Adjusting for Changes in Exposure Mix

Why Exposure Mix Matters

An insurer’s aggregate loss cost can change even when the experience within every class stays exactly the same. The reason may be that the portfolio contains a different proportion of exposures from higher- and lower-cost classes. Conversely, an aggregate result that appears stable can hide offsetting changes: one class may have worsening experience while the portfolio shifts toward a lower-cost class.

For ratemaking, the key is to distinguish changes in exposure mix from changes in class-level loss costs. This tutorial focuses on that distinction and on reweighting historical experience to a prospective mix. Assume throughout that losses have already been put on a consistent basis for the comparison—for example, the relevant development, trend, and coverage adjustments have been made. The loss and exposure bases must also match, as established in the earlier material on matching loss and premium bases.

Definition: For class $j$, let $e_j$ be the number of exposures, $L_j$ the corresponding losses on the selected common basis, and $q_j=L_j/e_j$ the class loss cost per exposure. The exposure weight is $w_j=e_j/\sum_k e_k$. The aggregate loss cost is the exposure-weighted average $\sum_j w_jq_j$.

An exposure weight is a share of exposures, not a share of losses or premium. The weights are nonnegative and sum to 1. A class with a higher loss cost contributes more to the aggregate when its exposure share is larger.

Standardizing Experience to a Target Mix

Suppose historical class loss costs are $q_j$, but the prospective portfolio is expected to have exposure weights $w_j^*$. The historical loss costs standardized to that prospective mix are $\sum_j w_j^*q_j$. This answers a focused question: what aggregate loss cost would these class-level results produce if the exposure distribution were the target distribution?

q_{\mathrm{standardized}}=\sum_j w_j^*q_j, \qquad \sum_j w_j^*=1

The standardized value is not the observed historical average unless the historical and target weights happen to match. It is a reweighted calculation, not a claim that the historical portfolio actually had the prospective mix.

Worked Example: Reweighting to a New Territory Mix

A hypothetical insurer has two territory classes. Historical losses on a common basis are $1,200,000 for 2,000 Coastal exposures and $900,000 for 3,000 Inland exposures. The prospective portfolio is expected to be 60% Coastal and 40% Inland. Find the historical aggregate loss cost and the result standardized to the prospective mix. Assume the class loss costs are suitable for this comparison.

Method. Calculate each class loss cost, then apply the prospective exposure weights. This avoids carrying forward the historical mix merely because it was observed.

Coastal loss cost:

q_C=\frac{\$1{,}200{,}000}{2{,}000}=\$600

Inland loss cost:

q_I=\frac{\$900{,}000}{3{,}000}=\$300

Historical exposure weights are $2{,}000/5{,}000=0.4$ Coastal and $3{,}000/5{,}000=0.6$ Inland. Thus the historical aggregate loss cost is:

0.4(\$600)+0.6(\$300)=\$240+\$180=\$420

The target weights are 0.6 Coastal and 0.4 Inland. Standardizing to that mix gives:

0.6(\$600)+0.4(\$300)=\$360+\$120=\$480

Check. The standardized result is $60 higher per exposure than the historical result. That is consistent with the target mix placing 20 percentage points more weight on the higher-cost Coastal class. The common mistake avoided here is using the historical weights to answer a prospective-mix question.

Why the Weighted-Average Formula Holds

The formula follows directly from the definitions. If class $j$ has loss cost $q_j$ per exposure and $e_j$ exposures, its losses are $e_jq_j$. Total losses are therefore the sum of those class losses. Dividing by total exposures produces the aggregate loss cost:

\frac{\sum_j e_jq_j}{\sum_j e_j} = \sum_j \left(\frac{e_j}{\sum_k e_k}\right)q_j = \sum_j w_jq_j

This also explains the reweighting procedure. To calculate the aggregate under a different mix, replace the historical exposure shares with the target shares while keeping the class loss costs being evaluated. The procedure is appropriate only when the class definitions are consistent and the class loss costs are comparable on the same basis. If a class’s losses have a different development, trend, or coverage basis, reweighting alone does not make them comparable.

