Two-Way Clustering with Non-Exchangeable Data

Abstract

Inference procedures for dyadic data based on two-way clustering rely on the data being exchangeable and dissociated. In particular, observations must be independent if they have no index in common. In an effort to relax this we consider, instead, data where Yij and Ypq can be dependent for all index pairs, with the dependence vanishing as the distance between the indices grows large. We establish limit theory for the sample mean and propose analytical and bootstrap procedures to perform inference.

Keywords

Reference

Koen Jochmans, “Two-Way Clustering with Non-Exchangeable Data”, TSE Working Paper, n. 26-1701, January 2026.