Stability of Credibility-Weighted Class Costs Over Time

Why Stability Over Time Needs a Careful Comparison

A credibility-weighted class cost is a blend of the class’s observed cost and a complement, such as a broader portfolio cost or a selected manual cost. Comparing that indication across experience periods can help identify whether a class’s indicated cost is changing. But a change in the indication does not, by itself, show that the class’s underlying experience has changed. The observed cost, complement, credibility weight, and even the definition of the measured cost may all differ by period.

The useful question is therefore not simply, “Did the indicated cost go up or down?” It is: which inputs changed, and does the class’s movement remain after putting the periods on a comparable basis? This tutorial develops a calculation framework for answering that question. The methods are aligned with the CAS Exam 8 (Advanced Ratemaking) syllabus.

Definition: For experience period $t$, let $O_t$ be the observed cost for a class, $C_t$ the complement applicable to that class, and $Z_t$ its credibility weight, with $0 \leq Z_t \leq 1$. The credibility-weighted class cost is $$ I_t = Z_t O_t + (1-Z_t)C_t. $$ Here $I_t$ is the indicated class cost for period $t$. The observed cost and complement must be expressed on compatible bases and in the same cost units before this blend is meaningful.

The formula is a weighted average: $Z_t$ is the share assigned to the class’s own observed cost, and $1-Z_t$ is the share assigned to the complement. The appropriate method for selecting $Z_t$ depends on the credibility procedure being used; in a time comparison, treat each period’s weight as a stated input or calculate it consistently under the selected procedure. A weight is not itself a measure of whether observed experience improved or worsened.

Make Periods Comparable Before Comparing Costs

A sound time comparison begins by checking what each cost represents. A class’s loss cost from one period should not be compared mechanically with a later cost if the two periods differ in coverage, exposure definition, claim treatment, or price level. For example, a later period may include a broader coverage, a different mix of insured risks, or losses valued at a more recent cost level. Those changes can move observed costs without representing a change in the class’s underlying relative experience.

A practical comparison should verify the following before interpreting a change:

Consistent class definition: the risks assigned to the class are defined the same way, or any definition change is identified.
Consistent cost measure: the observed cost and complement use the same loss or loss-cost basis, coverage, and treatment of expenses or adjustments.
Consistent exposure basis: costs are calculated using comparable exposure units and periods.
Common cost level: where needed, costs from different periods are adjusted to a common level so that general price or trend changes are not mistaken for class-specific movement.
Explicit weight and complement: record whether either changed. A different complement or weight changes the indicated cost even if the class’s observed cost does not change.

These checks do not prove that any remaining movement is persistent. They make the arithmetic comparison interpretable. A few successive periods can suggest that a pattern deserves attention, but the comparison alone does not establish that a change is statistically significant or will continue.

Separating the Sources of an Indication Change

When the weight is the same in both periods, the change in indicated cost is a weighted combination of the change in observed class cost and the change in the complement:

$$ I_2-I_1 = Z(O_2-O_1)+(1-Z)(C_2-C_1). $$

This identity is most useful when periods have been placed on a consistent basis and the same credibility weight $Z$ is being applied. It shows why the change in the observed cost is not generally the change in the indicated cost: only the fraction weighted by $Z$ passes through from observed experience, while the complement movement contributes with weight $1-Z$.

Derivation: With the same $Z$ in both periods, write $I_2=ZO_2+(1-Z)C_2$ and $I_1=ZO_1+(1-Z)C_1$. Subtract the second equation from the first and collect like terms: $$ \begin{aligned} I_2-I_1 &= ZO_2+(1-Z)C_2-ZO_1-(1-Z)C_1\\ &= Z(O_2-O_1)+(1-Z)(C_2-C_1). \end{aligned} $$ Thus the indication change is the sum of the weighted observed-cost movement and the weighted complement movement. If the complement is unchanged, the indication changes by $Z$ times the observed-cost change.

