Why the Exposure Denominator Matters
An observed claim count has no useful frequency interpretation until it is related to the amount of risk that produced it. Twelve claims could represent a low claim rate in a large portfolio or a high rate in a small one. Choosing the denominator—called the exposure basis—states what amount of insured risk each claim is being compared with.
In credibility work, experience is often summarized as a claim measure per exposure unit. That summary can then be compared with complementary information. But a rate is meaningful only when the exposure unit reflects the risk represented by the claims. A policy-year rate answers a different question from a vehicle-year rate, even when both use the same claims.
The unit is a choice about the interpretation of experience, not merely a convenient way to divide. Policy count may be suitable when each policy is the relevant unit of risk. Vehicle-years may be more informative when each insured vehicle contributes risk independently. Payroll or sales can be appropriate when covered risk is expected to vary with those quantities. No one denominator is automatically correct for every insurance product.
Rates per Exposure Unit
Let \(N\) be the number of claims counted under a stated claim definition, and let \(E\) be the amount of exposure measured on a specified basis. The observed claim frequency per exposure unit is \(N/E\). If \(L\) is aggregate loss on the same claims and coverage basis, \(L/E\) is the observed loss cost per exposure unit. These ratios describe different measures: one counts claims, while the other measures dollars of loss.
The division does not make unlike units interchangeable. “Claims per policy-year” and “claims per vehicle-year” have distinct meanings. A frequency can even exceed one: it is an average count per exposure unit, not necessarily the probability that an exposure unit has at least one claim. The numerator must also be defined consistently. For example, claim counts and occurrence counts are not necessarily the same measure.
| Exposure basis | Denominator represents | Typical interpretation |
|---|---|---|
| Policy-year | One policy in force for one year, or its earned fraction | Claims per policy-year |
| Vehicle-year | One vehicle insured for one year, or its earned fraction | Claims per vehicle-year |
| Payroll unit | A specified amount of covered payroll | Claims per stated payroll amount |
| Sales unit | A specified amount of insured sales | Claims per stated sales amount |
How Time and Insured Objects Enter Exposure
For time-based exposure, a partial term contributes only the fraction of a year for which coverage was in force, provided the chosen convention measures earned exposure in years. If the unit is vehicle-year, the time fraction is applied for each insured vehicle. Thus, a policy with several vehicles can contribute several vehicle-years during one policy-year. A policy count alone cannot capture this difference in the amount of vehicle exposure.
Some exposure bases are not time units. A payroll-based exposure might be measured in units of a fixed dollar amount of payroll, while a sales-based exposure might be measured in units of a fixed dollar amount of sales. In such cases, the recorded exposure quantity—not the number of policies or the length of each term—is the denominator.
Worked Example: Partial-Term Policy-Year Exposure
A hypothetical property portfolio has 60 policies in force for a full year and 40 policies that are each in force for six months. The policies provide the same coverage, and 12 claims are counted during their respective periods of coverage. Calculate claim frequency per policy-year and compare it with the rate obtained by dividing by the number of policy records.
Method: earned policy-year exposure. Each full-year policy contributes 1 policy-year, so those policies contribute \(60 \times 1=60\) policy-years. Each six-month policy contributes \(6/12=0.5\) policy-year, so those contribute \(40 \times 0.5=20\) policy-years.
Total earned exposure is \(60+20=80\) policy-years. Therefore, claim frequency is \(12/80=0.15\) claims per policy-year.
There are \(60+40=100\) policy records. Dividing by that count gives \(12/100=0.12\) claims per policy record, not claims per policy-year. The two calculations use different denominators and answer different questions.
Check. The exposure-weighted result is \(0.15 \times 80=12\) claims. The common mistake this method avoids is counting each six-month policy as a full year of exposure.
Worked Example: Policy-Years Versus Vehicle-Years
A hypothetical auto insurer observes 30 fleet policies, each covering four vehicles for a full year, and 20 individual policies, each covering one vehicle for six months. There are 26 claims, and each claim is assigned to one covered vehicle. Calculate frequency per vehicle-year and per policy-year.
Method: calculate each denominator on its own basis. The fleets contribute \(30 \times 4 \times 1=120\) vehicle-years. The individual policies contribute \(20 \times 1 \times 0.5=10\) vehicle-years. Total vehicle-year exposure is \(120+10=130\).
The vehicle-year frequency is \(26/130=0.20\) claims per vehicle-year.
For policy-years, the 30 full-year fleet policies contribute 30 policy-years, and the 20 half-year individual policies contribute \(20 \times 0.5=10\) policy-years. Total policy-year exposure is \(30+10=40\), so frequency is \(26/40=0.65\) claims per policy-year.
These results agree with the different exposure bases: there are \(130/40=3.25\) vehicle-years per policy-year on average, and \(0.20 \times 3.25=0.65\). The common mistake avoided is dividing vehicle-assigned claims by policy count and then labeling the result a vehicle-year frequency.
