基于小波聚类方法的股票收益率序列时间模式挖掘  

Mining Temporal Patterns of Stock Yield Sequences Based on Wave Cluster Method

作　　者：陈佐[1] 谢赤[1] 陈晖[1] 

机构地区：[1]湖南大学工商管理学院,湖南长沙410082

出　　处：《系统工程》2005年第11期102-107,共6页Systems Engineering

基　　金：国家社会科学基金资助项目(03BJY099);教育部博士点专项科研基金资助项目(20020532005);全国高校青年教师奖励基金资助项目

摘　　要：时间模式挖掘是指在重构的相空间中搜索能表征和预测的事件的区域。针对股票收益率序列重构相空间,以累计收益和累计密度作为聚类指标,应用小波聚类算法对序列进行时间模式挖掘。实证结果表明,以时间模式预测事件为指导的投资策略能获得高于持有策略的收益;时间模式挖掘能有效识别事件点,事件序列与非事件序列存在显著差别。A temporal pattern is a cluster of reconstructed phase space that is characteristic and predictive of events. This paper reconstructs the phase space based on stock yield sequence. We use WaveCluster method to mine temporal patterns by taking cumulative yield and density as clustering measure. The application shows that the investment strategy directed by events that was predicted from temporal patterns would get the yield higher than buy-and-hold strategy. There is significant difference between the event series and non-event series. Mining temporal pattern could identify event effectively.

关 键 词：小波聚类 时间模式 相空间重构 股票收益率 时间序列 

分 类 号：TP182[自动化与计算机技术—控制理论与控制工程]