FE PPML Gravity Model

[Marcelo Dolabella]

Dear All,

I have two questions concerning the gravity model of international trade estimated using a Fixed Effects PPML.

1) The Stata command xtpoisson ... , fe robust applies Hausman, Hall & Griliches (1984) procedure of conditional maximum likelihood in order to eliminate the country-pair fixed effects. As argued by Wooldridge (2010) in page 675, when y_it = 0 for all t, the cross section observation i does not contribute to the estimation.
Am I facing a selection bias problem if I estimate a fixed effects PPML gravity equation at a disaggregated level with a high incidence of country-pairs that did not trade within the panel? In other words, will the zero trade issue still be a problem while using PPML with a fixed effects panel framework?

2) The estimation of parameters by Maximum Likelihood requires that observations of the dependent variable are statistically independent from one another, which is not likely when considering trade flows. Is this independence assumption not relevant given that the PPML allows the Likelihood function to be misspecified? Or is this an issue while estimating FE Panel PPML?

Any comments are highly appreciated. Thank you in advance,

Marcelo


Hausman, J., Hall, B. H., & Griliches, Z. (1984). Econometric Models for Count Data with an Application to the Patents-R & D Relationship. Econometrica: Journal of the Econometric Society
Wooldridge, J. M. (2010). Econometric analysis of cross section and panel data. MIT press.

[Joao Santos Silva]

Dear Marcelo,

1) There is no selection bias; as you noted, the observations that are dropped are not informative about the parameters of interest, so whether you include them or not is irrelevant.

2) There is no issue with that; please check Wooldridge's paper on the fixed effects Poisson regression, or the summary in the book you mentioned.

Best wishes,

Joao

[Marcelo Dolabella]

Thank you João,

I will take a closer look at Wooldridge's work.

All the best,

Marcelo