Journal of Banking & Finance 37 (2013) 35293547

Contents lists available at SciVerse ScienceDirect

Journal of Banking & Finance

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

Corporate social responsibility in the banking industry: Motives

and nancial performance
Meng-Wen Wu a, Chung-Hua Shen b,
a
Department of Business Administration, National Taipei University, Taiwan, Republic of China
b
Department of Finance, National Taiwan University, Taiwan, Republic of China

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

Article history: The current study investigates the association between corporate social responsibility (CSR) and nancial
Received 21 May 2010 performance (FP), and discusses the driving motives of banks to engage in CSR. Three motives, namely,
Accepted 13 April 2013 strategic choices, altruism, and greenwashing, suggest that the relationship between CSR and FP is posi-
Available online 9 May 2013
tive, non-negative, and non-existent, respectively. We obtained our sample, which covered 20032009,
from the Ethical Investment Research Service (EIRIS) databank and Bankscope database. The data consists
JEL classication: of 162 banks in 22 countries. We then classied the banks into four types based on their degree of
C31
engagement in CSR. This study proposes the use of an extended version of the Heckman two-step regres-
C52
G21
sion, in which the rst step adopts a multinomial logit model, and the second step estimates the perfor-
M14 mance equation with the inverse Mills ratio generated by the rst step. The empirical results show that
CSR positively associates with FP in terms of return on assets, return on equity, net interest income, and
Keywords: non-interest income. In contrast, CSR negatively associates with non-performing loans. Hence, strategic
Corporate social responsibility choice is the primary motive of banks to engage in CSR.
Strategic motive 2013 Elsevier B.V. All rights reserved.
Altruistic motive
Extended Heckman two-stage model

1. Introduction reduction in social inequality. Micro-performance includes reputa-

The question of whether adopting corporate social responsibil-
ity (CSR) can improve a corporations nancial performance (FP) is
an old yet continually and heatedly debated issue.1 Thus, a better
understanding of the link between CSR and FP would be valuable
to corporate managers, stockholders, and stakeholders. For example,
what resources should managers direct to socially responsible activ-
ities? How should stockholders react to resource allocations for a so-
cial purpose? How can public policy best promote socially
responsible behavior (Simpson and Kohers, 2002)?
A corporation is generally encouraged to adopt CSR because of
its perceived benets to both macro and micro-performances.
Macro-performance includes environmental improvement and
Corresponding author.
E-mail addresses: mwwu@mail.ntpu.edu.tw (M.-W. Wu), chshen01@ntu.edu.tw
(C.-H. Shen).
1
According to the survey conducted by Grifn and Mahon (1997), explorations on
the relationship between CSR and FP date back to the works of Friedman (1962,
1970). His challenge, a corporations social responsibility is to make a prot,
triggered the debate on proving or disproving the relationship between CSR and FP.
Several studies found a negative relationship, including the study by Vance (1975),
whereas others found a positive link, such as the study by Fry and Hock (1976), as
well as mixed results, such as Coffey and Fryxell (1991).
tion enhancement, potential to charge a premium price for prod-
ucts as well as the enhanced ability to recruit and to retain high-
quality workers.2 The most attractive lure is that rms adopting
CSR can gain nancial benets that are greater than the ensuing
costs, thereby improving FP in the long run. Accordingly, adopting
CSR can be benecial to both corporate shareholders and stakehold-
ers, which creates a potential winwin situation.
2
Rather than using the concept of macro and micro benets, some studies suggest
that a company with strong CSR alleviates potential conicts of interest. For example,
Heal (2005) argues that CSR may reduce two types of conicts of interest between
corporations and society. The rst conict is that a competitive market may not
internalize all external costs, which yields discrepancies between private and social
cost and benet. This conict can be illustrated in the classic example of environ-
mental issues. The Big Oil Company that pollutes the seashore creates huge external
cost, making private cost considerably lower than social cost. The second conict is
that of distribution, which argues that the market is efcient but not necessarily fair
according to widely held opinions on distributive justice. For example, large
international companies pay low wages in poor countries, giving rise to the
sweatshop issue. Both government intervention and CSR can mitigate these two
conicts. However, the government is ineffective in solving the actual problem. Thus,
the role of CSR is still demanded. Harjoto and Jo (2011) also argue that managers may
use CSR as leverage to reduce conict with various stakeholders when managers
pursue their own interests at the expense of society and the nancial claimants of the
corporation.

