E-commerce Auction Agents and Online-auction

Dawn G. Gregg and Steven Walczak

DAWN G. GREGG AND STEVEN WALCZAK

INTRODUCTION The Web has dramatically changed how people buy and sell goods. In recent years new types of electronic marketplaces have been created to leverage information technology to create more efficient markets (Bakos 1998). Web usage and commerce is increasing dramatically, with an estimated $6.8 trillion in online retail sales by the year 2003 (Forrester Research 2001). One of the most successful types of electronic marketplaces has been the online auction. Online auctions allow businesses and consumers to easily buy or sell anything to anyone anywhere in the world. One real-world example of a successful online auction is eBay.com. eBay is the largest online-auction site currently operating on the Web and has nearly 7 million items available for sale on any given day. The success of online-auctions has given buyers access to greater product diversity with potentially lower prices. Conducting business online reduces transaction costs by eliminating the time and place aspects of offline markets (Bichler et al. 2002). It has also allowed sellers to reach a greater number of potential buyers. However, buyers must incur higher search costs to locate desired products within larger and larger numbers of products and

sellers face greater competition from other sellers in order to effectively reach the potential buyers (Bichler et al. 2002; Hahn 2001). Buyers and sellers would both benefit if the process of determining appropriate prices, where to participate, and how best to bid could be automated. Software agents provide a solution to the time and information demands of online-auction participants (Ye et al. 2001). Broadly defined, a software agent is a program that acts on behalf of a user to find and filter information, automate complex tasks, monitor events and procedures or negotiate for services (Maes 1994). Research is needed to examine the impact of software agents that can automatically collect large volumes of auction data and make recommendations based on that data. Currently online-auction sites provide software agents that can be used by both buyers and sellers (e.g. search agents and proxy bidding agents). In this article, the impact of Web agents operating in online B2C and C2C auctions is examined and likely agent developments from current and future research are explored. The remainder of this paper is organized as follows: the next section provides background information pertaining to B2C and C2C online auctions and agents operating in the B2C and C2C

A

b

s

t

r

a

c

t

Online-auctions are one of the most successful types of electronic markets. They bring together buyers and sellers on a massive scale. However, using an electronic medium for conducting auctions has fundamental differences from traditional English-style auctions. One difference is the availability of software agents that can facilitate many aspects of online-auction participation. The addition of software agents into onlineauctions is already having an impact on the dynamics of online-auctions. This study examines existing agent technologies with regard to their effect on online-auctions. In addition, future directions for research related to online-auction agents and the possible benefits of these agents are also discussed. Keywords: e-commerce, online auctions, intelligent agents DOI: 10.1080/1019678032000092237

E-commerce Auction Agents and Online-auction Dynamics

RESEARCH Agents E-commerce Auction

A

u

t

h

o

r

s

Copyright 䊚 2003 Electronic Markets Volume 13 (3): 242–250. www.electronicmarkets.org

242

Dawn G. Gregg ([email protected]) is an Assistant Professor at the University of Colorado, Denver. Her current research focus is on the use of artificial intelligent agents to organize and maintain Web-based content so that it can be better used to meet business needs. Steven Walczak ([email protected]) is an Associate Professor of Information Systems in the Business School at the University of Colorado at Denver. His current research interests are in applied artificial intelligence systems including agents and neural networks and 242 knowledge management systems.

Electronic Markets Vol. 13 No 3

online-auction domains, the third section analyses the effects of auction agents on online-auction dynamics, the penultimate section discusses the effect of future agent research and development on online B2C and C2C auctions.

