The Asymmetry

The theorems that crown the classical model — that a competitive equilibrium wastes nothing, that the invisible hand turns private greed to public good — carry a clause so quiet it is easy to miss: that everyone knows what they need to know. The buyer can judge the good, the lender the borrower, the insurer the risk; or if they cannot, the price stands proxy for the knowledge, summing into a single number all that the market has learned. Strike out that clause, and the theorems fall. This volume studies the trades for which the clause was always false — where one party holds, privately, some fact about the bargain that the other cannot see and that changes what the bargain is worth. Such information is called asymmetric, and the asymmetry is not a minor friction to be assumed away. It is a wedge driven into the foundations.

The wedge takes two shapes, and to keep them distinct is to hold the whole subject in order. The first is hidden type: a fixed fact about a party, settled before any deal, that the other cannot observe. The used car is sound or rotten and the seller knows which; the applicant is healthy or ailing and carries the knowledge in his body; the worker is diligent or idle by disposition. The type exists at the door, and the question is who will choose to come through it. The second is hidden action: a choice a party makes after the deal is struck, which the other cannot watch. The insured, once covered, locks his bicycle or does not; the borrower, once funded, builds the prudent business or stakes it on a long shot; the manager, once hired, labors or loafs. The action comes after the signature, and the question is what the protected party will do. From hidden type grows the failure of adverse selection, a sickness in who trades; from hidden action grows the failure of moral hazard, a sickness in how they behave once trading. One root — the unobservable — two diseases, striking at different hours and yielding to different cures.

That this belongs among the failures, and not among the ordinary frictions every market carries, is worth insisting on. Here is a market that may be perfectly competitive, crowded with buyers and sellers, no monopolist in sight and no cost spilling onto a bystander — and it fails all the same. Trades that would enrich both parties do not occur, because the gains cannot be proven; or the trades that do occur are the wrong ones, the bad driving out the good. The waste is real and the cause is singular: not too few sellers, not too much power, not a missing market for some side effect, but knowledge held unequally across the bargain. It is a failure of information, and it is distinctive in one hopeful respect that the later plates will dwell on — that unlike a monopoly, which has no wish to dissolve itself, a market crippled by asymmetry strains constantly to cure itself, breeding signals and screens and reputations in its own defense.

Economics was slow to admit the wedge, having built its rigor on a fiction of full information — the auctioneer who knew every value, the agent who knew every fact, the price that revealed all. The economics of information, arriving in a remarkable burst around nineteen-seventy, brought the gap inside the model and showed how much the older results had leaned on the assumption that abolished it: Akerlof's lemons, Spence's signals, the screening and the credit rationing that Stiglitz and his collaborators drew out of the same insight. The work earned its authors the Nobel three decades on, and remade labor economics, finance, insurance, and the economics of development in its image. The discipline had, in effect, discovered that the most interesting thing about a market is often not the price but what the price fails to say.

Consider the market that gave the failure its homely name. A man wishes to sell his used car, and he knows, as only an owner can, whether it is sound or a wreck — a peach or, in the American idiom that became the technical term, a lemon. The buyer knows no such thing. He sees a car of a certain make and age and mileage, and behind that visible surface a quality he cannot read. What, then, will he pay? Not the price of a peach, for the car before him might be a lemon; not the price of a lemon, for it might be a peach. He will pay, if he is sensible, the price warranted by the average — by his best guess of what the typical car offered at this price is worth. It is the only price his ignorance permits.

And here the trouble begins, in a turn of reasoning that is the engine of the whole subject. At a price set by the average, the owner of a car better than average will not sell. His peach is worth more to him than the averaged price on offer; he keeps it, or asks more and finds no taker. So the cars that remain on the market are the below-average ones — and the average quality of what is actually offered falls. The buyer, no fool, anticipates this and lowers what he will pay; which drives out the next tier of cars, those just above the new and lower average; which lowers the average again; which lowers the price again. The market does not settle. It unravels, each turn of the screw expelling the best of what remains, until in the limit only the worst is left to trade — the lemons, alone, as the name foretold — or until no price will support any trade at all and the market simply vanishes.

The result is more startling than any deadweight triangle. It is not that a few trades at the margin go unmade; it is that an entire market, in which every conceivable trade would have left both parties better off, can fail to exist. The buyer values each car above what its owner asks — gains from trade lie everywhere — and yet, because no one can prove which cars are which, none of the gains are reaped. Hidden quality has done what no monopolist and no tax could do: it has destroyed the market outright. And the mechanism is general. It is George Akerlof's lemons of nineteen-seventy, but its shadow falls across every market where one side knows a quality the other cannot: the insurance pool the healthy abandon, the credit market the safe borrowers flee, the labor market the able workers quit when all are paid the same averaged wage. Wherever a hidden type sorts the eager from the reluctant, the lemons logic is at work.

