Journal of Business Ethics

https://doi.org/10.1007/s10551-020-04496-3

ORIGINAL PAPER

Do Corporate Social Responsibility Reports Convey Value Relevant

Information? Evidence from Report Readability and Tone
Shuili Du1 · Kun Yu2

Received: 3 October 2019 / Accepted: 3 April 2020

© Springer Nature B.V. 2020

Abstract
Corporate social responsibility (CSR) reporting is becoming mainstream, yet there is limited research on whether and how
CSR reports communicate value relevant information. We examine the effects of CSR report readability and tone on future
CSR performance and the market reaction around the release of CSR reports. Using a hand-collected dataset of Fortune
500 companies that published stand-alone CSR reports from 2002 to 2014, we find that 1-year-ahead CSR performance is
positively associated with the changes in both CSR report readability and tone, suggesting that more readable text and more
optimistic tone in a firm’s CSR report are indicative of better future CSR performance. Furthermore, consistent with the
view that CSR reports communicate important value relevant information to the market, we document significant market
reactions to report readability and tone around the release of CSR reports. Additional analyses suggest that CSR report
readability enhances the association between the abnormal returns and the change in CSR report tone, and that the market
reaction to CSR report readability is more pronounced for firms with lower analyst following and higher financial opacity.
Taken together, our results substantiate the important roles of CSR report readability and tone in communicating future CSR
performance and imparting value relevant information to the market.

Keywords Corporate social responsibility · Corporate social responsibility reporting · Investor reaction · Textual analysis ·
Readability · Tone

Introduction more firms now release stand-alone CSR reports,1 which

provide comprehensive and in-depth information about
Disclosure of corporate social responsibility (CSR) activi- firm-level performance in various social, environmental,
ties has increased significantly over the years. Across the and governance-related domains. According to the Govern-
globe, CSR reporting requirements have more than doubled ance and Accountability Institute (2018), 85% of S&P 500
since 2013, with governments, financial market regulators, companies published CSR reports in 2017, up from less than
and stock exchanges issuing most of the reporting guidelines 20% in 2011.
(KPMG 2016). At the same time, both socially responsible CSR reporting is different from financial reporting in
investment fund and mainstream investors increasingly eval- several ways. While financial reporting is mandatory, veri-
uate and incorporate firms’ CSR performance when making fiable, and enforced through methods including external
investment decisions (CDP 2019; PWC 2019). More and audit, litigation, and regulatory oversight, CSR reporting is
voluntary in most countries, largely unregulated, and does
not have a widely enforced reporting framework (Perrini
* Kun Yu
kun.yu@umb.edu 2006; Tschopp and Huefner 2015). In addition, while finan-
cial reporting targets the investor community and focuses
Shuili Du
shuili.du@unh.edu primarily on financial data, the targeted audiences of CSR
reporting consist of various stakeholders, such as customers,
1
Peter T. Paul College of Business and Economics, University 1
Different terms are used to refer to company reports that communi-
of New Hampshire, 10 Garrison Avenue, Durham,
cate social and environmental performance, such as corporate social
NH 03824, USA
responsibility (CSR) reports, corporate responsibility/citizenship
2
College of Management, University of Massachusetts, 100 reports, sustainability reports, and so on. For simplicity, we use the
Morrissey Blvd, Boston, MA 02125, USA term “CSR reports” to broadly refer to these reports.

13
Vol.:(0123456789)
 S. Du, K. Yu

employees, business partners, advocacy groups, and inves- change in report readability, suggesting that more readable
tors (Perrini 2006). CSR reports primarily include textual, CSR reports spur trading by releasing more value relevant
non-quantifiable information regarding firms’ policies, prac- information to investors or increasing information preci-
tices, and performance in social, environmental, and govern- sion. In line with the finding that enhanced readability of
ance domains (Dhaliwal et al. 2011; Du et al. 2017). Given CSR reports is indicative of better future CSR performance,
the descriptive and non-financial nature of CSR reports, the there is a positive association between the abnormal returns
textual properties, such as readability and tone, naturally and the change in CSR report readability. With regard to
play a prominent role in determining the effectiveness of CSR report tone, we find that the abnormal returns around
CSR communication and in shaping the information content the release of CSR reports are positively associated with
of those reports. tone change, but there is no trading volume reaction to tone
Effective information disclosure and information trans- change.
parency are key aspects of corporate ethical behavior that This paper contributes to the literature on discretionary
could build stakeholder trust and sustainable competitive information disclosure in general and CSR reporting in par-
advantage (Das Neves and Vaccaro 2013; Jones et al. 2018). ticular. Truthful and effective information disclosure is an
Whether and how the textual characteristics of CSR reports integral part of corporate ethical behavior and stakeholder
convey useful information and improve transparency is a relationship management (Das Neves and Vaccaro 2013;
question of importance to investors, managers, regulators, Martinez-Ferrero et al. 2016; Cui et al. 2018; Jones et al.
and other stakeholders. The Securities and Exchange Com- 2018). As Jones et al. (2018, p. 375) stated, ethical firms
mission (SEC) is evaluating the importance and effective- engage in behaviors of, among others, “refraining from tak-
ness of CSR disclosure. In particular, the SEC states, “we ing advantage of power imbalances or information asym-
seek feedback on which, if any, sustainability and public metries, and willingly sharing relevant information.” Yet,
policy disclosures are important to an understanding of a how to truthfully and effectively communicate complicated
registrant’s business and financial condition and whether and multi-dimensional information such as CSR perfor-
there are other considerations that make these disclosures mance is not straightforward. By analyzing how both read-
important to investment and voting decisions” (SEC 2016). ability and tone of CSR reports convey information about
Furthermore, the Sustainability Accounting Standards Board future CSR performance and demonstrating the value rel-
(SASB) argues that sustainability accounting should have evance of these two textual attributes, this study substanti-
both confirmatory and predictive value, so that it can be ates the important roles of CSR report readability and tone
used for future planning and decision support (SASB 2017). in increasing information transparency and imparting value
While a few studies have examined how contemporaneous relevant information to the market.
CSR performance affects CSR report readability (e.g., Naz- As importantly, this study advances current understand-
ari et al. 2017; Wang et al. 2018) and how CSR disclosure ing of the information content of CSR reports. To the best of
quality influences analyst forecast (Muslu et al. 2019), it our knowledge, this is the first study that investigates stock
remains unknown whether and how the textual attributes of market reaction to CSR report readability and tone. Prior
CSR reports help predict future CSR performance and influ- studies have found that issuing stand-alone CSR reports
ence investors’ trading behavior. This study examines the reduces the cost of capital (Dhaliwal et al. 2011) and analyst
implications of CSR report readability and tone for future forecast error (Dhaliwal et al. 2012), and moves stock prices
CSR performance and the market reaction around the release (Du et al. 2017). Muslu et al. (2019) show that the overall
of CSR reports, and in so doing, sheds light on how firms disclosure score of CSR reports influences analyst forecast
could effectively reduce information asymmetry through accuracy. Our study complements this line of research by
CSR reporting. shedding light on the possible channel through which CSR
Using a hand-collected dataset of Fortune 500 companies reports affect analyst forecast and stock prices: the readabil-
that published stand-alone CSR reports from 2002 to 2014, ity and tone of CSR reports may help analysts and investors
we document a positive association between 1-year-ahead predict value relevant future CSR performance. The results
CSR performance and the changes in CSR report readabil- also provide support for the SASB’s view that CSR disclo-
ity and tone. The change in CSR report readability is also sure should have predictive value to be useful for decision
predictive of 2-year-ahead CSR performance. These results making.
suggest that increases in readability and tone in a firm’s CSR Furthermore, the significant market reaction to CSR
report are indicative of better future CSR performance. Fur- report readability and tone demonstrates not only investors’
thermore, we find that the stock market reacts significantly to demand for information contained in CSR reports, but also
the changes in report readability and tone around the release positive business returns, in the form of higher stock prices,
of a CSR report. Specifically, the abnormal trading volume for socially responsible firms that effectively communicate
around report issuance dates is positively associated with the their superior CSR performance using a positive tone and

13
Do Corporate Social Responsibility Reports Convey Value Relevant Information? Evidence from…

readable text. More generally, in line with instrumental performance. Socially responsible firms enjoy higher brand
stakeholder theory (Freeman 1999; Jones et al. 2018), our equity (Torres et al. 2012), greater customer satisfaction and
results suggest that there is a strong business case for sus- loyalty (Luo and Bhattacharya 2006; Ailawadi et al. 2014).
tainability reporting and that managers could reduce infor- Firms with superior CSR performance have advantages in
mation asymmetry by using textual properties (e.g., read- attracting, motivating, and retaining talented employees
ability and tone) to communicate future CSR performance (Turban and Greening 1997; Surroca et al. 2010). CSR is an
information. important means for boosting firm productivity (Hasan et al.
Our analysis is subject to several caveats. First, due to 2018) and innovation (Luo and Du 2015). More generally,
a lack of the word list specifically tailored to CSR reports, a positive record of CSR performance helps a firm attain
we calculate report tone using the word lists by Loughran legitimacy and the license to operate at local communities
and McDonald (2011), which are developed from financial as well as receive more favorable treatment from the media
reports and thus may not accurately capture the positive and the regulators (Fombrun et al. 2000). Furthermore, the
and negative words of CSR reports. Given the differences goodwill derived from CSR can act as “an insurance policy”
between financial and CSR reports, future research should that minimizes firm risk (Klein and Dawar 2004; Godfrey
generate word lists focused on the CSR context. Second, et al. 2009). Overall, prior literature has documented positive
we only study readability and tone of CSR reports in our effects of firm CSR performance on financial performance
analysis. Future research may examine the effects of other and market value (Khan et al. 2016; Margolis et al. 2007;
textual aspects of CSR reports, such as numerical and hori- Servaes and Tamayo 2013; Hasan et al. 2018), consistent
zon content, boilerplate, and specificity, on the stock market. with the value relevance of CSR performance information.
The rest of the paper is organized as follows. Section 2 Since information about firm CSR performance is value
reviews prior literature and develops key arguments for our relevant, publishing CSR reports could enable market par-
hypotheses. Section 3 outlines the research methodology. ticipants to have a better understanding of firm performance
We describe the sample and descriptive statistics in Sect. 4. and increase financial transparency. Prior studies have found
Section 5 reports main empirical results. Section 6 presents that the issuance of CSR reports reduces the cost of equity
additional analyses. Section 7 provides concluding remarks capital (Dhaliwal et al. 2011), enhances analyst forecast
and discussion. accuracy (Dhaliwal et al. 2012), and triggers significant
stock market reaction (Du et al. 2017). Nevertheless, it
remains, to date, largely a black box as to what information
Literature Review and Predictions in CSR reports has predictive value and influences inves-
tors’ trading decisions. We seek to shed light on these ques-
CSR Reporting tions by examining whether the textual attributes of CSR
reports—readability and tone—convey information about
CSR reporting has been on the rise over the past decades future CSR performance and influence market reaction to
(Cho et al. 2015). In their study on the evolution of CSR CSR reports.
reporting, Tschopp and Huefner (2015, p. 565) state, “(CSR
reporting) seems destined to become a key part of the overall Effects of CSR Report Readability on Future CSR
accounting reporting framework, joining external financial Performance and Market Reaction
reporting, income tax reporting, regulatory reporting, and
internal reporting.” CSR reports, ranging from several doz- The strategic reporting literature suggests that management
ens to several hundred pages in length, provide comprehen- tends to be more forthcoming in disclosure when the firm is
sive and in-depth information about firms’ social and envi- performing well, but has incentives to obfuscate information
ronmental performance. As compared to alternative CSR when firm performance is poor (Schrand and Walther 2000).
disclosure methods, such as communication on corporate Li (2008) finds that firms with less readable annual reports
websites, individual social/environmental data disclosure, have lower subsequent earnings, suggesting that manage-
or CSR information in financial reports, stand-alone CSR ment tries to hide poor future performance from investors
reports are unique in the sense that they provide greater by increasing the complexity of annual reports.
depth and breadth regarding information about corporate In the case of CSR reports, the incentive to obfuscate
social performance in all key domains (e.g., employee wel- CSR performance information is also likely to exist when
fare, diversity, community outreach, product safety, environ- managers anticipate inferior CSR performance in the future.
ment) and serve as a one-stop source of CSR performance Stakeholders react favorably to positive CSR performance
information for stakeholders. and unfavorably to negative CSR performance (Fombrun
CSR performance information could be value rel- et al. 2000). When learning about a firm’s poor CSR per-
evant because CSR practices can enhance firm financial formance, stakeholders are likely to sanction the firm by

