Website Usage and Sellers Listing in Internet Auctions
Upcoming SlideShare
Loading in...5
×
 

Website Usage and Sellers Listing in Internet Auctions

on

  • 1,125 views

 

Statistics

Views

Total Views
1,125
Views on SlideShare
1,125
Embed Views
0

Actions

Likes
0
Downloads
12
Comments
0

0 Embeds 0

No embeds

Accessibility

Categories

Upload Details

Uploaded via as Adobe PDF

Usage Rights

© All Rights Reserved

Report content

Flagged as inappropriate Flag as inappropriate
Flag as inappropriate

Select your reason for flagging this presentation as inappropriate.

Cancel
  • Full Name Full Name Comment goes here.
    Are you sure you want to
    Your message goes here
    Processing…
Post Comment
Edit your comment

Website Usage and Sellers Listing in Internet Auctions Website Usage and Sellers Listing in Internet Auctions Document Transcript

  • Website Usage and Sellers’ Listing in Internet Auctions Sangin PARK* Department of Economics SUNY at Stony Brook Stony Brook, NY 11794-4384 Tel: 631-632-7559 Fax: 631-632-7516 Email: sanpark@notes.cc.sunysb.edu Web: http://ms.cc.sunysb.edu/~sanpark First Draft: September 2001 This Draft: September 2002 In this paper, we provide empirical evidence on the network effect between website usage and the number of listings in Internet auctions. The (no) arbitrage condition of a seller’s listing behavior, combined with our unique weekly data on listings and website usage of eBay and Yahoo!Auctions as well as fee schedules and available auction mechanisms, indicates that a seller’s expected auction revenue increases with the number of potential bidders measured by (an increasing function of) page views per listing. In addition, our data analysis shows that an increase in the number of listings raises page views per listing (and thus the potential bidders). Therefore, we can infer the existence of the network effect as a positive feedback effect between listings and website usage. The existence of this network effect may explain why Yahoo!Auctions, which has substantially less listing, has a greater incentive to increase listings by setting lower fees. However, the low values of both the estimated elasticity of page views with respect to the number of listings (1.2) and the estimated elasticity of a seller’s auction revenue with respect to the number of potential bidders (0.1 in our sample of the Barber Quarter Dollar auctions) suggest that website usage per listing may not rise rapidly via this network effect in Internet auctions. Therefore, once a pioneering firm established a huge network of sellers and potential bidders, followers may not be able to catch up quickly even with substantially lower fees. KEYWORDS: Internet auctions; website usage; listing behavior; network effect. JEL Classification: L86; D44. * I thank Nielsen//NetRatings for providing the data of weekly website usage. I also appreciate the comments from the participants of the 2001 Telecommunications Policy Research Conference. All errors and omissions, of course, are my own.
  • 1. Introduction For the last three and a half years, the fluctuation of the market value of an e-commerce firm has been huge. First, the potential growth of e-commerce has hit the stock market: due to the Internet related stocks, the Nasdaq composite index went up more than 70 percent in 1999. Then, the enthusiasm of the Wall Street was struck by disappointing profits and worries about slowing economy. Between December of 1999 and August of 2001, the stock price of eBay peaked at $127.50, plunged to $26.75, and then rebounded to the range of $50 to $70. During the same time period, the stock prices of Yahoo! and Amazon.com continuously plunged from the level of $250 to the level of $10 and from the level of $110 to the level of $8, respectively. This turmoil of e-commerce firms’ stock prices highlights the worry of the long-term viability for these firms. Since the beginning, the e-commerce has been hailed as a frictionless competitive market (see, e.g., Bakos 1991, 1997). It is widely believed that the Internet drastically reduces buyers’ search costs (for prices, product offerings, and shop locations) and lowers barriers to entry and exit. The low search costs and lowered barriers to entry and exit will induce strong price competition, leading to low profit margins and low deadweight losses in the e-commerce. Consistently with these theoretical predictions, some case studies have found that e-commerce has on average lower prices than the conventional retail market (for the review of these studies, see Bailey 1998; Brynjolfsson and Smith 2000). However, as argued in Ellison and Ellison (2001), the overhead costs in e-commerce may not be as low as anticipated, and thus severe price competition may lead to the Bertrand paradox (with prices so low that firms cannot cover their overhead costs). More recently, however, several empirical studies (see, e.g., Brynjolfsson and Smith 2000; Clemons et al. 2002; Johnson et al. 2001; Loshe et al. 1999) indicate some frictions in e-commerce: (a) compared to the conventional retail market, e-commerce has low prices on average but high market concentrations; (b) the e-commerce firm with a higher market share usually charges a higher price; and (c) the price dispersion is higher than in the conventional retail market. These empirical observations 1
  • suggest that there may exist some first-mover advantage which explains why the pioneering firms, such as Amazon.com, Yahoo!, E*Trade and eBay, have higher market shares despite higher prices. Furthermore, some studies suggest that this first-mover advantage may be caused by switching costs in the on-line brokerage market (Chen and Hitt 2001) or by reputation in the Web portals (Goldfarb 2002). As another source of this first-mover advantage, this paper will examine network effect in a particular e-commerce called Internet auctions. As well known, eBay, the pioneer of the Internet Auctions, has been very profitable and dominated the Internet auctions market. It has been widely speculated that the positive feedback effect (or network effect) between buyers’ website usage and sellers’ listing behavior might be the reason for this eBay’s profitability and dominance.1 The idea of this network effect seems quite straightforward: more potential buyers will visit an Internet auctions site if there are more listed items for auctions, and more sellers will list their items on that site if more (potential) buyers visit the site. However, there are several non-trivial complications when we connect this (naïve) idea to the data and the auction theories. First of all, the number of potential bidders faced by a specific seller on an Internet auctions site may not be exactly equal to the website usage of the site. Different sellers may compete with each other because they list similar (or substitutable) items, and some potential buyers may not be interested in bidding the seller’s item. Moreover, the website usage data do not distinguish the buyers’ visits form the sellers’ visits, and is typically measured by ‘unique visitors’ or ‘page views’.2 Hence it is a question how to relate the usage data to the number of potential buyers faced by a seller. Second, whether a potential bidder will log on to an Internet auctions site may depend on not only the probability that the bidder to find an item she/he wants but also the expected sale price of the item. The probability that the bidder to find a wanted item may be higher if an Internet auctions site has more listings. However, the expected sale price of the item may also be affected by listings (via the number of potential bidders), and it depends 1 Refer to Katz and Shapiro (1994) for the review of the literature on network effects. For the recent empirical studies of network effects, refer to Gandal et al. (2000), Rysman (2000) and Park (2002). 2 ‘Unique visitors’ is the estimated number of different individuals who visit a firm’s website, and ‘page views’ is the number of unique visitors multiplied by the average unique pages viewed per visitor. 2
