Signaling Game

A signaling game is a game of incomplete information in which an informed player (the sender) chooses a signal to communicate private information to an uninformed player (the receiver), who then takes an action. The canonical model, developed by Michael Spence in 1973 to explain job-market signaling, reveals a counterintuitive principle: for communication to be credible, the signal must be costly, and the cost must differ across types. If all types could send the same signal at the same cost, no information would be transmitted. The signal works precisely because some types cannot afford to mimic others.

The Canonical Model

In the Spence model, a worker knows their own productivity (high or low) but the employer does not. The worker chooses an education level, which serves as a signal. Education is costly — in time, effort, and tuition — but it is more costly for low-productivity workers than for high-productivity workers. The employer observes the education level and offers a wage based on their updated belief about the worker's type.

The equilibrium analysis proceeds through perfect Bayesian equilibrium. The employer holds prior beliefs about the distribution of worker types, observes the signal, and updates beliefs using Bayes' rule. The worker's strategy must be optimal given the employer's wage schedule, and the employer's wage schedule must be optimal given their updated beliefs. The equilibrium is a fixed point in beliefs and strategies: the signal means what the receiver believes it means, and the receiver's beliefs are justified by the sender's behavior.

The mathematics is straightforward but the implications are deep. The signal — education — need not have any productive value whatsoever. It can be pure cost, a pure filter, and yet it coordinates expectations and determines outcomes. This is the essence of costly signaling: the signal works not because of what it says, but because of what it costs the sender to say it.

Equilibrium Types

Signaling games admit three classes of equilibrium, distinguished by how much information the signal conveys.

Separating Equilibrium. Each type sends a distinct signal, and the receiver learns the sender's type with certainty. Separation requires that the cost difference between types is large enough to prevent mimicking. In the Spence model, this means high-productivity workers choose enough education that low-productivity workers would rather accept the low wage than incur the education cost.

Pooling Equilibrium. All types send the same signal, and the receiver learns nothing. The signal is purely ceremonial. Pooling equilibria are often Pareto-dominated by separating equilibria — all parties would be better off if information were revealed — but the incentives to deviate prevent revelation.

Semi-separating equilibrium. Some types randomize between signals, and the receiver's posterior beliefs are partial. This is the realistic case: some information is transmitted, but not perfectly, and the noise is structural rather than accidental.

The multiplicity of equilibria is a feature, not a bug. It means that the same underlying fundamentals — the same cost functions, the same type distributions — can support radically different social outcomes depending on which equilibrium is selected. This is why signaling games are the natural framework for studying social conventions, credentials, and ritual: they capture how arbitrary practices become self-enforcing information structures.

Applications and Extensions

The signaling framework extends far beyond labor markets. In biology, the handicap principle — developed by Amotz Zahavi — explains why peacocks carry costly tails and why gazelles stot (jump high) when pursued by predators. The tail and the stot are signals of fitness precisely because they are costly; a weak peacock cannot afford the tail, and a tired gazelle cannot afford the stot. The biological and economic models are formally identical: both are costly signaling games in which differential costs sustain information transmission.

In political economy, signaling explains why regimes engage in costly demonstrations of commitment, why central banks endure recessions to prove their anti-inflation credibility, and why states fight wars rather than bargain. In each case, the cost is the point: the signal is credible because it would be too expensive to fake.

The relationship between signaling games and Bayesian persuasion is illuminating. In Bayesian persuasion, the sender can commit to an information structure; in signaling games, the sender cannot commit and must bear the cost of credibility. The gap between the two models is the gap between institutional communication (where commitment is maintained by reputation, law, or mechanism design) and interpersonal communication (where credibility must be earned through costly action).

The signaling game is often presented as a model of information asymmetry — a technical solution to a technical problem. This is wrong. The signaling game is a model of how social structure emerges from strategic interaction. Education, credentials, rituals, displays, and conventions are not arbitrary cultural artifacts. They are equilibrium strategies in games where information is scarce and trust is costly. To understand a society's signals is to understand the games its members are playing — and the costs they are willing to bear to play them.