Download as pdf or txt
Download as pdf or txt
You are on page 1of 48

Stakeholders and the Stock Price Crash Risk:

What Matters in Corporate Social Performance?


Ariadna Dumitrescu*a , Mohammed Zakriyab
a
ESADE Business School, Universitat Ramon Llull
Av. Pedralbes, 60-62, 08034 Barcelona, Spain
b
IESEG School of Management
3 rue de la Digue, 59000 Lille, France

September 2020

Abstract

This study provides evidence for the differential impacts of corporate social responsi-
bility (CSR) initiatives targeting different stakeholder groups on stock price crash risk.
In particular, it highlights CSR’s role in mitigating risk and creating shareholder value.
Our results reveal that managerial bad news hoarding and the resultant stock crashes
are largely determined by the social CSR dimension, and this effect is predominantly
seen in undervalued firms. Moreover, social CSR subcategories aimed at specific stake-
holder groups (such as the community, employees, or customers) tend to mitigate future
crashes. In contrast, firms’ environmental initiatives and governance characteristics seem
to have trivial effects on stock crashes. Using a quasi-natural experiment, we find that
the mitigating effect of social CSR dimension on crash risk is likely to be causal.

Keywords: CSR, ESG, Crash Risk, Social Performance, Environmental Performance

JEL Classification: G14, G30, M14, Q50

*
Corresponding author, Tel.: +34 934952191
Email addresses: ariadna.dumitrescu@esade.edu (Ariadna Dumitrescu* ), m.zakriya@ieseg.fr
(Mohammed Zakriya)

Electronic copy available at: https://ssrn.com/abstract=3305213


1. Introduction

In 2019, the Business Roundtable (BRT), a large group of major US companies’


CEOs, released a “Statement on the Purpose of a Corporation” in which these CEOs
“commit[ted] to lead their companies for the benefit of all stakeholders – customers, em-
ployees, suppliers, communities and shareholders.” Signed by 222 influential CEOs, this
statement controversially disregards shareholder primacy that dominated previous BRT
statements and has, thus, reinvigorated the debate on firms’ stakeholder vs. shareholder
orientations (see Harrison et al., 2020). However, stakeholder engagement through cor-
porate social responsibility (CSR) activities need not necessarily undermine shareholders’
interest if they can create long-term value for the firms. If this is indeed so, could risk
mitigation be a possible important channel through which stakeholder engagement/CSR
initiatives benefit the firms? To answer this question, this study departs from the tradi-
tional approaches of considering either the association between CSR and firm valuation or
the association between CSR and firm risk in isolation (Derwall et al., 2011; Krüger, 2015;
Ferrell et al., 2016). We study the relationship between CSR, stock price crash risk, and
firm valuation in a single framework to show that the mitigation of future price crashes
is an important channel through which firm value is impacted. CSR itself is a complex,
multidimensional construct and its influence on valuation is non-uniform because of the
opposing effects of its constituent dimensions (Galema et al., 2008). So, we also identify
and isolate the risk-mitigating roles of its constituent stakeholder groups.
Fatemi et al. (2015) theoretically show that CSR activities influence firm valua-
tion “through their effect on cash flows, the firm’s probability of survival, and its cost
of capital.” Empirically, El-Ghoul et al. (2011) find that CSR is negatively related to
cost of capital, whereas Ferrell et al. (2016) show that CSR is positively associated with
firm value. The recent literature largely ascertains the beneficial effects of CSR on firm
value by highlighting the roles of institutional investors (Buchanan et al., 2018), media
attention (Byun and Oh, 2018), and product market competition (Bardos et al., 2020).
However, the role of risk, and, more specifically, of crash risk, in driving valuation benefits
is relatively understudied. If CSR and stakeholder groups alleviate or aggravate crash

Electronic copy available at: https://ssrn.com/abstract=3305213


risk, they are expected to accordingly have a positive effect (Sharfman and Fernando,
2008; El-Ghoul et al., 2011) or negative effect (Garcia-Castro et al., 2010; Bhandari and
Javakhadze, 2017) on shareholder value as well. Since empirical evidence often contra-
dicts Friedman (1970) value-diminishing view of CSR, we aim to understand whether
CSR’s risk-mitigating role offers value benefits to both firms and their investors. Specif-
ically, we study how undervalued firms benefit from CSR activities through crash risk
mitigation. By focusing on stock price crash risk, we investigate managerial bad news
hoarding behavior and its ability to increase information asymmetry. In other words, we
examine whether and how stakeholders can influence managerial bad news hoarding and
the resultant price crashes, and how this relationship is affected by firms’ undervaluation.
We investigate individual categories (i.e., stakeholder groups) of firms’ CSR perfor-
mance to understand how they mitigate risk (and consequently, how they affect firm
valuation). Our main objective is to examine the differential effects of these categories
on managerial bad news hoarding behavior. When firms’ satisfactory CSR performance
translates into responsible financial and accounting reports (the stakeholder effect), such
transparent practices curb the likelihood of managers’ hoarding bad news (Kim et al.,
2014). This decreases the likelihood of future stock price crashes. However, satisfactory
CSR might merely be a whitewashing mechanism to conceal managerial misdemeanor
(Hemingway and Maclagan, 2004). This may be the case when managers indulge in earn-
ings management and financial misreporting (the agency cost effect), which increases the
likelihood of managers’ concealing bad news. Consequently, this increases the likelihood
of future stock price crashes. We expect these contrasting influences on crash risk (i.e.,
the stakeholder vs. agency cost effects) to vary when individual categories of CSR perfor-
mance are considered instead of the overall CSR performance because some of the CSR
categories or dimensions may incentivize managerial bad news hoarding, while others
might mitigate them.
The MSCI Environmental, Social and Governance (ESG) data has seven different
categories of CSR strengths and concerns— community, governance, diversity, employee
relations, environment, human rights, and product. While the community, diversity,

Electronic copy available at: https://ssrn.com/abstract=3305213


and human rights categories show firms’ CSR engagement targeted at the society at
large, the governance, employee relations, and product categories capture CSR initiatives
aimed at shareholders, employees, and customers respectively. Our empirical analysis
reveals that of the ESG dimensions, only the social dimension influences future stock
price crashes. This is important from the investors’ perspective because we show that
the mitigating influence of overall CSR performance on crash risk is mainly driven by
social CSR activities and CSR’s locally oriented subcategories. Reducing downside risk
is important for investors as they are known to be crash-averse (Chabi-Yo et al., 2018).
With respect to the governance dimension, investors seem to be well-informed about
the broad-based governance characteristics reported by the MSCI (such as corruption
and public accountability). Since investors are familiar with these characteristics, the
governance dimension has a negligible impact on crashes.1 As regards the environmental
dimension, while superstars (i.e., superlatively green firms) or villains (i.e. controversially
toxic firms) may be easy to recognize (Fernando et al., 2017), the average firm’s overall
environmental performance is not easily understood by investors. Therefore, it does not
influence extreme stock price declines or crashes.
At the categorical level, further analysis of the social dimension shows that firms
with stronger employee relations, more community initiatives, and superior product per-
formance exhibit a significantly lower likelihood of a future stock price crash, thus sup-
porting the stakeholder hypothesis. However, this association with crash risk is reversed
when human rights are considered. Firms with better ratings in this category exhibit a
significantly higher future stock price crash risk, indicating that human-rights-centered
CSR attributes may be used by firms merely for corporate greenwashing.
Risk and valuation are closely related (Litzenberger and Budd, 1970). Thus, next,
we examine whether valuation is an important driver of the association between CSR

1
Following Bereskin et al. (2018), we include governance within the CSR framework and, subse-
quently, omit it in our robustness checks. Andreou et al. (2016) showed that corporate governance
attributes such as ownership patterns, board characteristics, and managerial incentives influence future
price crash risk. However, the governance attributes covered by MSCI have a much wider scope as they
include social governance characteristics such as business ethics, political stability, and firms’ attitude
toward public policy. Therefore, our findings on the governance dimension differ from those on corporate
governance in previous literature.

Electronic copy available at: https://ssrn.com/abstract=3305213


performance and crash risk. Arguably, the negative relationship between CSR and future
crashes might be due to the value relevance of firms’ altruistic CSR motives. If this
is indeed the case, risk mitigation through CSR will be higher in undervalued firms.
Managers of undervalued firms are less likely to report inflated earnings (Sawicki and
Shrestha, 2014), so their incentives to hoard bad news are lower while also allowing them
to assign CSR expenses more prudently. In contrast, managers of overvalued firms are
more prone to earnings manipulation (Chi and Gupta, 2009), so they may strategically
employ CSR activities to hide their bad news hoarding behavior.
To examine the role of undervaluation on the relationship between CSR (as well as its
ESG dimensions and stakeholder groups) and crash risk empirically, we split the sample
into two by using the median level of the industry-adjusted Tobin’s Q and Rhodes-Kropf
et al.’s (2005) misvaluation measure. Our results show that, indeed, CSR’s mitigating
effect on crash risk is stronger for undervalued firms. Moreover, the social dimension of
ESG is the only one that indicates potential value benefits based on its crash risk miti-
gation, as the undervaluation dummy variable moderates the negative relation between
social dimension and future crashes. As regards the various stakeholder categories, we
find that only the community and employee relations subgroups are moderated by firms’
undervaluation. Taken together, these findings support the good governance view of CSR
(Ferrell et al., 2016) and suggest that CSR, its social ESG dimension, and some of the
social stakeholder groups are related to shareholder wealth maximization through their
mitigating effects on future price crashes.
The robustness of our results on the relationship between the social ESG dimension
and crash risk is corroborated by a quasi-natural experiment—which employs triple differ-
ence (difference-in-difference-in-difference or DDD) analyses—to correct for endogeneity.
Our experimental setup exploits an exogenous shock in the form of “other constituency
statutes” that allow the directors to consider all stakeholders’ interests (such as the em-
ployees’ and the community’s instead of only the shareholders’) when making business
decisions (Strine, 2014).2

2
The state of Maryland passed the other constituency statutes in 1999, whereas Delaware has still

Electronic copy available at: https://ssrn.com/abstract=3305213


To further explain the importance of social CSR dimension in crash likelihood and
its role as a value driver, we run supplementary tests for the agency cost effect by dividing
our sample into high and low agency cost firms by using several proxies. Since finding
a measurable proxy for stakeholders’ interest alignment remains a challenge, testing the
alternative stakeholder effect is difficult. As explained above, the agency cost effect and
the stakeholder effect have opposite influences on crash risk. Consequently, testing for
one effect indirectly allows us to infer the other. Our findings provide evidence for the
existence of the stakeholder effect for the social dimension, which consistently has a
mitigating influence on future crashes for low agency cost firms. This implies that only
social CSR complements good corporate governance—and the lack of agency problems
thereof—in reducing the likelihood of stock price crashes.3 Stock price crashes represent
extreme cases of information asymmetry. Marhfor et al. (2017) documented a direct link
between social CSR and price informativeness, that is, a higher CSR engagement in the
social dimension reduces information asymmetry. Bhandari and Javakhadze (2017) also
showed that social CSR drives the effect of CSR on firms’ capital allocation efficiency.
Thus, our findings on social CSR’s mitigating effect on crashes re-emphasize the benefits
of the social dimension to investors; these benefits could be further expanded by spillovers
of investment efficiency associated with the same social CSR initiatives.
Our findings contribute to the CSR, crash risk, and firm valuation literatures in
several ways. First, we report the isolated effects of individual ESG dimensions on future
stock price crashes to show that only social performance is a determinant of crashes, and
that its risk mitigation is largely driven by undervaluation.4 This implies that some of

not passed these statutes (Geczy et al., 2015). Thus, the treatment effect in our natural experiment
is computed as the change in the likelihood of a price crash when there are variations in the CSR
characteristics between Maryland- and Delaware-incorporated firms from 1999 to 2000. Cheng et al.
(2018) show that firms headquartered in states that adopt other constituency statutes show a significant
increase in CSR engagement. Similarly, Cremers et al. (2019) show that the adoption of these statutes
has a significant effect on shareholder value and that this effect is stronger for firms with a better CSR
performance. In our sample as well, we found qualitatively similar results by estimating the triple
difference for Maryland vs. Delaware firms before and after the adoption of these statutes. Together,
these results indicate that other constituency statutes do significantly impact the CSR orientation of
firms in the states where they are adopted.
3
Harjoto and Jo (2011), for instance, showed that when firm performance is being studied, managerial
entrenchment endogenously determines CSR. Thus, in Section 5, we use the managerial entrenchment
proxy E-Index to further examine the CSR–corporate governance–crash risk nexus.
4
It must be noted that we do not explore finer details of individual CSR activity because changing

Electronic copy available at: https://ssrn.com/abstract=3305213


the growth potential of undervalued firms is realized through the ability of social CSR ini-
tiatives to mitigate future stock price crashes. Second, some stakeholder groups, namely,
employees, the community, and customers, are critical for mitigating crash likelihood.
However, we identify that CSR activities targeting multiple stakeholder groups may have
no impact on crash risk (such as those covered in the diversity and environment cate-
gories) or may exacerbate this risk (the human rights category). We also show that the
strong stakeholder effect on crash risk is alleviated by the lack of its association with
shareholder-focused ESG activities (the governance category). In a recent study, Byun
and Oh (2018) showed that CSR activities that are locally oriented (with direct benefits
for stakeholders) create shareholder value, while the ones that are externality-oriented
do not create similar value. In a similar vein, Girerd-Potin et al. (2014) showed that
significantly larger risk premia are associated with the aforementioned locally oriented
activities when compared to other CSR activities. Our findings reveal that these lo-
cally oriented CSR categories are also negatively associated with firms’ future stock price
crashes. Third, we contribute to the literature by exploiting an exogenous legal shock
that allows us to capture the causal effects instead of merely the associations. We pro-
vide evidence of a causal relationship between some of the ESG dimensions/categories
and crash risk. Finally, we provide some preliminary insights on what drives the social
CSR’s relationship with crash risk. Exploring the stakeholder–agency cost interplay, we
identify the differential impacts of ESG activities on crashes, when manager and investor
perspectives are considered separately.
The rest of this paper is structured as follows: In Section 2, we discuss the rela-
tionship between CSR and crash risk, and highlight the role of firm valuation in this
relationship. In Section 3, we describe the sample, data, and variables employed. Next,
Section 4 presents our main results, including the identification strategy employed for
causal inferences. In Section 5, we explore a possible agency-based explanation for why
the social dimension is important to crash risk and firm valuation. Finally, Section 6
summarizes and concludes the paper.

MSCI data collection practices make it difficult to trace how each of them evolves with time. This issue
does not exist in the case of ESG dimensions.

Electronic copy available at: https://ssrn.com/abstract=3305213


2. CSR and Crash Risk

Notwithstanding managers’ fiduciary responsibility toward the shareholders, they


often have to indulge other stakeholders through trust-gaining activities (Eccles et al.,
2014; Lins et al., 2017). Generally, managers have more information regarding their
firms’ current and future operational performance and financial health than investors
and other stakeholders. Therefore, their decision about whether to withhold or disclose
this information (opacity versus transparency) is influenced by agency costs and signaling
motives (Healy and Palepu, 2001). Meanwhile, the strategic gamble of firm opacity is
governed by the tension between managerial misappropriation incentives and possible
future downside risks (Jin and Myers, 2006). When there are consecutive periods of bad
news hoarding, the decision of disclosing or withholding this information eventually goes
beyond the control of managers, thereby leading to its sudden release, which causes a price
crash. When CSR activities are negatively associated with reporting opacity (thereby
increasing financial transparency), a firm is less susceptible to crashes. Alternatively,
when CSR engagement is merely used to divert investors’ scrutiny, the firm is more
susceptible to crash risks.

