The Value of a Good Reputation Online:

An Application to Art Auctions.

José J. Canals-Cerdá1

First draft: November 16, 2007

Current draft: April 1, 2008

Abstract

Using a unique dataset of art auctions on eBay, we conduct an empirical analysis of thevalue of a seller�s online reputation. Several aspects distinguish our work from most existingresearch. We analyze a heterogeneous panel data consisting of a large number of observationsover a large period of time, including signi�cant variation in reputation across and withinsellers. The panel structure of our dataset allows us to employ �xed e¤ects techniquesto control for observed and unobserved di¤erences across auctions. The existing literaturesuggests a marginally signi�cant, and small in magnitude, impact of reputation on sale price.In contrast, our results point to a highly signi�cant, and sizable, impact of reputation onthe behavior of buyers and sellers and on market outcomes. The data suggest that sellers oneBay exert signi�cant e¤ort to avoid a negative feedback. Our analysis reveals a signi�cantimpact of the eBay feedback rating large enough to be consistent with this observation.

1 For useful comments I thank Chris Swann and seminar participants at The 2007 Meetings of theSouthern Economic Association. Excellent research assistance was provided by David Donofrio, TysonGatto, Woong Tae Chung, Jason Pearcy and Kelvin Tang. I am grateful to Kristen Stein, artist and eBaypower seller, for answering many questions on the functioning of eBay and on bidders�and artists�behavior.We are grateful to Katrina Beck for editorial assistance. These are the views of the author, and should notbe attributed to any other person or organization, including the Federal Reserve Bank of Philadelphia.Corresponding author: José J. Canals-Cerdá, Federal Reserve Bank of Philadelphia, 10 Independence

Mall, Philadelphia, PA 19106. Phone: (215) 574-4127, Fax: (215) 574-4146, e-mail: Jose.Canals-Cerda@phil.frb.org.

1 INTRODUCTION.

Online auction markets have attracted the attention of artists interested in o¤ering their

work for sale directly to potential buyers. Tens of thousands of works of art are available

for sale each day in online auction markets like eBay.com, currently the largest market

online; overstock.com; yahoo.com; or auction markets specialized on the sale of artwork,

like artbyus.com. Because each work of art is unique, its selling price may be di¢ cult

to determine ex-ante. Auctions relieve the seller of the responsibility of setting a price.

Not surprisingly, original art is frequently sold in an auction environment (Ashenfelter and

Graddy, 2002 and 2003). Furthermore, online auction markets like eBay reduce the cost of

the match between buyers and sellers and reduce the need of a middleman or art gallery.

Winning an auction on eBay is only the �rst step in an online transaction. Before the

sale is completed, the buyer has to pay the �nal auction price and trust that the seller

will deliver the product as agreed, at a later date. The seller may be more inclined to

carry out the agreement promptly, and to resolve any issues that may arise in a manner

agreeable to the buyer, if she wants to maintain a good reputation. At the end of each

transaction on eBay, the buyer and the seller have a chance to rate their level of satisfaction

with the transaction (positive, neutral, negative). The type of score, or feedback, received

by the seller in all previous transactions is readily available to potential buyers, along with

a summary measure representing positive feedbacks as a percentage of the positive and

negative feedback received. Ben-Ner and Putterman (2002) study the importance of trust

in economic transactions paying special attention to the New Economy. As their study

1

suggests, reputation mechanisms like the one implemented in eBay may increase the level of

trust among market participants and can result in signi�cant welfare gains.

In this paper we undertake an empirical study of the impact of an artist�s online repu-

tation on market outcomes using a unique dataset of art auctions on eBay. We analyze a

panel of art auctions from artists who post their work for sale on eBay regularly. Our data

consist of a large number of art auctions over a long period of time, including signi�cant

variation in reputation across and within sellers, as well as signi�cant variation in other rele-

vant characteristics, such as style (e.g., abstract, cubist), medium (e.g., acrylic, oil), ground

(e.g., stretch canvas, cardboard), and size. Our results suggest that, at least for auctions

for heterogeneous goods of uncertain value (the type considered in this study) the seller�s

reputation matters a great deal.

Several studies have analyzed the impact of a seller�s eBay reputation on the �nal auction

price using cross-sectional techniques. The overall conclusion from these studies is that the

impact of a bad reputation on the sale price is small. Bajari and Hortacsu (2003) �nd that

a negative reputation has a signi�cant negative impact on the number of bidders but does

not signi�cantly impact the �nal auction price. Melnik and Alm (2002) estimate that the

impact of negative feedback is signi�cant but very small in magnitude, and the same holds

for the impact of the overall rating. Houser and Wooders (2005) estimate that the average

cost to sellers stemming from neutral or negative reputation scores is less than one percent

(0.93%) of the �nal sales price. Lucking-Reiley, Bryan, Prasad, and Reeves estimate that a

one percent increase in positive/negative feedback increase/reduce sale price by only 0.03%

2

and 0.11% respectively. Bajari and Hortacsu (2004) review this evidence and conclude: "We

believe that these results are likely to signi�cantly understate the returns from having a good

reputation...Since getting positive feedback requires e¤ort on the part of sellers, it appears

that sellers are making e¤orts to avoid negative feedback..." Our analysis indicates that eBay

reputation has a signi�cant impact on market outcomes, large enough to be consistent with

their observation.

