The first chapter of this dissertation studies a principal-supervisor-agent model in which a privately informed supervisor is susceptible to collusion. We explore how the supervisory information helps the principal to extract information rent from the agent when the informational asymmetry between the supervisor and agent results in inefficient collusion on their part. The optimal collusion-proof mechanism is such that the principal receives expected payoff that is as high as in the direct supervision benchmark in which the principal publicly observes the supervisory signal. Under some parameters, the principal can extract more rent from the agent, and the equilibrium is more efficient. The second chapter analyzes an informed principal problem in which the principal privately observes the supervisory signal. The principal's expected payoff is strictly higher than the case in which the signal is publicly observed. This is a common feature of the independent type cases, except that this result holds even if preferences are quasi-linear. Unlike in the independent case, efficiency is constrained by the principal's incentive problem. The third chapter considers a collusion formation process in which the supervisor proposes the side mechanism. We show that the informed principal problem within the coalition does not affect the optimal design of the grand mechanism. We then consider the supervisor's incentive to exit once she obtains new information. The principal can take advantage of the implicit information revealed by the supervisor's exit decision. The direct supervision benchmark outcome can be attained by a simple mechanism if the coalition cannot write a contract based on the exit decision, or the principal can publicly observe the timing of the supervisor's exit decision.
