Kernel change point detection based on convergent cross mapping

Abstract

In this paper, a new method called kernel change point detection based on convergent cross mapping (KCP-CCM) is proposed to address the causality change detection problem. Specifically, KCP is used for running CCM statistics computed by sliding windows based on CCM. Thus, the main goal of this paper is to introduce KCP-CCM as a multivariate change point detection method to capture the evolution of complex systems. To this aim, we choose three types of data sets with different construction methods for simulation experiments to show that KCP-CCM offers a substantial gain in performance over KCP based on correlation and transfer entropy in chaotic system such as the deterministic model. In addition, KCP-CCM, as a causal measurement method, effectively captures the asymmetric dependence between subsystems and provides more accurate information for decision-makers. Finally, we apply it to the financial system under the influence of the 2008 financial crisis to supplement the research on the linkage of Chinese and American stock markets.