2.1. The Role of Firm Valuation

There is mixed evidence on the association between CSR and firm value (Van Beur-
den and Gössling, 2008). However, in recent years, the positive effect of CSR on firm
valuation has been firmly established (Ferrell et al., 2016; Fernando et al., 2017, 2019).
CSR activities can benefit firms through positive externalities, for example, by signaling
superior product quality (Bardos et al., 2020) or by attracting a larger media coverage
(Byun and Oh, 2018), especially when there is high customer awareness (Servaes and
Tamayo, 2013). However, little is known about internal mechanisms that drive the ben-
efits of CSR. The dominant governance view (Harjoto and Jo, 2011), which considers
CSR as complementary to good governance, thereby reducing information asymmetry
and increasing firm valuation, is largely employed to explain the benefits of CSR. How-
ever, valuation benefits from CSR may also arise from a decline in the cost of capital
(Fatemi et al., 2015) or from risk reduction (Fernando et al., 2017). This suggests that

Electronic copy available at: https://ssrn.com/abstract=3305213


risk mitigation is an alternative channel through which firms can benefit.
We expect the relationship between a firm’s CSR activities and its stock price crash
risk to be moderated by firm valuation. We conjecture that in overvalued firms, managers
use CSR activities to merely mask their earnings management activities. Chi and Gupta
(2009) show that managers of overvalued firms are more prone to earnings manipulation.
Thus, they can hoard more bad news by exploiting the outwardly feel-good image por-
trayal of their firms through CSR investments. This is not the case in undervalued firms,
where there are fewer incentives for managers to hoard bad news. Hence, in these firms,
CSR expenses are more prudently assigned. Therefore, the benefits of CSR activities in
terms of mitigating future crashes are expected to be greater for undervalued firms.

2.2. Stakeholders and Crash Risk

The mitigating effect of firm’s overall CSR performance ratings on crash risk has been
shown by Kim et al. (2014). However, the activities covered under the CSR performance
umbrella are directed toward several stakeholders, such as the employees, customers,
supply chain, and the society at large. This implies that different sets of CSR activities
may have different influences on crash risk depending on how much of an incentive they
provide toward managerial bad news hoarding behavior. We expect these influences to
depend on one or more of the following: the nature of the CSR activity, the power of
each stakeholder group, the extent of the managerial attention, and the market sensitivity
toward CSR.
Stock return variability and other risk measures capture the covariation between a
firm’s stock returns and the markets, but they do not represent the true downside risk
to which a firm’s shareholders are exposed. Crash risk is measured through isolated neg-
ative outliers experienced by the firm’s returns, and it can, therefore, effectively capture
the investors’ exposure. Thus, the consequences of bad news hoarding behavior by the
managers are better reflected in the crash risk measures than the second moments around
the mean returns (i.e., variances). Moreover, we expect CSR categories to have different
effects on managerial bad news hoarding, and hence, price crashes. Bouslah et al. (2013)
presented the following three arguments for the differential impacts of CSR categories on

Electronic copy available at: https://ssrn.com/abstract=3305213


firm risks: a) ambiguity in measuring and understanding CSR across these categories, b)
the need for isolating individual ESG dimensions’ (or CSR categories’) influences, and c)
within-industry CSR heterogeneity. These arguments can also be used for explaining the
differential impacts of CSR categories on managerial bad news hoarding, and, hence, on
the likelihood of future stock price crashes.

3. Sample and Data

The metrics related to CSR performance, its ESG dimensions, and categories are
obtained using MSCI (formerly Kinder, Lydenberg, and Domini Research & Analytics,
Inc. or RiskMetrics-KLD) ESG data. Our sample consists of almost 35,800 firm-year ob-
servations on US companies, spanning from 1991 to 2015. The sample size varies across
the years and has increased from approximately 650 companies in 1991 to over 3,000
in 2015. MSCI assesses each of these companies on multiple ESG strengths (exemplary
standards) and concerns (controversial issues). To measure CSR performance, most of
the CSR literature aggregates these strengths (+) and concerns (–) (e.g., Bhandari and
Javakhadze, 2017). However, following Kim et al. (2014) and Lins et al. (2017), we stan-
dardize these measures to keep year-on-year changes comparable as, over the years, MSCI
adds (removes) some of the indicators to (from) its ESG assessment. Since we apply an
industry-based standardization (see Table 1 for computational details), our measures of
CSR performance are unaffected by MSCI’s 2010 changes to the data collection and
assessment methodology (which gives prominence to industry-specific indicators).5 We
create the aggregate CSR performance measure CSP , which includes all the available
seven sets of ESG qualitative indicators (i.e., community, governance, diversity, em-
ployee relations, environment, human rights, and product). This provides us with a
holistic stakeholder view that includes shareholders (through governance characteristics)
to ensure that there are no possible biases due to the category omissions (about which

5
Until 2009, each of the firms assessed by MSCI was evaluated using all ESG strengths and concerns.
However, beginning in 2010, this methodology was updated to ensure that only a limited number of
industry-relevant indicators are used to evaluate each firm. Therefore, to address the indicator selection
issue, we additionally check the robustness of our results by restricting our sample up to 2009 and find
that all our inferences remain unaffected.

Electronic copy available at: https://ssrn.com/abstract=3305213


Ferrell et al., 2016 cautioned).6 Moreover, the governance characteristics covered by
MSCI cannot be omitted from the CSR measure because they include many stakeholder-
specific indicators (in this case, non-shareholder ones), such as business ethics, political
accountability, and corruption.
To identify the effects of individual ESG dimensions, we decompose the CSP mea-
sure into its three ESG sub-measures (i.e., EN V , SOC, and GOV ). This is done by
employing the same standardization as for CSP , but only for a subset of the relevant
categories (i.e., community, diversity, employees, human rights, and product for SOC;
governance for GOV ; and, environment for EN V ). Finally, each of the subcategories
covered by the SOC measure, namely, community (COM ), diversity (DIV ), employees
(EM P ), human rights (HU M ), and product (P RO) are also assessed separately.
All the annual accounting-based controls, analysts’ forecasts, CEO-specific data, in-
stitutional holdings details, entrenchment provisions, and the daily stock returns data for
computing the crash risk measures are obtained from COMPUSTAT, I/B/E/S, Execu-
Comp, Thomson Reuters, ISS-Riskmetrics, and the Center for Research in Stock Prices
(CRSP) database, respectively. Table 1 presents the definitions and computational de-
tails for each of the main variables and controls used in our analyses.

3.1. Crash Risk Measures

Crash risk, the main dependent variable, is measured as follows: As shown in Hut-
ton et al. (2009), we first estimate the firm-specific weekly returns (F SW Ri,t ) from the
residuals ϵi,t obtained using an expanded index model based on firm i’s Wednesday-to-
Wednesday returns for a week t, ri,t :

ri,t = αi + β1,i ∗ rm,t−2 + β2,i ∗ rm,t−1 + β3,i ∗ rm,t + β4,i ∗ rm,t+1 + β5,i ∗ rm,t+2 + ϵi,t (1)

where rm,t is CRSP value-weighted market return, with its additional one/two-weeks-
lagged and forward returns as controls.7 Subsequently, to obtain the F SW Ri,t , the
6
For a robustness check, we also create an alternative CSP measure that excludes the governance
and human rights indicator sets and compare our results to those in prior research that employs such
category omissions (e.g., Kim et al., 2014). Although untabulated, all the results using this alternative
CSR construct are largely similar to those obtained with our original CSP measure.
7
While standard asset pricing models perform well in measuring idiosyncratic risk, their utility is
limited when measuring crash risks (Chen et al., 2001). With index-based models that introduce lead

10

Electronic copy available at: https://ssrn.com/abstract=3305213


residuals from this model are transformed to ln(1 + ϵi,t ). Thus, crash risk (CRASH) is
measured through a dummy variable that takes the value of 1 if a firm has undergone at
least one crash in a given year. The crashes are identified when F SW Ri,t falls below 3.09
standard deviations of the average F SW Ri,t in a given year.8 We create an additional
variable, CRASHN U M , which measures the annual number of crash weeks.
CRASH and CRASHN U M have been extensively used in the literature to measure
crash risk. CRASH was introduced in Hutton et al. (2009), to measure the event of a
stock’s price falling below 5% of its mean value in any given year. It is common in the lit-
erature to use this proxy in logistic regressions, while using negative conditional skewness
(N CSKEW ) and down-to-up volatility (DU V OL) for robustness checks (Chang et al.,
2017; Deng et al., 2018; Al Mamun et al., 2020; Dumitrescu and Zakriya, 2020).9 Habib
et al. (2018), in an empirical review of the crash risk literature, show that CRASH is
used as a proxy for stock price crash risk as frequently as N CSKEW and DU V OL, if
not more frequently. Thus, to test the robustness of our results, we also use N CSKEW
and DU V OL (see Chen et al., 2001) as alternative stock price crash risk measures.

3.2. Firm Valuation Measures

As a proxy for firms’ misvaluation (i.e., under- or over-valuation), we use the Tobin’s
Q and Rhodes-Kropf et al. (2005) measures. Following Bebchuk et al. (2009), we compute
Tobin’s Q and its industry-adjusted values (calculated based on industry medians using
Fama and French’s (1997) 48 industry classification). Alternatively, following Rhodes-
Kropf et al. (2005), we measure misvaluation (RRV ) using a decomposition of the market-
to-book ratio. Specifically, we consider the model by Rhodes-Kropf et al.’s (2005), which
regresses market value on book value, net income, and leverage. The residuals from this
regression are then used as a proxy for misvaluation (RRV ). In other words, RRV is

and lag returns, we can control for possible autocorrelations in market indices. Therefore, the crash risk
literature always uses index-based models (e.g., Hutton et al., 2009).
8
The 3.09 threshold represents the 5th percentile of F SW R for any given year. Using the 5th or 1st
percentile as a threshold does not change our main results. In the natural experiment subsample (Section
4.3.3), the threshold used is 3.2 standard deviations, corresponding to the average of 5th percentile of
the F SW R values for that period.
9
Kim et al. (2011a), for instance, use CRASH to study the impact of tax avoidance on crash risk.
More recently, Li and Zeng (2019) and (Lee et al., 2020) have employed CRASH to study the effects of
CFO gender and firms’ customer concentration, respectively, on crash risk.

11

Electronic copy available at: https://ssrn.com/abstract=3305213


the abnormal valuation that is not explained by industry-specific valuation multiples (for
more details, see Adebambo and Yan, 2018). For both Tobin’s Q and RRV , we then
define an U ndervalued dummy variable, using their respective cross-sectional median
values in each year.

3.3. Summary Statistics

Table 2 shows summary statistics for the crash risk measures (CRASH and
CRASHN U M ), the CSR variables, and all the remaining controls. The CSR measures
are the overall CSR performance (CSP ), its three ESG dimensions (environmental EN V ,
social SOC, and governance GOV ), and the five subcategories of the social dimension
(COM , DIV , EM P , HU M , and P RO).

4. Stakeholders, Crash Risk, and the Moderating Role of Firm Valuation


4.1. Stakeholders and Crash Risk

We model the relationship between stakeholder management through CSR initiatives


and future stock price crashes through the following regression model:

SP CRi,t = β0 + β1 CSRi,t−1 + β2 Xi,t−1 + εi,t (2)

where the stock price crash risk variable SP CRi,t is either proxied by CRASH or
CRASHN U M for firm i in year t, as defined previously. CSRi,t−1 is either the over-
all CSR performance (CSP ), a vector of the three ESG dimensions (EN V , SOC, and
GOV ), or a vector of all the CSR subcategories for firm i in the previous year, t − 1. To
control for firm-specific characteristics that have been shown to determine future crashes
in prior literature (e.g., Hutton et al., 2009; Kim et al., 2014; Ben-Nasr and Ghouma,
2018), we include the previous year’s firm size, market-to-book ratio, leverage, return on
assets, average annual F SW R, the dispersion of F SW R, accounting opacity, differential
turnover, and negative conditional skewness (see Table 1 for details). When the dummy
variable CRASH is used to model crash likelihood, we use a Logit regression, and when
the number of crashes per year, that is, CRASHN U M , is used, we use a Tobit regression.
Table 3 summarizes the main results for our baseline CRASH (Logit) and
CRASHN U M (Tobit) regressions, which include year fixed effects to account for time

12

Electronic copy available at: https://ssrn.com/abstract=3305213


trends. The marginal effects from the Logit regressions for all the main variables are
given in Table 4. Concerning the relationship between future stock price crash risk and
the aggregated measure of CSR (Model 1), our results agree with the findings in Kim
et al. (2014). CSR has a mitigating effect on crash risk, which suggests that stakeholder
influence tends to discourage bad news hoarding behavior within our sample firms. How-
ever, our focal aim is to test whether one or all of the ESG dimensions are relevant to
crash risk. Our results for these three dimensions (Model 2) indicate that only the social
dimension affects the likelihood of future price crashes. This differs from the results of
Kim et al. (2014), who show that all CSR activities affect crash risk when taken as a
whole, and not as individual subcategories. Table 4 shows that the marginal negative
effect of SOC on crash risk is 3.5% (compared to the 4.4% marginal effect of the aggre-
gated CSP ). Given that the overall likelihood of a crash in our sample is 22.5%, these
marginal effects (20% for CSP and 16% for SOC) are economically significant to future
crashes. Additionally, we find that almost 80% of the crash likelihood captured by the
firms’ overall CSR performance is explained by its social dimension.
Furthermore, in Tables 3 and 4, we examine the social dimension of CSR (Model 3)
by identifying the impact of its five subcategories on crash risk. We find that three of the
categories, that is, community, employee relations, and product have almost equivalent
mitigating effects on the likelihood of a future crash. Based on Table 4, there is an
approximately 2.5% reduction in the probability of future crashes when all other variables
are kept at their average levels. This accounts for an economically significant influence
of over 10% out of the overall crash likelihood of 22.5%. The CSR activities covered
under the human rights category, meanwhile, have an opposing effect, which contributes
to future crashes instead of mitigating them. Its positive marginal impact on crash risk
is approximately 1.8%, which is again both statistically and economically significant.
When the proxy used for crash risk is CRASHN U M , we use a Tobit regression left-
censored at 0 and all our main results remain unchanged. A large part of the statistically
significant relationship between the aggregated CSP and the number of future crash
weeks is explained by the social dimension (i.e., approximately 80%), as observed for the

13

Electronic copy available at: https://ssrn.com/abstract=3305213


CRASH variable. With respect to the subcategories within the social dimension—the
community, employee relations, and product—we find that they again mitigate future
crash risk. Additionally, we run ordinary least squares (OLS) regressions using alterna-
tive measures of crash risk (i.e., N CSKEW and DU V OL). We find that the results for
N CSKEW and DU V OL are qualitatively similar to those in Table 3 for both the CSP
and its SOC dimension (see Appendix Table A.1) until the crisis years. In fact, the mag-
nitude of the SOC coefficient for these variables exceeds the 80% mitigating magnitude
of CSP observed with our main crash risk proxies.10
To further understand the impact of the financial crisis, following Lins et al. (2017),
we analyzed how CSR’s relationship with crash risk varies before, during, and after the
crisis periods by introducing dummy variables representing the crisis years (2008 and
2009) and post-crisis years (2010 onward). By taking the interactions of our main CSR
variables with these crisis and post-crisis dummy variables, we aim to identify whether the
crisis has any discernible impact on the relationship between future crashes and CSR.11
Further, we find that, for both our main crash risk measures, the coefficients of the two
interaction terms (i.e., for the crisis and post-crisis) are statistically insignificant, indi-
cating that the mitigating influence of CSR (and, its social dimension and subcategories)
on crash risk remains consistent across the crisis/non-crisis years (see Appendix Table
A.2).

4.2. Moderating Role of Firm Valuation

Next, we explore the triad of stakeholder groups (or CSR), firm valuation, and future
stock price crashes together. When stakeholder management initiatives are strategically
employed to increase a firm’s valuation, they benefit undervalued firms the most and this
can help in mitigating a firm’s crash risk. Therefore, we study the moderating role of

10
For both N CSKEW and DU V OL, the statistical significance of our results is restricted to the pre-
financial crisis period (i.e., before 2009). This implies that the mitigating effects of CSP and the SOC
dimension on future crashes are weakly captured by N CSKEW and DU V OL (see, also, Appendix Table
A.2). Using a different US sample, Utz (2017) showed similar variation in the statistical significance of
results across different measures of crash risk for the aggregated CSR measure.
11
Even when using the 2000–2002 Dot Com Bubble as the crisis event in place of the 2008–2009
finance crisis, we obtain similar results. We find that the crisis does not affect CSR’s relationship with
crash risk.

14

Electronic copy available at: https://ssrn.com/abstract=3305213


undervaluation on crash risk in the presence of CSR engagement and also test whether
the mitigating effects of social CSR activities (SOC) are mainly driven by their valuation
benefits for undervalued firms.
To understand the role that valuation plays in the relationship between stakeholder
management initiatives and the future stock price crashes empirically, we introduce the
undervaluation dummy variable U ndervalued in Equation 2:

SP CRi,t = γ0 + γ1 CSRi,t−1 + γ2 U ndervaluedi,t−1 + γ3 CSRi,t−1 ∗ U ndervaluedi,t−1

+γ4 Xi,t−1 + εi,t .


