In 1994 a Harvard employee who wanted the university's most generous health plan for a family, a preferred provider plan run by Blue Cross Blue Shield, paid about $500 a year more than for the cheapest option. Then Harvard changed one rule. Starting with half its employees in 1995 and the rest in 1996, it contributed the same number of dollars whichever plan a person chose, so the employee paid the whole difference in premium. In 1995 the extra cost of the generous plan rose to roughly $1,000. In 1996 it passed $2,000. Its share of employees fell from about 20 percent to 15 percent and then to about 9 percent, and for 1997 the plan was disbanded. David Cutler and Richard Zeckhauser, the Harvard economists who told the story, put it in one line: the adverse selection death spiral twisted swiftly.2
The plan's doctors did not change; its members did. The 1995 leavers were younger than those who stayed, and Blue Cross found that the 1996 leavers had been 20 percent healthier than the average employee the year before.2 Each year the premium covered the people who remained, and each year they were sicker. The story pulls apart two things that both sound like sound pricing. One is pricing the individual risk: letting each person pay, or at least weigh the price against, what they are likely to cost. The other is keeping the pool solvent: holding enough people who will not need much that the average claim stays near the average premium. My argument is that in a voluntary pool these pull against each other. A pool works only while the people who need it least stay in it, so decide which risks you will refuse to price, and pay for that choice openly.
That is not a claim that pricing risk is wrong. Car insurers and life insurers do it every day, often accurately. Nor is it a claim that every pool is near collapse, or that subsidies cost nothing. The claim is narrower. Wherever membership is a choice and members know their own risk, a price that tracks the pool's average is one the healthiest will eventually decline. Someone has to pay to keep them, and the real question is whether that someone is named.
Each year's extra price covered the people still in the plan, so it rose as the healthy left.
The state employees' plan paid most of any premium. Its generous plan was still there in 1997.
Why the price cannot match the risk
The puzzle is older than the Harvard plan. In 1970 George Akerlof published "The Market for 'Lemons,'" a paper mostly about used cars. Partway through, he turns to health insurance and notes that people over 65 have great difficulty buying it. Then he asks the obvious question: why doesn't the price rise to match the risk?1
His answer is that the people applying know more about their health than the insurer does. As the price rises, those who expect to need little care stop applying first, so a higher price buys a worse pool, and at the limit, Akerlof wrote, no insurance sales may take place at any price.1 He drew a policy conclusion that still sounds modern: if everyone would pay their expected cost but no insurer can afford to sell them a policy, then Medicare may pay off on a cost-benefit basis.1
That is the core of the problem. A pool is a bet that the lucky will pay for the unlucky, made before anyone knows which they are. Once people do know, or think they know, the lucky have every reason to leave.
The spiral, one renewal at a time
Liran Einav and Amy Finkelstein, in a 2011 paper that explains the economics of selection almost entirely in pictures, describe how an insurer that sets this year's price at last year's average cost can watch its market gradually shrink until it completely disappears.3 The arithmetic is simple enough to do on graph paper, so here it is.
Imagine 1,000 members split into ten equal groups whose expected claims run from $150 a year to $1,050, an average of $600. Each member will join if the price is no more than their own expected cost plus $100, the value of not facing a large bill alone. The insurer charges 110 percent of last year's average claims. In year one it charges $660, which is the population average plus its margin. Every group that expects to cost less than $560 decides the plan is a bad deal and leaves. The 500 who stay are the expensive half, averaging $850, so next year's premium is $935. Now only the three costliest groups stay, then two, then one. In the fifth year nobody buys at all.
The gold line is the pool when every member is left to price their own risk against a premium that tracks the pool. Each renewal removes the members who were holding the average down. The blue line adds one thing, a flat $300 contribution from the sponsor toward every member's premium, whichever group they belong to. Of the same $660 premium a member now pays $360, a good enough deal for eight groups out of ten. One more group leaves in the second year, and then the pool settles at 700 members, paying $525 each on an $825 premium. The contribution made nobody healthier. It changed the price the healthy compare with their own risk, which was enough to keep the pool solvent.
