Dynamic Implementation, Verification, and Detection

Abstract

We investigate implementation of social choice functions, where we impose severe restrictions on mechanisms, such as boundedness, permitting only tiny transfers, and uniqueness of an iteratively undominated strategy profile in the ex-post term. We assume that there exists some partial information about the state that is verifiable. We consider the dynamic aspect of information acquisition, where players share information, but the timing of receiving information is different across players. By using this aspect, the central planner designs a dynamic, not a static, mechanism, in which each player announces what he (or she) knows about the state at multiple stages with sufficient intervals. By demonstrating a sufficient condition on the state and on the dynamic aspect, namely full detection, we show that a wide variety of social choice functions are uniquely implementable even if the range of players’ lies that the verified information can directly detect is quite narrow. With full detection, we can detect all possible lies, not by the verified information alone, but by processing a chain of detection triggered by this information. This paper does not assume either expected utility or quasi-linearity.