(3)
The variable U ndervaluedi,t−1 is defined using industry-adjusted Tobin’s Q (T Q) or the
misvaluation measure (RRV ) by using an above/below median classification in each sam-
ple year (as explained in Section 3.2). The dependent variable SP CRi,t is either proxied
by CRASH or CRASHN U M . Similar to Equation 2, CSRi,t−1 is either CSP ; a vector
of the EN V , SOC, and GOV dimensions; or, a vector of the seven CSR subcategories.
In Table 5, the coefficients of both the T Q-based and RRV -based measures of
U ndervalued show that both these are significantly related to crash risk. However, the
sign is not consistent across the two measures. Although the overall CSR performance
exhibits statistically insignificant relationship with crash risk for overvalued firms (i.e.,
the coefficient of non-interacted CSP term), it is negatively related to crash risk for
undervalued firms (i.e., the coefficient of CSP interacted term). With the three ESG
dimensions, we find that only the SOC dimension shows consistent signs in its rela-
tionship with crash risk for undervalued firms. The interaction term between SOC and
U ndervalued is negative and statistically significant for both the firm valuation proxies.
Thus, our main results in Table 3 also hold for SOC (Model 2), with Table 5 showing
that the benefits of social CSR activities through mitigation of crash risk tend to ac-
company the valuation benefits of these activities for undervalued firms. For the five
subcategories of SOC (Model 3), only the community (COM ) and employee relations
(EM P ) categories reflect the mitigating effects on future crash likelihood for undervalued
firms. While the interaction term of EM P and U ndervalued is statistically significant

15

Electronic copy available at: https://ssrn.com/abstract=3305213


for both the proxies, the interaction term for COM is only significant with the Tobin’s
Q-based undervaluation measure.
Note that interaction terms should be interpreted with some caution since exogenous
variations in the interaction terms are difficult to isolate. Endogeneity concerns make it
hard to ascertain whether CSP /SOC initiatives causally mitigate crash risks conditional
on firms’ undervaluation. Nevertheless, since ESG measures generally tend to show little
variation over time allowing us to largely estimate cross-sectional effects, the endogeneity
issues related to the interaction terms are somewhat alleviated.

4.3. Quasi-Natural Experiment

To explore the causal relationship between CSR and crash risk, we employ a natural
experiment that exploits the ratification of the state-level “other constituency statutes”
as an exogenous shock to the firms’ CSR outlook and related stakeholder engagement
motives. These statutes empower firms’ directors to look beyond the shareholders and
consider other stakeholders’ interests when making business decisions (Strine, 2014; Geczy
et al., 2015). For example, in a US state, where the other constituency statutes apply,
even if a takeover bid with a hefty value premium is made, directors are allowed to decline
the bid if, for instance, the employment of a large number of the firm’s employees is at
stake. During our sample period, several states passed these statutes, while the others
did not. This provides us with a research setting that allows using a triple differences or
DDD estimation.12
Many studies have recently employed the adoption of these statutes in quasi-
experimental settings to study the outcomes of CSR. For example, Flammer and Kacper-
czyk (2016) and Ni (2020) show the causal effects of CSR on innovation and earnings
management, respectively. We set Delaware-based firms as the control group since the
state of Delaware has not passed the constituency statutes, whereas the Maryland-based

12
We assume that the other constituency statutes have a uniform effect on all the CSR dimensions.
However, this is a conservative assumption, considering that the purpose of these statutes is to pro-
mote stakeholder-inclusive decisions. Moreover, finding a shock for each of the dimensions separately is
problematic because of possible multiple treatment interference, for example, when the shocks targeting
different stakeholders overlap. Additionally, environmental or social shocks may not be strictly exoge-
nous or exclusive to a specific stakeholder group leading to pre-shock anticipations and spillovers in other
dimensions.

16

Electronic copy available at: https://ssrn.com/abstract=3305213


firms form the treatment group because the state of Maryland passed these statutes in
1999.13 Accordingly, for our causal estimation, we define the indicator variable treat,
which takes a value of 1 for Maryland-incorporated firms and a value of 0 for Delaware-
incorporated ones. We consider a four-year experimental period from 1998 to 2001, so that
the pre- and post-shock periods each cover a two-year period. The post-statute-passage
years are indicated by the dummy variable post. Thus, our DDD model estimates the
difference in the effect of CSR performance on future stock price crashes in Maryland
before and after 1999 (when the other constituency statutes were passed) relative to the
corresponding difference for the state of Delaware. By controlling for the unconditional
effects of all the relevant variables (i.e., treat, post, and the CSR measures) on crash risk,
in addition to including the respective double-interaction terms, we isolate the coefficient
on the triple-interaction term as the experimental effect on crash risk. Thus, all our DDD
estimations employ the following Logit specification:

CRASHi,t = a + b1 CSRi,t−1 + b2 posti,t + b3 treati,t + b4 CSRi,t−1 ∗ posti,t

+b5 CSRi,t−1 ∗ treati,t + b6 posti,t ∗ treati,t (4)

+b7 CSRi,t−1 ∗ posti,t ∗ treati,t + c Xi,t−1 + εi,t

As in Equation 2, CSRi,t−1 represents either the overall CSR performance (CSP ), a


vector of the three ESG dimensions (EN V , SOC, and GOV ), or a vector of all the CSR
subcategories for firm i in the previous year, t − 1. We apply propensity score (PS)
matching by using size, leverage, and profitability to counter selection biases. Finally, to
validate our results, we conduct placebo tests.
Table 6 shows the results of our DDD Logit regressions for the CRASH variable. Al-
ternative Tobit models using CRASHN U M as robustness check provided similar results.
In the baseline DDD estimations, while the triple interaction term for the aggregate CSP

13
For a complete list of the passage dates of the other constituency statutes in US states, see Barzuza
(2009). The state of Maryland exhibits the second-largest number of incorporated firms in our sample,
after Delaware, and its passage of stakeholder-orientated statutes in 1999 allows to capture pre- versus
post-shock treatments effectively. We also considered using other states but found that 30 of the 34
states that have adopted these statutes, did so before 1993. Of the remaining four, Maryland was the
only one with a sizable sample. The number of firms incorporated in the other three states, that is,
Connecticut, Vermont, and Texas, were less than 10% of the number of Maryland-incorporated firms.

17

Electronic copy available at: https://ssrn.com/abstract=3305213


fails to indicate a causal relationship with future crashes (Panel A), its social dimension
SOC shows evidence of a causal link with such crashes (Panel B). Furthermore, we find
that three of the subcategories of SOC (i.e., the community, human rights, and product)
have statistically significant triple interaction terms (Panel C), indicating that most of
the associations in Table 3 are causal in nature (except for employee relations).14 For PS-
matched DDD, we find that the economic and statistical significance of the coefficients
remain largely the same as those for the baseline. With the PS-matched estimation,
we confirm that our baseline results are not driven by selection biases, especially since
there is a large difference in the sample size between the control and treatment groups.
Table 7 summarizes the main firm-based control variables for each of these groups before
and after the PS-matching. The three firm characteristics that we employ for matching
the firms in the treatment and control groups are shown to sufficiently balance the two
groups on most of these variables. An additional placebo test that assumes an ad-hoc
treatment year (i.e., 2009 or 10 years after the actual treatment) for the two aforemen-
tioned groups confirms that our results are not random and did not exist during the
alternative non-experimental period.15

4.4. Why Does the Social Dimension Matter?

Managers cannot hide broad-based governance information covered by MSCI under


the governance category (such as ownership and accountability) from investors; hence,
the governance dimension does not have an impact on crashes. Since environmental
initiatives are long-term oriented and difficult for investors to understand, they do not
influence the possible declines in stock returns or crashes either. However, as regards so-
cial CSR activities, although some of them are long-term oriented, investors and markets
can assess their costs and benefits. This is further facilitated by the fact that SOC initia-
tives are aimed mostly at primary stakeholders (such as customers and employees). With
respect to these stakeholders, the firms and their managers hold considerably influential

14
It must be noted that the Logit regressions using CSR subcategories fail to converge, and, hence,
Panel C reports the OLS results for indicative purposes.
15
By considering a 10-year gap between the actual and placebo treatment years, we ensure that the
alternative placebo experiment is sufficiently distinct. However, similar “non-results” are also obtained
with other placebo tests in earlier periods.

18

Electronic copy available at: https://ssrn.com/abstract=3305213


power. Accordingly, Girerd-Potin et al. (2014) showed that the risk premium for primary
stakeholders is significantly larger than for “societal” stakeholders (i.e., the environment
and society at large). To the extent that firms can identify primary stakeholders’ utility
functions, we can expect stakeholder management to potentially drive firms’ competitive
advantage through reputational or similar supplemental value-enhancing effects (Harri-
son et al., 2010). Consequently, this will draw managers’ attention toward social CSR
activities, both in terms of their inclusion in decision-making and their effective com-
munication to the concerned stakeholders (Byun and Oh, 2018). Together, these will
improve firms’ information transparency and thereby discourage bad news hoarding.
Our comparisons between the stakeholder effect (i.e., stakeholder–shareholder inter-
est alignment ⇒ more transparency ⇒ lower possibility of bad news hoarding ⇒ lower
crash risk) and the agency cost effect (i.e., corporate greenwashing and more entrench-
ment ⇒ less transparency ⇒ higher possibility of bad news hoarding ⇒ higher crash risk)
have shown that the SOC dimension of CSR is a critical determinant of future crashes.
Moreover, when we explored the value channel of risk mitigation, we found that risk
mitigation through the SOC dimension is driven largely by undervalued firms (i.e., it is a
combined effect of a) undervalued firm ⇒ more transparency ⇒ lower possibility of bad
news hoarding ⇒ lower crash risk, and, b) undervalued firm ⇒ more transparency ⇒ less
managerial leeway ⇒ more prudent CSR investments). However, this does not shed light
on the underlying mechanisms that make the stakeholder effect important in mitigating
the crash likelihood. Thus, in this section, we further examine the non-existence of the
agency cost effect to understand why only the social dimension within ESG matters for
crash risk.
We run subsample tests to see if there are differences in the way investors and man-
agers react to ESG initiatives, which, in turn, affect crash risk. We identify subsamples
based on the institutional ownership percentage and the number of blockholders (to ob-
tain the investors’ perspective), and managerial entrenchment and earnings management
(to understand managers’ responses), by grouping firms into low (< median) and high
(≥ median) firms for each of these agency cost proxies.

19

Electronic copy available at: https://ssrn.com/abstract=3305213


The results are summarized in Table 8. From Panel A, we observe that aggregate
CSP reduces the future crash likelihood only when the agency costs are low, that is, in
the subsamples with low entrenchment, high institutional ownership, high blockholders,
and high transparency. Panel B provides further insights into this trend by exploring the
three ESG dimensions. As expected, the agency cost effect does not exist for any of these
three dimensions. The high entrenchment, low institutional ownership, low blockholders,
and low transparency subsamples have insignificant effects on future crashes. When
examining the low agency cost subsamples in the presence of higher investor monitoring
via institutional ownership and blockholding, we find that only the social dimension SOC
has the potential to determine future crashes. Similarly, when we consider manager-
specific measures, that is, entrenchment and earnings management, there is a marginal
effect only for the social CSR dimension. Hence, crash risk is only influenced by the social
dimension. Across the agency cost proxies and their subsamples, the environmental and
governance dimensions remain consistently unimportant for crash risk.
The results in Panel C largely support the inferences drawn from Panel B with respect
to the three broad ESG dimensions. For the five subcategories of the SOC dimension,
their relationship with crash risk is inconsistent across various agency cost proxies. This
indicates that although the overall SOC dimension mitigates future crashes in firms with
low agency costs, this influence is not driven by any single social subcategory. Moreover,
the coefficients of social activities targeting the society at large are either insignificant
(for diversity) or positive (for human rights), corroborating our previous results that
the stakeholder effect is largely driven by CSR activities that directly benefit specific
stakeholders (i.e., the community, employees, or customers).

Additional Tests: We run further tests using firms’ fundamental characteristics (i.e., size
and leverage) and sample sub-periods (see Appendix Table A.3). With respect to size,
we expect greater scrutiny by investors for larger firms’ ESG initiatives when compared
to those of smaller firms. Indeed, in Table A.3 Panel B, we see that larger firms derive
greater mitigating benefits from the social dimension of CSR. From the leverage sub-
grouping, consistent with the contract theory, we find that more leveraged firms benefit

20

Electronic copy available at: https://ssrn.com/abstract=3305213


from social CSR initiatives as these mitigate their future crash likelihood. Lins et al.
(2017) showed that CSR exhibited differential firm-specific during the crisis- and non-
crisis years. In our context, on the one hand, investors are expected to be more prudent
during the crisis years. On the other hand, managers are more likely to withhold bad
news during the crisis. Together, these factors would entail that crash risk should, there-
fore, be associated with other ESG dimensions in addition to social CSR during the crisis
years. Our results confirm this proposition. While only the social CSR dimension affects
future crashes during non-crisis years, all three ESG dimensions are influential in crisis
years. The positive coefficient of the environmental dimension during the crisis years
affirms our previous tests using agency cost proxies. In other words, the benefits and
costs of broader ESG dimensions or subcategories, such as the environment, are difficult
for the markets and investors to understand, which, during periods of low public trust,
is met with greater scrutiny, thereby increasing the likelihood of crashes. Finally, we
identify subsamples based on the “learning” phenomenon. Borgers et al. (2013) show
that investors’ attention to ESG issues increased until 2004. After that, the markets and
investors started to understand their value-relevance (this being attributed to the disap-
pearance of risk-adjusted returns for ESG-based portfolios after 2004). We expect this
learning process to have an impact on the mitigating effect of social CSR as well. Indeed,
the mitigating influence of social CSR is significantly lower in post-learning years (after
2004) than in pre-learning ones. As regards the social subcategories (Table A.3 Panel
C), the essence of much of our discussion is still applicable. The benefits of mitigation
from the social subcategories are observed for larger and more leveraged firms. In the
sub-period tests, the social subcategories provide additional insights into managerial bad
news hoarding. Several social subcategories mitigate future crashes during the non-crisis
and post-learning years than during the crisis and pre-learning years, respectively. This
may be indicative of the lower likelihood of greenwashing by managers when social trust
is higher (in non-crisis years) and when the magnitude of stock markets’ reaction to CSR
news is smaller (in the post-learning period).

21

Electronic copy available at: https://ssrn.com/abstract=3305213


5. Robustness Checks and Alternative Models
5.1. Additional Controls

There are other relevant controls, such as managerial entrenchment, institutional


ownership, CEO’s compensation, and CEO’s age, that have been shown to influence crash
risk (e.g., An and Zhang, 2013; Kim et al., 2011a; Andreou et al., 2017). We include these
as additional control variables as a robustness check against omitted variable bias. These
variables are not used to obtain our main results as the data on institutional owner-
ship, compensation structures, and entrenchment provisions are limited, which leads to
a considerable loss of sample size, especially when all these are included together in the
regressions. The inclusion of these variables does not change any of our main results on
the CSR measures (see Appendix Table A.4). In terms of the impact of these variables
on crash risk, we find evidence for the accentuating influence of entrenchment provisions
and total institutional ownership on future crashes, as shown in prior literature (An and
Zhang, 2013; Andreou et al., 2016). Conversely, we find marginal evidence for the miti-
gating influence of CEO’s age (Andreou et al., 2017) and ownership concentration or the
number of blockholders (Boubaker et al., 2014) on crashes. We also test other controls
such as tax avoidance (Kim et al., 2011b) and CFO’s compensation (Kim et al., 2011a)
and find that our results are unaffected.

5.2. Alternative CSR Measures

We employ two alternative measures for CSR (and its dimensions/subcategories).


First, we consider the raw CSR scores (i.e., a simple difference between overall CSR
strengths and concerns), unlike the industry-adjusted measures employed in the main re-
sults. Second, following Deng et al. (2013), we construct an equal-weighted measure that
scales each of the seven MSCI subcategories based on the number of available indicators
in each year. Appendix Table A.5 replicates our main results from Table 3 using the
ESG measures constructed with these two methodologies. We find that the coefficients
of these alternate ESG constructs remain qualitatively similar for both the overall CSR
performance (CSP ) and the social CSR dimension (SOC).

