My job market paper examines a scenario where a decision-maker trying to learn about an underlying state of the world must decide not only how much information to purchase, but also from which sources. I develop an approximate consumer theory for information, valid when information purchases are large and the relevant probabilities are well-described by large-deviations theory. I show that demand for information behaves as though preferences were kinked, and thus is poorly described by the benchmark smooth, convex preference model.

In this project, we generalize some of the results from my job market paper to a Bayesian estimation setting and consider implications for optimal experiment design. We show that the optimal design at large samples generally maximizes the Fisher information per dollar for the worst-case true value of the parameter of interest.

Optimal fast experimentation, with Lones Smith
Bargaining with multiple buyers: evidence from eBay