It is worth marking exactly what does the damage, because the cure must be aimed at it. The damage is not done by low quality; a market of nothing but lemons, openly known to be lemons, would function perfectly well, clearing at a lemon's price. The damage is done by hidden quality — by the buyer's inability to distinguish, which forces him to price by the average and so to punish the good along with the bad. It follows that every remedy this volume will examine is, at bottom, an attempt to make quality visible: to let the peach prove itself a peach, by a warranty its owner can afford and a lemon's owner cannot, by a reputation built over many sales, by an inspection the buyer trusts. The lemons market is not a market without hope. It is a market under pressure to invent the institutions that will save it.

Insurance is the native soil of adverse selection — the term is the insurers' own, older than the economics that formalized it — and the mechanism there wears its plainest face. To insure is to pool risk: the many who pay premiums cover the few who suffer losses, and each buys, for a small certain cost, escape from a large uncertain one. The arrangement works beautifully when the insurer can see the risk he is taking on. It begins to fail the moment the applicant knows his own risk better than the insurer does — which, for health, for life, for the soundness of a venture, is nearly always.

For then the eagerness to buy is itself a signal, and a damning one. The person who most wants health insurance is, on average, the person who most expects to use it — the one who feels the symptom the insurer cannot see. So the pool that actually forms is sicker than the population at large: selection has worked adversely, drawing in the bad risks and leaving out the good. The insurer, finding his pool more expensive than the average citizen, must raise the premium to cover it. But at the higher premium, the healthiest of his customers — those whose expected claims fall well below what they are now charged — conclude that the policy is no longer worth its price, and leave. Their departure makes the remaining pool sicker still; the premium must rise again; the next tier of the relatively healthy departs; and the market spirals, in the insurers' own grim phrase, toward death. At the end of the spiral sits a small pool of the very sickest, paying premiums near their own high expected losses, while the great middle of the population — moderate risks who would gladly have bought insurance at a fair price — go uncovered for want of a pool they could afford to join.

The diagram makes precise what the spiral does in time. The peculiarity of insurance, which no ordinary market shares, is that the cost of serving a customer depends on which customer it is — and the price, by determining who chooses to buy, determines the cost. So the cost curves slope downward: widen coverage by lowering the premium, and the people you draw in are progressively healthier, pulling the cost of the marginal customer down, and the average of the pool down behind it. Competition drives the premium to the average cost of whoever has chosen to insure; but the efficient extent of coverage is set by the marginal cost — by whether the next person values insurance above what it costs to cover him, not above the average dragged up by the sick. Because average cost sits above marginal cost, the premium sits above the efficient price, and a wedge of people who would have been worth insuring are priced out. That wedge is the signature of the failure: not collapse, usually, but chronic under-insurance.

Knowing the disease, one can read the institutions of insurance as so many treatments for it. The mandate — the requirement that everyone hold coverage — abolishes the selection at a stroke by denying the healthy their exit; it is the logic behind compulsory auto insurance and the contested individual mandate of health reform. Risk classification — sorting applicants by observable markers of risk, age and history and occupation, and pricing each class on its own — shrinks the asymmetry within each pool by making more of the hidden type visible. The group policy sold through an employer pools people selected by their jobs rather than their health. Waiting periods, exclusions, and medical underwriting all gnaw at the same problem from different sides. The architecture is elaborate because the force it resists is strong; an insurance market, like the lemons market, survives by inventing the devices that hold its pool together.

The lemons and the pool were failures of hidden type — of a fact fixed before the deal. We turn now to the second great failure, which springs from hidden action — a choice made after the deal, and unseen. Its name, borrowed like adverse selection from the insurance trade, is moral hazard, and its principle is one of the most reliable in all of economics: shield a person from the consequences of his actions, leave those actions unobserved, and his care will slacken in a direction one can name in advance. The insured against theft grows careless with his locks. The bank that knows it will be rescued reaches for riskier bets. The employee paid a flat salary, his effort unwatched, exerts rather less of it. In each case the party who bears the loss is not the party who controls the risk, and the wedge between them — the gap the contract cannot close, because effort cannot be seen — is where the waste accumulates.

The hazard comes in two tenses. Before the loss, it dulls prevention: the covered party takes fewer pains to keep the bad event from happening, since its cost will fall on another. After the loss, it swells consumption: the party whose bills are paid by the insurer uses more of whatever the insurer is buying, since it is free to him at the point of use. The second tense is the easier to draw, and the clinic supplies the example. A patient whose insurance covers the whole cost of care faces, at the moment of decision, a price of nothing — and a rational person facing a price of nothing consumes until the very last increment is worth nothing to him, far past the point where the care was worth what it cost to provide.

The remedy, the diagram suggests, is to put some of the price back. If the insured bears a part of the cost, the price he faces is no longer zero, and his consumption retreats toward the efficient quantity. So the deductible makes him pay the first slice of any loss; the co-payment and the co-insurance leave a fraction of every bill on his side of the ledger; the policy limit caps what the insurer will cover; experience rating raises tomorrow's premium after today's claim, restoring over time the cost of carelessness. Where consumption cannot be priced, it can sometimes be watched: the insurer requires the alarm, inspects the wiring, audits the claim. Every one of these devices works by the same logic — returning to the actor some share of the consequence his action sets in motion. And every one of them, in doing so, hands him back a piece of the very risk that insurance existed to take away. That tension — between disciplining the action and bearing the risk — is the knot the next plate must untie.

