Partial Credibility for Territorial Indications

Why Partial Credibility Is Used

A territory indication summarizes the experience available for that territory, but the amount of experience may be too small for the indication to be fully reliable. A territory with relatively few exposure units can have a high or low observed indication partly because of random variation. Partial credibility provides a systematic way to temper that indication with a stated complement.

The credibility weight answers a specific question: how much of the selected indication should come from the territory’s own experience? The rest of the weight is assigned to the complement. The weight measures the blending share; it does not change the observed indication itself, nor does it determine which complement is appropriate.

Definition: Let $I_{\text{obs}}$ be a territory’s observed indication, $I_{\text{comp}}$ a stated complement on the same scale, and $Z$ the territory’s credibility weight, with $0\leq Z\leq 1$. The partially credible indication is $$ I_{\text{cred}}=Z I_{\text{obs}}+(1-Z)I_{\text{comp}}. $$ The observed indication receives weight $Z$; the complement receives the remaining weight $1-Z$.

The Limited-Fluctuation Square-Root Rule

For a territory, a common limited-fluctuation approach starts with a full-credibility standard. Let $N$ denote the territory’s eligible exposure units and $N_0$ the number of comparable units required for full credibility under the selected standard. The partial-credibility weight is the square root of the ratio of those counts, capped at one:

Z=\min\left(1,\sqrt{\frac{N}{N_0}}\right).

The counts must use the same definition of an exposure unit and the same basis as the full-credibility standard. For example, if $N_0$ is expressed in earned exposure years, $N$ must also be in earned exposure years. Do not substitute a count of policies, claims, or vehicles unless the stated standard is defined in those units or a suitable conversion has been supplied.

Assumptions and scope: The square-root rule is a conventional limited-fluctuation credibility rule, not a universal formula for every credibility model. It assumes that the full-credibility standard applies to the exposure units being counted and that the experience is sufficiently comparable for that standard to be used. It does not correct for differences in exposure mix, trend, development, or coverage. Those issues must be addressed in constructing the indication before applying the weight. If $N\geq N_0$, use $Z=1$; the rule does not assign a weight greater than one.

The square root is important. If a territory has one quarter of the full-credibility exposure count, its weight is one half, not one quarter. Conversely, a weight of one half corresponds to one quarter of the full-credibility count. This relationship is a frequent source of exam errors.

Why the Square Root Appears

The rule is motivated by the way random fluctuation changes with the number of comparable exposure units. Suppose, as a simplifying model, that losses from individual exposure units are independent and identically distributed, with finite variance $\sigma^2$. The variance of the average loss over $N$ units is $\sigma^2/N$, so its standard deviation is $\sigma/\sqrt{N}$. Thus, relative to the standard deviation at the full-credibility count $N_0$, the standard deviation at count $N$ is:

\frac{\sigma/\sqrt{N}}{\sigma/\sqrt{N_0}} =\sqrt{\frac{N_0}{N}}.

The inverse of that relative fluctuation factor is $\sqrt{N/N_0}$, which motivates the conventional credibility weight. The full-credibility standard itself encodes a selected tolerance for random fluctuation and a confidence level; it is not derived from the exposure count alone. The square-root rule is therefore a practical scaling from a specified standard, not proof that every blend has a particular statistical optimality property.

Confirm the indication’s scale.
Determine whether the territory result is, for example, a relativity, a loss cost, or a loss ratio. The complement must use the same scale.

Match exposure units.
Check that the territory count $N$ and full-credibility count $N_0$ refer to comparable units, periods, and exposure bases.

Calculate and cap the weight.
Compute $\sqrt{N/N_0}$. If the result is at least one, set $Z=1$.

Blend and interpret.
Calculate $Z I_{\text{obs}}+(1-Z)I_{\text{comp}}$. Check that the result lies between the two indications when both weights are between zero and one.

Worked Applications

Worked Example: A Territory With One Quarter of the Full-Credibility Count

A hypothetical territory has 900 eligible exposure years. The applicable full-credibility standard is 3,600 exposure years. Its observed relativity is 1.18, and the stated complement is 0.94. Calculate the partial-credibility indication and interpret the weights. Round the final indication to three decimal places.

Method and unit check. Both counts are exposure years, and both indications are relativities, so the limited-fluctuation weight and the blending formula can be used directly.

Step 1: Calculate the weight.

\frac{N}{N_0}=\frac{900}{3{,}600}=0.25, \qquad Z=\sqrt{0.25}=0.50.

The complement receives weight $1-Z=1-0.50=0.50$.

Step 2: Blend the indications.

I_{\text{cred}} =(0.50)(1.18)+(0.50)(0.94) =0.590+0.470 =1.060.

Result and check. The partially credible relativity is 1.060. It lies halfway between 1.18 and 0.94, as expected from equal weights. This method avoids the mistake of using the exposure ratio, 0.25, as the credibility weight.

Worked Example: A Territory Above the Full-Credibility Standard

A hypothetical territory has 22,500 eligible exposure units, compared with a full-credibility standard of 10,000 units. Its observed indication is 1.11 and its stated complement is 0.97. Determine the credibility weight and blended indication. Round the indication to three decimal places.

Method. Apply the square-root rule, then enforce its full-credibility cap before blending.

Step 1: Calculate the uncapped value.

\sqrt{\frac{22{,}500}{10{,}000}} =\sqrt{2.25} =1.50.

A credibility weight cannot exceed one under this rule, so $Z=\min(1,1.50)=1$.

Step 2: Blend.