0378-4266/$ - see front matter 2013 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.jbankn.2013.04.023
3530 M.-W. Wu, C.-H. Shen / Journal of Banking & Finance 37 (2013) 35293547
Numerous empirical studies attempt to justify this positive
relationship between stock return3 and accounting performance.4
However, previous results often offer contradictory conclusions.
For example, the review of Margolis and Walsh (2003) of 127 empir-
ical studies, dating from 1972 to 2002, reveals the relationship be-
tween CSR and FP. Among those reviewed, 54 studies point toward
a positive relationship, 20 show mixed results, and 28 reveal insig-
nicant relationships. Only seven studies show a negative relation-
ship. The meta-analysis by Margolis et al. (2007) on 167 studies
over the past 35 years shows that the overall effect is positive yet
small. Moreover, they suggest that future studies should be redi-
rected to gain a better understanding of the reasons for companies
to pursue CSR, the mechanism connecting prior CSR to subsequent
FP, and the methods that companies employ to manage the pursuit
of both CSR and FP (see also Shen and Chang, 2009, 2012-a; Garcia-
Castro et al., 2010).
The conicting results may be attributed to the fact that corpo-
rations with having different motives in conducting CSR may exhi-
bit different CSR-FP relationships. According to Baron (2001), Dam
et al. (2009), and Bnabou and Tirole (2010), an engagement in CSR
reects the different motives of a corporation, which include altru-
ism, strategic choices, and greenwashing. The altruism motive
indicates that companies conduct CSR activities for their own sake
(Baron, 2001), thereby negatively affecting FP. However, the strate-
gic motive improves FP through CSR engagement. Finally, green-
washing attempts to enhance the corporate image without
signicantly changing the business (Frankental, 2001).5 According
to Dam et al. (2009), if no clear cost differences are observed be-
tween responsible and irresponsible corporations, then these rms
are merely greenwashing, hence no effects will become evident in
their earnings. Thus, the conicting conclusions on the relation be-
tween CSR and FP may be attributed to the different motives of cor-
porations. Attributing the differences in the results to the different
samples, methods, and periods used is plausible as well.
The banking system plays an important role in economic devel-
opment (Levine, 2005; Shen and Lee, 2005; Beck et al., 1999; 2010)
because its safety and soundness create several external benets to
society. Generally, for economic growth, banks serve as nancial
intermediaries by facilitating cash ow between lenders and bor-
3
Becchetti and Ciciretti (2009) investigate 372 rms included in the KLDs Domini
400 Social Index during 19902003 and nd that individual Socially Responsible
stocks have on average signicantly lower returns and unconditional variance than
non-Socially Responsible stocks. Cellier and Chollet (2011) analyze the impact of 1838
Vigeo corporate social rating announcements from 2004 to 2009 on short term stock
returns of 739 rms on European markets and nd a positive signicant inuence of
the announcement on the stock returns over two days prior to the announcement and
two days following. Brammer et al. (2006) examine the relation between corporate
social performance and stock returns of all rms in the UK that were constituents of
the EIRIS and FTSE All-Share Index as of July 2002, and nd that composite social
performance indicator are negatively related to stock returns, especially on the CSR
dimensions of Environmental and community aspects. Aktas et al. (2011) nd that the
stock market rewards acquirers who make socially and environmentally responsible
investments. Renneboog et al. (2008) conclude that existing studies hint but do not
unequivocally demonstrate that socially responsible investors are willing to accept
suboptimal nancial performance to pursue social or ethical objectives. El Ghoul et al.
(2011) nd that US rms with better CSR rankings, which are results of their
investment in improving responsible employee relations, environmental policies, and
product strategies, contribute substantially to reducing the cost of equity of rms. The
factors affecting stock price are complicated, and CSR may not be inuential in this
regard. For example, there is no signicant difference in the returns of socially
responsible and conventional funds (see Bauer et al., 2005; Bello, 2005; Statman,
2005; Galema et al., 2008).
4
For example, using Taiwanese data, Shen and Chang (2009, 2011a) suggest that
CSR rms exhibit a positive relation to CSR activities and earnings. Garcia-Castro et al.
(2010) use KLD data and nd that positive relationship found in most of the previous
research on the link between SP and FP becomes a non-signicant or even a negative
relationship when endogeneity is properly taken into account.
5 7
Greenwashing refers to the disingenuous act of companies to spin their products
and policies as environment-friendly, such as presenting cost cuts as reductions in
resource use.
rowers. According to Schumpeter (1912), well-functioning nan-
cial institutions, such as banks, enhance technological innovation
by supporting entrepreneurs with the best chances of successfully
introducing innovative products and production processes. There-
fore, a healthy banking system is the key to sustained prosperity
(King and Levine, 1993). Banks also play an important role during
a credit crisis. For example, when Basel I was implemented in
1989, many banks were compelled to increase capital, to decrease
lending, or both. Decrease in lending, or the credit crunch men-
tioned in the literature review, became one of the factors that
caused the US recession in 1991.
Furthermore, by using considerable resources from society,
banks are required to provide feedback to the community more of-
ten than other industries. For example, bank assets may mainly
come from depositors, but not from shareholders. Additionally,
when banks are in distress, governments bail them out or take over
at the expense of taxpayers. In March 2011, Bank of England Gov-
ernor Mervyn King accused banks of exploiting gullible or unsus-
pecting customers for short-term prot. He criticized the culture
of short-term prot and bonuses in the banking system, and sug-
gested that traditional manufacturing industries by contrast ad-
hered more to higher moral standards when conducting
business. According to King, They care deeply about their work-
force, about their customers and, above all, are proud of their prod-
ucts. Therefore, as banks employ public resources paid for by
society, they are highly scrutinized by the media, government,
and academe regarding their CSR activities. Scholtens (2009) men-
tions that in member countries of the Organization for Economic
Cooperation and Development (OECD), specialized banks offer sav-
ings accounts to the public while promising that the money will be
used to nance so-called community investments in the environ-
ment. Thus, in most countries, banks are involved in economic
activities aimed at sustainable development.
Banks are aware of their use of public resources, and this aware-
ness explains why many banks append a CSR section in their an-
nual reports to explain how they give back to society.6 However,
direct studies on the CSR-FP link in the banking industry and the mo-
tives for engaging in CSR are considerably scant. Simpson and Kohers
(2002) present one of the few studies using bank data but they do
not pursue the CSR issue. Other studies on the banking sector en-
gaged in CSR activities focus on a similar issue, but not on the asso-
ciation between CSR and FP. For example, Scholtens and Dam (2007)
compare CSR engagements between banks that adopt the Equator
Principles and those that do not. Their nding indicates that real cost
may be associated with the implementation of the said principles.
They observe that banks adopting the Equator Principles have signif-
icantly higher CSR policies and lower returns on assets (ROA). Chih
et al. (2010) examine the determinants of nancial rms that adopt
CSR. They observe that nancial rms with larger assets and worse
ROA adopt CSR. Scholtens (2009) provides a framework for assessing
CSR with international banks, but does not examine the concerned
relationship. Cuesta-Gonzlez et al. (2006) focus on social perfor-
mance but not on the nancial performance of top Spanish banks.7
These studies do not directly address the relationship between CSR
and FP, thus drawing a conclusion for the banking sector is difcult.
Hence, in contrast to the numerous studies using non-bank data,
empirical evidence for the CSR-FP link in the banking industry is
rare.
6
For example, the HSBC annual report states that the corporations CSR is related to
charitable donations, environment, and society as well as to decision-making that
maintains the right balance between the environment, society, and the business of
the bank.
Their social performance includes codes of ethics governing bribery, corruption,
and policy with stakeholders, employees, customers, suppliers, the community,
human rights, and so on.
 M.-W. Wu, C.-H. Shen / Journal of Banking & Finance 37 (2013) 35293547
Bank theories are typically related to delegated monitoring for
alleviating the moral hazard of borrowers and reducing risk shar-
ing. In addition, the theories are related to optimal contracts and
credit rationing for lenderborrower relationship8 as well as to
bank urn and liquidity shock (Freixas and Rochet, 1998). Few bank
theories are related to CSR. One way to incorporate CSR with modern
banking theories is to relate it to the role of bank reputation. In the
theoretical model of Chemmanur and Fulghieri (1994), high-reputa-
tion banks are shown to have incentives to perform more rigorous
pre-loan evaluations of the borrowers unobservable prospects com-
pared with low-reputation banks. Bushman and Wittenberg-Moer-
man (2012) document that high-reputation banks are associated
with the stronger protability and credit quality of borrowers in
the 3 years following the loan initiation as well as with the quality
of the reported accounting numbers of borrowers. Some studies
state that bank reputation certies borrower quality because bor-
rowers have positive abnormal stock returns at the time of loan
announcements from reputational banks (Billet et al., 1995; Ross,
2010). Hence, according to the above reputation assertion, banks
conducting CSR would select and attract more creditworthy borrow-
ers, which contribute higher prot and better asset quality to the
nancial institutions.
The current study aims to investigate the relationship between
CSR and FP in the banking sector, as well as the driving motive of
banks to conduct CSR. Our study differs from other studies and
contributes to the literature in four aspects.
First, we set up a theoretical model of banking prot function
considering three motives pertaining to the ways that CSR affects
FP. In this model, CSR is one of the determinants, thereby helping
us to understand the explicit channels and motives with regard
to CSR affecting FP. Similar to Dam et al. (2009), we establish the
prot functions to examine the three motives of altruism, strategic
choices, and greenwashing. While their study focuses on the non-
banking sector in the US, we examine the relationship between CSR
and FP in the banking sector by using global banking data. Employ-
ing these relationships, we investigate the motives that drive CSR
in the banking sector. The three motives of strategic choices, altru-
ism, and greenwashing suggest that the relationship between CSR
and FP is positive, non-negative, and non-existent, respectively.
Our model is a compromise between those companies that develop
CSR prot models in the non-banking sector and those that devel-
op non-CSR prot function in the banking sector.
Second, the current study adopts an extended version of the
two-step method of Heckman (1978) to minimize the selection
bias. This approach is different from those of previous studies,
which use the traditional Heckman two-stage method with a
dichotomous CSR variable. Although a number of studies consider
multi-degree CSR, they do not eliminate the self-selection bias as
we do in the current study. Studying the relationship between
CSR and FP is often subject to the criticism of causal inference.9
This criticism arises because the banks choice to conduct CSR activ-
ities may not be random, but rather a deliberate decision made by
managers to self-select their preferred options, thereby creating a
selection bias problem.10 Thus, we consider the endogeneity prob-
lem by combining the two-step method of Heckman (1978) and
multi-degree CSR. Specically, we classify banks into four types
based on their degree of engaging in CSR. The degree ranking is from
1 to 4. The ranking was derived by using a survey conducted by the
8
See Shen (2002) for the meaning of and empirical modeling of credit rationing.
9
Several studies challenge the idea that CSR endogeneity originates from selection
bias, which may affect the results. Thus, various methods are used to solve this
problem (see Hamilton and Nickerson, 2003; James, 2006; Margolis et al., 2009;
Garcia-Castro et al., 2010; Harjoto and Jo, 2008; Shen and Chang, 2009, 2012-a; etc.).
10
For example, if larger banks with a better net income tend to adopt CSR, then the
strategic behavior (i.e., banks adoption of CSR) becomes endogenous, which causes
bias in the estimated coefcients.
3531
Ethical Investment Research Service (EIRIS). In the current study, the
four-scale measure used is interchangeably referred to as the CSR in-
dex or CSR regime. Some studies dichotomize banks into those that
adopt CSR and those that do not, but this dichotomy ignores the real-
ity that the majority of banks fall between the two extremes.11 These
studies use the Financial Times Stock Exchange (FTSE) indices (e.g.,
FTSE4GOOD) or the Dow Jones Sustainability Indexes (DJSI) to clas-
sify corporations as CSR or non-CSR corporations. Scholtens (2009)
indicates that these indices use a best-in-class approach, which as-
sures that top corporations are selected in the indices. Accordingly,
rms included in FTSE4GOOD and DJSI are those that strongly en-
gage in CSR activities, otherwise, they are excluded. Banks that con-
duct only a certain degree of CSR are often classied as non-CSR
banks. Our four-scale CSR index enables us to investigate the perfor-
mance of banks engaged even in a medium degree of CSR activities.
The ndings may shed light on the reasons for the previous mixed
results.
Third, we assess the total effect of CSR, which is referred to as
the treatment effect. In the Heckman model, our study contributes
to the literature by demonstrating that the coefcient of a CSR
dummy variable detects only the direct effect of CSR. However, be-
cause the residuals of decision and performance equations are cor-
related, there is an indirect effect through the correlation. Past
studies typically focus on only the direct effect. The direct effect
is the inuence of adopting CSR, referred to as the average treat-
ment effect (ATE). Meanwhile, the indirect effect comprises the
correlation between unexplained parts in two equations that ap-
pear in the rst and second steps (the two equations are referred
to as the decision and the performance, respectively). Total effect
is the treatment effect on the treated group (ATT) (Clatworthy
et al., 2009). This study suggests further investigation of the indi-
rect effect to consider the full-edged effect of CSR, using the cal-
culation of the treatment effect by Greene (2003, p. 788) and by
Hamilton and Nickerson (2003) after conducting Heckmans two-
step method.
Finally, we used a large sample of banking data (from 2003 to
2009) from 22 countries. Our study extends the work of Simpson
and Kohers (2002), which focuses on US bank data as well as on
the binary CSR strategy to investigate global countries and multi-
ple CSR strategy. The large sample avoids the small sample bias,
which may vary on a particular area, sample period, and frequency.
The results of the current study therefore provide more reliable
data compared with those of others.
One caveat should be noted. Chatterji and Levine (2008), and
Chatterji et al. (2010) argue that the organizations rating the social
performance of enterprises cannot truly discern which rms are
socially responsible, resulting in metrics that are often invalid
and can be misleading to stakeholders. Thus, our results depend
on the correctness of the CSR rating. If studies that nd a positive
correlation between CSR metrics and nancial performance over-
state the relationship between the actual CSR and nancial perfor-
mance, the customers or other stakeholders are misled by the
erroneous CSR metrics (e.g., by successful greenwashing
campaigns).
The remainder of this paper is organized as follows. Section 2
