The Review of Economics and Statistics

VOL. CIV

NOVEMBER 2022

NUMBER 6

COLLECTIVE REPUTATION IN TRADE: EVIDENCE FROM
THE CHINESE DAIRY INDUSTRY

Jie Bai, Ludovica Gazze, and Yukun Wang*

Abstract—The existence of collective reputation implies an important ex-
ternality. Among firms trading internationally, quality shocks about one
firm’s products could affect the demand of other firms from the same origin
country. We study such a reputation spillover in the context of a large-
scale scandal that affected the Chinese dairy industry in 2008. Leveraging
detailed firm-product level administrative data and official quality inspec-
tion reports, we document sizable reputation spillovers on uncontaminated
firms. We further investigate potential mechanisms that could mediate the
strength of collective reputation, including information accuracy, observ-
ability of the supply chain, and prior export experience.

IO.

introduzione

IN markets with information frictions, quality shocks about

one firm’s products may impose an externality on other
firms selling similar products. In such settings, when an in-
cident damages the group’s reputation, firms can become
trapped in a low-trust–low-quality equilibrium, and new en-
trants may inherit the damaged reputation of their predeces-
sors. Tirole (1996) formalizes the theory of collective rep-
utation. Empirically, how important is collective reputation,
and what are the potential mechanisms that mediate its ef-
fects? We study collective reputation in the context of trade
and development. In international markets, producers from
the same origin country are often viewed as a group: for ex-
ample, we refer to Swiss watches, French wines, and “made-
in-China” products. For firms in developing countries, Quale
are mostly positioned at the lower end of the value-added
chain and export mainly nonbranded products, a long inter-
national supply chain can make it particularly difficult to trace
products to an individual producer. Di conseguenza, collective rep-
utation becomes especially important in determining firms’
export performance. Infatti, rising safety and quality con-

Received for publication September 23, 2019. Revision accepted for pub-

lication December 10, 2020. Editor: Amit K. Khandelwal.

∗Bai: Harvard Kennedy School; Gazze: University of Warwick, Depart-
ment of Economics; Wang: Cornell University, Department of Economics.
We thank Rodrigo Adao, Abhijit Banerjee, Chris Blattman, Oeindrila
Dube, Ben Faber, Rema Hanna, Asim Khwaja, Rocco Machiavello, Nina
Pavnick, Nancy Qian, Daniel Xu and participants at the HKS Growth Lab
seminar, Microsoft Research lab seminar, Entrepreneurship and Private En-
terprise Development (EPED) in Emerging Economies Workshop, and the
IGC/CDEP/Chazen Firms/Trade/Development conference for helpful com-
menti. We thank Hongyuan Xia for providing excellent research assistance.
All errors are our own.

A supplemental appendix is available online at https://doi.org/10.1162/

cerns regarding goods from developing countries in recent
years can significantly hinder firms from penetrating high-
end markets.1 In a recent survey of over 600 manufacturing
firms in China, firms report lack of reputation and mistrust as
one of the main challenges in exporting to global markets.2
Therefore, understanding how collective reputation spreads
within an industry or a geographic area is important for in-
forming trade and development policies.

We exploit a large-scale quality scandal that affected the
Chinese dairy industry in 2008. Similar to many industries
in developing countries and emerging markets, the Chinese
dairy industry was dominated by a large number of small
and nonestablished players that exhibited rapid growth prior
to the scandal. Using administrative data on quality inspec-
tions conducted by the Chinese government following the
scandal, we identify the firms and the products at each firm
that failed inspections (contaminated product-firm pairs) E
those products and firms for which no contamination was
ever detected.3 We merge the official inspection lists with
Chinese Customs data at the firm-product level and Man-
ufacturing Survey data at the firm level to examine the di-
rect effects of the scandal both on contaminated product-firm
pairs and on other products within the firms with contami-
nation (within-firm spillovers) as well as the effects on firms
with no contamination and on uninspected firms (cross-firm
spillovers).

We begin by showing that the scandal had a large impact
on the overall export performance of the entire dairy sec-
tor in China, thus providing an ideal setting for studying
within-sector spillovers. Using a difference-in-differences
(DD) framework, we find that the average value of dairy ex-
ports plummeted by 68% following the scandal and failed

1A list of food contamination incidents can be found at https://
en.wikipedia.org/wiki/List_of_food_contamination_incidents#2001_to_
present. Recent prominent cases include the Chinese dairy scandal of
2008 and the Brazilian meat scandal of June 2017 (see an Economist
article on the latter incident: https://www.economist.com/news/business/
21719416-chile-china-and-eu-have-banned-some-or-all-countrys-meat-
meat-scandal-brazil).

2The survey was administered by the Jinan Institute for Economic and So-
cial Research (IESR) and the Guangzhou General Administration of Qual-
ity Supervision, Inspection and Quarantine (GAQSIQ). We thank IESR for
sharing the survey instruments and the data.

3We conducted an extensive news search through LexisNexis to cross-

rest_a_01032.

validate the official inspection reports. See section IIIC for more details.

The Review of Economics and Statistics, novembre 2022, 104(6): 1121–1137
© 2021 The President and Fellows of Harvard College and the Massachusetts Institute of Technology
https://doi.org/10.1162/rest_a_01032

l

D
o
w
N
o
UN
D
e
D

F
R
o
M
H

T
T

P

:
/
/

D
io
R
e
C
T
.

M

io
T
.

e
D
tu
/
R
e
S
T
/

l

UN
R
T
io
C
e
–
P
D

F
/

/

/

/

1
0
4
6
1
1
2
1
2
0
5
8
1
9
3
/
R
e
S
T
_
UN
_
0
1
0
3
2
P
D

.

F

B
sì
G
tu
e
S
T

T

o
N
0
8
S
e
P
e
M
B
e
R
2
0
2
3

1122

THE REVIEW OF ECONOMICS AND STATISTICS

to recover even after five years. This estimate captures both
the direct impact on firms with contaminated products and
the spillovers to other firms and products. Decomposing
the direct versus indirect impacts, we estimate an aggregate
spillover effect of 57%, about four-fifths of the total effect
size. Our estimates are robust to various empirical specifica-
tions that relax the classical DD assumption. To the extent
that products by different dairy firms are imperfect substi-
tutes, the estimates provide a lower bound on the collective
reputation effect.

Prossimo, we investigate how the spillover effects are dis-
tributed across different firms and products. We leverage our
detailed micro data on export activities and inspection out-
comes at the firm-product level. Our results suggest that con-
taminated firms saw a drop of 84% in export revenue after the
scandal relative to the national industrial trend. These firms
were also 14.2% less likely to export following the scandal.
Inoltre, we find that both firms verified to be free of
contamination and uninspected firms experienced an equally
significant decline in export revenue of about 64% del
decline suffered by directly affected firms. Inoltre, firms
that successfully passed government inspections did not fare
any better than uninspected firms. Altogether, these findings
point to large reputation spillovers and highlight the potential
challenges of government actions to help firms signal quality
and restore trust.

Finalmente, we investigate potential mechanisms that may un-
derlie the observed spillover effects. Since a firm’s reputation
is essentially constituted by buyers’ beliefs and perceptions,
how consumers gather information and learn matters cru-
cially for reputation externalities. Specifically, we examine
the role of (1) information accuracy in global media reports,
(2) firms’ location and the traceability of suppliers affected
by contamination episodes, E (3) firms’ export experience,
which proxies for the strength of individual reputation.

Primo, to study the role of information accuracy in global
media, we construct a measure of consumers’ knowledge of
the scandal across different countries, using Google Trends
Search indices for phrases that reflect a more or less accurate
depiction of the event. We find that the spillover effects are
smaller in countries where people appear to have better in-
formation about the parties directly involved in the scandal,
reflected in more targeted internet search behavior. In partic-
ular, the cross-firm spillover effects are primarily driven by
exports to destinations with low information accuracy.

Secondo, we use the Chinese Customs data to identify the
sourcing locations of each exporting firm prior to the scan-
dal to examine the impact on firms that passed inspections
and uninspected firms of sourcing from the same locations
as firms affected by the contamination. È interessante notare, we find
that the spillover effects appear to be generalized rather than
contained within specific areas. This is consistent with in-
ternational buyers having coarse information about contam-
ination sources due to low traceability of the contaminated
inputs.

Third, to examine the interaction between collective and
individual reputation, we exploit baseline variation in firms’
export experience, as measured by the number of years a firm
has been exporting and the share of exports in total sales at
baseline. We find that young firms and firms with a smaller
baseline share of exports in total sales are more vulnerable to
the collective reputation shock. The results suggest that hav-
ing a more established individual reputation can (partially)
shield firms from the collective damage, whereas newcom-
ers suffer more from the “original sin” of the predecessors
(Tirole, 1996).

A growing empirical literature studies firm reputation
and quality provision in markets with information frictions
(Banerjee & Duflo, 2000; Jin & Leslie, 2009; Macchiavello,
2010; Cabral & Hortacsu, 2010; Björkman-Nyqvist, Svens-
figlio, & Yanagizawa-Drott, 2013; Bardhan, Mookherjee, &
Tsumagari, 2013). While information frictions appear to play
an important role in international trade (Allen, 2014; Mac-
chiavello & Morjaria, 2015; Startz, 2017), these frictions re-
main understudied (Atkin & Khandelwal, 2020). We build
upon these two bodies of research by examining the role of
group reputation in trade. Our results demonstrate that col-
lective reputation forces can have important implications for
a country’s trade patterns, contributing to a country’s com-
parative advantage.

Despite earlier works on the theory of collective reputa-
zione (Tirole, 1996; Winfree & McCluskey, 2005; Fleckinger,
2007; Levin, 2009; Fishman et al., 2010), empirical stud-
ies on the subject remain scarce (Bai, 2018; Zhao, 2018).
Exploiting a natural experiment, our results explicitly iden-
tify this important source of externalities and illustrate how
its effects are mediated by various informational and market
forces. Two other papers exploit similar natural experiments
to study collective reputation spillovers. Freedman, Kearney,
and Lederman (2012) examine toy recalls in the US and doc-
ument sizable industry-wide spillover effects independent of
origin country.4 Bachmann et al. (2019) find large spillover
effects a year after the Volkswagen scandal on US sales of
other German cars. To the best of our knowledge, our study
is the first one with a long enough post period to examine
the persistence of collective reputation damage. More work
is needed to compare developing and developed countries as
well as different market settings.5

4Our paper also speaks broadly to the literature on quality scandals and
product recalls. Most previous studies in this literature have either relied
on laboratory experiments to examine consumer reactions to hypothetical
product scandals (per esempio., Ahluwalia, Burnkrant, & Unnava, 2000) or focused
primarily on stock market outcomes using an event-study approach (per esempio.,
Davidson & Worrell, 1992). Inoltre, with the exception of Freedman
et al. (2012), most studies focus on losses in firms’ own sales and stock
market price (Van Heerde, Helsen, & Dekimpe, 2007; Jovanovic, 2021)
rather than cross-firm spillovers.

5There is also a literature in agricultural and resource economics that
studies collective reputation in the food and beverage industries in relation to
products and labels such as Bordeaux wine and regional appellations. Most
studies use hedonic price regressions to examine the role of group reputation
(per esempio., Castriota & Delmastro, 2014; Marchini et al., 2014; Gergaud et al.,

l

D
o
w
N
o
UN
D
e
D

F
R
o
M
H

T
T

P

:
/
/

D
io
R
e
C
T
.

M

io
T
.

e
D
tu
/
R
e
S
T
/

l

UN
R
T
io
C
e
–
P
D

F
/

/

/

/

1
0
4
6
1
1
2
1
2
0
5
8
1
9
3
/
R
e
S
T
_
UN
_
0
1
0
3
2
P
D

.