Harvard had a contrast next door. The Massachusetts Group Insurance Commission, which covers state employees, paid 90 percent of the premium, and from 1995 85 percent, whichever plan an employee chose. Enrollment in its most generous plan drifted from about 30 percent to 27 percent between 1994 and 1997, a decline but not a collapse. Cutler and Zeckhauser credit the commission's avoiding a spiral partly to not adopting an equal-contribution rule.2
None of this means Harvard was foolish. Cutler and Sarah Reber estimated that employees were very price sensitive, with a short-run elasticity of about minus 2, and put the welfare loss from the selection at about 2 percent of baseline spending. But the competition the rule created cut Harvard's premiums by nearly 10 percent relative to the Boston market, and they judged the reform welfare enhancing overall.4 The price signal did real work. It also destroyed a plan some employees valued. Both effects came from the same rule.
The best case for pricing the risk
The strongest argument for the gold side is that an accurate price can do more than divide a bill. In 1976 Michael Rothschild and Joseph Stiglitz worked out what competition does when insurers cannot see who is risky. Insurers can sort customers with a menu: full cover at a high price, which the high risks take, and thinner, cheaper cover that only low risks want. But then the low risks cannot buy complete insurance, the market may have no equilibrium at all, and the low risks are worse off merely because the high risks exist, a loss the authors called completely dissipative, since the high risks gain nothing from it.5 Better information about each person would dissolve the problem; that is the promise of pricing the individual risk.
Sometimes the promise is kept. Yizhou Jin and Shoshana Vasserman studied a large American auto insurer's voluntary program that tracks driving for a short period in exchange for a possible discount. Safer drivers self-selected into monitoring, as theory predicts, and the drivers who opted in became 30 percent safer while monitored.6 Here the precise price changed behavior, not just who paid. The authors also noticed the other edge: the insurer had to offer large discounts to win opt-in while keeping the unmonitored pool from unraveling.6 Even a good risk price is a withdrawal from somebody's pool.
Health differs from driving in one important way. A driver can brake more gently next month. A person with diabetes cannot make it go away by being watched. When the risk is mostly not a choice, a sharper price mostly moves the bill rather than shrinking it.
What a perfect price would cost
Suppose the insurer could see everyone's health and price it exactly. Benjamin Handel, Igal Hendel and Michael Whinston built a model of the exchanges created by the Affordable Care Act to ask what that would do. Under the ACA's pricing rules, they found substantial unraveling, with little coverage beyond the least generous plan, and in their main case everyone ending up in it, even when prices could vary by age. Allowing prices to vary with health status did reduce the adverse selection. But it exposed people to a different risk, that of being reclassified as expensive after falling ill, and they estimated that loss at about $3,082 a year for each person, roughly five times the $619 lost to adverse selection under a single community price.7
That is the deepest problem with the gold side. The insurance most people want most is insurance against becoming a bad risk, and a price that follows your health year by year cannot sell it. A precise annual price protects you against this year's bills and leaves you exposed to the diagnosis that will set next year's.
A pool losing its healthy members
This year has supplied a large example. The enhanced premium tax credits for the ACA marketplaces expired at the end of 2025. Effectuated marketplace enrollment, people who actually paid, fell from 22.1 million in 2025 to 19.2 million in February 2026, according to KFF, and enrollees faced an average 114 percent increase in their premium payments to keep the same plan.8 The causes are disputed; the administration attributes the drop to its efforts against fraud, and KFF notes it does not yet know how many people became uninsured.8
The insurers, though, are pricing the next year as if the healthy left first. In the rate filings for 2027 collected by Peterson-KFF, the median proposed increase is 15 percent, and insurers expect a sicker pool to push premiums about 4 percentage points higher than they would otherwise be, with healthier enrollees more likely to drop coverage. One Texas insurer, Antidote Health, wrote in its filing that healthier consumers are more likely to lapse coverage and added a 6 percent morbidity adjustment.9 That is Akerlof's mechanism, written into a rate filing.
New Mexico was the only state to fully replace the expired federal credits with its own subsidies, and the only state where effectuated enrollment grew, by 14 percent. In Ohio and Oklahoma it fell by just over 32 percent.10 That is not a controlled experiment, but it has the shape of Figure 2: where someone paid to keep the price low enough for the healthy, the pool held.
