Severity modelling with external loss data

Models of operational risk, based only on internal loss data, can exhibit wide confidence intervals for capital at 99.9 percentile of the aggregate loss distribution. Consequently, banks rely on scenario and/or external loss data to improve capital estimates and to narrow confidence intervals. Mikhail Makarov and Bahram Mirzai report

In this article we consider the use of external loss data for operational risk modelling, where examples of such pools are public data sources or loss data consortia. In the following we use the notion of pooled data instead of external data to stress that a banks own data can be part of the pooled data. We focus only on severity modelling, and, in particular, we show that a severity distribution fitted to the pooled data should not be used directly to estimate a bank's tail severity if the