Quantifying Operational Risks through MC simulations and Bayesian Networks

Keywords: Operational Risk, Bayesian Networks, Simulation

Abstract: Operational Risk Management imposes a structured approach to dealing with potential losses in complex operational processes and resources. In operational risk management, the absence of marked to market asset values, means that both qualitative beliefs and quantitative historical data about losses must be combined if we are to obtain a complete risk profile. Unlike the unidimensional risk factors used in financial risk management, operational risk management requires Loss Events incorporating both our uncertainty in the event's frequency and its impact. Because operational managers need to act on events (rather than being passive observers to change) emphasizes the importance of causation rather than correlation as being the primary representation of interdependence between risks. This is done by incorporating Events in a Causal Structure, modelled using a Bayesian Network.

The use of Bayesian Networks in our framework adds some very powerful capabilities to the analysis of Operational Risk. Learning algorithms can be used to validate and improve the causal structures by incorporating the best of prior beliefs and new evidence, facilitating consistency critical in view of the data challenges involved. After obtaining a preset level of consistency, Monte Carlo simulation is performed over the Bayesian Network, to generate a loss distribution from which we derive the expected, unexpected and catastrophic loss components of Operational Risk. Bayesian Inferencing can also be used for process diagnosis and resource allocation. We have implemented a prototype system of the framework. Preliminary results demonstrate the practical promise of the framework.