This paper studies identification and estimation of first-price auctions if the bidders face ambiguity about the distribution of valuations. Ambiguity is modeled using Gilboa and Schmeidler's (1989) Maxmin Expected Utility preferences. We exploit variation in the number of bidders to identify the essential primitives of the model. The identification result yields a closed form for the inverse bid function, which suggests a two-step estimation procedure. We study asymptotic and finite sample properties of the estimators. We find evidence of ambiguity in USFS timber auctions which leads to aggressive bidding for bidders with high valuations and has important implications for auction design.
We study the identification and estimation of first-price auctions with independent private values if bidders face ambiguity about the valuation distribution and have maxmin expected utility. Using variation in the number of bidders we nonparametrically identify the true valuation distribution and the lower envelope of the set of prior beliefs. We also allow for CRRA and unobserved auction heterogeneity, and propose a Bayesian estimation method based on Bernstein polynomials. Monte Carlo experiments show that our estimator performs well, and incorrectly ignoring ambiguity induces bias and loss of revenue. We find evidence of ambiguity in timber auctions in the Pacific Northwest.
The first chapter studies identification and estimation of first-price auctions if the bidders face ambiguity about the distribution of valuations. Ambiguity is modeled using Gilboa and Schmeidler's (1989) Maxmin Expected Utility preferences. We exploit variation in the number of bidders to identify the essential primitives of the model. The identification result yields a closed form for the inverse bid function, which suggests a two-step estimation procedure. We study asymptotic and finite sample properties of the estimators. We find evidence of ambiguity in USFS timber auctions which leads to aggressive bidding for bidders with high valuations and has important implications for auction design. The second chapter proposes a procedure to test restrictions on infinite-dimensional parameters (partially) identified by unconditional or conditional moment equalities. Our new method allows us to test restrictions involving a continuum of inequalities. Examples of such restrictions include weakly increasing, concavity and first-order stochastic dominance. We show that our testing procedure controls size uniformly and has power approaching 1 against fixed alternatives. We conduct Monte Carlo Experiments to study the finite sample properties of our procedure. The third chapter studies the inference problem of bidders' risk attitudes in Independent Private Value (IPV) first-price auctions with multiplicative auction-level unobserved heterogeneity. Bidders are assumed to have Constant Relative Risk Aversion. Under the exclusion restriction that bidders randomly select themselves into auctions given the auction-level unobserved heterogeneity, bidders' CRRA coefficient is point-identified from bid data of auctions with at least two different number of active bidders. Our exclusion restriction is consistent with a variety of models with endogenous entry. Empirical application to USFS timber auctions shows that we will conclude that timber firms are risk averse if we ignoring the unobserved heterogeneity. But once we take the unobserved heterogeneity into account, risk neutrality is consistent with the data.
This paper shows point identification in first-price auction models with risk aversion and unobserved auction heterogeneity by exploiting multiple bids from each auction and variation in the number of bidders. The required exclusion restriction is shown to be consistent with a large class of entry models. If the exclusion restriction is violated, but weaker restrictions hold instead, the same identification strategy still yields valid bounds for the primitives. We propose a sieve maximum likelihood estimator. A series of Monte Carlo experiments illustrate that the estimator performs well in finite samples and that ignoring unobserved auction heterogeneity can lead to a significant bias in risk-aversion estimates. In an application to U.S. Forest Service timber auctions we find that the bidders are risk neutral, but we would reject risk neutrality without accounting for unobserved auction heterogeneity.
This paper shows point identification in first-price auction models with risk aversion and unobserved auction heterogeneity by exploiting multiple bids from each auction and variation in the number of bidders. The required exclusion restriction is shown to be consistent with a large class of entry models. If the exclusion restriction is violated, but weaker restrictions hold instead, the same identification strategy still yields valid bounds for the primitives. We propose a sieve maximum likelihood estimator. A series of Monte Carlo experiments illustrate that the estimator performs well in finite samples and that ignoring unobserved auction heterogeneity can lead to a significant bias in risk-aversion estimates. In an application to U.S. Forest Service timber auctions we find that the bidders are risk neutral, but we would reject risk neutrality without accounting for unobserved auction heterogeneity.
