UC San Diego

This dissertation presents three stand-alone contributions to econometric inference and the analysis of strategic behavior. Chapter 1 develops a structural econometric framework for first-price auctions that generalizes the assumption of Bayesian Nash Equilibrium within the context of a level-k behavioral model. The level-k model nests equilibrium by allowing bidders to best respond to heterogeneous beliefs about opponents' bidding strategies. I characterize conditions for identification of the distribution over valuations and bidder-types in populations with heterogeneous behavioral strategies. I propose a semi-nonparametric maximum likelihood estimator, establishing nonparametric consistency with an upper- semicontinuous population likelihood function, which I compute using a generalized expectation maximization algorithm. Presenting evidence from a pilot study of vintage computer auctions, I find a high level of bidder sophistication in the field. I also characterize expected revenues in first price auctions with level-k bidders, establishing a partial identification result for expected revenues in unidentified behavioral models. Empirical evidence suggests a misspecified equilibrium optimal reserve price could reduce expected revenues 30% relative to an unbinding reserve price. Chapter 2 introduces new Bayesian methods adapted to estimating a large-dimensional covariance matrix. I analyze the return generating process using an unrestricted factor model of covariance, imposing structure on the covariance matrix through prior beliefs on the parameters governing this process. I use these methods to provide an empirical Bayesian foundation for a general class of shrinkage estimators and use the shrinkage interpretation to characterize prior beliefs that optimize a posterior objective function. This estimation strategy delivers lower finite-sample loss than existing estimators in Monte Carlo simulations and performs well in minimum variance portfolio selection exercises. Chapter 3 analyzes conformist tendencies for a population in which individuals gain utility by mimicking the average behavior, characterizing norms by the mean behavior, thus introducing an endogenous mechanism for establishing social norms. Under this specification, social preferences generally give rise to more concentrated behavior and a conformist pool forms when social preferences are sufficiently prominent. In addition to illustrating the determinants of conformist behavior with an endogenous reference point, these findings support applied work inferring social norms from average behavior