Journal of Banking & Finance 59 (2015) 409–422

Contents lists available at ScienceDirect

Journal of Banking & Finance

journal homepage: www.elsevier.com/locate/jbf

Corporate social responsibility and media coverage

Steven F. Cahan a,⇑, Chen Chen b,1, Li Chen a,2, Nhut H. Nguyen c,3
a
University of Auckland Business School, Private Bag 92019, Auckland 1142, New Zealand
b
Monash University Business School, 900 Dandenong Road, Caulfield East 3145, VIC, Australia
c
School of Economics and Finance, Massey University, Private Bag 102 904, Auckland 0745, New Zealand

a r t i c l e i n f o a b s t r a c t

Article history: In this study, we examine whether firms that act more socially responsible receive more favorable media
Received 27 January 2015 coverage, and we consider whether firms use CSR to actively manage their media image. We focus on all
Accepted 7 July 2015 news stories about a firm, not just those that report on specific CSR initiatives, and find that more socially
Available online 20 July 2015
responsible firms receive more favorable news reportage overall, i.e., they have a more positive media
image. These findings are robust after controlling for potential endogeneity. Further, consistent with
JEL classification: firms actively managing their media image, we find a stronger relation between CSR and media favora-
G10
bility when incentives to improve a firm’s media image are high, e.g., among firms in sin industries, dur-
L82
M14
ing periods of low investor sentiment, and prior to seasoned equity offerings. Finally, we find that for
firms that demonstrate superior social responsibility and receive more favorable news reporting, there
Keywords: is a significant interaction between social responsibility and media favorability that increases (decreases)
Media management a firm’s equity valuation (cost of capital). Our results are consistent with the media slanting their report-
Corporate social responsibility ing in favor of good performing CSR firms. Overall, we contribute to the literature by showing that firms
Business press can influence their media coverage through a relatively subtle channel, CSR performance.
! 2015 Elsevier B.V. All rights reserved.

1. Introduction 2011; Dougal et al., 2012; Kim et al., 2014a). Such studies assume
that media coverage is exogenous. In contrast, as Ahern and
It is well documented that the business press plays an impor- Sosyura (2014) suggest, causality could run in the opposite direc-
tant role as an information intermediary and that media coverage tion, i.e., firms may take actions to influence the media coverage
affects a firm’s information environment (e.g., Tetlock et al., 2008; they receive. Consistent with this notion, Gurun and Butler
Fang and Peress, 2009; Engelberg and Parsons, 2011; Griffin et al., (2012) find that the media respond to pressure from local advertis-
ers. Solomon (2012) finds that media coverage is more favorable
for firms that use an investor relations firm. Ahern and Sosyura
We thank Henk Berkman, Gary Biddle, Charl de Villiers, Fei Du, Torsten Jochem, (2014) find that firms with fixed stock exchange ratios issue fewer
Charlene Lee, John Lee, Donghui Li, Dimitri Margaritas, Michaela Rankin, Irene
negative press releases during the period when the exchange ratio
Tutticci, Anne Wyatt, Rencheng Wang, Xin Wang, Joyce Yu, Ying Zheng, and seminar
participants at the University of Auckland, University of Hong Kong, University of is set. We extend this limited line of research and examine whether
Otago, University of Queensland, University of Technology, Sydney, Sun Yat-Sen firms can manage their media coverage through a more subtle
University, 2014 AFAANZ Conference, 2014 AAA Annual Meeting, 2015 Financial channel, i.e., by performing better in terms of corporate social
Markets and Corporate Governance Conference, and 2015 FMA European Confer- responsibility (CSR).
ence for their comments and suggestions. We also thank Thomson Reuters for
We focus on CSR for practical and theoretical reasons. First,
providing access to the Thomson Reuters News Analytics database, and we
appreciate the assistance of Maciej Pomalecki and Xuehui Hou in accessing the investors and managers are increasingly concerned about CSR per-
data. We acknowledge research funding provided by the University of Auckland formance. For example, over 1250 institutional investors world-
Business School. wide have agreed to support the United Nation’s ‘Principles of
⇑ Corresponding author. Tel.: +64 9 923 7175.
Responsible Investment’ (PRI) project that encourages responsible
E-mail addresses: [email protected] (S.F. Cahan), [email protected]
investing by institutional investors based on factors including
(C. Chen), [email protected] (L. Chen), [email protected] (N.H.
Nguyen). environmental and social performance (PRI Annual Report 2014),
1
Tel.: +61 3 9903 4566. and in 2010, $3.07 trillion were invested in professionally managed
2
Tel.: +64 9 923 7484. US assets that ascribed to socially responsible investing
3
Tel.: +64 9 414 0800x43179.

http://dx.doi.org/10.1016/j.jbankfin.2015.07.004
0378-4266/! 2015 Elsevier B.V. All rights reserved.
410 S.F. Cahan et al. / Journal of Banking & Finance 59 (2015) 409–422

(Di Giuli and Kostovetsky, 2014). In 2012, large US firms spent $28 In the first part of the study, we consider whether good CSR
billion on sustainability and $15 billion on corporate philanthropy firms have a more positive overall media image. We regress media
(Di Giuli and Kostovetsky, 2014), and 93% of CEOs responding to a favorability on the firm’s CSR performance and find a positive rela-
2010 UN survey stated that CSR was ‘important’ or ‘very important’ tion, consistent with better performing CSR firms receiving more
to the future success of their firm (Cheng et al., 2013). Second, positive news coverage from the business press. However, we
Fombrun and Shanley (1990) develop a model of reputation build- acknowledge that the endogeneity issues, such as omitted vari-
ing where social responsiveness in the prior period is one factor ables and reverse causality, may confound our tests.
that can affect a firm’s media image in the current period. Supplemental analyses using instrumental variables and propen-
However, to our knowledge, the link between CSR and a firm’s sity score matching support our main results. Further, we conduct
overall media image has not been empirically examined. a quasi-natural experiment based on an executive order signed by
For CSR to affect a firm’s wider media coverage, the media must President Barack Obama on September 25, 2012 that strengthened
have incentives to report more positively about good CSR firms. protections against labor trafficking by federal contractors and
Mullainathan and Shleifer (2005) develop a demand-side model of subcontractors.5 This executive order would have affected CSR per-
media slant in which the media caters to the beliefs of their readers. formance in the human rights area, but it would not have had a
They define slanting as a process where, given a set of facts, the direct effect on overall media favorability. The results of this analysis
media portrays a subject favorably or unfavorably through selective support a causal relation where CSR performance affects media
reporting. This model assumes that readers prefer news that is con- favorability.
sistent with their beliefs – an assumption that is supported by In the second part of the study, we consider whether firms
research in communications (e.g., Graber, 1984; Severin and actively manage CSR in order to obtain more favorable media
Tankard, 1992) and psychology (e.g., Bartlett, 1932; Klayman, image. In other words, a positive relation between CSR and media
1995). Thus, slanting attracts readers, and Mullainathan and favorability could arise if firms engage in CSR for other reasons
Shleifer (2005) show that slant can persist even when readers have (e.g., altruism) in which case a positive media image is just an
homogeneous beliefs and when there are competing media sources. attractive side-benefit. To provide evidence on active media man-
Surveys of public opinion consistently show that corporate agement, we conduct three tests. First, we explore whether the
behavior matters (e.g., Epstein-Reeves, 2010; Grant Thornton, relation between CSR and media favorability is stronger in the
2011; Cone, 2013). For example, Epstein-Reeves (2010) finds that so-called ‘sin’ industries (i.e., alcohol, gambling, and tobacco). We
88% of consumers think companies should try to achieve their expect that sin firms would have greater incentives to promote a
business goals while improving society and the environment, and more positive image, and we find a stronger relation between
83% think companies should support charities and non-profits with CSR and media favorability for sin firms. Second, we expect that
financial donations. In our context, if the public prefers companies all firms would have more incentive to increase media favorability
that are good corporate citizens and wants them to succeed, when the prevailing market-wide investor sentiment is more pes-
Mullainathan and Shleifer’s (2005) catering theory predicts that simistic. Since investor sentiment reflects a broad social mood that
the media would provide more favorable coverage of firms with can affect investor behavior generally, firms would be motivated to
good CSR performance. Consistent with this notion, El Ghoul counteract investors’ pessimistic outlook. We find that the relation
et al. (2011, 2390) suggest that ‘‘the media are more inclined to between CSR and media favorability is stronger when investor sen-
spend time analyzing and reporting news about ‘good’ [CSR] timent is low, consistent with firms actively managing CSR during
firms’’, although they do not test this expectation. Further, if firms these periods. Third, we expect firms to use CSR to manage their
that are good CSR performers receive more favorable media cover- media image before a seasoned equity offerings. For a sample of
age, managers have incentives to actively manage their media SEO firms which have less favorable media coverage before SEO,
image by being more socially responsible, as long as a positive we compare their CSR performance two years before the SEO with
media image is economically beneficial. their CSR performance in the year before and year of the SEO. We
Consequently, we consider three research questions derived find that CSR performance is higher in the SEO period, consistent
from the discussion above. First, is there a link between good cor- with firms trying to generate a more positive media image in the
porate citizenship and favorable media coverage? Second, to the lead-up to the SEO.
extent there is a link, do firms actively manage CSR to gain more In the third part of the study, we explore the economic benefits
favorable media coverage? Third, to consider the economic incen- associated with better CSR performance and more favorable media
tives that motivate such behavior, do firms with better CSR perfor- coverage. If there are no economic benefits, it is difficult to argue
mance and more favorable media coverage benefit in terms of a that managers have incentives to use CSR to actively manage their
higher firm value or lower cost of capital? We divide our study into media image. El Ghoul et al. (2011) find that firms with better CSR
three parts to examine each of these questions. performance have a lower cost of capital. They speculate that
We rate news favorability using a measure of news tone or sen- media coverage plays a role because, based on Merton (1987),
timent from Thomson Reuters News Analytics (TRNA). TRNA investors cannot invest in securities that they do not know about.
employs a lexical analysis that uses a knowledge-driven neural If the media provides more coverage, and more favorable coverage,
network to score each news item released about a firm in terms of good CSR firms, investors’ awareness and interest in these firms
of its positivity and negativity. Using all news stories about a firm increase. While El Ghoul et al. (2011) examine the link between
in a period, we compute the difference between the positive and CSR performance and cost of capital, they do not explore the media
negative news sentiment scores from TRNA to proxy for media link.
favorability.4 Our media dataset includes 469,550 unique news sto- We use firm value and the implied cost of capital to measure
ries during the period 2003–2011. We use CSR performance data economic benefits. We find that for firms that have both high
from the KLD (now MSCI ESG STATS) database. CSR performance and more positive media coverage, the coefficient
of the interaction between CSR performance and media favorabil-
ity is positive (negative) and significant for the Tobin’s Q (cost of
4
Consistent with Gurun and Butler (2012), who examine the relation between
advertising and overall media coverage, our measure of media favorability is based on
5
all articles about a firm in a period, not only articles that report on CSR activity. The full content of the executive order is available on the following webpage:
However, in sensitivity tests, we exclude the CSR-related articles in computing media http://www.whitehouse.gov/the-press-office/2012/09/25/executive-order-strength-
favorability. Our results are qualitatively unchanged. ening-protections-against-trafficking-persons-fe.
 S.F. Cahan et al. / Journal of Banking & Finance 59 (2015) 409–422 411

