Introduction: The Market That Cannot Work Like a Market
In 1963, Kenneth Arrow published a paper in the American Economic Review titled “Uncertainty and the Welfare Economics of Medical Care.” It is widely regarded as the founding document of health economics, and its argument was startling for its time: healthcare is not a market failure because of some fixable friction. It fails the assumptions of the competitive model at nearly every point simultaneously.
Arrow catalogued the departures. Demand is irregular and unpredictable. The product cannot be tested before purchase. The seller advises the buyer on what to buy. Entry is legally restricted. Prices are frequently not posted. Uncertainty pervades both the incidence of illness and the efficacy of treatment. And insurance, the natural market response to uncertainty, creates its own distortions.
Seven years later, George Akerlof formalised the mechanism by which one of these problems — asymmetric information — can destroy a market entirely. This guide connects the two, works through the modern evidence, and shows how to deploy the material in an exam.
1. Akerlof’s Market for Lemons
George Akerlof’s “The Market for ‘Lemons’: Quality Uncertainty and the Market Mechanism” (Quarterly Journal of Economics, 1970) earned him a share of the 2001 Nobel Prize alongside Michael Spence and Joseph Stiglitz.
The setup is a used-car market. Sellers know whether their car is a “peach” (good) or a “lemon” (bad). Buyers cannot tell. Suppose peaches are worth $10,000 to buyers, lemons $4,000, and half of all cars are lemons.
A risk-neutral buyer will offer the expected value: $7,000. But at $7,000, owners of peaches — who value their cars at, say, $9,000 — refuse to sell. Only lemon owners transact. Buyers observe this, revise their beliefs, and lower their offer toward $4,000. The good cars have been driven out of the market by the bad.
This is adverse selection: the informed party’s private information causes the composition of the market to deteriorate. In the limit, the market unravels entirely, and mutually beneficial trades in high-quality goods never occur.
Why this describes health insurance exactly
Replace cars with insurance customers. The applicant knows their health status; the insurer does not. The insurer must set a premium based on the average risk of the pool.
- That premium is a bargain for the sick and a bad deal for the healthy.
- The healthy decline to insure.
- The pool’s average risk rises.
- The insurer raises the premium.
- The next-healthiest tranche exits.
This is the adverse selection death spiral. In principle it terminates only when the sole remaining customers are those certain to claim, at which point insurance has ceased to be insurance and become prepayment.
Note carefully the direction of the information asymmetry, because students routinely reverse it. In adverse selection, the buyer of insurance has private information about their type, held before the contract is signed. This is hidden information, and it is a pre-contractual problem.
2. Moral Hazard: The Post-Contractual Problem
Moral hazard is a distinct failure, and confusing the two is the single most common error in health economics exams.
Moral hazard arises after the contract is signed. Being insured changes behaviour. An insured person faces a marginal price of care far below its marginal cost, and therefore consumes more of it than they would if they paid the full price. This is hidden action.
Pauly (1968), responding critically to Arrow, argued that this response is not moral failing but ordinary demand behaviour: if you cut the price of a good, people buy more of it. Insurance is a price cut. The welfare loss from moral hazard is a standard deadweight loss triangle under the demand curve.
Two flavours are worth distinguishing:
- Ex ante moral hazard: insured people take fewer precautions. They exercise less, smoke more, drive faster. Empirical support for this is surprisingly weak.
- Ex post moral hazard: once ill, insured people consume more treatment. Empirical support for this is strong.
The RAND Health Insurance Experiment
The RAND HIE, running from 1974 to 1982, randomly assigned thousands of American families to insurance plans with differing coinsurance rates, from full coverage to a 95% coinsurance rate. It remains the only large-scale randomised experiment on cost-sharing in health insurance, and Newhouse and colleagues’ 1993 monograph Free for All? reports the results.
The central findings:
- Cost-sharing substantially reduced healthcare utilisation. The implied price elasticity of demand for medical care is roughly −0.2: inelastic, but decisively non-zero. Moral hazard is real.
