Centrality bias in inter-city trade

Using Japanese inter-city trade data, we find a substantial centrality bias in aggregate gravity estimations: Shipments from large cities (central places) to their hinterland are 50%–125% larger than predicted by gravity forces. We argue that this discrepancy results from aggregating across industries, that concentrate in a few central places, which predominantly serve their respective hinterlands. Decomposing the centrality bias along the margins of our data, we attribute most of the centrality bias to substantially larger extensive industry margins in exports from larger cities to their smaller hinterland cities than vice versa.