Credible implementation

The theory of mechanism design and implementation abounds with clever mechanisms whose equilibrium outcomes are optimal according to some social choice rule. However, the cleverness of these mechanisms relies on intricate systems of rewards and punishments off-the-equilibrium path. Generally, it is not in the designer's best interest to go through with the reward/punishment in the "subgame" arising from some disequilibrium play. This would make the mechanism's outcome function non-credible. In the context of exchange economies, we define an appropiate notion of "credible" implementation and show that (a) the non-dictatorial Pareto correspondence can be crediblyimplemented (b) there exists no credibly implementable Pareto-efficient and individually rational social choice rule and (c) there exists no credibly implementable fair social choice rules. We derive necessary and sufficient conditions for credible implementability of choice rules. The main implication is paradoxical: it is suboptimal for the designer to be endowed with "too much" information about the economy. Finally, we show that the negative results persist even under weaker credibility requirements .