What gives moral hazard its grip on the public imagination is its sheer reach. The bailed-out bank is its grandest instance — the institution too large to fail, gambling on the rescue it expects, is moral hazard written across the whole financial system. But the same shape recurs wherever protection meets an unwatched choice: the shareholder shielded by limited liability, the tenant who lets the maintenance slide, the contractor on a cost-plus deal with no reason to economize, the borrower who, once the loan is his, stakes it on the long shot. The phrase has escaped economics into ordinary speech precisely because the thing it names is everywhere — anywhere, that is, that someone has been insulated from a risk that remains his to influence.

Moral hazard, formalized, becomes the problem of the principal and the agent — the most worked-over relationship in modern microeconomics, and the one that joins this volume to the institutions that follow it. A principal engages an agent to act on his behalf: the owner hires a manager, the shareholder a chief executive, the landlord a tenant, the insurer takes on a policyholder. Two facts make the engagement fraught. The agent's interests are not the principal's — the manager would rather work less, the tenant tend the land less — and the agent's effort cannot be observed. The principal would dearly like to pay for effort and be done with it; but effort is precisely the thing he cannot see. He is forced to contract instead on what he can see, which is almost always the outcome — the profit, the harvest, the sale — and the outcome is a treacherous proxy for the effort behind it.

Treacherous because the outcome is the joint product of two things: the agent's effort, which the contract wants to reward, and luck, which it does not. The harvest reflects the tenant's labor and the season's rain; the firm's profit reflects the manager's diligence and the market's mood. To tie the agent's pay tightly to the outcome — to give him high-powered incentives — is to spur his effort, for now he keeps much of what his effort produces. But it is also to expose him to all the luck the outcome carries, to make his income lurch with rain and market in ways he cannot control. A risk-averse agent loathes that exposure and will not accept it without a premium; high-powered incentives, in other words, are expensive, purchased in the coin of the risk they dump on the agent and the compensation he demands for bearing it. Here is the trade-off that organizes the entire theory: incentive on one side, insurance on the other. The contract can motivate the agent, or it can shield him from risk, but it cannot fully do both, and every real contract is a settlement of the quarrel between them.

Where the settlement falls depends on a few intelligible forces. The more responsive the agent's effort is to reward, the more incentive is worth its cost; the more risk-averse the agent, the more the contract leans toward insuring him; the noisier the outcome — the more luck drowns out effort — the flatter the optimal contract, since a pay tied to a signal that is mostly noise buys little effort while imposing much risk. In the limit, when the outcome is almost all luck or the agent almost wholly unable to bear risk, the best contract is nearly flat: a near-wage that insures the agent and accepts that he will shirk somewhat, because the alternative — loading him with risk to buy an effort the noise would swamp anyway — is worse. This is why the agent is so rarely made the full owner of the outcome, the residual claimant who keeps every dollar his effort earns. That arrangement would give flawless incentives, but only by handing the agent every scrap of risk, which suits him only if he is risk-neutral or the outcome nearly certain. The optimal contract, Holmström showed, also reaches for any further signal that carries news of the agent's effort, however imperfect — judging the manager against his peers to strain out the shocks they share, so that he is rewarded for beating the field rather than for a rising tide that lifts them all.

The theory's reach is best seen in the institution it rescued from contempt. Sharecropping — the landlord taking a fixed share, classically a half, of what the tenant grows — was long dismissed as a feudal inefficiency, a drag on agriculture that gave the tenant only partial claim to his own labor. The principal-agent lens reads it instead as an intelligent compromise. A fixed rent would give the tenant perfect incentive — he keeps every extra bushel — but would saddle him, a poor and risk-averse man, with the whole risk of the harvest. A wage would insure him against the weather but leave him no reason to exert himself in a field no one is watching. The share splits the difference: the tenant bears some risk and keeps some incentive, the landlord bears some risk and supplies some insurance, and the fraction is the dial that tunes the one against the other. The same dial sets the salesman's commission, the executive's bundle of salary and stock, the franchisee's claim on his outlet's profit, and the firm's recurring choice between the employee it pays a steady wage and the contractor it makes a residual claimant. Each is the incentive-insurance trade-off, turned to a different setting — which is why this plate is the bridge from the failure of information to the firm and the contract that the institutions of the table are built from.

The party who holds the good news has every reason to share it — the able worker, the sound borrower, the maker of the reliable machine would all profit if only the other side believed them. The trouble is that belief cannot be bought with words. Anyone can say he is able; the lemon's owner will swear as readily as the peach's that his car is sound. Talk is cheap, in the precise sense that it costs the liar no more than the honest man, and a claim that costs nothing to make carries no information, because everyone makes it. To convey his type, the informed party must do something the other type would not — take an action whose cost is bearable to him but prohibitive to a pretender. Such an action is a signal, and Michael Spence's discovery, in nineteen seventy-three, was how it works and how strange its workings are.

