Improving the Robustness of Operational Risk Modeling Through External Data Scaling

Abstract

One of the biggest challenges that banks face in modeling operational risk is the instability of risk estimates. The key causes of this instability are the heavy-tailness of loss distributions and insufficient loss data. To address these issues, we propose a loss scaling method to combine internal loss data of a bank with external loss data of other banks. The method is based on our finding that the loss severity of tail losses is related to bank size. Using supervisory operational loss data from large U.S. bank holding companies, we demonstrate that our method of incorporating external data improves the robustness of operational risk estimates. In addition, we demonstrate that our scaling method produces statistically and economically stronger estimates of correlations between operational losses and the macroeconomic environment than estimates based on individual banks' data only. Thus, our method can improve the robustness of models for stress testing operational risk to severe macroeconomic shocks.