A Clustering Method of Non-Stationary Time Series and Its Application in CSI 300 Analysis
Special Issues Editor (Nottingham Tent University, United Kingdom (Great Britain))
Correlation analysis is the basic work for discovering the inner connection between sequences and has important implications to subsequent work like classification and dynamic analysis. This paper has studied the correlation measurement method of clustering non-stationary time series. Unlike most of the studies using Pearson's correlation coefficient, we propose to calculate the correlation between stocks combing the log-return and Spearman correlation coefficient. Using this method, the stocks in CSI 300, the most popular stock indicator in China, have been clustered and we use three evaluation methods to observe the clustering effect. The result shows this cluster method can come up with a satisfactory classification. Further analysis indicates that the implicit links between stocks, such as equity relationship, are implied in clusters. In general, the correlation between stocks is measured accurately and the proposed method can solve the correlation measurement problem of clustering non-stationary time series more reasonably.
Journal: International Journal of Simulation: Systems, Science & Technology, IJSSST V17
Published: Jul 14, 2016