Week 3: Cheap Talk vs. Costly Signals

What separates a claim anyone could make from one that actually reveals something

6 March 2028

Open the slides

“I love hiking” costs nothing to say and nothing to be wrong about — cheap talk. A half-marathon medal photographed on a real trail, on the other hand, is costly to fake convincingly, which is exactly what makes it informative. Signaling theory’s whole content is this asymmetry: a signal is only worth reading if it would have been expensive to send falsely.

The uncomfortable implication, which this week states outright rather than leaves implicit, is that most of what a profile or a first date communicates is cheap talk, and the skill this course is building is not “write a better bio” but “tell the difference” — a skill that transfers directly to Assignment 1.

Outline#

• Spence’s job-market signaling model, translated
• separating vs. pooling equilibria: when cheap talk still carries information, and when it can’t
• what would have to be true for a claim on a dating profile to be a costly signal rather than noise
• why a costly signal can be entirely credible and still tell you nothing about compatibility

One thing this framework won’t explain#

The model’s whole test for a claim is whether it would have been expensive to fake. That test can be passed by something completely irrelevant to whether you’d actually get along: a marathon medal is genuinely costly to fake, and genuinely proves discipline, follow-through, and free time — none of which is what either of you actually cares about finding out. A costly signal earns your trust in the claim. It does not earn your trust that the claim was worth verifying in the first place.

That’s a different gap from Week 1’s. Week 1’s model had no term for whether speed itself was worth anything. This model has no term for whether the thing being signaled was ever the right thing to ask about. Both gaps point the same direction: toward Week 12.

Further reading#

• Spence, A. M. (1973). “Job Market Signaling.” The Quarterly Journal of Economics, 87(3), 355–374. The paper this week’s model is translated from — Spence later shared a Nobel Memorial Prize for it, for the same reason it works here: it’s a general theory of when a costly action can substitute for information nobody can observe directly.