Deriving a Credibility-Weighted Modification Formula

Why Blend Actual and Expected Losses?

An experience modification compares a risk’s observed losses with the losses expected for its exposure. The observed amount may be informative but volatile, especially when experience is limited. A credibility-weighted modification allows a supplied weight to determine how much the actual experience influences the result, while the remaining weight is assigned to the expected result.

The calculation is meaningful only when actual and expected losses refer to the same exposure, experience period, valuation date, coverage, and loss basis. For example, a comparison is not coherent if actual losses include amounts above a limit while expected losses are limited to that amount. The earlier material on reconciling actual and expected loss bases addresses that consistency requirement. Here, assume the amounts have already been put on a consistent basis.

Definition: Let $A$ be actual losses on the selected basis, $E$ be expected losses on that same basis, and $Z$ be a supplied credibility weight with $0\leq Z\leq 1$. When $E>0$, the credibility-weighted loss amount is $Z A+(1-Z)E$. The credibility-weighted modification is this blended amount divided by $E$: $$ M=\frac{ZA+(1-Z)E}{E} $$ The weight $Z$ applies to actual losses; $1-Z$ applies to expected losses.

The modification is dimensionless. A value of $1.00$ indicates that the blended loss amount equals expected losses; a value above $1.00$ indicates a higher blended amount, and a value below $1.00$ indicates a lower blended amount. A modification is not itself a premium: it is a factor applied to an appropriate premium or expected-cost basis under the applicable rating plan.

Deriving the Modification Formula

The formula follows from two operations in a specific order: first blend the loss amounts, then compare that blended amount with expected losses. Dividing by $E$ expresses the blended amount relative to the expected baseline.

Assign credibility shares.
The actual amount receives weight $Z$, and the expected amount receives the remaining weight $1-Z$. The weighted loss amount is therefore $ZA+(1-Z)E$.

Express the blend relative to expected losses.
Divide the weighted loss amount by $E$, which must be positive for this ratio to be defined.

Simplify.
Separate the two terms in the numerator and cancel $E$ in the expected-loss term.

\begin{aligned} M &=\frac{ZA+(1-Z)E}{E}\\ &=Z\frac{A}{E}+(1-Z)\frac{E}{E}\\ &=Z\frac{A}{E}+(1-Z)\\ &=1+Z\left(\frac{A}{E}-1\right). \end{aligned}

This equivalent form shows how the supplied weight acts on the actual-to-expected ratio. The unmodified expected result corresponds to a ratio of $1$. The credibility-weighted modification starts at $1$ and moves a fraction $Z$ of the way from that baseline toward $A/E$. This is a consequence of the algebra, not an additional adjustment.

The endpoints provide a direct check on both the derivation and any calculation. At zero credibility, the actual amount receives no weight; at full credibility, the expected amount receives no weight.

\begin{aligned} Z=0 &: \quad M=\frac{E}{E}=1,\\ Z=1 &: \quad M=\frac{A}{E}. \end{aligned}

Thus, “the expected result at zero credibility” means a modification of $1$ relative to expected losses, not a dollar modification equal to $E$. At full credibility, the modification is the actual-to-expected ratio. If $A=E$, the modification is $1$ for every value of $Z$. If $A>E$, increasing $Z$ increases $M$; if $A<E$, increasing $Z$ decreases $M$. In either case, for $0\leq Z\leq1$, $M$ lies between $1$ and $A/E$.

Worked Calculations

Worked Example: A Partial-Credibility Modification

A hypothetical commercial account has actual losses of $92,000 and expected losses of $80,000 on the same coverage, period, and valuation basis. The supplied credibility weight is $Z=0.25$. Find the weighted loss amount and modification, then check both endpoints. The inputs are fixed aggregate amounts; no interest, mortality, or probability-distribution assumption is used.

Method and parameters. Use $A=\$92{,}000$, $E=\$80{,}000$, and $Z=0.25$. The remaining expected-loss weight is $1-Z=1-0.25=0.75$.

Blend the amounts. The credibility-weighted actual amount is $0.25(\$92{,}000)=\$23{,}000$. The expected component is $0.75(\$80{,}000)=\$60{,}000$. Therefore, the weighted loss amount is $\$23{,}000+\$60{,}000=\$83{,}000$.

Divide by expected losses. The modification is $\$83{,}000/\$80{,}000=1.0375$. Equivalently, $A/E=\$92{,}000/\$80{,}000=1.15$, so $M=1+0.25(1.15-1)=1+0.25(0.15)=1.0375$.

Check the endpoints. At $Z=0$, $M=1$. At $Z=1$, $M=A/E=1.15$. The calculated $1.0375$ lies between those values, as it should for a weight of $0.25$. The modification is 1.0375, or 1.038 to three decimal places if a three-decimal presentation is required.

Method and mistake avoided. The calculation blends dollar amounts and then divides by expected losses. Using $Z(A/E)$ alone would give $0.25(1.15)=0.2875$, incorrectly omitting the $0.75$ expected component.

Worked Example: Credibility Applied to a Below-Expected Result

A hypothetical fleet program has actual losses of $216,000 against expected losses of $240,000. The supplied credibility weight is 75%. Calculate the weighted loss amount and modification. Confirm that the result is closer to the expected baseline than the full-credibility result. Treat the supplied amounts as fixed aggregates; no interest, mortality, or probability-distribution assumption is used.

Method and parameters. Here $A=\$216{,}000$, $E=\$240{,}000$, and $Z=75\%=0.75$, not 75. The expected-loss weight is $1-0.75=0.25$.

