Adverse selection occurs when uninformed buyers cannot distinguish quality, inducing low-quality goods to crowd out high-quality: a market-unraveling problem. Uninformed buyers pay average quality value; high-quality sellers exit (their goods underpriced), lowering average quality. Equilibrium may feature only low-quality (pooling) or separate high/low markets if quality is observable. Costly signaling or screening breaks information asymmetry, but separating costs reduce surplus relative to full information.
Think of Akerlof's used car market. Sellers know whether their car is a "peach" (high quality) or a "lemon" (low quality), but buyers cannot tell the difference before purchase. A rational buyer, unable to distinguish, will only pay a price reflecting the *average* quality of cars on the market. If the average is, say, $10,000, that price is a great deal for lemon sellers (whose cars are worth only $6,000) but a bad deal for peach sellers (whose cars are worth $14,000). So peach sellers exit. Now the market is dominated by lemons, and the average quality — and therefore the price buyers will pay — falls further. More sellers exit. This self-reinforcing spiral is market unraveling: the information asymmetry causes the market to collapse toward low quality or disappear entirely.
The equilibrium that emerges depends on whether any separating mechanism exists. In a pooling equilibrium, all seller types participate at a single price equal to the average quality value, but high-quality sellers are systematically undercompensated. This equilibrium is unstable: if a high-quality seller could credibly communicate their type, they could command a premium. In a separating equilibrium, high- and low-quality goods trade in distinct markets at different prices, and each type is correctly priced. Separation requires that buyers can observe quality — either directly, or through a credible signal.
This is where signaling enters from your prerequisite knowledge of adverse selection. A signal is credible only if it is too costly for low-quality sellers to mimic. A car dealer offering a long warranty credibly signals quality because a lemon dealer would incur enormous repair costs under the same warranty. Education in labor markets works analogously: if acquiring credentials is genuinely harder for less productive workers, credentials credibly separate types. The key condition is the single-crossing property — the cost of the signal must differ enough across types that mimicry is not profitable for the low-quality type.
Screening is the buyer's side of the same problem. Instead of waiting for sellers to signal, an informed party (an insurer, employer, or lender) designs a menu of contracts that induces self-selection. A health insurer might offer a high-deductible plan at low premium and a low-deductible plan at high premium; healthy individuals self-select into the former, revealing their type through their choice. Both signaling and screening achieve separation, but at a cost: resources are spent on signals (education, warranties) or on distorting contracts away from the first-best, so total surplus is lower than under full information even when the market doesn't collapse. The equilibrium comparison is not "adverse selection vs. perfection" but "pooling with unraveling vs. separation with signaling costs vs. full information benchmark" — each with its own surplus level and distribution.