Selecting a Collective Complement for a Class

Why the Complement Matters

A credibility-weighted class indication combines the class’s own experience with a complement: a benchmark representing the cost expected when class experience is incomplete or noisy. The credibility weight determines how much the observed class experience contributes. The complement determines what fills the remaining share. Two analysts can use the same class experience and the same credibility weight yet obtain different indications if they select different complements.

Common choices include the full portfolio average, the average for a broader group of comparable classes, and a prior indicated cost. These are not interchangeable by definition. A full-portfolio average reflects the overall portfolio mix; a peer-group average emphasizes a selected segment; a prior cost expresses an expected level based on an earlier rating view, adjusted as appropriate to the current basis. The choice should fit the class being rated and the purpose of the indication.

Definition: Let $L_c$ be the observed loss cost for class $c$, $Z_c$ its credibility weight, and $C_c$ the selected complement, all expressed on the same per-exposure and rating basis. The credibility-weighted class indication is $$ I_c=Z_cL_c+(1-Z_c)C_c,\qquad 0\leq Z_c\leq 1. $$ The complement receives the share $1-Z_c$. This formula treats $Z_c$ as given; it does not determine which complement is appropriate.

When comparing candidate complements, hold the class experience and the credibility weight fixed. That isolates the effect of the complement rather than mixing two questions: how much weight to assign to experience, and what benchmark to use for the remaining weight. The credibility framework does not, by itself, guarantee that every plausible benchmark is equally suitable.

Three Common Complement Choices

A portfolio average summarizes a broader book of business. If classes have different exposure volumes, calculate it as an exposure-weighted average rather than as the unweighted mean of class loss costs. For class loss costs $L_j$ and exposures $E_j$, the portfolio average is

C_P=\frac{\sum_j E_jL_j}{\sum_j E_j}.

A broader grouping average is calculated over a selected set of classes, such as classes in a common territory or product segment. It can be useful when the full portfolio includes business unlike the target class. The selected group and its exposure basis should be stated. In particular, say whether the target class is included. If it is included, its own experience contributes to the complement as well as to the class estimate, so the benchmark is not independent of that experience.

A prior indicated cost is a prospective benchmark established before considering the current class experience in this blend. It may need adjustment to match the period, coverage, exposure definition, and rate level of the class indication. For example, a prior cost expressed at an earlier rating level may require a rate-level adjustment; an earlier-period cost may require a trend adjustment. Apply only adjustments justified by the stated assumptions and apply them consistently.

Complement	What it represents	Question to check
Portfolio average	The overall book’s exposure-weighted cost level	Does the portfolio mix represent the target class appropriately?
Broader grouping average	The cost level for a defined set of related classes	Are the grouped classes sufficiently comparable, and is the target included?
Prior indicated cost	An existing prospective estimate or rate level	Is it on the same period, coverage, and rating basis?

A complement is not selected merely because it produces a preferred result. The analyst should identify the benchmark’s population, its exposure weighting, and its basis, then explain why it is relevant. If the available data do not support a precise peer-group average, a broader benchmark may be more defensible than a narrowly defined but unstable one.

How Complement Choice Changes the Indication

With $L_c$ and $Z_c$ fixed, replacing complement $C_1$ with $C_2$ changes the indication by

$$ I_c(C_1)-I_c(C_2)=(1-Z_c)(C_1-C_2). $$

This result follows directly by subtracting the two credibility blends:

\begin{aligned} I_c(C_1)-I_c(C_2) &=\bigl[Z_cL_c+(1-Z_c)C_1\bigr] -\bigl[Z_cL_c+(1-Z_c)C_2\bigr]\\ &=Z_cL_c-Z_cL_c+(1-Z_c)C_1-(1-Z_c)C_2\\ &=(1-Z_c)(C_1-C_2). \end{aligned}

The class-experience terms cancel. Thus, if one complement is $10 per exposure higher than another and $Z_c=0.30$, the indication using the higher complement is $0.70(10)=$7$ higher. If $Z_c=1$, complement choice has no effect; if $Z_c=0$, the indication equals the complement. This comparison is exact when the same class experience and credibility weight are used for both calculations.

Worked Example: Comparing a Portfolio Average With a Peer-Group Average

A hypothetical class has an observed loss cost of $140 per exposure and a credibility weight of $0.40$. Its portfolio contains three classes: the target class has 200 exposures at $140 per exposure, another class has 300 exposures at $100, and a third class has 500 exposures at $120. For comparison, the latter two classes form a broader peer group. Calculate the indication using both the full portfolio and the peer group as complements.

Method. First calculate each exposure-weighted benchmark. Then blend each benchmark with the target class cost using the same credibility weight. The portfolio benchmark includes the target class; the peer-group benchmark excludes it.

Full-portfolio complement. The total loss cost amount represented by the three classes is $200(140)=$28{,}000$, $300(100)=$30{,}000$, and $500(120)=$60{,}000$. Their sum is $$28{,}000+$30{,}000+$60{,}000=$118{,}000$, over $200+300+500=1{,}000$ exposures. Therefore

C_P=\frac{118{,}000}{1{,}000}=$118.

The indication using this complement is $0.40($140)+0.60($118)=$56+$70.80=$126.80$ per exposure.

Peer-group complement. The other two classes have a combined loss cost amount of $$30{,}000+$60{,}000=$90{,}000$ over $300+500=800$ exposures. Thus

C_G=\frac{90{,}000}{800}=$112.50.

The indication using the peer-group complement is $0.40($140)+0.60($112.50)=$56+$67.50=$123.50$ per exposure.

