Rationalizable implementation of correspondences

We come close to characterizing the class of social choice correspondences that are implementable in rationalizable strategies. We identify a new condition, which we call set-monotonicity, and show that it is necessary and almost sufficient for rationalizable implementation. Set-monotonicity is much weaker than Maskin monotonicity, which is the key condition for Nash implementation and which also had been shown to be necessary for rationalizable implementation of social choice functions. Set-monotonicity reduces to Maskin monotonicity in the case of functions. We conclude that the conditions for rationalizable implementation are not only starkly different from, but also much weaker than those for Nash implementation, when we consider social choice correspondences.