If the credibility weight changes, applying the fixed-weight identity alone omits an effect. An exact decomposition, using period 1’s weight for the cost movements, is:

$$ I_2-I_1 = Z_1(O_2-O_1)+(1-Z_1)(C_2-C_1) +(Z_2-Z_1)(O_2-C_2). $$

The final term is the effect of changing the weight, evaluated using the period 2 difference between observed cost and complement. This decomposition is an accounting identity, not a separate credibility model. It helps describe the arithmetic; interpretation still depends on whether the weights, complements, and cost bases were selected consistently.

Worked Comparisons

Worked Example: Fixed Weight and a Noisy Observed Cost

A hypothetical class has the same complement and credibility weight in two comparable annual experience periods. In period 1, the observed cost is $1,100 per exposure, the complement is $1,000, and $Z=0.40$. In period 2, the observed cost is $900 per exposure. Calculate both indications and the change. Explain why using the full observed-cost change as the indicated-cost change is incorrect.

Method. Apply the same credibility-weighted blend in each period. Because the complement and $Z$ are fixed, the indication’s change must equal $Z$ times the observed-cost change.

Period 1. The complement weight is $1-0.40=0.60$. Therefore:

I_1=0.40(1{,}100)+0.60(1{,}000)=440+600=\$1{,}040.

Period 2. The complement and weight are unchanged:

I_2=0.40(900)+0.60(1{,}000)=360+600=\$960.

The observed cost fell by $900-1{,}100=-\$200$ per exposure. The indicated cost fell by $960-1{,}040=-\$80$ per exposure. The decomposition checks the result:

I_2-I_1=0.40(-200)+0.60(0)=-\$80.

The common mistake is to report a $200 reduction in the indication because the observed cost fell by $200. That calculation ignores the 60% complement weight. The corrected indicated reduction is $80. The result describes how the selected blend changes; it does not, by itself, determine whether the observed difference is ordinary sampling fluctuation.

Worked Example: Changing Observed Cost, Complement, and Weight

A hypothetical rating class is evaluated in two comparable periods. A selected credibility schedule is $Z=n/(n+300)$, where $n$ is the period’s exposure count. In period 1, $n_1=100$, observed cost is $1,200, and complement is $1,000. In period 2, $n_2=300$, observed cost is $1,100, and complement is $1,050. Calculate both indications and decompose their difference. Treat the schedule as specified for this example, not as a universal credibility formula.

Method. Calculate each period’s weight from the stated schedule, then calculate each blend. Since the weights differ, use the exact decomposition with the period 1 weight for the cost movements.

The weights are:

Z_1=\frac{100}{100+300}=0.25, \qquad Z_2=\frac{300}{300+300}=0.50.

The period 1 indication is:

I_1=0.25(1{,}200)+0.75(1{,}000)=300+750=\$1{,}050.

The period 2 indication is:

I_2=0.50(1{,}100)+0.50(1{,}050)=550+525=\$1{,}075.

Thus the indication increased by $1{,}075-1{,}050=\$25$, even though the observed cost decreased by $100. To identify the sources of the $25 increase, calculate the three terms:

\begin{aligned} Z_1(O_2-O_1)&=0.25(1{,}100-1{,}200)=-\$25,\\ (1-Z_1)(C_2-C_1)&=0.75(1{,}050-1{,}000)=\$37.50,\\ (Z_2-Z_1)(O_2-C_2)&=(0.50-0.25)(1{,}100-1{,}050)=\$12.50. \end{aligned}

The terms sum to $-25+37.50+12.50=\$25$, matching the direct calculation. The observed-cost movement reduced the indication by $25; the increased complement added $37.50; and the higher weight added $12.50 because period 2’s observed cost exceeded its complement by $50. The common mistake is to interpret the $25 increase as evidence that the class’s own observed cost rose. It did not; the complement and weight changes more than offset the observed-cost decrease.

Worked Example: Separating a Shared Cost-Level Movement From a Class Gap

A hypothetical class is compared over three periods, with all costs expressed per exposure on a consistent basis. The credibility weight is $Z=0.40$ in each period. The observed costs and complements are: period 0, $1,200 and $1,000; period 1, $1,450 and $1,100; period 2, $1,470 and $1,120. Calculate the indications, then decide which movement is class-specific relative to the complement.

