针对上海和深圳的日收益序列,采用重标级差(R/S分析)对其进行了实证研究.从统计结果来看,样本序列呈现出尖峰、胖尾等有偏特征,明显不满足正态分布的假设,表明收益序列可能具有长程相关或记忆性.采用ARFIMA模型对沪深股市收益率的长期记忆性进行了检验,根据分段检验的结果,得出了一些我国证券市场有效性的结论.
R/S series analysis is widely used as measures for long memory study in time series. As Hurst parameter estimated bias exists, the precision may be improved by using non-linear estimate, where ARFIMA model is proposed and used for verification. Finally long memory characters of Chinese stock index and stock samples are revealed by non-linear R/S analysis.