Equilibria & Countersignalling

You’ve built the whole machine: private information behind a wall, cheap talk that can’t cross it, and the one condition — the signal must be cheaper for the good type — that lets a costly signal cross it honestly. So far we’ve watched one type decide whether to signal. Now zoom out to the whole room at once. Everyone is choosing simultaneously — good types, bad types, and the receiver forming beliefs about all of them — and the question becomes: where does this whole system come to rest?

That resting point is the equilibrium, and it turns out there are only a few shapes it can settle into. Learn the four shapes and you can walk into any market — jobs, dating, luxury goods, diplomacy — and instantly read whether the signal is doing its job, doing nothing, or doing something delightfully backwards. Because here’s the punchline this lesson is building to: sometimes the smartest move is to not signal at all, and the person who figured that out is the one wearing the hoodie.

Equilibrium, in one line

An equilibrium is a stable state where nobody wants to change their move. More precisely: given what everyone else is doing, and given the beliefs the receiver has formed, no type can make itself better off by switching its signalling choice. Everyone is doing the best they can given what everyone else is doing — so the arrangement holds. Poke it and it springs back.

The analogy is a room full of people picking which side of the road to drive on. If everyone drives on the right, no individual driver wants to switch to the left — they’d be flattened. So “everyone drives right” is an equilibrium: stable, self-reinforcing, no unilateral incentive to deviate. (Notice “everyone drives left” is also an equilibrium — equilibria aren’t unique, and they aren’t necessarily the best outcome, just the stable ones. Hold that thought; it comes back with countersignalling.)

In signalling, the “move” is: what signal each type sends, plus what the receiver believes when they see each signal. An equilibrium is a full package — a strategy for every type and a belief for the receiver — that’s internally consistent: the types are choosing optimally given the beliefs, and the beliefs are correct given how the types are actually choosing. When all three lock together, the market has come to rest.

Pitfall: an equilibrium is not the same as a good outcome. It only means “no one wants to move.” A market can be stuck in a rotten equilibrium — everyone burning money on signals nobody needs — and stay there precisely because no single player can escape alone. Stability is not virtue.

When to use it

Reach for “equilibrium thinking” whenever you catch yourself asking “why doesn’t someone just stop?” — stop inflating credentials, stop the arms race, stop dressing to impress. The answer is almost always: because given everyone else’s choices, stopping alone makes you worse off. The state is stable, so it persists, even when it’s collectively wasteful. That reframing — from “people are foolish” to “the equilibrium is a trap” — is most of the value of this whole course.

Pooling equilibrium

A pooling equilibrium is one where all types send the same signal (or none), so the receiver can’t tell them apart and falls back on the average. The good types and bad types “pool” into one indistinguishable blob. The signal — if there even is one — carries no information, because everyone sends it identically. The receiver, unable to separate, treats each person as the population average: good types get underpaid, bad types get overpaid.

The analogy: imagine a coffee shop where every cup is labelled “premium roast” — the actual premium and the actual dishwater. Since the label costs nothing and everyone slaps it on, “premium roast” tells you nothing, and you’ll only ever pay the price of an average cup. The genuinely great beans are trapped selling at average; the swill is coasting at average. Nobody can escape by shouting louder, because everyone’s already shouting.

Why pooling happens: the cost gap is too small (the bad type can nearly afford to mimic the signal, so it stops separating anyone), or the benefit of separating is too low (even the good type can’t be bothered paying to stand out). Either way, the Spence window we built in lesson 1 is shut.

Worked example. A firm hires from a pool that’s half High workers (worth $120k to the firm) and half Low workers (worth $80k). The signal is a certificate. Suppose the certificate costs a High type $18k of effort and a Low type only $22k — a gap of just $4k. The benefit of being believed “High” instead of “average” is the wage jump.

• If the firm believed the certificate, it would pay a certificate-holder $120k and a non-holder $80k — a $40k swing.
• A Low type reasons: I can grab that extra $40k by paying $22k for the certificate. Net gain: $18k. So the Low type mimics.
• Now everyone holds the certificate. It no longer separates anyone. The firm, seeing that holders are a 50/50 mix, pays the average: $100k to everyone.
• Result: the High type is underpaid by $20k (worth $120k, paid $100k); the Low type is overpaid by $20k. And both burned real money on a certificate that told the firm nothing. Pure waste on top of pooling.

