ESTIMATION ISSUES IN SINGLE COMMODITY GRAVITY TRADE MODELS

Recently gravity trade models are applied to disaggregated trade data. Here many zeros are characteristic. In the presence of excess zeros usual Poisson Pseudo Maximum Likelihood (PPML) is still consistent, the variance covariance matrix however is invalid. Correct economic interpretation however requires also the last. So alternative estimators are looked for. STAUB &WINKELMANN (2010) argue that zero-inflated count data models (i.e. zero-inflated Poisson / Negative Binomial Pseudo Maximum Likelihood (ZIPPML / ZINBPML)) are no alternative since under model misspecification these estimators are inconsistent. Yet zeroinflated Poisson Quasi-Likelihood (PQL) is a reliable alternative. It is consistent even under model misspecifications and beyond that robust against unobserved heterogeneity. Another alternative is a log-skew-normal Two-Part Model (G2PM) which generalises the standard lognormal Two-Part Model (2PM). It is insofar advantageous as it adjusts for (negative) skewness and regression coefficients retain usual interpretations as in log-normal models. PQL is useful for multiplicative gravity model estimation and G2PM for log-linear gravity model estimation. Exemplarily the estimators are applied to intra-European piglet trade to assess their empirical performance and applicability for single commodity trade flow analysis. The empirical part favours PQL but G2PM is a reliable alternative for other trade flow analyses. PQL and G2PM should become standard tools for single commodity trade flow analysis.