April 2007

April 2007

Insertion Fees and Their Role on eBay

Abstract

Success of an online auction website is in its appeal to the bidders. Without the bidders there will be no transactions. In an auction, the reservation price can act as a deterrent to bidder entry. The auctioneer can induce lower reservation prices through a rising insertion fee scheme. This will in turn increase the probability of an auction resulting in a sale. This work examines the impact the insertion fees have on the seller and buyer behavior on eBay using data generated by a natural experiment observed on eBay. The results demonstrate that the rising insertion fee structure of eBay induces sellers to lower their reservation prices, which in turn increases the bidding activity.

Mikhail I. Melnik*

Commerce Department

PO BOX 2201

Niagara University

Niagara University, NY 14109-2201

* Please contact Mikhail Melnik at [email protected] with any questions or comments.

1 Introduction

In the last decade we have witnessed a rapid rise in online commerce. Particularly, there

has been a surge in online consumer-to-consumer auctions, largely due to the success of eBay, an

online auction website (www.ebay.com). Yet the eBay’s story is not competition free, making

the success of eBay particularly interesting. Early on in its existence, eBay attracted competition

from such large and widely recognized online firms as Amazon.com and Yahoo.com. The

competition from Yahoo, a popular search engine website, presents an especially interesting

case. And yet, despite all this competitive pressure, eBay not only endured but rapidly grew and

its brand became synonymous to online auctions.

In this research we examine the impact the insertion fee structure has on the seller’s

behavior. This may offer some explanation to the role the insertion fee structure may have played

in the competition between eBay and Yahoo. Insertion fees are imposed on the seller by the

auctioneer at the time the auction starts and may be a function of the opening price and other

auction characteristics. Opening price is a term used on eBay and is synonymous to the

reservation price in the theoretical literature on auctions. The opening price effectively acts as an

insurance against receiving a low bid, a bid below the opening price.

The two firms implemented very different insertion fee schemes producing different

incentives. Although the data from the period of early competition between these websites is no

longer available, a natural experiment conducted by eBay on March 14 of 2007 enables us to

conduct an empirical investigation into the effects of alternative insertion fee schemes. On March

14 of 2007, eBay altered its insertion fee scheme for only one day. This alteration represented a

substantial change in the fee scheme. The altered scheme used on that day was similar to the fee

scheme used by Yahoo throughout most of its competition with eBay.

The economic theory of optimal auction design is well established and goes back to the

revenue equivalence principle established by Vickery (1961). For an overview of auction

literature see Klemperer (1999). Reservation prices received substantial attention in the

theoretical literature. But, the literature’s approach has been centered on the seller’s revenue

1

maximization. Myerson (1981) demonstrated that selection of the opening price can impact the

revenues to the seller in a significant way. Laffont and Maskin (1980) and Riley and Samuelson

(1981) provided theoretical derivations of optimal reserve price strategies for auctions with

private values and risk neutral bidders. Since then, the theoretical modeling has become extended

to common value goods (Levin and Smith, 1996) and even to cases with collusive behavior

(Robinson, 1985).

However, the topic of insertion fees and opening prices seems to be largely overlooked

in the empirical literature. One noticeable exception is the work of Park (2002) which examines

the insertion fee competition between eBay and Yahoo. But Park’s analysis centers on network

externalities present in an online community. Park argues that Yahoo is induced to lower the fees

due to the fact that its network is smaller than that of eBay. Thus the seller should expect to

receive a higher price for her item on eBay, and the competition implies that the cost of listing

the item on Yahoo must be lower. However, a low insertion cost may induce a higher opening

price, which in turn can further reduce the network by deterring the entry by additional bidders, a

point that requires an empirical investigation into the bidders’ behavior as well.

A closely related topic of hidden reserve prices has indeed received substantial interest

from empirical researchers and has also provided some useful insight into the bidder’s behavior.

But this literature has two important caveats. One, the primary focus remains on the seller’s

revenue problem. Two, the results seem to suggest different directional impacts. Katkar and

Reiley (2000) employ a controlled experiment on eBay to investigate the effects of hidden

reserve prices and show that they reduce the probability of sale and hence reduce the realized

price. Alternatively, Bajari and Hortacsu (2003) use a structural econometric approach to

demonstrate that when an optimal secret reserve price is selected, the seller is expected to

increase her realized price. In fact a high opening value, whether hidden or not, may on one hand

indicate a high valuation of the item and hence increase bidding, but on the other hand act as a

deterrent to bidders by either signaling that the seller may not be serious or by setting the price

above the bidder’s valuation.

2

But when compared to the other areas of empirical research using eBay data, the area

of hidden reserve prices has received a disproportional interest from researchers. For example,

one of the main areas of focus of the current literature has been the asymmetry of information in

online auction and the eBay’s ability to reduce this problem through its rating (feedback)

mechanism. This topic was a subject of an international conference at MIT in 2003 (see the

Conference report by Dellarocas and Resnick, 2003). A comprehensive overview of this

literature can be found in Paul Resnick et al (2006) and in Bajari and Hortacsu (2004).

The lack of research in the area of opening prices may in part be attributed to the lack of

changes in the eBay’s insertion fee structure. The insertion fees are rarely changed in part

because of their unpopularity with eBay users1. And when changed, the changes may affect the

different ranges of the opening prices differently2. However, over the past couple of years eBay

has conducted several temporary changes in the fee structure that can be viewed as appropriate

natural experiments for any such investigation. This paper focuses on one such natural

experiment.

