This article was downloaded by: [134.50.218.9] On: 19 April 2019, At: 18:13Publisher: Institute for Operations Research and the Management Sciences (INFORMS)INFORMS is located in Maryland, USA
Management Science
Publication details, including instructions for authors and subscription information:http://pubsonline.informs.org
Peer Effects of Corporate Social ResponsibilityJie Cao, Hao Liang, Xintong Zhan
To cite this article:Jie Cao, Hao Liang, Xintong Zhan (2019) Peer Effects of Corporate Social Responsibility. Management Science
Published online in Articles in Advance 19 Apr 2019
. https://doi.org/10.1287/mnsc.2018.3100
Full terms and conditions of use: https://pubsonline.informs.org/page/terms-and-conditions
This article may be used only for the purposes of research, teaching, and/or private study. Commercial useor systematic downloading (by robots or other automatic processes) is prohibited without explicit Publisherapproval, unless otherwise noted. For more information, contact [email protected].
The Publisher does not warrant or guarantee the article’s accuracy, completeness, merchantability, fitnessfor a particular purpose, or non-infringement. Descriptions of, or references to, products or publications, orinclusion of an advertisement in this article, neither constitutes nor implies a guarantee, endorsement, orsupport of claims made of that product, publication, or service.
Please scroll down for article—it is on subsequent pages
INFORMS is the largest professional society in the world for professionals in the fields of operations research, managementscience, and analytics.For more information on INFORMS, its publications, membership, or meetings visit http://www.informs.org
Peer Effects of Corporate Social ResponsibilityJie Cao,a Hao Liang,b Xintong Zhana
aCUHK Business School, The Chinese University of Hong Kong, Hong Kong; b Singapore Management University, Singapore 188065Contact: [email protected], http://orcid.org/0000-0002-5245-8895 (JC); [email protected], http://orcid.org/0000-0003-1891-8453 (HL);[email protected] (XZ)
Received: July 12, 2016Revised: July 4, 2017; February 27, 2018Accepted: March 26, 2018Published Online in Articles in Advance:April 19, 2019
Abstract. We investigate how firms react to their product-market peers’ commitment toand adoption of corporate social responsibility (CSR) using a regression discontinuitydesign approach. Relying on the passage or failure of CSR proposals by a narrowmargin ofvotes during shareholder meetings, we find the passage of a close-call CSR proposal and itsimplementation are followed by the adoption of similar CSR practices by peer firms. Inaddition, peers that have greater difficulty in catching up with the voting firm in CSRexperience significantly lower stock returns around the passage, consistent with the notionthat the spillover effect of the adoption of CSR is a strategic response to competitive threat.Using alternative definitions of peers and examining underlying mechanisms, we furtherrule out alternative explanations, such as that based on propagation by financialintermediaries.
History: Accepted by Gustavo Manso, finance department.Funding: J. Cao and X. Zhan acknowledge generous financial support of the Research Grant Councilof the Hong Kong Special Administrative Region, China [Project No. CUHK 14501115].
Supplemental Material: The online appendices are available at https://doi.org/10.1287/mnsc.2018.3100.
1. IntroductionThe existence of peer effects on individual and house-hold financial decision making and behavior has beenwell documented (e.g., Kaustia and Knüpfer 2012,Bursztyn et al. 2014, Georgarakos et al. 2014, Agarwalet al. 2016, Agarwal et al. 2017, Bailey et al. 2018). Recentstudies have found substantial externalities and peereffects in corporate policies as well. For example,according to Leary and Roberts (2014), peer effects aremore important for determining capital structure thanmost previously identified determinants. Such peereffects are also present in corporate precautionarycash holdings (e.g., Hoberg et al. 2014), corporate in-vestment decisions (e.g., Foucault and Fresard 2014,Dessaint et al. 2018), dividend policies (e.g., Kaustia andRantala 2015, Grennan 2018), and financial misconduct(e.g., Parsons et al. 2018) as well as related stockmarketreactions during crucial corporate events, such as ac-quisitions and IPOs (e.g., Hsu et al. 2010, Servaes andTamayo 2014).
Despite the vast literature on peer effects in vari-ous corporate activities, an important yet largely un-explored aspect of such effects is its social component.The term “social component” refers to a firm’s en-gagement in activities that improve other stakeholders’welfare, from investing in environmental protection toincreasing workforce diversity and employee welfare.Corporate social responsibility (CSR) policies have
been widely adopted by companies worldwide, andmore than 80% of S&P 500 companies regularly publishCSR reports. This movement has fundamentally changedthe competitive landscape and industry structures ofeconomies around the world. Despite the prevalence ofCSR policies, relatively little is known about whetherthe surge in such policies is at least partially attribut-able to the spillover effect it has on other companies andtheir underlying mechanisms if such CSR peer effectsindeed exist. These are the questionswe aim to address inthis paper with a focus on competing peers on theproduct market.Underlying our empirical investigation is the argu-
ment that a firm’s adoption of CSR policies can affectpeer firms’ “utilities” in competition, leading the peersto respond strategically by adopting more CSR prac-tices. For example, one such strategy that firms canadopt is environment-friendly production technology,which can create a good image and builds social capitaltoward environmentally conscious consumers (Hongand Liskovich 2016, Lins et al. 2017). Because it isexpensive to develop such technology, firms may notinvest heavily in it. However, when one firm deviatesfrom the norm by adopting environment-friendlyproduction technology, it can gain a competitive ad-vantage over others in the product market (Flammer2015a). Consequently, its product-market peers mayget “hurt” in terms of profitability, leading them to
invest in similar environment-friendly technology tocatch up. This utility mechanism suggests that peerfirms’ adoption of CSR policies occurs as a strategicresponse to competitive threats, such that peer firmsmimic the deviating firm’s CSR strategy. Some peerfirms may also adopt a different type of CSR, that is,a differentiation strategy, depending on their com-petitive positions in the product market. Such a dif-ferentiation strategy may be particularly attractive fora market follower when evaluating uncertain in-vestment opportunities (Smit and Trigeorgis 2006,Stoughton et al. 2017) as this allows a firm to create itsown comparative advantage and avoid direct compe-tition with the market leader (Porter 1985).
Alternately, a firm’s CSR policy can spill over to itscompeting peers through managers’ herd behavior orpropagation by financial intermediaries. Specifically,managers of peer firms may have an incentive to herdout of concerns about their reputations in the labormarket regardless of whether doing so maximizesvalue for their own shareholders.1 Theymay also adoptmore CSR policies if pressured by external parties, suchas analysts and institutional investors, who usually“propagate” certain practices across the firms that theycover or hold.Moreover, if CSR plays little role in firms’interactions in the product market and managers makeinformed decisions for their shareholders, there may beno spillover whatsoever of the adoption of CSR.
Despite the growing importance of CSR and differenttheoretical predictions for peer effects, the empiricalevidence for CSR peer effects is limited. The mainreason is that both CSR and a firm’s interactions withits peers in the product market are arguably endoge-nous choices made by the firm, which posits an em-pirical challenge for testing the peer effects of CSR.Specifically, it is unknown whether firms and investorsreact to their peers’ CSR policies by changing their ownCSR practices or whether the preexisting differences inother unobservable firm characteristics lead differentpeer firms to adopt CSR policies to different extents. Itis also difficult to apply a typical quasi-natural ex-perimental approach by exploring exogenous legisla-tive changes because such changes usually affect allfirms’ CSR in the same industry or market.
We circumvent these empirical concerns and in-vestigate the peer effects of CSR by using a regressiondiscontinuity design (RDD) approach. We compare theeffects of a firm’s shareholder-sponsored CSR proposalsthat pass or fail by a small margin of votes (arounda 50% majority threshold) in annual meetings on itsproduct-market peer firms’ subsequent CSR practices.2
The passage of such close-call proposals is similar toa random assignment of CSR to companies and, hence,is not correlated with peer-firm characteristics. Con-ceptually, there is no reason to expect that the peerfirms of a company that passes a CSR proposal with
51% of the votes are systematically different from thepeer firms of a company in which a similar proposalfails with 49% of the votes. Therefore, close-call CSRproposals provide a source of random variation ofa firm’s commitment to CSR that can be used to esti-mate the causal effect on its peer firms’ CSR practices.3
Although a similar approach has been used byFlammer (2015a) and by Cuñat et al. (2012) to study theeffects of the passage of CSR proposals and corporategovernance proposals on stock returns, both of thosestudies examined the focal firm’s shareholder valuerather than peer effects induced by product-marketconnections. Our empirical setting focuses on peerfirms, which enables us to go beyond the focal firm’sperspective and study the dynamic interaction amongfirms. In addition, we investigate the proposals’implementation utilizing different data sources, in-cluding corporate news about the actual adoption ofCSR, to better understand the potential mechanisms.By empirically testing a sample of more than 3,000
U.S. public nonvoting peer firms over the period of1997–2011 using the RDD approach, we find strongeffects of the passage of close-call CSR proposals on thestock market reaction and the adoption of CSR policiesby the voting firm as well as subsequent adoption ofCSR policies by its peer firms. Specifically, if the votingfirm marginally passes a close-call CSR proposal, itexperiences higher three-day cumulative abnormalreturns (CARs) and, subsequently, a higher CSR score.The average CSR score for its competing peer firms inthe following year is significantly higher (30% of thestandard deviation) than that of the competing peerfirms inwhich the vote fails by a smallmargin. The effectis even stronger if the close-call CSR proposal was ac-tually implemented in the subsequent year, suggestingthat these effects go beyond pure signaling; they canmaterially influence peerfirms.Using global polynomialestimations, different measures of CSR and their sub-scores, and different competing peer samples, we foundthese results to be robust. Such effects are absent fromnonpeer groups and for non-CSR proposals. In addition,the effects are generally concentrated in the same do-main in which a competing voting firm passes a specifictype of CSR proposal although industry followers alsoadopt other types of CSR to some extent.In exploring the channels through which CSR peer
effects are transmitted, we find little evidence that theyare driven by financial intermediaries’ propagation assuch effects are absent from peer firms alternativelydefined by common analyst coverage or common in-stitutional investors. Similarly, we do notfind significantdifferences between two subsamples of product-marketpeers with and without common analysts or com-mon institutional owners with the voting firm. Instead,the peer effects of CSR policies are concentrated ina subsample in which the competitive pressure between
voting firm and peer firms is relatively strong, sug-gesting a channel throughwhich competitive pressure isexerted.
