Abstract
This paper is concerned with the design of a Bayesian network structure that is suitable for operational risk modelling. The model's structure is designed specifically from the perspective of a business unit operational risk manager whose role is to measure, record, predict, communicate, analyse and control operational risk within their unit. The problem domain modelled is a functioning structured finance operations unit within a major Australian bank. The network model design incorporates a number of existing human factor frameworks to account for human error and operational risk events within the domain. The design also supports a modular structure, allowing for the inclusion of many operational loss event types, making it adaptable to different operational risk environments.
Acknowledgements
The authors acknowledge the financial support provided through two research grants awarded by the Department of Accounting and Finance, Monash University, and the Melbourne Centre for Financial Studies. The second author is also supported by an ARC Discovery grant for which he is grateful.
Notes
1 The Bayesian network technology has been in existence for 20 years, and so a large number of tools, in both a commercial and research-based form, are now available for research, development and applications. The tool used for this research was Hugin Researcher version 7.0.
2 A document containing a detailed description of each node can also be obtained via e-mail by contacting the corresponding author.
3 The term ‘d-separation’ is short for ‘directed acyclic graph separation’. Nodes are said to be d-connected if their states are dependent.
4 If the state of a child node (effect) is observed, and this child node has multiple parent nodes (causes), then information about the state of one of the parent nodes can alter beliefs regarding the state of one of the other parent nodes. This is referred to as ‘explaining-away’.