Multi-unit online ascending price auctions: Mechanism design, evaluation, and calibration

Date of Completion

January 2002


Business Administration, Marketing|Economics, Commerce-Business




This dissertation comprises of three separate but related studies. The first study presents an analytical model that characterizes the revenue generation process for a popular kind of online auction, namely Yankee auctions. Such auctions sell multiple identical units of a good to multiple buyers using an ascending and open auction mechanism, which has its roots in the English auction, yet is significantly different. We put forward a portfolio of tools, varying in their level of abstraction and information intensity requirements, which help the auctioneers to maximize their revenues. The study sheds new light on how online auctions can be used to construct empirical demand curves for the auctioned goods. The methodologies used to validate the analytical model range from empirical analysis to simulation. A key contribution is the design of a partitioning scheme of the discrete valuation space of the bidders such that equilibrium points with higher revenue structures become feasible. Our analysis indicates that the auctioneers are, most of the time, far away from the optimal choice of key control factors such as the bid increment, resulting in substantial losses in a market with already tight margins. The second study uses available information to make inferences about bidder valuations. Using such information, we derive a priori estimates of the bidders valuations. We present an analytical model that predicts a consumer's valuation for a product, based on the joint consideration of the bidding strategy pursued and the bid values revealed, both of which are observable on the Internet. Subsequently, using an automated agent, we test the valuation prediction model against thousands of bids made on hundreds of real online auctions from Samsclub.com. Our data analysis is able to accurately “type” the bidding strategy based on observable variables, and is successful at predicting the bidder's eventual valuation. The third study is aimed at increasing the efficiency of multi unit ascending price auctions, armed with the ability to predict the bidder's valuation, we use such a priori valuation information to increase the efficiency of the auctions through realtime calibration. We develop an analytical model for the auctioneer's revenue and derive optimal dynamic bid increments. We compare the auction outcomes based on the analytically prescribed bid increments and heuristics that are motivated by a deterministic effort to order the bidders' sequence as they approach the final bidding round. Our empirical analysis indicates that auctioneers can derive the same or more expected revenue with fewer bidding cycles. ^