New regulations in the banking industry (Basel II) require banks to set aside risk capital to cover potential losses caused by such events as fraud, legal liabilities, disasters, system failures, and processing errors, collectively referred to as “operational risks.” Determining the appropriate level of operational risk capital involves statistical measurement of the annual frequency and severity of these various events. Through the Operational Riskdata eXchange (ORX) consortium, a group of over 50 international banks have pooled their operational loss data in an anonymized database to help them better meet the stringent measurement requirements imposed by regulators. A team of researchers at IBM Research - Zurich is working closely with operational risk managers from consortium member banks to determine how to apply consortium data to individual banks’ internal measures of risk. The team is developing innovative statistical techniques for benchmarking risk capital models and scaling loss distributions across a highly diverse set of member banks.
Key operational risk modeling steps when dealing with pooled heterogeneous data.