Research and Advances
Computing Applications

Insights and Analyses of Online Auctions

Exploring the structure and mechanisms for online mercantile processes and bidding strategies.
Posted
  1. Introduction
  2. Dimensions of Online Auctions
  3. Key Decisions for Auctioneers
  4. Future Trends in B2C Auctions
  5. References
  6. Authors
  7. Footnotes
  8. Figures
  9. Tables

The Internet-driven networked economy is evolving to the point where firms are beginning to realize the enormous business value possible from a richly connected, global network of consumers and producers. However, full exploitation of these seemingly limitless opportunities will depend largely on the design of efficient mercantile processes that facilitate a wide variety of mechanisms for the exchange of assets, goods, and services. Recognizing the increasing participation of consumers in the price-setting process, Kauffman and Riggins [5] have questioned whether fixed prices are a thing of the past. No longer is the time-tested posted-price mechanism the only choice available for the exchange of assets. Online auctions, brought about by the synergetic combination of Internet technology and traditional auction mechanisms, represent a significant new dimension of mercantile processes, many of which are not yet fully understood.

Auctions are as modern as today’s technology, yet as old as mankind. The Internet, however, has expanded the scope and reach of this market mechanism beyond the wildest dreams of its traditional purveyors. Unlike traditional auctions that were limited in scope, online auctions have brought this mechanism to the masses, providing them with an all-encompassing selection of goods they can buy or sell. Millions of globally dispersed consumers now engage in competitive exchange via bidding and can set prices that reflect real-time supply and demand as efficiently as any trading floor. The physical limitations of traditional auctions such as geography, presence, time, space, and a small target population virtually disappear in online settings. The Internet provides a critical mass of consumers located on different continents, who could, with a click of their mouse buttons, participate simultaneously in multiple auctions. Participants are also relieved of the social awkwardness and time-consuming nature of face-to-face haggling—many simply automate their bidding, letting the computer take care of negotiation.

While there is an immense body of literature [6–8] that analyzes traditional auction design and bidding strategies, the significant changes brought about by the Internet in this area are yet to be studied. Here, we empirically examine the various dimensions of online auctions, ranging from the key decisions that auctioneers must make to the variety of approaches used for consumers’ bidding strategies. In identifying these factors we note that many of the elegant and powerful theorems and some of the typical assumptions made in the classical analysis of auctions do not apply to the emerging online context.

The classical approach, perhaps unconsciously constrained by the physical limitations of traditional auctions, has been to model single auctions in isolation from the broader context of the markets in which such auctions take place. Without the physical constraints of traditional auctions the behavior of the different economic agents in auctions is heavily influenced by the (online) context in which they take place. For instance, the presence of simultaneous substitutable online auctions—which allows an individual shopping for, say a computer, to simultaneously bid at Onsale.com or Yahoo.com—affects the efficiency of not just the isolated auction under consideration but also the external market in which it takes place. Auction portals such as www.biddersedge.com are specifically designed to make tracking such simultaneous substitutable online auctions easy for the consumer.

Back to Top

Dimensions of Online Auctions

Table 1 categorizes online auctions into three main dimensions: business-to-consumer (B2C); consumer-to-consumer (C2C); and business-to-business (B2B). Much of the hype surrounding online auctions has centered on companies such as eBay that primarily serve the consumer-related dimensions. There are, however, success stories to report on the B2B front. Weirton Steel Corporation used to destroy odd lots of leftover steel until it realized it could globally auction this material on the Internet, where it was easy to find other firms having high valuations for their odd lots. Using auctions, firms have the opportunity to engage in dynamic and demand-driven production planning and control. Auctions provide information about demand that can in turn influence the nature of the input (capital, labor, and technology) to be employed to meet the demand. For instance, a company such as priceline.com could potentially present an airline company with 300 guaranteed prepaid consumers for a particular flight segment that may not exist but could be configured dynamically. While both C2C and B2B online auctions promise to occupy a prominent place in the emerging electronic marketplace, we focus our analysis on the B2C dimension.


Online auctions, brought about by the synergetic combination of Internet technology and traditional auction mechanisms, represent a significant new dimension of mercantile processes, many of which are not yet fully understood.


Mechanics of B2C online auctions. Consumers today face an interesting process choice in the electronic marketplace. They must decide between the age-old posted-price mechanism for buying goods where prices are fixed, and the bazaar-like competitive atmosphere of online auctions. To characterize such B2C online auctions, with respect to the theoretically known types of auctions, we conducted an extensive survey of the most popular auction sites (based on Web21.com ratings).

