Linearizing nonlinear gravity models: Biased BvOLS versus unbiased alternatives

Abstract

The use of high-dimensional fixed-effects estimation has become customary with the estimation of gravity models of bilateral trade, migration, or commuting as outcome. However, fixed-effects methods can be used without incidental-parameter bias in a very small set of stochastic models. Alternatives to fixed-effects estimation are iterative-structural model estimation or linearizations of the structural model. Baier and Bergstrand deployed such a linearization. While easy to implement, the approach has drawbacks related to the approximation point and lack of observability of ingredients needed for the linearization. This compromises empirical work. The present paper provides a remedy to this problem by linearizing at the observed trade equilibrium.