Why Exposure-Based Estimates Need a Common Basis
Two periods can have different expected ultimate losses for very different reasons: one may contain more exposure, while another may have a higher expected loss cost per exposure. A comparison of total dollars alone does not distinguish these effects. An exposure-based estimate makes the calculation explicit: estimate the number of exposure units, select the expected ultimate loss per unit, and combine them on a consistent basis.
This method is especially useful when periods have different exposure volumes or when premium is not the chosen measure of risk volume. The exposure might be earned vehicle-years, insured building-years, machine-hours, or another defined measure. The unit matters: an expected loss cost per vehicle-year cannot be multiplied by a count of vehicles unless each vehicle represents exactly one earned vehicle-year.
Here, “ultimate” means the estimate of the total loss for the period, including losses not yet paid or reported, rather than the amount already observed at a valuation date. The calculation itself does not estimate a development pattern. It applies an expected loss cost to exposure. If the selected cost includes loss adjustment expense (LAE), the resulting ultimate estimate includes that LAE; if it excludes LAE, the estimate excludes LAE as well.
The Calculation and Its Basis
The model is deterministic once the exposure and expected loss cost selections have been made. It assumes that \(c_j\) is an appropriate expected ultimate cost for each unit of exposure in period \(j\), on the stated basis. No independence assumption is needed to multiply the number of units by the expected cost: the expectation of a sum is the sum of the expectations. The exposure units must be sufficiently comparable for a single selected cost to be meaningful; when they are not, divide them into appropriate classes.
A short justification makes the formula clear. Suppose period \(j\) contains \(E_j\) comparable exposure units, and let \(L_{jk}\) be the ultimate loss associated with unit \(k\). If each unit has expected loss \(E[L_{jk}]=c_j\), then the aggregate loss is \(S_j=\sum_{k=1}^{E_j}L_{jk}\). By additivity of expectation,
The units do not have to be statistically independent for this expectation identity to hold. However, if expected costs differ materially by risk class, using one cost for all exposures can distort the estimate. In that case, calculate by class and sum the class estimates.
In this expression, \(E_{jk}\) is the exposure in class \(k\) during period \(j\), and \(c_{jk}\) is the expected ultimate loss cost for that class. If a single average cost is needed for a period, it is the exposure-weighted cost \(U_j/\sum_k E_{jk}\), not the simple arithmetic average of the class costs.
A Consistent Multi-Period Workflow
Before comparing periods, check that the exposure count and the expected loss cost use the same definition of an exposure unit, coverage, loss scope, and time basis. If exposure amounts are reported in thousands, for example, either convert them to individual units or use a loss cost stated per thousand units. Apply the same convention to every period.
State exactly what one unit represents, including its time basis. Distinguish, for example, a policy from an earned policy-year.
Confirm that the expected loss cost applies to the same coverage, exposure unit, currency, and loss or LAE scope as the estimate.
Multiply each period’s exposure by its applicable expected loss cost, or calculate by risk class and sum.
Report expected ultimate dollars and, where useful, the exposure-weighted expected cost per unit. Separate volume changes from cost changes.
Worked Example: Uneven Exposure Volumes and Different Costs
A hypothetical insurer estimates expected ultimate losses for three accident periods using earned vehicle-years. The selected ultimate loss costs, including allocated LAE, are \$184 per vehicle-year for 2023, \$197 for 2024, and \$205 for 2025. Earned exposures are 12,000, 9,500, and 15,000 vehicle-years, respectively. Calculate each period’s expected ultimate, the combined expected ultimate, and the combined cost per vehicle-year.
Method and assumptions. Apply the exposure-based expected loss method. Each selected cost is assumed appropriate for its stated period and already includes the same loss and LAE scope. This is a deterministic expected-cost calculation; it uses no interest discounting, mortality assumption, or parametric frequency-severity distribution.
