Why Claim-Count Standards Need an Exposure Conversion
A full-credibility standard is often stated as a required number of claims. For example, a frequency estimate may be considered fully credible under a specified standard once the expected claim count reaches \(N_0\). But a ratemaking or portfolio-planning question may instead ask how much exposure is needed to produce that count. The conversion depends on the expected claim frequency per exposure unit.
An exposure unit is the defined basis on which frequency is measured. It might be one vehicle-year, one dwelling-year, or another consistently defined unit. The units must match: a frequency stated per 1,000 machine-years cannot be divided into a claim count as if it were a frequency per machine-year.
The quantity \(N_0\) is the count standard; \(E_0\) is the exposure needed to meet it at the assumed frequency. This tutorial focuses on converting between those quantities. The count standard itself comes from the selected credibility criterion and its confidence and tolerance assumptions; it is not determined by the exposure conversion.
Model and Conversion
Assume claim counts follow a Poisson model with a constant expected frequency \(\lambda\) per exposure unit. For \(E\) units of exposure, the model is \(N\sim\operatorname{Poisson}(E\lambda)\), so \(E[N]=E\lambda\) and \(\operatorname{Var}(N)=E\lambda\). Setting the expected claim count equal to the required count gives \(E\lambda=N_0\), and therefore \(E_0=N_0/\lambda\).
This calculation is about the expected count, not a guarantee that the realized count will equal \(N_0\). Random variation means a portfolio with \(E_0\) exposure can produce fewer or more claims. The count standard is what the chosen full-credibility rule requires under its stated assumptions.
The units verify the direction of the division:
There is also a useful precision interpretation. The observed frequency is \(\hat{\lambda}=N/E\). Under the stated Poisson model, its variance is \(\operatorname{Var}(\hat{\lambda})=\lambda/E\), and its coefficient of variation is \(1/\sqrt{E\lambda}\). Since \(E\lambda\) is the expected claim count, the frequency estimate becomes more stable as the expected count grows. A count-based credibility criterion therefore translates into an exposure requirement by asking how much exposure is needed to reach its required expected count.
Use the required expected claim count \(N_0\) from the selected full-credibility criterion.
Express \(\lambda\) as expected claims per one of the exposure units used in the question.
Calculate \(E_0=N_0/\lambda\), preserving the exposure unit in the answer.
Multiply \(E_0\lambda\) to verify the required count. Round up only when exposure must be acquired in indivisible units.
Worked Conversions
Worked Example: Vehicle-Years at a Given Frequency
A hypothetical private-passenger auto class has a full-credibility count standard of 1,200 claims. Its expected frequency is 0.08 claims per vehicle-year. Find the required exposure in vehicle-years.
Method. Convert the count standard using \(E_0=N_0/\lambda\). The frequency is already expressed per one vehicle-year, so no further scale conversion is needed.
Substitute \(N_0=1{,}200\) claims and \(\lambda=0.08\) claims per vehicle-year:
Check. The expected count at this exposure is \(15{,}000(0.08)=1{,}200\) claims, exactly the count standard.
Common mistake avoided. Multiplying \(1{,}200\) by \(0.08\) gives 96, not exposure. It calculates the expected count associated with 1,200 vehicle-years, not the exposure required for 1,200 expected claims.
Worked Example: Find the Count Standard, Then Convert
A hypothetical equipment-insurance analysis uses a normal-approximation count standard with a two-sided confidence level of 95% and a relative error tolerance of 5%. Use \(z=1.96\) and \(N_0=(z/k)^2\), where \(k\) is the tolerance as a decimal. Expected frequency is 0.12 claims per equipment-year. Find the required exposure in whole equipment-years.
Method. First calculate the count standard under the stated approximation, round that required count up to the next whole claim, and then convert it to exposure.
The tolerance is \(k=0.05\). Thus:
Because the required count is a whole number and the count standard must not be undershot, use \(N_0=1{,}537\) claims. The exposure calculation is:
The question requires whole equipment-years, so round upward to \(12{,}809\). Rounding downward could leave the exposure below the calculated requirement. At the rounded exposure, the expected claim count is \(12{,}809(0.12)=1{,}537.08\), which is at least 1,537.
Common mistake avoided. Using 1,536.64 as though it were an exact whole-claim standard, then rounding exposure down, can undershoot the required count. Here the count and exposure are rounded upward at their respective whole-unit steps.
Worked Example: Frequency Stated per 1,000 Units
A hypothetical fleet of specialized service machines has an expected frequency of 2.4 claims per 1,000 machine-years. A claim-count standard requires 720 claims. Find the exposure in machine-years.
Method. Convert the frequency to claims per single machine-year before using the formula. The given frequency is \(2.4/1{,}000=0.0024\) claims per machine-year.
