Learning risk culture of banks using news analytics

Abstract

Risk culture is arguably a leading contributor to risk outcomes of a firm. We define risk culture indicators based on unstructured news data to develop a qualitative assessment of risk culture of banks. For US banks participating in an annual stress test program, we conduct a supervised learning ridge regression analysis to identify the most significant features to evaluate banks’ risk culture characteristics. These features are used for unsupervised clustering to determine the high to low quality of risk culture. The distinct groups obtained from clustering define and allow monitoring changes in the quality of risk culture in banks.