discusses the motivations of CSR and the model of CSR in banks
prot functions. Section 3 illustrates the construction of our CSR
index. Section 4 discusses the econometric model used in this
study and data sources and basic statistics. Section 5 presents
the empirical results for the extended Heckman two-stage model
and robustness testing. Finally, Section 6 summarizes the results.
11
Studies that dichotomize banks into CSR and non-CSR banks include those by
Simpson and Kohers (2002), Shen and Chang (2009, 2012-a), and Chih et al. (2008) to
name a few.
3532 M.-W. Wu, C.-H. Shen / Journal of Banking & Finance 37 (2013) 35293547
2. CSR motivations and bank prot function
An important aspect of CSR is how enterprises interact with
their internal and external stakeholders (employees, customers,
neighbors, non-governmental organizations, public authorities,
etc.). The stakeholder view states that effective management of
stakeholder relationships, the fundamental blocks of CSR, may also
result in better nancial performance (Ioannou and Serafeim,
2010). Thus, companies are dependent on stakeholders to obtain
the necessary resources for their survival and for their
development.
Following Baron (2001, 2009), Bnabou and Tirole (2010),
as well as Dam et al. (2009), we establish a bank prot function
incorporating CSR to illustrate how it affects FP within country.
The bank prot function is the sum of net interest income (NII)
and non-interest income (NonII) minus the overhead cost
(Overhead). That is,
Max p NIIiL ; iD ; CSR; Z NonIICSR; Z  Ov erheadCSR; Z
iL  Q L iL ; CSR; Z  iD  Q D iD ; CSR; Z NonIICSR; Z
 Ov erheadCSR; Z
where p refers to bank prot, NII denotes net interest income, iL and
iD are the loan and deposit rates, respectively, and QL and QD are the
loan and deposit amounts, respectively. The term NonII denotes
non-interest income, including fee earnings from duciary activi-
ties, service charges on deposits accounts, trading account gains
and fees, as well as commission. Overhead denotes overhead cost,
including salaries as well as premises and equipment expenses.
See Rose and Hudgins (2010) for detailed explanations for each
term. In addition, Z is the vector of other exogenous factors affecting
bank prot, such as asset size, nancial sector development, eco-
nomic growth, and legal structures, to name a few.
Our model suggests that CSR affects these three terms, namely,
NII, NonII, and Overhead to different extents. First, CSR primarily af-
fects NII positively through two amounts, QL and QD, and slightly
through two interest rates, iL and iD. Brnn and Vrioni (2001) argue
that the most obvious link of CSR to overall corporate performance
is through the aspect of reputation.
Thus, companies use CSR to increase customer loyalty and build
reputation. Fombrun and Shanley (1990) suggest that the greater
contribution of a rm to social welfare will improve its reputation.
Reputation, which closely relates to brand awareness, aids in brand
differentiation and ultimately helps a company gain competitive
advantage (Kay, 1993). Hence, CSR branding can draw consumers
away from competitors and thereby improve protability (Brine
et al., 2007). Accordingly, banks conducting CSR can attract more
loans and deposits than non-CSR banks because CSR creates a
brand name and a sense of identity among the customers. Hence,
we expect that oQL/oCSR > 0 and oQD/oCSR > 0 for both altruistic
and strategic motives.
However, the impacts of CSR on interest rates differ for various
motivations in different market structures. Banks engaging in CSR
with altruistic motives do not use interest rate to inuence the net
interest income because the goal of engaging in CSR is not to in-
crease prot. In contrast, banks conducting CSR with strategic mo-
tives use the rate to inuence net interest income that can
substantially raise prot. CSR could reduce consumer price sensi-
tivity (Sen and Bhattacharya, 2001). Kim et al. (2005) state that a
rm prefers to borrow from well-known banks even it has to pay
higher loan rates. The reason is that the loan can signify the credit
worthiness of a rm and mitigate asymmetric information ob-
tained by its other stakeholders. Customers may receive a low de-
posit rate because CSR banks provide a sense of trust. Baron (2009)
argues that strategic CSR allows rms to differentiate their prod-
ucts to soften the intensity of price competition, and to allow a
morally managed rm to obtain a price premium for its products.
However, market structure may affect the argument of charging
higher price strategy arising from product differentiation. When
the banking sector is in an oligopoly market, the argument is cor-
rect because few banks are in the market, and the existing banks
have stronger power to use CSR to differentiate the products. How-
ever, when the banking sector is in monopolistic competition, the
charging high loan rate strategy is weaker than that in oligopoly
market. Hence, CSR may only slightly affect the spread because cli-
ents can easily shop for the most favorable loan and deposit rates
among banks in this market.12 Therefore, considering the above two
different market structures, the effect of CSR on interest rates may be
positive or small. We expect that oiL/oCSR > 0 and oiD/oCSR < 0,
where > and < mean slightly larger and smaller, respectively.
Although the effect of CSR on rates may be nil, the net effect of
CSR on NII is still positive through the increase in loans and deposits.
Thus, our Hypothesis 1 is as follows.
Hypothesis 1. For both strategic and altruistic CSR banks, CSR
positively affects NII.
Brand differentiation can be applied to business-related non-
interest income in the banking sector. This brand differentiation
can be created because the quality of service differs across banks,
allowing CSR banks to charge higher commissions and fees. Hence,
customers are willing to pay a higher price for the socially
friendly good. Kim et al. (2005) state that product differentiation
in banking means that a bank differs from other banks in such a
way that all customers would consider it better than its competi-
tors (e.g., better services). Trotta et al. (2011) claim that CSR is an
important reputational driver that creates economic value over
time. The products of non-interest income are standardized or cus-
tomer-made. Standardized products include ATM fees, remittance
fee, letter of credit, and others. Customer-made products include
wealth management, among others. Accordingly, strategic CSR
banks tend to pursue prot maximization by promoting their rep-
utation and brand names through activities that include environ-
ment protection, charity behavior, and integrity to customers.
These activities result in higher prices and quantities of non-inter-
est income, especially for customer-made products. Bnabou and
Tirole (2010) label this perspective as the winwin vision of
CSR and they believe that strategic CSR involves taking a socially
responsible position to strengthen market position to increase
prot. The discussion of the effect of market structure on pricing
through brand differentiation in Hypothesis 1 could be applied to
the case of NonII.
The CSR activities themselves are the goals of the altruistic CSR
banks. The observed environmental improvement and reduced
tension in social problems, among others, are the rewards.
Although altruistic CSR may not increase the prices of non-interest
income, these activities create product differentiation, which at-
tracts customers and increases NonNII.
Hypothesis 2. CSR positively affects NonII for strategic and altru-
istic CSR banks.
The inuence of CSR on overhead is mostly likely to be positive.
A good CSR bank can reduce the information asymmetry with
stakeholder and win the respect of the community, reducing the
public relationship staff. Also, the employees, when they have
the choice, will prefer to work in a socially responsible company.
Thus CSR could reduce the overhead cost. For example, a rm with
a favorable work environment can decrease its hiring costs and in-
12
Banking market structure is rarely in perfect competition and monopoly, thus, we
do not discuss the impacts of these two conditions.
 M.-W. Wu, C.-H. Shen / Journal of Banking & Finance 37 (2013) 35293547
crease its employee retention rate. It can also decrease community
opposition and legal costs when opening a new factory, and more
easily lobby for tax breaks from local governments (Freeman,
1984; Waddock and Graves, 1997). However, conducting CSR con-
siderably increases the expenditure. Several studies argue that so-
cial responsibility detracts from a rms nancial performance,
because any discretionary expenditure on social betterment
unnecessarily raises the costs of a rm, thereby putting it at an eco-
nomic disadvantage in a competitive market (Friedman, 1970;
McWilliams and Siegel, 1997; Jensen, 2002). Given that CSR invest-
ments are costly, the potential nancial benets will be gained in
the long term, if they occur at all (Henderson, 2002). The net im-
pact of CSR on overhead is assumed to be positive, which indicates
that oOverhead/oCSR > 0. Strategic banks aim to maximize prot so
they carefully control CSR, suggesting that the impact of CSR on
prot is positive. In contrast, following the argument of Dam
et al. (2009) that altruistic CSR is prot-inefcient because of high-
er cost, banks that adopt altruistic CSR sacrice prot for social
benets. Hence, the impact of altruistic CSR on prot is close to
non-positive.
Hypothesis 3. The effect of CSR on prot is positive for strategic
banks and close to zero for altruistic banks.
Finally, greenwashing CSR may insignicantly affect the quan-
tity and quality of products. Greenwashing is comparable to lip
service, which induces no cost for charity, as well as no increase
in revenues in the long run. For example, banks may preach about
ethical loans when they commit to the Equator Principles but do
not actually follow them (Watchman et al., 2006).13 Greenwashing
banks engage in very few CSR activities having almost no effect on
interest income, non-interest income, and non-interest expenses.
Scholtens and Dam (2007) and Dam et al. (2009) suggest that CSR
based on greenwashing motives has no effect on the prot of
banks.14 That is, oNII/oCSR = 0, oNonII/oCSR = 0, and op/oCSR = 0.
Hypothesis 4. Greenwashing CSR exhibits no effect on NII, NonII,
and prot.
3. Construction of CSR index
Our aggregate CSR index covers all the CSR activities of each
bank, where the data of activities come from the EIRIS database,
a leading socially responsible investment research provider that
analyzes 2800 companies around the world.15 We choose the EIRIS
survey because, according to Brammer and Pavelin (2004) and Bram-
mer et al. (2006), this survey offers the largest and most complete
13
According to the Equator Principles Financial Institutions (EPFIs), banks should
not provide loans to projects that will not or cannot comply with their respective
social and environmental policies and procedures that implement the EPs.
14
Few empirical studies report the relationship between greenwashing and bank
performance. Some studies challenge that the banks that joined the Equator
Principles (EPs) did not grant loans to rms complying with EP (Orchard-Webb,
2008; Watchman et al., 2006). Scholtens and Dam (2007) do not examine the
greenwashing behavior of banks, but cite an accusation of greenwashing in banks
adopting EP as their motive. Missbach (2004) only mentions that Equator Principles
were labeled as Greenwash Principles in his footnote.
15
EIRIS was founded in 1983 by churches and charities interested in making
principled investment decisions. It is a global provider and gathers data on the basis
of a questionnaire. It analyzes the survey for clients, which include 80 major
institutional investors, socially responsible mutual funds, and the FTSE Group for their
leading SRI index, FTSE4Good. EIRIS specializes in the measurement of corporate
social performance against a consistent and objective set of criteria, principally for the
consumption of investors. EIRIS not only conducts surveys to evaluate the social
performance of rms and provides these data to the public, but also publishes CSR-
related research using survey data. As a result, they are able to provide social
performance scores for a rm irrespective of whether it participates in an EIRIS
survey.
3533
multidimensional social performance coverage, covering issues re-
lated to employment, the environment, community, human rights,
supply chain management, and so on. The EIRIS survey has been
widely used by academic CSR studies and bank practitioners (see
Cuesta-Gonzlez et al., 2006; Scholtens and Dam, 2007; Brammer
and Pavelin, 2004; Brammer et al., 2006; Cox et al., 2004).
Previous studies investigate the CSR effect on performance
using either a complete aggregate measure of CSR or dimensional
CSR measures. For example, Waddock and Graves (1997) use their
own weighting scheme to compute a weighted average CSP index
for each company, whereas Dam et al. (2009) consider aggregate
CSR as well as dissected CSR such as Product, Environment, Em-
ployee, Diversity and Community. Our CSR measure is close to
the aggregate measure of CSR for the banking sector. While differ-
ent CSR measures are used in the literature, Simpson and Kohers
(2002) assert that an ideal empirical measure of a comprehensive
conceptual construct of CSR does not exist.
The survey in EIRIS is complex. For example, for qualitative
questions, some are two-scale yes or no questions, whereas
some are three-scale (e.g., many, some, or none) and four- to
six-scale choices. What makes things even more complicated is
that some positive replies favor CSR,16 whereas others imply the
opposite.17 For simplicity, we refer to these two types of questions
as positive and negative attitude questions, respectively.
Our transformation of text answer in the survey into a single
aggregate value is similar to the approaches employed in the litera-
ture. Cox et al. (2004), Brammer and Pavelin (2004), and Brammer
et al. (2006) translate a text-grade rating into a number-grade rating
and normalize the number-grade rating for each attribute. Then, the
aggregate CSR index could range from 3 to 12 scores (Brammer et al.,
2006; Brammer and Pavelin, 2004) or 0 to 9 scores (Cox et al., 2004).
Our CSR index ranges from 1 to 5 when we originally construct the
index. Its calculation is based on the widely-accepted concept of
the Likert 5-point scale.18 However, after we construct the index,
there are only two observations in regime 5, making it unable to esti-
mate the multinomial logit regression in that regime. Hence, we nally
combine CSR indexes 4 and 5, resulting in a four-point scale.
Our approach on creating numerical ratings involves four steps.
First, we transform the text-grade rating for each reply into a num-
ber-grade rating. As mentioned above, both positive and negative
attitude questions are included. Taking the positive attitude ques-
tions of three-degree choice replies as an example, scores of 13
are assigned to different strength responses, such as Basic, Mod-
erate, and Good, in which the larger numbers indicate a higher
degree of social responsiveness. A similar concept is applied for the
opposite attribute. Subsequently, we extend the obtained numeri-
cal values initially into ve scales (i.e., 15) using the standardiza-
tion method, which assumes that the data follow a uniform
distribution.19 Next, we obtain a score for each bank by summing
up all standardized scores. At this step, each bank is assigned a
numerical value for each year. Finally, we again standardize each
banks score in each year. Later, we change the ve regimes into four
16
For example, the question How clear is the companys commitment to
community or charitable work? has the following reply choices: Advanced,
Good, Intermediate, Basic, Limited, and Little or none.
17
For another example, the question What is the level of potential exposure to
bribery issues? has the reply choices, High, Medium, and Low.
18
For example, Jehn et al. (1999) have used surveys with the 5-point Likert scale to
explore the differential effects of social category diversity, value diversity, informa-
tional diversity, task type, and task interdependence on workgroup performance. Al-
Hussami (2008) has investigated the relationship of the job satisfaction of nurses to
organizational commitment and perceived organization using surveys with the 5-
point Likert scale. Gil et al. (2005) have used surveys with the 5-point Likert scale to
analyze the effects of change-oriented leaders on group outcomes.
19
The comparison between our transformation and those used in previous studies is
difcult because they do not clearly explain how they do the transformations. Our
transformation method is reported in the Appendix.
3534 M.-W. Wu, C.-H. Shen / Journal of Banking & Finance 37 (2013) 35293547

