Mechanism design with tacit collusion

In the mechanism design literature, collusion is often modelled as agents signing side contracts. This modelling approach is in turn implicitly justified by some unspecified repeated-interaction story. In this paper, we first second-guess what kind of repeated-interaction story these side-contract theorists (would admit that they) are having in mind. We then show that, within this repeated-interaction story, there is a big difference between communicative and tacit collusion. While communicative collusion hurts the mechanism designer, tacit collusion is exploitable.