Manoj Kumar, Stuart Feldman. IBM Research Division, TJ Watson Research Center. November 1998.
http://www.ibm.com/iac/papers/auction_fp.pdf
Overview of different types of auctions: English (open-cry), sealed bid, Dutch auction (useful for airline tickets and other inventory based items), Yankee auction (winner pays what they bid), Vickrey auction (winner pays second highest bid).
Auctions built on the web need to be aware of the following unique issues: bidder collusion, bidder withdrawal capabilities, usability, bidding agents (agents must have similar response time as humans to not eliminate human bidding).
John Collins, Maksim Tsvetovat, Rashmi Sundareswara, Joshua van Tonder, Maria Gini. Department of CSE, University of Minnesota. 1999.
http://tiberius.cs.umn.edu/tech-reports/listing/
This article discusses an experiment on the behavior of buying and selling agents within a generalized multi-agent marketplace, called MAGNET. Similar to AuctionBot, MAGNET incorporates a buying agent's call-for-bids, the selling agents reply to bids and a buying agent acceptance of an appropriate bid. The paper presents a bid evaluation process for automated contracting that incorporates cost, task coverage, temporal feasibility, and risk estimation (not just price!). They showed that a buying agent can make tradeoffs between these parameters by adjusting the specifications of the task to be performed by the seller. They also show that the nature of these tradeoffs vary with the number of potential sellers in the market (the more sellers, the more chance of finding a flexible deal).
Jeffrey Kephart, Amy Greenwald. IBM Institute for Advanced Commerce, IBM Thomas J. Watson Research Center. April 1999.
http://www.research.ibm.com/infoecon/paps/html/ecsqaru99/shopbot.html
This paper proposes the following question: With the advent of shopbots, what is the expected impact of agent technology on the information economy?
Their conclusion: Nonlinear search costs (different search methods are more time consuming than others) can lead to complicated mixtures of buyer strategies and extensive searches. Assuming some buyers will not use search mechanisms at all will also result in the same conclusion.
Actions. Going, Going, Gone! A Survey of Auction Types.
Agorics, Inc. 1996.
http://www.agorics.com/new.html
Overview of Auction Types: English, Dutch, First-Price, Sealed-Bid, Vickrey, Double.
Pricing in agent economies using neural networks and muli-agent Q-learning
Gerald Terauro. IBM T. J. Watson Research Center. August 1999.
http://www.research.ibm.com/infoecon/paps/html/ijcai99_qnn/qnn.html
"This paper investigates how adaptive software agents may utilize reinforcement learning algorithms such as Q-learning to make economic decisions such as setting prices in a competitive marketplace. For a single adaptive agent facing fixed-strategy opponents, ordinary Q-learning is guaranteed to find the optimal policy. However, for a population of agents each trying to adapt in the presence of other adaptive agents, the problem becomes non-stationary and history dependent, and it is not known whether any global convergence will be obtained, and if so, whether such solutions will be optimal."
This paper concludes: "Simultaneous convergence to self-consistent optimal solutions is obtained in each model, at least for small values of the discount parameter. In some cases, such convergence is also found even at large discount parameters. Furthermore, the Q-derived policies increase profitability and damp out or eliminate cyclic price "wars" compared to simpler policies based on zero lookahed or short-term lookahead."