Introduction
Observed claims experience can suggest a rate level, but that does not tell us how much weight to give that experience when estimating a rate component. A high observed loss cost may be based on a stable body of experience or on a small, volatile set of observations. The dollar amount and the reliability weight answer different questions.
Credibility is the weight assigned to observed experience in estimating a rate component. It describes how much the experience is allowed to influence the estimate under a specified credibility approach; it is not the rate level itself, nor a direct measure of whether a particular observed rate is high or low. This distinction is foundational to credibility for ratemaking and is aligned with the CAS Exam 5 (Basic Ratemaking and Estimated Claim Liabilities) syllabus.
Two Different Questions
A credibility assessment separates two questions. First, what level does the observed experience indicate? For example, a portfolio might have an observed loss cost of $135 per exposure. Second, how much weight should that experience receive? Its assigned credibility might be 0.40. The loss cost is a rate-level quantity measured in dollars per exposure; the credibility is a dimensionless weight.
The credibility weight does not tell us whether the observed rate is adequate, excessive, or below a target. Nor does it say that the indicated rate is 40% accurate, or that there is a 40% probability the rate is correct. It describes the role of the experience in an estimation procedure. The indicated level and the weight must therefore be reported and interpreted separately.
| Quantity | Question answered | Typical form |
|---|---|---|
| Observed indication | What rate level does this experience suggest? | Dollars per exposure, or another rate component's units |
| Credibility \(Z\) | How much weight is assigned to this experience? | A dimensionless number from 0 to 1 |
| Estimated rate component | What value results after applying the selected estimation approach? | Same units as the rate component |
The last quantity depends on the estimation method and on the other information it uses. A credibility value alone cannot determine an estimated rate. In particular, it cannot transform a dollar indication into a new rate without knowing the rest of the procedure.
Interpreting the Weight
For a simple linear credibility approach, the estimated component responds to changes in the observed indication in proportion to \(Z\), provided the other inputs and the method remain fixed. This gives a direct interpretation of the weight without confusing it with the indicated rate. If the observed indication changes by \(\Delta r_o\), then:
Here \(r_o\) denotes the observed indication and \(\hat{r}\) denotes the resulting estimate. This relation assumes a linear weighting approach and unchanged other inputs; it is an interpretation of the weight, not a rule for selecting \(Z\). It follows by comparing the estimate before and after the observed indication changes. The part of the estimate associated with the observed indication changes in proportion to its assigned weight, so a change of \(\Delta r_o\) contributes \(Z\Delta r_o\). At \(Z=0\), that change contributes nothing; at \(Z=1\), the full change passes through. For intermediate values, only the corresponding fraction passes through.
This sensitivity interpretation is conditional. If the credibility procedure changes, other inputs change, or the estimation is not linear, the relation need not describe the total change in an estimate. It also does not specify how to calculate \(Z\). The appropriate method for assigning credibility depends on the problem and its assumptions.
Worked Examples
Worked Example: Separating Indication Level From Assigned Weight
A hypothetical equipment insurer reviews two groups on a consistent, fully earned basis. Group A has $144,000 of losses over 1,200 exposure units and an assigned credibility of 0.25. Group B has $90,000 of losses over 500 exposure units and an assigned credibility of 0.80. Calculate each observed loss cost and interpret the two quantities separately. Assume the loss amounts and exposures are on the same basis and require no trend or other adjustment.
Method. Calculate observed loss cost as total losses divided by exposure units, then compare that rate level with the separately provided credibility weight.
For Group A, the observed loss cost is:
The assigned credibility is \(Z=0.25\). Thus the experience indicates $120 per exposure unit, while the credibility assessment assigns that experience a 25% weight in the applicable estimation approach.
For Group B:
The assigned credibility is \(Z=0.80\). This experience indicates $180 per exposure unit and receives an 80% weight under the specified approach.
Check. Multiplying each calculated rate by its exposure reproduces its loss total: \(\$120(1{,}200)=\$144{,}000\) and \(\$180(500)=\$90{,}000\). Group B has both the higher observed indication and the higher assigned credibility in this example. That coincidence does not mean a high indicated rate causes high credibility; the problem supplied the weights separately.
Common mistake avoided. Do not call $180 “80% credibility” or treat 0.80 as a loss cost. The first is a rate level in dollars per exposure; the second is a dimensionless weight.
Worked Example: Same Indication, Different Credibility
Two hypothetical coverage groups each have an observed loss cost of $75 per earned exposure unit. A specified procedure assigns Group C a credibility of 0.35 and Group D a credibility of 0.90. For each group, identify the indicated level and the weight. Then explain how a $10 change in the observed loss cost would affect the estimate under a linear approach if all other inputs remained fixed.
Method. Keep the observed rate level separate from the given weight. Apply the sensitivity relation for a linear approach only to the stated change, holding the other inputs fixed.
