Efficient Auctions with Altruism∗ Ruggiero Cavallo Microsoft Research 1290 Avenue of the Americas, 6th floor New York, NY 10104
[email protected] September 6, 2012 Abstract We introduce a novel regret-based model of altruism, wherein agents are willing to forgo a small amount of value if doing so increases social welfare, and consider its implications in a single-item allocation setting. We demonstrate that even for very mildly altruistic agents classic approaches such as VCG are manipulable, but straightforward variants of known mechanisms that are strategyproof in the purely-selfish case succeed in achieving dominant strategy efficiency and strong budget-balance, i.e., full social welfare for the group, a result unachievable in the case of completely selfish agents. We contrast this positive result with a negative analysis of a more traditional (non-regret- based) altruism model, demonstrating that nothing short of complete altruism yields existence of a strongly budget-balanced efficient mechanism there. JEL Classification: D64, D44, D01, D82 1 Introduction Mechanism design is a cynical enterprise: the goal is to derive schemes under which selfish agents, unconcerned with the welfare of others, cannot benefit from doing other than what is considered optimal from a social perspective (e.g., truthfully reporting private information to allow identification of a social welfare maximizing action). When such schemes succeed in simultaneously satisfying the goals of the social planner and each agent, the assumption of selfishness makes them robust. But unfortunately often they cannot succeed; most notably, efficient mechanisms in ∗A significant portion of this work was completed while the author was a postdoctoral fellow at the University of Pennsylvania.