1.1 The Natural Experiment

The insertion fees are charged at the time of the listing of the auction, and are non-

refundable3. The fees themselves are a function of the opening price of the item, possible hidden

reserve price, and any additional auction features, such as positioning the auction on top of the

search page, converting the auction’s title to bold font and etc. The insertion fee has a rising step

functional structure with respect to the opening price. The seller is required to specify the

opening price at the start of the auction. The opening price is synonymous to the reservation

1 Large volume of evidence of this can be found online in various forums or on news websites. See Broersma (2005), Kerner (2005) for evidence of the unpopularity of eBay’s fee increases. 2 The insertion fee change implemented on 2/22/06 adjusted insertion fees applied to only those auctions where the opening price was below 1 USD. While the change applied on 2/2/04 was much more complex, with fee changes being different for the different ranges of the opening price. 3 Exceptions include two cases. One, in the event the item does not sell, the seller has an option to relist the item at or below the previous listing price. If the auction sells, the second insertion fee will be refunded, but the original insertion fee still remains non-refundable. Two, if the buyer refuses to complete the transaction, the seller may request the refund of all fees.

3

price in the context of the theory of auctions. Table 1 presents the insertion fee structure used by

eBay at the time of this research (March-April of 2007).

On March 14, eBay temporarily simplified the insertion fee structure to a flat scheme

which effectively reduced the fees for all but the first step category by applying the starting fee

of only 0.20 USD. Interestingly, the fee reduction was not announced in advance, but the

announcement to most eBay users came at 3:00 am Eastern Time (midnight Pacific Time) in the

form of email sent to all registered eBay users (the actual announcement is available online at:

http://pages.ebay.com/promo/20cent07) . This minor point may suggest that the eBay’s move

was not aimed at attracting more sellers but rather at understanding the demand for their services

from the existing user base. Note that eBay operates the listings on its US website based on the

Pacific Time schedule.

1.2 Insertion Fees and the Competition Between eBay and Yahoo

On September 14th of 1998, Yahoo launched its competitor to eBay, Yahoo Auctions.

Yahoo’s popularity among internet users was already high. Yahoo for several years had been a

popular search engine and now using its established brand recognition Yahoo was trying to enter

into other markets. And in 1998 Yahoo identified online consumer-to-consumer auctions as one

of these target markets for its expansion. In September of 1998, the eBay user base was less than

one million and the entry by a competitor whose brand was so widely recognized was a real

threat4.

From the start, Yahoo attempted to make its auction website more appealing to the

sellers by making the site free to both, sellers and bidders. The zero insertion fee became a

trademark of the Yahoo auctions. The perspective of listing their items on one of the most

popular websites free of charge attracted many sellers but it also lead to higher listing prices,

which scared away bidders. At its first anniversary of Yahoo Auctions, Yahoo pointed out the

4 According to Yahoo’s own press releases (http://docs.yahoo.com/docs/pr/release408.html) its user base (for the entire company, not just the auction segment of the business) exceeded 100 million in September of 1999, just one year after the launch of the auction website. This was more than thirteen times the multiple of the eBay user base at the end of 1999.

4

increasing popularity of the website with sellers. The number of active listings at any given date

in September of 1999 averaged at more than 725,000, a stunning comparison, when compared to

the 20,000 available in September of 19985. Reiley (2000) made a comparison between eBay,

Yahoo and Amazon auctions. Reiley reported that in the summer of 1999 eBay’s popularity was

much greater that that of its competition, and eBay averaged 340,000 auction closings per day,

while Yahoo was at only 88,000. Park (2002) provided a more favorable for Yahoo comparison

when reporting based on Nielsen/NetRatings surveys that Yahoo had about 4 million listings as

compared to 5.6 million on eBay in fall of 2000. But the volume of auctions did not suggest an

equal volume of successfully completed auctions. It’s important to note that unlike eBay, Yahoo

does not preserve the data on auctions that receive no bids, but examining the overall quantity of

available active listings on a given day versus a number of successfully completed listings on a

given day suggests a completion rate that is substantially less than 50%.

Eventually, Yahoo abandoned the practice of zero insertion fees but only for a while.

First, the insertion fees were introduced on optional features. On November 22 of 1999, Yahoo

introduced insertion fees applied to various optional features of the auction, but a basic insertion

fee associated with listing the item and structured as a function of the opening price, was not

introduced until January of 2001. And when introduced, the basic insertion fees on Yahoo were

considerably smaller than on eBay, ranging from $0.20 to $2.25 per auction, but some sellers

received credits and discounts resulting in free listings or in the insertion fee of $0.05 only (for

comparison with eBay fee structure used at the time please see Table 2). Interestingly enough,

there is plenty of evidence in various online public forums of how “failing” the policy of Yahoo

was6. This policy might have been effective had it been employed in September of 1998, but in

5 In the same press release Yahoo argued: “Recognizing that auctions provide them with an excellent way to develop new and lasting customer relationships and build their brand on the Web, Yahoo! Auctions' sellers gain immediate access Yahoo!'s 80 million unique users per month. Through dedicated programs, sellers can establish a compelling Web presence and access a sales channel that is free, international, and always open.” (See Yahoo press release #378, available online at http://docs.yahoo.com/docs/pr/release378.html). 6 One posting on www.AuctionBytes.com read: “Last summer I found out that Yahoo had a "secret" program that allowed sellers to list auctions for free and just pay a final value fee of 5%. This explains why some sellers have 100s of auctions starting at high values but hardly any sales” (posted by Reed on February 16, 2005 (http://www.auctionbytes.com/forum/phpBB/viewtopic.php?t=13504&view=next&sid=1fc553bfe8da54bf84de438ed37a326a)

5

January of 2001, when the eBay user base stood at nearly 30 million, it was too little too late. In

June of 2005, Yahoo abandoned the use of basic insertion fees and the auctions became free

again.

