The Cures — Signalling, Screening & Institutions

You’ve now watched adverse selection eat a market from the top down, and you’ve toured the arenas where it bites — insurance death spirals, credit, labour, marketplaces, IPOs. The natural question, the one every trader and regulator eventually asks, is: so how do we stop the bleeding? This lesson is the answer, and it comes with a single unifying idea that makes all of it click.

Here it is. Every cure works the same way: it shrinks the information gap. That’s the whole game. The disease was never dishonest sellers or a bad crop of cars — you proved that in the very first lesson when dragging the “hidden information” slider to zero cured the market by itself. So every remedy is just a different route to the same destination — make quality more observable, and the good types stop fleeing. There are exactly three families of route, sorted by who takes the first step:

1. Signalling — the informed side moves. The good type does something costly and hard to fake to prove it’s good.
2. Screening — the uninformed side moves. It designs a test or menu that makes the informed side reveal its type.
3. Institutions & mandates — a third party moves. Inspectors, ratings, disclosure rules and mandates make quality observable or force the good types to stay.

Keep the sorting question — who moves? — in your head the whole way through. It’s the skeleton key.

Signalling — the informed side proves it

The analogy. You’re the owner of a genuinely great used car, and you’re furious, because the buyer is offering you a lemon-adjusted average price and there’s nothing you can say to change their mind — every seller claims their car is great, so talk is worthless. So instead of talking, you do something a lemon owner would never dare do: you staple a two-year, bumper-to-bumper warranty to the windscreen. That warranty is a message the buyer can trust, precisely because it would ruin you if your car were secretly a lemon.

The definition. A signal is a costly, observable action taken by the informed (good) type that is not worth faking by the bad type. It works only when it satisfies the single-crossing condition — the intuitive heart of all signalling theory:

The signal must be cheaper for the good type than for the bad type.

Say that back to yourself, because everything hangs on it. If sending the signal cost the same for a peach and a lemon, the lemon would happily send it too, and the buyer would learn nothing. The signal separates the types only when the cost curves “cross” — when the action is affordable for the good type and punishingly expensive (or ruinous) for the bad one. A warranty is cheap for a peach owner (the car won’t break, so they’ll rarely pay out) and terrifyingly expensive for a lemon owner (the car will break, and the warranty makes them eat the repair). That cost gap is what makes “I offer a warranty” mean “this is a good car.” This is the machinery you met in signalling and costly signals — here it’s the cure rather than the phenomenon.

A worked example — the warranty a lemon would never write

Let’s put numbers on it. Two used cars look identical:

Now the peach owner offers a free 2-year warranty. To them, that promise costs an expected $400 — a small price to unlock the peach’s true value and escape the pooled lemon price. To a lemon owner, mimicking that same warranty costs an expected $3,200. If the price premium for looking like a peach is, say, $2,000, then:

• Peach owner: pays $400 in expected warranty cost to gain $2,000. Worth it.
• Lemon owner: would pay $3,200 in expected warranty cost to gain $2,000. Not worth it — they’d lose $1,200 pretending.

So only peach owners offer the warranty, the buyer reads “warranty = peach,” and the good cars stay on the lot. The single-crossing condition holds because the same warranty is cheap for the good type and ruinous for the bad one. That’s a separating equilibrium built by the seller, out of their own pocket.

Warranties are only the cleanest case. The same logic powers a whole family:

• Brands & reputation staking — a firm that has spent decades and millions building a name has something expensive to lose if it sells you junk. The sunk brand value is the hostage.
• Money-back guarantees — cheap to offer if your product is good (few returns), costly if it’s bad (everyone returns it).
• Credentials & education — Spence’s job-market signalling: a degree can raise your wage even if it taught you nothing, purely because it’s cheaper to earn for the high-ability type. (This is the flagship case in the signalling course — go there for the full model.)
• Certifications — a “board-certified” surgeon, an audited financial statement, a professional license: costly hoops the low type won’t clear.

Pitfall — signalling can be pure deadweight

Here’s the uncomfortable truth about signalling: it proves quality without creating it. The warranty doesn’t make the car better; the degree (in the pure Spence model) doesn’t make the worker more productive. The signal transfers information, but the resources burned sending it are often gone forever — economists call this deadweight loss.

Worse, signals invite arms races. If a bachelor’s degree once separated the able, and then everyone gets one, the signal stops separating — so the able escalate to a master’s, then a PhD, then an unpaid prestige internship, each one costlier and no more informative than the last. This is credential inflation: society spends more and more on the signal while learning exactly the same amount about who’s good. The signal did its job — it sorted the types — but it did it by setting money on fire.

