Why Exposure Must Match the Coverage
An exposure measure is the denominator used to describe how much risk was insured. The right denominator depends on the coverage being studied. For auto coverage, a vehicle-year measures insured vehicle exposure over time; for workers’ compensation, payroll is a common measure of covered work activity; for property coverage, insured value may be the relevant measure. The policy itself is not always the correct unit: one policy can contain several coverages, each with its own exposure basis.
This matters whenever claim counts or losses are compared across risks, periods, or segments. Dividing by a denominator with the wrong meaning may produce a number that looks precise but does not answer the intended question. Even correctly calculated rates are not automatically comparable if their exposure units differ.
The coverage determines which risk characteristic the denominator is intended to represent. The exposure field should therefore be selected at the coverage level, not merely because it is available in a policy record. A policy count, employee count, vehicle count, payroll amount, sales amount, and insured value are different measures; they are not interchangeable just because each can be summarized as a number.
For an exposure-based claim frequency, let \(N\) be the number of claims in a defined coverage segment and let \(E\) be the corresponding exposure amount. The basic observed frequency is \(N/E\). Its units are claims per unit of exposure, so the unit belongs in the answer: for example, claims per vehicle-year or claims per $100,000 of payroll. A frequency without its denominator’s unit is incomplete.
Matching Coverage to Exposure Fields
Selection begins by stating the coverage and the risk quantity that its exposure basis measures. Then verify that the field is measured for the same coverage and experience period as the claims. The following are common matches, not universal rules for every product or analysis:
| Coverage being measured | Potential exposure basis | Unit to preserve |
|---|---|---|
| Auto coverage | Insured vehicle exposure over time | Vehicle-years |
| Workers’ compensation | Covered payroll | Payroll dollars, or a stated payroll unit such as $100 |
| Property coverage | Insured property value over time | Insured-value-years, with the value scale stated |
| Sales-based liability coverage | Covered sales or receipts | Sales dollars, or a stated sales unit |
A listed match is meaningful only when the exposure definition is consistent. For example, “payroll” must refer to payroll in the relevant workers’ compensation coverage and measurement period, not a company-wide figure that includes excluded operations. Similarly, a property value field must describe value insured under the property coverage being analyzed, not an unrelated policy limit. Any scale factor—such as “per $100 of payroll”—must be carried through the calculation.
Identify which insured benefit or peril the claim count or loss measure represents.
Choose the field that represents the relevant insured quantity, such as vehicles, covered payroll, or insured property value.
Confirm the exposure refers to the same coverage, insured population, and experience period as the claims.
Write the frequency or loss measure with its exposure unit and any scaling convention.
Frequency per Exposure Unit
The ratio of claims to exposure is useful when the exposure units are consistently defined. Its interpretation relies on an important assumption: within the segment being analyzed, expected claim count is proportional to exposure. In other words, if exposure doubles while the underlying claim frequency per unit stays constant, expected claims double. This assumption can be unsuitable when the segment mixes materially different risks or when the exposure definition does not reflect the coverage being measured.
The formula follows directly from the model. For \(N=n\), the Poisson likelihood is proportional to \(e^{-\lambda E}(\lambda E)^n\). Its log-likelihood, omitting terms that do not depend on \(\lambda\), is \(n\log(\lambda E)-\lambda E\). Differentiating with respect to \(\lambda\) gives \(n/\lambda-E\). Setting this equal to zero yields \(\hat{\lambda}=n/E\); the second derivative is \(-n/\lambda^2<0\) when \(n>0\), so this is a maximum. When \(n=0\), the likelihood is maximized at the boundary \(\hat{\lambda}=0\), also equal to \(n/E\).
Worked Applications
Worked Example: Three Coverages, Three Denominators
A hypothetical commercial account has three coverage segments during one experience period. Auto coverage has 4 claims and 25 vehicle-years of exposure. Workers’ compensation has 5 claims and $1,000,000 of payroll. Property coverage has 2 claims and $8,000,000 of insured-value-years. Calculate the observed claim frequency for each coverage. State the rate in claims per $100,000 of payroll for workers’ compensation and per $1,000,000 of insured-value-years for property.
Method. For each coverage, divide its claim count by its matching exposure. Convert the payroll and property denominators to the requested units before reporting the rates.
Auto. The basis is vehicle-years, so \(4\div25=0.16\) claims per vehicle-year.
Workers’ compensation. $1,000,000 of payroll contains 10 units of $100,000. Therefore, \(5\div10=0.5\) claims per $100,000 of payroll.
Property. $8,000,000 of insured-value-years contains 8 units of $1,000,000. Therefore, \(2\div8=0.25\) claims per $1,000,000 of insured-value-years.
Check. Each rate uses only the claim count and exposure for its named coverage, and each reported denominator matches the requested unit. The rates are not directly comparable: one is per vehicle-year, one per payroll amount, and one per insured-value amount.
