Calculating Average Claim Severity

What Average Claim Severity Measures

Average claim severity summarizes the average eligible loss amount per eligible claim over a defined set of observations. The calculation is simple, but the numerator and denominator must describe the same claims, coverage, period, valuation basis, and currency. A correct division on mismatched data does not produce a meaningful average.

The eligible loss amount is the amount selected for the ratemaking target under the applicable loss definition. The eligible claim count is the number of claim observations whose amounts are included in that total. Use the observation unit and loss definition established for the analysis; do not switch from claims to payments, or from one loss basis to another, during the calculation.

Definition: For a set of $n$ eligible claim observations with eligible loss amounts $X_1,\ldots,X_n$, average claim severity is the total eligible loss divided by the number of eligible claims: $$ \bar{X}=\frac{\sum_{j=1}^{n}X_j}{n}, \qquad n>0. $$ The numerator and denominator must refer to the same eligible claims and the same defined coverage, period, valuation basis, and currency. If $n=0$, average severity is undefined; it is not zero.

The word eligible matters. A claim may have several recorded amounts, such as payments, recoveries, or components for different coverages. The loss definition determines which amount belongs in the numerator. The counting rule determines which claim observations belong in the denominator. Apply those rules consistently rather than choosing whichever amount or count is easiest to obtain.

Why the Average Uses This Ratio

Suppose each eligible claim contributes one loss amount. Adding those amounts gives the total eligible loss for the selected claims. An average per claim must allocate that total equally across the $n$ claim observations. If the common average is $\bar{X}$, then $n\bar{X}$ must equal the total loss. Solving this identity gives the severity formula:

n\bar{X}=\sum_{j=1}^{n}X_j \quad\Longrightarrow\quad \bar{X}=\frac{\sum_{j=1}^{n}X_j}{n}.

This also explains why averaging subgroup averages without regard to subgroup size can be wrong. A subgroup average represents its claims, not one equally weighted data point. When groups contain different claim counts, the pooled average must give more weight to a group with more claims.

Theorem: If groups $g=1,\ldots,G$ each have $n_g$ eligible claims and average severity $\bar{X}_g$, then the pooled average is $$ \bar{X}_{\text{pooled}} =\frac{\sum_{g=1}^{G} n_g\bar{X}_g}{\sum_{g=1}^{G}n_g}. $$ This holds when each group average and its count use the same eligible loss definition and a common basis. It is a claim-count-weighted average, not generally the simple average of group averages.

A Reliable Calculation Process

Before dividing, make the basis of the calculation explicit. A compact reconciliation helps prevent a technically correct arithmetic operation from answering the wrong question.

Fix the target basis.
State which claims, coverage, period, loss valuation, and currency are included. Use the loss amount already defined for that target.

Match the claim observations.
Confirm that each counted observation has exactly the eligible loss amount required by the counting and observation-unit rules. Combine claim payments first when the target is claim-level.

Reconcile amounts and counts.
Sum eligible loss amounts, count the corresponding eligible claims, and check that units and bases agree.

Divide and verify.
Calculate total eligible loss divided by eligible claim count. Multiply the resulting average by the count to check that it reproduces the total, subject to rounding.

Worked Calculations

Worked Example: Average Severity for One Coverage Group

A hypothetical insurer is summarizing eligible collision losses for one coverage group and one accident period. The selected valuation date is the same for all listed amounts. The eligible claim amounts are $1,200, $2,800, and $4,000. Calculate average claim severity. No interest or mortality assumption is used; these are deterministic example amounts, not a model of a loss distribution.

Method. Use the total eligible claim-level loss divided by the number of eligible claims. All three claims are on the stated coverage, period, valuation, and dollar basis, so each belongs in both numerator and denominator.

First sum the eligible amounts:

$$ 1{,}200+2{,}800+4{,}000=8{,}000. $$

There are $n=3$ eligible claim observations. Therefore:

\bar{X}=\frac{8{,}000}{3} =2{,}666.666\ldots \approx \$2{,}666.67.

The check is $3\times \$2{,}666.666\ldots=\$8{,}000$. Rounded to cents, the average is $2,666.67 per eligible claim. The common mistake avoided here is dividing by the number of policies in the portfolio instead of the three eligible claims; that would answer a different question.

Worked Example: Pooling Groups With Different Claim Counts

A hypothetical book has two rating groups with eligible losses on the same coverage, accident period, valuation date, and currency basis. Group A has four claim amounts: $1,000, $1,500, $2,500, and $3,000. Group B has two: $4,000 and $6,000. Find the average severity across all six claims. No interest or mortality assumption is used, and no probability distribution is assumed.

Method. Calculate each group total and count, then pool the loss totals and claim counts. Group A's total is:

1{,}000+1{,}500+2{,}500+3{,}000=8{,}000, \qquad n_A=4.

Its average severity is $8{,}000/4=\$2{,}000$. Group B's total and average are:

4{,}000+6{,}000=10{,}000, \qquad n_B=2, \qquad \bar{X}_B=\frac{10{,}000}{2}=\$5{,}000.

