平稳性检验方法的有效性研究

Research on the Validity of Stationary Testing Methods

平稳性检验是时间序列分析的重要研究内容,现有检验方法的性能缺乏系统的比较分析。文章从样本长度的视角研究平稳性检验方法的性能,采用ADF检验、PP检验、KPSS检验和LMC检验四种方法展开实证研究。仿真实验结果表明:时间序列数据长度会对检验方法的准确率产生明显的影响,数据长度较小时检验准确率偏低;数据长度增大时可以提升检验方法的准确率,但仍未能达到100%的上限值。当样本长度较小时,这些方法的检验统计量的渐进分布难以满足,因此其实际检验效果值得探究。样本长度是有限的,因此渐进分布检验方式的改进空间有限,新的检验方式值得探究。

Stationary test is one of the important parts of time series analysis. Although there exist some testing methods,systemic comparison of their validity is still lack. This paper studies their validity from the aspect of sample length by using ADF test,PP test,KPSS test and LMC test. The experimental results indicate that the data length of time series will dramatically influence the accuracy of these testing methods,which makes testing accuracy lower when it＇s shorter while improves accuracy through increasing itself but hardly reaches the upper limit of 100%. The progressive distribution of their test statistics is hardly satisfied when the sample length is short,which shows that the actual effect is worth exploring. The length of sample is limited,therefore the improvement of the asymptotic distribution test is restricted,and it＇s still a significant work to explore new test methods.