---
name: Adverse Selection
status: active
territory: strategic-interaction
host_mode: mechanism-design
also_loadable_in: []
msi_wired: true
msi_family: incentives
sources:
  - title: "Akerlof, George A. (1970), The Market for 'Lemons': Quality Uncertainty and the Market Mechanism, Quarterly Journal of Economics 84(3):488-500"
    url: https://doi.org/10.2307/1879431
  - title: Spence, Michael (1973), Job Market Signaling, Quarterly Journal of Economics 87(3):355-374
    url: https://doi.org/10.2307/1882010
  - title: Stiglitz & Weiss (1981), Credit Rationing in Markets with Imperfect Information, American Economic Review 71(3):393-410
    url: https://ideas.repec.org/a/aea/aecrev/v71y1981i3p393-410.html
---

# Adverse Selection

## Why it matters

When one side of a deal knows more than the other and everyone is offered the same terms, the good options quietly walk away — until the pool is mostly the bad ones nobody wanted.

For example: a health plan sets a single premium for everyone. The healthy do the math, decide it's a bad deal, and skip it; the people who know they're sick see a bargain and sign up. So the pool gets sicker, the premium has to rise, which chases out the next-healthiest tier — and round it goes. Nobody lied. The price itself sorted the wrong people in.

- **What it reveals.** Which way a pool will *drift* when one side can't tell quality apart and prices for the average — and that the drift is toward the worst participants, not a random mix.
- **How it changes the read.** You stop asking *"why is the quality here so bad?"* and start asking *"who finds these exact terms attractive, and who walks?"* The terms are the filter, and they're selecting against you.
- **When to foreground it.** Any pool people opt into on uniform terms when one side holds private information — insurance, lending, hiring funnels, marketplaces, warranties — and average quality is sliding.
- **What you'd miss without it.** That the fix isn't trying harder to attract good participants at the same price — it's changing the structure so the hidden information comes out: screening from your side, or a costly signal from theirs.
- **Where it misleads.** No hidden information, no adverse selection — if both sides know roughly the same thing, the pattern doesn't apply. And it's not the same as people behaving worse *after* they sign up; that's its sibling, moral hazard.

## Realtime examples

See real, dated analyses where this pattern shaped the read on the news → **[Adverse selection on Main Street Independent](https://mainstreetindependent.com/analyses/lens/incentives/adverse-selection)**

## How to invoke it in Ora

You're looking at a pool — an insurance plan, a lending book, a hiring funnel, a warranty — whose average quality keeps sliding, and you suspect the terms are drawing in exactly the people you didn't want. You want to know why, and what would stop it.

Describe who knows what the other side can't, and the terms everyone's offered, then ask:

> "Adverse selection: our pay-monthly phone warranty keeps attracting customers whose phones are already failing, and we cannot tell in advance. Why is the pool degrading, and how do we screen the risk types?"

Ora names the hidden information, traces who self-selects in and who exits, classifies it as adverse selection rather than its post-contract sibling, and works out the screening or signaling that would separate the types.

One thing to know: the words *adverse selection* are what route you here. A plain version — "our warranty keeps attracting failing phones, what's going on?" — gets a clarifying question back instead, because the bare situation could be a dozen different problems; naming the mechanism is what tells Ora which one you mean. *Adverse selection*, *moral hazard*, *hidden information*, *screening*, or *lemons* are the words that point it the right way.

Describe what one side privately knows that the other can't verify *before* the deal — that pre-contract information gap is the whole mechanism. If the knowledge gap only shows up in how people behave *after* signing, you're looking at moral hazard instead, and Ora will say so.

One thing Ora won't do: assume the market is doomed to collapse. It shows you the degenerative pull and the structural fixes that can halt it — screening, signaling, tiered terms — rather than declaring the pool unsalvageable.

## How it works

Imagine you're shopping for a used car, and so is everyone else in town. Some of the cars are solid; some are lemons that will fall apart in a month. The owners know which is which. You don't — from the curb, a cherry and a lemon look identical.

So what will you pay? Not top dollar — you'd be a sucker if the car turned out to be a lemon. Not bottom dollar either. You'll offer something in the middle: a price for a car of *average* quality, since average is your best guess for what you're getting.

Now watch what that middling price does to the people on the other side. The owner of a genuinely good car looks at your offer and says no thanks — the car is worth more than that, and they'd rather keep it. The owner of a lemon looks at the same offer and bites your hand off — it's far more than the wreck is worth. So the good cars come off the market, and the lemons stay on it. The average quality of what's for sale just dropped.

And here's the trap: you're not stupid, so you notice. If the good cars are gone, the average is now lower, so your fair price drops too. Which chases out the *next* tier of decent cars. Which lowers the average again. The market grinds downward, good cars vanishing at every step, until in the limit there's nothing left to buy but lemons — a market that should exist, full of cars people would happily trade, collapsing under its own information gap.

