From a Credibility Estimate to a Rate
A Bühlmann credibility estimate is an estimate of expected loss for a risk, expressed on the same basis as the observations used to construct it. It is not automatically a quoted premium. To use it in ratemaking, first identify the unit of exposure and the loss components being estimated. Then decide whether the requested amount is a loss cost, an aggregate expected loss, or a premium that also covers expenses.
For example, a hypothetical insurer might estimate expected claims per insured vehicle-year. That amount is a loss cost per vehicle-year. Multiplying it by the number of prospective vehicle-years gives expected aggregate losses, not necessarily the total premium. If expenses are to be included, they must be added or loaded according to how those expenses are defined.
The phrase “per unit” matters. If \(\bar X\) is measured per vehicle-year while \(\mu\) is per policy-year, the weighted average has no sound interpretation until both amounts are converted to a common basis. The same applies to historical and prospective loss levels: the observations and collective complement must represent the basis intended for the indicated rate.
In the standard Bühlmann setup, \(Z=n/(n+K)\), where \(n\) is the number of observation units and \(K\) is the Bühlmann constant. This tutorial uses the resulting factor to form a loss cost; it does not change how \(Z\) or \(K\) is estimated. The collective complement \(\mu\) is the mean for the reference collective, not an expense provision.
Why the Credibility Estimate Is a Loss Cost
The key justification is that credibility combines two estimates of the same quantity on the same scale: the risk's observed average and the collective mean. In the Bühlmann model, write \(\Theta\) for the risk parameter and \(m(\Theta)=E[X\mid\Theta]\) for that risk's expected loss per observation unit. The collective mean is \(\mu=E[m(\Theta)]\). The Bühlmann linear estimate of \(m(\Theta)\) is \(Z\bar X+(1-Z)\mu\).
For a prospective unit with the same expected loss structure, the estimated expected loss is that estimated mean. For \(q\) comparable prospective units, linearity of expectation gives the aggregate estimate below. This does not require the losses on different units to be independent for the expectation itself, although independence or other assumptions may matter for a variance calculation.
The interpretation depends on comparable units and a consistent loss definition. If the data measure losses after a particular deductible or limit, but the intended rate covers losses on a different basis, the estimate cannot be used unchanged. Likewise, a change in coverage, exposure definition, or loss level may require adjustments before the weighted estimate represents the prospective loss cost.
Worked Example: Converting a Weighted Estimate to Aggregate Expected Loss
A hypothetical small-commercial auto portfolio has a credibility factor of \(Z=0.40\). Its observed average incurred loss is $820 per vehicle-year, and the selected collective mean is $700 per vehicle-year. Calculate the indicated loss cost per vehicle-year and the expected losses for 125 prospective vehicle-years.
Method. Weight the two per-vehicle-year loss estimates, then multiply the resulting loss cost by prospective exposure. Both inputs are already on the same basis.
The complement weight is \(1-Z=1-0.40=0.60\). The two weighted amounts are $$ 0.40(820)=328,\qquad 0.60(700)=420. $$ Therefore, the indicated loss cost is $$ \hat m=328+420=\$748.00\text{ per vehicle-year}. $$ For 125 prospective vehicle-years, the estimated aggregate losses are $$ 125(748)=\$93{,}500.00. $$ Check. The estimate, $748, lies between the observed average of $820 and the collective mean of $700, as it must for a credibility-weighted average with \(0\leq Z\leq1\). The aggregate calculation also uses vehicle-years, not policies, as its multiplier.
Common mistake avoided: Applying \(Z\) to the observed amount but forgetting to apply \(1-Z\) to the collective complement would not produce the specified credibility estimate.
Adding Expenses Without Changing the Loss Estimate
A loss cost estimates losses per exposure unit. A premium generally also needs to provide for expenses, and sometimes other explicitly selected provisions. Keep that distinction visible: first calculate the credibility-indicated loss cost; then apply the expense method required by the problem.
