Signalling & Costly Signals

A peacock drags around a tail so big it can barely fly and might as well be wearing a “eat me” sign for every fox in the county. A student spends four years and a fortune on a degree that teaches them almost nothing they’ll use at work. A carmaker promises to fix your engine free for ten years. A country parades missiles it hopes never to fire. These look like four unrelated kinds of foolishness — vanity, waste, generosity, sabre-rattling. They are, in fact, the same move played on four boards. Each is a case of somebody who knows something others can’t check — “I’m a strong mate,” “I’m a capable worker,” “my cars don’t break,” “don’t attack us” — and who proves it not by saying it but by doing something that would be ruinously expensive to fake. That move is a signal, and this course is the anatomy of it.

The problem signalling solves is everywhere: one party has private information about their own quality, type, or intentions, and the other party would love to know it but has no way to verify a mere claim. And here’s the trap — talk is cheap. Anyone can say “I’m reliable,” “I’m the best candidate,” “I mean it.” Words cost nothing, so a liar can produce them exactly as fluently as an honest party; a claim that a faker can copy for free carries no information at all. The only claims worth believing are the ones a faker wouldn’t want to copy — the ones whose cost is high enough that only the genuine type finds them worth sending. That is the whole secret, and it has a razor-sharp condition attached: a signal separates the good type from the bad exactly when it is cheaper for the good type to send than for the bad type. Get that cost gap right and the signal is honest; let it narrow and the liars flood back in.

This is an expert course standing where biology & evolution meets strategy & competition, and it composes three models you’ve already built. From natural selection you bring the engine that carves honest signals into living things: a trait that reliably advertises fitness gets chosen, generation after generation, even when it looks suicidal — the peacock’s tail is not despite selection but because of it. From the principal–agent problem you bring the exact setting signalling lives in — hidden information, adverse selection, the market for lemons — and the two-sided toolkit introduced there: screening (the uninformed party builds a test) and signalling (the informed party pays to reveal itself), which we now turn inside out. And from credible commitment & deterrence you bring the strategic heart of it: a threat or promise is believed only when it’s backed by something costly and hard to reverse — burning a bridge, sinking a cost, tying your own hands. If any of the three feel shaky, shore them up first; everything here rests on them.

From that base we build the model floor to ceiling. We start with the sharp line between cheap talk and costly signals, and prove the separating condition that runs through the whole course. Then we watch it play out in biology — Zahavi’s handicap principle, the peacock and the stotting gazelle, honesty enforced by cost itself. We take it into markets — Spence’s Nobel-winning job-market signalling, warranties and brands that resolve the lemons problem, and the ugly flip side: credential inflation and signalling arms races that burn real resources to prove things that were already true. We turn to strategy and society — deterrence signals that tie hands and sink costs, and social signals from conspicuous consumption to virtue signalling to the costly initiation rites that bind groups. We map the equilibria — pooling, separating, semi-separating — and the delicious twist of countersignalling, where the very top skip the signal because they have nothing left to prove. And we stay ruthlessly honest about where the model lies: how honesty collapses into dishonest mimicry when the cost gap narrows, the pure deadweight waste of arms races where everyone spends more to stand still, and the trap of reading a signal as causing the quality it merely reveals.

Throughout, you’ll have a signalling separator to play with — drag the signal’s cost for a high type against its cost for a low type, dial how common the good type is, and watch the equilibrium snap between pooling (nobody can be believed, so the market pays everyone the average) and separating (only the good type signals, and is believed) — while a readout shows exactly how much pure waste gets burned to make the truth credible. By the end you’ll stop asking the naïve question — is this claim true? — and start asking the sharp one: would this cost the liar as much as it costs the honest party? If not, it’s just talk.