Experience Rating

Experience Rating in Health Insurance Pricing

Experience rating prices a health insurance contract from the claims that the insured unit, usually an employer group and occasionally a single person, has generated in earlier periods. The insurer starts from a manual rate for comparable buyers and moves the price towards the buyer's own record, further for larger records. For health economists the practice matters because it decides who pays for being sick. Rating on experience protects insurers against adverse selection but exposes buyers to premium rises after bad years, a trade-off behind most rules on pricing private health insurance. Writing at the end of the 1990s, Cutler and Zeckhauser observed that experience rating was rare for individuals but common for groups, and that most private health insurance in the United States was at least partly experience rated.

How a claims record enters the premium

Classical ratemaking treats every risk in a tariff class as alike and charges each the class mean. In a 1967 ASTIN Bulletin report, Bühlmann described experience rating as the devices by which a rate fixed at the outset is gradually altered during the contract, such as profit participation, no-claim bonuses and bonus-malus scales. He framed it as a sequence of estimates of a risk's unknown expected claims from its own observations, so that each class of risks with the same observed record pays its own way.

In employer health insurance the same idea works at the level of the group. ASOP No. 25, the US actuarial standard on credibility procedures, states that prospective experience rating formulas assign credibility to the actual experience of a single risk or a group of risks, and that the experience may be split, for example into primary and excess losses, with a different credibility for each part. Splitting matters in health cover because one catastrophic admission can dominate a small group's record, and Bühlmann's report already discussed improving credibility by excluding the largest claims.

Credibility weighting between the group record and the manual rate

The weight given to a group's own record is its credibility. ASOP No. 25 defines credibility as a measure of the predictive value that the actuary attaches to a particular set of data, and full credibility as the level at which that experience is given full predictive value. Most experience-rating formulas blend the two sources linearly.

$$P = Z\,\bar{X} + (1 - Z)\,M$$

where $P$ is the credibility-weighted estimate of expected claims per member-year, $\bar{X}$ is the group's observed mean claims per member-year, $M$ is the manual rate (expected claims for comparable groups in the insurer's wider book) and $Z$ is the credibility factor, between 0 and 1.

Bühlmann showed that this blend is the best linear approximation, in the least-squares sense, to the posterior mean of a risk's expected claims, with an optimal weight of a simple form. Treating each member-year as one observation from the group's claims distribution gives the following.

$$Z = \frac{n}{n + k}, \qquad k = \frac{E[\sigma^2(\theta)]}{\operatorname{Var}[\mu(\theta)]}$$

where $n$ is the group's volume in member-years, $\theta$ is the group's unknown risk parameter, $\mu(\theta)$ is its true expected claims per member-year, $\sigma^2(\theta)$ is the variance of claims per member-year around that mean, $E[\sigma^2(\theta)]$ is that variance averaged across groups, and $\operatorname{Var}[\mu(\theta)]$ is the variance of the true means between groups.

Bühlmann stressed that the result holds whatever the shape of the claims and risk-parameter distributions. The ratio $k$ carries the economics. When claims within a group are noisy relative to real differences between groups, $k$ is large and a group needs a long record before its own experience counts for much. For a health economist this is familiar shrinkage, the pull towards a common mean that a random-effects model applies to small units.

Greatest accuracy and limited fluctuation credibility

ASOP No. 25 calls the Bühlmann approach greatest accuracy credibility. It also describes the classical limited fluctuation method, which sets a full credibility standard and bases partial credibility on the square root of the ratio of actual volume to that standard.

$$Z = \min\left(1, \sqrt{n / n_F}\right)$$

where $n_F$ is the full credibility standard, the volume at which the group's experience receives full weight.

The standard allows full, partial or zero credibility to be assigned by judgement without a rigorous model. It also notes that credibility estimation is now built into hierarchical models and mixed-effects models.

Worked example with two illustrative employer groups

All figures in this example are illustrative. An insurer's manual rate for employer groups of this type is GBP 3,000 of expected claims per member-year. From its book it estimates the average within-group variance of claims per member-year at 81,000,000 (a standard deviation of GBP 9,000) and the variance of true group means at 202,500 (a standard deviation of GBP 450).