- For the average participant, reduced utilisation produced no detectable deterioration in health outcomes. People cut back on both effective and ineffective care roughly indiscriminately.
- For the poorest and sickest participants, cost-sharing did produce worse outcomes, notably in blood pressure control and vision.
The policy implication is uncomfortable and precise: cost-sharing controls spending effectively, and does so partly by deterring care that people needed. Patients are not good at distinguishing high-value from low-value care.
The Oregon Health Insurance Experiment
In 2008, Oregon expanded Medicaid via lottery — a randomised allocation of insurance created by the state’s budget constraint rather than by researchers. Finkelstein, Taubman, Wright and colleagues, publishing in the Quarterly Journal of Economics (2012) and subsequently in the New England Journal of Medicine (Baicker et al., 2013), exploited this natural randomisation.
Insurance coverage produced large increases in healthcare utilisation across every category, including emergency department visits — contradicting the widespread claim that insuring the uninsured would divert them from expensive emergency care. It substantially improved financial security, sharply reducing catastrophic out-of-pocket expenditure and medical debt. It produced a large, statistically robust improvement in self-reported mental health, notably reducing rates of depression.
What it did not produce, within two years, was statistically significant improvement in measured physical health markers such as blood pressure, cholesterol and glycated haemoglobin. The confidence intervals were wide enough that clinically meaningful effects could not be ruled out — a nuance frequently lost in the resulting political argument.
The Oregon experiment is the cleanest evidence we have on what health insurance does. It is, primarily, a financial product. It protects against ruin. Whether it extends life is a harder question than either side of the debate admits.
3. Supplier-Induced Demand: The Physician as Agent
Arrow’s most distinctive observation was that the seller in the healthcare market is also the buyer’s agent. The physician recommends the treatment and supplies it.
If physicians are perfect agents, this poses no difficulty. If they respond to financial incentives, it poses a serious one. Supplier-induced demand (SID) is the hypothesis that physicians can shift patients’ demand curves outward to compensate for income shortfalls.
Clean identification is genuinely difficult, because physician density correlates with population health, income and preferences. The most credible modern evidence comes from settings where physician payment rates change exogenously.
Clemens and Gottlieb (2014), in the American Economic Review, exploited a Medicare payment reform that raised reimbursement rates differentially across geographic areas. They found that physicians increased the volume of care they supplied when payment rates rose — but the elasticity was modest, and it was concentrated in discretionary, elective procedures rather than in urgent care. Financial incentives shape medicine at the margin of discretion, not at its centre.
This finding is important for exam evaluation. Supplier-induced demand exists; it is not so large that it swamps clinical judgement. Both the naive market model and the cynical account of medicine are wrong.
4. Institutional Responses to Asymmetric Information
Against adverse selection
| Mechanism | How it works | Problem |
|---|---|---|
| Mandates | Compel everyone to buy; the healthy cannot exit | Coercive; requires enforcement; unpopular |
| Employer-provided insurance | Groups formed for non-health reasons have near-random risk composition | Job lock; ties health to employment |
| Single-payer / universal systems | The pool is the entire population by construction | Requires tax financing; moral hazard remains |
| Screening / underwriting | Insurer invests in learning the applicant’s type | Excludes the sick — precisely those who need cover |
| Risk adjustment | Transfer payments across insurers based on enrollee risk | Depends on observable risk proxies; gameable |
| Community rating + guaranteed issue | Forbid pricing on health status | Without a mandate, accelerates the death spiral |
Rothschild and Stiglitz (1976), in the Quarterly Journal of Economics, formalised the insurance market with adverse selection and proved two devastating results. First, a pooling equilibrium cannot exist — an insurer can always profitably “cream-skim” low risks with a high-deductible contract. Second, a separating equilibrium, in which low risks accept inefficiently low coverage to signal their type, may not exist at all. Competitive insurance markets with hidden information may have no equilibrium whatsoever. This is a far stronger indictment of the market mechanism than the lemons model, and it is the theoretical justification for mandates.
