Systemic Risk Measurement in Banking Using Self-Organizing Maps

Abstract

This paper utilizes neural network mapping technology to assess the dynamic nature of systemic risk over time in the banking industry. We combine the nonparametric method of trait recognition with self-organizing maps (SOMs) to generate yearly pictures of the 16 largest U.S. banks’ financial condition from 2003 to 2012. Results show that systemic risk was gradually rising prior to the 2008-2009 financial crisis and peaked in 2009. Thereafter big banks were recovering but considerable systemic risk lingered. Implications to bank regulatory policy and credit risk measurement are discussed.