Picking Winners in Rounds of Elimination

Abstract

We study the optimal way to select projects or agents in environments where information arrives in well defined rounds. Examples include academic environments where review periods are set by policy, aptitude tests such as those given by software developers to programmers applying for jobs, venture capital protocols where the rounds of funding may be stopped before the project is complete, and FDA testing, where drugs can be dropped at well defined junctures. Sequential rounds of elimination reduce the cost of selection, but also reduce the average quality of surviving projects. We characterize the nature of the optimal screening process with and without "memory."

Keywords: log supermodularity, rounds of elimination, picking winners, screening

JEL Classification: C, L2