The Value of Data

Abstract

Data is an essential input of many modern industries. Yet, its value is hard to establish, since formal markets for data are still lacking. We study what value different types of data have for a principal---e.g., an e-commerce intermediary---who uses it to mediate the interaction between multiple agents---e.g., buyers and sellers. Our solution formulates this mediation as an information-design problem and uses linear-programming duality to characterize the principal's willingness to pay for each type of data. This reflects externalities between datapoints, which arise from how the principal optimally garbles them. Building on this, we study how the principal values information that refines existing datapoints, which we show can be zero or even negative. Our work establishes basic properties of the demand for data, a necessary step towards a full analysis of data markets and their welfare properties.