> ## Documentation Index
> Fetch the complete documentation index at: https://agentconcepts.io/llms.txt
> Use this file to discover all available pages before exploring further.
# adverse-selection
> When one side of a transaction privately knows the quality or risk it brings and the other cannot observe it before agreeing, the worse types are the ones most eager to transact at any pooled price, so the participant pool skews toward them and the market degrades or unravels.
computer-science economics sociology
# Adverse selection
## Description
Adverse selection is the market failure that arises when one side of a transaction privately knows the quality or risk it carries and the other side cannot observe it *before* agreeing. Facing one price for a mixed pool, the participants sort by their hidden type: those who know they are better than the pooled terms imply find the deal unattractive and withdraw, while those who know they are worse find it a bargain and stay. Each withdrawal of a good type worsens the average of who remains, which justifies worse terms, which drives out the next tier of good types. In the limit the market unravels — Akerlof showed used-car markets tending toward only lemons, and insurance markets can enter a death spiral where premiums chase an ever-sicker pool.
The load-bearing feature is that the selection happens *at the door* — it is about who chooses to transact, driven by information they hold that the counterparty does not. This is what distinguishes it from its twin, moral hazard: adverse selection is a hidden *type* revealed by who opts in; moral hazard is a hidden *action* taken after the deal is signed. The remedies target the information gap directly: screening (underwriting, inspection, risk-rated pricing) by the uninformed side, or credible signaling by the informed side, or structural fixes (mandates, warranties, certification) that keep good types in the pool.
## Triggers
**User-initiated:** User describes a market, pool, or platform where one side knows something about quality or risk the other cannot see beforehand, and worries the good participants are leaving. Vocabulary cues: "lemons," "the healthy people drop out," "death spiral," "who's actually signing up for this," "hidden information."
**Agent-initiated:** Agent notices a single price or set of terms offered to a pool whose members differ in a quality the offerer cannot observe pre-commitment. Candidate inference: "will this pool sort adversely — do the worse types have more reason to opt in than the better ones, and does that worsen the terms?"
**Situation-shape signals:** Insurance and lending pools; used or unverifiable goods markets; open enrollment without underwriting; any opt-in program whose value depends on who chooses to enroll; free tiers and platforms where the eager sign-ups may be the worst-behaved.
## Exclusions
* **Moral-hazard** — moral-hazard is hidden ACTION after the contract (the insulated party changes behavior once covered). Adverse-selection is hidden INFORMATION before the contract (the worse types select in). The axis is ex-ante hidden type versus ex-post hidden action; confusing them prescribes the wrong remedy (screening versus re-coupling incentives).
* **Selection-bias** — selection-bias is an inference error, an analyst generalizing from a non-representative sample. Adverse-selection is a real economic dynamic: the market's actual participant pool sorts by hidden type, changing who transacts and at what price. One corrupts a conclusion; the other corrupts a market. (The selection-mechanism-correlated-with-the-variable shape echoes across both, but the consequence and the fix differ.)
* **Symmetric or verifiable information** — when both sides can observe quality or risk equally (inspection, certification, verifiable history), the informed-side advantage vanishes, the pool does not skew, and the concept does not fire.
* **Effective screening or signaling in place** — when the uninformed side can screen (risk-rated pricing, underwriting) or the informed side can credibly signal type (costly-signaling), the separating mechanism prevents the unravelling. The information structure still exists, but the collapse dynamic is neutralized.
## Structure
## Relationships
* [moral-hazard](/concepts/moral-hazard) — the contrasting twin. Both run on information asymmetry, but adverse-selection is hidden type before signing and moral hazard is hidden action after. The pair is the backbone of information economics.
* [context-asymmetry](/concepts/context-asymmetry) — the required substrate: adverse-selection is what a pre-contract information asymmetry produces in a market with a pooled price.
* [costly-signaling](/concepts/costly-signaling) — the market's answer: the good type burns cost to separate itself from the pool, converting a pooling collapse into a separating equilibrium.
* [principal-agent](/concepts/principal-agent) — adverse-selection frequently sits at the entry of a delegation relationship, sorting the wrong type in before the principal-agent dynamics even start.
## Examples
Akerlof's used-car market is the founding case. Sellers know whether their car is sound or a "lemon"; buyers cannot tell before purchase. A buyer will therefore pay only the price a car of *average* quality is worth. But at that average price, owners of good cars are unwilling to sell, so they withdraw — which lowers the average quality of the cars still for sale, which lowers the price buyers will pay, which drives out the next-best cars. The process can cascade until only lemons remain, and in the limit the market for good used cars disappears despite willing buyers and willing sellers existing.
**Inference**: the failure is caused entirely by the *pre-purchase* information gap, not by anything either party does afterward. That locates the remedy at the information gap: warranties, third-party certification, dealer reputation, and lemon laws all work by letting a good type credibly reveal itself, holding good cars in the pool. Akerlof's paper (with Spence's and Stiglitz's) founded the economics of information and shared the 2001 Nobel.
The people most eager to buy generous health coverage are disproportionately those who privately know they are high-risk. If an insurer sets one community-rated premium for the pooled average, the coverage is a bargain for the sick and a bad deal for the healthy; the healthy decline, the pool gets sicker, the premium must rise to cover it, more of the remaining healthy people drop out, and the premium chases an ever-worsening pool upward — the insurance "death spiral." Underwriting, risk-rated pricing, waiting periods, and enrollment mandates are all counter-pressures that either observe the hidden type or keep good types in the pool.
In a dating market, a partner's true qualities as a long-term match are largely private and hard to observe up front, while continued availability is public. Participants reason about that gap — "if this person is as good as they present, why are they still in the pool?" — and the most desirable types tend to exit the open market fastest (matched early, or matching through trusted private channels), leaving the observable pool skewed toward those others have already passed over. The dynamic is self-reinforcing: as the visible pool worsens, more good types route around it, thinning it further.
**Inference**: the same separating devices appear as in any adverse-selection market — costly, hard-to-fake signals of commitment or quality, vetting through trusted intermediaries, and reputation — all working to let good types credibly distinguish themselves from a pool their mere presence in would otherwise tar.
A service that gives away resources for free cannot observe a new account's intent at signup. The users to whom unlimited free resources are worth the most are disproportionately the ones planning to abuse them — spammers, scrapers, crypto-miners, fraud rings — so the free-signup pool skews toward exactly the accounts the service least wants, degrading the tier for legitimate users and driving up the cost of offering it.
**Inference**: this is adverse selection with hidden intent as the hidden type, and it is why free tiers accrete friction — email/phone verification, rate limits, CAPTCHAs, or an explicit cost-to-enter like proof-of-work. Each is a screening or signaling device that makes the signup slightly costly in a way abusers (operating at volume) feel far more than legitimate users, separating the pool. The remedy is the same one costly-signaling names: raise the entry cost until only the type you want will pay it.