To better understand the underlying mechanism, wefurther investigate peer firms’ stock market reaction tothe passage of a CSR proposal. The passage of a CSRproposal should be bad news to competing peers if thevoting firm’s CSR policy forms a competitive threat tothese peers. We find significantly lower three-day CARsfor product-market peers around the passage of a close-call CSR proposal. Such negative CARs are more sig-nificant in firms with greater financial constraints andmore competitive pressures when the voting firm’s CSRproposal is more likely to be implemented (based on exante recommendations by the board) and when the gapin CSR between voting and peer firms widens after thevote. Our empirical results confirm that the unfavorablemarket reaction comes mainly from those who havegreater difficulty catching up in CSR; furthermore, theylend additional support to the competition-based ex-planation, that CSR can enhance a firm’s value by cre-ating a competitive advantage.
Two guideposts can be used to put our findings intocontext in the literature. First, our work contributes tounderstanding peer effects in economic activity. Theextant literature has documented substantial peer effectson corporate behavior, such as firms reacting to theirpeers’ financial policies by adjusting their capital struc-ture (Leary andRoberts 2014) and reducing cash holdingsand capital expenditure while increasing dividend pay-out and adopting more antitakeover devices followinghostile takeover threats in their industries (Servaes andTamayo 2014). What is theoretically less clear is whetherand how a firm’s social engagement can also spill over toother firms. Because it is an increasingly important aspectof corporate behavior, CSR provides us with an idealfoundation from which to test a different facet of peereffects—one that is related to social engagement—aswellas its implications for financial performance. Our resultsshow that the peer effects of CSR policies are a strategicresponse that relieves the competitive threats resultingfrom the voting firm’s CSR policy.
Second, our work contributes to the understandingof why firms participate in CSR, which has increas-ingly become a mainstream business activity despitestandard economic theories that predict the practiceshould be rather uncommon (Bénabou and Tirole 2010,Kitzmueller and Shimshack 2012). The neoclassicaleconomic paradigm usually considers these CSR ac-tivities to be unnecessary and inconsistent with profitmaximization (e.g., Friedman 1970). The extant litera-ture usually explains the determinants of CSR froma company’s own perspective by investigating howa firm’s decision to participate in CSR is motived byfinancial conditions (e.g., Hong et al. 2012), strategicand reputational concerns (Hong and Liskovich 2016),
shareholder engagement (Dimson et al. 2015), andagency problems (Di Giuli and Kostovetsky 2014,Masulis and Reza 2015, Cheng et al. 2016). Others haveinvestigated the inverse, namely the effects of CSR oncorporate policies and performance (Dowell et al. 2000,Edmans 2011, Deng et al. 2013, Flammer 2015a) and thecost of capital (Dhaliwal et al. 2011, El Ghoul et al. 2011,Albuquerque et al. 2017). Some recent studies haveshown that CSR is not merely driven by a firm’s owncharacteristics, but to a greater extent is determined byexternal factors, such as the legal and economic en-vironment (e.g., Liang and Renneboog 2017a, b). Ourwork extends the scope of this literature by doc-umenting that a firm’s CSR policy and financial per-formance can also be substantially changed by itspeers’ practices in the pursuit of competitive advan-tage, suggesting that the adoption of CSR policies isvalue-enhancing.
2. Data and Empirical Strategy2.1. Data, Measurement, and Sample ConstructionAs previously mentioned, we rely on a regressiondiscontinuity design approach as the key identificationstrategy and investigate the effect of the close-callpassage of a firm’s CSR proposal on its nonvotingpeers’ subsequent CSR performance. We obtain thedata on shareholder proposal voting results fromRiskMetrics and Factset’s SharkRepellent. The Risk-Metrics data covers shareholder proposals from 1997 to2011 for all 1,500 S&P companies and an additional400–500 widely held companies. The resolution type“SRI” in RiskMetrics identifies the proposals related toCSR. For each proposal, the data set provides the dateof the annual meeting, the proposal’s sponsor, thevoting requirements, and the vote outcome. We sup-plement the voting data from RiskMetrics with datafrom SharkRepellent, which covers about 4,000 com-panies in the Russell 3000 index from 2005 to 2011. TheSharkRepellent proposals related to CSR are catego-rized as “social/environmental issues.”We then cross-validate these proposals with the Institutional Share-holder Services voting outcomes data.To identify peer effects, we construct a sample of
competing peers using the Hoberg–Phillips indus-try classification based on firm pairwise similarityscores from textual analyses of firm 10-K productdescriptions.4 This approach essentially identifies peerfirms according to the similarity of their products,which better captures the competitive relationshipbetween peer firms and also matches the product-centric nature of CSR and the fact that CSR usuallyspans multiple industries, a point that may not beproperly captured by industry-based peer definitions.This peer (product-market rivals) database covers thefiscal years 1996–2011. In each fiscal year, two firms arerecorded as a pair of rivals if they exhibit a degree
of product similarity according to the product de-scriptions in their 10-K files. Our identification ofproduct-market peers is based on ex ante information—that is, the fiscal year in which the product-marketcompetitors from the Hoberg–Philips database areretrieved is the year before the actual event. After linkingthe shareholder proposal data with the Hoberg–Phillipspeer-firm database, we require no missing outcomevariables (discussed in the next paragraph) or relevantfirm fundamental variables (size, market-to-book, andleverage). When we later investigate the stock marketreaction, we also remove peer firms that have experi-enced stock and bond issuance, mergers and acquisi-tions (M&As) announcements, and dividend paymentsaround their affiliated voting firm’s voting date (Day −5to Day 5) to rule out potential confounding effects.5 Thisfiltering procedure leads to a final competing peer-firmsample of 38,630 (nonvoting peer-)firm vote observa-tions, corresponding to 3,452 unique nonvoting U.S.public firms associated with 1,407 unique firm votesfrom 1997 to 2011.6 In our robustness tests, we also usethe three-digit standard industrial classification (SIC)definitions of industry peers. Online Appendix C pro-vides the distribution of our sample with panel Ashowing a summary of the numbers of voting firm voteobservations and nonvoting competing peer-vote obser-vations in each year as well as the cumulative percentageand panel B showing the distribution of CSR proposals bytype, which are classified according to the general cate-gories (dimensions) as used in Kinder, Lydenberg, andDomini (KLD).
To test nonvoting peers’ reaction to the passage ofa CSR proposal in the voting firm, we retrieve the datafrom the MSCI ESG STATS database (formerly knownas the KLD database), which are the most comprehen-sive CSR scores used in the literature (e.g., Chatterji et al.2009; Flammer 2015a, b; Hong and Liskovich 2016;Cronqvist and Yu 2017; Lins et al. 2017). Developed bya for-profit company, KLD scores are similar to creditratings. The scores measure firm-level CSR along thelines of community relations, product characteristics,environmental impact, employee relations, workforcediversity, and corporate governance. KLD scans publicdatabases, such as those that have experienced employeestrikes and Environmental Protection Agency violationsand uses a team of analysts to measure these and othersocial-responsibility dimensions of firm production.We mainly rely on the CSR score of the nonvoting firmsin the year t+1 (the year after their peers’ vote) as theoutcomevariable. TheMSCI ESG STATS (KLD) databaseprovides detailed information on firms’ CSR activitiesaccording to 13 categories: community, diversity, em-ployment, environment, human rights, product, al-cohol, gaming, firearms, military, nuclear, tobacco, andcorporate governance. Within each category, the da-tabase shows whether the firm has performed a benefit
(“strength”) or effected aharm (“concern”) and awards onepoint for each relevant activity. The CSR score is cal-culated as strengths minus concerns. To measure theoverall CSR performance of a firm, we consider fourmain CSR categories (or dimensions) as classified byKLD: community, diversity, employee relations, andenvironment.7
Following Deng et al. (2013) and Servaes and Tamayo(2013), we count the number of strengths and concernswithin each of the four categories and subtract thenumber of concerns from the number of strengths toconstruct the raw score for each category in each year.The overall rawCSR score is the sum of the raw scores ofthe four categories. A higher raw CSR score indicatesa better CSR performance. However, as Manescu (2011)points out, the raw CSR score may not be helpful inevaluating a firm’s actual CSR activities over the yearsbecause the number of strengths and concerns withineach category can differ. To overcome this concern andobtain consistent comparisons in both the cross-sectionand time-series analyses, we scale the strengths andconcerns for each firm-year to a range of zero to one. Todo so, we divide the number of strengths (or concerns)for each firm-year within each CSR category by themaximum possible number of strengths (or concerns) ineach CSR category each year to get the adjusted strength(or concern) index. We then subtract the adjusted con-cern index from the adjusted strength index. For eachcategory, the adjusted CSR score ranges from −1 to +1.For the overall adjusted CSR score, we sum the fouradjusted scores. Therefore, in principle, the adjustedCSR score can range from −4 to +4. We use the rawCSR score and the change in the adjusted CSR per-formance score as alternative outcome variables fora robustness check. In robustness tests, we also col-lected information on the actual implementation of thepassed CSR proposals and tested it using the ASSET4U.S. ESG ratings, an alternative data source on firm-level CSR practices.The definitions and sources of our variables are
provided in Online Appendix B. The summary sta-tistics of our key outcome variables and control vari-ables are provided in Table 1.
