Dear Forum,

I have a panel dataset of US imports from 130 exporters over 48 years. I am running two models on the same dataset: Model 1 looks at export diversity at the extensive margin (#of exported products) as the dep var and Model 2 looks at export concentration (Herfindal-Hirschmann Index) as the dep var.

In Model 1, the dep var is a count variable, is highly skewed and therefore I used a poisson estimation.

In Model 2, the dep variable HHI instead takes on values between 0 and 1 and has a different distribution, closer but still very far from normal. (Mean = 0.36; Variance = 0.069)

Am I correct to say the OLS model is better suited to Model 2? Given HHI does not follow a Poisson, nor is it normally distributed. The standard errors are approx 65% smaller for the OLS.

I would go straight ahead with the OLS, however I have read that Santos Silva and Tenreyro (2006) that Poisson is also very robust to non-Poisson distributions.

Thank you in advance.

Kind regards,
Ray