Dai et al.2020. socially responsible corpoate customers

JID: FINEC
ARTICLE IN PRESS [m3Gdc;February 14, 2020;3:56]

Journal of Financial Economics xxx (xxxx) xxx
Contents lists available at ScienceDirect
Journal of Financial Economics

journal homepage: www.elsevier.com/locate/jfec
Socially responsible corporate customersR

Rui Dai a, Hao Liang b,∗, Lilian Ng c
a
Wharton Research Data Services (WRDS), Wharton School of Business, University of Pennsylvania, Suite 300, 3819 Chestnut St,
Philadelphia, PA 19104, USA
b
Lee Kong Chian School of Business, Singapore Management University, 50 Stamford Road, Singapore 178899, Singapore
c
Schulich School of Business, N223 SSB, York University, 4700 Keele Street, North York, Ontario M3J 1P3, Canada
a r t i c l e i n f o a b s t r a c t
Article history: Corporate customers are an important stakeholder in global supply chains. We employ
Received 12 August 2018 several unique international databases to test whether socially responsible corporate cus-
Revised 30 January 2019
tomers can infuse similar socially responsible business behavior in suppliers. Our findings
Accepted 1 July 2019
suggest a unilateral effect on corporate social responsibility (CSR) only from customers
Available online xxx
to suppliers, an evidence further supported by exogenous variation in customers’ close-
JEL classification: call CSR proposals and by product scandals. Customers exert influence on suppliers’ CSR
G23 through positive assortative matching and their decision-making process. Enhanced col-
G30 laborative CSR efforts help improve operational efficiency and firm valuation of both cus-
G34 tomers and suppliers but increase only the customers’ future sales growth.
M14
© 2020 Elsevier B.V. All rights reserved.
Keywords:
Corporate social responsibility
Corporate customers
Global supply chains
Economic benefits
More than 1,000 of the world’s largest companies... have

emissions-reduction targets for their own operations. Now,
they want the thousands of companies that supply them
R
We thank Ekkehart Boehmer, Richard Carney, Simba Chang, Zhen- with goods and services to reduce their own emissions.1
hui Chen, Tao Chen, Travers Child, Lauren Cohen, Hyunsoo Doh, Ita-
mar Drechsler, Ying Duan, Caroline Flammer, Aurobindo Ghosh, Sadok El
Ghoul, Xavier Giroud, Olga Hawn, Jie (Jack) He, Johan Hombert, Dashan 1. Introduction
Huang, Chuan Yang Hwang, Oguzhan Karakas, Andy Kim, Lloyd Kurtz,
Mauricio Larrain, Long (Jason) Li, Jun Li, Weikai Li, Karl Lins, Roger
Loh, Christopher Marquis, Ron Masulis, Roni Michaely, Clemens Otto, Today’s changing global businesses and corporate en-
Hoonsuk Park, Hyeong-sop Shim, Victor Song, Choi Hyun Soo, Laura vironments have brought about a new wave of corporate
Starks, Neal Stoughton, Alexander Verdarushko, Cong Wang, Heli Wang, social responsibility (CSR) activities that go beyond the
Chishen Wei, Qiping Xu, Frank Yu, Xiaoyun Yu, Bohui Zhang, Chendi regulatory requirement of the country (Bénabou and Tirole,
Zhang, and seminar participants at the China Europe International Busi-
ness School, China International Conference in Finance (CICF),City Univer-
2010; Kitzmueller and Shimshack, 2012). As corporations
sity of Hong Kong, Darla Moore School of Business-Hitotsubashi Univer-
sity Third International Conference on Corporate Finance, Deakin Univer-
sity, Massey University (Albany campus and Palmerston North campus), SSHRC research grant no. 435-2017-1231 from the Social Sciences and Hu-
Monash University, Nanyang Technological University, the Schulich School manities Research Council of Canada.
∗
of Business, Singapore Management University, Simon Fraser University, Corresponding author.
Sungykunkwan University, Telfer Annual Conference on Accounting and E-mail addresses: rdai@wharton.upenn.edu (R. Dai), hliang@smu.
Finance at the University of Ottawa, and WRDS of The Wharton School, edu.sg (H. Liang), lng@schulich.york.ca (L. Ng).
1
for their many helpful comments and suggestions. Ng acknowledges the https://www.greenbiz.com/article/how-get-suppliers-act-climate.
https://doi.org/10.1016/j.jfineco.2020.01.003
0304-405X/© 2020 Elsevier B.V. All rights reserved.
Please cite this article as: R. Dai, H. Liang and L. Ng, Socially responsible corporate customers, Journal of Financial Eco-
nomics, https://doi.org/10.1016/j.jfineco.2020.01.003
JID: FINEC
2 R. Dai, H. Liang and L. Ng / Journal of Financial Economics xxx (xxxx) xxx
face increasing societal demands for more CSR activities,2 and thereby their ability to access capital,5 among others.
there appear growing corporate efforts to integrate social, Also, research by practitioners has shown that environ-
environmental, and ethical concerns into their business mental and social scandals associated with suppliers
operations. Anecdotal evidence, as well as the opening increase not only the suppliers’ own reputation risk in-
quote, has suggested that many corporate customers are dex but also their customers’ and that such increases in
concerned not only with their own CSR standards but also reputation risks are accompanied by falling stock prices
with those of their suppliers. Some scholars argue that subsequent to the release of corporate scandals.6 There-
the growing popularity of CSR activities around the world fore, we hypothesize that customers are compelled to
is, in part, in response to the repeated failures of laws exert influence on suppliers for better CSR practices and
and regulations protecting stakeholders, raising the need that their actions have real economic consequences.
from stakeholders to protect their own interests through We exploit several unique international databases to
pushing the company to engage in CSR (Bénabou and test whether socially responsible corporate customers
Tirole, 2010; De Bettignies and Robinson, 2018). However, can infuse similar socially conscious business behavior in
it is not apparent whether corporate customers, one of the suppliers. The two primary international databases are: (1)
most important stakeholders, are really taking actions to a newly available FactSet Revere database that provides
push suppliers to engage in socially responsible business information on firm-level networks of customers and sup-
practices, or whether their public mention of CSR com- pliers around the world, and (2) Thomson Reuter’s ASSET4
mitment is simply a sideshow (see, e.g., Koehn and Ueng, Environmental (E), Social (S), and Corporate Governance
2010; Kitzmueller and Shimshack, 2012). Nevertheless, (G) database (ASSET4) that contains ASSET4 ratings (i.e.,
there is limited academic research on the role of corpo- a composite firm-level CSR rating) as well as more than
rate customers in influencing suppliers to conduct their 750 constituent ESG ratings of global publicly listed firms.
business operations in a socially and environmentally re- After merging these two databases, our sample consists
sponsible manner. Thus, the goal of our study is to explore of 34,117 unique corporate customer-supplier pairs from
whether and by which mechanisms corporate customers 50 countries worldwide for the period from 2003 to
drive CSR practices in global supply chains around the 2015. Using this large international sample of corporate
world, and consequently, their economic implications for customer-supplier relationships, we find evidence of a
both customers and suppliers. significant unilateral effect of customers’ socially responsi-
Increases in economic globalization, advancements ble behavior on their suppliers, suggesting that corporate
in production and information technologies, and im- customers make real efforts to ensure suppliers engage
provement in logistics have facilitated a dynamic growth in similar CSR standards. Suppliers, however, exhibit
in global supply chains. As corporations exploit these no influence on customers’ CSR activities. In terms of
expanding opportunities, they face new challenges to economic significance, a one-standard-deviation change
properly enforce their own CSR policies across their global in the customer CSR rating will generate about an 8%
and complex supplier networks. Investing in CSR initiatives aggregate increase in future CSR performance of suppliers
can be especially costly for corporations grappling with through the customer’s direct network. These results are
the difficulties posed by managing a global supply chain, robust to the inclusion of a multitude of firm-level control
but there are various arguments for why these firms variables, the log of a country’s gross domestic product
would want their suppliers to implement such initiatives. per capita (ln(GDPC)), as well as different combinations
For example, corporations may push suppliers to engage in of fixed effects. In addition, we find that locations of
CSR activities as a means of window dressing to appease customers and suppliers matter for the working of CSR in
various stakeholder groups and avoid negative publicity, supply chains. Customers play a crucial role in improving
may help recruit, motivate, and retain employees,3 may CSR standards at their suppliers when their countries have
attract new customers and increase market share,4 may similar standards of CSR. Finally, our key finding is robust
improve the firms’ image in the investment community to using (i) differential CSR measures between customers
and suppliers, (ii) alternative CSR databases (i.e., MSCI
Intangible Value Assessment and Sustainalytics), (iii) social
and environmental aspects of CSR, and (iv) falsification
2
As of 2017, more than 9,500 corporations from 160 developed and tests.
developing countries have become participants of the United Nations
Our finding of strong correlations between customer
Global Compact program, a global initiative to encourage “companies to
align strategies and operations with universal principles on human rights, CSR and subsequent improvement in supplier CSR might
labour, environment and anti-corruption, and take actions that advance not reflect a causal relation since customers’ CSR prac-
societal goals. ”https://www.unglobalcompact.org/what- is- gc/participants. tices might be correlated with their selective preference
3
For example, Deloitte Touche Tohmatsu provides its young managers toward suppliers who are more likely to cooperate and
the opportunity to participate in year-long education programs dedicated
to improving their skills and abilities. The company believes “the effort
commit to higher CSR standards. Thus, it is apparent
should help recruit top candidates... and increase retention rates of high- that identifying the impact of customers on supplier CSR
potential employees.” See “Deloitte focuses on ethics,” by Alina Dizik, Wall performance can be empirically challenging. One may
Street Journal, November 19, 2009.
4
For example, to court younger consumers and to reinvent their image,
5
Louis Vuitton buys stakes in organic clothing makers and has been de- There is an increasing number of institutional investors who now fac-
veloping eco-friendly products and supporting environmental causes. See tor CSR criteria into their selection processes to include socially responsi-
“Luxury-goods makersbrandish green credentials” by Rachel Dodes and ble companies (Starks et al., 2018).
6
Sam Schechner, The Wall Street Journal, July 2, 2009. https://www.reprisk.com/publications.
JID: FINEC
R. Dai, H. Liang and L. Ng / Journal of Financial Economics xxx (xxxx) xxx 3
tackle this identification issue by investigating the effect Our evidence suggests that a key mechanism by which
of an initial implementation of a country-level ESG regula- customers exert influence is through a positive assortative
tion on CSR propagation along the supply chain. But such matching of CSR attributes. Customers tend to establish
regulatory ESG mandates often have different implications relationships with suppliers that are likely to exhibit
from those arising from voluntary CSR engagements. As a socially and environmentally responsible behavior. On the
consequence, any mandatory response taken by customers other hand, customers may terminate these relationships if
that in turn impacts suppliers may not necessarily suggest suppliers are unable to meet the customers’ CSR demands
voluntary CSR actions made by customers beyond the (e.g., Banerjee et al., 2008). If the assortative matching is
regulation. To circumvent this identification problem, we the mechanism, a severance of the economic link does
examine the unilateral effect of customers on supplier CSR not imply a weakened CSR spillover effect, unless such
by using a regression discontinuity design (RDD) that re- terminations are exogenous. We test this mechanism on
lies on exogenous variation generated by voting outcomes a sample of newly linked and delinked customer-supplier
of customers’ shareholder-sponsored CSR proposals that relationships. Additionally, we construct a sample of target
pass or fail by a small margin of votes (around the actual customer and supplier firms that are acquired by corpo-
majority hurdle). The passage of these close-call proposals rations with no prior economic link with either the target
is similar to a random assignment of CSR to firms and customer or the target supplier. We consider such target
hence should not influence the supplier’s future CSR firms a source of exogenous variation of the CSR effect.
performance. Conceptually, there should be no systematic When a supplier or a customer is targeted (i.e., such firms
differences between suppliers whose customers pass and are targeted not necessarily by their own choice) and
those whose customers reject a CSR proposal by a small successfully acquired by another firm which is not part
margin of the votes. Therefore, close-call CSR proposals of the supply chain, we expect the stakeholder effect of
provide a source of random variation of a corporate cus- CSR to become weaker. The results are in line with our
tomer’s commitment to CSR that can be used to estimate expectations.
the causal effect on its supplier’s CSR practices.7 Our Another mechanism is through stakeholder bargaining
results suggest that the passage of a customer’s close-call power and/or through suppliers’ decision-making pro-
CSR proposal is followed by the adoption of similar CSR cess. We argue that the bargaining power of a customer
practices by its supplier. The latter’s CSR score in the fol- depends on its reliance on the relationship-specific invest-
lowing year is significantly higher (i.e., 24% of the standard ment (RSI) made by its supplier and the competition in-
deviation of CSR score) than the supplier’s in which the tensity of an industry. When the customer depends heavily
vote fails by a small margin. on its supplier’s RSI, it has less power to impose greater,
In another identification strategy, we employ exoge- typically costly, CSR commitment on the supplier. Prior
nous shocks related to unexpected product safety scandals literature suggests that customers from research-intensive
that have created global shocks to consumerism and the industries tend to be involved in specialized inputs that re-
general public,8 and find stronger customers’ influence on quire their suppliers to make investments consistent with
supplier CSR following these unexpected global shocks. their own (e.g., Armour and Teece, 1980; Levy, 1985; Allen
Since our analysis looks at the same customer-supplier and Phillips, 20 0 0; Dhaliwal et al., 2016; Chu et al., 2019).
pair before and after the shocks associated with the scan- Following this strand of literature, we employ a supplier’s
dals, the positive CSR effect should be attributed to the level of research and development (R&D) and number of
customer’s immediate push for suppliers to improve their patents registered as measures of RSI. Similarly, we expect
socially responsible behavior in response to the scandal. In a customer to be powerful when its industry is more con-
terms of economic significance, a one-standard-deviation centrated, or when its supplier’s industry is highly compet-
increase in the customer’s product responsibility rating itive. The results suggest that customers are less inclined to
in the scandal year will generate a 9.0%–10.9% rise in affect their supplier’s CSR performance when the supplier
the supplier’s mean product responsibility rating. An is highly innovative, or when the supplier’s Herfindahl-
additional test also suggests that such scandals result in Hirschman Index (HHI, a measure of industry competitive-
an increase in the supplier’s reputation risk index and ness) is low, or when the customer’s HHI is high.
ultimately lead to a rise in the customer’s reputation risk The extent to which a customer can push a supplier
index as well. Combined, these results stemming from the for more environmental and social responsibilities may
two identification strategies allay potential endogeneity possibly depend on network connectedness. One strand
concerns on the impact of corporate customers on the CSR of literature suggests that common ownership produces
performance of suppliers. positive externalities as shareholders aim to maximize the
value of firms in their portfolio as opposed to individual
firm value.9 Another strand documents that board net-
works via interlocked directors are conduits for common
7
Flammer (2015a) and Cuñat et al. (2012) also adopt a similar ap- behaviors across board-linked firms.10 We therefore expect
proach to study the effects of the passage of CSR proposals and of cor-
porate governance proposals on stock returns, respectively. Cao et al.
9
(2019) employ RDD to look at peer effects induced by product-market For example, Freeman (2017) provides evidence that common institu-
connections. tional ownership strengthens customer-supplier links and has synergistic
8 effects on the related firms.
Barrot and Sauvagnat (2016) find that natural disasters propagate in
10
production networks. Customers of suppliers hit by firm-level shocks ex- For example, Chiu et al. (2013) find that earnings management
perience a 2–3% fall in sales growth following the event. spreads between firms that share common directors.
JID: FINEC
common investor and board networks in a customer- and Renneboog, 2017) and interactions with other firms
supplier pair facilitate CSR propagation. In other words, (Flammer, 2015b; Cao et al., 2019), plays a role in CSR.
when an investor of a customer firm subsequently also However, little is known about how CSR is influenced by
holds a stake in a supplier firm, or a director from the economically linked stakeholders. Our focus on an impor-
customer’s board thereafter also serves on the supplier’s tant type of stakeholders, namely, corporate customers,
board, such connectness allows the customer to wield helps to reconcile some puzzles in the emerging CSR
influence. Our overall evidence that customers success- literature, especially why firms often engage in costly CSR
fully influence suppliers’ decisions on better responsible activity. The fact that such activity is increasingly preva-
business practices is consistent with this prediction. lent worldwide may be a result of forces by other market
Our findings, thus far, point to collaborative or cooper- players, such as powerful customers. This is especially
ative CSR efforts along the supply chain, where suppliers the case when societal demand for CSR becomes greater
are willing or are coerced to align their CSR standards following numerous CSR-related scandals in recent years.
with those of their customers. Such efforts perhaps reflect Our findings not only enhance our understanding on what
the fact that CSR decisions are not made in a vacuum drives CSR but also, more generally, shed light on non-
but, rather, are made through an informed understanding economic incentives and practices of modern corporations
of the benefits reaped and the costs incurred. We then around the world.
proceed to examine the economic implications of these Second, our research contributes to the understanding
collaborative CSR efforts between customers and suppliers. of how corporate policies and behavior spill over along
Previous studies show value enhancements in corporations global supply chains and the value implications of such
that implement CSR initiatives, such as issues related to spillovers. It also expands the supply chain literature, such
human rights, the community, the environment, and the as the spillover of corporate tax avoidance (Cen et al.,
treatment of employees (e.g., Dowell et al., 20 0 0; Gillan 2017b), innovation knowledge transfers (Chu et al., 2019),
et al., 2010; Edmans, 2011; Krüeger, 2015; Ferrell et al., and information diffusion along supply chains (Cen et al.,
2016). However, implementing these CSR initiatives is 2016; Cen et al., 2017a; Dai et al., 2020). By focusing on in-
costly and has negative financial implications (e.g., greater ternational corporate customer-supplier relationships, our
cost structure and agency problems) for their corporations study joins this strand of literature and further demon-
(Balotti and Hanks, 1999; Masulis and Reza, 2015). Unlike strates that some corporate behaviors, such as CSR, prop-
these studies that focus on corporations’ own CSR activities agate unidirectionally across countries. These institutional
and performances, our analyses look at the economic im- and firm-level nuances are often overlooked in the extant
pact of collaborative CSR efforts of customers and suppliers literature.
through their alignment of CSR standards. The increase of
collaborative efforts helps improve operational efficiency
and firm valuation for both the customer and supplier but 2. Data and summary statistics
enhances only the customer’s future sales growth.
Our research makes two significant contributions to the This study employs data from several different sources:
literature. First, our research examines the role of a specific (i) information on the global network of customer-supplier
group of stakeholders—corporate customers—in propagat- relationships from the FactSet Revere (‘Revere’) global
ing CSR along global supply chains, and shows evidence supply chain data obtained through the Wharton Research
of a strong unilateral influence on CSR from the customer Data Services (WRDS); (ii) information on firm-level CSR
to the supplier only. While this evidence is interesting on ratings provided by Thomson Reuters ASSET4 ESG (i.e.,
its own, our study further shows that there are economic Environment, Social, and Governance) database, together
benefits associated with an improved CSR along the supply with alternative ratings information from MSCI Intangible
chain. A contemporaneous study by Schiller (2018) investi- Value Assessment and Sustainalytics; country-level CSR
gates whether a global supply chain acts as a mechanism ratings are obtained from Vigeo Sustainability data; (iii)
through which CSR spills over from customers to suppliers. mergers and acquisitions (M&A) information from Securi-
But his study looks at changes in regulation only on ESG ties Data Company (SDC) Platinum from Thomson Reuters;
disclosures, whereas ours utilizes both quasi-randomized (iv) R&D and sales information for computing a firm’s
and quasi-natural experiments on real ESG actions to industry concentration intensity from Worldscope, and
establish the impact of customers’ voluntary rather than patent data from the European Patent Office’s worldwide
mandatory CSR practices on suppliers, and shows that Patent Statistical Database (PATSTAT); (v) international
shareholders’ proposals, expected product safety scandals, ownership information from the FactSet Global Owner-
industry structure, and network connectedness all play ship data; (vi) records of interlocking directorates from
a crucial role in propagating CSR. Our analyses offer BoardEx company-level networks data from WRDS, which
new insights on how CSR gets transmitted around the covers over 550,0 0 0 interlocking individuals worldwide;
world. (vii) voting data from Institutional Shareholder Services
Existing studies attribute CSR to a firm’s strategic pur- (ISS) Global and US Voting Outcomes databases; (viii)
suit for superior financial performance (Flammer, 2015a), information on firm-level reputation risk index is available
or a manifestation of agency problems (Masulis and Reza, from RepRisk data; (ix) news records from Ravenpack,
2015; Cheng et al., 2016). Recently, researchers began to Factiva, and Lexi-Nexis Bulk API, and (x) control variables
investigate how a firm’s surrounding environment, such as from Datastream Worldscope. The definitions of all key
national institutions (Ioannou and Serafeim, 2012; Liang variables are depicted in Appendix Table A.3.
JID: FINEC
87
suppliers
Fig. 1. A snapshot of FactSet Revere information on BMW and its worldwide suppliers.
This figure uses BMW and its suppliers from different countries as an example of a customer-supplier network reported in the FacetSet Revere database.
In 2007, FactSet Revere reports that BMW has 87 suppliers from the U.S., Europe, Canada, China, Japan, South Korea, and Mexico.
2.1. Global economic links names or shared similar names with other firms. Further-
more, the information is biased toward US suppliers that
Revere offers a unique database of supply chain rela- are Securities and Exchange Commission (SEC) filers.
tionships that identifies companies’ interrelationships and We merge Revere data with other sources of data,
their comprehensive geographic revenue exposures, start- mentioned above, and our final sample consists of 34,117
ing from April 2003. It covers about 23,400 global compa- unique corporate customer-supplier pairs, with customers
nies, whose information is culled from company regulatory and suppliers from 50 different countries worldwide.
filings, websites, and daily updates based on new filings, Columns 1–2 of Appendix Tables A.1 and A.2 report
press releases, and corporate actions releases. Revere gath- the numbers of suppliers and customers by year and
ers information on corporate direct relationships disclosed by country, respectively. The numbers of suppliers and
by the reporting company and on indirect relationships customers are monotonically increasing over time from
not disclosed by the reporting company but by companies 1,427 and 1,410 in 2003 to 14,427 and 13,227 in 2015,
doing business with the reporting company. For example, reflecting the growing coverage of firms by Revere and
their public sources of US firms include regulatory filings the expanding network of supply chain relationships. By
(e.g., 8-K, 10-Q, and 10-K), investor presentations, websites, country, Morocco has the smallest number of suppliers
and press releases. One advantage of Revere data is that with CSR ratings (i.e., 3), whereas the US has the largest
they contain information of both major and minor private (i.e., 38,546). On the other hand, the number of customers
and publicly listed customers, as well as their GVKEY iden- covered in Revere is the smallest in Egypt and Kuwait (i.e.,
tities for publicly listed firms. To illustrate the information 2) but is the largest in the US (i.e., 43,741).
contained in the Revere database, Fig. 1 shows a 2013
snapshot of BMW with some examples of its 87 suppliers 2.2. Global CSR ratings
from the US, the Euro markets, Canada, China, Japan,
South Korea, Mexico, and other countries worldwide. For Thomson Reuters’s ASSET4 database provides ESG
example, Alfa is BMW’s supplier in Mexico; Hankook, ratings of more than 6,0 0 0 publicly listed companies
Hyundai, and Mobis are its suppliers in South Korea; and worldwide and is employed in studies such as Ferrell
Baosteel in China. Under Statement of Financial Accounting et al. (2016), Dyck et al. (2019), Hsu et al. (2017), among
Standards Regulation Nos. 14 and 131, firms are required others. The ASSET4 database starts in 2002, but a more
to disclose any major customer that represents at least 10% extensive coverage of ESG ratings of firms whose stocks
of the firms’ total reported sales. Unlike Revere data, the are index members of SMI, DAX, CAC 40, FTSE 100, Stan-
Compustat segment data, which are commonly employed dard and Poors (S&P) 500, and Nasdaq 100 is available
in existing studies, obtain information on supply chain beginning from 2003. As the database expands, it also
relationships only from companies’ annual 10-K filings includes index members of DJ STOXX MSCI World in 2008,
and hence, contain a revenue distribution of firms’ major S&P/TSX Composite in 2009, Russell 1000, MSCI Emerging
customers. A critical limitation of Compustat segment data Markets in 2011, Bovespa in 2012, S&P ASX 300 in 2013,
is that it provides only names of customers instead of per- S&P NZX 50 in 2016, and Russell 20 0 0, IPC 35, IPSA 40,
manent identifiers. This further reduces the data coverage MERVAL, COLCAP, PERU General Index in 2017. The rat-
and accuracy, especially among firms that have changed ings consist of more than 750 environmental, social, and
JID: FINEC
corporate governance data points, including all exclusion among others) that will adversely affect the score. As a re-
(ethical screening) criteria and all aspects of sustainability sult, there is no overall systematic shift in the CSR effect.
performance. Every data point goes through a multi-step In Table 1, we present a more detailed distribution of
verification process, including a series of data entry checks, CSRS and CSRC , with their mean, median, standard devi-
automated quality rules, and historical comparisons. These ation, minimum and maximum values, and their values
data points reflect more than 280 key performance in- at 25th and 75th percentiles. In addition, we also report
dicators and are rated as both a normalized score (zero ESG component scores, particularly pertaining to envi-
to 100) and the actual computed value. Thomson Reuters ronmental (Env), Social (Soc), and product responsibility
employs the percentile score calculation methodology (Product) issues. It is evident that the mean and median
to compute their Environmental, Social, and Governance of CSR and ESG component scores are consistently greater
(ESG) scores,11 and evaluates the environmental and social for customers than for suppliers, further confirming that
ratings of a firm relative to those of their peers in the suppliers tend to be socially less responsible, compared to
same industry around the world and the firm’s governance customers.
rating against the country’s level. For robustness, our study also employs two alternative
For all companies, at least three years of historical in- firm-level ESG databases, namely, MSCI ESG Research
formation are available, and most companies have cover- Intangible Value Assessment (IVA) and Sustainalytics’ ESG
age for at least ten years. It is worth mentioning that firms Research & Ratings (Sustainalytics). Both databases provide
in the ASSET4 sample are rated based on both their ESG research, ratings, and analysis of companies’ risks and op-
compliance (regulatory requirements) and ESG engagement portunities arising from ESG factors. IVA industry-adjusted
(voluntary initiatives), and the effectiveness of their en- weighted average scores are between zero and ten and
deavor. Therefore, a firm’s CSR rating or score reflects Sustainalytics’ are between zero and 100. Both ratings
a comprehensive evaluation of how the firm engages in gauge how well companies manage CSR issues that are
stakeholder issues and complies with regulations. Columns related to their businesses and provide an assessment of
3–4 of Appendix Tables A.1 and A.2 show average compos- firms’ ability to mitigate risks and capitalize on opportu-
ite values of supplier CSR scores (CSRS ) and customer CSR nities. Similar to ASSET4, these two alternative ratings are
scores (CSRC ), by year and by country, respectively. The an- also industry-adjusted. In a subsequent section, we show
nual average CSRS score ranges from 52.69 in 2003 to 67.43 that our baseline evidence is robust to alternative CSR
in 2011, and the annual average CSRC score varies between ratings.
62.83 in 2003 and 75.25 in 2015. While these across- In addition, our analysis assesses a country’s CSR
year statistics suggest that customers, on average, are so- standard using Vigeo’s country-level sovereign ESG scores
cially more responsible than suppliers, there seems no sign and benchmarks based on 120 ESG risk and performance
of improved CSR ratings through time, despite the grow- indicators in three domains: (1) environmental protection,
ing expansion of CSR practices implemented by corpora- (2) social protection and solidarity, and (3) rule of law and
tions worldwide in this past decade. Hence, it is important governance. Countries are graded on a scale of one to 100
to understand how the widely employed ASSET4 ratings, on their commitment and performance in these indicators
as well as the alternative CSR databases, are constructed. (e.g., ratification of the Kyoto convention, the Vienna
These ratings are rescaled every year and that explains convention, the Stockholm convention, CO2 emissions per
why there seems no improvement in CSR across time. head, Gini index, etc.).
One concern is that rescaling of a firm’s rating rela-
tive to the ratings of its peers in the same industry may 2.3. Identification variables
possibly give rise to a mechanical adjustment of ESG rat-
ings of firms, particularly for large firms that first enter When implementing our identification strategies, we
the database. As Thomson Reuters adds more firms to the use information on CSR-related proposals from the ISS
ASSET4 database through the years, this addition of cor- voting database for both US and international firms, which
porations may shift the effect of ratings of all firms in also provides information on the threshold of passage,
the database. As smaller and more obscure firms enter the as some proposals are not necessarily passed by more
database, the large corporations that are already in the than 50% of support. For example, they may only be con-
database will result in having better scoring. While the ad- sidered as being passed if they receive supporting votes
dition of these new companies to an industry may, to some that are over 66.7% or 75% of the cumulative votes. Our
extent, have an impact on CSR ratings of firms, any adjust- analysis uses actual thresholds of passing instead of the
ment in ratings depends on the company’s and its peers’ conventional 50% threshold.
data transparency. There are some positive polarity data We also employ information on the firm-level reputa-
measures (i.e., policy emissions, biodiversity impact reduction risk index constructed by RepRisk, the only provider
tion, environmental products, among others) that will im- that systematically analyzes adverse ESG and business
prove the score, but there are also some negative polar- conduct data that can have a reputation and financial
ity data measures (i.e., self-reported environmental fines, impact on a firm. RepRisk’s research captures and analyzes
genetically modified organism products, animal testing, information from media, stakeholders, and other public
sources external to a firm. Its approach assesses whether
a firm’s policies and processes are translating into actual
11
https://www.refinitiv.com/content/dam/gl/en/documents/ performance. For example, a corporation may have a
methodology/esg- scores- methodology.pdf. human rights policy, and RepRisk would check whether
JID: FINEC
Table 1
Summary statistics.
The table reports the number of customer-supplier pair observations (NObs) and summary statistics of various corporate social responsibility (CSR) scores, firm-level characteristics, and GDP per capita employed
as control variables, and the variables employed for the mechanisms of CSR propagation effects. These variables are reported for both suppliers and customers. The CSR variables are ratings associated with the
composite CSR score, environmental (Env), social (Soc), and product responsibility CSR issues. Firm characteristics include leverage, return on assets (ROA), Tobin’s Q (Q), log of total assets (ln(TAssets)), % of
shares held by institutional investors and blockholders/insiders (Inst/Close holdings), sales growth, as well as the country’s log of gross domestic product per capita (ln(GDPC)). The channel variables include R&D,
R. Dai, H. Liang and L. Ng / Journal of Financial Economics xxx (xxxx) xxx