2023. Multiply 12,000 vehicle-years by \$184 per vehicle-year:
2024. Multiply 9,500 vehicle-years by \$197 per vehicle-year:
2025. Multiply 15,000 vehicle-years by \$205 per vehicle-year:
Combine periods. Total exposure is \(12{,}000+9{,}500+15{,}000=36{,}500\) vehicle-years. Total expected ultimate is \(\$2{,}208{,}000+\$1{,}871{,}500+\$3{,}075{,}000=\$7{,}154{,}500\). Therefore the combined exposure-weighted cost is
The period estimates sum to the combined estimate, and dividing that sum by total exposure gives the weighted cost. A simple average of the three selected costs would give \((184+197+205)/3=\$195.33\), which is not the combined cost because the periods have unequal exposure volumes.
When Expected Costs Differ by Class
If a period contains several classes with different expected costs, the total is the sum of class-level expected ultimates. The overall cost per exposure is then a weighted average, with each class’s exposure as its weight. This avoids treating a high-cost class and a low-cost class as though they had equal volume.
Worked Example: Changing Class Mix Across Periods
A hypothetical equipment insurer uses two risk classes, standard and heavy-use. For the first period, it has 4,000 standard equipment-years at an expected ultimate cost of \$120 per unit and 2,000 heavy-use equipment-years at \$260 per unit. For the second period, it has 2,500 standard equipment-years at \$126 and 4,500 heavy-use equipment-years at \$273. Calculate each period’s expected ultimate and weighted cost per equipment-year, then compare them.
Method and assumptions. Calculate expected ultimate by class and add the results. The selected costs are ultimate costs on the same coverage and expense basis. No interest, mortality, or parametric loss distribution is assumed or needed.
First period. Standard-class expected ultimate is \(4{,}000(120)=\$480{,}000\). Heavy-use expected ultimate is \(2{,}000(260)=\$520{,}000\). Thus total exposure is \(4{,}000+2{,}000=6{,}000\) equipment-years, and total expected ultimate is \(\$480{,}000+\$520{,}000=\$1{,}000{,}000\). Its weighted cost is \(\$1{,}000{,}000/6{,}000=\$166.6667\), or \$166.67 per equipment-year.
Second period. Standard-class expected ultimate is \(2{,}500(126)=\$315{,}000\). Heavy-use expected ultimate is \(4{,}500(273)=\$1{,}228{,}500\). Thus total exposure is \(2{,}500+4{,}500=7{,}000\) equipment-years, and total expected ultimate is \(\$315{,}000+\$1{,}228{,}500=\$1{,}543{,}500\). Its weighted cost is \(\$1{,}543{,}500/7{,}000=\$220.50\) per equipment-year.
The second period’s expected ultimate is \$543,500 higher, and its weighted cost is \$53.83 higher per equipment-year. Both the increase in total exposure, from 6,000 to 7,000 units, and the shift toward the higher-cost heavy-use class contribute to the larger total. The class calculations reconcile: the two class ultimates in each period add to that period’s total.
Common mistake avoided: Averaging the two class costs equally would give \$190 in the first period and \$199.50 in the second. Those are not the correct period costs because the class exposures are unequal; divide total expected ultimate by total exposure instead.
Converting Exposure Before Applying the Cost
A count of policies or insured objects is not necessarily the exposure measure. When coverage is in force for only part of a year, or the expected cost is stated per a different unit, first convert to the stated exposure basis. For earned annual exposure, a policy earning half a year contributes 0.5 policy-year, not one full policy-year. This conversion must be applied consistently across periods.
Worked Example: Partial-Year Policies and Annual Exposure Costs
A hypothetical small-business insurer selected expected ultimate costs per earned policy-year of \$210, \$230, and \$244 for three successive periods. The periods contain 10,000, 12,000, and 9,000 policies, but their average earned fractions are 0.80, 0.65, and 0.95, respectively. Find each period’s earned exposure and expected ultimate, then calculate the combined cost per earned policy-year.
Method and assumptions. Convert policy counts to earned policy-years by multiplying each count by its average earned fraction. Then apply the selected cost per earned policy-year. The costs are assumed to have matching coverage and loss scope. This calculation uses no interest discounting, mortality assumption, or parametric loss distribution.
The earned exposures are \(10{,}000(0.80)=8{,}000\), \(12{,}000(0.65)=7{,}800\), and \(9{,}000(0.95)=8{,}550\) earned policy-years.
