Operational Risk Management - Implementing a Bayesian Network for Foreign Exchange and Money Market Settlement

Abstract

Operational risk losses in the financial industry usually occur at the business unit (micro) level and are due to weak management oversight, weak internal controls or the lack of it, or to breakdown of procedures among others. It is therefore at the micro level that operational risk has to be managed. Models for managing operational risk at the micro level include Key Risk Indicators (KRI) and causal models among others. Bayesian Networks (BN) as part of the group of causal models is a tool which can be used to manage operational risk at the micro level.The thesis demonstrates with a real-world example how a BN can be used for managing operational risk at a business unit level by developing a BN model for the Foreign Exchange and Money Market settlement process of a bank. The BN developed shows the causal relationships and several levels of dependencies among risk factors, KRIs and other operational risk attributes in one complete model making it a powerful tool for detailed management of operational risk. Results from the model compared well with historical data for quantile values from 0 to 0.95 . Above this value, the model showed higher losses than the historical data. Finally, a complete practical guidance for implementing a BN for a desired process from the point of realising the network structure, quantifying the network - probability elicitations and managing operational risk with the model to maintaining the network is provided.