patents, Herfindahl-Hirschman Index (HHI), % of shares held by common institutional investors (%Shares ComInst), numbers of common owners (#ComInst) and directors in both customer and supplier firms,
and # of common directors holding multiple board positions at the supplier. All the variables are defined in Appendix Table A.3.
ARTICLE IN PRESS
Variables associated with Suppliers Variables associated with Customers
Variable NObs Mean Median Stdev Min 25th 75th Max NObs Mean Median Stdev Min 25th 75th Max
CSR composite and component scores

CSR 55,694 63.74 73.88 29.09 2.91 37.77 90.49 98.45 55,694 71.60 83.09 26.51 2.53 56.23 92.55 98.49
Env 51,192 60.88 73.78 32.03 8.27 25.68 91.19 97.49 55694 69.20 86.27 30.24 8.27 45.11 92.72 97.50
Soc 51,192 59.98 67.71 30.14 3.54 31.68 88.05 98.80 55,694 68.73 79.82 27.17 3.54 50.64 90.65 98.88
Product 51,027 57.51 60.02 29.14 2.19 33.32 85.79 99.20 55,399 57.73 57.76 27.91 2.19 34.63 85.44 99.22
The channel variables
R&D 35,501 0.05 0.03 0.05 0.00 0.01 0.08 0.18 32,892 0.03 0.02 0.04 0.00 0.00 0.05 0.14
Patents 1364 249.62 12.00 572.10 1.00 2.00 52 2803 2065 245.79 12.00 560.23 1.00 3.00 98 2803
HHI 55,691 0.18 0.08 0.22 0.01 0.04 0.23 1.00 52,553 0.22 0.11 0.25 0.01 0.05 0.30 1.00
%Shares ComInst 55,694 0.39 0.32 0.30 0.00 0.12 0.63 2.13 55,694 0.31 0.21 0.27 0.00 0.09 0.52 2.35
#ComInst 55,694 214.50 154 216.02 0.00 72 284 2210 55,694 214.50 154 216.02 0.00 72 284 2210
# of Common directors 55,694 0.06 0.00 0.49 0.00 0.00 0.00 26.00 55,694 0.06 0.00 0.49 0.00 0.00 0.00 26.00
# of Board positions 55,694 0.07 0.00 0.56 0.00 0.00 0.00 30.00 55,694 0.07 0.00 0.56 0.00 0.00 0.00 30.00
Control variables
Leverage 55,694 0.26 0.24 0.17 0.00 0.13 0.37 0.60 52,551 0.25 0.24 0.15 0.00 0.15 0.36 0.58
ROA 55,694 0.08 0.07 0.07 -0.10 0.04 0.12 0.21 52,201 0.08 0.07 0.07 -0.05 0.04 0.12 0.22
Q 55,694 1.80 1.52 0.88 0.86 1.16 2.15 4.27 52,551 1.61 1.35 0.71 0.89 1.08 1.89 3.58
MB 55,669 2.98 2.34 2.20 0.54 1.39 3.78 8.77 52,493 2.69 2.01 2.07 0.51 1.22 3.38 8.28
ln(TAssets) 55,694 22.94 22.78 1.57 14.87 21.89 23.97 28.86 52,551 24.13 24.15 1.57 15.63 23.03 25.30 28.96
Inst/Close holdings 55,694 0.15 0.05 0.20 0.00 0.01 0.22 1.00 49,875 0.17 0.06 0.22 0.00 0.00 0.27 1.00
SalesGrowth 55,694 0.07 0.06 0.14 -0.16 -0.01 0.14 0.51 51,751 0.05 0.04 0.12 -0.15 -0.02 0.11 0.40
[m3Gdc;February 14, 2020;3:56]

ln(GDPC) 55,694 10.76 10.79 9.47 6.90 10.66 10.91 11.69 55,029 10.72 10.78 9.58 6.90 10.63 10.87 11.69
7
JID: FINEC
stakeholders and sources at the local level properly report Indicators. The distribution of these control variables is
about how this corporation is handling human rights displayed in the bottom panel of Table 1. The descriptive
issues. statistics show that on average, suppliers have marginally
lower ROA and total assets, but are associated with greater
2.4. Channel variables Tobin’s Q, larger number of closely held shares, higher
sales growth, and larger ln(GDPC), compared with those of
We describe several proxies that we employ to ex- their customers.
plore a number of possible channels through which
customers affect suppliers’ CSR practices. (i) The supplier’s 3. CSR and customer-supplier relationships
relationship-specific investment is measured by its R&D
and number of patents registered. A firm’s R&D informa- In this section, we examine whether corporate cus-
tion is obtained from Datastream Worldscope, whereas tomers affect social and environmental engagements at
patent information is available from PATSTAT, which their economically linked firms—the suppliers. Specifically,
contains information on patents awarded to companies, we investigate the CSR effect along the supply chain and
individuals, and other institutions. (ii) A firm’s Herfindahl- determine whether any evidence of such effects depends
Hirschman Index (HHI) is employed to gauge the degree of on the locations of the customer and the supplier. We also
industry competitiveness. (iii) Network connectedness in conduct a host of additional tests to ensure robustness of
a customer-supplier relationship is measured by common- our baseline evidence.
ality in institutional ownership and directorates of both
the supplier and customer. Common ownership holdings
3.1. Stakeholder (‘corporate customer’) effects of CSR
information is obtained from the FactSet global ownership
database. FactSet gathers US institutional holdings from
To examine whether corporate customers, as stakehold-
mandatory quarterly 13F filings with the Securities and
ers, play a role in influencing suppliers’ CSR practices, we
Exchange Commission and holdings of non-US equities
estimate the following model.
from sources such as national regulatory agencies or
stock exchange announcements (e.g., the Regulatory News CSRS Score(t + 1 ) = a0 + a1 CSRC Score(t )
Service in the U.K.), local and offshore mutual funds,
K
mutual fund industry directories (e.g., European Fund + bk Xk (t ) + FE(t ) + (t + 1 ), (1)
Industry Directory), and company proxies and annual k=1
reports. Our analysis employs the percentage of shares
where CSRS Score(t + 1 ) denotesa supplier’s CSR rating
held in customer and supplier firms by common owners
at year t + 1, CSRC Score(t) is its customer’s CSR rating
and number of common owners in a customer-supplier
at year t. Our hypothesis predicts that the coefficient
pair. Our interlocking directorates (and executives) records
a1 is positive, which would suggest that the higher the
are obtained from BoardEx company-level networks data.
customer’s CSR rating, the better is the supplier’s subse-
Summary statistics of these channel variables are
quent CSR performance. Conversely, it is also likely that
shown in the second panel of Table 1. The statistics
socially responsible suppliers may likewise want to deal
suggest that suppliers tend to invest more in R&D than
with customers that engage in ethical business practices.
their customer counterparts. The mean (median) ratio of
In other words, suppliers can influence customers to meet
R&D relative to total assets is 0.05 (0.03) for suppliers and
their expectations of ethical and responsible behavior. To
0.03 (0.02) for customers. While there are significantly
test the effect of suppliers on customers’ CSR, we reverse
fewer firm-year observations for the number of registered
the roles of CSRS and CSRC in (1).
patents, suppliers have slightly larger R&D and number
Drawn from the existing literature,12 our analyses
of registered patents than customers. These statistics
also control for firm-specific characteristics or country
may reflect suppliers’ investment specific to the needs of
characteristics, as represented by Xk (t), which include a
their customers. The average percentage of shares held
firm’s leverage (Leverage), return on assets (ROA), To-
by common owners is 0.39% in a supplier and 0.31% in
bin’s Q, log of total assets (ln(TAssets)), percentage of
a customer, and the average number of common owners
shares held by institutional investors and closely held
is 214.5. The mean number of common directors is 0.06
shares (Inst/Close holdings), sales growth (SalesGrowth),
and the mean number of board positions held by common
as well as a country’s gross domestic product per capita
directors is 0.07.
(ln(GDPC)) that have previously been found to affect CSR.
All these variables are defined in Appendix Table A.3. Ad-
2.5. Control variables
ditionally, we incorporate different combinations of fixed
effects FE, such as the customer-supplier firm fixed effect,
Our analysis controls for firm-level covariates com-
customer-supplier industry fixed effect, customer-supplier
monly employed in the existing literature, such as lever-
country fixed effect, and year fixed effect. We employ the
age, return on assets (ROA), Tobin’s Q, log of total assets
random-effects generalized least square (GLS) approach
(ln(TAssets)), the percentage of shares held by institutional
to estimate the different model specifications in the table
investors and closely held shares (Inst/Close holdings),
and throughout the study.
and sales growth (SalesGrowth). Given the cross-country
variation of our data, we control for country-level log
of GDP per capita (ln(GDPC)) obtained from World Bank 12
See the references in the Introduction section.
JID: FINEC
Table 2
Corporate social responsibility (CSR) propagation along global supply chains.
This table reports results from the regression of supplier CSRS (t + 1 ) score on customer CSRC (t) score, or from the regression of customer CSRC (t + 1 )
score on supplier CSRS (t) score as follows.