2.2. MethodologyWe use a regression discontinuity framework to esti-mate the causal effect of shareholder proposals onpeer firms’ future CSR engagement and other out-come variables.8 Like Flammer (2015a), we use a votingfirm’s random passage of CSR proposals for iden-tification, but we focus on the CSR practices of thenonvoting peer firms instead of that of the voting firm.Ideally, to obtain a consistent estimate, we would wantthe passage of a CSR proposal to be a randomlyassigned variable with regard to peer firms’ charac-teristics, especially the firms’ CSR performance. The
RDD framework that exploits the vote shares helps usto approximate this ideal setup because the passage ofa CSR proposal is a random outcome in an arbitrarilysmall interval around the majority vote threshold (50%);for example, whether a proposal passes by 51% or failsby 49% is arguably random. These close-call CSR pro-posals, therefore, provide a source of random variationin commitment to CSR that can be used to estimate thecausal effect of passing a CSR proposal on peer firms’performance. Our estimate of the effect using RDD is notaffected by omitted variables, even if the variables arecorrelated with the vote, as long as the effects are con-tinuous around the threshold.
We perform the RDD by using a nonparametric,“local” linear estimation. Small “neighborhoods” on theleft- and right-hand sides of the threshold are used toestimate discontinuities in peer firms’ reactions. Wefollow Imbens and Kalyanaraman (2012) to derive theasymptotically optimal bandwidth under a squared-error loss. The choices of the neighborhood (band-width) are data-driven (determined by the datastructure) and different across samples and variables.By choosing the optimal bandwidth to the left and rightof the cutoff point (threshold), we are able to use thenonparametric linear estimation approach to capturethe difference in future CSR performance betweenpeers with respect to the passage and failure of a CSRproposal by their associated voting firm. In addition,the RDD requires no other observable covariates (controlvariables) for identification. The local linear regressionmodel can therefore be specified as
Yit � α + β ·Xit + ρ ·Passit + εit, (1)
where Yit is the CSR score in year t+1 of the peer firm i;Passit is a dummy equal to 1 if the peer firm’s associatedvoting firm passes a CSR-related proposal—that is, ifmore than 50% of the votes are in favor of adoptingthe CSR proposal—and 0 otherwise; and Xit is the
percentage of vote shares favoring the CSR proposal,centered at the 50% threshold. The estimate of ρ cap-tures the discontinuity at the majority threshold—thedifference in outcome between the peer firms of votingfirms that marginally pass a CSR proposal and the peerfirms of voting firms that marginally reject a CSRproposal—and, hence, provides a consistent estimateof the causal effect of passing a CSR proposal on peerfirms’ Yit. We also use alternative bandwidths that areeither narrower or wider than the optimal bandwidthto check the sensitivity of our results.
2.3. Tests for a Quasi-Randomized AssignmentOur identification strategy requires that passing orrejecting a close-call CSR proposal be nearly randomwith respect to peer-firm characteristics. In this sub-section, we perform two diagnostic tests for the RDDvalidity of the identifying assumption (randomnessassumption) that shareholders of the voting companycannot precisely manipulate the forcing variable (i.e.,vote shares) near the known cutoff (Lee and Lemieux2010). If this assumption is satisfied, the variation in thepassage of CSR proposals should be as good as thatfrom a randomized experiment.
2.3.1. Continuity in the Distribution of ShareholderVotes. We first test whether the distribution of share-holder votes is continuous around the majority thresh-old, that is, 50% of vote shares. We follow McCrary(2008) and provide a formal test of the discontinuityin the density, which checks for the smoothness of thedensity function around the threshold. A random as-signment of pass versus fail at a narrow margin impliesthat the distribution of vote shares should be smooth andcontinuous around the majority threshold. Online Ap-pendix D.1 visually confirms this. A more formal test isprovided in Online Appendix D.2, which plots thedensity of shareholder votes. The dots depict the density
Table 1. Summary Statistics
Variable No. of observations Mean Standard deviation P25 Median P75
Notes. This table reports the descriptive statistics of the key variables. Based on the 1,407 unique CSR proposals that were being voted on, oursample consists of 38,630 unique (nonvoting) peer-vote observations from 3,452 unique U.S. public firms over the period 1997–2011. Allcontinuous variables are winsorized at the fifth and 95th percentiles. Variable definitions are provided in Online Appendix B.
and the solid line the percentage of votes for CSR. Thedensity appears generally smooth with no evidence ofa discontinuous jump around the threshold. The p-valueis 0.1556, which fails to reject the null hypothesis ofcontinuity of the density function at the threshold. Weare able to confirm, in accordance with McCrary’s (2008)test result, that no precise manipulation exists and thatthe assumption of smoothness is validated.
2.3.2. Preexisting Differences. The randomness as-sumption of our RDD setting also requires that the peerfirms of companies whose voting shares are marginallybelow or above the majority threshold should be verysimilar on the basis of ex ante characteristics. In otherwords, if the passage of close-call CSR proposals isclose to a random assignment, it should be unrelated topeer-firm characteristics prior to the vote. There is littlereason to believe that such a voting outcome is directlyaffected by peer-firm characteristics. To justify this, weshow in Table 2 the differences of a few key firm-characteristic variables for these two peer groups(hereafter called “passing peers” and “failing peers,”which refer to peer firms of the voting firm that passesa close-call CSR proposal and those of the voting firmthat fails a close-call CSR proposal, respectively). Asshown in columns (1) and (2), before voting on CSRproposals, the firm’s characteristics—size, market-to-book ratio, book leverage, return on assets (ROA), andCSR scores—of passing peers and failing peers are notvery different. In column (3), the differences betweenpassing peers and failing peers in general are statisti-cally significant for firm size and market-to-book ratio,but such significance completely disappears in column (4)in which we compare the differences at the narrowmargin around the threshold.9 Overall, this evidencesuggests that no systematic or significant differenceexists between passing peers and failing peers aroundthe majority threshold, which lends support to ouridentification strategy.More preexisting differences testscan be found in Online Appendix E.
3. Results3.1. First-Stage Results of the Voting FirmHaving validated the randomness assumption of ourRDD setting, we first show the results for the first-stageof the CSR peer effects, namely the reaction of thevoting firm itself upon the passage of its close-call CSRproposal. We replicate Flammer’s (2015a) tests andreport the results in panel A of Table 3. We find that thevoting firms whose close-call CSR proposals are mar-ginally passed experience positive CARs. The eco-nomic magnitude for a voting firm’s CAR is 1.02%,slightly smaller than Flammer’s (2015a) result of 1.77%.This may be due to the fact that our sample firms onaverage are larger than Flammer’s (2015a). In addition,the passing voting firm experience an increase in itsCSR score one year later and higher ROA in year t+2,compared to a failing voting firm, also consistent withthe findings in Flammer (2015a).Panel B of Table 3 reports the probability of all
passed proposals being implemented in the followingyear based on several criteria.10 On average, theimplementation rate for all CSR proposals is around60%. This is comparable with Flammer (2015a), whofinds that 52% of the passing firms implemented theCSR proposal. Overall, the results in Table 3 confirmthat the passage of a CSR proposal indeed has a siz-able effect on the voting firm’s own stock marketperformance and subsequent CSR adoption, whichwill likely create a competitive threat to its product-market peers.
3.2. The Effects of CSR Commitment on Peer Firms’Following-Year CSR Levels
We then formally test the effect of the voting firm’spassage or rejection of a close-call CSR proposal onpeer firms’ subsequent-year CSR levels (using KLDscores). As previously mentioned, we start with com-peting peers based on the Hoberg–Phillips classificationand report the results of our baseline specifications(Equation (1)).
Table 2. Validity for CSR Vote as Regression Discontinuity Design (Preexisting Difference)
(1) Fail (2) Pass(3) Difference(fail vs. pass)
(4) Difference withinnarrow margin
(optimal bandwidth)
Observations Mean Observations Mean Estimate p-value Estimate p-value
Notes. This table shows differences in several observable characteristics—adjustedKLD score, firm size,market-to-book ratio, leverage ratio, andROA—between (nonvoting) peer firms that are associated with the passage (“pass”) of a CSR proposal in voting firms and those that areassociated with the rejection (“fail”) of a CSR proposal in voting firms by a small margin one year before the vote. We define the margin as theoptimal bandwidth following Imbens and Kalyanaraman (2012).
Panel A of Table 4 estimates the difference in CSRcommitment between passing peers and failing peers,as previously defined, with different bandwidths and
with rectangular as well as triangular kernels.11 Itclearly shows that the estimates are positive and sta-tistically significant above the 5% level across different
Table 3. Effect of the Passage of a Close-Call CSR Proposal on the Voting Firm
Panel A. Stock market reaction, subsequent CSR engagement, and profitability of the voting firm
Panel B. Implementation rate of all passed CSR proposals
Implementation rate All CSR proposals Environment Employee relationship Diversity
Panel B.1. Based on increase in CSR score 60% 60% 50% 20%Panel B.2. Based on increase in ASSET4 score 62.50% 100% 100% 25%Panel B.3. Based on news about implementation 57.14% 80% 66.70% 60%
Notes. Panel A reports the impact of CSR proposal passage on voting firms using the RDD approach. We examine voting firms’ future CSRperformance (year t + 1), immediate market reaction, and future profitability (year t+2). Panel B presents the implementation rate of CSRproposals based on different criteria. In panel B.1, we define implementation of an increase in adjusted CSR score from MSCI ESG STATS(formerly KLDdatabase). For each dimension, implementation is defined as an increase in related CSR dimensions. In panel B.2, implementationis the increase in ASSET4 score. In panel B.3, wemanually examine the news for passed CSR proposals and define implementation as (a) there isa news report about the proposal implementation, (b) there is a news report on the corporate website, or (c) management team vote “for” beforethe votes.