F

B
sì
G
tu
e
S
T

T

o
N
0
8
S
e
P
e
M
B
e
R
2
0
2
3

COLLECTIVE REPUTATION IN TRADE

1123

FIGURE 1.—EVENT STUDY GRAPH: CHINA’S DAIRY EXPORTS

This figure plots Chinese dairy export values and quantities from 2000 A 2013.

Finalmente, our study relates to the broad literature on firm
performance and quality upgrading in development and trade
(De Loecker & Goldberg, 2014). Previous studies have exam-
ined (1) supply-side constraints, including credit access, lack
of quality inputs, and managerial constraints (per esempio., De Mel,
McKenzie, & Woodruff, 2008; Kugler & Verhoogen, 2012;
Banerjee, 2013; Bloom et al., 2013), E (2) demand side
factors, including access to high-income markets (per esempio., Ver-
hoogen, 2008; Park et al., 2010; Manova & Zhang, 2012;
Atkin, Khandelwal, & Osman, 2017). Atkin & Khandelwal
(2020) highlight that information frictions may significantly
inhibit trading opportunities for firms in developing coun-
tries. Our paper is one of the first to document how a poor
collective reputation deriving from information frictions af-
fects firms’ export performance in developing countries.

The remainder of the paper is organized as follows. Sezione
II provides background information on the 2008 scandal, E
section III describes the data. Section IV presents evidence on
the overall impact of the scandal on the dairy industry. Sezione
V analyzes reputation spillovers across firms and products.
Section VI examines mechanisms. Section VII concludes.

II. Background on the 2008 Chinese Dairy Scandal

The Chinese dairy industry exhibited fast growth during
the early 2000s in terms of both domestic production and ex-
ports. Prior to 2008, the industry grew at an average annual

2017). We take advantage of a natural experiment that allows us to relax
the identification assumptions.

rate of almost 24%. Figura 1 shows that the industry’s an-
nual export value increased more than threefold from 2000
A 2007. The number of exporting firms increased from 150 In
2000 A 335 In 2007.6 Nonetheless, In 2007, dairy exports still
constituted a relatively small share of all dairy production in
China and accounted for only $300 million of the country’s $1.2 trillion export revenue. Like many other industries in
developing countries, the Chinese dairy industry was domi-
nated by a large number of small and nonestablished players.
Over the past decade, an increasing number of quality and
safety issues have affected Chinese food products. One of the
most widely known incidents is the distribution of contami-
nated baby formula in September 2008, hereinafter referred
to as “the scandal.” Infant formula had been illegally adulter-
ated with the industrial chemical melamine to mimic a higher
protein content. Melamine, commonly used in the manufac-
ture of plastics, has been linked to an increased risk of kidney
stones (Hau, Kwan, & Li, 2009). The incident led to 4 infant
deaths, 51,900 hospitalizations of children, and the nation-
wide recall of 700 tons of milk powder.

Following the outbreak, the Chinese government quickly
shut down the supplier of the contaminated milk powder,
Sanlu Group, one of the largest dairy firms in China. Tuttavia,

6Figura 1 shows a spike in export growth between 2006 E 2007, Quando
the total value of exports increased by about 138%. This spike was mainly
driven by new firms entering the export market. Specifically, 30 firms
contributed more than 80% of the growth spike; Sopra 50% of the spike
was driven by a single product—milk powder; finally, Sopra 50% del
spike was driven by exports to 6 destinations, namely, Thailand, Germany,
Bangladesh, Taiwan, UAE, and Hong Kong.

l

D
o
w
N
o
UN
D
e
D

F
R
o
M
H

T
T

P

:
/
/

D
io
R
e
C
T
.

M

io
T
.

e
D
tu
/
R
e
S
T
/

l

UN
R
T
io
C
e
–
P
D

F
/

/

/

/

1
0
4
6
1
1
2
1
2
0
5
8
1
9
3
/
R
e
S
T
_
UN
_
0
1
0
3
2
P
D

.

F

B
sì
G
tu
e
S
T

T

o
N
0
8
S
e
P
e
M
B
e
R
2
0
2
3

1124

THE REVIEW OF ECONOMICS AND STATISTICS

further investigations revealed that the adulteration stemmed
from malpractices of some upstream milk producers, rais-
ing the suspicion that more downstream dairy firms could
have been affected.7 This discovery led to three rounds of
government inspections: the first round targeted firms pro-
ducing infant formula—109 firms were inspected, E 22
were found to have traces of melamine in their products.
The next two rounds targeted producers of milk powder and
liquid milk, rispettivamente, and covered most dairy produc-
ers in China. During the second round, the government in-
spected 154 randomly sampled milk powder producers (A-
gether making up over 70% of the market share) out of 290
producers nationwide and found 20 to be contaminated. Dur-
ing the third round, the government targeted another 466 es-
tablished dairy brands with large market shares and found
Quello 9 plants of 3 major brands were contaminated. That said,
large firms appear to have been disproportionately targeted
in the second inspection round as well: in section IIID, we
find that inspected firms are significantly larger in terms of
baseline sales and employment. Our identification strategy
accounts for this imbalance by including a rich set of baseline
controls interacted with time to allow for differential growth
trajectories between large and small firms.8

These inspections uncovered contamination in several
dairy and dairy-related products, including yogurt, milk,
cheese, baby food, and cake. Product recalls were immedi-
ately issued. By the end of 2008, the Chinese government had
issued an official statement that the incident had been fully
addressed and that proper measures had been put in place to
ensure the safety of the dairy products on the market.9 Cor-
roborating the Chinese government’s statement, data from
the EU’s Rapid Alert System for Food and Feed (RASFF),
which publishes safety notifications and recalls for imported
food products, show that notifications related to melamine
contamination in dairy products imported from China surged
in the fall of 2008 but quickly subsided a few months after
the initial outbreak (first figure in appendix A).

the scandal

Despite the official statement,

triggered
widespread fears over food safety in China. Thousands of
Chinese dairy-related products were pulled from supermar-
ket shelves across the world. Some countries stepped up in-
spections for Chinese imports, while others issued explicit
import bans for products containing dairy ingredients from
China. For instance, EU authorities imposed tests on Chinese
imports containing more than 15% of milk powder and an-
nounced a ban on all Chinese products containing milk for
children; the U.S. Food and Drug Administration restricted

7According to the investigation reports, these malpractices were “open
secrets” in the industry. See https://www.wsj.com/articles/SB122567367
498791713.

8In addition to the three big rounds of inspections targeting downstream
dairy firms, the Chinese government conducted checks at upstream facilities
and shut down a number of milk stations. We do not have systematic data
on these upstream inspections, partly because many of the suppliers were
very small and informal.

9http://www.telegraph.co.uk/news/worldnews/3079146/China-claims-

tainted-milk-scandal-is-over.html

imports of all Chinese food products containing milk; India
imposed an import ban on milk and all milk-related products
from China that was extended until 2019. Our news search
identified 43 destinations (out of 157) that imposed explicit
regulatory bans on certain Chinese dairy imports (first table
in appendix B).

The scandal had a long-lasting impact on the dairy industry
in China. The GAQSIQ stopped issuing national exemption
status to domestic food producers10 and tightened inspec-
tions on domestically produced food products. Dairy firms
also tightened their standards for purchased raw milk, E
some started their own upstream milk farms to better control
quality. Tuttavia, figure 1 shows that despite these actions,
dairy exports sharply declined after 2008 and had not recov-
ered by the end of 2013, the end of our sample period. A
the same time, dairy imports in China rose rapidly after the
scandal (second figure in appendix A), suggesting that do-
mestic consumers also switched to foreign dairy products in
response to safety concerns about domestic producers.

III. Data

We merge three microlevel data sets: the Chinese Customs
Database, the Chinese Manufacturing Survey, and the list of
inspections conducted by the Chinese government.

UN. Chinese Customs Database (2000–2013)

The Chinese Customs Database provides information on
trade flows for the universe of China’s exports and imports.
We focus on exports for this study. We observe the exporting
firm’s name, location, export value, and export quantity, IL
HS eight-digit product code, the city in China from which
the product is sourced, and the final export destination. Noi
compute unit prices for exported products by dividing the
value of export by the quantity. For our industry-level analy-
sis in section IV, we aggregate the data to the HS two-digit
industry-year level; for our spillover analysis in sections V
and VI, we aggregate the data to the firm-product-year level.11
Third figure in appendix A shows that China’s overall exports
grew rapidly in the early 2000s following China’s entry into
the World Trade Organization (WTO).

We define the dairy industry using the HS eight-digit prod-
uct classification. Most dairy products fall under the HS two-
digit code 04, while infant dairy products fall under 19 E
milk protein products extracted from raw milk fall under 35.
Second table in appendix B provides the full list of the HS
eight-digit codes and descriptions for dairy products.

10This policy was previously known as the “inspection-free” program,
which gave exemption status to qualified food producers and waived various
quality inspections for these firms.

11For observations from years prior to 2006, the data include exact the
export date. For years after 2007, only monthly data are available. Noi
collapse the data to the year level to account for monthly seasonality in
exports.

l

D
o
w
N
o
UN
D
e
D

F
R
o
M
H

T
T

P

:
/
/

D
io
R
e
C
T
.

M

io
T
.

e
D
tu
/
R
e
S
T
/

l

UN
R
T
io
C
e
–
P
D

F
/

/

/

/

1
0
4
6
1
1
2
1
2
0
5
8
1
9
3
/
R
e
S
T
_
UN
_
0
1
0
3
2
P
D

.

F

B
sì
G
tu
e
S
T

T

o
N
0
8
S
e
P
e
M
B
e
R
2
0
2
3

COLLECTIVE REPUTATION IN TRADE

1125

B. Chinese Manufacturing Survey (2005–2009, 2011–2013)

D.

Summary Statistics

The Chinese Manufacturing Survey data are compiled
from annual surveys conducted by the National Bureau of
Statistics (NBS) and include all state-owned and non-state-
owned industrial firms with sales revenue above 5 million
RMB. Even though a large number of small to medium-size
industrial firms (80%) are excluded from the sample, Essi
account for only a small fraction of the total economic and
export activities in China. In particular, the excluded firms
employ 28.8% of the industrial workforce but produce only
9.3% of the total output and generate 2.5% of the export
revenue (Brandt, Van Biesebroeck, & Zhang, 2012). Nostro
spillover analysis in sections VB and VIC focuses on dairy
firms within the manufacturing sector. For each firm-year,
we observe production and financial information, including
firms’ four-digit industry code, years of operation, total sales,
employment, and export revenue. When export revenue is
missing, we complement the observation using the Customs
dati: we follow the standard procedure for matching firms
across the Manufacturing Survey and the Customs data us-
ing name and address information.12

C. Government Inspection Lists

Section II describes the three rounds of inspections imple-
mented by the Chinese government after the scandal. For
each round, the government released the list of products
inspected at each firm and the inspection results. We ob-
tained the inspection lists (at the firm-product level) from the
GAQSIQ website. To cross-validate the information in the of-
ficial reports, we conducted an extensive news search through
LexisNexis: all the media-reported cases of contamination
that we found appeared in the official inspection lists.

We merge the firm-product-level inspection lists with the
Customs data using firm names and product information and
with the Manufacturing Survey data using firm names (since
the latter data do not include product information). We clas-
sify merged firms into one of three categories: contaminated,
innocent, and uninspected. Contaminated firms have at least
one product found to be contaminated during one or more
rounds of inspections. Innocent firms passed the tests for all
of their inspected products. Uninspected firms were never
inspected. We analogously classify products into the follow-
ing categories: contaminated products are those found to be
contaminated in at least one of the inspected firms, and in-
nocent products are those that cleared inspection in all firms
inspected.