Key distinction: Changing $w_j$ while holding $q_j$ fixed changes the aggregate through portfolio composition. Changing $q_j$ changes the class-level experience. A sound analysis identifies which of these inputs changed rather than interpreting every aggregate movement as a change in underlying risk.

Separating Mix Effects from Class-Level Changes

When both the class loss costs and the weights change, a direct comparison of the two aggregate averages combines multiple effects. Let $w_j^{(0)}$ and $q_j^{(0)}$ denote the initial weights and loss costs, and $w_j^{(1)}$ and $q_j^{(1)}$ the later values. Define $\Delta w_j=w_j^{(1)}-w_j^{(0)}$ and $\Delta q_j=q_j^{(1)}-q_j^{(0)}$. The exact change in aggregate loss cost is:

\sum_j w_j^{(1)}q_j^{(1)}-\sum_j w_j^{(0)}q_j^{(0)} = \sum_j w_j^{(0)}\Delta q_j + \sum_j \Delta w_jq_j^{(0)} + \sum_j \Delta w_j\Delta q_j

The first term is the change in class loss costs evaluated at the initial mix. The second is the mix change evaluated at the initial class loss costs. The third is an interaction: the mix and class loss costs have both changed. Keeping that interaction visible makes the bridge exact rather than assigning it silently to one of the other effects.

Worked Example: Bridging an Aggregate Change

A hypothetical two-class portfolio has the following results. Initially, Class A is 60% of exposures with a loss cost of $500, and Class B is 40% with a loss cost of $200. Later, Class A is 30% of exposures with a loss cost of $550, and Class B is 70% with a loss cost of $220. Decompose the change in aggregate loss cost into class-cost, mix, and interaction components.

Method. Calculate both aggregate averages, then apply the exact decomposition using the initial weights and loss costs.

Initial aggregate:

0.6(\$500)+0.4(\$200)=\$300+\$80=\$380

Later aggregate:

0.3(\$550)+0.7(\$220)=\$165+\$154=\$319

The observed change is $\$319-\$380=-\$61$. Class-cost changes are $\Delta q_A=\$50$ and $\Delta q_B=\$20$. At the initial weights, the class-cost component is:

0.6(\$50)+0.4(\$20)=\$30+\$8=+\$38

Weight changes are $\Delta w_A=-0.3$ and $\Delta w_B=0.3$. At the initial loss costs, the mix component is:

(-0.3)(\$500)+(0.3)(\$200)=-\$150+\$60=-\$90

The interaction component is:

(-0.3)(\$50)+(0.3)(\$20)=-\$15+\$6=-\$9

Check. The components sum to $+\$38-\$90-\$9=-\$61$, the observed change. The lower aggregate is driven mainly by the shift toward Class B, whose loss cost is lower; the increases in both class loss costs partly offset that mix effect. The common mistake avoided is to call the full $61 decrease an improvement in class-level experience.

Using the Prospective Mix to Assess Loss-Only Rate Adequacy

A loss-only adequacy comparison asks whether current rates, at the relevant exposure distribution, are sufficient to cover expected losses. Let $r_j$ be the current rate per exposure in class $j$, and let $q_j^*$ be the selected expected loss cost per exposure for the prospective period. Using prospective weights $w_j^*$, the aggregate expected loss cost is $\sum_jw_j^*q_j^*$, while average current premium per exposure is $\sum_jw_j^*r_j$. Their ratio is a loss-only indication factor:

\text{Loss-only factor} = \frac{\sum_j w_j^*q_j^*}{\sum_j w_j^*r_j}

This ratio is a loss-only comparison, not a complete rate indication. It does not include expenses, profit provisions, or other premium components. Both the expected losses and current premium are weighted using the same prospective mix so that the numerator and denominator describe the same portfolio composition.

