Structural estimation of auction models: A simulation-based approach

This dissertation examines the issues of structural estimation of auction models using simulation based inference. Chapter 2 considers a first price procurement auction model with a random reservation price within the independent private values paradigm. The estimation techniques built in this chapter are applied to different auction models throughout this dissertation. The major contributions of this chapter are in three aspects. First, a two-moment econometric model is developed. The two-moment model is necessary to estimate the variance of the underlying distribution. Second, a numerical method of solving the differential equation is developed to facilitate the simulation based estimation method. Third, an appropriately adjusted simulated generalized method of moments (SGMM) criterion function is used to estimate the two-moment econometric model.; Chapter 3 is an empirical study of asymmetric auctions. In an asymmetric environment, bidders draw their costs from heterogeneous distributions. Asymmetry among bidders often arises in procurement bidding, where cost structure is different among bidders. A symmetric model, in which the asymmetric parameter is set to zero, is also estimated. The criterion difference test rejects the symmetric model. This finding demonstrates that the asymmetric model outperforms the symmetric model. In the standard auction forms with symmetric bidders, the expected price under the first price and English auctions are the same. In an asymmetric environment, or in an auction with a random reservation price, this may not be the case. The empirical distribution of winning bids in an asymmetric first price auction is compared with the distribution in an asymmetric English auction. It is found that the English auction is slightly better than the first price auction for the auctioneer in terms of the average winning bid. This finding suggests that the choice of the first price auction in procurement bidding cannot be explained by costs.; Chapter 4 compares a collusive model with the non-cooperative model introduced in Chapter 2. An exogenous switching regression model is introduced to study the collusive behavior in the highway construction auction market. The model allows one to construct an empirical model in which bidders stochastically switch between collusive and non-cooperative behavior. The purpose of the empirical analysis is to see whether the data support the commonly believed notion that bid rigging is a severe problem. Surprisingly, the results clearly favor the non-cooperative model as evidenced by the estimate of the probability of colluding being equal to zero.