  • on the Internet auction mechanisms whether more listings lead to a higher or lower expected sale price. Hence it is not certain that more potential buyers will visit an Internet auctions site if there are more listed items for auctions. Lastly, the literature of auction theory predicts that depending on whether potential bidders’ entries into a specific auction are endogenous or exogenous, a seller’s expected auction revenue is either decreasing or increasing in the number of the potential bidders (Levin and Smith 1994; Bulow and Klemperer 1996). Hence, there is no theoretical guarantee that more sellers will list their items on a site if more (potential) buyers visit the site. In the paper, to empirically examine the network effect in Internet auctions, we will use a unique weekly data for the number of listings and website usage on both eBay and Yahoo!Auctions (see section 2 for the details of these data). Due to the dominance of eBay, it is unusual to have the data of the number of listings and website usage on any auctions site other than eBay. Although our main data covers at most the first 17 weeks of the year 2001, it will provide us with a window of opportunity to take a snapshot of the empirical relationship between website usage and listings in Internet auctions. The paper is organized as follows. In section 2, based on the empirical observations of eBay and Yahoo!Auctions, we will discuss the measurement of the number of potential bidders faced by a seller on an Internet auctions site. Then in section 3, utilizing the (no) arbitrage condition of the seller’s listing behavior, we will empirically study the relationship between sellers’ listing behavior and the number of potential bidders. The arbitrage condition implies that a seller’s expected revenues from listing an item on any Internet auctions site must be the same. Hence, using this arbitrage condition, we can relate, for given auction mechanisms, the number of potential bidders to the difference of expected auction revenues implied by the difference of fees charged by the sites. For the completion of network effect as a positive feedback effect between website usage and listings, we further use a reduced-form analysis of the potential bidders’ website usage equation. We also, based on some specification assumptions, quantify a seller’s valuation of the number of potential bidders. In section 4, our empirical findings will be further discussed in contrast with the theoretical predictions of the auction literature, highlighting unique features 3
  • of the Internet auctions. Appendix describes the supplementary data of the Barber Quarter Dollar auctions on eBay and Yahoo!Auctions closed between November 28, 2001 and December 4, 2001. 2. Internet Auctions, Potential Bidders and Data 2.1. Internet Auctions Internet auctions began in 1995 and have been growing rapidly. As of fall 1999, Internet auctions sites are estimated to have almost $100 million revenues per month, while in May 2001, the estimated total revenue of Internet auctions reaches $556 million.3 Since the beginning of Internet auctions, eBay has maintained a dominant leading position although the popular attention and the profitability of eBay induced entries of two biggest e-commerce firms, Yahoo! in October 1998 and Amazon.com in March 1999.4 An eBay Vice President said in January 2000 that eBay’s market share in the Internet auctions market remained at approximately 90 percent (Lucking-Reiley 2000). As of fall 2001, it is still believed to be more than 80 percent.5 An Internet auctions site, such as eBay and Yahoo!Auctions, acts as a listing agent, allowing individual sellers to register their items for its website and running Web-based automatic auctions on their behalf.6 Actual exchanges including payment and shipment are worked out by the buyer and the seller on their own. The English auctions have been the most dominant format in Internet auctions.7 However, sellers usually have some control over these Web-based auctions, choosing a set of different parameters for each auction such as the duration days, an opening value, an optional secrete reserve price, etc. 3 See Lucking-Reiley (2000) and “eBay riding Net auction industry’s wave”, CNETNews.com on June 28, 2001. 4 Except these three largest generalist auctioneers, other small Internet auctions sites usually serve small niche markets. 5 See “The People’s Company”, Business Week on December 3, 2001. 6 In this paper, we narrowly define Internet auctions as consumer-to-consumer transactions via listing agents. Note that there are retail merchants such as Onsale and Egghead who use the auction format to sell their own products. 7 If the end time of an auction is fixed as in eBay, all bidders have incentives to bid only at the last minute. Hence, in this case, the apparent English auctions will be equivalent to the first-price sealed-bid auctions. 4
  • A variety of goods are auctioned in Internet auctions, but the largest category by far has been the collectibles. Each Internet auctions site has different categories, and there is usually no one top-level category that includes all the types of collectibles. Between September 27, 2001 and November 1, 2001, 59 percent of listings on eBay belonged to one of the categories such as ‘antiques & art’, ‘collectibles’, ‘books, movies, music’, ‘coins & stamps’, ‘dolls & doll houses’, or ‘toys, bean bag plush’. During the same time period, 54 percent of listings of Yahoo!Auctions were included in one of the following categories, ‘antique, art & collectibles’, ‘sports cards & memorabilia’, ‘toys & games & hobbies’, or ‘coins, paper money & stamps’. 2.2. eBay vs. Yahoo!Auctions Yahoo!Auctions has kept a distant second place in the Internet auctions market although it has offered a little bit more flexible auction mechanisms and lower listing fees compared to the leader, eBay. As shown in table 1, there has been no significant difference in available auction mechanisms (such as auction formats and auction parameters) between these two sites although Yahoo!Auctions has offered a little bit more options. Both sites employ ascending-bid (English) auctions and offer ‘proxy bid’ in which the Web-based auction automatically raises a bidder’s bid, as other bidders increase the bid price, to the maximum amount set secretly by the bidder in the beginning of the auction.8 On both sites, bidders can check sellers’ ratings before the auction and evaluate sellers after the auction (feedback-and-rating system). Options such as ‘secret reserve price’, ‘buy it now’, and ‘early close’ are available from both sites.9 Yahoo!Auctions, however, provides more flexible lengths of auction duration and an additional 8 A small portion of Internet auctions are multi-unit auctions, for which both sites also use ascending-bid auctions with specified quantity. However, on eBay, all winning bidders pay the lowest successful bid (uniform-price rule) while on Yahoo!Auctions, winning bidders have to pay the amount of their maximum bid (pay-your-bid rule). 9 ‘Buy it now’ is an optional feature in which a seller names a buy price, and if a bid is placed to match or exceed this price, the item is sold to that bidder and the auction is closed immediately. 5
  • Table 1: Auction Mechanism eBay Yahoo!Auctions auction format for a single item English auction English auction proxy-bidding Optional Optional secret reserve price Optional Optional duration (days) 3, 5, 7, or 10 2 - 14 feedback-and-rating Available Available buy it now Optional Optional early close Optional Optional auto extension Not Available Optional option called ‘auto extension’.10 On eBay, sellers can choose a length of 3, 5, or 7 days and a length of 10 days with an extra fee of $0.10 while on Yahoo!Auctions, sellers can choose a length between 2 and 14 days.11 FINDING 1: eBay and Yahoo!Auctions offer almost the same auction mechanisms although Yahoo!Auctions has more flexible options. Although there is no significant difference in available auction mechanisms between eBay and Yahoo!Auctions, these two sites charge distinctively different fees to sellers.12 During the first 17 weeks 10 In the case that a seller chooses ‘auto extension’, the auction closing time can be automatically extended for 5 minutes if a bid is placed within the last 5 minutes of the auction. Hence, ‘auto extension’ can avoid the problem of the fixed end time as discussed in footnote 7 and restore the English auction mechanism. 11 In our sample of the Barber Quarter Dollar auctions, the average duration day of the traded auctions is 9 days on Yahoo!Auctions and 6.7 days on eBay. In the survey of Lucking-Reiley (2000), a modal length of duration of Internet auctions is 7 days. 6