The pooled loss is $8{,}000+10{,}000=\$18{,}000$, and the pooled claim count is $4+2=6$. Thus:

\bar{X}_{\text{pooled}} =\frac{18{,}000}{6} =\$3{,}000.

As a check using the weighted-average formula, $(4\times 2{,}000+2\times 5{,}000)/6=(8{,}000+10{,}000)/6=\$3{,}000$. The simple average of the two group averages would be $(2{,}000+5{,}000)/2=\$3{,}500$, which is wrong for the pooled claims because it gives the four-claim and two-claim groups equal weight. The correct answer is $3,000 per eligible claim.

Worked Example: Reconciling Currency Before Averaging

A hypothetical portfolio has five eligible claims on the same coverage and period, all valued on the same date. Two claims are recorded in Canadian dollars: CAD 2,000 and CAD 4,000. For this exercise, convert both at the stated rate of USD 0.75 per CAD. Three claims are already in U.S. dollars: $2,500, $3,500, and $4,500. Calculate the average in U.S. dollars. No interest, mortality, or stochastic distribution assumption is used.

Method. Convert each Canadian-dollar amount to U.S. dollars before summing, so every numerator component has the same currency. The converted amounts are:

\text{CAD }2{,}000\times 0.75=\$1{,}500, \qquad \text{CAD }4{,}000\times 0.75=\$3{,}000.

The total eligible loss in U.S. dollars is:

$$ 1{,}500+3{,}000+2{,}500+3{,}500+4{,}500 =15{,}000. $$

All five listed claims are eligible, so the matching claim count is $n=5$. Average severity is:

\bar{X}=\frac{15{,}000}{5}=\$3{,}000.

The check is $5\times \$3{,}000=\$15{,}000$, the reconciled U.S.-dollar total. The correction to the common mistake of adding the two currencies directly is to convert amounts to a common currency first; adding CAD 2,000 to USD 2,500 has no valid dollar result. The average is $3,000 per eligible claim.

Common Traps and Checks

The arithmetic is usually not the difficult part. The main risk is that the total and count quietly refer to different data. Review the following issues before accepting a severity result:

Using the wrong denominator. Divide by the eligible claim count, not by policies, exposures, payment records, or an unrelated portfolio count. The appropriate observation unit and eligibility rule determine the count.
Averaging subgroup averages equally. This is only equivalent to pooling claims when all subgroup claim counts are equal. Otherwise, weight each subgroup mean by its eligible claim count.
Counting payments instead of claims. If the target is claim-level, multiple payments belonging to one claim do not create multiple claim observations. Combine the eligible payment amounts for that claim as required by the loss definition, then count the claim once.
Mixing loss bases. Do not combine amounts valued at different dates, for different coverages, or under different definitions without first reconciling them to the selected target basis. The ratio does not repair an inconsistent numerator.
Mixing currencies or units. Convert amounts to a common currency and unit before aggregation. If one amount is in thousands and another is in dollars, rescale before adding them.
Treating no claims as zero severity. With zero eligible claims, the ratio has a zero denominator and is undefined. A value of zero would falsely assert an observed average loss of zero per claim.

A zero-dollar claim requires particular care: include it in both the numerator and denominator only if the established eligibility and counting rules include that observation. Do not exclude it from one side while retaining it on the other. Similarly, a claim outside the selected coverage or period cannot be included in the loss total merely because a related record appears in the data.

Key takeaway: Average claim severity is eligible loss divided by the matching eligible claim count. Before interpreting the result, verify that both are aligned on claim eligibility, observation unit, coverage, period, valuation, and currency. For pooled groups, combine totals and counts or use claim-count weights.

Check Your Understanding

For each problem, calculate average severity using only the stated eligible amounts and matching claim count. Amounts are hypothetical; no interest, mortality, or loss-distribution assumptions are needed.

Three eligible claims on one coverage and valuation basis have losses of $900, $2,100, and $3,000. Find average severity. (Answer: $2,000.00 per claim)
Group A has three eligible losses totaling $4,800; Group B has five eligible losses totaling $18,000. Both groups use the same coverage and valuation basis. Calculate pooled average severity, not the unweighted average of group means. (Answer: $2,850.00 per claim)
Four eligible claims have losses of $1,000, $2,000, $3,000, and $4,000. A separate portfolio exposure count is 20 policies. Calculate claim severity and identify the correct denominator. (Answer: $2,500.00 per claim; denominator 4 eligible claims)
Two eligible claims are recorded as EUR 1,200 and EUR 2,400. Convert at USD 1.10 per EUR. Two other eligible claims are $1,800 and $3,240. Find the four-claim average in U.S. dollars. (Answer: $2,250.00 per claim)
Three eligible claims have payment records as follows: claim A has $700 and $300; claim B has $2,400; claim C has $600 and $900. The target is claim-level and all listed payments are eligible. Combine payments by claim and calculate average severity. (Answer: $1,633.33 per claim)

What You'll Learn