Why the Claim-Count Basis Matters
A loss amount at a valuation date is usually incomplete: some claims may develop to higher amounts, and additional claims may be reported later. Dividing that incomplete loss amount by a claim count can give a useful observed severity, but it is not automatically an estimate of ultimate severity. The numerator and denominator must describe the same stage of development.
For a cohort such as an accident period, define observed severity as loss dollars at the valuation date divided by claims reported by that date. Ultimate severity instead uses ultimate loss dollars and the corresponding ultimate claim count. Developing only the loss numerator while leaving a current claim count in the denominator answers a different question: ultimate dollars per claim already reported.
A claim count may be defined in different ways, such as reported claims or claims with an indemnity payment. The count definition used for severity must match the claims represented by the loss amount. For example, dividing indemnity losses by a count that includes unrelated expense-only claims changes the measure being calculated.
Developing Losses and Counts Together
In a deterministic development-factor approach, a selected loss development factor \(f_L\) converts valuation-date losses to estimated ultimate losses. A selected claim-count development factor \(f_N\) converts reported counts at the same valuation age to estimated ultimate counts. These factors are specific to the relevant cohort, age, and definitions; they need not be equal.
This gives a practical way to develop observed severity: multiply it by the ratio of the loss development factor to the claim-count development factor. If losses are expected to develop faster than counts, this ratio exceeds one and estimated severity increases. If counts develop faster than losses, the ratio is below one and estimated ultimate severity decreases.
Why the Severity Formula Holds
The formula follows directly from the definitions, rather than from an additional severity assumption. Estimate the ultimate numerator by applying the loss factor, and estimate the ultimate denominator by applying the count factor. Taking their ratio gives:
The identity also shows the model's limits. It is a development-factor estimate, not a claim-level prediction. It relies on selected factors that reasonably represent future loss and count development for the cohort. It can be misleading if the loss factor and count factor refer to different cohorts, valuation dates, claim definitions, or loss bases. No particular parametric distribution of individual claim sizes is required for this calculation.
Worked Example: Developing Losses and Claim Counts on Matching Bases
Assume a hypothetical accident-period cohort has $1,200,000 of cumulative paid losses and 80 reported claims at the valuation date. The selected paid-loss development factor is 1.25, and the selected reported-claim development factor is 1.10. Assume both factors apply to this same cohort and that the claim count represents the claims underlying the paid losses. Estimate ultimate losses, ultimate claim count, observed severity, and ultimate severity. This is a deterministic factor calculation; no interest rate, mortality assumption, or parametric severity distribution is used.
Step 1: Develop losses. The estimated ultimate paid losses are:
Step 2: Develop the count. The estimated ultimate number of claims is:
Step 3: Calculate observed severity.
Step 4: Calculate ultimate severity per ultimate claim.
As a check, the factor-ratio method gives \(\$15{,}000(1.25/1.10)=\$17{,}045.45\). By contrast, \(\$1{,}500{,}000/80=\$18{,}750\) is ultimate loss per currently reported claim, not ultimate severity per ultimate claim. The method avoids silently changing the denominator basis.
Pooling Developed Severity
When several cohorts have different development patterns, calculate their developed losses and counts before pooling. The pooled ultimate severity is total estimated ultimate loss divided by total estimated ultimate claim count. It is generally not the simple average of cohort severities, because cohorts can have different claim counts.
Worked Example: Pooling Two Cohorts After Development
Suppose two hypothetical accident-period cohorts have the following experience and selected factors. Losses are on the same basis, counts use the same claim definition, and each factor pair is selected for its own cohort. Estimate each cohort's ultimate severity and the pooled ultimate severity. The method is deterministic and assumes no particular distribution for individual claim sizes.
| Cohort | Losses at valuation | Reported claims | Loss factor | Count factor |
|---|---|---|---|---|
| A | $900,000 | 60 | 1.20 | 1.10 |
| B | $1,100,000 | 100 | 1.10 | 1.05 |
Cohort A. Developed losses are \(\$900{,}000(1.20)=\$1{,}080{,}000\). Developed claims are \(60(1.10)=66\). Its ultimate severity is \(\$1{,}080{,}000/66=\$16{,}363.64\).