12This matching is not 100% accurate, as documented in the literature
(per esempio., Kee & Tang, 2016). Of the 335 dairy firms appearing in the Customs
data in 2007, we can identify 151 in the Manufacturing Survey. The lack
of matches for the remaining 184 firms could be due to mismatches in firm
names or the fact that the Manufacturing Survey includes only above-scale
firms.

Our spillover analysis in section V uses two samples. Primo,
we use our linked Customs-Inspections sample to study the
scandal’s spillover effects on the export performance of dairy
firms at the firm-product level. This sample includes 1,868
firm products across 1,464 firms and 25 prodotti. Of these,
we identify 149 contaminated firm-product pairs in 67 con-
taminated firms and 413 innocent firm-product pairs in 95
innocent firms. Secondo, we use our linked Manufacturing
Survey-Inspections sample to study the scandal’s spillover
effects on the domestic performance of the dairy firms. Questo
sample includes 1,687 firms, 73 of which were contaminated
E 352 innocent. Restricting the sample to firms that appear
in the Manufacturing Survey both before and after the scan-
dal reduces the sample to 238 firms: 19 contaminated and
103 innocent.

Tavolo 1 presents firm-level baseline (2000–2007) sum-
mary statistics for export performance measures in the linked
Customs-Inspections sample (panel A) and for employment
and total sales in the Manufacturing Survey-Inspections sam-
ple (panel B). It compares contaminated (columns 1 E 2),
innocent (columns 3 E 4), and uninspected firms (columns
5 E 6). Column 7 calculates the difference in means be-
tween contaminated and uninspected firms (columns 2 E
6), and column 8 reports the p value of this difference. On av-
erage, contaminated firms have larger export revenue and are
more experienced than innocent firms; Tuttavia, they are not
systematically different from uninspected firms (panel A).
Contaminated firms are larger in terms of both employment
and sales revenue than both innocent firms and uninspected
firms (panel B). This pattern is consistent with the Chinese
government targeting larger firms in the third round of in-
spections.13 Even for the second round of inspections, Quale
were claimed to be random, inspected firms appear to differ
from uninspected firms on several observable characteristics,
including the value of exports in 2007 (second table in ap-
pendix A). Section V discusses how our empirical framework
addresses selection into inspection.

Tavolo 2 presents product-level baseline summary statistics
for the same export performance measures in the Customs-
Inspections sample for dairy (panel A) and nondairy (panel
B) prodotti. On average, contaminated products are exported
in larger quantities and for longer export periods than inno-
cent and uninspected products. Contaminated products also
appear to be less likely to be exported to countries in the
Organisation for Economic Cooperation and Development
(OECD) although the difference is not significant.

Fourth figure in appendix A plots the number of ex-
port products and destinations for contaminated, innocent,
and uninspected firms before and after the scandal. Many
firms exported a single product to a single destination. In

13First table in appendix A shows similar patterns for firms producing

nondairy food products.

l

D
o
w
N
o
UN
D
e
D

F
R
o
M
H

T
T

P

:
/
/

D
io
R
e
C
T
.

M

io
T
.

e
D
tu
/
R
e
S
T
/

l

UN
R
T
io
C
e
–
P
D

F
/

/

/

/

1
0
4
6
1
1
2
1
2
0
5
8
1
9
3
/
R
e
S
T
_
UN
_
0
1
0
3
2
P
D

.

F

B
sì
G
tu
e
S
T

T

o
N
0
8
S
e
P
e
M
B
e
R
2
0
2
3

1126

THE REVIEW OF ECONOMICS AND STATISTICS

TABLE 1.—BASELINE SUMMARY STATISTICS: DAIRY FIRMS

Contaminated

Innocent

Uninspected

Contaminated vs. Uninspected

Numero
(1)

Mean
(2)

Numero
(3)

Mean
(4)

Numero
(5)

Mean
(6)

Difference
(7)

p value
(8)

Panel A. Customs database

23
–
23
–
23
–

.49
(.91)
3.39
(2.37)
.39
(.42)

Panel B. Manufacturing survey

244
–
244
–
243
–
243
–

300
(554)
5.06
(1.01)
113
(222)
3.76
(1.26)

3.32
(6.16)
4.6
(2.92)
.51
(.44)

1087
(1979)
5.79
(1.57)
1038
(2193)
5.17
(2.02)

960
–
960
–
960
–

808
–
808
–
780
–
780
–

15
–
15
–
15
–

41
–
41
–
41
–
41
–

1.85
(6.62)
4.22
(2.43)
.6
(.39)

185
(292)
4.63
(1.01)
87
(184)
3.38
(1.33)

1.47
(1.55)
.38
(.73)
−.09
(.11)

902
(306)
1.16
(.25)
951
(339)
1.80
(.32)

.34
–
.61
–
.41
–

.003
–
0
–
.005
–
0
–

Avg. yearly export revenue

(in million dollars)

Years of exporting

% exports to OECD countries
(conditioning on exporting)

Employment

Log (employment)

Sales revenue (in million RMB)

Log (sales revenue)

For each firm, the sample includes only dairy products. Panel A uses the baseline (2000–2007) Customs database linked with the inspection list to identify contaminated, innocent and uninspected firms. Panel B
uses the baseline (2005–2007) Manufacturing Survey data, also linked with the inspection list. The unit of observation is collapsed to the firm level. Columns 1, 3, E 5 show the number of firms in each category.
Columns 2, 4, E 6 show the mean of selected variables in each subsample. For these variables, column 7 calculates the difference between contaminated firms (column 2) and uninspected firms (column 6), obtained
by regressing the outcome variable on a contaminated group dummy. Column 8 presents the p value of the difference. Standard deviations are in parentheses for columns 2, 4 E 6. For column 7, robust standard errors
are in parentheses.

TABLE 2.—BASELINE SUMMARY STATISTICS: DAIRY AND DAIRY-RELATED PRODUCTS

Contaminated

Innocent+Uninspected

Numero
(1)

Mean
(2)

Numero
(3)

Panel A. Dairy products

11
–
11
–
11
–

16
–
16
–
16
–

6.36
(11.12)
7.64
(1.21)
.55
(.25)

12
–
12
–
12
–

Panel B. Nondairy food products

38.53
(51.01)
7.88
(.50)
.80
(.13)

944
–
944
–
944
–

Mean
(4)

3.99
(6.41)
6.42
(1.98)
.65
(.33)

21.23
(67.02)
6.09
(2.51)
.76
(.26)

Difference
(5)

p value
(6)

2.37
(3.83)
1.22
(.68)
−.10
(.12)

17.30
(12.55)
1.79
(.15)
.04
(.03)

.542
–
.086
–
.437
–

.168
–
0
–
.204
–

Avg. yearly export revenue

(in million dollars)

Number of exporting years

% to OECD countries

Avg. yearly export revenue

(in million dollars)

Number of exporting years

% to OECD countries

The sample is obtained from the baseline (2000–2007) Customs database linked with the inspection list to identify contaminated, innocent and uninspected products. Panel A only includes dairy products, while
panel B includes nondairy food products. The unit of observation is collapsed to the product (HS eight-digit) level. Columns 1 E 3 show the number of products (HS eight-digit) in each category. Columns 2 E 4
show the mean of selected variables in each subsample. Column 5 calculates the difference between contaminated products (column 2) and innocent plus uninspected products (column 4), obtained by regressing the
outcome variable on a contaminated group dummy. Column 6 presents the p value of the difference. Standard deviations are in parentheses for columns 2 E 4. For column 5, robust standard errors are in parentheses.

section V, we discuss how these patterns generate the varia-
tion that we exploit for the firm-product-level analysis.

decreased the total value of dairy exports by 68% over the
course of five years motivates us to further explore within-
sector spillovers across firms and products in section V.

IV.

Impact of the Scandal: Industry-Level Analysis

UN. Empirical Specification

This section estimates the overall impact of the scan-
dal on the export performance of the Chinese dairy sec-
tor. Section IVA discusses our preferred empirical specifica-
zione, a difference-in-differences (DD) specification, anche
as threats to the validity of the DD assumptions and addi-
tional tests we perform that relax these assumptions. Sezione
IVB presents our DD estimates. Our finding that the scandal

Equazione (1) presents our baseline DD specification, Quale
compares the value of exports of the dairy industry (IL
treated industry) to the value of exports of other two-
digit-level industries before and after 2008, the year of the
scandal.

Yjt = βdairyDairy j

× Postt + γ j + δt + πX jt + (cid:2) jt .

(1)

l

D
o
w
N
o
UN
D
e
D

F
R
o
M
H

T
T

P

:
/
/

D
io
R
e
C
T
.

M

io
T
.

e
D
tu
/
R
e
S
T
/

l

UN
R
T
io
C
e
–
P
D

F
/

/

/

/

1
0
4
6
1
1
2
1
2
0
5
8
1
9
3
/
R
e
S
T
_
UN
_
0
1
0
3
2
P
D

.

F

B
sì
G
tu
e
S
T

T

o
N
0
8
S
e
P
e
M
B
e
R
2
0
2
3

COLLECTIVE REPUTATION IN TRADE

1127

Dep Var: Log (export value)

(1)

(2)

(3)

(4)

TABLE 3.—IMPACT OF THE SCANDAL ON EXPORTS: INDUSTRY-LEVEL ANALYSIS

DairyXPost

FoodXPost

R-squared
Observations

DairyXPost

FoodXPost

R-squared
Observations
Year, Industry FEs
YearXValue Share to different continents
Industry time trends
Excluding the food sector

Panel A. All dairy exports

−1.065***
(0.080)

0.964
1120

−0.892***
(0.114)
−0.174
(0.139)
0.963
1386

Panel B. Innocent+uninspected firm products only

−0.915***
(0.080)

0.964
1120
YES
NO
NO
YES

−0.741***
(0.114)
−0.174
(0.139)
0.963
1386
YES
NO
NO
NO

−0.685***
(0.238)

0.970
1107

−0.554**
(0.233)

0.970
1107
YES
YES
NO
YES

−1.140***
(0.086)

0.980
1120

−0.848***
(0.086)

0.980
1120
YES
NO
YES
YES

This table shows the regression results for equation (1). Panel A contains all exporters, collapsed to the industry-year level. Panel B excludes contaminated firm products to quantify the aggregate spillover effect. Noi
create a balanced panel at the industry (HS two-digit) and year level. The dependent variable is log annual export value for each industry. The baseline specification in column 1 includes only year and industry fixed
effects. Columns 3 E 4 build on this specification by adding time-varying controls, including the value share exported to different continents at baseline (2000–2007) interacted with year indicators and industry-specific
linear time trends. Columns 1, 3, E 4 exclude nondairy food industries; column 2 includes all HS two-digit industries. Standard errors clustered at the product (HS two-digit) level. *, **, E *** denote significance
at the 10%, 5%, E 1% level, rispettivamente.

Yjt is the natural logarithm of the value of exports for indus-
try j in year t; Dairy j is an indicator for the dairy industry;
and X jt includes time-varying controls at the industry level,
such as the value share exported to different continents at
baseline (2000–2007) interacted with year indicators. These
covariates control for differential trends in destination coun-
tries that may affect different industries differently. In our
preferred specification, we include 79 non-food control in-
dustries. Nondairy food exports may also be affected by the
scandal if foreign consumers update their perceptions about
the quality of Chinese food products in general. We cluster
standard errors at the industry level, allowing for arbitrary
correlation in error terms across time for a given industry.

