Large-Loss Sensitivity in Class Experience

Why Large Claims Matter in Class Experience

A class’s total losses can be strongly influenced by one unusually large claim, especially when the class has relatively few exposures. Since class loss costs are often used to indicate relativities, that claim can affect not only the class’s own loss cost but also its indicated position relative to other classes. Measuring this sensitivity helps an actuary see how much an indication depends on individual observations rather than on the broader experience pattern.

The analysis is a comparison of experience under clearly specified treatments. First calculate results using the stated losses. Then apply a defined large-loss adjustment consistently and recalculate. A cap is one possible analytical adjustment; it does not mean the insurance policy itself limits the claim. It is also not a substitute for considering whether losses and exposures have been placed on an appropriate, consistent basis.

Definition: For class $j$, let $E_j$ be its exposures and $L_j$ its total losses on the basis being analyzed. Its loss cost is $C_j=L_j/E_j$. With a per-claim cap $U$, a claim amount $X_{jk}$ contributes $\min(X_{jk},U)$, and the capped class loss cost is $C_j(U)=\sum_k \min(X_{jk},U)/E_j$. The cap’s effect on class $j$’s loss cost is $C_j-C_j(U)$.

A cap should be defined in terms of the claim amount it applies to: for example, each individual claim is limited to $U$ dollars for the sensitivity calculation. The claim amounts and cap must be expressed on the same dollar basis. If the analysis uses losses adjusted to a common cost level, the cap must be consistent with that basis as well. The exposure count is not reduced by capping losses.

From Loss Costs to Relative Indications

A simple comparison of two classes uses their loss-cost ratio. If classes $j$ and $k$ have positive loss costs, the indicated $j$-to-$k$ ratio is $C_j/C_k$. This comparison does not depend on other classes. If class indications are normalized to an overall average, however, the overall loss cost also changes when any class’s losses change.

Under exposure weighting, the overall loss cost is $\bar C=\sum_j L_j/\sum_j E_j$, and the normalized factor for class $j$ is $R_j=C_j/\bar C$. The same definitions apply after capping, using capped losses throughout. The normalized factors then have an exposure-weighted average of one, subject to rounding.

R_j=\frac{L_j/E_j}{\left(\sum_h L_h\right)/\left(\sum_h E_h\right)} \qquad\text{and}\qquad R_j(U)=\frac{L_j(U)/E_j}{\left(\sum_h L_h(U)\right)/\left(\sum_h E_h\right)}.

The denominator matters. If a large claim is capped in one class, the overall average usually decreases. Consequently, the normalized factors of classes whose losses were not changed can increase, even though their own loss costs stayed the same. That movement is a consequence of normalization, not evidence that those classes experienced a new loss.

Why One Claim Has a Measurable Effect

Suppose a class has $E_j$ exposures and total losses $L_j$, including one claim of amount $x$. If that claim alone is reduced to $U$, where $x>U$, the class loss total falls by $x-U$. Dividing this reduction by the unchanged exposures gives the change in loss cost.

C_j-C_j(U)=\frac{x-U}{E_j} \quad\text{when the only adjustment is capping that claim and }x>U.

This follows directly from the definition: the uncapped cost is $L_j/E_j$, while the adjusted cost is $(L_j-x+U)/E_j$. Subtracting gives $[L_j-(L_j-x+U)]/E_j=(x-U)/E_j$. If $x\le U$, the claim is not reduced and its direct effect is zero. If several claims exceed the cap, the total loss-cost reduction is the sum of their excesses divided by exposures.

For a two-class ratio, if only class $j$ is adjusted and class $k$ is unchanged, the ratio changes from $C_j/C_k$ to $C_j(U)/C_k$. In contrast, a normalized factor compares each class with the total experience, so its denominator may change too. Keeping these two comparisons distinct prevents misinterpretation.

Worked Example: Comparing Two Classes Before and After a Cap

In a hypothetical portfolio, classes P and Q each have 300 exposures. Class P has one claim of $90,000 and nine claims of $10,000 each. Class Q has twenty claims of $10,000 each. Calculate the P-to-Q loss-cost ratio with no cap and with a $40,000 per-claim cap. Name the method and identify the mistake the calculation avoids.

Method. Sum each class’s claim amounts, divide by its exposures, and then form the ratio. For the capped comparison, replace every claim by the smaller of its amount and $40,000 in both classes.

Uncapped P. Its total losses are $90{,}000+9(10{,}000)=180{,}000$. Its loss cost is $180{,}000/300=\$600$ per exposure.