K
CSRS Score(t + 1 ) = a0 + a1 CSRC Score(t ) + bk Xk (t ) + FE(t ) + (t + 1 ).
k=1
K
CSRC Score(t + 1 ) = a0 + a1 CSRS Score(t ) + bk Xk (t ) + FE(t ) + (t + 1 )
k=1
In Columns 1–4, the dependent variable is supplier CSR score and the key independent variable is customer CSR score. Conversely, in Columns 5–6, the
dependent variable is customer CSR score and the key independent variable is supplier CSR score. In Columns 7–8, the dependent variables, CSRS and
CSRC , are defined as the change of supplier CSR and of customer CSR from t to t + 2, respectively. Note that in Column 4, we include all customer-supplier
relationships with less than two years of relationship, but for the other columns, we require that the customer-supplier link must exist for at least two
years. Control variables Xk include firm characteristics of the supplier, customer, or both. They are leverage, return on assets (ROA), Tobin’s Q, log of total
assets (ln(TAssets)), % of shares held by institutional investors and blockholders/insiders (Inst/Close holdings), sales growth, as well as ln(GDPC). All the
variables are defined in Appendix Table A.3. NObs is the number of firm-year customer-supplier pair observations. The regressions also include intercepts
and combinations of different customer-supplier (CS) firm fixed effects, as well as industry, country, and year fixed effects (FE), and all standard errors
reported in parentheses are clustered at the customer-supplier-pair level. ∗ , ∗ ∗ , ∗ ∗ ∗ are significance levels denoted at the 10%, 5%, and 1% levels, respectively.
CSRS CSRC CSRS CSRC

Variable (1) (2) (3) (4) (5) (6) (7) (8)
CSRC score(t) 0.026∗ ∗ 0.031∗ ∗ ∗ 0.040∗ ∗ ∗ 0.038∗ ∗ ∗ 0.016∗ ∗

(0.011) (0.008) (0.007) (0.006) (0.008)
CSRS score(t) -0.002 -0.010 -0.008
(0.008) (0.008) (0.007)
LeverageS -8.888∗ ∗ ∗ -11.154∗ ∗ ∗ -10.617∗ ∗ ∗ -9.791∗ ∗ ∗ -2.012 2.172
(1.717) (1.341) (1.249) (1.073) (1.409) (1.515)
ROA S
32.886∗ ∗ ∗ 46.179∗ ∗ ∗ 46.159∗ ∗ ∗ 34.209∗ ∗ ∗ -1.984 29.201∗ ∗ ∗
(2.415) (2.498) (2.361) (2.127) (2.016) (3.392)
QS 0.454∗ 2.139∗ ∗ ∗ 2.150∗ ∗ ∗ 2.148∗ ∗ ∗ 0.185 -0.551∗
(0.265) (0.244) (0.233) (0.207) (0.221) (0.310)
ln(TAssets)S 4.014∗ ∗ ∗ 10.791∗ ∗ ∗ 10.697∗ ∗ ∗ 10.623∗ ∗ ∗ -0.618 -1.644∗ ∗ ∗
(0.545) (0.142) (0.131) (0.110) (0.452) (0.147)
Inst/Close holdingsS -9.853∗ ∗ ∗ -20.219∗ ∗ ∗ -20.031∗ ∗ ∗ -17.094∗ ∗ ∗ 0.818 5.433∗ ∗ ∗
(1.279) (1.079) (1.017) (0.827) (1.142) (1.402)
Sales growthS 2.838∗ ∗ ∗ -23.046∗ ∗ ∗ -22.555∗ ∗ ∗ -17.802∗ ∗ ∗ 1.253 12.370∗ ∗ ∗
(0.958) (1.040) (0.981) (0.751) (0.805) (1.583)
ln(GDPC)S 14.479∗ ∗ ∗ -1.810∗ ∗ ∗ -1.476∗ ∗ -0.238 1.123 -1.355
(2.000) (0.613) (0.576) (0.482) (1.604) (0.834)
LeverageC 1.287 0.323 -18.200∗ ∗ ∗ -8.840∗ ∗ ∗ 2.605∗
(2.178) (1.298) (1.734) (1.781) (1.518)
ROAC -1.111 -3.080 34.585∗ ∗ ∗ 39.558∗ ∗ ∗ 22.370∗ ∗ ∗
(3.207) (3.113) (2.579) (2.398) (3.721)
QC -0.811∗ ∗ 0.588∗ -0.935∗ ∗ ∗ -0.921∗ ∗ ∗ -2.358∗ ∗ ∗
(0.392) (0.330) (0.312) (0.314) (0.416)
ln(TAssets)C 0.736 0.114 3.830∗ ∗ ∗ 2.363∗ ∗ ∗ -2.091∗ ∗ ∗
(0.629) (0.166) (0.493) (0.618) (0.145)
Inst/Close holdingsC 3.045∗ 0.296 -8.660∗ ∗ ∗ -2.793∗ ∗ 5.171∗ ∗ ∗
(1.589) (0.968) (1.217) (1.203) (1.007)
Sales growthC -0.052 -0.016 1.045 -0.851 11.053∗ ∗ ∗
(1.146) (1.203) (0.916) (0.786) (1.566)
ln(GDPC)C -0.786 0.719 7.130∗ ∗ ∗ 9.955∗ ∗ ∗ -0.681
(1.668) (0.483) (1.333) (1.262) (0.635)
NObs 32,633 32,633 37,540 54,968 24,015 27,345 13,240 12,911
Year FE Yes Yes Yes Yes Yes Yes Yes Yes
CS-Firm FE Yes No No No Yes No No No
CS-Industry FE Yes Yes Yes Yes Yes Yes Yes Yes
CS-Country FE Yes Yes Yes Yes Yes Yes Yes Yes
Our results from estimating various specifications of customer-supplier firm fixed effect is to capture any time-
(1) are shown in Columns 1–4 of Table 2, with those for invariant unobservable factor that might simultaneously
the supplier effect of CSR in Columns 5–6. In Column 1, we affect both the customer and supplier ratings. However,
report our estimates of model (1) with all the above listed as discussed earlier, given that the ratings are rescaled
fixed effects incorporated, supplier and customer firm- each year and are fairly stable through time, the effect
specific characteristics, as well as ln(GDPC). In Column 2, of the time-invariant customer-supplier CSR relationship
we remove the customer-supplier firm fixed effect. It is would be unidentifiable in the model with firm fixed
important to stress that the purpose of incorporating the effects incorporated. As seen in Columns 1 and 2, the
JID: FINEC
coefficient of CSRC becomes statistically and economically influence on CSR from the customer to the supplier only.
stronger without the firm FE but with customer-supplier Suppliers do not have a similar influence on customers’
industry and country fixed effects in place; the estimate CSR activities. With corporate customers being closer to
increases from 0.026 (Column 1) to 0.031 (Column 2). end consumers in the supply chain, the finding of a sig-
Thus, consistent with the study of Dyck et al. (2019), our nificant unilateral CSR influence from the customer to the
subsequent analyses should only control for industry and supplier is not surprising. It is consistent with various re-
country fixed effects. cent food and safety scandals that show the dangers of not
In Column 3, we exclude customer characteristics from maintaining quality throughout the supply chain and how
the specification as these variables have virtually no robust corporate customers manage reputation crisis by ensuring
effect on supplier CSR (see Columns 1–2). Comparing the that suppliers commit to high social and ethical standards.
estimates of Columns 2 and 3 also suggests that customer- For instance, the 2008 melamine scandal in China demon-
specific variables are unable to subsume the strong effects strates how suppliers tainted milk to mitigate rising costs,
of the supplier’s characteristics on its own future CSR. The and the scandal severely destroyed China’s dairy industry.
coefficient of CSRC score (i.e., 0.04 with robust standard To rebuild consumer confidence and to restore their rep-
error of 0.007) maintains its strong and positive effect utation, it is corporate customers and not suppliers that
on future supplier CSR, implying that customers are able have placed new importance on product safety and on ac-
to push suppliers to commit to greater CSR standards. In countability and ethical management in the supply chain.
terms of economic significance, a one-standard-deviation However, our baseline results are also consistent with
increase in a customer’s CSRC score will generate, on the notion that customers with weak CSR standards can
average, a 1.67% (= 0.04 × 26.51/63.74) improvement in its lead to lower supplier CSR practices. To rule out this
suppliers’ next-period performance value of CSR (relative alternative explanation, in Column 7, we test the customer
to the mean supplier CSR rating). More importantly, given effect of CSR on the change of supplier CSR. Similarly, in
that for the entire sample period, the customers have, on Column 8, we also assess the supplier effect of CSR on the
average, 4.6 suppliers in their supply chain, the multiplier change of customer CSR. The coefficient of CSRC is positive
effect is 7.58% at the customer level. This economic mag- and statistically significant, whereas that of CSRS is neg-
nitude of our CSR effect is within the range of CSR effects ative and insignificant. These findings further corroborate
(2.1%–12.83%) reported in two related studies that employ our baseline evidence of a unilateral CSR effect from the
the same ASSET4 database (Dyck et al., 2019; Liang and customer to the supplier and that the customers exhibit a
Renneboog, 2017). Specifically, Liang and Renneboog report positive influence on the supplier’s CSR standards.
the difference of 8.33% in the environmental score and In summary, we find that, in aggregate, corporate
of 12.83% in the social score between civil and common customers, as a group of influential stakeholders, play an
law countries. Dyck et al., on the other hand, show that a important role in improving CSR performance of suppliers
one-standard-deviation change in institutional ownership across the world. This evidence also reflects customer ac-
is associated with a 4.5% increase in environmental perfor- tivism as a disciplining mechanism in suppliers’ corporate
mance score and a 2.1% increase in social performance. social responsible behavior.
In Column 4, we expand our sample to include all
customer-supplier links that have less than two years 3.2. Locations of customers and suppliers
of relationships, or the timing of the links is unclearly
reported in the database. The sample size increases from Local institutional environments, including socioeco-
37,540 to 54,968. With brief links established between the nomic, political, and cultural factors, play an important
customer and supplier, it is less likely that the customer role in CSR implementation (Ioannou and Serafeim, 2012).
will have the ability to coerce its supplier to commit to Firms in advanced countries typically take regulatory and
higher CSR standards.The finding shows only a slightly voluntary approaches to CSR issues and commit to simi-
weaker, while statistically significant, customer CSR effect larly high CSR standards. Firms in emerging markets, on
on the supplier; the estimate of customer CSRC score is the other hand, tend to adopt less CSR practices than their
0.038, compared to 0.040 when the customer and supplier counterparts from developed countries. For example, doing
have established at least two years of relationship. Never- business in these economies is probably challenging as
theless, to be conservative, our analyses in the subsequent many are characterized by “either bad or weak public gov-
sections shall report results based on a larger sample of ernance and administration, lack of public transparency,
customer-supplier links.13 high levels of bribery and corruption, poor records of
Next, in Columns 5–6, we switch the roles of CSRS and human rights, inadequate environmental, safety and labor
CSRC in model (1) and instead test the effect of supplier standards, and high levels of poverty and inequality”
CSR on the future CSR performance of the customer. But (Nelson, 2004, p. 31). The varying socio-cultural-political
we find no significant effect of supplier CSR on future contexts make it difficult for a customer to pressure its
customer CSR. Combined with the estimates shown in supplier to act responsibly. Therefore, CSR efforts along the
Columns 1–4, these results suggest a strong unilateral supply chain will vary, depending on the country in which
customers and suppliers do their business. We test this
13
conjecture by replicating our baseline regression results
Similar to the results of Table 2, the estimate of customer CSRC score
for the remaining tables is larger in magnitude when we require that the
shown in Column 4 of Table 2 on a sample of suppliers
customer-supplier link exists for at least two years. The results are avail- and customers located in countries with high vs. low
able upon request. national CSR standards. We employ Vigeo sustainability
JID: FINEC
Table 3
Location and country-level CSR.
This table reports results from the regression of supplier CSR score (CSRS score) on customer CSR score (CSRC score), where customers and suppliers are
located in countries with high vs. low Vigeo sustainable country index, residing in the US, or in different countries. Countries with high sustainable country
index are those that ranked above the median sustainable country index, and the remaining countries are considered low sustainable index countries.

K
CSRS Score(t + 1 ) = a0 + a1 CSRC Score(t) + bk Xk (t ) + FE(t ) + (t + 1 ).
k=1
Unreported control variables Xk include firm characteristics of the supplier (S), and they are leverage, return on assets (ROA), Tobin’s Q, log of total assets
(ln(TAssets)), % of shares held by institutional investors and blockholders/insiders (Inst/Close holdings), sales growth, as well as ln(GDPC). All variables are
defined in Appendix Table A.3. NObs is the number of customer-supplier pair observations. The regressions also include intercepts and customer-supplier
(CS) pair industry, country, and year fixed effects (FE), and robust standard errors reported in parentheses are clustered at the customer-supplier-pair level.
∗ ∗∗ ∗∗∗
, , are significance levels denoted at the 10%, 5%, and 1% levels, respectively.
High customer-country CSR Low customer-country CSR
Supplier-country CSR US Non-US
Low High Low High only countries

Variable (1) (2) (3) (4) (5) (6)
CSRC 0.014 0.043∗ ∗ ∗ 0.013∗ 0.011 0.018∗ 0.036∗ ∗ ∗

(0.014) (0.006) (0.008) (0.012) (0.010) (0.011)
NObs 11,820 13,787 27,987 8223 18,311 15,312
Controls Yes Yes Yes Yes Yes Yes
YearFE Yes Yes Yes Yes Yes Yes
CS-IndustryFE Yes Yes Yes Yes Yes Yes
CS-CountryFE Yes Yes Yes Yes Yes Yes
country index to rank the 50 countries in our sample. We in the US, such that the customer is able to influence the
classify the top 25 countries as those with high national supplier to engage in better CSR practices.14 In Column 6,
CSR standards and the rest as those with low national CSR when the customer and supplier are in non-US countries,
standards. Our analysis also includes tests on a sample of the CSR spillover is stronger. Combined, these results
customers and suppliers residing in the US and on another are suggestive that our baseline evidence is driven by
located in non-US countries.Results are shown in Table 3. customers and suppliers from both the US and non-US
Several notable findings emerge from the table. When countries.
both the supplier and customer are located in countries
with high or with low CSR standards (Columns 2–3), the 3.3. Additional tests
customer has the ability to influence its supplier to oper-
ate in a socially responsible way. It is plausible that both 3.3.1. The positive customer-supplier CSR relationship
parties are in countries with similar socio-cultural-political We conduct additional tests to rule out the alternative
environments and hence, recognize the economic and interpretation of our baseline result that customers with
social importance of committing to better CSR standards. weak CSR performance induce suppliers to become less
The coefficients on customer CSRC score are positive and socially responsible. We construct two additional CSR mea-
strongly significant at the 1% level; they are 0.043 (stan- sures that capture the differential CSR between customers
dard error = 0.006) in Column 2 and 0.013 (standard error and suppliers (CSR gap); one measures the difference
= 0.008) in Column 3. These findings may indicate that between CSRC and CSRS ratings, whereas another is a
while the customer and supplier are in countries with binary indicator that equals one if CSRC score is greater
similar CSR standards, there are still adequate variations than CSRS score and zero otherwise. We therefore test the
in their CSR ratings that induce the supplier to improve its following specification.
CSR performance. However, when the customer is located
in a country with high CSR standards, it faces challenges CSRS Score(t + 1 ) = a0 + a1 CSRC Score(t ) × CSR Gap(t )
to influence its suppliers located in countries with low CSR + a2 CSRC Score(t ) + a3 CSR Gap(t )
standards to implement similarly high CSR practices as the