13
 S. Du, K. Yu

engaging in negative word-of-mouth, boycotting or switch- the trading volume reaction to the release of CSR reports is
ing brand, and increased employee turnover (Klein and likely to be stronger when the reports are more readable.2
Dawar 2004; Godfrey et al. 2009; Surroca et al. 2010). To Furthermore, CSR report readability is likely to have
the extent that less readable CSR reports can hide the bad a favorable impact on the abnormal returns around the
news of poor future CSR performance by increasing stake- release of CSR reports. First, the large body of research
holders’ information processing costs, firms with poor future on the link between CSR performance and financial per-
CSR performance have incentives to issue less readable CSR formance reveals an overall positive relationship (Margolis
reports. In contrast, firms with favorable future CSR perfor- and Walsh 2003; Margolis et al. 2007; Servaes and Tamayo
mance are likely to issue CSR reports that are transparent 2013; Hasan et al. 2018),3 pointing to the value relevance
and easier to read, in order to reap goodwill and the asso- of CSR performance due to its positive effects on stake-
ciated business benefits from its various stakeholders (Du holder satisfaction (Luo and Bhattacharya 2006; Surroca
et al. 2010). Thus, the management obfuscation hypothesis et al. 2010), moral capital (Godfrey et al. 2009), produc-
suggests that lower CSR report readability indicates less tivity (Hasan et al. 2018), innovation (Luo and Du 2015),
favorable future CSR performance. and so on. If firms with more readable CSR reports tend
On the other hand, because CSR reports are voluntary, to have better future CSR performance, and consequently,
unaudited, and do not follow a mandatory reporting frame- better future financial performance, CSR report readability
work (Perrini 2006), managers have significant discretion in should be positively associated with the abnormal returns.
deciding what CSR information to report, or in some cases, Second, empirical evidence has pointed to lower firm risk
whether to release a CSR report or not. Instead of obfuscat- (Loughran and McDonald 2014) and cost of capital (Ertu-
ing information by using complex words/sentences, oppor- grul et al. 2017) associated with more readable financial
tunistic managers may choose to omit areas of concern or disclosure. Dhaliwal et al. (2011, 2012) suggest that CSR
disclose less when CSR performance is poor. Supporting this disclosure could reduce information asymmetry among
view of selective reporting, Clarkson et al. (2008) find that investors or between managers and investors, and decrease
firms with higher environmental performance have higher the cost of capital. Firms with more readable CSR reports
levels of discretionary environmental disclosures (i.e., a are likely to enjoy a greater reduction in information asym-
greater number of disclosure items). Similarly, Nazari et al. metry and the cost of capital, suggesting an increase in stock
(2017) find that CSR performance is positively related to the prices around the release of such reports.
length of CSR reports, suggesting that firms with superior Taken together, if higher CSR report readability reduces
CSR performance disclose more, whereas those with inferior information processing cost and increases information preci-
CSR performance disclose less. If managers resort to selec- sion, then more readable CSR reports should release more
tive reporting, instead of information obfuscation, to hide digestible and actionable information to investors, leading
poor future performance, then report readability is less likely to a positive association between the change in report read-
to be predictive of future CSR performance. ability and trading volume reaction to the release of CSR
In examining the stock market reaction to CSR report reports. With regard to price reaction to CSR report read-
readability, we look at both abnormal trading volume and ability, if improved report readability is indicative of better
abnormal returns around the release of CSR reports. Prior future performance or leads to a lower cost of capital, one
literature suggests that investors are less likely to rely upon would expect a positive association between the abnormal
less readable financial reports due to higher information returns around the release of CSR reports and the change in
processing cost and lower information precision (Kim and report readability.
Verrecchia 1991; Bloomfield 2002). Consistent with this
argument, several studies (Miller 2010; Franco et al. 2015)
have documented lower abnormal trading volume around the 2
An alternative stream of mostly theoretical research (e.g., Kim
release of less readable 10-Ks or analyst reports. Given the and Verrecchia 1994) suggests that less readable reports are likely
large amount of descriptive and non-quantifiable informa- to increase information asymmetry and divergence in investor belief,
tion regarding various CSR domains (e.g., employee welfare, leading to an increase in trading volume. Under this alternative view,
product safety, environment, and community relations) in there should be a negative association between abnormal trading vol-
ume and CSR report readability.
CSR reports, it would be difficult for investors to assess CSR 3
Margolis et al. (2007) conducted a meta-analysis of 167 studies and
information in less readable CSR reports and its implications found an overall positive link between CSR performance and finan-
for future financial performance. In contrast, investors may cial performance. Later studies have confirmed this overall positive
make greater use of CSR performance information in their link (e.g., Mishra and Suar 2010; Hasan et al. 2018). In addition, this
trading decisions if CSR reports are more readable and con- positive relationship is contingent upon a variety of factors, such as
consumer awareness (Servaes and Tamayo 2013), CSR type (Peng
tain more transparent information. Therefore, we expect that and Yang 2014), firm competence (Luo and Bhattacharya 2006), and
others.

13
Do Corporate Social Responsibility Reports Convey Value Relevant Information? Evidence from…

Effects of CSR Report Tone on Future CSR performance use biased language and tone (i.e., more
Performance and Market Reaction optimism and less certainty) to present a more favorable
depiction of their performance. If the opportunistic disclo-
To convey information about CSR performance, manag- sure motive prevails, then one would expect a negative or
ers can use both numerical/quantifiable indicators (e.g., non-significant association between tone and future CSR
key performance indicators, year over year comparisons) performance.
and textual, non-quantifiable indicators in the CSR reports. While the truthful and opportunistic disclosure motives
However, CSR performance information relies upon quali- are likely to coexist in CSR reporting (Clarkson et al. 2008;
tative, textual description to a much greater extent as com- Du et al. 2010), it is worth noting that there are mechanisms
pared to financial performance information. For instance, functioning to constrain managers’ opportunistic disclosure
in talking about employee wellbeing, a key aspect of CSR motive in CSR reporting. Stakeholders have access to not only
performance, 3M’s 2014 CSR report contains detailed tex- company-issued CSR reports, but also a variety of CSR infor-
tual description on how the company addresses issues such mation from independent third parties such as CSR ratings by
as employee benefits, education and career growth, health KLD, Newsweek Green rankings, environmental impact infor-
and wellness, employee engagement, and global safety. mation by CDP, and the mass media. These third-party infor-
Even when numerical indicators are used to describe mation intermediaries can deter managers from information
aspects of CSR performance (e.g., environmental per- distortion as stakeholders react negatively when they detect
formance), they are not as informative as the numerical false information in CSR communication from the corporate
indicators (e.g., sales, profit) in financial reports because sources. For example, Parguel et al. (2011) find that consum-
companies can report on different quantitative indicators ers rely upon independent, third-party sustainability ratings to
due to a lack of uniform reporting framework (Clarkson evaluate a firm’s corporate-controlled CSR communication.
et al. 2008). Thus, the predominantly textual nature of Turning to the effect of CSR report tone on the market
CSR reports entails that managers are more likely to use reaction, previous studies have provided evidence that the
tone to communicate hard-to-quantify information. market reacts positively to upward tone revision in financial
A priori, it is not straightforward whether CSR report reports. For example, Feldman et al. (2010) find that tone
tone is indicative of future CSR performance. The truthful change in the MD&As is positively associated with abnor-
disclosure hypothesis (e.g., Tetlock et al. 2008; Davis et al. mal returns surrounding the filing of the MD&As. Similarly,
2012 2015) suggests that managers use positive and nega- Davis et al. (2012) document a positive association between
tive words to convey private and hard-to-quantify infor- tone revision and abnormal returns around earnings press
mation, and to signal their expectation about future finan- releases. If more positive CSR report tone indicates better
cial performance. Supporting the view that CSR reports future CSR performance, there should be a positive associa-
contain credible information about CSR performance and tion between abnormal returns around the release of CSR
are relevant for assessing firm performance, Dhaliwal reports and tone change. Given that trading volume reac-
et al. (2011, 2012) find that the issuance of CSR reports tion is not conditional on the direction of news released by
reduces cost of equity capital and analyst forecast error; CSR reports, we expect a positive association between the
similarly, Muslu et al. (2019) document a positive associa- abnormal trading volume and the magnitude of tone revision
tion between CSR disclosure quality and analyst forecast in CSR reports.
accuracy. Under this view, an increase in CSR report tone In summary, if the overall sentiment of a CSR report con-
would be indicative of higher future CSR performance. veys value relevant information about future CSR perfor-
On the other hand, the opportunistic disclosure motive mance, tone change should be positively associated with the
is likely to play a role as well in CSR reporting. Prior abnormal returns, and the magnitude of tone change should
literature suggests that greenwashing and impression man- be positively linked with the abnormal trading volume.
agement is common when it comes to CSR communication However, if managers manipulate tone opportunistically, the
(Patten 1992; Cho et al. 2010; Du et al. 2010; Mahoney effect of CSR report tone on the market reaction would be
et al. 2013). Because CSR reporting remains unregulated weakened to the extent that investors may see through tone
and CSR performance information is not easily verifiable, management. Thus, whether and how CSR report tone may
opportunistic managers may manipulate tone to mislead affect future CSR performance and market trading activi-
investors about firms’ future CSR performance. Cho et al. ties remain open questions and demand further empirical
(2010) find evidence that firms with low environmental investigation.

13
 S. Du, K. Yu

Research Methodology To measure CSR report tone, we use the list of positive
and negative words compiled byLoughran and McDonald
Measuring Readability and Tone of CSR Reports (2011) (LM list).4 Prior research (see Loughran and McDon-
ald 2016 for a review) suggests that the LM list is more
We use the Fog Index to measure CSR report readability in appropriate and relevant for business disclosure than alterna-
our main analyses (Li 2008). The Fog Index, developed by tive lists, such as Diction and Harvard General Inquirer word
Robert Gunning, is a well-known and popular formula to lists. For example, many negative words in the Harvard Gen-
appraise readability. It captures text complexity as a func- eral Inquirer list are used to describe financial aspects (e.g.,
tion of two components: the average number of words per cost, tax, and foreign), corporate governance (e.g., board
sentence and percentage of complex words with more than and vice), and industries (e.g., mine, tire, and crude), and
two syllables. It is defined as follows. are typically not negative in business disclosure (Loughran
and McDonald 2011). In addition, the LM list also includes
Fog Index = 0.4 negative words that are used in the business context, such
× (words per sentence + percentage of complex words) as restated, litigation, and restructuring. Following prior
(1) research, we define POS (NEG) as the number of positive
The Fog Index reflects the number of years of formal (negative) words scaled by the number of total words in a
education that a reader of average intelligence would need to CSR report. TONE is used to gauge the overall sentiment of
read and understand the text. In general, a value of the Fog a report and is defined as the difference in the proportions of
Index above or equal to 18 indicates that the text is unread- positive and negative words (i.e., POS minus NEG).
able; 14–18 difficult to comprehend; 12–14 ideal; 10–12 Tetlock et al. (2008) and Davis et al. (2012) suggest that
acceptable; and 8–10 childish. For ease of presentation and investors use the textual properties of financial reports in
interpretation of the readability coefficients in our empirical the past year as the benchmark and react only to changes in
analyses, we scale the Fog Index by − 100 (i.e., FOG is the readability and tone. Furthermore, prior research (Feldman
Fog Index divided by − 100). As a result, higher values of et al. 2010) suggests that relative to the changes in readabil-
FOG indicate more readable CSR reports. ity and tone, the levels of textual properties are more likely
Loughran and McDonald (2014) argue that the Fog index to be affected by the boilerplate usage of certain words in
is a poor proxy for measuring readability of financial docu- an industry or a firm as well as the choice of a particular
ments, because, among others, many complex words may be word list. In addition, we recognize that the second compo-
quite easy to understand for investors and financial analysts. nent (percentage of complex words) of the Fog Index could
They recommend using file size as a readability measure for be a potentially misleading factor in measuring readability
10-K reports, since longer documents have higher informa- (Loughran and McDonald 2014), given that some multi-
tion processing cost and are more difficult to read. syllable words, such as sustainability and environment, are
While acknowledging the limitation of the Fog Index, common CSR terms and may be easily understood by stake-
we choose to use the Fog index as a measure for CSR report holders. If these multi-syllable words and the boilerplate
readability for two reasons. First, due to the discretionary usage of certain words are used in a similar way across years
nature of CSR reporting, file size is more likely to be cor- for a given company, using changes in readability and tone
related with the amount of CSR disclosure and thus is not should mitigate concerns about measurement error associ-
a good proxy for report readability. Prior research suggests ated with level measurements. Therefore, in our empirical
that disclosure length is a positive indicator of disclosure analysis, we focus on changes in report readability and tone,
transparency and informativeness, particularly in the case and examine their effects on future CSR performance and
of CSR disclosure (Lang and Stice-Lawrence 2015; Nazari the market reaction around the release of CSR reports.
et al. 2017; Muslu et al. 2019). Dhaliwal et al. (2011) argue
that the length of a CSR report is a proxy for firm’s efforts
and commitment to better disclosure. Second, unlike finan-
cial reports, CSR reports have targeted audiences including 4
For the complete list of positive and negative words, please refer to
not only investors and financial analysts, but also a variety https​://sraf.nd.edu/textu​al-analy​sis/resou​rces/#LM%20Sen​timen​t%20
Wor​d%20Lis​ts. Examples of the most frequent positive words in our
of other key stakeholders, such as consumers, employees,
sample of CSR reports are: opportunities, leadership, improve, effi-
business partners, community members, media, and so on. ciency, progress, better, innovation, improvement, success, achieve,
Complex words that could be easily understood by investors good, strong, benefit, innovative, leading, effective, highest, integrity,
or analysts may not be comprehensible for employees and excellence, achieved. Examples of the most frequent negative words
in our sample are: challenges, critical, concerns, against, incidents,
other shareholder groups. In this regard, using the Fog index
hazardous, prevention, injury, loss, questions, disaster, force, viola-
as a proxy for CSR report readability seems of less concern. tions, non-compliance, problems, concern, fines, corruption, chal-
lenging, crisis.