capital) test. These results suggest that firms that successfully The KLD rating is an annual measure where the rating, which is
manage their media favorability by performing better in the CSR typically released in June or July, is based on information pertain-
arena realize economic benefits from their positive media image. ing to the prior calendar year. For example, a KLD rating issued in
Our study contributes to the literature in two important ways. June 2011 is based on CSR data from the 2010 calendar year. Thus,
First, we contribute to an emerging line of research that considers the KLD rating is essentially a lagged measure relative to our mea-
how firms may proactively manage their media image (e.g., Gurun sure of media favorability (which is described in below). For exam-
and Butler, 2012; Solomon, 2012; Ahern and Sosyura, 2014). We ple, in 2011 for a firm with a December 31 year-end, we compute
add to this limited line of research by examining a less obvious media favorability using all articles from January 1, 2011 to
channel that firms can use to manage their media image, namely December 31, 2011 and we use the KLD rating released in 2011,
their CSR performance. Second, we contribute to an expansive lit- which is based on CSR data from January 1, 2010 to December
erature on CSR performance by documenting a previously unex- 31, 2010. For a small subset of firms (e.g., a firm with a
plored incentive for firms engaging in CSR, i.e., active media September 30 year-end), there could be some overlap between
management.6 our CSR and media measures. We address this concern in
The remainder of the paper is organized as follows. Section 2 Section 3.3.
describes the research design. Sections 3, 4, and 5 report the results
of parts 1, 2, and 3 of the study, respectively. Section 6 concludes. 2.2.2. Media favorability
For every firm in our sample, we compute an annual score of
media favorability (Media) equal to the aggregated positive senti-
2. Research design ment scores of press-initiated news less the aggregated negative
sentiment scores of press-initiated news for the year t scaled by
2.1. Data and sample total number of press-initiated news articles in that year. We use
an annual measure of media favorability since we are interested
We collect media sentiment data from the TRNA database from in whether good CSR improves the firm’s reputation among the
January 2003 to December 2011. A full description of the classifica- media in general.
tion and scoring process used by TRNA is provided in Appendix A.
TRNA provides three sentiment scores, ranging from 0 to 1, that
3. Part 1: Effect of CSR on media favorability
reflect the probability that a specific news item is positive, nega-
tive, or neutral (the scores sum to 1). Following Bushee et al.
3.1. Regression model
(2010), we classify news items based on their source, specifically,
press-initiated or firm-initiated. We classify news items carried
To examine the relation between CSR performance and media
on a press release wire (e.g., Business Wire, Prime Newswire, PR
favorability, we estimate the following model:
Newswire) as firm-initiated. Since our objective is to study the
impact of CSR on the news media, we focus on press-initiated news Mediait ¼ b0 þ b1 CSRit þ b2 NStorit þ b3 ROAit þ b4 BPit þ b5 MV it
items, although we control for firm-initiated news since the press
þ b6 Lev it þ b7 SP500it þ b8 IRiskit þ b9 Ret it
could be influenced by press releases issued by the firm (e.g.,
Solomon, 2012; Ahern and Sosyura, 2014). Finally, we reiterate þ b10 AdExpit þ b11 CGov it þ b12 FI-Newsit
that our measure of media favorability is based on all news reports þ b13 MktPEit þ eit ð1Þ
about a firm, not just those related to CSR activity, as we are inter-
ested in the media’s overall treatment of the firm, i.e., the firm’s where Media and CSR are defined above. If firms’ CSR behavior has a
media image.7 positive impact on firms’ media favorability, we expect b1, the
We extract data from KLD, Compustat, CRSP and I/B/E/S data- coefficient of interest, to be positive.
bases to calculate the remaining variables in our regression mod- We control for the number of news stories in year t (NStor)
els. The sample size varies depending on the analysis. We have a because we are interested in the firm’s media image as opposed
maximum of 12,749 firm-year observations. to the level of news coverage. We also include control variables
for the firm’s fundamentals, stock performance, risk, and visibility
since they may affect media favorability. Hence, we control for
2.2. Measurement of variables return on assets (ROA), the book-to-market ratio (BP), size (MV),
leverage (Lev), visibility (SP500), idiosyncratic risk (IRisk), and stock
2.2.1. CSR performance performance (Ret). Following Gurun and Butler (2012), we also
KLD evaluates firms’ CSR performance in seven qualitative issue include the firm’s advertising expenditure (AdExp) as they docu-
areas, i.e., corporate governance, community, diversity, employee ment a relation between advertising expense and media favorabil-
relations, environment, human rights, and product. We compute ity. In addition, as mentioned above, we control for KLD’s rating of
a total CSR score (CSR) for each firm at the end of year t by sum- the firm’s corporate governance (CGov). Following Ahern and
ming the net scores (strengths minus concerns) of community, Sosyura (2014), we control for firm-initiated news (FI-News) which
diversity, employee relations, environment, human rights, and we compute as the aggregated positive sentiment scores of
product.8 We include the corporate governance score as a control firm-initiated news less the aggregated negative sentiment scores
variable rather than as part of CSR (e.g., Kim et al., 2012; Di Giuli of firm-initiated news for year t scaled by total number of
and Kostovetsky, 2014). firm-initiated news articles in year t. Finally, we control for the
general state of the market which may affect the way journalists
6
Margolis et al. (2009) conduct a meta-analysis of 214 papers from 1972–2007 just interpret firm-specific events. To measure market-wide investor
on one aspect of CSR performance, i.e., the relation between CSR performance and sentiment, we follow Conrad et al. (2002) and use the
financial performance. value-weighted average of the market price-to earnings ratio of
7
In contrast, Kruger (2014) examines how investors respond to specific
all firms in the merged CRSP-Compustat database in year t
CSR-related news events.
8
Our results are robust to an alternative definition of CSR performance based on
(MktPE). We estimate Eq. (1) with industry and year fixed effects.
Deng et al. (2013) where the number of strengths and concerns are normalized across We include industry fixed effects because prior research indicates
each of the six CSR dimensions. that CSR is related to industry (e.g., Di Giuli and Kostovetsky, 2014)
412 S.F. Cahan et al. / Journal of Banking & Finance 59 (2015) 409–422

and press coverage is related to industry (e.g., Kothari et al., 2009).9 Table 1
Appendix B contains detailed definitions of all the variables. Descriptive statistics.

Variable Mean Median Q1 Q3 Std. dev.

3.2. Descriptive statistics and correlations Media 0.390 0.461 0.142 0.708 0.412
CSR %0.280 %1.000 %2.000 1.000 2.234
NStor 36.830 27.000 18.000 42.000 40.296
Table 1 summarizes the distribution of relevant variables for
ROA 0.073 0.091 0.037 0.146 0.152
the 2003–2011 period. We winsorize all the continuous variables BP 0.481 0.407 0.248 0.633 0.354
at 1% and 99%. Table 1 reports the descriptive statistics for vari- MV 7.099 6.888 6.001 7.999 1.525
ables in Eq. (1). The mean and median for Media are positive, indi- Lev 0.190 0.154 0.003 0.301 0.194
SP500 0.199 0.000 0.000 0.000 0.399
cating that the news media is generally positive towards our
IRisk 0.024 0.022 0.016 0.029 0.011
sample firms. The mean for CSR is %0.280 which indicates that con- Ret 0.072 %0.001 %0.194 0.242 0.438
cerns outweigh strengths on average, which is consistent with AdExp 0.012 0.000 0.000 0.009 0.028
prior studies (e.g., Kim et al., 2012; Deng et al., 2013). All the other CGov %0.176 0.000 %1.000 0.000 0.718
control variables are consistent with prior studies (e.g., Gurun and FI-News 0.019 0.000 %0.200 0.250 0.353
MktPE 23.193 17.715 16.198 24.153 23.103
Butler, 2012).
Table 2 reports the Pearson correlation matrix for the main vari- This table reports summary descriptive statistics for all the variables used to esti-
ables. The correlations between media favorability (Media) and CSR mate Eq. (1). See Appendix B for variable definitions. Sample size is 12,749. All
continuous variables are winsorized at 1% and 99%.
performance (CSR) are significantly positive as predicted. All of the
control variables are significantly correlated with Media. The high- idiosyncratic risk, firms with more favorable firm-initiated news,
est correlations are, not surprisingly, between MV and SP500. and firms with more advertising expense are treated more favor-
However, in our models, we are not interested in interpreting ably by the media. The latter finding is consistent with Gurun
either of these coefficients directly. and Butler (2012), and the second to last finding is consistent with
recent research on corporate relations (e.g., Solomon and Soltes,
3.3. Results for the effect of CSR performance on media favorability 2011; Solomon, 2012). Finally, NStor is positively and significantly
related to media favorability. Thus, the results in Table 3 indicate
Table 3 reports the multivariate regression result from estimat- that CSR is positively and significantly related to Media after con-
ing Eq. (1). Columns (1) and (2) report the main results using the trolling for the level of news coverage, i.e., number of stories.
CSR rating from the current year which is typically issued in
June. The result shows that the coefficient for CSR is significant 3.4. Endogeneity
and positive at 1% level, which indicates that firms with high CSR
performance receive more favorable news reporting. In terms of We acknowledge that endogeneity, particularly omitted vari-
economic magnitude, a shift from one standard deviation below ables and reverse causality, is an issue in our setting. We conduct
the mean of CSR to one standard deviation above the mean of three additional analyses to address these concerns.
CSR is associated with media favorability that is 8% more positive
based on the mean of Media (i.e., (0.007 & (1.954 % (%2.514))/ 3.4.1. Instrumental variables
0.390). This suggests that firms can use a strong CSR performance First, to explicitly address omitted variables as well as possible
to enhance the favorability of their media coverage, although we reverse causality (where firms’ CSR is driven by media favorabil-
acknowledge that this test does not address causality, a point that ity), we perform a 2SLS regression analysis using the ideological
we return to later. leaning of the state in which the firm is headquartered as instru-
As discussed in Section 2.2.1, for most firms, there will be no mental variables for the CSR scores. Di Giuli and Kostovetsky
overlap between CSR and Media in a temporal sense. However, (2014) find that firms which are headquartered in Democratic
for firms that have a fiscal year-end after the KLD release date rather than Republican-leaning states score higher on CSR activi-
and before December 31, there could be some overlap. For exam- ties. While the geographical location plays a pivotal role in firms’
ple, for a firm with a September 30 year-end, for 2011, Media is CSR operations, it is unlikely that location has a significant effect
based on articles October 1, 2010-September 30, 2011 while CSR on how the news media as a whole (i.e., including national and
would be based on CSR data from January 1, 2010-December 31, international media) treats firms, which satisfies the exclusion
2010. This could lead to a mechanical relation between CSR and condition of instrumental variables.11 Specifically, we use two
Media. We believe this possibility is remote since (i) CSR-related instrumental variables. The first one is Blue, which is a dummy vari-
news items are a fraction of the news items we consider, (ii) able equal to 1 if a firm’s headquarter is located in a state that is clas-
CSR-related news items are only one component of the KLD rating, sified as a blue state. The second one is Voting, which is a continuous
and (iii) for most firms, there is no overlap between CSR and variable measuring the average margin of victory/defeat in the five
Media.10 Still, we re-estimate Eq. (1) using the CSR rating from the presidential elections between 1992 and 2008 for the Democratic
prior year to ensure there will be no overlap between CSR and presidential candidate in the state where firm i has its headquarters.
Media for any firm. The disadvantage of this approach is that the To provide additional support for our choice of instruments in the
CSR rating may be stale. Columns (3) and (4) provide the results 2SLS regression, we perform the following two tests: (1) a Cragg
using the lagged CSR value. The results are highly similar. In fact, and Donald (1993) instrument relevance test to confirm the rele-
the coefficient for CSR is unchanged. vance of the instrumental variables (i.e., high correlations between
Turning to our control variables, we find firms with smaller the instrumental variables and adjusted CSR) and (2) a Sargan
book-to-market ratios, more visible firms, firms with lower (1958) overidentification test to examine the exogeneity of the
instrumental variables (i.e., no significant correlations between the
9
Although we include industry fixed effects in our main analyses, we also estimate instrumental variables and the error terms in the media favorability
our regressions using an industry-adjusted CSR score and an industry-adjusted media regression).
favorability variable to explicitly recognize that both have an industry component.
Our results are unchanged.
10 11
For example, 1663 of the 12,749 observations in our sample (13%) have fiscal Gurun and Butler (2012) find that national media slant is unrelated to a firm’
year-ends between July 1 and December 30. advertising expense.
 S.F. Cahan et al. / Journal of Banking & Finance 59 (2015) 409–422 413

