空间面板参数估计的小样本特性探究

Study on the Small Sample Property of Spatial Panel Estimation and its Test Statistics

本文通过蒙特卡罗模拟方法比较了GMM、QML和固定效应空间面板(SOLS)参数估计方法和相应模型的检验功效。模拟的结果表明:在参数估计的有效性与一致性方面,小样本情况下GMM估计优于QML和SOLS估计;空间效应的识别方面,LM检验能够有效地识别空间效应及相应的模型形式,而LR检验的功效比较底。Wald检验能够有效识别空间Durbin模型的潜在形式。在小样本情况下Hausman检验易于选择固定效应模型而不是随机效应模型。据此提出了空间面板数据模型实证研究中的对策建议。

Based on extensive Monte the small sample property of the GMM, well as its counterpart spatial fixed effect Carlo simulation, this paper investigates Quasi Maximum Likelihood （QML） as （SOLS） estimation methods. The related test power and test size property for the identification of spatial panel data are also included in the paper. The simulation results are as follows, the GMM method is superior to QML and SOLS in terms of effectiveness and consistence. As for test power and test size of the related spatial panel statistics, the simulation results reveal that the LM test can identify the spatial panel model more effectively than counterpart of LR test. The Wald based test for spatial Durbin model also has higher test power. The spatial Hausman test has the tendency of accept spatial fixed effect model while reject the spatial random effect model under small sample condition. Finally, this paper concluds with suggestion for carrying out empirical research based on these spatial panel methods.