The internal validity of the DD design rests on the assump-
tion that the treated and control industries would be on paral-
lel trends absent the scandal. This parallel trends assumption
may not hold in our context for two reasons. Primo, as dis-
cussed in section II, prior to the scandal, dairy exports grew
rapidly, a growth episode that might not be paralleled in other
industries. Secondo, the global financial crisis in 2008 could
have affected different industries differently. If so, we may
erroneously attribute to the scandal an export decline in dairy
products that was in fact due to the crisis. While perfectly
overcoming these concerns is challenging in the current con-
testo, we perform a series of robustness checks, detailed in
appendix C. Our estimates of the impact of the scandal are
robust to several specifications that relax the parallel trends
assumption, such as including a vector of industry-specific
linear time trends. Inoltre, an interactive fixed effect model
(Gobillon & Magnac, 2016) and a synthetic control model
produce qualitatively and quantitatively similar estimates to
those produced by the DD design, despite relying on different
identification assumptions.

B. Results: Difference-in-Differences

Tavolo 3 presents estimates of equation (1). We focus first
on panel A. The baseline specification in column 1 includes
only year and industry fixed effects and estimates a decline
Di 65.5% in the value of dairy exports following the scan-
dal.14 Columns 3 E 4 build on this specification by adding
time-varying controls and industry-specific linear trends. Nostro
preferred specification in column 4 estimates that the value
of dairy exports plummeted by 68% following the scandal.
Column 2 expands our sample to include nondairy food in-
dustries. The coefficient on the interaction between the food-
industry indicator and the post-scandal indicator suggests that
the scandal did not affect the nondairy food industry in an
economically and statistically significant way.

Panel A estimates the overall impact of the scandal on the
dairy industry, capturing both the direct impact on contami-
nated firm products and the spillovers to innocent and unin-
spected firms and products. Panel B disentangles the aggre-
gate spillover effect by excluding contaminated firm products
from the treated (dairy) industry. We estimate a 57% decrease
in exports for innocent and uninspected firm products, Di
four-fifths of the total effect of the scandal.15 The difference
between the spillover and the total effect is only statistically
significant in our preferred specification in column 4.

14Because most of the coefficients that we estimate are large in mag-
nitude, we compute elasticities using the following formula: Elasticity =
(eCoeff − 1) × 100, where Coeff is the estimated coefficient reported in the
tables.

15An analogous analysis shows that dairy imports significantly increased
relative to imports in other sectors after 2008 (third table in appendix
UN) as domestic consumers switched to foreign brands following the
crisis.

l

D
o
w
N
o
UN
D
e
D

F
R
o
M
H

T
T

P

:
/
/

D
io
R
e
C
T
.

M

io
T
.

e
D
tu
/
R
e
S
T
/

l

UN
R
T
io
C
e
–
P
D

F
/

/

/

/

1
0
4
6
1
1
2
1
2
0
5
8
1
9
3
/
R
e
S
T
_
UN
_
0
1
0
3
2
P
D

.

F

B
sì
G
tu
e
S
T

T

o
N
0
8
S
e
P
e
M
B
e
R
2
0
2
3

1128

THE REVIEW OF ECONOMICS AND STATISTICS

V. Reputation Spillover: Firm-Product-Level Analysis

Motivated by the findings in the previous section, we ex-
amine how the impact of the scandal spilled over across firms
and products within the dairy sector. We study both the direct
impact on contaminated firm products and the spillover ef-
fects within and across firms in the industry. To do so, noi usiamo
our Customs-Inspection sample described in section III to ex-
ploit within-industry variation in involvement in the scandal.
Our main regression specification is as follows:

Yf pt = βdirectCFirm-Product f p × Postt
+ βwithin-firmCFirm f × Postt
+ βacross-firmCProductp × Postt
+ λ1IFirm-Product f p × Postt
+ λ2IFirm f × Postt
+ γ f p + δt + (cid:2) f pt .

(2)

We restrict the analysis to the dairy industry. The depen-
dent variable Yf pt is an outcome for firm f ’s product p in year
T, including the inverse hyperbolic sine (IHS) transformation
of export revenue and export quantity, the natural logarithm
of export price, and an indicator for exporting.16 Except for
the price outcome, we first create a balanced panel at the
firm-product and year levels to capture extensive-margin re-
sponsorizzato (cioè., entry and exit into export). CFirm-Product f p is
an indicator for contaminated firm-product pairs directly in-
volved in the scandal: the indicator equals 1 if a given product
of a firm was inspected and failed the quality test. CFirm f is
an indicator for contaminated firms: the indicator equals 1 if
a firm was inspected and at least one of its products failed the
test. CProductp is an indicator for contaminated products: IL
indicator equals 1 if at least one of the inspected firms failed
the quality test for the given product. IFirm-Product f p and
IFirm f are defined similarly: IFirm-Product f p is an indica-
tor for innocent firm-product pairs that were inspected and
passed the quality test. IFirm f is an indicator for innocent
firms that passed the quality test for all of their inspected
prodotti.

Identification relies on the assumption that unobserved
firm-product-year-specific shocks that affect the outcomes
are uncorrelated with the initial inspection status. In other
parole, absent the scandal, all firm products would have seen
the same growth in export performance over time. Tuttavia,
the Chinese government did not choose which firms to in-
spect randomly. As discussed in section IIID, inspected firms
are on average larger than uninspected firms.

To assuage concerns of omitted variable or selection bias,
our preferred specification includes (1) firm-product (γ f p)
E (2) year (δt ) fixed effects as well as (3) an interaction of

16We use the IHS transformation for export revenues and quantity to
obviate the fact that we have missing firm-product-year cells when exports
are zero (Burbidge, Magee, & Robb, 1988).

baseline firm-product export volume with the post-scandal
indicator. Primo, this specification partials out time-invariant
firm-product unobservable characteristics, such as quality.
Secondo, it controls for common nationwide dairy industry
time trends, such as global demand shocks. Third, differen-
tial trends for firms of different sizes account for potentially
different growth trajectories during the dairy boom prior to
the scandal. Thus our analysis captures differential changes
in performance across firm products over time. To examine
the sensitivity of our results, we also estimate an alternative
specification including firm, product and year fixed effects
separately (fourth table in appendix A). We cluster standard
errors two ways at the product-year and firm level, allowing
for arbitrary correlation in error terms over time for a given
firm and across firms for a given product-year. This two-way
clustering allows for persistent shocks within a firm over time
as well as for cross-sectional yearly shocks affecting all firms
producing the same products.

The omitted category in equation (2) includes innocent
products and uninspected products from uninspected firms.
Therefore, βacross-firm identifies the impact of the scandal
on uninspected firms selling one of the products found to
be contaminated in other firms (cioè., cross-firm spillovers).
βwithin-firm identifies the overall impact on the contaminated
firms (cioè., within-firm spillovers), whereas βdirect captures the
additional impact on their directly involved products. Simi-
larly, λ1 and λ2 capture the impact on innocent firms and
innocent firm products relative to the omitted group. Given
that the regression is done at the firm-product-year level,
fifth table in appendix A presents an overview of the vari-
ation in the data by counting the number of observations
falling under different cells, namely, contaminated versus
noncontaminated firms and products before and after the
scandal.

We present our baseline estimation results in section VA
and discuss their interpretation in section VB as well as al-
ternative explanations and robustness checks in section VC.

UN. Results: Direct and Spillover Effects on Exports

Tavolo 4 reports the main estimates from equation (2).
Column 1 examines the impact of the scandal on the IHS
of export revenue and shows large within-firm and cross-
firm spillovers. Specifically, we estimate ˆβwithin-firm = −1.8,
which is significant at the 1% level; questo è, contaminated
firms experienced a drop of 84.1% in export revenue after
the scandal relative to the national trend and the firms’ av-
erage performance. Within contaminated firms, directly in-
volved products were affected more—the estimated coeffi-
cient βdirect is meaningful (−0.489, or −38.7%) but impre-
cisely estimated, suggesting that there may be heterogeneous
impacts among directly contaminated firm products. Another
possibility is that since most products within the contami-
nated firms were affected, the coefficient is only picking up
any differential impact of contaminated versus innocent and
uninspected products within those firms.

l

D
o
w
N
o
UN
D
e
D

F
R
o
M
H

T
T

P

:
/
/

D
io
R
e
C
T
.

M

io
T
.

e
D
tu
/
R
e
S
T
/

l

UN
R
T
io
C
e
–
P
D

F
/

/

/

/

1
0
4
6
1
1
2
1
2
0
5
8
1
9
3
/
R
e
S
T
_
UN
_
0
1
0
3
2
P
D

.

F

B
sì
G
tu
e
S
T

T

o
N
0
8
S
e
P
e
M
B
e
R
2
0
2
3

COLLECTIVE REPUTATION IN TRADE

1129

TABLE 4.—IMPACT OF THE SCANDAL ON EXPORTS: FIRM-PRODUCT-LEVEL ANALYSIS

IHS (Value)
(1)
−0.489
(1.180)
−1.838***
(0.437)
−0.773***
(0.281)
1.083
(0.978)
−0.944
(0.768)
0.285
13775
YES
YES
YES

IHS (Quantity)
(2)
−0.345
(1.157)
−1.811***
(0.456)
−0.757***
(0.257)
0.981
(0.929)
−0.847
(0.708)
0.299
13775
YES
YES
YES

Log (Price)
(3)
−0.122
(0.091)
0.209***
(0.051)
−0.157*
(0.087)
−0.211**
(0.083)
0.219**
(0.097)
0.903
1519
YES
YES
YES

Exporting (dummy)
(4)
−0.019
(0.078)
−0.153***
(0.031)
−0.064***
(0.023)
0.081
(0.071)
−0.081
(0.063)
0.211
13775
YES
YES
YES

CFirm-ProductXPost

CFirmXPost

CProductXPost

IFirm-ProductXPost

IFirmXPost

R-squared
Observations
Firm-product FE
Year FE
BaselineSizeXPost

This table shows the regression results for equation (2). The unit of observation is at the firm-product-year level. The sample contains all dairy exporters (excluding intermediaries) in the Chinese Customs data
(2000–2013). We rectangularize the data to create a balanced panel at the firm-product (HS eight-digit) and year level for the outcomes in columns 1, 2, E 4. Columns 1 E 2 present results for the inverse
hyperbolic sine (IHS) transformation of the outcome variables of interest, export value and export quantity. Column 3 presents results for the natural logarithm of unit price, while column 4 uses an indicator for
positive exports as the outcome variable. The interaction terms are the products of the post-scandal dummy (2009–2013) with the following five group indicators: (C)ontaminatedFirm-Product, (C)ontaminatedFirm,
(C)ontaminatedProduct, (IO)nnocentFirm-Product, E (IO)nnocentFirm. The omitted category includes innocent and uninspected products from uninspected firms. All regressions include firm-product and year fixed
effects. Baseline size measures a firm’s baseline (2000–2007) total export volume. Standard errors are two-way clustered at the firm and product-year level. *, **, E *** denote significance at the 10%, 5%, E 1%
level, rispettivamente.

In line with the industry-level analysis in section IV, we
also see a large negative spillover effect on firms selling con-
taminated products: the estimate for βacross-firm is −0.773 (O
−53.8%) and is significant at the 1% level. Finalmente, the ef-
fects on innocent firms and products are mixed: while the
coefficient on IFirm-Product×Post is large and positive, IL
overall impact on innocent firms, relative to uninspected
firms, is negative. Neither of these estimates is statistically
significant.

Column 2 examines the effects of the scandal on the IHS
of export quantity and finds similar results. Comparing the
estimates of βacross-firm in columns 1 E 2, the decrease in
quantity explains 97.9% of the cross-firm spillovers, after
entry and exit are taken into account.