Table 2
Pearson correlations for variables in Eq. (1).

Variable Media CSR NStor ROA BBP MV Lev SP500 IRisk Ret AdExp CGov FI-News
Media
CSR 0.153
NStor 0.239 0.347
ROA 0.071 0.089 0.019
BP %0.087 %0.119 %0.098 %0.315
MV 0.159 0.239 0.447 0.408 %0.312
Lev %0.076 %0.029 0.001 %0.034 %0.005 0.189
SP500 0.152 0.247 0.394 0.212 %0.149 0.652 0.129
IRisk %0.125 %0.148 %0.142 %0.384 0.130 %0.619 %0.130 %0.399
Ret 0.036 %0.014 %0.025 0.162 %0.275 0.193 %0.018 0.005 %0.049
AdExp 0.124 0.112 %0.009 0.122 %0.082 0.049 %0.012 0.083 %0.017 %0.015
CGov %0.045 %0.031 %0.121 0.031 %0.027 %0.289 %0.092 %0.168 0.126 0.021 %0.022
FI-News 0.175 %0.031 %0.001 0.342 %0.131 0.189 0.003 0.024 %0.239 0.257 %0.035 %0.031
MktPE 0.035 0.022 0.031 0.062 %0.030 0.029 0.011 0.005 %0.089 %0.102 %0.004 0.098 0.049

This table reports Pearson correlations for variables in Eq. (1). See Appendix B for variable definitions. Bold text indicates significance at the 5% level.

Table 3
Results for Eq. (1) examining the effect of CSR performance on media favorability.
Table 4
Variables (1) (2) (3) (4) Results for 2SLS test for Eq. (1) Using political leaning of home state as instruments.
Coeff. Std. err. Coeff. Std. err.
Variables First stage Second stage
Intercept 0.372⁄⁄ (0.162) 0.371⁄⁄ (0.223)
Coeff. Std. Coeff. Std.
CSR 0.007⁄⁄⁄ (0.002) 0.007⁄⁄ (0.003)
err. err.
NStor 0.001⁄⁄⁄ (0.000) 0.001⁄⁄⁄ (0.000)
ROA %0.029 (0.052) %0.052 (0.059) Intercept %5.608⁄⁄⁄ (0.882) 1.091⁄⁄⁄ (0.223)
BP %0.038⁄⁄ (0.019) %0.032 (0.021) Blue 0.305⁄⁄⁄ (0.048)
MV 0.001 (0.007) 0.005 (0.008) Voting 0.282⁄⁄ (0.116)
Lev %0.037 (0.036) %0.043 (0.039) ^CSR 0.116⁄⁄⁄ (0.023)
SP500 0.071⁄⁄⁄ (0.021) 0.070⁄⁄⁄ (0.022) NStor 0.012⁄⁄⁄ (0.001) 0.000 (0.000)
IRisk %1.840⁄⁄⁄ (0.619) %1.783⁄⁄⁄ (0.691) ROA 0.449⁄⁄⁄ (0.115) %0.057⁄ (0.035)
Ret %0.001 (0.008) %0.016 (0.009) BP 0.013 (0.057) %0.034⁄⁄ (0.013)
AdExp 0.668⁄⁄⁄ (0.234) 0.704⁄⁄⁄ (0.246) MV 0.309⁄⁄⁄ (0.024) %0.032⁄⁄⁄ (0.008)
CGov %0.005 (0.007) %0.005 (0.007) Lev %0.081 (0.089) %0.012 (0.022)
FI-News 0.165⁄⁄⁄ (0.014) 0.187⁄⁄⁄ (0.016) SP500 0.784⁄⁄⁄ (0.074) %0.018 (0.023)
MktPE 0.000 (0.000) 0.000 (0.000) IRisk 0.806 (2.124) %1.962⁄⁄⁄ (0.518)
Year fixed effects Yes Yes Ret %0.194⁄⁄⁄ (0.039) 0.022⁄⁄ (0.010)
Industry fixed effects Yes Yes AdExp 6.466⁄⁄⁄ (0.708) %0.047 (0.216)
Cluster by firm Yes Yes CGov 0.513⁄⁄⁄ (0.033) %0.061⁄⁄⁄ (0.013)
Adj-R2 0.189 0.207 FI-News %0.172⁄⁄⁄ (0.046) 0.182⁄⁄⁄ (0.012)
N 12,749 10,726 MktPE 0.004⁄⁄⁄ (0.001) %0.001⁄⁄⁄ (0.000)
Year fixed effects Yes Yes
This table reports the regression results for the effect of firms’ CSR on its media Industry fixed effects Yes Yes
favorability, i.e., Eq. (1). First-stage weak instrument test 13.78
Columns (1) and (2) report the results using CSRt, and columns (3) and (4) report the (F-value)
results using CSRt%1. See Appendix B for variable definitions. Year fixed effects and Overidentification test (p-value) 0.265
industry fixed effects are included in the model. Robust standard errors clustered at Adj-R2 0.283 0.185
the firm level are used to compute t-statistics. ⁄⁄ and ⁄⁄⁄ denote significance at the N 12,749 12,749
5% and 1% levels, respectively.
This table reports 2SLS regression results for Eq. (1).
Table 4 reports the 2SLS results. As predicted, in the first stage In the first stage, CSR is the dependent variable, and Blue and Voting are instruments.
regression, both instrumental variables, Blue and Voting, have pos- In the second stage, Media is the dependent variable, and the predicted value of CSR
(^CSR) is used in place of CSR. See Appendix B for variable definitions. Year fixed
itive and significant coefficients at 1% and 5% levels, respectively. effects and industry fixed effects are included in the models. Robust standard errors
The F-value for the Cragg and Donald (1993) instrument relevance clustered at the firm level are used to compute t-statistics. ⁄, ⁄⁄, and ⁄⁄⁄ denote
test is 13.78, rejecting the null hypothesis that the instruments are significance at the 10%, 5%, and 1% levels, respectively.
weak and confirming the relevance of our instrumental variables.
In the second stage regression, when the media favorability is
the dependent variable and the predicted value for CSR is used as
the independent variable, we find the coefficient estimate on the 3.4.2. Propensity score matching
predicted value of CSR is positive and significant at the 1% level. We further use propensity score matching to mitigate the endo-
Further, the p-value for the Sargan (1958) overidentification test geneity concern following prior literature (e.g., Gao et al., 2014). To
is 0.265, suggesting that our two instrumental variables do not vio- generate the propensity score, we construct a first-stage logistic
late the overidentifying restriction.12 regression model for CSR at the firm-year level:

12 ProbðCSR1i;t ¼ 1Þ
One might argue that most national media are located in blue states. If the
national media in these states exhibit local bias favoring local firms, the exclusion ¼ logitðb0 þ b1 RDit þ b2 Lossit þ b3 Lev it þ b4 ROAit
condition may not be satisfied. To address this concern, we repeated our instrumental
variable analysis after deleting firms located in California and New York, the two most þ b5 Salesit þ b6 Assetsit þ b7 AdExpit
populous blue states. The results are nearly identical. For example, the coefficient on
þ b8 Av gGrowthit þ b9 BPit þ b10 EP it þ b11 Volit
the instrumented CSR when firms in California and New York are omitted is 0.133
compared to 0.116 reported in Table 4. Both coefficients are highly significant at the þ b12 Analyst it þ b13 IOit þ eit Þ ð2Þ
1% level.
414 S.F. Cahan et al. / Journal of Banking & Finance 59 (2015) 409–422

We refer to firms with positive CSR scores as CSR-conscious Table 5

firms and define CSR1 equal to 1 for firm-year observations that Results for Eq. (2) using a propensity score matched sample.

