A0698
Title: Robust M-estimation for high dimensional regression on large panel data
Authors: Huanjun Zhu - Xiamen University (China) [presenting]
Abstract: Robust estimation of high dimensional regression coefficient vectors for large panel data is considered. Unlike the high dimensional regression model, cross-sectional dependence makes a robust estimation for large panel data more complicated. To describe cross-sectional dependence, some cross-sectional dependence is allowed even after taking out observed common factors. To achieve robustness against heavy-tailed sampling distributions, a robust M-estimator by assuming regular conditions on general loss functions is considered. The asymptotic Bahadur representation is provided, and simultaneously asymptotic joint distribution for the high-dimensional regression coefficient vector and factor loading vector is established. Thorough numerical results on simulated and real datasets illustrate commonly used M-estimators that outperform the least-squares method on heavy-tailed distributed data.