This paper discusses a proposed method for the estimation of loss distribution using information from a combination of internally derived data and data from external sources. The relevant context for this analysis is the estimation of operational loss distributions used in the calculation of capital adequacy. A robust, easy-to-implement approach that draws on Bayesian inferential methods has been presented. The principal intuition behind the method is to let the data itself determine how they should be incorporated into the loss distribution. This approach avoids the pitfalls of managerial choice on data weighting and cut-off selection and allows for the estimation of a single loss distribution.

This paper aims to estimate the operational loss distributions used in the calculations of capital adequacy. Capital adequacy may be either regulatory capital, as required under Basel II guidelines or economic capital. Throughout the paper, it has been referred to capital adequacy broadly as the methods and approaches here are generalizable. Under new Basel requirements, a number of institutions in the United States are likely to be required to use the so-called Advanced Measurement Approach (AMA) and a handful more look likely to `opt-in'. This AMA approach requires the inclusion of information from four `elements': internal data, external data, scenarios and business environment and internal control factors. The combination of internal and external data, are addressed though the methods could in principle be extended to include information from the other two.

External data are often used to `fill in' information on very low frequency, high severity losses where data are unavailable within the institution. By their nature, it is this group of losses that are both of issue in capital adequacy and the most difficult to measure. Measurement difficulties derive from the very fact that they are of such low frequency.