His example was education, and his assumption, made deliberately to shock, was that education might teach nothing whatever — add not one jot to a worker's productivity — and still command a wage. For suppose only that schooling is easier for the able than for the less-able: that the able sit through it with less effort, less expense, less pain. Then the able can afford a level of education the less-able cannot, and an employer who pays more to the educated is not being foolish. He is reading a signal that, by its very cost, none but the able would trouble to send. The educated worker earns his premium not because school made him productive but because reaching that rung of schooling proved he was the sort who could — a proof written in the currency of cost, conveying information without creating any. The diploma sorts the talent; it need not improve it.

The single crossing the figure turns on is the whole condition of a working signal. A signal separates the types only if it costs them differently — if the marginal pain of another year of school is lighter for the able than for the rest — for that is what makes the able willing to climb to a rung the less-able will not. A signal equally costly to all would be sent by all or by none, and would tell the employer nothing; the signal must be a handicap, and a handicap the good type can carry more easily. When the condition holds, the market can settle into a separating equilibrium, the able schooling themselves to the threshold and the less-able forgoing it, each paid the wage their revealed type commands. It may instead settle into a pooling equilibrium, all choosing the same level and all paid the average, the signal conveying nothing — and the theory, awkwardly, permits many of each, with no firm word on which will obtain. The multiplicity is a genuine embarrassment, patched by refinements that ask which beliefs are reasonable, never wholly cured.

Once the shape is in hand it appears everywhere, for the logic is indifferent to its setting. The long warranty is a signal: ruinous for the maker of junk, who will be called to honor it, and cheap for the maker of quality, who will not — so the warranty's length proves the product's soundness, and the warranty of the contract turns out to be a cousin of the diploma. The dividend is a signal: a firm confident of its earnings can afford to commit to paying one, a shaky firm cannot, so the dividend speaks for the balance sheet. The lavish, content-free advertisement signals a firm's own faith that buyers will return, for only a seller expecting repeat business would spend so freely to acquire them. And nature, as so often, was there first: the peacock's preposterous tail, useless and costly and borne only by birds fit enough to afford the burden, is a signal of fitness — Zahavi's handicap principle, which is Spence's theorem in feathers. A signal, wherever it is found, is a cost arranged so that only the genuine article would willingly pay it.

Signaling put the burden of proof on the party who knew — the able worker schooling himself to advertise his ability. Screening turns the arrangement inside out. Here the uninformed party moves first, and instead of waiting for the informed to disclose, he lays a trap of his own devising: a menu of contracts so constructed that each type, reaching for the one that suits it best, betrays which type it is. The customer is not asked to confess; he is offered a choice, and the choice confesses for him. Joseph Stiglitz and Michael Rothschild worked the idea out for insurance, and the device they described is the one every insurer now uses without naming it.

Suppose the insurer cannot tell the high-risk applicant from the low. Rather than post a single policy — which would draw the high-risk and repel the low, in the death spiral of an earlier plate — he posts two. The first offers full coverage at a steep premium; the second, partial coverage with a large deductible, at a low one. Now watch the applicants sort themselves. The high-risk, who fully expect to suffer the loss, will pay dearly to be wholly covered, and they take the full policy. The low-risk, who think the loss unlikely, balk at the steep premium and prefer to carry some risk themselves in exchange for the cheaper, thinner policy — and they take the partial one. No applicant was ever asked his risk; each revealed it in the act of choosing, and the menu accomplished what no inspection could. The insurer has screened his customers through their own preferences.

The diagram shows both the mechanism and its cost. The two types differ in how steeply they value coverage — the high-risk, certain of the loss, will pay much more premium for each added point of protection than the low-risk will — and their indifference curves accordingly cross but once, the same single crossing that made signaling work, here turned to the insurer's advantage. He chooses the two contracts so that the high-risk strictly prefer the full policy and the low-risk the partial one, and the sorting follows. But notice where the partial policy sits. To keep the high-risk from taking it — from posing as low-risk to grab cheap coverage — the insurer must make it thin enough that the high-risk disdain it, which means giving the low-risk less coverage than they would freely buy at a fair price. The low-risk are under-insured, and not for any fault or risk of their own: they are under-insured to keep the high-risk honest. This is the distortion that screening exacts, and it falls on the innocent — the careful, the healthy, the safe, made to carry risk they would rather shed, as the price of being told apart from those they are nothing like.

The same construction, once seen, is everywhere a seller faces buyers he cannot read. The firm that sells a crippled budget version beside its full-priced flagship has built the defects of the cheap one on purpose, to keep the wealthy from buying down — sorting the price-sensitive from the quality-sensitive exactly as the insurer sorts his risks. The airline's tangle of fares, with their Saturday-night stays and advance-purchase rules and non-refundable penalties, is a screen that separates the business traveler, who will not stay the weekend and will pay almost anything, from the holidaymaker, who will do both. The quantity discount, the product line, the menu of deductibles — all are screens, and they share their mathematics with the price discrimination of an earlier volume, for to sort hidden types by a menu of terms and to charge each type what it will bear are the same problem in two costumes. The Screen and the monopolist's graduated prices are one theory, wearing the mask of information here and of power there.

