Adverse selection occurs when one party has private information about their type/quality and this affects contracting. Example: insurance buyers know health risk better than insurers. High-risk types seek coverage eagerly; insurers cannot distinguish types. Standard contracts collapse (high risks drive out low risks). Solutions: screening (insurer offers menu of contracts) or signaling (informed party reveals type).
You have learned from Bayesian games and the principal-agent model that strategic interaction looks very different when one party holds private information. Adverse selection is the specific problem that arises when this information concerns a fixed characteristic — a *type* — that exists before any contract is signed. The classic setting is insurance: each buyer knows their own health risk, but the insurer can only observe the population distribution. A high-risk person knows they are likely to file large claims; a low-risk person knows they are unlikely to. The insurer knows that some buyers are high-risk and some are low-risk, but cannot tell them apart.
The trouble begins when the insurer tries to offer a single contract at a price reflecting the average risk in the population. For high-risk buyers, this is a great deal — they will likely collect more in claims than they pay in premiums. For low-risk buyers, it is a bad deal — they are effectively subsidizing the high-risk group. Rational low-risk buyers drop out of the market, leaving a riskier pool. The insurer now faces a pool that is worse than average and must raise prices, which pushes out more low-risk buyers. This unraveling logic — which Akerlof famously analyzed in the market for used cars ("lemons") — shows that adverse selection can cause markets to collapse entirely or serve only the worst risks.
Two solutions have been studied extensively. *Screening* is initiated by the uninformed party (the insurer): rather than offering one contract, the insurer designs a *menu* of contracts. Full coverage at a high premium is attractive to high-risk types; partial coverage at a low premium is designed to attract low-risk types. Crucially, the contracts are designed so that each type prefers the contract meant for them — this is the incentive-compatibility constraint. The resulting *separating equilibrium* effectively extracts the private information through self-selection, but at a cost: the low-risk contract must be distorted below full coverage to deter high-risk mimics, creating an efficiency loss relative to the full-information benchmark.
*Signaling*, by contrast, is initiated by the informed party. Rather than being screened, the high-quality agent voluntarily takes a costly action that only high types can afford (e.g., getting an expensive education, posting a bond, offering a warranty). If the signal is credible — if low-quality types cannot profitably mimic it — the signal separates types and credibly communicates private information. Signaling games, which you will study next, formalize the conditions under which such equilibria exist.
A key institutional implication is that mandatory participation can restore efficiency. If everyone must buy insurance (as in social insurance systems), the adverse selection death spiral is broken — low-risk types cannot exit, so the pooling premium is stable. This is part of the economic rationale for mandatory health insurance coverage requirements, even from a purely efficiency standpoint, separate from any distributional motivation.