From Credibility Estimates To Experience-Rated Premiums

From an Experience Estimate to a Premium

A credibility estimate is not yet a premium. It is an estimate of future loss cost, formed by combining a risk’s observed experience with a collective estimate. To turn it into a prospective premium, the actuary must put both estimates on a consistent basis, apply the credibility weight, and multiply by the exposure expected to be insured. If the requested premium includes expenses, those loads are added using their stated expense basis.

The credibility weight $Z$ is assumed to have already been determined by an appropriate credibility procedure. This tutorial uses that weight rather than deriving it. Its role here is to translate the resulting estimate into a prospective price. A valid calculation requires $0\leq Z\leq 1$, and the observed estimate and collective estimate must measure the same quantity.

Definition: Let $\widehat{L}_{\mathrm{obs}}$ be a risk’s observed loss-cost estimate, $L_{\mathrm{coll}}$ the collective loss-cost estimate, and $Z$ the credibility assigned to the observed estimate. The credibility-weighted prospective loss-cost estimate is $$ \widehat{L}_{\mathrm{cred}} =Z\widehat{L}_{\mathrm{obs}}+(1-Z)L_{\mathrm{coll}}. $$ If prospective exposure is $E_{\mathrm{fut}}$, the corresponding pure premium is $E_{\mathrm{fut}}\widehat{L}_{\mathrm{cred}}$. Here, a pure premium is the expected loss cost before expense and other premium loads.

A loss cost is typically expressed as dollars per exposure unit. An exposure unit might be a vehicle-year, payroll unit, insured location-year, or another measure specified for the coverage. The observed estimate is often total losses divided by historical exposure, but it might already be supplied as a loss cost. The denominator must represent the exposure that generated the losses, not the exposure expected in the prospective period.

Before blending, check that both estimates refer to the same coverage, limits, deductible, loss definition, currency, and exposure unit. If historical losses have not been adjusted for development or trend, they may not be comparable with a prospective collective estimate. Make any required adjustments first; do not blend estimates that are on different time or coverage bases.

Why the Credibility-Weighted Estimate Forecasts Losses

The blending formula has a direct forecasting interpretation under a standard conditional model. Let $\theta$ represent a risk’s underlying loss-cost level, $D$ its observed historical data, and $L_{\mathrm{future}}$ its future losses for one exposure unit. Suppose that, conditional on $\theta$, future losses are independent of the historical data and have conditional mean $\theta$. Then

E[L_{\mathrm{future}}\mid \theta,D]=\theta.

By the law of iterated expectations,

E[L_{\mathrm{future}}\mid D] =E\!\left[E[L_{\mathrm{future}}\mid\theta,D]\mid D\right] =E[\theta\mid D].

The conditional-independence assumption matters: it justifies replacing the inner conditional expectation by $\theta$ even after conditioning on $D$. Without that assumption, the replacement is not established merely by knowing $E[L_{\mathrm{future}}\mid\theta]=\theta$. A credibility procedure may estimate this conditional mean by a linear blend of observed and collective loss costs. The formula is not a claim that every possible model produces the same $Z$ or the same estimate; it is the pricing step once the estimates and weight have been selected.

The blend is a weighted average. When $Z=0$, the risk’s experience has no effect and the estimate equals the collective estimate. When $Z=1$, the estimate equals the observed estimate. For intermediate values, it lies between the two, provided both estimates are finite and on the same basis. This property is a useful arithmetic check.

Calculating the Prospective Pure Premium

The calculation is easiest to audit when separated into stages. First calculate or confirm the observed loss cost. Next apply any required adjustments to make it prospective. Then blend it with the collective estimate, and only then multiply by prospective exposure.

Confirm the basis.
Identify the coverage, loss measure, time period, and exposure unit for each estimate. Adjust historical experience when needed so that it is comparable with the prospective collective estimate.

Calculate observed loss cost.
When data are given as total losses and exposure, divide losses by the corresponding historical exposure. Do not substitute prospective exposure as the denominator.

Apply credibility.
Calculate $Z\widehat{L}_{\mathrm{obs}}+(1-Z)L_{\mathrm{coll}}$, using unrounded inputs when possible.

Calculate pure premium.
Multiply the credibility-weighted loss cost per exposure unit by the prospective number of exposure units.

Worked Example: Blending a Pooled Observed Loss Cost

A hypothetical account has losses of $54,000 over 60 exposure units in one historical period and $88,000 over 80 exposure units in another. Both periods are already adjusted to the same prospective coverage and loss basis. The credibility assigned to the observed estimate is $Z=0.40$, the collective loss cost is $1,100 per exposure unit, and prospective exposure is 100 units. Find the pooled observed loss cost, the credibility-weighted loss cost, and the pure premium.