13
Do Corporate Social Responsibility Reports Convey Value Relevant Information? Evidence from…

Measuring CSR Performance divided by total assets at the end of year t. We control for
financial leverage (LEV) because firms with constrained
Following prior research (e.g., Dhaliwal et al. 2011, 2012; financial resources are less likely to engage in CSR (Wad-
Kim et al. 2012; Servaes and Tamayo 2013), we use KLD dock and Graves 1997; Surroca et al. 2010). ­LEVt is calcu-
ratings provided by MSCI ESG Research (formerly KLD lated as total debt (i.e., short term debt plus long-term debt)
Research and Analytics Inc.) as the proxy for overall firm divided by total assets at the end of year t .
CSR performance. KLD ratings are among the most influen- Prior studies suggest that a firm’s financing needs, stock
tial and most widely accepted measures of CSR performance liquidity, research and development (R&D) investment, and
used by academics (Dhaliwal et al. 2011). Since 2003, KLD firm information environment might influence its incentive
ratings cover the 3,000 largest U.S. companies and provide to engage in CSR (McWilliams and Siegel 2000; Dhaliwal
CSR performance ratings in key social and environmental et al. 2011). We include corporate financing activities (FIN),
domains. stock liquidity (LIQUID), research and development inten-
KLD data include seven different CSR domains (envi- sity (RD), and analyst following (AF) as additional control
ronment, community, diversity, employee relations, product, variables. Specifically, ­FINt is calculated as the net amount
human rights, and corporate governance) and provide the of debt and equity capital raised by the firm (i.e., the net
numbers of strengths and concerns for each domain. Since sale of common and preferred shares plus the net issuance
the numbers of strength and concern indicators have changed of long-term debt) during the year scaled by total assets at
over the years in the KLD dataset, we scale the number of the end of year t . ­LIQUIDt is the number of shares traded
total strengths (concerns) for each firm-year by the maxi- divided by the number of shares outstanding for year t . ­RDt
mum possible number of strengths (concerns) in each year is R&D expenses deflated by sales for year t . A ­ Ft is defined
to obtain the corresponding strength and concern indices as the natural logarithm of one plus the number of analysts
that range from 0 to 1, and then subtract the concern index at the end of year t .
from the strength index to get the net CSR performance that Li (2008) shows that 10-K readability is related to future
ranges from − 1 to + 1 (for similar transformation of KLD financial performance, which in turn may be associated with
data, see Waddock and Graves 1997; Servaes and Tamayo future CSR performance. We thus control for the potential
2013). effect of 10-K readability on future CSR performance. Simi-
lar to the treatment of the CSR report readability measure,
Empirical Models our proxy for 10-K readability (READ_10K) is defined as
the natural logarithm of the file size in megabytes of the
We use the following model to test the effects of CSR report SEC EDGAR “complete submission text file” for the 10-K
readability and tone on future CSR performance. filing (Loughran and McDonald 2014) divided by − 100, so
that a higher value of READ_10K indicates more readable
CSRPt+1 = 𝛽0 + 𝛽1 CSRPt +𝛽2 CSRPt−1 + 𝛽 3 SIZEt + 𝛽4 ROAt
10-Ks. Finally, following Dhaliwal et al. (2011), we include
+ 𝛽5 LEV t + 𝛽6 FINt + 𝛽7 LIQUIDt + 𝛽8 RDt + 𝛽9 AFt fixed year and industry effects, with industry classifications
+ 𝛽10 READ_10Kt + 𝛽11 ΔREADt + 𝛽12 ΔTONEt based on Barth et al. (1998).
+ Industry and Year Fixed Effects + 𝜀t+1 To examine whether CSR report readability and tone
(2) affect the stock market reaction around the release of CSR
reports, we look at both abnormal trading volume and abnor-
The variables of interest are ΔREADt and ΔTONEt.
mal stock returns. Prior research (Cready and Hurtt 2002)
ΔREAD t is the change in CSR report readability, as
suggests that abnormal trading volume and abnormal returns
measured by FOG from year t − 1 to year t . ΔTONEt is
capture different aspects of market reactions to information
the change in CSR report tone from year t − 1 to year t.
events. In particular, abnormal returns reflect the changes
­CSRPt +1, ­CSRPt, and ­CSRPt − 1 are firms’ net CSR perfor-
in the expectations of the market as a whole, while abnor-
mance for year t + 1, t, and t − 1, respectively.
mal trading volume reflects the changes in the expectations
The control variables are taken from prior studies exam-
of individual investors (Beaver 1968). An increase in trad-
ining factors that affect CSR performance. We include firm
ing volume around the release of a CSR report does not
size (SIZE) as larger firms have greater visibility and face
indicate good or bad news, but suggests that the informa-
more intense stakeholder pressure to engage in CSR (Smith
tion contained in the report changes individual investors’
2003). ­SIZEt is the natural logarithm of total assets at the
expectations, leading to an altering of their optimal portfolio
end of year t. We include ­ROAt to control for the positive
positions. On the other hand, an increase in stock price (i.e.,
association between financial performance and CSR per-
positive abnormal returns) indicates that the CSR report
formance (Margolis and Walsh 2003). ­ROAt is the return
releases good news to the market as a whole, resulting in a
on assets, calculated as income before extraordinary items
higher equilibrium price. Using abnormal trading volume

13
 S. Du, K. Yu

and abnormal returns allows us to gauge the effects of new The Sample
information due to the release of CSR reports on the market.
The stock market would react to CSR reports to the extent We begin with Fortune 500 companies that published stand-
that the new information as conveyed by CSR report read- alone CSR reports during the period 2002 to 2015. CSR
ability and tone changes investors’ expectations of future reports are collected from various internet sources, includ-
firm performance. ing CSRwire.com, CorporateRegister.com, GlobalReport-
We use the following model to examine the effects of ing.org, SocialFunds.com, BusinessWire.com, and cor-
CSR report readability and tone on the cumulative tabnor- porate websites. We then match each CSR report with its
mal trading volume around the release of CSR reports. corresponding fiscal year. The initial sample includes 1,780
CABVOLt = 𝛽0 + 𝛽1 ΔCSRPt + 𝛽2 SIZEt + 𝛽3 ROAt + 𝛽4 LEVt
CSR reports for 340 firms for fiscal years 2002 to 2014. We
exclude the first observation (i.e., the first CSR report) for
+ 𝛽5 FINt + 𝛽6 LIQUIDt + 𝛽7 RDt + 𝛽8 AFt
each firm from our sample since the changes in readability
+ 𝛽9 READ_10Kt + 𝛽10 ΔREADt + 𝛽11 ABSΔTONEt and tone for the first CSR report cannot be calculated. We
+ Industry and Year Fixed Effects + 𝜀t then merge readability and tone information for CSR reports
(3) with CSR performance from KLD, financial information
CABVOLt is the cumulative abnormal trading volume from Compustat, analyst following from I/B/E/S, and 10-K
during the window (− 1, 1) centered on the release date readability from Loughran and McDonald 10X File Summa-
of CSR reports for year t, calculated as the logarithm of ries.5 Observations with missing data for required variables
the cumulative trading volume during the three-day event are deleted. The final full sample includes 1258 observations
window minus the logarithm of the firm-specific median for 262 unique firms from 2002 to 2014. This full sample
cumulative trading volume for contiguous three-day periods is used to assess how CSR report readability and tone are
over the estimation period from 100 trading days prior to related to future CSR performance.
the three-day event window (− 1, 1) to 21 trading days prior To examine the market response to CSR report readabil-
to this window (Franco et al. 2015). ∆CSRPt is the change ity and tone around the release of CSR reports, we start
in firm CSR performance from year t − 1 to year t . To with the full sample and search for the release dates of the
the extent that KLD ratings proxy for the factual/quantita- CSR reports, using various keywords (e.g., “CSR report,”
tive information about CSR performance in a CSR report, “sustainability report,” “corporate citizenship report,”
including ∆CSRPt allows us to control for the market reac- “release,” “announce,” “issue,” “today,” “becomes avail-
tion associated with the factual information in the report. able,” and other similar terms) and various Internet sites,
ABS∆TONEtis the absolute value of ∆TONEt, and captures including CSRwire, CorporateRegister.com, Business wire,
the magnitude of the change in CSR report tone from year Reuters, PRweb, and company websites (the newsroom or
t − 1 to year t. Additionally, we include the same set of con- investor relations section). We verify the report release dates
trol variables as in model (2) to make sure that the market by reading the press releases. CSR reports with unidentifi-
reaction to CSR report readability and tone is not driven by able release dates are excluded. To control for confounding
firm fundamentals correlated with CSR performance. events, we check for other major news concerning the firms
The following model is used to examine the effects of and eliminate the firm-date observations from our sample,
CSR report readability and tone on the cumulative abnormal if CSR report release dates are within the 5-day window
returns around the release of CSR reports. (− 2, 2) around the release of other major corporate events,
such as earnings announcements or merger and acquisition
CARt = 𝛽0 + 𝛽1 ΔCSRPt + 𝛽2 SIZEt + 𝛽3 ROAt + 𝛽4 LEVt announcements. The final market reaction sample includes
+ 𝛽5 FINt + 𝛽6 LIQUIDt + 𝛽7 RDt + 𝛽8 AFt 574 observations with identifiable release dates of CSR
+ 𝛽9 READ_10Kt + 𝛽10 ΔREADt + 𝛽11 ΔTONEt reports for 175 unique firms. Observations in the bottom or
top 1% regression residuals are deleted to mitigate the effect
+ Industry and Year Fixed Effects + 𝜀t
of outliers in the regression analyses.
(4)
Panel A of Table 1 reports the distribution of the full
CAR​t is the cumulative market-adjusted abnormal returns sample by year. Consistent with the trend that a growing
during the window (− 1, 1) around the release of CSR number of firms engage in CSR reporting in the recent
reports for year t. Throughout our regression analyses, stand- years (Tschopp and Huefner 2015), the number of CSR
ard errors are adjusted based on two-way clustering at firm reports among Fortune 500 firms increases from 7 to more
and year to address the concern that ordinary least squares
(OLS) may underestimate standard errors (Petersen 2009).
5
Available at https​://sraf.nd.edu/textu​al-analy​sis/resou​rces/#LM_10X_
Summa​ries.