The certificate pooled because $4k of cost gap couldn’t fence out a $40k prize. To separate, the gap has to be big enough that the Low type’s cost exceeds the wage jump — which it plainly doesn’t here.

When to use it

Diagnose pooling whenever a market pays the same despite obvious quality differences underneath — and everyone still bothers with the signal. That’s the tell of a signal that’s collapsed into cheap talk while keeping its costume on. It’s also the state that tempts fraud (lesson 6): once everyone pools, being the one type who could credibly separate is enormously valuable — which is exactly when a hard-to-fake new signal, or a mimic gaming the old one, appears.

Separating equilibrium

A separating equilibrium is the clean case: different types send different signals, so the receiver reads the type straight off the signal. High types signal, low types don’t (or signal less), and because the Spence cost gap is wide enough, no type wants to swap. See the signal, know the type. This is signalling working — information successfully crossing the wall.

The analogy: a chilli menu where the “extra hot” dish is genuinely, ruinously spicy. Ordering it and finishing it is a signal only real chilli-heads will send — the rest of us would suffer too much for the bragging rights. Because the cost of the dish differs by type (agony for the tourist, mild fun for the veteran), the order separates the crowd. The waiter can read your tolerance off your plate.

Worked example. Same firm, same $120k / $80k values, same 50/50 pool. But now widen the gap: the certificate costs a High type $15k and a Low type $50k.

• The firm’s belief: certificate ⇒ High ⇒ pay $120k; no certificate ⇒ Low ⇒ pay $80k.
• High type check: pay $15k, get paid $120k instead of $80k. Net gain $40k − $15k = +$25k. Worth it — the High type signals.
• Low type check: to grab the $40k jump they’d pay $50k. Net = 40 − 50 = −$10k. Not worth it — the Low type stays home.
• So only High types certify, the firm’s belief is confirmed, nobody wants to deviate. Certificate-holders earn $120k, non-holders $80k. The signal is honest. This is a separating equilibrium.

Notice the exact condition: separation survives only because the Low type’s cost ($50k) exceeds the prize ($40k), while the High type’s cost ($15k) falls below it. That’s the Spence single-crossing condition doing its one job — the good type’s cost line sits below the benefit, the bad type’s above it.

When to use it

You’re looking at a separating equilibrium when the signal reliably predicts the underlying quality and the two types visibly pay different prices to send it. That price difference is the load-bearing wall — the moment the low type’s cost falls near the high type’s, separation decays back toward pooling. Watch for that erosion: it’s the mechanism behind credential inflation (lesson 3) and dishonest mimicry (lesson 6).

Semi-separating (partial pooling) equilibrium

Real markets rarely give you a crisp yes/no. The semi-separating (or partial pooling) equilibrium is the messy middle: some but not all of a type separates, so the signal is only probabilistic evidence — it shifts the receiver’s belief without proving the type. Seeing the signal makes you think “probably high,” not “definitely high.”

The analogy: a degree from a top university. It doesn’t prove you’re brilliant — plenty of brilliant people skip it, and a few who muddle through aren’t. But it shifts the odds: a receiver who sees it rationally raises their estimate that you’re able, because the able are over-represented among degree-holders. The signal moves belief part of the way, not all the way. Belief goes from, say, 50% “high” to 80% “high” — better, not certain.

Intuitive example. Suppose earning a hard credential is worth it for the very ablest High types (cost is trivial for them) but a coin-flip for middling High types (cost roughly equals benefit, so some do it, some don’t). And a rare, unusually cheap Low type occasionally sneaks through too. Now:

• Everyone with the credential is a mix — mostly High, but not exclusively, and missing many High types who didn’t bother.
• The receiver updates toward “High” on seeing it, but can’t be sure. The signal is evidence, not proof.
• Someone without the credential isn’t certainly Low, either — they might be a middling High who opted out. So the absence is also just evidence.

This is where the real world lives. Most signals — a degree, a firm handshake, a well-run interview, a healthy credit score — are semi-separating: they tilt the posterior probability without collapsing it to certainty. Treating them as proof is the classic over-read (lesson 6 again).

When to use it

Default to semi-separating as your first guess for any real-world signal, and only upgrade to “fully separating” if the cost gap is genuinely enormous, or downgrade to “pooling” if the signal is universal. Most of the time you’re in the middle: the signal is a clue, worth updating on, not a verdict. Reading it as proof over-trusts it; ignoring it wastes real information.

