固定效应模型截面相关性检验的新方法  

A New Method for Testing Cross-sectional Dependence in the FE Panel Data Model

作　　者：徐凤[1] 黎实[2] 

机构地区：[1]西南财经大学经济数学学院 [2]西南财经大学统计学院

出　　处：《数量经济技术经济研究》2012年第6期115-126,139,共13页Journal of Quantitative & Technological Economics

基　　金：国家自然科学基金项目(71072160)的资助

摘　　要：本文考虑累积误差和有限样本问题,构造了固定效应模型截面相关性的新型检验统计量SchottP。研究表明,累积误差情形下,Schott(2004)的检验统计量存在与N/T和槡N/T^1/2相关的渐进偏差,引起检验统计量水平的扭曲和势的下降。据此,本文对存在的渐进偏差进行了修正,并采用回归的办法得到具体修正项的系数。本文提出的固定效应模型截面相关性检验统计量Schottp,在有限样本条件下,具有渐近正态分布以及良好的统计检验水平和势,并能用于异方差情形下的截面相关性检验。Considering accumulative error and finite sample problem, we pro- pose a new testing statistic for cross-sectional dependence in the FE panel data model. The results show that using Schott＇s statistics for testing cross-sectional dependence in a FE model, accumulative error will appear and produce asymptotic bias of the testing statistic in finite sample. The bias is found to be constants related to N/T and √Z-N/T. In view of this, this article suggests a bias-corrected Schott＇s test for cross-sectional dependence in FE model, and the bias-corrected coefficients are obtained by the linear regression. In the finite sample, the bias-corrected Schott＇s statistic named Schottp, converges to standard normal distribution asymptotically and performs well with respect to the size and the power, even under the het- eroscedasticity.

关 键 词：截面相关 累积误差 固定效应 有限样本 

分 类 号：F224.0[经济管理—国民经济]