Table 4. Responses of Nonvoting Peers to the Passage of a CSR Proposal: Baseline Results
Panel A. Following-year response of nonvoting peer firms to the passage of a voting firm’s CSR proposal
Panel B. Evidence from global polynomial regression
(1) (2)
Adj. KLD score t+1 Adj. KLD score t+1
Pass 0.24*** 0.086*(0.05) (0.047)
Constant −0.16*** −1.24***(0.03) (0.070)
Polynomial order 3 3Controls No YesObservations 38,630 37,634
Notes. This table presents peer firms’ future CSR performance as a response to the CSR votes. Panel A presents RDD estimations from a locallinear regression as specified in Equation (1) using the optimal bandwidth following Imbens and Kalyanaraman (2012). We report results acrossalternative bandwidths, including 50% of optimal bandwidth (narrower bandwidth) and 150% of optimal bandwidth (wider bandwidth).Results using both the rectangular and the triangular kernels are reported. Panel B shows the RDD estimations from a global polynomialregression. Column (1) does not include control variables, and Column (2) includes the control variables size, market-to-book, leverage, andROA. Variable definitions are provided in Online Appendix B. Standard errors are clustered at the firm level and reported in parentheses.
**, ***Significance at the 5% and 1% levels, respectively.
specifications of bandwidth and kernel. The point es-timate is approximately 0.16 under the data-drivenoptimal bandwidth (as in column (1)), indicating thatthe difference in CSR levels between passing peersand failing peers is as large as 0.16 point. Given thatthe adjusted KLD score has a mean of −0.13 anda standard deviation of 0.42, a difference of 0.16should be economically sizable. The results remainsignificant when we use 50% and 150% of the optionalbandwidth as shown in columns (2) and (3). When wereplace the adjusted KLD score with the change in ad-justed KLD score from year t−1 to year t+1 and estimateusing optimal bandwidth, we obtain quantitatively sim-ilar results (column (4)). Although all of these results areestimated using the rectangular kernel, a very similarresult is obtained when the difference in KLD scores isestimated using the triangular kernel as in column (5).12
These results imply that when a voting firm marginallypasses a CSR proposal, its peer firms’CSR practices in thefollowing year improve significantly. In otherwords, peerfirms follow their competitors’ potential adoption ofCSR proposals by engaging more in their own CSR,possibly because of competitive pressure created by thevoting firms.
In panel B we conduct a similar RDD test usinga different methodology to capture the discontinuity.Instead of relying only on the observations within theoptimal bandwidths, we extend the regression dis-continuity analysis with an estimation of a globalpolynomial series model by including polynomials oforder three on both sides of the threshold.13 Specifi-cally, we estimate the following model:
whereYit is the outcome variable of the (nonvoting) peerfirm—that is, an adjusted KLD score in year t+1; Passit isa dummy that equals 1 if the voting firm passes the CSRproposal—that is, if the vote percentage is higher than50%—and 0 otherwise; Pl(vit,γl) is a flexible polynomialfunction for observations on the left-hand side of themajority threshold γ (50% in our case) with differentorders; Pr(vit,γr) is a flexible polynomial function forobservations on the right-hand side of the threshold γwith different orders; and v is the percentage of sharesfavoring the CSR proposal. −Zit is a set of control variables.
The estimate of β is the variable of interest, and themagnitude shows the difference in these two smoothedfunctions at the cutoff, thereby capturing the effects ofpassing a CSR proposal on nonvoting peers’ sub-sequent CSR performance. As shown in panel B, theestimates of β are significantly positive both without(column (1)) and with (column (2)) controls with eco-nomic magnitudes similar to those in panel A, thusfurther confirming our baseline results.
3.3. Passage vs. ImplementationThese results demonstrate that peer firms in a com-petitive relationship follow the voting firms’ potentialCSR adoption and go on to improve their own CSRpractices. However, it is possible that some of theproposals that passed were not implemented (becausethe voting results are nonbinding) and some CSRpractices were carried out without a proposal’s beingpassed. Therefore, these results can also be interpretedas showing that the passage of a CSR proposal may, atleast, credibly signal a firm’s commitment to CSR. Weconduct a fuzzy RDD analysis to overcome the im-perfect compliance issue and disentangle the effect ofsignaling and actual adoption.Given the lack of pertinent data, we make use of
several resources to determine whether the proposalswere implemented. First, following Flammer (2015a),we define implementation based on the change of KLDscore (adjusted for the overall number) after the vote.Specifically, if the adjusted KLD score increased in theyear after the vote, we interpret it as suggestive thatthe proposal was implemented.14 Second, we defineimplementation based on another CSR data source,ASSET4, as an external validity check. As with the KLDdata, ASSET4 provides a firm’s engagement in severalenvironmental and social issues, and we focus on itsaggregate environmental and social scores. However,unlike KLD data, the ASSET4 scores for each dimensionrange fromzero to 100with amedian of 50.Wedefine anincrease of 10 of a firm’s ASSET4 score (the standarddeviation of the distribution within an industry) oneyear after the passage of a close-call CSR proposal as anindication that the proposal was likely to be imple-mented. Third, we manually collect data on the actualimplementation of those passed CSR proposals. To doso, we extensively searched various sources, includingthe corporate website, 10-K filing, third-party websitesthat follow corporate social responsibility issues orproxy votes, and the news through Google search. Wespecifically focus on the subsample of those passedproposals and consider that a proposal was passed andimplemented if (a) there was a clear indication of achange in corporate behavior, such as a news reportresponding to the proposal of the shareholders in theyear after the vote; (b) there was a corporate action,according to the news after the vote; or (c) the board ofdirectors recommended a “for” before the vote.We then apply a fuzzy RDD approach to estimate
CSR peer effects and use the threemethods described todefine the implementation of CSR. The first step offuzzy RDD is to estimate the probability of imple-mentation at the cutoff, which is 50%. The results aregiven in Table 5. The first stage of the fuzzy RDD showsthat the 50% cutoff has a significant impact on theprobability of implementation. This probability in-creased by 10% (panel A) if the voting percentage
surpassed 50%. The probability of proposal imple-mentation increases even more in other cases. If thereis a significant jump in the probability of implemen-tation, we run the second stage to obtain the estimate,adjusting for imperfect compliance. As reported inTable 5, the second-stage results are all significant andpositive with even larger economic magnitudes thanthe coefficients in Table 4. The results suggest that theeffects we have documented are robust after adjustingfor the probability of implementation.
The question naturally arises of whether an imple-mented proposal has a greater impact on the peer firms
than one that is not implemented. Unpacking thisquestion gives us some hints as to whether the effectdocumented herein is a pure signaling effect or hasmaterial meaning and impact. To answer it, we focuson all the passed proposals and then compare the re-sponses of peers, conditional on whether the focal firmimplemented the proposal or not. The results aretabulated in Table 6, inwhichwe find that the change insuch peer firms’ CSR performance is much stronger inthe subsample of implemented proposals than in thesubsample in which proposals went unimplemented.For example, the first two columns show a 0.17 increaseof adjusted KLD score in peer firms whose relatedvoting firms complied with the passed proposal. Yetthe change in adjusted KLD score is much lower in peerfirms whose related voting firms did not comply withthe passed proposal. The difference of 0.07 is significantat a level of 10%. The results hold and are even strongerfor the other compliance measures. This again suggeststhat the passage of a CSR proposal is not purely amatter of signaling, and its implementation can gen-erate significant CSR peer effects.As a caveat, we only searched for implementation
news regarding the passed proposals. As long as theprobability of implementation in the just-failing sampleis not zero, then the first-stage estimate would be bi-ased upward and the second-stage estimate would bebiased downward. However, given the robustness ofour results across the board, we do not think suchbiases will significantly change our conclusion.
3.4. Which CSR Matters?So far, we have documented an overall increase in CSRperformance of nonvoting peer firms after observingthe passage of a CSR proposal. A natural question arisesat this point: how do peer firms specifically respond tothe adoption of CSR policies? Do they more proac-tively engage in CSR by launching more initiatives, ordo they simply comply with regulations by reducingtheir production of negative societal externalities? To
Table 6. Response of Nonvoting Peers to Implementation and Nonimplementation
Notes. This table presents peer firms’ future CSR performance conditional on the compliance or implementation of the CSR proposals in thevoting firms. We compare the change of CSR scores of the peer firms in response to the compliance of the passed CSR proposals. Four measuresof compliance are used: (a) there is an increase in adjusted KLD score in the year after the vote, (b) there is an increase of more than 10 in theASSET4 social pillar score in the year after the vote, (c) there is an increase of more than 10 in the ASSET4 environmental pillar score in the yearafter the vote, and (d) there is news or information showing that the proposal was implemented. The optimal bandwidth is determined accordingto Imbens and Kalyanaraman (2012).
*, ***Significance at the 10% and 1% levels, respectively.
