A0167
Title: Factor-adjusted network analysis for high-dimensional time series
Authors: Haeran Cho - University of Bristol (United Kingdom) [presenting]
Matteo Barigozzi - Università di Bologna (Italy)
Dom Owens - University of Bristol (United Kingdom)
Abstract: A methodology is proposed for modelling network structures of high-dimensional time series exhibiting strong serial- and cross-sectional correlations. We adopt a factor-adjusted vector autoregressive (VAR) model where, after the factors account for pervasive co-movements of the variables, remaining idiosyncratic dependence between the variables is modelled by as a sparse VAR process. We propose methods for estimating the latent VAR model and thus learning a directed network representing the Granger causal linkages between the variables, an undirected one embedding the contemporaneous relationships among the residuals and finally, one that summarises both lead-lag and contemporaneous linkages by means of the long-run partial correlations. We provide consistency with rates of all estimated quantities without any specific distributional assumption but by requiring only the existence of fourth moments. As a by-product, the first complete treatment of factor-adjusted sparse VAR estimation is offered. Simulations and real data applications are provided to demonstrate the good performance of the proposed methodology.