  • Table 2: Fees eBay Yahoo!Auctions Insertion Fees Opening Value $0.01-$9.99 $0.30 $0.20 $10.00-$24.99 $0.55 $0.35 $25.00-49.99 $1.10 $0.75 $50.00-$199.99 $2.20 $1.50 $200.00and up $3.30 $1.50 Final Value Fees Closing Value $0-$25 5% Free $25-$1000 $1.25 + 2.5% Free over $1000 $25.63 + 1.25% Free of the year 2001, eBay charges two types of basic fees to sellers: insertion fees and final value fees.13 The insertion fees of eBay range from $0.30 to $3.30, depending on the opening values (called also reserve prices or minimum bid levels)14 while the final value fees are 5 percent of the final value (called also sale price or closing value) up to $24.99, 2.5 percent from $25.00 up to $1000.00, and 1.25 percent over $1000.00. On the other hand, Yahoo!Auctions charges only insertion fees ranging from $0.20 to $1.50. As indicated in table 2, eBay charges higher insertion fees for all the ranges of opening values. FINDING 2: eBay charges sellers higher insertion and final value fees. The basic fees of Internet auctions have not changed frequently. Indeed, Yahoo!Auctions began to charge insertion fees only from the beginning of the year 2001.15 At the same time, eBay raised its insertion fees 12 Both sites charge no fee to bidders. 13 eBay and Yahoo!Auctions also charge sellers non-refundable fees for several optional features, such as featured home page, highlight, bold, etc., which may promote the seller’s listing to receive more bids. 14 A seller can ex ante choose a secret reserve price as well. If the secret reserve price is not met by the close of the auction, the item will not be sold. The fees for the secret reserve price auctions are fully refundable if the item is sold. If an item is not sold, the seller can re-list the same item subject to insertion fees. On eBay, the insertion fee for the second listing is refundable if the item is sold in the second round. 15 Yahoo!Auctions once again changed its fee schedule on November 20, 2001. 7
  • Figure 1: Listings 7000000 6000000 5000000 eBay 4000000 3000000 2000000 Yahoo!Auctions 1000000 0 1 1 1 1 1 01 01 01 01 /0 /0 /0 /0 /0 7/ 4/ 4/ 1/ 21 18 18 15 29 1/ 2/ 3/ 4/ 1/ 2/ 3/ 4/ 4/ a little bit to the levels shown in table 2. As will be discussed below, these changes of fees had significant impacts on the number of listings on Yahoo!Auctions. Although Yahoo!Auctions has offered more flexible auction mechanisms and lower listing fees than eBay, eBay’s dominance is evident in terms of both the number of listings and website usage. The data employed in the paper is unique since it is unusual to obtain the data for the number of listings and website usage on any auctions site other than eBay.16 Our data of the number of listings are obtained from the Downtown Magazine’s Wednesday Report. The Downtown Magazine is an unbound magazine available via the Internet (www.dtmagazine.com) and reports its counts of the number of auction listings updated every Wednesday by noon Eastern Time.17 This weekly data of the number of listings for eBay and Yahoo!Auctions were collected from the first week of the year 2001. However, from the second week of April 2001, the Wednesday Report ceased counting the number of weekly listings on Yahoo!Auctions and began to report the number of listings on a monthly basis, and this is the main reason that the number 16 For instance, the Amazon.com auctions, the third largest Internet auctions site, is not included in our analysis because we have no data for listings or website usage. 17 In this counting, eBay listings do not include eBayMotors or Great Collections but include Business Exchanges and UltimateBid Tickets and Experiences while Yahoo!Auctions listings include Yahoo!’s Business Marketplace. 8
  • Figure 2: Unique Visitors 7000000 6000000 5000000 eBay 4000000 3000000 2000000 1000000 Yahoo!Auctions 0 1/7/2001 1/14/2001 1/21/2001 1/28/2001 2/4/2001 2/11/2001 2/18/2001 2/25/2001 3/4/2001 3/11/2001 3/18/2001 3/25/2001 4/1/2001 4/8/2001 4/15/2001 4/22/2001 4/29/2001 Figure 3: Page Views 900000000 800000000 700000000 eBay 600000000 500000000 400000000 300000000 200000000 100000000 Yahoo!Auctions 0 1/21/01 2/18/01 3/18/01 4/15/01 4/29/01 1/7/01 2/4/01 3/4/01 4/1/01 of our observations are very small. As illustrated in figure 1, during the first 17 weeks of 2001, the number of listings on eBay remained between 5 million and 5.7 million after an increase in the first week. On the other hand, the initiation of fees on Yahoo!Auctions in the beginning of the year 2001 had a dramatic impact on the listings of Yahoo!Auctions. The number of listings on Yahoo!Auctions declined drastically from more than 2 millions to about 250,000 by the 7th week of the year 2001, and then 9
  • stabilized around 220,000.18 In what follows, therefore, our data analysis will focus on the period of the 7th week to the 13th week (the third week of February to the first week of April) after the changes of the fee schedules were fully absorbed in sellers’ listing behavior. The data of weekly website usage (measured by ‘unique visitors’ and ‘page views’) of these two auctions sites are obtained from Nielsen//NetRatings for the first 17 weeks of the year 2001 (except in the first week of March for eBay and the first week of January for Yahoo!Auctions). This usage data from Nielsen//NetRatings is unique in the sense that the weekly usage of Yahoo!Auctions is counted separately from the entire site of Yahoo!. Unlike the number of listings, however, the changes of fees had no substantial effect on website usage. As illustrated in figures 2 and 3, between the 7th week and the 13th week of 2001, eBay had about 6.3 million unique visitors and 763.6 million page views on weekly average while Yahoo!Auctions had about 530,000 unique visitors and 1.7 million page views. It is quite puzzling that the initiation of fees on Yahoo!Auctions incurred the drastic decrease of listings but had no significant impact on website usage. One possible explanation is that before the initiation of fees, many sellers on Yahoo!Auctions might have kept listing their items over and over until they were sold and thus the number of the listings were substantially over-counted. The data of the number of listings and website usage shown in figures 1 - 3 clearly indicate the following. FINDING 3: eBay has much larger website usage (based on either unique visitors or page views) and much more listings than Yahoo!Auctions. 2.3. Number of Potential Bidders As discussed in Introduction, the number of potential bidders faced by a specific seller on an Internet auctions site may not be exactly equal to the number of website usage. Different sellers may 18 According to the Downtown Magazine, Yahoo!Auctions had highs of 2.5 million listings in mid 2000. 10
  • compete with each other because they list similar (or substitutable) items, and some buyers may not be interested in bidding the seller’s item. To reflect these two aspects, we assume that the number of potential bidders faced by a seller is an increasing, item-specific function of website usage per listed item. ASSUMPTION 1: Let Uj denote website usage and Lj denote the number of listings on Internet auctions site j. Let Nj i be the number of potential bidders faced by a seller listing item i and f i be an increasing real function. Then Nj = f i (Uj / Lj). From now on, for notational simplicity, we use Nj and f instead of Nj i and f i, but it is understood that Nj and f are indexed by the item which a seller is listing. Website usage, however, is typically measured by ‘unique visitors’ or ‘page views’. Hence it is a question which one is a more relevant to the number of potential bidders faced by a seller. To answer this question, we look into our data of unique visitors and page views as well as the number of listings discussed in the previous subsection. It is striking to note that during the period in concern (the 7th week to the 13th week of 2001), unique visitors per listing are on average 1.2 on eBay but 2.3 on Yahoo!Auctions19 while page views per listing are on average 148.4 on eBay but 7.3 on Yahoo!Auctions (see figures 4 and 5). Note that ‘unique visitors’ is the (estimated) number of different individuals who visit a website while ‘page views’ is the number of unique visitors multiplied by the average unique pages viewed per visitor. Considering that a unique seller will also be counted as a unique visitor, the small number of unique visitors per listing on both sites (especially on eBay) suggests the most unique feature of Internet auctions: a (same) unique visitor may act as a seller, a potential bidder, or both.20 ‘Page views’ 19 Note that unique visitors per listed item on Yahoo!Auctions exceeded those on eBay from the 5th week of 2001 (see figure 4). This is because the initiation of fees on Yahoo!Auctions reduced drastically the number of listings but had no significant impact on website usage on Yahoo!Auctions (see figures 1 – 3). 20 The unique visitors on eBay are much more active. The average page views by unique visitors are impressively higher on eBay. During the period in concern, the average page views by unique visitors were 124.7 on eBay and 3.2 on Yahoo!Auctions. 11