This dissertation consists of two chapters on nonparametrically identifying and estimating the sealed-bid k-double auction models between single buyer and single seller.Chapter 1: Nonparametric Identification and Estimation of k-Double Auctions Using Bid DataThis chapter studies the nonparametric identification and estimation of double auctions with one buyer and one seller. This model assumes that both bidders submit their own sealed bids, and the transaction price is determined by a weighted average between the submitted bids when the buyers offer is higher than the sellers ask. It captures the bargaining process between two parties. Working within this double auction model, we first establish the nonparametric identification of both the buyers and the sellers private value distributions in two bid data scenarios; from the ideal situation in which all bids are available, to a more realistic setting in which only the transacted bids are available. Specifically, we can identify both private value distributions when all of the bids are observed. However, we can only partially identify the private value distributions on the support with positive (conditional) probability of trade when only the transacted bids are available in the data. Second, we estimate double auctions with bargaining using a two-step procedure that incorporates bias correction. We then show that our value density estimator achieves the same uniform convergence rate as Guerre, Perrigne, and Vuong (2000) for one-sided auctions. Monte Carlo experiments show that, in finite samples, our estimation procedure works well on the whole support and significantly reduces the large bias of the standard estimator without bias correction in both interior and boundary regions.Chapter 2: Nonparametric Identification of k-Double Auctions Using Price DataThis chapter studies the model identification problem of k-double auctions between one buyer and one seller when the transaction price, rather than the traders bids, can be observed. Given that only the price data is available, I explore an identification strategy that utilizes the double auctions with extreme pricing weight (k=1 or 0) and exclusive covariates that shift only one traders value distribution to identify both the buyers and the sellers value distributions nonparametrically. First, as each exclusive covariate can take at least two values, the buyers and the sellers value distributions are partially identified from the price distribution for k=1 or k=0. The identified set is sharp and can be easily computed. I provide a set of sufficient conditions under which the traders value distributions are point identified. Second, when the exclusive covariates are continuous, it is shown that the buyers and the sellers value distributions will be uniquely determined by a partial differential equation that only depends on the price distribution, provided that the value distributions are known for at least one value of the exclusive covariates.
Handbook of Econometrics, Volume 7A, examines recent advances in foundational issues and "hot" topics within econometrics, such as inference for moment inequalities and estimation of high dimensional models. With its world-class editors and contributors, it succeeds in unifying leading studies of economic models, mathematical statistics and economic data. Our flourishing ability to address empirical problems in economics by using economic theory and statistical methods has driven the field of econometrics to unimaginable places. By designing methods of inference from data based on models of human choice behavior and social interactions, econometricians have created new subfields now sufficiently mature to require sophisticated literature summaries. Presents a broader and more comprehensive view of this expanding field than any other handbook Emphasizes the connection between econometrics and economics Highlights current topics for which no good summaries exist
This dissertation is a collection of three chapters on structural analysis of auctions. The first chapter studies nonparametric identification of the distribution of bidder valuations in auctions where valuations are independently and symmetrically distributed, the number of bidders follows a Poisson distribution, and the number is not known to the bidders. I consider both first and second-price sealed bid auctions. If the data set consists of all auctions, including auctions with no bids or only one bid, then I show that data on either the first or second highest bid is sufficient for the model to be identified. If the data set does not include auctions with no bids and only the highest bids are observed, then information on the number of bidders is also needed for identification. In the second chapter, I develop a method for identifying and estimating a dynamic model of auctions like eBay. The market is modeled as an infinite sequence of second-price, sealed bid auctions of a homogenous good. Bidders arrive randomly and, upon arrival, they enter a pool of potential bidders. The actual bidders in an auction are drawn randomly from the pool. Conditional on bidding, a bidder exits if she wins and returns to the pool if she loses. Then bidders in the pool exit with some probability each period. I define and solve for the oblivious equilibrium (Weintraub et al. (2008)). I prove the stochastic stability and the existence of an equilibrium. The equilibrium yields a closed form solution for the bid function in which bidders shade their bids by their continuation values. I demonstrate that the model is identified (modulo the discount factor) from the data of bidder identities and the second highest bid. Based on the identification result, an estimation procedure is developed. In the third chapter I apply the model to a data from a Japanese online auction website. The estimation results suggest that market dynamics are important. The estimate of the valuations obtained when each auction is treated independently is 23% smaller than the estimates obtained from the dynamic model.