For Group C, the observed indication is $75 per exposure unit and \(Z=0.35\). A $10 increase contributes:
For Group D, the observed indication is also $75 per exposure unit, but \(Z=0.90\). The same $10 increase contributes:
The groups have the same observed rate level, but different assigned weights. Under the stated linear and fixed-input conditions, a change in Group D's observed indication has a larger effect on its estimate. This comparison does not calculate either group's full estimated rate, because the other inputs and the full estimation procedure have not been specified.
Check. The calculated responses are smaller than the $10 change for both groups, as expected for weights between 0 and 1. Group D's response is larger because \(0.90>0.35\).
Common mistake avoided. Do not infer that Group D must have a higher rate because it has higher credibility. The two observed indications are identical; the difference is how strongly each experience would influence an estimate under the stated approach.
Worked Example: A Rising Indication Is Not a Rising Weight
A hypothetical repair portfolio has $100,000 of losses over 1,000 earned exposure units in one period and $180,000 over 1,200 earned exposure units in a later period. A consistent procedure assigns both periods a credibility of 0.60. Assume amounts are comparable, exposures are correctly earned, and no adjustments are needed. Calculate the two indications and, under a linear approach with other inputs held fixed, the effect of the change in indication.
Method. Divide losses by earned exposures for each period. Then apply the sensitivity relation to the difference between the indications, not to the credibility value.
The first period's indication is:
The later period's indication is:
The indication has increased by \(\$150-\$100=\$50\) per exposure unit. With \(Z=0.60\), and with other inputs fixed in the same linear approach, the change contributes:
Check. The observed indications are calculated using the period-specific exposure totals; using the $180,000 loss total divided by the earlier 1,000 exposures would give the wrong later-period rate. The procedure assigns the same credibility in both periods, even though the indicated level changes. The $30 is the change attributable to the observed indication under the stated conditions, not the full estimate for either period.
Common mistake avoided. A larger observed rate does not automatically mean larger credibility. Here, the indication rises by $50 while the assigned credibility remains 0.60.
What Credibility Does Not Establish
Credibility is meaningful only in relation to an estimation approach and the experience being considered. A value of 0.60 describes a weight in that framework; it is not a universal statement that the experience is “60% reliable” in every sense. It does not, by itself, identify why the weight has that value, prove that a portfolio is homogeneous, or establish that the indicated rate is appropriate for a future period.
In particular, avoid these interpretations:
- Not a rate level: \(Z=0.60\) is not $0.60 per exposure, a 60% loss ratio, or a 60% rate change.
- Not a probability of correctness: \(Z=0.60\) does not mean there is a 60% chance that the observed indication is correct.
- Not determined by the direction of the indication: A high or rising observed rate does not, on its own, imply high or rising credibility.
- Not a complete estimate: Knowing \(Z\) and the observed indication does not supply all information needed to determine an estimated rate component.
A sound interpretation reports both the observed level and its assigned credibility, with units for the rate and a clear identification of the procedure behind the weight. This is useful whenever experience is being assessed for ratemaking: it prevents a striking loss result from being mistaken for strong evidence, and it prevents a modest indication from being mistaken for weak evidence.
Check Your Understanding
For each problem, keep the rate level and credibility weight distinct. Assume all stated exposures and losses are on a comparable basis.
- A hypothetical roadside-assistance portfolio has $84,000 of losses over 700 earned exposure units and an assigned credibility of 0.30. What is the observed loss cost, and what does 0.30 describe? (Answer: $120 per exposure unit; the weight assigned to the observed experience under the specified approach.)
- Two hypothetical groups each indicate $240 per exposure unit. Their assigned credibility values are 0.20 and 0.75. Which group has the higher indicated rate, and which has the higher credibility? (Answer: Neither has the higher indicated rate; both indicate $240 per exposure unit. The second group has higher credibility, 0.75.)
- A hypothetical portfolio's observed indication increases from $90 to $110 per exposure unit while assigned credibility remains 0.50. Under a linear approach with all other inputs fixed, what change is attributable to the observed indication? (Answer: $10 per exposure unit, since \(0.50(\$110-\$90)=\$10\).)
- A group has $156,000 of losses over 1,300 earned exposure units and an assigned credibility of 0.40. A colleague says the indication is “40% reliable” because it is $120 per exposure. Identify the observed rate and correct the interpretation. (Answer: The observed rate is $120 per exposure unit; 0.40 is the assigned weight, not a probability or percentage measure of accuracy.)
- Two hypothetical periods indicate $100 and $160 per exposure unit. The assigned credibility is 0.25 in each period. Under a linear approach with other inputs fixed, calculate the change attributable to the observed indication and state whether credibility changed. (Answer: $15 per exposure unit; credibility did not change and remains 0.25.)