Dynamic Cost Per Action Mechanisms and Applications to On-line Advertising

We examine the problem of allocating a resource repeatedly over time amongst a set of agents. The utility that each agent derives from the consumption of the item is private information to that agent and, prior to consumption, may be unknown even to that agent. The problem is motivated by keyword auctions, where the resource to be allocated is a slot on a search page. We describe a mechanism based on a sampling-based learning algorithm that under suitable assumptions is asymptotically individually rational, asymptotically Bayesian incentive compatible and asymptotically ex-ante efficient.