Suppose \(L\) is the indicated loss cost per exposure unit, \(e\) is a fixed expense per exposure unit, and \(r\) is the fraction of gross premium allocated to variable expenses. If gross premium per unit is \(P\), the premium must cover the loss cost, the fixed expense, and the variable expense \(rP\). Solving this equation gives the gross premium formula below, provided \(r<1\).
The denominator reflects that a premium-based expense is a share of the premium itself. If instead a variable expense is stated as a percentage of losses, it is not \(rP\), and this gross-up formula does not apply as written. Always read the expense basis carefully. If the question asks only for a loss cost, do not add expenses at all.
Worked Example: Loading a Loss Cost for Fixed and Variable Expenses
A hypothetical equipment policy has 24 past exposure units and a Bühlmann constant \(K=36\). The total observed losses are $52,800, and the collective mean is $1,900 per exposure unit. For the prospective rating, assume 18 exposure units. Fixed expenses are $75 per exposure unit, and variable expenses are 12% of gross premium. Find the credibility-indicated loss cost, aggregate expected losses, and gross premium per exposure unit and in total.
Method. Calculate \(Z\), convert the observed total to an average per exposure unit, and form the loss cost. Apply expenses only after that estimate is obtained.
First, $$ Z=\frac{n}{n+K}=\frac{24}{24+36}=\frac{24}{60}=0.40. $$ The observed average loss per exposure unit is $$ \bar X=\frac{\$52{,}800}{24}=\$2{,}200. $$ Thus the credibility-indicated loss cost is $$ L=0.40(\$2{,}200)+0.60(\$1{,}900) =\$880+\$1{,}140 =\$2{,}020\text{ per exposure unit}. $$ For 18 prospective units, aggregate expected losses are $$ 18(\$2{,}020)=\$36{,}360.00. $$ The gross premium per exposure unit is $$ P=\frac{L+e}{1-r} =\frac{\$2{,}020+\$75}{1-0.12} =\frac{\$2{,}095}{0.88} =\$2{,}380.6818\ldots. $$ Rounded to cents, the premium is $2,380.68 per exposure unit. For 18 units, the total premium, calculated before rounding the per-unit amount, is $$ 18\left(\frac{\$2{,}095}{0.88}\right) =\$42{,}852.2727\ldots =\$42{,}852.27. $$ Check. At the unrounded per-unit premium, the variable expense is \(0.12(\$2{,}380.6818\ldots)=\$285.6818\ldots\). Adding the $2,020 loss cost and $75 fixed expense gives $2,380.6818…, the gross premium. The expenses have not been included in the credibility-weighted loss estimate.
Common mistake avoided: Treating 12% of premium as 12% of losses would use the wrong expense base and would not satisfy the stated premium equation.
Keeping the Collective Complement and Expenses Distinct
The collective complement is part of the credibility estimate because it supplies the portion of the expected loss estimate not assigned to the risk's own experience. Expenses serve a different purpose: they cover costs associated with issuing or servicing coverage, or other stated premium components. Unless the modeled observations explicitly include an expense component, adding expenses to the observed losses or to the collective mean mixes unlike quantities.
A useful order of operations is:
Determine whether the requested result is a per-unit loss cost, aggregate expected loss, or gross premium.
Check that experience, collective complement, and intended prospective rate use comparable exposure units and loss definitions.
Use \(Z\bar X+(1-Z)\mu\) to estimate expected loss per unit.
Multiply by prospective exposure for aggregate expected losses, or apply the stated expense provisions for premium.
This order also helps distinguish per-policy and per-exposure expenses. A fixed expense of $60 per policy cannot be added directly to a loss cost per vehicle-year if each policy covers a varying number of vehicle-years. Convert the expense to the same unit, or calculate policy premium using the appropriate number of units per policy.