13
Do Corporate Social Responsibility Reports Convey Value Relevant Information? Evidence from…

Table 1  Sample distribution

Panel A: Sample distribution by year
Year Frequency Percent Cumulative frequency Cumu-
lative
percent

2002 7 0.56 7 0.56

2003 15 1.19 22 1.75
2004 28 2.23 50 3.97
2005 29 2.31 79 6.28
2006 48 3.82 127 10.10
2007 61 4.85 188 14.94
2008 75 5.96 263 20.91
2009 107 8.51 370 29.41
2010 131 10.41 501 39.83
2011 167 13.28 668 53.10
2012 180 14.31 848 67.41
2013 208 16.53 1056 83.94
2014 202 16.06 1258 100
Panel B: Sample distribution by industry
Industry No. of CSR reports (%) No. of firms (%)

1. Mining and construction 44 3.50 10 3.82

2. Food 70 5.56 15 5.73
3. Textiles, printing, and publishing 55 4.37 13 4.96
4. Chemicals 79 6.28 15 5.73
5. Pharmaceuticals 66 5.25 10 3.82
6. Extractive industries 76 6.04 14 5.34
7. Durable manufacturers 220 17.49 46 17.56
8. Computers 117 9.30 19 7.25
9. Transportation 85 6.76 21 8.02
10. Utilities 127 10.10 27 10.31
11. Retail 136 10.81 30 11.45
12. Financial institutions 125 9.94 26 9.92
13. Insurance and real estate 4 0.32 1 0.38
14. Services 46 3.66 13 4.96
15. Other 8 0.64 2 0.76
Total 1258 100.00 262 100.00

than 200 in 2013 and 2014. Panel B of Table 1 reports CSR reports is 15.80 (16.08), which falls into the category of
the distribution of the full sample by industry based on “difficult to read,” but is lower than the mean Fog index for
Barth et al.’s (1998) industry classifications. The Durable 10-K reports (18.68) reported in Loughran and McDonald
Manufacturing industry contains the largest number of (2014). This indicates that although CSR reports are difficult
CSR reports (220) and firms (46), accounting for 17.5% to read, they are slightly easier to comprehend than 10-K
of all the CSR reports and 17.6% of all the firms. The retail reports. This is consistent with the view that, as compared
industry has the second largest number of CSR reports to financial reports, CSR reports target a greater variety of
(136) and firms (30). On the other hand, the Insurance stakeholders including less sophisticated audiences. The
and Real Estate industry and the Other industry have the mean (median) of TONE is 1% (0.9%), indicating that CSR
smallest number of CSR reports (4 and 8, respectively) in report tone is, in general, relatively positive. There is also a
our sample. relatively large variation in the changes in CSR report read-
Table 2 reports the descriptive statistics of our main vari- ability and tone (ΔFOG: Q1 = − 0.6%, Q3 = 0.6%; ΔTONE:
ables for the full sample. The mean (median) Fog index of Q1 = − 0.3%, Q3 = 0.3%). In addition, ΔPOS (SD = 0.6%)

13
 S. Du, K. Yu

Table 2  Descriptive statistics mean of C­ SRPt is 0.086, indicating an overall positive CSR
Variable Mean SD Q1 Median Q3 performance for the sample firms.
Table 3 presents the Pearson correlations of the main var-
FOG index 15.797 4.670 13.270 16.080 18.090 iables for the full sample. Correlation coefficients that are
FOGt − 0.158 0.047 − 0.181 − 0.161 − 0.133 significant at the 0.10 level or higher are in bold. The cor-
ΔFOGt − 0.001 0.038 − 0.006 0.000 0.006 relation between ΔFOG and ΔTONE (0.02) is insignificant,
POSt 0.019 0.005 0.016 0.018 0.022 suggesting that CSR readability and tone tap into distinct
NEGt 0.009 0.004 0.007 0.009 0.011 aspects of textual properties. In addition, future CSR per-
TONEt 0.010 0.007 0.006 0.009 0.013 formance ­(CSRPt +1) is positively correlated with ΔTONE
ΔPOSt 0.000 0.006 − 0.002 0.000 0.002 (correlation = 0.05), consistent with the argument that more
ΔNEGt 0.000 0.003 − 0.001 0.000 0.002 positive tone revision is associated with better future CSR
ΔTONEt 0.000 0.007 − 0.003 0.000 0.003 performance.
CSRPt +1 0.093 0.139 0.001 0.079 0.178
CSRPt 0.086 0.147 − 0.007 0.073 0.185
CSRPt − 1 0.079 0.153 − 0.022 0.067 0.185 Empirical Results
ΔCSRPt 0.007 0.116 − 0.038 0.000 0.072
SIZEt 10.299 1.295 9.340 10.187 10.986 Effects of Readability and Tone on Future CSR
ROAt 0.057 0.060 0.025 0.054 0.089 Performance
LEVt 0.255 0.142 0.152 0.246 0.351
FINt − 0.016 0.062 − 0.047 − 0.014 0.011 Table 4, Panel A, reports the effects of CSR report readabil-
LIQUIDt 2.454 1.780 1.445 1.998 2.903 ity (as measured by FOG) and tone on 1-year-ahead CSR
RDt 0.027 0.052 0.000 0.000 0.025 performance. Recall that FOG is the Fog Index divided by
AFt 2.883 0.434 2.708 2.944 3.135 − 100; thus higher values of FOG indicate more readable
READ_10Kt − 0.163 0.012 − 0.172 − 0.167 − 0.153 CSR reports. Column I presents the results based on model
Table 2 reports the descriptive statistics of the main variables for (2). The coefficient on ΔREAD is positive (coeff. = 0.044,
the full sample of 1258 observations. The Fog Index is calculated t stat = 2.19), suggesting that less readable CSR reports are
as 0.4* (words per sentence + percentage of complex words). ­FOGt indicative of lower future CSR performance. This result is
is the Fog Index for CSR reports for year t divided by − 100. Higher consistent with the management obfuscation hypothesis (Li
values of F­ OGt indicate greater report readability. P­ OSt is the num-
ber of positive words scaled by the number of total words in CSR 2008), suggesting that firms try to hide poor future CSR
reports for year t. NEGt the number of negative words scaled by the performance by decreasing CSR report readability. Also
number of total words in CSR reports for year t. ­TONEt is the dif- importantly, 1-year-ahead CSR performance is positively
ference in the proportions of positive and negative words in CSR associated with ΔTONE (coeff. = 0.806, t stat = 4.17), con-
reports for year t (i.e., ­POSt minus ­NEGt). ­CSRPt +1, ­CSRPt, and
­CSRPt−1 are firms’ net CSR performance for year t + 1, t, and t − 1, sistent with the truthful disclosure hypothesis that manag-
respectively. ­SIZEt is firm size, calculated as the logarithm of total ers use tone in CSR reports to convey credible information
assets at the end of year t. ­ROAt is the return on assets, calculated as regarding future CSR performance.
income before extraordinary items divided by total assets at the end In Columns II and III, we examine the separate effects of
of year t. ­LEVt is financial leverage, calculated as total debt (i.e.,
short term debt plus long-term debt) divided by total assets at the the proportions of positive and negative words by replacing
end of year t. ­FINt is calculated as the net amount of debt and equity ΔTONE with ΔPOS and ΔNEG, respectively. The coef-
capital raised by the firm (i.e., the net sale of common and preferred ficient on ΔPOS is positive (coeff. = 0.957; t stat = 3.63),
shares plus the net issuance of long-term debt) during the year while the coefficient on ΔNEG is negative (coeff. = − .934;
scaled by total assets at the end of year t. ­LIQUIDt is firm liquid-
ity, calculated as the number of shares traded divided by the number t stat = − 2.20), suggesting that the proportions of positive
of shares outstanding for year t. ­RDt is R&D expenses deflated by and negative words in CSR reports are informative about
sales for year t. ­AFt is defined as the natural logarithm of one plus future CSR performance. However, when both ΔPOS and
the number of analysts at the end of year t. ­READ_10Kt is 10-K ΔNEG are included as the independent variables in Col-
readability, defined as the natural logarithm of the file size in mega-
bytes of the SEC EDGAR “complete submission text file” for the umn IV, ΔPOS completely absorbs the explanatory power
10-K filing divided by − 100. Δ represents the change in variable of ΔNEG; the coefficient on ΔPOS remains significantly
from year t − 1 to year t. positive, whereas the coefficient on ΔNEG becomes insig-
nificant. While this result is in contrast with prior findings
in the case of financial reports that positive words provide
exhibits much higher variation than ΔNEG (SD = 0.3%), little incremental information compared to negative words
suggesting that managers tend to use changes in positive (Loughran and McDonald 2011), it seems to support the
words to convey information rather than negative words. The view that, given the voluntary and unregulated features of
CSR reports, negative words in CSR reports may be used

13
Do Corporate Social Responsibility Reports Convey Value Relevant Information? Evidence from…

Table 3  Correlations of the main variables

CSRPt +1 (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13)

CSRPt(1) 0.68
CSRPt − 1 (2) 0.55 0.70
SIZEt(3) − 0.02 − 0.01 0.01
ROAt(4) 0.11 0.12 0.11 − 0.09
LEVt(5) − 0.10 − 0.08 − 0.05 − 0.08 − 0.29
FINt(6) − 0.06 − 0.08 − 0.08 0.07 − 0.37 0.25
LIQUIDt(7) − 0.05 − 0.10 − 0.17 − 0.20 − 0.30 0.07 0.12
RDt(8) 0.27 0.26 0.27 0.02 0.22 − 0.17 − 0.09 − 0.10
AFt(9) 0.14 0.20 0.22 0.32 0.31 − 0.28 − 0.18 − 0.18 0.24
READ_10Kt(10) − 0.07 − 0.26 − 0.33 − 0.16 0.17 − 0.17 − 0.12 0.01 0.12 − 0.12
ΔFOGt(11) 0.01 0.00 − 0.04 − 0.01 − 0.01 0.03 0.02 0.03 0.00 − 0.02 0.02
ΔTONEt(12) 0.05 0.01 0.01 0.01 0.01 0.01 − 0.03 0.00 0.00 0.02 − 0.03 0.02
ΔPOSt(13) 0.04 0.00 0.01 0.01 0.00 0.00 − 0.02 0.01 0.00 0.02 − 0.02 0.03 0.89
ΔNEGt − 0.03 − 0.02 − 0.01 − 0.02 − 0.02 − 0.02 0.03 0.02 0.00 0.00 0.03 0.00 − 0.56 − 0.12

Table 3 reports the Pearson correlations of the main variables for the full sample. Correlation coefficients highlighted in bold are statistically sig-
nificant at the 10% level or better. All the variables are as defined in Table 2.

only in boilerplate format, and managers are more likely reports. However, the coefficient on ABSΔTONE is not sig-
to use changes in positive words to communicate informa- nificant, indicating that the volume reaction is not associated
tion about future CSR performance. However, an alternative with the magnitude of tone change in CSR reports. We fur-
explanation is that there is not enough power to detect the ther replace ABSΔTONE with ABSΔPOS and ABSΔNEG
association between future CSR performance and ΔNEG in Columns II and III to examine the volume reaction to the
due to relatively low variation in ΔNEG in our sample. magnitudes of the changes in the proportions of positive and
Given the long-term orientation of CSR reports, we test negative words. ABSΔPOS and ABSΔNEG are equal to the
whether CSR report readability and tone have implications for absolute value of ΔPOS and ΔNEG, respectively. The coeffi-
2-year-ahead CSR performance in Table 4, Panel B. The addi- cients on ABSΔPOS and ABSΔNEG are not significant. The
tional data requirement for 2-year-ahead CSR performance results consistently suggest that tone revision in CSR reports
reduces our sample size to 1235 observations. The results in does not affect the abnormal trading volume.6 Overall, the
Column I indicate that more readable CSR reports are also results reported in Panel A are consistent with the view that
associated with higher 2-year-ahead CSR performance (coeff. more readable CSR reports decrease information processing
on ΔREAD = 0.061, t stat = 2.75), although the coefficient on cost and increase information transparency, thus spurring
ΔTONE is insignificant. In Columns II and III, we replace trading activities around the release of these reports.
ΔTONE with ΔPOS and ΔNEG, respectively. There is weak Panel B of Table 5 reports the effects of CSR report
evidence that an increase in the proportion of positive words is readability and tone on the cumulative abnormal returns
indicative of more favorable 2-year-ahead CSR performance round the release of CSR reports. Column I shows the
(Column II: coeff. on ΔPOS = 0.583, t stat = 1.68). However, results based on model (4). The coefficient on ΔREAD is
when both ΔPOS and ΔNEG are added in the model in Col- positive (coeff. = 0.026, t stat = 2.84), suggesting that the
umn IV, the coefficient on ΔPOS becomes insignificant. market reacts positively to increase in report readability.
The coefficient on ΔTONE is also positive (coeff. = 0.188,
Market Reaction to Readability and Tone of CSR t stat = 2.07), indicating that the market reacts favorably
Reports to upward tone change. For a firm with the mean market
value of $56.3 billion in our sample, moving from the
Panel A of Table 5 reports the effects of CSR report read-
ability and tone on cumulative abnormal trading volume
around the release of CSR reports. Column I reports the 6
Following Franco et al. (2015), we also replace ABSΔTONE with
results based on model (3). The coefficient on ΔREAD is ΔTONEPOS and ΔTONENEG in model (5) to allow asymmetric
positive (coeff. = 0.334, t stat = 5.06), suggesting that firms volume reaction to positive vs. negative change in tone. ΔTONEPOS
(ΔTONENEG) is equal to ΔTONE if ΔTONE is positive (negative),
with a larger increase in CSR report readability experience and zero otherwise. We find no evidence that the trading volume
higher abnormal trading volume around the release of CSR reaction is related to positive or negative tone revision.

