Analysis and Design of Information Transmission in Networks of Strategic Agents
Nasimeh Heydaribeni · Deep Blue (University of Michigan) · 2021
Most of today's systems consist of strategic/selfish agents with some private information and uncertainty towards others' information and system states. Transmission and exchange of information in such networks have been the focus of many interesting research areas such as mechanism design, information design, and Bayesian learning. The information is directly exchanged in mechanism design and information design, and the goal is to steer agents' actions towards a desirable direction by putting incentives in place (mechanism design) or designing the appropriate information structure (information design). Information can also be spread more indirectly by agents who observe each other's actions (Bayesian learning). In this thesis, we follow two main directions of ``Analysis" and ``Design" to investigate the spread of information in networks of strategic agents. Specifically, we analyze dynamic systems with asymmetric information and characterize their equilibria and study the spread of information induced by these equilibrium behaviors. Furthermore, we study how incentives or information structures can be designed to shape the equilibrium behavior of agents. In part I of this thesis (analysis part), we study structured perfect Bayesian equilibria (structured PBE) in dynamic games with asymmetric information. While there is no general framework to characterize such equilibria, we can study them for some specific information structures. Specifically, we consider games with conditionally independent types. As an example of such games, we study a setting where there is a marketplace with a product that has an unknown value and privately informed agents coming to the market to decide on buying or not buying the product. The agents get multiple chances to enter the market, and in this sense, they act non-myopically. Characterization of structured PBE in this game enables us to analyze informational cascades and suggest settings that avoid such outcomes. In part II of this thesis (design part), we design distributed mechanisms for efficient resource allocation in networks. The message transmission is done locally in our mechanisms, and we investigate how appropriate information is propagated throughout the network so that the equilibrium outcome is efficient. We also study a joint information and mechanism design problem where agents with private types arrive at a queue with an unobservable backlog. We study how a planner that observes the queue backlog can design taxes and type-dependent admission signals for the agents to gain the most revenue. We further analyze an information design problem for a non-atomic service scheduling game. We investigate how a planner can give suggestions to users about the time to join a queue for a service with an unknown start time to minimize the social cost.