The period ultimates are:
Total earned exposure is \(8{,}000+7{,}800+8{,}550=24{,}350\) policy-years. Total expected ultimate is \(\$1{,}680{,}000+\$1{,}794{,}000+\$2{,}086{,}200=\$5{,}560{,}200\). The combined cost is therefore
The calculation checks because multiplying the combined exposure by the combined cost before rounding returns the combined expected ultimate. Using the raw policy count of 31,000 in place of 24,350 earned policy-years would overstate the exposure base and apply a cost to the wrong unit.
Interpreting Differences and Avoiding Basis Errors
For two periods, the change in expected ultimate depends on both exposure and expected cost. One exact way to express the change from period 1 to period 2 is:
This identity follows by expanding the right-hand side: \(E_2c_1-E_1c_1+E_2c_2-E_2c_1=E_2c_2-E_1c_1=U_2-U_1\). Under this ordering, the first term measures the exposure change at the first period’s cost, and the second measures the cost change applied to the second period’s exposure. Another ordering could allocate the interaction differently, so state the convention if presenting a decomposition.
Before accepting a multi-period calculation, verify the following:
- Exposure unit: Do not mix policy counts, earned policy-years, vehicle-years, or thousands of exposure units without conversion.
- Cost unit: A cost per 1,000 units must be paired with exposure in thousands, or converted consistently before multiplication.
- Scope: Confirm that the selected cost and requested estimate both include or exclude LAE and other included loss components on the same basis.
- Aggregation: Sum class or period ultimates first; calculate an overall cost by dividing by total exposure, not by taking an unweighted average of selected costs.
- Rounding: Carry full precision through intermediate calculations and round only the reported result.
An exposure-based expected ultimate is not automatically a reserve estimate. It estimates total ultimate losses from the selected cost and exposure. A reserve requires a separate comparison with amounts already paid or otherwise recognized, on a consistent basis. The distinction matters: calling the full expected ultimate “unpaid” would count amounts already paid as if they remained outstanding.
Check Your Understanding
For each problem, keep exposure units and expected loss costs on the same basis. Answers are rounded only where stated.
- A hypothetical liability portfolio has 3,200 earned account-years in one period at an expected ultimate cost of \$145 per account-year, and 4,100 in the next period at \$152. Find both period ultimates and the combined exposure-weighted cost per account-year. (Answer: \$464,000 and \$623,200; combined cost \(\$1{,}087{,}200/7{,}300=\$148.9315\), approximately \$148.93 per account-year.)
- A hypothetical property portfolio has 1,800 standard building-years at \$90 per unit and 1,200 high-protection building-years at \$210 per unit. Find the total expected ultimate and weighted cost per building-year. (Answer: \(\$162{,}000+\$252{,}000=\$414{,}000\); \$138 per building-year.)
- A hypothetical machinery book has 2,400 thousand operating hours at an expected ultimate cost of \$38 per thousand hours in one period, and 3,100 thousand operating hours at \$41 per thousand hours in another. Calculate each period’s ultimate and the combined cost per thousand operating hours. (Answer: \$91,200 and \$127,100; combined ultimate \$218,300 and combined cost \(\$218{,}300/5{,}500=\$39.6909\), approximately \$39.69 per thousand hours.)
- A hypothetical auto portfolio has 6,500 earned vehicle-years at \$310 per unit in one period and 7,200 at \$296 per unit in the next. Which period has the larger expected ultimate, and what is the combined weighted cost per vehicle-year? (Answer: \$2,015,000 and \$2,131,200, so the second is larger; combined cost \(\$4{,}146{,}200/13{,}700=\$302.6423\), approximately \$302.64 per vehicle-year.)
- A hypothetical insurer reports 8,000 policies averaging 0.75 earned policy-years each, at an expected ultimate cost of \$180 per earned policy-year. A later period has 7,500 policies averaging 0.90 earned policy-years each, at \$190 per earned policy-year. Find both earned exposures, both expected ultimates, and the combined weighted cost. (Answer: 6,000 and 6,750 earned policy-years; \$1,080,000 and \$1,282,500; combined cost \(\$2{,}362{,}500/12{,}750=\$185.2941\), approximately \$185.29 per earned policy-year.)