Why Exposure Allocation Matters

In Bühlmann-Straub credibility, total exposure determines the credibility assigned to a risk, while the allocation of that exposure across periods determines how the period observations combine into the risk’s experience mean. These are distinct effects. Two risks can have the same total exposure and therefore the same credibility factor, yet receive different estimates because high and low period costs are given different weights.

This distinction matters whenever period observations represent costs per exposure unit. A period with more exposure contributes more to the risk’s mean than a period with less exposure. Treating the periods equally can distort the experience summary before credibility is applied. Conversely, treating the weighted mean itself as the final estimate omits the credibility adjustment toward the collective mean.

Definition: In the Bühlmann-Straub model, let $X_{ij}$ be the observed cost per exposure unit for risk $i$ in period $j$, and let $w_{ij}>0$ be that period’s exposure. Write $W_i=\sum_j w_{ij}$ for total exposure. The exposure-weighted risk mean is $$ \bar X_i=\frac{\sum_j w_{ij}X_{ij}}{W_i}. $$ With collective mean $m$, process variance parameter $v$, and variance of hypothetical means $a$, define $K=v/a$. The Bühlmann-Straub credibility factor and estimate of the risk’s expected cost per exposure unit are $$ Z_i=\frac{W_i}{W_i+K}, \qquad \widehat{\mu}_i=Z_i\bar X_i+(1-Z_i)m. $$

The standard model assumes that, conditional on a risk’s hypothetical mean $\mu_i$, the period observations have mean $\mu_i$, variance $v/w_{ij}$, and are independent across periods. Across risks, the hypothetical means have common mean $m$ and variance $a$. Exposures must be measured consistently, and the observations must be on the same cost-per-exposure basis. The formulas do not directly apply if, for example, the period figures are aggregate losses rather than costs per exposure unit, or if the assumed conditional variance and independence structure is unsuitable.

Separate the Weighting From the Credibility Adjustment

There are two calculations, in order. First use exposure weights to combine a risk’s observations into $\bar X_i$. Then use total exposure to find $Z_i$ and blend that weighted mean with $m$. If two risks have equal $W_i$, they have equal $Z_i$ under the same model parameters; this does not imply that they have equal $\bar X_i$ or equal $\widehat{\mu}_i$.

Calculate total exposure.
For each risk, add its period exposures to obtain $W_i$.

Calculate the weighted risk mean.
Multiply each period cost by its exposure, add those products, and divide by total exposure.

Calculate credibility.
Use $Z_i=W_i/(W_i+K)$; exposure allocation within the periods does not change this factor if $W_i$ stays fixed.

Calculate the credibility estimate.
Blend the risk’s weighted mean with the collective mean, using weights $Z_i$ and $1-Z_i$.

Why the Period Costs Are Exposure-Weighted

The weighting follows from the model’s variance assumption. Conditional on $\mu_i$, a period with exposure $w_{ij}$ has variance $v/w_{ij}$: more exposure means a more precise observation. The weighted mean combines these observations in a way that preserves the risk mean and uses the total exposure to express the combined precision.

E[\bar X_i\mid\mu_i] =\frac{\sum_j w_{ij}E[X_{ij}\mid\mu_i]}{W_i} =\frac{\sum_j w_{ij}\mu_i}{W_i} =\mu_i.

Conditional independence gives the variance calculation:

\begin{aligned} \operatorname{Var}(\bar X_i\mid\mu_i) &=\frac{1}{W_i^2}\sum_j w_{ij}^2\operatorname{Var}(X_{ij}\mid\mu_i)\\ &=\frac{1}{W_i^2}\sum_j w_{ij}^2\frac{v}{w_{ij}}\\ &=\frac{v}{W_i^2}\sum_j w_{ij} =\frac{v}{W_i}. \end{aligned}

Thus, for the risk mean, the observations contribute according to their exposure, and the resulting conditional variance depends on total exposure. With $m$, $v$, and $a$ held fixed, this explains why the credibility factor depends on $W_i$, not on how that total is divided among periods. The allocation still affects the observed weighted mean whenever period costs differ.

