Credibility Premiums for Rates and Aggregate Losses

Why Rate-Based Credibility Requires Careful Units

A credibility premium may be expressed as a claim frequency, an expected loss per exposure unit, or a total amount for a future block of exposure. These quantities are related, but they are not interchangeable. The experience must first be put on a consistent exposure basis; credibility is then applied to that rate; and only then is the result scaled to the target exposure.

For example, two periods with different exposure amounts should not generally receive equal weight when calculating an overall claim frequency. A period covering more exposure contributes more information and more claims to the combined rate. The same principle applies when the observations are aggregate loss amounts rather than claim counts.

Definition: In the Bühlmann-Straub model, let $X_j$ be the observed outcome rate for period $j$, based on exposure $w_j>0$. Conditional on the risk parameter $\Theta$, assume $E[X_j\mid\Theta]=\mu(\Theta)$ and $\operatorname{Var}(X_j\mid\Theta)=\sigma^2(\Theta)/w_j$, with observations conditionally independent. Define the collective mean $m=E[\mu(\Theta)]$, the expected process variance $v=E[\sigma^2(\Theta)]$, and the variance of hypothetical means $a=\operatorname{Var}(\mu(\Theta))$. When $a>0$, the credibility constant is $K=v/a$.

From Totals to Experience Rates

When the data are claim counts $N_j$, the period rate is $X_j=N_j/w_j$, measured in claims per exposure unit. When the data are aggregate losses $L_j$, the period rate is $X_j=L_j/w_j$, measured in dollars per exposure unit. In either case, the combined experience mean is the exposure-weighted mean:

\bar X=\frac{\sum_j w_jX_j}{\sum_j w_j} =\frac{\sum_j \text{period total}_j}{W}, \qquad W=\sum_j w_j.

Thus the combined claim frequency is total claims divided by total exposure, and the combined loss rate is total losses divided by total exposure. The weighted-mean form and the total-over-total form are equivalent because $w_jX_j$ is the period total.

This calculation assumes that the exposure units are comparable across periods and that the recorded totals and exposures use a consistent basis. If exposures differ in risk meaning, or if the data need adjustment for changes such as coverage or price level, dividing by exposure alone does not make the periods comparable. Such adjustments must be addressed before applying the credibility formula.

The Bühlmann-Straub Rate and Target Premium

Given total historical exposure $W$, the Bühlmann-Straub credibility factor is $Z=W/(W+K)$. The credibility-weighted indicated rate is a blend of the observed experience rate and the collective rate $m$. If the target period has exposure $W^*$, the corresponding indicated aggregate amount is the indicated rate multiplied by $W^*$.

Theorem: For $W>0$, the Bühlmann-Straub credibility factor and indicated rate are $$ Z=\frac{W}{W+K}, \qquad \widehat{\mu}=Z\bar X+(1-Z)m. $$ For a target exposure $W^*$ on the same basis, the indicated aggregate amount is $$ \widehat{L}^{\,*}=W^*\widehat{\mu}. $$ Here $\widehat{L}^{\,*}$ is the credibility premium for the target aggregate, not a prediction of the realized aggregate with its random fluctuation included.

A short derivation explains both the exposure weighting and the credibility factor. Write $\bar X=\sum_jw_jX_j/W$. Conditional independence and the model variance give $\operatorname{Var}(\bar X\mid\Theta)=\sigma^2(\Theta)/W$, so $E[\operatorname{Var}(\bar X\mid\Theta)]=v/W$. Also, $E[\bar X\mid\Theta]=\mu(\Theta)$. The law of total variance therefore gives $\operatorname{Var}(\bar X)=a+v/W$, while $\operatorname{Cov}(\mu(\Theta),\bar X)=a$. The coefficient of $\bar X$ in the best linear predictor of $\mu(\Theta)$ from $\bar X$ is consequently

\frac{\operatorname{Cov}(\mu(\Theta),\bar X)} {\operatorname{Var}(\bar X)} =\frac{a}{a+v/W} =\frac{W}{W+v/a} =\frac{W}{W+K}.