  • Figure 4: Unique Visitors / Listings 3.0 2.5 2.0 Yahoo!A uctions 1.5 1.0 eBay 0.5 0.0 1/14/01 1/21/01 1/28/01 2/11/01 2/18/01 2/25/01 3/11/01 3/18/01 3/25/01 4/15/01 4/22/01 4/29/01 1/7/01 2/4/01 3/4/01 4/1/01 4/8/01 Figure 5: Page Views / Listings 180.0 160.0 140.0 120.0 eBay 100.0 80.0 60.0 40.0 20.0 Yahoo!Auctions 0.0 1/7/2001 1/14/2001 1/21/2001 1/28/2001 2/4/2001 2/11/2001 2/18/2001 2/25/2001 3/4/2001 3/11/2001 3/18/2001 3/25/2001 4/1/2001 4/8/2001 4/15/2001 4/22/2001 4/29/2001 will reflect these multiple roles of a unique visitor in Internet auctions. Hence, we can infer that ‘page views per listed item’ is more relevant to the number of potential bidders. FINDING 4: ‘Page views per listed item’ is more relevant to the number of potential bidders, and eBay has much more page views per listing. 12
  • Combining Assumption 1 and Finding 4, we now assume that the number of potential bidders on Internet auctions site j, Nj, is an increasing function of pages views per listed item. 3. Sellers’ Listing Behavior and Website Usage 3.1. Expected Auction Revenues and Potential Bidders As indicated in Findings 3 and 4, eBay has more listings and much more page views per listing. Hence we can infer a positive correlation between the number of listings and the number of potential bidders in Internet auctions. In this section, we proceed further to discuss empirical evidence on the positive feedback effect between website usage and the number of listings in Internet auctions. We begin by examining whether more sellers list their items on an Internet auctions site if more potential bidders visit the site. In principle, a seller’s expected revenue from listing an item on Internet auctions site j depends on the expected revenue from auctioning the item and the fees charged by site j. Furthermore, the expected auction revenue, say Rj, is affected by the auction mechanism and the number of potential bidders. As discussed above, there are two types of basic fees to sellers: insertion fees, say Fj, and final value fees, say αj. Then the seller’s expected revenue from listing his/her item on Internet auctions site j is: (1 − α j ) R j − F j . (1) The (no) arbitrage condition implies that the seller’s expected revenues from listing the item on any auction website must be the same. Then for any two auctions sites, say e and y, we have: (1 − α e ) Re − (1 − α y ) R y = Fe − Fy . (2) 13
  • Hence, the number of listings on each site will be determined by this arbitrage condition for a given (or expected) number of potential bidders, available auction mechanisms and fees charged by the sites. As shown in table 2, eBay charges sellers higher insertion fees for all the ranges of opening values. In addition, only eBay charges the final value fees. Hence, if a potential seller chooses the same (or a higher) range of opening values (for the same item) on eBay, the arbitrage condition in (2) indicates that the seller’s expected auction revenue must be greater on eBay. In other words, we expect eBay premium on auction revenues. A seller’s choice of an opening value (and thus an insertion fee), however, may be determined endogenously.21 In the Barber Quarter Dollar auctions on eBay and Yahoo!Auctions closed between November 28, 2001 and December 4, 2001 (see Appendix for details of this data set), we found that the average opening value of the traded coins was $26.47 with the average final value of $41.08 on eBay while the average opening value of the traded coins was $6.55 with the average final value of $14.65 on Yahoo!Auctions. Although we need a broader study of the comparison of auctions on eBay and Yahoo!Auctions for more conclusive arguments, this snapshot indicates that sellers on Yahoo!Auctions set on average a lower ratio of opening value to final value. Hence we assume as follows. ASSUMPTION 2: A seller will set no higher opening-value-to-final-value ratio if she/he lists an item on Yahoo!Auctions instead of eBay. As documented in Lucking-Reiley (1999) and Bajari and Hortacsu (2001), opening value is believed to be the most important determinant of potential bidders’ entry into a specific auction in Internet auctions. No higher ratio of opening value to final value on Yahoo!Auctions, therefore, seems consistent with the 21 In practice, however, sellers usually set low opening values to attract more bidders in Internet auctions. The conventional wisdom seems to be that a $1 opening value plus a $50 secret reserve price would be more profitable to the seller than a $50 opening value with no secret reserve price. 14
  • findings in Lucking-Reiley (1999) and Bajari and Hortacsu (2001) since Yahoo!Auctions has substantially less potential bidders. The difference in fees shown in table 2, together with Assumption 2 and the arbitrage condition of (2), indicates that a seller’s expected auction revenue is higher on eBay. As pointed out in the literature of the auction theory, a seller’s expected auction revenue, Rj, on site j depends on several factors such as the number of potential bidders, Nj, and auction mechanisms, say mj, such as available auction formats and a set of parameters which the seller can choose in each auction. That is, Rj = R(Nj, mj), where R is a real function. As discussed above, there is no significant difference in the choices of auction formats and auction parameters between eBay and Yahoo!Auctions although Yahoo!Auctions offers slightly more options. Hence, if the sizes of potential bidders are the same, a seller’s expected auction revenue on Yahoo!Auctions must be at least as high as that on eBay. However, as discussed above, the expected auction revenues are higher on eBay. Therefore, the implied eBay premium on auction revenues can be explained only by more potential bidders on eBay. FINDING 5: Under Assumptions 1 and 2, the eBay premium implied by differences in fees and available auction mechanisms indicates that a seller’s expected auction revenue increases with the number of potential bidders in Internet auctions. 3.2. Website Usage of Potential Bidders Findings 3 – 5 are suggestive of the existence of network effect in Internet auctions: website usage is positively correlated with the number of listings, and a seller’s expected auction revenue increases with website usage per listed item. Since a seller’s expected auction revenue increases with website usage per listed item, if 1 percent increase in listings induces more than 1 percent increase of 15
