Mechanisms to Manage Incentives in Online Systems a Dissertation Submitted to the Department of Management Science and Engineering and the Committee on Graduate Studies of Stanford University in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy

In large scale online systems, such as electronic marketplaces and peer-to-peer systems, users often act in their self interest without considering whether their actions lead to efficient outcomes for the system. This thesis studies two important classes of mechanisms that can be used to incentivize users to act in a way that promotes efficiency. Aggregation mechanisms provide aggregate information on the past behavior of a user to other users in the system. When there is such a mechanism in place, a user should expect that bad behavior now affects his future interactions within the system, and may be incentivized to act in a way that is beneficial for the system. Market mechanisms can be used to incentivize contribution to the system by using prices to identify value, and associating a budget with each user; the budget increases when the user contributes to the system and decreases when he uses system resources. By requiring that users have non-negative budgets, users can only use the system in return for valuable contributions. In Chapter 2 we address a basic question: how do we design an aggregation mechanism to encourage trustworthy behavior? Electronic marketplaces, such as eBay, are a natural setting to study this question, since they usually rely on mechanisms that collect ratings of sellers from past transactions, and provide aggregate information to potential buyers. First, we show that weighting all past ratings equally gives sellers an incentive to falsely advertise. We then study aggregation mechanisms that weight recent ratings more heavily, and show that under increasing returns to reputation the optimal strategy of a sufficiently patient and sufficiently high quality seller is to always advertise honestly. We suggest approaches for designing an aggregation mechanism that maximizes the range of parameters for which it is optimal for the seller to be truthful. We show that mechanisms that use information from a larger number of past transactions tend to provide incentives for patient sellers to be