BACKGROUND Online-auctions are an increasingly popular mechanism for exchanging goods and services via the Web (Ye et al. 2001). An online auction is a Web application that acts as an intermediary between sellers and buyers. In online auctions, resource allocation and prices are determined with an explicit set of rules based on bids from market participants (Bichler 2001; Hahn 2001). Online auctions can be categorized into three main dimensions: business-to-consumer (B2C); consumerto-consumer (C2C); and business-to-business (B2B). While all three types of online auctions can use similar auction mechanisms, the purchasing decision making processes of consumers and businesses differ substantially. The focus of this paper is on how software agents can benefit individual consumers participating in B2C or C2C online auctions. B2C or C2C online-auctions allow either single items or multiple items to be auctioned at one time. Singles item auctions work very similarly to traditional English-style auctions. Bidders bid against each other and the winning bidder is the one that bids the highest price first. Multiple item auctions offer multiple identical items for sale and bidders bid for the price and quantity they are interested in purchasing. The price winning bidders pay depends on the type of multiple-item-auction being conducted. In ‘Yankee auctions’ each winning bidder pays the final amount they bid for the item(s). In ‘Dutch or Vickery auctions’ all winning bidders pay the amount bid by the lowest successful winning bidder (Bapna et al. 2001a). Most online auctions are started with a low minimum bid price which is used to attract Web traffic (e.g., during a 1-week period in February 2002, 13.2% of 7,263 auctions at eBay had a minimum bid price of one dollar or less). Some online auctions also use a reserve price to specify the lowest price for which a seller is willing to sell an item. In order to win the auction, a bidder must meet or exceed the reserve price and have the highest bid (Teich, Wallenius, Wallenius and Zaitsev 1999). In addition, all online auctions have a bid increment that defines the minimum amount that can be bid next (Anonymous 2002). B2C and C2C auction durations typically range from a few days to three weeks (Hahn 2001). Some auction sites use an ascending bid protocol with a fixed end time (e.g. eBay.com and Yahoo.com). At these 243 sites the practice of last minute bidding is prevalent

243

(Matsubara 2001; Teich, Wallenius, Wallenius and Zaitsev 1999). Other auction sites more closely resemble traditional auctions and attempt to limit last minute bidding by providing a ‘Going, Going, Gone’ period. On these sites (e.g. Amazon.com and Ubid.com) the auction closing time is automatically extended by 10 minutes if a bid is received within 10 minutes of the current closing time. One of the inherent risks in participating in online auctions is the virtual anonymity of the transactions. In order to foster trust between unknown buyers and sellers, most auction sites provide ratings that reflect both the number of online auction transactions the participant has completed and amount of positive and negative feedback the participant has received. It has been suggested that these reputation systems help to foster better behaviour in both buyers and sellers because they seek to enhance their online-auction reputation (Resnick, et al. 2000). Conducting trade on the Web enables vendors to reach more consumers and at significantly lower cost (Deveaux et al. 2001). The use of the Web to establish virtual/online-auctions permits auctioneers to establish a necessary and critical mass of bidders as individuals may participate from any location around the world (Guttman et al. 1999). This will in turn optimize the auction outcomes for the seller. A successful online auction shows positive network effects; that is, the more traffic it has, the more desirable it is to new participants (Feldman 2000). In May 2001, research from Nielsen/NetRatings and Harris Interactive, indicated US auction websites generated $556 million in revenue. The revenue share for top US B2C & C2C auction sites is shown in Table 1. EBay, as the first-mover in the online-auction arena, has gained an advantage over other onlineauction sites. In addition to being a first-mover, EBay’s advantage also comes from developing a business strategy that integrates the asynchronous and global reach of the Internet, low overhead of a virtual marketplace, vertical integration of support markets, and a focus on customer satisfaction that promotes customer return (Kambil and Van Heck 2002). It has considerable market power, and the greatest revenue share with 64.3% of all online-auction revenue. In addition, its higher traffic has resulted in a conversion rate of 22.5%, more than double that of its nearest competitor (Anonymous 2001). Previous e-commerce agent research has utilized the consumer buyer behaviour model (Guttman et al. 1999), in which consumers are modelled as decision makers. Online-auction buyers must determine what they want to buy, from whom they want to buy, and how much they are willing to spend. Unfortunately, the easy access to auctions (via the Web) also increases the quantity of uninformed consumers (little auction


244

E-commerce Auction Agents

Table 1. Top US auction sites, ranked by revenue share Auction Site

Number of items for sale (1 day Feb 2002)

Revenue share† (May 2001) (%)

Conversion rate‡ (May 2001) (%)