Livingston (2005) addresses the issue from a di¤erent perspective by focusing the atten-

tion on the number of positive feedbacks rather than on the impact of a negative reputation.

His analysis suggests that sellers are strongly rewarded for the �rst few positive feedback

ratings but additional instances of positive feedback have a small impact. Perhaps surpris-

ingly, when the author considers the impact of negative and neutral feedback (see table 6)

on bidding behavior, he obtains the associated coe¢ cient to be of the same sign as that

associated with positive feedback.

A potential problem with cross-sectional studies is that they rely on variation in feedback

rating across sellers in order to identify the impact of the feedback rating on market out-

comes. This identi�cation strategy relies on the assumption that there are no confounding

unobserved di¤erences across sellers a¤ecting market outcomes. In general, within-sellers�

variation in feedback ratings would be a preferable source of identifying information. One

possible reason why previous studies have not considered this approach may be because

negative feedbacks is extremely rare. In particular, Melnik and Alm (2002) consider a sam-

ple of 450 auctions, and Bajari and Hortacsu (2003) consider a sample of 407 auctions. In

3

both studies, the probability of a bad reputation is about 0.0002. In our data, the proba-

bility of a bad feedback is 0.00026. Thus, to be able to conduct a panel data analysis with

su¢ cient variation in feedback ratings we need data on a large sample of auctions, from a

relatively large number of sellers, over a large period of time. Resnich, Zeckhauser, Swanson

and Lockwood (2006) conduct a controlled experiment in which a single seller sold matched

pairs of vintage postcards under his regular identity and under seven new seller identities.

Their results indicate that the stablished seller fared better. They also compared sales by

relatively new sellers with and without negative feedback. In their data, a small number

of negative feedback did not a¤ect buyers�willigness-to-pay. The analysis in our paper in-

dicates that a negative feedback often elicits a complex behavioral response on the part of

the seller that may be di¢ cult to replicate in a controlled experiment. Also, all the artists

in our dataset are seasoned sellers. Similar to our study, Cabral and Hortacsu (2005) use

panel data to analyze the impact of eBay�s reputation rating. Their paper considers data on

homogeneous goods, and for the most part the focus is on market outcomes other than the

ones considered in this paper. In particular, they consider the impact of feedback on sales

growth, the frequency of future negative feedback, and sellers�exit from the market. They

also �nd that negative feedback has a signi�cant and sizeable e¤ect on market outcomes.

We view their analysis and the analysis conducted in this paper as complementary.

The paper proceeds as follows. In section 2, we describe the data to be used in the

empirical analysis. In section 3 we present the empirical analysis and describe results.

Section 4 concludes. Tables with results are presented at the end of paper.

4

2 The Market and the Data.

A. The Market.

An auction on eBay includes a description of the item being auctioned, including pictures.

When posting a new auction, sellers can choose several auction characteristics (e.g., the

opening bid). Buyers can browse through thousands of auctions posted every day. Auctions

are organized by categories and subcategories, which simpli�es the buyer�s search. Buyers

can also use a powerful search engine and are also able to request additional information

about an auction from the seller anonymously via e-mail. Potential buyers can participate

in an active auction at any time over the duration of the auction by submitting a �proxy�

bid. This type of bidding works as follows: as long as a bidder�s proxy bid is higher than

the second highest bid, she will remain the highest bidder, at a price equal to the second

highest bid plus a small increment. When a new bid is submitted, the auction�s price is

updated accordingly. Proxy bids can be revised (increased) at any time prior to the end

of the auction. Only bids above the existing second highest bid, plus an increment, are

accepted and recorded. Thus, the number of potential bidders is likely to be larger than the

number of bidders recorded by eBay. The highest bidder at the end of the auction wins the

item at a price equal to the second highest bid, plus a small increment.2

B. The Data.

Between July and November of 2001 and again between August and December of 2004

we collected data on all auctions from a group of �self-representing�artists. This group is

2 Additional information about eBay auctions can be found in Bajari and Hortacsu (2004) :

5

composed of artists who sell their own artwork through eBay, without representation. The

most popular artwork up for auction is paintings, but other types of artwork, like collages,

ceramic tiles, or sculptures, are also common. In our analysis we use data on original

paintings only. The sample includes 42 artists chosen from a group of artists who posted at

least 25 paintings for sale in 2001 and who continued to sell on eBay until the end of 2004.

The dataset employed in our analysis includes 4514 auctions, with 2245 auctions from 2001

and 2269 auctions from 2004 (see table 1a). The data included auctions posted between

mid-July and mid-November 2001 and between mid-August and mid-December 2004.

For each auction, we collected four di¤erent kinds of information: object-speci�c charac-

teristics, other auction characteristics, bidding history, and artist reputation. Characteristics

speci�c to the object being auctioned include information on the height, width, style (ab-

stract, pop, whimsical, etc.), medium (acrylic, oil, etc.), and ground (stretch canvas, paper,

wood, etc.). Other auction characteristics include the opening bid, the shipping and han-

dling fees, and the �nal number of bidders and selling price. At the end of each transaction,

the buyer and the seller have a chance to rate their level of satisfaction with the transaction

(positive, neutral, negative). We collect data on the type of feedback received by the seller in

all previous transactions, which is readily available from eBay. We use these data to de�ne

several measures of feedback history: one represents the number of unique buyers prior to

the current auction, which could be interpreted as a measure of the artist�s customer base on

eBay, another one measures the number of negative feedback ratings received, and a third

one represents the feedback rating as reported on eBay, which is equal to the number of pos-

6

itive feedbacks divided by the number of positive and negative feedbacks from unique users;

neutral feedback does not count. Each of these variables can be interpreted by potential

buyers as measures of a seller�s reputation.

