摘要
在某些厚尾分布中,结构突变发生在分布尾部而非中心,传统的均值检验方法可能失效。在线性分位数回归上构建极差检验统计量,并给出统计量的渐进性质。模拟结果表明极差检验统计量是有效的,经验样本的检验水平是合理的,检验功效与断点位置,残差的分布,断点偏差,假设模型的形式,分位数位置,样本量等因素都存在密切关系。
In some heavy tailed distribution,the structural change occurs in the tail rather than the center,and the traditional mean test method may fail.The range test statistic is constructed on the linear quantile regression,and the asymptotic properties of the statistics are given.The simulated results show that the range test statistic is effective and the size of test is reasonable,The power of test is closely related with such factors as breakpoint location,distribution of residuals,breakpoint deviation,the form of model,position of quantile and the sample size.
出处
《衡阳师范学院学报》
2016年第3期15-18,共4页
Journal of Hengyang Normal University
关键词
极差检验
分位数回归
结构突变
empirical range test
quantile regression
structural change