K
suppliers could find such implementations to be too costly. + bk Xk (t ) + FE(t ) + (t + 1 ). (2)
Similarly, the results provide no evidence of CSR spillover k=1
from a customer residing in a low CSR country to a sup-
plier in a high CSR country. This finding is intuitive, as it If customers have a positive influence on suppliers’ sub-
is hard to impose high CSR standards on suppliers when sequent CSR performance, the coefficients of both the
the customer itself may not maintain a high CSR standard. interaction variable, CSR gap × CSRC score, and customer
In Column 5, when the locations of both the customer
and supplier are restrictedto the US, we find a marginally 14
We have also conducted a similar test based on non-US countries, but
significant and positive CSRC score of 0.018 with standard required that both the supplier and customer are in the same country.
error of 0.010. There appears some cross-sectional varia- Our untabulated results suggest no strong evidence of CSR spillover effect,
tion in the CSR rating across supplier and customer firms an indication that there is seemingly a larger degree of homogeneity in
their social responsibility and ethical behaviors.
JID: FINEC
CSRC score should be jointly positive. That is, the estimates 4 and 5 of Table 2 by replacing the composite CSR score
of a1 and a2 are positive. Results displayed in Columns 1–2 by each component rating (i.e., Env or Soc score) in turn.
of Panel A, Table 4 further confirm our baseline finding Results presented in Panel C of Table 4 further corroborate
that customers play a significant role in suppliers striving the robustness of unilateral CSR effects in environmen-
for high CSR standards. For example, the overall customer tal and social issues only from customers to suppliers.
effect of CSR on supplier CSR is 0.177 (=0.121+0.056) in Specifically, the greater the customer’s environmental and
Column 1 and is 0.820 (=0.016+0.804) in Column 2. All social ratings, the higher are its supplier’s subsequent
the coefficients are statistically significant at conventional environmental and social ratings. The coefficients of EnvC
levels. and SocC scores are 0.032 and 0.013, respectively, and
both coefficients are statistically significant at conventional
3.3.2. Alternative CSR databases levels. In contrast, those of EnvS and SocS scores display no
Oftentimes, one contends that the key evidence of a relation between the supplier’s environmental and social
study is specific to the sample of data it employs.15 In ratings and its customer’s.
other words, it is plausible that our baseline finding of
the unilateral CSR effect from customers to suppliers is 3.3.4. Falsification tests
attributed primarily to the ASSET4 data used in our study. We conduct another robustness check on the baseline
The coverage of ASSET4 data is fairly extensive, and the evidence through falsification tests. We designate pseudo
database is also widely employed in a number of impor- suppliers by matching true suppliers to pseudo suppliers
tant studies, as cited in the Introduction section. Still, it by industry and by country in terms of the number of
is arguable that the assignment of individual firm ratings overlapped sectors and of similar size. Revere provides
may be biased toward the methodology ASSET4 adopts. To information on a firm’s number of overlapped sectors to
rule out this possible bias, we employ two alternative CSR determine the firm’s closest rival. For example, Yahoo and
ratings databases, namely, the MSCI IVA and Sustainalytics Answers.com are the closest rivals to Google in that these
databases, which are also widely used in the literature. rivals and Google overlap substantially in the number of
We then repeat the estimation of Columns 4 and 5 in sectors they operate in. In a given industry,18 we match
Table 2 using firm-level ratings assigned by the two al- a pseudo supplier (SIndustry ) that is closest in size and in
ternative databases. Results in Panel B of Table 4 find the rivalry to a true supplier, but is not linked to the true
customer effect of CSR on supplier CSR, but not the sup- supplier’s customer. We repeat the same procedure by con-
plier effect of CSR on customer CSR.16 The results of our structing another pseudo supplier (SCountry ) from the same
key finding remain materially unaffected; the coefficients country as the true supplier. The results, reported in Panel
of CSRC are positive and strongly significant, whereas D of Table 4, show no spillover CSR effect from a true cus-
those of CSRS in the last two columns yield inconsistent tomer to a pseudo supplier, thereby reinforcing our key re-
signs and are statistically insignificant. Thus, these findings sult that customers can influence only their own supplier’s
further underscore the robustness of our baseline evidence CSR performance but not their suppliers’ competitors.
that the unilateral CSR effect along the supply chain is
4. Identification strategies using quasi-randomized and
not specific to the ASSET4 ratings employed in our earlier
quasi-natural experiments
analyses.
Identifying the impact of customers on their suppliers’

3.3.3. Environmental and social issues
future CSR performance poses an empirical challenge. The
Thus far, we employ firm-level composite CSR indexes
customer effect on supplier CSR might reflect the endoge-
to test the customer effect of CSR in the supply chain.
nous selection made by corporate customers. One might
But it is likely that the effect might be concentrated in
tackle this issue by examining the impact of government-
a specific aspect of corporate social initiatives, such as
mandated initiative or newly instituted country-level ESG
those pertaining to either environmental (Env) or social
regulation on CSR propagation in the supply chain. A criti-
(Soc) issues.17 We again repeat the analysis of Columns
cal problem with such approaches is that these regulatory
outcomes typically bear different implications from those
15
Chatterji et al. (2016) suggest that for every CSR research, one has arising from voluntary CSR engagements. As a result, any
to cross-validate the results using several different ESG samples/data mandatory action taken by customers that, in turn, affects
sources. suppliers does not necessarily reflect CSR, which is mostly
16
In Panel B of Table 4, the coefficient of IVA CSRC is 0.014. Using IVA
voluntary and beyond the regulation. To circumvent such
ratings, Liang and Renneboog (2017) find that, on average, firms in civil
law countries have a 7% higher CSR score (or a half-grade on a zero empirical issues, we instead employ two different identifi-
to six scale) than firms in common law countries (Columns 1 and 2 of cation strategies. First, we follow Flammer (2015a) and Cao
their Table 4, p. 873), whereas our economic significance is about 3.3% et al. (2019) by adopting the RDD approach to examine
(=0.014 × 2.34) which is about half of Liang and Renneboog’s. In other
the causal effect of voting outcomes of customers’ close-
words, the customer effect of CSR is about half of the legal origin effect
of CSR.
call CSR proposals on suppliers’ future CSR performance.
17
We do not consider the corporate governance aspect of CSR, even Second, we explore a couple of exogenous shocks related
though the databases contain ratings of firm-level governance. Our con- to unexpected product safety scandals that have created
cern is that such ratings may not reflect the typical corporate governance global shocks to consumerism and the general public.
issues examined by academic researchers (see Hong et al., 2012). To avoid
any misleading results, we focus strictly on the firm’s environmental and
18
social aspects of CSR. We use the Fama-French 30-industry classifications.
JID: FINEC
Table 4
Additional tests.
This table conducts a host of additional tests of CSR propagation along global supply chains.In Panel A, the key independent variables are the binary
indicator which equals one if (CSRC > CSRS ) and zero otherwise, the differential CSR defined by CSRC − CSRS , and the interaction between the CSR gap
and CSRC . Panel B repeats Columns 3–5 of Table 2 using two alternative CSR rating databases, namely, MSCI IVA and Sustainalytics (Salytics), while Panel C
repeats the same based on ratings associated with environmental (Env) and social issues (Soc). Finally, Panel D conducts falsification tests, where we match
a “false” supplier which is closest to the true supplier in terms of size, industry classification, and is the closest competitor to the true supplier (SIndustry ).
We construct another similar false supplier in terms of size and is the closest competitor to the true supplier from the same country (SCountry ). Across all
panels, the estimations include control variables, leverage, return on assets (ROA), Tobin’s Q, log of total assets (ln(TAssets)), % of shares held by institutional
investors and blockholders/insiders (Inst/Close holdings), sales growth, as well as ln(GDPC). All the variables are defined in Appendix Table A.3. NObs is the
number of firm-year customer-supplier pair observations. The regressions also include intercepts and customer-supplier- (CS-) industry, country, and year
fixed effects (FE), and all standard errors reported in parentheses are clustered at the customer-supplier-pair level. ∗ , ∗ ∗ , ∗ ∗ ∗ are significance levels denoted
at the 10%, 5%, and 1% levels, respectively.
Panel A: Differential CSR measures
Variable CSRS (t + 1 )
∗∗∗
CSR score x I(CSR > CSR )
C C S
0.121
(0.916)
CSR score x (CSR −CSR )
C C S
0.016∗ ∗ ∗
(0.001)
CSRC −CSRS -0.807∗ ∗ ∗
(0.458)
CSRC (t) score 0.056∗ ∗ ∗ 0.804∗ ∗ ∗
(0.003) (0.006)
NObs 48,283 48,283
Controls Yes Yes
YearFE Yes Yes
CS-IndustryFE Yes Yes
CS-CountryFE Yes Yes
Panel B: Alternative CSR rating databases
Variable CSRS (t + 1 ) CSRC (t + 1 )
IVA CSRC (t) score 0.014∗ ∗

(0.005)
Salytics CSRC (t) score 0.020∗ ∗ ∗
(0.008)
IVA CSRS (t) score 0.010
(0.006)
Salytics CSRS (t) score -0.002
(0.008)
NObs 28,056 22,173 23,600 20,805
Controls Yes Yes Yes Yes
YearFE Yes Yes Yes Yes
CS-IndustryFE Yes Yes Yes Yes
CS-CountryFE Yes Yes Yes Yes
Panel C: Environmental and social ratings
Variable CSRS (t + 1 ) CSRC (t + 1 )

C ∗∗∗
Env (t) 0.032
(0.006)
C
Soc (t) 0.013∗ ∗
(0.005)
EnvS (t) -0.005
(0.005)
S
Soc (t) -0.003
(0.005)
NObs 55,685 55,685 51,115 51,115
YearFE Yes Yes Yes Yes
CS-IndustryFE Yes Yes Yes Yes
CS-CountryFE Yes Yes Yes Yes
Panel D: Falsification tests
Variable CSRSIndustry (t + 1 ) CSRSCountry (t + 1 )
CSRC (t) score -0.011 -0.010

(-1.170) (-0.931)
NObs 30,998 26,349
Controls Yes Yes
YearFE Yes Yes
CS-IndustryFE Yes Yes
CS-CountryFE Yes Yes
JID: FINEC
4.1. Evidence from CSR-related shareholder-sponsored investigation of US firms reacting to their product-market
proposals peers’ commitment to CSR. The difference in the findings
probably reflects the larger US firms that are more explicit
Our study employs a sample of 31,634 customer CSR- about CSR practices (Matten and Moon, 2008), compared
related shareholder-sponsored proposals19 that allows us to our sample of international firms, which are gener-
to examine t + 1’s CSR scores of suppliers whose cus- ally smaller and would, expectedly, experience a smaller
tomers marginally pass a CSR proposal in year t (“Pass” reaction.
suppliers) with year t + 1’s CSR scores of suppliers whose While suppliers improve their CSR practices following
customers marginally reject a CSR proposal in year t (“Fail” the passage of the voting customer firms’ proposed CSR
suppliers). Using RDD, we estimate the difference in aver- adoption, it is unclear whether such passed proposals are
age CSR scores between Pass suppliers and Fail suppliers. implemented, or they are conducted without a proposal
RDD is based on a nonparametric, “local” linear estima- being passed. Thus, one may interpret that, to a certain
tion, where small “neighborhoods” (bandwidths) around extent, the passage of a CSR proposal simply conveys a
the threshold are used to estimate discontinuities in the signal about a customer’s commitment to CSR. To address
supplier’s reaction. Following Imbens and Kalyanaraman this imperfect compliance issue and to disentangle the
(2012) and Cao et al. (2019), we utilize a nonparametric effect of signaling and actual adoption, we follow Cao et al.
linear estimation approach to capture the difference in (2019) and perform a fuzzy RDD analysis to estimate the
future CSR performance between Pass and Fail suppliers customer CSR effect on a supplier. The first stage of a fuzzy
with respect to the passage and failure of a CSR proposal RDD is to estimate the probability of implementation at a
by their associated customer firm, as follows. firm-specific cutoff, whereas the second stage provides the
estimate that adjusts for the imperfect compliance. We
CSRS (t + 1 )Score = a0 + a1 Pass(t ) consider three definitions of CSR implementation:20 (1)
+ a2 %Votes For CSR Proposal News about CSR implementation; (2) Increase in ASSET4
+ (t + 1 ), (3) ratings by at least ten grades (out of 100) but still within
a reasonable range (such as within 20 grades, and we
where Pass is a binary indicator that equals one if the also enlarge this range in robustness tests to over 50
supplier’s customer passes a CSR-related proposal and zero and obtain similar results); (3) Board recommends “YES”
otherwise, and %Votes For CSR Proposal is the percentage (or “FOR”) to the proposal before the vote, as this is ex
of vote shares in favor of the CSR proposal, centered at the ante observable at the time of the vote and increases the
threshold (a level specified by the firm). In (3), a1 captures likelihood of the voting firm actually implementing the
the discontinuity at the majority threshold—the difference proposal. Results are reported in Panel C. This probability
in outcomes between Pass and Fail suppliers. increases by 23.9%–79.3% if the voting percentage sur-
We first verify the randomness assumption of our RDD passes a firm’s specific threshold. The second-stage results
setting, which requires that the suppliers of customers are all positive and statistically significant after adjusting
whose voting shares are marginally below or above the for the probability of implementation.
threshold should be similar on the basis of ex ante charac- We now turn to testing whether an implemented pro-
teristics. Otherwise, the passage of close-call CSR proposals posal has a greater impact on the supplier firms than one
would not be viewed as a random assignment if it is some- that is not implemented. We examine all the passed pro-
what associated with supplier firm characteristics prior to posals and then compare supplier responses, conditional
the vote. Results contained in Panel A of Table 5 suggest on whether the customer firm implements its proposal.
no systematic or significant difference between Pass and The results, as reported in Panel D, indicate greater and
Fail suppliers around the majority threshold, which lends statistically significant CSR improvements in the subsam-
support to our identification strategy. Panel B shows RDD ple of implemented than in that of non-implemented
estimates of the difference in CSR performances between proposals for both news of CSR implementation and board
Pass and Fail suppliers, with different bandwidths. Fig. 2 voted “For” categories, while not statistically significant for
provides a visual confirmation of the suppliers’ reactions the increase in ASSET4 ratings category. For example, the
to the outcome of their customers’ voting proposals. The adjusted CSR improvements are 5.778 based on the news
coefficient of the binary indicator Pass is positive and about CSR implementation and 3.990 based on the defini-
statistically significant at the 10% level across different tion that the Board recommends in favor of the proposal,
specifications of bandwidth and when the dependent and these improvements are statistically significant at
variable is measured by CSRS . The a1 estimate is 6.988 conventional levels. Overall, the evidence suggests that the
under the data-driven optimal bandwidth, indicating that passage of a customer’s CSR proposal has a real effect on
the CSR score gap between Pass and Fail suppliers is its suppliers’ subsequent CSR performance—the customer’s
almost 7 points. Given that the mean CSR score is 63.74 CSR implementation induces its suppliers to improve their
with a standard deviation of 29.09, a 6.988 gap is about CSR practices. In other words, if a proposal is subsequently
24% of the standard deviation of CSR score. In comparison, implemented, the customer CSR effect on a supplier is
Cao et al. (2019) estimate the gap to be about 38% of the even stronger than simply getting it passed.
standard deviation of CSR score (using KLD ratings) in their
20
Details of these definitions and construction are shown in Appendix
19
Shareholder-sponsored proposals are typically non-binding. Table A.3.
JID: FINEC
Table 5
Supplier responses to the passage of a CSR proposal and subsequent CSR implementation.
Panel A of this table reports pre-voting differences in the set of observable supplier firm-specific characteristics, namely, leverage, return on assets (ROA),
Tobin’s Q, log of total assets (ln(TAssets)), % of shares held by institutional investors and blockholders/insiders (Inst/Close holdings), sales growth, as well as
ln(GDPC) between supplier firms that are associated with the passage (“Pass” suppliers) of a CSR proposal in customer firms and those that are associated
with the rejection (“Fail” suppliers) of a CSR proposal in customer firms by a small margin. The margin is defined as the optimal bandwidth following
Imbens and Kalyanaraman (2012). Panel B presents the supplier’s future CSR performance (CSRS (t + 1 ) Score) in response to the customer’s CSR votes. It
reports RDD estimations from a local linear regression as specified in the model below using the optimal bandwidth.
CSRS Score(t + 1 ) = a0 + a1 Pass(t ) + a2 %Votes For CSR Proposal + (t + 1 ),
where CSRS (t + 1 ) score is the supplier CSR score in year t+1; Pass is a binary indicator that equals one if the supplier’s corporate customer passes a
CSR-related proposal—that is, if more than a certain specified number of votes are in favor of adopting the CSR proposal and zero otherwise; and %Votes
For CSR Proposal is the percentage of vote shares in favor of the CSR proposal, centered at the threshold (a level specified by the firm and it varies across
firms and countries). a1 captures the discontinuity at the majority threshold—the difference in outcome between customers that marginally pass a CSR
proposal and those that marginally reject a CSR proposal. CSR measures the difference in CSR between t − 1 and t + 1. Panel C presents supplier firms’
future CSR implementation as a response to the CSR votes estimated using a fuzzy RDD approach. We use three definitions of a CSR implementation,
namely, (1) there is news or information that the proposal has been implemented; (2) an increase of more than 10 points in the ASSET4 score; (3) Board
recommending “For” the proposal before the vote, as this is ex ante observable at the time of the vote and increases the likelihood of the customer firm
actually implementing the proposal. The first stage of fuzzy RDD is to estimate the probability of implementation at a firm’s specific cutoff. If there is a
significant jump in the probability of implementation, we perform the second stage to obtain the estimate, adjusting for imperfect compliance. Panel D
presents supplier firms’ future CSR performance conditional on the compliance or implementation of the CSR proposals in the voting customer firms. We
compare the CSR scores of the supplier firms in response to the compliance of the passed CSR proposals. ∗ , ∗ ∗ , and ∗ ∗ ∗ denote significance at the 10%, 5%,
and 1% levels, respectively.
Panel A: Validity for CSR vote (preexisting difference)
Difference in
Difference narrow margin
Fail Pass (Fail-Pass) (optimal band width)
NObs Mean NObs Mean Estimate P-value Estimate P-value
LeverageS 20,068 0.263 13,424 0.249 0.014 0.000 0.016 0.637

ROAS 20,068 0.08 13,424 0.073 0.007 0.000 0.001 0.945
QS 20,068 1.784 13,424 1.798 -0.014 0.136 -0.069 0.265
ln(TAssets)S 20,068 22.808 13,424 23.065 -0.257 0.000 -0.092 0.747
Inst/Close holdings 20,068 0.144 13,424 0.151 -0.007 0.002 -0.019 0.655
Sales growthS 20,068 8.731 13,424 6.805 1.926 0.000 -0.947 0.752
ln(GDPC)S 20,068 10.714 13,424 10.677 0.037 0.000 -0.137 0.165
Panel B: Supplier reactions to the passage of CSR vote
50% of 150% of CSR

Optimal bandwidth Optimal bandwidth Optimal bandwidthvariable Optimal bandwidth
Pass 6.988∗ ∗ ∗ 7.149∗ 10.742∗ ∗ ∗ 4.461∗ ∗

t−stat (3.42) (1.83) (8.44) (4.85)
NObs 4,180 802 12,818 2,169
Panel C: Supplier responses to the passage of a customer’s CSR proposal: A fuzzy RDD approach
News about implementation Increase in ASSET4 ratings Board recommending “For”
First stage Second stage First stage Second stage First stage Second stage
(1) (2) (3) (4) (5) (6)
Estimate 0.386∗ ∗ ∗ 32.857∗ ∗ ∗ 0.239∗ ∗ ∗ 36.309∗ ∗ ∗ 0.793∗ ∗ ∗ 15.995∗ ∗ ∗

t−statistic (34.87) (7.87) (13.11) (2.94) (102.36) (8.16)
Panel D: Supplier responses to implementation and nonimplementation of CSR
News about implementation Increase in ASSET4 ratings Board recommending “For”
Implementation Nonimplementation Implementation Nonimplementation Implementation Nonimplementation

(1) (2) (3) (4) (5) (6)
Supplier CSR score 69.272 63.495 60.276 59.760 68.856 61.031

Difference 5.778∗ 0.516 3.990∗ ∗
t−statistic (1.83) (0.11) (1.97)
4.2. Evidence from global product-safety scandals drawn global attention and heightened societal awareness
and activism demanding for more socially responsible
Our second identification strategy focuses on some practices along supply chains. While there are a number
of the worst product safety issues or scandals in history of human rights scandals in developing countries that
based on our extensive search on the Internet that have have also drawn public outcry and led to questions about
JID: FINEC
0 0 0
Fig. 2. Regression discontinuity plots of the stakeholder effect of CSR.

