Class Volume Shifts and the Effect of Recalculated Weights

Why Class Volume Shifts Matter

A classification indication can change even when the observed loss cost for every class is held constant. If a class has more exposure, its experience may receive more credibility; if the portfolio’s exposure mix changes, that class also contributes a different share to the portfolio indication. These are separate effects, and a sound calculation updates both.

The calculations below use a simple exposure-based credibility rule. It is an illustrative model, not a universal prescription: an actuary’s selected credibility procedure should reflect the data, assumptions, and applicable ratemaking framework. For clarity, all costs are dollars per exposure unit, and all exposures are comparable units. When a scenario says a class cost remains fixed as volume changes, its loss amount is understood to change proportionally with exposure.

Definition: For class $i$, let $n_i$ be its exposure volume, $x_i$ its observed loss cost per exposure unit, and $m_i$ its fixed collective complement. Under the exposure-based rule used here, with common positive constant $K$, $$ Z_i=\frac{n_i}{n_i+K},\qquad \hat{x}_i=Z_i x_i+(1-Z_i)m_i. $$ The indicated class loss cost is $\hat{x}_i$. Given a set of class indications, the portfolio indication at the current exposure mix is $$ \hat{x}_{\mathrm{port}}=\frac{\sum_i n_i\hat{x}_i}{\sum_i n_i}. $$

The credibility factor $Z_i$ answers how much weight this rule gives to the class’s own experience relative to its complement. The factor $n_i/\sum_j n_j$ answers how much of the portfolio exposure is in class $i$. Do not confuse them: one appears inside a class indication, while the other is used to aggregate class indications.

Why the Exposure-Based Credibility Rule Has This Form

A standard justification for this form comes from a simple random-effects model. Suppose a class has an unknown underlying cost $\theta$, with mean $m$ and variance $\tau^2$ across comparable classes. Suppose its observed cost $x$, conditional on $\theta$, is unbiased and has variance $\sigma^2/n$. Here $\sigma^2$ is the process variance per exposure unit, and exposures are assumed comparable and independent for this variance relationship.

Consider an estimate $a x+(1-a)m$, where $a$ is a weight between zero and one. Its mean squared error for estimating $\theta$ is the sum of the conditional sampling variance and the squared-complement contribution:

\operatorname{MSE}=a^2\frac{\sigma^2}{n}+(1-a)^2\tau^2.

Differentiating with respect to $a$, setting the derivative equal to zero, and solving gives

2a\frac{\sigma^2}{n}-2(1-a)\tau^2=0 \quad\Longrightarrow\quad a=\frac{\tau^2}{\tau^2+\sigma^2/n} =\frac{n}{n+\sigma^2/\tau^2}.

Thus the rule $Z=n/(n+K)$ follows when $K=\sigma^2/\tau^2$. More exposure reduces the sampling variance and increases the weight on observed experience. This derivation depends on the stated model and variance assumptions; it does not establish that the same $K$ is appropriate for every class or data set.

Recalculate the Class Factor and the Portfolio Mix

When volume changes, first calculate each affected $Z_i$ using the new $n_i$. Then calculate each class indication, and only then aggregate using the new exposure shares. This order makes it possible to see whether a portfolio movement comes from changed class credibility, changed portfolio mix, or both.

Record the fixed inputs.
Identify the observed class costs, collective complement, and credibility constant. Keep the units consistent.

Update credibility.
Use the revised exposure volume in $Z_i=n_i/(n_i+K)$ for every class whose volume changes.

Recalculate class indications.
Blend each class’s observed cost with its complement using its updated credibility factor.

Aggregate at the revised mix.
Weight the new class indications by revised exposures, not by old exposure shares.

Worked Example: Tripling One Class’s Exposure

A hypothetical two-class portfolio uses $K=200$ exposure units and a fixed collective complement of $1,000 per unit for both classes. Class A has observed cost $1,400 and increases from 100 to 300 exposure units. Class B remains at 100 units with observed cost $800. Calculate both portfolio indications, before and after the shift. Assume the observed class costs remain fixed.

Method. Recalculate the credibility factor for A at the new volume, retain B’s unchanged factor, and aggregate each set of class indications at its corresponding exposure mix.

Before the shift. For A, $Z_A=100/(100+200)=1/3$. Its indication is $ (1/3)(\$1{,}400)+(2/3)(\$1{,}000)=\$3{,}400/3=\$1{,}133.333333\ldots$. For B, $Z_B=100/(100+200)=1/3$, so its indication is $ (1/3)(\$800)+(2/3)(\$1{,}000)=\$2{,}800/3=\$933.333333\ldots$. The old portfolio indication, with 100 units in each class, is

\frac{100(3400/3)+100(2800/3)}{200} =\frac{6200}{6} =\$1{,}033.333333\ldots.

After the shift. Class A now has $Z_A=300/(300+200)=3/5$. Its new indication is $ (3/5)(\$1{,}400)+(2/5)(\$1{,}000)=\$840+\$400=\$1{,}240$. Class B is unchanged at $Z_B=1/3$, so its indication remains exactly $\$2{,}800/3=\$933.333333\ldots$. With 300 units in A and 100 in B, the new portfolio indication is

\frac{300(1240)+100(2800/3)}{400} =\frac{465{,}333.333333\ldots}{400} =\$1{,}163.333333\ldots.

Check. The new exposure shares are $300/400=0.75$ for A and $100/400=0.25$ for B. Weighting the indications directly gives $0.75(\$1{,}240)+0.25(\$933.333333\ldots)=\$1{,}163.333333\ldots$, as above. Rounded to the nearest cent, the old and new portfolio indications are $1,033.33 and $1,163.33.