have a positive CSR score in year t and 0 otherwise, following Variables Coeff. Std. err.
Gao et al. (2014). We include several control variables which are Panel A: Logistic regression to CSR
found to be correlated with firms CSR performance by previous lit- Intercept %5.378⁄⁄⁄ (0.662)
erature (e.g., Dhaliwal et al., 2012; Lys et al., 2015; Di Giuli and RD 0.000⁄⁄ (0.000)
Kostovetsky, 2014; Gao et al., 2014). In particular, we control for Loss 0.088 (0.070)
Lev %0.813⁄⁄⁄ (0.199)
research and development expense (RD), loss (Loss), leverage ROA 0.689⁄⁄ (0.346)
(Lev), return on assets (ROA), sales (Sales), total assets (Assets), Sales %0.037 (0.053)
firm’s advertising expenditure (AdExp), weighted average sales Assets 0.427⁄⁄⁄ (0.058)
growth over the past five years (AvgGrowth), book-to-market ratio AdExp 1.445⁄ (0.814)
AvgGrowth %0.004⁄ (0.003)
(BP), and earnings to book ratio (EP), the variance of daily stock
BM 0.003 (0.006)
returns (Vol), number of analyst following (Analyst), and institu- EP %0.027⁄⁄ (0.013)
tional ownership (IO). Finally, we control for industry and year Vol 0.001 (0.011)
fixed effects. Analyst 0.343⁄⁄⁄ (0.054)
Table 5 presents the propensity score matching results. Panel A IO %0.050 (0.141)
Year fixed effects Yes
shows the first-stage logistic regression results. The regression is
Industry fixed effects Yes
estimated using 12,221 firm-year observations, among which Cluster by firm Yes
3897 firm-year observations are classified as CSR-conscious firms. Pseudo-R2 0.246
We find that eight out of our thirteen controlling variables are sig- N 12,221
N of CSR1 = 1 3897
nificantly related to the CSR performance and the pseudo-R2 in our
estimation is 24.6%, which is fairly high compared to Gao et al. N Mean Median
(2014). All the significant control variables are consistent with Panel B: Propensity score matching results
our expected signs. After estimating the above logistic model, we CSR-conscious firms 3654 0.487 0.583
Matched firms 3378 0.399 0.471
calculate the propensity score using the predicted probabilities
Difference 0.001 0.000
and match our treatment sample (CSR-conscious firms) to the con-
trol sample with the closest propensity score, given the distance of This table reports results of propensity score matching. Panel A reports the
regression results for the first-stage CSR determinant model, i.e., Eq. (2).
the closet match is within 0.1. This procedure results in 3374 See Appendix B for variable definitions. Year fixed effects and industry fixed effects
matched firms (i.e., about 92.4% of the CSR-conscious firm-year are included in the model. Robust standard errors clustered at the firm level are
observations are matched). The untabulated parametric t-test used to compute t-statistics. Panel B reports the mean and median of media
and non-parametric Kolmogorov–Smirnov test show that the favorability for CSR-conscious firms vs. propensity-matched non-CSR-conscious
firms. ⁄, ⁄⁄, and ⁄⁄⁄ denote significance at the 10%, 5%, and 1% levels, respectively.
mean of the propensity score in our match sample (0.375) is not
statistically different from the one in our treatment sample
(0.369). We also examine whether the independent variables used
increases after the executive order was implemented. Specifically,
in the first-stage prediction model are different between the treat-
we identify 713 firm-year observations in the treatment sample
ment sample and the control sample. Untabulated results show
during the period between 2010 and 2013 where the firm’s major
there are no significant differences.
customer is the US government. As the US government is their
After confirming our treatment sample is similar to our
major customer, they are required to follow the executive order
matched sample except for their actual CSR performances, we then
starting from year 2012. We then form a one-to-one matched sam-
compare the media favorability between the two samples. Panel B
ple by choosing the firm that had the closest market capitalization
of Table 5 shows the results. Both the mean and median of Media in
in the same 2-digit SIC industry but did not have the US govern-
the treatment group (mean = 0.487, median = 0.583) are signifi-
ment as its major customer. We estimate the following regression
cantly higher than the matched group (mean = 0.399, med-
model:
ian = 0.471) at the 1% level, consistent with our main findings.
Mediait ¼ b0 þ b1 Treatment it & Postit þ b2 Treatmentit
þ b3 Postit þ b4 NStorit þ b5 ROAit þ b6 BPit þ b7 MV it
3.4.3. Quasi-natural experiment
We also conduct a quasi-natural experiment utilizing a group of þ b8 Lev it þ b9 SP500it þ b10 IRiskit þ b11 Ret it
firms that were forced to improve their performances in the area of þ b12 AdExpit þ b13 CGov it þ b14 FI-Newsit
employee relations. On September 25, 2012, President Obama
þ b15 MktPEit þ eit ð3Þ
signed an executive order to prohibit human trafficking-related
activities that applies to all federal contractors and subcontractors. We define two dummy variables in the model. Post equals to 1 if
This executive order requires compliance measures for govern- year is 2012 or 2013 and 0 otherwise. Treatment equals to 1 for the
ment contracts and subcontracts and provides federal agencies firm-year observations whose major customer is the US govern-
with additional tools to foster compliance. We expect that the ment. The coefficient, b1, of the interaction item between these
executive order, which forced the federal contractors and subcon- two dummy variables is our interest.
tractors to mandatorily improve their performances in the Table 6 tabulates the results of the estimation of Eq. (3). We find
employee relations area, would have a positive impact on the that the interaction item is significantly positive at the 5% level,
CSR performance of firms affected by the executive order; on the indicating that the treatment firms, which were forced to improve
other hand, the executive order would have had no direct effect their employee relations after 2012, received more favorable media
on overall media favorability of these firms.13 reporting compared to the control group in the post-executive
Using a difference-in-difference research design, we test order period. It is worth mentioning that the coefficient on
whether the media image for firms affected by the executive order Treatment is found to be insignificant, meaning that there is no sig-
nificant difference between our treatment sample and matched
13
The t-test shows the mean of CSR performance for our treatment sample sample in terms of media favorability before the regulation (i.e.,
increases by 0.415 after 2012, and it is significant at the 5% level. 2010 and 2011). The coefficient on Post is also found to be
 S.F. Cahan et al. / Journal of Banking & Finance 59 (2015) 409–422 415

Table 6 (JOB), and judicial (JUDIC). Based on these codes, 51,083 of the
Results for Eq. (3) using quasi natural experiment based on presidential executive 469,550 articles in our sample (10.9%) are classified as CSR-related.
order on human-trafficking.
Second, we identify 67 category labels based on definitions used
Variables Coeff. Std. err. by KLD. These labels are more specific that the Reuters code, e.g.,
Intercept 0.345⁄⁄⁄ (0.080) ‘charity’, ‘clean energy’, ‘employee benefits’, ‘support for educa-
Treatment & Post 0.021⁄⁄ (0.010) tion’, and ‘workplace safety’. We search all articles and classify
Treatment %0.002 (0.022) an article as CSR-related if one or more of the 67 category labels
Post 0.004 (0.009)
NStor 0.000⁄⁄ (0.000)
appear in the headline of the article. Using this approach, we clas-
ROA 0.076⁄ (0.044) sify 8834 of the 469,550 articles (1.9%) as CSR-related. Thus, we
BP %0.007 (0.048) view the Reuters-based approach as providing an upper bound of
MV %0.012 (0.016) CSR-related articles while the KLD-based approach provides a
Lev 0.006 (0.067)
lower bound.
SP500 0.044⁄ (0.025)
IRisk %3.623⁄ (2.005) We re-estimate Eq. (1) after deleting the CSR-related articles.
Ret 0.030 (0.042) When the Reuters-based CSR articles are omitted, the coefficient
AdExp 0.925⁄⁄ (0.423) for CSR is 0.006 (t-stat. = 2.29, p-value = 0.02). When the
CGov 0.009 (0.025) KLD-based CSR articles are omitted, the coefficient for CSR is
FI-News 0.047 (0.049)
0.006 (t-stat. = 2.37, p-value = 0.018). To summarize, our results
MktPE 0.001⁄ (0.000)
Industry fixed effects Yes do not change if we exclude the CSR related articles from our
Cluster by firm Yes sample.
Adj-R2 0.275 A related concern is that if firms have a greater propensity to
N 1426
disclose positive CSR performance than negative CSR performance,
This table reports the regression results for the effect of executive order signed by the media’s coverage could be overly optimistic. However, we do
President Obama in Sep 2012 on the affected firms’ media favorability, i.e., Eq. (3). not expect this to bias our tests in favor of finding a result for
See Appendix B for variable definitions. Industry fixed effects are included in the
two reasons. First, the firm’s own disclosures are not the media’s
model. Robust standard errors clustered at the firm level are used to compute t-
statistics. ⁄, ⁄⁄, and ⁄⁄⁄ denote significance at the 10%, 5%, and 1% level, respectively.
only source of information about CSR performance. For example,
the media can access reports and data from governmental and
industry sources, talk to activist groups and other engaged stake-
holders (e.g., community leaders, union officials, employees, char-
insignificant, indicating that the control firms’ media favorability ities), and observe actual activity (e.g., an oil spill and the firm’s
does not change after 2012. clean-up efforts). Second, if the media’s coverage is asymmetrically
Together, the results in Tables 4–6 address concerns about positive, it would reduce the correlation between actual CSR per-
endogeneity and provide support for a causal relationship where formance and media coverage, biasing the tests against finding a
CSR performance influences a firm’s media favorability. significant result. One implication from the asymmetrical repor-
tage of positive and negative events is that firms with the highest
CSR performance should have greater media favorability.
3.5. Alternative measures of media favorability Consequently, we divide our sample into quartiles based on CSR
performance and run results separately for subsets of firms in
As our objective is to examine the effect of CSR on the firm’s the top and bottom quartiles.17 We find that the positive relation
overall media image, we use all articles about the firm to compute between Media and CSR is concentrated among the firms with the
Media. This includes CSR and non-CSR related articles. Based on highest values for CSR performance as CSR is positive and significant
prior research which examines only CSR-related news, we expect for the top quartile but not the bottom quartile. This provides further
the number of CSR-related articles to be small relative to the support for the notion that firms actively manage CSR since it is pos-
non-CSR related articles.14 However, even if the number of itive events that they are most likely to actively manage.
CSR-related news items is small, it is possible that they may be driv-
ing media favorability. For example, if all non-CSR news is neutral 4. Part 2: Evidence on active media management
and CSR-related articles are positive, overall media favorability
would be positive, although in terms of the magnitude the effect As documented in the previous section, our evidence supports a
of the CSR-related articles would be small since Media is scaled by causal relation between firms’ CSR performance and media image.
the total number of articles in a year for that firm.15 However, it is possible that firms do not actively manage their CSR
To address this concern, we compute two alternative measures to improve their media image; instead, they may invest in CSR for
that exclude CSR-related articles. However, identifying other reasons (e.g., altruism), and favorable media coverage may be
‘‘CSR-related’’ articles is subjective. We use two definitions. First, an attractive side-benefit. We next consider whether managers
Reuters provides a topic code for every article. We identify eight proactively engage in media management. We provide three tests
topic codes16 that broadly relate to CSR issues: crime (Reuters code for different settings where firms have different incentives to man-
CRIM), disasters/accidents (DIS), environment/nature (ENV), geneti- age media image.
cally modified (GMO), health/medicines (HEA), labor/employment
4.1. Sin vs. non-sin industries
14
For example, Kruger (2014) examines CSR-related news events over a seven-year
period. He identifies 2116 such events. By contrast, our measure of media favorability
First, we consider ‘sin’ industries (i.e., alcohol, gambling, and
is based on 469,550 news articles over a nine-year period.
15
We thank an anonymous reviewer for raising this issue. tobacco). Hong and Kacperczyk (2009) argue firms in sin industries
16
There are well over 1000 topic codes in TRNA database and a news story usually are viewed negatively by the public because they can be addictive
has multiple topic codes assigned to it. Therefore, if an article has one of the and have undesirable social consequences when consumed exces-
CSR-related news codes but one or more non-CSR news codes, the article may actually sively. For example, in the US, the tobacco industry has had to
focus on the non-CSR topic(s) and the CSR portion could be relatively small. Thus, our
classification approach is extremely conservative as we classify an article as
17
CSR-related even if CSR is the focus of only a small part of the article. We thank an anonymous reviewer for suggesting this analysis.
416 S.F. Cahan et al. / Journal of Banking & Finance 59 (2015) 409–422

