Northgate Life: critical illness deterioration, IFRS 17 and an underwriting gap

The three are genuinely different and the label determines both the accounting and the message to the market.

Experience variance. Actual claims differ from expected by chance within the range the assumptions contemplate. Under the general measurement model this is a current-service effect: it flows through the liability for remaining coverage and, for a group that remains profitable, is largely absorbed by an adjustment to the CSM rather than hitting profit immediately.

Change in estimate. There is evidence of a systematic shift - 18 months at 121%, a falling age at claim, and a concentration in one distribution route is not consistent with random fluctuation. This is a change in the estimate of future cash flows relating to future service. It adjusts the fulfilment cash flows prospectively and, because the group is profitable, is absorbed by the CSM; it reduces future profit release rather than creating a current-year loss. But absorption is not neutral: the CSM of £38m is being consumed by the correction, and the release pattern will shorten.

Prior-period error. If, say, the simplified-underwriting block was modelled on the wrong incidence table or the claim count was misclassified, that is an error corrected retrospectively under IAS 8, with restated comparatives and disclosure. That is a very different conversation with the auditor.

Critical point on the CFO's argument. The risk adjustment covers non-financial uncertainty in the estimates. It is not a buffer for a demonstrated adverse trend, and treating it as one would understate the liability. If the revised best estimate makes the remaining coverage onerous, the CSM is written down to nil and a loss component recognised.

The concentration is the signal; the investigation must separate anti-selection from mix effects.

Analyses.

• Split incidence A/E by underwriting route, calendar year, duration, age band, sum assured band and cause of claim. The block has higher sums assured, so a raw comparison will overstate the effect.
• Build a control cohort: fully underwritten policies matched on age, sum assured, sex, smoker status and region. Compare incidence and average claim size.
• Examine claim timing: anti-selection usually shows early - claims in duration 1-3, and a shorter interval between inception and diagnosis.
• Review declined, postponed and repudiated claims, and post-claim non-disclosure investigations, separately for the route.
• Check the underwriting questions actually asked: whether family history, prior investigations, BMI or occupation were omitted, and whether the £250,000 limit was being used to split applications.
• Review the broker's remuneration and quality: volume-based incentive, persistency, complaint rate, and any pattern of applications structured below the evidence threshold.
• Assess the legal position on non-disclosure under the Insurance Act 2015 and the proportionate-remedies regime before assuming recoveries.

What I would not do is extrapolate the whole block from the claim concentration without the exposure data by sum assured. I would ask for the policy-level exposure file, the claim register with diagnosis codes and dates, the underwriting rules in force in 2021, and the broker's file. If the matched-cohort A/E is materially above the rest of the book, the assumption must be changed and the distribution arrangement reviewed commercially.

Indicative calculation. The block is 5,000/26,000 = 19.2% of policies but generates 40% of claims. Let total expected claims be E. Actual claims are 1.21E. Claims from the block are 0.40 × 1.21E = 0.484E; the rest are 0.726E against an expected 0.808E, an A/E of about 90%.

If the block's expected share of cost were also 19.2%, its implied A/E would be 0.484/0.192 ≈ 250%. That is not a credible steady state and almost certainly overstates the effect, because the block has higher average sums assured and a different age mix, so its expected share is above 19.2%. If, for example, its expected share were 30%, the implied A/E falls to about 161%. The point of the range is to show the board that the answer turns on data we do not yet have.

What I would report.

• The 121% is not consistent with random variance; a change in estimate is required for the 2024 valuation, whatever the size of the block effect.
• The CFM's 'absorbed by the risk adjustment' position is not supportable; the risk adjustment is for uncertainty, not for trend.
• Model the CSM roll-forward under the revised assumption and show the board the reduced future profit release, together with a sensitivity where the block experience persists.
• In the Solvency UK return, reflect the revised best estimate and the risk margin consequence.
• Recommend immediate remediation: re-underwrite or re-rate the block at renewal where contractually possible, review the broker arrangement, and strengthen underwriting limits.
• Document the judgement, the data requested and the uncertainty, so the auditor can see the basis of the estimate rather than only the number.