Table 1 and CSR regime 3 with 233 (23.02%). The lowest number (percent-
The number of bank-year observations based on CSR index and country. age) is for CSR regime 4, containing 64 (6.32%) observations. Thus,
CSR index most banks engage to some extent in CSR activities.
Country All 1 2 3 4 Next, we discuss bank-year observations in six regions. Asia,
Num Num (%) Num (%) Num (%) Num (%) especially Japan, has the largest number of bank-year observations,
North 157 10 (6.37) 110 35 (22.29) 2 (1.27)
followed by North America, Western Europe, Southern Europe,
America (70.06) Australia and Northern Europe. In addition, except for Northern
Canada 43 0 (0.00) 19 (44.19) 24 (57.14) 0 (0.00)
and Western Europe, which have the highest number of bank-year
USA 114 10 (8.85) 91 (79.82) 11 (9.73) 2 (1.77) observations in regime 3, other regions have the highest number in
North Europe 42 0 (0.00) 10 (23.81) 29 (69.05) 3 (7.14)
regime 2. It is interesting to note that most of the regions have a
Denmark 7 0 (0.00) 4 (57.14) 3 (42.86) 0 (0.00) single gure number of bank-year observations in regime 4, except
Norway 7 0 (0.00) 1 (14.29) 4 (57.14) 2 (28.57) for Western Europe, which has 46 bank-year observations.
Sweden 28 0 (0.00) 5 (17.86) 22 (78.57) 1 (3.57) Finally, in each country, the percentages of CSR banks in each
West Europe 154 15 (9.74) 29 (18.83) 64 (41.56) 46 regime are seen to differ. We illustrate the differences by discuss-
(29.87) ing only the highest percentages across four regimes for each coun-
Austria 14 5 (35.71) 5 (35.71) 4 (28.57) 0 (0.00) try. For example, in regime 1, banks from Austria, Hong Kong,
Belgium 14 0 (0.00) 2 (14.29) 10 (71.43) 2 (14.29) Israel, and South Korea exhibit the highest percentages, indicating
France 28 3 (10.71) 7 (25.00) 16 (57.14) 2 (7.14)
that the lions share of banks from these countries engage in lesser
Germany 21 2 (9.52) 8 (38.10) 6 (28.57) 5 (23.81)
Ireland 14 0 (0.00) 5 (35.71) 9 (64.29) 0 (0.00) CSR activities. Next, in regime 2, the highest percentages of banks
Switzerland 28 5 (17.86) 2 (7.14) 9 (32.14) 12 are from Australia, Denmark, Germany, Greece, Italy, Japan, Portu-
(42.86) gal, Singapore, and the United States. This suggests that most
UK 35 0 (0.00) 0 (0.00) 10 (28.57) 25
banks from these countries conduct some degree of CSR. Similarly,
(71.43)
in regime 3, banks from Belgium, Canada, France, Ireland, Norway,
South Europe 147 14 (9.52) 73 (49.66) 55 (37.41) 5 (3.40)
Spain, and Sweden show the highest percentages. Thus, banks from
Greece 28 2 (7.14) 14 (50.00) 12 (42.86) 0 (0.00) these countries conduct more CSR activities. In regime 4, the high-
Italy 56 5 (8.93) 33 (58.93) 17 (30.36) 1 (1.79)
est percentages of banks are from Switzerland and the United
Portugal 14 1 (7.14) 8 (57.14) 5 (35.71) 0 (0.00)
Spain 49 6 (12.24) 18 (36.73) 21 (42.86) 4 (8.16) Kingdom. Thus, banks in these two countries engage in more CSR
activities than other countries.20
Asia 463 191 238 33 (7.13) 1 (0.22)
(41.25) (51.40)
4. The extended Heckmans two-step method
Hong Kong 56 26 (46.43) 23 (41.07) 6 (10.71) 1 (1.79)
Japan 288 79 (27.53) 182 27 (9.41) 0 (0.00)
(63.07) 4.1. Multinomial logit model
South Korea 63 53 (84.13) 10 (15.87) 0 (0.00) 0 (0.00)
Singapore 21 3 (14.29) 18 (85.71) 0 (0.00) 0 (0.00) Studies of how CSR affects FP are often considered to suffer from
Israel 35 30 (85.71) 5 (14.29) 0 (0.00) 0 (0.00)
an endogeneity problem, because using OLS gives rise to the miss-
Australia 49 4 (8.16) 21 (42.86) 17 (34.69) 7 (14.29) ing third variable problem since non-random engagement in CSR
Australia 49 4 (8.16) 21 (42.86) 17 (34.69) 7 (14.29) affects performance. To avoid using OLS, the most widely-used
All (%) 1012 234 481 233 64 (6.32) method for obtaining consistent regression estimates in models
(23.12) (47.53) (23.02) with selectivity is a two-stage procedure developed by Heckman
Notes: The number in the blank refers to the number of bank-year observations
(1978). Theoretically, although the rst step of binary choice using
from 2003 to 2009. Heckmans two-step method can be extended to a multiple choices
% in the parentheses beside Num is the ratio of the number of bank-year obser- using extended Heckman method, in practice, this is difcult to
vations under specic CSR index in a country to the total number of bank-year implement. The correction of the endogeneity bias is seldom ap-
observations in the country. For example, in North America, 6.37% = (10/
plied in a multiple choices model. Hence, we discuss this in detail
157)  100%; 70.06% = (110/157)  100%; 22.29% = (35/157)  100%; 1.27% = (2/
157)  100%. In the last row, 23.12% = (234/1012)  100%; 47.53% = (481/ in this section.
1012)  100%; 23.02% = (233/1012)  100%; 6.32% = (64/1012)  100%. The conventional Heckmans two-step method proceeds as fol-
The 162 banks in 22 countries are 7 banks in Australia, 2 banks in Austria, 3 banks in lows. The rst step estimates the decision equation using a logit/
Belgium, 6 banks in Canada, 1 banks in Denmark, 4 banks in France, 3 banks in probit model to yield IMR, where the dependent variable in the
Germany, 4 banks in Greece, 8 banks in Hong Kong, 2 banks in Ireland, 5 banks in
decision equation is a binary number. The second step employs
Israel, 8 banks in Italy, 43 banks in Japan, 9 banks in South Korea, 1 banks in
Norway, 2 banks in Portugal, 3 banks in Singapore, 7 banks in Spain, 4 banks in the generated IMR as the additional explanatory variable in the
Sweden, 4 banks in Switzerland, 5 banks in United Kingdom and 31 banks in USA. performance equation (see Li and Prabhala, 2007; Hamilton and
Nickerson, 2003). Because of the presence of IMR in the perfor-
regimes because there are only two observations in the fth regime. mance equation, the estimated CSR effect of concern in the perfor-
See the Appendix for the detailed calculation steps. mance equation would be unbiased.21
By employing the EIRIS database, we collect 162 CSR-banks
from 22 countries. Table 1 reports the number of bank-year obser- 20
Our data are based on all observations based on replies to the EIRIS, instead of the
vations in each CSR regime across 22 countries, where the percent- number of banks included in the EIRIS. Thus, if fewer banks in a country reply to the
ages over each country are in parentheses. There are 234, 481, 233 EIRIS, the country may have fewer observations, even though it in fact has a large
number of banks. This may be the reason why Germany and Switzerland have few
and 64 bank-year observations found in CSR regimes 14, respec-
observations.
tively. Banks in higher CSR regimes are primarily to engage in 21
Previous studies in different issues employ binary probit/logit estimation as the
CSR. Thus, banks in CSR regimes 1 and 4 are those that are the least rst step of the Heckman two-stage model. Chaney et al. (2004) investigate whether
and primarily to conduct CSR, respectively. hiring or not hiring Big Five auditors commands a premium, while Booth (1993)
Three interesting ndings are summarized as follows. First, investigates whether taking or not taking employer-provided training inuences the
earnings of men and women graduates. Shaver (1998) and Cerutti et al. (2007)
judging by the aggregate number at the bottom of Table 1, the investigate the factors that inuence the banks type of organizational form when
highest number (percentage) of observations of 481 (47.53%) is operating in foreign markets, and Harjoto and Jo (2008) investigate whether
in CSR regime 2, followed by CSR regime 1 with 234 (23.12%) conducting or not conducting CSR activities affects a rms operating performance.
 M.-W. Wu, C.-H. Shen / Journal of Banking & Finance 37 (2013) 35293547
Our extended Heckmans two-step model extends the conven-
tional Heckmans model when the dependent variable in the deci-
sion equation is not a binary number, but is a multilevel choice.
The rst step uses the multinomial Logit model to estimate the
decision equation and generate IMR. The second step remains the
same by employing the resulting IMR in the performance equation.
Lee (1983) proposes a generalization of Heckmans two-step
selection bias correction method where selectivity is modeled as
a multinomial logit case. In their Monte Carlo experiment, Bourgui-
gnon et al. (2007) indicate that selection bias correction based on
the multinomial logit model provides good correction for the per-
formance equation, even when independent and identical assump-
tions of residuals are violated. This extended Heckmans two-step
approach for reducing the selection bias problem has rarely been
applied when the strategy is a multilevel choice.22
Without loss of generality, the CSR index s (s = 1, . . . , 4) is ob-
served only for those banks in the s degree of CSR. In addition,
for simplicity, vectors x and z represent two sets of exogenous
explanatory variables that inuence the performance and decision
equations represented by Eqs. (1) and (2), respectively.
Assuming that there are M (=4 here) categories, our decision
and performance equations can be written in a vector form as
IiS ziS cS giS i 1; . . . ; NS ; ; s 1; . . . ; M
IiS s iff Is > max Ij
j1;...;M; jS
yiS xiS bS liS
where the subscript s is a categorical indicator that describes the
choice of a bank among M alternatives and i denotes the ith bank;
NS denotes the number of banks that are in the sth degree of CSR;
yiS denotes the performance reecting the adoption of the sth CSR
by bank i; and IiS denotes a latent variable representing the indirect
utility of the ith bank in the selection of the sth degree of CSR. IiS
normalizes the value of the indirect utility derived from the sth
combination as IiS 0. IiS, which is the observed CSR index that
bank is selects as the sth degree of CSR. When the CSR index s is
chosen, the choice must yield the highest indirect utility compared
with any other combination. Vector z is a set of exogenous explan-
atory variables that includes variables that inuence the banks
choice of CSR; bS and cS are sets of unknown parameters and liS
is an unobservable random error; and giS is an unobservable ran-
dom component. In addition, E(liS|x1, . . . , xM, z1, . . . , xM) = 0 and
E(giS|x1, . . . , xM, z1. . . , xM) = 0. In the following discussion, we skip
the subscript i for simplicity.
To illustrate the reason why a coefcient is biased when errors
in the performance equation and those in the decision model are
correlated, we use CSR index 4 as an example. The mean perfor-
mance of that regime is conditional on the banks selection of
CSR index 4.
Ey4 jI4 > I3 ; I4 > I2 ; and I4 > I1 x4 b4 El4 jI4 > I3 ; I4
> I2 ; and I4 > I1
Hence, the conditional mean of the regression error will differ from
zero. Thus, a regression of yS on x alone will lead to biased and
inconsistent estimates of the parameter vectors. Not accounting
for the nonzero conditional mean of the error in the regression
function is equivalent to omitting a relevant variable from the esti-
mated model.
22
Garcia-Castro et al. (2010) consider the multilevel CSR index for studying the
relationship between social performance and FP in the nonbanking sector, where the
Kinder, Lydenberg, Domini & Co. Inc. (KLD) index is used. However, they do not use
the extended Heckman two-step method.
3535
To obtain consistent estimates of the quality equation, we in-
stead use a modication of Heckmans two-step estimator sug-
gested by Lee (1982, 1983). The principle underlying Lees
method is that a random variable from a non-normal distribution
can always be transformed into a standard normal random vari-
able. Hence, one need not assume normality in the errors of the
choice model. This procedure allows us to estimate the choice
model via a multinomial logit model, which is a computationally
simple procedure, and to construct the mean of the regression er-
ror conditional on the normal transformation of the estimated logit
choice model.23
When IS = s if and only if zScS > eS,
eS max zij ci gij  zis cs  gis
j1;...;M
js
Dening C = (zc1, zc2, zc3, zc4) and calling F eS jC the cumulative
distribution function of eS:
JjC U1 F eS jC
where U is the standard normal cumulative, assuming that lS and
JeS eS jC are jointly distributed and do not depend on C and
ElS jeS ; C rqS JeS eS jC, where qS denotes the correlation coef-
cient between eS and lS. Without loss of generality, the expected va-
1 lue of the disturbance term ls, conditional on the CSR index s being
chosen, can now be written as
2
/J eS 0jC
ElS jeS < 0; C rqS rqS  IMRS 4
F eS 0jC
3
where / is the standard normal density and IMRS = //U. Under this
hypothesized form for kC, a consistent estimator of bS is obtained
by performing a least squares analysis on the following equation:
yS xS bS  rqS IMRS mS 5
While we correct the aforementioned bias using Heckmans model
and Lees method, it has been noted that the resulting unbiased
coefcient of the CSR dummy variable is an incomplete measure
of the performance difference between the two groups of banks.24
Accordingly, to fully explore the CSR treatment effect, we adopt
the method of Clatworthy et al. (2009) to calculate ATE and ATT.
The counterfactual is calculated by inserting the means of the re-
gressed variables of the CSR benchmark into other CSR index regres-
sions. Thus, while the new specication includes IMR as the
additional explanatory variable, ATE simply calculates the perfor-
mance differences between CSR and non-CSR banks without consid-
ering the indirect effect resulting from IMR (see Clatworthy et al.,
2009), which is equal to d  CSRindex. However, the treatment effect
on the treated group (ATT) differs from ATE which is coefcients of
the CSR dummy variable by an adjustment term, making the esti-
mated coefcients of interest incomplete measures of the treatment
effect. Thus, ATT is calculated by inserting the means of variables of
CSR banks. That is
ATT Stoj EPerformS  Performj jI s; x xS ; z EyCSRS jx
xS ; z  EyCSRj jx xS ; z
/x xS
d rqS 6
F eS x xS 1  F eS x xS
23
The validity of the multinomial logit model relies on the assumption of the
independence of irrelevant alternatives (IIA) (Greene, 2012, 7th edition, page 807),
which means that all else being equal, a persons choice between two alternative
outcomes is unaffected by other available choices. Using Hasuman and McFaddens
test for IIA, we nd that the three Chi-square values are insignicant at the 5% level
regardless of the constrained models. Thus, our data fullls the IIA assumption. The
results are available upon request.
24
Chaney et al. (2004) investigate the effect of Big Five auditor premiums by using
the Heckman two-step method without calculating the treatment effect.
3536 M.-W. Wu, C.-H. Shen / Journal of Banking & Finance 37 (2013) 35293547
where PerformS and Performj denote the performances of the banks
choosing CSR index s and those choosing CSR index j, respectively.
Past studies have simply examined the coefcient of d and omitted
the adjusted term that arose from IMR. The existence of the last
term in Eq. (6) indicates that when considering only d, the estima-
tion of the treatment effect is incorrectly achieved.
We are still, however, confronted with a problem. Even when
ATT is obtained, the standard deviation is not easily derived. Thus,
a straightforward yet slightly difcult method is to use the boot-
strapping method to compute the condence intervals.25 The boot-
strapping is performed as follows. First, the residuals from
performance equation (i.e., Eq. (3)) are reshufed. Second, the
reshufed residuals are combined with xb to generate the predicted
performance (:y ^), where x represents the vector explanatory vari-
ables and b ^ is the vector of the estimated corresponding coefcients
in Eq. (3). Third, regressing y ^ on x creates new estimated coefcients
^
^ as well as new residuals and the estimated ATT. Fourth, the above
b and real estate, respectively. The three restrictions measure the de-
steps are repeated 10,000 times to yield 10,000 ATT values and a his-
togram. Intervals (2.5%, 97.5%) are used to decide the critical values.
In addition to using the extended Heckmans two-step estima-
tion, many researchers have used matching theory (Rubin, 1986)
to conduct similar studies by using banking or nonbanking data
in different countries.26 However, their matching method considers
only two categories of groups. The extension of the conventional
two-level categories into multilevel categories in the nancial eld
is still in its infant stage. This deserves further study.
4.2. Regression model
Our decision and performance equations, which comprise the
extended Heckman two-step method, are respectively specied
as follows.
PrCSRindex s a0 a1 log TAikt a2 Leverageikt
a3 LoanDepikt a4 CostIncikt a5 Coverageikt
a6 Res Skt a7 Res Ikt a8 Res Ekt
a9 Corruptkt a10 Credit GDPkt a11 HHIkt
a12 GDPGrowkt a13 GDPperkt likt
yikt dCSRIndex;ikt b0 b1 logTAikt b2 Leverageikt
b3 LoanDepikt b4 CostInckt b5 Res Skt b6 Res Ikt
b7 Res Ekt b8 Corruptkt b9 Credit GDPkt b10 HHIkt
b11 GDPgrowkt b12 GDPperkt kIMRikt eikt
where subscripts i, k, and t denote the ith bank in the kth country at
time t, respectively. In Eq. (7) (decision equation), CSRindex denotes
the CSR index (or regimes) discussed above, and s denotes the
CSR index and equals 1, 2, 3 and 4, respectively. In Eq. (8) (perfor-
mance equation), y denotes bank performance, which is proxied
by the return on assets (ROA), return on equity (ROE), nonperform-
ing loans ratio (NPL), net interest income ratio (NII), and non-inter-
est income ratio (NonII). In addition, l  iidN0; r2l and
e  iidN0; r2e . The current study does not examine the dynamic
inuence of CSR on performance. This should pose little problem
because the CSR ratings do not change to a large extent between
previous and subsequent periods. As an example, the study of Goss
and Roberts (2011) uses instrumental variable regressions and the
Heckman selection model without considering the dynamic process
to control the endogeneity of CSR.
25
To the best of our knowledge, Heitmueller (2004), in a study of job mobility, is the
only paper that computed condence intervals using the bootstrap method.
26
For example, see Shen and Chang (2009, 2012-a, 2012-b).
The controlled variables used in the decision and performance
equations follow our theoretical model, and those of McWilliams
and Siegel (2000, 2001), Shen and Chang (2009, 2012-a), Chih
et al. (2010) and Harjoto and Jo (2008). The explanatory variables
can be divided into three categories. The rst category is the bank
characteristics (logTA, Leverage, LoanDep, CostInc, Coverage)
where logTA refers to the total assets after the logarithmic trans-
formation, Leverage represents the ratio of equity to total assets,
LoanDep denotes the loan to deposit ratio, CostInc refers to the ra-
tio of overhead cost to total income, and Coverage denotes the ratio
of loan loss reserves to nonperforming loans.
Because our empirical estimation is conducted across countries,
our second category consists of the institutional factors, such as
Res_S, Res_I, Res_E, and Corrupt and nancial sector development
variable, Credit_GDP, where Res_S, Res_I, and Res_E denote the
restrictions on banking activities involving securities, insurance,
grees of regulatory restrictions for each activity from 1 to 4, with
the larger number representing greater restrictiveness. We con-
sider these institutional factors because these three restrictions
signicantly affect bank performance. For example, Wu and Shen
(2011) nd that more restrictions on insurance and real estate
activities strengthen the positive effect of market share on bank
performance (see also Shen and Chang, 2006). Corrupt, which re-
fers to control over corruption, is used since the variable is found
to affect bank performance (Shen and Lin, 2012). Following Beck
et al. (1999, 2010) and Shen and Lee (2005), we add Credit_GDP
to consider the inuence of nancial sector development on bank
performance, where Credit_GDP refers to the ratio of domestic
credit to the private sector over GDP. Finally, as suggested in Sec-
tion 2, we consider the concentration ratio of the HerndahlHir-
schman Index (HHI) as a measure of product market structure
(see Bikker and Haaf, 2002a, 2002b; Bikker and Spierdijk, 2010).
HHI is the sum of the squares of the market share percentages held
by banks. Bikker and Haaf (2002a, 2002b) and Bikker and Spierdijk
(2010) show that HHI is highly correlated to competition,27 which
affects the business decisions as well as the prot-making and cost-
reduction activities of banks (see Fisman et al., 2005, 2006). Fernn-
dez-Kranzl and Juan Santal (2010) as well as Bikker and Haaf
7 (2002a, 2002b) and Bikker and Spierdijk (2010) add HHI in the prot
function.
The third category comprises the macroeconomic variables
(GDPgrow and GDPper), where GDPgrow denotes the GDP growth
rate, and GDPper denotes GDP per capita. We consider the macro-
economic variables because the country development variables
8 may also affect the revenue and cost functions and CSR decision.
These macroeconomic variables are also used to capture country
economic heterogeneity. In addition, because many regulatory
and policy variables may share similar information to these macro-
economic variables, we include these two variables to reect the
information in this regard.
The coefcient of IMR has statistical and economic explana-
tions. First, the statistical signicance of the estimated coefcient
of IMR (often called the selectivity correction) tests for the pres-
ence of any selectivity bias and its sign indicates the direction of
the bias. If the coefcient of IMR is signicant, there is then a sig-
nicant correlation between the error terms obtained in the two
steps of the Heckman method (Magableh et al., 2011). An insignif-
icant effect of IMR on the dependent variables indicates that no
27
We thank the referee for reminding us that competition and concentration not
one-on-one concepts in the banking industry. They are two different but closely
related concepts; the former is typically estimated by using the method of Panzar and
Rosse, whereas the latter is often proxied by HHI. As the two concepts are highly
correlated, this study simply uses HHI as the proxy for competition. See Bikker and
Haaf (2002a, 2002b) and Bikker and Spierdijk (2010).
 M.-W. Wu, C.-H. Shen / Journal of Banking & Finance 37 (2013) 35293547
sample selection bias exists. Next, Clatworthy et al. (2009) provide
the economic explanation. They argue that IMR is a proxy for unob-
servable characteristics that affect both the decision and perfor-
mance equations. In our case, the unobservable characteristics
may be the banks knowledge of management, concerns over rep-
utation, and so on. The signicance of these characteristics indi-
cates that the performance equation is systematically related to
the unobservable characteristics of banks. Not adding it to the
equations results in the missing third variable problem, causing a
biased estimation.
To avoid the individual effect and heteroskedasticity, we con-
duct the least squares dummy variable (LSDV) regression by
including country dummies and calculate Whites heteroskedastic-
ity-consistent standard errors.
4.3. Data sources and basic statistics
Our sample is collected from ve different sources. As intro-
duced above, the CSR index of each bank is compiled from the EIRIS
database. The nancial ratios and relevant information of these
CSR banks are obtained from the Bankscope database. By consider-
ing some M&A events, credit functions and non-available data, we
consider only the depositary type of banks, including commercial
banks, savings banks, and cooperative banks, which total 162
banks in 22 countries from 2003 to 2009. Our data dates back to
2003 because the EIRIS provides more complete evaluations for
the banking industry beginning that year.28
Corruption data are taken from Kaufman et al. (2009). Three
restriction regulations are taken from Barth et al. (2006, 2008).
Financial sector development is taken from the World Bank data-
base. Finally, the macroeconomic variables including GDPper and
GDPgrow are taken from the International Financial Statistics data-
base. Table 2 lists detailed denitions of the study variables.
Table 3 reports the matrix of correlation coefcients among the
independent variables. Almost all correlation coefcients are less
than 0.5, except for the value (0.723) between Res_S and Res_E.
The result of the low correlations among our independent variables
may remove the problem of multicollinearity in our multilevel
choice model.
Table 4 lists the basic statistics for ve performance (Part I) and
eleven explanatory variables (Part II). In addition, Table 4 compares
the performance differences across four CSR regimes. The discus-
sion below is mainly based on median values because they are
not affected by outliers. We also report the means and standard
deviations. In addition, we perform a t-test to examine the signif-
icance of the differences.
There are four interesting ndings in our basic statistics. First,
contrasting performance differences are found in low CSR regimes
such as 1 and 2 and high CSR regimes such as 4. That is, in terms of
the medians of ROA, ROE, NPL and NonII, banks conducting the
least (regime 1 or 2) and the most CSR activities (regime 4) exhibit
the worst and best performances, respectively. Thus, the perfor-
mance of banks in CSR regime 3 falls between the two extremes.
28
We do not investigate different types of banks for two reasons. First, commercial
banks, savings institutions, and credit unions perform similar functions of accepting
deposits and making loans (Saunders and Cornett, 2010). Next, it is difcult to
distinguish them in a global study because the functions of banks are dynamic. Many
saving and cooperative banks are transformed into commercial banks but retain the
names savings and cooperative. For example, the Taiwan Cooperative Bank was
set up to help credit unions in the early stages of its establishment but gradually
became a pure commercial bank. In addition, some banks names contain the word
cooperative but this simply refers to cooperation and does not imply it has any
connection with a cooperative union. As another example, the Shanghai Commercial
and Savings Bank is a pure commercial bank because the word savings is purely
descriptive. Thus, databases such as Bankscope may wrongly classify the bank types
because they may not know the historical changes that have taken place in many
countries. Thus, we do not classily the bank types.
3537
Generally speaking, except for NII, better CSR reveals better perfor-
mance, suggesting a positive relationship between CSR and FP.
With respect to NII, banks in CSR regime 1 have the highest NII
and banks in CSR regime 4 the lowest one. These basic statistics
indicate that banks participating in fewer CSR activities are prone
to conducting more loan-deposit business and, by contrast, banks
favoring more CSR activities conduct more fee and commission
business. Hence, by using basic statistics, except for NII, the strate-
gic motive is the driving force behind banks conducting CSR.
The t-tests of the performance differences between the CSR re-
gimes with the fewest and most activities are all signicant. For
example, except for ROA and NII, the performance differences be-
tween 4 and 1 + 2, and 4 + 3 and 1 + 2 are signicant for ROE,
NPL and NonII, which is consistent with our theoretical model. This
supports the notion that banks engaging in more CSR activities
experience better performance.
Second, as to the characteristic variables, contrasting differ-
ences between the two extremes are also found. Banks in CSR re-
gime 1 have the smallest TA, CostInc and Coverage, and highest
LoanDep, whereas those in CSR regime 4 have the highest TA and
the lowest Leverage and LoanDep. Thus, small and large banks
have contrasting behavior related to engaging in CSR activities.
One of the reasons why large banks engage in more CSR activities
is probably because they have more capacity to do so. Our results
are consistent with the basic statistics of Scholtens and Dams
study (2007), which show that banks adopting the Equator Princi-
ple have a large asset size and lower equity/TA. In addition, large
banks demonstrate lower loan-to-deposit ratios, which is consis-
tent with the result that they have the lowest NII. Banks in regime
2 have the highest CostInc and Leverage, which is consistent with
the result that they have the lowest ROA and ROE.
The third component pertains to country institutional variables
and nancial sector development. Banks in CSR regimes 1 and 4 are
more likely to be located in countries with more and less corrup-
tion, respectively. In addition, banks in CSR regime 4 are located
in countries with fewer restrictions on banking activities, but
banks in CSR regime 1 are located in countries with more restric-
tions. Banks in CSR regime 2 are located in countries that depend
more on bank credit than banks in CSR regime 3. Banks in CSR re-
gimes 3 and 4 are more likely to be located in less competitive
markets.
Finally, as to macroeconomic variables, banks engaging in the
most CSR activities are from countries with second higher GDP
growth rates and the highest per capita income, while banks
engaging in the fewest CSR activities are from countries with the
highest GDP growth rate and lowest per capita income.
In short, for both performance and characteristic variables,
banks adopting more CSR activities are considerably different from
those adopting fewer CSR activities. Banks conducting more CSR
activities are more likely to be more protable (except for NII)
and to have fewer nonperforming loans. In addition, they have
more assets and lower loan-to-deposit and Leverage ratios. They
are more likely to be in countries with less corruption and fewer
restrictions on banking activities, as well as higher GDP per capita.
5. Empirical results
5.1. The extended Heckman two-stage model
Table 5 presents the estimated results of the decision equation
(i.e., Eq. (7)), where the CSR index is the dependent variable using
the multinomial logit method. Note that when using the multino-
mial logit method, we obtain N  1 sets of coefcients, where N is
the number of regimes and regime 1 is used as the benchmark.
3538 M.-W. Wu, C.-H. Shen / Journal of Banking & Finance 37 (2013) 35293547