EBay.com UBid.com Egghead.com* (Onsale.com) Yahoo! Auctions Amazon Auctions

6,957,080 1,133,661 * 455,570 936,231

64.3 14.7 4.0 2.4 2.0

22.5 11.0 8.0 4.4 6.5

* Since May 2001 Egghead.com contracted out website management to Amazon.com and no longer hosts independent auctions. † Revenue share is the percentage of the online auction market space captured by the corresponding site, where share is determined by auction close dollars. ‡ Conversion rate is the percentage of auction items listed that result in a sale (item receives a bid and bidder pays for the item). Source: eMarketer, online at: http://www.emarketer.com/estatnews/estats/ecommerce_b2c/20010706_nn_harris.html, July 6, 2001.

experience). These uninformed consumers are poor auction decision makers and have been found to pay an average of 18.5 percent more than experienced bidders in one research study (Bapna et al. 2001b). The use of agents to search for products and conduct automatic bidding helps to reduce time costs and certain other types of frictional costs (Ye et al. 2001). They can also serve to level the playing field between experienced and newer auction participants by providing information that can educate inexperienced buyers and sellers.

CURRENT ONLINE AUCTION AGENTS Online auctions are not just electronic markets that connect buyers and sellers; they also are services that can be used to improve the purchasing/sales efficiency and decision making of auction participants (Kambil and Van Heck 2002). Much literature has been written on the use of ‘comparison-shopping’ agents to automate the search for price and product information across multiple online merchants simultaneously (Clark 2000; Crowston 1996; Doorenbos et al. 1997). Online auctions provide potential buyers and sellers with a unique ability to use similar software agents to improve their purchase and sales outcomes. This section examines the availability and use of agents on several B2C and C2C online-auction sites to gain a better understanding of how agents impact online-auction processes. The first part of this study involved gathering data on auction site activity and the availability of different software agents at various online-auction sites. This data was used to determine what agent-based services are currently being provided at different online-auction sites. The various types of agent-based services are discussed in the remainder of this section.

Information retrieval agents There are currently several different types of services offered at the major online-auction sites. The most basic service provided by the auction sites is the search service. The search service is so common that it is difficult to classify it as an agent-based service at all. However, it is essential to the operation of onlineauction sites. For example, on a single day eBay had nearly 7 million auctions running at one time grouped into 9,522 different product categories. There is little differentiation between the information retrieval services currently provided by the different auction sites, however, there is room for improvement to these services that would substantially improve the efficiency of searches on different auction sites. Such improvements would include the ability to exclude terms, choosing inclusive or any search types, and specifying terms from particular fields such as location. Information retrieval agents can also be used to improve the decision making of auction participants by enabling them to determine appropriate prices (valuations) for products. Four of the seven online-auction sites provide the ability to search recently closed auctions. This feature enables online-auction users to determine what recent closing prices have been on items similar to ones they are considering purchasing (or selling). If utilized, this capability could help to level the playing field between new and more experienced auction participants and lessen the likelihood that bidders will pay more for an item than its value (known as the ‘winner’s curse’) (Mehta and Lee 1999).

Bidding agents Two different types of bidding agents are currently being used on most online-auction sites. The first type 244


245

of bidding agent is a ‘proxy-bidding agent’. Proxy bidding means that a bidder can submit a maximum bid amount they are willing to pay and the proxy-bidding agent will bid in their absence. Proxy bidding agents place their bids based on the previous high bid amount, the minimum bid increment and on any reserve price. In order to assess the impact of proxy bidding agents, the ratio of proxy bids placed to the total number of bids placed was examined. To do so, in-depth data from the eBay active online-auction market was collected using an information retrieval agent. The information retrieval agent gathered publicly available data on individual auctions including the item description and the bid history. The data was collected for all online-auctions ending during a single week (12 February 2002 to 18 February 2002) for 7 pre-specified product categories. The target categories were: Intel Celeron Desktop Computers, Fax Machines, DVD Drives, US Airmail Stamps, New Age Compact Discs, Business Database Software and Teeny Beanie Babies. Information on a total of 7,290 auction items gathered from the eBay auction site is summarized in Table 2. A total of 21,910 bids were placed on 4,262 different items across the seven product categories. Proxy bidding agents placed 34.3 % of the bids recorded on the 4,262 items. This indicates that proxy agent bidding plays a dramatic role in the outcome of most online-auctions. The data also indicates that the auction categories with a higher percentage of items being sold tend to have a higher percentage of bidding done by proxy agents. For example, the data in Table 3 show that 66.07% of DVD drives that were placed on auction at eBay were sold and that 55.22% of the bids placed on these items were placed by proxy bidding agents. Conversely, only 30.66% of the Teeny Beanie Babies that were placed on auction at eBay were sold and proxy-bidding agents placed only 14.66% of the bids placed in this product category. However, the correlation is weak and further research will be necessary to determine if this relationship exists across other categories.