I_{\text{cred}} =(1)(1.11)+(1-1)(0.97) =1.11+0 =1.110.

Result and check. The indication is 1.110: at or above the full-credibility standard, the territory’s own indication receives all the weight. The complement contributes zero. Leaving the uncapped value of 1.50 in the blend would assign a negative weight to the complement and is incorrect.

Worked Example: Calculating a Relativity Before Blending

A hypothetical territory has 900 eligible exposure years and $810,000 of standardized ultimate losses. The statewide benchmark loss cost is $750 per exposure year. The full-credibility standard is 3,600 exposure years, and the stated complement relativity is 1.00. Calculate the territory’s observed relativity and its partially credible relativity. Round the final result to three decimal places.

Method and assumptions. Treat the supplied losses as already standardized and ultimate for this calculation. Calculate the territory loss cost per exposure year, divide by the statewide benchmark to obtain an observed relativity, then apply the square-root rule.

Step 1: Calculate the territory loss cost.

\frac{\$810{,}000}{900}=\$900 \quad\text{per exposure year}.

Step 2: Calculate the observed relativity.

I_{\text{obs}}=\frac{\$900}{\$750}=1.20.

Step 3: Calculate credibility.

Z=\sqrt{\frac{900}{3{,}600}} =\sqrt{0.25} =0.50.

Step 4: Blend with the complement.

I_{\text{cred}} =(0.50)(1.20)+(0.50)(1.00) =0.600+0.500 =1.100.

Result and check. The partially credible relativity is 1.100. It is above the complement because the observed relativity is above the complement, and it is below the observed relativity because the complement has positive weight. This calculation avoids blending a dollar loss cost directly with a dimensionless relativity; the observed indication and complement must be expressed on the same scale.

Interpreting the Weight and Avoiding Traps

The credibility weight controls how strongly a change in the observed indication changes the blended result. Holding $Z$ and the complement fixed, an increase of $\Delta$ in the observed indication changes the blended indication by $Z\Delta$. For instance, when $Z=0.50$, a 0.10 increase in the observed indication raises the blended result by 0.05. This makes $Z$ a direct measure of the observed indication’s share, not a measure of how far the blended result should move toward the complement.

Condition	Credibility weight	Effect on the blend
$N=0$	$Z=0$	The result equals the complement.
$0<N<N_0$	$Z=\sqrt{N/N_0}$	The result lies between the observed indication and complement.
$N\geq N_0$	$Z=1$	The result equals the observed indication.

If the observed indication and complement are both on the same numerical scale, the blend is a weighted average. Because $Z$ and $1-Z$ are nonnegative and sum to one, the blended result cannot be outside the interval between the two inputs. This is a useful arithmetic check, although it does not establish that either input is appropriate.

Common exam trap: using the exposure fraction as the weight. Suppose $N/N_0=0.25$, $I_{\text{obs}}=1.18$, and $I_{\text{comp}}=0.94$. The incorrect calculation uses $Z=0.25$:

$$ (0.25)(1.18)+(0.75)(0.94)=0.295+0.705=1.000. $$

The correction is to take the square root first: $Z=\sqrt{0.25}=0.50$. The correct blend is $(0.50)(1.18)+(0.50)(0.94)=1.060$. Other frequent errors include failing to cap $Z$ at one, applying the observed weight to the complement instead, and blending values stated in different units or on different scales.

Partial credibility is a controlled blend, not a substitute for preparing a sound territory indication. The experience must still be expressed on an appropriate and comparable basis, and the complement must be stated in the same units as the indication. The calculation tells you the effect of the selected weight: a larger $Z$ gives more influence to territory experience, while a smaller $Z$ pulls the result more toward the complement. This treatment of partial credibility is aligned with the CAS Exam 5 (Basic Ratemaking and Estimated Claim Liabilities) syllabus.

Check Your Understanding

For each problem, use the stated full-credibility standard and complement. Round final indications to three decimal places unless the answer is exact.

A territory has 400 eligible exposure years, a full-credibility standard of 1,600 exposure years, an observed relativity of 1.30, and a complement of 1.00. Calculate $Z$ and the blended relativity. (Answer: $Z=0.500$; blended relativity $=1.150$.)
A territory has 1,800 eligible exposure units and a full-credibility standard of 7,200 units. Its observed indication is 0.90 and its complement is 1.10. Calculate the credibility weight and blend. (Answer: $Z=0.500$; blended indication $=1.000$.)
A territory has 6,400 eligible exposure units, a full-credibility standard of 1,600 units, an observed indication of 1.07, and a complement of 0.95. Apply the cap and calculate the blend. (Answer: $Z=1.000$; blended indication $=1.070$.)
A territory has 900 exposure years and standardized ultimate losses of $756,000. The statewide benchmark is $700 per exposure year, the full-credibility standard is 3,600 exposure years, and the complement relativity is 1.00. Calculate the observed relativity and blend. (Answer: observed relativity $=1.200$; $Z=0.500$; blended relativity $=1.100$.)
A territory needs a credibility weight of 0.75 under the square-root rule. The full-credibility standard is 6,400 exposure units. If its observed indication is 1.24 and its complement is 0.96, how many exposure units are required, and what is the blended indication? (Answer: 3,600 exposure units; blended indication $=1.170$.)

Condition	Credibility weight	Effect on the blend
\(N=0\)	\(Z=0\)	The result equals the complement.
\(0<N<N_0\)	\(Z=\sqrt{N/N_0}\)	The result lies between the observed indication and complement.
\(N\geq N_0\)	\(Z=1\)	The result equals the observed indication.

What You'll Learn