include health warnings on cigarette packages since 1965, and 4.2. Impact of investor sentiment
broadcast advertising of tobacco products has been banned since
1971. Hong and Kacperczyk (2009) find that a broad set of institu- We examine whether our findings vary with the level of
tions that are constrained by social norms (e.g., pensions, universi- market-wide investor sentiment. Investor sentiment is attractive
ties) invest less in sin industries. because it is exogenous to the firm and because it can affect inves-
Given a negative public image, we expect that firms in sin tor behavior. For example, Baker and Stein (2004) find that investor
industries would have more incentives to engage in CSR to improve sentiment affects trading activity, while Shiller (2000) and
their media image. We follow Hong and Kacperczyk (2009) classi- Nofsinger (2005) find investor sentiment affects expectations
fication scheme to identify firms in sin industries. Sin firms include about future firm performance. Mian and Sankaraguruswamy
the following: (1) firms with SIC codes 2100–2199 which are beer (2012) provide evidence that investor sentiment is associated with
and liquor producers, (2) firms with SIC codes 2080–2085 which the stock market’s response to unexpected earnings around the
are tobacco firms, and (3) firms with NAICS codes 7132, 71312, earnings announcement date, leading investors to overreact to
713210, 71329, 713290, 72112, and 721120 which are gambling bad earnings news during low sentiment periods. Not surprisingly,
firms. We create an indicator variable, Sin, that is equal to 1 for evidence suggests that firms take action to counteract investors’
firms that are members of a sin industry, and 0 otherwise, and pessimistic outlook. For example, Bergman and Roychowdhury
we estimate the following model: (2008) find that walk-up forecasts are more frequent during peri-
ods of low investor sentiment, consistent with managers correcting
investors’ overly low expectations. Thus, we expect that all firms
Mediait ¼ b0 þ b1 CSRit & Sinit þ b2 CSRit þ b3 Sinit þ b4 NStorit would have more incentives to increase media favorability when
þ b5 ROAit þ b6 BPit þ b7 MV it þ b8 Lev it þ b9 SP500it the prevailing investor sentiment is more pessimistic.
We use two market sentiment proxies. Following Conrad et al.
þ b10 IRiskit þ b11 Ret it þ b12 AdExpit þ b13 CGov it
(2002), MktPE is the value-weighted average of market price-
þ b14 FI-Newsit þ b15 MktPEit þ eit ð4Þ to-earnings ratio of all the firms in the merged CRSP-Compustat
In Eq. (4), the interaction, CSR & Sin, represents the incremental database in year t (MktPE is used as a control variable in our earlier
effect in CSR-Media relation for sin firms relative to non-sin firms. tests). BW is the average monthly Baker-Wurgler Sentiment Index
We expect that b1 will be positive and significant if firms in sin (Baker and Wurgler, 2006).18 We interact both measures with CSR
industries are more likely to engage in CSR in order to manage and rerun our main regression.
their media image. Table 8 provides the estimation results. When we use MktPE as
Table 7 contains the results for this analysis. CSR remains signif- a proxy for market sentiment, we find that the interaction item
icant with a coefficient of 0.007. More importantly, the interaction CSR & Sentiment is significantly negative at the 5% level. When
variable, CSR & Sin, has a positive and significant coefficient of we use BW as a proxy for market sentiment, we also find that
0.023. This indicates that the impact of a unit increase in CSR is the interaction item CSR & Sentiment is significantly negative at
4.29 times (0.030/0.007) higher if the firm is in a sin industry, con- 5%.19 These two results mean that during periods when the
sistent with these firms undertaking activities that will improve market-level investor sentiment is less favorable, firms’ CSR perfor-
their media image. Of course, this does not mean sin firms have mance is more correlated with media favorability. For example, for a
better media images. The coefficient of Sin is %0.130 and is signif- firm at the 75th percentile of CSR (a firm with superior CSR perfor-
icant, indicating that sin firms have much less favorable media mance), a move from one standard deviation below the mean of
images on average. BW to one standard deviation above the mean of BW decreases the
effect of CSR performance on media favorability by 2.37 times com-
Table 7 pared to when sentiment is neutral. This supports our hypothesis
Results for Eq. (4) examining the effect of CSR performance and sin industry that when market-level investor sentiment is pessimistic, firms have
membership on media favorability.
more incentives to perform well in the CSR area to improve their
Variables Coeff. Std. err. media image.
Intercept 0.379⁄⁄ (0.165)
CSR & Sin 0.023⁄⁄ (0.011)
4.3. Change around SEO offering
CSR 0.007⁄⁄⁄ (0.003)
Sin %0.130⁄⁄ (0.059)
NStor 0.001⁄⁄⁄ (0.000) Next, we conjecture that one important incentive for firms to
ROA %0.028 (0.052) use media management is to lower their financing costs. Prior
BP %0.040⁄⁄ (0.020) studies provide evidence that firms use various mechanisms to
MV 0.001 (0.007)
Lev %0.037 (0.036)
lower their financing cost around SEOs (e.g., Teoh et al., 1998;
SP500 0.071⁄⁄⁄ (0.021) Lang and Lundholm, 2000; Kim and Park, 2005; Shroff et al.,
IRisk %1.804⁄⁄⁄ (0.615) 2013), and media-related studies show that the tone of news sto-
Ret %0.002 (0.008) ries can affect firms’ cost of capital (e.g., Kothari et al., 2009). As
AdExp 0.712⁄⁄⁄ (0.236)
such, we expect that firms with poor media images prior to an
CGov %0.006 (0.008)
FI-News 0.165⁄⁄⁄ (0.014) SEO would have incentives to engage in active media management
MktPE 0.000 (0.000) in order to maximize the proceeds they obtain from the offering.
Year fixed effects Yes We use all SEO announcements between 2005 and 2011.
Industry fixed effects Yes Because Media has an upper bound, firms that already have high
Cluster by firm Yes
Adj-R2 0.194
media favorability have less ability (and incentives) to further
N 12,749 improve their media image. Accordingly, we divide the sample
based on the median media favorability two years before the SEO
This table report the regression results for the sin industry effect on the association
between firms’ CSR and its media favorability, i.e. Eq. (4).
18
See Appendix B for variable definitions. Year fixed effects and industry fixed effects We thank Jeffrey Wurgler for making the data available on his website,
are included in the model. Robust standard errors clustered at the firm level are http://pages.stern.nyu.edu/ ~jwurgler/.
19
used to compute t-statistics. ⁄, ⁄⁄, and ⁄⁄⁄ denote significance at the 10, 5, and 1% The sample size for the market-wide investor sentiment measure developed by
levels, respectively. Baker and Wurgler is smaller as their measure ends in 2010.
 S.F. Cahan et al. / Journal of Banking & Finance 59 (2015) 409–422 417

Table 8 Table 9
Results for regressions examining the effect of CSR performance and investor Results for Eq. (5) examining CSR performance prior to seasoned equity offerings.
sentiment on media favorability.
Variables Coeff. Std. err.
Variables Sentiment = MktPE Sentiment = BW
Intercept %1.884 (1.093)
Coeff. Std. err. Coeff. Std. err. Post1 0.200⁄⁄⁄ (0.073)
RD 0.010⁄ (0.006)
Intercept 0.371⁄⁄ (0.163) 0.378⁄⁄ (0.149)
Loss 0.341 (0.355)
CSR & Sentiment %0.000⁄⁄ (0.000) %0.012⁄⁄ (0.005)
Lev 0.134 (0.605)
CSR 0.009⁄⁄⁄ (0.003) 0.006⁄⁄ (0.002)
ROA %0.456 (0.848)
Sentiment 0.000 (0.000) %0.016 (0.081)
Sales 0.219⁄ (0.132)
NStor 0.001⁄⁄⁄ (0.000) 0.001⁄⁄⁄ (0.000)
Assets 0.165 (0.157)
ROA %0.030 (0.052) %0.048 (0.055)
AdExp 9.574⁄ (5.796)
BP %0.038⁄ (0.019) %0.054⁄⁄ (0.021)
AvgGrowth 0.166 (0.265)
MV 0.001 (0.007) %0.005 (0.011)
BM %0.736⁄⁄ (0.283)
Lev %0.037 (0.037) %0.039 (0.039)
EP %1.389⁄⁄⁄ (0.423)
SP500 0.071⁄⁄⁄ (0.021) 0.070⁄⁄⁄ (0.021)
Vol %0.046 (0.394)
IRisk %1.831⁄⁄⁄ (0.619) %2.000⁄⁄⁄ (0.634)
Analyst 0.708⁄⁄⁄ (0.221)
Ret %0.002 (0.008) 0.002 (0.009)
IO %0.327 (0.471)
AdExp 0.668⁄⁄⁄ (0.234) 0.697⁄⁄⁄ (0.246)
Year fixed effects Yes
CGov %0.005 (0.007) %0.005 (0.008)
Industry fixed effects Yes
FI-News 0.165⁄⁄⁄ (0.014) 0.166⁄⁄⁄ (0.014)
Cluster by firm Yes
Year fixed effects Yes Yes
Adj-R2 0.321
Industry fixed effects Yes Yes
N 1100
Cluster by firm Yes Yes
Adj-R2 0.190 0.280 This table reports the regression results for the CSR performance leading up to an
N 12,749 11,280 SEO, i.e., Eq. (5).
See Appendix B for variable definitions. Year fixed effects and industry fixed effects
This table report the regression results for the effect of market sentiment on the
are included in the models. Robust standard errors clustered at the firm level are
association between firms’ CSR and its media favorability. See Appendix B for
used to compute t-statistics. ⁄, ⁄⁄, and ⁄⁄⁄ denote significance at the 10%, 5%, and 1%
variable definitions. Year fixed effects and industry fixed effects are included in the
percent level, respectively.
model. Robust standard errors clustered at the firm level are used to compute t-
statistics. ⁄, ⁄⁄, and ⁄⁄⁄ denote significance at the 10, 5, and 1% levels, respectively.

calculate four measures of firms’ implied cost of capital based on

announcement (year T-2) and conduct our tests using firms in the the prior literature. The first measure (COCOJ) is developed by
lower half. The final sample consists of 1100 observations. Since Ohlson and Juettner-Nauroth (2005). Our second measure is devel-
more positive media reports can lower financing costs (e.g., oped by Easton (2004). Our third measure is from Gebhardt et al.
Kothari et al., 2009), we expect that firms will use CSR to improve (2001), and our fourth measure is based on Claus and Thomas
their media image leading up to SEO. (2001). The control variables are from Dhaliwal et al. (2005),
To examine this possibility, we estimate the following regres- Naiker et al. (2013), Gurun and Butler (2012).
sion model using the observations from year T-2 to year T: Following Dhaliwal et al. (2011), we focus on the economic ben-
CSRit ¼ b0 þ b1 Post1it þ b2 RDit þ b3 Lossit þ b4 Lev it þ b5 ROAit efits accruing to firms that had favorable media coverage and high
CSR performance. In other words, we consider whether the eco-
þ b6 Salesit þ b7 Assetsit þ b8 AdExpit þ b9 Av gGrowthit
nomic benefits are greater for firms that have a good media image
þ b10 BPit þ b11 EP it þ b12 Volit þ b13 Analyst it þ b14 IOit þ eit and that image is more credible (i.e., supported by superior CSR
ð5Þ performance). Specifically, we estimate:
We define a dummy variable Post1 equal to 1 if the firm-year TobinQ it ¼ b0 þ b1 CSRit & Mediait & H-Hit þ b2 CSRit & Mediait
observation falls into the period between T-1 and T year, and 0 if
þ b3 Mediait & H-Hit þ b4 CSRit & H-Hit þ b5 H-Hit
it belongs to T-2 year. We expect the coefficient of Post1 to be sig-
nificantly positive. The control variables are identical to Eq. (2). þ b6 Mediait þ b7 CSRit þ b8 FI-Newsit þ b9 CGov it
Table 9 presents the estimation results. The coefficient of Post1 þ b10 RD1it þ b11 CAPX it þ b12 IOit þ b13 SGrowthit
variable is significantly positive at 1% level, consistent with these
firms increasing their CSR performance prior to the SEO to lower þ b14 ROAit þ b15 MV it þ b16 Lev it þ b17 Analystsit
their financing costs. þ b18 IRiskit þ b19 Ret it þ b20 AdExpit þ eit ð6Þ
In sum, the three tests in Section 4 provide support for active
media management. or

