Implementing a Bayesian network for foreign exchange settlement: a case study in operational risk management

Kwabena Adusei-Poku, Gerrit Jan Van den Brink, Walter Zucchini

Abstract

ABSTRACT

Managing operational risk (OR) at the “micro” or business unit level is rather difficult due to the insufficiency and backward-looking nature of historical loss data. A Bayesian network is able to circumvent this difficulty. It can be used to model the causal relationship between risk factors, key risk indicators and a desired operational risk attribute for a specific process. This paper outlines the steps taken to construct such a model to manage OR within the foreign exchange settlement process of a bank. Direct checks on the model carried out using available historical loss data indicate that the frequency and severity distribution generated by the model had a slightly larger mean and fatter tail, respectively, than the observed. The model can be used to estimate the value-at-risk for the process and also to perform scenario and sensitivity analyses, which can reveal the types of changes and interventions that would reduce OR within the process.