Worked Example: Equal Total Exposure, Different Two-Period Allocations

A hypothetical insurer observes costs per exposure unit of $80 in Period 1 and $120 in Period 2 for each of two risks. Risk A has exposures 1 and 3; Risk B has exposures 3 and 1. For both risks, let the collective mean be $m=\$100$ and the structural constant be $K=4$. Calculate each weighted mean, credibility factor, and credibility estimate. Explain the mistake in calling the weighted mean the estimate.

Risk A weighted mean. Its total exposure is $W_A=1+3=4$. The exposure-weighted cost total is $1(\$80)+3(\$120)=\$80+\$360=\$440$, so $$ \bar X_A=\frac{\$440}{4}=\$110. $$

Risk B weighted mean. Its total exposure is $W_B=3+1=4$. Its weighted cost total is $3(\$80)+1(\$120)=\$240+\$120=\$360$, so $$ \bar X_B=\frac{\$360}{4}=\$90. $$

Credibility factors. Both risks have total exposure 4, so $$ Z_A=Z_B=\frac{4}{4+4}=\frac{4}{8}=0.5. $$

Credibility estimates. For Risk A, $$ \widehat{\mu}_A=0.5(\$110)+0.5(\$100)=\$55+\$50=\$105. $$ For Risk B, $$ \widehat{\mu}_B=0.5(\$90)+0.5(\$100)=\$45+\$50=\$95. $$ The estimates are $105 and $95, respectively. The weighted means, $110 and $90, are inputs to the credibility calculation, not the final estimates. The final estimates lie halfway between those means and the collective mean. The calculation avoids the mistake of reporting the experience mean as the credibility estimate.

Allocation Can Change the Indication Without Changing Credibility

When the same period costs are observed for multiple risks, changing which periods receive more exposure can move the weighted means in opposite directions. If total exposure is held constant, the credibility factor remains constant, so the change in the final indication comes from the change in the weighted mean. The next example makes this separation explicit across three periods.

Worked Example: Three Periods With the Same Total Exposure

A hypothetical portfolio has three period costs per exposure unit: $70, $100, and $160. Risk A has exposures 1, 2, and 3 in those periods; Risk B has exposures 3, 2, and 1. Both have collective mean $m=\$110$ and structural constant $K=3$. Find both weighted means and credibility estimates.

Total exposures. For each risk, $W=1+2+3=6$. Therefore each credibility factor is $$ Z=\frac{6}{6+3}=\frac{6}{9}=\frac{2}{3}. $$

Risk A weighted mean. Its weighted cost total is $1(\$70)+2(\$100)+3(\$160)=\$70+\$200+\$480=\$750$. Thus, $$ \bar X_A=\frac{\$750}{6}=\$125. $$ Its credibility estimate is $$ \widehat{\mu}_A=\frac{2}{3}(\$125)+\frac{1}{3}(\$110) =\$83.333\ldots+\$36.666\ldots=\$120. $$

Risk B weighted mean. Its weighted cost total is $3(\$70)+2(\$100)+1(\$160)=\$210+\$200+\$160=\$570$. Thus, $$ \bar X_B=\frac{\$570}{6}=\$95. $$ Its credibility estimate is $$ \widehat{\mu}_B=\frac{2}{3}(\$95)+\frac{1}{3}(\$110) =\$63.333\ldots+\$36.666\ldots=\$100. $$

The risks have identical total exposure and identical credibility, but Risk A places more exposure in the high-cost third period, while Risk B places more in the low-cost first period. Their different weighted means lead to different estimates. The estimates are checked by noting that each is two-thirds of the way from the collective mean toward its own weighted mean.

Why an Unweighted Average Can Mislead

An unweighted average gives each period the same influence regardless of the amount of exposure represented. In the Bühlmann-Straub model, that is generally inconsistent with the stated variance assumption: a period with larger exposure has lower conditional variance and should carry more weight. An unweighted average is appropriate only when the period exposures are equal, or when a different model specifically justifies equal period weights.

