Evaluating the conditions for robust mechanism design

We assess the strength of the different conditions identified in the literature of robust mechanism design. We focus on three conditions: ex post incentive compatibility, robust monotonicity, and robust measurability. Ex post incentive compatibility has been shown to be necessary for any concept of robust implementation, while robust monotonicity and robust measurability have been shown to be necessary for robust (full) exact and virtual implementation, respectively. This paper shows that while violations of ex post incentive compatibility and robust monotonicity do not easily go away, we identify a mild condition on environments in which robust measurability is satisfied by all social choice functions over an open and dense subset of first-order types. We conclude that there is a precise sense in which robust virtual implementation can be significantly more permissive than robust exact implementation.