Blend the amounts. The actual component is $0.75(\$216{,}000)=\$162{,}000$. The expected component is $0.25(\$240{,}000)=\$60{,}000$. Their sum is $\$162{,}000+\$60{,}000=\$222{,}000$.

Calculate the modification. Divide the blended amount by expected losses: $\$222{,}000/\$240{,}000=0.925$. As a check using the ratio form, $A/E=\$216{,}000/\$240{,}000=0.9$, so $M=1+0.75(0.9-1)=1+0.75(-0.1)=0.925$.

Check the direction and endpoints. At zero credibility, $M=1$; at full credibility, $M=0.9$. The result $0.925$ is between $0.9$ and $1$, and it is closer to $0.9$ because the credibility weight is 0.75. The modification is 0.925, or 0.925 to three decimal places.

Method and mistake avoided. Convert the percentage to a decimal before multiplying. Treating 75% as $75$ would not be a valid credibility weight and would produce a result outside the range between the expected and actual endpoints.

Worked Example: Aggregate Loss Components Before Applying One Weight

A hypothetical account has two loss components on a common basis. Expected primary losses are $65,000 and expected excess losses are $35,000. Actual primary losses are $52,000 and actual excess losses are $58,000. The plan supplies one overall credibility weight of $Z=0.40$. Find the aggregate modification using that single weight. No interest, mortality, or probability-distribution assumption is used; the component amounts are fixed aggregates.

Method and parameters. First aggregate the two components: $E=\$65{,}000+\$35{,}000=\$100{,}000$, and $A=\$52{,}000+\$58{,}000=\$110{,}000$. Use $Z=0.40$, so the expected-loss weight is $1-0.40=0.60$. The given weight applies to the total in this problem.

Blend the aggregate amounts. The actual component is $0.40(\$110{,}000)=\$44{,}000$. The expected component is $0.60(\$100{,}000)=\$60{,}000$. The weighted loss amount is $\$44{,}000+\$60{,}000=\$104{,}000$.

Calculate and check. The modification is $\$104{,}000/\$100{,}000=1.04$. The aggregate actual-to-expected ratio is $\$110{,}000/\$100{,}000=1.10$; the equivalent formula gives $1+0.40(1.10-1)=1+0.04=1.04$. At $Z=0$, the modification would be $1$, and at $Z=1$, it would be $1.10$. The result is between those endpoints. To three decimal places, the modification is 1.040.

Method and mistake avoided. This problem specifies one overall weight, so aggregate the actual and expected amounts before applying it. Applying separate weights to components would change the stated method and could produce a different answer.

Common Exam Traps and Practical Checks

The most frequent error is to multiply the actual-to-expected ratio by $Z$ and stop. That gives $Z(A/E)$, which is only the actual component of the weighted ratio. The expected component, $1-Z$, must also be included:

\underbrace{Z\frac{A}{E}}_{\text{actual component}} + \underbrace{(1-Z)}_{\text{expected component}} = M.

For example, if $A/E=1.15$ and $Z=0.25$, the incorrect calculation $0.25(1.15)=0.2875$ fails the zero-credibility check: it would imply a modification of zero when $Z=0$, rather than the expected-baseline modification of one. The correction is $0.25(1.15)+0.75=1.0375$.

Check the complement. The weights must add to one: $Z+(1-Z)=1$. Do not assign $Z$ to both actual and expected losses.
Keep the basis consistent. Confirm that actual and expected amounts match in period, coverage, valuation, limits, and units before taking their ratio.
Keep the denominator clear. The modification is relative to expected losses, so divide by $E$, not by the blended amount or actual losses.
Check the range. For $0\leq Z\leq1$ and $E>0$, the result must lie between $1$ and $A/E$. If it does not, check the weight, complement, and arithmetic.

The usual ratio formula requires $E>0$. If expected losses are zero, division by $E$ is undefined; the ordinary modification cannot be calculated with this formula. Do not insert a small denominator without an applicable plan rule. That situation requires separate treatment under the relevant rating procedure.

Key takeaway: Blend actual and expected losses using complementary weights, then divide by expected losses. The result $M=1+Z(A/E-1)$ makes the endpoints explicit: zero credibility gives $1$, and full credibility gives $A/E$.

Check Your Understanding

For each problem, assume actual and expected losses are on the same basis. Unless stated otherwise, the amounts are fixed aggregates, with no interest, mortality, or probability-distribution assumption.

A hypothetical contractor has actual losses of $48,000, expected losses of $60,000, and $Z=0.30$. Find the weighted loss amount and modification. (Answer: $56,400; $M=0.940$)
A hypothetical risk has $A/E=1.24$ and credibility of 60%. Find the modification using the ratio form, and state the zero- and full-credibility endpoint modifications. (Answer: $M=1.144$; endpoints $1$ and $1.24$)
A hypothetical program has expected losses of $150,000 and a weighted loss amount of $159,000. If $Z=0.30$, find the actual losses and the modification. (Answer: $A=\$180{,}000$; $M=1.060$)
A hypothetical account has expected losses of $90,000 and actual losses of $81,000. A calculation reports $M=0.20$ at $Z=0.20$, using only $Z(A/E)$. Correct the calculation and identify the omitted component. (Answer: $M=0.980$; the omitted expected component is $1-Z=0.80$)
A hypothetical account’s actual primary and excess losses are $44,000 and $36,000; expected primary and excess losses are $50,000 and $30,000. One overall weight $Z=0.50$ applies. Find the aggregate weighted loss amount and modification. (Answer: $80,000; $M=1.000$)

What You'll Learn