Validating Reserving Models: What Independent Review Should Actually Test
By Jonas Mohamed Osman Abdelghafour
A concrete validation programme for reserving and reserve-risk models, covering data, method, assumptions, implementation and use.
Executive answer
Effective validation tests five things: whether the data is fit for purpose, whether the method suits the class, whether assumptions are supported and current, whether the implementation does what the documentation claims, and whether the output is used as intended. Reviews that only re-perform the calculation address the least likely failure mode.
Evidence that carries weight
Actual-versus-expected analysis over several periods, held-out backtesting, sensitivity testing of the largest judgements, and independent re-estimation of one or two material classes with a different method. These produce conclusions a reviewer can defend; a checklist of assumption sign-offs does not.
Independence in practice
Independence means the reviewer can reach and publish an unwelcome conclusion. That requires separate reporting lines, an agreed scope set before the work begins, and a finding-tracking process with owners and dates. Without these, independence is nominal.
Governance considerations
Validation findings should feed a model inventory with risk tiering, so that review intensity matches materiality. Low-materiality models reviewed to the same depth as the largest reserve classes is a symptom of a framework that will run out of capacity.
Conclusion
Validation is a control, and a control that cannot fail anything is not a control. Designing it to be capable of producing bad news is the substantive step.
Primary sources
More in Validation & Governance
- Reporting Reserve Uncertainty to a Board: Making the Range Useful
By Jonas Mohamed Osman Abdelghafour · 8 min read