The topological structure of panel variance decomposition networks

In this paper we provide a framework to study the network topology of generalized forecast error variance decomposition (GFEVD) derived from multi-country, multi-variable time series models. Our dynamic variance decomposition network is based on a Bayesian Global Vector Autoregressive (GVAR) model, a suitable macroeconometric method to consider simultaneous multi-level interdependencies across variables. We demonstrate the usefulness of our methodology to analyze the network structure of shock propagation in longitudinal time series and, in particular: (a) the shortest paths of contagion; (b) the clusters of shock transmission; (c) the role of nodes in the risk transmission channels. We illustrate our method through an empirical application to a set of 12 European countries’ Industrial Production, Retail Trade and Economic Sentiment indices over the period 01/2000–11/2021.