关键词: 面板数据, 固定效应, 组群异方差, 惩罚伪最大似然估计法

Abstract: Panel data sets are aggregated from time series data of different individuals. A significant amount of studies have shown that there exists grouped structure between individuals in the panel data set, and the model heteroscedasticity is common. Drawing on the study of group heterogeneity, the paper assumes that the variance of the error term has a group structure. Based on the specific assumptions in the model, this paper proposes a penalty pseudo maximum likelihood function estimation method, which can simultaneously perform parameter estimation and structure recognition, and proves the Oracle asymptotic properties of the estimator. Subsequently, Monte Carlo simulation verifies its effective sample properties. Finally, this method is applied to the parameter estimation of the Fama-French three-factor model in the Chinese stock market, which shows the proposed model performs well.

Key words: Panel Data, Fixed Effect, Grouped Heteroscedasticity, Penalty Pseudo Maximum Likelihood Estimation