Watch agents Other agents are used to keep auction participants aware of what is going on in auctions in which they are interested. Several auction sites provide “Watch agents” which track auctions the participant might be interested in but at which they have not yet decided to place a bid. The three auction sites eBay.com, Ubid.com and Cityauction.com, provide auction notification agents that will search new auction postings and notify potential buyers when specific items they are interested in are posted on their online auctions. These sites also use another type of watch agent to notify bidders if they have been outbid or if they have won an auction. Yahoo.com extends this practice to allow notification of auction participants if an auction they are participating in has closed or was cancelled, if a previously losing bid was reinstated due to another bidders bid being retracted or if a feedback rating was posted for the participant. The ability to watch interesting auctions or to notify auction participants about important auction events can serve to stimulate more interest in the auction than would be possible without the use of such agents.

Seller agents Software agents can benefit both sellers and buyers. For example, Amazon.com, Ubid.com and AuctionAddict. com use agents to track bidding at their auctions. These agents automatically extend an online-auction if a bid occurs during some predetermined interval prior to the auction closing time (10 minutes for Amazon and Ubid, 24 hours for AuctionAddict). This more accurately simulates traditional English-style auctions that are kept open during active bidding. In addition, extending auctions can reduce ‘slamming’. Slamming is the practice of placing a last second bid so that other bidders will not have the opportunity to place another bid before the auction closes. Two online-auction

Table 2. eBay proxy bidding data summary

245

Celeron computer Fax machine DVD drive US airmail stamp New Age CD Database program Teeny Beanie Baby Total

Number of auctions closed

% items receiving bids

Number of bids placed

Percentage of bids placed by proxy

715 792 1176 1098 1874 425 1210 7290

73.15 69.44 66.07 63.02 62.01 44.00 30.66 58.46

4492 3223 6302 2675 3390 1105 723 21910

34.17 25.38 55.22 23.03 21.27 21.90 14.66 34.31


246


Table 3. Prevalence of last minute bidding at eBay

Celeron computer Fax machine DVD drive US airmail stamp New Age CD Database program Teeny Beanie Baby Total

Number of auctions closed

Percentage of items receiving bids

Percentage of items receiving bids bid on in the last 10 minutes

715 792 1176 1098 1874 425 1210 7290

73.15 69.44 66.07 63.02 62.01 44.00 30.66 58.46

96.94 73.45 87.00 50.00 30.38 89.30 35.58 60.65

characteristics enable slamming. First, the hard auction close times make it difficult to respond fast enough to a late bid, and second it is more difficult to judge the interest and availability of other bidders in the online environment. Table 3 shows how prevalent the practice of slamming is on eBay (an auction site that does not extend its auctions if there are last minute bids). The data was collected for all eBay auctions ending during a single week (12 February 2002 to 18 February 2002) for 7 pre-specified product categories. The data indicates that over 60% of auctions that receive any bids, receive bids within the last 10 minutes of the auction. The data also indicates that auction categories that have a greater percentage of their items receive bids; also tend to have a higher incidence of slamming. In the Celeron Computer category, 96.94% of the auctions that received bids, received at least one bid in the last 10 minutes of the auction! A second type of agent that is beneficial to auction sellers is one that allows them to control the types of bidders they will allow to bid at their auctions. For example, some online-auction bidders demonstrate a pattern of unacceptable bidding or fail to follow through on auction purchases. Some auction sites provide the

ability for sellers to ‘blacklist’ specific bidders (e.g. eBay and Yahoo in Table 4). Under this system an agent watches the seller’s auctions and automatically bounces bidders that appear on the sellers blacklist. While this system does not yet allow sellers to block all bidders with a specific feedback profile, it does allow them some mechanism for blocking undesirable bidders. Online auction sites currently provide differing agent-based services. The various services discussed above are summarized in Table 4 for seven different B2C and C2C online-auction sites.

