We show that mechanisms which generate endogenous asymmetric information fully
mitigate collusion. In our model, an agent has private information and a supervisor observes
a signal that is correlated with the agent's type. Agent and supervisor can form collusive side
agreements. We study the implementation of social choice functions that condition on the
agent's type and the supervisory signal. Our main result establishes that any social choice
function that is implementable if the signal is public can also be implemented when the …

This paper shows that the possibility of collusion between an agent and a supervisor
imposes no restrictions on the set of implementable social choice functions (SCF) and
associated payoff vectors. Any SCF and any payoff profile that are implementable if the
supervisor′ s information was public is also implementable when this information is
private and collusion is possible. To implement a given SCF we propose a one-sided
mechanism that endogenously creates private information for the supervisor vis-à …