Tutorials › CAS Exam 8 - Advanced Ratemaking › The Square-Root Rule for Partial Credibility

Home  /  Tutorials  /  CAS Exam 8 / Credibility and Limited Fluctuation

CAS Exam 8 · Credibility and Limited Fluctuation · Tutorial 296 of 989

The Square-Root Rule for Partial Credibility

Learn why partial credibility scales with the square root of exposure, how to calculate and apply the factor, and when the rule’s assumptions matter.

Beginner 10 min read

What You'll Learn

  • Define the full-credibility exposure and the partial-credibility factor
  • Derive the square-root scaling from how sampling error changes with exposure
  • Calculate partial credibility from available and required exposure
  • Cap the factor at one when available exposure meets or exceeds the standard
  • Apply credibility to blend an experience rate with a benchmark rate
  • Identify when the rule is inappropriate or the exposure bases do not match

Why Partial Credibility Uses a Square Root

A subject risk’s observed experience is less stable when it is based on fewer exposure units. Limited-fluctuation credibility expresses that difference with a credibility factor: a weight on the subject’s experience that increases with exposure, up to full credibility. The square-root rule gives a simple scaling when the required exposure for full credibility is known.

The essential distinction is between the exposure ratio and the credibility factor. If the subject has one quarter of the full-credibility exposure, the square-root rule assigns one half credibility, not one quarter. The factor is capped at one because more than full credibility is not assigned by this rule.

Definition: Let \(E\) be the subject’s available exposure, and let \(E_F\) be the exposure required for full credibility under a specified limited-fluctuation standard. When both quantities use the same exposure basis, the limited-fluctuation partial credibility factor is $$ Z=\min\left(1,\sqrt{\frac{E}{E_F}}\right). $$ If \(E<E_F\), then \(Z=\sqrt{E/E_F}\). If \(E\ge E_F\), then \(Z=1\).

The exposure basis must be consistent. For example, if \(E_F\) is stated in policy-years, the available exposure must also be measured in policy-years of the same type. This tutorial takes the full-credibility standard as given; the standard that establishes \(E_F\) depends on the specified fluctuation criterion and is a separate calculation.

Why the Square-Root Scaling Holds

The rule follows from the way uncertainty in an average changes as independent, comparable observations accumulate. Suppose each exposure unit produces a random observation with mean \(\mu\) and variance \(\sigma^2\), and the observations are independent and identically distributed. The sample average over \(E\) units has variance \(\sigma^2/E\), so its standard deviation is \(\sigma/\sqrt{E}\).

Compare the standard deviation at available exposure \(E\) with the standard deviation at full-credibility exposure \(E_F\). Their ratio is

$$ \frac{\sigma/\sqrt{E}}{\sigma/\sqrt{E_F}} =\sqrt{\frac{E_F}{E}}. $$

Thus, standard error decreases in proportion to the square root of exposure, not in proportion to exposure itself. Equivalently, precision relative to the full-credibility level scales as

$$ \frac{\sigma/\sqrt{E_F}}{\sigma/\sqrt{E}} =\sqrt{\frac{E}{E_F}}. $$

Limited-fluctuation partial credibility uses this relative-precision scaling as its factor, until the full standard is reached. At or above \(E_F\), the factor is set to one. This justification relies on independent, comparable exposure units with stable finite variance, so that uncertainty in the average declines at the square-root rate. Material dependence, substantial changes in the risk mix, or a different variance pattern can make the simple rule inappropriate.

Important condition: The square-root rule scales from a full-credibility standard already established on the same basis and under the same fluctuation criterion. It does not itself determine that standard, prove that the standard is suitable, or adjust for changes in the risk population.

Calculating and Using the Factor

The calculation has two steps: divide available exposure by full-credibility exposure, then take the square root and cap the result at one. If the factor is used to blend an observed subject rate with a benchmark rate, the credibility-weighted rate is

$$ \text{indicated rate}=Z(\text{subject rate})+(1-Z)(\text{benchmark rate}). $$

The subject experience receives weight \(Z\), and the benchmark receives the complement \(1-Z\). The complement is not another estimate of credibility; it is the remaining weight needed to make the two weights sum to one.

Worked Example: One Quarter of the Full-Credibility Exposure

A hypothetical small-business portfolio has 10,000 policy-years of experience. Its supplied full-credibility standard is 40,000 policy-years. The observed experience is 120 claims, and the benchmark claim rate is 0.0108 claims per policy-year. For illustration, assume independent Bernoulli claim indicators per policy-year with a common probability \(p=0.012\); the 120 observed claims are the data to be rated. Find the credibility factor and the blended claim rate.

Method. Use the square-root rule, then apply the factor to the subject claim rate. The supplied full standard is used as given; it is not recalculated from the Bernoulli model.

First calculate the subject claim rate:

$$ \text{subject rate}=\frac{120}{10{,}000}=0.0120. $$

Next calculate the exposure ratio and its square root:

$$ \frac{E}{E_F}=\frac{10{,}000}{40{,}000}=0.25, \qquad Z=\sqrt{0.25}=0.50. $$

The complement is \(1-Z=0.50\). Therefore,

$$ \text{indicated rate} =(0.50)(0.0120)+(0.50)(0.0108) =0.0060+0.0054 =0.0114. $$

The credibility factor is 0.50, and the blended rate is 0.0114 claims per policy-year. As a check, the result lies halfway between 0.0120 and 0.0108, as it should when the weights are equal. The common mistake to avoid is using the exposure ratio, 0.25, as the credibility factor; that would give the wrong subject weight.

