Bias Reduction in Nonlinear and Dynamic Panels in the Presence of Cross-Section Dependence

Fixed effects estimation of nonlinear dynamic panel models is subject to the incidental parameter issue, leading to a biased asymptotic distribution. While this problem has been studied extensively in the literature, a general analysis allowing for both serial and cross-sectional dependence is missing. In this paper we investigate the large-N,T theory of the profile and integrated likelihood estimators, allowing for dependence across both dimensions. We show that under stronger dependence types the asymptotic bias disappears, but a Op(1∕T) small-sample bias remains. We provide bias correction and inference methods, and also obtain primitive conditions for asymptotic normality under various dependence settings.