Credibility Weighting Rates and Loss Costs

Why the Basis of a Credibility Blend Matters

A credibility estimate combines experience with a complement, such as a broader portfolio estimate. The arithmetic may be short, but the result is only meaningful if both quantities being combined estimate the same thing on the same basis. A loss cost per policy-year cannot be blended directly with a loss cost per $1,000 of payroll, and a dollar rate cannot be blended directly with a dimensionless relativity.

This tutorial focuses on credibility weighting when the quantity of interest is expressed as a rate, loss cost, or relativity. It is aligned with the CAS Exam 5 (Basic Ratemaking and Estimated Claim Liabilities) syllabus. The key work is often not multiplying by the credibility factor; it is confirming that the experience estimate and complement use a consistent definition, period, exposure base, and unit.

Definition: A credibility-weighted estimate is a blend of an experience estimate and a complement: $$ \widehat{\theta}=Z\widehat{\theta}_{\mathrm{exp}}+(1-Z)\widehat{\theta}_{\mathrm{comp}}, $$ where $0\leq Z\leq 1$, $\widehat{\theta}_{\mathrm{exp}}$ is the estimate from the subject risk's experience, and $\widehat{\theta}_{\mathrm{comp}}$ is an estimate for the same quantity from a broader or otherwise selected source. Both estimates must have the same units and refer to the same exposure basis and target period.

In the Bühlmann-Straub setting, experience is commonly summarized as losses divided by exposure, with exposure weights accounting for differing amounts of data. The credibility factor $Z$ reflects the information in that experience relative to the model's uncertainty; its calculation and interpretation were established in the earlier material on Bühlmann-Straub credibility. Here, treat $Z$ as given or as already calculated. The blend is appropriate only when the subject experience and complement are comparable estimates of the same risk quantity.

Rates and Loss Costs Must Share an Exposure Unit

A loss cost is expected loss per unit of exposure. If total losses are $L$ and total exposure is $E$, the observed loss cost is $L/E$. The units of the result follow from the denominator: losses divided by policy-years produce dollars per policy-year; losses divided by thousands of dollars of payroll produce dollars per $1,000 of payroll.

When experience comes from several periods or groups with rates $r_j=L_j/E_j$, the combined observed rate is not generally the simple arithmetic average of those rates. It is the exposure-weighted average:

\frac{\sum_j L_j}{\sum_j E_j} = \frac{\sum_j E_jr_j}{\sum_j E_j}.

The identity follows because $L_j=E_jr_j$, so adding the losses gives $\sum_j L_j=\sum_j E_jr_j$. Dividing by total exposure yields the displayed expression. Thus each period contributes in proportion to its exposure, not equally merely because it is one row in a data table.

Worked Example: Blending a Loss Cost per Policy-Year

A hypothetical small commercial program has $640,000 of adjusted losses over 6,000 policy-years. Its selected complement is a loss cost of $120 per policy-year, and the credibility factor is $Z=0.60$. Estimate the credibility-weighted loss cost. Use exact arithmetic until the final result.

Method: calculate the subject loss cost, then blend quantities with matching units. The experience estimate is:

\widehat{\theta}_{\mathrm{exp}} =\frac{\$640{,}000}{6{,}000\text{ policy-years}} =\frac{\$320}{3}\text{ per policy-year}.

Both the experience estimate and complement are dollars per policy-year. Therefore the blend is valid:

\widehat{\theta} =0.60\left(\frac{\$320}{3}\right)+0.40(\$120) =\$64+\$48 =\$112\text{ per policy-year}.

The estimate is $112.00 per policy-year, rounded to cents. As a check, it lies between the experience estimate of approximately $106.67 and the complement of $120, as it must for a credibility factor between zero and one. The calculation uses the exact fraction $320/3$, so no equality relies on a rounded intermediate value.

Common mistake avoided: substituting a rounded observed rate such as $106.6667 into an equality that treats it as exact. That rounded value gives a slightly different intermediate product. Either retain the exact ratio, as above, or mark the rounded calculation as approximate and round only the final estimate.

Convert Before You Blend

Rate labels can conceal different denominators. A rate of $15 per $1,000 of payroll and a rate of $1.50 per $100 of payroll describe the same rate, but the numerical values cannot be added or averaged until they are expressed with the same denominator. Conversion changes the number, not the underlying economic quantity.

The same principle applies when the experience and complement use different exposure definitions. If one is per vehicle-year and the other per insured vehicle, first determine whether the exposures are equivalent for the purpose at hand. If they are not, a simple unit conversion cannot fix the mismatch: the estimates do not yet refer to the same quantity.

Worked Example: Converting a Payroll Rate Before Credibility Weighting

A hypothetical contractor program has $75,000 of adjusted losses on 5,000 units of $1,000 payroll exposure. The complement is $16.50 per $1,000 of payroll, and $Z=0.40$. Calculate the credibility-weighted rate per $1,000 of payroll and then express it per $100 of payroll.

Step 1: Calculate the experience rate on the stated exposure base.

\widehat{\theta}_{\mathrm{exp}} =\frac{\$75{,}000}{5{,}000\text{ units of \$1,000 payroll}} =\$15.00\text{ per \$1,000 of payroll}.

Step 2: Blend rates with matching units. Both values are stated per $1,000 of payroll:

\widehat{\theta} =0.40(\$15.00)+0.60(\$16.50) =\$6.00+\$9.90 =\$15.90\text{ per \$1,000 of payroll}.

Step 3: Convert the result. Since $100 is one-tenth of $1,000, the rate per $100 of payroll is one-tenth of the rate per $1,000:

\$15.90\div 10=\$1.59\text{ per \$100 of payroll}.

