232 | Amy Finkelstein on Adverse Selection and Hidden Information
What this covers
Amy Finkelstein, an economist at MIT and MacArthur Fellow, speaks with Sean Carroll about why insurance markets fail and what economic evidence reveals about attempted fixes. The core problem is adverse selection: buyers possess private information about their own risk that insurers cannot fully capture, even with big data and sophisticated algorithms. This asymmetry creates predictable failures—not because of greed, but because profitability in insurance depends on which customers buy, not just how many. When an insurer raises prices to cover losses, it inadvertently drives out healthier buyers and leaves only riskier ones, triggering a death spiral that can collapse the market entirely. Finkelstein illustrates the problem historically: 17th-century mathematicians Edmond Halley and Abraham de Moivre calculated mortality tables accurately but mispriced annuities because they failed to account for selection—buyers nominated unusually long-lived individuals, helping bankrupt the French crown.
The conversation spans how every policy intervention reshapes rather than solves the selection problem. Banning risk-based pricing through community rating, as New York and New Jersey did, creates fairness but triggers adverse selection on a larger scale. Government-run programs like Medicare Advantage face selection games with private insurers who cream skim the healthiest via gym memberships and other inducements, forcing the government into endless whack-a-mole adjustments. Mandates without sufficient teeth fail; randomized controlled trials—like the Oregon Medicaid lottery—reveal that subsidies and penalties matter far more than mandates themselves. Finkelstein distinguishes moral hazard (behavioral changes after buying insurance) from selection (who buys in the first place), using the Oregon experiment to show insurance increases both preventive care and emergency room use. Throughout, she emphasizes that empirical evidence, not theory or political hope, must adjudicate trade-offs—there is no free lunch, only hard choices between access, cost, and choice.
Finkelstein argues that insurance markets predictably fail not from greed but from adverse selection — buyers' private information about their own risk — and that every government remedy (community rating, mandates, subsidies, privatized Medicare) faces unavoidable trade-offs that empirical evidence, not theory or hope, must adjudicate.
- Insurance differs from ordinary goods because profitability depends on WHICH customers buy, not just how many
- Buyers retain private information about their own risk even in the age of big data, driving death spirals
- Each fix (guaranteed issue, mandates, risk-adjustment) creates new selection games, so empirical testing is needed to reveal real trade-offs
Insurance creates a distinctive market failure where buyer information about personal risk drives adverse selection that can destroy markets entirely despite mutual gains being available.
- Insurance differs fundamentally from ordinary goods like broccoli or smartphones because an insurer's costs and profitability depend not just on the price and number of customers, but on WHICH customers buy — high-risk, accident-prone buyers cost more than low-risk ones — which is why there may be a role for government mandating insurance purchase but not broccoli consumption.
“When the supermarket sells it's broccoli, the profits and the profitability of that broccoli depend only on the price it charges and at how many customers buy it, it does not depend on which customers buy it, that is not true for insurance”
- In a selection market, the seemingly sensible response to losses (raising prices) exacerbates the selection problem: each price increase drives out the relatively healthier buyers and leaves only riskier customers, which forces further price hikes in a 'death spiral' that can make the market disappear entirely.
“The problem in a selection market, like an insurance market, is when you raise your price, you just exacerbate the selection problem, now it's only the really, really high risk”
- An actuary may predict an individual's mortality more precisely than the individual can, so it is not that the buyer 'knows more' in absolute terms; the buyer's informational advantage is reactive — the insurer must move first by posting a price for a group, and the buyer then knows whether they are more or less risky than the typical member of that priced group, which is the source of selection.
“whatever the insurer knows, they use to offer the insurance... once they give me the price, I know that among that group of middle-aged female economists with great senses of humor, I know something about whether I'm more or less risky than the typical. That's where my informational advantage comes in”
- Adverse selection is a distinctive form of market failure because it can destroy a market even when everyone behaves rationally and even when every buyer would willingly pay an actuarially fair price plus a margin that lets the insurer profit — the market simply may not exist despite mutual gains being available on the table.