The market for credit is where the two failures of this volume meet and compound, for the lender suffers both at once: he cannot see, before he lends, which borrower carries the sound project and which the rash one — a hidden type — and he cannot govern, after he lends, what the borrower does with the money — a hidden action. It is also the market in which asymmetric information produces its most subversive result, one that strikes at a presumption the classical model holds almost sacred. That presumption is simple. If more people want loans than there are loans to be had, the interest rate should rise until the two are equal; a bank confronted with a queue of eager borrowers should be glad to charge them more, and the rate, like any price, should climb until the market clears. No one, on this view, who is willing to pay the going rate should ever be turned away.

Joseph Stiglitz and Andrew Weiss showed, in nineteen eighty-one, that the presumption fails — and fails because the interest rate is not merely a price but, like the insurer's premium, a thing that sorts and a thing that tempts. Raise it, and two effects work against the lender. The first is selection: the borrowers with safe, modest projects, who repay in nearly every outcome, find that a high rate eats their thin returns, and they withdraw — while the borrowers with wild, high-stakes ventures, who repay only if the gamble comes off and default otherwise, are scarcely deterred, since a higher rate only bites in the lucky state they are betting on. So the pool of borrowers, as the rate climbs, grows steadily more reckless. The second is incentive: a borrower who owes a great deal of interest has every reason to swing for the fences, for the safe project can no longer cover his obligation and only the long shot can — so the high rate pushes even the borrowers who remain toward riskier conduct. Between them, the two effects mean that beyond some point a higher rate brings the lender not more income but more default.

The picture follows at once, and it is strange. The lender's expected return, plotted against the rate he charges, does not climb without end; it rises, crests at a rate one may call r-star, and then falls away. The lender, being no fool, charges r-star — the rate that maximizes what he earns — and he will not be coaxed above it, because above it he earns less. But r-star may sit below the rate that would clear the market, the rate at which the throng of would-be borrowers just matches the supply of loans. At r-star, then, there is excess demand: more sound applicants want to borrow than the lender chooses to fund. The classical script says he should raise the rate and pocket the difference; the analysis says that raising the rate would lower his return, by spoiling his pool and inciting his borrowers. So he does neither. He keeps the rate at r-star and rations the credit — granting loans to some applicants and refusing others, who are no different and who would happily pay more, yet cannot buy a loan at any price. Identical borrowers, and some are financed and some are sent away.

What rationing overturns is among the load-bearing beliefs of the older economics: that markets clear, and that any excess demand is a fleeting thing the price will swiftly competed away. Here the excess demand is permanent, and the price will not budge it, because the price has work to do beyond balancing the books — it selects the borrowers and shapes their risk, and to move it for the sake of clearing would wreck those other functions. The equilibrium is one of quantity rationing, not price adjustment, and the lesson reaches far past the loan office. It tells us why credit can dry up in a downturn without any fall in the rate, why monetary policy works partly by loosening and tightening the spigot of available credit rather than only by moving its price, why the marginal borrower in a slump is rejected rather than merely charged more. The market for the price of money, studied in an earlier volume, turns out not to behave like a market for an ordinary good at all — and the reason is the asymmetry at its heart.

The signal and the screen were the parties' own devices, deployed within a single transaction. But a market does not face the asymmetry one deal at a time and unaided; it grows, over time and without any designer, a whole apparatus for coping with the gap — institutions that supply the trust the bare transaction lacks. This plate gathers them, for they are the answer to the question the volume opened with: how a market survives when one side knows what the other cannot. The short answer is that it builds witnesses — third things, standing outside the deal, that vouch for what the buyer cannot see.

The oldest and the strongest of these is reputation. A seller who must return to the market tomorrow has a reason to deal honestly today that a one-time stranger does not: he holds an asset, built slowly out of satisfied customers and destroyed quickly by a single betrayal, and the fear of losing it disciplines his conduct now. The mechanism is the repeated game of an earlier volume turned to commercial use. In a single encounter the seller gains by cheating and is gone; in an ongoing relationship the value of all the honest dealing still to come can exceed the one-time prize of a swindle, so that honesty enforces itself, held in place not by conscience or by law but by arithmetic. A reputation, on this account, is a bond — the discounted stream of future custom that the seller forfeits the moment he is caught cheating — and it explains why the branded good commands a premium over the anonymous one, why firms pour fortunes into names that confer no function, why the regular customer is treated better than the passing trade. The brand is a hostage the seller gives to his own future good behavior.

Where the parties cannot supply the trust between them, a third party can sell it. The inspector who certifies the building, the auditor who attests the accounts, the agency that rates the bond, the laboratory whose mark declares the wiring safe — each is a specialist in a particular kind of knowing, and each stakes its own reputation on the judgments it renders, so that its certification carries weight precisely because its name would suffer if the certification proved false. Near kin to the certifier is the intermediary, the middleman who makes a business of bridging the gap: the used-car dealer who inspects and reconditions and warrants, turning Akerlof's anonymous lemon lot into a branded forecourt with a name to protect; the broker, the market-maker, the platform. And the platform is the certifier and the intermediary reborn for an age of strangers — the marketplace that lets a buyer in one country trust a seller in another by staking its own reputation on the exchange, and by aggregating into a single star rating the remembered experience of ten thousand past customers, so that the reputation which once required personal acquaintance is now manufactured at scale and handed to a stranger for free.