Several characteristics of these data make them unique. First, the data collected comprise

all eBay transactions for a speci�c group of sellers for a long period of time, while the

data collected by other researchers usually represents only a narrow snapshot of market

activity from a cross-section of sellers. Second, by nature, the intrinsic value of artwork is

uncertain, especially in the case of less well-known artists. In contrast, much of the data

collected by other researchers refer to homogeneous items or items whose market value can

be determined with accuracy, like coins or stamps, which lessens the value of auctions as a

selling, price-�nding mechanism. Third, the panel structure of our data allows us to control

for sellers�speci�c �xed e¤ects. In contrast, most of the existing econometric research has

been conducted using cross-sectional analysis, the exception being Cabral and Hortacsu

(2005).

C. Descriptive Statistics.

We report sample descriptive statistics in Tables 1a and 1b. In Table 1a, we divide

auctions by year and according to their "Feature Plus!" status. For several years now,

eBay has been using a simple form of advertising, or as they call it, �Featured Plus!�(FP).

This type of advertising works as follows: at the time of listing the item on eBay, sellers

are given the option of incurring an extra fee in return for having their product listed

�rst when buyers search for speci�c items, or when buyers browse a speci�c category, like

7

�art/paintings/abstract�. This type of advertising is not cheap, the cost is $19.99 per auction

or about 40% of the average sale price of a standard auction. Canals-Cerda (2006) shows that

FP status has a signi�cant impact on auction outcomes, including the number of bidders, the

sale probability and the �nal sale price. This impact is also re�ected in Table 1. Furthermore,

FP auctions are signi�cantly di¤erent from other auctions. In particular, the average size

of an FP painting is more than twice that of a painting in a standard auction. To our

knowledge, the potential impact of FP has been ignored by the relevant literature. In our

empirical analysis we control for FP status as part of our �xed e¤ects strategy. This strategy

allows us to compare paintings from the same artist with similar characteristics and with

the same FP status.

Bid values in Table 1a are measured in real 2004 dollars and include shipping and handling

costs. The sale price of paintings, including shipping costs, ranges from as little as $0.01

(which would indicate a painting that was sold for one cent and free shipping) to as much

as $1441.41. The average probability of sale for standard auctions is 62% in 2001 and 48%

in 2004, or 95% and 87%, respectively, for FP auctions. The average selling price is about

$48 for the �rst group of auctions and $224 for the second group. Di¤erences in the sale

price between both groups of auctions are partly due to di¤erences in the characteristics of

objects being auctioned, like size, medium, or ground. There are also signi�cant di¤erences

in the number of bidders and bids received. FP auctions with at least one bidder receive

bids from more than �ve bidders on average, or more than double the number of bidders in

standard auctions.

8

Table 1b presents detailed information about the range of variation of several measures

of reputation available in our sample. Like eBay, we only count unique buyers. That is,

only the �rst positive feedback from a buyer counts, and only the �rst negative feedback

counts. The variables listed include the eBay rating and the overall number of feedback

ratings from unique buyers divided by category: positive, negative, and neutral. As with

existing research, negative feedbacks is extremely rare in our sample. In particular, the

average eBay rating across auctions is 99.9% in 2001 and 99.78% in 2004. The maximum

number of negative feedbacks for an artist in our sample is eight. From this �rst group of

artists, 36 have a perfect eBay rating in 2001, and 27 have a perfect rating in 2004. The

lowest eBay rating is 98.04 in 2001 and 94.32 in 2004. Most of the artists in our sample

have ample experience selling on eBay. The average number of unique feedback responses

in 2004 is 512. The feedback history for the artists in our sample was constructed from the

overall feedback history received by January 2005. These data overlap with the data on

auctions. The total feedback received by the artists in our sample was 30,305, including 50

negative and 55 neutral. Thus, the average frequency of a negative feedback is 0.16%. The

total feedback received from unique buyers was 17,844, including 39 negative and 55 neutral.

Thus, the average frequency of negative feedback from a unique buyer is 0.22%.

3 Empirical Models and Estimation Results.

Our reduced form econometric analysis is guided by economic theory. The type of eBay

auctions that we consider �t well within the framework of an independent private value,

9

ascending-bid, second-price auction subject to some speci�c rules. The seller of the object

being auctioned sets the opening bid, s0: Each potential bidder, k; assigns a value, vk; to the

object being auctioned. Bidders know and care only about their own valuation. Bids are

submitted electronically at any time within the [0; T ] time interval. At any particular time

the price of the auction is set at the current second highest bid, and the current, or active,

bidders are public information. New bids arrive sequentially at any time during the length

of the auction. Any new bid has to surpass the current second highest bid by a minimum

increment in order to be recorded.3 In this framework it is optimal for a bidder to bid her

true valuation. Thus, consider an auction with M observed bids, fbkgMk=1 ; from K active

bidders, with K � N; and N representing the total number of potential bidders; that is,

active bidders along with bidders who intended to bid but whose valuation was below the

second highest bid at the time they intended to bid. Within this framework, we have three

possible scenarios: 8>>>>>><>>>>>>:K = 0 : �vN�1 < �vN < s0

K = 1 : �vN�1 < p = s0 < �vN

K � 2 : p = �bK�1 = �vN�1

9>>>>>>=>>>>>>;(1)

with �vN�1; and �vN ; representing the second highest, and the highest, valuation from the

group of potential bidders. In order to understand the impact of a seller�s reputation on this

market it is important to di¤erentiate between the impact on the behavior of the seller and

the buyer.

