Actuarial Statistics · Credibility theory: Bayesian and empirical Bayes credibility premiums

Credibility Theory Basics and Limited Fluctuation Credibility

Updated 11 October 2026 · Fact-checked

Credibility theory combines an insurer's own claim experience with a wider benchmark: credibility premium = Z × own experience + (1 − Z) × benchmark. Under limited fluctuation, you find the full credibility standard, then set Z = 1 if exposure meets it, otherwise Z = √(n ÷ n_F).

Understand Credibility Theory Basics and Limited Fluctuation

Credibility theory answers one question: how much should you trust a group's own claim experience when setting its premium?

Suppose a small group of employees had low claims last year. Should you cut their premium to match? Probably not. Small groups have noisy experience. A better estimate blends the group's own experience with the average experience of a larger population, called the manual premium or prior estimate.

The blend is the credibility premium: P = Z × X̄ + (1 − Z) × M, where X̄ is the group's own mean experience, M is the manual (collateral) estimate, and Z is the credibility factor with 0 ≤ Z ≤ 1. A large Z means you trust the group's own data. A small Z means you lean on the benchmark. Z rises with more data and with less variable data.

The limited fluctuation (classical) approach sets a standard for how much data you need so that the observed experience is close to the true value with high probability. If the experience is at least within a fraction k of the true mean with probability p, you give it full credibility (Z = 1). If you have less data, you give partial credibility, with Z less than 1 and growing with the square root of the data volume.

The limited fluctuation method is simple but has weaknesses. The choice of p and k is arbitrary, and it does not use the benchmark's own variability. The Bayesian and Bühlmann methods in related topics fix these gaps.

Key rules to remember

Credibility premium: P = Z × X̄ + (1 − Z) × M; X̄ is the group's own mean experience, M is the manual or prior estimate, 0 ≤ Z ≤ 1.
Full credibility condition (mean of n i.i.d. observations): P(|X̄ − μ| ≤ k·μ) ≥ p; Using the normal approximation, this holds when n ≥ (y ÷ k)² × (σ² ÷ μ²), where y satisfies Φ(y) = (1 + p) ÷ 2.
Full credibility standard for claim numbers (Poisson): n_F = (y ÷ k)² (expected number of claims); Applies when the number of claims is Poisson with mean λ, so λ_F = (y ÷ k)². Here the standard is in expected claims.
Full credibility standard for aggregate claims (compound Poisson): λ_F = (y ÷ k)² × (1 + CV²), where CV = σ_Y ÷ μ_Y of the claim size; Equivalently λ_F = (y ÷ k)² × E[Y²] ÷ (E[Y])². The standard is in expected number of claims.
Partial credibility (square root rule): Z = √(n ÷ n_F), capped at 1; n and n_F must be in the same units, such as number of claims or exposure years.
Common normal values for y: p = 90%: y = 1.645; p = 95%: y = 1.960; p = 99%: y = 2.576; y is the standard normal point with Φ(y) = (1 + p) ÷ 2, a two-sided probability.

How to solve Credibility Theory Basics and Limited Fluctuation questions

Use this method for any limited fluctuation question. It works whether you are asked for the standard, Z or the premium.

1Identify the quantity being estimated: claim frequency, claim count, severity or aggregate claims.
2Read p and k from the question. Find y from Φ(y) = (1 + p) ÷ 2.
3Compute the base standard (y ÷ k)². This is the full credibility standard for expected Poisson claim numbers.
4If severity varies, multiply by (1 + CV²), or by E[Y²] ÷ (E[Y])². This gives the standard for aggregate claims.
5Convert the standard into the units of the data if asked, for example exposure years = λ_F ÷ claim rate per year.
6Compare the data volume n with the standard n_F. If n ≥ n_F, Z = 1. Otherwise Z = √(n ÷ n_F).
7Compute P = Z × X̄ + (1 − Z) × M and state the result with units.
8Check: Z lies between 0 and 1, and P lies between X̄ and M.

Quickest way: Standard, ratio, square root, blend

When to use it: Use when the question gives p, k, data volume, own experience and a manual premium, and asks for a credibility premium.

Write y for the given p: 1.645 for 90%, 1.960 for 95%, 2.576 for 99%.
Compute n_F = (y ÷ k)², times (1 + CV²) if severity is involved.
Compute Z = √(n ÷ n_F). If the ratio is 1 or more, stop and use Z = 1.
Compute P = M + Z × (X̄ − M). This form takes less arithmetic and shows the adjustment directly.

