Designing Intelligent Agents for Auctions with Limited Information Feedback

This paper presents analytical, computational, and empirical analysis of strategies for intelligent bid formulations in online auctions.  We present results related to a weighted-average ascending price auction mechanism that is designed to provide opaque feedback information to the bidders and presents a challenge in formulating appropriate bids.  Using limited information provided by the mechanism, we design strategies for software agents to make bids intelligently.  In particular, we derive analytical results for the important characteristics of the auction, which allow estimation of the key parameters; we then use these theoretical results to design several bidding strategies.  We demonstrate the validity of designed strategies using a discrete event simulation model that resembles the mechanisms used in treasury bills auctions, business-to-consumer (B2C) auctions, and auctions for environmental emission allowances.  In addition, using the data generated by the simulation model we show that intelligent strategies can provide high probability of winning an auction without significant loss in surplus.