Choosing The Rate Component To Credibilize

Why the Credibility Target Matters

A credibility-weighted estimate is meaningful only after the quantity being estimated has been identified. A territory may have its own claim frequency, a class may have a distinct average severity, and a coverage may have a loss cost that reflects both. Those quantities are related, but they are not interchangeable. Applying a credibility weight to frequency does not directly produce a credible severity or loss cost.

The practical question is therefore not only, “How much credibility does this experience receive?” It is also, “Credibility for which quantity?” The answer should follow the rate component that the estimate is intended to update. This distinction is aligned with the CAS Exam 5 (Basic Ratemaking and Estimated Claim Liabilities) syllabus.

Let \(\theta_{\rm obs}\) be the observed estimate of the target component, \(\theta_{\rm comp}\) its complementary estimate, and \(Z\) the credibility assigned to the observed estimate. The weighted estimate is:

This formula assumes that the two estimates refer to the same defined quantity on a compatible basis. For example, observed territory frequency should be blended with complementary territory frequency—not with a dollar severity or a premium amount. The value of \(Z\) is taken as given in the examples; selecting or deriving credibility weights is a separate question.

Frequency, Severity, and Loss Cost

Frequency and severity describe different aspects of claims. If \(N\) is the claim count for a group with \(E\) exposure units, observed frequency is \(N/E\). If the total loss is \(L\) across \(N\) claims, observed severity is \(L/N\), provided \(N>0\). Observed loss cost is \(L/E\). The units make their differences clear: frequency is claims per exposure, severity is dollars per claim, and loss cost is dollars per exposure.

When the claim definition and loss basis are consistent, the observed loss cost factors into observed frequency times observed severity:

This identity motivates component-level estimation, but it does not mean that every credibility estimate must be made separately for frequency and severity. The target may instead be loss cost directly. The choice depends on the rate component that is intended to vary and on whether the available observations and complementary information support that choice.

A useful selection rule is to name the target before calculating. If the question concerns how often a territory produces claims, estimate territory frequency. If it concerns the average cost of claims for a class, estimate class severity. If it concerns the expected covered loss per exposure under specified policy terms, estimate that coverage loss cost. In each case, the complementary estimate must have the same meaning, units, coverage, and relevant period basis as the observed estimate.

Credibilizing Frequency While Holding Severity Separate

Frequency may be the component of interest when a group is expected to differ in claim incidence. In that case, the observed frequency and complementary frequency are blended. A severity assumption can then be used separately to translate the frequency estimate into a loss cost. That translation is not a reason to treat the frequency credibility weight as a severity credibility weight.

Separate Component Estimates and Their Product

A rate analysis may call for separate frequency and severity estimates. In that case, each component is blended with its own complementary estimate and, where appropriate, its own credibility weight. Multiplying the resulting estimates gives an implied loss cost. This approach preserves the possibility that the available experience supports frequency differently from severity.

The resulting product should not be confused with a direct credibility blend of aggregate loss costs. In general, blending components separately and multiplying is not algebraically the same as blending observed and complementary loss costs with one weight. Each method answers a potentially different estimation question.

Credibilizing Loss Cost Directly

Sometimes the intended target is the combined loss cost for a particular coverage, rather than frequency or severity separately. Then observed losses per exposure are blended with a complementary loss cost per exposure. Both figures must reflect the same coverage and loss definition. For example, a complementary value based on losses before a deductible is not automatically compatible with observed payments after that deductible.

Choosing the Target in Practice

The choice of component should reflect the intended estimate, not simply which calculation is easiest. A useful sequence is to define the rate question, identify the relevant component, and check that observed and complementary figures share its definition. The same group can have a frequency estimate suitable for one question and a loss-cost estimate suitable for another.

The most consequential trap is a plausible-looking number that estimates the wrong quantity. A frequency blend cannot be reported as a loss cost; a severity complement cannot be blended with frequency; and a loss-cost complement under different policy terms may not be suitable even if its dollar units appear compatible. A second trap is assuming that separately credibilized frequency and severity must reproduce a direct loss-cost blend. The two approaches generally differ because they blend different targets, possibly with different weights.

For exam work and professional analysis, label each estimate with both its component and its units. “0.054 claims per exposure” is more informative than “0.054,” and “$131 per exposure under the stated deductible” is more precise than “$131.” Those labels make incompatible calculations easier to detect before they affect a rate indication.