Common mistakes in Credibility Theory Basics and Limited Fluctuation

Using the wrong y, such as 1.645 for 95%.
Students read p as a one-sided probability.
Fix: The condition is two-sided. Use Φ(y) = (1 + p) ÷ 2, so 95% gives 1.960 and 90% gives 1.645.
Forgetting to square: writing the standard as y ÷ k.
The formula looks like a simple ratio, so the square is dropped.
Fix: Write n_F = (y ÷ k)² every time. For p = 90% and k = 5%, it is (1.645 ÷ 0.05)² = 1,082.41, not 32.9.
Leaving out (1 + CV²) for aggregate claims.
The claim number standard is memorised and applied to all questions.
Fix: Check whether the question concerns claim numbers or total claim amounts. If amounts, include severity variability.
Mixing units, such as comparing exposure years with claim counts.
The standard is in expected claims but the data is given in policy-years.
Fix: Convert the standard into the data's units first, or convert the data into expected claims, before taking the ratio.
Applying Z = √(n ÷ n_F) when n is above n_F.
Students apply the formula mechanically.
Fix: Cap Z at 1. If n ≥ n_F, the data has full credibility.
Swapping the weights, giving Z to the manual premium.
The roles of X̄ and M are confused under time pressure.
Fix: Z always multiplies the group's own experience. Sense-check: more data should move P towards X̄.

Worked examples

Example 1

Claim numbers for a portfolio are Poisson. Full credibility requires the observed mean to be within 5% of the true mean with probability 90%. (a) Find the full credibility standard in expected claims. (b) A portfolio has 300 expected claims. Find Z. (c) Its observed mean aggregate claim is ₹80,000 per policy and the manual premium is ₹60,000. Find the credibility premium.

Show the solution

For p = 90%, Φ(y) = 0.95, so y = 1.645.
(a) n_F = (1.645 ÷ 0.05)² = 32.9² = 1,082.41, about 1,082 expected claims.
(b) Z = √(300 ÷ 1,082.41) = √0.27717 = 0.5265.
(c) P = Z × 80,000 + (1 − Z) × 60,000 = 60,000 + 0.5265 × 20,000.
P = 60,000 + 10,530 = ₹70,530.

Answer: (a) about 1,082 expected claims; (b) Z ≈ 0.527; (c) credibility premium ≈ ₹70,530.

Example 2

Aggregate claims are compound Poisson. Claim sizes have mean ₹20,000 and standard deviation ₹30,000. Full credibility requires aggregate claims to be within 10% of the mean with probability 95%. Find the expected number of claims needed for full credibility, and Z for a group with 150 expected claims.

Show the solution

For p = 95%, y = 1.960. k = 0.10.
Base standard: (1.960 ÷ 0.10)² = 19.6² = 384.16.
CV = 30,000 ÷ 20,000 = 1.5, so 1 + CV² = 1 + 2.25 = 3.25.
λ_F = 384.16 × 3.25 = 1,248.52.
Z = √(150 ÷ 1,248.52) = √0.12014 = 0.3466.

Answer: Full credibility needs about 1,249 expected claims; Z ≈ 0.347 for 150 expected claims.

Exam tips

Write y and the line Φ(y) = (1 + p) ÷ 2 at the start. Marks are often given for the correct y.
State your units for n_F: claims, policy-years or amount. Examiners look for consistent units.
Always say what the formula assumes, such as Poisson claim numbers and the normal approximation.
In written answers, comment on weaknesses of limited fluctuation: arbitrary p and k, and no use of the manual premium's variance. This links to Bühlmann credibility.
In MCQs, check the answer lies between X̄ and M. This eliminates wrong options quickly.

Practice questions from Credibility theory: Bayesian and empirical Bayes credibility premiums

Credibility Theory Basics and Limited Fluctuation: frequently asked questions

What is a credibility premium in actuarial statistics?

It is a weighted average of a group's own observed experience and a benchmark or manual premium. The weight on the group's own experience is the credibility factor Z. It lies between 0 and 1.

How do I calculate partial credibility Z using the square root rule?

Find the full credibility standard n_F, then compute Z = √(n ÷ n_F), where n is your data volume in the same units. If n is at least n_F, Z is capped at 1.

What is the full credibility standard in limited fluctuation?

It is the amount of data needed so the observed experience is within a fraction k of the true mean with probability p. For Poisson claim numbers it is (y ÷ k)² expected claims, where Φ(y) = (1 + p) ÷ 2.

Why is limited fluctuation credibility criticised?

The choices of p and k are arbitrary, and the method does not use the variability of the benchmark or a model of how risks differ. Bayesian and Bühlmann credibility address these points.