Abstract

Machine Learning (ML) automates prediction, making it cheaper and more accurate. The amount and variety of financial data will continue to increase, and with it the value of ML. A key implication for regulators is that the banking industry is likely to rely increasingly on ML methods for decisions that, by design, cannot be fully understood by their developers. As a result, regulators at all levels will increasingly confront ML models they can’t fully comprehend. Examination is impacted through the need for supervisors to opine on model risk. ML models contain more and complex features. Examiners may need to understand the implications of ML for transparency and associated operational risks. Use of historical data to train models may also have fair lending implications. Some banks and FinTech firms are already using ML for a broad range of banking services such as fraud detection, risk management and pricing. Policy may be impacted through at least two channels; operational risk and market behavior. ML has a direct impact on model risk, a component of operational risk. Banks are subject to model risk management regulatory guidance which has not been updated since April 2011. Some aspects of this guidance may be challenging to apply to ML tools due to their “black-box” nature. ML could also be changing the very nature of market behavior for some liquid assets. We provide an overview of ML and explore these and other implications for banking regulation.