Analyzing dynamic association of multivariate time series based on method of directed limited penetrable visibility graph

Abstract

In order to study the characteristics of the evolution behavior of the relationship among multivariate time series, this paper proposes a method of constructing Multivariate Time Series-Dynamic Association Network (MTS-DAN) which represents multivariate time series associated relationship in the specific period. Firstly, we adopt transfer entropy algorithm to measure the associated relationship among multivariate time series. Secondly, the temporal behavior of the relationship is constructed into a complex network by the directed limited penetrable visibility graph (DLPVG) method. Thirdly, we explore the potential patterns of multivariate time series according to the physical characteristics of the network. Artificially generated data, SST and financial time series data are as sample separately in this paper. The experimental results reveal some statistical evidences that the associated relationship among multivariate time series is in a dynamic evolution process. There are association patterns among multivariate time series and a few types of patterns play a significant role in the process, while the clustering effect appears in the long-term evolution process. Furthermore, the results also show that multivariate time series have a close relation with actual events, which indicates that the method is of great significance to the research and prediction of events.