Credibility As A Weighted Estimate

Introduction

A credibility estimate combines information from observed experience with complementary information. The general idea is simple: the result should reflect both sources, with the credibility weight determining how much of the estimate comes from the observed experience. Reading the blend correctly matters as much as performing the arithmetic; reversing the weights can move an estimate in the wrong direction.

This tutorial focuses on interpreting and using that general weighted-average form. It does not specify how to choose a credibility formula or decide which rate component to blend. Those decisions require additional considerations. Here, assume that the two estimates already refer to the same target, coverage, time basis, and units, and that a credibility weight has been supplied.

Definition: Let $O$ be an estimate based on the observed experience, $C$ be the complementary estimate, and $Z$ be the credibility weight assigned to the observed estimate. The general credibility estimate is $$ M=ZO+(1-Z)C, $$ where $0\leq Z\leq 1$. The weight on the complementary estimate is $1-Z$, so the two weights sum to one.

The symbols identify roles in the calculation, not particular types of insurance quantities. The observed estimate might be a loss ratio, a rate indication, or another measure, provided that the observed and complementary estimates are comparable. The value $Z$ is a weight, not an estimate of the quantity being blended. In particular, a value of $Z=0.30$ means that 30% of the weighted average comes from $O$, not that the final estimate is 30% of $O$.

Read the Weights Before Calculating

The phrase “credibility assigned to observed experience” means that $Z$ multiplies $O$. Its complement, $1-Z$, multiplies $C$. Thus, if $Z=0.40$, the observed estimate receives weight 0.40 and the complementary estimate receives weight $1-0.40=0.60$. The weights must total one:

$$ Z+(1-Z)=1. $$

This convention makes the result a weighted average. When both weights are between zero and one, the result lies between the two input estimates. If $O$ is larger than $C$, increasing $Z$ moves $M$ upward toward $O$. If $O$ is smaller than $C$, increasing $Z$ moves $M$ downward toward $O$. At the endpoints, $Z=0$ gives $M=C$, while $Z=1$ gives $M=O$.

The complement is not an extra adjustment added after the observed estimate. It is the other estimate in the blend, and its weight is the remainder after assigning $Z$ to observed experience. For that reason, a calculation should name both terms and their weights before substituting numbers.

Worked Example: Blending Two Hypothetical Loss Estimates

A hypothetical insurer is reviewing a prospective annual loss estimate for a small commercial book. Its observed-experience estimate is $420,000, its complementary estimate is $500,000, and the supplied credibility weight on observed experience is $Z=0.35$. Calculate the credibility estimate and check whether it is sensible.

Method. Use the general weighted-average form, assigning $Z$ to the observed estimate and $1-Z$ to the complement.

Step 1: Find the complementary weight. $$ 1-Z=1-0.35=0.65. $$ The weights are 0.35 on the observed estimate and 0.65 on the complementary estimate; they sum to $0.35+0.65=1.00$.

Step 2: Apply each weight. $$ 0.35(\$420{,}000)=\$147{,}000,\qquad 0.65(\$500{,}000)=\$325{,}000. $$

Step 3: Add the weighted amounts. $$ M=\$147{,}000+\$325{,}000=\$472{,}000. $$

Check. The result, $472,000, lies between $420,000 and $500,000. Since the complement is higher and has the larger weight, the result is closer to $500,000, as expected. The weighted differences from $500,000 are $28,000 below the complement and $52,000 above the observed estimate; their ratio is $28{,}000/52{,}000=0.5385$, matching the weight ratio $0.35/0.65=0.5385$ after rounding to four decimals.

Common mistake avoided. Giving 0.35 to the complement instead would produce $0.35(\$500{,}000)+0.65(\$420{,}000)=\$448{,}000$, which reverses the stated meaning of $Z$.

Why the Blend Has This Form

A weighted average assigns a share of the total weight to each input. If the observed estimate receives share $Z$, the remaining share must go to the complement, so it is $1-Z$. The contributions are therefore $ZO$ and $(1-Z)C$; adding the contributions gives the stated formula. An equivalent rearrangement makes the adjustment from the complement especially clear:

$$ M=ZO+(1-Z)C =ZO+C-ZC =C+Z(O-C). $$

This identity also explains how the blend moves. Start at $C$, then move a fraction $Z$ of the difference $O-C$ toward $O$. If $O>C$, the difference is positive and $M$ is above $C$. If $O<C$, the difference is negative and $M$ is below $C$. When $Z=0$, no part of the difference is applied; when $Z=1$, the full difference is applied and the result is $O$.

For $0\leq Z\leq1$, the result also satisfies the bounds property. If $O\geq C$, then $O-C\geq0$, so $$ C\leq C+Z(O-C)\leq C+(O-C)=O. $$ If $O\leq C$, the same reasoning with the order reversed gives $O\leq M\leq C$. This is a direct check that the result cannot fall outside the interval between the inputs under the stated weight condition.

Worked Example: Recovering the Observed Weight

For a hypothetical coverage review, the observed estimate of a prospective rate change is 9%, while the complementary estimate is 4%. The reported blended estimate is 7%. Determine the weight assigned to observed experience, and verify the result.

Method. Use $M=C+Z(O-C)$ and solve for $Z$. The inputs are percentages on the same rate-change basis.