Although this research is focused on the insertion fees, it is important to mention that it

is not the only dimension in the competition between Yahoo and eBay. Another important aspect

in this competition, at least at its start, was the reputation mechanism. The reputation mechanism

enabled eBay to overcome the trust problem in its community. Yahoo replicated this mechanism

on its own auction website from its very launch in September of 1998. But, as Melnik and Alm

(2005) argue that the rating mechanism can also establish a barrier to entry by creating a cost for

a seller with an established (high) rating in one online community to switch to another

community where he has no rating established yet. In their argument, the eBay rating acts as a

club public good, where the term club is confined to the eBay community. Yahoo seemed to

recognize that property and in 1998 attempted to invite eBay users to switch to their website by

offering them to transfer their rating from eBay, a move that was quickly blocked by eBay. After

being threatened by eBay with a possible legal repercussion, Yahoo abandoned that idea.

Although, established ratings can present a barrier to entry for a competing community,

it can be argued that in September of 1998, at that early stage in the development of eBay, when

eBay only had 916 thousand users, the barrier to entry was minimal.

2 Data

The opening price acts as a form of protection from a relatively low bid; a bid below the

opening price. The probably of receiving a bid below the opening price depends on the

distribution of prices. eBay keeps on its website all auctions that closed within the last two

weeks. Any eBay user, including a potential seller, has access to this information and can

therefore observe past prices and successful completion rates on these recently closed auctions.

To evaluate this possible impact the price distribution may have on the choice of the opening

6

price, a study must include observations drawn from different price distributions, i.e. different

items.

However, the inclusion of observations on different items inevitably will open a number

of issues. For instance, goods may have substantial differences in shipping methods, payment

methods, goods may differ in terms of the degree of heterogeneity. All of these issues, if not

adequately controlled for, can bias the results of the study. Perhaps the most difficult of these

issues to address is the degree of heterogeneity. Most goods sold on eBay will experience this

problem and it will be different for each good. Controlling for heterogeneity may be difficult, as

different control variables may be needed for different goods. And yet there is empirical

evidence that item specific heterogeneity has impact on other possible control variables, for

example, Melnik and Alm (2005) argue that the role of the seller’s reputation on eBay depends

on the degree of heterogeneity of the good. And since the seller’s reputation is an important

variable in this study as well, any inability to account for heterogeneity may bias the results.

Lastly, the one-day nature of the experiment requires that the good used in the study

generate a high volume of observations in a short period of time. Although the number of

transactions on eBay has surpassed 588 million listings in QI-20077, the number of transactions

for any particular good is considerably smaller, and a number of observations for a good that

does not exhibit substantial heterogeneity is even lower.

For these reasons, the study focuses on collectable US Morgan Silver Dollar coins in

Almost Uncirculated condition. These coins were minted between 1878-1904, and again in 1921

by four different US mints, presenting us with nearly 100 different price distributions; one for

each unique coin defined based on the year and mint mark. At the same time, all of these coins

do not exhibit any substantial variation in terms of the methods of payment and shipment. For

instance, 1880 and 1890 dollars are usually sent using the same shipping carriers and similar if

not identical methods.

7 See eBay’s quarterly financial reports available online at http://investor.ebay.com/results.cfm

7

In addition, by focusing on a particular grade (AU – Almost Uncirculated) we have

attempted to control for item-specific heterogeneity. All of the coins in the AU category exhibit a

very limited variation in terms of their quality8. This particular grade category was selected for

two reasons. One, the number of listings in this category was considerably larger than in most

other grade categories of Morgan Dollar coins. Two, the variation in catalog values within this

category appears to be much lower than in the Mint state category. Mint state coins have been

used previously in empirical studies of eBay auctions (see Reiley et al., (1999) for an example of

a study using coins in mint state).

And lastly, the US collectable coins are some of the most commonly traded items on

eBay. The Morgan Dollars represent one of the largest subcategories within the coin category,

with over 11,000 listings available on any day during the period of our data collection.

The data was collected on all AU Morgan dollar auctions started between March 11 and

March 31. Overall, the dataset includes 2641 observations generated by 674 unique sellers.

Tables 2 and 3 present the basic summary statistics for the dataset.

2.1 Construction of Variables

The dataset includes all auctions started between March 11 – 31 in the Morgan Dollar

category, with AU in the title of the auction. Most of these auctions are single-coin listings.

However, the dataset also includes 261 multi-coin auctions, where coins from different

years/mints are sold together as a lot.

FeeAlternative is the primary variable of interest in this study and is a binary variable

assuming the value of one if the auction was started on March 14, and zero otherwise. Simple

observation of the summary statistics shows the disproportional number of listings on March 14.

March 14 is just one of the twenty-one days covered in the study, and yet it accounts for about

14.3% of all single coin auctions.

8 The Standard Catalog of World Coins (Krause and Mishler, 2001) defines AU coins as coins where “all detail will be visible. There will be wear only on the highest point of the coin. There will often be half or more of the original mint luster present.”