Check and interpretation. The portfolio complement exceeds the peer-group complement by $5.50, so its indication should be higher by $0.60($5.50)=$3.30. Indeed, \($126.80-$123.50=$3.30$. The portfolio average is not automatically the better choice: it includes the target class and reflects the full portfolio mix. The peer group may be more relevant if those classes are the intended comparison set. A common mistake is to average the three class costs as $(140+100+120)/3$; that gives $120, incorrectly giving the 200-exposure class the same weight as the 500-exposure class.

Putting a Prior Indicated Cost on the Right Basis

A prior indicated cost can be a useful complement when it is the most relevant prospective benchmark, but the comparison is meaningful only if it is expressed on a consistent basis. Check the units (for example, dollars per exposure), coverage, period, and rating level. If the problem specifies an adjustment, apply it to the prior cost before blending. Do not adjust the observed class experience unless the problem also calls for that adjustment.

Worked Example: Comparing Three Complements After an Adjustment

A hypothetical class has observed loss cost $L_c=$130$ per exposure and credibility $Z_c=0.25$. Its portfolio average is $95 and a defined broader group has an average of $108, both on the current basis. A prior indicated cost is $102 on an earlier basis and is to be increased by 4% to match the current basis. Calculate all three indications and compare them.

Method. Adjust the prior complement first, then apply the same credibility blend to each candidate. The prior-cost adjustment is multiplicative:

$$ C_0^*=$102(1.04)=$106.08. $$

The share assigned to each complement is $1-Z_c=1-0.25=0.75$, and the class-experience contribution is $Z_cL_c=0.25($130)=$32.50$.

Portfolio complement: $I_c=0.25($130)+0.75($95)=$32.50+$71.25=$103.75$ per exposure.

Broader-group complement: $I_c=0.25($130)+0.75($108)=$32.50+$81.00=$113.50$ per exposure.

Adjusted prior complement: $I_c=0.25($130)+0.75($106.08)=$32.50+$79.56=$112.06$ per exposure.

Check and interpretation. The group indication exceeds the prior-based indication by $1.44, which agrees with $0.75($108-$106.08)=0.75($1.92)=$1.44$. The portfolio indication is lower because its complement, $95, is lower. If the stated objective is to use a benchmark formed from comparable classes, the broader-group indication is the relevant one here: $113.50 per exposure. That choice follows from the objective, not from the numerical size of the indication.

Common arithmetic trap. For the adjusted prior complement, the correct weighted terms are $32.50 and $79.56. Their sum is $112.06, not $112.81. Another error is to blend with the unadjusted $102; that would give $0.25($130)+0.75($102)=$109.00$, which fails to put the prior complement on the required current basis.

Choosing and Documenting a Complement

The selection is a judgment about relevance and consistency, not a mechanical consequence of the credibility formula. A defensible choice answers several questions:

Is the population relevant? A broad portfolio average may be less representative if its business mix differs materially from the target class. A peer-group benchmark is useful only if the group is defined on a meaningful basis.
Is the measure comparable? Confirm that the complement and class experience use compatible exposure definitions, coverage, loss-cost basis, and period. Adjust a prior cost when the stated assumptions require it.
Is the benchmark constructed correctly? Use exposure weighting for averages of class loss costs when exposures differ. State whether the target class contributes to a portfolio or group average.
Does the choice match the objective? A complement intended to represent a stable overall portfolio level may differ from one intended to represent similar neighboring classes or an existing prospective cost level.

A broader group that includes the target class may create a partial self-inclusion effect: some of the target’s own experience appears in $C_c$, then the blend also assigns it weight through $Z_cL_c$. This is not automatically invalid, but it changes the interpretation of the calculation. If the intention is to use experience from other classes as the complement, exclude the target class and calculate the group average from the remaining classes.

The comparison formula also helps explain when complement selection deserves particular attention. When $Z_c$ is small, $1-Z_c$ is large, so the indication is more sensitive to the complement. When $Z_c$ is large, the class’s own experience dominates and reasonable differences among complements have a smaller impact. This observation does not prescribe a new credibility weight; it shows how a fixed weight transmits differences between candidate benchmarks into the indication.

Key takeaway: State the complement’s population and basis, calculate it with the appropriate exposure weights, and use the complement share $1-Z_c$ in the blend. Compare alternatives with the credibility weight held fixed, and choose the benchmark that best represents the intended prospective comparison.

Check Your Understanding

For each problem, use the stated credibility weight and basis. Answers are rounded only where a cent is needed.

A class has an observed loss cost of $126 per exposure and credibility $0.20$. Calculate its indication using a portfolio complement of $100 and a peer-group complement of $115. (Answer: portfolio $105.20; peer group $117.20 per exposure)
A target class has observed loss cost $130 and credibility $0.30$. A peer group has two classes: 80 exposures at $90 and 120 exposures at $110. Find the peer-group complement and the target class indication. (Answer: complement $102; indication $110.40 per exposure)
A class has observed loss cost $120 and credibility $0.40$. Its prior complement is $96 and must be increased by 5% in each of two successive annual periods. Find the adjusted complement and the indication. (Answer: adjusted complement $105.84; indication $111.50 per exposure)
A class has observed loss cost $140 and credibility $0.25$. Compare indications using a portfolio complement of $100 and a broader-group complement of $124. By how much does the second indication exceed the first? (Answer: portfolio $110; broader group $128; difference $18 per exposure)
A class has observed loss cost $85 and credibility $0.50$. Candidate complements are a portfolio average of $110, a peer-group average of $95, and an adjusted prior cost of $100. Calculate the three indications and identify the lowest. (Answer: portfolio $97.50; peer group $90.00; adjusted prior $92.50; the peer-group indication is lowest)

What You'll Learn