Choosing an Exposure Basis for Class Experience

Why the Exposure Basis Matters

A class loss cost is meaningful only in relation to a stated amount of exposure. Losses divided by policy-years produce a cost per policy-year; the same losses divided by insured-unit-years produce a cost per insured unit-year. Those rates answer different questions, and the denominator also determines how class experience volume is measured for credibility.

Choosing an exposure basis is therefore more than a labeling decision. The basis should represent the quantity being rated, be measured consistently for the experience and the prospective class, and agree with the units used to calibrate the credibility parameters. This distinction is aligned with the CAS Exam 8 (Advanced Ratemaking) syllabus.

Definition: An exposure basis is the unit used to measure the amount of insurance exposure associated with a class. A class loss cost on that basis is aggregate losses divided by exposure on the same basis. Examples include dollars per policy-year, per vehicle-year, or per payroll unit. An earned policy-year may reflect a full policy term or a fraction of one; it is not necessarily the count of policies written.

Choosing and Matching a Denominator

A policy-year is often practical when the policy itself is the relevant rating unit and policies have reasonably comparable coverage and exposure. Insured-unit exposure can be more informative when the number of covered units varies substantially across policies and the rate is intended to apply per unit. Other measures, such as payroll, receipts, or area, may be appropriate when they better represent the quantity being rated.

The denominator alone does not make one basis universally correct. The selected measure should be available and auditable, align with the rating application, and be used consistently in historical experience, the complement, and the credibility calculation. Losses and exposures should also correspond to a consistent coverage period and loss definition.

In the Bühlmann credibility form used for class experience, $Z=n/(n+K)$, $n$ is the class volume on the selected basis and $K$ is the structural parameter on that same basis. The class indication blends its observed loss cost with a complement. The observed cost and complement must be expressed per identical exposure unit.

\text{Observed class loss cost}= \frac{\text{class losses}}{\text{class exposure}}, \qquad \text{Indicated class loss cost} =Z(\text{observed cost})+(1-Z)(\text{complement}).

The complement is the benchmark cost toward which limited class experience is tempered. In these calculations, it can be the collective mean or another selected benchmark expressed on the same basis. A credibility weight calculated from policy-years cannot automatically be applied to a cost per insured unit: the volume parameter must be on the policy-year basis too, or both must be converted consistently.

Worked Example: Converting a Fixed Exposure Basis

A hypothetical class has $7,200 of losses, 24 policy-years, and 36 unit-years. Each policy-year corresponds to 1.5 unit-years. The complement is $280 per policy-year. For the policy-year model, $K=12$ policy-years. Calculate the observed and indicated costs on both bases, using a consistently converted $K$.

Step 1: Calculate the observed costs. On the policy-year basis, the class cost is $7{,}200/24=\$300$ per policy-year. On the unit-year basis, it is $7{,}200/36=\$200$ per unit-year.

Step 2: Convert the complement and $K$. Since one policy-year corresponds to 1.5 unit-years, the complement is $280/1.5=\$186.6667$ per unit-year. The corresponding credibility parameter is $12(1.5)=18$ unit-years.

Step 3: Calculate credibility. On the policy-year basis, $Z=24/(24+12)=24/36=2/3$. On the unit-year basis, $Z=36/(36+18)=36/54=2/3$. The converted model retains the same weight.

Step 4: Calculate the indications. The policy-year indication is $(2/3)(300)+(1/3)(280)=200+93.3333=\$293.3333$ per policy-year. The unit-year indication is $(2/3)(200)+(1/3)(186.6667)=133.3333+62.2222=\$195.5555$ per unit-year. Rounding only at the end to the nearest cent gives $293.33 per policy-year and $195.56 per unit-year.

Check. Dividing the unrounded policy-year indication by 1.5 gives $293.3333/1.5=195.5555$, which agrees with the unit-year calculation. The method avoids changing the denominator while leaving the complement or $K$ in its old units.

Why Consistent Conversion Preserves Credibility

When exposures are in a fixed proportional relationship, changing units rescales both observed costs and the credibility parameter. Suppose every policy-year corresponds to $c$ unit-years. Then $n_{\text{unit}}=c n_{\text{policy}}$, and preserving the same model requires $K_{\text{unit}}=cK_{\text{policy}}$. The resulting credibility factor is unchanged:

Z_{\text{unit}} =\frac{c n_{\text{policy}}}{c n_{\text{policy}}+c K_{\text{policy}}} =\frac{n_{\text{policy}}}{n_{\text{policy}}+K_{\text{policy}}} =Z_{\text{policy}}.

Loss cost and complement per unit are each $1/c$ of their per-policy-year amounts. Since the same $Z$ weights both, the indicated cost per unit is also $1/c$ of the indicated cost per policy-year. This is a unit conversion, not a change in the underlying indication.

This invariance requires the fixed exposure relationship and consistent parameter conversion. If policies contain different numbers of insured units, a single conversion factor may not apply. In that case, choosing policy-years or unit-years may define a different model of class experience; it is not necessarily valid to convert a policy-year $K$ by multiplying by the class's average number of units.

Worked Example: Detecting a Mismatched Credibility Parameter

A hypothetical class has $5,400 of losses, 18 policy-years, and 27 unit-years. There are 1.5 unit-years per policy-year. The complement is $260 per policy-year, and the policy-year model uses $K=9$ policy-years. Compare the correctly converted unit-year indication with the result from mistakenly using $K=9$ unit-years.

Step 1: Find the observed costs and complements. The observed costs are $5{,}400/18=\$300$ per policy-year and $5{,}400/27=\$200$ per unit-year. The unit-year complement is $260/1.5=\$173.3333$.