8

ClosingPrice represents either the winning bid, in the event the auction receives at least

one bid, or is equal to the starting value otherwise. Table 2 also includes this statistic for

successfully completed auctions only. OpeningPrice represents the starting price chosen by the

seller.

Price distribution characteristics include the mean price (Pmean) and standard deviation

(Std_Dev) computed on successfully completed auctions for the coin (sold coins with the same

year/mint combination).

The dataset also includes two measures of the seller’s reputation on eBay, ln(Rating+1)

– the overall rating of the seller, and RatingPercent – percentage of total rating that is positive.

Other variables include; Certified – a binary variable which assumes the value of one if the coin

has been graded and certified by a third party, CoinCount – a measure of the scarcity of this

particular coin on eBay, and is computed as simply the total number of observations of auctions

for this particular coin in our dataset, Weekend – controlling whether the auction ended on a

weekend (Saturday and Sunday), or on any other day of the week. Numerous empirical

investigations demonstrate that the bidding activity increases sharply close to the expiration of

the auction (Wilcox, 2002; Schindler, 2003; Roth and Ockenfels, 2002). Although their research

investigates the bidding in the last minutes of the auction, there exists empirical evidence that

auctions closing on a given day of the week, even when controlled for the closing time also

exhibit different bidding activities (Melnik and Alm, 2002). Melnik and Alm argue that the

number of possible bidders may be different on weekends as opposed to any other day of the

week. See Tables 2 and 3 for summery statistics.

3 Empirical Investigation

The empirical analysis is divided into two parts. Firstly, the effect of the change in the

insertion fees on the seller’s choice of the opening price is investigated. Secondly, the effect the

opening price has on the buyer’s behavior is analyzed.

9

Table 4 presents the results of the OLS estimation with robust standard errors of the

model where the dependent variable is the number of listings started on a given day9. The control

variables in this regression are limited to the binary day of the week variables. In Specification II,

the Saturday and Sunday auctions are combined into one binary variable, Weekend..

Specifications I – V are limited to the single coin auctions only, while specification VI examines

the multi-coin auctions offered in the AU Morgan Dollar category. Specification IV is restricted

to the coins with Pmean of 50 dollars and lower, while Specification V is estimated on auctions

of the coins with Pmean exceeding 50 dollars.

The coefficient on FeeAlternative is positive and statistically significant at 99.9% level

and higher across all of these specifications. The magnitude of the coefficient remains nearly

constant in all specifications for the single-coin auctions. It is important to note that in addition to

the reported specifications a number of additional specifications were estimated including

LnRating, Rating_Percent, Pmean variables, and all of these provided similar results in terms of

the coefficient on FeeAlternative. The results suggest that the reduction in the fee resulted in a

sharp increase in the number of listings in the Morgan Dollar category. The coefficients on the

binary day of the week variables suggest that Saturdays and Sundays have the greatest number of

listings.

The average number of listings per day (excluding March 14) in the dataset is 99.8, and

given the average magnitude of the coefficient on FeeAlternative in Specifications I through III,

the change in the fee induced a rise in the response in the number of listings by about 240%.

Given the average value of the opening price in the dataset, the average applicable insertion fee

is 2.40 USD, which was reduced on March 14 by about 92% to 0.20 USD. This observation

implies that the elasticity of demand for the eBay service by the eBay’s customers, in this case

the sellers of Morgan Dollars on eBay, is about -2.6.

9 White’s (White, 1980) method is used in the construction of the robust standard errors and corrects for the heteroscedasticity problem, which in our setting arises from the use of observations drawn from different distributions of prices.

10

Multiple coin auctions are considerably less common in this category, and the coefficient

on FeeAlternative is substantially smaller in magnitude, however, the relative impact of the fee

reduction is actually much larger. One possible explanation is that the multi-item auctions have

higher values and hence the benefit of listing them on a day with lower fees is much higher. In

addition, there may be greater uncertainty associated with the completion of a multi-item auction.

For instance, multi-item auctions may include coins from different years/mints, which may

reduce the number of potential bidders, as some bidders may already have one or more of these

coins. In turn, this may induce the seller to increase the opening price, further underscoring the

benefit of the fee reduction.

Table 4 examines the choice of the opening price. Here, the opening price is estimated

using the OLS methodology with robust standard errors. FeeAlternative appears to be

statistically significant in all but the last specification.

Specification I has only one control variable, the mean price of the coin. The justification

for including the Pmean is that the average price may serve as the value of the coin, and the

seller may select to protect herself from receiving bids that are substantially lower than the

average price. Specification II includes a number of additional control variables. Specifications I

and II are performed on the entire single-item dataset.

Specification III is restricted to the coins with Pmean of 50 dollars and higher (the

relatively expensive Morgan dollar coins). Specifications IV – VII are matched to the insertion

fee structure of eBay in terms of the selection of the coins based on Pmean. Although the eBay

fee structure is rising, it is not a continuous function and assumes the form of a step function. In

Specification IV, the regression is limited to coins with Pmean between 0 and 25 USD. This

represents the first three steps in the insertion fee structure. The primary reason for combining

these three groups together is that there is an insufficient number of auctions to perform separate

estimations on the first two steps. Specification V is limited to the fourth step in the insertion fee

structure and is performed on the coins with Pmean ranging between 25 and 50 USD.