22

Electronic copy available at: https://ssrn.com/abstract=3305213


5.3. CSR Strengths and Concerns

In all our main analyses, we employ the aggregate CSR, each of the ESG dimensions,
and their respective categories. All these measures are calculated as the net score of a
difference between the MSCI strengths and concerns (see Table 1). The objective is to
understand how CSR, its ESG dimensions, and its constituent stakeholder groups affect
crash risk collectively and not in terms of firms’ CSR investments (positives/strengths)
or controversies (negatives/concerns). However, Krüger (2015) show that investors react
asymmetrically to the news of positive and negative CSR events. Likewise, Fernando
et al. (2019) study environmental policies and show that there is a “sharp asymmetry
between corporate policies that affect the firm’s exposure to environmental risk [i.e.,
concerns] and its perceived environmental friendliness [strengths].” Thus, to determine if
similar nonlinearity and asymmetry drive our main results, we consider the strengths and
concerns separately, and repeat the analysis in Table 3. Appendix Table A.6 summarizes
these results. We find that most of the mitigating effects of CSR and its social dimension
on crash risk are driven by CSR strengths, while CSR concerns or controversies do not
have any statistically significant effect on future crashes.

5.4. Panel Estimations

We further test the robustness of our results by using two different panel estimations.
First, we employ random effects panel Logit regression. For non-linear regressions, a fixed
effects model is biased by the incidental parameters problem, which is further aggravated
when there are limited observations for each group within the panel (Greene, 2004; Bar-
tolucci and Nigro, 2010). Specifically, in the case of CRASH, which is a binary variable,
most of the changes within a firm occur over a short two-year period. This amplifies the
aforementioned bias, resulting in inconsistent estimates (Wooldridge, 2005). Moreover,
the random effects specification is preferred over the more constrained firm fixed effects
one as several sample firms experience multiple or zero crashes in consecutive years, and
these are omitted when measuring the within-firm variation. Second, we run a dynamic
panel Logit regression by including one-year lagged CRASH as an additional control.
As per the suggestion by Wooldridge (2005), we also control for the unobserved hetero-

23

Electronic copy available at: https://ssrn.com/abstract=3305213


geneity in this specification by including the firms’ CRASH measures at the beginning of
our sample period (i.e., 1990), as additional regressors. This allows us to solve the initial
conditions problem in our non-linear Logit regressions. Thus, the dynamic panel Logit
of this type collectively corrects for the unobserved firm heterogeneity, reverse causality,
and state dependence in a single specification.
In Appendix Table A.7, we report the two panel-based variations of the results shown
in Table 3 for the dependent variable CRASH: (a) panel Logit regressions with random
effects and (b) dynamic panel Logit regressions. In both these static and dynamic panel
regressions, the coefficients of all the CSR variables remain robust and support our main
results. As expected, in the dynamic regressions, the estimates are more conservative
when the unobserved heterogeneity is controlled for in addition to the past stock price
crash. However, as in Table 3, the SOC dimension (Model 2) shows a significant mitigat-
ing effect on future crashes, which accounts for a large portion of the overall mitigating
power of the aggregate CSP (Model 1). The significant influences of the community,
employee, and product subcategories on crash risk are also retained (Model 3).

5.5. Instrumental Variables

Lastly, to test the robustness of our causal estimates in Section 4.3, we apply instru-
mental variables (IV) approach. Following El-Ghoul et al. (2011) and Kim et al. (2014),
we first employ the industry-average CSR scores (excluding the focal firm) with Fama-
French 48-industry classification as instruments. Next, following Jiraporn et al. (2014),
we include geographical-peer-average CSR scores along with the industry-averages as
instruments. In this case, geographical peers are identified as firms with the same three-
digit zip codes as the focal firm. The results for both these IV estimations are shown
in Appendix Table A.8, where Model 1 uses only the industry-average CSR scores as
IVs and Model 2 additionally instruments the geographical-peer-average CSR scores. We
find that all our causal inferences from the quasi-natural experiment are confirmed for
the social CSR dimensions as well as for many of its subcategories.

24

Electronic copy available at: https://ssrn.com/abstract=3305213


6. Conclusion

The association between CSR and financial performance has been widely debated in
recent years (e.g., Galema et al., 2008; Derwall et al., 2011; Borgers et al., 2013; Krüger,
2015). While the importance of CSR for firm valuation has been firmly established in
the literature (Ferrell et al., 2016; Byun and Oh, 2018; Bardos et al., 2020), the present
study is the first to show that CSR’s risk-mitigating role is crucially determined by
firm valuation. By studying stakeholder, crash risk, and firm valuation together, we
investigate whether and how CSR can mitigate or contribute to future stock price crash
risk, conditional on firms being undervalued. We find that satisfactory CSR engagement
by undervalued firms mitigates crash risk more than such engagement by overvalued
firms. In other words, firm undervaluation has a significant moderating effect on the
relationship between CSR and future price crashes.
In prior literature, all ESG characteristics are usually combined into a single CSR
measure when assessing the importance of CSR for crash risk and valuation. In contrast,
this study also examines the influence of each of the ESG activities targeted at different
stakeholder groups on future price crashes.16 By analyzing the different stakeholder
groups separately, we are also able to identify the stakeholders that are critical for firm
valuation due to their risk-mitigating properties. Our analysis shows that the social
CSR dimension is the most important determinant of future crashes and managerial bad
news hoarding behavior. This result is confirmed in a natural experiment setting that
exploits the passage of state-level other constituency statutes. We further examine how
undervaluation (and its expected effects on future shareholder value) drives the social
CSR dimension’s effect on crash risk. We find that the moderation effect of valuation on
the CSR–crash risk relationship can be completely explained by the social CSR dimension.
We further explore the different stakeholder groups within the social dimension and
find that CSR categories do not have a homogenous effect on mitigating crash risk.
The social subcategories that target specific stakeholder groups, such as the community,

16
Very few studies have explored the impact of individual CSR dimensions on risk, and, specifically,
on crash risk. Ben-Nasr and Ghouma (2018), for example, show that superior employee welfare practices
lead to higher stock price crash risks.

25

Electronic copy available at: https://ssrn.com/abstract=3305213


employees, and customers (i.e., the product category), mitigate future crashes, whereas
those targeting the society at large either increase the crash likelihood (human rights)
or do not have a bearing on it (diversity). We show that when managers care for the
community, employees, and customers, they are more likely to be perceptive of investors’
needs as well; this helps the managers to abstain from bad news hoarding. Investors, in
turn, associate positive value with the CSR activities that target these stakeholder groups
(Byun and Oh, 2018). In this regard, our results show that investors may be more prudent
in their interpretation of these activities, hence limiting the possibility of managerial bad
news hoarding. Additionally, we find that there is a causal relationship for most of the
CSR categories within the social ESG dimension for which an association with crash risk
exists. Finally, our supplementary tests support the view that the social CSR dimension
is compatible with low agency costs and good corporate governance. This may be driving
firms to maintaining better accounting transparency, thereby lowering the likelihood of
future crashes. Social CSR initiatives targeted at specific stakeholder groups benefit
investors as the latter can effectively scrutinize the associated costs and benefits, thereby
reducing the likelihood of managers’ using these initiatives as a means of concealing bad
news.

26

Electronic copy available at: https://ssrn.com/abstract=3305213


References
Adebambo, B. N., Yan, X., 2018. Investor overconfidence, firm valuation, and corporate
decisions. Management Science 64 (11), 5349–5369.

Al Mamun, M., Balachandran, B., Duong, H. N., 2020. Powerful CEOs and stock price
crash risk. Journal of Corporate Finance 62, Forthcoming.

An, H., Zhang, T., 2013. Stock price synchronicity, crash risk, and institutional investors.
Journal of Corporate Finance 21, 1–15.

Andreou, P. C., Antoniou, C., Horton, J., Louca, C., 2016. Corporate governance and
firm-specific stock price crashes. European Financial Management 22 (5), 916–956.

Andreou, P. C., Louca, C., Petrou, A. P., 2017. CEO age and stock price crash risk.
Review of Finance 21 (3), 1287–1325.

Bardos, K. S., Ertugrul, M., Gao, L. S., 2020. Corporate social responsibility, product
market perception, and firm value. Journal of Corporate Finance 62, Forthcoming.

Bartolucci, F., Nigro, V., 2010.


√ A dynamic model for binary panel data with unobserved
heterogeneity admitting a n-consistent conditional estimator. Econometrica 78 (2),
719–733.

Barzuza, M., 2009. The state of state antitakeover law. Virginia Law Review, 1973–2052.

Bebchuk, L., Cohen, A., Ferrell, A., 2009. What matters in corporate governance? Review
of Financial Studies 22 (2), 783–827.

Ben-Nasr, H., Ghouma, H., 2018. Employee welfare and stock price crash risk. Journal
of Corporate Finance 48, 700–725.

Bereskin, F., Byun, S. K., Officer, M. S., Oh, J.-M., 2018. The effect of cultural similarity
on mergers and acquisitions: Evidence from corporate social responsibility. Journal of
Financial and Quantitative Analysis 53 (5), 1995–2039.

Bhandari, A., Javakhadze, D., 2017. Corporate social responsibility and capital allocation
efficiency. Journal of Corporate Finance 43, 354–377.

Borgers, A., Derwall, J., Koedijk, K., Ter Horst, J., 2013. Stakeholder relations and
stock returns: On errors in investors’ expectations and learning. Journal of Empirical
Finance 22, 159–175.

Boubaker, S., Mansali, H., Rjiba, H., 2014. Large controlling shareholders and stock price
synchronicity. Journal of Banking and Finance 40, 80–96.

Bouslah, K., Kryzanowski, L., M’Zali, B., 2013. The impact of the dimensions of social
performance on firm risk. Journal of Banking and Finance 37 (4), 1258–1273.

Buchanan, B., Cao, C. X., Chen, C., 2018. Corporate social responsibility, firm value,
and influential institutional ownership. Journal of Corporate Finance 52, 73–95.

27

Electronic copy available at: https://ssrn.com/abstract=3305213


Byun, S. K., Oh, J.-M., 2018. Local corporate social responsibility, media coverage, and
shareholder value. Journal of Banking and Finance 87, 68–86.

Chabi-Yo, F., Ruenzi, S., Weigert, F., 2018. Crash sensitivity and the cross section of
expected stock returns. Journal of Financial and Quantitative Analysis 53 (3), 1059–
1100.

Chang, X., Chen, Y., Zolotoy, L., 2017. Stock liquidity and stock price crash risk. Journal
of Financial and Quantitative Analysis 52 (4), 1605–1637.

Chen, J., Hong, H., Stein, J. C., 2001. Forecasting crashes: Trading volume, past returns,
and conditional skewness in stock prices. Journal of Financial Economics 61 (3), 345–
381.

Cheng, C., Li, X., Sun, Z., Xie, J., 2018. Corporate social responsibility and shareholder-
stakeholder conflict: Evidence from a quasi-natural experiment. Unpublished working
paper, http://ssrn.com/abstract=2914724.

Chi, J. D., Gupta, M., 2009. Overvaluation and earnings management. Journal of Banking
and Finance 33 (9), 1652–1663.

Cremers, M., Guernsey, S. B., Sepe, S. M., 2019. Stakeholder orientation and firm value.
Unpublished working paper, http://ssrn.com/abstract=3299889.

Deng, X., Hung, S., Qiao, Z., 2018. Mutual fund herding and stock price crashes. Journal
of Banking and Finance 94, 166–184.

Deng, X., Kang, J.-k., Low, B. S., 2013. Corporate social responsibility and stakeholder
value maximization: Evidence from mergers. Journal of Financial Economics 110 (1),
87–109.

Derwall, J., Koedijk, K., Ter Horst, J., 2011. A tale of values-driven and profit-seeking
social investors. Journal of Banking and Finance 35 (8), 2137–2147.

Dumitrescu, A., Zakriya, M., 2020. Governance, information flow, and stock returns.
Unpublished working paper, http://ssrn.com/abstract=3170881.

Eccles, R. G., Ioannou, I., Serafeim, G., 2014. The impact of corporate sustainability on
organizational processes and performance. Management Science 60 (11), 2835–2857.

El-Ghoul, S., Guedhami, O., Kwok, C. C., Mishra, D. R., 2011. Does corporate social
responsibility affect the cost of capital? Journal of Banking and Finance 35 (9), 2388–
2406.

Fama, E. F., French, K. R., 1997. Industry costs of equity. Journal of Financial Economics
43 (2), 153–193.

Fatemi, A., Fooladi, I., Tehranian, H., 2015. Valuation effects of corporate social respon-
sibility. Journal of Banking and Finance 59, 182–192.

Fernando, C. S., Sharfman, M. P., Uysal, V. B., 2017. Corporate environmental policy and

28

Electronic copy available at: https://ssrn.com/abstract=3305213


shareholder value: Following the smart money. Journal of Financial and Quantitative
Analysis 52 (5), 2023–2051.

Fernando, C. S., Uysal, V. B., Abeysekera, A. P., 2019. An investor perspective on the
black box of corporate social responsibility. Journal of Applied Corporate Finance
31 (2), 92–104.

Ferrell, A., Liang, H., Renneboog, L., 2016. Socially responsible firms. Journal of Finan-
cial Economics 122 (3), 585–606.

Flammer, C., Kacperczyk, A., 2016. The impact of stakeholder orientation on innovation:
Evidence from a natural experiment. Management Science 62 (7), 1982–2001.

Friedman, M., 1970. The social responsibility of business is to increase its profits. New
York Times Magazine: 13th September.

Galema, R., Plantinga, A., Scholtens, B., 2008. The stocks at stake: Return and risk in
socially responsible investment. Journal of Banking and Finance 32 (12), 2646–2654.

Garcia-Castro, R., Arino, M. A., Canela, M. A., 2010. Does social performance really
lead to financial performance? Accounting for endogeneity. Journal of Business Ethics
92 (1), 107–126.

Geczy, C., Jeffers, J. S., Musto, D. K., Tucker, A. M., 2015. Institutional investing when
shareholders are not supreme. Harvard Business Law Review 5, 73.

Girerd-Potin, I., Jimenez-Garcès, S., Louvet, P., 2014. Which dimensions of social re-
sponsibility concern financial investors? Journal of Business Ethics 121 (4), 559–576.

Greene, W., 2004. The behaviour of the maximum likelihood estimator of limited de-
pendent variable models in the presence of fixed effects. Econometrics Journal 7 (1),
98–119.

Habib, A., Hasan, M. M., Jiang, H., 2018. Stock price crash risk: Review of the empirical
literature. Accounting and Finance 58, 211–251.

Harjoto, M. A., Jo, H., 2011. Corporate governance and CSR nexus. Journal of Business
Ethics 100 (1), 45–67.

Harrison, J. S., Bosse, D. A., Phillips, R. A., 2010. Managing for stakeholders, stakeholder
utility functions, and competitive advantage. Strategic Management Journal 31 (1),
58–74.

Harrison, J. S., Phillips, R. A., Freeman, R. E., 2020. On the 2019 Business Roundtable
“Statement on the Purpose of a Corporation”. Journal of Management 46 (7), 1223–
1237.

Healy, P. M., Palepu, K. G., 2001. Information asymmetry, corporate disclosure, and the
capital markets: A review of the empirical disclosure literature. Journal of Accounting
and Economics 31 (1-3), 405–440.

29

Electronic copy available at: https://ssrn.com/abstract=3305213


Hemingway, C. A., Maclagan, P. W., 2004. Managers’ personal values as drivers of cor-
porate social responsibility. Journal of Business Ethics 50 (1), 33–44.

Hutton, A. P., Marcus, A. J., Tehranian, H., 2009. Opaque financial reports, R2 , and
crash risk. Journal of Financial Economics 94 (1), 67–86.

Jin, L., Myers, S., 2006. R2 around the world: New theory and new tests. Journal of
Financial Economics 79 (2), 257–292.

Jiraporn, P., Jiraporn, N., Boeprasert, A., Chang, K., 2014. Does corporate social re-
sponsibility (CSR) improve credit ratings? Evidence from geographic identification.
Financial Management 43 (3), 505–531.

Kim, J.-B., Li, Y., Zhang, L., 2011a. CFOs versus CEOs: Equity incentives and crashes.
Journal of Financial Economics 101 (3), 713–730.

Kim, J.-B., Li, Y., Zhang, L., 2011b. Corporate tax avoidance and stock price crash risk:
Firm-level analysis. Journal of Financial Economics 100 (3), 639–662.

Kim, Y., Li, H., Li, S., 2014. Corporate social responsibility and stock price crash risk.
Journal of Banking and Finance 43, 1–13.

