Why Class Indications Reconcile to the Collective Mean

Why Reconciliation Requires a Condition

Credibility-weighted class indications combine a class’s own experience with a collective complement. A useful portfolio check is whether the exposure-weighted average of those indications equals the collective mean. That equality is not automatic: it depends on how the complement is defined in relation to class experience and credibility.

This tutorial focuses on that relationship. The key is to distinguish an ordinary exposure-weighted mean from a mean weighted by both exposure and credibility. They are generally different when class credibility factors differ.

Definition: For class $j$, let $e_j$ be its positive exposure on a consistent basis, let $\bar X_j$ be its observed loss cost per exposure unit, and let $Z_j$ be its credibility factor, with $0\leq Z_j\leq 1$. Given a common collective complement $\mu_C$, the class indication is $$ I_j=Z_j\bar X_j+(1-Z_j)\mu_C =\mu_C+Z_j(\bar X_j-\mu_C). $$

The expression on the right highlights the adjustment from the complement. A class indication starts at $\mu_C$, then moves toward the class’s observed loss cost by the fraction $Z_j$. The total indicated amount for a class is $e_j I_j$, so its exposure-weighted portfolio average is $\sum_j e_j I_j/\sum_j e_j$.

The Reconciliation Condition

For indications to average to $\mu_C$ using ordinary exposure weights, the credibility-weighted deviations from the complement must sum to zero. In other words, classes above the complement must offset classes below it after their deviations have been multiplied by both exposure and credibility.

\frac{\sum_j e_j I_j}{\sum_j e_j} =\mu_C+ \frac{\sum_j e_jZ_j(\bar X_j-\mu_C)}{\sum_j e_j}.

Consequently, the exposure-weighted average indication equals $\mu_C$ precisely when $\sum_j e_jZ_j(\bar X_j-\mu_C)=0$. One way to satisfy that condition is to define the collective mean using the effective credibility-weighted exposures $e_jZ_j$:

\mu_C= \frac{\sum_j e_jZ_j\bar X_j}{\sum_j e_jZ_j}, \qquad \text{provided } \sum_j e_jZ_j>0.

The weights $e_jZ_j$ are called effective here because they measure each class’s contribution to the credibility-weighted deviation. They are not the same as the exposures $e_j$ used to average the final indications. This distinction is essential: the complement is calculated using $e_jZ_j$, while reconciliation of the indications uses $e_j$.

Calculate effective weights.
For each class, multiply its exposure by its credibility factor: $e_jZ_j$.

Calculate the complement.
Average the observed class loss costs using the effective weights to obtain $\mu_C$.

Calculate class indications.
Blend each class loss cost with the same complement using that class’s $Z_j$.

Check the portfolio average.
Average the indications using the original exposures $e_j$; the result is $\mu_C$, subject to the stated condition.

Why the Formula Works

The justification follows directly from the definition of an indication. Subtract the complement from every class indication and then add those deviations using exposure weights:

\begin{aligned} \sum_j e_j(I_j-\mu_C) &=\sum_j e_j Z_j(\bar X_j-\mu_C)\\ &=\sum_j e_jZ_j\bar X_j-\mu_C\sum_j e_jZ_j. \end{aligned}

If $\mu_C=\sum_j e_jZ_j\bar X_j/\sum_j e_jZ_j$, the last line is zero. Thus $\sum_j e_j(I_j-\mu_C)=0$, which is equivalent to $\sum_j e_jI_j=\mu_C\sum_j e_j$. Dividing by the positive total exposure proves that the exposure-weighted average indication equals the complement.

This derivation also shows the limitation. If a complement is supplied independently, reconciliation holds only if the weighted deviations happen to sum to zero. Calling a number the “collective mean” does not, by itself, establish that condition.

Worked Example: Reconciling Three Class Indications

A hypothetical insurer reviews three classes. Exposures are 100, 200, and 300 units; observed loss costs are $80, $100, and $140 per unit; and credibility factors are 0.25, 0.50, and 0.75. Use the common credibility-weighted complement, calculate each indication, and verify the exposure-weighted average. Assume all loss costs are on the same basis, with no separate trend or discount adjustment.

