Choosing an Expected Loss Cost

What an Expected Loss Cost Represents

An expected loss cost expresses expected losses for a defined coverage and population per unit of exposure. It is a useful benchmark when a candidate must choose among estimates based on historical ultimate losses. The central question is not simply which number is closest to an average: it is whether the underlying losses, exposure counts, and exposure unit are relevant to the risk being evaluated.

For example, a cost per insured-device-year is not automatically suitable for a rating plan that charges per contract-year if one contract can cover several devices. Likewise, a cost for all covered causes of loss is not a direct estimate of a cost for one cause alone. In each case, the denominator or loss scope changes the meaning of the result.

Definition: An expected loss cost is expected ultimate loss divided by the corresponding number of exposure units. For a period with ultimate losses $U$ and exposure count $E$, the observed ultimate loss cost is $U/E$, provided $E>0$. It is expressed in dollars per stated exposure unit.

An observed historical cost is not automatically the expected cost for future business. It is evidence used to assess that expected cost. Before comparing candidate values, confirm that the historical losses are ultimate (including estimates of amounts not yet paid or reported), that loss scope is consistent, and that exposures use the same definition and time basis as the target rating unit. Any necessary adjustments to make the data comparable must be made before selecting a value; the calculations below assume that the supplied inputs are already on a suitable common basis.

Pooled Cost and the Role of Exposure Volume

Suppose period $j$ has ultimate losses $U_j$ and exposure count $E_j$. Its individual-period cost is $U_j/E_j$. If the periods are comparable, the pooled cost divides total ultimate losses by total exposures:

\widehat{c}_{\mathrm{pooled}} = \frac{\sum_{j=1}^{n} U_j}{\sum_{j=1}^{n} E_j}.

This is generally not the same as the unweighted arithmetic mean of period costs. The pooled calculation gives more influence to periods with more exposure units. That is appropriate when each exposure unit is the unit being rated and the periods are sufficiently comparable. It is not a substitute for checking whether the periods actually represent the target risk.

Derivation: Write each period's cost as $c_j=U_j/E_j$, so $U_j=E_jc_j$. Substituting into the pooled formula gives $$ \frac{\sum_j U_j}{\sum_j E_j} = \frac{\sum_j E_jc_j}{\sum_j E_j} = \sum_j\left(\frac{E_j}{\sum_k E_k}\right)c_j. $$ The weights $E_j/\sum_k E_k$ are nonnegative and sum to one. Thus, the pooled cost is an exposure-weighted mean of the period costs. An unweighted mean instead assigns every period the same weight, regardless of its exposure volume.

A simple model explains why this pooled measure is useful. Assume that comparable exposure units have a common expected ultimate loss cost $c$, and that period $j$ has $E_j$ such units. Then

\operatorname{E}[U_j\mid E_j]=E_jc.

Summing over periods gives expected total ultimate losses of $c\sum_j E_j$, so dividing by total exposures gives $c$. This justification relies on a common expected cost per unit across the periods. If the risks, coverage, or exposure definitions differ materially, pooling can obscure rather than estimate the cost relevant to the target.

Worked Calculations

Worked Example: Pooling Periods With Different Exposure Counts

A hypothetical equipment-repair program reports ultimate losses of $240,000 on 8,000 repair-years in one period and $330,000 on 15,000 repair-years in a second period. The periods have the same coverage and are already on a comparable basis. Calculate the pooled cost and the unweighted mean of the period costs. Which is the exposure-weighted estimate?

Method and assumptions. Use ultimate losses divided by repair-years. Assume both periods represent the same target repair-year exposure and loss scope. No discounting, mortality assumption, or parametric loss distribution is used; the calculation uses the supplied ultimate losses and exposure counts.

First-period cost:

\frac{\$240{,}000}{8{,}000\ \text{repair-years}} = \$30.00\ \text{per repair-year}.

Second-period cost:

\frac{\$330{,}000}{15{,}000\ \text{repair-years}} = \$22.00\ \text{per repair-year}.

The unweighted mean is $(\$30.00+\$22.00)/2=\$26.00$ per repair-year. For the pooled cost, total losses are $\$240{,}000+\$330{,}000=\$570{,}000$, and total exposure is $8{,}000+15{,}000=23{,}000$ repair-years. Therefore:

\frac{\$570{,}000}{23{,}000\ \text{repair-years}} = \$24.7826\ldots \approx \$24.78\ \text{per repair-year}.

Check and trap avoided. The weighted-mean check is $(8{,}000\times\$30+15{,}000\times\$22)/23{,}000=\$570{,}000/23{,}000$, again $24.78 per repair-year. The pooled result is exposure-weighted; $26.00 is the unweighted mean, not the pooled cost. Both are correctly calculated summaries, but they answer different questions.

Worked Example: Comparing a Pooled Cost With an Annual Average

A hypothetical service plan has three comparable periods. Their ultimate losses and exposures are $300,000 and 10,000 exposure-units, $420,000 and 14,000 exposure-units, and $280,000 and 8,000 exposure-units. Calculate the pooled cost and the unweighted mean of annual costs, then state which one gives each period equal weight.

Method and assumptions. Calculate each period's cost, then calculate both summaries. The exposure-unit definition and loss scope are the same in all three periods. No discounting, mortality assumption, or parametric loss distribution is assumed.

The period costs are:

\frac{\$300{,}000}{10{,}000}=\$30.00,\qquad \frac{\$420{,}000}{14{,}000}=\$30.00,\qquad \frac{\$280{,}000}{8{,}000}=\$35.00

The unweighted mean is $(\$30.00+\$30.00+\$35.00)/3=\$31.6667\ldots$, or $31.67 per exposure-unit to the nearest cent. Each period receives one-third of this average, regardless of its exposure count.

