Boston Fed publishes new op risk paper

Boston -- The Federal Reserve Bank of Boston published another ground-breaking paper on operational risk in mid-June.

Implications of alternative operational risk modeling techniques was authored by Patrick de Fontnouvelle and Eric Rosengren of the Boston Fed, and John Jordan, now with FitchRisk. It uses data supplied by six large internationally active US banks to determine if loss data is regular enough to make consistent modelling of operational losses possible. The authors found that “there are similarities in the results of models of operational loss across institutions, and that our results are consistent with publicly reported op risk capital estimates produced by banks’ internal economic capital models”.

The authors began the analysis by plotting the tail of each bank’s loss data by business line and event type. By doing this, the authors found that “loss data for most business lines and event types may be well modelled by a Pareto-type distribution, as most of the tail plots are linear when viewed on a log-log scale. Second, the severity ranking of event types is consistent across institutions. Clients, Products and Business Practices is the highest severity event type, while External Fraud and Employment Practices are the lowest severity event types. Third, the tail plots suggest that losses for certain business lines and event types are very heavy-tailed.” The authors noted that this last finding “highlights that while basic measurement approaches such as the tail plot are easy to implement and intuitively appealing, overly simplistic approaches may yield implausible estimates of economic capital”.

The authors also modelled the distribution of loss amounts using a “full data” approach, as well as using extreme value theory. As a result of these techniques, the authors estimate that the level of capital required for operational risk for the median bank in their sample would be equivalent to 5-9% of the bank’s current minimum regulatory capital requirement. Currently, most banks are allocating 12-15%, but the authors note that the banks’ models “tend to have a broader set of model inputs than those used in this analysis, including, external data, scenarios and qualitative risk assessments”.