Shopbots -- software agents that automatically query multiple on-line vendors to gather information about prices and other attributes of consumer goods and services -- herald a future in which automated agents are an essential component of electronic commerce [2, 5, 11, 16]. Shopbots outperform and out-inform humans, providing extensive product coverage in just a few seconds, far more than a patient and determined human shopper could ever achieve, even after hours of manual search. Rapidly increasing in number and sophistication, shopbots are helping more and more buyers minimize expenditure and maximize satisfaction.
Since the launch of BargainFinder [12], a CD shopbot, on June 30, 1995, the range of products represented by shopbots has expanded dramatically. A shopbot available at shopper.com claims to compare 1,000,000 prices on 100,000 computer-oriented products. Another shopbot, DealPilot.com (formerly acses.com), gathers, collates and sorts prices and expected delivery times of books, CDs, and movies offered for sale on-line. One of the most popular shopbots, mysimon.com, compares office supplies, groceries, toys, apparel, and consumer electronics, just to name a few of the items on its product line. As the range of products covered by shopbots expands to include more complex products such as consumer electronics, the level of shopbot sophistication is rising accordingly. On August 16th, 1999, mysimon.com incorporated technology that, for products with multiple features such as digital cameras, uses a series of questions to elicit multi-attribute utilities from buyers, and then sorts products according to the buyer's specified utility. Also on that day, lycos.com licensed similar technology from frictionless.com.
Shopbots are clearly a boon to buyers who use them. Moreover, when shopbots become adopted by a sufficient portion of the buyer population, it seems likely that sellers will be compelled to decrease prices and improve quality, benefiting even those buyers who do not shop with bots. How the widespread utilization of shopbots might affect sellers, on the other hand, is not quite so apparent. Less established sellers may welcome shopbots as an opportunity to attract buyers who might not otherwise have access to information about them, but more established sellers may feel threatened. Some larger players have even been known to deliberately block automated agents from their web sites [4]. This practice seems to be waning, however; today, sellers like Amazon.com and BarnesandNoble.com tolerate queries from agents such as DealPilot.com on the grounds that buyers take brand name and image as well as price into account as they shop.
As more and more buyers are relying on shopbots to increase their awareness about products and prices, it is becoming advantageous for sellers to increase flexibility in their pricing strategies, perhaps by using pricebots -- automated agents that employ price-setting algorithms in an attempt to maximize profits. A primitive example of a pricebot is available at books.com, an on-line bookseller. When a prospective buyer expresses interest in a given book, books.com automatically queries Amazon.com, Borders.com, and BarnesandNoble.com to determine the price that is being offered at those sites. books.com then slightly undercuts the lowest of the three quoted prices, typically by 1% of the retail price. Such dynamic pricing on millions of titles is virtually impossible to achieve manually, yet can easily be implemented with a modest amount of programming.
As more and more sellers automate price-setting, pricebots are going to interact with one another, yielding unexpected price and profit dynamics. This paper reaches toward an understanding of strategic pricebot dynamics via analysis and simulation of four candidate price-setting algorithms that differ in their informational and computational needs: game-theoretic pricing (GT), myoptimal pricing (MY), derivative following (DF), and Q-learning (Q). Previously, we studied the price and profit dynamics that ensue when shopbot-assisted buyers interact with homogeneous collections of pricebots that utilize these algorithms [9, 10, 15]. In this work, we first establish that our previous results are not significantly altered when buyers' valuations are inhomogeneous rather than identical. Later, we examine the behavior that ensues when pricebots employing different strategies are pitted against one another.
This paper is organized as follows. The next section presents our model of an economy that consists of shopbots and pricebots. This model is analyzed from a game-theoretic point of view in Sec. 3. In Sec. 4, we discuss the price-setting strategies of interest: game-theoretic, myoptimal pricing, derivative following, and Q-learning. Secs. 5 and 6 describe simulations of homogeneous and heterogeneous collections of pricebots that implement these algorithms. Finally, Sec. 7 presents our conclusions and discusses ideas for future work.