Specification VI is performed on the fifth step (50 < Pmean < 200), and Specification VII

11

combines the last two steps (200 < Pmean < infinity). Again, the limited number of observations

forces these two groups to be combined together.

The last two specifications are performed on two particular coins that have the highest

number of observations and also come from two different ends of the distribution of Pmean.

Specification VIII is limited to the observations on 1884 S AU Morgan Dollar coin, which has

the catalog value of 550 USD and the mean price on eBay of 140.30 USD10. Specification IX is

limited to the observations on 1921 (P) AU Morgan Dollar coin, with the catalog value of 16

USD and the mean price on eBay of 19.10 USD.

The results suggest that the reduction in the insertion fee has a positive and statistically

significant impact on the opening price. However, the overall explanatory power of these

regressions remains relatively low (R2 values do not exceed 0.436). The results also suggest that

the impact is rising as the mean price, a proxy for the coin’s value, increases. This is seen from

the difference in the coefficient on FeeAlternative between Specifications II and III, between

Specifications IV through VII, and also between Specifications VIII and IX. The lack of

statistical significance in Specification IX suggests that the change in the fee structure has little

impact on the choice of the opening price by sellers of low-value coins.

The results also suggest that sellers take into account the characteristics of the closing

price distribution when making their listing decisions. The coefficient on Pmean is statistically

significant and positive in Specifications I – III, suggesting that more valuable coins tend to be

listed with higher opening prices. Recall that the opening price acts as protection against a low

bid, and the need for such protection increases with the valuation of the item. Indeed, some

selection is observed in terms of coins listed on March 14. Pmean for auctions listed on March

14th is greater than Pmean for the overall dataset, suggesting that the coins listed for sale on the

day with the lower insertion fee tend to have higher values.

The coefficient on the Std_Dev is also positive and statistically significant in

Specification II and III. As Std_Dev increases, the uncertainty about the auction outcome

10 The catalog value is obtained from the PCGS website (www.PCGS.com/prices). PCGS (Professional Coin Grading Service) is one of the major US coin grading companies.

12

increases, and the results seem to show that the sellers respond to that by increasing the opening

values. Both, the coefficients on Pmean and on Std_Dev lose their statistical significance in

Specifications IV – VII. One explanation for that is the restriction that is imposed on Pmean in

these specifications. The specifications IV – VII are restricted to a much smaller number of price

distributions and the variability in Pmean and Std_Dev is reduced. Note that coins with

CoinCount=1 are excluded from all of the specifications in Table 4.

Auctions ending on Saturdays and Sundays appear to start with lower opening values.

The coefficient is statistically significant in all specifications except for the low value coins in

Specification IV and IX. Numerous studies show that the bidding activity on eBay is a function

of the day of the week, and it tends to increase on Saturdays and Sundays. Most auctions in our

dataset are 7 day listings, and hence an auction listed on a Saturday is expected to end on the

following Saturday. Increased bidding activity on weekends may suggest that the probability of

the auction closing with a low bid is reduced, and hence the need to raise the opening price

(which can result in a higher fee), is reduced also.

Certified coins tend to be listed with much higher opening prices. For the average coin in

this dataset, certification by a third party induces the seller to raise the opening value by 68.17

USD, which is more than double the average price on successfully completed auctions in the

dataset (55.66 USD). Certification does not just guarantee authenticity of the coin, but also

insures that the coin is graded professionally. The cost of certification services differs from

company to company, but the most common grading service providers in the dataset (ANACS

and PCGS) charge in excess of 16 USD per coin not inclusive of shipping and insurance

charges11. But it is important to note that the value of the certification is not in its cost but in the

fact that it insures the grade and authenticity of the coin. Certification is omitted in Specification

IX due to the lack of certified 1921 coins in the dataset.

11 PCGS charges 30 USD/coin for its regular service (http://www.pcgs.com/grading_list.chtml), while ANACS charges 16 USD/coin for its regular service (http://www.anacs.com/contentPages/Services.aspx). These charges do not include shipping and insurance. In each case, the certification process is expected to take 15 business days.

13

The estimations also control for the reputation of the seller. The expectation is that

sellers with higher reputation values will be less likely to receive low bids and hence the need for

higher opening values will be reduced. The results don’t seem to offer much support to this

notion. The coefficient on RatingPercent does not seem to be statistically significant in almost all

of the specifications, and the coefficient on ln(Rating+1) is not significant in the case of coins

with Pmean<25, but more interestingly, changes the sign in Specification V.

CoinCount is included as a measure of the overall supply of the coin on eBay. Higher

values of CoinCount indicate increased competitiveness and may induce the seller to lower the

opening price. Although the sign of the coefficient is consistent with this hypothesis, the

coefficient is not statistically significant in all of the specifications.

Table 5 examines the impact the opening price has on the probability of sale. Probit

model is used in estimating these effects. The first specification is estimated on all single-coin

auctions. Specification II includes an additional variable, Shipping, which is likely to influence

the buyer’s decision. The buyer is likely to evaluate the total cost of the transaction rather than

just the price – the winning bid. However, nearly 30% of all observations are missing the

shipping data. This is due to the fact that many sellers do not specify the exact shipping charges

in the auction description. This reduction in the observations makes the inclusion of Shipping

into specifications III and IV performed on two most frequently observed coins impossible.

