Much of the repeated game literature is concerned with proving Folk Theorems. The logic of the exercise is to specify a particular game, and to explore for that game specification whether any given feasible (and individually rational) value vector can be an equilibrium outcome for some strategies when agents are sufficiently patient. A game specification includes a description of what agents observe at each stage. This is done by defining a monitoring structure, that is, a collection of probability distributions over the signals players receive (one distribution for each action profile players may play). Although this is simply meant to capture the fact that players don’t directly observe the actions chosen by others, constructed equilibria often depend on players precisely knowing these distributions, somewhat unrealistic in most problems of interest. We revisit the classic Folk Theorem for games with imperfect public monitoring, asking that incentive conditions hold not only for a precisely defined monitoring structure, but also for a ball of monitoring structures containing it. We show that efficiency and incentives are no longer compatible.
