using OLS, PPML (xtpoission) and Negative Binomial (xtnbreg), for all three doing 1) one without any FE's, 2) one with country and year fixed effects, and 3) one with country-pair and year effects. I have generated time-specific yearly dummies, country-spec dummies and pair-spec dummies. I am however running into trouble, getting unexpected results and STATA saying "cannot compute an improvement -- discontinuous region encountered" for both xtpoisson and xtnbreg. Perhaps you could help me out explaining what I do wrong?
Please note; I am very new to stata.

The commands I have tried for xtpoission and xtnbreg are:
1) xtpoisson without any FE's:
[xtpoisson imports LnGDPexp LnGDPimp Lndist LnPOPexp LnPOPimp contig comlang_off col45 smctry GSPplus, vce(robust)]

2) xtpoisson with country-spec and year FE's:
[xtpoisson imports LnGDPexp LnGDPimp Lndist LnPOPexp LnPOPimp contig comlang_off col45 smctry GSPplus FE_imp* FE_exp* FE_year*, vce(robust)]

3) xtpoisson with pair-specific and year fixed effects:
[xtpoisson imports LnGDPexp LnGDPimp Lndist LnPOPexp LnPOPimp contig comlang_off col45 smctry GSPplus FE_year*, fe vce(robust)]

4) xtnbreg without any FE's
[xtnbreg imports LnGDPexp LnGDPimp Lndist LnPOPexp LnPOPimp i.contig i.comlang_off i.col45 i.smctry i.GSPplus, vce(oim)]

5) xtnbreg with country-spec and year FE's:
[xtnbreg imports LnGDPexp LnGDPimp Lndist LnPOPexp LnPOPimp i.contig i.comlang_off i.col45 i.smctry i.GSPplus FE_imp* FE_exp* FE_year*, vce(oim)]

6) xtnbreg with pair-specific and year fixed effects:
[xtnbreg imports LnGDPexp LnGDPimp Lndist LnPOPexp LnPOPimp i.contig i.comlang_off i.col45 i.smctry i.GSPplus FE_year*, fe vce(oim)]



Joao Santos Silva


Many thanks in advance,
Hannes