Against moral hazard
- Deductibles, copayments, coinsurance — restore some marginal price. Effective (RAND), but blunt.
- Managed care and utilisation review — the insurer second-guesses the physician.
- Capitation and bundled payments — pay providers a fixed sum per patient, reversing the volume incentive. Introduces the opposite risk: under-provision.
- Gatekeeping — primary care physicians control specialist access.
- Value-based insurance design — set copayments inversely to the clinical value of the treatment, so that high-value care is free and low-value care is expensive. This directly addresses the RAND finding that patients cut back indiscriminately. Chernew, Rosen and Fendrick have developed this framework extensively.
5. Case Study: The UK NHS and the Absence of Prices
The National Health Service abolishes the insurance market entirely. Care is financed from general taxation and free at the point of use.
This solves adverse selection by construction — there is one pool, containing everyone, and no one can exit. It does nothing about moral hazard; the marginal price of care is zero, which is the moral hazard problem in its purest form.
The NHS therefore rations care not by price but by waiting time and by clinical prioritisation. Waiting lists are the shadow price. This has a specific and underappreciated economic property: waiting is a cost borne by the patient that generates no revenue for anyone. It is pure deadweight loss, whereas a monetary price at least transfers surplus. Against this, rationing by waiting is distributionally neutral in a way that rationing by price is not — a rich patient and a poor patient wait the same.
Explicit cost-effectiveness rationing is conducted by the National Institute for Health and Care Excellence (NICE), which assesses treatments against a threshold expressed in cost per quality-adjusted life year (QALY). A QALY weights a year of life by a quality factor between 0 (death) and 1 (perfect health).
NICE’s threshold has historically sat in the region of £20,000–£30,000 per QALY. Claxton and colleagues (2015), in Health Technology Assessment, argued on the basis of NHS spending data that the true opportunity cost of NHS expenditure — the health forgone elsewhere when money is spent on a new drug — implies a substantially lower threshold, in the region of £13,000 per QALY. If correct, this means NICE has been systematically approving treatments that reduce total population health, because the health they generate is less than the health displaced by the money spent.
This is one of the most consequential empirical claims in health economics, and it turns entirely on the distinction between willingness to pay for health and the opportunity cost of health spending in a fixed budget.
6. The QALY and Its Discontents
The QALY makes health commensurable. It allows a hip replacement to be compared with a cancer drug. Without it, cost-effectiveness analysis is impossible.
It is also ethically contested, for reasons students should be able to articulate:
- Ageism: a treatment for a young person generates more remaining QALYs than the identical treatment for an old person, mechanically favouring the young.
- Disability discrimination: a person with a permanent disability has a quality weight below 1. Restoring them to their baseline generates fewer QALYs than restoring a previously healthy person, so their life-years are valued less. The US Affordable Care Act explicitly prohibits Medicare from using QALY thresholds partly on these grounds.
- Aggregation: QALY maximisation is utilitarian. It is indifferent between giving one QALY to each of a thousand people and a thousand QALYs to one person. It contains no concept of severity, fairness, or rule of rescue.
Alternatives — the DALY, the equity-weighted QALY, multi-criteria decision analysis — each trade one problem for another. There is no neutral metric.
7. Exam Technique
The distinction that earns the marks
Write it explicitly, every time:
Adverse selection = hidden information, arising before the contract. The buyer knows their type.
Moral hazard = hidden action, arising after the contract. The buyer changes behaviour.
Cambridge A-Level structure
For “Evaluate whether healthcare should be provided by the market or the state”:
- Establish healthcare as a merit good with positive externalities (vaccination, herd immunity) and imperfect information.
- Apply the lemons model to insurance to demonstrate adverse selection.
- Note Rothschild–Stiglitz: competitive equilibrium may not exist at all.
- Concede that state provision does not solve moral hazard, and creates rationing by waiting.
- Use evidence: RAND on cost-sharing, Oregon on what insurance actually delivers.
- Evaluate: the choice is not market versus state but which failure you prefer, and for whom.
Common errors
- Describing moral hazard as “people being irresponsible.” It is a rational response to a price of zero.
