Why Adjust for Exposure?
A portfolio with more policies will often produce more claims simply because it has more opportunities for claims to occur. Comparing raw claim counts can therefore be misleading. Exposure-adjusted metrics put claim counts and losses on a common basis, making comparisons more informative when portfolios have different sizes or observation periods.
Three quantities are useful and distinct: claim frequency measures claims relative to exposure; average severity measures loss per claim; and aggregate loss per exposure measures total loss relative to exposure. The denominators matter. A frequency denominator is exposure, a severity denominator is claim count, and an aggregate-loss denominator is exposure.
An exposure unit must be specified. Depending on the hypothetical portfolio, it might be one insured vehicle-year, one policy-year, or a stated amount of payroll. A fractional exposure is allowed: for example, half a policy-year contributes 0.5 policy-years. If a rate is reported per 1,000 exposure units, multiply the frequency \(N/E\) by 1,000. The unit is part of the rate: 8 claims per 1,000 vehicle-years is not the same description as 8 claims per 1,000 policy-years.
Three Metrics, Three Interpretations
| Metric | Calculation | What it measures |
|---|---|---|
| Claim frequency | \(N/E\) | Claims per exposure unit |
| Average severity | \(S/N\) | Average loss amount per claim |
| Aggregate loss per exposure | \(S/E\) | Loss amount per exposure unit |
These measures answer different questions. Frequency is a count rate, severity is a dollar amount per claim, and aggregate loss per exposure is a dollar amount per exposure unit. Frequency can exceed one claim per exposure unit if multiple claims can occur during that unit; it is not generally a probability.
For observed data with at least one claim, the definitions give an exact relationship:
The claim count cancels. Thus, observed aggregate loss per exposure equals observed frequency times observed average severity. This is an arithmetic identity for the same portfolio, observation period, claim definition, and loss basis. For instance, one cannot combine a frequency based on reported claims with a severity based on a different set of settled claims and expect the identity to describe those mismatched figures.
Worked Comparisons
Worked Example: Comparing Portfolios of Different Sizes
Two hypothetical books of vehicle insurance are observed for one year. Book A has 12,000 vehicle-years, 96 claims, and $480,000 of aggregate loss. Book B has 8,000 vehicle-years, 72 claims, and $540,000 of aggregate loss. Calculate each book’s frequency per 1,000 vehicle-years, average severity, and aggregate loss per vehicle-year. Which book has the higher claim rate, and which has higher loss per exposure?
Model and assumptions. These are hypothetical observed totals. No interest or mortality assumption is relevant, and no probability distribution is assumed for the counts or claim amounts. Loss totals are measured on the same basis for both books, and each exposure unit is one vehicle-year.
Book A frequency. Divide claims by exposure: \(96/12{,}000=0.008\) claims per vehicle-year. Per 1,000 vehicle-years, this is \(0.008\times1{,}000=8\) claims.
Book A severity. Divide aggregate loss by claims: \(\$480{,}000/96=\$5{,}000\) per claim.
Book A aggregate loss per exposure. \(\$480{,}000/12{,}000=\$40\) per vehicle-year. Check the identity: \(0.008\times\$5{,}000=\$40\).
Book B frequency. \(72/8{,}000=0.009\) claims per vehicle-year, or \(0.009\times1{,}000=9\) claims per 1,000 vehicle-years.
Book B severity. \(\$540{,}000/72=\$7{,}500\) per claim.
Book B aggregate loss per exposure. \(\$540{,}000/8{,}000=\$67.50\) per vehicle-year. Check: \(0.009\times\$7{,}500=\$67.50\).
Book B has the higher claim rate, 9 rather than 8 claims per 1,000 vehicle-years, and the higher loss per exposure, $67.50 rather than $40. Its higher loss per exposure reflects both a higher observed frequency and a higher observed severity. The calculations are checked by the exact frequency-times-severity identity for each book. The method avoids comparing raw counts, which would favor the smaller book simply because it has fewer exposures.
Worked Example: Pooling Groups With Different Exposure Volumes
A hypothetical insurer reviews two groups with the same type of exposure and the same observation period. Group R has 2,000 exposure units, 30 claims, and $150,000 aggregate loss. Group T has 6,000 exposure units, 42 claims, and $420,000 aggregate loss. Calculate the pooled frequency, pooled severity, and pooled aggregate loss per exposure. Explain why taking simple averages of the two groups’ rates is not the correct pooling method.
Model and assumptions. The figures are hypothetical descriptive totals, not simulated outcomes. No interest or mortality assumption or probability distribution is used. Both groups use the same exposure unit and consistent claim and loss definitions.
Combine totals first. Total exposure is \(2{,}000+6{,}000=8{,}000\). Total claims are \(30+42=72\). Total aggregate loss is \(\$150{,}000+\$420{,}000=\$570{,}000\).
Pooled frequency. \(72/8{,}000=0.009\) claims per exposure unit, or 9 claims per 1,000 exposure units.
Pooled severity. \(\$570{,}000/72=\$7{,}916.666\ldots\) per claim, which is $7,916.67 to the nearest cent.
