Machine Learning for Automating Monitoring, Review and Testing at Financial Institutions

Abstract

There is increasing deployment of machine learning algorithms by financial institutions during and after the coronavirus pandemic. However, majority of these models are being implemented for credit risk management, anti-fraud and anti-money laundering use cases. Moreover, previous research and existing industry papers on machine learning applications in financial and non-financial risk overlook the potential use cases for monitoring, review and testing performed by the second-line-of-defence. This paper bridges the gap in theory and practice by investigating, evaluating and demonstrating the viability of a new vector of use case for deploying machine learning algorithms to automate controls testing for Volcker Rule compliance. This research presents robust evidence on the effectiveness of logistic regression, linear discriminant analysis and neural network-based models for accurately predicting and classifying whether a financial transaction meets the positions-excluded, non-trading account and trading outside the US exemptions under the Volcker Rule. Crucially, this paper offers a proof-of-concept, scalable minimum viable product and pioneering solution to an existing robotics process automation problem facing financial institutions when optimising internal controls monitoring, review and testing processes.

Keywords: Artificial Intelligence, AI, Machine Learning, ML, Controls Testing, Auditing, Compliance, Volcker, Volcker Rule, Monitoring, Review, Testing, Second Line of Defence, Three Lines of Defence, Synthetic Data, Generative Models, Regression Models, Neural Networks, Linear Discriminant Analysis

JEL Classification: A10, A19, C00, C02, C15, C19, C35, C38, C39, C67, C61, C88, C89, K20, K22, K23, K29, Z18