The predictor has mean $m$, so its form is $m+Z(\bar X-m)=Z\bar X+(1-Z)m$. Multiplying by $W^*$ converts a rate per exposure unit to an aggregate amount. This multiplication assumes the target exposure is measured on the same basis and that the indicated rate is appropriate for it.

Worked Example: Credibility-Weighted Claim Frequency

A hypothetical coverage has the following claim counts and exposures. The collective frequency is $m=0.6$ claims per exposure unit and the credibility constant is $K=5$ exposure units. Find the experience frequency, the credibility-weighted frequency, and the indicated number of claims for a target exposure of 15 units.

Method and units. First combine claim counts and exposures, rather than averaging the period frequencies equally. The frequency is measured in claims per exposure unit.

\begin{array}{c|c|c|c} \text{Period} & w_j & N_j & X_j=N_j/w_j\\ 1 & 2 & 1 & 1/2=0.5\\ 2 & 5 & 4 & 4/5=0.8\\ 3 & 3 & 3 & 3/3=1.0 \end{array}

Total exposure is $W=2+5+3=10$ units, and total claims are $1+4+3=8$. The experience frequency is therefore $\bar X=8/10=0.8$ claims per exposure unit. The credibility factor is $Z=10/(10+5)=10/15=2/3$.

The indicated frequency is $ (2/3)(0.8)+(1/3)(0.6)=8/15+1/5=11/15$, or approximately $0.7333$ claims per exposure unit. For 15 target exposure units, the indicated number of claims is $15(11/15)=11$.

Check. The indicated frequency lies between the observed rate $0.8$ and the collective rate $0.6$, as it must because $0\leq Z\leq1$. Multiplying the exact rate by the target exposure gives 11 indicated claims. The common mistake avoided here is taking the unweighted mean of $0.5,0.8,1.0$; that gives $0.7667$, not the total-claims-over-total-exposure rate $0.8$.

Worked Example: Credibility Premium for Aggregate Losses

A hypothetical portfolio reports aggregate losses of $36,000 over 4 exposure units and $54,000 over 6 exposure units. The collective loss rate is $8,000 per exposure unit, and $K=10$ exposure units. Calculate the indicated loss rate and the indicated aggregate amount for a target exposure of 12 units.

Method. Convert the loss totals to rates, then use total historical exposure in the credibility factor.

The period rates are $36{,}000/4=\$9{,}000$ and $54{,}000/6=\$9{,}000$ per exposure unit. Total losses are $\$36{,}000+\$54{,}000=\$90{,}000$, and total exposure is $W=4+6=10$. Thus the weighted experience rate is $\bar X=\$90{,}000/10=\$9{,}000$ per exposure unit.

The credibility factor is $Z=10/(10+10)=1/2$. The indicated loss rate is $ (1/2)(\$9{,}000)+(1/2)(\$8{,}000)=\$8{,}500$ per exposure unit. For 12 target units, the indicated aggregate amount is $12(\$8{,}500)=\$102{,}000$.

Check. The indicated rate is the midpoint of $9,000 and $8,000 because the credibility factor is one-half. Multiplying dollars per exposure unit by 12 exposure units gives dollars. A common error is to apply the credibility formula to the $90,000 total as if it were a rate; the collective input is a rate too, so the units would not match.

Worked Example: Scaling a Credibility Rate to a Larger Block

A hypothetical insurer observes $18,000 of losses over 3 exposure units and $42,000 over 7 exposure units. The collective loss rate is $8,000 per exposure unit, and $K=5$ exposure units. Find total historical exposure, the credibility factor, the indicated loss rate, and the indicated aggregate amount for a target exposure of 20 units.