  • website usage, then the expected auction revenue will be raised by these increased listings.22 Therefore, for the completion of the network effect as a positive feedback effect between listings and website usage, we have to check whether 1 percent increase in listings induce more than 1 percent increase of website usage. The comparison of the number of listings and page views on eBay and Yahoo!Auctions provides us with some idea on this quantitative relationship. As discussed in section 2, eBay has more listings and more page views. However, the difference between eBay and Yahoo!Auctions is much bigger in page views than the number of listings. In fact, during the time period in concern, page views per listing are on average 148.4 on eBay but 7.3 on Yahoo!Auctions. Hence, considering that eBay has more listings, these numbers suggest that the elasticity of page views with respect to listings is likely to be greater than 1.23 To have a more accurate idea of the elasticity of page views with respect to listings, we need to conduct a regression analysis. As discussed in Introduction, whether a potential bidder will log on to an Internet auctions site may depend on the probability that the bidder to find an item she/he wants and the expected sale price of the item. The probability that the bidder to find a wanted item may be higher if an Internet auctions site has more listings. However, the expected sale price of the item may also be affected by listings (via the number of potential bidders), and it depends on the Internet auction mechanisms whether more listings lead to a higher or lower expected sale price. As will be discussed in section 4, since, to our knowledge, appropriate structural models for the Internet auction mechanisms are not yet available, we use a reduced-form analysis of the potential bidders’ website usage. To quantify the effects of listings on potential bidders’ website usage, we use a logarithm specification as follows. c +η j U j = Lβj1 Lβ2j e − , (3) 22 Let eUL denote the elasticity of website usage with respect to listings, that is, eUL = (∂Uj / Uj) / (∂Lj / Lj). Since ∂(Uj/Lj) / ∂Lj = (eUL - 1)(Uj / Lj2), Uj/Lj will increase in Lj if eUL > 1. 23 As shown in footnote 22, if ∂(Uj/Lj) / ∂Lj > 0, then eUL > 1. Our observations indicate that (Ue/Le – Uy/Ly) / (Le – Ly) > 0. However, unlike the following regression analysis, this calculation does not consider the influence of the rival site’s listings on the potential bidders’ website usage. 16
  • where Uj is the website usage of potential bidders, Lj is the number of listings, L-j is the number of listings of the rival site, (β1, β2, c) is a vector of parameters to estimate, and ηj is an error term of this reduced- form estimating equation reflecting a measurement error of Uj or the other factors which are not correlated with the number of listings. Note that β1 and β2 measure the elasticity of website usage on an Internet auctions site with respect to its own listings and its rival’s listings, respectively. Since ηj is not correlated with Lj and L-j, we will apply an OLS estimation procedure to the reduced-form estimating equation of (3). Table 3 reports these estimation results. As discussed in section 2, ‘page views’ is a more relevant measurement of website usage, and thus we use page views as the website usage of potential bidders, Uj, in (3). Since the page views data also include sellers’ usage, we also employ page views minus the number of listings as the website usage of potential bidders in our regressions. As shown in table 3, however, there is no significant difference in our estimates between these two alternative measurements of the potential bidders’ website usage. Table 3 indicates that the logarithm specification of (3) fits the data very well although the constant, c, is a little bit imprecisely estimated. Note that our data size is only 14 but the OLS estimates are the best linear unbiased estimates. In addition, since our data covers a short period of time, we do not have to worry about the effects of any structural changes on our estimates of the reduced-form equation. As reported in table 3, our key coefficient, the elasticity of page views with respect to listings (β1), is estimated to be 1.2. That is, 1 percent increase of sellers’ listings on a site induces 1.2 percent increase of page views.24 Hence, this estimation result indicates that an increase in listings will eventually raise the seller’s expected auction revenue in Internet auctions. Since increased expected auction revenues will induce more listings, we can infer a positive feedback effect between the number of listings and website usage. The existence of this network effect may explain the different fee schedules between eBay and Yahoo!Auctions. Yahoo!Auctions, which has substantially less listings, may 24 As an alternative, we also considered a linear specification of the potential bidders’ website usage equation. However, we ended up with an imprecise estimate of the propensity (and thus the elasticity) of page views with respect to listings under this linear specification. 17
  • Table 3: Website Usage Equations Measurement of Potential Page Views Page Views - Listings Bidders' Website Usage logarithm linear logarithm linear specification specification specification specification Estimate Estimate Estimate Estimate Independent Variables [St. Error] [St. Error] [St. Error] [St. Error] constant 11.20 7.64*e8 11.25 7.64*e8 [7.35] [3.29*e8] [8.32] [3.29*e8] number of listings 1.20 12.84 1.22 11.84 [0.26] [59.37] [0.30] [59.37] number of the rival site's listings -0.76 -143.59 -0.78 -143.59 [0.26] [59.37] [0.30] [59.37] R squared value 0.99 0.99 0.99 0.99 Number of observations 14 14 have greater incentives to increase listings via this feedback effect by setting lower fees. 3.3. Seller’s Valuation of the Number of Potential Bidders We now proceed to quantify the effects of the number of potential bidders on the seller’s expected auction revenue, based on specific functional forms of the seller’s expected auction revenue, R(Nj, mj), and the number of potential bidders, Nj. Due to no significant difference in the choices of auction formats and auction parameters between eBay and Yahoo!Auctions, we will, without modeling specific auction mechanisms, assume a (reduced-form) logarithm specification of a seller’s expected auction revenue as follows.25 ξ R j = aN b e j , j (4) 25 If the seller’s expected auction revenue is specified as a linear function, then we may not obtain a simple (linear) estimating equation as in equation (6) or (7). 18
  • where a and b are parameters, and ξj represents any exogenous factors which may affect a seller’s auction revenues week by week but are not correlated with the sizes of potential bidders. In equation (4), a reflects the influences of available auction mechanisms, while b measures the elasticity of the seller’s expected auction revenue with respect to the number of potential bidders. In principle, a and b are understood to be indexed by the listed item. As discussed in section 2, Yahoo!Auctions offers a little bit more flexible auction parameters. If these more flexible options make a substantial difference in expected auction revenues, the parameter a must differ across these two auctions sites. Later on, we will discuss how the value of b may be affected by this difference in the values of a. Based on the seller’s expected auction revenue function in (4), we will use the arbitrage condition of listing behavior in (2) to calculate b (the elasticity of the seller’s expected auction revenue with respect to the number of potential bidders). For the calculation of b, we can first consider an estimation procedure directly from the arbitrage condition of (2). However, the expected auction revenue, Rj, is not observed in this non-linear estimating equation. To calculate Rj, we need a structural econometric model for the Internet auctions which, to our knowledge, is not yet available (see section 4). Hence, to keep our analysis manageable, we will rewrite the arbitrage condition of (2), calculating an eBay premium equivalent to the difference in insertion and final value fees between eBay and Yahoo!Auctions for some representative items. As discussed in section 2, the most popular category by far has been the collectibles in Internet auctions. In the U.S. mint/proof coin sets auctions on eBay, as reported in Bajari and Hortacsu (2001), the average opening value of the traded coins was $16.28 while the average final value was $47. The Goldman Sachs study, based on more than 1000 closed auctions of eBay, estimated the average sale price (final value) to be $40 in the third quarter of 2001.26 Hence, in the paper, we will consider as a focal case that a seller lists an expected auction value of $50 on eBay with an 26 See “eBay Revenue Estimates Raised”, www.internetnews.com on August 23, 2001. 19
