Adverse Selection

Adverse Selection in Health and Life Insurance

Adverse selection explains how an insurance market can unravel even when every buyer and seller acts rationally. When the people choosing cover know more about their own expected costs than the insurer can observe or price, the insured pool becomes more expensive than the population the premium was set for, and lower-risk buyers may drop out. The problem shapes the design of health insurance markets, employer plan menus and public schemes, and it is also one reason life insurers underwrite applicants. It is distinct from moral hazard and from insurer risk selection, and the policy responses to it aim to maintain broad risk pooling.

Why the Composition of an Insurance Pool Matters

Insurance combines people with different expected healthcare needs so that uncertain individual costs can be shared across a group. The premium required to finance that coverage depends partly on the expected costs of the people who enrol. When enrolment is systematically related to information that individuals possess but insurers cannot fully observe or use, the insured group may become more costly than the wider eligible population.

The idea was formalised by Akerlof in his analysis of the used-car market for "lemons", where buyers cannot tell good cars from bad ones and the average quality on offer falls as the price falls. Akerlof applied the same logic to medical insurance for people over 65 and argued that as the price rises the people still willing to buy are increasingly those most certain they will need care, so no sale may take place at any price. He treated this as one argument for public insurance through Medicare.

Four linked features create this problem. Individuals differ in their expected healthcare costs; they may know more than insurers about symptoms, family history or anticipated needs; they can use that knowledge when deciding whether to enrol and how much cover to buy; and a common premium reflects the expected cost of the people who actually enrol, not necessarily that of everyone eligible.

The result is a composition effect. Insurance choices do not merely redistribute people across plans; they can change the expected cost of each plan and therefore the premiums and benefits that can be sustained.

A Simple Insurance Mechanism

A basic model shows how individual risk and enrolment interact. Suppose person $i$ faces a probability $p_i$ of incurring a healthcare cost $L$. The person’s expected healthcare cost is the probability of the cost multiplied by its size.

$$E(C_i)=p_iL$$

Here $E(C_i)$ is the expected cost that person $i$ brings to the insurer.

An insurer that cannot charge each person according to their complete risk may set a common premium using the average expected cost of those who select the plan. If $S$ represents the group that enrols, the break-even premium is the average expected cost of that group.

$$P=\frac{\sum_{i \in S}p_iL}{|S|}$$

Here $P$ is the premium that just covers expected claims and $|S|$ is the number of people who enrol; administrative loading is ignored.

The premium therefore depends on the composition of $S$. If people with comparatively low expected costs leave while people with higher expected costs remain, the average value of $p_i$ rises and the insurer must increase the premium or reduce the generosity of coverage.

This mechanism does not require concealment, fraud or irresponsible behaviour. Individuals can create the selection effect simply by making rational choices using information available to them.

How a Premium Spiral Can Develop

An insurance market does not unravel merely because people have different risks. Instability develops when private risk information affects enrolment strongly enough that changing premiums repeatedly alter the risk pool. Each round can make coverage less attractive to people with lower expected costs.

A potential premium spiral follows the sequence set out below.

1. Initial premium. The insurer sets a premium using the expected cost of the anticipated membership.
2. Selective take-up. A disproportionate number of people who expect to use more healthcare, or who value generous coverage most strongly, enrol.
3. Higher claims. Claims exceed the cost assumed when the premium was calculated.
4. Repricing. The insurer increases the premium or reduces benefits to restore financial balance.
5. Exit of lower risks. Some members with comparatively lower expected costs leave because the revised plan offers them less value.
6. Recalculation. The premium is recalculated using the higher average expected cost of the remaining membership.

This process is sometimes described as a death spiral, but complete market collapse is not inevitable. Risk aversion, subsidies, employer contributions, enrolment restrictions, preferences for security and differences between insurance products can keep lower-risk people enrolled.

The best-documented case in health insurance comes from Harvard University in the mid-1990s. When the university moved to contributing an equal amount towards each health plan, employees had to pay the full extra cost of its generous preferred provider organisation (PPO) plan. Younger and healthier members left, the plan lost money and raised its premium, more members left, and the PPO was withdrawn within three years of the change.

Illustrative example with two risk groups and one premium

The figures in this example are hypothetical and chosen for easy arithmetic. An insurer offers one plan to 1,000 people. Half are low risk, with a 10% chance of a £10,000 healthcare cost, and half are high risk, with a 30% chance of the same cost. Each person knows which group they are in, but the insurer cannot tell them apart and must charge everyone the same premium.

The expected cost of a low-risk person is 0.10 × £10,000 = £1,000, and that of a high-risk person is 0.30 × £10,000 = £3,000. If everyone enrols, the break-even premium is the average expected cost, which is (500 × £1,000 + 500 × £3,000) ÷ 1,000 = £2,000.