Method. First calculate effective weights $e_jZ_j$: $100(0.25)=25$, $200(0.50)=100$, and $300(0.75)=225$. Their sum is $25+100+225=350$.

The weighted loss-cost numerator is $25(80)+100(100)+225(140)=2{,}000+10{,}000+31{,}500=43{,}500$. Therefore, the common complement is $\mu_C=43{,}500/350=\$124.2857$ per unit.

For class 1, $I_1=124.2857+0.25(80-124.2857)=124.2857-11.0714=\$113.2143$. For class 2, $I_2=124.2857+0.50(100-124.2857)=124.2857-12.1429=\$112.1429$. For class 3, $I_3=124.2857+0.75(140-124.2857)=124.2857+11.7857=\$136.0714$.

The exposure-weighted numerator of the indications is $100(113.2143)+200(112.1429)+300(136.0714)=11{,}321.43+22{,}428.58+40{,}821.42=\$74{,}571.43$, subject to rounding of the displayed indications. Dividing by total exposure $100+200+300=600$ gives $\$124.2857$ per unit before rounding. The result agrees with the complement. The common mistake avoided is using exposure weights to calculate the complement while expecting the credibility-weighted reconciliation identity to hold automatically.

Common Credibility Patterns and Their Limits

A familiar Bühlmann-style credibility specification uses $Z_j=e_j/(e_j+K)$, where $K>0$ is common across classes and $e_j$ is measured on a consistent exposure basis. Under that specification, the effective weight is $e_jZ_j=e_j^2/(e_j+K)$. The reconciliation result still applies if the common complement is calculated from those effective weights.

The presence of a common $K$ does not make $e_jZ_j$ equal to $e_j$, nor does it imply that the complement is the ordinary exposure-weighted average of observed loss costs. Common parameters support a consistent credibility structure; they do not remove the need to check the complement condition.

Worked Example: Applying a Common Bühlmann-Style Parameter

For three hypothetical classes, exposures are 100, 200, and 400 units, and observed loss costs are $60, $90, and $120 per unit. Suppose the common parameter is $K=200$, and use $Z_j=e_j/(e_j+K)$. Calculate the complement and indications, then verify their exposure-weighted average. The loss costs are already on a common basis.

The credibility factors are $100/(100+200)=1/3$, $200/(200+200)=1/2$, and $400/(400+200)=2/3$. The effective weights are $100/3$, $100$, and $800/3$, with total $100/3+100+800/3=400$.

The complement numerator is $(100/3)(60)+100(90)+(800/3)(120)=2{,}000+9{,}000+32{,}000=43{,}000$. Hence $\mu_C=43{,}000/400=\$107.50$ per unit.

The indications are $I_1=107.50+(1/3)(60-107.50)=\$91.6667$, $I_2=107.50+(1/2)(90-107.50)=\$98.75$, and $I_3=107.50+(2/3)(120-107.50)=\$115.8333$. The exposure-weighted numerator is $100(91.6667)+200(98.75)+400(115.8333)=9{,}166.67+19{,}750+46{,}333.32=\$75{,}249.99$, with the one-cent difference caused by rounding. Using unrounded indications gives $\$75{,}250$. Dividing by 700 exposure units gives $\$107.50$, as required.

Check. The common $K$ generates different class credibility factors because exposures differ. The reconciliation uses the resulting effective weights to calculate the complement, not equal weights or the original exposures alone.

When an Ordinary Collective Average Does Not Reconcile

The ordinary exposure-weighted mean is $\mu_E=\sum_j e_j\bar X_j/\sum_j e_j$. If it is used as the complement, the average indication equals $\mu_E$ only when $\sum_j e_jZ_j(\bar X_j-\mu_E)=0$. That can occur, for example, if all $Z_j$ are equal and $\mu_E$ is the exposure-weighted mean. With unequal credibility factors, it need not hold.

