Blending Experience and Exposure Indications

Why Blend Two Indications?

Experience rating uses the subject account’s adjusted loss history to estimate its future cost. Exposure rating uses exposure characteristics and an appropriate loss model to estimate that cost. Each method can be informative, but each can also be uncertain: the account’s history may be short or volatile, while an exposure model may not fully reflect the particular account. When both indications address the same expected cost, a weighted blend makes the contribution of each method explicit.

A blend is not automatically more accurate just because it uses two methods. The indications must first be comparable, and the weight must have a defensible interpretation. A weight selected by judgment is a judgmental credibility weight; it is not, by itself, a statistical estimate. This treatment of combining indications is aligned with the CAS Exam 9 (Risk Management for Actuaries) syllabus.

Definition: Let $P_E$ be the experience-rated indication and $P_X$ the exposure-rated indication, both expressed on the same basis. If $Z$ is the weight assigned to experience, where $0 \leq Z \leq 1$, the blended indication is $P_B=ZP_E+(1-Z)P_X$. The exposure weight is the complement $1-Z$. The resulting indication is a weighted average, not a third independent estimate.

Make the Indications Comparable First

The arithmetic is simple; establishing a fair comparison is the actuarial work that comes first. Before blending, confirm that the two indications refer to the same reinsured coverage, attachment point, limit, loss definition, and time period. They should also use a consistent currency and price basis. For example, an experience indication adjusted to current terms should not be combined directly with an exposure indication for a different layer.

Where appropriate, each indication should reflect the same treatment of loss adjustment expenses and other included costs. The analyst should state whether the figures are expected loss costs or premiums that already include other pricing components. This tutorial’s examples use indicated expected loss costs, with no expense, brokerage, profit, or uncertainty load added. A blended expected loss cost is not necessarily the final amount to quote.

A weight is attached to an indication, not to a method in the abstract. If experience rating produces a cost of $1.20 million and exposure rating produces $1.50 million, a 60% experience weight gives greater influence to the $1.20 million figure. It does not mean that 60% of the losses will arise from experience or that the experience estimate has a 60% probability of being correct.

Worked Example: Blending Two Comparable Layer Indications

A hypothetical insurer is evaluating one year of a specified reinsurance layer. After adjustment to the same terms and period, its experience-rated expected loss cost is $1,260,000 and its exposure-rated expected loss cost is $1,500,000. The pricing team assigns experience a weight of $Z=0.60$. Assume both figures include the same loss components and that no additional pricing loads are being applied. Calculate the blended indication and check its position relative to the two inputs.

Method. Apply the stated weight to experience and the complementary weight to exposure. The exposure weight is $1-Z=1-0.60=0.40$.

Experience contribution. $0.60 \times \$1,260,000=\$756,000$.

Exposure contribution. $0.40 \times \$1,500,000=\$600,000$.

Blend. Add the two weighted contributions: $\$756,000+\$600,000=\$1,356,000$. The blended expected loss cost is therefore $1,356,000.

Check. The result lies between $1,260,000 and $1,500,000, as a weighted average with nonnegative weights summing to one must. The common mistake this method avoids is applying the 60% experience weight to exposure instead: that reversal would produce $0.40(\$1,260,000)+0.60(\$1,500,000)=\$1,404,000$, a different indication corresponding to the opposite weighting.

Why a Weighted Average Has This Form

The blend can be justified as an estimate of one underlying expected cost. Suppose the true expected cost is $\theta$, and the two indications are random estimates $P_E$ and $P_X$. Assume each is unbiased, so $E[P_E]=E[P_X]=\theta$. For any fixed $Z$, linearity of expectation gives:

E[P_B] =E[ZP_E+(1-Z)P_X] =ZE[P_E]+(1-Z)E[P_X] =Z\theta+(1-Z)\theta =\theta.

Thus the blend remains unbiased under those assumptions. The weight controls precision and influence, not the expected value of an unbiased estimate. In practice, an indication may be biased, or the methods may target slightly different quantities; in those cases this conclusion does not follow, and averaging does not cure the mismatch.

A common statistical justification for a particular weight comes from minimizing the blend’s variance. Assume the indications are unbiased, have variances $\sigma_E^2$ and $\sigma_X^2$, and are independent. Independence is important: it means their covariance is zero. Then:

\operatorname{Var}(P_B) =Z^2\sigma_E^2+(1-Z)^2\sigma_X^2.

Differentiate with respect to $Z$, set the derivative to zero, and solve. The second derivative is positive when the variances are positive, so the solution minimizes the variance:

\frac{d}{dZ}\operatorname{Var}(P_B) =2Z\sigma_E^2-2(1-Z)\sigma_X^2=0,

Z\sigma_E^2=(1-Z)\sigma_X^2 \quad\Longrightarrow\quad Z=\frac{\sigma_X^2}{\sigma_E^2+\sigma_X^2}.

The estimate with the lower variance receives the larger weight. For example, if experience has smaller variance than exposure, then $\sigma_E^2<\sigma_X^2$, which makes $Z>0.5$. This formula is not a universal credibility rule: it relies on unbiased estimates, known variances, and independence. If the indications are dependent, the variance includes a covariance term, $2Z(1-Z)\operatorname{Cov}(P_E,P_X)$, and this independent-estimate formula does not generally apply.

