Signaling in deterministic and stochastic settings

► We present a generic signaling model with deterministic (invertible) and stochastic (noisy) actions and run experiments. ► The data are more congruent with noise, in part because noise is akin to imperfect play. ► Comparative static predictions also hold for noise. We contrast a standard deterministic signaling game with one where the signal-generating mechanism is stochastic. With stochastic signals a unique equilibrium emerges that involves separation and has intuitive comparative-static properties as the degree of signaling depends on the prior type distribution. With deterministic signals both pooling and separating configurations occur. Laboratory data support the theory: in the stochastic variant, there is more signaling behavior than with deterministic signals, and less frequent types distort their signals relatively more. Moreover, the degree of congruence between equilibrium and subject behavior is greater in stochastic settings compared to deterministic treatments.