The figure presents regression discontinuity plots of the average CSR score in year t+1 of corporate customers using a fitted quadratic polynomial estimate
with a 95% confidence level. The full sample covers CSR-related voting outcomes in 3478 shareholder meetings of a total 1610 firms from January 2003 to
December 2015. The x-axis is the distance (in percentage of votes) from the majority threshold by which a proposal can pass. The dots represent reactions
of suppliers in terms of their CSR scores in the subsequent year. The figure is graphed based on 20 equally spaced bins (bandwidth being 5% of the vote
shares), and the grey lines represent their confidence intervals.
the CSR across global supply chains,21 it is conceptu- had been recalled for defective airbags made by Takata,
ally difficult to comprehend why customer stakeholders and potentially 42 million cars in the United States were
would care about their suppliers’ human rights issues affected. The problem was the defective metal airbag
in developing countries.22 Thus, in this subsection, we inflators. When deployed, the huge explosive force may
focus on the Chinese milk scandal and Toyota car/Takata disintegrate the canister of the inflator sending metal
airbag recalls that shed light on the influence of corporate shrapnel into the passenger cabin and possibly injuring
customers towards the CSR practices of their suppliers. or killing the occupants in the vehicles. According to
The 2008 Chinese milk scandal involving infant milk Report Consumers,24 this problem resulted in 16 deaths
formula adulterated with melamine exploded when 16 and about 180 injuries reported worldwide, and affected
infants in China’s Gansu Province were diagnosed with 19 automakers. Toyota recalled more than one million
kidney stones after consuming the melamine-tainted milk vehicles sold in the United States over faulty airbags and
powder.23 Subsequently, an estimated 30 0,0 0 0 babies in windshield wipers. The National Highway Traffic Safety
China were sick from consuming the contaminated milk, Association has called this “the largest and most complex
and the kidney damage led to six deaths. The World safety recall in US history.”25
Health Organization claimed this incident was one of the To examine the impact of such exogenous shocks
largest food safety events it has had to deal with in recent arising from product safety scandals on the customer-
years, and the incident raised serious concerns about food supplier CSR propagation, we utilize a subcomponent of
safety, not only in China but also across the world. the ASSET4 ESG ratings, namely, the rating on product
The other event is the recalls issued by Takata and responsibility (Product), and follow Liang and Renneboog
Toyota in 2013. At least seven million cars worldwide (2017) difference-in-differences approach, as given by
CSRS ProductScore(t + 1 ) = a0 + a1 CSRC ProductScore(t )

× Event + a2 CSRC ProductScore(t ) + a3 Event
21
For example, Nike has long been accused of abusive labor practices
since the early 1970s when outsourcing its manufacturing to developing
economies with labor practices which would be considered illegal in the
K
US Other examples include the 2012–2013 devastating workplace disas- + bk Xk (t ) + FE(t ) + (t + 1 ). (4)
ters in Pakistan and the collapse of a factory in Bangladesh, which to- k=1
gether claimed more than 1,300 lives.
22
Nevertheless, we have analyzed the infamous Nike worker abuse
24
scandal in 2005 and the Rana Plaza building collapse (Bangladesh) in https://www.thestar.com/business/2017/02/27/attorneys-say-five-
2013, and our unreported results show stronger positive CSR spillover ef- automakers- knew- takata- airbags- were- dangerous.html.
fects following these unexpected shocks. 25
https://www.theverge.com/2016/5/4/11591724/takata- air- bags-
23
https://en.wikipedia.org/wiki/2008_Chinese_milk_scandal. largest-recall-nhtsa.
JID: FINEC
10.00
8.00
Abnormal reputaon risk idex level (RRI)
6.00
4.00
2.00
0.00
-5 -4 -3 -2 -1 0 1 2 3 4 5
Months relave to event month
Customer RRI Supplier RRI
Fig. 3. Reputation risk indexes of customers and suppliers around ESG scandal announcement periods.
The figure shows plots of equal-weighted customer and supplier RepRisk indexes around month t = 0 when a supply chain related ESG scandal broke. The
sample contains 1086 scandals covered in the RepRisk database for the period from June 2008 to December 2015. An equal-weighted supplier RepRisk
index (RRI) is constructed by taking the average monthly RepRisk indexes of all suppliers that faced an ESG scandal. We adjust the average monthly
RepRisk indexes five months prior to and following the month a scandal occurred, by subtracting the average mean computed over the past 12 months
from t − 17 to t − 6 from the index level. We repeat the procedure when constructing the supplier’s customer’s corresponding RepRisk index. The supply-
chain relationship is from Factset Revere database.
In (4), CSRS ProductScore is the supplier’s firm-level rating positive and statistically significant in both cases. Their
on product responsibility, CSRC ProductScore is its cus- coefficient is 0.250 (robust standard error= 0.145) in
tomer’s product responsibility rating, Event is a binary Column 1 and 0.268 (robust standard error= 0.122) in
indicator that equals one if it is the specific event year Column 2.27 In terms of economic significance, a one-
(i.e., the event year for the China milk scandal is 2008, standard-deviation increase in the customer’s product
and for the Takata and Toyota recalls is 2013) and zero responsibility rating (i.e., 27.91, Table 1) in the scandal
otherwise.26 Column 1 of Table 6 shows estimates of year will lead to a 9.0% (= (0.268 − 0.082 ) × 27.91/57.51)
model (4) on a subsample of the existing linked firms in to 10.9% (= (0.250 − 0.026 ) × 27.91/57.51) rise in the
the global food-related industry suppliers from China. Col- supplier’s mean product responsibility rating in Columns
umn 2 of the same table reports those using a subsample 2 and 1, respectively. These findings suggest that the
of firms in the global auto industry with suppliers from effect of a customer’s product responsibility rating on
Japan and either customers or suppliers from the auto the supplier’s next-period product responsibility rating
industry in all countries. is more pronounced in the event year, an indication that
The variable of interest is the interaction between CSRC the pressure from corporate customers driving suppliers
product score and Event indicator, and its coefficient is to improve their CSR becomes more intense when public
26
Note that the Event indicator does not capture post-event years be- 27
It is worthwhile to mention that when we interact CSRC product
cause we expect the CSR gap would not be long lasting after such large-
score with other year dummies, our unreported results show that none
scaled scandals and media exposure.
of the coefficients of the interaction term are statistically significant. Such
placebo tests provide further reinforcing evidence of stakeholder effects
in CSR activities along the global supply chain.
JID: FINEC
Table 6
Product safety scandals and reputation risk.
This table tests the effect of product safety scandals on CSR propagation along the supply chain using the following two models.

K
CSRS ProductScore(t + 1 ) = a0 + a1 CSRC ProductScore(t ) × Event + a2 CSRC ProductScore(t ) + a3 Event + bk Xk (t ) + FE(t ) + (t + 1 ).
k=1

K
RepRisk (t + 1 ) = a0 + a1 RepRisk (t ) × Month + a2 RepRisk (t ) + a3 Month + bk Xk (t ) + FE(t ) + (t + 1 ).
C S S
k=1
In the first model, we regress the supplier’s product responsibility rating (Product score) on its customer’s counterpart, the event year (Event; i.e., the 2008
food safety scandal in China and the 2013 Takata airbag and Toyota car recalls), and the interaction between Event and customer CSR product score. In the
second model, we regress the customer’s reputation risk index (RepRisk) on its supplier’s counterpart associated with the product safety scandals, Month
(i.e., the month of the year when the scandal was announced), and the interaction between Month and RepRisk. Unreported control variables Xk include the
supplier’s leverage, return on assets (ROA), Tobin’s Q, log of total assets (ln(TAssets)), % of shares held by institutional investors and blockholders/insiders
(Inst/Close holdings), sales growth, and ln(GDPC). All the variables are defined in Appendix Table A.3. NObs is the number of customer-supplier (CS) pair
observations. In Columns 1–2, the frequency is annually, whereas in Columns 3–4, the frequency is monthly. All regressions also include intercepts and
CS-industry, CS-country, and year fixed effects (FE), and robust standard errors reported in parentheses are clustered at the customer-supplier-pair level. ∗ ,
∗∗ ∗∗∗
, are significance levels denoted at the 10%, 5%, and 1% levels, respectively.
CSRS (t + 1 ) RepRiskC (t + 1 )
Food safety Airbag & car Food safety Airbag & car
scandal recalls scandal recalls
Variable (1) (2) (3) (4)
Product responsibilityC x 2008 0.250∗

(0.145)
Product responsibilityC x 2013 0.268∗ ∗
(0.122)
Product responsibilityC -0.026 -0.082
(0.057) (0.076)
RepRiskS x Aug 2008 0.080∗ ∗
(0.036)
RepRiskS x April 2013 0.112∗ ∗ ∗
(0.018)
RepRiskS 0.016 -0.072∗ ∗ ∗
(0.012) (0.011)
NObs (CS-Years) 753 423
NObs (CS-Months) 32,821 88,653
Year FE Yes Yes Yes Yes
CS-Industry FE Yes Yes Yes Yes
CS-Country FE Yes Yes Yes Yes
demands for responsible business practices are greater Liang and Renneboog, who measure the average effect
following major product safety scandals. five years after the event. Overall, the evidence lends
We also compare our results on the 2008 Chinese further credence to our causal inferences of our base-
milk scandal with that of Liang and Renneboog (2017), line finding (based on cross-sectional analyses) that the
whose study also includes this particular scandal. The two stakeholder orientation of CSR is only from customers to
authors investigate firms’ reactions to product safety issues suppliers.
across legal regimes, whereas ours focuses on whether Oftentimes, scandals hurt not only the stock price,
customers respond to such issues by forcing suppliers sales, and brand reputation of the corporation in question
to comply with higher CSR standards. Specifically, they but also those of its customers. RepRisk, a leading research
find that firms in civil law countries react more strongly and business intelligence provider which specializes in
towards the Chinese milk scandal than firms in common ESG and responsible business conduct risks, evaluates
law countries by about 5% (see their Table 8, Panel A, Col- firm-level risk exposure to ESG issues and dynamically
umn 1, p. 884) post 2009 (i.e., 2009–2014). As discussed quantifies reputation risk exposure pertaining to all ESG
above, the economic magnitude associated with the 2008 scandals to construct a firm’s reputation risk (RepRisk)
Chinese milk scandal is a 10.9% rise in the supplier’s mean index at the monthly frequency. Based on the firm-level
product responsibility rating. In other words, the supplier’s RepRisk index, we graph in Fig. 3 the RepRisk of both sup-
average product responsibility rating will increase from pliers and their customers following all ESG scandals that
57.51 to 63.78 in the following year. Our results therefore occur at t = 0, identified by RepRisk. The graphs provide
suggest that the customers must have compelled their anecdotal evidence that ESG scandals not only increase
suppliers to react immediately to such a global scandal, the suppliers’ own reputation risk index but also their
as their resulting reaction in improved CSR performance customers’. To formally examine the effect of suppliers’
is about twice the size of the average effect shown in scandals on their customers’ reputation risk, we replicate
JID: FINEC
the regression model (4) by replacing Product score by and a customer no longer report their relationship which
RepRisk index as follows. has previously existed in the sample period, we consider
the chain to be delinked. It is important to highlight
RepRiskC (t + 1 ) = a0 + a1 RepRiskS (t ) × Month
that customer-supplier pairs that do not experience a
+ a2 RepRiskS (t ) + a3 Month change in the relationship during our sample period (i.e.,

K always linked or never linked firms) do not enter into
+ bk Xk (t ) + FE(t ) + (t + 1 ). (5) the sample. Understandably, two firms being first linked
k=1 or first delinked in a supply chain relationship can be
Specifically, in (5), we test whether the reputation risk of due to either endogenous reasons (such as selection and
suppliers associated with the above two product safety matching) or exogenous reasons such as the supply chain
scandals that occurred in a specific month (Month) of the is interrupted by external forces.
scandal year affects the reputation risk of customers. Re- To capture the change-status effect, we use two binary
sults from Columns 3–4 of Table 6 suggest strong evidence indicators, namely, Post-link and Post-delink. Post-link
of a positive association between the supplier’s RepRisk (delink) is a binary indicator that equals one if the
index and the customer’s next-month RepRisk index. The customer and supplier first establish (sever) the rela-
transmission of RepRisk from the supplier to the customer tionship and zero otherwise. We then run the following
provides a strong incentive for the customer to strengthen regression.
its influence on its supplier’s CSR when a global ESG
scandal floods worldwide news media. Our unreported CSRS Score(t + 1 ) = a0 + a1 CSRC Score(t )
results show that consistent with the graph shown in Fig. × Post Link/Delink(t ) + a2 CSRC Score(t )
3, the RepRisk transmission effect fades away three to five
K
months after the scandal broke, an indication that our + a3 Post Link/Delink(t ) + bk Xk (t )
finding is scandal-specific rather than a result of increasing k=1
CSR over time. + FE(t ) + (t + 1 ). (6)
Collectively, all the above tests bolster a causal in-
terpretation of our baseline evidence that customer CSR In model (6), the key independent variable is the inter-
exhibits a significant impact on the subsequent supplier action between CSRC score and Post-link (or Post-delink).
CSR. We now turn to analyzing the potential mechanisms Results are shown in Table 8. We find that in newly
by which corporate customers can exert influence. established economic links, the customer CSR is positively
associated with the subsequent supplier CSR. The estimate
5. The mechanisms of a1 is positive and statistically significant at the 1%
level after the customer and supplier firms establish an
One key mechanism by which customers might influ- economic relationship. Column 1 shows the overall cus-
ence a supplier’s CSR is through assortative matching. For tomer CSR effect of 0.013 (=0.020–0.007) on supplier CSR,
example, customers with high CSR are more likely to se- suggesting some selection associated with the initial es-
lect suppliers who will meet their expectations of socially tablishment of the relationship. On the other hand, when
responsible behavior. Alternatively, it is also possible that the supply chain is terminated, the customer no longer
the customers actively push suppliers to get more involved exhibits any significant influence on suppliers during the
in CSR and embrace it. In this section, we examine these post-delink period. The interaction term of CSRC score and
two possible avenues. Post-delink (Column 2) is statistically insignificant in the
post-delinked period, even though Post-delink itself is still
5.1. Assortative matching between customers and suppliers significant at conventional levels. If the assortative match-
ing is the channel through which customers facilitate CSR
A positive assortative matching between customers and propagation along the supply chain, a severance of the
suppliers is a possible channel when customers seek to relationship between the customer and its supplier does
form a relationship with suppliers that have the propen- not imply a weakened CSR spillover effect, unless such a
sity to engage in similar CSR practices as theirs. On the severance is exogenous. In other words, socially conscious
other hand, the customers may sever their relationships customers choose firms with high CSR potential to form a
with suppliers when the latter fail to meet the customers’ supply chain relationship, and dissolve the relationship if
CSR standards (e.g., Banerjee et al., 2008). To test this the supplier is constrained by its ability to further improve
potential matching or selection channel, we compare two its CSR standards.
cases of the formation/severance of customer-supplier While the baseline evidence could be driven by cus-
relationships—one is due largely to potentially endogenous tomers having preferences for suppliers who would be
reasons and the other results mostly from exogenous willing to assume greater corporate social responsibilities,
shocks. our unreported results suggest that the customer effect
First, we construct a sample of customer-supplier of CSR still holds even after we remove the observations
pairs that have experienced a change in relationship at when customers and suppliers first establish their rela-
any point in our entire sample period. For example, if a tionship, indicating that there is an additional effect that
supplier (customer) reports a relationship with a customer goes beyond pure selection.
(supplier) for the first time in the sample period, we Second, we focus on target firms of M&As as a relatively
classify the two firms as linked. Alternatively, if a supplier exogenous reason for the termination of customer-supplier
JID: FINEC
relationships. The reason why we examine only targets, Table 7

Linked and delinked customer-supplier relationships.
and not acquirers, is that target firms are typically not
This table reports results from the regression of supplier CSR score on
being acquired by their own choices. For example, when customer CSR score, linked/delinked indicator, and their interaction as fol-
a supplier is acquired by a third party, the economic link lows.
between the supplier and its existing customer could
CSRS Score(t + 1 ) = a0 + a1 CSRC Score(t ) × PostLink/Delink(t )
be weakened as there is no economic link between the
+ a2 CSRC Score(t ) + a3 PostLink/Delink(t )
supplier and acquirer prior to the acquisition. Such an
acquisition would be different from an M&A where a
K
+ bk Xk (t ) + FE(t ) + (t + 1 ).
customer (i.e., the acquirer) intentionally merges with k=1
the supplier to establish an upstream integration. Simi-
The dependent variable is supplier CSR score (CSRS score) and the key
larly, when a customer is targeted by a supplier with no independent variables are the customer CSR score (CSRC score), Post-
prior link with the customer, the customer’s relationship link/delink variable, and their interaction. Post-link (delink) is a binary
with its existing supplier might diminish following the variable that equals one if the customer and supplier first establishes
acquisition. (severs) the relationship and zero otherwise. Unreported control variables
Xk include supplier firm characteristics, namely, leverage, return on assets
Our sample contains 839 customer or supplier firms (ROA), Tobin’s Q, log of total assets (ln(TAssets)), % of shares held by in-
with ASSET4 ratings that are M&A targets and eventually stitutional investors and blockholders/insiders (Inst/Close holdings), sales
acquired by another firm. To implement our test, we define growth, as well as country’s gross domestic product per capita (ln(GDPC)).
a binary indicator, Post M&A, which captures the supply All the variables are defined in Appendix Table A.3. NObs is the number
of customer-supplier pair observations. The regressions also include inter-
chain relationship following the completed acquisition
cepts and customer-supplier (CS) industry, country, and year fixed effects
of a target supplier or customer, and run the following (FE), and robust standard errors reported in parentheses are clustered at
regression. the customer-supplier level. ∗ , ∗ ∗ , ∗ ∗ ∗ are significance levels denoted at the
10%, 5%, and 1% levels, respectively.
CSRS Score(t + 1 ) = a0 + a1 CSRC Score(t )
Variable (1) (2)
× Post M&AS(or C ) (t ) + a2 CSRC Score(t ) C ∗∗∗
CSR score x Post-link 0.020
S (or C )

K (0.004)
+ a3 Post M&A (t ) + bk Xk (t ) + FE(t ) + (t + 1 ). Post-link 0.209
k=1 (0.275)
CSRC score x Post-delink 0.001
(7) (0.004)
Post-delink 0.808∗ ∗ ∗
In columns 1–2 of Table 8, we estimate the impact of (0.296)
all acquisitions on the supplier’s subsequent CSR perfor- CSRC score -0.007∗ 0.001
mance. Columns 3–4 and 5–6 of the table show separate (0.003) (0.003)
results of suppliers as targets and customers as targets, NObs 233,881 233,881
respectively. Columns 2, 4, and 6 report the effect of cus- Controls Yes Yes
Year FE Yes Yes
tomers whose CSR ratings are above the median level of
CS-Industry FE Yes Yes
the sample customer CSR rankings. If a customer or sup- CS-Country FE Yes Yes
plier is targeted and then acquired by a third-party firm
with no associated economic link, the CSR propagation
effect ought to become weaker following the acquisition.
For instance, in Column 1, the combined customer’s CSR
effect on the supplier’s next-period CSR score is -0.038 delinks do not mitigate the customer effect of CSR on the
(= −0.051 + 0.013). A closer analysis suggests that the supplier’s CSR (i.e., the interaction term bears an insignif-
negative effect is driven mainly by acquisitions of suppli- icant coefficient, a1 ). In contrast, when the severance of a
ers. As seen in Columns 3–4, when a supplier is the target, relationship is more likely due to exogenous reasons (i.e.,
its subsequent CSR practices become negatively associated acquisitions by a third party), the propagation effect is sig-
with its customer’s, especially when the latter is above nificantly reduced. This potentially points to the presence
the median level. It is plausible that the supplier pays less of a selection process in explaining the CSR spillover along
attention to its CSR practices during and immediately after the supply chain.
the acquisition process. Alternatively, this may imply that
the supplier’s acquirer is unwilling to expend more re- 5.2. How corporate customers actively influence suppliers’
sources to maintain the sustainability of the supplier’s CSR CSR activities
standard, or that the supply chain is severed following the
acquisition. In contrast, when a customer is the target, the As stakeholders–governments, legislators, consumers,
acquisition has no bearing on the supplier’s subsequent employees, activist groups–intensify their expectations of
CSR performance. The implication is that such acquisitions corporate responsibility to society, such demands shed
do not break down the supply chain and hence, display no light on the influence of corporate customers towards the
effect on the existing CSR collaboration. CSR practices of their suppliers. Our analysis evaluates
Taken together, the results of Tables 7 and 8 suggest two ways that corporate customers can push suppliers
more concrete inferences on the selection or matching to act; one is through bargaining power and another is
mechanism. When two firms are economically delinked through directly influencing the decision-making process
due to either exogenous or endogenous reasons, such of suppliers.
JID: FINEC
Table 8
Stakeholder effects of CSR in target supplier and customer firms.
This table reports results from the regression of supplier CSR score on customer CSR score, binary indicator for post-M&A event of either a customer or
a supplier being the target firm, and the interaction between post-M&A indicator (Post M&A) and customer CSR score as follows.

