Signalling in Markets

So far we’ve watched signalling in the wild: peacocks dragging absurd tails, gazelles pronking to advertise “you’ll never catch me,” students burning years on a diploma. Beautiful theory. Now we drag it into the fluorescent-lit world of labour markets, used cars, and warranty cards — where the same math decides who gets hired, which fridge you buy, and why a master’s degree in a subject you’ll never use might still be a rational purchase.

The through-line from Lesson 1 holds: a signal only separates the good types from the bad ones when it is cheaper for the good type. Markets are just where that condition gets a price tag.

Spence’s job-market signalling

Spence’s job-market signalling is the model that started all of this: education can raise your wage even if it builds no productive skill, purely by acting as a costly signal of your underlying ability.

The setup. Employers can’t see how productive a worker is before hiring — ability is hidden. Workers know their own ability. Both types would love to be paid like a high type, so simply saying “I’m high-ability” is cheap talk (Lesson 1) and gets ignored. Education breaks the tie because it is a costly action, and — critically — its cost differs by type: studying is less painful, faster, and less likely to fail for a high-ability person. That’s the Spence separating condition wearing a graduation gown.

Here’s the elegant, slightly cynical claim: the content of the degree can be irrelevant. If high types find each year of school cheaper than low types do, then a diploma of the right length is a signal only high types will bother to earn — so employers can read “graduate” as “high-ability” and pay accordingly.

A worked example

Two worker types. A high type produces value 90k/year; a low type produces 60k/year. Employers pay a worker their expected productivity given the signal. The signal is years of school, and the cost per year differs:

• High type: studying costs $10,000 per year (they breeze through).
• Low type: studying costs $25,000 per year (it’s a slog).

For a signal to separate the types, we need a number of years y such that:

1. The high type is willing to get y years to be seen as high (the 30k wage gap beats their cost): 30,000 ≥ 10,000 × y → y ≤ 3.
2. The low type is NOT willing to mimic (the wage gap doesn’t beat their cost): 30,000 < 25,000 × y → y > 1.2.

So any y between 1.2 and 3 years separates them. Pick y = 2. The high type spends $20,000 in study cost to unlock a $30,000/year wage bump — worth it. The low type would have to spend $50,000 in study cost to grab the same $30,000 bump — not worth it, so they don’t pretend. The market splits cleanly: graduates earn 90k, non-graduates earn 60k, and no one lied.

Notice what did the work: not one hour of that schooling had to teach anything. The cost gap alone carries the information. That insight — asymmetric information can be resolved by costly, differentially-cheap signals — is what earned Michael Spence a share of the 2001 Nobel Prize in Economics.

Pitfall. People hear “education is just a signal” and conclude schooling is useless. That’s not what the model says. It says schooling can be informative even with zero skill-building — the productivity number is doing nothing there by assumption to isolate the mechanism. In reality, the wage premium is a blend of “I learned things” and “finishing proved I could.” The model’s job is to prove the signalling channel exists, not to claim it’s the only one.

When to use it

Reach for Spence whenever you see a costly credential that gates access to a reward, and you suspect the credential’s difficulty — not its curriculum — is what’s being priced: professional licenses, elite internships, “10,000 LeetCode problems,” a hard-to-get security clearance. Ask: is this cheaper for exactly the people we want to select? If yes, it’s a separating signal whether or not anyone learned anything.

The sheepskin effect

If Spence is right that credentials signal, we should see a fingerprint in the data — and we do. The sheepskin effect (named for old diplomas printed on sheepskin) is the empirical finding that the wage jump from finishing a degree is far larger than the wage return to the equivalent amount of learning along the way.

The intuition. Suppose each year of college adds real skill worth, say, a 5% wage bump. Then a four-year degree should be worth roughly four of those bumps — smooth, incremental. Instead, studies repeatedly find a discontinuous leap in the final year: the person who completes year 4 and gets the sheepskin earns dramatically more than the person who completed the same coursework but stopped one credit short. Same learning, wildly different wage. That extra jump can’t be human capital — the two workers learned nearly the same amount. It’s the credential certifying pre-existing ability: only finishers cross the line, and finishing is the signal.

A worked example

Imagine two students. Both complete 3.9 years of identical coursework. Alex crosses the finish line and gets the diploma; Sam has a family emergency and drops out with one seminar left.

That +12% for effectively zero extra learning is the sheepskin. It’s the market paying for the certified fact of completion, not for the last seminar. Employers can’t see Sam’s transcript is 99% identical; they see “no degree,” and the signal is binary.

Misconception (be fair here). The sheepskin effect does not prove education is all signalling. Look at the table again: 30 of Alex’s 42 points are human capital. Both channels are real. The honest reading is: degrees build skill AND certify ability, and the mix varies by field — a welding certificate is mostly skill; a philosophy degree headed into consulting is more signal. Anyone who tells you it’s 100% one or the other is selling something.

When to use it

Use the sheepskin lens whenever a reward attaches to a binary completion rather than continuous progress — bootcamp graduates vs. dropouts, “certified” vs. “took the course,” a finished vs. abandoned side project. A large gap between “almost done” and “done” is a tell that the market is buying the certification, not the incremental skill.

Market signals that resolve the lemons problem

Now leave the labour market and walk onto a used-car lot. In Lesson 5’s territory — the principal–agent problem and Akerlof’s market for lemons — hidden quality can unravel a market entirely: buyers can’t tell peaches from lemons, so they only pay the average price, good sellers refuse to sell at a lemon-adjusted price, quality drains out, prices fall further, and the market collapses toward all-lemons. Nasty little death spiral.