Table 5. Responses of Nonvoting Peers to the Passage ofa CSR Proposal: A Fuzzy RDD Approach
Second stage First stage
Panel A. KLD score
Estimate 0.95*** 0.10***t-statistic 3.60 4.49
Panel B. ASSET4 social pillar score
Estimate 0.50*** 0.43***t-statistic 2.66 6.11
Panel C. ASSET4 environmental pillar score
Estimate 0.90*** 0.24***t-statistic 2.61 3.43
Panel D. News about implementation
Estimate 0.18*** 0.67***t-statistic 3.93 25.70
Notes. This table presents peer firms’ future CSR implementation asa response to the CSR votes estimated using a fuzzy RDD approach.Specifically, the true “treatment” is a dummy variable if (a) there is anincrease in adjusted KLD score in the year after the vote (panel A),(b) there is an increase of more than 10 in the ASSET4 score (theenvironmental pillar and the social pillar) in the year after the vote(panels B and C), or (c) there is news or information showing that theproposal was implemented (panel D). The optimal bandwidth isdetermined following Imbens and Kalyanaraman (2012).
answer this question, we break down the overall KLDscore into strengths and concerns for each KLD di-mension. Items defined as strengths capture a firm’svoluntary engagement in CSR issues whereas concernscapture (potential) negative externalities producedby the firm. For example, in the environment category,strengths include environmentally beneficial productsand services (that promote the efficient use of energy),pollution prevention, recycling, clean energy, and com-munication on environmental issues (e.g., a signatory tothe Ceres Principles, a notably substantive environ-mental report, an effective internal communicationssystem for environmental best practices, etc.) as well asproperty, plans, and equipment that have an above-average environment performance. Concerns in thiscategory include hazardous waste, regulatory prob-lems, ozone-depleting chemicals, substantial emissions,agricultural chemicals, and climate change (substantialrevenues from the sale of coal or oil and its derivativefuel products).15
We conduct the same tests on peers’ following-yearCSR as before except that we replace the adjusted KLDscore with KLD strengths and concerns. As shown inpanel A of Table 7, the RDD estimate of strengths isstatistically significant at a level of 5%, and the estimateof concerns is insignificant. This may imply that theeffects on the change of passing peers’CSR seem to comefrom their focus on strengths (launching new initiativesaimed at strengthening firms’ social engagement) ratherthan on concerns (reducing negative externalities).Another, related, question is whether the adoption
of peer firms’ follow-up CSR is a mimicking strategyor a differentiation strategy. In other words, do peerfirms adopt CSR practices similar to the voting firm’smarginally passed proposal (mimicking), or do theyadopt a different type of CSR so as to avoid directcompetition (i.e., differentiation)? As mentioned inour theoretical framework, peer firms may adoptdifferent strategies depending on their position in theproduct market (i.e., whether they are market leaders
Table 7. Breakdown of KLD Score
Panel A. Following-year KLD strengths score and concerns score of nonvoting peers
Pass vs. fail
(1) Strengths score (2) Concerns score
RDD estimate 0.07** −0.02t-statistic 2.16 −0.79
Panel B. Environment-related proposals and nonvoting peers’ following-year KLD subscores
Notes. This table presents the effects of the passage of a CSR proposal on (nonvoting) peers’ following-year CSR performance by breaking downthe KLD score into different dimensions. Panel A shows the RDD estimates for the adjusted KLD strengths score (column (1)) and the adjustedKLD concerns score (column (2)). Panel B shows the RDD estimates for three major subdimensional KLD scores: environment (column (1)),employee relations (column (2)), and workforce diversity (column (3)). Panel C shows the RDD estimates on (nonvoting) peers’ following-yearworkforce diversity scores following the passage of an environment-related CSR proposal conditional on whether the (nonvoting) peer is an industryfollower or a leader (defined in different ways). In column (1), follower and leader are defined based on their market shares (sales revenue): a peerfirm is defined as a follower if its market share is lower than the voting firm’s market share prior to the voting and is otherwise defined as leader.Similarly, follower and leader are defined by their profitability (ROA) relative to the voting firm prior to the voting in column (2) and definedbased on their adjusted environmental score relative to the voting firm prior to the voting in column (3). We follow Imbens and Kalyanaraman(2012) and estimate the effects of the passage of a close-call CSR proposal using local linear regression with the optimal bandwidth. Variabledefinitions are provided in Online Appendix B.
*, **, ***Significance at the 10%, 5%, and 1% levels, respectively.
or followers). To test these, we further break downthe voting firms’ CSR proposals into different typesby classifying the proposals as environment-relatedproposals, workforce diversity–related proposals, orproposals related to employee relationships becausethese are the most important aspects of CSR and covermost firms. We then break down the overall adjustedKLD score into a few subdimensions, such as environ-ment, employee relations, and workforce diversity, tomatch the proposal type and analyze these subsampleswith respect to different types of CSR proposals.16
For reasons of brevity,we report only the results of thepassage of an environment-related proposal on peer firms’CSR as we delineate in the theoretical framework. Asshown in panel B of Table 7, we indeed find evidencethat peer firms match the voting firm’s specific CSRstrategies. The difference in the environmental scorebetween passing and failing peers is significant and hasthe greatest magnitude of all subdimensional CSRscores, supporting the mimicking hypothesis.17 We alsoobserve some evidence of differentiation: for example,the passage of an environment-related proposal alsorelates to an increase in peer firms’ diversity score al-though the magnitude of the increase is much smaller.This can be interpreted as peer firms adopting a mim-icking strategy, which appears to be economically moreimportant than a differentiating strategy.
Moreover, based on our argument, peer firms thatare industry followers will be more likely to adopt adifferentiating strategy, besides mimicking the votingfirm’s specific CSR practice. We define industry fol-lowers and leaders based on their product-market shares(using sales revenue), profitability (ROA), and the CSRscore (using the environmental score as an example)prior to the vote, all relative to the voting firm. For ex-ample, if a peer firm’s environmental score is lower thanthat of its voting firm, it is considered a follower; oth-erwise, it is considered a leader. We then test the “dif-ferentiating strategy” conjecture by performing the RDDtests on nonvoting peers’ following-year workforcediversity score following the passage of a close-callenvironment-related proposal conditional on whetherthe peer firm is an industry follower or a leader. Asshown in panel C of Table 7, the differentiation effectsshow up only in the subsamples of industry followersbased on all three criteria. This is consistent with ourtheoretical arguments and the conjectures by the in-dustrial organization literature.
3.5. Disentangling the Mechanisms of CSRPeer Effects
Our theoretical arguments are based on the idea thata firm’s CSR practices create competitive pressure on itspeer firms, which triggers the CSR peer effects we havedocumented. However, as mentioned earlier, an alter-native explanation of the aforementioned results may be
that the spillover of CSR among product-market peers ispropagated by financial intermediaries. If the peer effectsof CSR are driven by financial intermediaries’ propa-gation, we should expect similar reactions from peerswithout any economic links—for example, peer firmscovered by a common financial analyst or held by thecommon institutions. The external forces may propagateCSR although the voting firm’s CSR strategy has littleimpact on the utility of the peer firms.We investigate the propagation explanation inpanelA
of Table 8 by conducting an analysis similar to that we
Table 8. Mechanisms of Peer Effects: DisentanglingPropagation and Competition
Panel A. Redefining peer firms
Pass vs. fail
Common analystCommon institutional
investors
Estimate 0.05 0.01t-statistic 1.62 0.47
Panel B. Peer effects of CSR: Conditional on external propagation
Panel C. Peer effects of CSR: Conditional on competitive pressure
Pass vs. fail
High competitive pressure Low competitive pressure
Estimate 0.17** 0.05t-statistic 2.02 0.84
Notes. This table reports RDD results disentangling propagation andcompetition explanations. In panel A, we perform the RDD tests onsamples of alternative definitions of peers using common analystsand common institutional investors. In panel B, we perform the RDDtests on subsamples of product-market peers divided by commonanalysts and by common institutional investors. Common analysts(or common institutional investors) are a binary variable that equalsone if two firms are covered by the same financial analyst (or held bythe same institutional investor). In panel C, we perform RDD tests onsubsamples of product-market peers divided by the voting firm’scompetitive pressure (high vs. low), which measures similaritiesbetween a change in a firm’s product space and an aggregatechange in the competitors’ product description. This competitivepressure measure is taken from the fluidity measure as describedinHoberg et al. (2014), which argues that, when fluidity is greater, thefirm’s products are more similar to its peers’, and thus, the competitivethreat is greater. We follow Imbens and Kalyanaraman (2012) andestimate the effects of the passage of a close-call CSR proposal usinglocal linear regression with the optimal bandwidth. Variabledefinitions are provided in Online Appendix B.
did in Table 4, comparing CSR scores of passing andfailing peers, but we use alternative definitions of peerfirms based on whether they are covered by commonanalysts or held by common institutional investors.That is, a firm is considered a peer of the voting firm ifthese two firms have at least one analyst who is thesame or the same institutional investor. This alternativedefinition also helps shed light on the mechanisms ofpeer effects. However, the RDD results for peers basedon common analysts and common institutional in-vestors, as shown in columns (1) and (2), are all in-significant, indicating that even in firms with a higherinfluence of financial intermediaries’ propagation, thepeer effects of CSR are absent. Propagation, therefore,is unlikely to be the driving force for CSR peer effects.
To cross-validate the results in panel A, we alsodivide our sample of product-market peers (using theHoberg–Phillips classification) into two subsamplesbased on whether the peer firm and the voting firm arecovered by a common analyst or held by the sameinstitutional investor. As shown in panel B, we againfail to find statistically and economically significantdifferences between the subsamples. Combining allthese results in panels A and B with our earlier resultsbased on product-market peers, we see that CSR peereffects are mostly driven by competitive pressure andare absent in those without competitive relationshipsalthough these are more likely to induce propagationby financial intermediaries (common analysts andcommon institutional investors).
