Measuring Systemic Risk in the U.S. Banking System

Abstract

This paper develops a novel measure of systemic risk that combines mapping technology and regression methods. Self-organizing maps (SOM) and lasso logistic regressions are employed to estimate default probabilities for individual U.S. commercial banks from 2001 to 2017. Subsequently, these probabilities are aggregated into a size-weighted measure of systemic risk dubbed SYSTEM. Empirical results show that, due primarily to large banks, volatility in systemic risk increased in 2005 followed by a very large spike from late 2006 to 2008 related to the financial crisis. Comparative tests to the popular systemic risk measure SRISK reveal that SYSTEM: (1) provided earlier warning signals of the impending 2008−2009 crisis; and (2) indicated relatively lower systemic risk after 2012. Further tests show that SYSTEM and SRISK are useful in predicting industry-wide nonperforming loans and numbers of bank failures. We conclude that micro- and macro-prudential measures of bank condition are useful in assessing and predicting systemic risk.