Modeling the Effects of Distance and Spatial Dependence in International Trade

The gravity model has been widely used estimate the effect that distance has in international trade; however, two important areas have seen little attention in the literature, namely, the influence of using a more accurate measure of distance and how distance effect estimates change when controlling for spatial dependence in observed trade flows. Using transportation networks to measure distance and estimating both a spatial lag and spatial error gravity model, Canadian provincial exports to the lower 48 states in the United States were analyzed to address these previously ignored issues. It was found that the traditional distance measure of great circle distance underestimated the distance between any given province and state pair by an average of 20%, however, it did so consistently across all pairs, therefore having no significant influence on the distance effect estimate, regardless of estimation technique. When trade was disaggregated into mode of transportation distance effect estimates differed significantly, reflecting the most efficient uses of each transport mode and also the commodities that flowed across them. Finally, when using the spatial lag gravity model, distance effect estimates decreased substantially compared to traditional least squares estimation, while the spatial error model provided asymptotically equivalent parameter estimates to least squares, but with overall increased predictive power.