partial credibility

Most real books of business are not big enough to be fully credible — you almost never have the thousand-plus claims the full-credibility standard demands. Does that mean your own data are worthless and you must use the benchmark alone? No. Partial credibility is the sensible middle: you give your own data a fractional weight Z, somewhere strictly between 0 and 1, that grows as your data grow.

Under classical (limited fluctuation) credibility, partial credibility is set by the square-root rule: Z equals the square root of (your number of claims divided by the number needed for full credibility), capped at 1. So if full credibility needs 1,082 claims and you have only 270, then Z = sqrt(270/1082), about 0.50 — you blend half your own data with half the benchmark. The square root appears because random error shrinks with the square root of the sample size, so to cut the error in half you need four times the data, and the weight you can safely assign scales accordingly.

Partial credibility is what actuaries use day to day, since full credibility is the exception, not the rule. The practical art is to pick the right complement of credibility — the benchmark that gets weight (1 minus Z) — so that when your own data are thin, you fall back on something genuinely relevant (a similar class, the manual rate) rather than an arbitrary number. A poorly chosen complement can do more harm than a low Z.