Estimating Gravity Models of International Trade: A Survey of Methods

Abstract

Gravity models, which relate volume of exports from one country to another to the economic size of those countries and various trade costs, have long been successfully used in empirical analysis of international trade flows and their determinants. Recent decades saw a substantial increase in a number of methods to estimate such models due to the emergence of theoretically founded modifications of the gravity equation. This paper surveys these advances in the estimation methodology. First, alternative techniques to account for structural multilate­ral resistance terms (introducing remoteness measures, using price indices, non-linear estimation and its linear approximation, importer and exporter fixed effects) are examined. Variants of gravity specifications on panel data (country-pair fixed and random effects, Hausman – Taylor model) are then reviewed. Next, common errors in gravity modeling associated with atheoretical calculations of bilateral trade and economic size variables are analyzed. Finally, ways to consistently estimate gravity models in the presence of zero trade flows using Poisson regression, Tobit and Heckman models are discussed.

All the methods considered have their own advantages and drawbacks. The choice of estimator in an application is not obvious, and depends on properties of available data and goals of research. A good practice in the modern empirical literature is to utilize simultaneously several techniques, at least for robustness checks.