Agents' valuations are interdependent if they depend on the signals, or types, of all agents. Under the implicit assumption that agents cannot observe their outcome-decision payoffs, previous literature has shown that with interdependent valuations and independent signals, efficient design is impossible. This paper shows that an efficient mechanism exists in an environment where first the final outcome (e.g., allocation of the goods) is determined, then the agents observe their own outcome-decision payoffs, and then final transfers are made.

We examine a dynamic model of up-or-down problem solving. A decision maker can either spend resources investigating a new problem before deciding what to do or decide on the basis of similarity with precedent problems. Over time, a decision-making framework, or jurisprudence, develops. We focus on the model’s application to judge-made law. We show that judges summarily apply precedent in some cases. The law may converge to efficient or inefficient rules. With positive probability, identical cases are treated differently. As the court learns over time, inconsistencies become less likely. We discuss the existing empirical evidence and the model’s testable implications.