Pooled aggregate loss per exposure. \(\$570{,}000/8{,}000=\$71.25\) per exposure unit. Using the unrounded pooled severity, \(0.009\times\$7{,}916.666\ldots=\$71.25\). The results reconcile.
A simple average of the group frequencies would be \((30/2{,}000+42/6{,}000)/2=(0.015+0.007)/2=0.011\), which is not the pooled rate of 0.009. The larger group must contribute more weight because it has more exposure. Similarly, the simple average of group severities, \((\$5{,}000+\$10{,}000)/2=\$7{,}500\), is not the pooled severity. Pooling uses total claims and total losses, not equal weights for groups of unequal size. This method avoids the trap of averaging ratios without accounting for their denominators.
Worked Example: Recovering Counts and Loss From Rates
A hypothetical property portfolio has 5,000 exposure units. Its observed claim frequency is reported as 6.4 claims per 1,000 exposure units, and its observed average severity is $4,200 per claim. Find the implied claim count, aggregate loss, and aggregate loss per exposure unit. Verify the result in two ways.
Model and assumptions. Treat the stated rate and severity as exact descriptive values for this hypothetical portfolio. No interest or mortality assumption or probability distribution is specified or needed. The exposure unit and loss basis are consistent throughout.
Convert the frequency to a per-unit rate. \(6.4/1{,}000=0.0064\) claims per exposure unit.
Find the implied claim count. \(0.0064\times5{,}000=32\) claims.
Find aggregate loss. \(32\times\$4{,}200=\$134{,}400\).
Find aggregate loss per exposure. \(\$134{,}400/5{,}000=\$26.88\) per exposure unit.
For the first check, multiply frequency by severity: \(0.0064\times\$4{,}200=\$26.88\). For the second check, reconstruct aggregate loss from the per-exposure result: \(\$26.88\times5{,}000=\$134{,}400\). Both checks agree. The important unit conversion is dividing the quoted “per 1,000” frequency by 1,000 before multiplying by the exposure count.
Comparison and Interpretation
Exposure adjustment makes portfolio comparisons more meaningful, but it does not make unlike data comparable automatically. Before interpreting a difference, check that the exposure unit, time period, covered population, claim-counting rule, and loss basis are aligned. A rate based on policy-years should not be directly compared with one based on vehicle-years unless the relationship between those units is known and relevant.
Also distinguish a rate from a probability. If a book has 12 claims per 1,000 exposure units, its frequency is 0.012 claims per exposure unit. That does not by itself say that 1.2% of exposure units had a claim: an exposure unit could have multiple claims, and the frequency calculation counts every claim. A probability of at least one claim requires a model or data summary specifically about whether an exposure unit had any claim.
Observed severity is an average across claims, not an average across all exposure units. If some exposure units have no claim, they contribute zero to aggregate loss per exposure but do not appear in the denominator of \(S/N\). This explains why average severity can be large while aggregate loss per exposure remains modest when frequency is low.
When combining groups, add the underlying totals before calculating a pooled metric. For frequency, divide total claims by total exposure. For severity, divide total loss by total claims. For aggregate loss per exposure, divide total loss by total exposure. If the component group ratios are averaged directly, the result generally gives each group equal weight regardless of its exposure or claim count.
Check Your Understanding
For each hypothetical portfolio, use the stated exposure unit and keep units visible in your calculations.
- A portfolio has 7,500 policy-years, 45 claims, and $315,000 aggregate loss. Find frequency per 1,000 policy-years, average severity, and aggregate loss per policy-year. (Answer: 6 claims per 1,000 policy-years; $7,000 per claim; $42 per policy-year.)
- Group A has 2,400 exposure units, 30 claims, and $180,000 loss. Group B has 6,000 exposure units, 42 claims, and $378,000 loss. Find the pooled frequency per 1,000 exposure units, pooled severity, and pooled loss per exposure unit. (Answer: 8.5714 claims per 1,000 exposure units; $7,750 per claim; $66.4286 per exposure unit.)
- A hypothetical portfolio has 10,000 exposure units, a rate of 7.5 claims per 1,000 exposure units, and average severity of $3,600. Find the implied number of claims, aggregate loss, and aggregate loss per exposure unit. (Answer: 75 claims; $270,000; $27 per exposure unit.)
- Two portfolios each have 4,000 exposure units. Portfolio P has 36 claims and $288,000 aggregate loss; portfolio Q has 44 claims and $286,000 aggregate loss. Calculate each portfolio’s frequency per 1,000 exposure units, severity, and loss per exposure unit. Identify which has the higher frequency and which has the higher loss per exposure. (Answer: P: 9 claims per 1,000, $8,000 per claim, $72 per exposure unit; Q: 11 claims per 1,000, $6,500 per claim, $71.50 per exposure unit. Q has higher frequency; P has higher loss per exposure.)
- A group has 1,200 exposure units and 18 claims, with average severity of $2,500. A second group has 2,800 exposure units and 42 claims, with average severity of $4,000. Find the pooled frequency per 1,000 exposure units, pooled severity, and pooled loss per exposure unit. (Answer: 15 claims per 1,000 exposure units; $3,550 per claim; $53.25 per exposure unit.)