The results indicate that as the opening price increases relative to the coin valuation, the

probability of sale diminishes. Based on the average opening and winning bids for auctions

started on all days except March 14, the ratio is 0.876. A 10% increase from the mean in the ratio

of the opening price to the average sales price to 0.964 should be expected to reduce the

probability of successful completion by about 2.8%12. Specifications III and IV suggest that the

effect is diminishing with the value of the coin. Specification III, performed on the coin with the

average sales price on eBay of 140.30 USD suggests that for a five dollar increase in the opening

value (3.6% of the average price), the probability of sale will decrease by only 1%. While

12 Computed using the average value of the coefficient on OpeninBid/Pmean from Specifications I and II of Table 3.

14

Specification IV performed on the coin with the average sales price of 19.10 USD, suggests that

a 5 USD increase in the opening value (26% increase relative to the average price) will reduce

the probability of a successful sale by 1.5%.

Auctions closing on a weekend are expected to be more likely to close successfully. The

results of the Specifications I – III offer some support for that hypothesis, but the last

specification has a negative coefficient on Weekend. The sign on the Ln(Rating+1) coefficient is

positive, suggesting that the rise in the level of the reputation of the seller increases the

probability of a successful completion of the auction. This topic has received substantial attention

in empirical literature over the last few years. However, the coefficient is significant in only one

of the specifications. Certification also does not appear to have a significant impact on the

probability of sale. Recall that certified coins have much higher opening values. Shipping does

appear to have the expected effect. The coefficient is negative and statistically significant at 99%.

It appears that an increase in the shipping costs of one dollar, which represents 1.8% of the

average price of successfully completed auctions, will reduce the probability of sale by 2.4%,

suggesting that the demand is elastic with respect to shipping charges.

Table 6 presents the analysis of the impact the opening price has on the number of bids.

Specification I is estimated using the OLS methodology, while Specifications II and III are

estimated using the Poisson model. All estimations are performed using robust standard errors.

Intuitively, the OLS model is not a preferred choice as the distribution of bids does not satisfy the

assumption of normality, with a disproportional concentration of observations with low numbers

of bids. The Pseudo R2 comparison also favors the Poisson model. The results demonstrate that

the number of bids is indeed a function of the opening price. As the ratio of the opening price to

the mean price increases, the number of bids decreases. Although the coefficient is statistically

significant at 99% confidence level, the magnitude of the coefficient suggests that the impact

near the mean value in the dataset is relatively small. Based on the average magnitude of the

coefficient from specifications II and III, a 10% increase in the ratio from its mean value of 0.84

will be expected to reduce the number of bids by only 0.16.

15

4 Conclusions

Success of an online auction website is in its appeal to the buyers. Without the buyers

there will be no transactions. The insertion fee structure of eBay creates the incentives for sellers

to list their items with lower opening prices, which in turn attracts buyers. This insertion fee

scheme may have played an important role in the growth of eBay and in the success in its

competition with Yahoo, as it made the eBay marketplace more attractive to the buyers.

The empirical findings demonstrate that the sellers on eBay respond to the rising

insertion fee scheme by lowering their opening values. The findings also show that the bidders

respond to lower opening prices by increased bidding activity. As a result, the likelihood of a

transaction taking place increases.

However, it is also important to point out that these results need to be interpreted with

caution, as the change in the fee structure was only limited to one day. From this temporary, and

most importantly, known to be temporary in advance change in the fee structure we cannot

conclude with certainty that the observed changes in behavior will hold in the event of a

permanent change in the insertion fee scheme.

16

References:

AuctionBytes.com, cited forum page can be found on line at http://www.auctionbytes.com/forum/phpBB/viewtopic.php?t=13504&view=next&sid=1fc553bfe8da54bf84de438ed37a326a

Bajari, P., and Hortacsu, A. (2003). “The Winner’s Curse, Reserve Prices and Endogenoue Entry: Empirical Insights from eBay Auctions,” RAND Journal of Economics, Volume 34, Summer 2003, pp. 329-355.

Bajari, P., and Hortacsu, A. (2004). “Economic Insights from Internet Auctions,” Journal of Economic Literature, Volume XLII, pp. 457-486.

Broersma, M. February 7, (2005). “Amid Customer Backlash, eBay Reduces Some Fees,” eWeek.com, http://www.eweek.com/article2/0,1895,1761416,00.asp

Dellarocas, C., and Resnick, P. (2003). “Online Reputation Mechanisms: A Roadmap for Future Research” Summary Report of the First Interdisciplinary Symposium on Online Reputation Mechanisms. Available online at http://www2.sims.berkeley.edu/research/conferences/p2pecon/papers/s8-dellarocas.pdf

eBay, March 13, (2007), Fee Announcement, http://pages.ebay.com/promo/20cent07

eBay, (1998 – 2007)Quarterly Financial Reports, http://investor.ebay.com/results.cfm

Katkar, R. and Reiley, D. (2006). “Public versus Secret Reserve Prices in eBay Auctions: Results from a Pokemon field Experiment,” Advances in Economic Analysis & Policy: Vol. 6 : No. 2.

Kerner, S. “eBay Reduces Some Fees” (February 7, 2005). Internetnews.com, http://www.internetnews.com/ec-news/article.php/3469471

Klemperer, P. (1999). “Auction theory: A Guide to the Literature,” Journal of Economic Surveys, 13, 227-260.

Krause, C., Mishler, C. (2001). Standard Catalog of World Coins: 1801-1900. Iola, WI: Krause Publications, 3rd edition.