Uncapped Q. Its total losses are $20(10{,}000)=200{,}000$. Its loss cost is $200{,}000/300=\$666.6667$ per exposure. The uncapped P-to-Q ratio is $600/666.6667=0.9000$, equivalently $180{,}000/200{,}000=0.9000$.

Capped P. The $90,000 claim becomes $40,000; each $10,000 claim remains unchanged. Capped losses are $40{,}000+9(10{,}000)=130{,}000$, so the capped loss cost is $130{,}000/300=\$433.3333$.

Capped Q. All twenty claims are below the cap, so losses remain $200,000 and the loss cost remains $200{,}000/300=\$666.6667$. The capped ratio is $433.3333/666.6667=0.6500$, or $130{,}000/200{,}000=0.6500$.

Check and common mistake. The adjustment reduces P’s losses by $90{,}000-40{,}000=\$50{,}000$, giving a cost reduction of $50{,}000/300=\$166.6667$. Subtracting this from $600 gives $433.3333, as calculated. A common error is to report a P-to-Q ratio above one simply because P has the single largest claim. The actual total losses show that P’s uncapped ratio is 0.9000, and the capped ratio is 0.6500.

Normalizing Factors After a Consistent Adjustment

The following example shows why all classes must be recalculated under a consistent treatment when normalized factors are compared. The normalization uses exposure weights: each class’s cost is divided by the overall loss cost for the same version of the experience.

Worked Example: Recalculating Exposure-Weighted Factors

Consider three hypothetical classes. Class A has 1,000 exposures and $200,000 of losses. Class B has 800 exposures and $300,000 of losses, including one $100,000 claim. Class C has 400 exposures and $200,000 of losses. Apply a $40,000 per-claim cap, assuming the other claims in B are below the cap. Calculate the uncapped and capped normalized factors. State the common mistake avoided.

Uncapped loss costs. Class A has $200{,}000/1{,}000=\$200$. Class B has $300{,}000/800=\$375$. Class C has $200{,}000/400=\$500$.

The exposure total is $1{,}000+800+400=2{,}200$, and the loss total is $200{,}000+300{,}000+200{,}000=\$700{,}000$. Thus the overall loss cost is $700{,}000/2{,}200=\$318.1818$. Dividing each class cost by this overall cost gives factors A $=200/318.1818=0.6286$, B $=375/318.1818=1.1786$, and C $=500/318.1818=1.5714$, rounded to four decimals.

Capped losses and costs. The $100,000 claim in B is reduced to $40,000, a reduction of $60,000. Capped losses are therefore $200,000 for A, $300{,}000-60{,}000=\$240{,}000$ for B, and $200,000 for C. The corresponding costs are A $200, B $240{,}000/800=\$300$, and C $500.

Total capped losses are $200{,}000+240{,}000+200{,}000=\$640{,}000$. The capped overall loss cost is $640{,}000/2{,}200=\$290.9091$. The normalized capped factors are A $=200/290.9091=0.6875$, B $=300/290.9091=1.03125$, and C $=500/290.9091=1.71875$. To four decimals these are 0.6875, 1.0313, and 1.7188.

Check and common mistake. Using unrounded factors, the exposure-weighted average is $(1{,}000(0.6875)+800(1.03125)+400(1.71875))/2{,}200=(687.5+825+687.5)/2{,}200=1.0000$. A common mistake is to cap B’s losses but retain the uncapped overall average in the denominator. That would mix two different experience bases and would not produce factors with an exposure-weighted average of one. The uncapped factors also pass the check: $(1{,}000(0.6285714)+800(1.1785714)+400(1.5714286))/2{,}200\approx1.0000$.

Apply the Treatment Consistently

Consistency means more than using the same numeric threshold in a spreadsheet. Before comparing results, specify the claim unit to which the threshold applies, the dollar basis for the amounts, and the classes included. Then apply the stated rule to each class in the comparison. If the analysis is per claim, do not cap a class’s aggregate losses once as though the aggregate were a single claim.

If a class has claims $X_{j1},\ldots,X_{jn_j}$, its capped losses are calculated claim by claim as $\sum_k\min(X_{jk},U)$. The reduction is $\sum_k\max(X_{jk}-U,0)$. This adjustment leaves both the number of exposures and the claim count as observed; it changes the amount assigned to claims that exceed the threshold.

Worked Example: Capping Claims in Both Classes

Two hypothetical classes each have 500 exposures. Class M has claims of $50,000, $30,000, and $20,000. Class N has claims of $40,000, $30,000, and $30,000. Compare the M-to-N loss-cost ratio before and after a $35,000 per-claim cap. Explain why capping only one class would be inappropriate for this comparison.