Worked Example: Converting a Partial Factor Into a Rate Indication

A hypothetical equipment insurer has 18,000 exposure units on a particular class, with a full-credibility standard of 72,000 units. The class has 186 reported claims. Its benchmark rate is 0.0095 claims per exposure unit. Assume independent Bernoulli claim indicators per unit with a common probability \(p=0.010\), and use the supplied full standard without attempting to derive it. Calculate the credibility-weighted rate.

Method. Calculate the observed rate, find \(Z\), and blend the two rates using \(Z\) and \(1-Z\).

The observed rate is

$$ \frac{186}{18{,}000}=0.010333\ldots $$

The exposure ratio is \(18{,}000/72{,}000=0.25\), so

$$ Z=\sqrt{0.25}=0.50, \qquad 1-Z=0.50. $$

The credibility-weighted rate is

$$ \begin{aligned} \text{indicated rate} &=(0.50)(0.010333\ldots)+(0.50)(0.0095)\\ &=0.0051666\ldots+0.00475\\ &=0.0099166\ldots \end{aligned} $$

To five decimal places, the indicated rate is 0.00992 claims per exposure unit. It is between the observed rate and benchmark rate, as a weighted average must be. The common mistake to avoid is rounding the observed rate too early; keeping the unrounded value through the blend gives the stated final result.

Worked Example: Combining Comparable Periods and Applying the Cap

A hypothetical watercraft portfolio’s experience is reported in three successive, comparable periods: 14,000, 9,000, and 2,000 exposure units. Treat the periods as independent and assume the same risk definition and exposure basis throughout. The supplied full-credibility standard is 40,000 units. There were 320 claims in total, and the benchmark rate is 0.0115 claims per unit. For illustration, model each unit’s claim indicator as Bernoulli with common probability \(p=0.012\). Find the credibility-weighted rate, and state what happens if a further 20,000 comparable units are added.

Method. Add comparable exposure and claims before calculating the aggregate observed rate. Apply the cap separately after calculating the factor.

Total available exposure and the observed rate are

$$ E=14{,}000+9{,}000+2{,}000=25{,}000, \qquad \text{subject rate}=\frac{320}{25{,}000}=0.0128. $$

The exposure ratio is

$$ \frac{E}{E_F}=\frac{25{,}000}{40{,}000}=0.625. $$

Taking the square root gives \(Z=\sqrt{0.625}=0.790569\ldots\), which is below one, so the cap does not change it. The complementary weight is \(1-Z=0.209430\ldots\). The blended rate is

$$ \begin{aligned} \text{indicated rate} &=(0.790569\ldots)(0.0128) +(0.209430\ldots)(0.0115)\\ &=0.0115+(0.790569\ldots)(0.0013)\\ &=0.01252774\ldots \end{aligned} $$

The indicated rate is 0.01253 claims per unit to five decimal places. If 20,000 additional comparable units are added, total exposure becomes \(25{,}000+20{,}000=45{,}000\). The uncapped square root would be \(\sqrt{45{,}000/40{,}000}=\sqrt{1.125}=1.06066\ldots\), so the capped factor is \(Z=1\). The common mistake to avoid is either leaving the factor above one or treating only the added period as the subject’s total exposure.

Exam Traps and Practical Limits

Several checks help prevent errors when applying the rule:

  • Do not omit the square root. If \(E/E_F=0.36\), the factor is 0.60, not 0.36.
  • Do not let the factor exceed one. Once \(E\ge E_F\), use \(Z=1\), not a value above one.
  • Keep the exposure bases aligned. A numerator and denominator stated in different kinds of exposure cannot be divided meaningfully without a valid conversion.
  • Use the correct weights. In a blend, the subject experience receives \(Z\); the benchmark receives \(1-Z\).
  • Do not confuse the factor with a claim rate. The factor is a dimensionless weight between zero and one; the indicated rate retains the units of the rates being blended.

The square-root rule is a scaling method, not a universal correction for any small data set. Its result depends on the relevance of the supplied full standard and on the comparability of exposure. If a class’s underlying risks or claim process change across periods, simply adding exposures may not produce a sound credibility assessment. A computed factor should therefore be interpreted within the standard and assumptions that support it.

Key takeaway: For comparable exposure measured on the same basis, calculate \(Z=\min(1,\sqrt{E/E_F})\). The square root reflects how standard error changes with exposure; the cap ensures credibility does not exceed full credibility under this rule.

Check Your Understanding

For each problem, use the stated full-credibility standard as given. Unless otherwise specified, assume comparable independent exposure units and use the square-root rule.

  1. A hypothetical roadside-assistance portfolio has 3,600 exposure units and a full standard of 14,400 units. What is its partial credibility factor? (Answer: 0.50)
  2. A hypothetical artisan-liability segment has 8,000 exposure units and a full standard of 32,000. Its observed rate is 0.015, and the benchmark is 0.012. Find the factor and blended rate. (Answer: \(Z=0.50\); rate \(=0.0135\))
  3. A hypothetical travel policy group has 27,000 units and a full standard of 48,000 units. Find the factor to four decimal places. (Answer: \(Z=\sqrt{27{,}000/48{,}000}=0.7500\))
  4. A hypothetical marine class has 30,000 units against a full standard of 20,000. What factor should be used, and why? (Answer: \(Z=1\), because the uncapped ratio exceeds one and the rule caps credibility at full credibility.)
  5. A hypothetical device-protection portfolio has two comparable periods with 12,000 and 13,000 units, a full standard of 50,000 units, 275 claims, and a benchmark rate of 0.010 claims per unit. Assuming Bernoulli claim indicators with common probability \(p=0.011\), calculate the factor and blended rate. (Answer: \(Z=\sqrt{25{,}000/50{,}000}=0.7071\); subject rate \(=0.0110\); blended rate \(=0.0107\) claims per unit.)

Previous

Next

Back to Tutorials
Clinical Biostats