Reputation and adverse selection: theory and evidence from eBay

• DOI: http://doi.org/10.1111/1756-2171.12297
• Published date: 01 December 2019
• Author: Maryam Saeedi
• Date: 01 December 2019

RAND Journal of Economics

Vol.50, No. 4, Winter 2019

pp. 822–853

Reputation and adverse selection: theory and

evidence from eBay

Maryam Saeedi∗

How can a marketplace introduce mechanisms to overcome inefficiencies caused by adverse

selection? In this article, I use a unique data set that follows eBay sellers to show that reputation

is a major determinant of price variations. I develop a model of sellers’ dynamic behavior

where sellers have heterogeneous qualities unobservable by buyers. Using reputation as a signal

of quality, I structurally estimate the model to uncover buyers’ utility and sellers’ costs and

underlyingqualities. I show that removing the reputation mechanism increaseslow-quality sellers’

marketshare, lowers prices, and consequently reduces sellers’ profitby 66% and consumer surplus

by 35%.

1. Introduction

Asymmetric information is known to lead to inefficiencies in markets. In particular, in

online markets where trading is decentralized and with little repeat interactions, asymmetric

information problems can be magnified.1In these markets, the sellers are better informed about the

characteristics of the items they sell or their level of expertise than are their customers. The sellers

on eBay,for example, can misrepresent the objects they sell or mishandle the shipping of the items

sold. Reputation mechanisms are often used to mitigate these asymmetric information problems.

∗Carnegie Mellon University; msaeedi@andrew.cmu.edu.

I am indebted to Patrick Bajari and Thomas Holmes for their valuable advice. I would also like to thank V.V. Chari,

Alessandro Dovis, Konstantin Golyaev, Hugo Hopenhayn, Larry Jones, Patrick Kehoe, Kiyoo-il Kim, Tina Marsh, Ellen

McGrattan, Minjung Park, Amil Petrin, Erick Sager, Jack Shen, Ali Shourideh, Neel Sundaresan, SteveTadelis, Robert

Town,Joel Waldfogel, Thomas Youle,and Ariel Zetlin-Jones, as well as the participants of the Applied Micro Workshop

and Applied Micro Seminar of the University of Minnesota (2009–2011), IIOC (April 2013), the Ohio State University

(January 2012), Carnegie Mellon University (January 2012), Arizona State University (February 2012), State University

of New York at Stony Brook (February 2012), UCLA (November 2012), Charles River Associates (January 2012),

Revolution Analytics Inc. (January 2012), eBay Research Labs’ Reading Group (June 2011), Universidad Catolica de

Chile (December 2015), and Said Business School, Oxford University (August 2016). I am also grateful to the Editor,

Marc Rysman, for his insightful suggestions and guidance and to three anonymous referees for their excellent comments.

Any remaining errors are my own.

1Examples of these markets include eBay,Amazon Marketplace, Alibaba, and Taobao in retail; Airbnb and VRBO

in room and house sharing; Uber and Lyftin transportation; Care.com in child care; Rover in pet care; and Upwork (former

oDesk) in the freelance and labor market, as studied in Cai et al. (2013); Fan, Ju, and Xiao (2016); Zervas, Proserpio,

and Byers (2015); and Filippas, Horton, and Golden (2017). However, asymmetric information problems are prevalent

in offline marketplaces as well. They have been shown to exist in insurance markets, Finkelstein and McGarry (2006);

credit markets, Crawford,Pavanini, and Schivardi (2018); and financial markets, Ivashina (2009); among many others.

822 C2019, The RAND Corporation.

SAEEDI / 823

However, although the positive role of reputation in overcoming asymmetric information problems

is known, its quantitative effect on marketplaces is still not well understood.2To quantify the

effect of reputation mechanisms, in this article, I develop and estimate a dynamic model for

the eBay reputation mechanism. My quantitative analysis highlights the main channel through

which reputation affects eBay. In par ticular, I show that the equilibrium effects of removing the

reputation mechanism will greatly outweigh the static effect of reputation in terms of the price

premium reputable sellers receive.

The eBay marketplace is plagued by information asymmetries, and the problem is partially

alleviated by the eBay reputation system.3Moreover, as Bar-Isaac and Tadelis (2008) mention,

eBay providesa very good environment for economists to study the effects of reputation on sellers’

actions and profits. First, economists can observe all of the sellers’ characteristics observable by

buyers. Second, sellers and buyers have little to no interactions with each other outside the eBay

website; thus, buyers do not have additional information, unobservable to a researcher, about

sellers. Third, economists can track sellers over time, which gives them extra information about

sellers that is unobservable to buyers. This information can then be used to estimate the underlying

model parameters.

To quantify the value of reputation, I examinesellers on eBay and use a unique data set that

follows them over time. First, I analyze the determinants of price variation in a set of relatively

homogeneous goods (iPods).Second, I develop and estimate a model of sellers’ behaviorover time,

in which sellers have heterogeneous unobserved qualities and build up their reputation by selling

objects and acquiring the eBay registered-store status (eBay store from now on), Powerseller

status, or both. Finally, using the estimated model, I perform counterfactuals to analyze the effect

of reputation on profits and market outcomes, as well as possible alternativemethods to overcome

adverse selection. The counterfactuals highlight the dynamic role of reputation in reducing the

market share of high-quality sellers. Specifically, absent any reputation mechanism, consumer

surplus declines by 35%, most of which is due to a change in market structure, as the market

share of high-quality sellers decreases and the market share of low-quality sellers increases.

