Econometrics Seminar
Speaker: Xavier d'Haultfoeuille (CREST ENSAE)
Title: Analytic inference with multiway clustering
Format: Hybrid
Homepage: https://faculty.crest.fr/xdhaultfoeuille/
Organisers: Julien, Pietro
Abstract: This paper studies analytic inference with several dimensions of clustering. In such setups, the commonly used variance formula has two drawbacks. First, it is not necessarily positive. Second, it does not lead to asymptotically valid inference in non-gaussian regimes, where the estimator of the parameter of interest is not asymptotically gaussian. We consider a simple modification of the usual variance estimator that addresses both issues. In gaussian regimes, this modified estimator is asymptotically equivalent to the usual one, and thus leads to exact inference. Otherwise, the corresponding inference is asymptotically conservative. This variance estimator also leads to uniformly valid inference over a certain class of data generating processes. Finally, we highlight problems that can affect nonlinear estimators with several dimensions of clustering and discuss the related case of dyadic data.