1. Credibility constant. The ratio is 81,000,000 / 202,500 = 400, so $k$ is 400 member-years.

2. Group A. Group A has 400 members observed for three years, a volume of 400 × 3 = 1,200 member-years, and mean claims of GBP 3,600. Its credibility is 1,200 / 1,600 = 0.75, and the blended estimate is 0.75 × 3,600 + 0.25 × 3,000 = 3,450.

3. Group B. Group B has 20 members over three years, a volume of 20 × 3 = 60 member-years, and mean claims of GBP 4,500 after one member's long hospital stay. Its credibility is 60 / 460 = 0.1304, and its estimate, written as the manual rate plus credibility times the excess, is 3,000 + 0.1304 × 1,500 = 3,195.6.

4. Limited fluctuation alternative. With an illustrative full credibility standard of 2,000 member-years, Group A's credibility is the square root of 0.6, which is 0.7746, giving 3,000 + 0.7746 × 600 = 3,464.8. Group B's is the square root of 0.03, which is 0.1732, giving 3,000 + 0.1732 × 1,500 = 3,259.8.

5. Interpretation. Group A's record ran 20% above the manual rate and its estimate is 15% above, so three quarters of the excess passes through. Group B's record ran 50% above, yet its estimate rises only about 6.5%, because 60 member-years of health claims say little about the group's true mean and the long stay is mostly treated as chance. Group A, with twenty times Group B's volume, is priced much closer to its own record, in line with Cutler and Zeckhauser's remark that bigger groups are more likely to be experience rated. The limited fluctuation rule puts Group B about GBP 64 higher than the Bühlmann rule does, so the choice of method matters most for small groups. Trend and loadings for administration and margin then turn these expected claims into a premium.

Why insurers rate on experience and what buyers lose

Experience rating is a response to information asymmetry. Cutler and Zeckhauser trace adverse selection to individuals knowing their likely use of care when insurers do not, or may not use that knowledge, and note that insurers increasingly monitor past utilisation to forecast future costs. Once a claims record is priced, older and sicker groups are charged more per head for the same cover. Handel, Hendel and Whinston note that unrestricted pricing on health status would remove adverse selection entirely, at a very high price for sick consumers.

The cost falls on buyers through reclassification risk. Cutler and Zeckhauser point out that people whose premiums depend on their sickness lose the ability to buy at the same price as their peers, and that judged before anyone knows who will fall ill, this distributional effect is an efficiency loss. With annual contracts the risk of being found to be high risk cannot be insured, a missing market behind their fourth lesson of health insurance. Handel, Hendel and Whinston estimated that allowing even partial pricing of health status in exchange-type markets would create a welfare loss from reclassification risk substantially larger than the gain from wider within-year coverage, provided consumers can borrow when young or premiums can vary by age.

One proposed remedy keeps risk-based pricing at entry but blocks later reclassification. Pauly, Kunreuther and Hirth proposed guaranteed renewable insurance, with a premium schedule that lets the insurer break even and is chosen by both low-risk and high-risk buyers whether or not they have suffered a loss.

Rules that limit experience rating in the United States

The Affordable Care Act moved US individual and small-group cover away from experience rating. Section 2701 of the Public Health Service Act (42 U.S.C. 300gg) lets individual and small group premiums vary only by individual or family cover, rating area, age and tobacco use, each within set limits, and by no other factor. The single risk pool rule in 45 CFR 156.80 requires an insurer to treat the claims experience of all its members in a state's individual market, and in its small group market, as one risk pool, so a small employer's own claims cannot move its price. The rating rules reach the large group market only if a state lets large-group cover into its Exchange, and even then exclude self-insured group health plans.

Within a group, 42 U.S.C. 300gg-4 bars a plan or insurer from charging an individual more than a similarly situated member on the basis of health status-related factors, a list that includes claims experience. The same section says this does not restrict the amount an employer may be charged, apart from a ban on adjusting the group's premium for genetic information. A large group can therefore be priced on its members' combined record while no single member pays for their own.