Worked Example: Detecting a Non-Reconciling Complement

Two hypothetical classes have exposures of 100 and 300 units, observed loss costs of $40 and $100 per unit, and credibility factors of 0.50 and 0.25. An analyst uses the ordinary exposure-weighted mean as the complement. Calculate the class indications and test reconciliation. All figures are on a consistent loss-cost basis.

The ordinary complement is $\mu_E=[100(40)+300(100)]/400=(4{,}000+30{,}000)/400=\$85$ per unit. The indications are $I_1=85+0.50(40-85)=\$62.50$ and $I_2=85+0.25(100-85)=\$88.75$.

Their exposure-weighted average is $[100(62.50)+300(88.75)]/400=(6{,}250+26{,}625)/400=\$82.1875$ per unit, not $85. The deviation check gives $100(0.50)(40-85)+300(0.25)(100-85)=-2{,}250+1{,}125=-\$1{,}125$, which is nonzero. The failure is therefore expected; it is not an arithmetic error.

To make these same indications reconcile by the stated method, use effective weights $100(0.50)=50$ and $300(0.25)=75$. The corresponding complement is $[50(40)+75(100)]/(50+75)=(2{,}000+7{,}500)/125=\$76$. This is a different complement, so it produces different indications. The example illustrates the required condition, not a rule that every analysis must replace its chosen complement.

Common Exam Trap

A frequent error is to state that credibility-weighted class indications always reconcile to the ordinary exposure-weighted mean because the class deviations are “being blended back” to the collective. The blend only shrinks each deviation by $Z_j$; it does not guarantee that the exposure-weighted sum of the shrunken deviations is zero.

The incorrect shortcut is to set $\mu_C=\sum e_j\bar X_j/\sum e_j$, calculate $I_j=\mu_C+Z_j(\bar X_j-\mu_C)$, and assert $\sum e_jI_j/\sum e_j=\mu_C$. The corrected calculation checks $\sum e_jZ_j(\bar X_j-\mu_C)$. If that sum is nonzero, the asserted reconciliation is false. Either report the resulting weighted indication average, or use a complement that satisfies the required weighted-deviation condition when that is the intended modeling assumption.

Key takeaway: Class indications $I_j=\mu_C+Z_j(\bar X_j-\mu_C)$ average to $\mu_C$ under ordinary exposure weights exactly when $\sum_j e_jZ_j(\bar X_j-\mu_C)=0$. Defining $\mu_C$ as the mean of class loss costs weighted by $e_jZ_j$ satisfies that condition; an ordinary exposure-weighted mean generally does not.

Check Your Understanding

For each problem, distinguish the exposures used to average indications from the effective weights used to define a reconciling complement.

Two classes have exposures 50 and 150, loss costs $40 and $100, and credibility factors 0.40 and 0.80. Calculate the effective-weight complement, both class indications, and their exposure-weighted average. (Answer: complement $91.4286; indications $70.8571 and $98.2857; exposure-weighted average $91.4286 per unit.)
Two classes have exposures 80 and 120, loss costs $30 and $90, and credibility factors 0.25 and 0.50. Calculate the reconciling complement and the exposure-weighted average indication. (Answer: complement $75; exposure-weighted average $75 per unit.)
Three classes have exposures 100, 200, and 300, loss costs $50, $80, and $110, and credibility factors 0.20, 0.40, and 0.60. Find the effective-weight complement. (Answer: $97.1429 per unit.)
Two classes have exposures 100 and 300, loss costs $40 and $100, and credibility factors 0.50 and 0.25. If the ordinary exposure-weighted mean is used as the complement, what is the exposure-weighted average indication, and does it reconcile? (Answer: $82.1875 per unit; no, it does not equal the $85 complement.)
State the condition that a fixed complement $\mu_C$ must satisfy for exposure-weighted class indications to average to it. (Answer: $\sum_j e_jZ_j(\bar X_j-\mu_C)=0$.)

What You'll Learn