For instructors

Teaching notes

An applicant completes a degree that teaches little she will use, because employers read it as evidence of ability. A manufacturer offers a long warranty that only a maker of reliable goods could afford. An incumbent keeps prices low to persuade a potential entrant that its costs are low. In each case an informed party acts and an uninformed party draws a conclusion.

What the game shows

The game tests whether the actions of an informed party reveal what that party knows. It shows the difference between separating outcomes, in which types act differently and can be told apart, and pooling outcomes, in which they act alike and nothing is learned. It also brings out the role of beliefs about moves that are not supposed to occur, which is where the equilibrium concepts for such games differ.

The theory

Spence (1973) showed that a costly action can separate types when it is cheaper for one type than for the other, even if the action has no value of its own. In a perfect Bayesian equilibrium the sender's choice is optimal for each type given the receiver's response, and the receiver's response is optimal given beliefs that follow from Bayes' rule wherever it applies. After a move that no type makes in equilibrium, Bayes' rule is silent, and many outcomes can be supported by a suitable belief.

The default game is the beer and quiche example of Cho and Kreps (1987). It has two pooling equilibria, one in which both types choose beer and the receiver would fight after quiche, and another in which both choose quiche and the receiver would fight after beer. The intuitive criterion of the same authors rejects the second of these. The weak type already obtains its best payoff with quiche and could not gain from beer under any response, so a deviation to beer can only come from the strong type, and the receiver should then not fight, which makes the deviation profitable for the strong type. The job market preset has a separating equilibrium, in which only the able worker takes the degree, and a pooling equilibrium without degrees.

What to expect

Play in signaling games converges on an equilibrium more often than not, though which one depends on early experience. Brandts and Holt (1992) found that about four fifths of choices in their baseline game matched the equilibrium selected by the intuitive criterion. After they altered the payoffs so that early play pointed the other way, most choices matched the equilibrium that the criterion rejects. Banks, Camerer and Porter (1994) report that most choices are part of some Nash equilibrium and that participants tend toward the more refined equilibria only up to a point. Cooper, Garvin and Kagel (1997) study how play adjusts with experience in a game of limit pricing.

For a class this means that the outcome is open. Receivers form beliefs from what senders of each type did in earlier rounds, and those early rounds are shaped by senders who choose what is best for their type without regard to the inference it invites.

Tips and things to look out for

References