Defining Claim Measures Before Assessing Credibility

Why the Claim Measure Must Come First

Before assessing whether experience deserves weight in a rate estimate, specify what quantity the experience is intended to estimate. A claim count per vehicle-year, average dollars per accident, and losses per policy-year answer different questions. Two groups can have identical underlying results but appear different if one group’s claim count, loss amount, or exposure is defined on a different basis.

This is not just a matter of labeling a column. The definition determines which observations belong in the calculation and what the denominator means. A credibility comparison is interpretable only if the experience and the quantity being estimated refer to the same coverage, claim unit, loss basis, and exposure unit.

Definition: Let $E$ be exposure measured in a stated unit, $N$ the number of qualifying claim units, and $L$ the total loss amount for those units on a specified basis. Then claim frequency is $N/E$, claim severity is $L/N$ when $N>0$, and loss cost (also called pure premium in this context) is $L/E$. State the claim unit, exposure unit, and loss basis whenever reporting one of these measures.

For example, a claim unit might be a reported accident, a coverage claim, or a policy claim. Those units are not automatically interchangeable: one accident can generate multiple coverage claims. Exposure might be a policy-year, vehicle-year, or another appropriate unit. The loss basis also needs a definition, such as which coverage and loss components are included, the applicable limits and deductibles, and the evaluation date. If the experience has been adjusted to a common dollar level, use the same adjustment basis for all groups being compared.

A loss cost is not the same as a premium rate that includes expenses, profit, or contingencies. It measures losses per exposure on the stated loss basis. Likewise, a claim frequency does not reveal average loss size, and a severity does not reveal how often claims occur. Define the target component before selecting or comparing experience.

How the Measures Fit Together

When the same qualifying claim units appear in both the count and the loss total, and all three measures use the same exposure basis, loss cost can be decomposed into frequency and severity. This is an arithmetic identity, not an assumption that frequency and severity are statistically independent.

\frac{L}{E} = \frac{N}{E}\times\frac{L}{N}, \qquad N>0.

The cancellation of $N$ gives the result: $(N/E)(L/N)=L/E$. This explains why a loss cost can be interpreted as the product of claims per exposure and dollars per claim. The interpretation is useful only if the count $N$ describes the same claims whose losses sum to $L$. If $N=0$, severity $L/N$ is undefined; loss cost $L/E$ can still be computed when $E>0$.

Key condition: The frequency-severity identity is meaningful as a decomposition only when claim counts and loss amounts refer to the same claim unit, inclusion rules, and evaluation basis, and when exposure $E$ has the same definition in both factors.

Worked Measures on a Consistent Basis

The first example illustrates a matched set of definitions. For the count calculations, assume a Poisson model $N\sim\operatorname{Poisson}(\lambda E)$, where $E$ is exposure and $\lambda$ is the claim frequency per exposure unit. The observed frequency $N/E$ is the estimate of $\lambda$. No probability distribution is assumed for individual claim sizes; the supplied loss totals are treated as observed amounts. Interest and mortality assumptions are not applicable.

Worked Example: Comparing Frequency, Severity, and Loss Cost

A hypothetical insurer reviews two groups of liability coverage. Exposure is earned policy-years, and a claim unit is one reported accident. All qualifying accident losses are included in the total loss amount on the same evaluation and coverage basis. Group A has 400 policy-years, 24 accidents, and $180,000 of losses. Group B has 600 policy-years, 30 accidents, and $270,000 of losses. Calculate each group’s frequency, severity, and loss cost, then calculate the portfolio measures.

Method and assumptions. Use $E$ in policy-years, $N$ in accidents, and $L$ in dollars. For frequency, use $N/E$; for severity, use $L/N$; for loss cost, use $L/E$. The count model is $N\sim\operatorname{Poisson}(\lambda E)$, with the observed frequency estimating $\lambda$. No claim-size distribution, interest rate, or mortality assumption is needed for these calculations.

Group A. Frequency is $24/400=0.0600$ accidents per policy-year. Severity is $\$180{,}000/24=\$7{,}500.00$ per accident. Loss cost is $\$180{,}000/400=\$450.00$ per policy-year. The product check is $0.0600\times\$7{,}500.00=\$450.00$.

Group B. Frequency is $30/600=0.0500$ accidents per policy-year. Severity is $\$270{,}000/30=\$9{,}000.00$ per accident. Loss cost is $\$270{,}000/600=\$450.00$ per policy-year. The product check is $0.0500\times\$9{,}000.00=\$450.00$.

