异方差问题伪检验研究  

Research of Spurious Test Concerning Heteroscedasticity

作　　者：刘田[1] 谈进[2] Liu Tian;Tan Jin(School of Statistics,Southwestern University of Finance and Economics;School of Economic Information Engineering, Southwestern University of Finance and Economics)

机构地区：[1]西南财经大学统计学院 [2]西南财经大学经济信息工程学院

出　　处：《数量经济技术经济研究》2018年第8期140-156,共17页Journal of Quantitative & Technological Economics

基　　金：教育部人文社会科学研究规划基金项目(15YJA910003);中央高校基本科研业务费专项资金创新团队项目"时间序列前沿研究团队"(JBK150501)的资助

摘　　要：研究目标:探讨异方差相关的伪检验及参数可分辨问题。研究方法:运用理论分析和蒙特卡洛仿真方法。研究发现:同方差或异方差情形使用错误异方差的WLS的假设检验是错误的,易导致伪回归,常比直接使用OLS更糟;异方差时虽估计是一致的,即使大样本也不可直接使用OLS进行假设检验,因得到伪回归或将显著参数误判为不显著的可能性不会随样本增大而改善;稳健方差估计可得到OLS实际方差的较好估计,对原假设误判通常无影响,但参数分辨力变差,检验功效明显下降;WLS的BLUE性很不稳健,只识别出引起异方差的变量而不了解其完整结构没有意义。研究创新:推导出参数分辨率公式,发现避免异方差伪检验的思路。研究价值:为正确评价存在异方差的模型提供了有益的建议。Research Objectives:This article addresses the circumstances which give rise of spurious test concerning heteroscedasticity and identification of a parameter which can be distinguished.Research Methods:Theoretical analysis and Monte Carlo simulation.Research Findings:In the situation of homoscedasticity or heteroscedasticity,the hypothesis test of WLS using improper heteroscedasticity assumption is wrong and prone to spurious regression,often worse than the direct use of OLS;Although the estimator is consistent in heteroscedasticity,the hypothesis test cannot be used directly even if samples are large,because the possibility of obtaining spurious regression or mistaking the significant parameter for insignificant one is unable to be improved with the increase of sample size;Robust variance estimation can obtain good estimation of actual variance of OLS,usually having no effect on misjudgment of the null hypothesis,but the ability of parameter distinguish becomes worse and the test power decreases obviously;The BLUE property of WLS is not robust,it makes no sense to only recognize the variables without knowing the complete structure that causes heteroscedasticity.Research Innovations:The formula of parameter resolution ratio is derived,and the idea of avoiding pseudo test with heteroscedasticity is offered.Research Value:This paper provides useful suggestions for the correct evaluation of models with heteroscedasticity.

关 键 词：异方差 参数分辨率 检验功效 伪检验 蒙特卡洛仿真 

分 类 号：F224.0[经济管理—国民经济]