Worked Example: Diagnosing an Unweighted-Average Error

A hypothetical risk has costs per exposure unit of $60 and $140 in two periods, with exposures 4 and 1. Let $m=\$100$ and $K=5$. Calculate the proper credibility estimate, then compare it with the result from incorrectly averaging the two period costs equally.

Proper exposure-weighted mean. Total exposure is $W=4+1=5$. The weighted cost total is $4(\$60)+1(\$140)=\$240+\$140=\$380$. Therefore, $$ \bar X=\frac{\$380}{5}=\$76. $$

Credibility factor and estimate. $$ Z=\frac{5}{5+5}=\frac{1}{2}, \qquad \widehat{\mu}=\frac{1}{2}(\$76)+\frac{1}{2}(\$100)=\$38+\$50=\$88. $$

Incorrect equal-period weighting. The unweighted mean would be $(\$60+\$140)/2=\$100$, leading to an incorrect estimate of $0.5(\$100)+0.5(\$100)=\$100$. This calculation gives the one-exposure period the same influence as the four-exposure period. Under the model, the correct estimate is $88. The check is that the weighted mean of $76 is below the collective mean, so a credibility estimate with $0<Z<1$ must lie between $76 and $100; $88 does.

Interpretation and Exam Traps

Exposure allocation is not a second credibility factor. It determines the weighted mean; total exposure determines $Z_i$. Keeping those roles separate makes comparisons easier to audit and prevents a common arithmetic error: stopping after calculating $\bar X_i$ and labeling it the credibility estimate.

Do not substitute an unweighted average. First check whether the period exposures are equal. If not, use the exposure-weighted mean required by the model.
Do not confuse total exposure with the weighted cost total. The denominator of $\bar X_i$ is $W_i$; the numerator is the sum of exposure times cost per exposure unit.
Do not stop at the weighted mean. Apply $Z_i$ and blend with the collective mean to obtain the credibility estimate.
Do not infer equal estimates from equal total exposures. Equal totals imply equal credibility factors under common $K$, not equal weighted means.
Check the direction. For $0\le Z_i\le1$, the estimate must be between the weighted mean and $m$. If it is outside that interval, recheck the arithmetic or the formula.

Key takeaway: In Bühlmann-Straub credibility, the allocation of exposure across periods controls how much each period cost contributes to $\bar X_i$, while the sum of those exposures controls $Z_i$. The final estimate uses both pieces: $\widehat{\mu}_i=Z_i\bar X_i+(1-Z_i)m$.

Check Your Understanding

For each problem, calculate the exposure-weighted mean before applying credibility. Report dollar amounts to the nearest cent.

A hypothetical risk has costs per exposure unit of $90 and $150, with exposures 2 and 2. Given $m=\$120$ and $K=4$, find its weighted mean, credibility factor, and estimate. (Answer: weighted mean $120; $Z=0.5000$; estimate $120.00)
A hypothetical risk has period costs of $75 and $135, with exposures 1 and 3. Given $m=\$100$ and $K=2$, calculate its weighted mean and credibility estimate. (Answer: weighted mean $120; $Z=0.6667$; estimate $113.33)
A hypothetical risk has costs of $50, $110, and $170, with exposures 1, 2, and 1. Given $m=\$90$ and $K=4$, calculate its weighted mean and estimate. (Answer: weighted mean $110; $Z=0.5000$; estimate $100.00)
Two hypothetical risks have period costs of $80, $100, and $140. Risk A has exposures 4, 1, and 1; Risk B has exposures 1, 1, and 4. Given $m=\$110$ and $K=6$, calculate each weighted mean and estimate, and state whether their credibility factors differ. (Answer: A weighted mean $93.33 and estimate $101.67; B weighted mean $123.33 and estimate $116.67; both have $Z=0.5000$)
A hypothetical risk has costs of $40, $100, and $160, with exposures 2, 1, and 3. Given $m=\$100$ and $K=3$, calculate its weighted mean, credibility factor, and estimate. (Answer: weighted mean $110; $Z=0.6667$; estimate $106.67)

Why Exposure Allocation Matters

What You'll Learn