When to use it

Signalling is the right lens when the informed side has both the ability and the incentive to prove itself — when a good type can take a hard-to-fake action and reap the reward. If you’re the seller of something genuinely good in a market that’s pricing you like a lemon, your move is to find the warranty — the costly action a fraud wouldn’t dare copy. And if you’re the buyer, ask which sellers are voluntarily doing expensive things to prove themselves, and why the bad ones aren’t.

Screening — the uninformed side extracts the truth

The analogy. Flip the table. Now you’re the uninformed one — an insurer who can’t see whether an applicant is a careful low-risk driver or a reckless high-risk one. You can’t wait for them to signal; instead you design a clever menu: two policies, priced so that each type, acting purely in its own self-interest, walks straight to the one that reveals what it is. You didn’t learn their type by asking. You built a trap that makes them sort themselves.

The definition. Screening is when the uninformed side offers a menu of contracts, tests, or terms designed so that the informed side’s own choice reveals its private type. Same single-crossing logic as signalling — a difference in costs across types — but now the uninformed party constructs it. The classic instrument is the insurance deductible: the amount of a claim you pay out of your own pocket before coverage kicks in.

A worked example — the deductible menu

An insurer faces two hidden types. Low-risk drivers crash rarely; high-risk drivers crash often. The insurer offers two policies and lets applicants pick:

Now watch the self-selection. A high-risk driver expects to crash often, so a $1,500 deductible would hurt repeatedly — they gladly pay the high premium for full cover (Policy A). A low-risk driver rarely crashes, so that $1,500 deductible is a threat that almost never materialises — they happily take the cheap premium (Policy B). Each type sorts itself by choosing what’s cheapest for its own risk level, and the insurer, without ever seeing a driving record, ends up with the low-risks in B and the high-risks in A — priced correctly. That’s a separating equilibrium the insurer engineered.

This is the accessible core of the Rothschild–Stiglitz model: in a competitive insurance market with hidden risk types, firms can’t offer one pooled policy (the high-risks would swamp it and it’d lose money), so they offer a menu where the low-risk contract carries a deductible painful enough that high-risks won’t touch it. The deductible is the screening device; the self-selection is the cure.

Screening shows up far beyond insurance:

• Tiered contracts & versioning — software “Basic vs Pro,” coach vs business class: menus that sort willingness-to-pay.
• Probation periods — a job offer that pays fully only after 6 months; a confident good hire accepts, a bluffer balks.
• Credit checks & collateral requirements — asking a borrower to post collateral screens out those who privately doubt they’ll repay.
• Interviews, tests, and take-home assignments — the uninformed employer builds a task the strong candidate finds easy and the weak one finds prohibitive.

Pitfall — screening can immiserate the good types

Screening has a dark cost that’s easy to miss: to prove they’re good, the good types often have to distort their own choices and end up worse off than they’d be under perfect information. In the Rothschild–Stiglitz world, the low-risk driver would love full coverage at a fair low price — but no such contract survives, because high-risks would grab it too. So the low-risk driver is stuck accepting a big deductible they don’t want, bearing risk they’d happily pay to shed, purely to separate themselves from the high-risks. The information problem taxes the innocent. The good type isn’t driven out of the market (that was the raw lemons problem) — instead they’re kept in, but on distorted, worse-than-first-best terms.

When to use it

Screening is your move when you’re the uninformed side and the informed side won’t volunteer the truth — when you can’t wait for a signal but you can design the terms of the deal. If you’re hiring, lending, insuring, or pricing, ask: what menu or test would the good type accept and the bad type refuse? Build that, and let self-interest do the sorting.

Third-party institutions & mandates — someone outside closes the gap

The analogy. Sometimes neither side can fix it alone. The good seller’s warranty is too expensive, the buyer’s screening menu too crude — so a third party steps in and simply makes quality observable to everyone. Before you buy that used car, you pull its vehicle-history report: every accident, every owner, the odometer truth. Now the hidden information isn’t hidden anymore, and the peach and the lemon wear their true colours. Nobody had to signal; nobody had to screen; an outside institution dissolved the asymmetry.

The definition. Institutional cures are third-party mechanisms — inspections, ratings, disclosure rules, licensing, and mandates — that either make quality observable or force the good types to stay in the pool by law. They’re often the cheapest cure per unit of information revealed, because one inspection or rating serves the whole market at once, rather than every good type paying to signal individually.

