Estimating Gravity Equation Models in the Presence of Sample Selection and Heteroskedasticity

Gravity equation models are widely used in international trade to assess the impact of various policies on the patterns of trade. Although recent literature provides solid micro-foundations for the gravity equation model, there is no consensus on how to estimate a gravity equation model in the presence of the two stylized features of trade data: frequent zeros and heteroskedasticity. We propose a Two-Step Nonlinear Least Square estimator that satisfactorily deals with both problems. Monte-Carlo experiments show that the proposed estimator strictly outperforms the Poisson Pseudo Maximum Likelihood (PPML), the Heckman sample selection model, and the E.T.-Tobit estimators, and that it weakly dominates the Truncated PPML model in the estimation of the intensive margin of trade. An empirical study of world trade in 1986 suggests that currency union and regional trade agreements facilitate trade primarily through improving market access, as opposed to intensifying pre-existing trade.