Clustering in Dynamic Causal Networks as a Measure of Systemic Risk on the Euro Zone

Abstract

In this paper, we analyze the dynamic relationships between ten stock exchanges of the euro zone using Granger causal networks. Using returns for which we allow the variance to follow a Markov-Switching GARCH or a Changing-Point GARCH, we first show that over different periods, the topology of the network is highly unstable. In particular, over very recent years, dynamic relationships vanish. Then, expanding on this idea, we analyze patterns of information transmission. Using rolling windows to analyze the topologies of the network in terms of clustering, we show that the nodes' state changes continually, and that the system exhibits a high degree of flickering in information transmission. During periods of flickering, the system also exhibits desynchronization in the information transmission process. These periods do precede tipping points or phase transitions on the market, especially before the global financial crisis, and can thus be used as early warnings of phase transitions. To our knowledge, this is the first time that flickering clusters are identified on financial markets, and that flickering is related to phase transitions.

Keywords: Causal Network, Topology, Custering, Flickering, Desynchronisation, Phase transitions

JEL Classification: C18, D85, G01