Title: Time-varying correlation network analysis of non-stationary multivariate time series with complex trends
Authors: Lujia Bai - Tsinghua University (China) [presenting]
Weichi Wu - Tsinghua University (China)
Abstract: A unified approach is proposed for the inference of time-varying cross and autocorrelation curves of multivariate time series, which are observed once at a time and are non-stationary with piece-wise smooth trends. The dimension of the time series and the number of lags of cross and autocorrelation considered are allowed to diverge. The framework enables us to visualize the evidence of connections in the network based on various asymptotically correct multiple hypothesis testing of correlation functions, which is implemented via a difference-based and nonparametric estimator as well as bootstrap-assisted procedures for generating critical values. Therefore, the inferred network is able to capture functional relationships among second-order structures of time series. We prove the asymptotic validity of the correlation network inference procedure, and demonstrate its effectiveness in finite samples by simulation studies and empirical applications in finance.