That is adverse selection, the mechanism the economist George Akerlof laid out in 1970 in a paper about exactly this — the market for "lemons." The engine is simple and merciless: when one side can't tell good from bad and so prices for the average, the good side gets a bad deal and leaves, the bad side gets a good deal and stays, and the average everyone is pricing against keeps falling. It is not a story about dishonest people. Every player is being perfectly reasonable. The structure does the damage.

And that points straight at the only real fixes, because you can't repair this by hoping for better participants at the same price. You have to break the information gap itself — either the uninformed side builds a way to *screen* (inspections, credit checks, medical exams, a probation period) so good and bad can be told apart and priced apart, or the informed side finds a *signal* it can send that a faker couldn't afford (a warranty, a certification, a guarantee). Change what the terms can see. Until they can tell quality apart, the terms will keep selecting for the worst of it.

## Framework & implementation

*This section uses Ora's own terms for the parts of an analysis, so that if you open the actual mode and lens files they line up. Each is glossed in plain language on first use.*

### Pipeline execution

Adverse selection is one of the mental models in Mechanism and Incentive Analysis's **`ANALYTICAL PERSPECTIVES`** block, listed under "always loaded" — so it is active on every run of that mode, whether or not the prompt names it. (Mechanism and Incentive Analysis is the information-and-incentive sibling of Strategic Interaction: where that mode reads a game of observable moves, this one handles the case where what each party privately *knows* or privately *does* is the crux.) It runs at **Gear 4**, Ora's most thorough setting: a **Depth analyst** and a **Breadth analyst** read the situation independently, each critiques the other's reading, both revise under that critique, and a consolidator merges what survives. The lens threads through those stages like this.

**Detection.** The lens engages on the cases in its **Detection Signals** — a pool whose average quality is declining with no obvious external cause; participants with the most to gain self-selecting in disproportionately; an insurance pool, lending book, or hiring funnel attracting the people the designer did *not* target; flat terms applied across heterogeneous risk. The precondition is the mode's first **Critical Question** (CQ1, *asymmetry named*): one side holds material private information the other cannot verify *before* the deal.

**The Depth and Breadth analysts.** Two models read the situation in parallel. The **Depth analyst** commits to one reading and defends it — it names who holds the private information, then runs the lens's **Application Steps**: confirm the terms are uniform across heterogeneous quality, predict which subset finds those terms attractive, and compare that prediction against how the pool has actually moved over time. Its load-bearing job is the mode's CQ2 (*selection vs. hazard*): classify this as **adverse selection** — hidden *type*, known *before* contracting — and not its post-contract sibling, moral hazard. The **Breadth analyst** works the same situation at the same time, scanning the mode's other information-and-incentive failure modes — moral hazard, the winner's curse, signaling, screening, the principal-agent split — and ruling each in or out rather than assuming this one. Neither analyst sees the other's work.

**Cross-adversarial evaluation.** Each analyst's reading is handed to the *other* to critique against the mode's criteria. The lens's signature failures are caught here, keyed to its **Critical Questions**: invoking the lens where the parties actually have similar information (*asymmetry overclaim* — the mode's *assume-away-asymmetry* failure in reverse), conflating pre-contract sorting with post-contract behavior change (*selection-vs-hazard confusion* — the mode's CQ2, the one distinction it will not let blur), and fingering the wrong party as the informed one (*reverse-direction error*, caught when the predicted self-selection doesn't match the observed drift).

**Revision and claim-check.** The reviser addresses the fixes. Where the reading rests on a factual claim — the actual composition of the pool, real loss ratios, who is genuinely exiting — that claim is marked a **flagged claim** and sent to a web-search tool; it has to resolve against outside sources before the revised draft moves forward.

**Consolidation and output.** The consolidator merges the two revised readings into one information-and-incentive corpus, and the formatter places it into the mode's set sections. Who holds the hidden information lands in **Parties and the asymmetry**. The classification — adverse selection, hidden type, pre-contract — lands in **Selection vs. hazard**. The lemons spiral (good types exit, average falls, terms re-adjust down) lands in **The distortion**. Screening and signaling are grounded in **Named mechanisms in play** (the mode's CQ4 forbids name-dropping — each must be shown operating). When the prompt asks for a fix, the screening/tiering/signaling design lands in **Mechanism (for a design)**, which must carry *both* the **participation constraint** (will good types still join once screened?) and the **incentive-compatibility constraint** (does the mechanism make honest self-sorting each party's best move?) — a design that satisfies only one is reshaped. The conclusion lands in **Read**, with the mode's CQ5 posture marker stating plainly whether it is *explaining* a failure or *proposing* a fix.

**What the analysis will not assert.** It reports the degenerative pull and what would halt it. It does not assume the market collapses — screening, signaling, and segmentation routinely stabilize a pool short of the lemons limit, and the lens's *corrective-overshoot* caution warns the other way too: screening so aggressive that verification costs more than the cleaner pool is worth shrinks the market instead of saving it. And it holds adverse selection strictly apart from moral hazard, because the two demand different fixes.

