A Suitable Parametric Model For Operational Risk Applications

This paper considers the question of the most appropriate severity distribution estimator for Loss Distribution Analysis (LDA) on operational risk data. We compare the performance of four severity distribution estimators, three well known and one relatively new and assess their suitability for quantitative operational risk analysis. We start with a simulation study assessing the distance between the estimating density function and the true function. The different estimators are then evaluated on a real operational loss dataset where we focus on the risk measures Value at Risk (VaR) and Tail Value at Risk (TVaR). We conclude that the Generalized Champernowne Distribution (GCD) demonstrates the greatest flexibility and is therefore an appropriate choice for LDA in operational risk analysis