Table 2
Variable denitions.

Mnemonics Denitions Source

Performance variable
ROA (Net income/Total Asset)  100% BankScope
ROE (Net income/Total Equity)  100% BankScope
NPL (Non-Performing Loan/Total Loan)  100% BankScope
NII (Net interest income/(Net interest income + Non-interest income)  100% BankScope
NonII (Non-interest income/(Net interest income + Non-interest income)  100% BankScope
Independent variable
CSR CSR index is an aggregate measure the degree that a bank engage in CSR activity and ranging from 1 to 5. CSR index is 1 when a bank EIRIS
engages in least CSR while CSR index is 5 when a bank engages in most CSR.
logTA log(total asset) BankScope
Leverage (Equity/Total Asset)  100% BankScope
LoanDep (Loan/Deposit)  100% BankScope
CostInc (Overhead cost/total income)  100% BankScope
Coverage (Loan Loss Reserve/Non-Performing Loan)  100% BankScope
Res_S The degree of restriction on banking activities in securities, ranging from 1 (less restriction) to 4 (higher restriction). Barth et al. (2006,
2008)
Res_I The degree of restriction on banking activities in insurance, ranging from 1 (less restriction) to 4 (higher restriction). Barth et al. (2006,
2008)
Res_E The degree of restriction on banking activities in real estate, ranging from 1 (less restriction) to 4 (higher restriction). Barth et al. (2006,
2008)
Corrupt The degree of Control of Corruption, ranging from 3 (higher corruption) to +3 (less corruption). KKM (2009)
Credit_GDP Domestic credit to private sector/GDP The World Bank
P
HHI HerndahlHirschman Index. HHI ni1 S2i Si is market share for bank i. Higher HHI indicates less market competition and more BankScope
monopoly or oligopoly.
GDPper GDP based on current price/population IFS
GDPgrow GDP per capita growth rate IFS

Notes: Barth et al. (2006) characterize the regulatory situation at the end of 2002 with 152 respondents. Barth et al. (2008) provide similar data in the summer of 2007 with
responses from 142 countries.
KKM (2009): Kaufman et al. (2009).
IFS: International Financial Statistics.

Table 3
Correlation matrix.

logTA Leverage LoanDep CostInc Coverage Res_S Res_I Res_E Corrupt GDPgrow GDPper Credit_GDP EIRIS HHI
logTA 1 0.376 0.460 0.021 0.021 0.298 0.025 0.370 0.249 0.284 0.345 0.029 0.596 0.132
Leverage 0.376 1 0.365 0.067 0.285 0.113 0.124 0.297 0.106 0.079 0.005 0.113 0.214 0.079
LoanDep 0.460 0.365 1 0.189 0.208 0.328 0.238 0.452 0.368 0.011 0.255 0.037 0.330 0.131
CostInc 0.021 0.067 0.189 1 0.101 0.030 0.025 0.010 0.055 0.190 0.217 0.061 0.056 0.071
Coverage 0.021 0.285 0.208 0.101 1 0.020 0.124 0.091 0.082 0.123 0.083 0.010 0.046 0.010
Res_S 0.298 0.113 0.328 0.030 0.020 1 0.221 0.723 0.184 0.083 0.064 0.209 0.370 0.469
Res_I 0.025 0.124 0.238 0.025 0.124 0.221 1 0.057 0.372 0.025 0.378 0.613 0.053 0.380
Res_E 0.370 0.297 0.452 0.010 0.091 0.723 0.057 1 0.446 0.009 0.140 0.067 0.424 0.421
Corrupt 0.249 0.106 0.368 0.055 0.082 0.184 0.372 0.446 1 0.015 0.576 0.242 0.343 0.335
GDPgrow 0.284 0.079 0.011 0.190 0.123 0.083 0.025 0.009 0.015 1 0.369 0.207 0.273 0.046
GDPper 0.345 0.005 0.255 0.217 0.083 0.064 0.378 0.140 0.576 0.369 1 0.440 0.469 0.0003
Credit_GDP 0.029 0.113 0.037 0.061 0.010 0.209 0.613 0.067 0.242 0.207 0.440 1 0.076 0.486
EIRIS 0.596 0.214 0.330 0.056 0.046 0.370 0.053 0.424 0.343 0.273 0.469 0.076 1 0.222
HHI 0.132 0.079 0.131 0.071 0.010 0.469 0.380 0.421 0.335 0.046 0.0003 0.486 0.222 1

Thus, all the estimated coefcients are interpreted relative to the
benchmark regime.29
We only highlight the main results because the estimates of the
decision equations generate IMR, which is then inserted into the
performance equation to avoid bias. The coefcients of TA in CSR
regimes 2, 3, and 4 are all signicantly positive. This result indi-
cates that banks with larger total assets engage in more CSR activ-
ities. The coefcients of Leverage and CostInc are signicantly
positive in regime 2, and this nding suggests that banks with high
leverage and cost to income ratios conduct the second degree of
CSR activities. The coefcients of Coverage are signicantly posi-
29
The interpretation would not change, even if we use another number as the
benchmark (see Greene, 2003). This is similar to the situation where the interpre-
tation will not change in a simple logit model when either event is used as the
benchmark of unity.
tive in regimes 2, 3, and 4, suggesting that banks provide more loan
loss reserves when engaging in CSR activities to a considerable ex-
tent. The above results are also consistent with the basic statistics
to some extent. However, the insignicant positive coefcients of
LoanDep in CSR regimes 2, 3, and 4 indicate that banks that lend
more (relative to deposits) would not engage in more CSR activi-
ties. The results are similar to the basic statistics.
In addition, the coefcients of Res_I and Res_E are signicantly
negative, whereas the coefcients of Res_S are insignicant in most
of the specications. Thus, restrictions on bank activities in terms
of engaging in other related nancial sectors reveal different inu-
ences on the adoption of CSR. For example, banks permitted to en-
gage in insurance and real estate activities engage in less CSR
activities. The coefcients of Corrupt are signicantly negative,
indicating that banks in countries with less corruption engage in
more CSR activities. The coefcients of HHI in regimes 2, 3, and 4
 M.-W. Wu, C.-H. Shen / Journal of Banking & Finance 37 (2013) 35293547 3539

Table 4
Basic statistics of variables under various CSR indices: whole sample in 20032009.