The quiet defenses
Pools that cannot compel membership defend themselves in other ways, as dental benefits show. Many dental plans still cap what they pay each year at about $1,000, which the American Dental Association describes as a level established some 40 years ago; by the National Association of Dental Plans' figures, 32.8 percent of in-network annual maximums sit between $1,000 and $1,500, and only 3.4 percent of patients reach the typical maximum.11 Dentists read the cap as a benefit that has not kept up with inflation, and it is. It is also a defense of a voluntary pool. A low cap makes the plan worth less to the person who joins the month before a crown, and so does any wait before major work is covered. That is my reading, not the plans' stated reason, but it explains why the caps last: a richer benefit is exactly what would draw in the members a voluntary pool can least afford.
Those defenses keep the pool solvent by making it worth less. Better ways have a visible price.
Pay for the pool on purpose
The canonical answer, in Einav and Finkelstein's words, is to mandate that everyone buy insurance; the other common remedy is to subsidize it, and a large enough subsidy brings everyone in.3 An employer does the same thing with a contribution that does not track each plan's cost. A state does it with reinsurance, which pays a share of each enrollee's claims above a threshold and so takes the most expensive members' costs out of everyone's premium. Seventeen states run such programs under federal waivers, and from 2018 through 2023 they lowered benchmark marketplace premiums by between 3.75 and 41.17 percent compared with having no waiver; Alaska's program, which pays for enrollees with listed high-cost conditions, accounts for the top of that range.12
Societies also decide which risks may not be priced at all. In 2011 the European Court of Justice ruled in the Test-Achats case that, from December 21, 2012, insurers in the European Union could no longer use sex as a factor that changed premiums and benefits.13 The court decided that a difference insurers could measure would be pooled anyway. Whether one agrees, it is the honest form of the choice: a line drawn in public, with its cost borne in plain view.
Objections, considered
Risk pricing is fairer. Each person pays for what they are likely to cost. That is a coherent view of fairness at a single moment, with real force for risks people control, like driving. Over a lifetime it is less attractive. The healthy 30-year-old who pays a precise price today is buying no protection against the diagnosis at 45 that will reprice them, and Handel, Hendel and Whinston's estimate is that this reclassification risk is the larger loss.7 Before anyone knows their luck, a pool is the only way to buy that protection.
Pools do not really unravel. The mandate repeal proved it. When the federal penalty for going uninsured fell to $0 in 2019, the individual market shrank by only about 5 percent, to 13.7 million people. So, the objection runs, death spirals are a textbook curiosity. But look at where the decline happened. All of it was among unsubsidized enrollees, whose numbers fell 10 percent; the people receiving federal subsidies went from 9.2 million to 9.3 million.14 The subsidized were insulated from the price their own risk implied. The pool held where someone was paying to keep it, which is the argument of this essay, not a refutation of it.
Cross-subsidy is a hidden tax. A flat contribution or a uniform price moves money from the healthy to the sick without saying so, and it blunts the price signals that made Harvard's premiums fall. Partly right. The remedy is not to stop pooling but to stop hiding it. A contribution or a reinsurance fund has a budget line and a number someone can debate. A spiral simply happens, and the people left in the plan pay for it.
A pool stays solvent only while the people who need it least choose to stay.
Before the next renewal
- Who would leave first? If the price your members pay rose 10 percent, name the group most likely to go, and estimate what their leaving would do to next year's average claim.
- Does the sponsor's money follow the plan's cost? A fixed dollar contribution sends the full price difference to the employee; a percentage shares it. Know which you have chosen.
- Which risks have you decided not to price, and who pays for that? Write the answer down, with a budget line beside it.
- Do you measure who leaves, not just how many? A count of departures says the pool is smaller. The claims history of the people who left says whether it is still solvent.
Back to the receipt. The gold lines are perfectly reasonable: each year the extra price reflected what the plan's members cost. The blue lines are what that reasonableness bought. Twenty percent of the staff became fifteen, then nine, and the healthiest leavers took the plan's average with them. Harvard and Blue Cross did finally agree to close the PPO, but by then the rule had already decided, one renewal at a time.