COC it ¼ b0 þ b1 CSRit & Mediait & H-Hit þ b2 CSRit & Mediait

5. Part 3: Economic benefits of media management
þ b3 Mediait & H-Hit þ b4 CSRit & H-Hit þ b5 H-Hit
Our tests assume that firms would have an economic motiva- þ b6 Mediait þ b7 CSRit þ b8 FI-Newsit þ b9 CGov it
tion to manage their media image. That is, firms that make a pos- þ b10 I-COC it þ b11 BPit þ b12 Disperseit þ b13 MV it þ b14 LTGit
itive effort in the CSR arena and that receive favorable media þ b15 Beta % SMBit þ b16 Beta % Mktit þ b17 Beta-HMLit þ eit
coverage realize economic benefits. Consequently, to provide evi- ð7Þ
dence on this underlying assumption, we examine the effect of
the interaction between CSR performance and media favorability where H-H equals 1 if firm i receives more favorable media report-
on the firm’s value and cost of capital. ing and has strong CSR performance in year t, and 0 otherwise. We
define high media favorability as firm-year observations where
5.1. Models and variables Media exceeds the sample median and the high CSR performance
as firm-year observations where CSR is above the sample median.
To measure firm value, we calculate TobinQi,t as the log of Thus, the interactions with H-H allow us to estimate separate coef-
market-to-book ratio for firm i at the end of year t. We also ficients for the high favorability/high CSR performing firms. All
418 S.F. Cahan et al. / Journal of Banking & Finance 59 (2015) 409–422

other variables are defined above. We expect b1 to be positive (neg- performance is rewarded only if it is supported by favorable media
ative) and significant if firms with good CSR and favorable media coverage, which is consistent with favorable media coverage
coverage realize a higher firm valuation (lower cost of capital). enhancing the overall reputation of good CSR firms.
We assess the incremental economic effect of CSR & Media &
5.2. Results for the economic benefits of media management H-H by considering how a shift from below the mean of Media to
one standard deviation above the mean of Media affects firm value
Table 10 reports the regression results from estimating Eq. (6) at the 75th percentile of CSR (i.e., a firm with superior CSR perfor-
that considers the joint effect of media favorability and firms’ mance). We find such a shift is associated with an increase in
CSR performance on firm value. Before estimating Eq. (6), we esti- Tobin’s Q of 6.79% (7.15%) based on the mean (median) of
mate two reduced models. In the first reduced model, we regress Tobin’s Q. Thus, the incremental economic effect of CSR & Media &
TobinQ on CSR and the control variables, i.e., we exclude Media H-H on Tobin’s Q is economically significant.
and H-H and the interactions between those variables and CSR. Table 11 reports the results for Eq. (7) which uses cost of capital
We use this as a baseline model because an extensive body of as the dependent variable. We only tabulate the results for the
research examines the effect of CSR on firm attributes, such as firm measure COCOJ as we obtain similar results for the other three cost
value, with mixed results (e.g., see Margolis et al., 2009, for a of capital measures. Similar to Table 10, we estimate two reduced
review). In the second reduced model, we add Media and models before estimating the full model. In columns (1) and (2), we
CSR & Media to the model, but exclude H-H and the interactions find no evidence that CSR affects cost of capital. In columns (3) and
with H-H. In this way, we are able to link Eq. (6) to the prior (4), when we include CSR, Media, and the interaction CSR & Media,
research and can assess whether it is likely that the omission of we find that their coefficients are not significant. On the other
media-related variables impacts the results of those earlier studies. hand, in columns (5) and (6), when Eq. (7) is estimated, the coeffi-
In columns (1) and (2) which reports the results for the first cient for CSR & Media & H-H is negative and significant (coeffi-
reduced model, we find an insignificant coefficient for CSR, indicat- cient = %0.011, p-value < 0.05), indicating that cost of capital is
ing that better CSR performance does not increase firm value. lower for firms that have both better CSR performance and higher
Further, in the second reduced model, shown in columns (3) and media favorability scores. In terms of economic significance, for a
(4), we find insignificant coefficients for CSR, Media, and the inter- firm at the 75th percentile of CSR, a shift from one standard devi-
action CSR & Media, indicating that CSR and media favorability do ation below the mean of Media to one standard deviation above
not individually or jointly affect firm value. However, when Eq. the mean of Media is associated with an average decrease in cost
(6) is estimated, columns (5) and (6) show that the coefficient for of capital of 97.5 basis points across our four cost of capital mea-
the three-way interaction, CSR & Media & H-H, is positive and sta- sures. Based on the mean (median) of the four measures, this is
tistically significant (coefficient = 0.086, p-value < 0.05), indicating equivalent to a 9.39% (10.59%) reduction in cost of capital, which
that, for firms with superior CSR performance and superior media is economically significant.
favorability (i.e., when H-H = 1), better CSR performance and more To summarize, Tables 10 and 11 provide consistent results that
positive media favorability are jointly associated with a higher firm CSR performance and media favorability can jointly lower a firm’s
value. In other words, superior CSR performance by itself is insuf- cost of capital and increase its firm value when its CSR perfor-
ficient to generate a lower cost of capital. Instead, superior CSR mance is relatively good and its media image is relatively

Table 10
Results for Eq. (6) Examining the Effect of Media Favorability and CSR Performance on Tobin’s Q.

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

Coeff. Std. err. Coeff. Std. err. Coeff. Std. err.
Intercept %0.705⁄⁄⁄ (0.187) %0.716⁄⁄⁄ (0.187) %0.721⁄⁄⁄ (0.194)
CSR & Media & H-H 0.086⁄⁄ (0.041)
CSR & Media 0.008 (0.014) 0.045 (0.039)
Media & H-H %0.032 (0.119)
CSR & H-H 0.077⁄⁄ (0.033)
H-H 0.025 (0.079)
Media 0.001 (0.030) %0.013 (0.029)
CSR %0.007 (0.006) %0.011 (0.010) 0.005 (0.009)
FI-News 0.057⁄⁄ (0.023) 0.054⁄⁄ (0.024) 0.054⁄⁄ (0.023)
CGov 0.052⁄⁄⁄ (0.015) 0.051⁄⁄⁄ (0.015) 0.053⁄⁄⁄ (0.015)
RD1 0.097⁄⁄⁄ (0.016) 0.113⁄⁄⁄ (0.018) 0.095⁄⁄⁄ (0.016)
CapX %0.255⁄⁄ (0.113) %0.267⁄ (0.154) %0.254⁄⁄ (0.113)
IO 0.207⁄⁄⁄ (0.063) 0.198⁄⁄⁄ (0.063) 0.212⁄⁄⁄ (0.063)
SGrowth 0.263⁄⁄⁄ (0.029) 0.249⁄⁄⁄ (0.027) 0.260⁄⁄⁄ (0.028)
ROA 0.375⁄⁄ (0.151) 0.457⁄⁄⁄ (0.157) 0.377⁄⁄ (0.151)
MV %0.112⁄⁄⁄ (0.015) %0.122⁄⁄⁄ (0.021) %0.122⁄⁄⁄ (0.021)
Lev 0.347⁄⁄⁄ (0.086) 0.332⁄⁄⁄ (0.086) 0.351⁄⁄⁄ (0.085)
Analysts 0.102⁄⁄⁄ (0.026) 0.102⁄⁄⁄ (0.026) 0.099⁄⁄⁄ (0.025)
IRisk 3.765⁄⁄⁄ (1.381) 4.122⁄⁄⁄ (1.394) 3.863⁄⁄⁄ (1.375)
Ret 0.401⁄⁄⁄ (0.017) 0.401⁄⁄⁄ (0.017) 0.399⁄⁄⁄ (0.017)
AdExp 0.049⁄⁄⁄ (0.018) 0.053⁄⁄⁄ (0.016) 0.053⁄⁄⁄ (0.015)
Year fixed effects Yes Yes Yes
Industry fixed effects Yes Yes Yes
Cluster by firm Yes Yes Yes
Adj-R2 0.367 0.367 0.369
N 12,332 12,332 12,332

This table reports the regression results for the joint effect of superior CSR performance and high media favorability on firm value, i.e., Eq. (6).
Results for the full model are reported in columns (5) and (6). Columns (1)–(4) report results for two reduced models that exclude the media-related variables. See Appendix B
for variable definitions. Year fixed effects and industry fixed effects are included in the models. Robust standard errors clustered at the firm level are used to compute t-
statistics. ⁄, ⁄⁄, and ⁄⁄⁄ denote significance at the 10%, 5%, and 1% levels, respectively.
 S.F. Cahan et al. / Journal of Banking & Finance 59 (2015) 409–422 419

Table 11
Results for Eq. (7) examining the effect of media favorability and CSR performance on cost of capital.