Common mistake avoided. Adding the three claim counts and dividing by the sum of the three exposure amounts would combine incompatible units. The expression \(11/(25+10+8)\) has no meaningful exposure denominator and is not an overall claim frequency.
Worked Example: Choosing Payroll Rather Than Employee Count
A hypothetical workers’ compensation segment reports 18 claims, $2,400,000 of covered payroll, and 300 employees. The analysis requires observed claim frequency per $100,000 of payroll. Calculate the rate and explain why employee count is not the denominator for this requested measure.
Method. Use payroll because the requested basis is payroll. Express the payroll in $100,000 units before dividing.
The number of exposure units is \(2{,}400{,}000\div100{,}000=24\). The frequency is \(18\div24=0.75\) claims per $100,000 of payroll.
The employee count would give \(18\div300=0.06\) claims per employee. That is a different statistic, with a different denominator and interpretation; it does not answer the requested payroll-based question. It cannot be substituted for the payroll measure simply because employee count is also available.
Check. Multiplying the reported frequency by the 24 payroll units gives \(0.75\times24=18\) expected claims at the observed rate, matching the claim count used in the calculation.
Common mistake avoided. Dividing by 300 and labeling the result “per $100,000 of payroll” would give the correct arithmetic for the wrong exposure basis. The unit in the label must agree with the denominator actually used.
Worked Example: Matching Fields in a Package of Coverages
A hypothetical business reports 3 property claims, 2 auto claims, and 4 sales-based liability claims. Its corresponding exposure records are $12,000,000 of property insured-value-years, 10 vehicle-years, and $5,000,000 of covered sales. Calculate property frequency per $1,000,000 of insured-value-years, auto frequency per vehicle-year, and liability frequency per $1,000,000 of sales.
Method. Match each claim count to its coverage-specific field, then convert the dollar-based fields to the requested million-dollar units.
Property. The exposure is \(12{,}000{,}000\div1{,}000{,}000=12\) units. Thus \(3\div12=0.25\) claims per $1,000,000 of insured-value-years.
Auto. The rate is \(2\div10=0.2\) claims per vehicle-year.
Sales-based liability. The exposure is \(5{,}000{,}000\div1{,}000{,}000=5\) units. Thus \(4\div5=0.8\) claims per $1,000,000 of covered sales.
Check. Reversing each calculation gives \(0.25\times12=3\) property claims, \(0.2\times10=2\) auto claims, and \(0.8\times5=4\) liability claims.
Common mistake avoided. Using the $5,000,000 sales field for property would produce \(3\div5=0.6\) claims per $1,000,000 of sales. That is not a property frequency per insured value: it answers a different, generally inappropriate denominator question. A numerically computable ratio is not necessarily a valid coverage measure.
Common Interpretation Traps
A frequent error is to treat all exposure as a generic count of policies. A policy may cover several vehicles, a building and its contents, or multiple lines with distinct terms. If the selected denominator is one policy for every record, it may not reflect how much of the relevant insured risk was actually present. The coverage-specific field and its definition must guide the calculation.
Another trap is to compare two frequencies without checking their units. A value of 0.5 claims per $100,000 of payroll cannot be ranked directly against 0.2 claims per vehicle-year. Neither number is inherently larger in actuarial meaning; the denominators represent different exposure quantities. Comparisons require a common, appropriate basis or a separate analysis that accounts for the different risk measures.
The same caution applies when reporting aggregate results. Counts and exposures can be summed across records when they represent the same coverage and use a consistent unit. They should not be pooled across coverage types merely to obtain a single denominator. If risk characteristics differ substantially within a coverage, an overall frequency may also conceal important differences; the constant-frequency assumption should be considered before interpreting the aggregate ratio.
Check Your Understanding
For each problem, identify the relevant exposure basis, calculate the requested frequency, and keep its unit in the answer.
- A hypothetical auto segment has 6 claims and 30 vehicle-years. What is its claim frequency per vehicle-year? (Answer: 0.2 claims per vehicle-year.)
- A hypothetical workers’ compensation segment has 8 claims and $1,600,000 of covered payroll. Calculate frequency per $100,000 of payroll. (Answer: 0.5 claims per $100,000 of payroll.)
- A hypothetical property portfolio has 5 claims and $20,000,000 of insured-value-years. Calculate frequency per $1,000,000 of insured-value-years. (Answer: 0.25 claims per $1,000,000 of insured-value-years.)
- A hypothetical sales-based liability segment has 9 claims, $3,000,000 of covered sales, and 150 employees. What is frequency per $1,000,000 of sales, and why is dividing by employees not the answer to that question? (Answer: 3 claims per $1,000,000 of sales; dividing by employees gives a per-employee measure instead.)
- A hypothetical account has 4 auto claims and 20 vehicle-years, plus 3 workers’ compensation claims and $600,000 of payroll. Calculate each frequency, using $100,000 payroll units. Can the two rates be added to form one overall frequency? (Answer: 0.2 claims per vehicle-year and 0.5 claims per $100,000 of payroll; no, the units are incompatible.)