非平稳和长记忆时间序列主频率估计方法研究  

Study on Estimation of Dominating Frequency of Non-Stationary and Long Memory Time Series

作　　者：徐梅[1] 张世英[1] 樊智[1] 

机构地区：[1]天津大学管理学院,天津300072

出　　处：《天津大学学报（自然科学与工程技术版）》2003年第4期507-511,共5页Journal of Tianjin University：Science and Technology

基　　金：国家自然科学基金资助项目(70171001).

摘　　要：许多经济时间序列表现出非平稳性和长记忆性,对这样的序列进行谱分析并估计其主频率,采用传统的谱分析方法已不适用.基于平稳和非平稳分数阶差分自回归滑动平均模型,研究了具有非平稳性和长记忆性序列的主频率的估计方法,同时提出了包括模型定阶在内的基于极小化残差平方和的近似极大似然参数估计算法.通过对上海和深圳证券交易所综合指数收益序列的分析,说明了此方法的可行性和有效性.Many economic time series are nonstationary and have long memory. Traditional spectral analytical methods are not fit for the spectral analysis or the estimation of the dominating frequency of such series. The estimation method of the dominating frequency of such series is studied on the basis of the invertible short and long memory autoregressive integrated moving average model. An algorithm including the determination of the model orders on approximate maximum likelihood parameters estimation method of minimizing the sum of the squared residuals is put forward. The method suggested is proved to be effective and reasonable by the analysis of the income series of aggregate index of Shanghai and Shenzhen Stock Markets.

关 键 词：经济时间序列 非平稳序列 长记忆序列 主频率 估计方法 谱分析 差分自回归滑动平均模型 近似极大似然估计 

分 类 号：F224.7[经济管理—国民经济]