Mechanism design is the science of designing the rules of a game (e.g., auction or election) so that good outcomes ensue despite each participant (human or computational) acting based on self-interest. Intuitively, allowing individuals or organizations to express richer preferences should yield better outcomes. Billions of dollars of annual savings from outcome efficiency improvements due to increased expressiveness have indeed been observed, for example, in combinatorial sourcing auctions by the PI and others. What is missing is a rigorous methodology for designing appropriately expressive mechanisms. This project combines new theoretical results in mechanism design - including computational measures of expressiveness - with custom search algorithms and machine learning techniques. The goal is to create knowledge about mechanism design with varying levels of expressiveness. The work also involves developing an operational methodology to guide the design of appropriately expressive mechanisms across a broad class of combinatorial and multi-attribute domains. Furthermore, the work will yield new theory and computational methodologies for bundling. The objects of study are rational agents and agents with forms of bounded rationality.
While many of the results will be general, the work will be validated in qualitatively different applications such as combinatorial auctions, advertising markets, and catalog-offer-based (web) commerce. In the US alone, combinatorial multi-attribute sourcing auctions give rise to tens of billions of dollars in annual trade. Annual volumes associated with advertising markets, spectrum auctions, consumer-to-consumer auctions, and catalog-based commerce are all in hundreds of billions or trillions of dollars. Improvements would thus offer substantial economic and societal benefits.
