Bayesian mechanism design, which requires explicit assumptions on the agents’ beliefs, has recently proved useful in algorithmic and automated mechanism design (mostly for internet applications). We want to capitalize on these advances and go beyond the two extreme assumptions on beliefs that have been privileged in previous work, full-freeness (or independence) and no-freeness (or Beliefs Determine Preferences). Using the notion of "independence class of beliefs", we introduce a general condition covering all cases between no-freeness and full-freeness. The set of mechanisms that can be implemented under this general condition is very large and can be characterized by the property of "independence class efficiency" both under imposed and voluntary participation. (joint work with Jacques Crémer).

Язык: английский