K
CSRS Score(t + 1 ) = a0 + a1 CSRC Score(t ) × Post M&AS (or C ) (t ) + a2 CSRC Score(t ) + a3 Post M&AS (or C ) (t ) + bk Xk (t ) + FE(t ) + (t + 1 ).
k=1
The dependent variable is supplier CSR score (CSRS score) and the key independent variables are customer CSR score (CSRC score) and its interaction with
the Post M&A indicator when a customer or a supplier is the target. The table also shows results when evaluating a customer whose CSR is above the
median rating (CSRCHigh ) in the sample of customer firms. Unreported control variables Xk include supplier firm characteristics, namely, leverage, return
on assets (ROA), Tobin’s Q, log of total assets (ln(TAssets)), % of shares held by institutional investors and blockholders/insiders (Inst/Close holdings), sales
growth, as well as its country’s gross domestic product per capita (ln(GDPC)). All the variables are defined in Appendix Table A.3. NObs is the number of
customer-supplier (CS) pair observations. The regressions also include intercepts and customer-supplier (CS) industry, country, and year fixed effects (FE),
and robust standard errors reported in parentheses are clustered at the customer-supplier-pair level. ∗ , ∗ ∗ , ∗ ∗ ∗ are significance levels denoted at the 10%,
5%, and 1% levels, respectively.
Target=Supplier/Customer Target=Supplier Target=Customer
Variable (1) (2) (3) (4) (5) (6)

∗∗ ∗∗
CSR score × Post M&A
C
-0.051 -0.075 -0.028
(0.023) (0.032) (0.032)
C
CSR score 0.013∗ ∗ 0.013∗ ∗ 0.012∗ ∗
(0.006) (0.006) (0.006)
CSRCHigh x Post M&A -1.595∗ -2.307∗ ∗ -0.038
(0.896) (1.153) (1.419)
CSRCHigh 0.382∗ 0.372∗ 0.336
(0.219) (0.218) (0.218)
Post M&A 4.492∗ ∗ 1.514∗ 7.894∗ ∗ ∗ 3.381∗ ∗ ∗ 1.603 -0.504
(1.860) (0.830) (2.482) (1.044) (2.639) (1.308)
NObs 52,555 52,555 52,555 52,555 52,555 52,555
Controls Yes Yes Yes Yes Yes Yes
YearFE Yes Yes Yes Yes Yes Yes
CS-Industry FE Yes Yes Yes Yes Yes Yes
CS-Country FE Yes Yes Yes Yes Yes Yes
5.2.1. Degree of stakeholder bargaining power then conduct the following regression.
The extent to which customers can exert considerable
CSRS Score(t + 1 ) = a0 + a1 CSRC Score(t )
pressure on suppliers and demand for improved CSR
depends on the customers’ bargaining power. We argue × InnovS (t ) + a2 CSRC Score(t ) + a3 InnovS (t )
that when customers have low bargaining power, suppliers
K
can afford to withhold and avoid incurring the cost of + bk Xk (t ) + FE(t ) + (t + 1 ). (8)
concession to meet the customers’ demand for better k=1
CSR practices. To test this bargaining power channel of In (8), we evaluate the impact of a supplier’s innova-
CSR effect along the supply chain, we look at suppliers’ tion (InnovS ) on the supplier’s CSR score,28 conditional
investments specific to customers as well as the intensity on its customer’s CSR. We expect the coefficient of the
of industry competition for customers and suppliers. interaction between InnovS and the customer’s CSR score
(i.e., CSRC score) to be negative if the customer relies
(a) Relationship-specific investments heavily on the supplier’s innovation. In other words, when
Existing studies have established that customers from suppliers make large investments specific to customers’
research-intensive industries tend to be involved in needs, they would be in a better bargaining position to
specialized inputs that require their suppliers to make decide whether or not to align their CSR practices with
transaction-specific investments, consistent with their those of their customers. The results reported in Columns
own investments (e.g., Armour and Teece, 1980; Levy, 1–2 of Table 9 are consistent with our expectations.
1985; Allen and Phillips, 20 0 0; Dhaliwal et al., 2016; Chu We find that the interaction effect of CSRC score and
et al., 2019). Extending this strand of research, we argue InnovS is negative and statistically significant for both
that suppliers with more investments in innovation are proxies of the customer-specific investment. For instance,
more likely to engage in customer-specific investments. the coefficient of CSRC score x R&DS is -0.323 and is
We therefore hypothesize that the greater the supplier’s statistically different from zero. This finding suggests that
innovation capacity, the customers would have less power when a supplier’s customer-specific investment is low, the
to influence their suppliers to be more socially responsible. supplier is more inclined to meet the CSR standard of its
Following Chu et al. (2019), we employ the amount of
R&D relative to total assets and the log of the number of
28
We have also examined whether a customer’s innovation activity has
registered patents as proxies for a supplier firm’s invest-
any effect on a supplier’s CSR performance. Unreported results suggest
ments in innovation specific to the customer’s needs. We that investment in innovation does not give customers strong bargaining
power to drive CSR in suppliers.
JID: FINEC
Table 9 less of a need to learn from their customers. Hence, in

The effect of bargaining power in the supply chain and supplier CSR.
line with the evidence, it would be harder for customers
This table reports results from the regression of supplier CSR score
(CSRS score) on customer CSR score (CSRC score), a measure of the bar- to influence highly innovative suppliers.
gaining power (Power) in the customer-supplier relationship, and the in-
teraction between Power and CSRC score as follows. (b) Intensity of industry competition
CSR Score(t + 1 ) = a0 + a1 CSR Score(t ) × Power(t )
S C The intensity of competition in an industry is one
determining force that drives the bargaining power of
+ a2 CSRC Score(t ) + a3 Power(t )
suppliers and customers. A supplier has strong bargaining

K
+ bk Xk (t ) + FE(t ) + (t + 1 ). power if there are barriers to entry, fewer threats of
k=1 substitutes, the industry is highly concentrated or low in-
Our analysis examines the degree of bargaining power in a supply chain tensity of industry rivalry, and customers have weak or no
through the supplier’s level of innovation, measured by its R&D to to- power. Conversely, a customer has high bargaining power
tal assets and the log of the number of patents, in Columns 1–2, and if it has fewer competitors, substitutes are available, little
the industry competitiveness of suppliers and customers, as measured
product differentiation, and its purchases comprise a large
by the Herfindahl-Hirschman Index (HHI), in Columns 3–4.The key in-
dependent variables are CSRC score, the proxy for Power, and their in-
portion of the supplier’s sales. Thus, the degree to which
teraction. Unreported control variables Xk include supplier firm charac- customers are able to pressure suppliers to achieve high
teristics, namely, leverage, return on assets (ROA), Tobin’s Q, log of total CSR standards is determined by the intensity of industry
assets (ln(TAssets)), % of shares held by institutional investors and block- competition of the supplier and customer. We measure the
holders/insiders (Inst/Close holdings), sales growth, as well as country’s
intensity of competition using the Herfindahl-Hirschman
gross domestic product per capita (ln(GDPC)). All the variables are defined
in Appendix Table A.3. NObs is the number of customer-supplier (CS) Index (HHI). A customer’s HHI is measured by the sum-
pair observations. The regressions also include intercepts, CS-industry, CS- mation of squared market share (based on sales) of each
country, and year fixed effects (FE), and robust standard errors reported firm within the same industry, whereas a supplier’s HHI is
in parentheses are clustered at the customer-supplier-pair level. ∗ , ∗ ∗ , ∗ ∗ ∗
measured within the same global industry to approximate
are significance levels denoted at the 10%, 5%, and 1% levels, respectively.
the level of competition suppliers encounter globally by
Variable (1) (2) (3) (4) squaring the market share of the firm’s sales in the indus-
C
CSR score x R&D S
-0.323 ∗∗∗ try and then summing the squares.29 We then estimate
(0.114) (8) by replacing the innovation variable with HHI and
R&D S
65.482∗ ∗ ∗ present the results in Columns 3–4 of Table 9.
(9.280)
Column 3 evaluates the effect of a supplier’s HHIS on
CSRC score x PatentsS -0.013∗
(0.008) supplier CSRS score, whereas Column 4 assesses the im-
PatentsS 0.795 pact of a customer’s HHIC . When the supplier is in a highly
(0.737) competitive industry, it tends to align its CSR practices
CSRC score x HHIS -0.567∗ ∗ ∗
with that of its customer. The coefficient of the interaction
(0.191)
HHIS 32.811∗ ∗
variable, CSRC score x HHIS is negative and statistically
(13.939) significant at the 1% level. This finding suggests that the
CSRC score x HHIC 0.048∗ ∗ supplier may lose its customer if it does not improve its
(0.023) CSR standard. On the other hand, when a customer is in
HHIC -4.235∗ ∗
a less competitive industry, it has strong bargaining power
(2.004)
CSRC score 0.037∗ ∗ ∗ 0.042 0.034∗ ∗ ∗ 0.010 to influence its suppliers to be more socially responsible.
(0.009) (0.042) (0.007) (0.007) The coefficient of CSRC score x HHIC is positive and sta-
NObs 35,493 1,364 55,682 52,552 tistically significant at the 5% level (i.e., 0.048 with robust
Controls Yes Yes Yes Yes standard error of 0.023; Column 4). Combined, these re-
YearFE Yes Yes Yes Yes sults suggest that the stakeholder bargaining power plays
CS-Industry FE Yes Yes Yes Yes
a critical role in propagating CSR along the supply chain.
CS-Country FE Yes Yes Yes Yes
5.3. Influencing the decision-making process of suppliers
customer. Alternatively, when there is less resource depen-

Extant literature has shown that corporate behavior
dence in the customer-supplier relationship, the customer
propagates through networks of decision-making bodies,
would have more power to demand a higher CSR standard
namely, owners and directors. Common ownership exists
in its supplier. Thus, the extent of the customer-specific
between two firms whenever an investor owns shares
investment is one mechanism that drives the supplier’s
of both firms, and two firms are board-linked whenever
desire to align its CSR with that of its more socially
they have shared directors on their boards. Such network
responsible customer.
structures are one potential mechanism for spreading
We recognize that our findings are also consistent with
corporate policies from firm to firm. This subsection exam-
a learning channel. A significant part of CSR initiatives
ines whether shared ownership and interlocking corporate
relates to product innovation and production process by
boards affect CSR efforts along supply chains.
using environmentally friendly technologies and engaging
in R&D that enhances product safety and responsibility.
Such processes involve learning from customers and feed- 29
An industry is defined using Fama and French’s 48 industry classifi-
back to suppliers; more innovative suppliers would have cations.
JID: FINEC
Table 10
Common institutional ownership and supplier CSR.
This table reports results from the regression of supplier CSR score on customer CSR score, common institutional ownership (ComInst), and the interaction
between ComInst and customer CSR score as follows.

K
CSRS Score(t + 1 ) = a0 + a1 CSRC Score(t ) × ComInstS(or C ) (t ) + a2 CSRC Score(t ) + a3 ComInstS(or C ) (t ) + bk Xk (t ) + FE(t ) + (t + 1 ).
k=1
The dependent variable is supplier CSR score (CSRS score) and the key independent variables are the customer CSR score (CSRC score) and common
ownership variable. ComInst is measured using the percentage of shares held by common institutional investors in a supplier or customer firm (%Shares
ComInst), or the log number of institutional investors holding both the supplier and customer firms (#ComInst). In Columns 1–3, the estimates include all
existing common ownership in both supplier and customer firms. In Columns 4–5, the specifications only consider institutional owners of a customer then
owns stocks of its supplier for the first time. Unreported control variables Xk include supplier firm characteristics, namely, leverage, return on assets (ROA),
Tobin’s Q, log of total assets (ln(TAssets)), % of shares held by institutional investors and blockholders/insiders (Inst/Close holdings), sales growth, as well
as country’s gross domestic product per capita (ln(GDPC)). All the variables are defined in Appendix Table A.3. NObs is the number of customer-supplier
(CS) pair observations. The regressions also include intercepts, CS-industry, CS-country, and year fixed effects (FE), and robust standard errors reported in
parentheses are clustered at the customer-supplier-pair level. ∗ , ∗ ∗ , ∗ ∗ ∗ are significance levels denoted at the 10%, 5%, and 1% levels, respectively.
Existing cross-holdings First-time holdings
Variable (1) (2) (3) (4) (5)

C S ∗∗ ∗∗∗
CSR score x %Shares ComInst 0.033 0.383
(0.015) (0.117)
%Shares ComInstS 3.421∗ ∗ ∗ -1.205
(1.244) (6.702)
CSRC score x %Shares ComInstC 0.035∗ ∗
(0.014)
%Shares ComInstC 4.160∗ ∗ ∗
(1.078)
CSRC score x #ComInst 0.068∗ ∗ ∗ 0.033∗
(0.024) (0.017)
#ComInst 0.040∗ ∗ 5.341∗ ∗ ∗
(0.018) (1.297)
CSRC score -0.010 -0.004 -0.026 0.038∗ ∗ ∗ 0.027∗ ∗ ∗
(0.008) (0.008) (0.018) (0.006) (0.009)
(0.008) (0.008) (0.018)
NObs 52,555 52,555 55,685 51,745 51,745
Controls Yes Yes Yes Yes Yes
Year FE Yes Yes Yes Yes Yes
CS-Industry FE Yes Yes Yes Yes Yes
CS-Country FE Yes Yes Yes Yes Yes
5.3.1. Common ownership invest in both firms in a given year and is represented
There is substantial robust evidence that institutional by “#ComInst.” Using these measures, we estimate the
investors, who are large equityholders, play a critical following model.
role in corporate policies of the firms they invest in. CSRS Score(t + 1 ) = a0 + a1 CSRC Score(t )
In recent years, several studies find that institutional
cross-ownership influences the outcomes of mergers and × ComInstS(or C ) (t ) + a2 CSRC Score(t ) + a3 ComInstS(or C ) (t )
acquisitions (e.g., Matvos and Ostrovsky, 2008; Harford
K
et al., 2011), industry competitiveness (e.g., Azar et al., + bk Xk (t ) + FE(t ) + (t + 1 ), (9)

2016; Azar et al., 2018; He and Huang, 2017), return corre- k=1
lations (Antòn and Polk, 2014), chief executive officer (CEO) where ComInst is either %Shares ComInst or #ComInst
pay incentives (Antòn et al., 2018), and customer-supplier in a supplier or customer firm. If common institutional
links (Freeman, 2017). In particular, Freeman shows that investors can influence their supplier firm to engage in
common institutional ownership in both the customer and socially responsible practices as their customer firm does,
supplier strengthens their supply chain relationship. Based we expect the interaction between CSR score and mea-
on her finding, it is plausible that common ownership is a sures of common institutional ownership to be positively
channel through which a supplier’s CSR is aligned with its correlated. Results of model (9) are reported in Columns
customer’s. 1–3 of Table 10.
Our analysis employs two different measures of com- The table reveals distinct evidence—the important role
mon institutional ownership. Our first measure computes of common institutional investors in aligning the CSR of
the maximum percentage of ownership held by all com- their portfolio of customer and supplier firms. We find
mon owners in the supplier (customer) in a given year, that the interaction of the two measures of common
and this measure is labeled, “%Shares ComInstS .” We institutional ownership with customer CSRC score yields
compute the same for the customer firm and label it positive and statistically significant effects on the next-
“%Shares ComInstC .” The other measure is based on the period’s supplier CSRS score. For example, in Column 1, the
log of the number of common institutional investors that interaction term, CSRC Score × %Shares ComInstS is 0.033
JID: FINEC
Table 11
Common directors and supplier CSR.
This table reports results from the regression of supplier CSR score on customer CSR score, common directors (CDirectors), and the interaction between
CDirectors and customer CSR score as follows.