In contrast to the existing body of literature that mainly focuses on auctions of single items, a vast majority of the auctions conducted on the Web sell multiple identical units of an item using a mechanism analogous to, but not the same as, the first-price ascending English auction. Rothkopf and Harstad [10] point out that single-item results do not carry over into multiple-item settings. In such auctions, say of 10 Sony monitors, the 10 highest bidders win and the price they pay is equivalent to their highest bids. Usually, a very low opening bid, such as one dollar, is set by the auctioneer as a way to attract Web traffic. In addition, all auctions have a bid increment that defines the minimum step size for bidders. Bids that fall in between bid increments are automatically rounded down to the nearest step. This discretization of the process challenges the common auction theory assumption that individuals’ valuations can be drawn from a known, continuous distribution. The bid increment also helps determine the minimum required bid at any time during the auction. This is equal to the lowest winning bid plus the bid increment, provided the current size of the winners list is equal to the lot size. Bids are ranked by bid amount and by time within amount.

The list of current winning bidders, the bid increment, the minimum required bid, and the auction closing time are all continuously updated on the Web. Auction durations typically range from one-hour express auctions to day-long regular auctions—Figure 1 depicts such an auction. Unlike traditional, single-item English auctions, a new bidder’s high bid does not automatically displace the existing winner from the winner’s list: if the current number of bidders is less than the lot size, then the new high bid does not affect any of the existing winners.

On the other hand if the current number of winning bidders is equal to the lot size then the new high bidder displaces only the lowest of the many (equal to the lot size) current winners. The displaced consumer then has a decision to make: whether to drop out or to place a higher bid to re-enter the winning list. This process continues until the pre-announced auction closing time, which is preceded by a “going, going, gone” period. Auctioneers typically close the auction if the pre-announced closing time has passed and there are no new bids in the last five minutes. Onsale.com coined the term “Yankee Auctions” to name such auctions but we prefer to describe them as Multiple Item Progressive Electronic Auctions (MIPEA).

Common assumptions and beliefs that do not hold. Researchers using the classical game theoretic approach to model single, isolated auctions commonly make use of assumptions, such as the independent private values assumption, that do not hold in multi-item online settings. Such an assumption implies that a single indivisible object is to be sold to one of several bidders. Each bidder is risk-neutral and knows the value of the object to himself, but does not know the value of the object to other bidders (hence the phrase private values). It also implies there is a finite population of bidders, with each bidder drawing an independent valuation from some given continuous distribution (for a detailed description see [7, 9]).

In order to test assumptions like these and to better understand the important parameters that characterize such auctions we designed an automated agent that conducted around-the-clock monitoring of 90 such auctions hosted by a popular online auction site.

We tested the finite, known set of bidders assumptions by examining the auctions we tracked for the presence of new arrivals during sequential time periods. Figure 2 depicts this information for 90 such auctions, 30 each for bid increment equal to $5, $10 and $20. Each point on a line in Figure 2 represents the average percentage of new bidders in a given time period as a percentage of total bidders in a given auction.

Typically, a large percentage of new bidders arrive during the early course of an auction. Subsequently, a smaller but continuous stream of new bidders is observed. The evidence of continuous new bidders throughout the course of an auction makes it difficult to justify the assumption of a known, finite set of bidders in a globally dispersed consumer base.

One could also suggest bidding intensity rises toward the end of an auction and the probability rises that high valued consumers will bid higher as the duration increases. However, as shown in Figure 3, our examination of the number of active bidders in an auction suggests this argument has little validity. A bidder is considered active in a given period if the bidder appears in the winners list in any subsequent period. Significantly, all three curves have a tendency to level out toward the end of the auction. This contradicts the notion that longer auction duration will lead to increased revenues for the auctioneer. Figure 3 clearly indicates the bid increment seems to have a significant role in number of active participants. Since Figure 2 indicates there is a steady arrival of new bidders in a given auction, Figure 3 suggests:

  • At lower bid increments the number of active participants increases during the intermediate phase of an auction, implying there is more competition and bidding activity; and
  • At higher bidding increments, as many bidders drop off from the auctions as there are new arrivals, implying individual bidders go through fewer rounds of bidding activity as compared to auctions with lower bid increments.

In addition to disproving traditional assumptions, we conducted an exploratory study to identify control factors that actually affect auctioneers’ revenues. Our multivariate regression analysis (described in more detail later and thoroughly in [2]) of these multi-item auctions revealed that to a large extent the valuation of the marginal consumer and the bid increment set by the auctioneer determine the range of revenues for the auctioneer.

