基于Sieve Bootstrap方法的长记忆过程均值变点的检验

Sieve bootstrap test for mean chang in long memory time series

本文提出一个新的统计量来检测长记忆时间序列中可能存在的均值变点,在原假设下推导出了检验统计量的极限分布在备择假设下证明了检验方法的一致性.为便于实际应用还提出了一种Sieve Bootstrap方法来近似统计量的临界值.模拟结果表明本文方法不仅可以很好的控制经验水平,而且相比已有均值变点检验的方法经验势也有了一定幅度的提高.

In this paper,we propose a new statistic to detect for a change point in the mean under long memory,The null distribution of the test statistic is derived and the consistency is proved under the alternative hypothesis. In order to facilitate the practical application, we present a sieve bootstrap procedure which can give asymptotic correct critical value. Simulations indicate that the proposed methods can not only control the empirical size well but also test power improved compared with the existing detect method of mean change point.