Adverse selection

Why does a market where buyers cannot judge quality fill up with poor goods?

Here is the reasonable-sounding answer to why used cars are cheap: buyers cannot tell a good one from a bad one, so the price settles at what an average car is worth, and everyone takes their chances. Tidy, and wrong — or at least incomplete in a way that matters. It assumes the cars on offer are still an average sample. But the person choosing whether to sell knows exactly which car they have. What does that do to what is on the lot?

Follow one round. Buyers cannot inspect quality, so the most they will pay is what an average unit is worth — call it the average price. Now put yourself on the other side, holding a genuinely good car. Would you sell it for the price of an average one? No: you would keep it, or sell it privately to someone who knows you. Only owners of cars worth less than the offered price have a reason to trade.

So the pool on sale is no longer average. It is the below-average part of the population. Buyers are not fools — they work this out and revise downward. The price falls to the average of what is actually on offer. And now the same argument runs again on the smaller pool: the best cars remaining are once more worth more than the new price, so they withdraw too.

That is the unravelling Akerlof described in 1970, in the paper that named the market for lemons. Each round of honest repricing removes the best of what is left, which justifies the next repricing. In the extreme, the market shrinks toward the worst goods, or disappears entirely — and note what has been lost: trades that both sides would have wanted, at prices both would have accepted, simply because one side could not verify what it was buying.

Why does that not happen everywhere? Two reasons worth holding onto. Some sellers trade for reasons unrelated to quality — emigrating, inheriting, divorcing — so the pool never becomes purely lemons. And, more importantly, people build institutions specifically to break the information barrier, which is where most of the interesting design lives.

Insurance is the cleanest case, because there the hidden quality is the customer themselves. Who most wants comprehensive cover? The person who expects to claim. Price a policy at the average risk and the low risks decline — they would rather carry the risk themselves — so the people who remain are costlier than average, so the premium must rise, so the next tier of low risks drops out. The word the industry uses for this is a death spiral, and it is the lemons argument with the sign of the hidden quality reversed.

Now the responses, which fall into three families. Screening: the uninformed side offers a menu designed so that different types choose differently — a policy with a high deductible and a low premium, which appeals to someone who does not expect to claim, and thereby separates the risks without ever asking. Signalling: the informed side takes an action that is only worth taking if the quality is real — a warranty, which is cheap to offer on a good car and ruinous on a bad one; a certification; a brand with something to lose; in Spence's original case, education. Pooling by rule: mandate the purchase, or forbid the exit, so the low risks cannot leave — which is what a compulsory motor insurance law or a universal health scheme does. Alongside these sit plain verification: inspection reports, disclosure requirements, audited accounts, ratings.

The signalling family has a subtle requirement that is easy to miss. A signal only works if it is cheaper for the genuinely good type. Anyone can say "this car is excellent"; only the owner of an excellent car can afford a three-year warranty on it. Costless claims separate nothing.

The most valuable thing to keep straight is the one most often blurred: adverse selection is not moral hazard. Adverse selection is hidden information, and it acts before the contract — who chooses to trade with you, given what only they know about themselves. Moral hazard is hidden action, and it acts after — what they do differently once covered, like driving less carefully because the damage is someone else's. Different timing, different hidden thing, different cures. They get conflated partly because a deductible treats both: it screens cautious buyers in advance and keeps the insured exposed to some of the loss afterwards.

Two further honesty notes. The unravelling argument is a logical mechanism, not a prediction that every opaque market collapses — how strong the effect is empirically varies a great deal by market, and in some insurance lines researchers find the opposite pattern, sometimes called advantageous selection, where the cautious people are both the more careful drivers and the keener buyers of cover, because risk aversion pushes in the same direction on both. And "the informed side has the advantage" is not always the seller: in insurance and lending it is the customer who knows more.

• How screening menus separate types formally, and why some separations are impossible at any price.
• Whether online ratings and platform escrow genuinely solve the lemons problem or relocate it.

Every term the collection defines is gathered in the glossary.