Suppose each low-risk person would pay up to £1,500 for cover (their expected cost plus £500 for protection against risk) and each high-risk person would pay up to £3,600. At £2,000 the plan costs more than low-risk people are willing to pay, so they leave. The remaining pool is entirely high risk, the break-even premium rises to £3,000, and high-risk people still buy at that price. The market settles with only the high-risk group insured.

The outcome is inefficient. Each low-risk person values cover at £1,500 but costs only £1,000 to insure, so leaving them uninsured forgoes £500 of surplus per person, or 500 × £500 = £250,000 across the group. The trade fails because the insurer cannot observe risk, even though every low-risk person would pay more than their own expected cost.

Pooling, Separation and Plan Choice

Insurance markets can respond to hidden risk through pooling or separation. A pooling arrangement offers similar terms to people with different risks, supporting broad risk sharing but potentially making coverage unattractive to some lower-risk participants. A separating arrangement offers contracts with different premiums, benefits or cost sharing so that different risk groups select different plans.

Rothschild and Stiglitz showed that in a competitive market with hidden risk types, a single pooled contract cannot survive, because a rival insurer can offer less cover at a lower price that appeals only to low risks. The competitive outcome, if there is one, is a separating equilibrium in which high risks buy full cover at a premium reflecting their risk and low risks accept partial cover that high risks find unattractive. When low-risk buyers make up a large enough share of the market, no competitive equilibrium may exist at all.

Pooling shares costs between healthier and less-healthy members, but the pooled premium may be hard to sustain if lower-risk members leave. Separation eases the insurer's information problem, but may weaken redistribution and leave higher-risk people on less favourable terms. Sorting can also have other causes. High-deductible plans may attract people expecting lower use, and plan choice may reflect income, provider preferences or risk tolerance as well as private information about health risk.

Observed sorting between plans is therefore not sufficient evidence of inefficient adverse selection. Analysts must establish that private information about expected risk is materially influencing choice and market outcomes.

A Health-Insurance Example

Consider two plans offered to the same workforce. Plan A has a higher premium but low cost sharing, while Plan B has a lower premium but a high deductible. Employees who expect regular specialist visits, prescription use or planned treatment may be more likely to choose Plan A.

If those expected costs are not fully captured by the information used to set premiums, Plan A may attract a membership with substantially higher claims than anticipated. Its premium may then rise, encouraging some healthier members to move to Plan B and further increasing the average cost of Plan A.

The example shows why spending differences between plans need careful interpretation. Higher spending in Plan A may reflect health risks present before enrolment, which is adverse selection, or extra use caused by its lower cost sharing, which is moral hazard. Both can occur together and must be separated empirically.

Adverse Selection in Life Insurance

Life insurance faces the same problem in a different form. An applicant may know about symptoms, medical history, family history or lifestyle that affect their chance of dying during the policy term, and the insurance industry often calls the resulting behaviour anti-selection. Life insurers respond mainly by underwriting. They ask about medical and family history and price or decline cover on that basis, which reduces the private information left over.

Evidence on how much adverse selection remains in life insurance is mixed. A study of United States term life insurance found patterns that are hard to reconcile with the standard theory, since unit prices fell as the amount of cover rose, and higher-risk individuals did not buy more cover. A later study that focused on people deciding whether to buy a new policy, rather than on everyone already holding one, did find a positive link between purchase and later mortality. People who died within 12 years of a base year were 19% more likely to have taken up life insurance in that year than those who survived.

Genetic testing shows how regulation interacts with this risk. In the United Kingdom, the Code on Genetic Testing and Insurance agreed between the government and the Association of British Insurers allows insurers to take a predictive genetic test result into account only where it is on an approved list. The only approved test is for Huntington's disease, and only for life insurance cover above £500,000 per person. The Code names significant anti-selection as a development that could undermine insurers' ability to offer affordable cover.

Distinguishing Related Concepts

Adverse selection sits within the broader field of information asymmetry, but it should not be treated as interchangeable with every information problem. It is a hidden-information problem, and its distinguishing feature is the use of privately held characteristics in participation or contract choice before an agreement. Other concepts describe different timings, actors or mechanisms.

The most important distinctions are set out below.

• Information asymmetry is the broader condition in which parties possess unequal information; adverse selection is one possible consequence of that condition.
• Moral hazard concerns behaviour or healthcare use after insurance changes the price or consequences faced by the insured person.
• Risk selection describes actions taken by an insurer to attract lower-cost members or discourage people expected to generate higher claims.
• Cream skimming is a form of insurer risk selection aimed at securing favourable risks rather than improving the efficiency of care.
• Provider-induced demand concerns the influence of healthcare professionals over patient demand and is not a form of consumer adverse selection.