Column 3 examines changes in unit price on the unbal-
anced panel of firm-product-year observations with posi-
tive export activity. The estimate of βacross-firm is −0.157
(−14.5%) and is significant at the 10% level. The direct
impact on contaminated firm products is −0.122 (−11.5%)
but is not statistically significant, whereas the within-firm
spillover ˆβwithin-firm is positive at 0.209 (18.9%) with a stan-
dard error of 0.051. One possible explanation for this positive
within-firm price coefficient is that contaminated firms dis-
proportionately raised the prices of their noncontaminated
products to make up for lost revenue. Alternatively, con-
taminated firms may have incurred larger cost shocks (per esempio.,
costs of recalling products or clearing government inspec-
zioni) than noncontaminated firms and partially passed those
costs through to final consumers. The estimated βwithin-firm
captures the net of these additional spillover effects in addi-
tion to potential reputation spillovers.

Column 4 examines the impact of the scandal on the exten-
sive margin and finds that contaminated firms are 14.2% less
likely to export after the scandal. The estimate for βwithin-firm

is significant at the 1% level, whereas the one for βdirect is
indistinguishable from 0. All Chinese dairy firms carrying
products found to have been contaminated during the scan-
dal, regardless of whether the firms themselves are innocent,
contaminated or not inspected, are 6.2% less likely to export
those products, and the estimate of βacross-firm is significant at
IL 1% level. The results on innocent firms and products are
again inconclusive.

We examine how persistent the direct and spillover effects
are by fully interacting the firm-product group dummies in
equation (2) with year dummies. Figura 2 plots the regression
coefficients with 95% confidence intervals for two outcome
variables: log value of exports and exporting indicator. Both
the within-firm and cross-firm spillover effects persist more
than five years after the scandal and display little sign of
recovery.

Our findings contrast with estimates of reputation spillover
effects in developed countries. Per esempio, Freedman et al.
(2012) and Bachmann et al. (2019) document cross-firm rep-
utation spillover effects for US and German manufacturers,
rispettivamente, that are much smaller than what we estimate
for Chinese dairy firms. This difference in magnitude sug-
gests that information frictions may be larger in developing
countries.17 Mistrust of government agencies in developing
countries may make it more difficult for firms to recover from
damages to collective reputation. More work is needed to
draw systematic comparisons across different countries and
market environments.

17Specifically, Bachmann et al. (2019) finds that other German auto man-
ufacturers experienced a 9.2 percentage point reduction in sales growth rate
from 2015 A 2016 following the VW scandal. Freedman et al. (2012) finds
that sales revenue during the Christmas holiday season declined by 30%
for unaffected toy manufacturers in the US after a related scandal.

l

D
o
w
N
o
UN
D
e
D

F
R
o
M
H

T
T

P

:
/
/

D
io
R
e
C
T
.

M

io
T
.

e
D
tu
/
R
e
S
T
/

l

UN
R
T
io
C
e
–
P
D

F
/

/

/

/

1
0
4
6
1
1
2
1
2
0
5
8
1
9
3
/
R
e
S
T
_
UN
_
0
1
0
3
2
P
D

.

F

B
sì
G
tu
e
S
T

T

o
N
0
8
S
e
P
e
M
B
e
R
2
0
2
3

1130

THE REVIEW OF ECONOMICS AND STATISTICS

FIGURE 2.—EFFECTS OVER TIME: EXPORT VALUE (LEFT) AND EXPORT DUMMY (RIGHT)

l

D
o
w
N
o
UN
D
e
D

F
R
o
M
H

T
T

P

:
/
/

D
io
R
e
C
T
.

M

io
T
.

e
D
tu
/
R
e
S
T
/

l

UN
R
T
io
C
e
–
P
D

F
/

/

/

/

1
0
4
6
1
1
2
1
2
0
5
8
1
9
3
/
R
e
S
T
_
UN
_
0
1
0
3
2
P
D

.

F

B
sì
G
tu
e
S
T

T

o
N
0
8
S
e
P
e
M
B
e
R
2
0
2
3

This figure plots the regression coefficients of the following three group dummies interacted with year dummies: ContaminatedFirm-Product, ContaminatedFirm and ContaminatedProduct. The same regressions also
include the interaction terms between year dummies and InnocentFirm-Product as well as InnocentFirm dummies; these coefficients are not plotted. The outcome variable for the left column is the inverse hyperbolic
sine (IHS) transformation of export value, and the one for the right column is the export dummy. All regressions include firm-product and year fixed effects. The dotted lines plot the 95% confidence intervals, based
on two-way clustered standard errors at the firm and product-year level.

B.

Interpretation: The Role of Government Inspections

We find large cross-firm spillovers in export performance,
including for uninspected firms. Tuttavia, we find mixed evi-
dence of reputation spillovers to innocent firms and products.
These findings highlight the potential challenges of govern-
ment inspection efforts in helping firms signal quality. Primo,
the public may perceive government inspections to be a neg-
ative signal. Being inspected may have been taken to indicate
that something was not right (“bad signaling effect”), even
though some inspections were allegedly targeted at random.
Secondo, being mentioned in news reports on the scandal may

impose a stigma if customers do not pay attention to the de-
tails of the news. Mamma, Wang, and Khanna (2016) discuss this
“reminder (salience) effect.”

Both the bad signaling and reminder effects may also be
present in the domestic market, potentially making govern-
ment inspections ineffective. The second figure in appendix
A and the third table in appendix A show that the scandal
increased imports, suggesting mistrust of domestic brands.
To shed light on domestic spillover effects, we turn to the
Manufacturing Survey data, which are only available at the
firm level. We restrict the analysis to a balanced sample of

COLLECTIVE REPUTATION IN TRADE

1131

TABLE 5.—IMPACT OF THE SCANDAL ON FIRMS’ DOMESTIC PERFORMANCE

Log (total sales
revenue)
(1)
−0.315
(0.255)
−0.036
(0.089)
0.884
1664
YES
YES
YES

Log (domestic
sales revenue)
(2)
−0.309
(0.255)
−0.027
(0.091)
0.882
1665
YES
YES
YES

Log
(employment)
(3)
−0.181
(0.235)
−0.099
(0.067)
0.839
1666
YES
YES
YES

CFirmsXPost

IFirmsXPost

R-squared
Observations
Firm FE
Year FE
BaselineSizeXPost

This table shows the regression results for the effects of the scandal on firms’ sales and employment.
The unit of observation is at the firm-year level. The sample includes the balanced sample of dairy firms
in the Manufacturing Survey data (2005–2009 and 2011–2013). We compute a firm’s domestic sales by
subtracting export sales from total sales; in cases where export information is missing in the Manufacturing
Survey data, we merge the sample with the Customs data and fill in the missing export sales information
whenever we can. The interaction terms are the post-scandal indicator (2009–2013) with the following two
group indicators: (C)ontaminatedFirm and (IO)nnocentFirm. The omitted category is uninspected firms. Tutto
regressions include firm and year fixed effects. Baseline size measures a firm’s total sales revenue from
2005 A 2007. Standard errors are clustered at the firm level. *, **, E *** denote significance at the 10%,
5%, E 1% level, rispettivamente.

firms to account for survey composition changes.18 In total,
238 dairy firms appeared in all years between 2005–2009
and 2011–2013, out of 1,687 that ever appeared during this
period. These firms account for 49.8% of the total dairy pro-
duction during this period.

Tavolo 5 estimates the impact of the scandal on this sample
of continuing firms. We cluster standard errors at the firm
level, as product-level information is not available in this
sample. The coefficients are not precisely estimated. Quali-
tatively, we see a negative impact of the scandal on contam-
inated firms, while innocent firms do not appear to perform
better than uninspected ones. These findings are consistent
with the bad signaling and reminder effects acting on the do-
mestic market and highlight the challenges that governments
may face in restoring trust. These results speak to the impor-
tance of understanding how consumers acquire information.
We come back to this point in section VI when we discuss
the mechanisms underlying the reputation spillover.

C. Alternative Explanations and Robustness Checks

This section considers several alternative explanations
aside from reputation spillovers that may contribute to the re-
sults in section VA and presents additional robustness checks.

Differential time trends. Different subindustries within the
dairy sector may have followed different growth trajectories
in the absence of the scandal, leading to biased estimates of
the spillover effects. Fourth table in appendix A allows for
differential time trends across subindustries at the HS two-
digit level. Reassuringly, these results are qualitatively very
similar to the results in table 4.

Reversion to the mean could also bias our results. If con-
taminated firms were growing faster prior to the scandal, our
estimates may be partly driven by these fast-growing firms
mechanically scaling down their production and reducing ex-
ports after the scandal. To alleviate this concern, we allow for
differential time trends with respect to baseline sales in our
baseline specification. In sixth table in appendix A, we fur-
ther exclude firms and destination countries that account for
most of the export growth spike between 2006 E 2008 (Vedere
the discussion in section II). The results are very similar.19

Confounding supply-side factors. Collective reputation
represents a demand-side force, but supply-side forces may
also have contributed to the observed spillover effects. For ex-
ample, the scandal disrupted the activities of some upstream
suppliers: some milk farmers and milk stations exited the
market as a result. Allo stesso modo, stronger government regulations
may have imposed additional costs on firms, raising their pro-
duction costs. All these supply-side forces could have led to
reductions in export revenue and quantity. Tuttavia, a pure
upward shift of the supply curve would have resulted in an un-
ambiguous increase in price (conditional on exporting), con-
trary to what we estimate in table 4, column 3: the coefficient
on CProductXPost, the key collective reputation spillover ef-
fect, is negative and significant at the 10% level.20 This re-
sult alone implies that the demand curve must have moved
downward and offset the supply curve movement. While we
cannot completely rule out supply-side movements, we con-
clude that the demand-side force due to collective reputation
effects played an important role in this context.

Confounding foreign demand shocks. Different firms may
be subject to idiosyncratic demand shocks depending on the
conditions in the destination countries that they export to. If
contaminated firms were more likely to export to countries
that happened to demand more or fewer imported dairy prod-
ucts after 2008, the estimated coefficients would be biased.
To examine this possibility, we construct a measure of firm-
specific demand shocks using a firm’s baseline export value
share for each destination country multiplied by each des-
tination country’s yearly dairy imports from the rest of the
world excluding China and summed over all destinations. IL
fourth table in appendix A shows that our results are robust
to including these firm-specific demand shocks as additional
controls.

Foreign import regulations due to protectionist motives.

Im-
port regulations targeted at all firms from the same origin
industry can result in patterns similar to those estimated in
table 4. The first table in appendix B lists the 20 countries

18Section IIIB explains that the Manufacturing Survey only includes
firms with sales revenue exceeding 5 million RMB. Thus we cannot dis-
tinguish between true exits due to the scandal versus mere reductions in
scala.

19As discussed in section II, a few predominant firms and export destina-

tions drove the growth spike in the pre-scandal period (2006–2007).

20We can estimate the price regression on a balanced sample of exporters
(cioè., firms that exported both before and after the scandal) to account for
any sample composition change; the results are robust.

l

D
o
w
N
o
UN
D
e
D

F
R
o
M
H

T
T

P

:
/
/

D
io
R
e
C
T
.

M

io
T
.

e
D
tu
/
R
e
S
T
/

l

UN
R
T
io
C
e
–
P
D

F
/

/

/

/

1
0
4
6
1
1
2
1
2
0
5
8
1
9
3
/
R
e
S
T
_
UN
_
0
1
0
3
2
P
D

.

F

B
sì
G
tu
e
S
T

T

o
N
0
8
S
e
P
e
M
B
e
R
2
0
2
3

1132

THE REVIEW OF ECONOMICS AND STATISTICS

plus the EU that imposed explicit import bans on Chinese
dairy products after the 2008 scandal by the countries’ value
share in total dairy exports from China.