Third-party agents In addition to agents provided by specific online-auction sites, third-party agents have been developed to provide information retrieval and watch services. Agents found at ‘BidXS.com’ and ‘McFind.com’ may be used to compare offerings across multiple online-auction sites. McFind.com allows simultaneous category browsing or searching of both Amazon.com and Yahoo.com. BidXS.com provides a more complete buyer’s agent that allows keyword searches, searches by category and

Table 4. Agent services currently provided by selected online-auction sites Auction site

Search

Search closed

Activity manager

Proxy bid

Watch

Bidder blacklist

Auction agent*

Notification**

Auction extension

EBay.com Ubid.com Amazon.com Yahoo.com

yes yes yes yes

yes – – yes

yes yes yes yes

yes yes yes yes

yes yes – yes

yes – – yes

yes yes – –

– 10 min 10 min no bids

Auctionaddict.com Cityauction.com Intershopzone.com

yes yes yes

yes yes –

yes yes yes

yes yes yes

– yes yes

– – –

– yes –

ob, win, cls ob, win ob, win ob, win, can, cls, resb, rat ob, win ob, win ob, win

* An auction agent will wait for a specific item to be listed & notify buyer. ** outbid (ob), win, cancellation (can), auction close (cls), bid resubmit (resb), rating posting (rat).

24 hrs – –

246


searches by price. In addition, it can simultaneously conduct searches on 118 different auction sites. BidXS provides other agent-based services to auction buyers. BidXS has partnered with StrongNumbers to make past price trends available for online-auction items. The process for obtaining past-price histories is simpler on BidXS than that provided by the individual auction sites examined. On BidXS, users need only click on a price tag icon that appears on the bottom of every search results page and the recent price history for that item is displayed. This allows potential buyers (and sellers) to determine what other people paid for the same item at previous auctions. BidXS also provides an auction agent that has the ability to look at multiple auction sites to determine when a desired item comes up for auction. The agent can then alert participants by email when it does. This type of agent is currently only provided by ebay.com, Ubid.com and Cityauction.com. The advantage of third-party auction agents is that they allow users to locate items and compare prices and terms across multiple online-auction sites. One hurdle to developing and implementing more sophisticated third-party auction agents is that some auction sites have placed restrictions on the use of agents to gather information on their websites. For example, both eBay.com and Amazon.com have agent exclusion policies. Anybody interested in developing auction agents for use on these sites must get written permission before using the agents to gather any data.

FUTURE DIRECTIONS FOR AUCTION AGENTS For online auctions to remain competitive, they must continue to improve the services they provide. Onlineauction sites that provide additional services to end-users and allow them to increase their efficiency and decisionmaking are the ones that will be the most successful in the future (Kambil and Van Heck 2002). In this section, directions for further improvements in agent technology and capabilities are explored and the corresponding effect on online-auction dynamics is inferred.

247

all major auction sites provide only rudimentary search services (Table 4). The search capabilities of all onlineauction sites could be improved substantially. Often auction searches retrieve hundreds of products – many of which an auction participant is not interested in. Improved search agents could allow participants to check the products they are and are not interested in and then a more accurate search could be constructed automatically – greatly improving the efficiency of auction searches. When evaluating past auctions, the maximum, minimum and average selling price for items could also be calculated automatically. In addition, the starting prices of items that did not sell could also be tabulated. This would provide valuable information to aid auction decision-making. For example, assume a consumer is interested in purchasing a digital camera at auction and finds a camera with a starting price of $780 plus a $20 shipping fee. If an improved information retrieval agent indicates that similarly featured digital cameras have closed at prices as low as $550, the consumer would know that the seller has set too high of a price. Similarly sellers can use auction history information to determine optimal reserve pricing to maximize the opportunity for sale and still achieve the desired profit. Auction shopping software agents could also access multiple online-auction sites to determine if a desired item is up for auction (e.g BidXS and StrongNumbers.com). Combining the auction-shopping agent with an information retrieval agent would allow the retrieval of additional auction data like the time left in the auction until it closes and relative bidder activity. A buyer would then be able to choose from among multiple auctions at multiple sites and evaluate which auction would most likely result in a winning bid within the buyer’s budget. Another addition to the auction shopping agent would be to retrieve the online retail price from the manufacturer. This information would allow the consumer to know if they should drop out of an auction due to uninformed bidders raising the price above standard retail.