Uncapped costs. M has total losses $50{,}000+30{,}000+20{,}000=\$100{,}000$, so its loss cost is $100{,}000/500=\$200$. N also has total losses $40{,}000+30{,}000+30{,}000=\$100{,}000$, so its loss cost is $200. The uncapped ratio is $200/200=1.0000$.

Capped costs. In M, $50,000 becomes $35,000; the other claims remain unchanged. Capped losses are $35{,}000+30{,}000+20{,}000=\$85{,}000$, and the cost is $85{,}000/500=\$170$. In N, $40,000 becomes $35,000, while both $30,000 claims remain unchanged. Capped losses are $35{,}000+30{,}000+30{,}000=\$95{,}000$, and the cost is $95{,}000/500=\$190$.

The capped M-to-N ratio is $170/190=0.8947368$, or 0.8947 to four decimals. The check is that the loss totals fell by $15,000 in M and $5,000 in N, exactly the amounts above the cap. Capping only M would compare adjusted M experience with unadjusted N experience, so the resulting ratio would not represent a consistent treatment of the two classes.

Interpretation and Exam Traps

A large-loss sensitivity result answers a limited question: how much does the indicated class experience change under a specified treatment? It does not, by itself, establish that the adjusted indication is more appropriate. A cap can reduce the influence of individual large claims, but the chosen threshold and the reason for using it must be explained. Results should be shown both before and after adjustment so the sensitivity remains visible.

Confusing a claim cap with an aggregate cap. A per-claim cap is applied separately to each claim. The calculation $\min(\sum_k X_{jk},U)$ is an aggregate cap and produces a different result.
Changing the denominator inconsistently. A class loss cost uses that class’s exposures. A normalized factor uses an overall loss cost calculated on the same adjusted or unadjusted basis.
Applying a cap to one class only. Unless the comparison explicitly calls for that asymmetric treatment, apply the same specified rule to every class being compared.
Treating a sensitivity adjustment as an actual coverage feature. An analytical cap changes the experience used in the calculation; it does not alter the hypothetical policy’s contractual coverage.
Rounding too early. Keep full precision through the loss-cost ratios and normalization, then round the displayed results. Rounded factors may not average to exactly one.

For a strong analysis, report the cap definition, exposures, uncapped and adjusted losses, class loss costs, and the resulting ratios or normalized factors. State the size of the loss-cost reduction attributable to the claims above the cap. This makes the effect auditable and prevents a changed overall denominator from being mistaken for a change in every class’s underlying losses.

Key takeaway: Large-loss sensitivity is measured by recalculating class experience under a precisely stated treatment and comparing it with the unadjusted result. Apply the treatment consistently, preserve the appropriate exposure denominators, and normalize each version to its own overall loss cost when normalized factors are required. This approach is aligned with the CAS Exam 5 (Basic Ratemaking and Estimated Claim Liabilities) syllabus.

Check Your Understanding

For each hypothetical problem, calculate the requested result using the treatment stated. Answers are rounded to four decimals where applicable.

Class R has 200 exposures and claims of $50,000, $20,000, and $10,000. Class S also has 200 exposures and claims of $30,000, $30,000, and $20,000. Find the uncapped R-to-S loss-cost ratio and the ratio with a $25,000 per-claim cap applied to both classes. (Answer: uncapped ratio 1.0000; capped ratio 0.7857.)
Class A has 400 exposures and $160,000 in losses, including one $80,000 claim. Class B has 400 exposures and six claims of $20,000 each. Find the uncapped A-to-B ratio and the ratio after a $30,000 per-claim cap is applied to both classes. (Answer: uncapped ratio 1.3333; capped ratio 0.9167.)
Class A has 100 exposures and $20,000 in losses. Class B has 200 exposures and $60,000 in losses, including one $50,000 claim. Calculate both classes’ exposure-weighted normalized factors before and after applying a $20,000 cap to B’s claim. (Answer: uncapped factors A 0.7500 and B 1.1250; capped factors A 1.2000 and B 0.9000.)
Classes K and L each have 250 exposures. K has claims of $60,000, $40,000, and $20,000; L has claims of $50,000, $25,000, and $25,000. Calculate the K-to-L ratio before and after applying a $35,000 cap to every claim. (Answer: uncapped ratio 1.2000; capped ratio 1.0588.)
Class U has 100 exposures and $100,000 in losses, including one $70,000 claim. Class V has 100 exposures and $80,000 in losses, including one $50,000 claim; assume each class’s remaining losses are below $40,000. Calculate the U-to-V ratio before and after applying a $40,000 cap to both classes. (Answer: uncapped ratio 1.2500; capped ratio 1.0000.)

What You'll Learn