Regardless of its vocation, practically every economically motivated agent will have a component that is involved in negotiation over price and other attributes of information goods and services. Agents will use a wide variety of negotiation mechanisms, including auctions (of which there are at least several million enumerable types, many of which can be further parametrized [33]), one-on-one negotiation, exchanges, and posted pricing. Research on all of these types of agent-mediated negotiation is already proceeding, and a few simple examples of these technologies can already be found on the Internet today. Thus automated negotiation stands on its own as an important field of study, quite apart from its eventual importance in the information economy.
A simple present-day example of an agent that participates in auctions is the Bid-Click proxy agent available at amazon.com's auction site. A buyer interested in an auctioned item merely specifies a minimum and maximum bid, and the BidClick agent increments the bid automatically in response to challenging bids until either another bidder exceeds the agent's maximum or the agent wins the auction on behalf of the buyer. Kasbah [4] is an agent-mediated marketplace developed by researchers at MIT's Media Lab, in which human users delegate the responsibility for buying or selling physical goods to agents that engage in one-on-one negotiations with another agent. Park et al. [22] at the University of Michigan have studied the use of Markov modeling to create successful bidding strategies for agents participating in a continuous double auction; this can be regarded as a sort of exchange in that it supports trade among multiple buyers and sellers that freely enter and leave the market. Books.com is an online bookseller that employs posted pricing (i.e. it announces a nonnegotiable, take-it-or-leave-it price), but uses a price-comparison agent that dynamically undercuts the price offered by its chief competitors: Amazon, Barnes and Noble, and Borders.
Despite the popularity of the online auctions that one finds at Amazon, eBay, and hundreds of other sites, posted pricing is the dominant form of pricing on the Internet today, and certainly it is the most common retail pricing model -- the one with which we are most familiar in our everyday purchases of physical goods. Another feature of posted pricing that makes it attractive for agents is that price determination is quicker than it is for other forms of negotiation. A human buyer may tolerate waiting two weeks for the close of an auction of a Willie Mays baseball card at amazon.com. However, consider a CD-review-finder agent looking for reviews of the Karajan performances of the Nine Beethoven Symphonies on behalf of a human or another agent. Part of its strategy might be to issue specially-formulated queries to search engines in an effort to locate likely candidate reviews. The review-finder agent will not want to have a protracted negotiation with the Excite search engine over the price of 100 hits. It will need to settle the price and make the purchase within a fraction of a second. Multi-round auctions or negotiations will almost certainly be too slow in such a situation. Continuous double auctions may be viable, but posted pricing is guaranteed to be the fastest mechanism.
In this paper, we focus on what we call dynamic posted pricing -- that is, take-it-or-leave-it pricing in which the seller may change the price at any time. We particularly emphasize the collective interaction among multiple sellers that attempt to maximize profits by employing price-setting algorithms that are more--or less--sophisticated than what Books.com is using today. We use the adjective ``posted'' to emphasize that we are not considering online auctions, which constitute another important class of dynamic pricing mechanisms. Technically, posted pricing can be regarded as a simple form of auction, but there is an important distinction: in auctions the price changes occur during the course of each transaction, while in dynamic posted pricing the price changes occur across different transactions.