Limited Fluctuation Credibility Calculator
Enter a confidence level, tolerance, claim count, severity variation, and complement estimate to calculate the limited fluctuation credibility factor and the blended estimate.
Credibility factor0.61%
- Normal critical value
- 1.64
- Frequency full-credibility standard
- 271
- Selected full-credibility standard
- 542
- Exposures for full credibility
- 10,823
- Weight on complement
- 0.39%
- Credibility-weighted estimate
- $448.64
- Claims required for target credibility
- 305
- Additional claims required
- 105
- Exposures required for target credibility
- 6,088
Uses a normal approximation to limited fluctuation credibility with Poisson claim frequency.
Aggregate-loss credibility assumes claim counts and claim severities are independent.
How to use this calculator
- Choose the confidence level and acceptable deviation for the full-credibility standard.
- Select claim-count or aggregate-loss credibility, then enter the severity coefficient of variation if aggregate losses matter.
- Enter observed claims, expected claim frequency, the subject estimate, and the complement estimate.
- Set the target credibility to see the total and additional claims or exposures required.
How limited fluctuation credibility is calculated
Limited fluctuation credibility asks how much experience is needed so an observed result is likely to stay within a chosen tolerance of its expected value. The calculation starts with a confidence probability and the acceptable deviation. A 90% confidence level with a 10% tolerance uses the standard-normal critical value at (1 + 0.90) / 2, which is about 1.645.
For claim-count credibility under a Poisson model, the full-credibility standard is:
Nfreq = (z / k)^2
- z is the standard-normal critical value.
- k is the tolerance as a decimal.
- Nfreq is the claim count needed for full claim-count credibility.
For aggregate loss, claim-size variation increases the standard. The calculator applies the common independent frequency and severity adjustment:
Nagg = Nfreq x (1 + c^2)
Here c is the coefficient of variation of claim severity. A coefficient of variation of 1 doubles the claim-count standard. A value of 0 makes the aggregate-loss standard equal to the claim-count standard.
How partial credibility and blending work
The selected standard is the claim-count standard for frequency work, or the severity-adjusted standard for aggregate-loss work. Partial credibility follows the square-root rule:
Z = min(1, sqrt(n / Nfull))
The weighted estimate is then P = Z x X + (1 - Z) x M, where X is the subject experience estimate and M is the complement. The complement should be measured on the same basis as the subject estimate, such as a manual rate, class rate, prior estimate, or larger credible pool.
What moves the result most
A smaller tolerance sharply increases the full-credibility standard because the tolerance is squared in the denominator. Higher confidence also raises the standard through the normal critical value. For aggregate losses, higher severity variability raises the standard directly through 1 + c^2. More observed claims increase the credibility factor, but only by the square root, so doubling the claims does not double credibility.
What this calculator leaves out
The formulas assume a normal approximation, Poisson claim frequency, independent claim counts and severities, and a single severity coefficient of variation. They do not model trend, development, exposure mix, policy limits, deductibles, catastrophe load, dependence between claims, or Buhlmann and Bayesian credibility procedures. The exposure conversion also assumes the expected claim frequency you enter is appropriate for the same exposure unit used in the analysis.
Worked example
Suppose the selected basis is aggregate loss, confidence is 90%, tolerance is 10%, the severity coefficient of variation is 1, observed claims are 200, expected claim frequency is 5%, the subject estimate is $480, the complement is $400, and the target credibility is 75%.
The normal critical value is 1.645. The claim-count full-credibility standard is (1.645 / 0.10)^2, or 271 claims after rounding up. The aggregate-loss standard doubles that to 542 claims because 1 + 1^2 = 2. With 200 observed claims, the credibility factor is about 60.8%, leaving 39.2% on the complement. The blended estimate is about $448.64.
For a 75% target, the required claim count is 0.75^2 x 541.11, or 305 claims after rounding up. Since 200 claims are already observed, 105 additional claims are needed. At a 5% claim frequency, full credibility needs 10,823 exposures, and the 75% target needs 6,088 exposures.
Common questions
What is limited fluctuation credibility?
Limited fluctuation credibility is a way to decide how much experience is needed before an observed result is likely to be close enough to its expected value. The user chooses the confidence level and tolerance, and those choices define the full-credibility standard.
How many claims are needed for full credibility?
For claim-count credibility, the standard is the square of the normal critical value divided by the tolerance. At 90% confidence and 10% tolerance, that standard is 271 claims after rounding up. Aggregate-loss credibility can require more claims when claim severities are variable.
Why does severity variability increase the credibility standard?
Aggregate losses vary because claim counts vary and because claim sizes vary. The adjustment multiplies the claim-count standard by one plus the square of the severity coefficient of variation, so less stable claim sizes require more claims for the same fluctuation limit.
What is the square-root rule for partial credibility?
The square-root rule sets credibility equal to the square root of observed claims divided by the selected full-credibility standard, capped at 100%. It gives partial weight to the subject experience before the full standard is reached.
What should I use as the complement of credibility?
Use a relevant estimate measured on the same basis as the subject experience, such as a manual rate, class average, prior rate level, or larger credible pool. The complement should represent the part of the estimate not supported by the subject experience.
Why is credibility capped at 100%?
Once observed claims meet or exceed the selected full-credibility standard, the square-root rule would otherwise produce a value above one. Capping the factor at 100% means the estimate receives full weight but never more than full weight.