Krüger, P., 2015. Corporate goodness and shareholder wealth. Journal of Financial Eco-
nomics 115 (2), 304–329.

Lee, S. M., Jiraporn, P., Song, H., 2020. Customer concentration and stock price crash
risk. Journal of Business Research 110, 327–346.

Li, Y., Zeng, Y., 2019. The impact of top executive gender on asset prices: Evidence from
stock price crash risk. Journal of Corporate Finance 58, 528–550.

Lins, K. V., Servaes, H., Tamayo, A., 2017. Social capital, trust, and firm performance:
The value of corporate social responsibility during the financial crisis. Journal of Fi-
nance 72 (4), 1785–1824.

Litzenberger, R. H., Budd, A. P., 1970. Corporate investment criteria and the valuation
of risk assets. Journal of Financial and Quantitative Analysis, 395–419.

Marhfor, A., Bouslah, K., M’Zali, B., 2017. Corporate social responsibility and stock price
informativeness: Evidence from the Canadian market. Unpublished working paper,
https://www.st-andrews.ac.uk/business/rbf/workingpapers/RBF17_011.pdf.

Ni, X., 2020. Does stakeholder orientation matter for earnings management: Evidence
from non-shareholder constituency statutes. Journal of Corporate Finance 62, Forth-
coming.

Piotroski, J. D., Roulstone, D. T., 2004. The influence of analysts, institutional investors,
and insiders on the incorporation of market, industry, and firm-specific information into
stock prices. Accounting Review 79 (4), 1119–1151.

Rhodes-Kropf, M., Robinson, D. T., Viswanathan, S., 2005. Valuation waves and merger

30

Electronic copy available at: https://ssrn.com/abstract=3305213


activity: The empirical evidence. Journal of Financial Economics 77 (3), 561–603.

Sawicki, J., Shrestha, K., 2014. Misvaluation and insider trading incentives for accrual-
based and real earnings management. Journal of Business Finance & Accounting 41 (7-
8), 926–949.

Servaes, H., Tamayo, A., 2013. The impact of corporate social responsibility on firm
value: The role of customer awareness. Management Science 59 (5), 1045–1061.

Sharfman, M. P., Fernando, C. S., 2008. Environmental risk management and the cost of
capital. Strategic Management Journal 29 (6), 569–592.

Strine, L. E., 2014. Making it easier for directors to do the right thing. Harvard Business
Law Review 4, 235.

Utz, S., 2017. Over-investment or risk mitigation? Corporate social responsibility in


Asia-Pacific, Europe, Japan, and the United States. Review of Financial Economics,
https://doi.org/10.1016/j.rfe.2017.10.001.

Van Beurden, P., Gössling, T., 2008. The worth of values – A literature review on the
relation between corporate social and financial performance. Journal of Business Ethics
82 (2), 407–424.

Wooldridge, J. M., 2005. Simple solutions to the initial conditions problem in dynamic,
nonlinear panel data models with unobserved heterogeneity. Journal of Applied Econo-
metrics 20 (1), 39–54.

31

Electronic copy available at: https://ssrn.com/abstract=3305213


Table 1: Variable Definitions
Variable Notation Definition
Overall CSR Performance CSP (CSR − M inCSR)/(M axCSR − M inCSR), where CSR is the total ESG
strengths minus the concerns for a year. MinCSR and maxCSR are the re-
spective minimum and maximum CSR score for each of the corresponding
industry classified using Fama and French’s (1997) classification.
Environmental Performance EN V Similar standardized score as CSP , but only using environmental strengths
and concerns from within the ESG.
Social Performance SOC The standardized social score using the same formula as CSP , but only in-
cluding the Community, Diversity, Employee Relations, Human Rights and
Product indicator sets.
Corporate Governance GOV Similar to EN V , but using governance strengths and concerns from within
the ESG, in place of environmental ones.
Community/Diversity/ COM/DIV / Similar to standardized SOC score but using the respective subset of CSR
Employee Relations/ EM P/HU M/ category from within the social characteristics.
Human Rights/Product P RO
Price Crash CRASH A dummy variable that takes value 1 if the firm undergoes at least one crash
week in a given year. The crash weeks are identified when firm-specific weekly
return (F SW R, log of one plus the residuals from extended market model)
drops below 5% of average annual F SW R.
Number of Price Crashes CRASHN U MThe number of crash weeks annually experienced by a firm. Crash week is
defined in the same manner as for measuring CRASH.
Tobin’s Q TQ The ratio of market value (MV) of assets and book value (BV) of assets
(Compustat data item 6). MV of assets is found as: (BV of assets + MV of
common stock) – (BV of common stock + deferred taxes). T Q is industry-
adjusted using Fama and French’s (1997) classification.
Misvaluation Proxy RRV Firm-specific valution error as defined in Rhodes-Kropf et al. (2005). See
Section 3.2. for more details.
Differential Turnover DIF T U RN The average weekly share turnover for current year minus the same for the
previous year, where the weekly turnover is computed as the the sum total of
a week’s trading volume divided by the total shares outstanding.
Average FSWR AV G The average firm specific weekly return F SW R for a firm in a given year.
Standard Deviation of FSWR SD The standard deviation of FSWR for a year.
Return on Assets ROA Operating income before depreciation (Compustat data item 13) / Total assets
(Compustat data item 6)
Leverage LEV The ratio of long term debt (Compustat data item 9) to total assets (Compu-
stat data item 6).
Firm Size SIZE The logarithmic transformation of market value of equity.
Market-to-Book Ratio MB The log of market value of common equity divided by its book value. Book
value (BV) is the sum of common equity BV and deferred taxes (Compustat
items 60 and 74).
Opacity OP Q Annual opacity as defined in Hutton et al. (2009) using a modified Jones
model, taking the absolute value of one year’s discretionary accruals.
Book-Tax Difference BT D Following Kim et al. (2011b), it is the difference between book income and
taxable income standardized using the last year’s total assets.
Analysts’ Forecasts N REV It is the number of analysts’ earnings forecasts + revisions for a firm in a given
year (Piotroski and Roulstone, 2004)
Negative Conditional N CSKEW Following Chen et al. (2001), the normalized negative third moment of F SW R
Skewness in a given year.
Down-to-Up Volatility DU V OL Following Chen et al. (2001), it is the logarithmic transformation of the ratio
of SDs of down weeks to SDs of up weeks using FSWRs in a given year.
Managerial Entrenchment EIN DEX Bebchuk et al.’s (2009) E-Index which is the sum of the presence of six anti-
takeover provisions (staggered boards, poison pills, golden parachutes, limits
to bylaws, limits to charters, and supermajority for mergers) in the firm in a
given year.
Institutional Ownership IO This is the percentage of shares owned by institutional investors, computed
as the number of shares for a firm reported in quarterly 13f filings (for insti-
tutional holdings) divided by the total number of outstanding shares. Annual
IO computed as an average of quarterly IO.
Blockholding BHN The number of blockholders in a firm, with the blockholders defined as the
institutional investors that have a minimum of 5% ownership in the firm.
Quarterly data averaged over the year.
Earnings Management T P CY The previous two-year moving average of absolute values of discretionary ac-
cruals as defined in OP Q. Using OP Q from multiple years helps in identifying
consistent earnings management policies (Hutton et al., 2009).
CEO’s Age CEOAGE In years, the firm’s CEO’s age.
CEO’s Compensation CEOCOM P In logs, the sum total of CEO’s compensation, which includes annual salary,
bonuses, stocks/options granted, any other payments and long term incentives.
The option value is computed using Black and Scholes formula.

32

Electronic copy available at: https://ssrn.com/abstract=3305213


Table 2: Descriptive Statistics
This table shows the summary statistics for the two main stock price crash risk measures, the aggregate
CSR variable (CSP ) –its three dimensions (EN V , SOC and GOV ) and five social subcategories (COM ,
DIV , EM P , HU M and P RO)– and all the control variables.
For variable definitions and details, see Table 1.
N Mean Median SD 5th 95th
Main Crash Risk Measures
CRASH 29398 0.26 0.00 0.44 0.00 1.00
CRASHNUM 29398 0.27 0.00 0.47 0.00 1.00
CSR Measures
CSP 35788 0.39 0.36 0.25 0.00 0.89
ENV 35788 0.37 0.33 0.31 0.00 1.00
SOC 35788 0.37 0.33 0.24 0.00 0.86
CG 35788 0.56 0.50 0.30 0.00 1.00
COM 35788 0.29 0.25 0.31 0.00 1.00
DIV 35788 0.31 0.25 0.29 0.00 1.00
EMP 35788 0.41 0.40 0.27 0.00 1.00
HUM 35788 0.52 0.50 0.44 0.00 1.00
PRO 35788 0.63 0.67 0.29 0.00 1.00
Undervaluation Measures
TQ 35649 -0.04 -0.18 6.02 -1.69 2.35
RRV 35164 0.23 0.17 0.89 -1.03 1.78
Main Controls
DIFTURN 28342 0.68 0.43 26.10 -29.50 32.12
AVG 29398 -0.18 -0.16 0.78 -1.34 0.92
SD 29390 0.05 0.04 0.03 0.02 0.10
LEV 35788 0.19 0.14 0.21 0.00 0.57
SIZE 35788 7.37 7.23 1.60 5.06 10.24
MB 35490 4.77 4.07 2.90 0.87 9.92
ROA 35788 0.11 0.11 1.58 -0.08 0.28
OPQ 31022 0.11 0.06 0.28 0.00 0.33
BTD 34885 0.18 0.01 1.88 -0.13 0.82
NREV 35788 39.06 25.00 46.67 0.00 133.00
NCSKEW 29390 0.14 0.10 1.05 -1.49 1.98
Additional Controls and/or Grouping Variables for Subsampling
EINDEX 19663 3.10 3.00 1.44 0.00 5.00
IO 20233 0.76 0.77 1.91 0.35 1.00
BHN 20236 2.56 2.50 1.61 0.00 5.50
TPCY 35859 0.16 0.10 0.40 0.00 0.48
CEOAGE 19465 56.08 56.00 7.26 45.00 68.00
CEOCOMP 19816 8.13 8.19 1.14 6.48 9.76

33

Electronic copy available at: https://ssrn.com/abstract=3305213


Table 3: CSR and Crash Risk Predictability
This table summarizes the Logit (for CRASH) and Tobit (for CRASHN U M ) regressions for CSR
measures on stock price crash risk. Model 1 reports the coefficients when aggregate CSR measure
(CSP ) is employed. Models 2 and 3 give the results for the three ESG dimensions and the seven broad
CSR categories respectively. Standard errors clustered two-dimensionally for both the firm and year, are
given in parentheses. All the variables are defined in Table 1. Significance levels at 10%, 5%, and 1%
are shown using *, ** and *** respectively.
CRASH (Logit) CRASHN U M (Tobit)
(1) (2) (3) (1) (2) (3)
CSPt−1 -0.1300** -0.1032**
(0.065) (0.049)
ENVt−1 0.0706 0.0333 0.0478 0.0192
(0.057) (0.049) (0.042) (0.037)
SOCt−1 -0.1350** -0.0992**
(0.067) (0.050)
GOVt−1 0.0273 -0.0072 0.0133 -0.0127
(0.051) (0.049) (0.039) (0.037)
COMt−1 -0.1475*** -0.1116***
(0.049) (0.037)
DIVt−1 -0.0024 -0.0041
(0.052) (0.040)
EMPt−1 -0.1199** -0.0901**
(0.057) (0.043)
HUMt−1 0.1014*** 0.0743***
(0.036) (0.027)
PROt−1 -0.1309*** -0.1000***
(0.050) (0.038)
DIFTURNt−1 0.0066*** 0.0066*** 0.0065*** 0.0047*** 0.0047*** 0.0047***
(0.001) (0.001) (0.001) (0.001) (0.001) (0.001)
AVGt−1 0.0937*** 0.0927*** 0.0984*** 0.0697*** 0.0690*** 0.0764***
(0.026) (0.026) (0.026) (0.019) (0.020) (0.020)
SDt−1 3.6121*** 3.6505*** 5.2969*** 2.8676*** 2.8946*** 4.1659***
(0.725) (0.726) (0.685) (0.538) (0.539) (0.514)
LEVt−1 -0.0037 -0.0017 -0.2127*** 0.0037 0.0048 -0.1548***
(0.076) (0.076) (0.071) (0.057) (0.057) (0.054)
SIZEt−1 0.0186 0.0203 0.0323** 0.0146 0.0155 0.0242**
(0.014) (0.014) (0.013) (0.010) (0.011) (0.010)
MBt−1 0.0189*** 0.0188*** 0.0391*** 0.0143*** 0.0142*** 0.0299***
(0.006) (0.006) (0.005) (0.004) (0.004) (0.004)
ROAt−1 0.4317*** 0.4300*** 0.7117*** 0.3189*** 0.3176*** 0.5366***
(0.108) (0.108) (0.109) (0.080) (0.080) (0.082)
OPQt−1 -0.0377 -0.0351 0.2606** -0.0199 -0.0190 0.2191**
(0.137) (0.137) (0.131) (0.104) (0.104) (0.098)
BTDt−1 -0.0051 -0.0045 0.0187 -0.0036 -0.0032 0.0150
(0.016) (0.016) (0.016) (0.012) (0.012) (0.012)
NREVt−1 0.0003 0.0003 -0.0003 0.0002 0.0002 -0.0002
(0.000) (0.000) (0.000) (0.000) (0.000) (0.000)
NCSKEWt−1 0.0629*** 0.0628*** 0.0791*** 0.0453*** 0.0452*** 0.0594***
(0.017) (0.017) (0.017) (0.012) (0.012) (0.013)
Year FEs Yes Yes Yes Yes Yes Yes
Industry FEs Yes Yes Yes Yes Yes Yes
# Observations 27360 27360 27360 27360 27360 27360
Pseudo R-squared 0.0315 0.0315 0.0291 0.0231 0.0231 0.0241

34

Electronic copy available at: https://ssrn.com/abstract=3305213


Table 4: Marginal Effects for Crash Predictability
The table presents marginal effects from Logit regressions shown in Table 3 for the variable CRASH.
These marginal effects represent the ability of different CSR measures to predict one-year forward stock
price crash. Variable definitions are available in Table 1. For details on the Logit regressions for all the
three models, see Table 3. All controls are excluded for brevity. Standard errors for each of the marginal
coefficients are shown in parenthesis, and ***, **, and * show statistical significance at the 1%, 5%, and
10% levels respectively.
(1) (2) (3)
CSPt−1 -0.0246**
(0.012)
ENVt−1 0.0134 0.0064
(0.011) (0.009)
SOCt−1 -0.0255**
(0.013)
GOVt−1 0.0052 -0.0014
(0.010) (0.009)
COMt−1 -0.0283***
(0.009)
DIVt−1 -0.0005
(0.010)
EMPt−1 -0.0230**
(0.011)
HUMt−1 0.0195***
(0.007)
PROt−1 -0.0251***
(0.010)
# Observations 27360 27360 27360

35

Electronic copy available at: https://ssrn.com/abstract=3305213


Table 5: CSR, Firm Valuation and Crash Risk Predictability
For each of the two undervaluation proxies using T Q and RRV , this table summarizes the Logit (for CRASH) and Tobit (for CRASHN U M ) regressions for
CSR measures and its interaction with firm misvaluation dummy (U ndervalued) on stock price crash risk. Model 1 reports the coefficients when aggregate CSR
measure (CSP ) is employed. Models 2 and 3 give the results for the three ESG dimensions and the seven broad CSR categories respectively. Standard errors
clustered two-dimensionally for both the firm and year, are given in parentheses. All the variables are defined in Table 1. Using *, ** and *** respectively,
significance levels at 10%, 5%, and 1% are shown.
Undervalued = High Tobin’s Q (T Q) High Misvaluation Proxy (RRV )
CRASH CRASHNUM CRASH CRASHNUM
(1) (2) (3) (1) (2) (3) (1) (2) (3) (1) (2) (3)
CSPt−1 0.003 -0.006 0.010 -0.001
(0.087) (0.065) (0.102) (0.076)
Undervalued 0.448*** 0.410*** 0.369*** 0.334*** 0.305*** 0.266*** -0.186*** -0.173** -0.200** -0.139*** -0.130** -0.154**
(0.056) (0.079) (0.100) (0.042) (0.059) (0.075) (0.055) (0.081) (0.101) (0.041) (0.060) (0.076)
CSPt−1 * Undervalued -0.252** -0.180** -0.205* -0.148
(0.117) (0.087) (0.122) (0.091)
ENVt−1 0.011 -0.005 -0.004 -0.015 0.115 0.006 0.088 0.004
(0.075) (0.076) (0.055) (0.056) (0.081) (0.076) (0.061) (0.058)
ENVt−1 * Undervalued 0.137 0.139 0.120* 0.120 -0.087 0.008 -0.076 -0.003
(0.096) (0.099) (0.072) (0.074) (0.095) (0.097) (0.071) (0.074)
SOCt−1 0.047 0.039 0.038 0.032
(0.090) (0.067) (0.105) (0.078)
SOCt−1 * Undervalued -0.384*** -0.290*** -0.253** -0.189**
(0.123) (0.091) (0.129) (0.096)

