IDENTIFICATION AND ESTIMATION OF A NONPARAMETRIC TIME-VARYING PANEL DATA MODEL WITH COMPLETELY MISSING REGRESSORS IN SOME PERIODS

This article considers a nonparametric time-varying panel data model in which the primary regressor of interest is completely missing in some periods. Under standard assumptions, identification results are established for the unknown functions and the distributions of the missing regressor. Specifically, one period of completely observed data suffices to identify the random effects model, whereas two periods are required for the fixed effects model. Following identification, conditional deconvolution kernel estimators are developed, and their convergence rates are derived. Simulation results show that these estimators perform well in small samples, and an empirical exercise demonstrates the relevance and practical potential of the proposed method.