One way to think about such trade policies is that for-
eign governments react on behalf of domestic consumers in
light of rising safety concerns about products imported from
a particular country. Such blanket regulations represent an
underlying channel for the collective reputation effect. Alter-
natively, these regulations may arise from protectionist mo-
tives. In other words, foreign countries could take advantage
of the scandal to raise import barriers. Empirically distin-
guishing these two stories is challenging. The seventh table
in appendix A presents estimates only for the sample of des-
tinations without explicit import bans. These results are very
similar to our main estimates, suggesting that explicit gov-
ernment regulations cannot fully explain the spillover effects.
Market-based forces due to collective reputation do matter.

VI. Mechanisms

In this section, we investigate three potential mechanisms
that may mediate the strength of the reputation spillover ef-
fects documented in sections IV and V. Since a firm’s repu-
tation consists essentially of buyers’ beliefs, how consumers
gather information and learn matters crucially for reputation
externalities. Thus the mechanisms that we investigate relate
to what may shape consumers’ information sets. Specifically,
we examine the roles of (1) information accuracy in global
media reports, (2) firms’ location and the traceability of con-
taminated suppliers, E (3) firms’ export experience. All of
these forces can act jointly and interact with one another.
Rather than trying to disentangle and quantify the impact of
each, our goal is to examine whether a particular force has
bite.

UN.

Information Accuracy in Global Media Reports

A large literature has shown that the media influence peo-
ple’s perceptions, thereby affecting a wide range of social
and economic outcomes (DellaVigna & Gentzkow, 2010).
In the context of food scares, Adda (2007) and Luong, Shi,
and Wang (2019) show that news information alters con-
sumers’ perceived risk of encountering low-quality products
and thereby affects demand. Therefore, how the media report
an event shapes the event’s impacts. We investigate an im-
portant aspect of media reports on the scandal: informational
accuracy, questo è, the level of detail with which media outlets
described the involvement of different firms in the scandal.
The fifth figure in appendix A shows a typical Chinese media
report on the scandal (left panel), which includes a full list of
contaminated firms and products, and an example from the
Western media, the New York Times (right panel), which only
reported an estimated number of contaminated firms without
mentioning any specific names.

Such heterogeneity in media reports across countries can
generate different information sets among local consumers.

We can imagine two scenarios: one in which consumers per-
fectly understand the evolution of the scandal and are able to
closely keep track of the inspection outcomes and another in
which consumers have trouble identifying the contaminated
firms and worry about Chinese dairy products in general as
a result of the scandal. Collective reputation forces would be
stronger in the latter scenario than in the former.

To construct a systematic measure of consumers’ informa-
tion accuracy across different export destinations, we lever-
age Google Trends data. Google Trends provides public time
series indices based on Google Search data, which capture
how often a search term is entered relative to the total search
volume in a given geographical area. To allow comparisons
of relative popularity across search terms, each data point in
Google Trends indices is divided by the total searches in the
corresponding geographical area and time range and scaled
on a range of 0 A 100 for any given period. We collect data for
31 countries available on Google Trends and construct the rel-
ative search intensity ratio for two keywords—“Sanlu” versus
“2008 Chinese milk scandal”—for each country. Figura 3 dis-
plays the relative search intensity across countries. Web users
in Japan, China, Hong Kong, and New Zealand, for exam-
ple, searched for “Sanlu” much more than the generic phrase,
suggesting that consumers in these locations may have been
more informed about the parties directly involved in the scan-
dal. In comparison, the searched information appears to have
been much less specific in countries such as Myanmar, Pak-
istan, Austria, and Vietnam.21 We classify countries into two
groups based on the relative search intensity: high indicates
a higher ratio (top quartile) and thus higher information ac-
curacy (figure 3).22

Tavolo 6 reports the effect of the scandal on export per-
formance to destinations with high and low search intensity
ratios. Consistent with consumers in countries with low in-
formation accuracy not being able to distinguish innocent and
contaminated firms, the cross-firm spillover effect ( ˆβacross-firm)
is driven by exports to destinations with low information ac-
curacy (−0.639 compared to −0.025, with a p value for test-
ing equality of 0.0016). In eighth table in appendix A, we
further divide countries into quartiles of the information ac-
curacy measure. We find suggestive evidence of bigger cross-
firm spillover effects as information worsens.

Overall, these results show that information accuracy plays
an important role in mediating the force of collective repu-
tazione. Information accuracy may be particularly relevant in
the context of international trade, as media coverage of events

21The sixth figure in appendix A shows the search behavior across
provinces in China. Not surprisingly, Hebei province, where the headquar-
ters of Sanlu was located, has the highest search intensity for the keyword
“Sanlu.”

22This measure of information accuracy may be correlated with countries’
baseline market share of Chinese dairy exports, potentially confounding
the estimation results. Regressing the Google search index on countries’
baseline market share yields a low R-square of 0.007 (0.005), suggesting
that most variation in information accuracy cannot be explained by market
condividere. One factor affecting information accuracy could be how the scandal
was covered in the local media, which may depend on political attitudes.

l

D
o
w
N
o
UN
D
e
D

F
R
o
M
H

T
T

P

:
/
/

D
io
R
e
C
T
.

M

io
T
.

e
D
tu
/
R
e
S
T
/

l

UN
R
T
io
C
e
–
P
D

F
/

/

/

/

1
0
4
6
1
1
2
1
2
0
5
8
1
9
3
/
R
e
S
T
_
UN
_
0
1
0
3
2
P
D

.

F

B
sì
G
tu
e
S
T

T

o
N
0
8
S
e
P
e
M
B
e
R
2
0
2
3

COLLECTIVE REPUTATION IN TRADE

1133

FIGURE 3.—RATIO OF GOOGLE SEARCH INDEX: “SANLU” VERSUS “2008 CHINESE MILK SCANDAL”

l

D
o
w
N
o
UN
D
e
D

F
R
o
M
H

T
T

P

:
/
/

D
io
R
e
C
T
.

M

io
T
.

e
D
tu
/
R
e
S
T
/

l

UN
R
T
io
C
e
–
P
D

F
/

/

/

/

1
0
4
6
1
1
2
1
2
0
5
8
1
9
3
/
R
e
S
T
_
UN
_
0
1
0
3
2
P
D

.

F

B
sì
G
tu
e
S
T

T

o
N
0
8
S
e
P
e
M
B
e
R
2
0
2
3

This figure plots the ratio of the Google search index for the keyword “Sanlu” versus “2008 Chinese milk scandal” by country between 09/01/2008 E 10/31/2008. The black bars mark countries with high information
accuracy (the top quartile), while the grey ones mark countries with low information accuracy.

that happen in foreign countries, where information needs to
travel far, may be less precise.

B. Firms’ Sourcing Location

Prossimo, we investigate how firms’ location mediates the
strength of reputation spillovers. In the case of the Chinese
dairy scandal, contamination stemmed from wrongdoings in
the upstream sector, as discussed in section II. Given that
dairy firms from the same location tend to source inputs
from the same local upstream farms, one may expect to see
stronger spillover effects on innocent and uninspected firms
located in the same areas as contaminated firms.23 To exam-
ine this aspect, we take advantage of the Chinese Customs
dati, in which firms are required to report the sourcing lo-
cation for each of their export transactions. We leverage this

23Generalmente, the source of contamination and the public’s knowledge
of this information matter for the degree of reputation spillovers. If the
root cause of a quality defect is limited to a well-known individual firm,
consumers may not be concerned about other firms in the same origin-
industry. One example of a defect that involved a single firm only is the
case of the Samsung Galaxy battery fire, the aftermath of which did not
affect any other South Korean phone brands. By contrast, if the quality
defect stems from upstream production processes, as was the case in the
Chinese dairy scandal, all downstream firms may suffer from reputation
spillover effects, especially if inputs are hard to trace. A scandal about the
product quality of one firm may cause consumers to worry about the quality
of other firms that source from the same upstream sources.

information to identify whether a city hosts any contaminated
firms for a given product; if so, we call it a “contaminated
city.” We define an indicator variable CSourceCity for firm
i that equals 1 if i sourced (any product) from a contam-
inated city prior to the scandal. We define another indica-
tor variable CSourceCity-Product that equals to 1 if i ever
sourced contaminated product j from the same city as a con-
taminated firm. We interact these indicator variables with the
post-Scandal indicator to study how baseline sourcing pat-
terns affected firms’ post-scandal export performance.

Tavolo 7 reports the impact of sourcing from a contam-
inated location for innocent and uninspected firms. While
we estimate large overall cross-firm spillovers (the negative
coefficient on CProductXPost), the exact sourcing location
of a firm does not appear to matter: the point estimates on the
two additional interaction terms are very close to zero, Anche se
the standard errors are fairly large. The result suggests that
the spillover effects were generalized rather than contained
around contaminated sources. One potential explanation is
that international buyers may have very coarse information
about contamination sources due to the low traceability of
the contaminated inputs; as a result, the spillover effects are
not localized. This echoes our previous findings that informa-
tion accuracy matters for the strength of reputation spillover
effects. Inoltre, to the extent that firms in one location com-
pete for the same labor and upstream supplier inputs, innocent

1134

THE REVIEW OF ECONOMICS AND STATISTICS

TABLE 6.—HETEROGENEOUS IMPACT BASED ON THE GOOGLE SEARCH INDEX

IHS (value)

IHS (quantity)

Exporting (dummy)

High
(1)
−1.101
(1.391)
−0.571
(0.639)
−0.025
(0.153)
0.391
(0.496)
−0.259
(0.368)
0.418
13775
YES
YES
YES

Basso
(2)
−0.068
(0.963)
−1.678***
(0.595)
−0.639***
(0.195)
0.713
(0.756)
−0.452
(0.382)
0.294
13775
YES
YES
YES

Rest
(3)
−0.208
(0.452)
−0.739***
(0.230)
−0.367***
(0.141)
0.676
(0.543)
−0.810
(0.637)
0.255
13775
YES
YES
YES

High
(4)
−0.958
(1.349)
−0.650
(0.657)
−0.057
(0.138)
0.363
(0.494)
−0.231
(0.362)
0.440
13775
YES
YES
YES

Basso
(5)
−0.026
(0.929)
−1.585***
(0.586)
−0.591***
(0.183)
0.649
(0.723)
−0.410
(0.346)
0.296
13775
YES
YES
YES

Rest
(6)
−0.180
(0.420)
−0.690***
(0.219)
−0.341***
(0.130)
0.621
(0.513)
−0.725
(0.592)
0.254
13775
YES
YES
YES

High
(7)
−0.061
(0.094)
−0.055
(0.047)
−0.001
(0.013)
0.023
(0.038)
−0.026
(0.037)
0.341
13775
YES
YES
YES

Basso
(8)

0.005
(0.062)
−0.132***
(0.049)
−0.053***
(0.016)
0.047
(0.059)
−0.035
(0.031)
0.242
13775
YES
YES
YES

Rest
(9)
−0.014
(0.033)
−0.062***
(0.018)
−0.031***
(0.011)
0.062
(0.042)
−0.069
(0.049)
0.224
13775
YES
YES
YES

CFirm-ProductXPost

CFirmXPost

CProductXPost

IFirm-ProductXPost

IFirmXPost

R-squared
Observations
Firm-product FE
Year FE
BaselineSizeXPost

This table shows the regression results for the heterogeneous effects of the scandal across export destinations with different information accuracy. The unit of observation is at the firm-product-year level. The sample
contains all dairy exporters in the Chinese Customs data (2000–2013). We create a balanced panel at the firm-product (HS eight-digit) and year level. Columns 1–6 present results for the inverse hyperbolic sine (IHS)
transformation of the outcome variables of interest, export value and export quantity. Columns 7–9 use an indicator for positive exports as the outcome variable. We categorize export destinations by high and low
information accuracy, using the Google search intensity ratio. High-information-accuracy destinations display a high ratio of searches for the word “Sanlu” relative to searches for “2008 Chinese milk scandal”. We also
include results for countries without a Google search index (“Rest”). The interaction terms are the products of the post-scandal dummy (2009–2013) with the following five group indicators: (C)ontaminatedFirm-Product,
(C)ontaminatedFirm, (C)ontaminatedProduct, (IO)nnocentFirm-Product, E (IO)nnocentFirm. The omitted category is innocent and uninspected products from uninspected firms. All regressions include firm-product
and year fixed effects. Baseline size measures a firm’s baseline (2000–2007) total export revenue. Standard errors are two-way clustered at the firm and product-year level. *, **, E *** denote significance at the 10%,
5%, E 1% level, rispettivamente.