Panel A: CSR index Panel B: t value

Variable 1 2 3 4 (4)  [(1) + (2)] [(4) + (3)]  [(1) + (2)]
Part I ROA Mean 0.59 0.57 0.63 0.61 0.34 0.99
Med 0.54 0.53 0.64 0.65
Std 0.65 1.09 0.53 0.57
ROE Mean 8.76 8.16 12.44 12.93 2.27*** 4.80***
Med 9.76 8.31 14.13 16.05
Std 17.06 10.63 9.71 13.96
NPL Mean 3.87 2.64 2.21 1.58 7.51*** 6.18***
Med 3.14 2.50 1.40 1.34
Std 2.91 1.96 1.95 1.16
NII Mean 1.97 1.81 1.56 1.27 6.01*** 6.98***
Med 1.84 1.59 1.47 1.08
Std 0.65 0.73 0.65 0.68
NonII Mean 0.77 1.01 1.06 1.13 2.41*** 2.84***
Med 0.67 0.79 0.94 1.19
Std 0.57 0.91 0.64 0.56

Part II TA(billion) Mean 68 151 551 1055 8.38*** 12.11***

Med 44 73 312 782
Std 79 232 612 800
Leverage Mean 6.12 7.29 5.73 4.27 10.21*** 3.58***
Med 5.84 6.89 4.77 4.22
Std 1.78 2.71 6.97 1.70
LoanDep Mean 72.52 69.77 63.64 51.01 6.12*** 7.15***
Med 71.78 70.77 65.46 49.57
Std 11.30 15.26 17.35 22.81
CostInc Mean 56.92 62.11 59.87 64.25 0.95 0.24
Med 57.41 59.97 57.96 58.35
Std 11.51 31.36 16.62 28.02
Coverage Mean 88.50 108.87 108.98 87.57 1.65 0.36
Med 57.27 69.42 73.02 68.53
Std 83.72 99.74 91.19 53.70
Res_S Mean 1.77 1.73 1.52 1.19 10.72*** 9.00***
Med 2.00 2.00 2.00 1.00
Std 0.42 0.44 0.50 0.39
Res_I Mean 2.61 2.28 2.47 2.48 1.43 2.35***
Med 2.00 2.00 2.00 2.00
Std 0.71 0.47 0.50 0.50
Res_E Mean 3.40 3.35 2.37 1.56 12.11*** 13.10***
Med 4.00 4.00 3.00 1.00
Std 1.18 1.15 1.34 1.14
Corrupt Mean 1.10 1.34 1.50 1.81 10.88*** 8.07***
Med 1.15 1.34 1.54 1.88
Std 0.49 0.49 0.59 0.37
Credit_GDP Mean 137.00 157.99 140.97 156.26 1.08 2.39***
Med 139.66 175.46 128.25 159.64
Std 43.74 39.70 41.97 35.07
HHI Mean 0.155 0.109 0.180 0.197 4.75*** 5.85***
Med 0.11 0.052 0.143 0.140
Std 0.198 0.120 0.156 0.115
GDPgrow Mean 2.57 0.79 0.49 0.77 1.87* 4.33***
Med 2.29 1.61 1.56 1.63
Std 1.82 2.83 2.83 2.44
GDPper Mean 27450.80 35988.28 38488.70 43746.19 8.76*** 9.91***
Med 27651.80 35641.12 38267.91 40556.05
Std 8253.67 7343.66 9725.95 11801.17

Notes: CSR index denotes the degree that banks engage in CSR, with 1 being the least, and 4 being the strongest engagement in CSR.
Mean, med, and std denote mean, median, and standard deviation of variables.
t value of (4)  [(1) + (2)] is the t-statistics used to examine the difference between CSR index 4 and the sum of CSR indices 1 and 2.
t value of [(4) + (3)]  [(1) + (2)] is the t-statistics used to examine the difference between the sum of CSR indices 3 and 4 and the sum of CSR indices 1 and 2.
*
Denote signicance at the 10% level.

Denote signicance at the 5% level.

***
Denote signicance at the 1% level.

are all signicantly positive, indicating that the increase of market activities (Fisman et al., 2006; Fernndez-Kranzl and Juan Santal,
concentration would encourage more CSR activities. This result 2010). Moreover, banks in countries with slow GDP growth and
contradicts the nding that concentration would increase CSR high GDP per capita engage in more CSR activities.
3540 M.-W. Wu, C.-H. Shen / Journal of Banking & Finance 37 (2013) 35293547

Table 5 the greenwashing motive is ruled out and only the strategic and
Multinomial logit model: First Step of the Extended Heckman Model (20032009) altruistic motives may drive banks to conduct CSR (see Hypotheses
PrCSRindex s a0 a1 log TAikt a2 Leverageikt a3 LoanDepikt a4 CostIncikt
a5 Coverageikt a6 Res Skt a7 Res Ikt a8 Res Ekt a9 Corruptkt a10 Credit GDPkt 1, 2 and 4). Furthermore, because CSR and FP (i.e., ROE and ROA)
a11 HHIkt a12 GDPGrowkt a13 GDPperkt likt .. are strongly positively linked, altruism is less possible for the driv-
ing motive (see Hypothesis 3). Therefore, we cannot reject the stra-
2 3 4
tegic motive argument, indicating that the strategic motive is the
Constant 15.521*** 33.045*** 51.550*** most likely reason for banks to engage in CSR (Baron, 2001; Dam
(3.149) (5.616) (5.044)
logTA 0.750*** 1.663*** 2.365***
et al., 2009).
(4.867) (8.147) (5.244) In addition, the coefcients of IMR are signicantly positive for
Leverage 0.291*** 0.071 0.057 the protability measures and negative for nonperforming loans.
(3.468) (0.649) (0.285) The most often used intuition behind IMR is the control of the
LoanDep 0.008 0.022 0.003
unobserved factor.30 Clatworthy et al. (2009) argue that IMR is a
(0.572) (1.093) (0.103)
CostInc 0.032*** 0.019 0.014 proxy for unobservable characteristics that affect both selection
(2.598) (1.338) (0.801) and performance equations. They then highlight that previous re-
Coverage 0.004*** 0.006*** 0.010* search does not entirely clarify what such unobservable characteris-
(2.187) (2.719) (1.814) tics represent or how important they are in systematically affecting
Rest_S 0.146 0.111 3.191
CSR selection and performance. Ireland and Lennox (2002) empha-
(0.141) (0.103) (1.615)
Rest_I 1.716*** 1.752*** 1.382 size that: . . .however, the unobservable characteristic is rather an
(2.912) (2.480) (1.582) abstract concept and very difcult to be materialized. In our study,
Rest_E 0.328 0.699** 0.069 the unobservable characteristics could be the banks knowledge of
(0.998) (2.033) (0.106)
management, concerns for reputation, and so on. The signicance
Corrupt 1.323* 2.924*** 0.123
(1.949) (3.919) (0.106) of these characteristics indicates that the performance equation is
Credit_GDP 0.010 0.007 0.007 systematically related to the unobservable characteristics of banks.
(1.062) (0.682) (0.603) Not adding this factor into the equations results in the lack of the
HHI 10.924*** 20.812*** 16.445*** third variable problem in econometrics, thus resulting in a biased
(2.731) (4.762) (3.091)
estimation. Then, including IMR becomes meaningful.
GDPgrow 0.307*** 0.246*** 0.200
(3.776) (2.416) (1.500) We adopt the method of Clatworthy et al. (2009) for calculating
GDPper 0.0002*** 0.0003*** 0.0003*** ATE and ATT to explore fully the CSR effect, which is referred to as
(4.778) (5.841) (4.389) the treatment effect. The counterfactual is calculated by inserting
Nobs 756 the means of the regresses of the CSR benchmark to other CSR in-
Pseudo-R2 0.698 dex regressions. Although the new specication includes IMR as
Notes: This table reports the estimated results of the multinomial logit model, the additional explanatory variable, ATE simply calculates the per-
which uses CSR index as the dependent variable. The CSR index, ranging from 1 to 4, formance differences between CSR and non-CSR banks without
denotes the degree that banks engage in CSR, with 1 being the least and 4 being the considering the indirect effect resulting from IMR (see Clatworthy
strongest engagement in CSR. Moreover, the CSR index of 1 is used as the et al., 2009, Table 8), which is equal to d  CSRindex. However, the
benchmark.
The Extended Heckman model involves two steps. The rst step is the multinomial
treatment effect (ATT) differs from the coefcients of the CSR dum-
logit model used in the study. The second step is the performance equation. The my variable (ATE) by an adjustment term, making the estimated
Extended Heckman model is similar to the conventional Heckman model, but the coefcients of interest incomplete measures of the treatment ef-
latter uses binary dependent variables. fect. Thus, ATT is calculated by inserting the means of the variables
The least square dummy variable (LSDV) approach is used to estimate the model.
of CSR banks.
Country dummies are skipped to save space. The data are available upon request.
Nobs denotes the number of observations. We use 2007 as the dividing period for the following reasons. In
Pseudo-R2 is the pseudo-R square of multinomial logit model. July 2007, As mentioned earlier, ATE simply calculates the perfor-
*
Denote signicance at the 10% level. mance differences between CSR and non-CSR banks without con-
**
Denote signicance at the 5% level. sidering the indirect effect resulting from IMR, which is equal to
***
Denote signicance at the 1% level.
d  CSRindex. Because d is signicant, ATE is also signicant. At the
bottom of Table 6, we further report ATT, which considers the com-
plete CSR effect calculated from Eq. (6). ATT is positively correlated
Once we estimate the decision equation, IMR is generated and for prot measures and negatively correlated for asset quality
then put into the performance equation (i.e., Eq. (8)). Table 6 pre- measures.
sents the estimated results of the performance equation with con- Coefcients of controllable variables are similar to those of past
sideration for IMR. Because the data are unbalanced, we do not use empirical results with few exceptions. First, the coefcients of TA
the panel xed or random effects. The t-statistics are generated by are signicantly positive for ROA and NPL, but negative for NII
considering a White heteroskedasticity-consistent estimate and and NonII. The results are similar to those in the literature, where
the clustering effect (Petersen, 2009). large assets commonly have a positive impact on prot (Goddard
Interestingly, Table 6 shows that the coefcients of the CSR in- et al., 2004; Bonin et al., 2005). In addition, our results show that
dex (i.e., d) are overwhelmingly signicantly positive for four prof- restrictions on bank activities in securities and insurance improve
itability measures (ROA, ROE, NII, and NonII) and negative for performance, but restrictions on real estate hurt performance.
nonperforming loans (NPL). Thus, CSR and FP are positively associ-
ated in our sample banks. These results are consistent with the 30
To our best knowledge, IMR does not have signicant economic interpretation
idea that more CSR activities improve performances (Shen and but econometric interpretation. IMR is added because it is the correct term for self-
Chang, 2009, forthcoming-a; Simpson and Kohers, 2002; Scholtens selection bias and is produced by the rst-stage regression of the Heckman two-stage
and Dam, 2007; Scholtens, 2009; Harjoto and Jo, 2008). That is, CSR model. Accordingly, the signicance of the IMR coefcient justies that the bias is
may affect both the cost and revenue functions, which is consistent removed. This econometric background was developed by the Nobel Prize winner,
Heckman (1978), and is well documented in a large number of econometric books,
with our theoretical argument. When a bank engages in CSR, such as that by Greene (2003). Application studies typically cite Heckmans work to
although costs increase, revenues increase even more. Accordingly, justify the validity of empirical models. A number of applied studies add simple
because of the positive relationship between CSR and NII and NonII, economic intuition similar to our previous version.
 M.-W. Wu, C.-H. Shen / Journal of Banking & Finance 37 (2013) 35293547 3541

Table 6
Performance equation: second step of the extended Heckman model (20032009) yikt dCSRIndex;ikt b0 b1 logTAikt b2 Leverageikt b3 LoanDepikt
b4 CostInckt b5 Res Skt b6 Res Ikt b7 Res Ekt b8 Corruptkt b9 Credit GDPkt b10 HHIkt b11 GDPgrowkt b12 GDPperkt kIMRikt eikt .

ROA ROE NPL NII NonII

CSR 0.097*** 2.841*** 1.236*** 0.135*** 0.291***
(2.931) (4.426) (9.761) (4.096) (7.074)
Constant 2.098*** 16.094 0.542 2.795*** 1.845***
(3.513) (1.382) (0.235) (4.758) (2.479)
logTA 0.038*** 0.199 0.302*** 0.048*** 0.079***
(3.303) (0.889) (6.845) (4.207) (5.490)
Leverage 0.005*** 0.030 0.030*** 0.013*** 0.012***
(3.809) (1.102) (5.535) (9.436) (6.830)
LoanDep 0.058*** 0.024 0.120*** 0.084*** 0.046***
(5.511) (0.118) (2.988) (8.040) (3.494)
CostInc 0.027*** 0.261*** 0.020*** 0.004*** 0.011***
(32.749) (16.636) (6.594) (4.392) (11.307)
Rest_S 0.216*** 4.591*** 0.252 0.451*** 0.518***
(3.513) (3.849) (1.070) (7.371) (6.752)
Rest_I 0.382*** 5.467*** 0.222 0.497*** 0.921***
(6.779) (5.005) (1.032) (8.883) (13.128)
Rest_E 0.058*** 1.661*** 0.226*** 0.066*** 0.036
(2.257) (3.320) (2.286) (2.560) (1.104)
Corrupt 0.104 1.367 1.799*** 0.047 0.558***
(1.649) (1.122) (7.483) (0.746) (7.111)
Credit_GDP 0.0004 0.003 0.003 0.002*** 0.005***
(0.543) (0.178) (1.222) (2.745) (5.568)
HHI 0.988*** 9.712 8.674*** 2.198*** 3.563***
(2.951) (1.498) (6.780) (6.607) (8.545)
GDPgrow 0.104*** 1.674*** 0.471*** 0.071*** 0.105***
(8.599) (7.177) (10.229) (5.918) (7.050)
GDPper 0.00001 0.0001 0.0001*** 0.00003*** 0.0001***
(1.262) (1.105) (6.751) (6.530) (9.195)
IMR 0.165*** 2.047** 1.844*** 0.435*** 0.794***
(3.136) (2.011) (9.184) (8.336) (12.157)
R_square 0.688 0.427 0.354 0.556 0.394
Nobs 756 756 756 753 753
White Test for Heteroskedasticity 63.744 85.928 106.462 121.073 100.522
[0.966] [0.482] [0.067] [0.008] [0.136]
ATT 2.240 19.795 6.402 3.420 3.107

Notes: Performance Equation estimates CSR effects on bank performances, which are proxied by ROA, ROE, NPL, NII, and NonII. This estimate is the
second step of the Extended Heckerman model.
The Performance Equation considers IMR as the additional variable to remove the estimate bias, where the IMR is generated by the rst step of
the Extended Heckman model.
The least square dummy variable (LSDV) approach is used to estimate the model. Country dummies are skipped to save space. The data are
available upon request.
We examined the heteroskedasticity using the White method. We regress the transformed dependent variables on the transformed cross-product
terms of all independent variables, where the transformed variable denotes each relevant variable to be divided by residuals. The likelihood
multiplier of Nobs  R2 is reported. The t-values are reported in brackets.
All explanatory variable denitions are shown in Table 2.