Step 2: Convert $K$ and calculate the correct weight. The matching unit-year parameter is $9(1.5)=13.5$ unit-years. Thus $Z=27/(27+13.5)=27/40.5=2/3$. On the policy-year basis, $Z=18/(18+9)=2/3$, as expected.

Step 3: Calculate the correct indication. The policy-year indication is $(2/3)(300)+(1/3)(260)=200+86.6667=\$286.6667$. The unit-year indication is $(2/3)(200)+(1/3)(173.3333)=133.3333+57.7778=\$191.1111$. To the nearest cent, these are $286.67 per policy-year and $191.11 per unit-year.

Step 4: Show the error. If $K=9$ is incorrectly treated as unit-years, $Z=27/(27+9)=0.75$. The resulting unit-year indication is $(0.75)(200)+(0.25)(173.3333)=150+43.3333=\$193.33$, not $191.11. This incorrect calculation assigns more credibility to the class simply because the exposure was counted in smaller units. The correction is to convert $K$ to 13.5 unit-years before calculating $Z$.

When Different Bases Represent Different Models

A constant conversion is not always available. For example, a hypothetical equipment insurer may cover one, three, or six machines under different policies. Policy-years give each policy-year the same exposure count, while machine-years count the insured units directly. The two measures emphasize different features of the portfolio.

If the rating objective is a cost per machine-year, unit exposure may be the more direct denominator. A policy-year measure could obscure the difference between a policy covering one machine and one covering six. Conversely, if the policy is the relevant rating and its coverage package is the unit being priced, policy-years may be appropriate even when policy sizes vary. Selection requires judgment about the rating unit and the assumptions supporting the credibility model.

Worked Example: Interpreting Two Calibrated Bases

A hypothetical equipment class has 8 policy-years, 32 machine-years, and $9,600 of losses. Policies have varying numbers of machines. A policy-year model has a complement of $1,100 per policy-year and $K=8$ policy-years. A separately calibrated machine-year model has a complement of $270 per machine-year and $K=24$ machine-years. Calculate each model's indication and explain why the results need not be simple unit conversions.

Step 1: Calculate observed costs. The policy-year observed cost is $9{,}600/8=\$1{,}200$ per policy-year. The machine-year observed cost is $9{,}600/32=\$300$ per machine-year.

Step 2: Calculate policy-year credibility and indication. $Z=8/(8+8)=0.5$. The indication is $(0.5)(1{,}200)+(0.5)(1{,}100)=600+550=\$1{,}150$ per policy-year.

Step 3: Calculate machine-year credibility and indication. $Z=32/(32+24)=32/56=4/7$. The indication is $(4/7)(300)+(3/7)(270)=1{,}200/7+810/7=2{,}010/7=\$287.1429$ per machine-year, or $287.14 to the nearest cent.

Check and interpretation. The aggregate ratio is four machine-years per policy-year, but the models are separately calibrated, and policy sizes vary. Dividing the policy indication by four gives $287.50 per machine-year, not $287.14. That difference is not an arithmetic contradiction: the complements and $K$ values do not represent a single model converted by a fixed factor. The method avoids assuming that a class-average unit count automatically converts every model parameter.

Common Mistakes and Practical Checks

The denominator should be stated alongside every loss cost and complement. A bare figure such as “$300” is ambiguous if it is not clear whether it is per policy-year or per insured unit-year. Before calculating credibility, verify the units of $n$ and $K$; before blending, verify the units of the observed cost and complement.

Using written policies as earned policy-years. A policy in force for half a year ordinarily contributes half a policy-year, not one full policy-year.
Changing the exposure count but not $K$. If exposures are converted by a fixed factor and the same model is intended, convert $K$ by that factor too.
Blending costs with different denominators. Convert one cost to the other's basis before applying the credibility weight.
Treating average unit count as a universal conversion. This can be misleading when policy sizes vary or their mix changes across experience and prospective periods.
Assuming the largest denominator is automatically best. A basis should reflect the rating objective and the model calibration, not merely produce a larger numerical exposure count.

A useful final check is dimensional: $Z$ must be dimensionless, while the observed cost, complement, and indicated cost must all have the same monetary-per-exposure units. When a constant unit conversion is claimed, the final rates should convert by the corresponding reciprocal factor.

Check Your Understanding

For each problem, keep the exposure basis and the credibility parameter in matching units.

A hypothetical class has $4,500 of losses, 15 policy-years, a complement of $280 per policy-year, and $K=10$ policy-years. Calculate $Z$ and the indicated cost per policy-year. (Answer: $Z=0.6000$; $292.00 per policy-year.)
Each policy-year corresponds to 2 unit-years. A class has 14 policy-years and 28 unit-years, and the policy-year model uses $K=7$ policy-years. What is the matching unit-year $K$, and what credibility weight results on either basis? (Answer: $K=14$ unit-years; $Z=0.6667$ on both bases.)
A class has $12,600 of losses and 42 unit-years. Its complement is $260 per unit-year and $K=28$ unit-years. Calculate the indicated cost per unit-year. (Answer: $Z=0.6000$; $284.00 per unit-year.)
A class has $4,800 of losses, 16 policy-years, and 40 unit-years. Each policy-year corresponds to 2.5 unit-years. The policy-year complement is $280 and $K=8$ policy-years. Calculate the indication per unit-year using a converted complement and $K$. (Answer: $Z=0.6667$; $117.33 per unit-year.)
A class has $8,100 of losses, 9 policy-years, and 27 unit-years, with a fixed conversion of 3 unit-years per policy-year. The complement is $750 per policy-year, and $K=9$ policy-years. Calculate the correctly converted unit-year indication, then state the result if someone incorrectly uses $K=9$ unit-years. (Answer: Correct $K=27$, $Z=0.5000$, and indication $275.00 per unit-year; incorrect $Z=0.7500$ and indication $287.50 per unit-year.)

What You'll Learn