13
 S. Du, K. Yu

Table 4  The effects of CSR report readability and tone on future CSR performance
Panel A: 1-year-ahead CSR performance

ΔREAD = ΔFOG

Dependent Var = CSRPt+1

I II III IV

Coeff t stat Coeff t stat Coeff t stat Coeff t stat

CSRPt 0.517 (5.39)*** 0.517 (5.38)*** 0.521 (5.52)*** 0.517 (5.38)***

CSRPt− 1 0.219 (5.14)*** 0.220 (5.13)*** 0.214 (5.31)*** 0.219 (5.14)***
SIZEt − 0.003 (− 1.11) − 0.003 (− 1.09) − 0.003 (− 1.08) − 0.003 (− 1.10)
ROAt 0.015 (0.58) 0.016 (0.60) 0.015 (0.58) 0.016 (0.59)
LEVt − 0.022 (− 1.93)* − 0.022 (− 1.97)** − 0.020 (− 1.80)* − 0.022 (− 1.96)**
FINt − 0.007 (− 0.14) − 0.008 (− 0.16) − 0.011 (− 0.25) − 0.007 (− 0.14)
LIQUIDt − 0.000 (− 0.19) − 0.000 (− 0.19) − 0.000 (− 0.18) − 0.000 (− 0.19)
RDt 0.092 (2.04)** 0.094 (2.12)** 0.091 (1.99)** 0.093 (2.08)**
AFt − 0.001 (− 0.26) − 0.002 (− 0.27) − 0.001 (− 0.16) − 0.002 (− 0.27)
READ_10Kt 0.384 (0.60) 0.377 (0.59) 0.368 (0.57) 0.383 (0.60)
ΔREADt 0.044 (2.19)** 0.043 (2.15)** 0.046 (2.23)** 0.043 (2.18)**
ΔTONEt 0.806 (4.17)***
ΔPOSt 0.957 (3.63)*** 0.922 (3.48)***
ΔNEGt − 0.934 (− 2.20)** − 0.479 (− 1.49)
Fixed industry and year effects Yes Yes Yes Yes
Adjusted R2 0.690 0.690 0.688 0.689
No. of observations 1258
Panel B: 2-year-ahead CSR performance

ΔREAD = ΔFOG

Dependent Var = CSRPt+2

I II III IV

Coeff t stat Coeff t stat Coeff t stat Coeff t stat

CSRPt 0.425 (7.07)*** 0.425 (7.12)*** 0.425 (7.05)*** 0.425 (7.13)***

CSRPt–1 0.126 (3.32)*** 0.126 (3.36)*** 0.126 (3.28)*** 0.126 (3.36)***
SIZEt − 0.003 (− 0.65) − 0.003 (− 0.64) − 0.003 (− 0.64) − 0.003 (− 0.63)
ROAt − 0.044 (− 0.70) − 0.044 (− 0.69) − 0.044 (− 0.69) − 0.044 (− 0.69)
LEVt − 0.045 (− 1.59) − 0.045 (− 1.57) − 0.045 (− 1.59) − 0.045 (− 1.58)
FINt − 0.048 (− 0.87) − 0.048 (− 0.88) − 0.049 (− 0.90) − 0.048 (− 0.88)
LIQUIDt − 0.000 (− 0.12) − 0.000 (− 0.13) − 0.000 (− 0.11) − 0.000 (− 0.13)
RDt 0.149 (2.42)** 0.150 (2.42)** 0.148 (2.37)** 0.150 (2.40)**
AFt − 0.007 (− 1.13) − 0.007 (− 1.15) − 0.006 (− 1.09) − 0.007 (− 1.14)
READ_10Kt 0.877 (1.93)* 0.873 (1.90)* 0.863 (1.88)* 0.869 (1.89)*
ΔREADt 0.061 (2.75)*** 0.060 (2.68)*** 0.062 (2.78)*** 0.059 (2.65)***
ΔTONEt 0.384 (1.40)
ΔPOSt 0.583 (1.68)* 0.599 (1.64)
ΔNEGt 0.071 (0.11) 0.213 (0.30)
Fixed industry Yes Yes Yes Yes
and year
effects
Adjusted R2 0.598 0.598 0.597 0.598
No. of observa- 1235
tions

Panels A and B report the effects of CSR report readability and tone on 1-year-ahead and 2-year-ahead CSR performance, respectively, using
FOG as a proxy for readability. C ­ SRPt+2 is firms’ net CSR performance for year t + 2. The other variables are as defined in Table 2. Variables of
interest are highlighted in bold. t statistic is based on robust standard errors clustered by firm and year
*, **, and *** denote significance of coefficients at the 10%, 5%, and 1% levels, respectively, using a two-tailed test.

13
Do Corporate Social Responsibility Reports Convey Value Relevant Information? Evidence from…

Table 5  The effects of CSR report readability and tone on the market reaction around the release of CSR reports
Panel A: The effects of readability and tone on cumulative abnormal trading volume

Dependent Var = CABVOLt

ΔREAD = ΔFOG

I II III

Coeff t stat Coeff t stat Coeff t stat

ΔCSRPt 0.170 (0.88) 0.170 (0.88) 0.166 (0.91)

SIZEt − 0.027 (− 7.76)*** − 0.026 (− 9.75)*** − 0.029 (− 5.13)***
ROAt 0.298 (1.70)* 0.290 (1.71)* 0.298 (1.69)*
LEVt − 0.246 (− 2.39)** − 0.250 (− 2.37)** − 0.240 (− 2.32)**
FINt 0.669 (2.01)** 0.664 (2.04)** 0.655 (2.02)**
LIQUIDt − 0.009 (− 1.11) − 0.009 (− 1.11) − 0.010 (− 1.14)
RDt 0.263 (0.78) 0.273 (0.80) 0.250 (0.73)
AFt − 0.030 (− 0.59) − 0.031 (− 0.61) − 0.031 (− 0.61)
READ_10Kt − 5.410 (− 2.13)** − 5.398 (− 2.14)** − 5.716 (− 2.15)**
ΔREADt 0.334 (5.06)*** 0.340 (5.09)*** 0.344 (5.24)***
ABSΔTONEt 1.799 (0.52)
ABSΔPOSt 3.417 (0.61)
ABSΔNEGt − 7.961 (− 1.18)
Fixed industry and year effects Yes Yes Yes
Adjusted R2 0.039 0.039 0.039
No. of observations 574
Panel B: The effects of readability and tone on cumulative abnormal returns

Dependent Var = CAR​t

ΔREAD = ΔFOG

I II III

Coeff t stat Coeff t stat Coeff t stat

ΔCSRPt − 0.005 (− 1.17) − 0.005 (− 1.09) − 0.004 (− 1.06)

SIZEt 0.001 (1.70)* 0.001 (1.66)* 0.001 (1.63)
ROAt 0.009 (0.45) 0.009 (0.49) 0.012 (0.68)
LEVt − 0.002 (− 0.48) − 0.002 (− 0.51) − 0.001 (− 0.28)
FINt 0.012 (0.77) 0.013 (0.80) 0.008 (0.56)
LIQUIDt − 0.001 (− 1.29) − 0.001 (− 1.31) − 0.001 (− 1.15)
RDt 0.018 (0.97) 0.019 (1.03) 0.015 (0.82)
AFt − 0.003 (− 2.19)** − 0.003 (− 2.23)** − 0.005 (− 3.35)***
READ_10Kt 0.117 (0.67) 0.111 (0.63) 0.160 (0.95)
ΔREADt 0.026 (2.84)*** 0.025 (2.79)*** 0.028 (2.99)***
ΔTONEt 0.188 (2.07)**
ΔPOSt 0.322 (3.08)***
ΔNEGt 0.188 (0.94)
Fixed industry and Yes Yes Yes
year effects
Adjusted R2 0.015 0.016 0.017
No. of observations 574

Panels A and B report the effects of CSR report readability and tone on the cumulative abnormal trading volume and the cumulative abnor-
mal returns, respectively, around the release of CSR reports. C ­ ABVOLt is the cumulative abnormal trading volume during the window (− 1,
1) centered on the release date of CSR reports for year t, calculated as the logarithm of the cumulative trading volume during the three− day
event window minus the logarithm of the firm− specific median cumulative trading volume for contiguous three-day periods over the estimation
period from 100 trading days prior to the event window to 21 trading days prior to this window. CAR​t is the cumulative market-adjusted abnor-
mal returns during the window (− 1, 1) around the release of CSR reports for year t. ABS∆TONEt is the absolute value of ∆TONEt. ABSΔPOS
and ABSΔNEG are equal to the absolute value of ΔPOS and ΔNEG, respectively. The other variables are as defined in Table 2. Variables of
interest are highlighted in bold. t statistic is based on robust standard errors clustered by firm and year
*, **, and *** denote significance of coefficients at the 10%, 5%, and 1% levels, respectively, using a two-tailed test

13
 S. Du, K. Yu

bottom decile of ΔREAD to its top decile could increase Effect of CSR Report Tone on Market Reaction
firm value by $40 million.7 Similarly, moving from the Conditional on Report Readability
bottom decile of ΔTONE to its top decile could increase
firm value by $127 million.8 Overall, the results are con- Franco et al. (2015) find that analyst report readability and
sistent with the view that investors treat increases in the tone reinforce each other such that the effect of tone on
readability and tone of CSR reports as credible signals the market reaction to analyst reports is stronger for more
of higher future CSR performance, thus leading to more readable analyst reports. We examine whether CSR report
favorable market reactions to CSR reports with such tex- readability moderates the association between the abnormal
tual characteristics. returns around the release of CSR reports and tone revi-
When we replace ΔTONE with ΔPOS and ΔNEG in sion in Panel A of Table 7. More specifically, in Column
Columns II and III, respectively, the coefficient on ΔPOS I, we add the interaction between ΔREAD and ΔTONE
is positive (coeff. = 0.322, t stat = 3.08), but the coefficient (ΔREAD*ΔTONE) into model (4). Consistent with Franco
on ΔNEG is not significant (t stat = 0.94). The results are et al. (2015), the results indicate that, ceteris paribus,
consistent with the those documented in Table 4 and suggest improving CSR report readability could enhance the effect
that the positive association between the cumulative abnor- of tone revision on the abnormal returns (coeff. on ΔREAD*
mal returns and ΔTONE in Column I is likely to be driven ΔTONE = 7.224; t stat = 1.89). We replace ΔTONE with
by the change in the proportion of positive words. ΔPOS in Column II. The results are even stronger (coeff. on
ΔREAD* ΔPOS = 8.074; t stat = 3.88), suggesting that the
market reaction to the change in the proportion of positive
Additional Analysis words is more pronounced for more readable CSR reports.