Step 1: Find the difference between the observed and complementary estimates. $$ O-C=9\%-4\%=5\text{ percentage points}. $$

Step 2: Find how far the blend is above the complement. $$ M-C=7\%-4\%=3\text{ percentage points}. $$

Step 3: Divide the movement achieved by the full movement available. $$ Z=\frac{M-C}{O-C} =\frac{3\text{ percentage points}}{5\text{ percentage points}} =0.60. $$ The observed estimate has weight 0.60, and the complementary weight is $1-0.60=0.40$.

Step 4: Verify by substitution. $$ M=0.60(9\%)+0.40(4\%)=5.4\%+1.6\%=7.0\%. $$

Check. The answer is between 4% and 9%, and it is closer to 9%, which has the larger weight.

Common mistake avoided. Dividing $M$ by $O$, giving $7\%/9\%=0.7778$, does not recover the credibility weight. The weight measures the share of the difference from $C$ to $O$ that the blend travels, so the correct ratio is $(M-C)/(O-C)$.

When the Weighted-Average Interpretation Applies

The interpretation depends on both the weights and the meaning of the inputs. First, the two weights must be $Z$ and $1-Z$, with $Z$ between zero and one. Second, $O$ and $C$ must estimate the same quantity on a comparable basis. Blending a dollar amount with a percentage, or estimates for different coverage periods, does not produce a meaningful weighted estimate merely because the arithmetic can be done.

If a supplied “weight” is outside the interval from zero to one, then $ZO+(1-Z)C$ is no longer a convex weighted average and the result can lie outside the input estimates. That may describe some other mathematical calculation, but it is not the ordinary credibility blend defined here. Do not silently treat a negative value or a value greater than one as a conventional credibility weight.

The weight alone does not explain why a particular value was selected. It describes how much influence the observed estimate has in the stated blend. Determining or justifying the weight requires additional methods and information; it should not be inferred from the formula alone. This distinction is aligned with the CAS Exam 5 (Basic Ratemaking and Estimated Claim Liabilities) syllabus.

Worked Example: Checking Units and Direction

A hypothetical specialty insurer has an observed estimate of 68% and a complementary estimate of 74% for the same prospective loss ratio. A note states that the credibility weight on observed experience is 0.25. Calculate the blend, state the direction of movement, and identify a tempting but incorrect calculation.

Method. Apply the observed weight to 68% and the complementary weight to 74%.

Step 1: Calculate the complementary weight. $$ 1-Z=1-0.25=0.75. $$

Step 2: Calculate the two weighted contributions. $$ 0.25(68\%)=17.0\%,\qquad 0.75(74\%)=55.5\%. $$

Step 3: Add the contributions. $$ M=17.0\%+55.5\%=72.5\%. $$

Check. The result lies between 68% and 74%. It is closer to the complement, 74%, because that estimate has weight 0.75. Equivalently, $M=74\%+0.25(68\%-74\%)=74\%-1.5\%=72.5\%$.

Common mistake avoided. Reversing the weights gives $0.75(68\%)+0.25(74\%)=69.5\%$, incorrectly pulling the result much closer to the observed estimate despite the stated observed weight being only 0.25.

Common Exam Traps

Assigning $Z$ to the complement. By convention here, $Z$ is the weight on the observed estimate; the complement receives $1-Z$.
Using only $Z O$. A credibility estimate includes the complementary contribution $(1-Z)C$ as well.
Confusing weight with value. A 40% weight on observed experience does not mean the credibility estimate is 40% of the observed estimate.
Ignoring direction. If $O<C$, more weight on observed experience pulls the result downward, not upward.
Blending incomparable quantities. Confirm that both estimates refer to the same target and units before combining them.
Failing the bounds check. With $0\leq Z\leq1$, the blend must lie between $O$ and $C$. A result outside that interval is a prompt to recheck the inputs, weights, and arithmetic.

Key takeaway: In the general credibility blend, $Z$ is the share assigned to observed experience, and $1-Z$ is the share assigned to complementary information. Check that the weights sum to one, the inputs are comparable, and the result lies between them.

Check Your Understanding

For each problem, identify the observed and complementary estimates before calculating. Treat all stated weights as weights on observed experience.

A hypothetical insurer has an observed estimate of $360,000 and a complementary estimate of $440,000. If $Z=0.20$, what is the blended estimate, and is it closer to the observed estimate or the complement? (Answer: $424,000; closer to the complement.)
A hypothetical rate review gives an observed indication of 12% and a complementary indication of 7%. The blend is 10%. Find $Z$ and the complementary weight, then verify that the result is between the two indications. (Answer: $Z=0.60$, complementary weight $=0.40$; the blend is between 7% and 12%.)
For a hypothetical loss-ratio estimate, $O=81\%$, $C=69\%$, and $Z=0.35$. Calculate the blend and state whether it moves upward or downward from the complement. (Answer: $M=73.2\%$; it moves upward from 69% toward 81%.)
A hypothetical analysis reports $O=52$, $C=64$, and a blend of 59.2, with all three values on the same scale. Recover $Z$, find the complement weight, and check the direction. (Answer: $Z=0.40$, complement weight $=0.60$; the blend is below 64 and moves toward 52.)
A hypothetical indicated cost is $O=\$1.08$ million, while the complement is $C=\$0.96$ million. A calculation uses 0.30 on the complement and 0.70 on the observed estimate, although the stated observed credibility is $Z=0.30$. Give the correct blend and the incorrectly calculated reversed-weight result. (Answer: correct blend $=\$996{,}000$; reversed-weight result $=\$1{,}044{,}000$.)

What You'll Learn