If competitive pressure is driving CSR peer effects,peer firms’ responses should be stronger when they arein amore competitive relationship with the voting firm.To test this hypothesis, we define competitive pressure,which captures the similarity of two competing firms’products by measuring changes in the competingfirms’ products relative to the focal firm’s products(Hoberg et al. 2014). This competitive-pressure variableis constructed according to the way in which com-petitors change the wording used to describe theproduct, which overlaps with the focal firm’s vocab-ulary of the product description section in the 10-Ks.
When a firm’s products aremore similar to those of itspeers, the competitive pressure among its peers is greater;thus, peer firms may have a stronger desire to mimictheir competitors’ commitment to CSR.18 Specifically, wedivide the peer firms into two groups according to theirassociated voting firms’ competitive pressure level in theyear before the vote. A high-competitive-pressure groupincludes peers of the voting firms whose competitive-pressure scores are higher than those of the samplemedian among all voting firms. We expect that peerfirms in the high-competitive-pressure group are subjectto more pressures than those in the low-competitive-pressure group. The results of this competitive-pressuremeasure are reported in panel C of Table 8. Consistent
with this conjecture, we find that the CSR peer effectis present mainly in the high-competitive-pressuregroup—namely peer firms whose products are similarto their competing voting firms’ products and, thus, facegreater pressure to engage in CSR.
3.6. RobustnessIn addition to using a global polynomial approach andadjusting for imperfect compliance issues, we conducta few robustness tests with alternative peer samples andplacebo tests. First, we replace CSR proposals with(probable) actual CSR practices. To do so, we manuallycollected data fromFactiva and an online news search ona peer firm’s actual environmental initiatives, such asdeveloping an environmental R&D project.19 We thenverified these news datawith data from theASSET4U.S.data set. The overlapped part (when the company ac-tually adopted a certain CSR practice as reported by thenews andwhen this information is adjusted inASSET4’sraw data) forms our sample for testing the actual CSRpractices of peer firms. With these data, we conduct thesame RDD tests, and our results remain.Second, we redefine our peer firms according to
(a) the three-digit SIC and (b) a randomly selectedsample with 50 arbitrarily chosen peers from a votingfirm’s peer pool.20 Consistent with evidence accordingto the Hoberg–Phillips definition, we find similar re-sults in these alternative peer groups.Third, we conduct two placebo tests to ensure that the
documented evidence reflects the peers’ response toCSR proposals and to rule out potential confoundingeffects. We first replace the competing peers withsimilar nonpeer firms,21 and from this, we find thatthe differences between passing nonpeers’ and failingnonpeers’ adjusted KLD scores are not statisticallysignificant and that the point estimates are muchsmaller. This indicates that a voting firm’s potentialadoption of CSR does not affect nonpeer firms, and theobserved differences in peer firms’ CSR are induced bythe voting firm’s CSR strategy rather than by otherconfounding factors. In addition, we purify the CSRvotes by excluding the events if there is a vote ona corporate governance proposal on the same day in theannual meeting.We find that the estimated difference inCSR remains statistically significant, indicating that ourprevious findings were not entirely driven by con-founding proposals regarding corporate governance.These results are reported in Online Appendix F.
4. The Value Implications of CSRPeer Effects
The aforementioned findings suggest the existence ofstrong CSR peer effects. As a final step, we study thevalue implications of such peer effects by examining the
peer firms’ stock market reactions to the voting firms’passage of close-call CSR proposals. If the passage ofa CSR proposal by voting firms forms a competitivethreat to its peers, we would expect the investors of thecompeting peer firms to react negatively to such news.However, the strategic response from peers shouldrelieve the pressure and weaken the negative reactionof the investors.
We gauge peer firms’ stock market reaction to thevoting firm’s increased CSR engagement by calculatingtheir CARs over the three-day event window (−1, +1)around the voting date using a market model.22 Weconduct similar RDD analysis between passing peersand failing peers, and the results are reported in Table 9.Panel A shows that, in general, product-market peersexperienced significantly lower CARs. That is, the dif-ference in CARs between passing peers and failing peersis negative. To confirm the result of stock market re-actions, we also examine peer firms’ changes in theproduct market share one year after the vote. Becausevoting firms will implement the CSR proposal in yeart+1, we expect the competitive threats on market sharesto be materialized in year t+2 (Flammer 2015a).Therefore, we focus on changes in peer firms’ marketshare from year t+1 to year t+2. Consistent with theevidence from the stock market (column (1)), we findthat the passing peers on average experience a decreaseinmarket share relative to the failing competing peers incolumn (2). These results suggest that shareholders ofcompetitors interpret the voting outcome as bad news,which is consistent with the notion that a firm’s adop-tion of CSR can attract a greater market share, thushurting its competitors.
Because the stock market is forward-looking, thenegative effect on passing peers’ CARs should bestronger if these peer firms’ investors perceive theircompanies as being less likely to catch up in terms ofCSR in the future. We further use four proxies to testthis “lost in competition” prediction. First, we directlytest the change in CSR difference between the votingfirm and its peer firm after the vote. A significant in-crease in the CSR of the voting firm relative to that of itspeers suggests that the vote can be regarded as “ef-fective” in a competition relationship. If the voting firmwas already a leader in CSR competition before thevote, the passage of the proposal will enlarge itscompetitive advantage over its peers. If the voting firmlags behind its peers in CSR, the passage of the pro-posal will reduce the gap with its peers. We, thus,partition our sample into two groups: those with asignificant increase in the CSR gap (the voting firm’sCSRminus the peer firm’s CSR) and thosewithout suchan increase. As shown in panel B of Table 9, the neg-ative and significant effect is present only in the sub-sample of peers that experience an increase in the gapbetween their CSR and their voting firms’ CSR.
Second, we test the conjecture that if peer firms aremore financially constrained, they are less likely tohave the capacity to catch up by investing in CSR. We
RDD estimate CAR [−1, +1] −0.51%*** 0.01%t-statistic −3.06 0.07
Panel C. Financial constraints
Pass vs. fail
High Low
RDD estimate CAR [−1, +1] −0.54%** −0.17%t-statistic −2.38 −1.24
Panel D. Competitive pressure
Pass vs. fail
High Low
RDD estimate CAR [−1, +1] −1.33%*** −0.29%t-statistic −5.48 −0.6
Panel E. Board recommendations
Pass vs. fail
For (not against) Against
RDD estimate CAR [−1, +1] −0.93%*** −0.13%t-statistic −4.72 −1.07
Notes. This table reports the regression discontinuity design estimatesof peer firms’ cumulative abnormal returns (cumulative abnormalreturns) on the sample of product-market peers. Panel A reportsthe results for three-day cumulative abnormal returns around thecorporate social responsibility vote and the one-year-later change inproduct-market shares for the overall sample. We measure thethreat of competition by classifying sample firms based on (a)whether the corporate social responsibility gap between the votingfirm and its peer firms is further enlarged after the passage ofa corporate social responsibility proposal (panel B), (b) whetherthe level of financial constraints of competing peers is high or low(using Hadlock and Pierce 2010, panel C), (c) competitive pressurebetween the voting firm and peer firms (panel D), and (d) whethervoting firms’ board vote “for” for the corporate social responsibilityproposal (ex ante probability of implementing the proposal, panel E).We follow Imbens and Kalyanaraman (2012) and estimate the effectsof the passage of a close-call corporate social responsibility proposalusing local linear regression with the optimal bandwidth. Variabledefinitions are provided in Online Appendix B.
**, ***Significance at the 5% and 1% levels, respectively.
capture a peer firm’s financial constraint using themeasure developed by Hadlock and Pierce (2010)(hereafter HP),23 in which a higher index value in-dicates greater financial constraint (in terms of the peerfirm’s ability to catch up with respect to CSR in-vestment). As shown in panel C of Table 9, the negativeeffect shows up only in the subsample of high-HPpeers, that is, peers with greater financial constraints.This is consistent with our conjecture.
The immediate response from the market also de-pends on the competitive pressure on the peer firmand the ex ante likelihood of a CSR proposal’s beingimplemented as perceived by peer firms’ investors.The loss will translate into immediate market reactiononly when the investors believe the pressure fromcompetition is high and the signal of the voting firmimplementing the passed CSR proposal is credible.Therefore, we conduct a further test by splitting thesample into two subgroups based on the competitivepressure of the peer firm as measured in Table 8. Theresults in panel D of Table 9 show that only peers withhigh competitive pressures experience negative marketreactions.
Finally, we split our sample based on whether theboard recommended “yes” (or “for”) to the proposalbefore the vote as this is ex ante observable at the timeof the vote and increases the likelihood of the focal firmactually implementing the proposal. We expect thepeer firms’ reactions to be stronger if their investorsbelieve that the focal firms would implement the CSRimprovement. Data on board recommendation prior tothe vote are obtained from ISS U.S. voting outcomesand aremanuallymatchedwith our original CSR votes.As shown in panel E of Table 9, the negative marketreaction shows up only in the subsample whose boardsrecommended yes (or not against) to the proposal priorto the vote; this is again consistent with our conjecture.These results are also consistent with the empiricalliterature on the value-enhancing effect of CSR (e.g.,Flammer 2015a, Ferrell et al. 2016, Lins et al. 2017) inthat they show one company’s gain in value byadopting CSR is its competitors’ loss in value.
