How do banks collect operational risk loss data and why is it so challenging?

Loss data collection is the foundation of operational risk measurement, but it's one of the most practically difficult aspects of risk management. Here's a breakdown:

Internal Loss Data Collection (ILD)

Banks maintain internal loss databases that capture every operational loss above a defined threshold (typically $10,000–$20,000). Each event records:

• Loss amount (gross and net of recoveries)
• Event type (using Basel's 7 event categories)
• Business line
• Date of occurrence vs. date of discovery vs. date of accounting impact
• Root cause description

Basel's 7 Operational Risk Event Types:

1. Internal fraud
2. External fraud
3. Employment practices & workplace safety
4. Clients, products & business practices
5. Damage to physical assets
6. Business disruption & system failures
7. Execution, delivery & process management

Key challenges with internal data:

• Under-reporting: Business units may not report small losses or near-misses to avoid scrutiny. Pinnacle Trust Corp discovered that its retail division was absorbing small fraud losses into operating budgets without flagging them to the risk team.
• Boundary events: Is a fine for mis-selling a credit loss or an operational loss? Classification disputes are common.
• Timing gaps: A rogue trading event may have occurred over months but is discovered in a single day, creating date ambiguity.
• Insufficient history: Most banks only have 5–10 years of clean data, which is not enough to model tail events.

External Loss Data (ELD)

To supplement thin internal tails, banks use consortia data (like ORX) or public loss databases. The challenge is scaling — a $2 billion trading loss at a mega-bank may not be relevant to a regional bank with $50 billion in assets.

Exam tip: FRM Part II tests whether you understand the limitations of loss data and how scenario analysis complements historical data for capital modeling.

Join our FRM community for more operational risk discussions on AcadiFi.