36
GOVt−1 0.015 0.016 0.006 0.007 -0.004 -0.059 -0.017 -0.055
(0.070) (0.070) (0.052) (0.053) (0.078) (0.077) (0.058) (0.058)
GOVt−1 * Undervalued 0.062 0.047 0.043 0.032 0.054 0.078 0.051 0.063
(0.095) (0.096) (0.071) (0.072) (0.097) (0.097) (0.073) (0.074)
COMt−1 0.045 0.041 -0.060 -0.053
(0.071) (0.053) (0.079) (0.059)
COMt−1 * Undervalued -0.202** -0.157** -0.135 -0.089
(0.095) (0.071) (0.099) (0.074)
DIVt−1 -0.033 -0.032 0.051 0.033
(0.074) (0.055) (0.082) (0.062)
DIVt−1 * Undervalued -0.033 -0.024 -0.085 -0.057
(0.100) (0.074) (0.103) (0.077)
EMPt−1 0.030 0.017 -0.154* -0.108
(0.081) (0.060) (0.088) (0.066)
EMPt−1 * Undervalued -0.211** -0.145* -0.233*** -0.179***
(0.107) (0.080) (0.067) (0.051)
HUMt−1 0.071 0.056 0.223*** 0.168***
(0.054) (0.041) (0.053) (0.040)

Electronic copy available at: https://ssrn.com/abstract=3305213


HUMt−1 * Undervalued 0.032 0.018 0.065 0.037
(0.066) (0.049) (0.109) (0.082)
PROt−1 -0.011 -0.014 -0.178** -0.140**
(0.075) (0.056) (0.080) (0.061)
PROt−1 * Undervalued 0.077 0.070 0.084 0.071
(0.102) (0.076) (0.101) (0.077)
Controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Industry + Year FEs Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
# Observations 27357 27357 27357 27357 27357 27357 27069 27069 27069 27069 27069 27069
Pseudo R-squared 0.0353 0.0355 0.0357 0.0258 0.0260 0.0261 0.0337 0.0339 0.0316 0.0247 0.0248 0.0258
Table 6: Triple Difference Estimates for CSR and Crash Risk Predictability
This table reports the Diff-in-Diff-in-Diff (DDD) estimation results for the causal effects of CSR measures
on future stock price crash risk. The dependent variable is CRASH, and the main regressors and controls
for each of the models 1 to 3 are the same as those in Table 3 (in a Logit regression). The experimental
period is from 1998 to 2001, with post indicating years after other constituency statutes are passed in
Maryland (i.e. 2000 and 2001) and treat representing Maryland-based firms with the Delaware-based
ones as control group. Baseline DDD results are in column 1. Column 2 corrects for possible selection
biases in the baseline results by using propensity score (PS) matched treatment and control groups. PS
matching is done using SIZE, ROA and LEV , to find a comparable Delaware firm for every Maryland
firm (0.001 calliper of nearest neighbor) in a given year. Column 3 shows the validation test for baseline
DDD, by assuming a placebo experimental period i.e., 2008 to 2011 such that post dummy is 1 for the
years 2010 and 2011. Only the main effects and the triple interaction effects are reported for brevity.
Significance levels at 10%, 5%, and 1% respectively are indicated by *,**, and ***.
Baseline DDD PS Matched DDD Placebo DDD
Panel A: Aggregate CSR Performance
post 0.4929 1.2429 -0.4897***
(0.489) (1.493) (0.177)
treat 1.0452 1.2784 -0.2646
(0.822) (1.260) (0.535)
CSPt−1 -0.1662 1.4052 -0.4856**
(0.518) (1.420) (0.229)
post * treat * CSPt−1 1.6255 5.2630 -1.1433
(4.578) (5.404) (1.527)
Controls + Year FEs Yes Yes Yes
# Observations 691 90 4843
Pseudo R-squared 0.048 0.173 0.018
Panel B: ESG Dimensions of CSR
post 0.3366 0.8832 -0.2454
(0.564) (1.895) (0.255)
treat 0.9889 2.5402** -0.6177
(0.777) (1.224) (0.793)
ENVt−1 -0.0185 1.3730 0.3393*
(0.477) (1.693) (0.180)
post * treat * ENVt−1 -7.4373 -12.0851* -1.3968
(5.216) (6.574) (1.183)
SOCt−1 -0.1498 2.1868 -0.7449***
(0.507) (1.569) (0.243)
post * treat * SOCt−1 13.0024** 21.1229** 0.0010
(6.044) (8.387) (1.623)
GOVt−1 -0.2638 -4.5924* 0.0406
(0.456) (2.470) (0.197)
post * treat * GOVt−1 1.3789 3.9229 -0.6349
(5.343) (6.743) (1.373)
Controls + Year FEs Yes Yes Yes
# Observations 691 90 4843
Pseudo R-squared 0.051 0.239 0.021
Panel C: Subcategories of SOC
post 0.0663 0.2724 -0.1388***
(0.093) (0.317) (0.051)
treat 0.4535** 0.4171* -0.1249
(0.182) (0.212) (0.147)
ENVt−1 0.0430 0.2532 0.0037
(0.044) (0.184) (0.022)
COMt−1 -0.0269 0.1417 -0.0268
(0.053) (0.226) (0.039)
post * treat * COMt−1 -0.7306** -0.7596** 0.0573
(0.314) (0.443) (0.327)
DIVt−1 0.0404 0.3496 -0.0285
(0.067) (0.324) (0.041)
post * treat * DIVt−1 -0.3312* -0.8648 -0.0044
(0.188) (0.577) (0.186)
EMPt−1 -0.1139 0.2997 -0.0449
(0.070) (0.251) (0.042)
post * treat * EMPt−1 -0.5965 -0.2326 -0.4686
(0.789) (1.346) (0.289)
HUMt−1 -0.0191 -0.3610* 0.0009
(0.049) (0.183) (0.024)
post * treat * HUMt−1 2.7529*** 2.8661** -0.1873
(0.982) (1.172) (0.229)
PROt−1 -0.0246 -0.3770 -0.0649*
(0.064) (0.298) (0.038)
post * treat * PROt−1 -2.0066** -2.3146* 0.2436
(0.898) (1.187) (0.297)
GOVt−1 -0.0308 0.0372 0.0040
(0.045) (0.207) (0.025)
Controls + Year FEs Yes Yes Yes
# Observations 691 90 4843
Adj R-squared 0.056 0.391 0.021
37

Electronic copy available at: https://ssrn.com/abstract=3305213


Table 7: Summary Statistics for Unmatched and PS Matched Samples
In this table, we report the means of all the main firm characteristics important for crash risk (i.e.,
the control variables) for both the treatment (treat = 1) and control (treat = 0) groups. To compare
the two means along, we report their differences along with the significance level for two sample t-tests
represented using *,**, and *** for 10%, 5%, and 1% respectively. For definitions of each variable, see
Table 1. First set of the means summarizes the variables for the full experimental sample (Unmatched)
starting from 1998 and ending in 2001 (see section 4.3.3). The second set of means consider propensity
score matched (PS Matched) sample. The PS matching employs three fundamentally important firm
characteristics i.e., size (log of assets), profitability (return on assets) and leverage (debt to asset ratio).
For the mean differences, significance levels at 10%, 5%, and 1% are represented using *,**, and ***
respectively.
Unmatched Sample PS Matched Sample
treat=0 treat=1 difference treat=0 treat=1 difference
DIFTURN 3.183 1.846 1.337* 5.290 2.920 2.370*
AVG -0.198 -0.023 -0.175 -0.182 -0.216 -0.034
SD 0.064 0.048 0.016*** 0.065 0.037 0.028**
LEV 0.191 0.229 -0.039* 0.211 0.236 -0.025
SIZE 8.512 7.71 0.802*** 8.486 8.194 0.293
MB 4.594 4.771 -0.177 4.615 5.530 -0.915
ROA 0.135 0.087 0.048*** 0.154 0.172 -0.018
OPQ 0.065 0.016 0.049** 0.060 0.000 0.060
BTD -0.013 -0.003 -0.01 -0.018 -0.016 -0.001
NREV 46.410 35.649 10.761* 44.854 42.000 4.854
NCSKEW 0.163 0.084 0.079 0.145 -0.332 0.477

38

Electronic copy available at: https://ssrn.com/abstract=3305213


Table 8: Why Social Dimension Matters for Crash Risk Predictability?
This table reports the Logit regression results when subsamples (grouped on various agency cost
proxies) are used to study the effects of CSR measures on future stock price crash risk. The
dependent variable is CRASH, and the main regressors and controls are the same as those shown
in Table 3. Agency costs are proxied using managerial entrenchment (E-Index), institutional
ownership (IO - in percentage), blockholding (numbers) and earnings management (accounting
transparency/opacity). For each of these four proxies, our sample is subdivided using median
values for each year to get high/low classification, and corresponding grouping variable (lagged)
is included as an additional control. Only the coefficients of CSR measures (Panel A: Aggregate
CSR, Panel B: ESG Dimensions, and Panel C: SOC subcategories) are reported along with
two-dimensionally clustered standard errors (at firm and year levels, given in parenthesis) for
the sake of brevity. Significance levels at 10%, 5%, and 1% are shown with *,**, and ***
respectively.
Panel A: Aggregate CSR Performance
Agency Cost Proxy = Managerial Institutional Blockholding Earnings
Entrenchment Ownership Management
(Grouping Variable) (EIN DEX) (IO) (BHN ) (T P CY )
LOW HIGH LOW HIGH LOW HIGH LOW HIGH
CSPt−1 -0.2548** -0.0714 -0.1179 -0.1328* -0.0733 -0.1665* -0.0611 -0.2210**
(0.105) (0.134) (0.113) (0.008) (0.108) (0.100) (0.087) (0.111)
Controls Yes Yes Yes Yes Yes Yes Yes Yes
(Grouping Variable)t−1 Yes Yes Yes Yes Yes Yes Yes Yes
Industry + Year FEs Yes Yes Yes Yes Yes Yes Yes Yes
# Observations 9123 6143 8661 10566 8833 10397 13854 10691
Pseudo R-squared 0.0367 0.0364 0.0359 0.0363 0.0311 0.0353 0.0312 0.0393

Panel B: ESG Dimensions of CSR


Agency Cost Proxy = Managerial Institutional Blockholding Earnings
Entrenchment Ownership Management
(Grouping Variable) (EIN DEX) (IO) (BHN ) (T P CY )
LOW HIGH LOW HIGH LOW HIGH LOW HIGH
ENVt−1 -0.1414 0.0565 0.0341 0.0906 -0.0366 0.1218 -0.0571 0.1228
(0.096) (0.112) (0.098) (0.088) (0.095) (0.091) (0.076) (0.100)
SOCt−1 -0.1370* -0.1324 -0.1571 -0.1768* -0.0511 -0.2285** -0.0094 -0.3719***
(0.008) (0.137) (0.118) (0.103) (0.111) (0.107) (0.091) (0.110)
GOVt−1 -0.0864 0.1607 0.1027 0.0018 0.0581 0.0500 0.0399 0.0482
(0.087) (0.107) (0.093) (0.079) (0.089) (0.082) (0.071) (0.082)
Controls Yes Yes Yes Yes Yes Yes Yes Yes
(Grouping Variable)t−1 Yes Yes Yes Yes Yes Yes Yes Yes
Industry + Year FEs Yes Yes Yes Yes Yes Yes Yes Yes
# Observations 9123 6143 8661 10566 8833 10397 13854 10691
Pseudo R-squared 0.0368 0.0368 0.0361 0.0365 0.0311 0.0356 0.0312 0.0400

Panel C: Subcategories of SOC


Agency Cost Proxy = Managerial Institutional Blockholding Earnings
Entrenchment Ownership Management
(Grouping Variable) (EIN DEX) (IO) (BHN ) (T P CY )
LOW HIGH LOW HIGH LOW HIGH LOW HIGH
ENVt−1 -0.1617* 0.0351 0.0078 -0.0397 -0.0601 0.0346 0.1180 -0.0365
(0.089) (0.101) (0.098) (0.083) (0.095) (0.084) (0.104) (0.079)
COMt−1 -0.1825** -0.2374** -0.0474 -0.1977** -0.0084 -0.2404*** 0.0288 -0.1511**
(0.088) (0.106) (0.093) (0.078) (0.089) (0.080) (0.088) (0.076)
DIVt−1 -0.0305 0.0044 0.0037 -0.0468 0.0055 -0.0802 -0.0948 -0.0401
(0.092) (0.114) (0.098) (0.083) (0.094) (0.086) (0.091) (0.079)
EMPt−1 -0.0195 -0.0699 0.0667 -0.1180 0.0962 -0.1321 -0.1876* 0.0415
(0.100) (0.120) (0.104) (0.091) (0.100) (0.094) (0.100) (0.085)
HUMt−1 0.1163* -0.0070 0.1346* 0.0132 0.1434** 0.1019* -0.0940 0.2514***
(0.065) (0.077) (0.074) (0.057) (0.073) (0.058) (0.067) (0.064)
PROt−1 -0.2981*** -0.1026 -0.0932 -0.1027 -0.0446 -0.1576* 0.0508 -0.1358*
(0.088) (0.101) (0.098) (0.079) (0.094) (0.080) (0.092) (0.079)
GOVt−1 -0.0580 0.1567 0.0925 0.0624 0.0546 0.0542 0.0440 0.0902
(0.086) (0.106) (0.093) (0.078) (0.089) (0.080) (0.085) (0.074)
Controls Yes Yes Yes Yes Yes Yes Yes Yes
(Grouping Variable)t−1 Yes Yes Yes Yes Yes Yes Yes Yes
Industry + Year FEs Yes Yes Yes Yes Yes Yes Yes Yes
# Observations 9123 6143 8661 10566 8833 10397 12829 10088
Pseudo R-squared 0.0371 0.0371 0.0364 0.0365 0.0316 0.0356 0.0337 0.0399

39

Electronic copy available at: https://ssrn.com/abstract=3305213


Appendix: Supplementary Results

Table A.1: CSR and Crash Risk Predictability using N CSKEW and DU V OL
This table replicates the estimations shown in Table 3 using two alternate measures of crash risk i.e.,
N CSKEW and DU V OL in OLS regressions. For variable definitions, see Table 1 and Section 3. All
regressions employ control variables similar to Table 3, but are omitted for brevity. For each dependent
variable, we report the results separately for the Full Sample, and the two sub-periods 1995-2009 and
2010-2015 as indicated on top of each panel. The sub-period 1995-2009 replicates the results for Kim
et al.’s (2014) sample. Panel A shows results for aggregate CSR measure (CSP ). Panels B and C report
the results for the three ESG dimensions and the seven broad CSR categories respectively. For each
coefficient, standard errors clustered two-dimensionally on firms and years, are given in parentheses.
Significance levels at 10%, 5%, and 1% are indicated by *, ** and *** respectively.