Method. Calculate the observed loss cost using the losses and exposure that generated them. Then blend that rate with the collective rate and apply the prospective exposure.

The combined historical losses are

$$ 54{,}000+88{,}000=142{,}000. $$

The combined historical exposure is

60+80=140\text{ units}.

Thus, the pooled observed loss cost is

\widehat{L}_{\mathrm{obs}} =\frac{142{,}000}{140} =1{,}014.285714\ldots \text{ dollars per unit}.

The credibility-weighted loss cost is

\begin{aligned} \widehat{L}_{\mathrm{cred}} &=0.40(1{,}014.285714\ldots)+0.60(1{,}100)\\ &=405.714285\ldots+660\\ &=1{,}065.714285\ldots \text{ dollars per unit}. \end{aligned}

The pure premium for 100 prospective units is

100(1{,}065.714285\ldots)=106{,}571.4285\ldots.

Rounded to the nearest cent, the pooled observed loss cost is $1,014.29 per unit, the credibility-weighted loss cost is $1,065.71 per unit, and the pure premium is $106,571.43. The weighted cost lies between $1,014.29 and $1,100, as required.

Common mistake avoided: Multiplying the pooled observed cost by 100 would give about $101,428.57, but that ignores the credibility complement. The prospective pure premium uses the credibility-weighted cost, not the unblended observed cost.

Adjusting Experience to the Prospective Period

When historical loss costs are not already on a prospective basis, an adjustment may be needed before credibility is applied. For example, if a stated annual trend rate $t$ applies for $k$ years, a historical cost $L$ is trended to $L(1+t)^k$, assuming the rate compounds annually and the stated number of years is the correct interval. Exposure weights matter when the adjusted costs come from periods with different exposure. Use the observed estimate specified by the rating method; do not take an unweighted average of rates when the available data represent unequal amounts of exposure.

Worked Example: Trending Before Credibility Blending

A hypothetical risk has a loss cost of $700 per exposure unit from a period two years before the prospective period and $900 per unit from a period one year before it. Each period has 100 exposure units. Annual trend is 5%, compounded once per year. The credibility weight is $Z=0.50$, the prospective collective loss cost is $850 per unit, and the prospective period has 200 exposure units. Find the trend-adjusted observed estimate and pure premium.

Method. Trend each historical cost to the prospective period, combine the two equally exposed adjusted costs, apply credibility, and multiply by prospective exposure. The annual compounding assumption means the two-year factor is $1.05^2$, not $1+2(0.05)$.

The cost from two years before the prospective period becomes

$$ 700(1.05)^2=700(1.1025)=771.75. $$

The cost from one year before becomes

$$ 900(1.05)=945. $$

Since both periods have 100 exposure units, the exposure-weighted trend-adjusted observed cost is

\widehat{L}_{\mathrm{obs}} =\frac{100(771.75)+100(945)}{100+100} =\frac{171{,}675}{200} =858.375.

Apply credibility:

\begin{aligned} \widehat{L}_{\mathrm{cred}} &=0.50(858.375)+0.50(850)\\ &=429.1875+425\\ &=854.1875 \text{ dollars per unit}. \end{aligned}

The pure premium is

$$ 200(854.1875)=170{,}837.50. $$

The trend-adjusted observed estimate is $858.375 per unit, the credibility-weighted cost is $854.1875 per unit, and the pure premium is $170,837.50. These results are checked by noting that $854.1875 lies between the adjusted observed cost and the collective estimate.

Common mistake avoided: The untrended average of $700 and $900 is not the prospective observed estimate. Applying credibility to that historical average would mix a historical-basis risk estimate with a prospective-basis collective estimate.

From Pure Premium to Gross Premium

A pure premium covers expected losses only. A gross premium may also include expenses, taxes, assessments, commissions, or other loads, depending on the rating problem. Read each expense definition carefully. In particular, a variable expense stated as a percentage of gross premium must be solved for; it is not simply added as that percentage of the loss cost.

Suppose the credibility-weighted loss cost is $L_{\mathrm{cred}}$ per unit, fixed expense is $F$ per unit, and variable expense is a fraction $q$ of gross premium per unit. If there are no other loads, the gross premium per unit $G$ satisfies

G=L_{\mathrm{cred}}+F+qG.

Collecting the gross-premium terms gives

G(1-q)=L_{\mathrm{cred}}+F, \qquad G=\frac{L_{\mathrm{cred}}+F}{1-q}, \quad 0\leq q<1.

