French (2011) can analytically show that the standard Anderson and van Wincoop (2003) gravity trade model is only correctly specified for disaggregate data; gravity trade model analysis should be done at product level and then estimation results should be reaggregated. If however gravity trade model analysis is to be done at product level then also estimation issues in disaggregate gravity trade models should come to the fore. As is shown, previous estimators suffer under different statistical problems. This paper proposes a zero-in ated Poisson Quasi-Likelihood (PQL) and a Gamma Two-Part Model (G2PM) as reliable alternatives. Estimated within a Generalised Estimating Equation (GEE) framework, both estimators are consistent and have more or less conservative test statistics. Further, for model selection a Quasi-Likelihood under the Independence Model Criterion (QIC) is recommend since this statistic is conform with GEE approaches. Both estimators PQL and G2PM and the model selection technique QIC should become standard tools for disaggregate gravity trade model estimation.
gravity model excess zeros Poisson Quasi-Likelihood Gamma Two-Part Model Generalised Estimating Equation Approach