Title: Bias-corrected robust estimation of dynamic panel data models
Authors: Pavel Cizek - Tilburg University (Netherlands) [presenting]
Abstract: Panel data are increasingly used due to their wider availability. The so-called fixed-effect panel models are difficult to estimate in the presence of outliers especially when the lagged values of the dependent variable are included and the number of time periods is small or moderate. Except for the median-ratio estimator, only locally robust methods based on the generalized method of moments (GMM) adjusted to have a bounded influence function were studied. To design robust regression estimators with positive breakdown points, we first generalize some existing robust regression estimators to preserve their robust properties when applied in dynamic models. However similarly to other linear-regression estimators, the proposed robust-regression estimators exhibit bias when applied to the first-differenced panel data. To address this, we next derive their asymptotic biases to facilitate the bias-correction procedures. We analyze the applicability and robustness of the bias correction procedures based on the derived asymptotic bias and on the computational methods such as jackknife, bootstrap, and indirect inference.