基于残差的分位数回归模型结构突变检验  

Residuals-based Quantile Regression Model Structural Mutation Test

作　　者：伍兴国[1] 肖霞[1] Wu Xingguo;Xiao Xia(Dongguan Polytechnic,Dongguan Guangdong 523808,China)

机构地区：[1]东莞职业技术学院,广东东莞523808

出　　处：《统计与决策》2020年第14期32-36,共5页Statistics & Decision

基　　金：东莞市老年社会工作服务及评估项目(政201724);金融与区域经济发展科研创新团队项目(201804)。

摘　　要：CUSUM等基于残差的方法来检验回归模型的结构突变,其是在最小二乘法下进行的,仅能检验条件分布均值下突变。因此,文章将其扩充到分位数回归中,提出了基于分位数回归残差的检验统计量,同时给出了它的极限分布,并进行了蒙特卡洛模拟。模拟结果显示:该统计量具有良好的检验水平和检验功效;如果回归模型是位置突变,检验的分位数越靠近中位数,其检验功效越强;如果回归模型是尺度突变,则刚好相反。The methods of CUSUM et al.based on residuals used to test the structural mutation of the regression model,which are carried out under the least square,can only test the mutation under the mean value of conditional distribution.Therefore,this paper extends it to quantile regression and proposes a test statistic based on quantile regression residuals.At the same time,the paper gives the limit distribution and performs the Monte Carlo simulation.The results show that the statistic has good testing level and testing efficacy,that if the regression model is a positional mutation,the closer the quantile of the test is to the median,the stronger the test efficacy will be,and that if the regression model is a scale mutation,the reverse is true.

关 键 词：分位数回归 Kiefer过程 残差 结构突变 

分 类 号：F224.0[经济管理—国民经济]