Why a Full Standard Leads to Partial Credibility
A full-credibility standard specifies how much experience is needed to meet a chosen precision criterion. In practice, an estimate may be based on less than that required volume. The partial-credibility factor translates the shortfall in volume into a proportional weight for the experience estimate. The key is that sampling error decreases with the square root of volume, not in direct proportion to volume.
This tutorial derives that relationship from a common sampling model. The derivation is aligned with the CAS Exam 5 (Basic Ratemaking and Estimated Claim Liabilities) syllabus. It assumes the full standard and the observed experience refer to the same measure, unit of volume, and precision criterion.
The volume may be a count of comparable observations or an exposure measure; what matters here is that the full standard and the observed experience use the same volume basis. This tutorial uses \(N\) generically and does not assume a particular claim-count, severity, or loss-ratio standard.
Deriving the Square-Root Relationship
Suppose an estimate is the sample mean of \(N\) independent, identically distributed observations. Let each observation have mean \(\mu\) and variance \(\sigma^2\), with \(\sigma\) finite. Under this model, the sample mean has variance \(\sigma^2/N\), so its standard error is \(\sigma/\sqrt{N}\). If a normal approximation is used, a fixed confidence multiplier \(k\) gives a precision half-width of \(k\sigma/\sqrt{N}\).
At volume \(N_0\), suppose the specified precision half-width is \(h_0=k\sigma/\sqrt{N_0}\). The confidence multiplier and the variability are fixed for this comparison.
At volume \(N\), the same model and criterion give \(h_N=k\sigma/\sqrt{N}\).
For \(N<N_0\), the ratio of the full-standard half-width to the observed half-width is $$ \frac{h_0}{h_N} =\frac{k\sigma/\sqrt{N_0}}{k\sigma/\sqrt{N}} =\sqrt{\frac{N}{N_0}}. $$ The common factors \(k\) and \(\sigma\) cancel.
The experience has the fraction \(\sqrt{N/N_0}\) of the full-standard precision on this scale. Thus the partial factor is \(Z=\sqrt{N/N_0}\) below the standard. It reaches 1 at \(N=N_0\); additional volume does not produce a credibility factor above 1.
The square root appears because variance decreases as \(1/N\), while standard error and a normal-approximation precision half-width decrease as \(1/\sqrt{N}\). If observed volume is one quarter of the required volume, for example, its standard error is twice the full-standard standard error, and its partial factor is one half—not one quarter.
In a common credibility-weighting application, the experience estimate is combined with a complement \(C\) as \(Z\) times the experience estimate plus \(1-Z\) times \(C\). The formula for \(Z\) determines the weight; it does not by itself specify how to choose the complement or establish that the experience and complement are unbiased.
Worked Calculations
Worked Example: A Quarter of the Required Volume
A hypothetical portfolio’s normalized loss measure is estimated from independent, comparable observations \(X_i\), where \(X_i\sim N(\mu,100)\). The variance is 100, so the standard deviation is 10. A two-sided 95% normal criterion uses \(k=1.96\), and the full standard requires \(N_0=400\) observations. The current experience contains \(N=100\). Find the partial factor and compare the precision half-widths. No interest or mortality assumption applies.
Method. Derive the factor by comparing the normal-approximation half-width at observed volume with the half-width at the full standard.
At the full standard, the standard error is $$ \frac{\sigma}{\sqrt{N_0}}=\frac{10}{\sqrt{400}}=\frac{10}{20}=0.5. $$ Thus the 95% half-width is $$ h_0=1.96(0.5)=0.98. $$
At the observed volume, the standard error is $$ \frac{\sigma}{\sqrt{N}}=\frac{10}{\sqrt{100}}=\frac{10}{10}=1, $$ so the half-width is \(h_N=1.96(1)=1.96\). It is twice the full-standard half-width.
The partial factor is $$ Z=\sqrt{\frac{100}{400}}=\sqrt{0.25}=0.5. $$ The precision comparison checks the result: \(h_0/h_N=0.98/1.96=0.5\). Common mistake avoided: using the volume ratio \(100/400=0.25\) as the factor. That ratio describes the fraction of the required volume, not the fraction of the full-standard precision.
Worked Example: Relating the Factor to a Precision Target
For a hypothetical risk group, independent comparable observations follow \(X_i\sim N(\mu,400)\), so \(\sigma=20\). The full standard is \(N_0=2{,}500\) observations under a two-sided 90% normal criterion with \(k=1.645\). The group has \(N=900\) observations. Calculate its factor and its precision half-width, then compare that half-width with the full-standard half-width. No interest or mortality assumption applies.
Method. Calculate both standard errors under the same model and criterion, and use the square-root volume ratio.
