Full and Partial Credibility

Why the Distinction Matters

Credibility describes how much weight an estimate gives to a particular set of experience. The distinction between full credibility and partial credibility concerns whether that experience meets a specified standard for being relied on fully. It does not, by itself, tell an actuary how to calculate every partial weight.

This matters when experience volume is near a threshold. An actuary must first identify the applicable standard and compare the available experience with it. Only then should the actuary apply the prescribed credibility weight. Confusing these steps can lead to a rate indication that treats insufficient experience as fully reliable, or to a partial weight calculated by a method the problem did not authorize.

Definition: A set of experience is fully credible under a specified standard when it meets or exceeds the stated requirement for full reliance. In the usual credibility-weight notation, its credibility $Z$ is 1. Experience that does not meet the requirement may receive partial credibility, meaning it is assigned a weight $Z$ less than 1 and greater than or equal to 0. The applicable standard and the method for assigning a partial weight must be stated or otherwise specified.

Separate the Standard From the Weight

A full-credibility standard is a rule defining what experience volume or fluctuation performance is sufficient for full reliance. Depending on the application, the standard may be expressed in exposure units, claim counts, or another measure. Its numerical requirement and the conditions under which it applies must come from the problem or the selected actuarial method.

For this tutorial, examples supply the standard directly. They do not derive a claim-count threshold or a rule for converting a shortfall in volume into a partial weight. Those are separate calculations. In particular, knowing that experience falls short of a threshold does not alone determine a unique value of $Z$.

Question	What it establishes
What full-credibility standard applies?	The required experience volume or performance criterion
Does the observed experience meet it?	Whether the experience qualifies for full credibility under that standard
If it does not, what partial-weight method applies?	The value of $Z$, if partial credibility is assigned
How is the indicated estimate calculated?	The blend of subject experience and the complement using $Z$

In the standard linear credibility framework, let $E$ be the estimate based on the subject experience and $M$ the complement estimate. The indicated estimate $C$ is:

C=ZE+(1-Z)M,\qquad 0\leq Z\leq 1.

Here $1-Z$ is the weight on the complement, not an additional measure of experience volume. If $Z=1$, then $C=E$: the complement has zero weight. If $Z=0$, then $C=M$: the subject experience has zero weight. For partial credibility, $0<Z<1$, so both estimates contribute.

Why the Weighted Estimate Has This Form

The formula is a weighted average: the two weights are $Z$ and $1-Z$, and they sum to 1. Rewriting the same expression shows exactly how the subject estimate moves the complement:

\begin{aligned} C&=ZE+(1-Z)M\\ &=ZE+M-ZM\\ &=M+Z(E-M). \end{aligned}

Thus the indicated estimate starts at $M$ and moves a fraction $Z$ of the difference between $E$ and $M$. When $E>M$, the result is between $M$ and $E$; when $E<M$, it is also between them. This follows because $C$ is a convex combination of the two estimates for $0\leq Z\leq1$. The formula describes how a given weight is applied; it does not establish that a particular weight is statistically optimal.

Worked Example: Experience Meets the Full-Credibility Threshold

A hypothetical specialty contractor program uses a stated standard of at least 800 earned exposure units for full credibility. A contractor has 920 earned exposure units. Its subject-experience indicated annual cost is $1,260 per unit, and the complement estimate is $1,140 per unit. Determine the credibility classification and the indicated estimate.

Assumptions and method. The full-credibility threshold of 800 units is stipulated for this problem. The applicable partial-weight method is not needed because the experience may qualify as fully credible. No interest or mortality assumptions apply. No probability distribution or claim-count model is assumed; the given threshold is treated as the governing standard.

Step 1: Compare the experience with the threshold. The program has 920 units, and:

920\geq 800.

It meets the stated standard, so $Z=1$.

Step 2: Apply full credibility. With subject estimate $E=\$1{,}260$ and complement $M=\$1{,}140$:

\begin{aligned} C&=ZE+(1-Z)M\\ &=1(\$1{,}260)+0(\$1{,}140)\\ &=\$1{,}260\text{ per unit}. \end{aligned}

Check. Full credibility assigns all the weight to subject experience, so the indicated estimate equals $1,260 per unit. The result is consistent with $Z=1$.

Method and mistake avoided. The method is to test the stated volume criterion before applying the credibility formula. A common mistake is to blend the subject and complement estimates even after the problem says the experience meets the full standard. Doing so would incorrectly assign a positive complement weight.

Partial Credibility Requires a Weighting Rule

When experience does not meet the full standard, an actuary may assign it partial credibility. The subject estimate is then blended with its complement, but the weight must come from an applicable rule or be supplied in the problem. Different partial-credibility methods can assign different weights to the same volume. Therefore, a threshold comparison alone supports the classification “not fully credible under this standard”; it does not justify a particular numerical $Z$.

The distinction is especially important in exam calculations. If a question supplies $Z$, use that value. If it supplies a method for deriving $Z$, apply that method and its assumptions. If neither is supplied, do not invent a partial weight from the volume shortfall.

Worked Example: Applying a Supplied Partial Weight

A hypothetical small marine equipment class has 750 earned exposure units. The stated full-credibility standard is 1,200 units. Under the company’s stipulated partial-weight method, the class is assigned $Z=0.65$. Its subject estimate is $96 per exposure unit, and its complement estimate is $80 per exposure unit. Find the classification and indicated estimate.

Assumptions and method. The 1,200-unit threshold and $Z=0.65$ are both given. The weight is not derived from the volume. No interest or mortality assumptions apply, and no distribution or claim-count model is specified or needed for this calculation.