The remaining devices put money where the word is. The warranty, met before as a signal, is also a bond: in promising to bear the cost of any failure, the seller stakes his own purse on the quality he claims, and the longer and fuller the warranty the larger the stake. The money-back guarantee, the service contract, the maker's pledge to repair — all convert a claim about quality into a financial hostage. On the borrower's side the corresponding device is collateral: the security he forfeits if he defaults, which at once signals his confidence in his own project and binds him to the prudent path, since it is now his house and not merely the bank's money at risk. Bonds, deposits, escrow, the contractor's surety — each is a pledge that aligns the pledger's interest with the truth of his representations. Taken together, reputation and certification and intermediation and bonding compose a remarkable fact, and the hopeful counterweight to the gloom of the lemons: a market crippled by asymmetric information does not passively accept its fate but spontaneously evolves the institutions that hold it together, so that the failure which the blackboard pronounces fatal is, in any market that actually works, already half-cured by contrivances the blackboard never drew.

The witnesses of the last plate were the market's own, grown without a designer and answerable to no one but the custom they hoped to keep. Where they fail, or fall short, or arrive too late, the state may step in — supplying by command the information or the assurance the market could not supply itself. The interventions form a ladder of increasing force, and the lightest rung is also the favorite, for it neither bans a trade nor fixes a price but merely drags a hidden fact into the light: mandated disclosure. The seller of a used car must declare its known defects; the lender must state the true annual rate; the label must list the ingredients, the calories, the side effects, the energy the appliance will burn. The grandest disclosure regime of all governs the issue of securities, where the market in a company's shares is a lemons problem of vast scale — the insiders knowing the firm's true worth, the public guessing — and the law's answer, since the reforms of the nineteen-thirties, has been to compel the firm to open its books: the audited statement, the prospectus, the continuous filing. Disclosure is the economist's preferred remedy because it is the least distorting, leaving the market to work but furnishing it the facts.

Above disclosure stand the heavier rungs. Where quality is too hard for any buyer to judge and the cost of error too grave — in medicine, in aviation, in the safety of food and drugs — the state licenses the provider, its certificate standing as a public warrant of competence where private reputation would form too slowly or too late. And where even a licensed and disclosed market cannot be trusted to protect the buyer, the state regulates the quality directly, decreeing that the drug be proven effective before it is sold, that the building meet the code, that the food be safe — mandating the thing itself, and not merely its disclosure. Each rung trades a measure of the market's freedom for a measure of the buyer's protection, and the choice of rung is the substance of a long argument.

That argument has three corners, and an honest umpire will not award it outright to any. In the first corner stands the interventionist: asymmetric information is a failure, and the state should mandate, license, and regulate to mend it. In the second stands the skeptic, who points to the previous plate — the market already breeds reputation and certification and the warranty, and the state that intervenes often merely duplicates these, or crowds them out, while adding the failures peculiar to regulators, who are captured by the industries they oversee and who know less than the market they correct. In the third, and lately the loudest, stands the technologist, who observes that a new machinery of information is dissolving the very asymmetries that justified the old interventions: the database that yields a car's whole history for a few dollars, the reviews that pool ten thousand buyers' experience, the rating and the comparison site and the device in every pocket, all making hidden facts cheaply visible — so that the market reveals what the law once had to compel. The truthful verdict is that it depends: on whether the private remedies actually work in this market, on whether buyers can read and act on what is disclosed, on the stakes and the cost of being wrong, and on whether the state can really do better than the market it proposes to fix. There is no rule, only the case at hand.

Asymmetric information has, besides its failures and its remedies, a dark and stubborn consequence that no disclosure cures, and it deserves its own paragraph. When a party cannot observe the trait he cares about in an individual — the worker's productivity, the borrower's reliability, the applicant's risk — he is tempted to predict it from a trait he can observe: the group the individual belongs to. The employer who cannot see this applicant's diligence judges him by the average diligence of his kind; the lender prices the loan by the neighborhood; the insurer by the sex or the age. Kenneth Arrow and Edmund Phelps named this statistical discrimination, and its disturbing feature is that it need spring from no hatred at all. It is the cold, profit-seeking use of a group's average as a proxy for the person, rational given the gap in knowledge — and unjust precisely because it is rational, penalizing the able member of a disfavored group not for anything he is or does but for the company the statistics keep him in. Worse, it can fulfill itself: a group that knows it will be judged by its average, whatever its members achieve, has less reason to invest in the achievements that would distinguish them, and the slackened investment confirms the average that justified the discrimination — a trap that locks itself. It is the mirror image of screening turned cruel: where the screen lets the individual reveal himself, statistical discrimination assigns him his group's mean, and the weaker the individual signal, the stronger the pull toward the group proxy. The law's response has been to forbid the use of certain traits — race, sex — as proxies, even where they predict, choosing to bar the rational use of some information rather than tolerate the injustice it works, the same hard choice met in the banning of genetic tests from the insurer's hands. The choice has only grown sharper as the proxy passes from the human prejudice to the trained algorithm, which will find in a postal code or a purchase history a serviceable stand-in for the very traits the law forbids.