Using Alternative Readability Measures Effect of CSR Report Readability on Market Reaction
Conditional on Analyst Following and Financial
We use two alternative readability measures, FLESCH and Opacity
ARI, based on the FLESCH Reading Ease Score and the
Automated Readability Index, respectively, to check the The literature (Ayers and Freeman 2003) has provided evi-
robustness of our results. FLESCH Reading Ease Score and dence consistent with the intermediary role of financial ana-
Automated Readability Index are calculated as follows: lysts in information generation and capitalization. To the
extent that analysts acquire information in a CSR report from
FLESCH Reading Ease Score = 206.835
alternative sources before its release and accelerate infor-
− (1.015 × words per sentence) − (84.6 × syllables per word) mation pricing, market reaction to CSR reports should be
(5)
less pronounced for firms with more analysts following. Fur-
Automated Readability Index = −21.43 thermore, Dhaliwal et al. (2012) find that issuance of CSR
+ (4.71 × characters per word) + (0.5 × words per sentence) reports reduces analyst forecast error to a greater extent for
(6) firms with a higher level of financial opacity. This suggests
Similar to the construction of FOG, FLESCH is defined that CSR reports play a complementary role in enhancing
as the FLESCH Reading Ease Score divided by 100, and financial transparency and are more useful for firms with
ARI is defined as the Automated Readability Index divided greater financial opacity. We thus posit that market reaction
by − 100, so that higher values of FLESCH and ARI indi- to CSR report readability should be more pronounced for
cate more readable CSR reports. Panels A and B of Table 6 firms with less analyst following and greater financial opac-
report the results using FLESCH and ARI, respectively, as ity. We examine this conjecture using the following model.
the proxy for readability. The results are largely consistent CARt = 𝛽0 + 𝛽1 ΔCSRPt + 𝛽2 SIZEt + 𝛽3 ROAt + 𝛽4 LEV t
with those based on FOG.
+ 𝛽5 FINt + 𝛽6 LIQUIDt + 𝛽7 RDt + 𝛽8 AFt
+ 𝛽9 READ_10Kt + 𝛽10 ΔREADt + 𝛽11 ΔPOSt
+ 𝛽12 AFt ∗ ΔREADt + 𝛽13 FFINt + 𝛽14 FFINt *ΔREADt
+ Industry and Year Fixed Effects + 𝜀t
7 (7)
The increase in market value is computed as follows: the coefficient
on ΔREAD * the interdecile range of ΔREAD * the mean market FFIN is financial opacity, equal to 1 if the absolute value
value (i.e., 0.026*0.028*56.3 = 0.04). of a firm’s scaled accruals averaged over the past three
8
The increase in market value is computed as follows: the coefficient years is higher than the corresponding industry-year mean,
on ΔTONE * the interdecile range of ΔTONE * the mean market
and 0 otherwise (Dhaliwal et al. 2011; Muslu et al. 2019).
value (i.e., 0.188*0.012*56.3 = 0.127).

13
Do Corporate Social Responsibility Reports Convey Value Relevant Information? Evidence from…

Table 6  The effects of CSR report readability and tone on market reaction using alternative readability measures
Panel A: Using FLESCH as the proxy for CSR report readability
ΔREAD = ΔFLESCH
CSRPt +1 CSRPt +2 CABVOLt CAR​t
I II III IV
Coeff t stat Coeff t stat Coeff t stat Coeff t stat

ΔREADt 0.031 (1.81)* 0.036 (2.11)** 0.135 (3.81)*** 0.011 (2.48)**

ΔTONEt 0.808 (4.19)*** 0.388 (1.43) 0.180 (1.83)*
ABSΔTONEt 1.765 (0.51)
Fixed industry and year effects Yes Yes Yes Yes
Control variables Yes Yes Yes Yes
Adjusted R2 0.690 0.598 0.039 0.015
No. of Obs 1258 1235 574 574
Panel B: Using ARI as the proxy for CSR report readability
ΔREAD = ΔARI
CSRPt +1 CSRPt +2 CABVOLt CAR​t
I II III IV
Coeff t stat Coeff t stat Coeff t stat Coeff t stat

ΔREADt 0.039 (1.97)** 0.053 (2.39)** 0.245 (5.33)*** 0.022 (2.65)***

ΔTONEt 0.806 (4.17)*** 0.385 (1.40) 0.188 (2.08)**
ABSΔTONEt 1.808 (0.52)
Fixed industry Yes Yes Yes Yes
and year
effects
Control vari- Yes Yes Yes Yes
ables
Adjusted R2 0.690 0.598 0.039 0.015
No. of Obs 1258 1235 574 574

Panel A (B) of Table 6 reports the results using FLESCH (ARI) as the proxy for readability. Columns I and II present the results based on model
(2). Columns III and IV present the results based on models (3) and (4), respectively. ­FLESCHt is FLESCH Reading Ease Score for CSR reports
for year t divided by 100, where FLESCH Reading Ease Score is calculated as 206.835—(1.015 × words per sentence)—(84.6 × syllables per
word). ­ARItis the Automated Readability Index for CSR reports for year t divided by − 100, where Automated Readability Index is calculated
as − 21.43 + (4.71 × characters per word) + (0.5 × words per sentence). Higher values of ­FLESCHt and ­ARIt indicate greater report readability.
­CSRPt +2 is firms’ net CSR performance for year t + 2. The other variables are as defined in Tables 2 and 5. The control variables are included
but not reported. t statistic is based on robust standard errors clustered by firm and year
*, **, and *** denote significance of coefficients at the 10%, 5%, and 1% levels, respectively, using a two-tailed test

Scaled accruals are computed as follows: (ΔCA – ΔCL is less pronounced for firms with more analyst following.
– ΔCASH + ΔSTD – DEP + ΔTP)/LAGTA, where ΔCA In Column II, the coefficient on FFIN*ΔREAD is positive
(ΔCL) is the change in total current assets (liabilities); (coeff. = 0.239; t stat = 2.38), consistent with the argument
ΔCASH is the change in cash; ΔSTD is the change in the cur- that the market relies upon CSR reports to a greater extent
rent portion of long-term debt; DEP is depreciation and amor- for firms with a higher level of financial opacity. Column III
tization expense; ΔTP is the change in income taxes payable; reports the results based on the full model (7). The results are
and LAGTA is total assets at the end of the previous year. qualitatively similar to those reported in Columns I and II.
The results are reported in Panel B of Table 7. In Column
I, the abnormal returns around the release of CSR reports Addressing Alternative Explanations
are positively associated with ΔREAD, but negatively asso-
ciated with AF*ΔREAD (coeff. = – 0.15; t stat = – 2.11), An alternative explanation for the positive association
suggesting that the market reaction to CSR report readability between future CSR performance and the change in CSR

13
 S. Du, K. Yu

Table 7  Conditional effects of CSR report readability and tone on the abnormal returns
Panel A: The effect of CSR report tone on the abnormal returns conditional on CSR report readability
Dependent Var = CAR​t
ΔREAD = ΔFOG
I II III
Coeff t stat Coeff t stat Coeff t stat

ΔREADt 0.061 (2.72)*** 0.069 (4.14)***

ΔTONEt 0.177 (2.48)**
ΔREADt*ΔTONEt 7.224 (1.89)*
ΔPOSt 0.315 (3.47)***
ΔREADt*ΔPOSt 8.074 (3.88)***
Control variables Yes Yes
Adjusted R2 0.013 0.014
No. of observations 574
Panel B: Market reaction to CSR report readability conditional on analyst following and financial opacity
Dependent Var = CAR​t
ΔREAD = ΔFOG
I II III
Coeff t stat Coeff t stat Coeff t stat

AFt − 0.004 (− 3.25)*** − 0.004 (− 3.83)*** − 0.004 (− 3.09)***

ΔREADt 0.474 (2.21)** 0.023 (3.29)*** 0.672 (3.45)***
ΔPOSt 0.325 (2.84)*** 0.322 (2.79)*** 0.328 (2.88)***
AFt *ΔREADt − 0.150 (− 2.11)** − 0.217 (− 3.35)***
FFINt − 0.007 (− 3.65)*** − 0.008 (− 4.31)***
FFINt *ΔREADt 0.239 (2.38)** 0.268 (2.67)***
Control variables Yes Yes Yes
Adjusted R2 0.016 0.022 0.024
No. of observations 574

Panel A presents the effect of tone revision on the abnormal returns around the release of CSR reports conditional on report readability. Panel B
presents the effect of readability change on the abnormal returns conditional on analyst following and financial opacity. FFIN is financial opac-
ity, equal to 1 if the absolute value of a firm’s scaled accruals averaged over the past three years is higher than the corresponding industry-year
mean, and 0 otherwise. Scaled accruals are computed as follows: (ΔCA – ΔCL – ΔCASH + ΔSTD – DEP + ΔTP)/LAGTA, where ΔCA (ΔCL)
is the change in total current assets (liabilities); ΔCASH is the change in cash; ΔSTD is the change in the current portion of long-term debt; DEP
is depreciation and amortization expense; ΔTP is the change in income taxes payable; and LAGTA is total assets at the end of the previous year.
The other variables are as defined in Tables 2 and 5. The control variables are included but not reported. t statistic is based on robust standard
errors clustered by firm and year
*, **, and *** denote significance of coefficients at the 10%, 5%, and 1% levels, respectively, using a two-tailed test

report tone is that KLD may assess CSR performance by to corporate disclosure, KLD utilizes 100 + specialized
fixating on the tone of prior-year CSR disclosure even if datasets from governments and NGOs, and engages in daily
tone is manipulated upwards, leading to a mechanic positive monitoring of 1600 + media sources (global and local news
relationship between CSR disclosure tone and future CSR sources, government, NGO, and other stakeholder sources);
performance. While we cannot fully rule out this possibil- furthermore, it relies upon systematic communication with
ity, we notice that, to arrive at the KLD ratings, experienced issuers to verify data accuracy and conducts in-depth quality
research analysts apply a same set of criteria to related com- review processes (e.g., specialized research, formal commit-
panies and use data gathered from a wide range of sources, tee review) at all stages of rating.
both internal and external to the firm (Waddock and Graves To the extent that sophisticated KLD analysts could use
1997; Kim et al. 2012). In explaining its KLD rating meth- other information sources to verify information disclosed in
odology, MSCI ESG Research (2018) states that, in addition CSR reports and see through tone management, KLD ratings

13
Do Corporate Social Responsibility Reports Convey Value Relevant Information? Evidence from…

Table 8  The predictability of Dependent Var = RETt+1

CSR report readability and tone ΔREAD = ΔFOG
for 1-year-ahead returns
I II
Coeff t stat Coeff t stat

CAPDt − 0.131 (− 2.75)*** − 0.131 (− 2.76)***

BETADt 0.004 (0.57) 0.004 (0.57)
BTMt − 0.024 (− 0.43) − 0.024 (− 0.43)
PEt − 0.000 (− 2.55)** − 0.000 (− 2.53)**
TACC​t − 0.543 (− 1.61) − 0.544 (− 1.61)
NOAt − 0.114 (− 1.68)* − 0.114 (− 1.68)*
DTAt − 0.047 (− 0.51) − 0.048 (− 0.52)
ΔREADt 0.223 (3.76)*** 0.223 (3.76)***
ΔTONEt − 0.668 (− 0.66)
ΔPOSt − 0.551 (− 0.47)
Fixed industry and year effects Yes Yes
Adjusted R2 0.226 0.226
No. of observations 1,197

Table 8 presents the effects of CSR report readability and tone on future stock returns. ­RETt +1 is 1-year-
ahead stock returns following the release year t of CSR reports. ­CAPDt and ­BETADt are size and beta
deciles at the end of year t from the CSRP database. ­BTMt is the book to market ratio at the end of year t
, calculated as the book value of equity divided by the market value of equity. ­PEt is the price to earnings
ratio at the end of year t , calculated as the fiscal year end stock price divided by the EPS. TACC​t is total
accruals for year t , calculated as the difference between earnings before extraordinary items and cash flow
before extraordinary items scaled by lagged total assets. ­NOAt is net operating assets at the end of year t ,
calculated as the difference between operating assets and operating liabilities scaled by lagged total assets.
­DTAt is the debt to assets ratio, defined as total liabilities divided by total assets. The other variables are as
defined in Table 2. Variables of interest are highlighted in bold. t statistic is based on robust standard errors
clustered by firm and year
*, **, and *** denote significance of coefficients at the 10%, 5%, and 1% levels, respectively, using a two-
tailed test

are less likely to be affected by tone management. Further- RETt+1 is 1-year-ahead stock returns following the release
more, if CSR reports are subject to substantial tone manipu- year t of CSR reports. We control for well documented risk
lation and KLD ratings are driven by the manipulated tone of factors and market anomaly. More specifically, C ­ APDt and
CSR reports, then KLD ratings should not be informative of ­BETADt are size and beta deciles at the end of year t from
financial performance. However, prior studies have provided the CRSP database. B ­ TMt is the book to market ratio at the
ample evidence that KLD ratings are positively associated end of year t , calculated as the book value of equity divided
with financial performance (e.g., Margolis and Walsh 2003), by the market value of equity. ­PEt is the price to earnings
which does not support the view that KLD ratings are based ratio at the end of year t , calculated as the fiscal year end
on distorted tone information, if any, in CSR reports. stock price divided by the EPS. TACC​t is total accruals for
One may argue that the market overreacts to CSR report year t , calculated as the difference between earnings before
readability and tone due to functional fixation. As a result, extraordinary items and cash flow before extraordinary items
the market reaction to CSR report readability and tone as scaled by lagged total assets. ­NOAt is net operating assets
documented in Table 5 may be attributable to market mis- at the end of year t, calculated as the difference between
pricing rather than the value relevance of readability and operating assets and operating liabilities scaled by lagged
tone. To address this concern, we examine whether the total assets. ­DTAt is the debt to assets ratio, defined as total
changes in CSR report readability and tone are negatively liabilities divided by total assets at the end of year t .
predictive of future returns using the following model. The results are reported in Table 8. We find no evidence
that the market overreacts to CSR report readability and
RETt+1 = β0 + β1 CAPDt + β2 BETADt + β3 BTMt + β4 PEt
tone. Instead, 1-year-ahead returns are positively associated
+ β5 TACCt + β6 NOAt + β7 DTAt + β8 ΔREADt with ΔREAD (coeff. = 0.223; t stat = 3.76), but not asso-
+ β9 ΔTONEt + IndustryandYearFixedEffects + εt ciated with ΔTONE or ΔPOS, suggesting that investors
(8)