  • opening value between $10.00 and $24.99.27 We also assume that the seller will choose an opening value less than $25.00 on Yahoo!Auctions.28 Then, the arbitrage condition in (2) implies that a seller will expect about 4 percent less revenues from auctioning the same item on Yahoo!Auctions than on eBay.29 In other words, a seller expects about 4.2 percent (= 1/0.96 − 1) of the eBay premium on auction revenues. Hence, the arbitrage condition in (2) can be rewritten as follows. 1 − α = R y / Re , (5) where α = 0.04. If the seller’s expected revenue from auctioning his/her item on eBay is $40 or $100 with an opening value between $10.00 and $25.00, then α = 0.043 or 0.033, respectively.30 Note that these calculated eBay premiums are minimal because of a (likely) lower opening-value-to-final-value ratio and more flexible auction parameters on Yahoo!Auctions. As another reference case, therefore, we will consider that the eBay premium is 10 percent (and thus the corresponding α = 0.091), which is the eBay premium calculated in our sample of the Barber Quarter Dollar auctions (see Appendix for details). As discussed in section 2, we assume that the number of potential bidders is an increasing function of page views per listing: i.e., Nj = f(Uj / Lj). In what follows, we will use both logarithm and linear specifications, Nj = c ln(Uj / Lj) and Nj = c (Uj / Lj), where c is a positive constant and understood to be indexed by the listed item. Note that our sample of the Barber Quarter Dollar auctions suggests that a logarithm specification is a better fit (see Appendix for details). Substituting each of these specifications 27 The survey of Lucking-Reiley (2000) indicates that as of fall 1998, most of collectibles traded on Internet auctions are relatively inexpensive with median prices well below $100. 28 Recall that in our sample of the Barber Quarter Dollar auctions, sellers on Yahoo!Auctions set lower opening- value-to-final-value ratios. 29 If the final value on eBay auctions is $50 with an insertion fee of $0.55, then the seller will pay the total fee equal to $2.425 (= $25(0.05) + $25(0.025) + $0.55) on eBay auctions, while the seller will pay only an insertion fee of $0.35 (or $0.20) on Yahoo!Auctions. Hence the seller will pay $2.075 (or $1.925) more fees on eBay, which is approximately 4 percent of the final value. 30 For instance, if the final value on eBay auctions is $100 with an insertion fee of $0.55, then the seller will pay the total fee equal to $3.675 (= $25(0.05) + $75(0.025) + $0.55) on eBay auctions, while the seller will pay only an insertion fee of $0.35 (or $0.20) on Yahoo!Auctions. Hence the seller will pay $3.325 (or $3.175) more fees on eBay, which is approximately 3.3 percent of the final value. 20
  • into the seller’s expected auction revenue in (4), we can rewrite equation (5) to obtain the following estimating equations for the logarithm and the linear specifications: − ln(1 − α ) = b ln((ln(U e / Le ) / ln(U y / L y )) + ε , (6) and − ln(1 − α ) = b{ln(U e / Le ) − ln(U y / L y )} + ε , (7) where ε = ξe − ξy. Since ξj is not correlated with the numbers of potential bidders, we can apply an OLS estimation procedure for (6) and (7). Recall that, in principle, b depends on the listed item. As discussed above, for a given expected auction revenue of an item on eBay, we can calculate a corresponding eBay premium or α . Hence, in our estimating equations of (6) and (7), we will understand that b is indeed indexed by α . Our estimation results are reported in table 4. Before we proceed, we have to concede a couple of limitations in these estimations. First, we have only 7 weekly observations in our estimation.31 However, the standard errors of our estimates are very small even with this small number of observations. As an alternative, we also calculated the values of b (ignoring the error term ε) from the average values of the number of listings and page views, which turned out to be almost the same with our estimates of b (see table 4). Second, the dependent variable of the estimating equations (6) and (7) is a constant. In other words, both ln(Ue / Le) / ln(Uy / Ly) and (Ue / Le) / (Uy / Ly) are expected to fluctuate around a constant, and our data are supportive of this implication. Since our estimating equations are based on the reduced-form expected auction revenue function, the short coverage of the time period may help the model to fit the data. 31 We have 6 weekly observations for the period in concern (the third week of February to the first week of April) except the first week of March due to missing data of website usage for eBay. In addition, we add the observations of the forth week of April. 21
  • Table 4: Seller's Valuation of the Number of Potential Bidders Elasticity w.r.t. potential bidders regression results: calculations based estimate (st. error) on mean values eBay premium = 4.2% N_j = c ln(U_j/L_j) 0.044 (0.001) 0.044 N_j = c (U_j/L_j) 0.013 (0.0002) 0.014 eBay premium = 3.4% N_j = c ln(U_j/L_j) 0.036 (0.001) 0.036 N_j = c (U_j/L_j) 0.011 (0.0001) 0.011 eBay premium = 10% N_j = c ln(U_j/L_j) 0.102 (0.002) 0.103 N_j = c (U_j/L_j) 0.031 (0.0004) 0.032 eBay premium = 15% N_j = c ln(U_j/L_j) 0.150 (0.003) 0.152 N_j = c (U_j/L_j) 0.046 (0.0006) 0.046 Number of observations 7 We now discuss our estimation results. There are a couple of interesting patterns in table 4. First, the elasticity of the seller’s expected revenue with respect to the number of potential bidders (b) is increasing in the eBay premium (or decreasing in the corresponding α ). Due to the fee schedules shown in table 2, eBay premium is increasing in the expected auction revenue (see footnote 30), and thus we can infer that the elasticity of the seller’s expected revenue with respect to the number of potential bidders is higher for a more expensive item. Consistently, we can usually find more expensive items listed on eBay. Second, b is estimated to be higher under the logarithm specification of the number of potential bidders. In our focal case of the eBay premium equal to 4.2 percent, we find that 1 percent increase of the number of potential bidders induces 0.04 percent increase of a seller’s expected auction revenue under the logarithm specification but 0.01 percent increase under the linear specification. However, note that an 22
  • increase of page views per listing may induce more potential bidders under the linear specification. Hence the higher elasticity of the seller’s expected auction revenue with respect to the number of potential bidders does not necessarily mean a higher elasticity of the seller’s expected auction revenue with respect to page views per listing under the logarithm specification. Since Yahoo!Auctions offers more flexible options, the parameter a in the seller’s expected auction revenue in (4) may be different across these two auctions sites, and thus we will have (1 − α )(a e / a y ) instead of (1 − α ) in equations (5) to (7), where ay are ae the parameters for Yahoo!Auctions and eBay, respectively. If more options in auction parameters, such as more flexible duration days and auto extension, significantly increase the seller’s expected auction revenues, we have: ay > ae, and thus the eBay premium must be higher than 4.2 percent in our focal case. As pointed out in table 4, b is increasing in the eBay premium, and thus the above estimate of b in our focal case can be considered a lower bound for the true value of b. As discussed in Appendix, the eBay premium calculated in our sample of the Barber Quarter Dollar auctions is about 10 percent with the average sale price of $41.1 for the traded eBay auctions. Table 4 indicates that if the eBay premium is 10 percent, then 1 percent increase of the number of potential bidders induces 0.10 percent increase of a seller’s expected auction revenue (based on the logarithm specification) or 0.03 percent increase (based on the linear specification). As discussed in the previous subsection, the existence of the network effect between website usage and listings may explain why Yahoo!Auctions, which has substantially less listing, has a greater incentive to increase listings by setting lower fees. However, the low values of the estimated elasticity of a seller’s auction revenue with respect to the number of potential bidders in table 4, together with the low value (1.2) of the estimated elasticity of page views with respect to the number of listings in table 3, suggest that website usage per listing (and the number of potential bidders) may not rise rapidly via this network effect in Internet auctions. Therefore, once a pioneering firm established a huge network of 23