### Origin and evidence

The mechanism is George Akerlof's, from his 1970 *Quarterly Journal of Economics* paper "The Market for 'Lemons,'" which used the used-car market to show that quality uncertainty plus uniform pricing can unravel a market entirely — work that helped found the economics of information and earned Akerlof a share of the 2001 Nobel Memorial Prize. The two standard correctives arrived alongside it. Michael Spence's 1973 *signaling* model showed how the *informed* side can break the spiral by taking a costly action a low type couldn't afford to imitate — the move detailed in this lens's sibling, Signaling. Joseph Stiglitz and Andrew Weiss (1981) worked the *screening* side, showing in credit markets how the *uninformed* side can design terms (and, in their case, ration quantity rather than only raise price) to separate the types it cannot directly observe. Akerlof, Spence, and Stiglitz shared the 2001 Nobel for the economics of markets with asymmetric information — adverse selection is one of its founding results.

### Applications and common uses

Adverse selection is a workhorse diagnosis wherever a pool is assembled on uniform terms under hidden information — used both to explain why a pool is degrading and to design the screen or signal that stops it.

- **Insurance.** The textbook case: flat premiums draw the high-risk and repel the low-risk, pushing the pool toward a death spiral. Risk-rating, medical underwriting, waiting periods, and mandates are all answers to it — each a way to either screen risk or force participation so the pool can't unravel.
- **Lending and credit.** A single interest rate attracts the borrowers most likely to default and deters the safe ones; lenders screen with credit scores, collateral, and — as Stiglitz and Weiss showed — by rationing credit rather than simply raising the rate, which would only worsen the mix.
- **Labor markets and hiring.** A funnel with undifferentiated terms can over-attract exactly the candidates who can't do better elsewhere; the response is screening (work samples, probation, references) and reading the costly signals candidates send (credentials, portfolios).
- **Used goods and marketplaces.** Akerlof's own domain — platforms fight the lemons dynamic with warranties, certified-pre-owned programs, ratings, and return policies, each a structural device that lets quality reveal itself.
- **Mechanism and contract design.** Run in reverse, the lens is a design brief: build the menu, the deductible schedule, or the screening test whose terms make good types willing to identify themselves — separating the pool by construction instead of pricing for a sinking average.

In every case the diagnosis is the same: the terms are the filter, the filter is selecting against you, and the fix is to make the hidden information visible — not to wish for better participants at a price that drives them away.

### Failure modes and when not to use it

The lens's characteristic ways of going wrong are catalogued in its **Common Failure Modes**:

- **Asymmetry overclaim.** Invoking the lens when both sides actually know roughly the same thing. The tell is that each side can produce equivalent diagnostic data on the deal. With no real information gap there is nothing to select on, and a different lens (incentive misalignment, moral hazard) fits better.
- **Selection-vs-hazard confusion.** Treating a *post-contract* change in behavior as adverse selection. The tell is that the problem behavior began *after* the deal was signed — that's moral hazard, a different mechanism with different fixes. Adverse selection is pre-contract sorting of *types*; keep the two apart.
- **Corrective overshoot.** Screening so aggressively that the cost of verification exceeds the value of the cleaner pool. The tell is a market that *shrinks* rather than improves once screening is added. The fix is to size the screen to the asymmetry's actual cost, not to maximize purity.
- **Reverse-direction error.** Identifying the wrong party as the informed one, so the predicted self-selection runs backwards. The tell is that the pool isn't drifting the way the lens predicted; the correction is to re-check who actually holds the private information.

**When not to reach for it.** When information is roughly symmetric, the lens predicts a degradation that won't happen. When the interaction is one-shot with no pool and no pricing feedback, there is no spiral to trace. And when the behavior in question changed *after* contracting rather than sorting the pool *before* it, the mechanism is moral hazard — reach for that sibling instead, because screening the entry pool won't fix a problem that lives in post-signing behavior.

## Related

- **Mechanism and Incentive Analysis** — the analysis that hosts this lens; reads situations where hidden information and incentive structure drive the outcome, and designs the rules that fix them.
- **Moral Hazard** — the post-contract sibling: behavior turns riskier *after* the deal, once someone else bears the downside. Adverse selection sorts the pool before signing; moral hazard changes behavior after.
- **Signaling** — the corrective from the informed side: a costly action only a genuine type can afford, which lets quality reveal itself.
- **Principal-Agent Problem** — the broader frame of one party acting for another under hidden information and hidden action, of which adverse selection is the pre-contract face.

## Sources

- [Akerlof, George A. (1970), The Market for 'Lemons': Quality Uncertainty and the Market Mechanism, Quarterly Journal of Economics 84(3):488-500](https://doi.org/10.2307/1879431)
- [Spence, Michael (1973), Job Market Signaling, Quarterly Journal of Economics 87(3):355-374](https://doi.org/10.2307/1882010)
- [Stiglitz & Weiss (1981), Credit Rationing in Markets with Imperfect Information, American Economic Review 71(3):393-410](https://ideas.repec.org/a/aea/aecrev/v71y1981i3p393-410.html)