Keeping these mechanisms separate matters because they require different evidence and policy responses. Enrolment rules may address adverse selection, while cost-sharing or payment design may address moral hazard. Where premiums are community rated, plans that cannot charge high-risk members more have an incentive to attract predictably profitable members, which risk adjustment of payments to plans is designed to counter.

Adverse and Advantageous Selection

Selection does not always move in the direction predicted by the simplest insurance model. People who are more cautious, financially organised or risk averse may be both more likely to purchase insurance and less likely to generate claims. This can create advantageous selection, in which comparatively lower-risk individuals are more strongly represented among purchasers.

The direction of selection depends on characteristics that affect both insurance demand and expected cost. Risk aversion and health awareness may raise demand for insurance while also encouraging prevention and treatment adherence, and income, financial literacy, access and trust may shape purchase independently of health risk.

For this reason, adverse selection should be tested rather than assumed. A positive relationship between insurance generosity and claims is consistent with adverse selection but does not prove that private risk information caused the relationship, because moral hazard can produce the same correlation. The reverse also holds. Evidence from the United States long-term care insurance market found that people with private information that they were high risk bought more cover, but so did more cautious people who turned out to be lower risk, so the two effects offset and those with more insurance were not higher risk on average. The absence of a positive correlation does not show that private information is absent.

Measuring Adverse Selection

Empirical analysis must separate differences that existed before enrolment from changes caused by insurance after enrolment. Simple comparisons between insured and uninsured people are unreliable because the groups may differ in health, income, preferences, access and risk tolerance. The central challenge is to identify whether private expected risk affected the choice of coverage.

Several kinds of evidence can help.

• Pre-enrolment diagnoses, spending or medicine use show whether people choosing generous cover already had greater expected needs.
• Randomised premiums or subsidies reveal whether higher-risk and lower-risk groups respond differently to price.
• Employer benefit changes can act as natural experiments when new plan choices arise for reasons unrelated to health.
• Plan-switching patterns show whether people moving into or out of a plan later generate different claims.
• Changes in average claims after premium increases indicate whether lower-cost members left disproportionately.
• Structural models estimate private information and welfare effects where direct observation is impossible, but rest on modelling assumptions.

Risk-adjustment models help predict expected cost, but they cannot capture information that individuals hold and administrative or clinical data lack.

Reducing Adverse Selection with Policy Responses and Their Trade-Offs

Adverse selection is reduced by keeping lower-risk people in the pool and by stopping people from waiting until they need care before buying cover. Measures that compensate insurers for high-cost members work on the insurer's side. They let a plan that attracts sicker members hold its premium, and they also counter cream skimming, the insurer selection that community rating invites.

• Mandates and coverage requirements. A requirement to hold cover brings in people who expect to claim little. The US Affordable Care Act (ACA) required individuals, with exemptions, to hold qualifying health coverage from 2014 or pay a tax penalty; a 2017 amendment cut the penalty to zero from 2019, so the requirement remains without a federal penalty. Einav and Finkelstein call a mandate the canonical solution but show that, once insurance loading costs are allowed for, it can also cover people whose expected costs exceed their willingness to pay, so its welfare effect is an empirical question.
• Premium subsidies. A subsidy lowers the price that lower-risk buyers weigh against their expected costs, and in Einav and Finkelstein's textbook case, with no loading costs, any positive subsidy reduces the welfare loss from under-insurance. US Marketplace premium tax credits, based on household income, are one example. The cost falls on public funds, and the subsidy also goes to buyers who would have insured anyway.
• Risk rating and underwriting. Pricing on observable risk, such as age or medical history, narrows the private-information gap. Einav and Finkelstein show that where buyers hold no further private information within each priced group, that segment is efficient, and banning the price difference creates adverse selection where none existed. The trade-off is higher premiums for higher-risk groups; the ACA limits individual and small-group premium variation to age, tobacco use, rating area and individual or family cover.
• Late-entry (lifetime) loadings. Community rating sharpens the incentive to wait, because a person who joins only when ill pays the same premium as everyone else, and a late-entry loading penalises that delay. Ireland's Lifetime Community Rating, introduced after 30 April 2015, adds 2% of the gross premium for each year a person spent aged 35 or over without cover, up to 70%, for at most 10 years. Australia's Lifetime Health Cover works in a similar way, as described under community rating. The loading falls equally on people who delayed by choice and on those who could not afford cover earlier.
• Enrolment windows and waiting periods. Limiting when people can join, or when new members can claim, stops people buying cover only once they need care. US Marketplace plans are sold in a yearly open enrolment period and otherwise only in special enrolment periods, for example after losing other coverage. Australian hospital cover has maximum waiting periods of 12 months for pre-existing conditions and pregnancy, which the Commonwealth Ombudsman's privatehealth.gov.au says stop people making a large claim shortly after joining and then cancelling; the ACA instead prohibits pre-existing condition exclusions. The cost is a gap in protection for anyone who needs care during a waiting period or misses the window.
• Risk adjustment and risk equalisation. These schemes move money between insurers according to the risk profile (and, in Ireland, partly the actual hospital use) of their members. Under the ACA, plans with lower than average actuarial risk pay a charge and plans with higher than average risk receive a payment; Ireland's Risk Equalisation Scheme pays insurers credits for older members and those with hospital admissions from a stamp duty on premiums. Van de Ven and colleagues found that in European markets even the best formulas leave insurers substantial selection incentives, and diagnosis-based payments reward intensive coding. Geruso and Layton found risk scores 6% to 16% higher in private Medicare plans than under fee-for-service Medicare.
• Reinsurance. Reinsurance pays part of the cost of very expensive members, reducing the loss an insurer expects from enrolling them. The ACA ran a transitional reinsurance programme for plan years 2014 to 2016 for issuers covering high-risk individuals in the individual market, and Ireland's scheme includes a High Cost Claims Pool credit. Because such payments follow actual costs, they also weaken insurers' incentives to control cost.

