Inference without smoothing for large panels with cross-sectional and temporal dependence

This paper addresses inference in large panel data models the presence of both cross-sectional and temporal dependence unknown form. We are interested making inferences that do not rely on choice any smoothing parameter as is case with often employed “ H A C ” estimator for covariance matrix. To end, we propose a cluster asymptotic estimators valid bootstrap schemes require selection bandwidth or accommodate nonparametric nature dependence. Our approach based observation spectral representation fixed effect model such errors become approximately temporally uncorrelated. proposed can be viewed wild bootstraps frequency domain. present some Monte Carlo simulations to shed light small sample performance our inferential procedure.