Portal de Periódicos da CAPES

An Experimental Test of Equilibrium Dominance in Signaling Games

1992; American Economic Association; Volume: 82; Issue: 5 Linguagem: Inglês

1944-7981

Jordi Brandts, Charles A. Holt,

Experimental Behavioral Economics Studies

Many economic situations with asymmetric information can be modeled as signaling games. Even simple signaling games can have sequential equilibria that are considered unintuitive. For example, in a wellknown model due to Michael Spence (1973), workers with good information about their own ability levels have to decide whether to obtain additional education or not. An employer, knowing that education is more costly for low-ability workers, observes the education signal, but not the worker's ability, prior to deciding on a wage offer. As shown below, it is possible to construct examples in which all types of workers decide against the educational investment because the employer will interpret education as a signal of low ability, even though the signal is not less costly for low-ability workers. Despite the unintuitive nature of these out-of-equilibrium beliefs, this equilibrium outcome survives the tests imposed by all of the commonly used, strengthened versions of the Nash concept. In particular, this outcome is a sequential equilibrium in the sense of David M. Kreps and Robert Wilson (1982); it satisfies a backwards-induction rationality requirement that decisions be optimal from any nonterminal stage until the end of the game, given the equilibrium beliefs at that stage. Despite the fact that unintuitive outcomes can pass the test imposed by a sequential equilibrium, there is no consensus on exactly how this equilibrium concept should be strengthened or refined. The recent debate centers on refinements that place more restrictions on players' beliefs about what would happen off the equilibrium path. Several refinements have been proposed, each of which is motivated by specific games in which a weaker refinement permits unreasonable equilibria. For example, In-Koo Cho and Kreps (1987) discuss their intuitive criterion, which is based on a dominance notion that they call equilibrium dominance. They also discuss a number of stronger refinements, including strategic stability (Elon Kohlberg and JeanFrancois Mertens, 1986) and divinity (Jeffrey Banks and Joel Sobel, 1987).1 We are particularly interested in equilibrium dominance, since it is widely discussed and since it is implied by stronger refinements. The arguments about which refinement is appropriate are typically based on subjective opinions about how reasonable individuals would behave. There is an apparent need for empirical work that directly tests the validity of these arguments; such work could guide the theoretical debate toward alternative refinements. Although game theory is sometimes used as a normative theory about how rational agents ought to behave, the theory is also widely applied in