Leveraging Alternative Data and Machine Learning to Predict bank Risk
Predictive models of bank risk have proliferated in recent years with the advent of modern machine learning methodologies and capabilities. While the focus has been on methodological and algorithmic innovation to drive improved performance, less attention has been paid to the range of data used to build these models, with scholars largely using financial information derived from balance sheets and financial statements. In this research, I demonstrate improved performance by expanding the variable set to include information that is known by bank supervisors to be important for identifying risk but has not yet been included in model builds –- namely, non-financial data which speaks to a bank’s organisational culture, the quality of its management and governance, and operational resilience.
Dr Joel H. Suss is a Research Data Scientist at the Bank of England. He obtained his PhD from the London School of Economics, Department of Psychological and Behavioural Science. His research applies advanced quantitative methodologies to pressing public policy issues, covering financial regulation and economic inequality amongst other topics.