Through every plate of this volume a single remedy has hovered, unspoken: if only the ignorant party knew. If the buyer could see the car's quality, the lender the borrower's risk, the employer the worker's ability, the failures would dissolve and the market would clear as the classical model promised. The obvious thought, then, is that the market should supply the missing knowledge itself — that where information is scarce and valuable, someone should produce it and sell it, and a flourishing market for information should furnish the asymmetric markets the very thing they lack. The thought is sound, and it fails, for the most instructive of reasons: information is the strangest commodity in the whole of economics, and the market for it is itself among the markets that fail. The cure, it turns out, is made of the same stuff as the disease.

The strangeness begins with a paradox that Kenneth Arrow set down and that bears his name. To know what a piece of information is worth to you, you must know the information — for only then can you judge how it bears on your decisions. But once you know it, you have no reason to pay, for you already possess it. The seller of information thus confronts a buyer who cannot value the goods until he has been shown them, and who, having been shown them, no longer needs to buy. A fact cannot be inspected before purchase like a melon, then bought or declined; to inspect it is to acquire it. This is why information is so awkward to sell in any ordinary transaction, and why those who deal in it resort to such contrivances — selling reputations rather than facts, bundling the knowledge into a service, guarding it as a secret disclosed only under contract — none of which quite escapes the paradox at the root.

The paradox is only the first of information's peculiarities, and the others compound it. Information is non-rival: my knowing a fact does not consume it or keep you from knowing it too, so that once produced it can serve everyone at no further cost — which means its efficient price is nothing, and a price of nothing pays no one to produce it. It leaks: revealed once, or sold once, it escapes into the world, copied and passed on and reverse-engineered, so that its producer captures only a fraction of the value it ultimately yields. And it is costly to make — the research, the search, the experiment all consume real resources before any fact is in hand. Put these together and the verdict is the figure's: the market under-produces information, because the producer, able to capture only the private sliver of a value that is mostly social, will stop far short of the amount worth making. It is the under-provision of the public good, met before with lighthouses and basic science, here visited upon knowledge itself — and the devices that answer it, the patent that grants a temporary monopoly so the producer can appropriate his return, the public funding of the research no firm would undertake, are the same imperfect contrivances, trading one distortion for another in a good the market cannot handle cleanly.

From this peculiar good follows the deepest result of the whole subject, and the one that fixes the Asymmetry's place among the failures past any doubt. Bruce Greenwald and Joseph Stiglitz proved that an economy with imperfect and asymmetric information — which is to say every real economy — is essentially never efficient in the way the classical theorem promised: not efficient, and not at a single point of failure to be patched, but generically, pervasively, as a matter of course. The famous efficiency of the invisible hand turns out to be the knife-edge special case, requiring information complete and symmetric and markets entire; the ordinary case, the actual case, is inefficiency. This inverts the inherited presumption at its foundation. The textbook had treated efficiency as the rule and market failures as a short list of exceptions — the monopoly, the externality, the public good. Information economics reveals that the list was the world: once knowledge is admitted to be unequal and costly, as it always is, failure is the rule and efficiency the exception, and the question is never whether a market falls short of the frictionless ideal but only by how much and what, if anything, can be done.

One question the subject raises it cannot answer, and the honest theorist says so. If information is costly, then the right amount to gather is finite — perfect knowledge is not the goal, for the last increments cost more than they are worth. But how much, exactly, should a buyer learn before he trades, a firm research before it invests, a society know before it acts? The question has no clean answer, and Arrow's paradox is the reason: to weigh a piece of information against its cost, one must already have it, so the calculation that would tell us the optimal amount of knowledge requires the very knowledge it is meant to ration. The economics of information ends, fittingly, at a fact about itself — that the value of knowing is the one thing knowing cannot settle.

We began with a clause the classical model kept quiet — that the parties to a trade know what they need to, or that the price will tell them — and we have spent eleven plates watching what happens when the clause is false. The secret, in the first plate, split into two: the hidden type, fixed before the deal, and the hidden action, taken after it; and from the two grew the two great failures of the subject. From hidden type came adverse selection — the lemons that drive the good cars from the market, the insurance pool the healthy abandon to the sick — a sickness in who comes forward to trade. From hidden action came moral hazard — the careless insured, the gambling bank, the shirking agent — a sickness in how the protected behave once trading. We watched the failures, and then we watched the market refuse to accept them.