In order to empirically analyze the role of reputation, I examine the data on sellers of iPods

between 2008 and 2009. The data set follows sellers on eBay and includes the number of items

sold, the final price of items sold, and items’ and sellers’ characteristics. Consistent with other

studies on eBay, considerable variation exists in the prices of iPods sold. In this context, there

are two main variables of interest related to reputation: Powerseller status and eBay store status.4

Using these two variables as proxies for reputation, I show that reputation has a significant role

in explaining price variations. In particular, the prices of new iPods are positively correlated

with reputation. Among sellers of new iPods, being a Powerseller, ceteris paribus, increases

prices by approximately 3%, whereas being an eBay store, ceteris paribus, increases prices by

approximately 4%.

Although this empirical evidence is suggestive of the value of reputation, it ignores its effect

on sellers’ incentives to build their reputation over time (i.e., achieve Powerseller or eBay store

status). In particular, a high-quality seller without Powerseller or eBay store status might sell a

higher quantity of goods than a low-quality seller in order to become a Powerseller or eBay store,

which results in a change in the market share of sellers of different quality. To quantify this effect,

I develop a dynamic structural model of sellers’ behavior where sellers are privately informed

about their quality and acquire reputation over time.

2For example, see Holmstrom (1999), Mailath and Samuelson (2001), Board and Meyer-ter Vehn (2014), and

Board and Meyer-ter Vehn(2013).

3See, for example, Resnick et al. (2006), Brown and Morgan(2006), Lucking-Reiley et al. (2007), Kollock (1999),

and Yamagishiand Matsuda (2002). Bajari and Hortac¸su (2004) surveys the literature on eBay’s feedback system.

4In this article, I do not use the percentage positive feedback as a measure of reputation. As I show in the article,

sellers on eBay receive predominately positive feedback, which leads to little variation in observable feedback ratings.

Moreover,there is no significant cor relation betweenthis measure and the prices sellers receive.

C

The RAND Corporation 2019.

824 / THE RAND JOURNAL OF ECONOMICS

Formally, the model consists of two sets of agents: buyers and sellers. Buyers are short-lived,

derive utility from purchased goods, and do not observe the quality of the objects bought by

previous buyers. Sellers are long-lived, can sell different quantities of goods over time, and are

heterogeneous in the quality of the goods they are selling. The higher the quality of a good, the

higher the buyer’s utility from purchasing one unit of that good. Sellers’ quality has a persistent

component and transitory independent and identically distributed (i.i.d.) component for each

time period. To capture adverse selection, I assume that qualities are privately known to sellers;

therefore, buyers do not observe the quality of an object.

In this environment, I introduce eBay’s reputation system: eBay store and Powerseller sta-

tuses. High-quality sellers can choose to pay a monthly fee to become an eBay store, and they

must fulfill two requirements to become a Powerseller: sell more than the quantity threshold set

by eBay and havea quality higher than the quality threshold. These statuses are valued by sellers,

because buyers use them as signals to distinguish high-quality sellers from low-quality ones. As

buyers prefer high-quality sellers to low-quality sellers and are willing to pay a higher price for

items sold by high-quality sellers, sellers are willing to pay the monthly fee or sell more than the

static optimum in order to achieve these statuses.

In this article, I solve for a dynamic game between many sellers. Specifically, my data set

contains 769 sellers whose decisions depend on their quantity choices in the past. Due to this

high number of sellers and the large state space at the individual level, the standard methods of

estimation of dynamics games, namely, the estimation of Markov Perfect Equilibrium by Pakes

and McGuire (2001) is impossible—due to the “curse of dimensionality” present in dynamic

models with many state variables, as discussed by Pakes and McGuire (2001). To overcome this

problem and achieve tractability, I use the oblivious equilibrium concept introduced by Weintraub,

Benkard, and Van Roy (2008). Under this equilibrium concept, the state of the game is given by

market aggregates. Roughly speaking, the assumption is that sellers are small and, as a result,

their actions do not affect the state of the industry. This means that any given seller does not

need to take into account other sellers’ response to her decision during this period or the next,

and the seller does not need to keep track of the history of her opponents’ actions. This greatly

reduces the dimensionality of the state space that one needs to keep track of.5Given that the

data set has a large number of sellers and the largest seller accounts for only 5% of the market,

this equilibrium concept is reasonable and closely approximates Markov Perfect Equilibrium, as

discussed by Weintraub, Benkard, and Van Roy (2006).

I estimate the model in two steps. First, I identify the stochastic process for qualities byusing

an important theoretical insight from the model. In particular, I show that in the model, higher-

quality sellers (as measured by their persistent level of quality) must always sell higher quantities.

This relationship allows the parametric estimation of qualities using the quantity choices of sellers

over time. In the next step, using the approach of Bajari, Benkard, and Levin (2007), I identify

sellers’ cost parameters. This procedure is based on the assumption that the observed data are

the outcome of sellers’ maximization problem based on their actual cost function; therefore, the

conditions for optimality of policy functions can be used to back out cost parameters.

Using the above model, I perform two counterfactuals to estimate the added value of having

a reputation system and the effect of substituting it with a warranty mechanism. In the first

counterfactual, I remove eBay’s reputation system altogether and solve the equilibrium in which

sellers have no means of signalling their quality to buyers. Absent any reputation mechanism,

the model becomes a static model with adverse selection. As a result, higher-quality sellers face

lowerprices and therefore lower market shares. In contrast, the market share of low-quality sellers

increases. This decline in average quality in the market leads to a further decrease in prices and

consequently,a shrinkage of the market and its unravelling. Specifically, removing the reputation

system decreases buyers’ surplus by 35%, total sellers’ profit by 66%, and eBay’s profit by 38%.

5To some extent, oblivious equilibrium is similar to monopolistic competition when firms are infinitesimal and

their decisions change their price and profits but do not affect the market as a whole.

C

The RAND Corporation 2019.