Advanced bidding agents Advanced information retrieval agents Information retrieval agents have been developed for other domains, but are still in their infancy for the online-auction domain. The purpose of information retrieval agents is to make the seller or consumer more informed about the auction process and also the pricing of goods. Agents would automatically retrieve bidding and auction close information. With detailed information regarding the historical performance of specific auction items, buyers and sellers will be better informed regarding 247 an optimal price and even a bidding strategy. Currently,

A further extension of the auction shopping agent is to enable consumers to engage in simultaneous auctions for the same item on multiple online-auction sites. Currently, bidding agents work on a single site for a single item. Building agents that can bid for the same item on multiple simultaneous auctions is significantly more complex (Greenwald and Stone 2001), since the consumer may desire to win only a single item auction. Benyoucef et al. (2001) have demonstrated a model for negotiating for multiple items from multiple retailers in the B2C marketplace. Modifications to this B2C negotiation model are required for efficient handling

248

of a single item bid on multiple online-auction sites. The agent would monitor selected auction sites and maintain active bids so that only one bid at a time was current and would be the minimum cost across all auctions. Another problem with current online-auctions is that buyers and sellers are matched with regard to price alone (Teich, Wallenius, and Wallenius 1999; Teich et al. 2001). A buyer or seller may have other criteria. Agents are not limited to only bidding or looking for items with regard to price, but instead may be programmed with multiple search/bidding criteria. Criteria of interest to a buyer in addition to price may include: seller rating, cost of shipping, willingness to ship outside of the local area of the seller, shipping method, time left in the auction, quality or evaluation of the item. All of these pieces of information are available at most online-auction sites. The new multi-criteria bidding agent would then maximize the buyer’s return with regard to all specified criteria, perhaps paying a little more to get an item sooner or from a seller with a better reputation (rating). Optimizing return for buyers and sellers is the goal of auction agents. All the auction sites evaluated in Table 4 permit proxy or automatic bidding through simple single-item bidding agents. Current bidding agents can artificially increase the current bid for an item by competing against each other. Future bidding agents could incorporate more sophisticated bidding strategies to minimize costs and maximize returns. Deveaux et al. (2001) examined three different behaviours (conceding, boulware, and imitative) in bidding agents and found that different behaviours produce optimal results under different situations, implying the need for an adaptive approach. Elements needed to develop a bidding strategy for an agent are the rules of the auction, the valuation of the item by the bidder, and an estimate of other bidder’s strategies and valuations (Reynolds 1996b). Embedding machine learning into advanced bidding agents would allow the agents to modify bidding strategies and valuations on future auctions based on the results of current and past auctions. Other re-estimates of the current valuation and consequently the maximum bid may be performed if the identity of another bidder (from their email or account name) indicates an individual with special knowledge of the object, such as an art critic bidding on a particular piece of art (Reynolds 1996a).

Advanced watch agents Watch agents notify a user when a particular item comes up for auction at an online-auction site. Current watch agents (Table 4) are limited to ‘watching’ for specific items detailed by the user. However, preference grouping agents are already employed at various



e-commerce sites (e.g., Amazon.com) to suggest retail (B2C) items for purchase based on the shopper’s current and past purchases or inquiries. Preference-learning agents should be incorporated into auction decision support systems (Hess et al. 2000), and would advance the state of auction watch agents. These preferencebased watch agents would monitor auction activity of a user and may additionally monitor, at the user’s option, other activities such as web-browsing to gain a better perspective on the interests of the user. The watch agents would then be able to notify a user whenever any item of potential interest comes up for auction, not just those items explicitly specified by the user.