Panel A: Aggregate CSR Performance


DV= N CSKEW DU V OL
Full Sample 1995-2009 2010-2015 Full Sample 1995-2009 2010-2015
CSPt−1 -0.0152 -0.0646* -0.0121 -0.0028 -0.0221 -0.0039
(0.028) (0.037) (0.050) (0.010) (0.014) (0.019)
Lagged DV + Controls Yes Yes Yes Yes Yes Yes
Industry + Year FEs Yes Yes Yes Yes Yes Yes
# Observations 27356 13496 13212 27356 13496 13212
Adj. R-squared 0.0213 0.0222 0.0270 0.0255 0.0313 0.0314

Panel B: ESG Dimensions of CSR


DV= N CSKEW DU V OL
Full Sample 1995-2009 2010-2015 Full Sample 1995-2009 2010-2015
ENVt−1 0.0219 0.0558* -0.0252 0.0057 0.0218* -0.0166
(0.024) (0.032) (0.042) (0.009) (0.012) (0.016)
SOCt−1 -0.0081 -0.0671* 0.0738 -0.0027 -0.0282** 0.0357**
(0.029) (0.037) (0.049) (0.011) (0.014) (0.018)
GOVt−1 0.0440* 0.0373 -0.0182 0.0139 0.0178 -0.0183
(0.023) (0.033) (0.037) (0.009) (0.012) (0.014)
Lagged DV + Controls Yes Yes Yes Yes Yes Yes
Industry + Year FEs Yes Yes Yes Yes Yes Yes
# Observations 27356 13496 13212 27356 13496 13212
Adj. R-squared 0.0214 0.0223 0.0270 0.0255 0.0316 0.0317

Panel B: Subcategories of SOC


DV= N CSKEW DU V OL
Full Sample 1995-2009 2010-2015 Full Sample 1995-2009 2010-2015
ENVt−1 0.0069 0.0528 -0.0269 -0.0011 0.0190 -0.0177
(0.024) (0.033) (0.042) (0.009) (0.012) (0.016)
COMt−1 0.0303 -0.0658** 0.0695* 0.0157* -0.0303** 0.0396***
(0.024) (0.033) (0.038) (0.009) (0.013) (0.014)
DIVt−1 -0.0523** -0.0470 -0.0477 -0.0334*** -0.0298** -0.0303**
(0.025) (0.036) (0.038) (0.009) (0.013) (0.014)
EMPt−1 0.0098 0.0075 0.0001 0.0132 0.0104 0.0085
(0.026) (0.035) (0.042) (0.010) (0.013) (0.016)
HUMt−1 0.0701*** 0.0080 0.1394*** 0.0296*** 0.0062 0.0586***
(0.019) (0.024) (0.035) (0.007) (0.009) (0.013)
PROt−1 0.0697*** 0.0230 0.1139*** 0.0293*** 0.0179 0.0445***
(0.024) (0.034) (0.040) (0.009) (0.013) (0.015)
GOVt−1 0.0358 0.0350 -0.0334 0.0104 0.0159 -0.0249*
(0.023) (0.033) (0.037) (0.009) (0.012) (0.014)
Lagged DV + Controls Yes Yes Yes Yes Yes Yes
Industry + Year FEs Yes Yes Yes Yes Yes Yes
# Observations 27356 13496 13212 27356 13496 13212
Adj. R-squared 0.0223 0.0223 0.0289 0.0271 0.0321 0.0347

40

Electronic copy available at: https://ssrn.com/abstract=3305213


Table A.2: CSR and Crash Risk Predictability Around the Crisis Years
This table replicates the estimations shown in Table 3 (for CRASH and CRASHN U M ) and Table A.1
(for N CSKEW and DU V OL) by introducing the interactions of Crisis and P ost − Crisis dummies
with the main independent variables (i.e. CSR and its three dimensions). The estimation period begins
three years before the crisis (2005) and ends in 2016. Crisis dummy represents years 2008 and 2009,
while P ost−Crisis dummy represents the years 2010 onwards. See Table 1 for other variable definitions.
All regressions employ control variables similar to Tables 3 and A.1. Panel A shows results for aggregate
CSR measure (CSP ), while Panel reports the results for the three ESG dimensions. For brevity, the
results for seven broad CSR categories are not reported in this table. The reported standard errors (in
parentheses) are clustered two-dimensionally for firms and years. Significance levels at 10%, 5%, and 1%
are indicated by *, ** and *** respectively.

Panel A: Aggregate CSR Performance


CRASH CRASHN U M N CSKEW DU V OL
Logit Tobit OLS OLS
Crisis -0.0273 -0.0279 -0.0457 -0.0529**
(0.136) (0.100) (0.065) (0.024)
Post-Crisis 0.3170*** 0.2399*** -0.2174*** -0.1356***
(0.111) (0.082) (0.059) (0.022)
CSPt−1 -0.1651 -0.1099 -0.0246 -0.0246
(0.181) (0.135) (0.085) (0.032)
CSPt−1 * Crisis -0.0203 -0.0155 -0.0817 -0.0068
(0.249) (0.184) (0.111) (0.041)
CSPt−1 * Post-Crisis 0.1980 0.1277 0.1005 0.0578
(0.202) (0.150) (0.095) (0.035)
Controls Yes Yes Yes Yes
Industry + Year FEs Yes Yes Yes Yes
# Observations 20260 20260 20260 20260
Pseudo/Adj. R-squared 0.0331 0.0239 0.0236 0.0286

Panel B: ESG Dimensions of CSR


CRASH CRASHN U M N CSKEW DU V OL
Logit Tobit OLS OLS
Crisis -0.0266 -0.0230 -0.0784 -0.0664**
(0.193) (0.144) (0.091) (0.033)
Post-Crisis 0.3611** 0.2753** -0.1894** -0.1185***
(0.155) (0.116) (0.079) (0.029)
ENVt−1 -0.0075 0.0036 -0.0224 -0.0234
(0.155) (0.117) (0.071) (0.027)
ENVt−1 * Crisis 0.1424 0.1053 0.0835 0.0467
(0.202) (0.152) (0.087) (0.033)
ENVt−1 * Post-Crisis 0.0468 0.0127 0.0615 0.0384
(0.173) (0.130) (0.079) (0.030)
SOCt−1 -0.2404 -0.1681 -0.0650 -0.0392
(0.182) (0.136) (0.085) (0.032)
SOCt−1 * Crisis 0.0436 0.0507 -0.0559 -0.0072
(0.252) (0.187) (0.113) (0.042)
SOCt−1 * Post-Crisis 0.1955 0.1407 0.1846* 0.0919***
(0.204) (0.152) (0.095) (0.035)
GOVt−1 0.1693 0.1237 0.1202* 0.0545**
(0.146) (0.111) (0.072) (0.027)
GOVt−1 * Crisis -0.1533 -0.1318 -0.0303 -0.0117
(0.205) (0.154) (0.093) (0.035)
GOVt−1 * Post-Crisis -0.0916 -0.0698 -0.1359* -0.0757**
(0.162) (0.123) (0.079) (0.030)
Controls Yes Yes Yes Yes
Industry + Year FEs Yes Yes Yes Yes
# Observations 20260 20260 20260 20260
Pseudo/Adj. R-squared 0.0333 0.0240 0.0295 0.0240

41

Electronic copy available at: https://ssrn.com/abstract=3305213


Table A.3: Additional Tests for Why Social Dimension Matters for Crash
Risk
This table reports the additional Logit regressions which extend the results shown in Table 7. We first
use firm fundamentals, namely the size and leverage (see Table 1), to subdivide the samples based on
the median values for each of these characteristics. Next, we analyze the subsamples by separating the
financial crisis 2007-2008 from the non-crisis years (since the dependent variable is future crash likelihood,
accordingly the crisis effect are defined for the years 2008 and 2009). Last, we subgroup the sample years
using pre- and post-learning years as defined in Borgers et al. (2013). All the variables are same as
those shown in Table 3. Only the coefficients of CSR measures (Panel A: Aggregate CSR, Panel B: ESG
Dimensions, and Panel C: SOC subcategories) are reported along with double clustered standard errors
(for firm and year groupings, given in parenthesis). Levels of significance at 10%, 5%, and 1% are shown
using *,**, and *** respectively.

Panel A: Aggregate CSR Performance


Grouping by: Fundamentals Sub-periods
(Variable) Size Leverage Financial Crisis Learning
LOW HIGH LOW HIGH Crisis Non-crisis Pre Post
CSPt−1 -0.2082* -0.2874*** -0.2636*** -0.2505*** -0.4149** -0.2355*** -0.3550** -0.2237***
(0.108) (0.084) (0.102) (0.085) (0.180) (0.070) (0.143) (0.073)
Controls Yes Yes Yes Yes Yes Yes Yes Yes
Industry + Year FEs Yes Yes Yes Yes Yes Yes Yes Yes
# Observations 13385 14523 13910 13998 3798 24110 3733 24175
Pseudo R-squared 0.0388 0.0161 0.0233 0.0195 0.0102 0.0231 0.024 0.0181

Panel B: ESG Dimensions of CSR


Grouping by: Fundamentals Sub-periods
(Variable) Size Leverage Financial Crisis Learning
LOW HIGH LOW HIGH Crisis Non-crisis Pre Post
ENVt−1 -0.0037 0.0297 -0.0768 0.0765 0.2731** -0.0207 -0.1225 0.0640
(0.077) (0.068) (0.073) (0.070) (0.134) (0.054) (0.113) (0.056)
SOCt−1 -0.1257 -0.2282*** -0.1160 -0.2668*** -0.3330* -0.1812** -0.3805*** -0.1488**
(0.108) (0.086) (0.102) (0.088) (0.187) (0.071) (0.144) (0.075)
GOVt−1 -0.0397 -0.0185 -0.1164 0.1116 -0.2792* 0.0255 0.0157 -0.0086
(0.079) (0.067) (0.075) (0.068) (0.155) (0.053) (0.120) (0.055)
Controls Yes Yes Yes Yes Yes Yes Yes Yes
Industry + Year FEs Yes Yes Yes Yes Yes Yes Yes Yes
# Observations 13387 14583 13928 14042 3798 24172 3794 24176
Pseudo R-squared 0.0387 0.0156 0.0231 0.0197 0.0111 0.0228 0.0238 0.0179

Panel C: Subcategories of SOC


Grouping by: Fundamentals Sub-periods
(Variable) Size Leverage Financial Crisis Learning
LOW HIGH LOW HIGH Crisis Non-crisis Pre Post
ENVt−1 -0.0195 0.0433 -0.0983 0.1165 0.2715** -0.0131 -0.1246 0.0781
(0.079) (0.068) (0.074) (0.072) (0.138) (0.055) (0.115) (0.057)
COMt−1 -0.0481 -0.2290*** -0.0957 -0.1983*** -0.1523 -0.1533*** -0.0762 -0.1692***
(0.081) (0.066) (0.077) (0.068) (0.161) (0.054) (0.128) (0.055)
DIVt−1 0.0667 0.0176 0.0204 0.0009 -0.0102 0.0168 -0.1553 0.0395
(0.088) (0.071) (0.082) (0.072) (0.179) (0.057) (0.134) (0.059)
EMPt−1 -0.1490 -0.0714 0.0492 -0.3182*** 0.0380 -0.1695*** -0.2190 -0.1189*
(0.097) (0.077) (0.090) (0.079) (0.169) (0.063) (0.138) (0.066)
HUMt−1 0.1376** 0.0713 0.1173** 0.0900* 0.0131 0.1149*** 0.1480 0.0976**
(0.055) (0.052) (0.055) (0.052) (0.096) (0.041) (0.101) (0.040)
PROt−1 -0.1751** -0.1319** -0.1688** -0.1178* -0.3963*** -0.1026* -0.3232*** -0.0851
(0.082) (0.067) (0.081) (0.067) (0.153) (0.055) (0.123) (0.057)
GOVt−1 -0.0323 0.0006 -0.0949 0.1319* -0.2255 0.0327 0.0450 0.0014
(0.080) (0.067) (0.076) (0.069) (0.156) (0.054) (0.122) (0.056)
Controls Yes Yes Yes Yes Yes Yes Yes Yes
Industry + Year FEs Yes Yes Yes Yes Yes Yes Yes Yes
# Observations 13387 14583 13928 14042 3798 24172 3794 24176
Pseudo R-squared 0.0396 0.0164 0.0237 0.0213 0.0122 0.0236 0.0260 0.0186

42

Electronic copy available at: https://ssrn.com/abstract=3305213


Table A.4: CSR and Crash Risk Predictability with Additional Controls
This table replicates the Logit (for CRASH) and Tobit (for CRASHN U M ) regressions shown in Table
3 by including additional controls for managerial entrenchment (EIN DEX), Institutional Ownership
(IO), Number of Blockholders (BHN ), CEO age and CEO compensation (CEOCOM P ). Models 1 to 5
consider each of these additional controls separately, whereas Model 6 specifies them all together. All the
other control variables are retained from Table 3, but not reported for brevity. Panel A shows results for
aggregate CSR measure (CSP ). Panels B and C report the results for the three ESG dimensions and the
seven broad CSR categories respectively. For each coefficient, standard errors clustered two-dimensionally
on firms and years, are given in parentheses. See Table 1 for variable definitions. Significance levels at
10%, 5%, and 1% are indicated by *, ** and *** respectively.
Panel A: Aggregate CSR Performance
CRASH CRASHN U M
(1) (2) (3) (4) (5) (6) (1) (2) (3) (4) (5) (6)
CSPt−1 -0.156** -0.102 -0.119 -0.149* -0.133* -0.173** -0.128** -0.096* -0.095* -0.124** -0.112** -0.155**
(0.078) (0.074) (0.074) (0.076) (0.075) (0.087) (0.058) (0.055) (0.055) (0.056) (0.056) (0.063)
EINDEXt−1 0.046*** 0.046*** 0.036*** 0.040***
(0.016) (0.018) (0.012) (0.013)
IOt−1 0.435*** 0.735*** 0.007*** 0.006***
(0.094) (0.173) (0.002) (0.001)
BHNt−1 0.008 -0.054*** 0.007 0.001
(0.012) (0.020) (0.009) (0.012)
CEOAGEt−1 0.000 0.001 -0.000 0.000
(0.002) (0.003) (0.002) (0.002)
CEOCOMPt−1 -0.010 -0.029 -0.008 -0.013
(0.020) (0.024) (0.014) (0.017)
Other Controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Industry + Year FEs Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
# Observations 16603 19260 19262 18263 18594 14089 16603 19260 19262 18263 18594 14089
Pseudo R-squared 0.0335 0.0308 0.0297 0.0291 0.0290 0.0351 0.0246 0.0221 0.0220 0.0214 0.0214 0.0250

Panel B: ESG Dimensions of CSR


CRASH CRASHN U M
(1) (2) (3) (4) (5) (6) (1) (2) (3) (4) (5) (6)
ENVt−1 -0.015 0.063 0.060 -0.005 -0.002 -0.014 -0.059 -0.009 -0.011 -0.042 -0.043 -0.035
(0.069) (0.065) (0.065) (0.067) (0.066) (0.076) (0.046) (0.043) (0.043) (0.044) (0.044) (0.050)
SOCt−1 -0.150* -0.141* -0.148* -0.150* -0.153** -0.198** -0.161*** -0.143*** -0.142*** -0.151*** -0.153*** -0.196***
(0.080) (0.076) (0.076) (0.078) (0.076) (0.088) (0.058) (0.055) (0.055) (0.056) (0.055) (0.063)
GOVt−1 0.015 0.055 0.038 0.038 0.055 0.039 -0.020 -0.002 0.007 -0.004 0.015 0.010
(0.063) (0.059) (0.059) (0.061) (0.060) (0.070) (0.046) (0.043) (0.043) (0.044) (0.044) (0.050)
EINDEXt−1 0.047*** 0.047*** 0.033*** 0.037***
(0.016) (0.018) (0.011) (0.013)
IOt−1 0.444*** 0.739*** 0.009*** 0.006***
(0.094) (0.173) (0.003) (0.001)
BHNt−1 0.009 -0.053*** 0.026*** 0.020*
(0.013) (0.020) (0.009) (0.012)
CEOAGEt−1 0.000 0.001 -0.002 -0.001
(0.002) (0.003) (0.002) (0.002)
CEOCOMPt−1 -0.007 -0.027 0.010 0.002
(0.020) (0.024) (0.015) (0.018)
Other Controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Industry + Year FEs Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
# Observations 16603 19260 19262 18263 18594 14089 16603 19260 19262 18263 18594 14089
Pseudo R-squared 0.0335 0.0309 0.0298 0.0291 0.0291 0.0352 0.0146 0.0136 0.0138 0.0132 0.0131 0.0155