The final rates are $15.90 per $1,000 or $1.59 per $100 of payroll. As a check, the weighted rate is between $15.00 and $16.50 and converts consistently: ten units of $100 payroll exposure cost $10 \times \$1.59=\$15.90. A common error is to divide the losses by 5,000 and label the result per $100; the denominator is 5,000 units of $1,000, so the initial result is per $1,000.

Credibility-Weighting Relativities

A relativity is dimensionless: it compares a rate for a class with a specified base rate. A class relativity of 1.20 means the class rate is 1.20 times the base rate under the stated definitions. A credibility blend of relativities is valid when both the experience relativity and complement relativity use the same base class, coverage, time period, and rating basis.

If the common base loss cost is $B$, the experience class loss cost is $C_{\mathrm{exp}}$, and the complement class loss cost is $C_{\mathrm{comp}}$, the corresponding relativities are $C_{\mathrm{exp}}/B$ and $C_{\mathrm{comp}}/B$. Credibility-weighting those relativities gives:

Z\frac{C_{\mathrm{exp}}}{B}+(1-Z)\frac{C_{\mathrm{comp}}}{B} = \frac{ZC_{\mathrm{exp}}+(1-Z)C_{\mathrm{comp}}}{B}, \qquad B>0.

The equality holds because the denominator $B$ is common to both terms and can be factored out. It shows why blending consistent relativities agrees with first blending the corresponding class loss costs and then dividing by that same base. If the base differs between the two estimates, this equivalence does not apply; express them on one common base before comparing or blending.

Worked Example: Blending a Class Relativity

A hypothetical equipment-insurance class has an experience loss cost of $216 per policy-year. The common base-class loss cost is $180 per policy-year. A selected complement gives a class loss cost of $207 per policy-year on that same basis, and $Z=0.40$. Find the credibility-weighted relativity.

Step 1: Convert each class loss cost to a relativity using the same base.

R_{\mathrm{exp}}=\frac{\$216}{\$180}=1.20, \qquad R_{\mathrm{comp}}=\frac{\$207}{\$180}=1.15.

Step 2: Blend the dimensionless relativities.

\widehat{R} =0.40(1.20)+0.60(1.15) =0.48+0.69 =1.17.

The credibility-weighted relativity is 1.17. As a second check, blend the class costs first:

0.40(\$216)+0.60(\$207) =\$86.40+\$124.20 =\$210.60, \qquad \frac{\$210.60}{\$180}=1.17.

The two routes agree because both relativities use the same $180 base. The common mistake is to blend a dollar loss cost directly with a relativity, for example $0.40(\$216)+0.60(1.15)$. That sum has no coherent unit and is not a valid estimate. Convert both inputs to the same type of quantity before blending.

Conditions and Practical Checks

A credibility factor does not make incomparable inputs comparable. Before calculating, define exactly what the estimate represents: for example, expected loss per policy-year for a specified coverage and period, or a class factor relative to a specified base. Confirm that the experience has been adjusted to the same target period and that its exposure denominator matches the complement's basis.

Name the target quantity.
Specify whether the blend is for dollars per exposure unit, a rate with a stated denominator, or a dimensionless relativity with a stated base.

Construct the experience estimate.
For a loss cost, divide the applicable losses by the matching exposure. Combine unequal exposure volumes through total losses divided by total exposure, not an unweighted average of period rates.

Align the complement.
Check its coverage, period, exposure definition, and units. Convert units where appropriate; do not treat a substantive exposure-basis mismatch as a mere conversion.

Apply credibility and verify.
Use $Z\widehat{\theta}_{\mathrm{exp}}+(1-Z)\widehat{\theta}_{\mathrm{comp}}$, then check units, arithmetic, and whether the result lies between the two inputs when $0\leq Z\leq1$.

The between-inputs check is a useful diagnostic, not evidence that the inputs are appropriate. A blend can fall between two numbers and still be invalid if one is per policy-year and the other per $1,000 of payroll. Likewise, a relativity with one base cannot be blended with a relativity using a different base without first reconciling the bases.

Key takeaway: Credibility weighting preserves the type of quantity being estimated. Blend loss costs with loss costs, rates with rates on the same exposure basis, and relativities defined against the same base. Retain unrounded values through the calculation and round only the final reported result.

Check Your Understanding

For each problem, identify the quantity and units before applying credibility. Round money to cents where needed.

A hypothetical benefit program has $90,000 of adjusted losses over 5,000 policy-years. The complement is $20 per policy-year and $Z=0.30$. Find the credibility-weighted loss cost. (Answer: $19.40 per policy-year)
A hypothetical liability class has $84,000 of losses on 4,200 units of $1,000 payroll exposure. Its complement is $22 per $1,000 of payroll and $Z=0.65$. Find the estimate per $1,000 and per $100 of payroll. (Answer: $20.70 per $1,000; $2.07 per $100)
A hypothetical class's experience loss cost is $132 per vehicle-year, its complement is $120 per vehicle-year, and $Z=0.25$. Find the credibility-weighted loss cost. (Answer: $123.00 per vehicle-year)
A hypothetical class has an experience loss cost of $198 per policy-year. Its complement relativity is 1.15 against a common base loss cost of $160 per policy-year. With $Z=0.40$, find the credibility-weighted class loss cost and its relativity to the $160 base. (Answer: $189.60 per policy-year; relativity 1.185)
A hypothetical portfolio has $48,000 of adjusted losses on 3,000 exposure units in one period and $70,000 on 4,000 units in another. The complement is $18 per exposure unit and $Z=0.50$. Calculate the combined experience loss cost using total losses divided by total exposure, then calculate the credibility-weighted estimate. (Answer: experience loss cost $16.8571 per exposure unit; weighted estimate $17.43 per exposure unit)

What You'll Learn