Incomplete contracting models often neglect to consider revelation mechanisms that can condition contracts on all relevant information by eliciting it from the involved parties. This paper considers a simple buyer-seller relationship under one-sided asymmetric information, where the incompleteness of information leads to ine±ciencies. We show that in the presence of renegotiation the gain from writing a complete contract is zero. The paper also provides an intuition that contracting may lead parties to undertake ine±cient actions on the equilibrium path above the ine±ciency created through the absence of a contract. Furthermore, any contract involves on-the-equilibrium-path renegotiation and thus in this context the Renegotiation-Proofness-Principle does not apply. I am very grateful to John Moore, Oliver Hart and Leonardo Felli for their advice and encouragement. Many thanks also go to Alexander Muermann and Daniel Sturm. Financial support by the German Academic Exchange Service (DAAD) and the TMR (Marie Curie Fellowship) is gratefully acknowledged. y

In a two period decision problem, we study individuals who, in the second period, may be tempted by ex ante inferior choices. Individuals have preferences over sets of alternatives that represent the feasible choices in the second period. Our axioms yield a representation that identifies the individual's ex ante ranking, her temptation ranking, and her cost of self-control. We provide an axiomatic model of temptation to justify the main assumption of our representation theorem and to analyze second period behavior. An agent has a preference for commitment if she strictly prefers a subset of alternatives to the set itself. An agent has self-control if she resists temptation and chooses an option with higher ex ante utility. We introduce comparative measures of preference for commitment and self-control and relate these measures to our representations.

(First Version: April 2008) We model a DM as a collection of utility functions (selves, rationales) and an aggregation rule (a theory of how selves are activated by choice sets) on which we impose five simple axioms of social choice. This framework encompasses many multi-self models proposed in the existing literature. For a broad class of aggregators we show that with sufficiently many selves the resulting model can rationalize any choice function. We define an accounting procedure for IIA violations and show that for any fixed number of selves, a lower bound on the set of choice functions that these aggregators can rationalize is given by the set of choice functions that exhibit no more IIA violations than a certain linear function of the number of selves.