K
CSRS Score(t + 1 ) = a0 + a1 CSRC Score(t ) × CDirectors(t ) + a2 CSRC Score(t ) + a3 CDirectors(t ) + bk Xk (t ) + FE(t ) + (t + 1 ).
k=1
The dependent variable is supplier CSR score (CSRS score) and the key independent variables are the customer CSR score (CSRC score), CDirectors variable,
and their interaction. CDirectors is measured using the log number of directors who serve on the boards of both the supplier and the customer, or the
log number of common directors holding multiple positions in the supplier firm. Columns 1–2 look at the sample of directors who already have positions
in both the supplier and customer firms. Column 3 tests the effect of a board member from the customer who later serves on the supplier board for
the first time. Unreported control variables Xk include supplier firm characteristics, namely, leverage, return on assets (ROA), Tobin’s Q, log of total assets
(ln(TAssets)), % of shares held by institutional investors and blockholders/insiders (Inst/Close holdings), sales growth, as well as country’s gross domestic
product per capita (ln(GDPC)). All the variables are defined in Appendix Table A.3. NObs is the number of customer-supplier (CS) pair observations. The
regressions also include intercepts, CS-industry, CS-country, and year fixed effects (FE), and robust standard errors reported in parentheses are clustered at
the customer-supplier-pair level. ∗ , ∗ ∗ , ∗ ∗ ∗ are significance levels denoted at the 10%, 5%, and 1% levels, respectively.
Existing appointments First-time appointment
Variable (1) (2) (3)

∗∗∗
C
CSR score x # of Common directors 0.035 0.095∗ ∗ ∗
(0.010) (0.029)
# of Common directors -2.453∗ ∗ ∗ -6.356∗ ∗ ∗
(0.687) (2.335)
CSRC score x # of Board positions 0.028∗ ∗ ∗
(0.008)
# of Common positions -1.989∗ ∗ ∗
(0.569)
CSRC score 0.036∗ ∗ ∗ 0.036∗ ∗ ∗ 0.035∗ ∗ ∗
(0.006) (0.006) (0.006)
NObs 55,642 55,642 55,642
Controls Yes Yes Yes
Year FE Yes Yes Yes
CS-Industry FE Yes Yes Yes
CS-Country FE Yes Yes Yes
with robust standard error of 0.015. Column 3 produces and customer’s boards. To cite an example of interlocking
a qualitatively similar finding based on the log number of directorates, in year 2006, Rudy Provoost was an Executive
common institutional investors in both the supplier and Vice President, division CEO, and board member of Philips,
customer firms. The coefficient of the interaction term a Dutch technology company headquartered in Amster-
is 0.068 and is statistically significant at the 1% level. dam, and at the same time, he was the head of the board
Thus, common institutional investors do seek to mobilize of directors (Chairman) at Philips’s supplier, LG Display
investor voice towards positive social impact along the Company. An alternative measure is the log number of po-
supply chain. sitions held by common board members. For example, in
In Columns 4–5, we test the effect of an institutional 2003, Garo Armen was CEO and Chairman of Agenus, Inc.,
investor who has an existing stake in the customer but a Lexington, Massachusetts-based biotechnology company
later also holds a stake for the first time in the already focused on immunotherapy, and was also Board Chair
linked supplier firm. The effect seems to be much stronger and Acting CEO of Elan Corp, Agenus’s supplier located
for the first-time holding, especially when common own- in Dublin, Ireland. In this case, Armen held two positions
ership is measured by the % of shares held; the coefficient on the supplier’s board of directors. Results are presented
of CSRC score x %Shares ComInst increases from 0.033 in in Table 11. Consistent with the evidence of Table 10,
Column 1 to 0.383 in Column 4. These findings suggest the results also suggest that common directors serve as
that common ownership paves the way for the subsequent another channel through which the customer firm is able
CSR propagation along the supply chain. to influence better CSR practices at its supplier firms. The
coefficients of both interaction terms (CSRC score × #
5.3.2. Board interlocks of Common directors and CSRC score × # of Board
Economically linked firms can also be connected positions) in Columns 1 and 2 are positive and statistically
through shared directors, where directors serve on boards significant at the 1% level.
of both the customer and supplier. One advantage of We also investigate the impact of the customer’s di-
having shared directors is that directors can act in concert rectors who subsequently serve on the supplier’s board
to promote similar CSR practices at both corporations. We for the first time. As shown in Column 3 of Table 11, the
therefore examine whether board connections through estimate of the interaction term in question is positive
shared directors can, in part, influence CSR policies. To and statistically significant at the 1% level. In fact, the
evaluate this channel, we construct two measures of com- sensitivity of the first-time board effect is stronger in
mon directors. The first measure is based on the log of Column 3, compared to that in Column 1, suggesting
the number of directors who serve on both the supplier’s that board members help customers to push suppliers to
JID: FINEC
implement better CSR policies. Our analysis, however, does consequences of the stakeholder (customer) effect of CSR
not extend to evaluate a first-time board member who on suppliers and not of firms’ own CSR activities.
will hold multiple positions on the supplier’s board. Our We postulate that collaborative CSR efforts between
sample suggests that it is rarely the case that a director customers and suppliers would lead to increased operating
who newly comes on board holds more than one position. efficiency, sales growth, and firm value, through enhancing
Thus, we are unable to conduct such an analysis. branding and reputation effects, attracting more con-
Overall, the results provide corroborating evidence sumers, and generating greater sales. Improved CSR stan-
that the stakeholder bargaining position and collaboration dards along a supply chain may involve the increased focus
among common institutional owners and common direc- of product responsibility and safety, which in turn lower
tors are instrumental in propagating CSR practices from discretionary expenses, such as selling, general, and ad-
customers to suppliers. Such influences go beyond a pure ministrative expenses,30 through improved operational ef-
selection channel. ficiency. Customers and suppliers have desires to promote
a sustainable relationship, through greater CSR efforts, if
6. Economic consequences of customer effects of CSR such efforts produce better future sales growth and en-
hanced firm valuation for both the customer and supplier.
In the preceding sections, we have shown that cus- Our analysis uses the ratio of selling, general, and admin-
tomers have a positive impact on suppliers’ CSR practices, istrative expenses to total assets (SG&A), 3-year annualized
suggesting that suppliers do respond to their customers future sales growth, and market-to-book ratio as mea-
and behave similarly in socially responsible ways. However, sures of firm performance (Performance) associated with
a question that remains is whether there is any economic increased collaborative CSR efforts in a customer-supplier
benefit arising from customers pushing suppliers for relationship. To evaluate the economic benefits associated
greater social responsibilities. with such efforts, we run the following regression model.
Existing studies dispute whether the benefits of CSR
Performance ( ) (t + 1 ) = a0 + a1 CSRC Score(t − 1 )
S or C
outweigh its costs. Some studies find that CSR initiatives
can help firms build a social reputation (e.g., Fombrun × CSRS Score(t ) + a2 CSRC Score(t − 1 )
and Shanley, 1990), attract more productive employees
K
(Burbano, 2016), exploit new markets for environmentally + a3 CSRS Score(t ) + bk Xk (t ) + FE(t ) + (t + 1 ). (10)
friendly products (Arora and Gangopadhyay, 1995), and k=1
can be financially profitable through branding/reputation
The variable of interest is the interaction of CSRC score
effects on different stakeholders (Baron, 2001). Bénabou
(t − 1 ) and CSRS score(t) in (10). If the CSR effect on
and Tirole (2010) argue that CSR engagements are ben-
suppliers benefits both the supplier and customer, we
eficial to firms in the long run and help strengthen
should expect the coefficient of a1 to be negative for SG&A
their market positions. Other studies, however, show that
whereas positive for future sales growth and market-to-
adopting CSR policies is likely to increase costs and hurt
book ratio. The interpretation is that high supplier CSR
firm performance, as firms redefine their corporate social
leads to better firm performance only conditioned on
responsibilities under the pressure of various stakeholders.
the customer’s CSR being high. Such an interpretation
CSR costs include major investment costs involving con-
is consistent with our above finding of CSR propagation
struction, equipment, or new environmental technologies
along the supply chain, implying some degree of coop-
and processes, permanent contributions such as scholar-
erative or collaborative efforts between the supplier and
ships, and other operating costs of CSR implementation.
the customer. Estimates of (10) are shown in Table 12.
Margolis et al. (2010) provide a meta-analysis of the
The dependent variables are SG&A, sales growth, and
relationship between corporate social and financial perfor-
market-to-book ratio of the supplier in Columns 1, 3, and
mance, and document that the overall effect is positive but
5, while those of its customer counterparts are shown in
small.
Columns 2, 4, and 6.
Here we ask whether customers are financially incen-
We find that the stakeholder effect of CSR gener-
tivized to impose better CSR practices on their suppliers,
ates favorable economic outcomes. It therefore pays for
and whether suppliers benefit from taking greater social
customers to influence their suppliers to act socially re-
responsibilities. The economic implications of these collab-
sponsibly, as such behavior has an overall positive impact
orative CSR efforts between customers and suppliers are
on the customer’s future performance. Customers enjoy
closely related to the growing literature linking CSR to firm
not only improved operational efficiency in terms of lower
financial performance (e.g., Gillan et al., 2010; Edmans,
SG&A, but also greater future sales growth and, albeit
2011; Deng et al., 2013; Servaes and Tamayo, 2013; Flam-
small,31 firm valuation. For example, the a1 coefficient is
mer, 2015a; Krüeger, 2015; Lins et al., 2017). Most of these
negative in Column 2 but is positive in Columns 4 and 6.
studies consider CSR engagements as a firm’s own strate-
gic choice, and investigate their direct and indirect effects
on the firm’s profitability and valuation. Some recent 30
Kalwani and Narayandas (1995) show that maintaining long-term re-
work also studies the role of various stakeholders, such lationships with their customers decreases discretionary expenses, such
as selling, general, and administrative expenses, and hence, improves
as institutional investors (Chen et al., 2020; Dimson et al.,
profitability.
2015; Dyck et al., 2019) and competitors (Cao et al., 2019) 31
In their meta-analysis of several existing empirical studies, Margolis
in driving CSR engagements and their value implications. et al. (2010) conclude that the overall correlation between CSR and cor-
Unlike these studies, our analysis focuses on the economic porate firm performance is positive but small.
JID: FINEC
Table 12
Firm performance and customer effects of CSR.
This table reports results from regressing firm performance, measured by SG&A, 3-year annualized sales growth, and firm valuation on supplier CSR
score, customer CSR score, and the interaction between customer CSR score and supplier CSR score as follows.
K
Performance ( ) (t + 1 ) = a0 + a1 CSRC Score(t − 1 ) × CSRS Score(t ) + a2 CSRC Score(t − 1 ) + a3 CSRS Score(t ) + bk Xk (t ) + FE(t ) + (t + 1 ).
S or C
k=1
The dependent variable is the supplier’s performance (Performance) in Columns 1, 3, and 5, and of customer’s in Columns 2, 4, and 6, and key explanatory
variables include supplier CSR score (CSRS score), customer CSR score (CSRC score), and their interaction. Performance is measured by SG&A, 3-year an-
nualized sales growth, and market-to-book equity value. Unreported control variables Xk include supplier firm characteristics, namely, leverage, return on
assets (ROA), Tobin’s Q, total assets (ln(TAssets)), % of shares held by institutional investors and blockholders/insiders (Inst/Close holdings), sales growth, as
well as the country’s gross domestic product per capita (ln(GDPC)). All the variables are defined in Appendix Table A.3. NObs is the number of customer-
supplier (CS) pair observations. The regressions also include intercepts, CS-industry, country, and year fixed effects (FE), and robust standard errors reported
in parentheses are clustered at the customer-supplier-pair level. ∗ , ∗ ∗ , ∗ ∗ ∗ are significance levels denoted at the 10%, 5%, and 1% levels, respectively.
SG&A 3-Year annualized sales growth Market-to-book value
Supplier Customer Supplier Customer Supplier Customer

Variable (1) (2) (3) (4) (5) (6)
CSRC ∗ CSRS -0.0003∗ ∗ ∗ -0.0003∗ ∗ ∗ 0.021 0.029∗ 0.004∗ ∗ 0.0003∗

(0.0000) (0.0000) (0.149) (0.017) (0.001) (0.0001)
CSRS 0.0548∗ ∗ ∗ 0.0257∗ ∗ ∗ 0.006 0.025 0.270∗ ∗ -0.065∗ ∗ ∗
(0.00747) (0.00868) (0.122) (0.015) (0.121) (0.013)
CSRC 0.0235∗ ∗ ∗ 0.0341∗ ∗ ∗ 0.340∗ ∗ ∗ -0.021∗ -0.211∗ -0.014
(0.00771) (0.00720) (0.108) (0.012) (0.109) (0.014)
NObs 31,232 28,238 32,242 32,244 33,202 32,242
Supplier controls Yes No Yes No Yes No
Customer controls No Yes No Yes No Yes
Year FE Yes Yes Yes Yes Yes Yes
CS-Industry FE Yes Yes Yes Yes Yes Yes
CS-Country FE Yes Yes Yes Yes Yes Yes
Similarly, it is also worthwhile for suppliers to strive for also shows that in response to global shocks associated
better CSR standards that adhere to that of their socially with product-safety scandals, customers exert pressure to
responsible corporate customer—suppliers also experience accelerate their suppliers’ product responsibility practices.
decreased SG&A and enhanced firm value following their Our study offers new insights on the real effects of CSR
adoption of improved CSR practices. However, unlike their initiatives and actions.
customers, suppliers do not experience any statistically The evidence suggests that corporate customers tend
significant increase in 3-year annualized future sales to establish supply chain relationships with firms that are
growth, even though the sign of the coefficient is positive. inclined to engage in responsible social and environmen-
In summary, this section shows evidence of economic tal practices. Customers with greater bargaining power
benefits associated with the customer effect of CSR along and with their own directors and investors also having
the supply chain. stakes in supplier firms make them an influential voice
in decisions pertaining to suppliers’ responsible business
7. Conclusion operations.
Finally, the collaborative CSR efforts resulting from the
Many large corporate customers around the globe alignment of CSR standards deliver economic value to both
increasingly recognize the importance of integrating social suppliers and customers. Customers have incentives to
responsibility initiatives into their business model to build aim for better CSR from their suppliers as a higher CSR
a sustainable competitive advantage in the marketplace. standard results in improved operational efficiency, sales
However, the impact of their power as customers to growth, and firm value, possibly through socially and en-
drive improvements in responsible business practices vironmentally friendly production and through enhancing
through their global supply chains has not been widely branding and reputation effects. Similarly, suppliers also
studied. Our research exploits several unique international have the desire to engage in responsible business practices
databases to examine whether and how their large stakes that adhere to those of their customers, as such adherence
might give them an active role in suppliers’ CSR initiatives contributes to improving both operational efficiency and
and standards. firm valuation.
To investigate whether corporate customers affect What we document in this paper is not merely how
suppliers’ CSR, and if so, through which channels, we one firm’s CSR practices can affect another’s; in fact, the
employ a regression discontinuity design that depends on multiplier effect that we stress has significant policy impli-
exogenous variation in customers’ close-call CSR proposals cations. That is, increasing one firm’s CSR can potentially
that pass or fail by a small margin of votes. Passing a have a ripple effect across extensive global supply chains.
CSR proposal results in about a 7% improvement in the If we view the evidence of our study at face value, several
suppliers’ subsequent year’s CSR performance. Our analysis ideas emerge for the improvement of social welfare and
JID: FINEC
firm performance. As multinational corporations around Table A.2

the world are increasingly interconnected via global sup- Distribution of numbers of customers and suppliers, together with their
ply chains and making significant impacts on billions of mean CSR scores by country. This table shows the numbers of suppliers
and customers, as well as their average CSR (CSRS and CSRC , respectively)
consumers, their value creation process and social respon-
scores, by year.
sibilities have become foremost issues in public debate.
Despite their importance, we still have limited knowledge Number of Average
Suppliers Customers CSRS score CSRC score
on how their CSR practices are influenced by each other
and spilled over along supply chains. Therefore, a firm’s Country (1) (2) (3) (4)
own socially responsible practices might have a multiplier Australia 1580 2195 81.30 63.05
effect through economic links and generate bigger positive Austria 115 186 69.63 62.97
social payoffs. However, our findings of the unidirectional Belgium 178 193 76.61 72.43
Bermuda 107 303 21.95 22.40
CSR effect from customers to suppliers and only in some
Brazil 1338 768 70.17 59.98
subsamples of countries suggest that such effects are Canada 2450 3195 75.20 57.79
bounded by a firm’s relative position in the global network Chile 458 293 38.64 34.32
and its socio-cultural and institutional environment. This China 874 469 50.11 41.04
potentially indicates that policies aiming at promoting Colombia 61 18 80.45 74.22
Czech Republic 19 4 45.02 49.64
socially responsible practices among public companies
Denmark 246 238 75.24 72.71
cannot be universally applied. Egypt 6 2 4.86 5.46
Our findings also shed light on some fundamental is- Finland 437 760 90.39 90.99
sues in industrial organization and strategic management, France 3857 4500 88.55 83.90
Germany 3622 3061 82.36 73.41
such as why some firms are incentivized to coerce others
Greece 104 45 77.83 45.13
to adopt certain practices. The answer might simply be Hong Kong 871 639 60.75 62.45
that it makes economic sense because as corporate cus- Hungary 35 12 50.62 56.90
tomers, they can benefit from increased sales throughout India 713 398 76.58 74.54
the supply chain. It is therefore unsurprising that these Indonesia 348 244 58.05 58.63
Ireland 215 261 62.43 64.56
companies are also more responsive to product-safety
Israel 274 185 38.98 56.75
scandals, as these events are closely related to consumer Italy 811 503 82.80 88.58
perception and future purchases of their products. These Japan 5786 3997 70.73 67.26
are important strategic considerations for managers, es- Kuwait 10 2 75.13 24.99
pecially pertaining to indirect costs and benefits incurred Luxembourg 346 219 80.14 66.48
Malaysia 174 225 62.56 46.07
by other upstream and downstream firms, when trading Mexico 298 288 40.02 43.98
off their social investment and other capital expenditures Morocco 3 5 26.38 26.38
with limited corporate resources. Netherlands 1,665 1,224 88.33 79.34
New Zealand 135 113 59.06 26.50
Appendix A Norway 435 287 89.01 82.95
Peru 43 19 22.07 35.90
Philippines 101 61 49.10 47.32
Poland 172 267 46.75 26.23
Table A.1 Portugal 162 39 86.15 87.05
Distribution of the numbers of customers and suppliers, together with Qatar 13 5 7.12 9.45
their mean CSR scores, by year. Russia 554 434 51.06 51.25
This table shows numbers of suppliers and customers, as well as their Saudi Arabia 62 52 62.64 73.71
average CSR scores (CSRS and CSRC ), respectively, by year. Singapore 329 678 65.80 58.92
South Africa 591 651 82.01 67.96
Number of Average South Korea 2,235 1,903 68.35 64.44
Suppliers Customers CSRS score CSRC score Spain 799 630 90.64 87.32
Year (1) (2) (3) (4) Sweden 693 1,052 86.36 80.74
Switzerland 956 1,300 86.53 64.85
2003 1427 1410 52.69 62.83 Thailand 197 170 81.07 83.93
2004 2125 2097 58.86 71.69 Turkey 163 84 56.83 49.62
2005 2425 2396 62.40 74.90 United Arab Emirates 4 3 20.00 15.38
2006 2538 2523 60.31 75.17 United Kingdom 5,632 6,192 87.58 78.67
2007 2707 2675 63.66 75.14 United States of America 38,546 43,741 71.13 59.12
2008 3446 3351 62.59 72.91
2009 4690 4551 61.17 72.20
2010 6772 6534 66.06 73.52
2011 9622 9241 67.43 74.78
2012 9402 9961 66.94 75.12
2013 11,261 10,366 65.02 74.28
2014 11,271 10,492 65.52 73.43
2015 14,427 13,227 62.17 75.25
JID: FINEC
Table A.3
Variable definition and data source.
Variable Definition and data source
CSR component scores and RepRisk index