Costly signals are the escape hatch. A high-quality seller needs some action that is cheaper for them than for a low-quality seller — the Spence condition again, now sold at retail. Several classics:

• Warranties / money-back guarantees — a promise to eat the cost of failure. Cheap for a reliable maker (few claims), ruinous for a lemon-maker (many claims). Separating.
• Brand-name capital — a costly-to-build reputation you’d forfeit by cheating. The brand is a hostage: misbehave and the market executes it.
• Skin in the game — a founder keeping their own money at risk, an owner-operator, a chef eating at their own restaurant. Costly to fake because the fraudster doesn’t want their capital exposed to the thing they know is bad.
• Costly advertising — a giant Super Bowl budget that only pays back if customers return. A one-shot scammer can’t recoup it; a repeat-quality firm can. The waste is the point: burning money credibly says “we plan to be here next year.”

A worked warranty example

A washing machine sells for $800. Two makers:

• Reliable Co.: 3% of units fail in the warranty window. A repair costs $400. Expected warranty cost per machine: 0.03 × 400 = \$12.
• Lemon Co.: 40% of units fail. Same $400 repair. Expected warranty cost per machine: 0.40 × 400 = \$160.

Now offer a long, generous warranty. Reliable Co. adds $12 to its cost — trivial, easily absorbed into the price. Lemon Co. would have to eat $160 per machine — that obliterates its margin. So Lemon Co. can’t afford to imitate the signal. The warranty separates: buyers correctly read “long warranty” as “this thing rarely breaks,” because only the reliable maker could survive offering it. The signal works precisely because it is differentially costly — cheap for the honest type, back-breaking for the fraud.

Misconceptions, three for one.

• “A warranty is just customer service / a nice perk.” No — it’s a separating device. Its economic function is to be unaffordable for lemon-makers, which is why it certifies quality. The customer-service warmth is a side effect.
• “Brands are just marketing fluff.” No — a brand is accumulated hostage capital. Its value is that cheating destroys it, which is exactly what makes the brand’s promise credible. Fluff can’t be a hostage.
• “Costly ads just inform people the product exists.” Sometimes — but a conspicuously expensive ad also signals commitment to repeat business, because only a firm expecting return customers could ever earn the ad spend back.

When to use it

Deploy this whenever you’re the buyer facing hidden quality: look for signals that would be too expensive for a bad actor to fake. Long warranty, real skin in the game, a reputation with something to lose, a firm that’s obviously spent money it could only recoup by being good repeatedly. Conversely, if you’re a high-quality seller stuck in a lemons market, your job is to find a signal you can afford and your rivals can’t — and pay it loudly.

Screening vs signalling (recall + sharpen)

You met both concepts before; here’s the crisp distinction, because people mix them up constantly. Both are responses to the same asymmetric-information problem (the informed party knows something the uninformed party doesn’t) — they just differ in who moves.

• Signalling — the informed party spends first, voluntarily, to reveal what they know. The worker gets the degree. The seller offers the warranty. “I’ll prove it to you.”
• Screening — the uninformed party designs a test or a menu of options so that the types sort themselves. The employer sets an entrance exam. The insurer offers a menu of deductibles so cautious drivers self-select the cheap high-deductible plan.

They’re two sides of the same coin — both exploit the fact that a good type and a bad type face different costs, whether that cost is paid by the mover (signalling) or triggered by the menu (screening).

Pitfall. Don’t ask “is this signalling or screening?” as if it’s one or the other — many real institutions are both at once. A university is a screen (admissions filters applicants) wrapped around a signal (the diploma the graduate then emits). The question isn’t the label; it’s who’s paying the differential cost and why.

Credential inflation & signalling arms races

Here’s the ugly footnote to the whole beautiful theory. Signals are positional. Their value comes from relative rank, not absolute achievement — and that makes them prone to an arms race.

Credential inflation is what happens when everyone acquires the signal: a credential that used to separate the top slice from the rest now merely marks the baseline, so people pile on more education to climb back above the crowd. The bar ratchets up. Real resources — years of life, tuition, foregone earnings — get burned, and yet the relative ordering barely moves. Everyone runs faster to stay in the same place. (Yes — that’s the Red Queen dynamic: you sprint just to hold your position.)

This is the deadweight-waste critique of signalling. If the degree is pure signal — no skill added — then when the whole population escalates its schooling to out-signal each other, society pays the full cost of all that education and gets zero extra productive skill in return. It’s a zero-sum tournament with a very real bill.

Crucial nuance (the fair version). “The degree is just a signal, so it’s worthless” is wrong on its own terms. Even a pure signal that adds no skill still creates value by improving matching — it puts the right people in the right jobs, which is genuinely useful even when the signalling itself is wasteful at the margin. The critique isn’t “signalling is worthless.” It’s “signalling is simultaneously valuable (better matching) and wasteful (resources burned to occupy the same rank).” Both are true. Expert-tier means holding both without flinching.

Check yourself

The hinge

Zoom out. Every market institution in this lesson — the diploma, the sheepskin jump, the warranty, the brand, the founder’s skin in the game, the Super Bowl ad, even the treadmill of credential inflation — is the same Lesson-1 idea wearing a business suit: a costly action that is cheaper for the good type separates the good types from the bad ones. Markets just attach a currency to the cost. The theory doesn’t change; the units become dollars.

But notice what’s been lurking. In the arms race we saw signals turn wasteful — resources burned to hold rank. In the warranty we saw signals turn cooperative — a promise that binds two strangers into a deal. That tension between signals that inform vs. signals that merely posture is about to get personal. Because most of your signalling isn’t for markets at all. It’s for other people — your friends, your rivals, your tribe.

Next up: Lesson 4, Strategic & Social Signals — where costly signals stop being about wages and warranties and start being about status, loyalty, virtue, and the elaborate, expensive theatre of proving who you are to the people watching.