Choosing a Denominator for the Question
A useful selection test is: What unit of insured risk does one counted claim represent? Then ask whether the exposure measure counts that unit over the same coverage and experience period. For a motor portfolio, one policy can cover multiple vehicles, so the policy-year denominator may conceal variation in vehicle count. For a liability coverage whose relevant exposure is tied to payroll, policy count may conceal large differences in payroll volume.
The exposure base also affects comparisons between groups or periods. If one group has more vehicles per policy, its claims per policy-year may differ from another group’s even when claims per vehicle-year are equal. A change in the mix of policy sizes can therefore change a policy-based rate without a corresponding change in the rate per insured object. Conversely, choosing an exposure base that is not meaningfully related to the risk can give a misleading impression of comparability.
Worked Example: Payroll as the Exposure Unit
A hypothetical workers’ compensation portfolio has 12 small accounts, each with $100,000 of payroll, and 12 large accounts, each with $300,000 of payroll. The small accounts produce 6 claims and the large accounts produce 30 claims. Calculate the aggregate frequency per $100,000 of payroll and compare it with claims per policy.
Method: convert payroll to the stated exposure unit. Small-account payroll is \(12 \times \$100{,}000=\$1{,}200{,}000\), or 12 units of $100,000. Large-account payroll is \(12 \times \$300{,}000=\$3{,}600{,}000\), or 36 units. Total payroll is \(\$4{,}800{,}000\), equal to 48 exposure units.
There are \(6+30=36\) claims. Frequency per $100,000 of payroll is \(36/48=0.75\) claims per exposure unit. There are \(12+12=24\) policies, so claims per policy is \(36/24=1.50\).
The segment calculations confirm the aggregate: the small accounts have \(6/12=0.50\) claims per payroll unit, while the large accounts have \(30/36\approx0.8333\). Weighting by their respective exposure units gives \((12 \times 0.50+36 \times 0.8333\ldots)/48=36/48=0.75\). The common mistake avoided is treating every account as the same amount of payroll exposure.
Why Claim Frequency Scales With Exposure
A basic justification for using a rate per exposure unit is a proportionality model. Suppose that, over the period and population being summarized, each unit of exposure has the same expected claim count per unit, denoted by \(\lambda\), and that expected claims scale proportionally with exposure. This is a model assumption, not a consequence of merely observing a denominator. It is most plausible when the exposure unit represents comparable risk and the claim definition is consistent.
Under this model, for \(E\) exposure units, the expected claim count is \(\lambda E\). The observed rate \(N/E\) follows naturally as an estimate of the expected count per unit: its expected value is \(\operatorname{E}[N/E]=\operatorname{E}[N]/E=(\lambda E)/E=\lambda\), when \(E\) is treated as fixed. Thus, dividing by exposure puts the count on a common per-unit scale.
If expected claims do not scale with the selected exposure—for example, if risk is driven by the number of policies while the denominator is payroll—then the proportionality assumption may fail. A calculated ratio is still arithmetic, but its interpretation as a rate per relevant unit is weakened. It is also unsafe to compare rates whose claim definitions, coverage periods, or exposure conventions differ without first reconciling those differences.
Common Mistakes and Practical Checks
- Using policy count as if it were earned policy-years. A half-year policy contributes half a policy-year under a time-based convention, not one full policy-year.
- Ignoring multiple insured objects. A policy covering several vehicles can contribute several vehicle-years for each year in force.
- Comparing rates without naming the denominator. “0.2 claims” is incomplete; state the rate per vehicle-year, policy-year, payroll unit, or other defined basis.
- Assuming one exposure basis fits every question. A denominator should reflect the unit of risk represented by the claims, not merely the data field that is easiest to obtain.
- Mismatching numerator and denominator. Claims must be counted for the same coverage and experience period represented by the exposure, using an explicit and consistent claim-count definition.
Check Your Understanding
For each problem, identify the appropriate denominator and interpret the resulting rate.
- A hypothetical book has 50 policies in force for a full year and 30 policies in force for three months. There are 23 claims. Calculate claims per policy-year. (Answer: \(23/57.5=0.400\) claims per policy-year.)
- An auto portfolio has 18 full-year policies covering two vehicles each and 12 half-year policies covering three vehicles each. There are 27 claims, each assigned to one vehicle. Calculate frequency per vehicle-year and per policy-year. (Answer: 0.500 claims per vehicle-year and 1.125 claims per policy-year.)
- A hypothetical portfolio has $6,300,000 of payroll and 42 claims. Calculate claims per $100,000 of payroll. (Answer: \(42/63=0.6667\) claims per $100,000 of payroll.)
- A business coverage has 15 policies with $2,000,000 of total sales and 9 claims. Calculate claims per policy and claims per $100,000 of sales. Explain why they are different measures. (Answer: 0.600 claims per policy and 0.450 claims per $100,000 of sales; the denominators represent different exposure units.)
- A hypothetical auto portfolio has 25 policy-years, 70 vehicle-years, and 14 claims, each assigned to one vehicle. Calculate both rates and state which rate directly describes claims per insured vehicle-year. (Answer: 0.560 claims per policy-year and 0.200 claims per vehicle-year; the latter directly describes claims per insured vehicle-year.)