BoE published a working paper on predicting bank distress in the UK using machine learning techniques. In the analysis, the main input variables come from confidential regulatory returns while the measure of distress is derived from supervisory assessments of bank riskiness from 2006 through to 2012. Overall, the paper demonstrates practical benefits of machine learning and ensembling methods for providing regulators with advance warning of firm distress. Supervisors can apply these findings to aid in anticipating problems before they occur, thus helping them in their mission to keep financial institutions safe and sound.
The authors compare a number of machine learning and classical statistical techniques, implementing a rigorous, double-block randomized cross-validation procedure to evaluate out-of-sample performance. The random forest algorithm was found to be superior in terms of ranking test observations, while also having relatively better calibrated probabilities than the other techniques. The performance results indicate that the random forest should be used to build an early warning system. To improve the transparency of the algorithm, the study examined the drivers of the predicted probabilities of the model, utilizing an aggregation of Shapley values per test set observation and Shapley regression framework. The Shapley regression reveals the importance of macroeconomic variables and a firm’s sensitivity to market risk, capital buffer, and net interest margin. Finally, the authors also performed simple ensembling techniques to combine all the model outputs, demonstrating substantive and statistically significant improvements relative to the random forest on its own.
Future research might extend this analysis in a number of ways. First, scholars might seek to incorporate additional data beyond financial ratios and macroeconomic variables. Second, future work might delve into more complex configurations of diverse underlying models to reap substantive improvements. Third, the analysis relies on data from a highly unusual period in economic history. Future research might seek to establish whether the documented relationship between input variables and measures of distress persist in relatively benign economic environments. It is likely that in such periods macroeconomic variables are less important in predicting firm distress and, therefore, an early warning system might be better if it were based on data that encompasses more or all of an economic cycle.
Keywords: Europe, UK, Banking, Machine Learning, Statistical Techniques, Ensembling Techniques, Research, Technology, Bank Distress, BoE
Previous ArticleBDE Updates Technical Instructions for Reporting by Banks
BoE published a statistical notice (Notice 2020/9) explaining the approach for treatment of payment holidays on the profit and loss return or Form PL.
BoE updated the known issues document for the statistical reporting Forms AS and FV.
FED announced individual capital requirements for 34 large banks and these requirements go into effect on October 01, 2020.
SRB published a set of documents to give operational guidance to banks on implementation of the bail-in tool.
BIS published an update on the G20 TechSprint Initiative, which was launched in April 2020 and aims to highlight the potential for technologies to resolve regulatory compliance (regtech) and supervisory (suptech) challenges.
OSFI published a letter that provides an update on the milestones for the implementation of the IFRS 17 standard on insurance contracts.
EBA updated the report on the implementation of selected COVID-19 policies.
The Financial Stability Institute (FSI) of BIS published a brief note that examines the supervisory challenges associated with certain temporary regulatory relief measures introduced by BCBS and prudential authorities in response to the COVID-19 pandemic.
BCBS is consulting on the principles for operational resilience and the revisions to the principles for sound management of operational risk for banks.
BoE updated the reporting template for Form ER as well as the Form ER definitions, which contain guidance on the methodology to be used in calculating annualized interest rates.