Machine Learning Based Panel Data Models

We examine nonparametric panel data regression models with fixed effects and cross-sectional dependence through a diverse collection of machine learning techniques. We add cross-sectional averages and time averages as regressors to the model to account for unobserved common factors and fixed effects respectively. Additionally, we utilize the debiased machine learning method by Chernozhukov et al. (2018) to estimate parametric coefficients followed by the nonparametric component. We comprehensively investigate three commonly used machine learning techniques - LASSO, random forests, and neural network - in finite samples. Simulation results demonstrate the effectiveness of our proposed method across different combinations of the number of cross-sectional units, time dimension sample size, and the number of regressors, irrespective of the presence of fixed effects and cross-sectional dependence. In the empirical part, we employ the proposed machine learning-based panel data model to estimate the total factor productivity (TFP) of public companies of Chinese mainland and find that the proposed machine learning methods are comparable to other competitive methods.