Estimating gravity coefficients with multiple layers of heterogeneity

We estimate a gravity model for 205 countries, with data from 1954 to 2014, allowing for multiple layers of heterogeneity. The first layer arises from the interactions between a set of traditional gravity variables. The second one comes from country pairs that differ in values of the binary time-independent gravity variables. Further layers come from different income groups and regions. Our results show the importance of heterogeneities at various levels. For instance, landlockedness is less restrictive when trading partners are contiguous or colonially linked. We also find, for example, that the positive interaction between landlockedness and contiguity is more (less) pronounced when non-high-income countries import (export) or when Middle East and North Africa (MENA) countries import, and the negative interaction between landlockedness and language similarities is stronger when MENA countries export.