Portfolio. Add the underlying totals before dividing: exposure is $400+600=1{,}000$ policy-years, accidents are $24+30=54$, and losses are $\$180{,}000+\$270{,}000=\$450{,}000$. Portfolio frequency is $54/1{,}000=0.0540$. Portfolio severity is $\$450{,}000/54=\$8{,}333.33$, rounded to cents. Portfolio loss cost is $\$450{,}000/1{,}000=\$450.00$ per policy-year. Using the unrounded severity, $0.0540\times\$8{,}333.333\ldots=\$450.00$. The calculation checks.

Common mistake avoided. Averaging the two group severities gives $(\$7{,}500+\$9{,}000)/2=\$8{,}250$, which is not the portfolio severity because the groups have different claim counts. The correct pooled severity weights each group by its number of accidents: total losses divided by total accidents, or $\$450{,}000/54$.

Claim Unit and Exposure Unit Are Part of the Answer

An unqualified statement such as “frequency is 0.05” is incomplete. It should say what was counted and per what exposure unit. Similarly, “average claim is $8,000” needs a definition of a claim and the losses included. If one data set counts accidents and another counts coverage claims, their frequencies and severities are not directly comparable—even if both columns happen to be named “claim count.”

Exposure choice matters as well. A policy-year describes a policy in force for a year; a vehicle-year describes one insured vehicle for a year. If policies cover different numbers of vehicles, the rates per policy-year and vehicle-year need not move together. Neither denominator is universally correct: select the one that matches the intended rating or analysis unit, then use it consistently.

Worked Example: One Accident Can Produce Multiple Claims

A hypothetical commercial auto portfolio has two groups. Group A has 1,000 vehicle-years, 40 accidents, 50 coverage claims arising from those accidents, and $360,000 of total accident losses. Group B has 800 vehicle-years, 32 accidents, 40 coverage claims, and $320,000 of total accident losses. The intended target is accident frequency and average loss per accident. Calculate those measures and show what changes if coverage claims are mistakenly substituted for accidents. For frequency, assume $N\sim\operatorname{Poisson}(\lambda E)$; the count unit in the model must be the specified event. No severity distribution, interest rate, or mortality assumption is used.

Correct accident-based measures for Group A. There are 40 accidents over 1,000 vehicle-years, so accident frequency is $40/1{,}000=0.0400$ per vehicle-year. Average loss per accident is $\$360{,}000/40=\$9{,}000.00$. Accident loss cost is $\$360{,}000/1{,}000=\$360.00$ per vehicle-year. Check: $0.0400\times\$9{,}000.00=\$360.00$.

Correct accident-based measures for Group B. Accident frequency is $32/800=0.0400$ per vehicle-year. Average loss per accident is $\$320{,}000/32=\$10{,}000.00$. Accident loss cost is $\$320{,}000/800=\$400.00$ per vehicle-year. Check: $0.0400\times\$10{,}000.00=\$400.00$.

Effect of substituting coverage-claim counts. Group A would show $50/1{,}000=0.0500$ coverage claims per vehicle-year and $\$360{,}000/50=\$7{,}200.00$ per coverage claim. Group B would show $40/800=0.0500$ coverage claims per vehicle-year and $\$320{,}000/40=\$8{,}000.00$ per coverage claim. Multiplying still returns the same loss costs: $0.0500\times\$7{,}200=\$360.00$, and $0.0500\times\$8{,}000=\$400.00$. That arithmetic does not make coverage-claim frequency the same target as accident frequency, or per-coverage-claim severity the same as per-accident severity. The chosen count unit changes the interpretation of both factors.

Common mistake avoided. Do not compare an accident frequency from one group with a coverage-claim frequency from another, or label both simply “claims per vehicle-year.” The totals can reconcile while the component measures answer different questions.

Choose the Denominator to Match the Target

A change in exposure unit changes the rate’s meaning, not merely its presentation. If an analyst wants losses per policy-year, dividing by vehicle-years answers a different question. When groups have different average numbers of insured vehicles per policy, the two comparisons may even suggest different patterns.

Worked Example: Policy-Years Versus Vehicle-Years

Two hypothetical fleets have consistent accident counts and losses. Fleet X has 300 policy-years covering 450 vehicle-years, 27 accidents, and $216,000 of losses. Fleet Y has 250 policy-years covering 300 vehicle-years, 21 accidents, and $168,000 of losses. The intended claim unit is an accident. Calculate frequency and loss cost using both exposure bases, along with severity. For frequency, assume a Poisson count with mean $\lambda E$ on whichever exposure basis is explicitly used. No severity distribution, interest rate, or mortality assumption is used.