Self-funding takes the logic furthest. The 2026 KFF Employer Health Benefits Survey, published on 22 October 2026, reports that 71% of covered workers are in self-funded plans, in which firms pay for their workers' care from their own funds, some buying stop-loss cover to limit their liabilities. It also reports that 38% of covered workers in firms with 10 to 199 workers are in level-funded plans, which are sold as self-funded and, for employers with fewer than 50 employees, are not subject to the ACA rating standards for small firms. KFF notes that, unlike insured plans, level-funded plans may use health status in rating and underwriting.

Experience rating beside community rating and risk adjustment

Experience rating is one form of risk rating. Medical underwriting prices predicted risk from health information given at application, whereas experience rating prices the realised claims of the insured unit. Community rating, the opposite rule, removes reclassification risk at the price of adverse selection, which Handel, Hendel and Whinston found produced substantial unravelling under the ACA pricing rules in their model. Age rating lets price vary with a characteristic other than health. Risk pooling is the wider principle that experience rating limits, since a group priced at full credibility shares risk only among its own members.

Risk adjustment works on the other side of the market, moving money between insurers so that premiums which cannot vary with health do not reward enrolling healthy people. Cutler and Zeckhauser warn that adjusters built on diagnoses, past claims and actual use, the information experience rating draws on, are open to distortion by plans.

Reading experience-rated premiums in health economic work

Premiums from experience-rated markets carry each group's own history. A comparison of employer premiums across plans or regions therefore mixes the price of cover with the composition and luck of each workforce, so an analyst using premiums as a price measure has to allow for group size and age mix or work from claims data directly. In models of insurance regulation, the credibility formula gives a tractable rule for how quickly premiums follow claims, which governs how much reclassification risk a policy option leaves with buyers.

Misreadings of experience rating

Most errors come from treating the credibility blend as settled fact or the legal limits as uniform. The blend rests on estimates and choices made by the insurer, and the law treats market segments differently.

1. Treating credibility as a fixed property of the data. The constant $k$ is estimated from the insurer's portfolio, so different books give different weights, and ASOP No. 25 leaves the choice of procedure to professional judgement.
2. Assuming the ACA ended experience rating. Its rating rules govern individual and small group cover, and large groups and self-funded plans remain free to be priced on their own experience.
3. Assuming each group's risk is stable. Bühlmann's model holds each risk constant over time, and he called for work on trend-type changes, so claims from a workforce that has since changed carry less information.

Sources

• Actuarial Standards Board. Actuarial Standard of Practice No. 25, Credibility Procedures. Adopted December 2013, effective 1 May 2014. https://www.actuarialstandardsboard.org/asops/credibility-procedures/
• Bühlmann H. Experience rating and credibility. ASTIN Bulletin. 1967;4(3):199-207. https://doi.org/10.1017/S0515036100008989
• Cutler DM, Zeckhauser RJ. The anatomy of health insurance. In: Culyer AJ, Newhouse JP, eds. Handbook of Health Economics. Vol 1A. Elsevier; 2000:563-643. https://doi.org/10.1016/S1574-0064(00)80170-5
• Handel BR, Hendel I, Whinston MD. Equilibria in health exchanges: adverse selection vs. reclassification risk. NBER Working Paper 19399; September 2013. https://www.nber.org/papers/w19399
• KFF. 2026 Employer Health Benefits Survey, Summary of Findings and Section 10 Plan Funding. Published 22 October 2026. https://www.kff.org/health-costs/2026-employer-health-benefits-survey/
• Pauly MV, Kunreuther H, Hirth R. Guaranteed renewability in insurance. Journal of Risk and Uncertainty. 1995;10(2):143-156. https://ideas.repec.org/a/kap/jrisku/v10y1995i2p143-56.html
• 42 U.S.C. 300gg, Fair health insurance premiums (Public Health Service Act section 2701, as amended in 2010). https://www.law.cornell.edu/uscode/text/42/300gg
• 42 U.S.C. 300gg-4, Prohibiting discrimination against individual participants and beneficiaries based on health status. Text as in force, accessed 25 October 2026. https://www.law.cornell.edu/uscode/text/42/300gg-4
• 45 CFR 156.80, Single risk pool. Text as in force, accessed 25 October 2026. https://www.law.cornell.edu/cfr/text/45/156.80