CProductXPost

CSourceCity-ProductXPost

CSourceCityXPost

R-squared
Observations
Firm-product FE
Year FE
BaselineSizeXPost

TABLE 7.—HETEROGENEOUS IMPACT BY FIRMS’ SOURCING LOCATION

IHS (value)
(1)

−0.587**
(0.274)
0.030
(0.554)
−0.090
(0.274)
0.313
8768
YES
YES
YES

IHS (quantity)
(2)

Log (price)
(3)

Exporting (dummy)
(4)

−0.551**
(0.259)
0.050
(0.520)
−0.113
(0.257)
0.319
8768
YES
YES
YES

−0.164
(0.110)
0.053
(0.096)
0.027
(0.083)
0.898
1098
YES
YES
YES

−0.045**
(0.021)
0.023
(0.045)
−0.002
(0.022)
0.248
8768
YES
YES
YES

This table shows the regression results for the heterogeneous effects of the scandal on exports across firms with different baseline sourcing locations. The unit of observation is at the firm-product-year level. IL
sample contains innocent and uninspected dairy exporters in the Chinese Customs data (2000–2013). We create a balanced panel at the firm-product (HS eight-digit) and year level for the outcomes in columns 1, 2,
E 4. Columns 1 E 2 present results for the inverse hyperbolic sine (IHS) transformation of the outcome variables of interest, export value and export quantity. Column 3 presents results for the natural logarithm
of unit price, while column 4 uses an indicator for positive exports as the outcome variable. The interaction terms are the products of the post-scandal dummy (2009–2013) with the following three group indicators:
(C)ontaminatedProduct, (C)ontaminatedSourceCity-Product and (C)ontaminatedSourceCity. All regressions include firm-product and year fixed effects. Baseline size measures a firm’s baseline (2000–2007) total
export revenue. Standard errors are two-way clustered at the product-year and firm level. *, **, E *** denote significance at the 10%, 5%, E 1% level, rispettivamente.

and uninspected firms may have benefited when their rivals
were hit by the scandal. The estimated coefficients thus reflect
the net of the reputation and competition effects.

C. Firms’ Export Experience

A firm’s reputation can have both an individual component
and a collective component: Per esempio, consumers may
observe a quality signal from each firm as well as a noisy
signal of the industry’s average quality. In questo caso, a strong
individual reputation may mitigate the impact of a collective
reputation shock. To examine this possibility, we proxy a
firm’s individual reputation in the global market by its export
experience, measured by the number of years the firm had
been exporting prior to the scandal and the firm’s baseline
share of exports in total sales.

Tavolo 8 shows the heterogeneous impact of the scandal
based on export experience. Due to our short baseline period,
we define new (young) firms as those that had just started
exporting in 2008 and established firms as those that had ex-
ported for one or more years prior to 2008. Consistent with
individual reputation shielding a firm from collective reputa-
tion shocks, the cross-firm spillover effect is larger for new
exporters. Here, pure, we allow for differential time trends with
respect to firm size, which is likely to differ between new
and established exporters. A test of equality of ˆβacross-firm in
columns 1 E 2 (or columns 3 E 4) has a p value of 0.0191
(O 0.0734). The test of equality of the spillover effects on the
extensive margin, ˆβacross-firm in columns 5 E 6 (or columns
7 E 8), has a p value of 0.0335 (O 0.1313). These findings
suggest that in light of a collective reputation shock, a more
established individual reputation can (partially) shield firms

l

D
o
w
N
o
UN
D
e
D

F
R
o
M
H

T
T

P

:
/
/

D
io
R
e
C
T
.

M

io
T
.

e
D
tu
/
R
e
S
T
/

l

UN
R
T
io
C
e
–
P
D

F
/

/

/

/

1
0
4
6
1
1
2
1
2
0
5
8
1
9
3
/
R
e
S
T
_
UN
_
0
1
0
3
2
P
D

.

F

B
sì
G
tu
e
S
T

T

o
N
0
8
S
e
P
e
M
B
e
R
2
0
2
3

COLLECTIVE REPUTATION IN TRADE

1135

TABLE 8.—HETEROGENEOUS IMPACT BY FIRMS’ EXPORT EXPERIENCE

IHS (value)

Exporting (dummy)

CFirm-ProductXPost

CFirmXPost

CProductXPost

IFirm-ProductXPost

IFirmXPost

R-squared
Observations
Firm-product FE
BaselineSizeXPost
HS2digitXYear
HS2digitXPost

Established
(1)

Nuovo
(2)

Established
(3)

Nuovo
(4)

Established
(5)

Nuovo
(6)

Established
(7)

Nuovo
(8)

−0.678
(1.419)
−1.119**
(0.502)
−0.390
(0.358)
−0.456
(0.831)
−0.597
(0.798)
0.339
9830
YES
YES
YES
NO

−1.251***
(0.433)
−2.098***
(0.712)
−1.520***
(0.438)
2.979*
(1.764)
−1.714
(1.209)
0.228
3945
YES
YES
YES
NO

−0.760
(1.423)
−1.069**
(0.508)
−0.214
(0.411)
−0.463
(0.821)
−0.583
(0.794)
0.340
9830
YES
YES
NO
YES

−0.943***
(0.332)
−2.112**
(0.832)
−1.160***
(0.382)
2.787
(1.843)
−1.375
(1.209)
0.230
3945
YES
YES
NO
YES

−0.032
(0.092)
−0.084**
(0.035)
−0.034
(0.028)
−0.026
(0.061)
−0.047
(0.067)
0.252
9830
YES
YES
YES
NO

−0.091***
(0.034)
−0.190***
(0.063)
−0.115***
(0.034)
0.199*
(0.120)
−0.147
(0.093)
0.206
3945
YES
YES
YES
NO

−0.036
(0.092)
−0.082**
(0.035)
−0.026
(0.033)
−0.026
(0.060)
−0.048
(0.067)
0.253
9830
YES
YES
NO
YES

−0.068*
(0.039)
−0.192***
(0.073)
−0.089***
(0.030)
0.185
(0.126)
−0.122
(0.093)
0.208
3945
YES
YES
NO
YES

This table shows the regression results for the heterogeneous effects of the scandal on exports across firms with different lengths of export experience. The unit of observation is at the firm-product-year level. IL
sample contains all dairy exporters in the Chinese Customs data (2000–2013). We create a balanced panel at the firm-product (HS eight-digit) and year level. Columns 1–4 present results for the inverse hyperbolic
sine (IHS) transformation of the outcome variable of interest, export value. Columns 5–8 use an indicator for positive exports as the outcome variable. Columns 1, 3, 5, E 7 use the subsample of established firms,
which are firms that had exported for more than 1 year before 2008. Columns 2, 4, 6, E 8 use the subsample of new firms, which are firms that had not exported before 2008. The interaction terms are the products of
the post-scandal dummy (2009–2013) with the following five group indicators: (C)ontaminatedFirm-Product, (C)ontaminatedFirm, (C)ontaminatedProduct, (IO)nnocentFirm-Product, E (IO)nnocentFirm. The omitted
category is innocent and uninspected products from uninspected firms. All regressions include firm-product and year fixed effects. Baseline size measures a firm’s baseline (2000–2007) total export revenue. Standard
errors are two-way clustered at the firm and product-year level. *, **, E *** denote significance at the 10%, 5%, E 1% level, rispettivamente.

from the collective damage, whereas newcomers are more
likely to suffer from the “original sin” of their predecessors.
Ninth table in appendix A explores an alternative measure
of export experience based on firms’ fraction of exports in
total sales in 2007. For this exercise, we merge the Chinese
Customs data with the Manufacturing Survey data and use
the total sales information in the latter. Of the 335 dairy firms
in the Chinese Customs sample in 2007, 151 are identified
in the Manufacturing Survey. The relatively low match rate
could be due to the imperfect matching of firms’ names and
addresses across the Manufacturing Survey and the Chinese
Customs data or the fact that the Manufacturing Survey only
includes above-scale firms (see section IIIB). Seventh figure
in appendix A plots the distribution of the fraction of exports
in total sales in the baseline year. On average, conditional on
exporting, exports account for 10.3% of the firms’ total sales
(the median is 9.8%) with a standard deviation of 0.35. Noi
classify firms as large exporters if their fraction of exports
exceeds the median and small otherwise. The results again
suggest that small firms tend to suffer more from collective
reputation damage.

Summary. This section explores three mechanisms poten-
tially driving collective reputation spillover effects. Primo, In-
formation accuracy plays an important role in mediating the
strength of collective reputation—the spillover effects are
smaller in destinations where people appear to have had bet-
ter information about the parties involved in the scandal. Sez-
ond, the spillover effects appear to be generalized rather than
localized to contaminated sources, consistent with interna-
tional buyers having coarse information about contamination
sources due to low traceability of the contaminated inputs.
Finalmente, individual reputation can mitigate collective reputa-

tion damage, and new exporters and firms with larger baseline
export shares appear to be the most vulnerable to collective
reputation shocks.

VII. Conclusione

Understanding how reputation spreads within an industry
or a geographic area is key for informing trade and develop-
ment policy, as the existence of collective reputation implies
important externalities. We study this question in the con-
text of Chinese dairy firms’ exports following the 2008 scan-
dal. We document strong reputation spillover effects on firms
whose products were not contaminated. Surprisingly, firms
that cleared formal inspections do not appear to have fared
any better than uninspected firms. These findings highlight
the role of collective reputation in international trade and the
challenges that governments may face in signaling quality
and restoring trust. Analyses of potential mechanisms high-
light the role of information accuracy, supply chain traceabil-
ity and firms’ individual reputations in mediating the collec-
tive reputation effect.

Our study has two broad policy implications concerning
(1) the role of government and third-party certifications and
(2) the role of market structure. That said, the external validity
of the results is an empirical question, as the exact magnitudes
of spillovers vary across industries and countries and depend
on whether directly affected firms are large industry leaders or
small players. Our approach can be applied to other contexts.
Primo, collective reputation may call for government in-
terventions, but government-led inspection efforts may gen-
erate counterproductive signals, depending on the reputa-
tion of the inspection body itself. Private third parties and
international certification bodies may act as an effective

l

D
o
w
N
o
UN
D
e
D

F
R
o
M
H

T
T

P

:
/
/

D
io
R
e
C
T
.

M

io
T
.

e
D
tu
/
R
e
S
T
/

l

UN
R
T
io
C
e
–
P
D

F
/

/

/

/

1
0
4
6
1
1
2
1
2
0
5
8
1
9
3
/
R
e
S
T
_
UN
_
0
1
0
3
2
P
D

.

F

B
sì
G
tu
e
S
T

T

o
N
0
8
S
e
P
e
M
B
e
R
2
0
2
3

1136

THE REVIEW OF ECONOMICS AND STATISTICS

complement to or substitute for government regulations, es-
pecially in developing-country settings. Tuttavia, based on
our interviews with firms in the Chinese dairy industry, third-
party certification has not been adopted in this sector. Questo
could be due either to high costs or logistical hurdles in ob-
taining these certifications or to perceived low returns to cer-
tification. Understanding the barriers to adoption as well as
the effectiveness of these programs is crucial for designing
policies that could assure a high quality standard and break
the low-quality-low-reputation equilibrium.