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

Coeff. Std. err. Coeff. Std. err. Coeff. Std. err.
Intercept 0.046⁄⁄⁄ (0.006) 0.057⁄⁄⁄ (0.001) 0.049⁄⁄⁄ (0.006)
CSR & Media & H-H %0.011⁄⁄ (0.005)
CSR & Media %0.000 (0.000) 0.000 (0.000)
Media & H-H 0.021 (0.023)
CSR & H-H 0.006 (0.004)
H-H %0.022 (0.021)
Media %0.001 (0.001) %0.001 (0.001)
CSR %0.000 (0.000) %0.011 (0.010) 0.005 (0.009)
FI-News %0.007⁄⁄⁄ (0.001) %0.006 (0.001) %0.005⁄⁄⁄ (0.001)
CGov 0.000⁄⁄⁄ (0.000) 0.000 (0.000) 0.001 (0.000)
I-COCOJ 0.620⁄⁄⁄ (0.041) 0.621⁄⁄⁄ (0.041) 0.095⁄⁄⁄ (0.016)
BP 0.008⁄⁄⁄ (0.014) 0.007⁄⁄⁄ (0.001) 0.009⁄⁄ (0.002)
Disperse 0.176⁄⁄⁄ (0.016) 0.194⁄⁄⁄ (0.017) 0.173⁄⁄⁄ (0.016)
MV %0.002⁄⁄⁄ (0.000) %0.003⁄⁄⁄ (0.000) %0.002⁄⁄⁄ (0.000)
LTG 0.050⁄⁄⁄ (0.002) 0.051⁄⁄⁄ (0.002) 0.049⁄⁄⁄ (0.002)
Beta-SMB 0.005⁄⁄⁄ (0.001) 0.005⁄⁄⁄ (0.001) 0.005⁄⁄⁄ (0.001)
Beta-Mkt 0.004⁄⁄⁄ (0.001) 0.005⁄⁄⁄ (0.001) 0.005⁄⁄⁄ (0.001)
Beta-HML 0.003⁄⁄⁄ (0.001) 0.003⁄⁄⁄ (0.001) 0.003⁄⁄⁄ (0.001)
Year fixed effects Yes Yes Yes
Industry fixed effects Yes Yes Yes
Cluster by firm Yes Yes Yes
Adj-R2 0.446 0.450 0.452
N 9983 9983 9983

This table reports the regression results for the joint effect of superior CSR performance and high media favorability on cost of capital, i.e., Eq. (7).
Results for the full model are reported in columns (5) and (6). Columns (1)–(4) report results for two reduced models that exclude the media-related variables. Cost of capital
is estimated following Ohlson and Juettner-Nauroth (2005). See Appendix B for variable definitions. Year fixed effects and industry fixed effects are included in the models.
Robust standard errors clustered at the firm level are used to compute t-statistics. ⁄⁄ and ⁄⁄⁄ denote significance at the 5% and 1% levels, respectively.

favorable. Said differently, a firm with good CSR performance can- example, the effect of CSR performance on media favorability is
not realize economic benefits unless its overall media image is pos- 4.29 times greater for sin firms compared to non-sin firms, sug-
itive. This suggests that good CSR firms rely on the media in gesting that sin firms have more incentives to manage their CSR
building its reputation, consistent with Fombrun and Shanley’s performance to improve their media image. Similarly, for a firm
(1990) reputation-building model. Indeed, our results suggest that at the 75th percentile of CSR performance, a move from one stan-
a superior media image may be a missing link in prior studies that dard deviation below to one standard deviation above the mean of
find insignificant or mixed results for the effect of CSR performance investor sentiment decreases the effect of CSR performance on
on firm attributes such as firm value. media favorability by 2.37 times compared to when investor senti-
ment is neutral. Further, we document a significant increase in CSR
6. Conclusion performance in the two-year period prior to a SEO for a sample of
low media favorability firms, suggesting that these firms use CSR
We examine whether firms manage the favorability of their performance to improve their media favorability and lower their
news coverage through better CSR performance. We use data from financing costs.
TRNA where news items are rated based on their positive or nega- In part 3 of the study, we show that firms with good CSR perfor-
tive tone and data on CSR performance from KLD. mance only realize a higher firm value or lower cost of capital if
In part 1 of the study, we find that firms that perform better in they also have favorable media coverage, consistent with these
CSR areas are viewed more favorably in the media. In terms of the firms benefiting from the positive tone in the media’s general
economic consequences, a shift from one standard deviation below reportage about the firm. We show that for a firm at the 75th per-
to one standard deviation above the mean of CSR performances centile of CSR performance, a shift from one standard deviation
increases media favorability by 8%. These results are robust after below to one standard deviation above the mean of media favora-
controlling for the endogeneity issues and eliminating any tempo- bility increases Tobin’s Q by 6.79% and decreases cost of capital by
ral overlap between our media favorability and CSR measures. 9.39%. These findings suggest that the media can play an active role
Results from a quasi-natural experiment support a causal relation- in the reputation building process. Further, our analyses that link
ship from CSR performance to media favorability. Thus, we provide our models that incorporate media favorability to reduced models
support for Mullainathan and Schleifer’s (2005) catering theory of that exclude media favorability suggest that a firm’s overall media
the media where slants its news coverage to coincide with the image is a missing link in prior studies that examine the effect of
beliefs and preferences of its readers. In our setting, the public’s CSR performance on firm attributes such as firm value.
preference for socially responsible business practices provides Our study contributes to the literature in two important ways.
the media with incentives to provide more favorable coverage of First, we extend the growing literature on the business press.
good performing CSR firms. Our empirical results support El While prior research on the business press generally treats news
Ghoul et al.’s (2011, 2390) conjecture that the media spends more as exogenous, we examine whether firms can actively influence
time ‘‘analyzing and reporting news about ‘good’ CSR firms’’. their own media coverage. Our evidence suggests that firms can
In part 2 of the study, consistent with the notion that managers manage their media image through their CSR performance, a chan-
actively manage their CSR activity to affect their media image, we nel that is more subtle and indirect than the direct channels docu-
find our results are more pronounced in firms in sin industries and mented by Gurun and Butler (2012), Solomon (2012), and Ahern
when market-wide investor sentiment is more negative. For and Sosyura (2014). Second, we contribute to an expansive
420 S.F. Cahan et al. / Journal of Banking & Finance 59 (2015) 409–422

literature on CSR performance. Although this literature is at least (low) to 1 (high). A news item that mentions several firms may
40 years old, academic interest in CSR has not abated. For example, not be equally relevant for all the firms if the item focuses on just
recent studies examine the relation between CSR and cost of capi- one of those firms. Consistent with Sinha (2011), we use the rele-
tal (El Ghoul et al., 2011), cost of bank loans (Goss and Roberts, vance score to filter news items and exclude news items that have
2011), earnings management (Kim et al., 2012), analyst forecast a relevance score of less than 1.
errors (Becchetti et al., 2013), access to finance (Cheng et al.,
2013), stock price crash risk (Kim et al., 2014b), and local political Appendix B
orientation (Di Giuli and Kostovetsky, 2014). We add to this liter-
ature by documenting a previously unexplored incentive for firms Variable definitions
engaging in CSR, i.e., active media management. Overall, our
results suggest that the media slants their reporting in favor of Variables in Eq. (1)
good CSR firms, giving firms an opportunity to enhance their media
image by improving CSR performance. Media = The aggregated positive sentiment
scores of press-initiated news less
the aggregated negative sentiment
Appendix A scores of press-initiated news for
year t scaled by total number of
A.1. TRNA Rating Procedure press-initiated news articles in year t
CSR = Total CSR score for firm i at the end of
TRNA analyzes news items in real-time to determine the senti- year t, calculated by aggregating the
ment of the item using a lexical analysis that uses a net scores (strengths minus
knowledge-driven neural network to rate each news item released concerns) for the community,
about a firm in terms of the tone of the news coverage. The output diversity, employee relations,
of this analytical process is three probability scores that reflect the environment, human rights, and
positive, negative, and neutral tones of the news item. According to product categories in KLD
Thomson Reuters (2013, 6), these scores capture the ‘‘sentiment NStor = The total number of news articles for
expressed by the author about the subject matter being discussed’’. firm i during year t
Thus, TRNA ratings are useful in our context as they capture the ROA = Ratio of the operating income after
journalistic slant embedded in the news reports. depreciation to the total assets for
TRNA uses a three-stage process to assign sentiment scores: firm i at year t
pre-processing, feature extraction, and classification. In the BP = Ratio of the book value of equity to
pre-processing phase, the news item is divided into sentences, the market value of equity for firm i
and each sentence is broken down by part of speech, e.g., noun, at the end of year t
verb, or adjective. In the feature extraction phase, TRNA identifies MV = Log of market value of equity for firm
the words or phrases that convey sentiment. These words and i at the end of year t
phrases are compared to a dictionary of 16,000 words and 2500 Lev = Ratio of debt to total assets for firm i
phrases convey sentiment based on the ratings of three human at the end of year t
annotators. In this stage, the words and phrases are not just ana- SP500 = Indicator variable being 1 if firm i is
lyzed alone but are analyzed in conjunction with each other. For in the S&P500 index at year t, and 0
example, TRNA recognizes negation (e.g., ‘‘not well’’ or ‘‘did not otherwise
go well’’) as well as intensification where favorability can differ IRisk = Standard deviations of the
for the subject and object of the same verb (e.g., ‘‘X outperformed market-model residuals of daily
Y’’ which is positive for X and negative for Y). stock returns measured over the year
In the classification phase, a sentiment score is determined for Ret = Market adjusted buy-and-hold stock
each entity named in the news item. Specifically, TRNA considers returns measured over year t
the sentiment attached to the individual words and phrases in AdExp = The total advertising expense scaled
the news item and assigns three overall sentiment scores that cap- by total sales for firm i at the end of
ture the positive, negative, and neutral tones of the entire news year t
item. Importantly, these overall sentiment scores are not just a CGov = Total score for firm i at the end of
count of the number of positive, negative, and neutral elements year t, calculated by aggregating the
in the news item since some elements carry more weight than net scores (strengths minus
others in terms of setting the tone. To assess the relative weights concerns) for the corporate
of different elements, TRNA employs a three-layer governance in KLD
back-propagation neural network with weight relaxation that has FI-News = The aggregated positive sentiment
been trained using 5000 news articles that have been scored by scores of firm-initiated news less the
three humans. aggregated negative sentiment
According to Thomson Reuters (2013), 75% of the ratings scores of firm-initiated news over
assigned by TRNA agree with ratings made by humans, which is year t scaled by total number of
only slightly lower than the 82% agreement rate when two humans press-initiated news articles in year t
rate the same articles. Aside from its classification accuracy, a key MktPE = The value weighted average of
feature of TRNA is the speed at which the process takes place. The market price-to-earnings ratio of all
entire rating process takes 0.1 s per news item. the firms in the merged
If the item refers to more than one firm, each firm receives its CRSP-Compustat database in year t,
own set of sentiment scores. In addition, TRNA provides a rele- following Conrad et al. (2002)
vance score for each news item that reflects the degree to which
the item focuses on a particular firm. Relevance is scored from 0
 S.F. Cahan et al. / Journal of Banking & Finance 59 (2015) 409–422 421

Appendix B (continued) Appendix B (continued)

Variables in Eq. (1) Variables in Eq. (1)

