长记忆时间序列均值变点的比率检验

Ratio tests for mean change-point in long memory time series

本文基于两种比率方法研究了长记忆时间序列均值变点的检验问题,在无变点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.由于检验统计量的极限分布依赖未知的长记忆参数,还提出了一种避免估计长记忆参数的Sieve AR Bootstrap方法来近似计算检验统计量的临界值,数值模拟结果表明提出的检验方法具有较好的检验效果.最后通过分析一组尼罗河年径流量数据说明了方法的可行性.

This paper applieds two ratio statistics detecting mean change-point in long memory time series.The null distributions of test statistics as well as their consistency are proved.Since the limiting distributions depend on unknown long memory parameter,we propose a Sieve AR Bootstrap method,which can avoid estimating long memory parameter,to compute their critical values.Simulations showed the proposed methods perform well.Finally,we illustrated our methods by a set of annual volume of discharge from the Nile river.