This interdisciplinary project investigates the detailed structure of information sources and processing that underly strategic interactions in games of incomplete information. Traditional economic models typically treat private information, or signals, as generated from some underlying state. However, multiagent interactions often pivot around situations where signals are actually based on alternative interpretations of available information, and such processes often produce qualitatively different results for the game. By developing a common language using probabilistic graphical models, the effort seeks insights about alternative models of signals, and computationally effective representations in support of strategic analysis. Specific technical developments will include extension of qualitative probabilistic reasoning methods, tailored to capture important distinctions in signal structure. The project further develops an expanded view of signals, that encompasses deliberate choices about what information to consider in complex decisions under uncertainty.
Results from this project will have applications to many situations of economic significance. An example is the domain of auctions, which encompasses markets in electronic commerce (e.g., eBay and B2B exchange), internet advertising (keyword search, social media), financial securities (equities, commodities), and energy (electricity, fuels). Auctions are conventionally modeled in the game-theoretic framework employed in this research, but typically using much more restricted models than necessary. By bringing to bear new computational methods and extending the scope of analysis, the project may lead to innovative new market designs, and a better understanding of human and automated market behavior. The techniques employed bridge across computational and economic disciplines, and open the door to more comprehensive and operational models of decision making.
