Estimation of Partially Linear Panel Data Model With Double/Debiased Machine Learning

ABSTRACT

This paper studies the estimation of partially linear static/dynamic panel models with two‐way fixed effects. The nonlinear component is estimated by machine learning. We extend the cross‐sectional Double/Debiased Machine Learning (DML) framework to panel data to deal with the bias of the machine learning estimator. The parameter of the linear component is estimated by Generalized Method of Moments (GMM) based on Neyman orthogonal conditions and cross‐fitting. We establish the asymptotic properties of the proposed DML‐GMM estimator under mild conditions, and provide a consistent estimator for its asymptotic variance. Simulations with complex settings show that the DML‐GMM estimator consistently outperforms other common estimators. Lastly, we apply the proposed method to investigate the effect of K‐12 school openings on the spread of COVID‐19 in the US.