Other advanced agents Agents may also be used to assist auction sites. A technique that is sometimes employed to raise the interest and current bid level is the use of a shill who makes bids on merchandise to bid up the price and keep the action alive, but with the foreknowledge that the shill will not have to purchase the auctioned item (Reynolds 1996a). At auction houses, bidders may watch for items that are set aside after close to indicate the use of a shill, but online a seller could employ one or more shills without any visible signs. To minimize this practice, agents can be used to determine if a seller posts an item for auction that is the same as one they recently sold (e.g. in the last two-weeks). If the seller repeats this pattern this may be an indication to the online-auction site that the seller is employing a shill. Finally, an ongoing challenge for Web agents is their ability to communicate with other agents, systems, and users (Devaux et al. 2001; Hess et al. 2000). If multiauction bidding agents are employed and if an item has M agents and individuals bidding for a specific item, but there are N of these items for sale on various auction sites (where M ≤ N), then from an economic perspective supply exceeds demand and all agents and individuals should be able to obtain the item. Agents that detect only minimal bidding activity on a group of the same item across multiple online-auction sites should then open a dialogue with the other agents and individual users to negotiate an equitable price and allow all of the M agents and individual bidders to acquire one of the items for the negotiated price. Unfortunately, although several agent communication protocols have been developed (e.g., KQML (Finin et al. 1997)) they are not used and consequently agents developed under different platforms cannot communicate with each other. Developing inter-agent communication is imperative to future agent technology development. Communication between agents and users is the other aspect of the agent communication problem. Natural language processing and production can be embedded within agents to enable users to specify 248


search/watch criteria and bidding strategies using natural language. The development of better natural language parsing algorithms will facilitate the use of auction agents by all users, regardless of background.

SUMMARY The software agents already in use on auction websites have the potential to alter dramatically the dynamics of online-auctions. For example, proxy bidding agents place bids for potential buyers up to a maximum predetermined bid amount. However, these agents currently do not react to changes in the bidding environment as a real bidder would. If another bidder were repeatedly bidding against the proxy agent, it would immediately increase its bid to the next possible minimum bid amount. A real bidder, on the other hand, might wait to raise their bid and perhaps avoid getting into a bidding war with the other bidder. Agents that can be used to provide price histories and other auction information help to educate auction participants. Buyers will know what a reasonable price for a given item is and can avoid paying too much. Sellers can use the same information to determine appropriate minimum and reserve prices for their auctions. The availability of this information can help to even the playing field between experienced and newer auction players and reduce the occurrence of ‘winner’s curse’, thus opening online auctions as a viable and satisfactory commerce alternative for a much larger consumer population. Agents in general have great potential for helping buyers and sellers to set-up auctions, find specific items, and place appropriate bids. Future agents will be able to automatically find an item by description or category across multiple auction sites and actively monitor bids that maximize a user’s value (e.g. including cost, shipping, and feedback ratings). However, introducing agents into online-auctions will fundamentally change the way these auctions operate and the outcomes for both buyers and sellers. As online-auction buyers and sellers increase their use of software agents to automate auction decisions, the outcome for both buyers and sellers will be improved. Agents can be used to optimally match buyers and sellers for specific products. With sufficient information an optimal price can be determined to maximize the return for all auction participants.

References Anonymous (2001) ‘US Auction Sites: Where the B2C Action is Online’, online at: http:// www.emarketer.com/estatnews/estats/ecommerce_b2c/ 20010706_nn_harris.html [accessed 10 December 2001]. 249

