Learning From Diverse Sources and Models

University Of Pennsylvania, Philadelphia PA

Investigators

Abstract

This award funds research in economic theory. Markets are increasingly saturated with very detailed and precise information, such as data about consumer preferences, social networks or product attributes. An important but understudied aspect of these settings is how the multiplicity of sources of information available to consumers and firms affects their decisions, as well as the diverse ways in which this information is interpreted. For example, different past experiences and views of the world may lead individuals to interpret information in starkly different ways. A central objective of the proposed work is to identify the inefficiencies of information production and aggregation that emerge when the informational environment is complex, and to understand the ways in which these inefficiencies can be ameliorated via appropriate restructuring and regulation of the informational environment. There are many possible interventions. For example, a policymaker can release information by making a public announcement or seeding information with specific individuals, restructure the environment by restricting certain kinds of data sharing, or influence how individuals interpret information by providing information about others' preferences or beliefs. The projects funded by this award develop theoretical frameworks for clarifying the differences between these interventions and identifying which are most effective. These findings will be of practical use to inform policy regarding the design, dissemination, and regulation of information. Further, the problem of how to regulate data markets is of recent interest in many disciplines, including computer science, law, and sociology. The research will contribute to this interdisciplinary literature by focusing on the economic consequences to 'big data' production and usage. The research plan proposal consists of three projects. The first project focuses on the production of data when agents have access to many interrelated sources of information. The second proposes a model in which information about consumers is pooled across various markets and examines the consequences of sharing this information with firms. The third project considers how agents learn from their peers when they have mis-specified models of inference and seeks to characterize how an information planner can release information or modify the informational environment to facilitate learning. Although there is a classic and well-developed economics literature on learning and information acquisition, consideration of heterogeneity in sources and information-processing models has only recently begun to attract interest. For example, the learning literature has mostly focused on agents with correctly-specified models (i.e. Bayesian learners with priors that put positive weight on the true model). While more recent work has focused on the case where agents have mis-specified models, the question of what happens when agents with different mis-specified models interact is an important open channel for research. Similarly, the information acquisition literature has mostly studied learning from a single source, focusing on questions such as how much information should an agent acquire, and at what level of precision. Recent work introduces diverse information sources and studies dynamic information acquisition patterns. But study of the economic interactions between agents who aggregate information across diverse sources, e.g. in market settings, where firms aggregate information across consumers and markets for screening, remains incomplete. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.