Publisher:
University of Göttingen, Center for European, Governance and Economic Development Research (cege), Göttingen
Abstract:
This paper evaluates the performance of alternative estimation methods for multiplicative and log models with heteroskedasticity. Contrary to Santos Silva and Tenreyro (2006), the results of a simulation study indicate that the Pseudo Poisson Maximum Likelihood estimator (PPML) is not always the best estimator. New estimates of the gravity equation are obtained for three different datasets with traditional methods (OLS and FGLS) and with the PPML. We find that the PPML assumption concerning the pattern of heteroskedasticity is, in most cases, rejected by the data and PPML estimates are outperformed by OLS and FGLS estimates in out-of-sample forecast.