Identifying the Risk Culture of Banks Using Machine Learning

Although risk culture is a key element in the risk management process, identifying the riskculture of a firm can be challenging due to the abstract concept of culture. We introduce textmining and unsupervised machine learning algorithms to define the risk culture for U.S. bankholding companies and examine the relation between risk culture and performance. Applyingprincipal component analysis on textually extracted features from 10-K filings identifies uncertainty,litigious and constraining sentiments among risk culture features to be significant in definingrisk culture of banks. Cluster analysis of these features proposes three distinct risk cultureclusters which we label as good, fair and poor. Consistent with regulatory expectations, soundrisk culture in banks is characterized by high profitability ratios, bank stability, lower defaultrisk and good governance