Worked Example: Separating Mix Change from Loss-Only Adequacy

A hypothetical insurer has High-Risk and Low-Risk classes. Selected prospective loss costs are $900 and $300 per exposure, respectively. Current rates are $850 and $320 per exposure. The old portfolio was 50% in each class, while the prospective portfolio is 25% High-Risk and 75% Low-Risk. Calculate the loss-only factor at the prospective mix, and show why the old mix gives a different conclusion.

Method. Weight both selected losses and current rates at each mix before taking the ratio.

At the prospective mix, expected loss cost is:

0.25(\$900)+0.75(\$300)=\$225+\$225=\$450

Average current rate at that same mix is:

0.25(\$850)+0.75(\$320)=\$212.50+\$240=\$452.50

The prospective-mix loss-only factor is:

\frac{\$450}{\$452.50}=0.994475\ldots\approx 0.9945

To check the effect of retaining the old mix, its weighted expected loss cost would be $0.5(\$900)+0.5(\$300)=\$600$. Its weighted current rate would be $0.5(\$850)+0.5(\$320)=\$585$, giving $\$600/\$585=1.025641\ldots$, or approximately 1.0256. That historical-mix comparison suggests a loss shortfall, while the prospective-mix comparison is slightly below 1.0000. The difference arises because the prospective portfolio has a much smaller share of the high-cost class.

Check. At the prospective mix, current rates exceed selected expected loss cost by $2.50 per exposure, so a loss-only factor below 1 is consistent. The common mistake avoided is mixing the expected losses at one exposure distribution with current premium at another.

Common Traps and Practical Checks

Using loss shares as exposure weights. Loss shares are influenced by the class loss costs themselves. The weights for standardizing an average are exposure shares.
Applying the wrong mix. Historical weights answer what the historical portfolio experienced. Prospective weights are needed to estimate the aggregate at the expected future composition.
Changing the mix but not the denominator. In a loss-only adequacy ratio, weight both class expected losses and class current rates using the same mix.
Calling a mix effect a class trend. A changing aggregate average does not by itself establish that the underlying class loss costs changed.
Reweighting incomparable class results. Standardization does not correct inconsistent loss development, trend, exposure definitions, or class boundaries. Resolve basis differences before attributing a result to mix.

Practical sequence: Confirm a common loss and exposure basis; calculate class loss costs; calculate exposure weights; select the target mix; reweight the class results; and, for a loss-only adequacy comparison, weight current rates with the same target mix. Interpret any remaining class-level changes separately from the composition effect.

Check Your Understanding

For each problem, use exposure weights rather than loss shares, and keep the loss and rate bases consistent.

A portfolio has 700 Low-Risk exposures with a loss cost of $200 and 300 High-Risk exposures with a loss cost of $800. A target portfolio is 80% Low-Risk and 20% High-Risk. Find the observed aggregate loss cost and the loss cost standardized to the target mix. (Answer: observed $380; standardized $320 per exposure.)
A historical portfolio is 40% Coastal, with a loss cost of $1,000, and 60% Inland, with a loss cost of $400. The prospective mix is 20% Coastal and 80% Inland. Find the historical aggregate loss cost and the prospective-mix standardized loss cost. (Answer: historical $640; standardized $520 per exposure.)
Prospective loss costs for Classes A and B are $720 and $240. Their prospective exposure weights are 35% and 65%, and current rates are $700 and $260. Calculate the prospective expected loss cost, average current rate, and loss-only factor. (Answer: expected loss cost $408; average current rate $414; factor 0.9855.)
Initially, Class A has weight 70% and loss cost $400; Class B has weight 30% and loss cost $200. Later, the weights are 40% and 60%, and the loss costs are $450 and $250. Calculate the initial and later aggregates and decompose the change into class-cost, mix, and interaction components using the initial values as the reference. (Answer: initial $340; later $330; class-cost effect +$50; mix effect −$60; interaction $0; total change −$10.)

What You'll Learn