When One Side Knows More

George Akerlof's paper on the used car market was rejected by three journals, one of which explained that if the argument were correct, economics would be different. It was correct. He shared the Nobel for it.

The setup is minimal. Some used cars are good and some are bad: lemons, in American usage. The seller knows which they have. The buyer cannot tell.

A buyer who cannot distinguish will only pay something like the average value. But the owners of genuinely good cars know theirs is worth more than the average, and at the average price they would rather keep it. So good cars are withdrawn. The quality of what remains falls. Buyers, noticing, lower what they will offer. Which withdraws the next tier of cars. And so on.

In the limit the market unravels completely, even though good cars exist, and buyers who would happily pay for one exist, and both sides would gain from the trade. Nothing here requires anybody to behave badly. The seller need not lie; they simply decline to sell. The structure alone does the damage, and asymmetric information turns out to be enough to destroy a market that has every other ingredient for success.

The subsequent literature separated two distinct problems that are often confused, and the distinction is about timing.

Adverse selection is hidden information before the contract. The people most eager to buy health insurance are disproportionately those expecting to need it, and they know this while they are filling in the form. If the insurer prices for the average person, the average person does not turn up. The pool is worse than the population, so premiums rise, which drives out the healthiest remaining buyers, which worsens the pool again. It is Akerlof's spiral with a policy instead of a car.

Moral hazard is hidden action after the contract. Once insured, people take marginally more risk and claim marginally more readily, not through fraud but because the cost is now shared. The insurer cannot observe how carefully you lock the door.

They call for different remedies, which is why the distinction earns its keep. Adverse selection is addressed by screening, waiting periods, pre-existing condition rules, or a mandate that forces the whole population into the pool. Moral hazard is addressed by deductibles, co-payments, no-claims discounts and monitoring: all devices that leave the insured party carrying some of the consequence of their own behaviour.

Then there are the two ways a market can fight back, and again the difference is about who acts.

Michael Spence described signalling: the informed party takes a costly action precisely because it would be too expensive for a low-quality party to imitate. The condition is essential. The cost of the signal must be lower for the high type, or it conveys nothing at all. Anyone can claim their product is reliable. Only a firm that believes its product will last can afford to offer a ten-year unconditional warranty, because for a firm making rubbish the warranty is ruinous.

This is why a warranty's main economic work is done before anything breaks. Its compensation function is real and secondary; its signalling function is why the length of it carries information for a buyer who never intends to claim, and why a warranty stuffed with exclusions signals almost nothing, having made itself cheap for everybody.

Spence's own application was education, and it is the uncomfortable one. A degree can function as a signal of ability and persistence that existed beforehand, quite independently of whether the course taught anything an employer needs. To the extent that is what is happening, the private return to the degree is entirely real (you do earn more) and the social return is much smaller, because the economy has spent three years and a great deal of money sorting people rather than improving them. How much of the education premium is signalling and how much is genuine human capital is a live empirical question with respectable people at both ends.

Joseph Stiglitz described the mirror image: screening, where the uninformed party builds a mechanism that sorts the informed ones. An insurer offering a menu (high deductible with a low premium, or low deductible with a high one) is screening. Low-risk customers prefer the first, high-risk ones the second, and the choice reveals what no questionnaire would extract, because nobody has to answer a question honestly. The information comes out of the structure of the options.

Which brings us to why online marketplaces spend so much on things that are not the product. The lemons mechanism runs entirely on the buyer's inability to distinguish quality before paying, so anything that makes quality visible in advance attacks the cause. Persistent seller reputation does it by converting a one-shot game, where cheating pays, into a repeated one, where it does not. Provided identities are expensive to abandon, which is why cheap new accounts undo the whole apparatus and why platforms guard identity so jealously. Verified purchases stop the reviews being manufactured. Escrow removes the moment where one party must trust the other.

Notice what does not work. A rule requiring sellers to describe items accurately adds no information, because the dishonest seller was already lying. A warning telling buyers to be careful transfers the risk without reducing it. The interventions that work are the ones that change what a buyer can know, not the ones that announce what a seller ought to do.