关键词: 动态面板, 蒙特卡洛模拟, 弱工具变量, 检验功效

Abstract: Dynamic panel estimation methods are widely applied in all the fields of Economics. This paper is to evaluate the performance of different estimation methods for dynamic panel data and their test power of these test estimators from the Monte Carlo Simulation. The conclusion shows the Sys-GMM and Diff-GMM estimation methods which are used extensively have obvious estimation biases for small samples, together with their misleading test power. The biases are more significant when the fix effect variance ratio becomes larger. On this occasion, the bias correction and maximum likelihood estimation methods can be more efficient in parameter estimation. When the dependent variable is truncated, the Sys-GMM and Diff-GMM also have larger estimation bias comparing with other approaches, but the transformed Tobit model can offer a robust estimator. The larger the fix effect variance ratio is, the more robust the test power of LM and CLR with weak instrumental variables has. The robustness of the Monte Carlo Simulation is justified in the case study of inter-regional labor migration effect on regional wage differentials by means of different dynamic panel estimation methods.

Key words: Dynamic Panel data, Monte Carlo Simulation, Weak Instrumental Variables, Test Power