Title: Graphical models for multivariate time series using wavelets
Authors: Matthias Eckardt - HU Berlin (Germany)
Maria Grith - Erasmus University Rotterdam (Netherlands) [presenting]
Abstract: Local partial dependence and Granger causality graphs are defined for nonlinear and locally stationary multivariate time series processes using wavelet-based methods. In these graphs, nodes denote component processes, and edges describe pairwise conditional dependence between two processes, after removing the contemporaneous, lag and lead influences of the remaining variables. Local dependence is characterized by the wavelet partial coherence measures, defined in the time-frequency domain. Based on these measures, we define undirected, directed and mixed (multi)graphs, which describe specific interactions between time processes. We recover the graphs structure and the edge weights for Gaussian and non-Gaussian settings. We illustrated our methodology for simulated data and apply it to the realized volatilities of the ten largest equity indexes in the world.