3 In Ebay, the value of the minimum increment varies with s0: The minimum increment is $0.05 for bidsunder $1.00 dollar and increases up to $100.00 for bids above $5000.00.

10

We always observe s0: In addition, when the opening bid is low enough we observe

�vN�1 = p; K � 2; otherwise we observe s0. Note that s0 is set by the seller while p, when

K � 2; is set by the bidders�willingness to pay. Sellers and bidders may have con�icting

objectives. Thus, in our empirical analysis we will analyze s0 and �vN�1 separately. Because

s0 is always observed we can employ simple linear regression techniques in its analysis.

However, �vN�1 is only observed when s0 is low enough to attract at least two bidders;

otherwise, s0 represents an upper bound for �vN�1. Thus, in this second case we will employ

censored regression techniques. In addition, we also study the impact of seller feedback

on the probability of sale, the number of bidders, and the probability of future negative

feedbacks.

Given the heterogeneous nature of the data employed in our analysis, we divide the data

in groups along several di¤erent dimensions: artist, medium, ground, and Feature Plus!

status. More precisely, all paintings from a particular artist, using a particular medium (e.g.

oil), on a particular ground (e.g., canvas), and for a speci�c Feature Plus! status will be

assigned to a group. This data segmentation skims along with the use of �xed e¤ects (FE)

techniques allow us to identify the average e¤ect of explanatory variables (e.g. measures of

feedback) from variation in the data within relatively homogeneous groups.

All econometric models considered share a common structure. The endogenous outcome

of interest, yij; representing the outcome associated with auction i from group j is repre-

sented as a function of three components: observed exogenous explanatory variables, Xij; a

group-speci�c �xed e¤ect, �j; and a residual component, "ij; representing random variation

11

unobserved by the econometrician. Analytically, we consider models of the form

yij = �0 + �j + �Xij + "ij (2)

that are estimated using group-speci�c �xed e¤ect (FE) techniques appropriate for each spec-

i�cation to control for potentially endogenous �xed e¤ects.4 The results from our econometric

analysis are presented in Tables 2-7 and are discussed next.

A. Feedback Dynamics.

Understanding how feedback is generated can help us better understand the impact of an

artist�s feedback on market outcomes. However, sellers on eBay are far from a homogeneous

group and it is unlikely that a single theoretical model will be able to explain all types of

seller behaviors.

Livingston (2005) considers a model of bidder and seller behavior in which there are

two possible seller types� honest and dishonest� that decide whether to cooperate with or

betray the winning bidder. Honest sellers choose to cooperate with a higher probability

than dishonest sellers. If the seller cooperates, the buyer receives a positive payo¤ from the

purchase; otherwise she receives a negative payo¤. The winning bidder reports on how the

seller behaves, and future bidders update their beliefs about the seller�s type, and the new

beliefs a¤ect their bidding. Cabral and Hortacsu (2005) ; consider several models of buyers

and sellers�behavior and �nd that it is not easy to empirically distinguish between theories.

4 Style is included in Xij rather than in the group FE de�nition because, by eBay�s policy, each auctioncan be associated with up to three di¤erent styles, and, as a result, we have a very large number of possiblestyle combinations.

12

In order to analyze feedback dynamics in our data, we estimate simple linear probability

models with artists�speci�c �xed e¤ects and controls for lagged feedback measures and a

measure of past sales.5 The models are estimated using feedback data from unique users

(this includes the �rst positive, negative, and neutral feedback received from a buyer, when

applicable), which are the exact same data used by eBay to generate its feedback ratings.

Table 2 presents results from simple linear probability models, in the spirit of equation (2),

where the probability of an artist�s receiving a negative feedback in auction t is a function of

her feedback history at the beginning of the auction. The estimated coe¢ cients from a simple

linear probability model (model 1) and an artist�s speci�c �xed e¤ects version of it (model

3) are very di¤erent. Estimation results from model 1 suggest that a one-unit increase in

the number of feedback responses increases the probability of negative feedback by 0.0016,

a 73% increase with respect to the average unconditional probability 0.00219 (table 1b).

This result is highly signi�cant. In contrast, estimation results from model 3 suggest that

a one-unit increase in the number of negative feedback responses increases the probability

of future negative feedback by 0.00043, a 20% increase with respect to the average. This

result is insigni�cant. The associated R-square values are very small as expected, as negative

feedback is very unusual and, as a result, very di¢ cult to predict. Models 2 and 4 expand

on models 1 and 3 by including a control for the number of feedback received in the past.

As one would expect, there is a positive correlation between the feedback received and the

negative feedback received. In model 4 we control for the overall total feedback received from

5 Similar results are obtained using conditional FE logit models. We present results from linear probabilitymodels because they accept an easier interpretation.

13

unique users. The results indicate that a one-unit increase in negative feedback increases the

probability of future negative feedback by 0.00012, a 5% increase, and that an average artist

is more likely to receive negative feedback in a transaction if she already has a large number

of feedbacks, 10% more likely for an increase in 100 feedbacks; both e¤ects are statistically

insigni�cant.

