Rationing through Classification

Basing object-assignment on recipients' actions can improve allocations but also distort action choices. The degree to which a planner should utilize such information to make allocation decisions is addressed in previous work on "decentralized" settings, i.e. where individual recipients strategically choose their own actions. We consider less competitive situations (e.g. organ allocation) where strategic agents (transplant centers) choose actions (medical treatments) on behalf of multiple potential recipients (patients). When there are many such agents, we show that equilibrium behavior resembles that of the decentralized setting and that a reduction in competition reduces the threat of distortions, improving welfare. With sufficiently few agents, however, a second form of equilibrium arises where agents inefficiently "invert" action choices for multiple recipient types, diverting resources towards high-value recipients. Computational analysis suggests that such equilibria are welfare-optimal only in some low-stakes allocation problems.