Social Welfare Maximization in Dynamic Strategic Decision Problems
Advisor: David Parkes
Full dissertation pdf
Chapter 1: Introduction
Chapter 2: Static mechanism design
Chapter 3: Redistribution mechanisms
Chapter 4: Optimal decision-making in dynamic settings
Chapter 5: Dynamic mechanism design
Chapter 6: A dynamic redistribution mechanism
Chapter 7: Periodic inaccessibility and dynamically changing populations
Chapter 8: Efficient metadeliberation auctions
Chapter 9: Conclusion
Deriving effective group decision-making procedures for complex environments is hard but of fundamental importance, and the challenges grow significantly more daunting when individuals are self-interested. There is an inherent tension in striving to achieve social goals in decisions that will impact individuals who are only concerned with selfish objectives. Innumerable scenarios fit this mold, from resource allocation to coordinating behavior in the presence of global constraints. The field of mechanism design addresses such problems via specific payment schemes that disarm agent self-interest. This thesis attacks two fundamental issues in this area.
First: How can one implement a decision-making mechanism that maximizes the net welfare of a group of self-interested agents? Classic solutions typically require agents to make large payments to a central coordinator which, from the agents' perspective, purely detracts from social welfare. This thesis provides a mechanism applicable to arbitrary group decision-making problems that yields drastically higher group welfare in important settings, including resource allocation. The redistribution mechanism uses structure inherent in the domain to give payments required in the classic solution back to the agents in a way that does not yield a budget deficit or distort their incentives to participate truthfully.
Second: How can social welfare maximizing outcomes be reached with selfish agents in a setting that is dynamic and uncertain? In the real world, decisions do not exist in isolation, but rather are situated in a temporal context with other decisions. Individuals will act to maximize their utility over time, and decisions in the present influence how the world will look in the future, but rarely in completely predictable ways. This thesis addresses the problem of dynamic mechanism design for such settings and provides key results including: a characterization of the social welfare maximizing dynamic mechanisms that can be implemented in an ex post equilibrium; an extension to handle dynamically changing agent populations; an application to coordinating research preceding allocation of a resource. Finally, a dynamic redistribution mechanism unifies the two main focuses of the thesis, providing a solution with near-optimal social welfare properties for an array of important dynamic problems.