Double/Debiased Machine Learning for Treatment Effects in Dynamic Panels

We introduce a novel framework for causal inference in dynamic panel data that extends the cross-sectional Double/Debiased Machine Learning (DML) approach. The method rests on a partially linear dynamic panel model with two-way fixed effects for potential outcomes and readily accommodates binary, multi-valued, or continuous treatments. To translate potential outcome models into an estimable form, we introduce dynamic conditional independence assumptions. By integrating cross-fitting, the nonparametric component—which captures nonlinear confounding—is estimated using machine learning, while the linear parameter is identified via a Generalized Method of Moments (GMM) estimator based on Neyman-orthogonal moment conditions. We define a range of causal estimands and estimate them using a plug-in approach, establishing the consistency and asymptotic normality of the estimators for both model parameters and causal estimands. Simulation studies demonstrate the favorable finite-sample performance of the proposed estimators. Finally, we illustrate the practical utility of the framework through an empirical application examining the causal effect of the number of children on women’s labor supply.