The estimated e¤ect of feedback history on future feedback observed in model one and

two are identi�ed from variation in feedback histories over the 17,827 feedback events from

unique buyers, while the estimated e¤ect observed in model three and four are identi�ed

from within-artists� variation in feedback histories. The results in these last two models

suggest that receiving negative feedback today does not have a signi�cant impact on the

probability of receiving negative feedback in the future. This suggests that the artists in

our sample do not signi�cantly alter the level of e¤ort invested in serving their clients after

receiving negative feedback.

B. Artists�Response to Negative Feedback.

If the artist anticipates a change in buyers�behavior as a result of receiving negative

feedback, then we would expect a strategic response on her part. This is what we observe

in the data.

In table 3 we present estimation results from several log-linear regression model speci�-

cations of the opening bid, which is set by the artist, as a function of feedback history and

other characteristics speci�c to the artwork being auctioned. All models have been estimated

using a �xed e¤ects strategy to control for artist, medium, and ground speci�c e¤ects, as

14

well as FP status. All estimated models have signi�cant explanatory power, as indicated

by R-square values around 0.71. The size of the artwork is an important determinant of

open bid value; the overall number of feedback responses received does not have a signi�cant

e¤ect, while the number of negative feedback responses received has a signi�cant impact

on the opening bid. The results from models one and two suggest that a one-unit increase

in the negative feedback received increases the opening bid by 20%, other things the same.

This result is highly signi�cant. Similar results are observed when we use a standardized

eBay feedback measure instead of the number of negative feedbacks. In this case a higher

eBay feedback value decreases the opening bid signi�cantly. One possible interpretation of

this empirical result is that it measures the strategic response of the artist to a decrease in

the number of potential buyers.

C. Buyers�Response to an Increase in a Seller�s Negative Feedback Rate.

In this subsection we analyze the impact of an increase in an artist�s negative feedback

rating on the behavior of potential buyers. In particular, we consider the impact on the

number of bidders and the probability of sale and the impact on the second highest bid�

and indirectly on prices. Overall, we observe that the response of potential buyers to an

increase in an artist�s negative feedback rating is large in magnitude, has the expected

sign, and is statistically signi�cant in most cases. Our results contrast with results from the

existing literature. We investigate the signi�cance of speci�c econometric model assumptions

on results, and highlight the importance of controlling for sellers�speci�c �xed e¤ects.

C.1 The Impact on the Number of Bidders and the Probability of Sale.

15

The objective here is to analyze results from a reduced form speci�cation of the overall

impact of an increase in negative feedback on the number of bidders participating in an

auction. However, we should point out that the arrival of potential bidders to an auction is

a complex function of artist and artwork characteristics, opening bid, bidders�arrival process,

and bidders�bidding behavior. A structural analysis of this process has been conducted in

Canals-Cerda and Pearcy (2006) and is not the objective of this study.

In Table 4 we present results from log-linear regression models of the number of bidders,

plus one, as a function of characteristics of the artwork being auctioned and artist feedback

measures. All models have been estimated using a �xed e¤ects strategy to control for artist-,

medium-, and ground-speci�c e¤ects, as well as Feature Plus! status. All estimated models

have signi�cant explanatory power, as indicated by R-square values around 0.55. Size is

not an important determinant of the number of bidders. The overall number of feedback

responses received does not have a signi�cant e¤ect either. In contrast, the average e¤ect

of an additional negative feedback is an 8.5% reduction in the number of bidders. This

e¤ect is highly signi�cant and consistent across di¤erent model speci�cations. In particular,

we observe similar results when we use the standardized eBay feedback measure instead

(models 3 and 4). In that case, a higher eBay feedback value increases the number of

bidders signi�cantly.

Similarly, in Table 5 we present �xed e¤ects linear probability models for the probability

of sale with the same speci�cations as these in Table 4. The models suggest that size has

a negative impact on the sale probability. Thus, this result, along with these from Table 4,

16

suggests that bigger works of art may be more di¢ cult to sell, but that the ones that get

sold attract a large number of bidders. With regard to the impact of negative feedback, the

coe¢ cient associated to the number of negative feedback responses (models 1 and 2) is large

in magnitude and suggests that one additional negative feedback reduces the probability of

sale by about 8.5%, in average; however the e¤ect is statistically insigni�cant. In contrast,

if we consider the standardized eBay feedback measure instead (models 3 and 4), we also

observe a relatively large e¤ect, suggesting that a higher eBay feedback rating is good for

sales, and the e¤ect is statistically signi�cant in this case.

C.2 The Impact on the Second Highest Bid.

