Abstract: We study when and how randomization can help improve the seller's revenue in the sequential screening setting. In a model with discrete ex ante types and a continuum of ex post valuations, the standard approach based on solving a relaxed problem that keeps only local downward incentive compatibility constraints often fails. Under a strengthening of first-order stochastic dominance ordering on the valuation distribution functions of ex ante types, we introduce and solve a modified relaxed problem by retaining all local incentive compatibility constraints, provide necessary and sufficient conditions for optimal mechanisms to be stochastic, and characterize optimal stochastic contracts. Our analysis mostly focuses on the case of three ex ante types, but our methodology of solving the modified problem, as well as the necessary and sufficient conditions for randomization to be optimal, can be extended to any finite number of ex ante types.
Keywords: Sequential Screening, Stochastic Mechanism
JEL Classification: D83; D82