At the full standard, \(\sqrt{N_0}=50\), so the standard error is \(20/50=0.4\). The half-width is $$ h_0=1.645(0.4)=0.658. $$ At the observed volume, \(\sqrt{N}=30\), so the standard error is \(20/30=0.6667\) to four decimals. The half-width is \(1.645(20/30)=1.0967\), rounded to four decimals.
The partial factor is $$ Z=\sqrt{\frac{900}{2{,}500}}=\sqrt{0.36}=0.6. $$ The half-width ratio checks this calculation: \(h_0/h_N=0.658/1.0967\approx0.6\), allowing for rounding. Equivalently, the observed half-width is \(1/0.6=1.6667\) times the full-standard half-width. Common mistake avoided: treating the 36% volume ratio as a 36% credibility factor; the factor is 60%.
Worked Example: Volume Above the Full Standard
A hypothetical portfolio’s independent observations follow \(X_i\sim N(\mu,64)\), with standard deviation 8. Under a two-sided 95% normal criterion, \(k=1.96\), and the full standard is \(N_0=625\). The observed volume is \(N=1{,}000\). Determine the uncapped square-root ratio, the applicable credibility factor, and whether the observed precision is better or worse than the full standard. No interest or mortality assumption applies.
Method. First calculate the raw ratio, then apply the limit that credibility cannot exceed 1.
The uncapped ratio is $$ \sqrt{\frac{1{,}000}{625}}=\sqrt{1.6}=1.2649 $$ to four decimals. Therefore, $$ Z=\min(1,1.2649)=1. $$
At the full standard, the standard error is \(8/\sqrt{625}=8/25=0.32\), giving a half-width of \(1.96(0.32)=0.6272\). At observed volume, the standard error is \(8/\sqrt{1{,}000}\approx0.2530\), and the half-width is approximately \(1.96(0.2530)=0.4959\). The observed half-width is smaller, as expected with more volume. Common mistake avoided: assigning a factor of 1.2649. The factor is capped at 1, even though the experience is more precise than the full standard.
Assumptions and Common Traps
The derivation depends on a stable relationship between sampling error and volume. In the sample-mean model, each added observation is independent and drawn from the same distribution, with unchanged finite variance. The observed volume and \(N_0\) must measure comparable information. A full standard developed for one measure, population, or unit basis should not be applied automatically to another.
If observations are dependent, their average need not have variance \(\sigma^2/N\). If variability changes across the experience, or the unit of volume does not represent comparable information, a simple square-root comparison may not describe the precision. The formula is also not a substitute for checking the assumptions used to establish the full standard. The partial factor inherits that standard’s model and precision criterion.
| Situation | What the square-root calculation says | Check before using it |
|---|---|---|
| Observed volume below the full standard | \(Z=\sqrt{N/N_0}\) | Confirm a common volume basis and unchanged precision model. |
| Observed volume equal to the full standard | \(Z=1\) | The specified full-credibility criterion is just met. |
| Observed volume above the full standard | The raw square root exceeds 1, but \(Z=1\) | Do not use a credibility factor greater than 1. |
| Observed and required volumes are not comparable | The ratio has no warranted interpretation | Establish a suitable common measure or standard first. |
A frequent exam trap is to confuse the fraction of required volume with the fraction of full-standard precision. For example, if \(N/N_0=0.16\), then \(Z=\sqrt{0.16}=0.4\), not 0.16. Another is to compare unlike volumes or to report an uncapped value above 1. Before calculating, identify the full standard, the observed volume, and whether they share a basis; then take the square root and apply the cap.
Check Your Understanding
For each question, assume the full standard and observed volume are comparable unless the question says otherwise.
- A full standard requires 576 comparable observations. A hypothetical group has 144. Find its partial-credibility factor. (Answer: \(0.5\))
- A full-standard criterion corresponds to a precision half-width of 4%. Observed volume is 500 and required volume is 2,000. Find the partial factor and the implied half-width under the square-root model. (Answer: \(Z=0.5\); half-width \(=8\%\))
- A full standard is 3,600 comparable units. What volume is required for a partial factor of 0.75? (Answer: \(2{,}025\) units)
- A portfolio has 1,300 units against a full standard of 1,000. Calculate the uncapped ratio to four decimals and the applicable factor. (Answer: uncapped ratio \(=1.1402\); \(Z=1\))
- An analyst has calculated \(\sqrt{N/N_0}\) using a full standard based on independent observations, but the observed values are materially dependent. Can the analyst conclude that this is a valid partial factor from the volume ratio alone? (Answer: No; dependence may change the variance-with-volume relationship, so the stated assumptions and precision model must be reassessed.)