Denote signicance at the 10% level..
**
Denote signicance at the 5% level.
***
Denote signicance at the 1% level.

These results differ from Shen and Chang (2006), where restric-
tions on bank activities in securities and insurance reduce perfor-
mance. This result can be attributed to the selection of the 46
countries that are spread across different regions. In contrast, we
only considered 22 countries that are mostly located in Europe be-
cause of the data provided by the EIRIS survey. These differences
merit future studies. The coefcients of HHI are signicantly nega-
tive for ROA, ROE, NII, and NonII, and positive for NPL. The results
are consistent with a past study that indicated banks would gain
more prot in highly concentrated markets (Wu and Shen, 2011).
We also check the residuals using Whites (1980) heteroskedas-
ticity test. We nd that all the chi-square statistics are insigni-
cant, suggesting a lack of heteroskedasticity. Moreover, the
country dummies are not reported here, but are available upon re-
quest. Most of our country dummy variables are signicant, indi-
cating the existence of an individual effect. Hence, the
performance of banks in each country is different. Because adding
these dummies does not affect the coefcients with which we are
concerned, our results are robust to the country individual effect.
5.2. Robustness testing
We consider three types of robust testing to examine the esti-
mated results. One is the impact of the 2008 subprime crisis. The
theoretical model of Brilius (2010) sets up a theoretical model to
investigate the effect of a nancial crisis on CSR. The study of Karai-
brahimoglu (2010) uses Fortune 500 companies as the sample for
the years 2007 and 2008, and concludes that decreases in CSR pro-
jects are due to nancial downturn. The decrease is larger in the
USA than in Europe and other countries. Given that the perfor-
mance of many large banks deteriorated considerably during the
crisis, we wonder whether our results would hold when the subs-
amples are used to examine the issues. Hence, it is also interesting
to know whether our results are robust or not in times of crisis. We
3542 M.-W. Wu, C.-H. Shen / Journal of Banking & Finance 37 (2013) 35293547

Table 7 Table 8
Robustness test 1: performance equation before the sub-prime crisis (20032006). Robustness test 1: performance equation after the subprime crisis (20072009).

ROA ROE NPL NII NonII ROA ROE NPL NII NonII
*** *** *** *** *** *** ***
CSR 0.100 2.334 0.941 0.129 0.215 CSR 0.006 2.837 0.635 0.024 0.111**
(2.835) (2.453) (5.803) (2.998) (4.227) (0.127) (4.444) (4.609) (0.553) (2.045)
Constant 0.096 23.711 7.666*** 3.244*** 2.117** Constant 3.271*** 60.608*** 15.138*** 1.235 0.608
(0.130) (1.193) (2.275) (3.669) (1.997) (2.592) (3.747) (4.372) (1.092) (0.434)
logTA 0.001 0.098 0.131* 0.067*** 0.030 logTA 0.099*** 0.871*** 0.094*** 0.012 0.004
(0.069) (0.247) (1.934) (3.712) (1.408) (6.492) (4.430) (2.225) (0.859) (0.263)
Leverage 0.087*** 0.210 0.040 0.082*** 0.050*** Leverage 0.041*** 0.080 0.167*** 0.124*** 0.122***
(7.709) (0.685) (0.766) (5.939) (3.044) (2.619) (0.395) (3.844) (8.951) (7.145)
LoanDep 0.001 0.048 0.027*** 0.015*** 0.011*** LoanDep 0.008*** 0.030 0.025*** 0.011*** 0.010***
(0.349) (1.049) (3.448) (7.295) (4.272) (3.299) (0.942) (3.619) (4.918) (3.641)
CostInc 0.018*** 0.317*** 0.046*** 0.004 0.013*** CostInc 0.026*** 0.226*** 0.001 0.000 0.004***
(8.834) (5.671) (4.776) (1.645) (4.256) (29.437) (20.069) (0.451) (0.025) (4.391)
Rest_S 0.217*** 6.068*** 0.094 0.492*** 0.443*** Rest_S 0.126 2.564** 0.068 0.387*** 0.391***
(2.952) (3.051) (0.276) (5.415) (4.116) (1.307) (2.051) (0.250) (4.514) (3.691)
Rest_I 0.356*** 6.901*** 0.321 0.611*** 0.883*** Rest_I 0.185*** 2.296*** 0.299 0.168*** 0.483***
(4.876) (3.501) (0.956) (6.846) (8.337) (2.486) (2.376) (1.435) (2.535) (5.902)
Rest_E 0.095*** 2.294*** 0.552*** 0.036 0.022 Rest_E 0.050 1.611*** 0.011 0.148*** 0.137***
(3.185) (2.841) (4.015) (0.995) (0.516) (1.323) (3.283) (0.104) (4.411) (3.302)
Corrupt 0.171** 0.614 1.361*** 0.191* 0.777*** Corrupt 0.129 1.070 2.368*** 0.333*** 0.266***
(2.072) (0.275) (3.576) (1.875) (6.431) (1.252) (0.801) (8.218) (3.634) (2.352)
Credit_GDP 0.001 0.007 0.013*** 0.002 0.006*** Credit_GDP 0.002 0.051*** 0.016*** 0.001 0.005***
(0.762) (0.217) (2.548) (1.306) (3.931) (1.442) (3.355) (4.768) (1.113) (3.665)
HHI 1.528*** 6.112 15.599*** 4.125*** 5.464*** HHI 0.698 2.028 2.064 0.609 1.660***
(3.079) (0.456) (6.829) (6.771) (7.568) (1.294) (0.290) (1.368) (1.271) (2.800)
GDPgrow 0.056*** 0.625 0.336*** 0.101*** 0.032 GDPgrow 0.102*** 2.017*** 0.552*** 0.021 0.086***
(2.828) (1.169) (3.693) (4.164) (1.110) (5.847) (8.911) (11.301) (1.356) (4.461)
GDPper 0.00001* 0.0001 0.0002*** 0.00004*** 0.0001*** GDPper 0.000005 0.0002*** 0.0001*** 0.000002 0.00002***
(1.930) (0.348) (5.520) (5.253) (8.351) (0.680) (2.115) (6.784) (0.305) (2.458)
IMR 0.176*** 1.418 1.759*** 0.372*** 0.699*** IMR 0.047 2.485*** 1.317*** 0.113** 0.439***
(3.777) (1.127) (8.205) (6.536) (10.364) (0.735) (3.015) (7.411) (2.004) (6.281)
R_square 0.471 0.194 0.469 0.559 0.453 R_square 0.783 0.700 0.455 0.560 0.424
Nobs 421 421 421 418 418 Nobs 335 335 335 335 335
ATT 1.084 22.181 11.991 2.865 2.359 ATT 3.190 27.431 2.521 0.258 6.650

Notes: Performance Equation estimates CSR effects on bank performances, which
are proxied by ROA, ROE, NPL, NII, and NonII. This estimate is the second step of the
Extended Heckerman model.
The Performance Equation considers IMR as the additional variable to remove the
estimate bias, where the IMR is generated by the rst step of the Extended Heck-
man model.
The least square dummy variable (LSDV) approach is used to estimate the model.
Country dummies are skipped to save space. The data are available upon request.
We examined the heteroskedasticity using the White method. We regress the
transformed dependent variables on the transformed crossproduct terms of all
independent variables, where the transformed variable denotes each relevant var-
iable to be divided by residuals. The likelihood multiplier of Nobs  R2 is reported.
The t-values are reported in brackets.
All explanatory variable denitions are shown in Table 2.
* 
Denote signicance at the 10% level.
**
Denote signicance at the 5% level.
***
Denote signicance at the 1% level.
Notes: Performance Equation estimates CSR effects on bank performances, which
are proxied by ROA, ROE, NPL, NII, and NonII. This estimate is the second step of the
Extended Heckerman model.
The Performance Equation considers IMR as the additional variable to remove the
estimate bias, where the IMR is generated by the rst step of the Extended Heck-
man model.
The least square dummy variable (LSDV) approach is used to estimate the model.
Country dummies are skipped to save space. The data are available upon request.
We examined the heteroskedasticity using the White method. We regress the
transformed dependent variables on the transformed cross-product terms of all
independent variables, where the transformed variable denotes each relevant var-
iable to be divided by residuals. The likelihood multiplier of Nobs  R2 is reported.
The t-values are reported in brackets.
All explanatory variable denitions are shown in Table 2.
Denote signicance at the 10% level.
**
Denote signicance at the 5% level.
***
Denote signicance at the 1% level.

FP. The coefcients of the controllable variables to some extent

separate our sample into two subsamples, 20032006, and 2007
2009.31
Tables 7 and 8 report the estimated results by using two subs-
amples: 20032006 and 20072009, respectively. Our coefcients
of concern for the CSR index remain signicantly positive for two
protability measures (ROA and ROE) and earning measure (NII
and NonII) and negative for non-performing loans, regardless of
the sub-sample periods. The CSR effect on ROA and NII turns to
be insignicantly positive and negative during subprime period.
In addition, ATT is positive. Thus, CSR is positively linked with
31
We use 2007 as the dividing period for the following reasons. In July 2007, the
collapse of Bear Stearns Companies, Inc. shocked the market because it was one of the
top ve investment banks in the United States. Despite its efforts to recover, Bear
Stearns was eventually acquired by JP Morgan at a price of $2 per share under the
supervision of the Federal Reserve.
also become more consistent with previous studies.
The second robustness test examines whether the CSR coef-
cients in four regimes would be signicantly different from each
other. We create four CSR dummy variables, namely, DCSR1, DCSR2,
DCSR3 and DCSR4, which denote that CSR is equal to unity if a bank
is ranked as CSR index 1, 2, 3, and 4, respectively.
Table 9 reports the estimated results of using the four CSR dum-
my variables (DCSR) in the performance equation. The new results
are still consistent with our hypotheses. For example, when ROA
is used as the dependent variable, the coefcients of the four dum-
mies are signicant to be 2.292, 2.545, 2.666, and 2.859, respec-
tively. Hence, the coefcient size increases with the increase in
CSR index, supporting the argument that a larger CSR index has a
larger positive impact on ROA. Moreover, the increasing coefcient
size suggests that the impacts of CSR on the performance are lin-
ear. Similar results are found when ROE, NII, and NonII are used
as the dependent variables. The coefcient size of NPL decreases
 M.-W. Wu, C.-H. Shen / Journal of Banking & Finance 37 (2013) 35293547 3543

Table 9
Robustness test 2: four CSR dummy variables in performance equation yikt a1 DCSR1 a2 DCSR2 a3 DCSR3 a4 DCSR4 b1 logTAikt b2 Leverageikt b3 LoanDepikt b4 CostInckt
b5 Res Skt b6 Res Ikt b7 Res Ekt b8 Corruptkt b9 Credit GDPkt b10 HHIkt b11 GDPgrowkt b12 GDPperkt kIMRikt eikt .

ROA ROE NPL NII NonII

DCSR1 2.292*** 19.510 0.812 3.129*** 1.627***
(3.682) (1.604) (0.340) (5.144) (2.101)
DCSR2 2.545*** 23.129* 2.474 3.289*** 1.290
(3.914) (1.820) (0.991) (5.173) (1.594)
DCSR03 2.666*** 26.579** 3.266 3.524*** 1.156
(3.945) (2.012) (1.258) (5.335) (1.375)
DCSR4 2.859*** 29.852*** 4.711* 4.014*** 0.969
(4.027) (2.151) (1.727) (5.783) (1.097)
logTA 0.059** 0.530 0.293*** 0.175*** 0.002
(2.004) (0.924) (2.598) (6.090) (0.043)
Leverage 0.019 0.233 0.153*** 0.045*** 0.064***
(1.322) (0.827) (2.757) (3.185) (3.570)
LoanDep 0.007*** 0.043 0.031*** 0.012*** 0.011***
(4.635) (1.458) (5.391) (8.234) (5.846)
CostInc 0.029*** 0.279*** 0.021*** 0.007*** 0.010***
(27.755) (13.637) (5.291) (6.423) (7.414)
Rest_S 0.181*** 4.276*** 0.303 0.411*** 0.557***
(2.903) (3.505) (1.266) (6.710) (7.142)
Rest_I 0.455*** 5.945*** 0.298 0.544*** 0.899***
(7.685) (5.136) (1.308) (9.378) (12.162)
Rest_E 0.033 1.463*** 0.238*** 0.039 0.056*
(1.253) (2.804) (2.322) (1.499) (1.687)
Corrupt 0.165*** 1.866 1.755*** 0.033 0.499***
(2.534) (1.466) (7.019) (0.518) (6.131)
Credit_GDP 0.001 0.005 0.004 0.002*** 0.005***
(1.045) (0.335) (1.415) (2.798) (5.535)
HHI 1.860*** 16.364** 8.483*** 3.367*** 2.805***
(4.494) (2.023) (5.337) (8.309) (5.436)
GDPgrow 0.124*** 1.810*** 0.489*** 0.089*** 0.096***
(9.498) (7.096) (9.748) (6.960) (5.942)
GDPper 0.00002*** 0.0002* 0.0001*** 0.00005*** 0.00005***
(3.255) (1.707) (5.612) (8.032) (6.233)
IMR 0.362*** 3.448*** 1.908*** 0.672*** 0.659***
(4.892) (2.386) (6.718) (9.295) (7.159)
Pseudo-R2 0.707 0.707 0.707 0.707 0.707
Nobs 756 756 756 756 756
t-test or F-test for the equality of coefcients of CSR dummies
H0: a1 = a2 t(739) = 4.22*** t(739) = 3.08*** t(739) = 7.20*** t(736) = 2.70*** t(736) = 4.50***
H0: a2 = a3 t(739) = 2.09*** t(739) = 3.05*** t(739) = 3.56*** t(736) = 4.17*** t(736) = 1.86*
H0: a3 = a4 t(739) = 2.13*** t(739) = 1.84*** t(739) = 4.14*** t(736) = 5.51*** t(736) = 1.65*
H0: a1 = a2 = a3 = a4 F(3, 739) = 7.55(6E05)*** F(3, 739) = 6.43(2E04)*** F(3, 739) = 22.82(1E09)*** F(3, 736) = 16.92(1E09)*** F(3, 736) = 7.72(4E05)***

