Flexible Unobserved Heterogeneity in Panel Data: Discretization and Inference

Series

Econometrics Seminars and Workshop Series
Speaker(s)

Wei Miao (KU Leuven, Belgium)
Field

Econometrics, Data Science and Econometrics

Location

Tinbergen Institute, Roeterseiland Campus, E5.22
Amsterdam
Date and time

September 17, 2026
12:00 - 13:00

Abstract

We study inference in panel models with flexible nonseparable unit and time heterogeneity. Rather than imposing additive or factor structures, we approximate latent heterogeneity through discretization. We consider two settings. In high-dimensional short panels, rich covariate information helps recover latent unit heterogeneity when T is fixed. After clustering and group demeaning, we use double machine learning for inference. In nonlinear panels, a grouped-additive approximation creates a trade-off between approximation accuracy and incidental-parameter bias. We derive a bias correction and establish root-NT asymptotic normality. Simulations and empirical applications illustrate the performance of both approaches.