Fleet X. Accident frequency per policy-year is $27/300=0.0900$; per vehicle-year it is $27/450=0.0600$. Severity is $\$216{,}000/27=\$8{,}000.00$ per accident. Loss cost per policy-year is $\$216{,}000/300=\$720.00$; per vehicle-year it is $\$216{,}000/450=\$480.00$. Checks: $0.0900\times\$8{,}000=\$720.00$, and $0.0600\times\$8{,}000=\$480.00$.

Fleet Y. Accident frequency per policy-year is $21/250=0.0840$; per vehicle-year it is $21/300=0.0700$. Severity is $\$168{,}000/21=\$8{,}000.00$ per accident. Loss cost per policy-year is $\$168{,}000/250=\$672.00$; per vehicle-year it is $\$168{,}000/300=\$560.00$. Checks: $0.0840\times\$8{,}000=\$672.00$, and $0.0700\times\$8{,}000=\$560.00$.

Interpretation. Per policy-year, Fleet X has higher frequency and loss cost. Per vehicle-year, Fleet X has lower frequency and loss cost. Both conclusions follow from the given exposure totals; they address different units. The difference arises because the fleets have different vehicles per policy: $450/300=1.5$ for X and $300/250=1.2$ for Y. The correct denominator is the one that matches the intended target, not the one that produces a preferred comparison.

Common mistake avoided. Do not compare a per-policy-year rate for one fleet with a per-vehicle-year rate for another. State the denominator alongside every reported measure.

Definition Checks Before Comparing Experience

A useful comparison starts with a compact specification of the measure. Record the target (frequency, severity, loss cost, or another component), the claim unit, the exposure unit, and the loss basis. Then verify that the numerator and denominator include the same population and period. Earlier material on exposure and aligning experience periods is relevant here: matching time periods alone does not correct a mismatch in claim or exposure definitions.

For loss cost, also state which losses are included. If one group’s losses reflect a different coverage scope, limit, deductible, or evaluation basis, a difference in $L/E$ cannot automatically be attributed to a difference in underlying risk. Those features must either be made comparable or explicitly treated as part of the target being estimated. Do not quietly change the definition during the comparison.

Practical checklist:

Target: Which rate component is being estimated?
Claim unit: Is the count of accidents, coverage claims, or another defined unit?
Exposure unit: Is the denominator policy-years, vehicle-years, or something else?
Loss basis: Which claims and loss amounts are included, and on what common basis?
Consistency: Do the count, loss total, and exposure refer to the same experience population?

A frequent exam trap is to calculate correctly from numbers that do not represent the stated target. For instance, suppose losses are accident-based but the count is the number of coverage claims. Computing “severity” as total losses divided by that count gives dollars per coverage claim, not dollars per accident. The arithmetic is valid for its stated numerator and denominator; the error is giving the result the wrong interpretation. Define first, calculate second, and check that frequency multiplied by severity reproduces loss cost only when the definitions align.

Check Your Understanding

For each problem, identify the stated unit before calculating. Answers use the precision shown.

A hypothetical property group has 240 policy-years, 12 qualifying fire claims, and $96,000 of losses on a common basis. Calculate claim frequency, severity, and loss cost. (Answer: 0.0500 claims per policy-year; $8,000.00 per claim; $400.00 per policy-year)
A liability portfolio has 500 vehicle-years, 20 accidents, 26 coverage claims, and $150,000 of accident losses. The target is accident frequency and severity per accident. Calculate both target measures and the accident loss cost. (Answer: 0.0400 accidents per vehicle-year; $7,500.00 per accident; $300.00 per vehicle-year)
A fleet has 200 policy-years covering 300 vehicle-years, with 18 accidents and $144,000 of losses. Calculate loss cost per policy-year and per vehicle-year, and state why the two figures differ. (Answer: $720.00 per policy-year; $480.00 per vehicle-year; the exposure denominators differ)
Two groups have the same claim definition and loss basis. Group A has 10 claims, $80,000 of losses, and 400 exposure units; Group B has 30 claims, $240,000 of losses, and 1,200 exposure units. Calculate pooled frequency, severity, and loss cost. (Answer: 0.0250 claims per exposure unit; $8,000.00 per claim; $200.00 per exposure unit)
A group has 0 qualifying claims, positive exposure of 350 policy-years, and $0 of losses. Which measures can be calculated, and which is undefined? (Answer: frequency is 0 claims per policy-year; loss cost is $0.00 per policy-year; severity is undefined because the claim count is zero)

What You'll Learn