As indicated in section three, the �nal sale price of an auction is the result of a combina-

tion of sellers and buyers�choices. However, we can measure the impact of sellers�feedback

on buyers�willingness to pay by concentrating our attention in the second highest bid. As

illustrated in equation (1) the second highest bid is only observed in auctions with two or

more bidders. Otherwise, the opening bid, for auctions with a single bidder or with no

bidders, represents an upper bound to the second highest bid. Thus, it is appropriate in this

framework to employ censored regression techniques to analyze the impact of a seller�s feed-

back on the �nal auction price. Formally, consistent with equation (2), consider a log-linear

speci�cation for the second highest bid,

ln �vN�1;ij = �0 + �j + �Xij + "ij

where the second highest bid, �vN�1;ij; is observed when �vN�1;ij � s0;it; with s0;ij representing

17

the opening bid. Thus consider yit = ln �vN�1;ij and cit = ln s0;it: We observe Max (yit; cit) ;

or after a transformation, y�it = yit� cit = �0+ �j� cit+�Xij+ "ij; the equivalent expression

Max (y�it; 0) : We estimate this �xed e¤ects censored regression model using semiparametric

estimation techniques developed in Honore (1992) and surveyed in Arellano and Honore

(2005) ; and use the bootstrap method with 100 repetitions to generate t-values.6 The basic

idea of the estimation technique is to �de�ne pairs of �residuals�that depend on the individual

speci�c e¤ect in exactly the same way. Intuitively, this implies that di¤erencing the residuals

will eliminate the �xed e¤ects�(Arellano and Honore, 2005). A detailed description of the

approach can be found in these references.

Estimation results are presented in Tables 6a and 6b with associated �xed e¤ects for

groups de�ned by artist, medium, and ground, as well as Feature Plus! status, as with other

prior speci�cations. For comparison purposes with previous research that has not considered

�xed e¤ects strategies to control for sellers�speci�c e¤ects, we also estimate a model that

includes �xed e¤ects to control for medium, and ground, and Feature Plus! status, in order

to reduce heterogeneity across observations, but not artist-speci�c e¤ects; results from this

exercise are presented in Table 7.

Table 6b presents estimation results from several model speci�cations that di¤er on

the set of control variables.7 As with most previous speci�cations, size is an important

6 We use the same approach described in Stata manual v.9 "bootstrap - Bootstrap sampling andestimation."

7 Model 4 is inspired by the work of Rezende (2006). Under the assumption of a exogenous number ofbidders, Rezende has shown that one can identify the valuation distribution with simple OLS methods byincluding dummies to control for the number of bidders in a linear regression framework. eBay auctionslikely violate some of the assumptions in Rezende (2006). As such, the estimation results are included forillustrative purposes.

18

determinant of value. A one square foot increase in size is associated with a 19% average

statistically signi�cant increase in the second highest bid. A one-unit increase in negative

feedback is associated with an average loss in value between 6.3% and 8%, depending on the

model speci�cation; this e¤ect is statistically signi�cant. The estimation results from Table

7a are very similar, except for the caveat that the measured e¤ect of the standardized eBay

feedback rating is not statistically signi�cant, although it is large in magnitude in all model

speci�cations. With regard to this last result, observe that two sellers with the same number

of negative feedback will have di¤erent eBay feedback ratings unless their overall number of

feedback is the same. Thus, one possible interpretation of the non-signi�cant result is that

the standardized eBay feedback rating represents a noisy measure of reputation and as such

the associated impact is not estimated accurately.

D. How Do Our Results Compare with the Existing Literature?

Our analysis shows that the impact of reputation on the behavior of buyers and sellers

and on market outcomes is signi�cant statistically and in magnitude. As we indicate in the

introduction most existing studies of the impact of reputation on eBay have been conducted

using cross-sectional analysis techniques. The results in general suggest that reputation has

a signi�cant impact on the sale price, but this impact is small in magnitude. Livingston

(2005) focuses on the impact of good feedback; his analysis suggests that sellers are strongly

rewarded for the �rst few positive feedback responses, but additional positive feedback re-

sponses have a small impact. Besides our work, only Cabral and Hortacsu (2005) have

conducted an econometric analysis of reputation on eBay using panel data. The focus of

19

their analysis is on the impact of feedback on sales growth, the frequency of future negative

feedback, and sellers�exit from the market. Their analysis also shows that negative feedback

has a signi�cant and sizeable e¤ect on market outcomes. We view their analysis and the

analysis conducted in this paper as complementary.

4 Conclusions.

The existing literature on the impact of reputation using eBay data suggests a marginally sig-

ni�cant, and small in magnitude, impact of reputation on sale price. In contrast, our results

point to a signi�cant statistically-signi�cant, and large in magnitude, impact of reputation

on the behavior of buyers and sellers and on market outcomes. More precisely, on average

a negative feedback is associated with an 8.5% reduction in the number of bidders and the

probability of sale; it is also associated with a signi�cant reduction in buyers�willingness

to pay, anywhere between 6.3% and 8%. Furthermore, after receiving negative feedback,

the artist increases the opening bid in upcoming auctions by about 20% on average. One

possible interpretation of this empirical result is that it measures the strategic response of

the artist to a decrease in interest for her work. Thus, our results suggest that, at least for

auctions of heterogeneous goods of uncertain value, the seller�s reputation matters.

20

References

Ashenfelter, O. and K. Graddy (2002): �Art Auctions: A Survey of Empirical Studies,�CEPR Discussion Paper 3387.

Bajari, P. and A. Hortacsu (2003): �The Winner�s Curse, Reserve Prices, and Endoge-nous Entry: Empirical Insights from eBay Auctions,�The Rand Journal of Economics,Vol. 34, No. 2, pp. 329-355.

Bajari, P. and A. Hortacsu (2004): �Economic Insights from Internet Auctions,�Jour-nal of Economic Literature, Vol. XLII, 457-486.

Luis, Cabral and Ali Hortacsu (2005). "The Dynamics of Seller Reputation: Evidencefrom eBay." Working Paper, University of Chicago.