For the responses were the larger half of the story. The party who knew his type tried to prove it, by a signal too costly for a pretender to send — the schooling that separates the able even if it teaches them nothing. The party who did not know tried to extract it, by a screen of self-sorting terms — the menu of policies through which the high-risk and the low confess themselves in the act of choosing. Where price could neither signal nor screen, the market abandoned price and turned to quantity, the lender rationing credit rather than raising a rate that would draw the reckless and tempt the rest. And around these contrivances the market grew a standing apparatus — the reputation it could ruin, the certifier it could shame, the warranty it could call, the collateral it could seize — witnesses that let strangers deal across a gap no single transaction could close. Where even these fell short, the state added its hand, compelling the disclosure the seller would not volunteer, licensing what the buyer could not judge, and confronting the darkest issue of all, the use of a group's average as a proxy for a person it could not read. And beneath everything lay the final paradox: that the knowledge which would cure the asymmetric markets is itself a good the market cannot make or price, so that information is at once the remedy and another instance of the disease.

What the whole accumulates to is a correction of the model at its root, and the correction is simple to state. Markets run on knowledge. They run on the buyer knowing what he buys and the lender whom he lends to and the employer whom he hires, and this knowledge is unequally held, costly to gather, impossible to complete — which the classical picture, for the sake of its theorems, had assumed away. The economics of information put the knowledge back, and found that almost everything turned on it: that a market could vanish for want of it, that the most ordinary institutions — the diploma, the deductible, the brand, the credit limit — were elaborate machinery for coping with its absence, and that the efficiency the older theory had taken for the rule was in truth the rarest of special cases. The Asymmetry is the name for the failure of knowing, and the failure of knowing, it turns out, is not one market's misfortune but the condition of them all.

Seen so, the volume sends its lines out across the whole table. The warranty that signaled quality here is the warranty that bonds a promise in the volume of the contract; the agent whose hidden effort the principal could not see is the agent at the heart of the firm, which is itself a knot of incentive contracts tied against moral hazard. The private types that the signal revealed and the screen sorted are the private types of the games of incomplete information; the screen that sorted insurance risks shares its mathematics with the graduated prices of the monopolist, so that the power to discriminate and the asymmetry of information are one theory in two masks. The credit that rationed rather than cleared is the rate that an earlier volume took for an ordinary price and that this one finds will not behave like one; the pool that adverse selection unravelled is the premium's pool, and the moral hazard that swells a claim is the premium's hazard too. The information that the market under-produces is the public good of the commons, non-rival and unowned and therefore too little made. Even the reputation that disciplines the seller is the norm enforced by repeated dealing, and the incentive that crowds out a craftsman's pride is a warning the norm will repeat. The failure of knowing touches every element it neighbors, because knowing underlies them all.

Within its own family the Asymmetry takes the second place. The Failures are the ways a market can betray us with no villain in sight — no monopolist, no thief, no fool — and each member names a different treachery. The Power was the failure of number, of too few sellers; the Externality, still to come, is the failure of the missing market, of the cost that falls on a bystander no price counts; the Commons, also ahead, is the failure of the resource that all may use and none will tend. The Asymmetry is the failure of knowledge, of the fact held by one side and hidden from the other. With it the group stands two of four complete, and the two that remain — the spillover and the shared resource — wait their volumes. What unites the four is the lesson that a market may be perfectly competitive and perfectly lawful and fail all the same, for reasons written into its structure rather than charged to anyone's account.

The achievement deserves its full due, for it is among the largest in the economics of the last century. Information economics humbled a model that had grown complacent, showing that its central promise held only under conditions the world does not meet; and in humbling it the new theory built rather than merely tore down, explaining as the rational fruit of the informational problem a vast catalogue of institutions that the older economics had to treat as given or as folly — the credential, the warranty, the deductible, the rationed loan, the brand, the rating, the disclosure law. It reshaped the study of labor and of finance, of insurance and of development and of regulation; it gave its founders, Akerlof and Spence and Stiglitz, the field's highest honor, and drew into the work the deepest minds of a generation. Above all it taught economics to look past the price to the knowledge the price conceals — to ask, of any market, not merely what things cost but who knows what, and what the not-knowing does.

And yet the volume has insisted, plate by plate, on the limits of its own instrument, and the insistence is part of the achievement. The elegant models breed equilibria they cannot choose among and lean on a rationality no one quite possesses, so that they identify forces more surely than they predict outcomes. The remedies, private and public alike, are partial and corruptible — reputations that fail the one-shot deal, certifiers bought by those they rate, disclosures no one reads, mandates that redistribute under the name of efficiency, screens that punish the innocent to discipline the guilty. Markets prove more resourceful than the gloomy theorems allow, breeding cures and harnessing a technology of information that dissolves asymmetries the models held fatal — even as that same technology hands the discriminating proxy to an algorithm. The policy question stays genuinely open, the case for the state's hand always shadowed by the state's own ignorance and capture; and the deepest question of all, how much it is worth knowing, the subject confesses it cannot answer, since the value of knowledge is the one thing knowledge will not reveal. The Asymmetry is the failure that recognizes the plainest truth about markets and the one the model most often forgets — that they run on knowledge, unequally held, costly to share, and never complete. It humbles the model, it explains the institutions, and it leaves to judgment, market by market, whether the gap is a crippling failure or a managed friction, and whether its cure is best left to the market's ingenuity or supplied by the hand of the state. The argument remains open.