The standard practice in the literature is to define the marginal consumer as either the highest unsuccessful bidder or the lowest successful bidder [3, 4]. Both definitions characterize the price-setting consumer and are equally useful in examining the structural characteristics of these auctions.

Structural characteristics of online auctions. Given that consumers choose to participate in such auctions in the presence of other fixed price alternatives, it is fair to assume they do so with an objective of maximizing their net worth. We also assume a rational, net worth-maximizing consumer then will know his or her valuation and will try to win the auction at some lower value. We contend that, taken to the brink, such a consumer will be willing to bid up to one bid increment less than whatever his or her valuation is (or the alternative fixed price option); bidding any higher would mean the bidder would fail to derive any positive surplus and would be as well off without obtaining the item being auctioned. In the context of multiple-item auctions this behavior can help us characterize the structure of the winning bids and at the same time examine the revenue bounds for the sellers.

Depending on the temporal position of the marginal consumer’s bid (the highest losing bid) the total revenue for the auctioneer will range between (see [1] for more details):

  • A lower bound = (lot size) × (marginal consumer’s value – bid increment); and
  • An upper bound = (lot size) × (marginal consumer’s value)

A hypothetical scenario characterizing the discrete and sequential nature of such auctions and depicting how the lower and upper bound instances can materialize is described here.

Example bidding scenario. Assume an auctioneer is selling three identical objects and the bid increment is $1. Consider that at the end of the auction there are four bidders A, B, C, and D with true valuations of $51, $52, $53, and $54. Let A be the marginal consumer, with the fourth highest valuation. A MIPEA upper bound instance can occur if the marginal consumer is the last person to exit at the level of ‘his value less one bid increment.’ For example consider the following sequence of progressive bids: A(50[marginal consumer’s value less one bid increment])—C(49)—D(49)—B(50)—D(50)—C(51)—B(51)—D (51)—STOP. The auction will terminate because A will have to bid $51 to get in at this stage, which he would be reluctant to do because that would equal his valuation. Observe that since A bid $50 earliest he was the last person to be kicked out at that level. Therefore, the auctioneer’s revenue is $153.

A MIPEA lower bound instance can occur if the marginal consumer is the first person to bid at the level of ‘his value less two bid increments.’ For example consider the following sequence of progressive bids: D(48)—C(48)—B(48)—A(49[marginal consumer’s value less two bid increments])—B(49)—C(49)—D(50)—C(50)—B(50)—STOP. The auction will terminate because A will have to bid $51 to get in now, which he would be reluctant to do because that would equal his valuation. Therefore, the auctioneer’s revenue is $150.

Consumer bidding strategies. The preceding structural characterization, driven by the rationality assumption, implies that all consumers employ the same strategy while bidding. Such a strategy involves active participation that bids the minimum required bid at any stage during the auction. While such behavior is exhibited in reality, it is not the only kind of strategy employed by bidders. Our empirical investigation of 90 such auctions identified three distinct types of bidders [1], which are summarized in Table 2.

In order to compare the performance of these categories we introduced a metric [1] based on loss of surplus. This is the difference between an individual’s winning bid and the minimum winning bid. Loss of surplus evaluates the performance of an individual or a group with respect to the bidder who had the minimum winning bid in a given auction. In [1] we compared the relative performance of these three groups with respect to loss of surplus. We found the evaluators as a group fared worst, the participators were best off, and the opportunists lay in between. Interestingly, the use of automated data-collecting agents, such as ours, makes possible the hitherto ignored, detection and analysis of novice strategies pursued by less-informed bidders. Figure 4 presents the percentage of each bidder type among the winners of the auctions that we tracked at different bid increment levels.

Figure 4 indicates that as the bid increment increases from 5 to 20, there is a significant decrease in the number of evaluators. Typically, large bid increments correlate with larger absolute magnitude auctions. As mentioned earlier, evaluators as a group are the worst off. Therefore, Figure 4 suggests that consumers are more likely to take a cautious approach toward considering bigger-ticket items when the bid increment is large. In contrast, there is evidence of a more cavalier approach toward bidding considerations when the bid increment is small.

Back to Top

Key Decisions for Auctioneers

In order to identify the key revenue impacting decision variables we utilized a multivariate regression model that had normalized revenue as the dependent variable and the various key controllable attributes for online auctioneers as independent variables. The four independent variables and their impact on the revenue are presented in Table 3. The initial regression was carried out using all possible interaction terms between the independent variables. Subsequently, the higher order terms were dropped using backward selection, as they did not contribute toward explaining the variation in normalized revenue.