Laffont, J., and Maskin, E. (1980). “Optimal Reservation Price,” Economic Letters, 6, 309-313.

Levin, D., and Smith, J. (1996) “Optimal Reservation Prices in Auctions,” Economic Journal, 106, pp. 1271-1283.

Melnik, M. and Alm, J. (2002). “Does a Seller’s eCommerce Reputation Matter? Evidence from eBay Auctions.” Journal of Industrial Economics, Volume L, Number 3, pp. 337-349.

Melnik, M., and Alm, J. (2005).”Seller Reputation, Information Signals, and Prices for Heterogeneous Coins on eBay,” Southern Economic Journal, Volume 72, Number 2, pp 305-327.

Myerson, R. (1981). “Optimal Auction Design,” Mathematics of Operations Research. 6, pp. 58-73.

17

Park, S. (2002). “Website Usage and Seller’s Listing in Internet Auctions,” working paper, SUNY Stony Brook.

Reiley, D. (2000). “Auctions on the Internet: What’s Being Auctioned and How?” Journal of Industrial Economics, Volume 48, Number 3, pp. 227-252.

Reiley, D., Bryan, D., Prasad, N., Reeves, D. (1999) “Pennies from eBay: the Determinants of Price in Online Auctions.” Journal of Industrial Economics, Forthcoming.

Resnick, P., Zeckhauser, R., Swanson, J., Lockwood, K. (2006). “The Value of Reputation on eBay: A Controlled Experiment,” Experimental Economics, Volume 9, Number 2, pp. 79-101.

Riley, J. and Samuelson, W. (1981). “Optimal Auctions,” American Economic Review, 71, pp 381-392.

Roth, A. and Ockenfels, A. (2002). “Last-Minute Bidding and the Rules for Ending Second-Price Auctions: Evidence from eBay and Amazon Auctions on the Internet,” American Economic Review, 92:4, pp. 1093-103.

Schindler, J. (2003). “Late Bidding on the Internet.” University of Vienna, working paper.

Vickery, W. (1961). “Counterspeculation, Auctions, and Sealed tenders,” Journal of Finance, 16, pp. 8-37.

Wilcox, R. (2000). “Experts and Amateurs: The Role of Experience in Internet Auctions,” Marketing Letters, Volume 11, Number 4, pp. 363-74

Yahoo.com, (November 1, 1999). “Yahoo Auctions Surpasses One Million Simultaneous Daily Auctions” Yahoo press release, http://docs.yahoo.com/docs/pr/release408.html

18

Table 1: Insertion Fee Schedule on eBayStarting or Reserve Price

Insertion Fee

$0.01 - $0.99 $0.20

$1.00 - $9.99 $0.40

$10.00 - $24.99 $0.60

$25.00 - $49.99 $1.20

$50.00 - $199.99 $2.40

$200.00 - $499.99 $3.60

$500.00 or more $4.80

Table 2: Descriptive Statistics: Single-Coin AuctionsAll Observations Auctions Starting on March 14

variable meanStd. dev min max mean

Std. dev min max

                 ClosingPrice 87.109 141.301 0.01 999.99 163.728 175.368 0.99 999.99     ClosingPrice (sold coins) 55.656 87.741 5.12 999 71.225 90.000 9.99 510     ClosingPrice* (sold coins) 54.661 87.529 5.12 999       OpeningPrice 62.131 138.489 0.01 999.99 150.153 178.752 0.01 999.99     OpeningPrice* 47.879 125.172 0.01 999.99       Pmean 72.899 105.931 16.871 999 96.881 94.416 16.871 714.997     Std_Dev 33.398 37.267 0 225.676 46.027 38.958 0 147.569     CoinCount 41.007 28.130 1 131 37.962 21.671 1 131     ln(Rating+1) 7.433 1.695 0.693 10.816 7.848 1.509 2.944 10.758     Rating_percent 99.788 0.347 96.7 100 99.771 0.363 97.4 100     Bids 5.085 5.306 0 41 2.170 4.879 0 41     Certified 0.078 0 1 0.142 0 1     Completed 0.746 0 1 0.322 0 1     Weekend 0.308 0 1       FeeAlternative 0.143 0 1       Number of observations 2318   323                   

* Excludes auctions listed on March 14.

19

Table 3: Descriptive Statistics: Multi-Coin AuctionsAll Observations Auctions Starting on March 14

variable meanStd. dev min max mean Std. dev min max

                 ClosingPrice 183.312 160.533 0.01 1499 215.405 116.591 20.95 592     ClosingPrice (sold coins) 114.573 122.484 14.25 549.99 121.063 79.252 20.95 295     OpeningPrice 157.055 166.549 0.01 1499 214.739 117.638 0.99 592     ln(Rating+1) 8.513 1.772 0.693 9.59 9.164 1.455 2.944 9.589     RatingPercent 99.822 0.160 98.3 100 99.798 0.083 99.4 100     Bids 3.107 6.145 0 34 0.150 0.702 0 6     Certified 0.000 0 0 0.000 0 0     Completed 0.414 0 1 0.080 0 1     Weekend 0.215 0 1       FeeAlternative 0.383 0 1       Number of observations 261   100                   

20

Table 4: Number of Listingsvariable I II III IV V VI

             FeeAlternative 238.173*** 249.114*** 232.644*** 233.439*** 230.723*** 85.08***  (0.558) (0.471) (0.636) (0.760) (1.133) (1.003)

