News and Networks: Using Text Analytics to Assess Bank Networks During COVID-19 Crisis

Abstract

We study the 'interconnectedness' of stress-tested banks by exploiting how they are mentioned together in the context of financial news. We start by constructing weekly co-occurrence network matrices following Ronnqvist and Sarlin (2015) text-to-network approach. Using the COVID-19 pandemic as an external shock, we examine how bank networks behave during high stress periods. We find that banks become more interconnected during peaks of COVID-19 induced stress. We put forth a new measure of systemic risk that utilizes text-based eigenvector centrality. This measure provides a more stable ranking system than the traditional SRISK measure during both high and low stress periods.