重尾时间序列环境下持久性变点的稳健检验

Robust test of persistence change in heavy-tailed time series environment

  • 摘要: 本文通过构造基于M估计的Ratio双侧检验统计量研究了方差无穷重尾序列持久性变点检验.证明了在原假设下统计量的渐近分布是布朗运动的泛函,与尾指数无关,并得到备择假设下统计量的相合性.利用Bootstrap抽样方法逼近原假设下统计量的渐近分布以获取精确的临界值.数值模拟结果表明基于M估计的Ratio检验具有良好的经验水平,没有出现显著的扭曲,且相比基于最小二乘估计的检验明显的提高了经验势,尤其是在观察序列尾部特征越厚的情况下.最后通过一组黄金ETF波动率指数数据进一步验证了本文所提方法的有效性和可行性.

     

    Abstract: In this paper, we study the persistence change problem of heavy-tailed observations with infinite variance by constructing Ratio two-sided test statistic based on M-estimates. It is shown that the asymptotic distribution of the statistic under the null hypothesis is functional for Brownian motion, which is independent of the tail index, and its consistency is given under the alternative hypothesis. The Bootstrap sampling method is used to approximate the asymptotic distribution to obtain the accurate critical values. The numerical simulation results show the Ratio test based on M-estimates has a satisfactory empirical size without significant distortion, and significantly improves empirical power compared with the test based on least square estimate, especially when the tail features of time series are thicker. Finally, a set of gold ETF volatility index data verifies the validity and feasibility of our proposed method.

     

/

返回文章
返回