Why Unequal Period Exposures Change the Calculation
A risk’s experience over several periods may not carry equal information in every period. A period with more exposure generally produces a more reliable estimate of the risk’s underlying expected loss cost than a period with less exposure. Bühlmann-Straub credibility accounts for this by weighting each period’s experience according to its exposure, then blending the risk’s weighted mean with a collective mean.
The calculation has several distinct steps. First calculate the risk’s weighted mean, using each period’s exposure. Next add the exposures to obtain total weight. Then calculate the credibility factor from total weight and the structural constant. Finally combine the weighted mean and the collective mean. Keeping these steps separate helps prevent an attractive-looking but incorrect unweighted average or credibility factor.
Model Assumptions and Calculation Sequence
The standard Bühlmann-Straub model describes period observations conditional on a risk’s hypothetical mean \(\theta\). The period loss costs have conditional means \(E[X_j\mid\theta]=\theta\), and conditional variances \(\operatorname{Var}(X_j\mid\theta)=\sigma^2/w_j\), where \(\sigma^2\) is the process variance per unit of exposure weight. Period observations are conditionally independent given \(\theta\). Across risks, \(E[\theta]=\mu\) and \(\operatorname{Var}(\theta)=\tau^2\), so \(K=\sigma^2/\tau^2\).
These are moment assumptions; a particular probability distribution, such as a normal distribution, is not required for the credibility formula. There is no interest or mortality adjustment in this calculation: the observations are treated as being on a consistent, comparable basis, and weights must use the exposure unit for which the process variance and \(K\) are specified. If period means have shifted, observations are dependent, or variance is not inversely proportional to exposure, the standard formula may not be appropriate without adjustments.
The weighted risk mean is not the same as the collective mean. The first summarizes one risk’s experience, with more exposed periods receiving more weight; the collective mean represents the prior or portfolio-level expectation used in the credibility blend. In an applied problem, the collective mean and \(K\) may be supplied. Unless the problem provides the broader portfolio data and asks for their calculation, do not attempt to infer them from the individual risk’s periods alone.
Why the Credibility Factor Has This Form
The credibility factor balances two sources of uncertainty: process variation in the observed experience and variation in underlying means across risks. The conditional independence and variance assumptions imply that the weighted mean has conditional expectation \(\theta\) and conditional variance \(\sigma^2/W\). Thus, more total exposure reduces the process variance of the risk’s mean.
To justify the factor, consider a linear estimate \(Z\bar{X}_w+(1-Z)\mu\) of \(\theta\). Its deviation from \(\theta\) is \(Z(\bar{X}_w-\theta)+(1-Z)(\mu-\theta)\). The cross-product has expectation zero because the experience error has conditional mean zero given \(\theta\). The mean squared error is therefore
Minimize this expression with respect to \(Z\). Differentiating and setting the result equal to zero gives
This also explains the direction of the result. As total exposure \(W\) increases, the experience mean becomes less noisy and \(Z\) increases. As \(K\) increases, process variation is larger relative to between-risk variation, so the estimate places less weight on the individual risk’s experience.
Worked Example: Three Periods With Unequal Exposure
A hypothetical commercial account has loss costs of $500, $600, and $700 in three consecutive rating periods. The corresponding exposure weights are 1, 2, and 3 units. The supplied collective mean is $550 per unit and the structural constant is \(K=4\) exposure units. Calculate the credibility-weighted estimate. Assume the observations satisfy the Bühlmann-Straub moment assumptions, use no discounting or mortality adjustment, and require no specified distribution.
Method and weighted risk mean. Multiply each period loss cost by its own weight. The weighted amounts are \(1(500)=500\), \(2(600)=1{,}200\), and \(3(700)=2{,}100\). Their total is \(500+1{,}200+2{,}100=3{,}800\). Total exposure is \(W=1+2+3=6\), so
Credibility and estimate. The credibility factor is \(Z=6/(6+4)=6/10=0.6\). The complementary weight is \(1-Z=0.4\). Therefore,
The estimate is $600.00 per unit. As a check, it lies between the weighted risk mean of $633.33 and the collective mean of $550. The common mistake is to average the three period costs equally: that would give \((500+600+700)/3=600\) before any credibility blend, incorrectly treating the exposure weights as irrelevant. Here the correctly weighted risk mean is $633.33, not $600.
Worked Example: Fractional Exposure Weights
A hypothetical specialty portfolio record shows a risk’s period loss costs of $840, $760, and $700, with exposure weights of 0.5, 1.5, and 2 units. The collective mean is $680 and \(K=6\) units. Find the credibility-weighted estimate. Assume the standard conditional mean, variance, and independence conditions; no interest or mortality adjustment applies, and no named distribution is assumed.
Method and total weight. The total exposure is \(W=0.5+1.5+2=4\). Fractional weights are valid: they represent exposure in the unit used to define \(K\); they are not rounded to whole periods.