Canals-Cerdá, José J. (2005). �Congestion Pricing in an Internet Mar-ket.� NET Institute Working Paper No. 05-10. Available at SSRNhttp://papers.ssrn.com/sol3/papers.cfm?abstract_id=848004

Canals-Cerdá, José J. (2006). �Advertising as a Signal in an Internet Auctions Market.�Available at SSRN http://papers.ssrn.com/sol3/papers.cfm?abstract_id=935093.

Canals-Cerdá, José J. and Pearcy J. (2006). �Econometric Analysis of Eng-lish Auctions: Applications to Art Auctions on eBay.� Available at SSRNhttp://papers.ssrn.com/sol3/papers.cfm?abstract_id=947605.

Houser, D. and Wooders, J. (2005). "Reputation in Auctions: Theory, and Evidencefrom eBay", Journal of Economics and Management Strategy 15 (2006), 353-369.

Livingston, J.A. (2005). "How Valuable is a Good Reputation? A Sample SelectionModel of Internet Auctions." The Review of Economics and Statistics, Aug. 2005,87(3): 453-465.

Melnik, M. and J. Alm (2002): �Does a Seller�s eCommerce Reputation Matter? Evi-dence from eBay Auctions,�Journal of Industrial Economics, Vol. 50 (3), 337-349.

Milgrom, P. R. and R. J. Weber (1982). �A Theory of Auctions and CompetitiveBidding.�Econometrica, 50, 1089-1122.

Resnick, P, R. Zeckhauser, J. Swanson, K. Lockwood (2006), �The Value of Reputationon Ebay: A Controlled Experiment.�Experimental Economics, 2006, vol. 9, issue 2,pages 79-101.

L. Rezende (2006) : "Econometrics of Auctions by Least Squares." W.P. Departmentof Economics, university of Illinois.

21

Table 1a: Descriptive Statistics for Auction Specific Characteristics. Standard Auctions 2001 2004 Mean Std. Min. Max. Mean Std. Min. Max. Open Bid 33.88 31.75 0.01 436.75 47.81 65.94 3 595.55 Sale Price** 46.11 45.25 2.23 426.02 49.80 58.01 3.01 595.55 # Square Feet 2.06 2.34 0.01 20.25 1.95 2.40 0.02 24 # Bidders* 2.56 1.82 1 13 2.35 1.62 1 10 # Bids* 4.25 4.20 1 38 3.75 4.11 1 29 # auctions 2226 1758 % Sold 0.620 0.484 Feature Plus! Auctions 2001 2004 Mean Std. Min. Max. Mean Std. Min. Max. Open Bid 40.92 51.96 0.01 235.10 69.92 75.48 6.6 595.55 Sale Price** 225.00 159.82 40.08 697.62 222.91 132.05 21.10 1441.41# Square Feet 5.69 3.43 0.29 12 5.40 3.09 0.44 24 # Bidders* 6.28 3.41 1 14 5.26 2.71 1 17 # Bids* 15.44 13.03 1 52 12.27 8.22 1 47 # auctions 19 511 % Sold 0.947 0.873

(*) For auctions with at least one bidder. (**) For sold auctions.

Table 1b: Feedback Information. 2001 2004 AUCTIONS Mean Std. Min. Max. Mean Std. Min. Max. Ebay Rating 99.90 0.307 98.04 100 99.78 0.342 94.32 100 Positive 137.33 120.74 0 576 508.85 260.78 39 1650 Negative 0.234 0.676 0 4 1.39 2.06 0 8 Neutral 0.164 0.427 0 2 2.14 2.40 0 9 AUCTIONS BY # OF NEGATIVE FEEDBACK negative feedback 0 1 2 4 0 1 or 2 3 or 5 7 or 8 Frequency 1946 128 143 28 1189 669 278 133 % Frequency 86.68 5.70 6.37 1.25 52.40 29.48 12.26 5.86 ARTISTS Max. negative # 0 1 2 4 0 1 2,3 5,8 Frequency 36 4 1 1 27 7 2,3 2,1 % Frequency 0.86 0.10 0.02 0.02 0.64 0.17 0.12 0.07 FEEDBACK Overall Positive Negative Neutral All feedbacks 30305 30199 50 55 % Frequency 100 99.65 0.165 0.181 Unique Buyers 17827 17733 39 55 % Frequency 100 99.47 0.219 0.309

22

Table 2: The Dynamics of Negative Feedback. Linear Probability Models F.E. Linear Probability Models Model 1 Model 2 Model 3 Model 4 Coef. T-val Coef. T-val Coef. T-val Coef. T-val # Neg. Feedback 0.00161 5.16 0.00176 5.37 0.00043 0.92 0.00012 0.23 # feedbks/100 -.00017 -1.48 0.00021 1.27

R-sq 0.0015 0.0016 0.0099 0.0099 # Obs. 17827 17827 17827 17827 Fixed effects models include control for artist specific effects.

Table 3: FE Log-linear regression Models for the opening bid. Model 1 Model 2 Model 3 Model 4 Coef. T-val Coef. T-val Coef. T-val Coef. T-val FEEDBACK Std. eBay Fbk. -0.251 -7.39 -0.238 -6.96 # Neg. Feedback 0.208 13.62 0.206 13.08 # Feedbks/100 0.010 0.75 0.042 3.06 DIMENSION Square-Feet 0.248 27.29 0.248 27.28 0.244 26.44 0.243 26.41 Square-Feet2 -0.009 -15.53 -0.009 -15.52 -0.009 -15.12 -0.009 -15.09DUMMIES Style Yes Yes Yes Yes Year & Month Yes Yes Yes Yes R-sq 0.7127 0.7128 0.7118 0.7125 # Obs. 4514 4514 4514 4514 Fixed effects include control for artist specific effects, medium, ground and Feature Plus! Status. eBay feedback has been standardized.