For the pooled cost, total losses are $\$300{,}000+\$420{,}000+\$280{,}000=\$1{,}000{,}000$. Total exposures are $10{,}000+14{,}000+8{,}000=32{,}000$. Thus:

\frac{\$1{,}000{,}000}{32{,}000\ \text{exposure-units}} = \$31.25\ \text{per exposure-unit}.

Check and trap avoided. The exposure weights are $10/32$, $14/32$, and $8/32$. Applying them to the respective costs gives $(10/32)(30)+(14/32)(30)+(8/32)(35)=31.25$. Using $31.67 as the pooled result would incorrectly give the smaller third period the same weight as the larger periods.

Worked Example: Selecting a Cost With the Correct Loss Scope

A hypothetical insurer is evaluating an expected loss cost for water-leak coverage per covered-device-year. A comparable historical dataset limited to water-leak losses has ultimate losses of $144,000 on 4,800 device-years in one period and $180,000 on 6,000 device-years in another. A separate, broader dataset has $450,000 of ultimate losses on 10,000 device-years, but includes all covered causes. Candidate costs are $30, $45, and $32.50 per device-year. Assess which candidate is supported for the stated target.

Method and assumptions. Compare the candidates with costs calculated on the target loss scope and exposure unit. Assume both leak-only periods are comparable to each other and to the target; the broad dataset is explicitly not leak-only. No discounting, mortality assumption, or parametric loss distribution is used.

The leak-only costs are:

\frac{\$144{,}000}{4{,}800}=\$30.00,\qquad \frac{\$180{,}000}{6{,}000}=\$30.00

Their pooled cost is $(\$144{,}000+\$180{,}000)/(4{,}800+6{,}000)=\$324{,}000/10{,}800=\$30.00$ per device-year. By contrast, the broad dataset gives $\$450{,}000/10{,}000=\$45.00$ per device-year, but that figure includes losses outside the target coverage. The candidate of $32.50 is not the pooled cost of the supplied leak-only data.

Selection and check. Of the three candidates, $30 per device-year is supported by the comparable leak-only experience. The $45 candidate is a valid calculation for a broader loss scope but does not estimate the stated target. Matching the denominator alone does not make loss data relevant if the losses cover different causes.

How to Assess Candidate Values

A defensible selection starts by defining exactly what one exposure unit represents and what losses are included. Then calculate the period costs and pooled cost from comparable data. These calculations reveal which candidate is consistent with the observed experience, but they do not remove the need for actuarial judgment. Historical variation, changes in the risks represented, and the relevance of each period may affect how much weight a candidate deserves.

State the target.
Specify the coverage or loss scope and the exposure unit whose expected cost is being estimated.

Check the inputs.
Confirm that losses are ultimate and that loss scope, exposure definitions, and valuation basis are comparable across the candidate data.

Calculate period and pooled costs.
Divide each period's ultimate losses by its exposure count; also divide aggregate losses by aggregate exposures when pooling is appropriate.

Evaluate each candidate in context.
Distinguish a pooled estimate from an unweighted average and reject a value whose scope or exposure unit does not fit the target.

The pooled cost is a useful benchmark when the periods describe comparable units, but it is not automatically the correct selection in every situation. For instance, if one period represents a materially different coverage or population, pooling its losses with the others may produce a precise-looking number that answers the wrong question. Conversely, an unweighted mean may be appropriate only when equal influence by period is intentional; it should not be mislabeled as exposure-weighted.

This distinction is aligned with the CAS Exam 5 (Basic Ratemaking and Estimated Claim Liabilities) syllabus: candidates should be able to calculate an expected loss cost, interpret its denominator, and assess whether a proposed value is supported by relevant experience. The essential check is always contextual: what losses are counted, and what does one exposure unit mean?

Key takeaway: For comparable historical periods, the pooled expected loss cost is total ultimate losses divided by total exposure units. It weights each period by its exposure volume. Select a candidate only after confirming that the loss scope and exposure unit match the target being rated.

Check Your Understanding

For each problem, use the stated loss scope and exposure unit when assessing the cost.

A hypothetical coverage has $196,000 of ultimate loss on 7,000 comparable exposure-years. What is the observed cost per exposure-year? (Answer: $28.00 per exposure-year)
Two comparable periods have losses of $200,000 on 5,000 units and $330,000 on 11,000 units. Calculate the pooled cost and the unweighted mean of the two period costs. (Answer: pooled $33.13 per unit; unweighted mean $35.00 per unit)
Three comparable periods have losses of $250,000 on 5,000 units, $330,000 on 6,000 units, and $315,000 on 7,000 units. What is the pooled cost per unit, to the nearest cent? (Answer: $49.72 per unit)
A target cost is per insured-item-year. One dataset has $100,000 of relevant ultimate losses on 2,000 item-years; another has $150,000 on 5,000 customer-years, where a customer may insure multiple items. What cost is supported directly for the target, and can the two datasets be pooled as stated? (Answer: $50.00 per item-year from the first dataset; no, the denominators are different exposure units)
A target is glass-only coverage per covered-property-year. Comparable glass-only experience is $210,000 on 7,000 property-years. A broader all-cause dataset is $400,000 on the same 7,000 property-years. Which cost is supported for the target, and why? (Answer: $30.00 per property-year; the broader dataset's $57.14 cost includes losses outside the target scope)

What You'll Learn