Step 1: Classify the experience. Since:

$$ 750<1{,}200, $$

the experience does not meet the stated full-credibility standard. It is assigned partial credibility under the stipulated method.

Step 2: Find the complement weight.

$$ 1-Z=1-0.65=0.35. $$

Step 3: Blend the estimates.

\begin{aligned} C&=0.65(\$96)+0.35(\$80)\\ &=\$62.40+\$28.00\\ &=\$90.40\text{ per exposure unit}. \end{aligned}

Check. The indicated estimate is between $80 and $96, as it must be for a weighted average with weights 0.35 and 0.65. It is closer to the subject estimate because the subject receives the greater weight.

Method and mistake avoided. The method is to classify against the threshold, then apply the supplied weight. A common mistake is to assume that 750 units out of 1,200 automatically means $Z=750/1{,}200$. The problem instead specifies $Z=0.65$; the volume ratio is not a substitute for the stated method.

Boundary Cases and Interpretation

Because the standard is a threshold, equality matters. If a standard requires “at least 500 claims,” exactly 500 meets it; 499 does not. If it instead requires “more than 500 claims,” exactly 500 falls short. Read the condition literally and preserve its unit: a requirement stated in claims is not interchangeable with one stated in policies or exposure years.

Also distinguish an experience classification from an estimate. “Fully credible” describes whether the experience satisfies a specified standard; “$X per unit” is an indicated estimate. The classification alone does not tell you the subject estimate, the complement, or the resulting premium level.

Worked Example: Two Classes With Different Credibility Status

A hypothetical equipment warranty program stipulates a full-credibility standard of 400 reported claims for each class. Class A has 430 claims, while Class B has 300. For Class B, the program explicitly specifies $Z=0.75$. Class A’s subject estimate is $510 per contract-year, Class B’s is $460, and the common complement estimate is $500. Classify each class and calculate both indicated estimates.

Assumptions and method. The threshold and Class B’s partial weight are stipulated. For this calculation, meeting the standard gives Class A $Z=1$. No interest or mortality assumptions apply. No distribution or claim-count model is assumed; claim counts are compared directly with the supplied standard.

Step 1: Classify each class. Class A has $430\geq400$ claims and meets the standard. Class B has $300<400$ claims and does not. Its partial weight is provided as $Z=0.75$.

Step 2: Calculate Class A. Full credibility means $Z=1$:

\begin{aligned} C_A&=1(\$510)+0(\$500)\\ &=\$510\text{ per contract-year}. \end{aligned}

Step 3: Calculate Class B. Its complement weight is $1-0.75=0.25$, so:

\begin{aligned} C_B&=0.75(\$460)+0.25(\$500)\\ &=\$345+\$125\\ &=\$470\text{ per contract-year}. \end{aligned}

Check. Class A’s result equals its subject estimate because it is fully credible. Class B’s result is between $460 and $500, and it is closer to $460 because the subject weight is 0.75. The two classes are treated differently because only Class A meets the stated standard.

Method and mistake avoided. Compare each class separately with the same threshold, then apply its applicable weight. A common mistake is to pool the classes’ claims before classifying them. The problem specifies a standard for each class, so a class cannot borrow claim volume from another class to meet its requirement.

Common Exam Traps

Assuming the threshold determines the partial weight. The threshold identifies whether the full standard is met. A separate rule or supplied value is needed to determine $Z<1$.
Using the complement weight as the subject weight. If $Z=0.30$, the subject receives 30% and the complement receives $1-Z=0.70$.
Reversing the estimates. In $C=ZE+(1-Z)M$, $E$ is the subject estimate and $M$ is its complement.
Rounding before classifying. Compare the actual experience amount with the threshold before rounding, particularly when the amounts are close.
Ignoring the standard’s wording or unit. “At least,” “more than,” claim counts, and exposure units are not interchangeable.
Calling every estimate based on experience fully credible. The relevant question is whether that experience meets the specified standard, not whether the estimate was calculated from actual data.

Key takeaway: First identify the full-credibility standard, then determine whether the experience meets it. Full credibility gives the subject experience weight 1. Partial credibility gives it a weight below 1, but that weight must come from an applicable method or from the problem; it cannot be inferred from the threshold comparison alone.

Check Your Understanding

For each problem, use only the standard and method stated in the question. Dollar amounts are hypothetical.

A property class has 615 exposure units. Its stated full-credibility standard is 600 units. Its subject estimate is $720 per unit. What is its credibility classification and indicated estimate? (Answer: Full credibility; $720 per unit.)
A hypothetical fleet program has 360 exposure units against a stated full standard of 500. Its prescribed method gives $Z=0.60$; the subject and complement estimates are $1,050 and $950 per vehicle-year. Find the complement weight and indicated estimate. (Answer: Complement weight 0.40; $1,010 per vehicle-year.)
A warranty class must have at least 250 claims for full credibility. It has exactly 250 claims. Its subject estimate is $84 per contract and its complement is $90 per contract. What is its classification and indicated estimate? (Answer: Full credibility; $84 per contract.)
Two hypothetical segments each face a full standard of 1,000 exposure units. Segment P has 900 units and a supplied partial weight of 0.80; its subject estimate is $430 and its complement is $400. Segment Q has 1,050 units; its subject estimate is $390. Calculate each indicated estimate. (Answer: P is partially credible at $424; Q is fully credible at $390.)
A class has 480 claims against a stipulated threshold of more than 480 claims. Its subject estimate is $125 and its complement is $110. No partial-weight method or weight is supplied. State what can be concluded and whether a numerical blended estimate can be calculated. (Answer: It does not meet the full standard; a numerical partial estimate cannot be determined without a partial-weight method or supplied $Z$.)

What You'll Learn