13
 S. Du, K. Yu

underreact to CSR report readability. The results seem MKTSHARE, AGE, and CAPX are included in model (9)
consistent with the view that investors do not fully under- but excluded from the second stage models. These vari-
stand the long-term implications of CSR report readability ables impose important exclusion restrictions on the sec-
for future CSR performance as documented in Panel B of ond stage estimation. The results are presented in Panel
Table 4. B of Table 9 and are consistent with those documented in
Tables 4 and 5.
Addressing Sample Selection Bias

In this section, we further test the robustness of our results Summary and Discussion
by addressing the potential sample selection bias. Since
our sample only includes firms that issue stand-alone CSR We examine the information content of CSR report readabil-
reports, the OLS estimation may be subject to the potential ity and tone. Using a hand-collected dataset of Fortune 500
sample selection bias. We perform the Heckman two-stage companies that published stand-alone CSR reports for years
procedure (Heckman 1979) to account for the endogenous 2002 to 2014, we find that future CSR performance is posi-
nature of firms’ decision to publish a CSR report or not. tively associated with changes in both readability and tone
Specifically, in the first stage, we estimate the following of CSR reports, suggesting that CSR reports with higher
Probit model. readability and more optimistic tone are indicative of better
future CSR performance. In addition, the positive associa-
DISCt = 𝛽0 + 𝛽1 CSRPt + 𝛽2 SIZEt + 𝛽3 ROAt + 𝛽4 LEVt
tion between future CSR performance and tone change is
+ 𝛽5 FINt + 𝛽6 LIQUIDt + 𝛽7 RDt + 𝛽8 AFt primarily due to the change in the proportion of positive
+ 𝛽9 READ_10Kt + 𝛽10 MKTSHAREt + 𝛽11 AGEt words, suggesting that managers tend to use positive, rather
+ 𝛽12 CAPXt + 𝛽13 ROAVOLt + 𝛽14 FFINt than negative, words to convey information about future
CSR performance.
+ Industry and Year Fixed Effects + 𝜀t
The stock market appears to treat CSR report readability
(9)
and tone as credible signals of future CSR performance and
DISCt is a dummy variable, equal to one if the firm reacts accordingly. Specifically, the change in report read-
releases a CSR report for year t and zero otherwise. The ability is positively associated with both abnormal trading
independent variables are largely taken from prior litera- volume and abnormal returns around the release of CSR
ture on the determinants of CSR disclosure (Dhaliwal et al. reports, consistent with the argument that improved report
2011, 2012). In additional to the control variables specified readability not only reduces information ambiguity, thus
in model (2), we include market share (MKTSHAREs), firm spurring trading activities, but also indicates better future
age (AGE), capital expenditure (CAPX), earnings volatil- CSR performance, thus leading to higher abnormal returns.
ity (ROAVOL), and financial opacity (FFIN) in model (9). Similarly, tone change is positively associated with abnor-
­MKTSHAREst is the firm’s fraction of sales in its two-digit mal returns, and this positive association is primarily due to
SIC industry. A­ GEt is the number of years since a firm’s first the effect of the change in the proportion of positive words,
appearance in CRSP. C ­ APXt is capital expenditure scaled by in line with the view that managers use positive, but not
total assets. R
­ OAVOLt is computed as the standard devia- negative, words to communicate future CSR performance.
tion of the return on assets over the most recent 5 years; at Our results are robust to using alternative readability
least three non-missing observations are required to calcu- measures and further controlling for the sample selection
late ­ROAVOLt. bias. Additional analysis suggests that CSR report readabil-
The results are reported in Panel A of Table 9. DISC is ity influences the effect of report tone such that the market
positively associated with CSRP, SIZE, AF, MKTSHARE, reaction to CSR report tone is more pronounced for firms
AGE, and CAPX, suggesting that larger and older firms as with more readable CSR reports. Furthermore, consist with
well as firms with better CSR performance, more analyst the view that CSR disclosure plays a complementary role
following, higher market share and capital expenditure in improving financial transparency, we find that the mar-
are more likely to issue CSR reports. In addition, DISC is ket reaction to CSR report readability is stronger for firms
negatively associated with LEV, FIN, RD, and READ_10K, with less analyst following and higher financial opacity. In
indicating that firms with higher financial leverage and con- addition, we find no evidence that the market overreacts to
straints, higher R&D intensity, and more readable 10-K CSR report readability and tone. Investors appear to under-
reports are less likely to issue CSR reports. react to CSR report readability, as evidenced by the positive
In the second stage, we add the inverse Mills ratio association between future stock returns and the change in
­(LAMBDAt) computed from model (9) into models (2), CSR report readability.
(3), and (4) as an additional control variable. Note that

13
Do Corporate Social Responsibility Reports Convey Value Relevant Information? Evidence from…

Table 9  Correcting for self-selection bias

Panel A: Determinants of issuance of CSR reports
Dependent Var = DISCt
Coefficient pvalue

CSRPt 2.640 0.000

SIZEt 0.382 0.000
ROAt 0.171 0.618
LEVt − 0.640 0.000
FINt − 0.671 0.053
LIQUIDt 0.014 0.269
RDt − 1.183 0.098
AFt 0.118 0.025
READ_10Kt − 7.822 0.058
MKTSHAREst 1.448 0.000
AGEt 0.004 0.000
CAPXt 1.555 0.022
ROAVOLt 0.265 0.565
FFINt − 0.001 0.990
Industry and year dummies Yes
Pseudo R2 0.447
Likelihood ratio 1879.6
No. of observations with dep. var. = 1 1543
No. of observations 4900
Panel B: The second stage estimation results
ΔREAD = ΔFOG
Dependent variable
CSRPt +1 CSRPt +2 CABVOLt CAR​t
I II III IV

Coeff t stat Coeff t stat Coeff t stat Coeff t stat

ΔREADt 0.045 (2.16)** 0.067 (2.75)*** 0.341 (6.11)*** 0.025 (3.02)***
ΔTONEt 0.783 (4.29)*** 0.289 (1.10) 0.203 (2.16)**
ABSΔTONEt 0.839 (0.28)
LAMBDAt − 0.037 (− 2.03)** − 0.047 (− 1.79)* 0.232 (2.57)** 0.004 (0.76)
Control variables Yes Yes Yes Yes
Fixed industry and year effects Yes Yes Yes Yes
Adjusted R2 0.691 0.599 0.040 0.015
No. of observation 1258 1235 574 574

Panel A reports the determinants of firms’ decision to issue CSR reports in the first stage. Panel B reports the results in the second stage. D­ ISCt
is a dummy variable, equal to one if the firm releases a CSR report for year t and zero otherwise. M ­ KTSHAREst is the firm’s fraction of sales in
its two-digit SIC industry for year t . ­AGEt is the number of years since a firm’s first appearance in CRSP. ­CAPXt is capital expenditure scaled
by total assets. R
­ OAVOLt is computed as the standard deviation of the return on assets over the most recent 5 years. At least three non-missing
observations are required to calculate R ­ OAVOLt. ­FFINt is financial opacity, equal to 1 if the absolute value of a firm’s scaled accruals averaged
over the past 3 years is higher than the corresponding industry-year mean, and 0 otherwise. ­CSRPt+2 is firms’ net CSR performance for year
t + 2. The other variables are as defined in Tables 2 and 5. Variables of interest are highlighted in bold. t statistic is based on robust standard
errors clustered by firm and year
*, **, and *** denote significance of coefficients at the 10%, 5%, and 1% levels, respectively, using a two-tailed test

This study contributes to the literature on discretionary reports, suggesting that CSR reports play an important role
information disclosure, particularly CSR reporting (Mar- in reducing information asymmetry by imparting value
tinez-Ferrero et al. 2016; Muslu et al. 2019). Our results relevant information to investors. Our results highlight
provide direct evidence for the information content of CSR the importance of examining the textual properties (i.e.,

13
 S. Du, K. Yu

readability and tone) of CSR reports as they serve as cred- References

ible signals of future CSR performance and affect the effec-
tiveness of CSR disclosure. This study also contributes to Ailawadi, K. L., Neslin, S. A., Luan, Y. J., & Taylor, G. A. (2014).
the broad literature on business ethics, in particular, that of Does retailer CSR enhance behavioral loyalty? A case for benefit
segmentation. International Journal of Research in Marketing,
stakeholder management. Information asymmetry gives rise 31(2), 156–167.
to moral hazard problems and is fertile ground for unethi- Ayers, B. C., & Freeman, R. N. (2003). Evidence that analyst following
cal corporate behavior (Kulkarni 2000; Jones et al. 2018). and institutional ownership accelerate the pricing of future earn-
Stakeholders demand information transparency to make ings. Review of Accounting Studies, 8(1), 47–67.
Beaver, W. H. (1968). The information content of annual earnings
informed decisions and monitor corporate behaviors. In announcements. Journal of Accounting Research, 6, 67–92.
line with instrumental stakeholder theory (Freeman 1999), Bloomfield, R. J. (2002). The “incomplete revelation hypothesis” and
our findings suggest that there are tangible financial benefits financial reporting. Accounting Horizons, 16, 233–243.
associated with providing transparent CSR information to Barth, M. E., Beaver, W. H., & Landsman, W. R. (1998). Relative
valuation roles of equity book value and net income as a func-
the market. More generally, we not only document the busi- tion of financial health. Journal of Accounting and Economics,
ness returns to CSR reporting, but also highlight the means 25(1), 1–34.
to effective CSR disclosure (i.e., communicating good future CDP. (2019). Becoming an Investor Member. Retrieved July 20, 2019
CSR performance through a positive tone and readable text). from www.cdp.net/en/inves​tor/signa​torie​s-and-membe​rs.
Cho, C. H., Michelon, G., Patten, D. M., & Roberts, R. W. (2015). CSR
This research provides important implications for com- disclosure: the more things change…? Accounting, Auditing and
panies, investors, and regulators. Our results highlight the Accountability Journal., 28(1), 14–35.
important role of CSR disclosure in reducing information Cho, C. H., Roberts, R. W., & Patten, D. M. (2010). The language of
asymmetry between firms and investors, especially for firms US corporate environmental disclosure. Accounting, Organiza-
tions and Society, 35(4), 431–443.
with lower analyst following and higher financial opacity. Clarkson, P. M., Li, Y., Richardson, G. D., & Vasvari, F. P. (2008).
Given the significant market reaction to CSR report read- Revisiting the relation between environmental performance and
ability and tone, firms with superior CSR performance can environmental disclosure: An empirical analysis. Accounting,
maximize the benefits of CSR disclosure by improving CSR Organizations and Society, 33(4–5), 303–327.
Cready, W. M., & Hurtt, D. N. (2002). Assessing investor response to
report readability. Furthermore, regulators (e.g., SEC and information events using return and volume metrics. The Account-
SASB) interested in assessing and improving the effective- ing Review, 77(4), 891–909.
ness of CSR disclosure could look into the textual proper- Cui, J., Jo, H., & Na, H. (2018). Does corporate social responsibil-
ties of CSR reports and provide guidelines and toolbox on ity affect information asymmetry? Journal of Business Ethics,
148(3), 549–572.
how firms could use simple language and appropriate tone Das Neves, J. C., & Vaccaro, A. (2013). Corporate transparency: A
to truthfully communicate value relevant CSR information. perspective from Thomas Aquinas’ summa theologiae. Journal
There are several avenues for future research. First, while of Business Ethics, 113(4), 639–648.
we focus on CSR reports issued by Fortune 500 companies Davis, A. K., Ge, W., Matsumoto, D., & Zhang, J. L. (2015). The
effect of manager-specific optimism on the tone of earnings
and the US stock market, future research can extend our conference calls. Review of Accounting Studies, 20(2), 639–673.
inquiry by examining readability and tone of CSR reports Davis, A. K., Piger, J. M., & Sedor, L. M. (2012). Beyond the num-
issued by smaller companies and the implications of these bers: Measuring the information content of earnings press
textual characteristics for other stock markets. Second, release language. Contemporary Accounting Research, 29(3),
845–868.
despite the popular use of a pre-existing word list for tone Dhaliwal, D. S., Li, O. Z., Tsang, A., & Yang, Y. G. (2011). Voluntary
analysis, such dictionary-based approach does not consider nonfinancial disclosure and the cost of equity capital: The initia-
the individual context of a negative or positive word (Li tion of corporate social responsibility reporting. The Accounting
2010; Loughran and McDonald 2016). Future research Review, 86(1), 59–100.
Dhaliwal, D. S., Radhakrishnan, S., Tsang, A., & Yang, Y. G. (2012).
should employ alternative methods for tone analysis to cor- Nonfinancial disclosure and analyst forecast accuracy: Interna-
roborate our findings, such as a statistical approach (e.g., tional evidence on corporate social responsibility disclosure. The
Bayesian machine learning method) that could use algo- Accounting Review, 87(3), 723–759.
rithms to analyze the statistical correlations between key Du, S., Bhattacharya, C. B., & Sen, S. (2010). Maximizing business
returns to corporate social responsibility (CSR): The role of CSR
words and more precisely classify the sentiments of the text. communication. International Journal of Management Reviews,
12(1), 8–19.
Acknowledgements We appreciate the research assistance from Jia Du, S., Yu, K., Bhattacharya, C. B., & Sen, S. (2017). The business
Gao and Minyi Chen in data collection. case for sustainability reporting: Evidence from stock market reac-
tions. Journal of Public Policy and Marketing, 36(2), 313–330.
Compliance with Ethical Standards Ertugrul, M., Lei, J., Qiu, J., & Wan, C. (2017). Annual report read-
ability, tone ambiguity, and the cost of borrowing. Journal of
Financial and Quantitative Analysis, 52(2), 811–836.
Conflict of interest The authors declare that they have no conflict of
interest.