5. ConclusionIdentifying the causal effect of peer effects amongcorporations is notoriously difficult. Equally difficult isidentifying whether this peer effect has a social com-ponent. Existing evidence shows that CSR is de-termined largely by a firm’s operating environment(Ioannou and Serafeim 2012, Liang and Renneboog2017b). Despite the growing literature on the de-terminants and value consequences of CSR, little isknown about the influence of other firms on a firm’sCSR. In this paper, we present evidence on the peereffects of CSR using a regression discontinuity designapproach.We rely on CSR proposals that pass or fail by
a small margin of votes during shareholder meetings asa source of locally exogenous variation in CSR com-mitment. By focusing on the reactions of peer firmscompeting in the same product markets (Hoberg andPhillips 2016) to this potential adoption of CSR, ourpaper provides new insight into the motivations forcorporate engagement in social issues and sheds lighton the recent phenomenon of the surge in CSR.We find strong evidence of the adoption of CSR
practices among peer firms following the passage ofa voting firm’s CSR proposal. On average, the differ-ence in CSR scores between passing peers and failingpeers is 0.16 point. Stronger results are found forpassed proposals that were actually implemented,which we call CSR peer effects. These results are robustfor alternative samples of peers with economic links.These peer firms actively follow (more strengths ratherthan fewer concerns) and specifically (in the samedomain) follow the voting firm’s signaled commitmentto CSR. However, peers that are market followers mayalso adopt CSR in different domains in addition tomimicking the voting firm’s CSR. Such effects are notlikely to be attributed to financial intermediaries’propagation as we don’t find similar results in samplesof common analysts and common institutional in-vestors but rather are more consistent with a utility-based explanation. We further find that on the daysaround a shareholder meeting, a close-call CSR pro-posal is related to lower CARs in competing peers,especially those that have more difficulty catching up.All these results are consistent with a competition-based explanation. As a whole, our analysis iden-tifies an important, yet so far unexplored, determinantof CSR practice and highlights the importance ofstrategic interaction in understanding peer effects ineconomic activity.
AcknowledgmentsThe authors thank the editor, GustavoManso, an anonymousassociate editor, two anonymous reviewers, Renee Adams,Rui Albuquerque, Tamas Barko, Bo Becker, Ye Cai, PeterCziraki, Jay Dahya, Elroy Dimson, Ljubica Djordjevic,Ofer Eldar, Caroline Flammer, Thierry Foucault, NicolaeGarleanu, Stuart Gillan, Sadok El Ghoul, JarradHarford, BingHan, Oguzhan Karakas, Michael Kisser, Philipp Krueger,Inessa Liskovich, M. Fabricio Perez, Patrick Verwijmeren,Michael Weisbach, and Andrew Winton, as well as seminarparticipants at Chinese University of Hong Kong, CheungKong Graduate School of Business, Erasmus School of Eco-nomics, National University of Singapore, Norwegian Schoolof Economics, Queens University, Singapore ManagementUniversity, Swiss Finance Institute–Lugano, University ofManchester, University of Toronto, and Wilfrid LaurierUniversity for helpful discussions and suggestions. The au-thors have benefited from the comments of participants at the28th Australasian Finance & Banking Conference, the 2016Asian Bureau of Finance and Economic Research Conference,
the 2016 China International Conference in Finance, and the2017 AFAAnnual Meeting. We also acknowledge the ZephyrPrize for best corporate finance paper from the 28th Aus-tralasian Finance & Banking Conference. All errors are ourown.
Endnotes1As argued by Scharfstein and Stein (1990), an unprofitable decisionis not as bad for a manager’s reputation when others make the samemistake (the “sharing the blame” effect). Therefore, even if a man-ager’s private information suggests that investing in CSR will pro-duce negative expected value, the manager may still pursue it ifothers have adopted similar strategies. Bustamante and Fresard(2017) further argue that when firms’ managers have imperfect in-formation about fundamentals, they may use peers’ investment asa source of information and make similar investments.2The need for shareholder proposals on CSR arises from the fact that,with limited firm resources, self-interested managers are not alwayswilling to voluntarily invest in CSR even though doing somay benefita broader group of stakeholders, possibly including shareholders(Flammer 2015a).3Online Appendix A shows two examples of voting on CSR pro-posals that help illustrate our empiricalmethod. The example in panelA is a case of a marginally rejected CSR proposal during the MasseyCompany shareholder meeting on May 19, 2010. The proposal oncarbon dioxide emissions was rejected with 45.6% supporting votes.The example in panel B is a case of a marginally approved CSRproposal during the IDACORP, Inc., shareholder meeting onMay 21,2010. The proposal on reducing total greenhouse gas emissions waspassed with 51.2% supporting votes. Our objective is to examine thedifference in nonvoting peer-firm reactions. In our sample, MasseyEnergy has 49 peer firmswith an average adjustedKLD score of −0.62in the year after the vote, that is, 2010. IDACORP, Inc., has 55 peerswith an average adjusted KLD score of −0.20.4The text-based product-market peer data can be obtained fromhttp://cwis.usc.edu/projects/industrydata/. Refer to Hoberg andPhillips (2010) for detailed descriptions.5Our stock and bond issuance data comes from the SDC database.The M&A announcement data are obtained from the Zephyr andSDC databases. Dividend payment data are obtained from the CRSP.6Our sample has fewer votes than Flammer (2015a) because the datacoverage of the Hoberg–Phillips database is smaller than that of theCompustat universe. Nevertheless, as we show later, our results arerobust to different peer definitions, such as the SIC, which includesbroader coverage in Compustat.7We exclude corporate governance from our CSR performanceconstruction because it is perceived as a mechanism for mitigatingconflict between principles and managers (Shleifer and Vishny 1997)rather than a concern about other stakeholders, such as the com-munity and employees. We also exclude the dimension of productsafety and quality because it is more likely to be subject to legalrestrictions and regulations and because it is hardwired into product-market competition, which may obscure the interpretation of ourresults based on product-market peers.8 Several papers have used the regression discontinuity design, in-cluding Cuñat et al. (2012), Flammer (2015a), and Bradley et al. (2017).9We conduct the tests using optimal bandwidth following Imbensand Kalyanaraman (2012). The number of observations varies acrossdifferent variables because the optimal bandwidths are different. Ourresults do not change when we test the preexisting difference withinsome other specified small margins, such as (48%, 52%) or (49%, 51%).10We define implementation according to three criteria. First of all,we define the implementation of an increase in adjusted CSR scorefromMSCI ESG STATS (formerly KLD database). For each dimension,
implementation is defined as an increase in related CSR dimensions.Second, implementation is defined as the increase in ASSET4 score.Finally, we manually checked the news for passed CSR proposals anddefine implementation as (a) there is a news report about the proposalimplementation, (b) there is a news report on the corporate website, or(c) the management team voted “for” before the votes.11For these baseline specifications, we test the discontinuity at themajority threshold, that is, 50%. For placebo tests, we conduct thesame analysis at other cutoffs (e.g., 45%, 35%, 55%, 65%, etc.); we findno evidence of discontinuity for the subsequent CSR activities, whichsupports our argument that the effects on peer firms’ CARs aregenerated by the exogenous increase of the CSR level of the votingfirm caused by marginally passing the CSR proposal.12According to Imbens and Lemieux (2008), the choice of kernel haslittle impact on the estimation in practice although using a rectan-gular kernel is more common.13The global polynomial approach, however, fails to take into con-sideration RDD’s strong locality and weak externality, which areimportant features of the approach (Bakke and Whited 2012).14The rule also applies to firms that rejected a CSR proposal.15For a detailed description of strengths and concerns in each cate-gory, please refer to https://wrdsweb.wharton.upenn.edu/wrds/support/Data/_001Manuals%20and%20Overviews/_070KLD/_001General/_002Rating%20Criteria%20Definitions.pdf.cfm.16 In addition to the environment dimension, as explained in the text,the employee-relations dimension considers company engagement intreating a unionized workforce fairly, maintaining a consistent no-layoff policy, implementing a cash profit-sharing program, employeestock-option plans, retirement benefits, health and safety programs,and so forth. The workforce-diversity dimension considers whethera company engages in promoting a female or minority CEO andboard of directors; provides childcare, elder care, or flexible time;contracts with women and minorities; develops innovative hiringprograms for the disabled; puts into place progressive policies re-garding gay and lesbian employees; and so forth.17 In unreported results on peer firms associated with proposalsrelated to workforce diversity, the reaction to the diversity di-mension is the strongest. Similarly, in a subsample that includesonly proposals related to employee relations, only the estimate ofthe employee relationship score is significant; that is, peer firmssignificantly improve their engagement only in issues related toemployee relations.18We do not use a traditional Herfindahl–Hirschman index measureor market share because peers are identified by the product ratherthan by a specified industry. Instead, we use a measure called“fluidity,”which is obtained from 10-K files and compiled by Hoberget al. (2014), and it shows the competitive dynamics between a firmand its peers identified through a text-based analysis to measurecompetitive pressure. More detailed description of this variable canbe found in Online Appendix B: Variable Definitions.19 In particular, we focus on (a) whether the company reported on itsimpact on biodiversity, on activities to reduce its impact on nativeecosystems and species, and on the biodiversity of protected andsensitive areas (report on biodiversity impact: yes or no); (b) whetherthe company took any initiative to restore or protect native ecosys-tems or the biodiversity of protected and sensitive areas (restoringecosystems: yes or no); (c) whether the company report made pro-active environmental investments or expenditures (environmentalinvestment initiative: yes or no); (d) whether it had been reported inthe news that the company invested in R&D on new environmentallyfriendly products or services intended to limit the amounts of emis-sions and resources needed during product use (yes or no); (e) whetherthe company invested in R&D on new environmentally friendlyproducts (environmental R&Ds: yes or no); and (f) whether the com-pany trained its employees on environmental issues (environmental
training on employees: yes or no). We further supplement the newsinformationwith two ratings fromASSET4’s environmental R&D scoreand renewable/clean energy products score, both of which range fromzero to 100.20This is because we notice that the number of competing peers(based on the Hoberg–Phillips definition) in our sample differs sig-nificantly across different voting firms—from one to 272. To addressthe concern that our results might be biased owing to an imbalance inthe distribution of peer firms, we conduct similar tests on sampleswith a predetermined number of peer firms. In unreported tables, wealso randomly select 20, 30, or 40 peers. The results are consistent andare available upon request.21To do so, we replace each nonvoting peer with onematched firm byrequiring the two firms to be the same in size, market-to-book ratio,and leverage ratio decile. If more than one non-peer firm is found, wekeep the one closest in size to the peer firm.22We also validate the results based on the market model by esti-mations using the Fama and French (1993) three-factor and Carhart(1997) four-factor models. The results are available upon request.23 Several measures of financial constraints exist, but according toHadlock and Pierce (2010), most of these suffer from too much noisefrom firm attributes other than size and age. Therefore, we use thefinancial-constraints index developed by HP and divide our com-peting peer sample into two groups, one with high financial con-straints and the other with low financial constraints. To check therobustness of this result, we conduct the same analysis on the sub-samples partitioned by alternative measures of financial constraints,including the Whited and Wu (2006) index and an indicator ofwhether the nonvoting peer firm distributed dividends in year t−1(Denis and Sibilkov 2010). The results are similar to those using theHP index.