- Claiming the Oregon experiment showed insurance “does not improve health.” It showed no significant effect on three specific physical markers within two years, alongside large effects on mental health and financial security.
- Treating the QALY threshold as a willingness-to-pay figure rather than an opportunity cost.
Summary
Healthcare violates the assumptions of the competitive model comprehensively, and it does so in ways that no institutional design fully repairs. Insurance solves the problem of uncertainty and creates the problem of moral hazard. Mandates solve adverse selection and create coercion. Public provision solves the pool problem and creates queues.
Arrow’s 1963 conclusion has held for six decades: the institutions we observe in healthcare — professional licensing, non-profit hospitals, physician ethics, universal coverage — are not market distortions to be swept away. They are the accumulated responses of societies to the fact that this market cannot clear on its own.
Exercises for Further Thought
1. The RAND experiment found that cost-sharing reduced healthcare utilisation without harming the average person’s health, but harmed the health of the poorest and sickest. Value-based insurance design proposes to solve this by setting copayments inversely to clinical value. But clinical value varies across patients: a statin is high-value for one person and near-worthless for another. Design an insurance contract that achieves the RAND spending reduction without the harm to the poorest, and identify precisely what information the insurer would need. Is that information obtainable? If not, is the harm unavoidable?
Suggested reading: Baicker, K., Taubman, S. L., Allen, H. L., et al. (2013). “The Oregon Experiment — Effects of Medicaid on Clinical Outcomes.” New England Journal of Medicine, 368(18), 1713–1722. Attend closely to the reported confidence intervals rather than the point estimates.
2. Claxton and colleagues argue that NICE’s threshold exceeds the true opportunity cost of NHS spending, meaning that approving a new drug at £30,000 per QALY destroys more health elsewhere than it creates. If this is correct, then a cost-effectiveness body acting in good faith has been systematically reducing population health for years. Why might a threshold set by willingness-to-pay diverge from one set by opportunity cost? Which should a public health system use, and what does your answer imply about whether health budgets should be fixed at all?
Suggested reading: Claxton, K., Martin, S., Soares, M., et al. (2015). “Methods for the Estimation of the National Institute for Health and Care Excellence Cost-Effectiveness Threshold.” Health Technology Assessment, 19(14), 1–503.
References
- Akerlof, G. A. (1970). The Market for “Lemons”: Quality Uncertainty and the Market Mechanism. Quarterly Journal of Economics, 84(3), 488–500.
- Arrow, K. J. (1963). Uncertainty and the Welfare Economics of Medical Care. American Economic Review, 53(5), 941–973.
- Baicker, K., Taubman, S. L., Allen, H. L., et al. (2013). The Oregon Experiment — Effects of Medicaid on Clinical Outcomes. New England Journal of Medicine, 368(18), 1713–1722.
- Claxton, K., Martin, S., Soares, M., et al. (2015). Methods for the Estimation of the NICE Cost-Effectiveness Threshold. Health Technology Assessment, 19(14), 1–503.
- Clemens, J., & Gottlieb, J. D. (2014). Do Physicians’ Financial Incentives Affect Medical Treatment and Patient Health? American Economic Review, 104(4), 1320–1349.
- Finkelstein, A., Taubman, S., Wright, B., et al. (2012). The Oregon Health Insurance Experiment: Evidence from the First Year. Quarterly Journal of Economics, 127(3), 1057–1106.
- Newhouse, J. P., & the Insurance Experiment Group (1993). Free for All? Lessons from the RAND Health Insurance Experiment. Harvard University Press.
- Pauly, M. V. (1968). The Economics of Moral Hazard: Comment. American Economic Review, 58(3), 531–537.
- Rothschild, M., & Stiglitz, J. (1976). Equilibrium in Competitive Insurance Markets: An Essay on the Economics of Imperfect Information. Quarterly Journal of Economics, 90(4), 629–649.
- Spence, M. (1973). Job Market Signaling. Quarterly Journal of Economics, 87(3), 355–374.