Panel B: Subcategories of SOC


CRASH CRASHN U M
(1) (2) (3) (4) (5) (6) (1) (2) (3) (4) (5) (6)
ENVt−1 -0.058 -0.017 -0.007 -0.040 -0.046 -0.049 -0.054 -0.010 -0.012 -0.037 -0.041 -0.035
(0.061) (0.058) (0.057) (0.059) (0.059) (0.068) (0.046) (0.043) (0.043) (0.045) (0.044) (0.051)
COMt−1 -0.181*** -0.144*** -0.152*** -0.139** -0.138** -0.151** -0.139*** -0.114*** -0.112*** -0.103** -0.103** -0.123**
(0.060) (0.056) (0.056) (0.057) (0.056) (0.064) (0.045) (0.042) (0.042) (0.043) (0.042) (0.048)
DIVt−1 -0.033 -0.034 -0.031 0.003 -0.016 -0.094 -0.028 -0.026 -0.026 -0.004 -0.018 -0.080
(0.065) (0.061) (0.061) (0.062) (0.061) (0.070) (0.050) (0.046) (0.046) (0.047) (0.046) (0.053)
EMPt−1 -0.043 -0.015 -0.028 -0.086 -0.074 -0.031 -0.033 -0.022 -0.022 -0.068 -0.058 -0.037
(0.069) (0.065) (0.065) (0.066) (0.066) (0.075) (0.052) (0.049) (0.049) (0.050) (0.050) (0.057)
HUMt−1 0.075* 0.079* 0.084** 0.078* 0.083** 0.068 0.056 0.062** 0.061* 0.056* 0.059* 0.053
(0.045) (0.042) (0.042) (0.042) (0.042) (0.048) (0.034) (0.031) (0.031) (0.032) (0.032) (0.036)
PROt−1 -0.180*** -0.117** -0.135** -0.166*** -0.155*** -0.181*** -0.137*** -0.104** -0.101** -0.130*** -0.122*** -0.146***
(0.060) (0.058) (0.057) (0.058) (0.057) (0.066) (0.046) (0.044) (0.044) (0.044) (0.044) (0.050)
GOVt−1 0.007 0.063 0.033 0.023 0.046 0.054 -0.002 0.008 0.017 0.009 0.026 0.023
(0.060) (0.057) (0.056) (0.058) (0.057) (0.066) (0.046) (0.043) (0.043) (0.044) (0.044) (0.051)
EINDEXt−1 0.042*** 0.041** 0.033*** 0.037***
(0.015) (0.017) (0.011) (0.013)
IOt−1 0.650*** 0.999*** 0.009*** 0.006***
(0.091) (0.169) (0.003) (0.001)
BHNt−1 0.031** -0.054*** 0.025*** 0.018
(0.012) (0.020) (0.009) (0.012)
CEOAGEt−1 -0.002 -0.000 -0.001 -0.001
(0.002) (0.003) (0.002) (0.002)
CEOCOMPt−1 0.011 -0.018 0.008 0.002
(0.020) (0.024) (0.015) (0.018)
Other Controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Industry + Year FEs Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
# Observations 16603 19260 19262 18263 18594 14089 16603 19260 19262 18263 18594 14089
Pseudo R-squared 0.0305 0.0313 0.0290 0.0282 0.0281 0.0339 0.0152 0.0140 0.0141 0.0136 0.0136 0.0160

43

Electronic copy available at: https://ssrn.com/abstract=3305213


Table A.5: CSR and Crash Risk Predictability with Alternative ESG Measures
This table replicates the Logit (for CRASH) and Tobit (for CRASHN U M ) regressions shown in Table 3 by using alternative constructs (alt) of ESG-based
measures. First set of results are estimated by employing the raw ESG scores (i.e., the difference between strengths and concerns) for CSP , its three dimensions,
and subcategories. The second set of results are estimated with ESG scores computed using category-weighting approach following Deng et al. (2013). Model
1 reports the coefficients when aggregate CSR measure (CSP ) is employed. Models 2 and 3 give the results for the three ESG dimensions and the seven
broad CSR categories respectively. All control variables are retained from Table 3, but not reported for brevity. For each coefficient, standard errors clustered
two-dimensionally on firms and years, are given in parentheses. See Table 1 for variable definitions. Significance levels at 10%, 5%, and 1% are indicated by *,
** and *** respectively.
Raw ESG Scores Category-weighted ESG Scores
CRASH CRASHN U M CRASH CRASHN U M
(1) (2) (3) (1) (2) (3) (1) (2) (3) (1) (2) (3)
CSPalt
t−1 -0.0080 -0.0062 -0.0065 -0.0051
(0.006) (0.005) (0.023) (0.017)
ENValt
t−1 0.0100 0.0120 0.0045 0.0431*** 0.0240 0.0371 0.0039 0.1690**
(0.020) (0.020) (0.015) (0.014) (0.107) (0.108) (0.080) (0.077)
SOCalt
t−1 -0.0203** -0.0143** -0.0232* -0.0142*

44
(0.009) (0.006) (0.013) (0.008)
GOValt
t−1 0.0379 0.0398* 0.0259 -0.0181 0.0687 0.0444 0.0476 -0.1022*
(0.024) (0.024) (0.018) (0.018) (0.069) (0.070) (0.052) (0.053)
COMalt
t−1 -0.0869*** -0.0546** -0.1709** -0.0709
(0.032) (0.024) (0.085) (0.063)
DIValt
t−1 0.0005 0.0119 0.0680 0.0886**
(0.014) (0.010) (0.056) (0.041)
EMPalt
t−1 -0.0248 -0.0157 -0.1639* -0.0986
(0.017) (0.012) (0.091) (0.067)
HUMalt
t−1 0.0772 0.0378 0.2805** 0.1159
(0.058) (0.041) (0.141) (0.104)

Electronic copy available at: https://ssrn.com/abstract=3305213


PROalt
t−1 -0.0374 -0.0192 -0.0218 0.0148
(0.026) (0.019) (0.070) (0.053)
Controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Industry + Year FEs Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
# Observations 27360 27360 27360 27360 27360 27360 27360 27360 27360 27360 27360 27360
Pseudo R-squared 0.0314 0.0316 0.0319 0.0230 0.0231 0.0137 0.0314 0.0313 0.0318 0.0230 0.0229 0.0136
Table A.6: CSR and Crash Risk Predictability
(With Strengths and Concerns Segregated)
This table replicates the estimations shown in Table 3 (i.e., Logit for CRASH and Tobit for
CRASHN U M ) by segregating the MSCI strengths (str) and concerns (con) indicators separately for
each of the aggregate CSR measure (Model 1), three ESG dimensions (Model 2), and all seven stakeholder
subcategories (Model 3). All the ESG variables are constructed as defined in Table 1, but by considering
the set of strength and concern indicators separately. All regressions employ control variables similar
to Table 3. Standard errors (in parentheses) are reported by clustering two-dimensionally for both the
firms and years. Significance levels at 10%, 5%, and 1% are indicated by *, ** and *** respectively.
CRASH (Logit) CRASHNUM (Tobit)
(1) (2) (3) (1) (2) (3)
CSPstrt−1 -0.1550** -0.1220**
(0.072) (0.053)
CSPcont−1 -0.0518 -0.0333
(0.067) (0.050)
ENVstrt−1 0.0840 0.0115 0.0620 0.0014
(0.078) (0.072) (0.059) (0.054)
ENVcont−1 -0.0715 -0.0526 -0.0509 -0.0370
(0.075) (0.079) (0.056) (0.059)
SOCstrt−1 -0.1589* -0.1222*
(0.093) (0.069)
SOCcont−1 -0.0292 -0.0214
(0.039) (0.029)
GOVstrt−1 -0.0760 -0.0655 -0.0621* -0.0532
(0.048) (0.048) (0.036) (0.036)
GOVcont−1 -0.0951* -0.1110** -0.0679* -0.0783*
(0.054) (0.055) (0.040) (0.041)
COMstrt−1 -0.1877** -0.1307**
(0.078) (0.058)
COMcont−1 0.0382 0.0290
(0.076) (0.057)
DIVstrt−1 0.0242 0.0052
(0.073) (0.054)
DIVcont−1 -0.0163 -0.0152
(0.039) (0.029)
EMPstrt−1 -0.1041 -0.0772
(0.065) (0.048)
EMPcont−1 0.0699 0.0484
(0.058) (0.043)
HUMstrt−1 0.1268 0.1008
(0.139) (0.104)
HUMcont−1 -0.1326 -0.0998
(0.085) (0.063)
PROstrt−1 -0.1265** -0.0899**
(0.062) (0.046)
PROcont−1 0.0687 0.0584
(0.064) (0.048)
Industry + Year FEs Yes Yes Yes Yes Yes Yes
# Observations 27360 27360 27360 27360 27360 27360
Pseudo R-squared 0.0316 0.0318 0.0317 0.0232 0.0233 0.0232

45

Electronic copy available at: https://ssrn.com/abstract=3305213


Table A.7: Panel Estimates for CSR and Crash Risk Predictability
This table shows the results for static and dynamic panel Logit regressions for stock price crash likelihood
(CRASH) on CSR measures. Model 1 has the aggregate CSR measure (CSP ) as main regressor. Models
2 and 3 have the three ESG dimensions and the seven broad CSR categories as the regressors respectively.
Simple panel Logit includes random effects, whereas the dynamic panel Logit additionally corrects for
initial conditions problem and past crash likelihood. Standard errors clustered at firm level, are given in
parentheses. All the variables definitions are given in Table 1. *, ** and *** show significance levels at
10%, 5%, and 1% respectively.
Panel Logit Dynamic Panel Logit
(1) (2) (3) (1) (2) (3)
CSPt−1 -0.2124*** -0.2157***
(0.064) (0.071)
ENVt−1 0.0352 0.0367 0.0082 0.0180
(0.051) (0.051) (0.056) (0.056)
SOCt−1 -0.1674** -0.1485**
(0.066) (0.073)
GOVt−1 -0.0001 0.0083 0.0088 0.0243
(0.051) (0.051) (0.057) (0.057)
COMt−1 -0.1317** -0.1587***
(0.052) (0.058)
DIVt−1 -0.0022 0.0679
(0.055) (0.061)
EMPt−1 -0.1132* -0.0873
(0.060) (0.067)
HUMt−1 0.1010** 0.0716
(0.040) (0.044)
PROt−1 -0.1029* -0.1287**
(0.053) (0.058)
DIFTURNt−1 0.0066*** 0.0066*** 0.0066*** 0.0064*** 0.0064*** 0.0064***
(0.001) (0.001) (0.001) (0.001) (0.001) (0.001)
AVGt−1 0.0735*** 0.0723*** 0.0715*** 0.0772*** 0.0761** 0.0771**
(0.026) (0.026) (0.026) (0.030) (0.030) (0.030)
SDt−1 4.7825*** 4.7957*** 4.6953*** 4.7608*** 4.7776*** 4.7090***
(0.745) (0.746) (0.749) (0.839) (0.839) (0.845)
LEVt−1 -0.2260*** -0.2280*** -0.2205*** -0.2168*** -0.2173*** -0.2032**
(0.075) (0.075) (0.075) (0.081) (0.081) (0.081)
SIZEt−1 0.0412*** 0.0412*** 0.0376** 0.0324** 0.0326** 0.0256
(0.015) (0.015) (0.015) (0.016) (0.017) (0.017)
MBt−1 0.0428*** 0.0434*** 0.0414*** 0.0404*** 0.0411*** 0.0400***
(0.005) (0.005) (0.006) (0.006) (0.006) (0.006)
ROAt−1 0.6698*** 0.6664*** 0.6808*** 0.7352*** 0.7314*** 0.7377***
(0.119) (0.119) (0.120) (0.125) (0.125) (0.126)
OPQt−1 0.2231* 0.2238* 0.2194* 0.2390 0.2412 0.2386
(0.131) (0.131) (0.131) (0.147) (0.147) (0.146)
BTDt−1 0.0216 0.0222* 0.0200 0.0186 0.0191 0.0177
(0.013) (0.013) (0.013) (0.014) (0.014) (0.014)
NREVt−1 -0.0000 -0.0000 0.0000 0.0001 0.0001 0.0002
(0.000) (0.000) (0.000) (0.000) (0.000) (0.000)
NCSKEWt−1 0.0439** 0.0436** 0.0449*** 0.0234 0.0231 0.0248
(0.017) (0.017) (0.017) (0.024) (0.024) (0.024)
CRASHt−1 0.1089** 0.1090** 0.1084**
(0.049) (0.049) (0.049)
CRASH1990 0.0989** 0.0984** 0.0946*
(0.049) (0.049) (0.049)
Year FEs Yes Yes Yes Yes Yes Yes
# Observations 27360 27360 27360 21814 21814 21814
# Groups 3415 3415 3415 2679 2679 2679
Chi-squared 461.5 461.0 475.7 423.7 421.7 438.2

46

Electronic copy available at: https://ssrn.com/abstract=3305213


Table A.8: CSR and Crash Risk Predictability Using Instrumental Variables
This table shows the results for instrumental variables (IV) regressions for CRASH on different CSR
measures. The Model (1) employs Fama-French industry averages as IVs (El-Ghoul et al., 2011; Kim
et al., 2014), while the Model (2) includes geographical peers’ averages along with industry averages as IVs
(Jiraporn et al., 2014). The first set of columns shows the IV estimations for CSP as the main regressor.
The next two sets show the results for ESG dimensions and the seven broad CSR categories. For brevity,
only the main variables and controls from the second stage of two-stage least squares estimations are
reported. Standard errors, clustered two-dimensionally at firm and year levels, are given in parentheses.
*, ** and *** show significance levels at 10%, 5%, and 1% respectively.
Aggregate CSR Performance ESG Dimensions of CSR Subcategories of SOC
Model (1) Model (2) Model (1) Model (2) Model (1) Model (2)
CSPt−1 -0.4390*** -0.2867***
(0.078) (0.060)
ENVt−1 -0.0095 0.0132 -0.0139 0.0184
(0.038) (0.039) (0.039) (0.040)
SOCt−1 -0.2031** -0.1454**
(0.079) (0.062)
GOVt−1 -0.0774* -0.1120* -0.0590 -0.0792*
(0.047) (0.065) (0.048) (0.046)
COMt−1 -0.0640* -0.0528
(0.039) (0.039)
DIVt−1 -0.0024 -0.0283
(0.053) (0.047)
EMPt−1 -0.0642 -0.0591
(0.065) (0.054)
HUMt−1 0.0588** 0.0475*
(0.024) (0.026)
PROt−1 -0.1178*** -0.1880***
(0.045) (0.044)
DIFTURNt−1 0.0038*** 0.0038*** 0.0038*** 0.0038*** 0.0038*** 0.0038***
(0.000) (0.000) (0.000) (0.000) (0.000) (0.000)
AVGt−1 0.0585*** 0.0596*** 0.0600*** 0.0616*** 0.0605*** 0.0607***
(0.015) (0.015) (0.015) (0.015) (0.016) (0.016)
SDt−1 3.2764*** 3.2976*** 3.2774*** 3.2340*** 3.1328*** 3.0463***
(0.407) (0.407) (0.410) (0.410) (0.414) (0.414)
LEVt−1 -0.1210*** -0.1249*** -0.1306*** -0.1364*** -0.1241*** -0.1223***
(0.042) (0.042) (0.042) (0.042) (0.042) (0.042)
SIZEt−1 0.0299*** 0.0264*** 0.0234*** 0.0191** 0.0155* 0.0118
(0.008) (0.008) (0.008) (0.008) (0.009) (0.009)
MBt−1 0.0227*** 0.0234*** 0.0240*** 0.0242*** 0.0236*** 0.0242***
(0.003) (0.003) (0.003) (0.003) (0.003) (0.003)
ROAt−1 0.4301*** 0.4209*** 0.4155*** 0.4070*** 0.4113*** 0.4056***
(0.064) (0.064) (0.064) (0.064) (0.065) (0.065)
OPQt−1 0.1685** 0.1702** 0.1680** 0.1659** 0.1631** 0.1617**
(0.080) (0.080) (0.080) (0.080) (0.080) (0.080)
BTDt−1 0.0113 0.0118 0.0121 0.0122 0.0105 0.0106
(0.010) (0.010) (0.010) (0.010) (0.010) (0.010)
NREVt−1 -0.0001 -0.0002 -0.0002 -0.0002 -0.0001 -0.0001
(0.000) (0.000) (0.000) (0.000) (0.000) (0.000)
NCSKEWt−1 0.0462*** 0.0464*** 0.0469*** 0.0470*** 0.0472*** 0.0475***
(0.010) (0.010) (0.010) (0.010) (0.010) (0.010)
Year FEs Yes Yes Yes Yes Yes Yes
# Observation 27360 27360 27360 27360 27360 27360
Model Chi-Square 558.6 543.2 535.0 535.9 544.1 554.8
Model p-value 0.000 0.000 0.000 0.000 0.000 0.000
Exogeneity Test (p-value) 0.000 0.002 0.104 0.030 0.509 0.003

47

Electronic copy available at: https://ssrn.com/abstract=3305213

You might also like