中文摘要：

文章通过严格的理论证明发现，传统的长记忆检验理论在非平稳框架下失效了。Andrew虽然将长记忆检验原假设拓展到混合性，但并未放弃平稳性假设：虽然考虑到了序列相依性对传统检验统计量的影响，却并未发现非平稳性的影响。文章给出更一般意义的零假设，不同于其他研究之处，拓展后的零假设更符合实际；并在Andrew的基础上提出一种非平稳稳健且更具一般意义的检验方法。最后运用该方法讨论了中国股票市场中的长记忆性，从而解释了长记忆实证检验结果令人疑惑的地方，该结论对金融领域基础研究具有重要价值。

英文摘要：

We discuss the long memory theory under the nonstationary framework, and test long memory characteristics of China stock market by the adjusted method. Based on the limitations of the existing null hypothesis, we put forward a more general null hypothesis which conforms to reality better. This is different from other researches. Through a strict and theoretic test, we find that the traditional statistics is out of effect. Based on Andrew, we put forward a nonstationary robust and general long memory testing approach. Finally, we discuss the long memory characteristic of China stock market. The conclusions have great practical value and provide a more scientific model for us to analyze the long memory characteristics of stock market.