  • sellers and potential bidders, followers may not be able to catch up quickly even with substantially lower fees. 4. Implications from Our Findings In Internet auctions, a seller cannot restrict the number of potential bidders of a certain site but can choose an auction site based on the sizes of potential bidders. Finding 5 in section 3 indicates that the seller’s expected auction revenue increases with the number of potential bidders. This empirical finding, however, contrasts with the theoretical prediction in Levin and Smith (1994). Levin and Smith (1994), based on the equilibrium analysis of a symmetric endogenous entry with fixed costs, concluded that the expected revenue of any seller who uses his/her optimal mechanism, in either private-value or common- value auctions, decreases with the number of potential bidders in a mixed-strategy entry equilibrium when there are too many potential bidders. Hence, if the number of potential bidders is too big, the seller can be better off ex ante by restricting the number of potential bidders. Bulow and Klemperer (1996), on the other hand, showed a positive correlation between the seller’s expected auction revenue and the number of bidders if the number of bidders is exogenously given.32 However, the stochastic endogenous entry with fixed costs in Internet auctions is supported by Lucking-Reiley (1999) and Bajari and Hortacsu (2001) as well as our sample of the Barber Quarter Dollar auctions. Our intuition for this discrepancy between the theoretical prediction and our empirical findings in Internet auctions is as follows. The theoretical prediction depends on whether there is a sufficient number of potential bidders as well as whether potential bidders’ entry to specific auctions is endogenous with fixed costs of entry. We first suspect that any Internet auctions site, even eBay, may not have reached the 32 Bulow and Klemperer (1996) showed that very generally in a private-value auction and also in a wide class of common-value auctions, a simple ascending auction with no reserve price and N+1 (exogenously given) symmetric bidders is more profitable to the seller than any realistic auction with N of these bidders. 24
  • sufficient number of potential bidders (or n* in Levin and Smith 1994).33 As reported in Bajari and Hortacsu (2001), the average number of bidders for U.S. mint/proof coin sets was only 3 on (traded) eBay auctions. In our sample of the Barber Quarter Dollar auctions, the average number of bidders in the traded auctions turns out to be 4.5 on Yahoo!Auctions and 4 on eBay while the conjectured average number of bidders in all the closed auctions (including non-traded auctions) is 0.9 on Yahoo!Auctions and 1.9 on eBay. In reality, sellers pay more fees for featured auctions and set low opening values (usually with higher secret reserve prices) in order to attract more bidders in Internet auctions. However, more importantly, the discrepancy between the theoretical prediction and our empirical findings may be caused by the unique feature of the Internet auctions. Potential bidders on an Internet auctions site may choose to enter specific auction(s) among the (many) auctions listing similar items,34 while the same seller may list multiple similar items for auctions at the same time. As reported in Bajari and Hortacsu (2001), there were 100 to 400 U.S. mint/proof set auctions closed on eBay every day between September 28 and October 2, 1998. Hence the potential bidders’ entry behavior in Internet auctions may not be consistent with the underlying assumption of Levin and Smith (1994) in which potential bidders enter a single auction (at a time).35 This unique feature of Internet auctions invites more theoretical studies as well as empirical studies on the mechanism of Internet auctions. 33 In addition, our finding implies that Internet auctions may not be pure common-value auctions. In pure common- value auctions, expected auction revenues decrease with the number of potential bidders even if it is smaller than the sufficient number of potential bidders (n*). Refer to Matthews (1984) and Levin and Smith (1994). 34 The costs of entry may increase with the number of listings, which may lead to lower expected auction revenues for sellers on eBay than on Yahoo!Auctions. Therefore, like the effects of more flexible auction parameters in Yahoo!Auctions, the eBay premium calculated for the focal case in section 3 can be considered the lower bound for the true values in this case. On the other hand, if there are not many potential bidders and entry does not incur zero profit of bidders, then the costs of entry increasing with the number of listings may cause a negative effect of the number of listings on website usage. However, as indicated by our data on listings and website usage, the positive effect (higher probability to find a matching item due to more listings) seems to dominate this negative effects in Internet auctions. 35 In addition, the stochastic entry reported in Lucking-Reiley (1999) and Bajari and Hortacsu (2001) may be generated by things happening in bidders’ lives or by different numbers of competing auctions over time. 25
  • 5. Conclusion In this paper, we empirically examine the network effect in Internet auctions as a positive feedback effect between website usage and the number of listings. The (no) arbitrage condition of a seller’s listing behavior, combined with our unique data for eBay and Yahoo!Auctions, indicates that a seller’s expected auction revenue increases with page views per listing. Our data analysis further indicates that increased listings raises page views per listing. Hence, we can infer a positive feedback effect between the number of listings and website usage. The existence of this network effect may explain the different fee schedules between eBay and Yahoo!Auctions. Yahoo!Auctions, which has substantially less listings, may have greater incentives to increase listings via this feedback effect by setting lower fees. However, the low values of both the estimated elasticity of page views with respect to the number of listings and the estimated elasticity of a seller’s auction revenue with respect to the number of potential bidders suggest that website usage per listing (and the number of potential bidders) may not rise rapidly via this network effect in Internet auctions. Therefore, once a pioneering firm established a huge network of sellers and potential bidders, followers may not be able to catch up quickly even with substantially lower fees. Appendix As a supplementary data for the comparison of Internet auctions on eBay and Yahoo!Auctions, we collected the data of the Barber Quarter Dollar auctions closed between November 28, 2001 and December 4, 2001.36 As well known, coins are a popular collectible, and collectibles are the most popular category of Internet auctions. The Barber Quarter Dollar is a particular kind of Barber Quarter coins which have 74 regular issues between 1892 and 1916. The comparison of Internet auctions on eBay and 36 I thank the students who participated in this project in the course of E-Commerce at SNUY at Stony Brook in fall 2001. 26