The division by $1-q$ follows because the variable expense is calculated as a share of the gross premium itself. The expression does not apply unchanged if the expense percentage is instead defined as a share of loss cost, or if the rating problem specifies other loads or a different expense base.

Worked Example: Grossing Up for Variable Expense

A hypothetical class has an observed loss cost of $1,500 per unit, a collective estimate of $1,200 per unit, and $Z=0.60$. Fixed expense is $80 per unit, and variable expense is 20% of gross premium. Find the gross premium for 40 exposure units, assuming there are no other loads.

Method. First find the credibility-weighted loss cost. Add fixed expense, then solve the gross-premium equation for the variable expense.

The credibility-weighted loss cost is

\begin{aligned} L_{\mathrm{cred}} &=0.60(1{,}500)+0.40(1{,}200)\\ &=900+480\\ &=1{,}380\text{ dollars per unit}. \end{aligned}

Before variable expense, the loss cost plus fixed expense is

1{,}380+80=1{,}460\text{ dollars per unit}.

Let $G$ denote gross premium per unit. Since variable expense is 20% of $G$,

G=1{,}460+0.20G, \qquad 0.80G=1{,}460, \qquad G=\frac{1{,}460}{0.80}=1{,}825.

For 40 exposure units, the gross premium is

$$ 40(1{,}825)=73{,}000. $$

The gross premium is $1,825 per unit, or $73,000.00 for 40 units. As a check, the total per-unit variable expense is $0.20(1{,}825)=$365$; adding $1,380 of loss cost, $80 of fixed expense, and $365 of variable expense gives $1,825.

Common mistake avoided: Adding 20% of $1,460 produces $1,752, but that calculates variable expense as a percentage of the pre-expense amount, not of gross premium. The specified denominator is gross premium, so the correct calculation divides by $1-0.20$.

Checks Before Reporting a Premium

A well-presented result identifies both the rate and the total premium. For example, a loss cost of $900 per unit is not the same quantity as a $90,000 pure premium; the latter also depends on prospective exposure. Report the units and basis explicitly so that the calculation can be reviewed.

Use these checks before finalizing an experience-rated premium:

Basis check: Both components of the credibility blend refer to the same coverage and prospective loss-cost basis.
Weight check: $Z$ and $1-Z$ are nonnegative and sum to one.
Range check: The weighted cost lies between the observed and collective estimates.
Exposure check: Historical exposure is used to calculate historical loss cost; prospective exposure is used to calculate the prospective total.
Expense check: Each expense is applied to the stated base, and any percentage of gross premium is solved algebraically.

A credibility-weighted estimate does not eliminate judgment about the underlying assumptions. If experience and collective estimates refer to different deductibles or limits, a numerical blend does not repair the mismatch. If the prospective exposure is uncertain, the rate per unit and total premium should be distinguished rather than treating them as interchangeable.

Key takeaway: Calculate the credibility-weighted loss cost on a common prospective basis first. Multiply that cost by prospective exposure to obtain pure premium; add expenses only afterward, using the exact expense bases specified.

Check Your Understanding

For each problem, calculate the requested prospective rate or premium. Round money to the nearest cent where needed.

A hypothetical risk has $64,800 in losses over 72 historical exposure units. The collective loss cost is $760 per unit, $Z=0.25$, and prospective exposure is 90 units. Find the observed loss cost, credibility-weighted cost, and pure premium. (Answer: $900.00 per unit; $795.00 per unit; $71,550.00)
A risk’s loss costs are $600 per unit from two years before the prospective period and $720 per unit from one year before. The periods have 120 and 80 exposure units, respectively. Trend is 4% annually, compounded annually. If $Z=0.40$, the collective cost is $700 per unit, and prospective exposure is 150 units, find the pure premium. (Answer: $104,333.76)
A hypothetical class has observed loss cost $1,250 per unit, collective cost $1,000 per unit, and $Z=0.70$. Fixed expense is $55 per unit, variable expense is 15% of gross premium, and prospective exposure is 25 units. Find the gross premium. (Answer: $36,176.47)
A risk has $96,000 of historical losses over 120 exposure units, a collective cost of $680 per unit, and $Z=0.50$. Prospective exposure is 160 units. Find the credibility-weighted cost and pure premium. (Answer: $740.00 per unit; $118,400.00)
A hypothetical account’s observed loss cost is $2,100 per unit, its collective estimate is $1,800 per unit, and $Z=0.35$. There are no expenses, and prospective exposure is 45 units. Find the pure premium and state whether the weighted cost is closer to the observed or collective estimate. (Answer: $1,905.00 per unit; $85,725.00; closer to the collective estimate)

What You'll Learn