Common mistake to avoid. Keeping A’s old $Z_A=1/3$ would leave its indication at $1,133.333333… and produce a new portfolio indication of $1,083.333333…. That calculation uses the wrong credibility factor for A’s new volume. The corrected factor is $3/5$; the exact new portfolio indication is $1,163.333333…, not a sum of prematurely rounded class values.

Separating Credibility Changes From Mix Changes

A volume change need not alter both components equally. If all class exposures are multiplied by the same factor, the exposure shares stay constant, but each class’s credibility factor changes under this rule. If exposure is shifted from one class to another while total exposure stays constant, both the individual factors and the portfolio shares can change. The next examples isolate these patterns.

Worked Example: Redistributing a Fixed Total of Exposure

A hypothetical portfolio has classes A and B, each with complement $1,000 and $K=200$. Their observed costs are $1,500 and $700, respectively. Exposure changes from A = 100 and B = 300 to A = 300 and B = 100. The total remains 400 units. Find both portfolio indications, assuming the two observed class costs remain fixed.

Method. Compute both class factors and indications at each allocation; do not treat a fixed total as a fixed class mix.

Initially, $Z_A=100/300=1/3$, so $\hat{x}_A=(1/3)(1500)+(2/3)(1000)=\$1{,}166.666666\ldots$. For B, $Z_B=300/500=3/5$, so $\hat{x}_B=(3/5)(700)+(2/5)(1000)=\$820$. The initial portfolio indication is

\frac{100(3500/3)+300(820)}{400} =\frac{116{,}666.666666\ldots+246{,}000}{400} =\$906.666666\ldots.

After redistribution, $Z_A=300/500=3/5$, giving $\hat{x}_A=(3/5)(1500)+(2/5)(1000)=\$1{,}300$. For B, $Z_B=100/300=1/3$, giving $\hat{x}_B=(1/3)(700)+(2/3)(1000)=\$900$. The revised portfolio indication is

\frac{300(1300)+100(900)}{400} =\frac{390{,}000+90{,}000}{400} =\$1{,}200.

Check and trap. Total exposure is 400 in both cases, but A’s share rises from 25% to 75%, and its credibility rises as well. The changed portfolio indication is therefore not explained by a change in total volume alone. Rounded to the nearest cent, the indications are $906.67 and $1,200.00. A common error is to preserve the old class weights simply because aggregate exposure is unchanged.

Worked Example: Scaling All Class Volumes

A two-class portfolio has a $1,000 complement for each class and $K=200$. Class A’s observed cost is $1,400 and class B’s is $800. Both classes increase from 100 to 300 exposure units. Calculate the portfolio indication before and after the change.

Method. The portfolio shares remain equal, but recalculate both credibility factors because the class volumes have changed.

Initially, each factor is $100/(100+200)=1/3$. The class indications are $3400/3=\$1{,}133.333333\ldots$ for A and $2800/3=\$933.333333\ldots$ for B. With equal exposure, the portfolio indication is

\frac{(3400/3)+(2800/3)}{2} =\$1{,}033.333333\ldots.

At 300 units each, both factors are $300/(300+200)=3/5$. A’s indication becomes $(3/5)(1400)+(2/5)(1000)=\$1{,}240$; B’s becomes $(3/5)(800)+(2/5)(1000)=\$880$. The new portfolio indication is

\frac{1240+880}{2}=\$1{,}060.

Check. The raw average observed cost is $(1400+800)/2=\$1{,}100$. Both portfolio indications lie between that average and the $1,000 complement, and the higher-volume indication is closer to the observed average. The exposure mix did not change, so the shift from $1,033.333333… to $1,060 comes from recalculated credibility, not portfolio reweighting. Rounded to the nearest cent, the results are $1,033.33 and $1,060.00.

Common Interpretation Traps

Holding credibility fixed after a class-volume change. Under the stated rule, $Z_i$ depends on $n_i$; use the revised volume to calculate it.
Updating credibility but retaining old portfolio shares. The portfolio average uses current exposures, so its weights must also be recalculated.
Changing only the total and overlooking the class mix. Equal proportional growth preserves shares; redistributing volume does not.
Confusing exposure with loss cost. Exposure is the volume used in the credibility rule and portfolio weighting. The observed cost is loss per exposure unit.
Rounding class indications too early. Carry full precision through aggregation and round the requested final result.

These distinctions matter because a portfolio indication is a statement about both class experience and the portfolio’s current composition. With costs and complements fixed, an exposure change can still move the indication by increasing the credibility of a class, changing that class’s portfolio share, or doing both. The calculation is aligned with the CAS Exam 8 (Advanced Ratemaking) syllabus.

Check Your Understanding

For each problem, use the exposure-based rule stated above and carry full precision until the final answer.

A class has $n=150$, $K=300$, observed cost $1,200, and complement $900. Find its credibility factor and indicated cost. (Answer: $Z=0.3333$; indicated cost $1,000.00.)
A class has $n=400$, $K=200$, observed cost $750, and complement $900. Find its credibility factor and indicated cost. (Answer: $Z=0.6667$; indicated cost $800.00.)
Two classes have $K=200$ and complement $1,000. A has exposure 100 and observed cost $1,300; B has exposure 300 and observed cost $700. Find the exposure-weighted portfolio indication. (Answer: $890.00.)
Two equally exposed classes have $K=100$, complement $1,000, and observed costs $1,500 and $500. Both volumes increase from 100 to 200. Calculate the portfolio indication before and after the change. (Answer: $1,000.00 before and $1,000.00 after; the individual class indications change.)
A portfolio has $K=100$, complement $1,000 for each class, and fixed observed costs of $1,200 for A and $900 for B. Exposure shifts from A = 200, B = 300 to A = 400, B = 100. Calculate the old and new portfolio indications. (Answer: $1,008.33 before and $1,118.00 after.)

What You'll Learn