HETEROGENEOUS EFFECTS OF ONLINE REPUTATION FOR LOCAL AND NATIONAL RETAILERS

• Author: Xiaolu Zhou,Peter Newberry
• Published date: 01 November 2019
• Date: 01 November 2019
• DOI: http://doi.org/10.1111/iere.12397

INTERNATIONAL ECONOMIC REVIEW

Vol. 60, No. 4, November 2019 DOI: 10.1111/iere.12397

HETEROGENEOUS EFFECTS OF ONLINE REPUTATION FOR LOCAL AND

NATIONAL RETAILERS∗

BYPETER NEWBERRY AND XIAOLU ZHOU1

The Pennsylvania State University, U.S.A.; School of Economics, WISE,

Xiamen University, China

We study the heterogeneous effect of online reputation for sellers that differ in their national presence and

examine how this heterogeneity affects the distribution of sales on a large Chinese platform. We estimate a

demand model that incorporates a learning process and allow for the process to vary across sellers who are

differentiated by their national presence. The estimates suggest that the impact of reputation is larger for local

sellers. Using these estimates, we find that removing the reputation system would result in large shift of demand

from local sellers to national sellers.

1. INTRODUCTION

To lessen the impact of information asymmetries that are due to the impersonal nature of

the Internet, most online markets have peer review systems that provide consumers a signal of

a seller and/or product’s quality (i.e., reputation).2Survey and empirical evidence suggest that

these systems are an important part of a consumer’s decision-making process.3At the same

time, there exists heterogeneity in online sellers in terms of their experience, size, prevalence

in offline markets, etc. In this article, we study the variation in the impact of reputation across

heterogenous sellers in an online marketplace.

Many of the sellers on Alibaba’s Tmall (our data source) are small sellers who only have

a local offline outlet, whereas other sellers are large manufacturers or well-known retailers

who have an extensive offline presence. Absent any measures of reputation, consumers may

concentrate their demand on the offline firms with whom they have interactions outside of

Tmall, making it difficult for lesser-known firms to be able to compete in the marketplace.

However, it may also be the case that consumers view the online version of the national retailer

as different from the offline version, implying that the role of online reputation might not differ

across this dimension. In this article, we ask: To what extent does an online reputation system

determine the distribution of demand across sellers who differ in their national presence?

To answer this question, we quantify the impact of the rating system on Tmall, a branch of

China’s largest e-commerce company, Alibaba. Tmall is a business-to-consumer (B2C) platform

that features thousands of professional sellers offering a wide variety of products. Although

∗Manuscript received April 2017; revised January 2019.

[Correction added on 9 August 2019, after first online publication: The affiliation of Xiaolu Zhou was changed to

School of Economics, WISE, Xiamen University, China.]

1This research is supported by the National Natural Science Foundation of China (grant 71803162). We would like

to thank Paul Grieco, Charles Murry, Mark Roberts, Chris Parker, Xiang Hui, and two anonymous referees for their

helpful comments and suggestions. Please address correspondence to: Peter Newberry, Department of Economics, 510

Kern Building, Pennsylvania State University, University Park, PA 16802. E-mail: pwnewberry@psu.edu.

2For example, Amazon.com displays the distribution of seller and product ratings (1 through 5) given by previous

shoppers, whereas ebay.com has a feedback system in which users indicate whether their experience was positive

or negative.

3For survey evidence, see http://marketingland.com/survey-customers-more-frustrated-by-how-long-it-takes-to-res

olve-a-customer-service-issue-than-the-resolution-38756. Both Chevalier and Mayzlin (2006) and Dellarocas (2003)

provide empirical evidence.

1565

C

(2019) by the Economics Department of the University of Pennsylvania and the Osaka University Institute of Social

and Economic Research Association

1566 NEWBERRY AND ZHOU

there have been numerous previous papers studying Taobao, Alibaba’s consumer-to-consumer

marketplace, this is the first article to use data from Tmall to the best of our knowledge. This

platform is a leader in China’s online B2C market, as it had 54% market share and total

transactions reached $39 billion in Q3 2015.4Tmall features a rating system that is similar to

that of Amazon.com, where each customer purchasing a product from a given seller rates the

quality of the transaction on a scale of 1–5. Customers who arrive thereafter can observe both

the average rating score and the total number of ratings for a given seller.5

Our data include monthly prices and quantities for tablets sold on Tmall between September

2014 and April 2015. We observe both product and seller characteristics, where the latter include

the rating score (i.e., average rating) and the complete distribution of ratings. Additionally, we

observe a classification of sellers, as defined by Tmall, which is based on a seller’s national

presence. The known nationally (hereafter “national sellers”) and those who are local or online

only retailers (hereafter “local sellers”). Our aim is to estimate the impact of the rating system

in determining the concentration of sales across these two types.

We do this in three steps. First, we show descriptive evidence that seller ratings have a larger

impact for local sellers. When accounting for product characteristics and utilizing the covariation

between ratings and sales within a seller, we find evidence that ratings positively affect demand

for local sellers but not national sellers.

Next, we estimate a discrete choice model of demand where the consumer chooses to buy a

tablet from one of the sellers on Tmall, but is uncertain about the quality of the sellers prior to

making the purchase. This uncertainty is a key dimension of differentiation in online markets,

as many sellers offer similar products. Therefore, we enrich the demand model by assuming

that the consumer infers the expected quality of each seller utilizing the average rating and the

number of ratings and a Bayesian learning process. We allow the updating process (i.e., the

learning parameters) to be functions of the seller’s type, along with other seller characteristics.

The results show that consumers have a higher prior belief about national sellers, implying

that consumers believe that these sellers are of higher quality, ex ante. Additionally, we find

that the weight placed on the rating score is significantly higher for local sellers, meaning the

rating score has a larger effect on demand for these sellers. This suggests that consumers rely

more heavily on their prior belief about quality instead of the rating score for national sellers,

whereas they use the rating score to infer quality for local sellers.

Finally, using the estimates of the model, we quantify the effect of the reputation system

on demand across seller types by removing ratings and forcing consumers to use only their

prior beliefs about seller quality to make their purchase decision. When removing ratings and

keeping prices and all other aspects of the market fixed, we find that the total market share

of national sellers increases from 20% to 39%, whereas the total market share of local sellers

decreases from 66% to 31%.6This implies that the lack of ratings leads to an increase of the

outside share, but the primary effect is to redistribute sales between the two different types. The

largest effects are for the biggest sellers, as the total market share of the top 20% of national

sellers increases from 17% to 35% and the total market share for the top 20% of local sellers

drops from 61% to 27%. This implies that the reputation system on Tmall allows the “best”

local sellers to compete with, and even outsell, their national counterparts.

To take a closer look at the substitution between local and national sellers, we calculate the

effect of removing ratings across different quality groups for each type, where a quality group

is defined by the sellers’ rating score at the end of the sample. We find that removing the rating

score increases market share for national sellers at all levels of ratings, even those with a perfect

rating score. On the other hand, the lowly rated local sellers are better off and the highly rated

local sellers are hurt by the removal of ratings, suggesting that the ratings allow the high-quality

4Information from http://www.chinainternetwatch.com/15957/chinas-b2c-sales-q3-2015/.

5See Figures 1 and 2, below, for screen shots of a search page and a seller’s home page.

6In a robustness check, we have computed all exercises allowing sellers to have a price response via a reduced-form

pricing function and the results change very little. See Section 5 for details.