Worked Example: Avoiding Expense Double-Counting
For a hypothetical liability class, \(Z=0.60\), the observed average loss is $1,050 per exposure unit, and the collective mean is $900 per exposure unit. Fixed expenses are $45 per exposure unit, while variable expenses are 10% of gross premium. Calculate the loss cost and gross premium. Then identify the error in adding expenses to both credibility inputs before applying credibility.
Method. Calculate the loss estimate using loss amounts only, then solve the premium equation. The expense assumptions are outside the credibility calculation.
The credibility estimate is $$ L=0.60(\$1{,}050)+0.40(\$900) =\$630+\$360 =\$990\text{ per exposure unit}. $$ The gross premium is $$ P=\frac{\$990+\$45}{1-0.10} =\frac{\$1{,}035}{0.90} =\$1{,}150.00. $$ The premium check is \( \$990+\$45+0.10(\$1{,}150)=\$990+\$45+\$115=\$1{,}150\).
An incorrect approach would add the fixed expense to each credibility input first: \(0.60(\$1{,}050+\$45)+0.40(\$900+\$45)=\$1{,}035\). That result already includes $45 of fixed expense. If it is then treated as a loss cost and the fixed expense is added again, the mistaken premium becomes $$ \frac{\$1{,}035+\$45}{0.90}=\$1{,}200.00, $$ which is $50 too high. The $45 was counted twice and then grossed up by the variable-expense factor. Keep the loss estimate and expense provisions separate unless the problem explicitly defines expenses as part of the modeled amounts.
Interpretation and Exam Traps
A credibility-indicated loss cost is an estimate, not a guarantee of the losses that will occur. Its interpretation relies on using the specified credibility model and an appropriate collective complement. Even a correctly calculated estimate can be unsuitable for a rate if the exposure units, coverage terms, or loss basis do not match the prospective policy.
- Mixing units: Do not average a loss per policy with a collective mean per vehicle-year. Convert first.
- Confusing aggregate losses with premium: Multiplying a loss cost by exposure produces expected losses, not a gross premium with expenses.
- Using the wrong expense denominator: An expense percentage of premium requires a premium gross-up. A percentage of losses has a different base.
- Credibility-weighting expenses by accident: Apply credibility to the modeled loss quantity, not to a total that includes separately rated expenses.
- Rounding too early: Keep full precision through the premium calculation and round the final money amount as requested.
Check Your Understanding
For each problem, keep the loss estimate separate from expenses and use the stated exposure basis.
- A hypothetical property class has \(Z=0.25\), an observed average loss of $1,200 per location-year, and a collective mean of $1,000 per location-year. What is the indicated loss cost, and what aggregate expected loss does it imply for 40 location-years? (Answer: $1,050 per location-year; $42,000 aggregate expected loss.)
- A risk has \(n=18\), \(K=27\), observed total losses of $43,200 over those 18 exposure units, and a collective mean of $2,100 per exposure unit. Calculate its loss cost and aggregate expected losses for 30 prospective exposure units. (Answer: \(Z=0.40\); $2,220 per exposure unit; $66,600 aggregate expected loss.)
- A credibility calculation has already produced a loss cost of $1,450 per exposure unit. Fixed expenses are $80 per unit, and variable expenses are 8% of gross premium. Calculate gross premium per exposure unit, rounded to cents. (Answer: $1,663.04.)
- A hypothetical class has \(Z=0.70\), observed losses averaging $680 per exposure unit, and a collective mean of $740 per exposure unit. Find the indicated loss cost and aggregate expected losses for 250 prospective units. (Answer: $698 per unit; $174,500 aggregate expected loss.)
- A risk has \(n=32\), \(K=48\), observed total losses of $96,000, and a collective mean of $2,700 per exposure unit. Fixed expenses are $110 per unit, variable expenses are 15% of gross premium, and the prospective book has 12 units. Find the indicated loss cost, gross premium per unit, and total premium, rounding money to cents. (Answer: \(Z=0.40\); loss cost $2,820 per unit; gross premium $3,447.06 per unit; total premium $41,364.71.)