Countersignalling

Now the delicious twist. Everything so far says: good types signal, bad types don’t. But watch what happens at the very top. Countersignalling is when the highest types deliberately skip the costly signal — because they have nothing left to prove. The genuinely rich dress down. The world-class expert is disarmingly casual. The truly secure never boast. And the reason is exquisite: only the anxious middle, desperate not to be mistaken for the bottom, bothers to signal hard.

The logic runs in three tiers. Picture Low, Medium, and High types, and a receiver who has an independent channel — some other way of telling High from Medium — that the Low type lacks access to.

• Low types can’t afford the signal, so they don’t send it. Easy.
• Medium types can afford it and desperately want to — because without it they look like they might be Low. So they signal loudly. The suit, the watch, the name-drop.
• High types could easily afford the signal, but they don’t — because the independent channel already reveals them as High, and because sending the loud signal would lump them in with the anxious, striving Mediums. Not signalling becomes the signal. It says: “I’m so obviously top that I don’t need this — and only someone unsure of their standing would.”

So you get a backwards, U-shaped pattern: the bottom can’t signal, the middle signals hard, the top signals by refusing to. The costly display becomes a marker of the middle, not the top — which is why, past a certain altitude, sending it can actively mark you as insecure and middling.

Worked example — the hoodie and the suit. A billionaire founder walks into a conference in a $40 grey hoodie. A mid-level manager arrives in a $5,000 bespoke suit and a $30,000 watch. To an outsider with no other information, the suit “wins.” But the room has an independent channel: everyone already knows who the founder is — their face is on magazine covers, their company on every screen. Given that channel, the founder gains nothing from a suit (their status is common knowledge) and loses something by wearing one (they’d blend into the striving crowd). The manager, whose status is not common knowledge, must spend to avoid being mistaken for junior. Net result: the hoodie signals more status than the suit — because in context, only the person with nothing to prove can afford to look like they’re not trying.

The same pattern, everywhere: the tenured genius who lectures in a rumpled t-shirt; the old-money family in worn tweed while the new-money neighbour buys a Lamborghini; the senior engineer whose CV is one line while the junior’s is three pages of buzzwords; the actually-tough country that stays quiet while the insecure regime holds a missile parade (lesson 4’s deterrence signals, now with a countersignalling top tier).

Pitfall: don’t mistake countersignalling for a claim that “signalling is fake” or that “the humble are always the best.” It’s a precise equilibrium move available only to types whose quality is already revealed by an independent channel. Remove the channel and the ordinary separating logic snaps back: without other evidence, the top must signal like everyone else, and dressing down just makes them look Low.

When to use it

Invoke countersignalling when you notice a status ordering that’s inverted — where the loud signallers are visibly not the top, and the top are conspicuously understated. Then check for the independent channel: how does the receiver already know who’s top? If there’s a strong one (fame, reputation, a track record everyone can see), countersignalling is stable. If there isn’t, what looks like countersignalling is usually just someone at the bottom who can’t afford the signal — and you should read them accordingly.

Reading it all together

Four shapes, one table. For each, ask the same two questions — who sends the signal? and what does the receiver actually learn? — and a real example snaps into place.

Read left to right and you can see the arc of the whole course. Pooling is signalling failing — cheap talk in a costume. Separating is signalling working — the cost gap wide enough to fence out fakers. Semi-separating is signalling in the real world — a clue, not a verdict. And countersignalling is signalling transcended — the point where quality is so evident by other means that the costly signal becomes redundant, even self-incriminating. The same variable — the cost gap, plus whether an independent channel exists — drives all four.

The hinge

You now have the full grammar of signalling equilibria: pooling when the gap is too narrow to fence out fakers, separating when it’s wide enough to be honest, semi-separating for the probabilistic middle where most real signals live, and countersignalling for the rarefied top where an independent channel makes the costly signal redundant — even a liability. The same two levers — how wide is the cost gap? and is there another channel already revealing quality? — decide all four.

But notice how much of this lesson kept pointing at the same crack: signals that stop separating. Pooling is that crack. Credential inflation is that crack widening. The badge collapsing to noise is that crack in motion. Every one of these is a case where the beautiful honest machine slips — where the cost gap narrows, fakers mimic, and the signal lies. It’s time to stop admiring the model and start auditing it.

Next up: lesson 6, Where the Model Lies — dishonest mimicry when the cost gap closes, the deadweight waste of signalling arms races, the trap of confusing a signal that reveals quality with one that causes it, and the danger of over-reading a noisy, one-shot signal as if it were proof.