Weighted risk mean. The weighted amounts are \(0.5(840)=420\), \(1.5(760)=1{,}140\), and \(2(700)=1{,}400\). Their sum is \(420+1{,}140+1{,}400=2{,}960\). Thus,
Credibility-weighted estimate. The credibility factor is \(Z=4/(4+6)=0.4\), and the collective weight is \(1-Z=0.6\). Hence,
The estimate is $704.00. It is between $740 and $680, as it must be for a credibility factor between zero and one. The common mistake is to use \(Z=4/6\), dividing by \(K\) rather than by \(W+K\). That erroneous factor is about 0.6667 and would produce a different, overly experience-weighted estimate. The specified formula gives \(Z=4/10=0.4\).
Worked Example: Exposure Given in Policy-Years
A hypothetical liability risk has period loss costs of $48, $60, and $44 per policy-year, based on exposures of 100, 250, and 150 policy-years. The collective mean is $50 per policy-year, and \(K=250\) policy-years. Calculate the estimated loss cost. Assume the loss costs are on the same basis in every period, the Bühlmann-Straub moment assumptions apply, and there is no interest or mortality adjustment or specified probability distribution.
Method and weighted risk mean. Use the stated policy-year weights, not the number of periods. The weighted amounts are \(100(48)=4{,}800\), \(250(60)=15{,}000\), and \(150(44)=6{,}600\). Total weighted amount is \(4{,}800+15{,}000+6{,}600=26{,}400\), while total exposure is \(W=100+250+150=500\) policy-years. Therefore,
Credibility and estimate. The risk’s credibility is \(Z=500/(500+250)=500/750=2/3\), and the collective weight is \(1/3\). The estimate is
Rounded only at the final step to two decimal places, the estimate is $51.87 per policy-year. The calculation checks: it is between the experience mean of $52.80 and the collective mean of $50.00, and the larger experience weight gives the experience more influence than the collective mean. A common error is to use the three period costs as though they had equal weights. That gives an unweighted mean of \((48+60+44)/3=50.6667\), which discards the fact that the 250-policy-year period carries more weight than the 100-policy-year period.
Checks That Prevent Calculation Errors
Before reporting an estimate, check each input and intermediate result. The credibility factor is dimensionless, while \(W\) and \(K\) must use the same exposure unit. If \(W\) is in policy-years, \(K\) must also be in policy-years; adding, for example, 500 policy-years to a structural constant measured in hundreds of policy-years would be inconsistent until the units are converted. The weighted risk mean and collective mean must likewise have matching units and be stated on a comparable basis.
A correctly calculated estimate is a weighted average of \(\bar{X}_w\) and \(\mu\), since \(0\leq Z\leq1\) for nonnegative \(W\) and positive \(K\). It must lie between those two values. If it does not, check the credibility formula, the complement \(1-Z\), and the arithmetic. Also check that every period’s exposure is paired with its own period result. The sum of the weights belongs in both the weighted-mean denominator and the credibility factor; it does not replace the weighted sum in the numerator.
Match every period loss cost with its own exposure weight and confirm that all costs and exposures use consistent units.
Calculate each product \(w_jX_j\), sum those products, and divide by \(W=\sum_jw_j\).
Use the supplied structural constant in \(Z=W/(W+K)\); verify that \(Z\) is between zero and one.
Calculate \(Z\bar{X}_w+(1-Z)\mu\) and confirm the answer lies between the two means.
The method is aligned with the CAS Exam 8 (Advanced Ratemaking) syllabus. The central exam skill is not just substituting into a formula: it is recognizing which quantities are period-level results, which are exposure weights, and which supplied values are collective parameters. A correct calculation preserves those roles throughout.
Check Your Understanding
For each problem, apply the Bühlmann-Straub calculation and check that the estimate lies between the risk’s weighted mean and the collective mean. Assume the stated moment model applies and no interest or mortality adjustment is needed.
- A hypothetical risk has period loss costs $90 and $120 with weights 1 and 3. The collective mean is $100 and \(K=4\). Calculate the weighted risk mean, \(Z\), and estimate. (Answer: Weighted mean = $112.50; \(Z=0.5000\); estimate = $106.25.)
- Period costs are $30, $42, and $36, with weights 2, 1, and 3. The collective mean is $35 and \(K=6\). Calculate the estimate. (Answer: Weighted mean = $35.00; \(Z=0.5000\); estimate = $35.00.)
- A hypothetical account has costs of $210 and $180 at exposure weights 0.5 and 1.5. The collective mean is $200 and \(K=2\). Find total weight, weighted mean, credibility, and estimate. (Answer: \(W=2\); weighted mean = $187.50; \(Z=0.5000\); estimate = $193.75.)
- A risk has period loss costs of $72, $80, and $68 per exposure unit, with respective weights 100, 200, and 100. The collective mean is $75 and \(K=200\) in the same exposure units. Calculate the estimated loss cost. (Answer: Weighted mean = $75.00; \(W=400\); \(Z=0.6667\); estimate = $75.00.)
- For a hypothetical risk, period costs are $400, $520, and $460, with weights 2, 1, and 1. The collective mean is $450 and \(K=4\). Calculate the weighted mean, credibility, and estimate; explain why the unweighted mean is not the correct experience input. (Answer: Weighted mean = $445.00; \(Z=0.5000\); estimate = $447.50; the unweighted mean of $460.00 ignores the unequal exposure weights.)