249 Anonymous (2002) ‘EBay Help Basics’, online at: http:// pages.ebay.com/help/basics/ [accessed 5 February 2002]. Bakos, Y. (1998) ‘The Emerging Role of Electronic Marketplaces on the Web’, Communications of the ACM 41(9), September, 35–42. Bapna, R., Goes, P. and Gupta, A. (2001a) ‘Insights and Analyses of Online Auctions’, Communications of the ACM 44(11), November, 42–50. Bapna, R., Goes, P. and Gupta, A. (2001b) ‘Comparative Analysis of Multi-item Online Auctions: Evidence from the Laboratory’, Decision Support Systems 32(2), 135–53. Benyoucef, M., Alj, H., Vézeau, M. and Keller, R. F. (2001) ‘Combined Negotiations in E-Commerce: Concepts and Architecture’, Electronic Commerce Research 1(3), 277–99. Bichler, M. (2001) The Future of e-markets: Multidimensional Market Mechanisms, Cambridge: Cambridge University Press. Bichler, M., Kalagnanam, J., Katircioglu, K., King, A. J., Lawrence, R. D., Lee, H. S., Lin, G. Y. and Lu, Y. (2002) ‘Application of Flexible Pricing in Business-tobusiness Electronic Commerce’, IBM Systems Journal 41(2), 287–302. Clark, D. (2000) ‘Shopbots Become Agents for Business Change’, Computer, February, 18–21. Crowston, K. (1996) ‘Market Enabling Web Agents’, in J. I. DeGross, S. Jarvenpaa and A. Srinivasan (eds), Proceedings of the Seventeenth International Conference on Information Systems, Cleveland, OH, December, 381–90. Deveaux, L., Paraschiv, C. and Latourrette M. (2001) ‘Bargaining on an Web Agent-based Market: Behavioral vs. Optimizing Agents’, Electronic Commerce Research 1(4), 371–401. Doorenbos, R. B., Etzioni, O. and Weld, D. S. (1997) ‘A Scalable Comparison-shopping Agent for the WorldWide Web’, Proceedings of the First International Conference on Autonomous Agents, 5–8 February 1997, Marina del Rey, CA, 39–48. Feldman, S. (2000) ‘Electronic Marketplaces’, IEEE Web Computing, July/August, 93–5. Finin, T., Labrou, Y. and Mayfield, J. (1997) ‘KQML as an Agent Communication Language’, in J. M. Bradshaw (ed.), Software Agents, Menlo Park, CA: AAAI Press. Forrester Research (2001) ‘Web Commerce’, online at: http://www.forrester.com/ER/Press/ForrFind/ 0,1768,0,00.html [accessed 5 February 2002]. Greenwald, A. and Stone, P. (2001) ‘Autonomous Bidding Agents in the Trading Agent Competition’, IEEE Web Computing 5(2), 52–60. Guttman, R., Moukas, A. and Maes, P. (1999) ‘Agents as Mediators in Electronic Commerce’, in M. Klusch (ed.), Intelligent Information Agents, Berlin: Springer. Hahn, J. (2001) ‘The Dynamics of Mass Online Marketplaces: A Case Study of an Online-auction’, Proceedings of the SIGCHI Conference on Human Factors in Computing Systems, Seattle, WA, 317–24.

250 Hess, T. J., Rees, L. P. and Rakes, T. R. (2000) ‘Using Autonomous Software Agents to Create the Next Generation of Decision Support Systems’, Decision Sciences 31(1), 1–31. Kambil, A. and van Heck, E. (2002) Making Markets: How Firms Can Design and Profit from Online Auctions and Exchanges, Boston, MA, Harvard Business School Press. Maes, P. (1994) ‘Agents that Reduce Work and Information Overload’, Communications of the ACM 37(7), July, 32–40. Matsubara, S. (2001) ‘Accelerating Information Revelation in Ascending-bid Auctions: Avoiding Last Minute Bidding’, Proceedings of the 3rd ACM Conference on Electronic Commerce, Tampa, FL, 29–37. Mehta, K. and Lee, B. (1999) ‘An Empirical Evidence of Winner’s Curse in Electronic Auctions’, in Proceeding of the 20th International Conference on Information Systems, Charlotte, NC, 465–71. Resnick, P., Zeckhauser, R., Friedman, E. and Kuwabara, K. (2000) ‘Reputation Systems’, Communications of the ACM 43(12), December, 45–8.



Reynolds, K. (1996a) ‘Collusions and Tricks’, Agorics, Inc, online at: http://www.agorics.com/Library/Auctions/ auction11.html [accessed 21 January 2002]. Reynolds, K. (1996b) ‘Auction Strategies’, Agorics, Inc, online at: http://www.agorics.com/Library/Auctions/ auction8.html [accessed 21 January 2002]. Teich, J., Wallenius, H. and Wallenius, J. (1999) ‘Multiple Issue Auction and Market Algorithms for the World Wide Web’, Decision Support Systems, 26, 49–66. Teich, J., Wallenius, H., Wallenius, J. and Zaitsev, A. (1999) ‘A Multiple Unit Auction Algorithm: Some Theory and a Web Implementation,” Electronic Markets 9(3), 199–205. Teich, J., Wallenius, H., Wallenius, J. and Zaitsev, A. (2001) ‘Designing Electronic Auctions: A WeB2-Based Hybrid Procedure Combining Aspects of Negotiations and Auctions’, Electronic Commerce Research 1(3), 301–14. Ye, Y., Liu, J. and Moukas, A. (2001) ‘Agents in Electronic Commerce’, Electronic Commerce Research 1(1–2), 9–14.

250