Total historical losses are $\$18{,}000+\$42{,}000=\$60{,}000$, and total historical exposure is $W=3+7=10$. The weighted experience rate is $\bar X=\$60{,}000/10=\$6{,}000$ per exposure unit. Equivalently, the two period rates are both $\$6{,}000$: $\$18{,}000/3=\$6{,}000$ and $\$42{,}000/7=\$6{,}000$.

The credibility factor is $Z=10/(10+5)=2/3$. The indicated loss rate is $ (2/3)(\$6{,}000)+(1/3)(\$8{,}000)=\$4{,}000+\$2{,}666.666\ldots=\$6{,}666.666\ldots$ per exposure unit. Rounded to the nearest cent, this is $6,666.67 per exposure unit.

Using the unrounded indicated rate, the target amount is $20(\$20{,}000/3)=\$400{,}000/3=\$133{,}333.333\ldots$, or $133,333.33 to the nearest cent. Check. The rate is between $6,000 and $8,000, and the target amount is 20 times that unrounded rate. Rounding the rate before scaling can create a small difference in the final cents.

Common Exam Traps and Practical Checks

The arithmetic is often straightforward; the difficult part is keeping the quantity and its units clear through every step. Before finalizing a result, identify whether each value is a total or a rate, and whether it is based on historical or target exposure.

Using an ordinary mean of period rates. With unequal exposures, the relevant experience rate is $\sum_j w_jX_j/W$, not the unweighted average of the rates.
Credibility-weighting a total against a rate. Both inputs to $Z\bar X+(1-Z)m$ must be rates with matching units.
Using the wrong exposure in $Z$. $W$ is the sum of historical exposures used to form the experience mean. Target exposure $W^*$ scales the indicated rate afterward.
Scaling twice or not at all. The indicated rate is already per exposure unit. Multiply it by $W^*$ once to obtain the target aggregate amount.
Rounding too early. Retain the exact fraction or full precision through the credibility blend and target scaling; round only the requested final result.

A further modeling check is whether the target exposure and historical exposure are genuinely comparable. If a target has a different coverage basis, expected severity level, or exposure definition, a direct multiplication may not be appropriate without adjustments. The credibility calculation itself does not correct mismatched data.

Key takeaway: Calculate the experience rate as total experience divided by total exposure; apply Bühlmann-Straub credibility to that rate using historical total exposure; then multiply the indicated rate by target exposure to obtain an indicated aggregate amount.

Check Your Understanding

For each problem, keep rates and totals separate, and use unrounded values until the final step.

A hypothetical program reports 2 claims over 4 exposure units and 3 claims over 6 units. The collective frequency is 0.4 claims per unit and $K=5$ units. Find the indicated frequency and indicated claims for a target exposure of 15 units. (Answer: $0.4667$ claims per exposure unit; 7 indicated claims)
A hypothetical pool reports losses of $12,000 over 2 exposure units and $32,000 over 4 units. The collective loss rate is $8,000 per unit and $K=6$ units. Find the indicated rate and amount for a target exposure of 9 units. (Answer: approximately $7,666.67 per exposure unit; $69,000)
A hypothetical book has 2 claims over 3 exposure units and 8 claims over 7 units. The collective frequency is 0.8 claims per unit and $K=10$ units. Find the indicated frequency and the indicated number of claims for a target exposure of 25 units. (Answer: $0.9$ claims per exposure unit; 22.5 indicated claims)
A hypothetical coverage reports $18,000 over 3 exposure units and $42,000 over 7 units. Given a collective loss rate of $8,000 per unit and $K=5$ units, calculate the indicated loss rate and aggregate amount for a target exposure of 20 units. (Answer: approximately $6,666.67 per exposure unit; approximately $133,333.33)
A hypothetical portfolio has 12 claims over 20 exposure units. Its collective frequency is 0.5 claims per unit and $K=10$ units. Assume the expected severity is fixed at $2,400 per claim and use the indicated frequency to price a target of 30 exposure units. Find the indicated frequency and aggregate amount. (Answer: approximately $0.5667$ claims per exposure unit; $40,800)

What You'll Learn