Robust Estimation of Large Panels with Factor Structures

This article studies estimation of linear panel regression models with heterogeneous coefficients using a class weighted least squares estimators, when both the regressors and error possibly contain common latent factor structure. Our theory is robust to specification such structure because it does not require any information on number factors or itself. Moreover, our efficient, in certain circumstances, nests GLS principle. We first show how unfeasible weighted-estimator provides bias-adjusted estimator conventional limiting distribution, for situations which OLS affected by first-order bias. The technical challenge resolved consists showing these properties are preserved feasible double-asymptotics setting. illustrated extensive Monte Carlo experiments an empirical application that investigates link between capital accumulation economic growth international Supplementary materials this available online.