   Weekend 31.658***  

  (1.034)  

Monday -57.824*** -58.032*** -57.415*** -1.354

  (0.861) (1.073) (1.444) (1.243)

Tuesday -34.246*** -33.722*** -35.378*** -6.142***

  (0.709) (0.879) (1.185) (1.133)

Wednesday -21.424*** -22.066*** -19.779*** -0.606

  (0.747) (0.903) (1.313) (1.427)

Thursday -41.987*** -41.836*** -42.256***-11.617***

  (0.899) (1.111) (1.532) (1.060)

Friday -36.229*** -36.940*** -33.579*** -5.115***

  (0.978) (1.119) (2.036) (1.533)

Saturday -15.013*** -14.551*** -16.251*** -7.288***

  (2.001) (2.299) (4.099) (1.210)

   Constant 108.827*** 120.479 135.78*** 135.626*** 136.056*** 15.526***

  (0.557) (93.865) (0.392) (0.487) (0.664) (1.015)

   N 2349 2349 2349 1548 801 267

   R2 0.928 0.955 0.966 0.945 0.982 0.993

21

Table 5: Opening Pricesvariable I II III IV V VI VII IV V

FeeAlternative 77.548*** 63.509*** 82.859*** 27.003*** 51.187** 79.045*** 129.392*** 89.571* 6.949

  (8.465) (8.123) (13.849) (7.106) (21.615) (13.700) (49.418) (51.986) (5.449)

Weekend-

20.552*** -53.528*** -1.536 -15.578*** -57.572*** -41.108 -95.255** -1.175

  (4.081) (12.431) (2.132) (5.239) (10.033) (60.250) (39.889) (2.919)

ln(Rating+1) 2.405** 11.733*** 0.985 -4.178** 5.448** 48.957*** 7.283 3.150

  (1.183) (3.104) (0.784) (1.765) (2.780) (15.222) (9.535) (2.160)

RatingPercent 6.841 0.881 6.658*** 19.538*** 16.043 -161.597 -16.791 2.307

  (4.289) (20.151) (1.410) (6.870) (17.878) (162.660) (52.980) (1.879)

Certified 68.168*** 85.873*** 27.680 44.524** 66.768*** 151.935*** 254.829**  

  (12.645) (17.050) (19.196) (20.595) (15.031) (50.293) -108.41  

Pmean 0.858*** 0.654*** 0.552*** -1.308 1.136 0.639*** 0.324  

  (0.066) (0.122) (0.161) (0.890) (0.709) (0.092) (0.401)  

Std_Dev 0.445** 1.006** 0.092 0.658 1.032*** 1.338  

  (0.207) (0.399) (0.215) (0.453) (0.374) (1.112)  

CoinCount -0.063 -0.616 -0.068 -0.275 -0.638 -5.249  

  (0.090) (0.506) (0.084) (0.468) (0.537) (3.243)  

Constant -8.604** -704.83* -191.744 -629.32*** -1940.59*** -1660.526 1582.529 1773.596 -240.229

  (3.565) (426.076) (2012.728) (132.669) (687.673) 1781.168 1622.222 (5321.994) (189.096)

N 2305 2305 758 1088 459 639 119 85 131

R2 0.403 0.436 0.383 0.061 0.115 0.282 0.341 0.215 0.038

22

Table 6: Probability of Sale  I II III IV

variable dy/dx dy/dx dy/dx dy/dx

OpeningBid/Pmean -1.174*** -0.381 -0.929** -0.247  

  (0.352) (0.364)  

OpeningBid -0.005** -0.002 -0.012* -0.003

  (0.002) (0.007)  

Weekend 0.188** 0.059 0.086 0.023 0.833*** 0.304 -0.502* -0.12

  (0.089) (0.087) (0.316) (0.283)  

ln(Rating+1) 0.004 0.001 0.051** 0.014 0.017 0.006 0.082 0.018

  -0.023 -0.025 -0.092 -0.106  

RatingPercent 0.048 0.016 -0.046 -0.012 -1.083 -0.421 0.558 0.121

  (0.098) (0.116) (0.794) (0.343)  

Certified 0.017 0.006 0.164 0.041 0.428 0.155  

  (0.221) (0.210) (0.889)  

Shipping -0.091*** -0.024  

  (0.029)  

Constant -3.245 6.083 108.551 -54.822  

  (9.702) (11.527) (79.369) (34.303)  

N 2318 1606 85 130  

Pseudo R2 0.357 0.264 0.321 0.098  

23

Table 7: Number of Bidsvariable I II III       OpeningBid/Pmean -0.996*** -1.967*** -1.871***  (0.153) (0.030) (0.036)   Weekend 0.789*** -0.082*** 0.008  (0.228) (0.019) (0.023)   

Rating 8.3x10-5*** 7.4x10-7 3.7x10-6***

  (1.3x10-5) (7.9x10-7) (7.9x10-7)

   RatingPercent -0.277 0.044* 0.019  (0.287) (0.025) (0.027)   Certified -0.55 0.183*** 0.119**  (0.437) (0.041) (0.050)   Coin_Count -0.008** -0.002*** -0.002***

  (0.003) (3.3x10-4) (3.9x10-4)

   Shipping -0.025***  (0.007)   Constant 33.362 -1.9 0.633  (28.651) (2.580) (2.700)   N 2318 2318 1606   

Pseudo R2 0.148 0.418 0.371       

24