Combining Scenario Analysis and Loss Data for Operational Risk Capital

Summary

The document raises a methodological question about hybrid approaches to operational risk capital under advanced measurement models. It contrasts three possible interpretations: combining historical internal losses with scenario estimates in a severity distribution, using scenarios to stress an existing model, or building frequency and severity distributions from a pool that includes both observed losses and scenario analysis. Its central concern is how scenario analysis changes a loss distribution when the model otherwise relies on historical frequency and severity data.

The text does not provide an answer, worked example, or empirical comparison of these approaches. It is useful as a framing of the distinction between incorporating scenario-derived losses into model inputs and using scenarios to adjust or stress model assumptions. It also highlights a practical limitation of purely historical data: rare, severe events may be poorly represented. Readers should treat the alternatives as questions posed for clarification, rather than as established implementation guidance.

Key ideas

• The document distinguishes possible ways to combine scenario analysis with loss distribution analysis.
• Scenario estimates might be added to severity data or used to stress model assumptions.
• A hybrid model may draw on both historical losses and scenario-based inputs for frequency and severity.
• The document asks how scenarios can alter a model when its distributions otherwise rely on internal loss history.
• It presents no resolution or evidence comparing the proposed methods.

Tags

# Using a hybrid approach to calculate operational risk capital


# Using a hybrid approach to calculate operational risk capital












I've read that a hybrid approach combing scenario analysis and loss distribution analysis can be used to calculate operational risk capital under the advanced models approach. I've read a couple ways this can be implemented and am not clear on the difference.

They sound the same to me:

- Generate a hybrid distribution (I'm assuming this is the loss severity distribution) where expected losses are comprised of historical internal losses (commensurate with the LDA approach) and use scenario analysis for the right hand side of the distribution.

- Use scenario analysis to "stress" the model producing a more appropriate distribution

- Develop loss frequency and severity distributions from the data pool based on a combination of internal historical losses and scenario analysis.

Can someone explain the difference? In LDA, we use a loss frequency and a loss severity distribution created from internal historical data. If we are not using different data to create the distributions, what other means are there to shock the model using scenario analysis?

Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.