Behavior in Multi-Unit Demand Auctions: Experiments with Uniform Price and Dynamic Vickrey Auctions

Abstract

We experimentally investigate the sensitivity of bidders demanding multiple units of a homogeneous commodity to the demand reduction incentives inherent in uniform price auctions. There is substantial demand reduction in both sealed bid and ascending price clock auctions with feedback regarding rivals' drop-out prices. Although both auctions have the same normal form representation, bidding is much closer to equilibrium in the ascending price auctions. We explore the behavioral process underlying these differences along with dynamic Vickrey auctions designed to eliminate the inefficiencies resulting from demand reduction in the uniform price auctions.Key words: multi-unit demand auctions, uniform price auction, dynamic Vickrey auction, demand reduction, experiment.JEL classification: D44 (Auctions) * Research has been partially supported by grants from the Economics Division of the National Science Foundation. We thank Larry Ausubel and Peter Cramton for helpful discussions, Glen Archibald, James Peck, and seminar participants at Cornell University, University of Houston, University of Montreal, Ohio State University, University of Pittsburgh, Purdue University, and the University of Maryland conference on auctions, and three referees and the editor for helpful comments. Hui-Jung Chang, Juping Jin, and Scott Kinross provided valuable research assistance. We alone are responsible for any remaining errors. 2 1 Treasury bill auctions are often considered the canonical example of multi-unit demand auctions in which bidders have non-increasing demands. Policy debates regarding the optimal structure of Treasury bill auctions reveal a long history of confusion by a number of prominent economists regarding the incentive effects of uniform price auction rules (see Ausubel and Cramton, 1996). 3Spurred by the recent FCC spectrum auctions, theoretical research in multi-unit demand auctions reveals two distinctly different behavioral forces at work in auctions of this sort. In uniform price auctions, such as employed in Treasury bill auctions or in the recent FCC spectrum auctions, when bidders have non-increasing demand for homogeneous goods there is an incentive to reduce demand on some units in an effort to win other units at more favorable prices (see, for example, Ausubel andCramton, 1996 and Englebrecht-Wiggans andKahn, 1998). In contrast, when there are complementarities between items so that the value of a package of items exceeds the sum of its parts there are incentives for agents to bid above the value they place on any individual item (see, for example, Krishna and Rosenthal, 1996). Indeed, the recent FCC spectrum auctions have provided examples of both types of incentives: In the nationwide narrowband auction bidders appear to have had non-increasing demands, while in the broadband MTA auction there appear to have been complementarities between items (for analysis of the FCC auctions, see Cramton, 1995;McAfee and McMillan, 1996; Ausubel, Cramton, McAfee and McMillan, 1997).In this paper we experimentally...

Sequential Auctions with Capacity Constraints: An Experimental Investigation

“…9 Design choices vary sharply regarding whether to allow bids above value. In many experiments those bids are prevented [19][20][21], discouraged by design [22][23][24] or presented as "wrong" during instructions [10,25,26]. In the rare cases in which bids above value are allowed and not framed any way, some are observed although fewer than in our paper [27][28][29].…”

Anxiety and Learning in Dynamic and Static Clock Game Experiments

“…Their average RT in static trials was also less than 2 seconds. 6 Computerized agents have been used in many experimental studies on bargaining and auctions [13,14,15,16]. 7 This was indeed a small challenge for BM since mouse-clicking noise could serve as an auditory cue about selling time to other subjects.…”

“…The shortcoming of these types of models in games like this, with many strategies and relatively few strategies, is that reinforcement does not learn fast enough to match the empirical pace of human learning observed in experimental data. 15 For reinforcement learning to work, each possible strategy has to be sampled several times. In this game, there are 300 choices of times when a player can sell, and only 100 trials, so simple reinforcement is likely to learn too slowly without modification.…”

“…Fictive learning simply refers to adjustment of strategy likelihoods from model-based knowledge of the payoffs that would have been earned from a strategy 15 This limit of reinforcement learning has been known for many years. It produces especially poor results in many-person games in which only one person earns a reward (e.g., "market games" [21]; LUPI lottery games [22]; and auctions).…”

Demand reduction and preemptive bidding in multi-unit license auctions

“…They …nd evidence for demand reduction in the open ascending auction, which generally yields lower revenues than the sealed-bid format. Kagel and Levin (2001) consider environments without strategic uncertainty where a single human bidder competes against robot bidders in a sealed-bid uniform-price auction, an open ascending auction, and the Vickrey/Ausubel auction. Demand reduction occurs in both the sealed-bid uniform-price and the open ascending auction, but the level of demand reduction is more pronounced in the latter.…”