CSR score An equal-weighted CSR rating score (ASSET4)
Env A score associated with the environmental pillar of CSR Rating (ASSET4)
Soc A score associated with the social responsibility pillar of CSR Rating (ASSET4)
Product A score associated with the product responsibility pillar of CSR Rating (ASSET4)
RepRisk index A firm-level reputation risk index that assesses a firm’s risk exposure to ESG issues (RepRisk)
Mechanism variables
R&D Research and development expense (Worldscope item 01201)/Total assets (Worldscope item 02999)
Patents Natural logarithm value of number of successful patent applications filed plus one in a year (PATSTAT)
HHI Herfindahl-Hirschman Index measured by the summation of squared market share (based on sales) of each firm within
the same industry)(Datastream Worldscope)
%Shares ComInst The maximum % of ownership held by all common institutional owners in the supplier (or customer) in a given year
(FactSet Global Ownership Data)
#ComInst # of common owners who own shares of both the supplier and customer in a given year (FactSet Global Ownership
Data)
# of Common directors # of common directors who serve on the boards of both the supplier and customer in a given year (BoardEx)
# of Board positions # of board positions held by common directors at the supplier in a given year (BoardEx)
Control variables
Leverage Book value of debt (WorldScope item 03255) / Total assets (Worldscope item 02999)
ROA Earnings before interest and taxes (Worldscope item 18191)/Total assets (Worldscope item 02999)
Q Market value of common equity (Worldscope item 08001) + Total assets (Worldscope item 02999) - Book value of
common equity (Worldscope item 03501) and then divided by the net value of Total assets (Worldscope item 02999)
ln(TAssets) Natural logarithm value of total assets (Worldscope item 02999) plus one
Inst/Close holdings Percentage of outstanding shares owned by institutional investors and blockholders/insiders (Worldscope item
08021 + FactSet Global Ownership Data)
Sales growth One-year net sales growth (Worldscope item 08631)
ln(GDPC) Log ofgross domestic product in USD divided by domestic population (World Bank Indicators)
Performance variables
SG&A Selling, general and administrative expenses (Worldscope item 01101) divided by total assets (Worldscope item 02999)
3-Yearannualizedsales 3-year net sales growth (Worldscope item 01101)
growth
Market-to-book ratio Market value of common equity(Worldscope item 08001)divided by Book value of common equity (Worldscope item
03501)
CSR implementation definitions
News about We conduct a news search to determine whether firms actually implement their proposed CSR practices within 12
implementation months following the passage of the proposals.We first match our CSR-related proposals with news contained in
Ravenpack News Analytics, based on the following screening criteria: 1) news is announced within one calendar year
after the proposal date, and 2) news headlines and the text in the CSR proposal share any words that are not
prepositions, articles, and pronouns. To obtain the maximum number of news matches, we use Ravenpack full package,
which includes articles from over 150,000 press releases, regulatory disclosures, web aggregators, and blog sites. We
remove a few news categories that are likely unrelated to CSR proposals, such as order-imbalances and
technical-analysis. This matching procedure yields 23,381 proposal-to-news pairs from 8,888,788 pieces of Ravenpack
news for firms with CSR-related proposals. We then calculate term frequency-inverse document frequency (TF-IDF) for
each lemmatized common word shared in the voting proposal and news headlines. The TF-IDF is a numerical statistic
that is intended to reflect how important a word is in a collection of sentences. We also compute the cosine similarity
measure between proposals and news headlines. The cosine similarity is a textual analysis statistic, which is commonly
used in the finance literature to measure the similarity among documents. Based on the two textual analysis metrics,
we are able to narrow our sample to 1,723 proposal-news pairs with the maximum number of same words matched.
We then manually check the 1,723 pairs and further remove 1,175 pairs due to duplications and incorrect matches. As
a result, through Ravenpack, our final sample ends up with 548 proposals with implementation-related news. Since we
only have access to news headlines through Ravenpack, it is possible that we may miss certain news, of which the
proposals are mentioned in its content but not headlines. To mitigate this problem, we use Factiva to check all the
proposals that are not matched to Ravenpack news but are within -25% to 25% around the voting pass threshold. We
find proposal-related news for an additional 93 proposals. To check our proposal and news match quality, we further
use Lexis-Nexis bulk API to download more than 20,000 news articles for randomly selected near-the-margin (-10% to
10%) proposals and apply a similar textual analysis method mentioned above. However, we do not find any missed
news, or wrongly matched news through Ravenpack and Factiva news matches. (Various news media sources)
Increase in ASSET4 An actual CSR implementation based on the change of ASSET4 ratings after the vote. Specifically, if the voting
ratings customer’s ASSET4 rating increases by more than 10 (on a scale of 1 to 100) in the year after the vote, we interpret it
as an indication that the proposal has been implemented. (ASSET4)
Board recommendation A proposal is implemented if the board recommends “YES” (or “FOR”) to the proposal before the vote, as this is ex ante
“For” observable at the time of the vote and increases the likelihood of the supplier firm actually implementing the proposal.
(ISS Global and US Voting Outcomes)
JID: FINEC
References Edmans, A., 2011. Does the stock market fully value intangibles? employee
satisfaction and equity prices. J. Financ. Econ. 101, 621–640.
Allen, J., Phillips, G., 20 0 0. Corporate equity ownership, strategic alliances, Ferrell, A., Liang, H., Renneboog, L., 2016. Socially responsible firms. J. Fi-
and product market relationships. J. Finance 55, 2791–2815. nanc. Econ. 122, 585–606.
Antòn, M., Ederer, F., Ginè, M., Schmalz, M. C., 2018. Common ownership, Flammer, C., 2015a. Does corporate social responsibility lead to superior
competition, and top management incentives. Unpublished working financial performance? a regression discontinuity approach. Manag.
paper, University of Navarra. Sci. 61, 2549–2568.
Antòn, M., Polk, C., 2014. Connected stocks. J. Finance 69, 1099–1127. Flammer, C., 2015b. Does product market competition foster corporate so-
Armour, H., Teece, D., 1980. Vertical integration and technological innova- cial responsibility? evidence from trade liberalization. Strat. Manag. J.
tion. Rev. Econ. Stat. 62, 470–474. 36, 1469–1485.
Arora, S., Gangopadhyay, S., 1995. Toward a theoretical model of voluntary Fombrun, C., Shanley, M., 1990. What’s in a name? reputation building
overcompliance. J. Econ. Behav. Organ. 28, 289–309. and corporate strategy. Acad. Manag. J. 33, 233–258.
Azar, J., Raina, S., Schmalz, M. C., 2016. Ultimate ownership and bank com- Freeman, K.M., 2017. The effects of common ownership on customer-
petition. Unpublished working paper, University of Navarra. supplier relationships. Unpublished working paper, Indiana University.
Azar, J., Schmalz, M.C., Tecu, I., 2018. Anticompetitive effects of common Gillan, S.L., Hartzell, J.C., Koch, A., Starks, L.T., 2010. Firms’ environmen-
ownership. J. Finance 73, 1513–1565. tal, social and governance (ESG) choices, performance and managerial
Balotti, R., Hanks, J., 1999. Giving at the office: a reappraisal of charitable motivation. Unpublished working paper, University of Texas at Austin.
contributions by corporations. Bus. Lawyer 54, 956–966. Harford, J., Jenter, D., Li, K., 2011. Institutional cross-holdings and their
Banerjee, S., Dasgupta, S., Kim, Y., 2008. Buyer-supplier relationships and effect on acquisition decisions. J. Financ. Econ. 99, 27–39.
the stakeholder theory of capital structure. J. Finance 63, 2507–2552. He, J., Huang, J., 2017. Product market competition in a world of cross
Baron, D., 2001. Private politics, corporate social responsibility, and inte- ownership: evidence from institutional blockholdings. Rev. Financ.
grated strategy. J. Econ. Manag. Strat. 10, 7–45. Stud. 30, 2674–2718.
Barrot, J., Sauvagnat, J., 2016. Input specificity and the propagation of Hong, H.G., Kubik, J.D., Scheinkman, J.A., 2012. Financial constraints on
idiosyncratic shocks in production networks. Quart. J. Econ. 131, corporate goodness. NBER Working Paper No. 18476.
1543–1592. Hsu, P., Liang, H., Matos, P., 2017. Leviathan inc. and corporate environ-
Bénabou, R., Tirole, J., 2010. Individual and corporate social responsibility. mental engagement. Unpublished working paper, Singapore Manage-
Economica 77, 1–19. ment University.
Burbano, V.C., 2016. Social responsibility messages and worker wage re- Imbens, G., Kalyanaraman, K., 2012. Optimal bandwidth choice for the re-
quirements: field experimental evidence from online labor market- gression discontinuity estimator. Rev. Econ. Stud. 79, 933–959.
places. Organ. Sci. 27, 1010–1028. Ioannou, I., Serafeim, G., 2012. What drives corporate social performance?
Cao, J., Liang, H., Zhan, X., 2019. Peer effects of corporate social responsi- the role of nation-level institutions. J. Int. Bus. Stud. 43, 834–864.
bility. Manag. Sci. 65, 5487–5503. Kalwani, M.U., Narayandas, N., 1995. Long-term manufacturer-supplier re-
Cen, L., Doidge, C., Schiller, C. M., 2016. Do countries matter for informa- lationships: do they pay off for suppliers? J. Mark. 59, 1–16.
tion diffusion in financial markets? evidence from global supply-chain Kitzmueller, M., Shimshack, J., 2012. Economic perspectives on corporate
networks. Unpublished working paper, University of Toronto. social responsibility. J. Econ. Literat. 50, 51–84.
Cen, L., Hertzel, M., Schiller, C. M., 2017a. Speed matters: limited attention Koehn, D., Ueng, J., 2010. Is philanthropy being used by corporate wrong-
and supply-chain information diffusion. Unpublished working paper, doers to buy goodwill? J. Manag. Governance 14, 1–16.
University of Toronto. Krüeger, P., 2015. Corporate goodness and shareholder wealth. J. Financ.
Cen, L., Maydew, E.L., Zhang, L., Zuo, L., 2017b. Customer-supplier re- Econ. 115, 304–329.
lationships and corporate tax avoidance. J. Financ. Econ. 123, 377– Levy, D.T., 1985. The transactions cost approach to vertical integration: an
394. empirical examination. Rev. Econ. Stat. 67, 438–445.
Chatterji, A., Durand, R., Levine, D., Touboul, S., 2016. Do ratings of Liang, H., Renneboog, L., 2017. On the foundations of corporate social re-
firms converge? implications for managers, investors and strategy responsibility. J. Finance 72, 853–909.
searchers. Strat. Manag. J. 37, 1597–1614. Lins, K.V., Servaes, H., Tamayo, A., 2017. Social capital, trust, and firm per-
Chen, T., Dong, H., Lin, C., 2020. Institutional shareholders and corporate formance: the value of corporate social responsibility during the fi-
social responsibility. J. Financ. Econ. 135, 483–504. nancial crisis. J. Finance 72, 1785–1824.
Cheng, I.H., Hong, H., Shue, K., 2016. Do managers do good with other Margolis, J.D., Elfenbein, H.A., Walsh, J.P., 2010. Does it pay to be good? a
peoples’ money? Unpublished Working Paper, Dartmouth College. meta-analysis and redirection of research on the relationship between
Chiu, P.-C., Teoh, S.H., Tian, F., 2013. Board interlocks and earnings man- corporate social and financial performance. Unpublished working pa-
agement contagion. Account. Rev. 88, 915–944. per, Harvard University.
Chu, Y., Tian, X., Wang, Z., 2019. Corporate innovation along the supply Masulis, R.W., Reza, S.W., 2015. Agency problems of corporate philan-
chain. Manag. Sci. 65, 2445–2945. thropy. Rev. Financ. Stud. 28, 592–636.
Cuñat, V., Gine, M., Guadalupe, M., 2012. The vote is cast: the effect of Matten, D., Moon, J., 2008. “Implicit” and “Explicit” CSR: a conceptual
corporate governance on shareholder value. J. Finance 67, 1943–1977. framework for a comparative understanding of corporate social re-
Dai, R., Ng, L., Zaiats, N., 2020. Short seller Attention. Unpublished work- sponsibility. Acad. Manag. Rev. 33, 404–424.
ing paper, WRDS of the Wharton School, University of Pennsylvania. Matvos, G., Ostrovsky, M., 2008. Cross-ownership, returns, and voting in
De Bettignies, J., Robinson, D.T., 2018. When is social responsibility so- mergers. J. Financ. Econ. 89, 391–403.
cially desirable? J. Labor Econom. 36, 1023–1072. Nelson, J., 2004. Leadership, Accountability, and Partnership: Critical
Deng, X., Kang, J.K., Low, B.S., 2013. Corporate social responsibility and Trends and Issues in Corporate Social Responsibility, Report of the CSR
stakeholder value maximization: evidence from mergers. J. Financ. Initiative Launch Event, Corporate Social Responsibility Initiative, Re-
Econ. 110, 87–109. port No. 1. Kennedy School of Government, Harvard University, Cam-
Dhaliwal, D.S., Shenoy, J., Williams, R., 2016. Common auditors and bridge, Mass.
relationship-specific investment in supplier-customer relationships. Schiller, C., 2018. Global supply-chain networks and corporate social re-
Unpublished working paper, University of Arizona. sponsibility. Unpublished working paper, University of Toronto.
Dimson, E., Karakas, O., Li, X., 2015. Active ownership. Rev. Financ. Stud. Servaes, H., Tamayo, A., 2013. The impact of corporate social responsi-
12, 3225–3268. bility on firm value: the role of customer awareness. Manag. Sci. 59,
Dowell, G., Hart, S., Yeung, B., 20 0 0. Do corporate global environmental 1045–1061.
standards create or destroy market value? Manag. Sci. 46, 1059–1074. Starks, L., Venkat, P., Zhu, Q., 2018. Corporate ESG profiles and investor
Dyck, A., Lins, K.V., Roth, L., Wagner, H.F., 2019. Do institutional investors horizons. Unpublished working paper, University of Texas-Austin.
drive corporate social responsibility? international evidence. J. Financ.
Econ. 131, 693–714.

Dai et al.2020. socially responsible corpoate customers

Uploaded by

Dai et al.2020. socially responsible corpoate customers

Uploaded by

JID: FINEC

ARTICLE IN PRESS [m3Gdc;February 14, 2020;3:56]

Contents lists available at ScienceDirect

Journal of Financial Economics

Socially responsible corporate customersR

More than 1,000 of the world’s largest companies... have

2 R. Dai, H. Liang and L. Ng / Journal of Financial Economics xxx (xxxx) xxx

R. Dai, H. Liang and L. Ng / Journal of Financial Economics xxx (xxxx) xxx 3

4 R. Dai, H. Liang and L. Ng / Journal of Financial Economics xxx (xxxx) xxx

R. Dai, H. Liang and L. Ng / Journal of Financial Economics xxx (xxxx) xxx 5

6 R. Dai, H. Liang and L. Ng / Journal of Financial Economics xxx (xxxx) xxx

R. Dai, H. Liang and L. Ng / Journal of Financial Economics xxx (xxxx) xxx

CSR composite and component scores

[m3Gdc;February 14, 2020;3:56]

8 R. Dai, H. Liang and L. Ng / Journal of Financial Economics xxx (xxxx) xxx

R. Dai, H. Liang and L. Ng / Journal of Financial Economics xxx (xxxx) xxx 9

CSRS CSRC CSRS CSRC

CSRC score(t) 0.026∗ ∗ 0.031∗ ∗ ∗ 0.040∗ ∗ ∗ 0.038∗ ∗ ∗ 0.016∗ ∗

10 R. Dai, H. Liang and L. Ng / Journal of Financial Economics xxx (xxxx) xxx

R. Dai, H. Liang and L. Ng / Journal of Financial Economics xxx (xxxx) xxx 11

High customer-country CSR Low customer-country CSR

Supplier-country CSR US Non-US

Low High Low High only countries

CSRC 0.014 0.043∗ ∗ ∗ 0.013∗ 0.011 0.018∗ 0.036∗ ∗ ∗

12 R. Dai, H. Liang and L. Ng / Journal of Financial Economics xxx (xxxx) xxx

Identifying the impact of customers on their suppliers’

R. Dai, H. Liang and L. Ng / Journal of Financial Economics xxx (xxxx) xxx 13

Panel A: Differential CSR measures

Panel B: Alternative CSR rating databases

Variable CSRS (t + 1 ) CSRC (t + 1 )

IVA CSRC (t) score 0.014∗ ∗

Panel C: Environmental and social ratings

Variable CSRS (t + 1 ) CSRC (t + 1 )

Panel D: Falsiﬁcation tests

Variable CSRSIndustry (t + 1 ) CSRSCountry (t + 1 )

CSRC (t) score -0.011 -0.010

14 R. Dai, H. Liang and L. Ng / Journal of Financial Economics xxx (xxxx) xxx

R. Dai, H. Liang and L. Ng / Journal of Financial Economics xxx (xxxx) xxx 15

CSRS Score(t + 1 ) = a0 + a1 Pass(t ) + a2 %Votes For CSR Proposal + (t + 1 ),

Panel A: Validity for CSR vote (preexisting difference)

NObs Mean NObs Mean Estimate P-value Estimate P-value

LeverageS 20,068 0.263 13,424 0.249 0.014 0.000 0.016 0.637

Panel B: Supplier reactions to the passage of CSR vote

50% of 150% of CSR

Pass 6.988∗ ∗ ∗ 7.149∗ 10.742∗ ∗ ∗ 4.461∗ ∗

News about implementation Increase in ASSET4 ratings Board recommending “For”

Estimate 0.386∗ ∗ ∗ 32.857∗ ∗ ∗ 0.239∗ ∗ ∗ 36.309∗ ∗ ∗ 0.793∗ ∗ ∗ 15.995∗ ∗ ∗

Panel D: Supplier responses to implementation and nonimplementation of CSR

News about implementation Increase in ASSET4 ratings Board recommending “For”

Implementation Nonimplementation Implementation Nonimplementation Implementation Nonimplementation

Supplier CSR score 69.272 63.495 60.276 59.760 68.856 61.031

16 R. Dai, H. Liang and L. Ng / Journal of Financial Economics xxx (xxxx) xxx

Fig. 2. Regression discontinuity plots of the stakeholder effect of CSR.

CSRS ProductScore(t + 1 ) = a0 + a1 CSRC ProductScore(t )

R. Dai, H. Liang and L. Ng / Journal of Financial Economics xxx (xxxx) xxx 17

Customer RRI Supplier RRI

18 R. Dai, H. Liang and L. Ng / Journal of Financial Economics xxx (xxxx) xxx

Product responsibilityC x 2008 0.250∗

R. Dai, H. Liang and L. Ng / Journal of Financial Economics xxx (xxxx) xxx 19

20 R. Dai, H. Liang and L. Ng / Journal of Financial Economics xxx (xxxx) xxx

relationships. The reason why we examine only targets, Table 7

R. Dai, H. Liang and L. Ng / Journal of Financial Economics xxx (xxxx) xxx 21

Target=Supplier/Customer Target=Supplier Target=Customer

Variable (1) (2) (3) (4) (5) (6)

22 R. Dai, H. Liang and L. Ng / Journal of Financial Economics xxx (xxxx) xxx

Table 9 less of a need to learn from their customers. Hence, in

5.3. Inﬂuencing the decision-making process of suppliers

customer. Alternatively, when there is less resource depen-

R. Dai, H. Liang and L. Ng / Journal of Financial Economics xxx (xxxx) xxx 23

Existing cross-holdings First-time holdings

Variable (1) (2) (3) (4) (5)

et al., 2011), industry competitiveness (e.g., Azar et al., + bk Xk (t ) + FE(t ) + (t + 1 ), (9)

24 R. Dai, H. Liang and L. Ng / Journal of Financial Economics xxx (xxxx) xxx

Existing appointments First-time appointment

Variable (1) (2) (3)

R. Dai, H. Liang and L. Ng / Journal of Financial Economics xxx (xxxx) xxx 25