Dynamic communities in multichannel data: An application to the foreign exchange market during the 2007--2008 credit crisis

TL;DR

This paper analyzes the evolution of communities in multichannel financial data, specifically foreign exchange rates during the 2007-2008 credit crisis, using dynamic network community detection to reveal market structural changes.

Contribution

It introduces a node-centric dynamic community detection method applied to multichannel time series, providing insights into individual node roles during market crises.

Findings

Persistent community attachment of exchange rates identified

Important transfer nodes tend to be on community edges

Major trading changes during the crisis were uncovered

Abstract

We study the cluster dynamics of multichannel (multivariate) time series by representing their correlations as time-dependent networks and investigating the evolution of network communities. We employ a node-centric approach that allows us to track the effects of the community evolution on the functional roles of individual nodes without having to track entire communities. As an example, we consider a foreign exchange market network in which each node represents an exchange rate and each edge represents a time-dependent correlation between the rates. We study the period 2005-2008, which includes the recent credit and liquidity crisis. Using dynamical community detection, we find that exchange rates that are strongly attached to their community are persistently grouped with the same set of rates, whereas exchange rates that are important for the transfer of information tend to be positioned on the edges of communities. Our analysis successfully uncovers major trading changes that occurred in the market during the credit crisis.