Secondo, this study takes a first step in investigating various
mechanisms that may affect the transmission of reputation
spillovers. Understanding these mechanisms can help inform
policies in response to a collective reputation crisis. Follow-
ing the milk scandal, many firms integrated vertically with
upstream farms. One rationale for vertical integration is the
ability to enforce stronger quality controls (Hansman et al.,
2017); an equally important rationale is to signal quality, so
that a firm is better shielded from wrongdoing by other firms’
suppliers. It is expected that by 2020, Sopra 70% of Chinese
raw milk will be produced from vertically integrated milk
farms.24 Future work is needed to better understand how col-
lective reputation affects firms’ quality investment incentives.
Acting as an important externality, collective reputation could
also have rich interactions with other market forces, ad esempio
market entry and competition.

24News source accessed on 09/28/2020: http://www.gov.cn/zhengce/

content/2018-06/11/content_5297839.htm.

REFERENCES
Adda, J., “Behavior towards Health Risks: An Empirical Study Using the
“Mad Cow” Crisis as an Experiment,” Journal of Risk and Uncer-
tainty 35:3 (2007), 285–305. 10.1007/s11166-007-9026-5
Ahluwalia, R., R. E. Burnkrant, and H. R. Unnava, “Consumer Response to
Negative Publicity: The Moderating Role of Commitment,” Journal
of Marketing Research 37:2 (2000), 203–214. 10.1509/jmkr.37.2
.203.18734

Allen, T., “Information Frictions in Trade,” Econometrica 82:6 (2014),

2041–2083. 10.3982/ECTA10984

Atkin, D., e A. Khandelwal, “How Distortions Alter the Impacts of In-
ternational Trade in Developing Countries?” Annual Review of Eco-
nomics 12 (2020), 213–238. 10.1146/annurev-economics-081919
-041554

Atkin, D., UN. K. Khandelwal, e A. Osman, “Exporting and Firm Perfor-
mance: Evidence from a Randomized Trial,” The Quarterly Journal
of Economics (2017), 551–615.

Bachmann, R., G. Ehrlich, Y. Fan, and D. Ruzic, “Firms and Collective Rep-
utation: A Study of the Volkswagen Emissions Scandal,” National
Bureau of Economic Research working paper 26117 (2019).
Bai, J., “Melons as Lemons: Asymmetric Information, Consumer Learning

and Quality Provision,” working paper (2018).

Banerjee, UN. V., “Microcredit under the Microscope: What Have We
Learned in the Past Two Decades, and What Do We Need to Know?"
Annual Review of Economics 5:1 (2013), 487–519. 10.1146/annurev
-economics-082912-110220

Banerjee, UN. V., and E. Duflo, “Reputation Effects and The Limits of
Contracting: A Study of The Indian Software Industry,” The Quar-
terly Journal of Economics 115:3 (2000), 989–1017. 10.1162/
003355300554962

Bardhan, P., D. Mookherjee, and M. Tsumagari, “Middlemen Margins and
Globalization,” American Economic Journal: Microeconomics 5:4
(2013), 81–119. 10.1257/mic.5.4.81

Björkman-Nyqvist, M., J. Svensson, and D. Yanagizawa-Drott, “The Mar-
ket for (Fake) Antimalarial Medicine: Evidence from Uganda,” Cen-
ter for International Development at Harvard University working
paper 242 (2013).

Bloom, N., B. Eifert, UN. Mahajan, D. McKenzie, and J. Roberts, “Does
Management Matter? Evidence from India,” The Quarterly Journal
of Economics 128:1 (2013), 1–51. 10.1093/qje/qjs044

Brandt, L., J. Van Biesebroeck, and Y. Zhang, “Creative Accounting or
Creative Destruction? Firm-Level Productivity Growth in Chinese
Manufacturing,” Journal of Development Economics 97:2 (2012),
339–351. 10.1016/j.jdeveco.2011.02.002

Burbidge, J. B., l. Magee, e A. l. Robb, “Alternative Transformations to
Handle Extreme Values of the Dependent Variable,” Journal of the
American Statistical Association 83:401 (1988), 123–127. 10.1080/
01621459.1988.10478575

Cabral, L., e A. Hortacsu, “The Dynamics of Seller Reputation: Evidence
from eBay,” The Journal of Industrial Economics 58:1 (2010), 54–
78. 10.1111/j.1467-6451.2010.00405.x

Castriota, S., and M. Delmastro, “The Economics of Collective Reputation:
Evidence from the Wine Industry,” American Journal of Agricultural
Economics 97:2 (2014), 469–489. 10.1093/ajae/aau107

Davidson, W. N., and D. l. Worrell, “Research Notes and Communi-
cations: The Effect of Product Recall Announcements on Share-
holder Wealth,” Strategic Management Journal 13:6 (1992), 467–
473. 10.1002/smj.4250130606

De Loecker, J., and P. K. Goldberg, “Firm Performance in a Global Mar-
ket,” Annual Review of Economics 6:1 (2014), 201–227. 10.1146/
annurev-economics-080113-104741

De Mel, S., D. McKenzie, and C. Woodruff, “Returns to Capital in Mi-
croenterprises: Evidence from a Field Experiment,” The Quar-
terly Journal of Economics 123:4 (2008), 1329–1372. 10.1162/
qjec.2008.123.4.1329

DellaVigna, S., and M. Gentzkow, “Persuasion: Empirical Evidence,” An-
nual Review of Economics 2:1 (2010), 643–669. 10.1146/annurev
.economics.102308.124309

Fishman, A., UN. Simhon, IO. Finkelshtain, and N. Yacouel, “The Eco-
nomics of Collective Brands,” Bar-Ilan University Department of
Economics Research Paper 2010–2011 (2010).

Fleckinger, P., “Collective Reputation and Market Structure: Regulating the
Quality vs. Quantity Trade-Off,” Centre National De La Recherche
Scientifique working paper 00243080 (2007).

Freedman, S., M. Kearney, and M. Lederman, “Product Recalls, Imperfect
Information, and Spillover Effects: Lessons from the Consumer Re-
sponse to the 2007 Toy Recalls,” this REVIEW, 94:2 (2012), 499–
516.

Gergaud, O., F. Livat, B. Rickard, and F. Warzynski, “Evaluating the Net
Benefits of Collective Reputation: The Case of Bordeaux Wine,"
Food Policy 71 (2017), 8–16. 10.1016/j.foodpol.2017.07.002
Gobillon, L., and T. Magnac, “Regional Policy Evaluation: Interactive Fixed
Effects and Synthetic Controls,” this REVIEW, 98:3 (2016), 535–
551.

Hansman, C., J. Hjort, G. León, and M. Teachout, “Vertical Integration,
Supplier Behavior, and Quality Upgrading among Exporters,” Na-
tional Bureau of Economic Research working paper 23949 (2017).
Hau, UN. K.-C., T. H. Kwan, and P. K.-T. Li, “Melamine Toxicity and the
Kidney,” Journal of the American Society of Nephrology 20:2 (2009),
245–250. 10.1681/ASN.2008101065, PubMed: 19193777
Jin, G. Z., and P. Leslie, “Reputational Incentives for Restaurant Hygiene,"
American Economic Journal: Microeconomics 1:1 (2009), 237–267.
10.1257/mic.1.1.237

Jovanovic, B., “Product Recalls and Firm Reputation,” American Eco-
nomic Journal: Microeconomics 13:3 (2021), 404–442. 10.1257/
mic.20180396

Kee, H. L., and H. Tang, “Domestic Value Added in Exports: Theory
and Firm Evidence from China,” American Economic Review 106:6
(2016), 1402–36. 10.1257/aer.20131687

Kugler, M., and E. Verhoogen, “Prices, Plant Size, and Product Quality,"
The Review of Economic Studies 79:1 (2012), 307–339. 10.1093/
restud/rdr021

Levin, J., “The Dynamics of Collective Reputation,” The BE Journal of

Theoretical Economics 9:1 (2009), 1–25.

Luong, T. A., C. M. Shi, and Z. Wang, “The Impact of Media on Trade: Evi-
dence from the 2008 China Milk Contamination Scandal,” Available
at SSRN 3164244 (2019).

l

D
o
w
N
o
UN
D
e
D

F
R
o
M
H

T
T

P

:
/
/

D
io
R
e
C
T
.

M

io
T
.

e
D
tu
/
R
e
S
T
/

l

UN
R
T
io
C
e
–
P
D

F
/

/

/

/

1
0
4
6
1
1
2
1
2
0
5
8
1
9
3
/
R
e
S
T
_
UN
_
0
1
0
3
2
P
D

.

F

B
sì
G
tu
e
S
T

T

o
N
0
8
S
e
P
e
M
B
e
R
2
0
2
3

COLLECTIVE REPUTATION IN TRADE

1137

Mamma, J., Z. Wang, and T. Khanna, “Milk Safety in China: A Field Experi-

Startz, M., “The Value of Face-to-Face: Search and Contracting Problems

ment,” working paper (2016).

Macchiavello, R., “Development Uncorked: Reputation Acquisition in the
New Market for Chilean Wines in the UK,” CEPR discussion paper
DP7698 (2010).

Macchiavello, R., e A. Morjaria, “The Value of Relationships: Evidence
from a Supply Shock to Kenyan Rose Exports,” American Economic
Review 105:9 (2015), 2911–2945. 10.1257/aer.20120141
Manova, K., and Z. Zhang, “Export Prices across Firms and Destina-
zioni,” The Quarterly Journal of Economics 127:1 (2012), 379–436.
10.1093/qje/qjr051

Marchini, A., C. Riganelli, F. Diotallevi, and C. Paffarini, “Factors of
Collective Reputation of the Italian PDO Wines: An Analysis on
Central Italy,” Wine Economics and Policy 3:2 (2014), 127–137.
10.1016/j.wep.2014.12.003

Park, A., D. Yang, X. Shi, and Y. Jiang, “Exporting and Firm Performance:
Chinese Exporters and the Asian Financial Crisis,” this REVIEW, 92:4
(2010), 822–842.

in Nigerian Trade,” working paper (2017).

Tirole, J., “A Theory of Collective Reputations (with Applications to the
Persistence of Corruption and to Firm Quality),” The Review of Eco-
nomic Studies 63:1 (1996), 1–22. 10.2307/2298112

Van Heerde, H., K. Helsen, and M. G. Dekimpe, “The Impact of a Product-
Harm Crisis on Marketing Effectiveness,” Marketing Science 26:2
(2007), 230–245. 10.1287/mksc.1060.0227

Verhoogen, E. A., “Trade, Quality Upgrading, and Wage Inequality in
the Mexican Manufacturing Sector,” The Quarterly Journal of Eco-
nomics 123:2 (2008), 489–530. 10.1162/qjec.2008.123.2.489
Winfree, J. A., and J. J. McCluskey, “Collective Reputation and Quality,"
American Journal of Agricultural Economics, 87:1 (2005), 206–213.
10.1111/j.0002-9092.2005.00712.X

Zhao, Y., “Your (Country’s) Reputation Precedes You: Information Asym-
metry, Externalities and the Quality of Exports,” working paper
(2018).

l

D
o
w
N
o
UN
D
e
D

F
R
o
M
H

T
T

P

:
/
/

D
io
R
e
C
T
.

M

io
T
.

e
D
tu
/
R
e
S
T
/

l

UN
R
T
io
C
e
–
P
D

F
/

/

/

/

1
0
4
6
1
1
2
1
2
0
5
8
1
9
3
/
R
e
S
T
_
UN
_
0
1
0
3
2
P
D

.

F

B
sì
G
tu
e
S
T

T

o
N
0
8
S
e
P
e
M
B
e
R
2
0
2
3

Scarica il pdf