An equivalent calculation keeps the stated rate scale intact: at 2.4 claims per 1,000 machine-years, the exposure needed for 720 claims is \(720/2.4=300\) groups of 1,000 machine-years. Therefore \(300(1{,}000)=300{,}000\) machine-years.
Check. \(300{,}000/1{,}000=300\) groups of 1,000 machine-years, and \(300(2.4)=720\) expected claims.
Common mistake avoided. Dividing 720 by 2.4 and reporting 300 machine-years forgets that the frequency is per 1,000 machine-years. The result is 300 groups of 1,000, not 300 individual machine-years.
Assumptions, Units, and Interpretation
The direct conversion \(E_0=N_0/\lambda\) is appropriate when exposure units are defined consistently and the expected frequency applies to those units. The simple Poisson model also assumes a stable frequency and independent claim occurrences over the exposure being combined. It is a model for the count; it does not establish that the assumed frequency is accurate or that the data are suitable for ratemaking.
If the frequency estimate is stated on a different scale, convert it before dividing. For example, a rate of \(r\) claims per \(b\) exposure units corresponds to \(\lambda=r/b\) claims per exposure unit. The required exposure is then \(E_0=N_0/(r/b)=N_0b/r\). The factor \(b\) is essential; omitting it changes the exposure by a factor of \(b\).
| Given frequency basis | Frequency per one exposure unit | Exposure conversion |
|---|---|---|
| \(r\) claims per 100 exposure units | \(r/100\) | \(E_0=100N_0/r\) |
| \(r\) claims per 1,000 exposure units | \(r/1{,}000\) | \(E_0=1{,}000N_0/r\) |
| \(\lambda\) claims per exposure unit | \(\lambda\) | \(E_0=N_0/\lambda\) |
A low expected frequency requires more exposure to reach the same expected claim count. For instance, if the count standard is fixed, halving \(\lambda\) doubles \(E_0\), because \(N_0/\left(\lambda/2\right)=2N_0/\lambda\). This is a direct consequence of the conversion, not a change in the credibility standard.
The exposure requirement is not automatically a number of policies. It is the amount of the stated exposure measure. Converting it to a policy count is only valid if the exposure contributed by each policy is known and expressed on the same basis. In addition, combining different risks under one frequency requires that the selected \(\lambda\) reasonably represents the combined exposure. If expected frequencies differ materially across the included risks, a single homogeneous-rate conversion may not describe the portfolio.
When exposure is divisible, such as a measured amount of exposure on a continuous scale, report the calculated value at suitable precision. When the requested units are indivisible, round upward to ensure the stated exposure meets or exceeds the threshold. Do not round a result downward merely to obtain a convenient whole number.
Common Exam Traps
- Reversing the ratio. Required exposure is count divided by frequency. Frequency divided by count does not have exposure units.
- Ignoring the rate scale. A rate per 100 or per 1,000 units must be converted or carried through with its scale factor.
- Changing the basis without changing the number. A vehicle-year, dwelling-year, and policy-year are not interchangeable unless the problem explicitly defines them as equivalent.
- Treating expected count as a guaranteed realized count. \(E_0\lambda=N_0\) matches the expected count to the standard; Poisson variation remains.
- Rounding down when units must be whole. Round the required exposure upward when a smaller whole-unit amount would fall short.
- Assuming more exposure fixes every issue. This conversion addresses count volume under the model. It does not correct an unsuitable frequency assumption, poor data quality, or a mismatch between the experience and the intended application.
Check Your Understanding
For each problem, use the stated frequency basis carefully and give exposure in the requested units.
- A hypothetical rental-property class has a count standard of 600 claims and an expected frequency of 0.075 claims per dwelling-year. What exposure is required in dwelling-years? (Answer: 8,000 dwelling-years.)
- A hypothetical watercraft class uses \(z=1.645\), a relative error tolerance of 5%, and \(N_0=(z/k)^2\). Its frequency is 0.09 claims per boat-year. Find the required exposure in whole boat-years, rounding the count standard and exposure upward as needed. (Answer: 12,034 boat-years.)
- A hypothetical agricultural machine portfolio has an expected frequency of 2.4 claims per 1,000 machine-years and a count standard of 720 claims. How many machine-years are required? (Answer: 300,000 machine-years.)
- A hypothetical commercial property class requires 1,250 expected claims. Its frequency is 0.16 claims per insured-year. Find the required exposure in whole insured-years. (Answer: 7,813 insured-years.)
- A hypothetical courier-vehicle class has a standard of 1,800 claims and a frequency of 4.5 claims per 100 vehicle-years. Calculate the required exposure in vehicle-years and state why dividing 1,800 directly by 4.5 gives the wrong unit. (Answer: 40,000 vehicle-years; \(1{,}800/4.5=400\) groups of 100 vehicle-years, not 400 vehicle-years.)