Nonparametric Identication and Structural Estimation of Auction Models

Thesis (Ph.D.)--University of Washington, 2016-06

This dissertation contributes to the structural auction literature in two different auction models, namely the pure common value model and the affiliated private value model. The goal of structural analysis of auction data is to recover the model primitives and to provide policy guidance for welfare analysis. In Chapter 1, we study identification in the first-price and the second-price sealed-bid auctions within the pure common value framework. In Chapter 2, we apply the identification results and estimation method in Chapter 1 to analyze the U.S. Outer Continental Shelf (OCS) wildcat auction data and provide policy guidance for welfare analysis. In Chapter 3, we develop identification and partial identification results for the first-price and the second-price sealed-bid auction models with affiliated private values and incomplete sets of bids. Chapter 1: In this chapter, we establish novel identification results for both the first-price and the second-price sealed-bid auction models within the pure common value framework. We show that the policy parameters, including the expected total welfare, the seller's expected revenue, and the bidders' expected surplus under any reserve price are identified for a general nonparametric class of latent joint distributions when the ex-post common value is unobserved. Moreover, we establish that these policy parameters are nonparametric identified without normalization assumption when the ex-post common value is observed. We propose a semiparametric estimation method and establish consistency of the estimator. Results from Monte Carlo experiments reveal good finite sample performance of the estimator. Chapter 2: In this chapter, we employ the identification strategy and estimation method in Chapter 1 to analyze data from the U.S. Outer Continental Shelf (OCS) wildcat auctions in the pure common value framework. We study the welfare implication of different counterfactual reserve prices, focusing on the cases with two and three bidders. The empirical results suggest that if the U.S. government had set reserve prices optimally using the newly-developed econometric method in Chapter 1, its expected revenue can be increased by around $34\%$ and $30\%$ for these two cases, respectively. Lastly, we compare our results with those estimated under the affiliated private value framework, and find that the estimated welfare curves under the two different frameworks are very different. Chapter 3: In this chapter, we address the identification issue in the first-price sealed-bid affiliated private value model when an incomplete set of bids is observed. In the simple case with symmetric bidders and non-binding reserve price, we establish identification or partial identification results in two scenarios of practical interest. First, when the two highest bids are observed, we achieve identification of the joint distribution function of private values by assuming the copula function of private values to be a nonparametric Archimedean copula with weak requirement. Second, when only the highest bid is observed, we establish partial identification for the quantile function of private value and several policy parameters by parameterizing the copula function. Further, we extend the identification/partial identification results to the cases with asymmetric bidders and/or binding reserve price. We also extend our identification/partial identification results to the second-price sealed-bid auction.