We tested the importance of the lot size as a decision variable with the underlying theory that appropriately choosing the lot size and holding several auctions for the same product may yield higher revenues. The intuitive appeal of this argument becomes apparent if a group of consumers is rank-ordered in a descending manner based on their valuation. By cutting down the lot size, an auctioneer may avoid selling to consumers whose valuations are lower. This creates the potential for capturing new customers with higher valuations in a later auction. However, we found lot size did not have a significant effect on an auctioneer’s revenue. This may sound surprising, however, we believe the lot-sizing decision has to be made with consideration of the alternatives available for the consumers. In online auction settings, most consumers have access to information regarding the comparable product alternatives and their posted prices. This causes consumer valuations to cluster around the prevailing market price, which in turn neutralizes the effect of lot sizing. Furthermore, for many products (for example, computer equipment, electronic goods, and computer software), holding on to inventory may be risky because of downward pressure on prices. Therefore, we believe auctioneers should avoid lot-sizing strategies and should get rid of all the inventory as soon as possible.

Most auctioneers prefer to keep low opening bids in order to attract Web traffic, at the risk of selling an item below its cost. In regard to the impact of the magnitude of the final bids on the auctioneer’s revenue, we found the magnitude of the final bids has a significant positive correlation with the bid increment. However, the bid increment showed a much stronger relationship in the regression analysis and explained a larger degree of the variance. While it is counterproductive to keep very low bid increments for high valued items and very high bid increments for low valued items, the role of the expected magnitude of the final bids is simply to make a decision regarding the range of feasible bid increments.

The most important attribute found in our analysis is the bid increment. In the auctions of comparable magnitudes we tracked, we found the coefficient for the bid increment had a positive sign in the regression equation and was significant at the a = 5% level. This suggests that increasing the bid increment can result in higher relative revenue. However, the choice of bid increment needs to be carefully considered as higher values lead to increased revenues only up to a point, after which they act as deterrents to marginal bidders who may otherwise choose to participate. Therefore, the auctioneers have an optimization problem with respect to the bid increment. This is, in our opinion, an important dimension of future research in online auctions. The fact that most online auctioneers are aware of this can be discerned from observing their strategies. For instance, during the last few months of our year-long data collection, we observed that the same Kodak digital camera was sold on a popular auction site with bid increments equal to $5 on one occasion, $10 on another. To the best of our knowledge, we have been the only researchers to highlight the importance of bid increment while others have been focusing on lot sizing and time span of auctions [11].

Auctioneers also must realize the mix of consumer strategies changes with bid increment. As shown in Figure 2, the proportion of evaluators decreases from 59% to 23% and that of participators increases from 21% to 50% as the bid increment increases from $5 to $20. From the auctioneer’s perspective a larger proportion of evaluators is desirable as these bidders often over-value the object and have been shown to be statistically worse off then the other categories [1].

Back to Top

Future Trends in B2C Auctions

While the majority of the online auction sites started with the objective of clearing aging or perishable inventory, auction-based dynamic pricing has expanded well beyond the realm of collectibles or surplus goods and in fact is a legitimate complement to the conventional notion of posted prices.


Most auctioneers preferto keep low opening bids in order to attract Web traffic, at the risk of selling an item below its cost.


Choice of auction mechanism. There is a great deal of theoretical work that compares the efficiency of alternative mechanisms. However, much of it is restricted to the single-item setting. Myerson [9] and Bulow and Roberts [3] describe the theorem of revenue equivalence, which states that if the winner under one type of auction (say an English auction) is also the winner under a second type of auction (Dutch auction, for example), than the two auctions will yield the same expected revenue. The extension of this result into multi-item settings is not trivial even with the most simplistic of assumptions regarding the consumer type. As the B2C online auction market matures one can already see considerable, but at times directionless, experimentation concerning the choice of auction mechanism.

Hybrid mechanisms. There are increasing signs of overlap (on eBay) between C2C and B2C auctions. Just as with other B2C auctions, these hybrid consumer-to-consumer (hC2C) auctioneers also sell multiple units of items, often providing links to their own Web sites for future transactions. This saves them advertising and promotion expenses and they gain access to a large, savvy, and targeted consumer base, genuinely interested in their wares.

