Capstone: a one-page model risk assessment

Three ways to see it

1. 1

   A working one-page model risk assessment has seven sections. Model identity: name, owner, version, purpose. Inputs: data sources and refresh cadence. Methodology: model family and high-level approach. Performance: key metrics, when last measured. Risks: top three model risks (drift, bias, opacity, etc.) with severity. Mitigations: what is being done about each. Sign-off: model owner, validator, CRO, dates. Each section gets two to four lines.

2. 2

   Way one to do this fast: start from the SR 11-7 inventory if it exists. If your bank has any model inventory, pull the row for your chosen model and expand. Each cell in the inventory row corresponds to a section in the one-page assessment. If there is no inventory, this one-pager will become the seed for one.

3. 3

   Way two: write the risks in plain language. 'This model has a 12 percent disparate decline rate against female applicants in our SME book, controlling for declared revenue' is a usable risk. 'Fairness risk' is not. Plain risks attract plain mitigations and plain accountability.

The why-tree

Try this with Claude

AI-edge prompt to try: 'Acting as an SR 11-7 model validator, draft a one-page model risk assessment for the credit scoring model I describe. Identify three top risks specifically relevant to a Pakistani consumer bank and propose mitigations.' Edit before sending.