Instrumental Post1 = 1 if the firm-year observation is one
variables in 2SLS year before or the year when SEO
tests announcement is made, and 0 if it is
Blue = 1 if the state where firm i’s two year before the SEO is
headquarter is located is defined as a announced
Blue state that supports President
Additional variables in Eqs. (6), (7)
Obama in 2008 presidential election
TobinQ = The log of market-to-book ratio for
and 0 otherwise. The Blue state is
firm i at the end of year t
defined by the following webpage:
COCOJ = Implied cost of capital calculated
http://www.electoral-vote.com/
following Ohlson and
evp2010/Info/red-blue.html
Juettner-Nauroth (2005)
Voting = The average margin of victory in the
I-COCOJ = The mean of implied cost of capital
five presidential elections between
calculated following Ohlson and
1992 and 2008 for the democratic
Juettner-Nauroth (2005) for the
president candidate in the state
industry that the firm belongs to
where firm i’s headquarter is located
using 2-digit SIC industry
in. The statistics is from the following
H-H = Dummy variable and it equals to 1
webpage:
when the firm-year observation is
http://en.wikipedia.org/wiki/File:
both above the sample median of
Red-and-Blue-States-Map-(Average-
media favorability and above the
Margins-of-Presidential-Victory).svg
sample median of CSR performance
Additional variables in Eq. (2) and 0 otherwise
CSR1 = 1 if a firm has positive CSR score in RD1 = Spending on research and
year t and 0 otherwise development expenses as reported
RD = 1 if a firm has positive R&D expenses by firm i in year t scaled by sales
in year t and 0 otherwise CapX = Capital expenditure as reported by
Loss = 1 if a firm reports negative earnings firm i in year t scaled by sales
before extraordinary items in year t SGrowth = The percentage change in sales
and 0 otherwise compared to prior year’s sales
Sales = Sales revenue in year t divided by LTG = Analyst long-term growth in
total assets at the end of year t earnings forecast from I/B/E/S
Assets = The natural logarithm of total assets Disperse = Forecast dispersion defined as the
at the end of year t coefficient of variation of I/B/E/S
AvgGrowth = The weighted average sales growth earnings forecasts
over the past five years Beta-Mkt = Market factor loading from Fama and
EP = The ratio of net income before French (1996) three-factor model on
extraordinary items for fiscal year to risk factors
market value of equity in year t Beta-SMB = Size factor loading from Fama and
Vol = The variance of daily stock returns French (1996) three-factor model on
over year t risk factors
Analyst = The natural logarithm of 1 plus the Beta-HML = Book-to-market factor loading from
number of analysts following the Fama and French (1996) three-factor
firm in year t model on risk factors for firm i at the
IO = The percentage of institutional end of year t
holdings at the end of year t

Additional variables in Eq. (3)

Treatment = 1 if a firm’s major customer is the US
government and 0 otherwise
References
Post = 1 if year is 2012 or 2013 and 0
otherwise Ahern, K., Sosyura, D., 2014. Who writes the news? Corporate press releases during
merger negotiations. Journal of Finance 69, 241–291.
Additional variables in Section 3.3 Baker, M., Stein, J., 2004. Market liquidity and a sentiment indicator. Journal of
Sin = 1 if firms with SIC codes 2100–2199 Financial Markets 7, 271–299.
or SIC codes of 2080–2085 or NAICS Baker, M., Wurgler, J., 2006. Investor sentiment and the cross-section of returns.
Journal of Finance 61, 1645–1680.
codes: 7132, 71312, 713210, 71329, Bartlett, F., 1932. Remembering: A Study in Experimental and Social Psychology.
713290, 72112, and 721120 and 0 Cambridge University Press, Cambridge.
otherwise Becchetti, L., Ciciretti, R., Giovannelli, A., 2013. Corporate social responsibility and
earnings forecasting unbiasedness. Journal of Banking and Finance 37, 3654–
BW = Average monthly Baker-Wurgler 3668.
Sentiment Index during one fiscal Bergman, N., Roychowdhury, S., 2008. Investor sentiment and corporate disclosure.
year Journal of Accounting Research 46, 1057–1083.
Bushee, B., Core, J., Guay, W., Hamm, S., 2010. The role of the business press as an
information intermediary. Journal of Accounting Research 48, 1–19.
422 S.F. Cahan et al. / Journal of Banking & Finance 59 (2015) 409–422

Cheng, B., Ioannou, I., Serafeim, G., 2013. Corporate social responsibility and access Kim, Y., Park, M., 2005. Pricing of seasoned equity offers and earnings management.
to finance. Strategic Management Journal 35, 1–35. Journal of Financial and Quantitative Analysis 40, 435–463.
Claus, J., Thomas, J., 2001. Equity premia as low as three%? Evidence from analysts’ Kim, Y., Park, M., Wier, B., 2012. Is earnings quality associated with corporate social
earnings forecasts for domestic and international stock markets. Journal of responsibility? The Accounting Review 87, 761–796.
Finance 56, 1629–1666. Kim, J.-B., Zhang, H., Li, L., Tian, G., 2014a. Press freedom, externally-generated
Cone, 2013. Cone Communications Social Impact Study. Available at: http:// transparency, and stock price informativeness: international evidence. Journal
www.conecomm.com/stuff/contentmgr/files/0/e3d2eec1e15e858867a5c2b1a2 of Banking and Finance 46, 299–310.
2c4cfb/files/2013_cone_comm_social_impact_study.pdf. Kim, Y., Li, H., Li, S., 2014b. Corporate social responsibility and stock price crash risk.
Conrad, J., Cornell, B., Landsman, W., 2002. When is bad news really bad news? Journal of Banking and Finance 43, 1–13.
Journal of Finance 57, 2507–2532. Klayman, J., 1995. Varieties of confirmation bias. In: Busemeyer, J., Hastie, R., Medin,
Cragg, J., Donald, S., 1993. Testing identifiability and specification in instrumental D. (Eds.), Decision Making from a Cognitive Perspective: The Psychology of
variable models. Econometric Theory 9, 222–240. Learning and Motivation, vol. 32. Academic Press, San Diego, pp. 365–418.
Deng, X., Kang, J., Low, B., 2013. Corporate social responsibility and stakeholder Kothari, S., Li, X., Short, J., 2009. The effect of disclosure by management, analysts,
value maximization: evidence from mergers. Journal of Financial Economics and business press on cost of capital, return volatility, and analysts forecasts: a
110, 87–109. study using content analysis. The Accounting Review 84, 1639–1670.
Dhaliwal, D., Krull, L., Li, O., Moser, W., 2005. Dividend taxes and implied cost of Kruger, P., 2014. Corporate goodness and shareholder wealth. Journal of Financial
equity capital. Journal of Accounting Research 43, 675–708. Economics, forthcoming.
Dhaliwal, D., Li, O., Tsang, A., Yang, Y., 2011. Voluntary nonfinancial disclosure and Lang, M., Lundholm, R., 2000. Voluntary disclosure and equity offerings: reducing
the cost of equity capital: the initiation of corporate social responsibility information asymmetry or hyping the stock? Contemporary Accounting
reporting. The Accounting Review 86, 59–100. Research 17, 623–662.
Dhaliwal, D., Radhakrishnan, S., Tsang, A., Yang, Y., 2012. Nonfinancial disclosure Lys, T., Naughton, J., Wang, C., 2015. Signaling through corporate accountability
and analyst forecast accuracy: international evidence on corporate social reporting. Journal of Accounting and Economics, forthcoming.
responsibility disclosure. The Accounting Review 87, 723–759. Margolis, J., Elfenbein, H., Walsh, J., 2009. Does it pay to be good and does it matter?
Di Giuli, A., Kostovetsky, L., 2014. Are red or blue companies more likely to go A meta-analysis of the relationship between corporate social and financial
green? Politics and corporate social responsibility. Journal of Financial performance. Working paper, Harvard University.
Economics 111, 158–180. Merton, R., 1987. A simple model of capital market equilibrium with incomplete
Dougal, C., Engelberg, J., Garcia, D., Parsons, C., 2012. Journalists and the stock information. Journal of Finance 42, 483–510.
market. Review of Financial Studies 25, 639–679. Mian, G., Sankaraguruswamy, S., 2012. Investor sentiment and stock market
Easton, P., 2004. PR ratios, PEG ratios, and estimating the implied expected rate of response to earnings news. The Accounting Review 87, 1357–1384.
return on equity capital. The Accounting Review 79, 73–95. Mullainathan, S., Shleifer, A., 2005. The market for news. American Economic
El Ghoul, S., Guedhami, O., Kwok, C., Mishra, D., 2011. Does corporate social Review 95, 1031–1053.
responsibility affect the cost of capital? Journal of Banking and Finance 35, Naiker, V., Navissi, F., Truong, C., 2013. Options trading and the cost of equity
2388–2406. capital. The Accounting Review 88, 261–295.
Engelberg, J., Parsons, C., 2011. The causal impact of media in financial markets. Nofsinger, J., 2005. Social mood and financial economics. Journal of Behavioral
Journal of Finance 66, 67–97. Finance 6, 144–160.
Epstein-Reeves, J., 2010. Consumers overwhelmingly want CSR. Forbes. Available at: Ohlson, J., Juettner-Nauroth, B., 2005. Expected EPS and EPS growth as determinants
http://www.forbes.com/sites/csr/2010/12/15/new-study-consumers-demand- of value. Review of Accounting Studies 10, 349–365.
companies-implement-csr-programs/. Sargan, J., 1958. The estimation of economic relationships using instrumental
Fama, E., French, K., 1996. Multifactor explanations of asset pricing anomalies. The variables. Econometrica 26, 393–415.
Journal of Finance 51, 55–84. Severin, W., Tankard Jr., J., 1992. Communication Theories: Origins, Methods, and
Fang, L., Peress, J., 2009. Media coverage and the cross-section of stock returns. Uses in Mass Media, third ed. Longman Group Ltd, New York.
Journal of Finance 64, 2023–2052. Shiller, R., 2000. Irrational Exuberance. Princeton University Press, Princeton, NJ.
Fombrun, C., Shanley, M., 1990. What’s in a name? Reputation building and Shroff, N., Sun, A., White, H., Zhang, W., 2013. Voluntary disclosure and information
corporate strategy. Academy of Management Journal 33, 233–258. asymmetry: evidence from the 2005 securities offering reform. Journal of
Gao, F., Lisic, L., Zhang, I., 2014. Commitment to social good and insider trading. Accounting Research 51, 1299–1345.
Journal of Accounting and Economics 57, 149–175. Sinha, N., 2011. News articles and momentum. Working paper, University of Illinois,
Gebhardt, W., Lee, C., Swaminathan, B., 2001. Toward an implied cost of capital. Chicago.
Journal of Accounting Research 39, 135–176. Solomon, D., 2012. Selective publicity and stock prices. Journal of Finance 67, 599–
Goss, A., Roberts, G., 2011. The impact of corporate social responsibility on the cost 638.
of bank loans. Journal of Banking and Finance 35, 1794–1810. Solomon, D., Soltes, E., 2011. The determinants of coverage in the business press.
Graber, D., 1984. Processing the News: How People Tame the Information Tide. Working paper, University of Southern California.
Longman Press, New York. Teoh, S., Welch, I., Wong, T., 1998. Earnings management and the
Grant Thornton, 2011. Public opinion driving CSR activity. Available at: http:// underperformance of seasoned equity offerings. Journal of Financial
www.internationalbusinessreport.com/Press-room/2011/CSR.asp. Economics 50, 63–99.
Griffin, Hirschey, J.N., Kelly, P., 2011. How important is the financial media in global Tetlock, P., Saar-Tsechansky, M., Mackassy, S., 2008. More than words: quantifying
markets? Review of Financial Studies 24, 3941–3992. language to measure firms’ fundamentals. Journal of Finance 63, 1437–1467.
Gurun, U., Butler, A., 2012. Don’t believe the hype: local media slant, local Thomson Reuters, 2013. Thomson Reuters News Analytics: Sentiment analysis,
advertising, and firm value. Journal of Finance 67, 561–597. relevance, novelty. White paper, Thomson Reuters, New York.
Hong, H., Kacperczyk, M., 2009. The price of sin: the effects of social norms on
markets. Journal of Financial Economics 93, 15–36.