23

Table 4: FE Log-linear regression Models for the number of bidders. Model 1 Model 2 Model 3 Model 4 Coef. T-val Coef. T-val Coef. T-val Coef. T-val FEEDBACK Std. eBay Fbk. 0.134 3.87 0.131 3.76 # Neg Feedback -0.085 -5.43 -0.085 -5.24 # Feedbks/100 -0.003 -0.18 -0.009 -0.66 DIMENSION Square-Feet 0.010 1.07 0.010 1.07 0.013 1.43 0.014 1.44 Square-Feet2 0.000 -0.49 0.000 -0.49 0.000 -0.71 0.000 -0.72 DUMMIES Style Yes Yes Yes Yes Year & Month Yes Yes Yes Yes R-sq 0.5452 0.5452 0.5499 0.5499 # Obs. 4514 4514 4514 4514 Fixed effects include control for artist specific effects, medium, ground and Feature Plus! Status. eBay feedback has been standardized.

Table 5: FE linear probability Models for the Sale probability.

Model 1 Model 2 Model 3 Model 4 Coef. T-val Coef. T-val Coef. T-val Coef. T-val FEEDBACK Std. eBay Fbk. 0.067 2.63 0.066 2.57 # Neg. Feedback -.105 -1.09 -0.084 -0.86 # feedbks/100 -0.082 -1.21 -0.002 -0.20 DIMENSION Square-Feet -.095 -2.05 -0.094 -2.04 -0.015 -2.02 -0.015 -2.02 Square-Feet2 0.005 1.72 0.005 1.71 0.001 1.60 0.001 1.60 DUMMIES Style Yes Yes Yes Yes Year & Month Yes Yes Yes Yes LLF or R-sq -1960.84 -1960.11 0.34 0.34 # Obs. 4514 4514 4514 4514 Fixed effects include control for artist specific effects, medium, ground and Feature Plus! Status. eBay feedback has been standardized.

24

Table 6a: Honore’s FE Models for the log-second highest bid. Model 1 Model 2 Model 3 Model 4 Coef. T-val Coef. T-val Coef. T-val Coef. T-val FEEDBACK Std. eBay Fbk. # Neg Feedback -0.077 2.47 -0.080 2.37 -0.063 2.11 -0.076 2.26 # Feedbks/100 0.015 0.30 0.029 0.92 0.063 1.76 DIMENSION Square-Feet 0.195 8.01 0.194 7.83 0.189 7.75 0.179 6.46 Square-Feet2 -0.006 3.32 -0.006 3.26 -0.006 3.17 -0.006 3.03 DUMMIES Style No No Yes Yes Year & Month Yes Yes Yes Yes # bidders No No No Yes LLF -23715.17 -23711.71 -22810.78 -17442.12 # Obs. 4514 4514 4514 4514 Fixed effects include control for artist specific effects, medium, ground and Feature Plus! Status. eBay feedback has been standardized. T-values computed using bootstrap and 100 repetitions.

Table 6b: Honore’s FE Models for the log-second highest bid. Model 1 Model 2 Model 3 Model 4 Coef. T-val Coef. T-val Coef. T-val Coef. T-val FEEDBACK Std. eBay Fbk. 0.129 0.96 0.131 0.93 0.124 1.06 0.096 0.72 # Neg Feedback # Feedbks/100 0.005 0.11 0.029 0.91 0.060 1.71 DIMENSION Square-Feet 0.196 8.13 0.196 7.96 0.190 7.85 0.180 6.56 Square-Feet2 -0.006 3.44 -0.006 3.38 -0.006 3.25 -0.006 3.10 DUMMIES Style No No Yes Yes Year & Month Yes Yes Yes Yes # bidders No No No Yes LLF -23821.61 -23821.61 -22864.78 -17520.65 # Obs. 4514 4514 4514 4514 Fixed effects include control for artist specific effects, medium, ground and Feature Plus! Status. eBay feedback has been standardized. T-values computed using bootstrap and 100 repetitions.

25

26

Table 7: Honore’s FE Models without Artists’ FE for the log-second highest bid. Model 1 Model 2 Model 3 Model 4 Coef. T-val Coef. T-val Coef. T-val Coef. T-val FEEDBACK Std. eBay Fbk. # Neg Feedback 0.0670 4.46 0.0235 1.55 0.0246 1.57 0.0236 1.41 # Feedbks/100 0.0461 4.68 0.0429 3.99 0.0491 4.18 DIMENSION Square-Feet 0.2214 11.92 0.2196 12.01 0.2292 11.56 0.2384 9.52 Square-Feet2 -.0087 6.03 -.0088 5.84 -.0092 5.56 -.0095 5.30 DUMMIES Style No No Yes Yes Year & Month Yes Yes Yes Yes # bidders No No No Yes LLF -205348.17 -201397.95 -174640.66 -126195.74 # Obs. 4514 4514 4514 4514 Fixed effects include control for medium, ground and Feature Plus! Status but not for artist specific effects. eBay feedback has been standardized. T-values computed using bootstrap and 100 repetitions.