  • Yahoo!Auctions, however, may not be straightforward because eBay stored all the closed (traded or not) auctions data while Yahoo!Auctions kept the data of only traded auctions. Keeping this restriction in mind, we will proceed to describe our observations. During the time period, we observe 188 closed auctions but only 90 traded auctions on eBay while there were 24 traded auctions on Yahoo!Auctions. In our sample, therefore, eBay has about 4 times as many traded auctions as Yahoo!Auctions, and the conversion rate (the number of traded auctions divided by the number of closed auctions) on eBay auctions is a little bit less than 50 percent. Note that the Goldman Sachs study found about 50 percent conversion rate on eBay auctions.37 Although we do not have any direct information of the conversion rate on Yahoo!Auctions, we can infer that the conversion rate is much lower on Yahoo!Auctions. It is reported that 22.5 percent of website users actually made purchases on eBay but only 4.4 percent on Yahoo!Auctions.38 Since, as discussed in section 2, unique visitors per listing are on average 1.2 on eBay and 2.3 on Yahoo!Auctions, we suspect that the conversion rate on Yahoo!Auctions is only 37.5 percent of the conversion rate on eBay. Hence, if the conversion rate on Yahoo!Auctions is about 19 percent (50 percent times 0.375), the total number of closed auctions on Yahoo!Auctions will be about 126. Considering that eBay has about 25 times as many total listings as Yahoo!Auctions (see section 2), we suspect that the ratio of traded auctions on Yahoo!Auctions to those on eBay is higher in our sample although eBay has a much greater variety of listings.39 In our sample, the average sale price (final value) of the traded auctions is $41.1 on eBay and $14.65 on Yahoo!Auctions, respectively. On average, the minimum bid level (opening value) is $26.47 on eBay and $6.55 on Yahoo!Auctions, respectively. Hence we can infer that more expensive coins of the Barber Quarter Dollar are traded on eBay, and the opening-value-to-final-value ratio is higher on eBay. In addition, since the average sale price and the minimum bid level of all the closed (including non-traded) eBay auctions are $54.2 and $47.13, respectively, we can infer that auctions with more expense coins or 37 See “eBay Revenue Estimates Raised”, www.internetnews.com on August 23, 2001. 38 See Nielsen//NetRatings & Harris Interactive eCommercePulse, May 2001. 39 At the top-level categories, eBay has 26 different categories while Yahoo!Auctions has 14 categories. 27
  • higher opening values are less likely to attract more bids. Although eBay has a higher average sale price, it does not necessarily imply that a seller’s expected auction revenue is higher on eBay since the quality of coins auctioned on these two sites may be different. To control quality differences of auctioned coins, we calculate the ratio of sale price to book value. Since the data of closed auctions usually contain the description of the auctioned coins, we can find matching book values from the Official National Bestseller 2002 Blackbook, Price Guide to United States Coins. We found these matching book values for 17 Yahoo!Auctions and 73 eBay traded auctions. The average sale-price-to-book-value ratio turns out to 1.35 on Yahoo!Auctions and 1.36 on eBay.40 At a first glance, this result seems surprising. However, if we consider a possible big difference of the conversion rate between these two sites, the expected sale-price-to-book-value ratio (prior to auctions) must be higher on eBay. For instance, as discussed above, suppose that the conversion rate is 19 percent on Yahoo!Auctions (and 50 percent on eBay). Suppose also that a seller’s reservation value for the coin is the book value. Then, the expected (prior-to-auction) sale-price-to-book-value ratio will be 1.07 on Yahoo!Auctions and 1.18 on eBay. Hence, in this case, the eBay premium will be about 10 percent. In our sample, the average number of bidders in the traded auctions turns out to be 4.5 on Yahoo!Auctions but 4 on eBay.41 Note that, in Bajari and Hortacsu (2001), the average number of bidders for the U.S. mint/proof coin sets was 3 on eBay auctions. Yahoo!Auctions, ceteris paribus, may attract more entries of potential bidders since it offers longer duration days and has less expensive coins auctioned with lower opening-value-to-final-value ratios.42 In our sample, the average duration day of the traded auctions is 9 days on Yahoo!Auctions and 6.7 days on eBay. However, if we consider a possible big difference of the conversion rate between these two sites, the average number of bidders in all the closed auctions may be higher on eBay. For instance, as discussed above, suppose that the conversion rate 40 In the case of eBay, it reduces to 1.08 after throwing out two outliers. 41 Note that the maximum number of bidders is 9 on Yahoo!Auctions and 13 on eBay. Note also that many traded auctions received only 1 bid on both sites: 45.6 percent (41.7 percent) of traded auctions on eBay (Yahoo!Auctions) had only 1 bidder. 42 As documented in Lucking-Reiley (1999) and Bajari and Hortacsu (2001), opening value is believed to be the most important determinant of bidders’ entry in Internet auctions. 28
  • is 19 percent on Yahoo!Auctions (and 50 percent on eBay). Then the average number of bidders in all the closed auctions will be 0.9 on Yahoo!Auctions and 1.9 on eBay. Since eBay has much more page views per listing (which were on average 148.4 on eBay but only 7.3 on Yahoo!Auctions), these observations of the number of bidders in our sample suggest that a better specification for the functional form of the number of potential bidders may be a (increasing) concave function such as a logarithm function, i.e., Nj = c ln(Uj / Lj) where c is a positive constant. 29
  • References Bailey, J. P. 1998. Electronic commerce: Prices and consumer issues for three products: Books, compact discs, and software. Working paper OCDE/GD(98)4, Organization for Economic Co-Operation and Development. Bajari, P., A. Hortacsu. 2001. Winner’s Curse, reserve prices, and endogenous entry: Empirical insights from eBay auctions. Working paper, Stanford University, Stanford, CA. Bakes, J. Y. 1991. A Strategic Analysis of Electronic Marketplaces, MIS Quarterly. 15(3) 295-310. . 1997. Reducing buyer search costs: Implications for electronic marketplaces. Management Sci. 43(12) 1676-1692. Brynjolfsson, E., M. Smith. 2000. Frictionless commerce? A comparison of Internet and conventional retailers. Management Sci. 46(4) 563-585. Bulow, J., P. Klemperer. 1996. Auctions versus negotiations. Amer. Econom. Rev. 86(1) 180-194. Chen, P., L. Hitt. 2001. Measuring switching costs and their determinants in Internet-enabled businesses: A study of the on-line brokerage industry. Working paper, The Wharton School, University of Pennsylvania, Philadelphia. PA. Clemons, E. K., I. Hann, L. Hitt. 2002. Price dispersion and differentiation in online travel: An empirical investigation. Management Sci. 48(4) 534-549. Ellison, G., S. F. Ellison. 2001. Search, obfuscation, and price elasticities on the Internet. Working paper, MIT, Cambridge, MA. Gandal, N., M. Kende, R. Rob. 2000. The dynamics of technological adoption in hardware/software systems: The case of compact disk players. Rand J. Econom. 31(1) 43-62. Goldfarb, A. 2002. Using household-specific regressions to estimate true state dependence at Internet portals. Working paper, University of Toronto, Ontario, Canada. Harstad, R. M. 1990. Alternative common values auction procedures: Revenue comparisons with free entry. J. Political Econom. 98(2) 421-429. 30
  • Johnson, E. J., W. W. Moe, P. S. Fader. 2001. On the depth and dynamics of online search behavior. Working paper, The Wharton School, University of Pennsylvania, Philadelphia. PA. Katz, M., C. Shapro. 1994. Systems Competition and Network Effects. J. Econom. Perspectives, 8(2) 93-115. Klemperer, P. 1999. Auction theory: A guide to the literature. J. Econom. Surveys. 13(3) 227-286. Levin, D., J. L. Smith. 1994. Equilibrium in auctions with entry. Amer. Econom. Rev. 84(3) 585-599. Loshe, G., S. Bellman, E. J. Johnson. 1999. Consumer buying behavior on the Internet: Findings from panel data. Working paper, The Wharton School, University of Pennsylvania, Philadelphia. PA. Lucking-Reiley, D. 2000. Auctions on the Internet: What’s being auctioned, and how? J. Industr. Econom. 48(3) 227-254. . 1999. Experimental evidence on the endogenous entry of bidders in Internet auctions. Working paper, Vanderbilt University. Matthews, S. A. 1984. Information acquisition in discriminatory auctions. M. Boyer, R. E. Kihlstrom, eds. Bayesian Models in Economic Theory. North Holland, New York. Park, S. 2002. Quantitative analysis of network externalities in competing technologies: the VCR case. Working paper, SUNY at Stony Brook, Stony Brook, NY. Rysman, M. 2000. Competition between networks: A study of the market for Yellow Pages. Working paper, Boston University, Boston, MA. 31