CIRJE-F-178. Hitoshi Matsushima, "Honesty-Proof Implementation", October 2002.

We investigate implementation of social choice functions that map from states to lotteries and may depend on factors other than agents' preferences. We assume that agents are not only purely self-interested but also honesty-oriented in a lexicographical way. We define iterative honesty-proofness by iteratively removing messages dominated by more honest messages. We show that in the complete information environments with small fines, every social choice function is implementable in iterative honesty-proofness. This is in contrast with the standard implementation model, because any 'normative' social choice function depending on non-preference factors is never implementable when agents are not influenced by factors other than pure self-interest. We extend this result to the incomplete information environments with quasi-linearity and with correlated private signals.

Next, we assume that it is costly for each agent to report dishonestly and this cost may be close to zero. We show that in the incomplete information environments, every incentive compatible social choice function can be implemented by the mechanism that is universal in the sense that it does not depend on the private signal structure.