Interpreting the Policy Problem

The policy question is whether private information is causing coverage to become inefficiently expensive, incomplete or unstable, and whether intervention can improve welfare after accounting for its costs. A widely used graphical framework makes this question concrete by plotting the demand for insurance against the average and marginal cost of the people who buy it. Under adverse selection the people most willing to pay are also the most costly to cover, so a price set at average cost leaves uninsured some people whose willingness to pay exceeds the cost of covering them, as in the illustrative example above. Under advantageous selection the pattern reverses and the market can provide more insurance than is efficient. Estimating the size of either welfare loss requires variation in price, such as premium changes or subsidies that shift enrolment.

Sources

• Akerlof GA. The market for "lemons": quality uncertainty and the market mechanism. Quarterly Journal of Economics. 1970;84(3):488-500.
• Rothschild M, Stiglitz J. Equilibrium in competitive insurance markets: an essay on the economics of imperfect information. Quarterly Journal of Economics. 1976;90(4):629-649.
• Cutler DM, Reber SJ. Paying for health insurance: the trade-off between competition and adverse selection. Quarterly Journal of Economics. 1998;113(2):433-466.
• Cutler DM, Zeckhauser RJ. Adverse selection in health insurance. In: Garber AM, editor. Frontiers in Health Policy Research, volume 1. MIT Press. 1998:1-32.
• van de Ven WPMM, Ellis RP. Risk adjustment in competitive health plan markets. In: Culyer AJ, Newhouse JP, editors. Handbook of Health Economics, volume 1A. Elsevier. 2000:755-845.
• Finkelstein A, McGarry K. Multiple dimensions of private information: evidence from the long-term care insurance market. American Economic Review. 2006;96(4):938-958.
• Einav L, Finkelstein A. Selection in insurance markets: theory and empirics in pictures. Journal of Economic Perspectives. 2011;25(1):115-138.
• Cawley J, Philipson T. An empirical examination of information barriers to trade in insurance. American Economic Review. 1999;89(4):827-846.
• He D. The life insurance market: asymmetric information revisited. Journal of Public Economics. 2009;93(9-10):1090-1097.
• HM Government and Association of British Insurers. Code on Genetic Testing and Insurance. 2018, updated 19 March 2026.
• United States Code, 26 U.S.C. section 5000A (requirement to maintain minimum essential coverage), as amended by Public Law 115-97, section 11081, 2017.
• United States Code, 42 U.S.C. sections 300gg, 300gg-3, 18061 and 18063 (Patient Protection and Affordable Care Act 2010).
• HealthCare.gov. Glossary entries: Fee; Open Enrollment Period; Special Enrollment Period; Premium tax credit. Accessed 2 October 2026.
• Health Insurance Authority (Ireland). Lifetime community rating (LCR). Accessed 2 October 2026.
• Department of Health and Health Insurance Authority (Ireland). Public Consultation: Community-rated health insurance market in Ireland and proposed changes to the Risk Equalisation Scheme. 2025.
• Commonwealth Ombudsman. Waiting periods. privatehealth.gov.au. Accessed 2 October 2026.
• van de Ven WP, van Kleef RC, van Vliet RC. Risk selection threatens quality of care for certain patients: lessons from Europe's health insurance exchanges. Health Affairs. 2015;34(10):1713-1720.
• van de Ven WPMM, Hamstra G, van Kleef RC, Reuser M, Stam PJA. The goal of risk equalization in regulated competitive health insurance markets. European Journal of Health Economics. 2023;24(1):111-123.
• Geruso M, Layton T. Upcoding: evidence from Medicare on squishy risk adjustment. Journal of Political Economy. 2020;128(3):984-1026.