Quantification of operational losses using firm-specific information and external database

ABSTRACT The revised Basel Capital Accord requires banks to reserve capital for operational risk as part of an overall risk-based capital framework. This has motivated researchers to attempt to quantify operational risk. Losses from operational risk can be categorized into two types: one is the highfrequency and low-severity losses that form the body of the distribution and are referred to as expected losses and the other type is the lowfrequency and high-severity losses that form the tail of the distribution and are referred to as unexpected losses. While many banks should have adequate data for modeling the body of the distribution, few have sufficient internal data to estimate the tail distribution. For these rare events, external databases are extremely valuable because they can pool together the industry’s experiences and provide an estimate of exposure for an average bank in the industry. In this paper, we estimate the aggregate operational risk exposure for large losses of a bank by separately estimating the frequency and severity distribution from external data. We utilize a model from Bayesian credibility theory to estimate the frequency distribution and introduce covariates in the estimation of severity distribution. This framework allows the use of both external data and firm specific information to quantify operational risk so that firm specific capital calculation can be made. Different model selection criteria provide conflicting conclusions as to whether the model from extreme value theory outperforms the traditional actuarial models in estimating the tail of the severity distribution. A simulation procedure is provided to construct the entire aggregate distribution of operational risk exposure.