“you could have a situation in which everyone values insurance. They're actually willing to pay a price that's actuarially fair based on their own risk... and yet that market may not exist”
- Adverse selection is a global, market-driven phenomenon not confined to insurance or any country: it arises in any market where different customers impose different costs on the seller and the customer knows something the seller doesn't — examples include all-you-can-eat restaurants and college income-sharing agreements, where Hendren and Herbst show that even controlling for college and major, students with private knowledge that they'll earn little are the ones who want the income-sharing agreement.
“selection markets can occur whenever the different customers are going to impose different costs on the seller and the customer knows something that the seller doesn't”
Empirical randomized trials reveal insurance's actual causal effects, overturning intuitive claims and showing real trade-offs that complicate simple expansion arguments.
- The Oregon Medicaid randomized trial showed that covering low-income uninsured people increases both primary and preventive care (more doctor visits, more mammograms, more diabetes blood-sugar checks) AND emergency room use, contradicting the widely-claimed argument that insurance gets people out of the ER — so one should not advocate insurance expansion on the grounds that it reduces ER use.
“covering the low-income uninsured with Medicaid does get them into more primary and preventive care, and they have more doctor visits. Their rates of mammograms go up... and yet they also go to the emergency room more”
- Moral hazard is distinct from adverse selection: it is the behavioral effect whereby identical people who are randomly given insurance subsequently use more (drive less carefully, visit the doctor more) — confirmed by the randomized Oregon Medicaid experiment — whereas selection is about who chooses to buy in the first place; the two can be disentangled by randomizing the PRICE of insurance, since buyers willing to pay a higher price show higher healthcare utilization than low-price buyers even though both groups are insured.
“the people who have the insurance will now put their foot on the gas pedal a little more and drive a little less carefully... once people get health insurance, including randomly assigned health insurance, we did a randomized evaluation of this in the state of Oregon”
- Empirical economics has undergone a 'credibility revolution' (Angrist's term) emphasizing isolating causal effects from spurious correlation, with the randomized experiment as one of the most valuable tools — applying medicine's randomized-trial logic to policy, as in the 2008 Oregon Medicaid lottery, where the state randomly allocated Medicaid to some low-income uninsured adults and not others to measure true causal effects.
“there's been a real revolution, I would say, in empirical work. My colleague, Josh Angrist, who won the Nobel Prize a couple of years ago for some of this work, refers to it as the credibility revolution”
- While a quarter of all medical spending on the elderly does occur in the last 12 months of life, the common inference that we waste money on people known to be dying is false: using sophisticated machine learning on past health records, it is very hard to identify, at the time care is given, patients with very high (90-95%) annual mortality — even late-stage 80-year-old cancer patients rarely exceed a 70% predicted annual mortality — so high death rates among the sick reflect that we spend on sick people, not that we knowingly spend on the soon-to-be-dead.
“a quarter of all medical spending on the elderly occurs in the last 12 months of life, and it turns out that is actually true... it's very, very hard to identify a group of people that we're spending a lot of money on who are actually going to die with 90-95% probability”
Insurers and governments respond to adverse selection through mechanisms like pricing, underwriting, risk-adjustment, and product bundling that attempt to manage or exploit selection dynamics.
- An insurance applicant must answer questions truthfully but has no obligation to volunteer information not asked; lying is possible since insurers rarely verify upfront, but it is a very bad idea because verification happens at claim time and false statements can lead to denial after years of paying premiums — Finkelstein illustrates with answering all auto-insurance questions truthfully while not disclosing that she barely drove.
“you could lie. I wouldn't advise it because it's very expensive to verify all this information... it's a very, very bad idea to try and shade the truth because you'll end up paying all these premiums. Then in the event that you actually need the insurance, then you could be denied”
- Insurers offering discounted gym memberships and employers offering workplace wellness programs do so not because these make people healthier but because they attract customers who are already healthier at baseline — Cooper and Trivedi show gym-membership bundles draw measurably healthier buyers, and the Jones-Molitor-Reif randomized trial at the University of Illinois found wellness programs produced no health or productivity gains but were preferentially chosen by already-healthy employees, revealing the true purpose is selection rather than behavior change.
“when insurers start offering these discounted gym memberships bundled with the health insurance product, they attract customers who are already healthier. You can measure their health before they buy the insurance”
- When the government intervenes, the selection problem shifts from a game between customer and insurer to one between government and insurer, which the government tends to lose: in privatized Medicare Advantage, the government pays private insurers to take customers, insurers cream-skim the healthiest by age and gender (e.g. via gym memberships), the government counters by risk-adjusting on past medical claims, and insurers stay a step ahead in a continuing game of whack-a-mole.