13
Do Corporate Social Responsibility Reports Convey Value Relevant Information? Evidence from…

Feldman, R., Govindaraj, S., Livnat, J., & Segal, B. (2010). Manage- Loughran, T., & McDonald, B. (2016). Textual analysis in accounting
ment’s tone change, post earnings announcement drift and accru- and finance: A survey. Journal of Accounting Research, 54(4),
als. Review of Accounting Studies, 15(4), 915–953. 1187–1230.
Franco, G., Hope, O. K., Vyas, D., & Zhou, Y. (2015). Analyst report Luo, X., & Bhattacharya, C. B. (2006). Corporate social responsibility,
readability. Contemporary Accounting Research, 32(1), 76–104. customer satisfaction, and market value. Journal of Marketing,
Freeman, R. E. (1999). Divergent stakeholder theory. Academy of Man- 70(4), 1–18.
agement Review, 24(2), 233–236. Luo, X., & Du, S. (2015). Exploring the relationship between corpo-
Fombrun, C. J., Gardberg, N. A., & Barnett, M. L. (2000). Opportunity rate social responsibility and firm innovation. Marketing Letters,
platforms and safety nets: Corporate citizenship and reputational 26(4), 703–714.
risk. Business and Society Review, 105(1), 85–106. Mahoney, L. S., Thorne, L., Cecil, L., & LaGore, W. (2013). A research
Godfrey, P. C., Merrill, C. B., & Hansen, J. M. (2009). The relationship note on standalone corporate social responsibility reports: Sign-
between corporate social responsibility and shareholder value: aling or greenwashing? Critical Perspectives on Accounting,
An empirical test of the risk management hypothesis. Strategic 24(4–5), 350–359.
Management Journal, 30(4), 425–445. Margolis, J. D., Elfenbein, H. A., & Walsh, J. P. (2007). Does it pay
Governance and Accountability Institute. (2018). 85% of S&P 500 to be good? A meta-analysis and redirection of research on the
Index Companies Published Sustainability Reports in 2017. relationship between corporate social and financial performance.
Retrieved July 20, 2019 from https​://www.ga-insti​tute.com/press​ Annals of Arbor, 1001, 48109–51234.
-relea​ses/artic​le/flash​-repor ​t-85-of-sp-500-index​R-compa​nies- Margolis, J. D., & Walsh, J. P. (2003). Misery loves companies:
publi​sh-susta​inabi​lity-repor​ts-in-2017.html. Rethinking social initiatives by business. Administrative Science
Hasan, I., Kobeissi, N., Liu, L., & Wang, H. (2018). Corporate social Quarterly, 48(2), 268–305.
responsibility and firm financial performance: The mediating role Martínez-Ferrero, J., Ruiz-Cano, D., & García-Sánchez, I. M. (2016).
of productivity. Journal of Business Ethics, 149(3), 671–688. The causal link between sustainable disclosure and information
Heckman, J. J. (1979). Sample selection bias as a specification error. asymmetry: The moderating role of the stakeholder protection
Econometrica, 47(1), 153–161. context. Corporate Social Responsibility and Environmental Man-
Huang, X., Teoh, S. H., & Zhang, Y. (2014). Tone management. The agement, 23(5), 319–332.
Accounting Review, 89(3), 1083–1113. McWilliams, A., & Siegel, D. (2000). Corporate social responsibil-
Jones, T. M., Harrison, J. S., & Felps, W. (2018). How applying instru- ity and financial performance: Correlation or misspecification?
mental stakeholder theory can provide sustainable competitive Strategic Management Journal, 21(5), 603–609.
advantage. Academy of Management Review, 43(3), 371–391. Miller, B. P. (2010). The effects of reporting complexity on small and
Khan, M., Serafeim, G., & Yoon, A. (2016). Corporate sustainabil- large investor trading. The Accounting Review, 85(6), 2107–2143.
ity: First evidence on materiality. The Accounting Review, 91(6), Mishra, S., & Suar, D. (2010). Does corporate social responsibility
1697–1724. influence firm performance of Indian companies? Journal of Busi-
Kim, O., & Verrecchia, R. E. (1991). Trading volume and price reac- ness Ethics, 95(4), 571–601.
tions to public announcements. Journal of Accounting Research, MSCI ESG Research. (2018). MSCI ESG Ratings Methodology. https​
29(2), 302–321. ://www.msci.com/docum​ents/10199​/123a2​b2b-1395-4aa2-a121-
Kim, O., & Verrecchia, R. E. (1994). Market liquidity and volume ea14d​e6d70​8a.
around earnings announcements. Journal of Accounting and Eco- Muslu, V., Mutlu, S., Radhakrishnan, S., & Tsang, A. (2019). Corpo-
nomics, 17(1–2), 41–67. rate social responsibility report narratives and analyst forecast
Kim, Y., Park, M. S., & Wier, B. (2012). Is earnings quality associ- accuracy. Journal of Business Ethics, 154(4), 1119–1142.
ated with corporate social responsibility? The Accounting Review, Nazari, J., Hrazdil, K., & Mahmoudian, F. (2017). Assessing social
87(3), 761–796. and environmental performance through narrative complexity in
Klein, J., & Dawar, N. (2004). Corporate social responsibility and CSR reports. Journal of Contemporary Accounting and Econom-
consumers’ attributions and brand evaluations in a product–harm ics, 13, 179–191.
crisis. International Journal of Research in Marketing, 21(3), Parguel, B., Benoît-Moreau, F., & Larceneux, F. (2011). How sustain-
203–217. ability ratings might deter ‘greenwashing’: A closer look at ethi-
KPMG. (2016). Carrots and Sticks: Global Trends in Sustainability cal corporate communication. Journal of Business Ethics, 102(1),
Reporting Regulation and Policy. Retrieved July 20, 2019 from 15–28.
https​://asset​s.kpmg.com/conte​nt/dam/kpmg/pdf/2016/05/carro​ts- Patten, D. M. (1992). Intra-industry environmental disclosures in
and-stick​s-may-2016.pdf. response to the Alaskan oil spill: A note on legitimacy theory.
Kulkarni, S. P. (2000). Environmental ethics and information asym- Accounting, Organizations and Society, 17(5), 471–475.
metry among organizational stakeholders. Journal of Business Peng, C. W., & Yang, M. L. (2014). The effect of corporate social
Ethics, 27(3), 215–228. performance on financial performance: The moderating effect
Lang, M., & Stice-Lawrence, L. (2015). Textual analysis and inter- of ownership concentration. Journal of Business Ethics, 123,
national financial reporting: Large sample evidence. Journal of 171–182.
Accounting and Economics, 60(2–3), 110–135. Perrini, F. (2006). SMEs and CSR theory: Evidence and implications
Li, F. (2008). Annual report readability, current earnings, and earn- from an Italian perspective. Journal of Business Ethics, 67(3),
ings persistence. Journal of Accounting and Economics, 45(2), 305–316.
221–247. Petersen, M. A. (2009). Estimating standard errors in finance panel
Li, F. (2010). Survey of the literature. Journal of Accounting Literature, data sets: Comparing approaches. The Review of Financial Stud-
29, 143–165. ies, 22(1), 435–480.
Loughran, T., & McDonald, B. (2011). When is a liability not a lia- PWC. (2019). Responsible investment: Aligning interests. Retrieved
bility? Textual analysis, dictionaries, and 10-Ks. The Journal of July 20, 2019 from https​://www.pwc.com/gx/en/servi​ces/susta​
Finance, 66(1), 35–65. inabi​lity/publi​catio​ns/respo​nsibl​e-inves​tment​.html.
Loughran, T., & McDonald, B. (2014). Measuring readability in finan-
cial disclosures. The Journal of Finance, 69(4), 1643–1671.

13
 S. Du, K. Yu

Schrand, C. M., & Walther, B. R. (2000). Strategic benchmarks in Torres, A., Bijmolt, T. H., Tribó, J. A., & Verhoef, P. (2012). Generat-
earnings announcements: The selective disclosure of prior-period ing global brand equity through corporate social responsibility to
earnings components. The Accounting Review, 75(2), 151–177. key stakeholders. International Journal of Research in Marketing,
Sustainability Accounting Standards Board (SASB). (2017). SASB 29(1), 13–24.
conceptual framework. https​://www.sasb.org/wp-conte​nt/uploa​ Tschopp, D., & Huefner, R. J. (2015). Comparing the evolution of
ds/2019/05/SASB-Conce​ptual​-Frame​work.pdf. CSR reporting to that of financial reporting. Journal of Business
Securities and Exchange Commission (SEC). (2016). Concept release: Ethics, 127(3), 565–577.
Business and financial disclosure required by regulation S-K. https​ Turban, D. B., & Greening, D. W. (1997). Corporate social perfor-
://www.sec.gov/rules​/conce​pt/2016/33-10064​.pdf. mance and organizational attractiveness to prospective employees.
Servaes, H., & Tamayo, A. (2013). The impact of corporate social Academy of Management Journal, 40(3), 658–672.
responsibility on firm value: The role of customer awareness. Waddock, S. A., & Graves, S. B. (1997). The corporate social perfor-
Management Science, 59(5), 1045–1061. mance-financial performance link. Strategic Management Jour-
Smith, N. C. (2003). Corporate social responsibility: Whether or how? nal, 18(4), 303–319.
California management review, 45(4), 52–76. Wang, Z., Hsieh, T., & Sarkis, J. (2018). CSR performance and the
Surroca, J., Tribó, J. A., & Waddock, S. (2010). Corporate responsibil- readability of CSR reports: Too good to be true? Corporate Social
ity and financial performance: The role of intangible resources. Responsibility and Environmental Management, 25, 66–79.
Strategic Management Journal, 31(5), 463–490.
Tetlock, P. C., Saar-Tsechansky, M., & Macskassy, S. (2008). More Publisher’s Note Springer Nature remains neutral with regard to
than words: Quantifying language to measure firms’ fundamen- jurisdictional claims in published maps and institutional affiliations.
tals. The Journal of Finance, 63(3), 1437–1467.

13