Claim-Count Methods for Unreported Liabilities

Why Separate Claim Counts From Claim Costs?

An unreported claim liability depends on two distinct judgments: how many claims have occurred but are not yet reported, and how much those claims are expected to cost. Separating these questions makes the estimate easier to explain and review. A count projection can be challenged using exposure, frequency, or reporting development; an expected claim cost can be assessed using severity assumptions and the coverage being valued.

This tutorial uses unreported claim liability in the narrow sense: the expected ultimate cost of claims that have occurred by the valuation date but have not yet been reported. It does not include case reserves for reported claims. Broader reserve presentations may use IBNR to include additional items; the scope must always be made explicit.

The estimate of \(N_U\) can be based on exposure multiplied by an expected claim frequency, or on development of reported counts. The expected cost per claim is a separate severity assumption. Both inputs are estimates; a fractional projected count is therefore acceptable. For example, 54.3 expected unreported claims is an expected value, not a claim inventory containing a fraction of a claim.

Two Ways to Project Ultimate Claim Counts

An exposure-frequency projection starts with a measure of exposure for the period. Exposure might be policies, vehicles, payroll, or another suitable unit. If expected frequency is measured as claims per exposure unit, the basic projection is:

Here \(E\) is exposure and \(\lambda\) is expected ultimate claim frequency per exposure unit. The frequency and exposure units must match. For example, a frequency per vehicle-year cannot be multiplied by a count of individual vehicles unless the exposure period is one year for each vehicle.

A second approach develops reported counts toward ultimate. If \(F_N\) is a selected count development factor from the current reported count to ultimate, then:

The factor must be a count development factor: it projects claim numbers, not dollars. Its suitability depends on the claim definition and on whether historical count reporting development is relevant to the period being estimated. The estimate can be sensitive to changes in reporting practices or claim handling. A development factor should not be applied mechanically when those changes make the historical pattern a poor guide.

Why Count Times Expected Cost Gives the Estimate

Suppose the unreported claims have ultimate costs \(Y_1,\ldots,Y_{N_I}\). Their total ultimate cost is the sum of the individual costs. If each unreported claim has the same conditional expected cost \(\bar{Y}_I\), then:

The middle equality follows because, conditional on there being \(N_I\) claims, the expected sum of their costs is \(N_I\bar{Y}_I\). Thus, when the selected count is an expected count and the expected cost per claim is applicable to that unreported population, multiplying the two gives the expected aggregate cost. Independence between claim count and severity is not required for this result if the conditional expected cost per claim remains constant. If the expected cost varies with the number or mix of claims, that condition does not hold and the estimate should reflect the relationship or be segmented more carefully.

This justification concerns the expected amount, not the uncertainty around it. Two portfolios can have the same expected claim count and expected cost but different variability. This method alone does not quantify that variability.

Keep the Estimate on a Consistent Basis

A count projection and its severity assumption must describe the same claims. Before combining them, specify the period, coverage, claim definition, and valuation basis. In particular, decide whether the expected claim cost includes only indemnity or also allocated claim adjustment expense. If expense is included, it must be included consistently in the intended liability; if it is excluded, the resulting amount is a loss estimate rather than a combined loss-and-expense estimate.

The same consistency is required for gross and net estimates. A gross claim count paired with a net expected cost is not automatically a valid gross or net estimate. The cost assumption must reflect the reinsurance basis being measured. Similarly, a cost assumption for one class should not be applied to a different class without a reason to believe it represents that class’s expected cost.

Expected ultimate cost per claim should refer to the cost of a claim, not a payment, unless the problem explicitly defines the unit as a payment. An unreported claim may generate more than one payment. Treating each payment as a separate claim would change the unit of count and could overstate the liability. The claim definition used to count reports must be compatible with the severity measure.

Interpretation and Common Exam Traps

A count-based estimate is especially transparent when uncertainty about the number of claims can be assessed separately from uncertainty about cost per claim. That transparency does not guarantee that either assumption is appropriate. A high projected ultimate count may reflect a higher expected frequency, a longer development from reported to ultimate, or both. The actuary should know which mechanism the selected method represents.

The most frequent calculation error is to apply expected cost to projected ultimate claims rather than to the projected unreported claims. Another is to subtract a reported count from a projection for a different period or claim definition. A third is to use a severity assumption for a different population or cost basis. Each error can produce a mathematically tidy result that answers the wrong question.