ReferencesAgarwal S, Mikhed V, Scholnick B (2016) Does inequality cause fi-
nancial distress? Evidence from lottery winners and neighboringbankruptcies. Working paper, Georgetown University, Wash-ington, DC.
Agarwal S, Qian W, Zou X (2017) Thy neighbor’s misfortune: Peereffect on consumption. Working paper, Georgetown University,Washington, DC.
Albuquerque R, Koskinen Y, Zhang C (2018) Corporate social re-sponsibility and firm risk: Theory and empirical evidence.Management Sci. Forthcoming.
Bakke T-E, Whited T (2012) Threshold events and identification: Astudy of cash shortfalls. J. Finance 67(3):1083–1111.
Bailey M, Cao R, Kuchler T, Stroebel J (2018) The economic effects ofsocial networks: Evidence from the housing market. J. PoliticalEconom. 126(6):2224–2276.
Bénabou R, Tirole J (2010) Individual and corporate social re-sponsibility. Economica 77(305):1–19.
Bradley D, Kim I, Tian X (2017) Do unions affect innovation? Man-agement Sci. 63(7):2251–2271.
Bursztyn L, Ederer F, Ferman B, Yuchtman N (2014) Understandingmechanisms underlying peer effects: Evidence from a field ex-periment on financial decisions. Econometrica 82(4):1273–1301.
BustamanteMC, Frésard L (2017) Does firm investment respond to peers’investment? Working paper, University of Maryland, College Park.
Carhart MM (1997) On persistence in mutual fund performance. J.Finance 52(1):57–82.
Chatterji AK, Levine DI, Toffel MW (2009) Howwell do social ratingsactually measure corporate social responsibility? J. Econom.Management Strategy 18(1):125–169.
Cheng I-H, Hong H, Shue K (2016) Do managers do good with otherpeople’s money? Working paper, Dartmouth College, Hanover, NH.
Cronqvist H, Yu F (2017) Shaped by their daughters: Executives,female socialization, and corporate social responsibility. J. Fi-nancial Econom. 126(3):543–562.
Cuñat V, Gine M, Guadalupe M (2012) The vote is cast: The effect ofcorporate governance on shareholder value. J. Finance 67(5):1943–1977.
Denis DJ, Sibilkov V (2010) Financial constraints, investment, and thevalue of cash holdings. Rev. Financial Stud. 23(1):247–269.
Deng X, Kang JK, Low BS (2013) Corporate social responsibility andstakeholder value maximization: Evidence from mergers. J. Fi-nancial Econom. 110(1):87–109.
Dessaint O, Foucault T, Frésard L, Matray A (2018) Noisy stock pricesand corporate investment. Working paper, University of Tor-onto, Toronto.
DhaliwalDS, Li OZ, TsangA, Yang YG (2011) Voluntary nonfinancialdisclosure and the cost of equity capital: The initiation of cor-porate social responsibility reporting. Accounting Rev. 86(1):637–667.
Di Giuli A, Kostovetsky L (2014) Are red or blue companies morelikely to go green? Politics and corporate social responsibility. J.Financial Econom. 111(1):158–180.
Dimson E, Karakas O, Li X (2015) Active ownership. Rev. FinancialStud. 28(12):3225–3268.
Dowell G, Hart S, Yeung B (2000) Do corporate global environmentalstandards create or destroy market value?Management Sci. 46(8):1059–1074.
Edmans A (2011) Does the stock market fully value intangibles?Employee satisfaction and equity prices. J. Financial Econom.101(3):621–640.
El Ghoul S, Guedhami O, Kwok CCY, Mishra DR (2011) Does cor-porate social responsibility affect the cost of capital? J. BankingFinance 35(9):2388–2406.
Fama EF, French KR (1993) Common risk factors in the returns onstocks and bonds. J. Financial Econom. 33(1):3–56.
Ferrell A, Liang H, Renneboog L (2016) Socially responsible firms. J.Financial Econom. 122(3):585–606.
Flammer C (2015a) Does corporate social responsibility lead to su-perior financial performance? A regression discontinuity ap-proach. Management Sci. 61(11):2549–2568.
Flammer C (2015b) Does product market competition foster corpo-rate social responsibility? Evidence from trade liberalization.Strategic Management J. 36(10):1469–1485.
Foucault T, Frésard L (2014) Learning from peer firms’ stock pricesand corporate investment. J. Financial Econom. 111(3):554–577.
Friedman M (1970) The social responsibility of business is to increaseits profits. New York Times Magazine (September 13), 122–126.
Georgarakos D, Haliassos M, Pasini G (2014) Household debt andsocial interactions. Rev. Financial Stud. 27(5):1404–1433.
Grennan JP (2018) Dividend payments as a response to peer influ-ence. J. Financial Econom. 131(3):549–570.
Hadlock CJ, Pierce JR (2010) New evidence on measuring financialconstraints: Moving beyond the KZ index. Rev. Financial Stud.23(5):1909–1940.
Hoberg G, Phillips G (2010) Product market synergies and compe-tition in mergers and acquisitions: A text-based analysis. Rev.Financial Stud. 23(10):3773–3811.
Hoberg G, Phillips G (2016) Text-based network industries and endog-enous product differentiation. J. Political Econom. 124(5):1423–1465.
Hoberg G, Phillips G, Prabhala N (2014) Product market threats,payouts, and financial flexibility. J. Finance 69(1):293–324.
Hong HG, Liskovich I (2016) Crime, punishment, and the value ofcorporate social responsibility. Working paper, Columbia Uni-versity, New York.
Hong HG, Kubik JD, Scheinkman JA (2012) Financial constraints oncorporategoodness.Workingpaper,ColumbiaUniversity,NewYork.
Hsu H-C, Reed AV, Rocholl J (2010) The new game in town: Com-petitive effects of IPOs. J. Finance 65(2):495–528.
Imbens G, Lemieux T (2008) Regression discontinuity designs: Aguide to practice. J. Econometrics 142(2): 615–635.
Imbens G, Kalyanaraman K (2012) Optimal bandwidth choice for theregressiondiscontinuity estimator.Rev. Econom. Stud. 79(3):933–959.
Ioannou I, Serafeim G (2012) What drives corporate social perfor-mance? The role of nation-level institutions. J. Internat. Bus. Stud.43(9):834–864.
Kaustia M, Knüpfer S (2012) Peer performance and stock marketentry. J. Financial Econom. 104(2):321–338.
Kaustia M, Rantala V (2015) Social learning and corporate peer ef-fects. J. Financial Econom. 117(3):653–669.
KitzmuellerM, Shimshack J (2012) Economic perspectives on corporatesocial responsibility. J. Econom. Literature 50(1):51–84.
Leary MT, Roberts MR (2014) Do peer firms affect corporate financialpolicy? J. Finance 69(1):139–178.
Lee DS, Lemieux T (2010) Regression discontinuity design in eco-nomics. J. Econom. Literature 48(2):281–355.
Liang H, Renneboog L (2017a) Corporate donations and shareholdervalue. Oxford Rev. Econom. Policy 33(2):278–316.
Liang H, Renneboog L (2017b) On the foundations of corporate socialresponsibility. J. Finance 72(2):853–910.
Lins K, Servaes H, Tamayo A (2017) Social capital, trust, and firmperformance: The value of corporate social responsibility duringthe financial crisis. J. Finance 72(4):1785–1824.
Manescu C (2011) Stock returns in relation to environmental, socialand governance performance: Mispricing or compensation forrisk? Sustainable Develop. 19(2):95–118.
Servaes H, Tamayo A (2013) The impact of corporate social re-sponsibility on firm value: The role of customer awareness.Management Sci. 59(5):1045–1061.
Servaes H, Tamayo A (2014) How do industry peers respond tocontrol threats? Management Sci. 60(2):380–399.
Shleifer A, Vishny R (1997) A survey of corporate governance. J.Finance 52(2):737–783.
Smit HTJ, Trigeorgis L (2006) Strategic planning: Valuing andmanaging portfolios of real options. R&D Management 36(4):403–419.
Stoughton N, Wong KP, Yi L (2017) Competitive corporate socialresponsibility. Working paper, Vienna University of Economicsand Business, Vienna.