Worked Example: Converting a Cost per Exposure Unit Before Blending

A hypothetical reinsurer is estimating the annual expected loss cost for a defined layer. Experience rating, already adjusted to the target period and layer terms, indicates $450,000. Exposure rating gives an expected cost of $20 per exposure unit, and the account has 26,000 exposure units on the same basis. Assign experience a weight of $Z=0.35$. Assume the indications cover the same losses and include no additional loads. Calculate the blended annual expected loss cost.

Method. First convert the exposure rate to an annual dollar indication. Then blend dollar amounts, using the stated experience weight and its complement.

Exposure indication. Multiply the rate by the number of exposure units: $26,000 \times \$20=\$520,000$. This is an annual expected loss cost, matching the experience figure’s unit.

Weights. The experience weight is 0.35, so the exposure weight is $1-0.35=0.65$.

Weighted contributions. Experience contributes $0.35 \times \$450,000=\$157,500$. Exposure contributes $0.65 \times \$520,000=\$338,000$.

Blend. The total is $\$157,500+\$338,000=\$495,500$. The blended annual expected loss cost is $495,500.

Check. The answer is between $450,000 and $520,000. The common mistake avoided is blending the $20 rate directly with a dollar amount: a cost per exposure unit and an annual total have different units and cannot be added or averaged. Convert both indications to the same unit first.

Testing Alternative Weights

When the methods disagree, the selected weight can materially affect the result. A sensitivity analysis holds the two indications fixed and recalculates the blend at several plausible weights. This does not prove which weight is correct; it shows how much the indicated cost depends on that judgment. The direction is predictable: increasing $Z$ moves the blend toward experience, while decreasing $Z$ moves it toward exposure.

$$ P_B=P_X+Z(P_E-P_X). $$

This rearrangement makes the sensitivity clear. For fixed indications, a change $\Delta Z$ changes the blend by $\Delta P_B=\Delta Z(P_E-P_X)$. The larger the disagreement between the two methods, the larger the effect of a given change in weight. If the indications agree, changing the weight has no effect.

Worked Example: Measuring Sensitivity to the Experience Weight

For a hypothetical one-year catastrophe layer, the experience indication is $720,000 and the exposure indication is $1,080,000, on the same coverage and cost basis. A reviewer asks for blends using experience weights of 0.25, 0.50, and 0.75. No expense or other load is included. Calculate all three results and quantify the change between the lowest and highest experience weights.

Method. For each case calculate $ZP_E+(1-Z)P_X$. Keep the same two indications and change only $Z$.

At $Z=0.25$. The exposure weight is 0.75. The contributions are $0.25(\$720,000)=\$180,000$ and $0.75(\$1,080,000)=\$810,000$, giving a blend of $990,000.

At $Z=0.50$. The exposure weight is also 0.50. The contributions are $0.50(\$720,000)=\$360,000$ and $0.50(\$1,080,000)=\$540,000$, giving a blend of $900,000.

At $Z=0.75$. The exposure weight is 0.25. The contributions are $0.75(\$720,000)=\$540,000$ and $0.25(\$1,080,000)=\$270,000$, giving a blend of $810,000.

Sensitivity check. Raising the experience weight from 0.25 to 0.75 changes $Z$ by 0.50. The indication difference is $\$720,000-\$1,080,000=-\$360,000$, so the change in blend is $0.50(-\$360,000)=-\$180,000$. This matches $\$810,000-\$990,000=-\$180,000$. The common mistake avoided is assuming that more weight on experience must increase the premium: here experience is the lower indication, so increasing its weight reduces the blend.

Interpreting and Communicating the Result

A useful pricing record should identify both source indications, their basis, the weight and its rationale, the calculation, and any sensitivity results. If the weight is judgmental, say so. If it is derived from an assumed variance model, state the assumptions and explain whether dependence or potential bias has been considered. Avoid presenting a chosen weight with more precision than the supporting evidence warrants.

The arithmetic provides several immediate checks:

The weights must be between zero and one and sum to one.
The blended result must lie between the two indications when both weights are nonnegative.
A larger weight on one indication must move the blend toward that indication.
The calculation must combine values on the same basis and in the same units.

If the blend falls outside the range of its two inputs, either the weights are not valid convex weights, the arithmetic is wrong, or the reported result includes an adjustment beyond the blend. State such an adjustment separately rather than obscuring it in the weighting step.

Key takeaway: Blending is a transparent way to combine comparable experience and exposure indications, but the weight is a substantive assumption. Show the weight, its rationale, the resulting calculation, and how the indication changes when the methods disagree.

Check Your Understanding

For each problem, assume the indications address the same period and coverage basis and that no additional loads apply unless stated.

An experience indication is $640,000 and an exposure indication is $800,000. Experience receives weight 0.65. What is the blended indication? (Answer: $696,000)
Experience indicates $1,100,000 and exposure indicates $900,000. If experience receives weight 0.40, calculate the blend and state which indication pulls it lower. (Answer: $980,000; the exposure indication pulls it lower.)
Experience indicates $600,000. Exposure rating indicates $32 per exposure unit for 15,000 units. If experience receives weight 0.25, calculate the annual blended expected loss cost. (Answer: $510,000)
A hypothetical account has experience and exposure indications of $2,400,000 and $1,800,000. Calculate the blends at experience weights 0.20, 0.50, and 0.80. (Answer: $1,920,000; $2,100,000; $2,280,000)
Assume unbiased, independent indications. Experience has variance $100^2$ (in squared thousands of dollars) and exposure has variance $200^2$. Using the minimum-variance weight, calculate the experience weight. If the indications are $900,000 and $1,100,000, respectively, calculate the corresponding blend. (Answer: experience weight 0.80; blended indication $940,000)

What You'll Learn