Method. Calculate the blend and the observed-cost excess $O_t-C_t$ for each period. The excess measures the class’s distance from its contemporaneous complement in dollars per exposure; it is not itself the indicated cost.

Period 0:

I_0=0.40(1{,}200)+0.60(1{,}000)=480+600=\$1{,}080, \qquad O_0-C_0=\$200.

Period 1:

I_1=0.40(1{,}450)+0.60(1{,}100)=580+660=\$1{,}240, \qquad O_1-C_1=\$350.

Period 2:

I_2=0.40(1{,}470)+0.60(1{,}120)=588+672=\$1{,}260, \qquad O_2-C_2=\$350.

From period 0 to period 1, the complement increased by $100 and the observed cost increased by $250. The indication increased by $160:

0.40(250)+0.60(100)=100+60=\$160.

From period 1 to period 2, both observed cost and complement increased by $20, while the observed-cost excess stayed at $350. Accordingly, the indication increased by $20:

0.40(20)+0.60(20)=\$20.

The class’s excess over the complement rose from $200 to $350 between periods 0 and 1, then remained at $350 in periods 1 and 2. The later $20 increase in the indication is shared with the complement, not a further widening of the class’s relative gap. The common mistake is to treat every increase in indicated dollars as evidence of a worsening class differential. These calculations show a persistent elevated gap in the two later observations, but they do not establish that the pattern will continue or that it is statistically significant.

Common Mistakes in Time Comparisons

Comparing raw observations instead of indications. The observed-cost change is not the indicated-cost change unless $Z=1$. Apply the blend in each period.
Ignoring a changed complement. If the complement moves, part of the indication can move even when the observed class cost is unchanged.
Holding the weight fixed in the arithmetic when it changed. Use each period’s actual weight and account for the weight effect, or the comparison will not reconcile to the calculated indications.
Comparing unlike cost bases. A change in coverage, exposure, or cost level can create an apparent movement unrelated to a change in class experience. Check comparability before interpreting the result.
Calling a short sequence proof of persistence. Repeated elevated observations can motivate review, but a time comparison alone does not show that the movement is statistically distinguishable from fluctuation.

Key takeaway: For each period, calculate the indication from its own observed cost, complement, and credibility weight. Then reconcile the change to those inputs. A stable relative class gap, a changing complement, and a changing weight are different explanations and should not be conflated.

Check Your Understanding

For each problem, calculate the requested indications and interpret the change using the stated inputs.

A class has $Z=0.30$ and complement $1,200 in both comparable periods. Its observed costs are $1,500 and $1,200. Calculate both indications and the indicated-cost change. (Answer: $1,290 and $1,200; the change is $-\$90$ per exposure.)
A specified schedule is $Z=n/(n+200)$. Period 1 has $n=100$, observed cost $900, and complement $1,000. Period 2 has $n=200$, observed cost $1,100, and complement $1,050. Calculate both indications and their difference. (Answer: $Z_1=1/3$, $Z_2=0.50$; $I_1=\$966.67$, $I_2=\$1,075.00$; increase $=\$108.33$ per exposure.)
With $Z=0.50$ in both periods, observed cost rises from $1,000 to $1,080 and complement rises from $800 to $880. Calculate both indications and compare the class’s excess over the complement. (Answer: indications are $900 and $980, an $80 increase; the excess is $200 in both periods, so the class gap did not widen.)
A class has fixed $Z=0.25$ and complement $1,000 in each of three comparable periods. Observed costs are $1,600, $1,400, and $1,500. Calculate the three indications and the successive changes. (Answer: $1,150, $1,100, and $1,125; changes are $-\$50$ and $+\$25$, which do not show a steadily increasing indication.)
Period 1 has $Z_1=0.20$, observed cost $1,400, and complement $1,000. Period 2 has $Z_2=0.60$, observed cost $1,300, and complement $1,100. Calculate the indications and verify the three-part decomposition of the change. (Answer: $I_1=\$1,080$, $I_2=\$1,220$, increase $=\$140$; the terms are $-\$20$ for observed cost, $+\$80$ for the complement, and $+\$80$ for the weight change.)

What You'll Learn