A Management Override to a Catastrophe Model Before a Large Account Renewal
Arin Insurance uses a vendor catastrophe model for its property portfolio. Independent validation, completed last month, found that the model's convective storm hazard in one inland region is materially understated for short-return-period events and that the region's loss costs are therefore unreliable, though the model performs acceptably at the tail. A large commercial account - 9% of the property book's premium - is renewing in two weeks. The Chief Underwriting Officer wants to cut the modelled loss cost for that region by 20%, arguing that current market pricing proves the model is too conservative overall and that the account will be lost otherwise. Under the firm's model risk policy, management adjustments to model output are permitted up to 10% with documented rationale and Model Risk Committee approval. A 20% cut would need the Committee to waive the cap. The CUO has asked the CRO to support the waiver.
1. Should the CRO support the waiver? Assess the request on its merits rather than on the policy threshold.
Read the worked answer
No, not on the evidence presented - and the reason is that the request uses the wrong argument to reach the right conclusion.
The CUO's case is that market pricing proves the model too conservative. That is a commercial observation, not a validation finding. Market pricing reflects capacity, capital costs, competitor models and appetite; it is not independent evidence about hazard. If it were, the firm would never need a model. The validation finding actually cuts the other way: the model understates short-return-period losses in this region. A 20% cut to the modelled loss cost would move the price further from the validated view of risk, not closer to it.
The cap exists for a reason. A 10% management adjustment accommodates documented model limitations and judgement. A waiver to 20% is not an adjustment; it is a decision to price below the firm's own view of risk. That is an underwriting decision and should be taken as one, by the underwriting authority, with the capital consequences visible - not laundered through a model parameter.
The CRO should also check the CUO's premise. Does the account really turn on a 20% rate movement? What is the actual break-even price, and what is the expected return at that price after reinsurance and capital? A 9% premium concentration is significant, and losing it has real cost. But the question is whether it is profitable at the price the market will bear. If it is, the firm can write it and accept a lower return. If it is not, the model is not the obstacle.
A defensible position: decline the waiver, and offer two genuine routes forward - a documented management adjustment within the 10% cap supported by the account's own loss experience and any independent exposure data, and a separate underwriting decision to accept a lower return if the strategic case is strong. Keep the two conversations distinct.
2. What evidence and controls would you require if the firm nevertheless decides to write the account at the lower price?
Read the worked answer
Separate the two decisions cleanly: what the model says, and what the firm chooses to do about it.
The model output stays as validated. The region's loss cost is what it is. The validation finding - understated short-return-period hazard - should if anything lead to an upward adjustment, and the Model Risk Committee should consider that separately. Do not let a commercial negotiation set a model parameter.
The pricing decision is taken by the underwriting authority. It should be documented as an approved deviation: the account, the price agreed against the modelled technical price, the gap expressed in rupees and as a percentage, the expected return on capital at that price, and the named executive accountable. That record is what allows the firm to learn from the outcome.
Evidence required before the decision.
- A bottom-up exposure and loss estimate for this specific account: construction, occupancy, protection, and the account's own claims history. If the account is genuinely better than the regional average, price it as a better risk - that is legitimate differentiation and does not require a model override.
- The modelled loss at the short return periods, where the validation flagged the weakness, not just the tail.
- Terms and conditions that reduce the gap: higher deductibles, sub-limits for storm, reinstatement provisions, or a facultative reinsurance slice.
- Capital consumption at the agreed price.
Controls.
- A hard accumulation limit: the account must not push the firm's regional storm aggregate beyond the appetite limit, tested against a realistic event.
- Mandatory referral of any account priced more than a set percentage below technical price, applied consistently and not waived for large accounts.
- Reinsurance structure reviewed for the concentration it creates, including counterparty exposure.
- Reporting of all below-technical-price accounts to the Board Risk Committee quarterly, with outcomes.
- A post-placement review of the account's loss experience at 12 and 24 months, so the debate about whether the model was too conservative is settled by data.
3. The validation finding itself - understated short-return-period hazard - needs a response. How should the firm handle a model limitation that affects pricing but not the tail?
Read the worked answer
This is a limitation with a specific shape, and the response should match it rather than being a general caveat.
The model is acceptable at the tail, where capital is set, and unreliable at the short return periods, where pricing is set. Those are two different uses of the same model, and they should be governed separately.
For capital and solvency. The finding does not invalidate the model's tail behaviour. No capital add-on is required on this evidence, but the Model Risk Committee should record the limitation and test whether repeated small understatements of attritional storm loss feed through to a capital effect over time. If the firm's own regional experience diverges from the model, that is a signal to revisit the calibration.
For pricing. The limitation matters directly. Until the vendor fixes it, the firm should apply a documented regional loading to short-return-period loss costs for this region, reviewed at each model update. That is a conservative adjustment - the opposite direction from the CUO's request - and it should be applied consistently to all accounts in the region, not negotiated account by account.
Ownership and documentation.
- Record the limitation in the model inventory with its scope, severity, affected uses, and the compensating controls, signed off by the Model Risk Committee.
- Notify the vendor formally and log the issue; keep a register of outstanding vendor issues and their age.
- Set a revalidation date before the next renewal cycle.
- Disclose the limitation in the underwriting guidance issued to the region, so underwriters are not relying on a number the firm knows to be wrong.
The broader point for the Board. The governance failure here would not be the understated hazard; it would be discovering the understatement and then permitting a change that moves pricing further in the same direction. The test of a model risk framework is what happens when a validated finding collides with a commercial deadline. If the answer is 'the deadline wins', the framework is decoration.