Cohort B. Developed losses are \(\$1{,}100{,}000(1.10)=\$1{,}210{,}000\). Developed claims are \(100(1.05)=105\). Its ultimate severity is \(\$1{,}210{,}000/105=\$11{,}523.81\).
Pool the developed amounts. Total ultimate losses are \(\$1{,}080{,}000+\$1{,}210{,}000=\$2{,}290{,}000\). Total ultimate claims are \(66+105=171\). Thus:
As a reasonableness check, the pooled result lies between the two cohort severities and is weighted more toward Cohort B, which has more estimated ultimate claims. The simple average of the two cohort severities would not use those claim-count weights and would not give the pooled measure.
Interpreting the Denominator
Sometimes the intended quantity is not ultimate severity per ultimate claim. An analyst may instead need estimated ultimate dollars associated with each claim already reported. That measure keeps the current count \(N_t\) in the denominator:
This can be useful for questions explicitly framed per currently reported claim, but it should not be labeled ultimate severity per claim without qualification. The count factor is needed when the target is severity per ultimate claim. A sound calculation therefore states the denominator in words as well as in notation.
Worked Example: Distinguishing Two Severity Measures
A hypothetical cohort has $720,000 of losses and 60 reported claims at the valuation date. The selected loss development factor is 1.25 and the selected claim-count development factor is 1.20. Assume both factors apply to the same cohort and aligned claim definitions. Calculate ultimate loss per currently reported claim and ultimate severity per ultimate claim. No interest rate, mortality assumption, or parametric severity distribution is required.
Step 1: Calculate observed severity. \(\$720{,}000/60=\$12{,}000\) per reported claim.
Step 2: Develop losses and counts. Ultimate losses are \(\$720{,}000(1.25)=\$900{,}000\), and ultimate claims are \(60(1.20)=72\).
Step 3: Calculate each target. Ultimate loss per currently reported claim is \(\$900{,}000/60=\$15{,}000\). Ultimate severity per ultimate claim is \(\$900{,}000/72=\$12{,}500\). The latter also equals \(\$12{,}000(1.25/1.20)=\$12{,}500\).
Check. The two answers differ because their denominators are 60 and 72, respectively. Calling both simply “ultimate severity” would obscure the claim-count basis.
Common Mistakes and Practical Checks
The most common exam trap is to develop losses and then divide by the current reported count while describing the result as severity per ultimate claim. For the first worked example, that incorrect calculation is:
The arithmetic is correct, but the label is not: the denominator remains 80 reported claims. If the requested measure is ultimate severity per ultimate claim, the correction is to develop the count as well:
Before using a developed severity estimate, verify the following:
- The losses and claims refer to the same cohort and valuation date.
- The loss basis is explicit, such as paid or incurred, and the selected loss factor applies to that basis.
- The claim-count definition matches the claims represented in the loss amount.
- The denominator is identified as reported claims or ultimate claims, not left implicit.
- When pooling cohorts, add developed losses and developed counts separately, then divide the totals.
Check Your Understanding
For each problem, state which claim-count basis is used and retain full precision until the final dollar amount.
- A hypothetical cohort has $640,000 of losses and 40 reported claims. Its loss development factor is 1.25 and its claim-count development factor is 1.10. Find ultimate severity per ultimate claim. (Answer: $18,181.82 per ultimate claim)
- Observed severity is $12,000 per reported claim. The selected loss and claim-count development factors are 1.18 and 1.10, respectively. Estimate ultimate severity per ultimate claim. (Answer: $12,872.73 per ultimate claim)
- A cohort has $2,400,000 of losses and 150 reported claims, with loss factor 1.12 and count factor 1.05. Calculate estimated ultimate loss, estimated ultimate claim count, and ultimate severity. (Answer: $2,688,000; 157.5 expected claims; $17,066.67 per ultimate claim)
- Cohort A has $480,000 of losses and 40 reported claims, with factors 1.20 for losses and 1.10 for counts. Cohort B has $700,000 of losses and 50 reported claims, with factors 1.10 for losses and 1.04 for counts. Find pooled ultimate severity by developing each cohort first. (Answer: $14,020.83 per ultimate claim)
- A cohort has $720,000 of losses and 60 reported claims. Its loss factor is 1.25 and count factor is 1.20. Give both ultimate loss per currently reported claim and ultimate severity per ultimate claim. (Answer: $15,000 per currently reported claim; $12,500 per ultimate claim)