Constrained Poisson pseudo maximum likelihood estimation of structural gravity models

This paper reconsiders the estimation of structural gravity models. It introduces a constrained, projection-based Poisson pseudo maximum likelihood estimation procedure (constrained PPML) that exploits the equilibrium conditions introduced by Anderson and Van Wincoop (2003) for estimation and inference. The constrained PPML estimator avoids the estimation of the large number of exporter and importer fixed effects, and provides more reliable inference than the unconstrained PPML estimator. Moreover, based on the delta method the paper drives confidence intervals of counterfactual changes. Monte Carlo simulations yield encouraging results on the performance of the constrained PPML estimator.