The family:

• Inspections & history reports — CARFAX and vehicle-history reports, home inspections, pre-purchase mechanic checks. They make the hidden quality directly verifiable.
• Ratings & reputation systems — eBay/Uber/Airbnb star ratings, Yelp reviews, credit scores. A reputation system aggregates many past interactions into a public score, so a seller’s history becomes a hostage: cheat once and the rating punishes you across all future trades. This turns a one-shot lemons market into a repeated game where honesty pays.
• Licensing — doctors, electricians, lawyers must clear a certified bar to trade at all, putting a floor under quality.
• Mandatory disclosure — securities law forcing companies to publish audited financials before selling stock; nutrition labels; house-sale disclosure forms. The law compels the informed side to reveal what it knows.
• Insurance mandates — the direct cure for the death spiral. Pair community rating (everyone charged the same premium regardless of risk) with a mandate that forces the low-risks to buy in anyway. Without the mandate, community rating would collapse — the healthy would flee the pooled price, exactly the unravelling you know. With the mandate, the good risks are held in the pool by law, so the pool doesn’t rot.
• Escrow & money-back platforms — a marketplace that holds the buyer’s payment until the good arrives removes the seller’s ability to exploit hidden quality.

Watch a cure freeze the unravelling

Time to see two of these cures actually work. Below is the same lemons market you’ve been driving — good cars up top, lemons down below, pooled pricing driving the peaches off the lot round after round. This time, use the remedy switch underneath the simulator.

Start on No cure and scrub the rounds: the familiar collapse, good cars going dim tier by tier. Now click Warranty signal and watch the good-type sellers certify — they post a costly warranty a lemon would never buy, so the peaches stop leaving and sell for what they’re truly worth. That’s signalling, from the informed side. Then click Mandatory inspection: now every car’s quality is observable, the whole information gap slams shut, and the entire market survives — not just the peaches, everyone trades at their true value. That’s the institutional cure, imposed from outside. Toggle between the three and watch the good cars stop fleeing.

Pitfall — institutions can misfire too

None of these is free or foolproof. Inspections and ratings can be gamed (fake reviews, odometer rollbacks, “teaching to the certification”). Mandatory disclosure can bury the real signal under a flood of boilerplate nobody reads — a 200-page prospectus can hide more than it reveals. Licensing can curdle into rent-seeking, where incumbents raise the bar not to protect quality but to block competitors. And mandates carry a political and liberty cost: forcing people to buy something is coercive, and setting the community-rated price wrong can over- or under-insure the whole population. An institution is a cure, not a miracle — it too can be captured, gamed, or misdesigned.

When to use it

Reach for an institutional cure when the private cures are too weak or too wasteful — when good types can’t afford to signal, buyers can’t screen finely enough, or the signalling arms race has become pure deadweight. A shared inspection, rating, or disclosure rule reveals quality once for the whole market, and a mandate can hold a pool together when voluntary sorting would collapse it. This is the level a regulator or platform designer operates on.

Sort the real cures

Now prove you can tell the three families apart by who moves first. Sort each real-world cure into signalling (informed side acts), screening (uninformed side builds a menu/test), or an institution/mandate (a third party or law).

Signalling vs screening — who moves?

Signalling and screening are so easily confused that it’s worth a side-by-side. They use the same single-crossing physics; they differ only in choreography.

And the third family stands apart: institutions & mandates move from outside the deal entirely, making quality observable to everyone at once or compelling the good types to stay — often the cheapest route when the two private cures are too weak or too wasteful.

Where this leaves you

Step back and the map is clean. A lemons market is a market with a hole in it where information should be, and there are exactly three ways to fill the hole: the informed side can signal through it, the uninformed side can screen across it, or a third party can institutionalise it shut. All three obey the same law — they raise the cost of faking quality until only the truth is worth telling. You watched the simulator go from a collapsing lemons pile to a fully surviving market the instant a warranty or an inspection closed the gap.

You came into this lesson from where adverse selection strikes — the arenas where hidden type poisons trade. Next, in Transfer & where the model lies, you’ll turn all of this into a daily thinking tool and, just as importantly, learn the sharp line between adverse selection (hidden type, cured here) and its sibling moral hazard (hidden action), plus the places where markets don’t actually unravel. A cure aimed at the wrong disease is worse than none.