Mistaken terminology. Based on the preceding reasons it is not surprising that many small businesses are utilizing C2C auction sites like eBay and Amazon to auction their goods. The so-called Dutch auction used by eBay is an ascending open uniform price auction that differs from the original version of Dutch auctions, which originated in Dutch Flower markets. In the original scheme, an auctioneer announces successively lower prices until a bidder bids and thereby wins the right for a sale [8]. While purists may not like the amorphous and free-spirited cross-mingling of attributes, methods, and nomenclature pursued by these online auctioneers, none can criticize the verve and tenacity with which they continue to bring new mercantile processes to the electronic marketplace.

The Dutch auctions of eBay and Amazon.com can be most loosely identified with a multi-item extension of what is classically known as a Vickrey auction—in honor of the seminal work of Nobel laureate, William Vickrey [12]. It should be noted that Vickrey’s original version of this auction was designed to be a sealed-bid auction, in contrast to eBay’s open and progressive version where all bids are posted and can be revised. In eBay’s Dutch auctions arrangement, the current standings are always displayed after the item description, and the complete bidding history (including unsuccessful bids) is available as well. Current research shows that to prevent demand reduction in multi-item auctions, of say m items, the top m bids are declared winners and for the jth unit won by a bidder, who pays an amount equal to the jth highest of the rejected bids submitted by others. Hence, this revised mechanism offers discriminating prices in contrast to the original mechanisms’ uniform pricing. An interesting question is, does eBay’s so-called Dutch mechanism, which has little theoretical basis and unexplored incentive characteristics, lead to higher clearing prices than, for example, an equivalent Yankee auction? All said, examining the efficiency and applicability of the emerging dynamic pricing mechanism in the electronic marketplace promises to be an exciting area of future research.

Back to Top

Back to Top

Back to Top

Back to Top

Figures

F1 Figure 1. Snapshot of a multi-item online auction.

F2 Figure 2. Average percentage of new arrivals overtime by bid increment.

F3 Figure 3. Active bidders overtime grouped by bid increment.

F4 Figure 4. Bidder classification aggregated by bid increment.

Back to Top

Tables

T1 Table 1. Dimensions of online auctions.

T2 Table 2. Bidder classification.

T3 Table 3. Decision variables for auctioneers and thier significance.

Back to top

    1. Bapna, R., Goes, P., and Gupta, A. A theoretical and empirical investigation of multi-item online auctions. Information Technology and Management 1, 1 (Jan. 2000), 1–23.

    2. Bapna, R. Economic and Experimental Analysis and Design of Auction-based Online Mercantile Processes. Ph.D. dissertation, Department of Operations and Information Management, University of Connecticut, 1999.

    3. Bulow, J. and Roberts, J. The simple economics of optimal auctions. Journal of Political Economy 7, 5 (May 1989), 1060–1090.

    4. Harris, M. and Raviv, A. Allocation mechanism and the design of auctions. Econometrica 49, 6 (Nov. 1981), 1477–1499.

    5. Kauffman, R.J. and Riggins, F. J. Information systems and economics. Commun. ACM 41, 8 (Aug. 1998), 32–34.

    6. McAfee, R.P. and McMillan, J. Auctions and bidding. Journal of Economic Literature 25 (1987), 699–738.

    7. Milgrom, P. Auctions and bidding: A primer. Journal of Economic Perspectives 3 (1989), 3–22.

    8. Milgrom, P. and Weber, R. A theory of auctions and competitive bidding. Econometrica, 50 (1982), 1089–1122.

    9. Myerson, R.B. Optimal auction design. Mathematics of Operations Research 6 (1981), 58–73.

    10. Rothkopf, M.H. and Harstad, R.M. Modeling competitive bidding: A critical essay. Management Science 40, 3 (1994), 364–384.

    11. Vakrat, Y. and Seidmann, A. Implications of the bidders' arrival process on the design of Online Auctions. In Proceedings of the 33d Annual Hawaii International Conference on System Sciences (HICSS-33), (Jan. 2000, Maui, Hawaii).

    12. Vickrey, W. Counterspeculation, auctions, and competitive sealed tenders. Journal of Finance 41 (1961), 8–37.

    This research was supported in part by TECI—the Treibick Electronic Commerce Initiative, OPIM/SBA, University of Connecticut. Alok Gupta's research is supported by NSF Career grant #IIS-0092780, but does not necessarily reflect the views of the NSF.

Join the Discussion (0)

Become a Member or Sign In to Post a Comment

The Latest from CACM

Shape the Future of Computing

ACM encourages its members to take a direct hand in shaping the future of the association. There are more ways than ever to get involved.

Get Involved

Communications of the ACM (CACM) is now a fully Open Access publication.

By opening CACM to the world, we hope to increase engagement among the broader computer science community and encourage non-members to discover the rich resources ACM has to offer.

Learn More