Poisson regression controls

[Camilo Castro]

Dear all,

I am looking for a little bit of assistance on a project of trade. I started building the dependent variable to be estimated which is the number of occurrences of an event between two countries.

Then through a gravity model, I proposed the equation:

y_ijk= Irate_it + Irate_jt + GDPpc_it + GDPpc_jt + GDist_ij + Home_i + Host_j +2009

y is the number of transactions made from the same country (i) to the same country (j) in the a year (t) or the sum of the value of the transactions.

Irate is the interest rate of the country (i) or (j) in the year (t)
GDPpc is the GDP per capita of the country (i) or (j) in the year (t)
Home is a dummy variable of the country (i)
Host is the dummy variable of the country (j) (But is coded in 4 groups because there are 65 different countries)
2009 is a dummy variable to capture the effects of the economic crisis

The first question is: Am I capturing the variance of the relationship between i and j countries since the moment I built the dependent variables? Should I use or not the Home and Host control variables?

The second question is: Since the event variable is a positive integer count and must be modeled through a Poisson-type regression, What command is more suited for panel data analysis? PPML or ZTP?

The third question, I have the information of 18 years so I declared the panel using xtset Year But because I have the information of two countries, I do not know what id should I use to declare the panel.

. ztp NMA RINTRATEHOME RINTRATEHOST lnGDPPChome lnGDPPChost lnGDIST i.Home i.Hostgroup AL

Thank you in advance for your help

[Joao Santos Silva]

Dear Camilo Castro,

1. Only you know what you are trying to do, so you have to decide on the appropriate specification.
2. Is it impossible that the number of events is zero? Anyway, just use PPML; I recommend the user-contributed ppmlhdfe command.
3. You may not even have to xtset your data, but if you do, the standard is to do xtset pairid year.

Best wishes,

Joao

[Camilo Castro]

Thank you for your valuable insights Joao. I discussed this with my main advisor and we changed not only the way we are assessing the phenomenon but how we are presenting the results.