“the selection problem can cease being a game between the customer and the insurer and can become a game between the government and the insurer. And then the government is often on the losing side of this”
- Pet health insurance is priced exorbitantly relative to the risk — a 12-year-old bulldog can cost $4,300 in annual premium against a maximum payout of $5,000 — because the owners who insure aging dogs are precisely those who know the dog will need expensive care AND who are the type to pursue high-tech invasive surgery rather than euthanasia, so the high price, waiting periods and deductibles are the insurer's response to severe adverse selection.
“a 12-year-old bulldog, if you want to insure him, the annual premium is $4300. And here's the kicker, the maximum payout is $5000”
- Despite insurers having vast data, sophisticated actuaries, and algorithms, buyers still possess private information about their own risk that insurers cannot capture — documented cases include people at risk for Huntington's chorea buying nursing home insurance at higher rates (especially after testing positive), and people who buy life insurance dying sooner than observationally identical non-buyers even after medical exams, family-history questions and lifestyle screening — so big data does not eliminate adverse selection.
“the people who buy life insurance are more likely to die in the next 10 years... they actually come and do a medical exam... they ask about that... they ask about your smoking and your skydiving... So there are cases where we don't know what it is, but the data are very clear that there is something that people know above and beyond what the insurance company is able to figure out”
- Finkelstein deliberately bought a homeowner's policy with the highest possible deductible ($10,000 versus the typical $500-$1,000) because insurance is most valuable for catastrophic risk, and crucially because she reasons that people who choose low deductibles do so precisely because they privately know their home has many problems and will generate many claims — so the low-deductible premium is loaded with adverse selection, making the high-deductible option cheaper and, over time, worth it even after two large losses.
“I would argue based on the type of research I do that the reason the premium was a lot lower is because people who know that there's a lot of problems with their home and they're gonna have a lot of claims want a low deductible”
Every policy response to selection involves genuine trade-offs with no facile solutions; mandates, subsidies, price regulation, and choice restrictions each create their own problems.
- To solve the selection problem through subsidies rather than penalties, the economically correct policy is to subsidize the young and healthy people — those who seem to need help least — rather than the old and sick, because drawing healthy buyers into the pool is what stabilizes pricing, even though this is politically unpalatable and unlikely to pass 'the sniff test with Joe Public.'
“if you're trying to design those subsidies to deal with the selection problem, the people that you should be subsidizing are the people who, in some sense, you feel like might need the subsidies the least... You want to subsidize the young, healthy people. That's actually how you'd solve the selection problem”
- There are no facile solutions to insurance market problems; every policy involves real trade-offs — for example, mandating a low minimum coverage recreates the selection problem on the margin of how much insurance to buy, while mandating high minimum coverage forces people to buy more than they want, and restricting choice to prevent market collapse can leave individuals (like a man who got sick and could no longer obtain comprehensive Medicare supplemental coverage) without the products they need.
“there aren't facile solutions. There really are trade-offs”
- Although forcing community rating looks economically nutty because it creates artificial private information, it can be justified if what people want is not insurance for this year's claims but insurance against the lottery of being born an unhealthy person or bad driver — i.e. insurance behind a Rawlsian veil of ignorance — which markets cannot provide in utero but the government can approximate by mandating equal pricing.
“if what you want is not insurance for my claims this year... but I want insurance against being an unhealthy individual, behind the Rawlsian veil of ignorance, then it makes perfect sense to have community rating charge everyone the same”
- Mandating insurance does not by itself solve selection because, like telling a toddler to eat broccoli, saying people must buy insurance doesn't make it so; evidence from Massachusetts shows that the penalties or subsidies attached to the mandate matter far more than the mandate itself for getting healthy customers into the market — a mandate without teeth is like a Bobby shouting 'Stop or I'll say stop again.'
