The Econometrics of Gravity Models

Melbourne Institute Working Paper No. 05/98

Date: February 1998

Author(s):

Mark N. Harris
László Mátyás

Abstract

Gravity type models have often been used to analyse trade flows between countries and trading blocs. Previously however, these models were only applied to either cross-section data, or to single country time-series data, which imposed severe explicit (or implicit) restrictions on the specification of the model. Recently Gravity models have been generalised and adapted to a panel data setting, where several time-series of cross-section data sets were pooled. This approach not only increases the degrees of freedom, it also enables the proper specification of source and target country effects and time (or business cycle) effects. In this paper, we review in a unified framework, the recent developments in the econometric methodology of Gravity models, and refine the estimation techniques to account for any possible simultaneity bias. Although a fully specified fixed effects Gravity model has been estimated previously, this paper contains the first ever results of its random effects counterpart. We also suggest an extension to the basic model, which accounts for the fact that contemporaneous trade flows are likely to be strongly related to previous ones. Once more, this appears to be the first application of such a model in the literature. Finally, all of these various models and methods are illustrated with an application to export flows in the APEC region. The results clearly suggest that it is important to properly specify the model, in terms of source, target and business cycle effects. If this is not the case, policies could be instigated that do not take into account, for example, that some countries have “naturally” higher propensities to import than others. Moreover, if these effects are not properly specified the affect of other important driving factors, e.g. population will be wrongly estimated. In both cases, policy will be misguided. Important explanatory variables are found to be domestic and target country GDP, and dependent upon specification, local and domestic population, the exchange rate and foreign currency reserves. Also, there is strong evidence that current export flows are highly correlated with those of the previous year.

Download Paper