Multi-way clustering estimation of standard errors in gravity models

•Illustrates the role of multi-level clustering in structural trade gravity models.•Shows that using simple robust standard errors generates misleading inference.•Shows that multi-level clustering makes a big difference in practice.

This paper analyzes the consequences of ignoring the multi-indexed structure with cross-sectional and panel-data gravity models of bilateral trade for inference. We estimate log-linear and generalized-linear gravity equations of bilateral trade. Ignoring multi-way clustering in the data at hand leads to misleading inference regarding the relevance of preferential trade-agreement memberships of different kinds.