“what turns out to matter a lot, and there's evidence of this from Massachusetts, is what kinds of penalties are put in place for people who don't buy insurance. And when those penalties are in place, you see that that's much more effective at getting the healthy customers into the market than simply saying you must buy insurance”
- Banning risk-based pricing through community rating and guaranteed issue (as New York and New Jersey did in the 1990s private health markets) makes coverage fairer but, by forcing the same price on everyone, draws older sicker buyers, drives prices up, pushes out the healthiest, and can spiral the market completely out of existence — so 'guaranteed issue is guaranteed to create issues' and equal treatment can leave everyone equally without access.
“New York and New Jersey did this in the 1990s... you can't price health insurance differently for people based on their preexisting conditions... In the case of New York and New Jersey, it drove the market completely out of existence”
Insurance fundamentally offers valuable risk pooling that reduces individual uncertainty, yet market failures and selection problems often prevent it from delivering those gains.
- Reinsurance exists because insurers face correlated shocks: an insurer concentrated in earthquake or flood coverage in Florida or California may have no claims some years and claims from 80% of customers in others, so it lays off some risk to a larger, more diversified pool with a bigger capital stock that can bear it — unlike the independent coin-flip case where a large pool nearly eliminates the chance everyone is wiped out.
“if you're, say, an insurer who's insuring against natural disasters, earthquakes, floods... and you're sort of geographically concentrated in, say, Florida or California, some years you may have no costs and some years you may have 80% of your customers have a claim. So, then in some sense, you may want to lay off some of that risk to a larger pool”
- Insurance offers something economists call a free lunch — a measure of security and certainty in a dangerous and uncertain world from which everyone can benefit — yet despite this real and valuable potential, insurance often fails to deliver, with consequences including ruined businesses, policies, and lives.
“that's why I think it's exciting, it's the tantalizing what economists call a free lunch, everyone can benefit. But what our book is about is the fact that despite I think this very real and potentially valuable service, insurance often fails to deliver on its promise”
- Buying insurance is rational for most people even though on average you pay more than you get back, because risk-averse individuals will willingly pay slightly above the actuarially fair price (e.g. $11 to avoid a 10% chance of losing $100, expected cost $10) to secure certainty and peace of mind, which raises rather than lowers their expected utility.
“you might be willing to pay $10 for sure, and then 90 times out of 100, you're actually not gonna lose any money... in fact, maybe you'd even be willing to pay like $11 bucks for that”
- Insurance works by pooling risk: if a million people each face a one-in-ten chance of losing $100, the law of large numbers ensures roughly one in ten actually lose it, so if everyone contributes a little the pool can cover the unlucky few, smoothing individual risk across the community rather than buying from a small entity that cannot absorb fluctuations.
“you take 100 people or a million people, each of whom has a one in 10 chance of losing that $100 and you spread that risk around. The law of large numbers will work... if we all give a little bit of money, that can cover the losses for the few”
Historical cases from 17th-century annuities to modern pet insurance illustrate how adverse selection has repeatedly undermined insurance markets across contexts and centuries.
- Resistance to causal empirical results in economics often comes from well-intentioned people who hope expanding health insurance will both improve lives and save money, but the honest finding is that there is no free lunch — insurance expansion may be worth its cost but it is not free, and we face real trade-offs many people wish away.
“having to tell people that actually, "No, there's no free lunch. We face real trade-offs." And it may well be worth it to spend that money, but it's not gonna be free”
- Edmond Halley and Abraham de Moivre correctly computed mortality tables for the general population in the 17th century, but mispriced survival annuities because they failed to anticipate adverse selection: the people on whom annuities were purchased were not randomly drawn — buyers nominated unusually long-lived individuals (young French girls who had survived smallpox, hardy old Scotsmen), which helped bankrupt the French crown.
“they calculated the mortality tables correctly for the population in that they had data for, but what they weren't thinking of is the fact that once you start selling these survival contingent streams of payment, these things called annuities... the set of people whose... On whom those bets are placed are not randomly drawn from the population”
The United States handles health insurance selection problems differently and problematically compared to other high-income countries.
- Among high-income countries, the United States' approach to dealing with the selection problem in health insurance is unique and, in Finkelstein's view, uniquely problematic relative to how other high-income countries handle it.
“in the particulars of health insurance, the way the United States has tried to deal with this problem is unique. And we have come to conclude uniquely problematic relative to how other high-income countries have tried to deal with this”