摘要
本文检测非参数回归模型均值函数结构变点,针对均值函数跃度的长期均值为零时,基于残量的CUSUM统计量对均值函数结构变点检验无效的问题,本文提出了一种基于均值函数的核估计的检验统计量,得到统计量在原假设和备择假设下的极限分布,并构造Bootstrap方法对非参数回归模型均值函数结构变点进行检验,证明了检验和估计的一致性;模拟结果表明本文方法明显优于已有方法。
This paper tests the mean function＇s structural break in nonparametric models. For the residual based CUSUM test has local power equal to size when the long run average mean of the mean function＇s jump is small. A new test based on the mean function＇s kernel estimation is considered, and it＇s limit behaviors under the null hypothesis and alternative are proved. A bootstrap procedure is proposed and the consistency both of the test and estimation are also proved. Simulation results show that our method is more powerful in detecting the structural break of mean function in nonparametric time series.
出处
《数理统计与管理》
CSSCI
北大核心
2011年第4期629-638,共10页
Journal of Applied Statistics and Management
基金
国家自然科学基金(60375003
60972150)
西北工业大学科技创新基金(2007KJ01033)
