Multiparty large-scale network formation: economic models and mechanisms

Networks permeate our social and economic lives. In industry, it has been long posited that research findings spill over to other firms. Peer-to-peer file-sharing networks provide online markets for media content that rival traditional media sources for total share of consumption. The Internet itself consists of a loose federation of independent service providers. In all these settings, the well-being of participants depends on social, geographic, or trading relationships. The growing interest in networks partly reflects a growing body of empirical work in a multitude of settings suggesting definite structure to the pattern of externalities between agents. It is also partly due to new opportunities for community formation and connectivity afforded by the Internet. In large part, growing interest in network analysis of social, organizational, and economic settings stems from a desire to influence behavior, provide robustness, or improve efficiency. The large and very diverse class of network-based problems makes this a rich area of study, while also requiring a flexible approach to modeling and a familiarity with diverse techniques. In this thesis, I study the derivation of networked economic models specially suited to large-scale decentralized systems with competing interests, with a particular focus on two problem domains: communication networks and collaboration networks. For each of these models I study the interplay between the structure of externalities, as captured by a network structure, and network formation dynamics, on the one hand, and economic efficiency, on the other. I also study the impact of the interaction protocol on efficiency and characterize a number of simple policies or algorithmic procedures for regulating system behaviour. The models that I develop address different empirical features of the systems. Accordingly, while the models have the self-interested utility-maximizing behaviour of participants in common, they differ in the decision space of agents and in the interaction protocol between the agents. The latter features determine the tractability (equivalently, the complexity) of computing economic outcomes in the model, as well as its suitability for different analysis techniques. In particular, the different formulations I investigate in this thesis allows me to leverage a number of different mathematical programming techniques ( mostly interior point methods in regards to LCP) in the computation of outcomes.