A Mixing Severity Model Incorporating Three Sources of Data for Operational Risk Quantification

Abstract

To meet the new Solvency II Directive for operational risk capital assessment, an insurance company's internal model should make use of internal data and relevant external information. One of the unresolved challenges with operational risk quantification is combining different information sources appropriately (e.g. internal data, consortium data and publicly reported losses). This paper develops a systematic approach that apart from internal data, incorporates two sources of prior knowledge into internal loss distribution modelling. The standard statistical model resembles the idea with credibility theory and Bayesian methodology, in the sense that the sources of prior knowledge are weighted more when internal data is scarce than when internal data is abundant.

Keywords: Operational risk, Mixing data sources, Actuarial loss models, Transformation, Multiplicative bias reduction, Pre- and Post insurance Loss Distribution

JEL Classification: C13, C14