Notes: The Extended Heckman model involves two steps. The rst step is the multinomial logit model used in the study. The second step is the performance equation. The
least square dummy variable (LSDV) approach is used to estimate the model. Country dummies are skipped to save space. The data are available upon request.
Nobs denotes the number of observations.
Pseudo-R2 is the pseudo-R square of multinomial logit model.
For the test of H0: a1 = a2 = a3 = a4, the values in the parentheses are the signicant level for F-test.
*
Denote signicance at the 10% level.
**
Denote signicance at the 5% level.
***
Denote signicance at the 1% level.

when CSR index increases, supporting the argument that a larger
CSR has a larger negative impact on NPL.
Next, we examine the equality among the four coefcients. The
equal coefcients would reject the hypothesis that a larger CSR has
a larger effect on performance. Two types of tests are conducted.
One is the three-pair tests among the four coefcients, namely,
a1 = a2, a2 = a3, and a3 = a4, and the other is the joint test of
a1 = a2 = a3 = a4. The bottom part of Table 9 reports the testing re-
sults. Still employing ROA as the example, the t-values for the
three-pair tests are 4.22, 2.09, and 2.13, respectively, and the F-va-
lue for the joint test is 7.55. All tests are signicant at the 5% level.
Similar results are obtained when other performance measures are
used. Hence, the CSR coefcient in each regime is statistically dif-
ferent and we cannot reject the argument that a larger CSR has a
larger inuence on performance.
We started from the linear relationship in Section 4 because
most of the relevant studies adopt this specication except for
few exceptions. For example, Barnett and Salomon (2002, 2006)
found a curvilinear relationship between CSR and the types of so-
cial screenings in mutual funds by using 12 social screening in-
dexes. Our third robustness test considers the nonlinear
possibility in our regression analysis. We consider two types of
nonlinear forms. First, we classify CSR banks into 12 regimes, in
which the numbers of observations in these regimes are 10, 230,
172, 204, 136, 99, 79, 47, 24, 9, 1, and 1, respectively. Considering
that the multinomial logit model requires a sufcient degree of
freedom in each regime; therefore we combine regimes 1 and 2,
and combine regimes 8, 9, 10, 11, and 12, resulting in seven CSR re-
gimes to ensure that each regime has at least 50 observations.
Table 10 report the estimated results of performance equations.
The estimated result of decision equation is provided upon request.
In Table 10, the estimation results also remain similar to those
using four regimes, except that the ROA coefcient slightly changes
from signicantly positive to insignicantly positive. The reason
3544 M.-W. Wu, C.-H. Shen / Journal of Banking & Finance 37 (2013) 35293547

Table 10
Robustness test 3: seven CSR regimes in performance equation.

ROA ROE NPL NII NonII

*** *** ***
CSR 0.014 1.053 0.507 0.068 0.071 ***
(0.774) (3.025) (7.106) (3.714) (2.904)
Constant 0.181 2.408 10.518*** 0.815* 5.310***
(0.412) (0.285) (6.195) (1.828) (9.517)
logTA 0.069 0.114 0.718 *** 0.317 *** 0.092
(1.021) (0.087) (2.673) (4.575) (1.001)
Leverage 0.057 *** 0.432 *** 0.147 *** 0.018 * 0.007
(5.529) (2.175) (3.622) (1.736) (0.497)
LoanDep 0.084 *** 0.251 0.044 0.104 *** 0.133 ***
(7.701) (1.194) (1.027) (9.372) (9.025)
CostInc 0.006 *** 0.036 0.021 *** 0.011 *** 0.016 ***
(4.425) (1.332) (3.790) (7.765) (8.297)
Rest_S 0.025 *** 0.244 *** 0.013 *** 0.003 *** 0.006 ***
(25.767) (13.275) (3.398) (3.374) (5.007)
Rest_I 0.218 *** 4.589 *** 0.579 *** 0.576 *** 0.598 ***
(3.449) (3.769) (2.319) (8.906) (6.991)
Rest_E 0.222 *** 3.665 *** 0.773 *** 0.298 *** 0.382 ***
(4.298) (3.674) (3.781) (5.649) (5.474)
Corrupt 0.083 *** 1.858 *** 0.546 *** 0.148 *** 0.167 ***
(3.351) (3.888) (5.576) (5.843) (4.977)
Credit_GDP 0.006 0.024 0.785 *** 0.358 *** 0.081
(0.107) (0.022) (3.462) (6.067) (1.036)
HHI 0.001 0.005 0.007 *** 0.00001 0.00002
(1.073) (0.445) (2.711) (0.018) (0.023)
GDPgrow 0.256 8.756 3.359 0.729 0.263
(0.477) (0.845) (1.581) (1.327) (0.362)
GDPper 0.075 *** 1.416 *** 0.210 *** 0.014 0.017
(8.158) (8.015) (5.792) (1.497) (1.375)
IMR 0.00001 ** 0.00003 0.00001 0.00001 0.000002
(1.978) (0.501) (0.789) (1.576) (0.388)
R_square 0.684 0.426 0.301 0.524 0.277
Nobs 756 756 756 753 753
ATT 0.164 1.347 2.352 0.883 0.306

Notes: Performance Equation estimates CSR effects on bank performances, which are proxied by ROA, ROE, NPL, NII, and NonII. This estimate is the second step of the Extended
Heckerman model.
The Performance Equation considers IMR as the additional variable to remove the estimate bias, where the IMR is generated by the rst step of the Extended Heckman model.
The least square dummy variable (LSDV) approach is used to estimate the model. Country dummies are skipped to save space. The data are available upon request.
We examined the heteroskedasticity using the White method. We regress the transformed dependent variables on the transformed cross-product terms of all independent
variables, where the transformed variable denotes each relevant variable to be divided by residuals. The likelihood multiplier of Nobs  R2 is reported. The t-values are
reported in brackets.
All explanatory variable denitions are shown in Table 2.
The result of decision equation is provided upon request.
*
Denote signicance at the 10% level.
**
Denote signicance at the 5% level.
***
Denote signicance at the 1% level.

can be attributed to a lesser degree of freedom.32 Next, we add CSR
squares in our regression analysis. The coefcients of the CSR
squares are insignicant, rejecting the nonlinear form.33
6. Conclusion
This study investigates the association between CSR and FP in
global banking sector and discusses the motives driving banks to
engage in CSR. Our theoretical model for banking industry suggests
that CSR could facilitate brand differentiation for strategic and
altruistic banks, but not for greenwashing banks. Strategic banks
32
There is a fundamental difference between Barnett and Salomons (2006, BS) CSR
social funds and our CSR banking study. Funds produce a large amount of data
because they are transacted daily in the market. With monthly data from 1972 to
2000, fund-month observations reach 22,512 fund-month observations (67
funds  28 years  12 months). Thus, each regime can have 1876 observations on
average. Our accounting nal ratios of CSR banks are annual data, which were only
recently available from 2003 to 2009, resulting in a total of only 1012 observations.
Then, we have 234 observations in regime 1, 481 observations in regime 2, 233
observations in regime 3 and 64 observations in regime 4). Hence, it is not easy to
apply Barnett and Salomons approach in this study.
33
Results are available upon request.
utilize this advantage to increase NII, NonII, and net prots (ROA
and ROE), as well as decrease NPL. Although altruistic banks are
not interested in utilizing these advantages, their products are dif-
ferent from the perspectives of clients, thus resulting in an increase
in NII and NonII. However, engaging in the altruistic CSR also in-
creases cost; hence, the net effect on prot (ROA and ROE) is uncer-
tain. Greenwashing banks provide only lip service, which does not
affect their income and cost. Depending on the driving motives of
strategic, altruistic, and greenwashing, CSR could generate a posi-
tive, uncertain, or no inuence on net income.
To explore this issue, we extend the Heckman two-step method
to minimize the selection bias given the multi-level CSR data. Past
studies using the similar type of data often ignore this bias. Our
empirical results nd that CSR positively affects NII, NonII, ROA,
and ROE, and negatively affects NPL. Our evidence is consistent
with existing theories of nancial intermediarys reputation, be-
cause we show that bank with higher CSR would have higher
nancial earnings and asset quality. Hence, in terms of prot mea-
sures, banks conducting more CSR activities outperform those
banks that do not engage in CSR. The strategic motive seems to
be predominant in the banking sector that engages in CSR. By using
the aggregate CSR index, we can conclude that strategic motives
mainly drive the CSR activities of banks.
Appendix A. Example of CSR index transformation from EIRIS survey

Questionnaire Q11 Q22 Q33 Q44 Q55 Q66

Reply choices by text-grade Major 3 Excep- 3 Advanced 5 No convictions 0 Advanced 5 Up to 1

ratings and their improve tional 100 m
transformed number-
grade ratings
Signicant 2 Good 2 Good 4 Fines in 1 Good 4 More 2
improve lower half than
100 m
Minor 1 Moderate 1 Inter- 3 Fines in second 2 Inter- 3
improve mediate quartile mediate
Basic 2 Fines in top 3 Basic 2
quartile
Limited 1 Limited 1
Step 1: Transform banks reply from text-grade rating to number-grade rating
HSBC Holdings 2005 Major improve 3 Moderate 1 Advanced 5 No convictions 0 Advanced 5 More than 100 m 2
HSBC Holdings 2006 Major improve 3 Good 2 Good 4 No convictions 0 Advanced 5 More than 100 m 2
Standard Chartered 2005 Minor improve 1 Good 2 Intermediate 3 No convictions 0 Advanced 5 More than 100 m

M.-W. Wu, C.-H. Shen / Journal of Banking & Finance 37 (2013) 35293547
2
Standard Chartered 2006 Minor improve 1 Good 2 Good 4 No convictions 0 Advanced 5 More than 100 m 2
Barclays 2005 Minor improve 1 Exceptional 3 Intermediate 3 Fines in lower half 1 Advanced 5 Up to 100 m 1
Barclays 2006 Minor improve 1 Exceptional 3 Good 4 No convictions 0 Advanced 5 Up to 100 m 1
Step 2: Extend the obtained numerical values into ve scales 15 by standardization method7
HSBC Holdings 2005 [(3  1)/(3  1)]  4 + 1= 5 [(1  1)/ 1 [(5  1)/ 5 0 0 [(5  1)/ 5 (1)  {[(2  (1))/ 5
(3  1)]  4 + 1= (5  1)]  4 + 1= (5  1)]  4 + 1= ((2)
 (1))]  4 + 1}=
HSBC Holdings 2006 [(3  1)/(3  1)]  4 + 1= 5 [(2  1)/ 3 [(4  1)/ 4 0 0 [(5  1)/ 5 (1)  {[(2  (1))/ 5
(3  1)]  4 + 1= (5  1)]  4 + 1= (5  1)]  4 + 1= ((2)
 (1))]  4 + 1}=
Standard Chartered 2005 [(1  1)/(3  1)]  4 + 1= 1 [(2  1)/ 3 [(3  1)/ 3 0 0 [(5  1)/ 5 (1)  {[(2  (1))/ 5
(3  1)]  4 + 1= (5  1)]  4 + 1= (5  1)]  4 + 1= ((2)
 (1))]  4 + 1}=
Standard Chartered 2006 [(1  1)/(3  1)]  4 + 1= 1 [(2  1)/ 3 [(4  1)/ 4 0 0 [(5  1)/ 5 (1)  {[(2  (1))/ 5
(3  1)]  4 + 1= (5  1)]  4 + 1= (5  1)]  4 + 1= ((2)
 (1))]  4 + 1}=
Barclays 2005 [(1  1)/(3  1)]  4 + 1= 1 [(3  1)/ 5 [(3  1)/ 3 (1)  {[(1  (1)) 1 [(5  1)/ 5 (1)  {[(1  (1))/ 1
(3  1)]  4 + 1= (5  1)]  4 + 1= / (5  1)]  4 + 1= ((2)
((3)  (1))]  4 + 1}=  (1))]  4 + 1}=
Barclays 2006 [(1  1)/(3  1)]  4 + 1= 1 [(3  1)/ 5 [(4  1)/ 4 0 0 [(5  1)/ 5 (1)  {[(1  (1))/ 1
(3  1)]  4 + 1= (5  1)]  4 + 1= (5  1)]  4 + 1= ((2)
 (1))]  4 + 1}=
Step 3: Sum all standardized scores
HSBC Holdings 2005 5 + 1 + 5 + 0 + 5 + (5) = 11
HSBC Holdings 2006 5 + 3 + 4 + 0 + 5 + (5) = 12
Standard Chartered 2005 1 + 3 + 3 + 0 + 5 + (5) = 7
Standard Chartered 2006 1 + 3 + 4 + 0 + 5 + (5) = 8
Barclays 2005 1 + 5 + 3 + (1) + 5 + (1) = 12
Barclays 2006 1 + 5 + 4 + 0 + 5 + (1) = 14
Step 4: Standardize each banks score in each year to a 15 grading scale, referred to as the CSR index
HSBC Holdings 2005 (11  7)/(14  7)  4 + 1  3
HSBC Holdings 2006 (12  7)/(14  7)  4 + 1  4
Standard Chartered 2005 (7  7)/(14  7)  4 + 1  1
Standard Chartered 2006 (8  7)/(14  7)  4 + 1  2
Barclays 2005 (12  7)/(14  7)  4 + 1  4
Barclays 2006 (14  7)/(14  7)  4 + 1  5

Q1: What level of improvements in environmental impact can the company demonstrate?
Q2: How does EIRIS rate the companys environmental reporting?
Q3: What is the extent of policy addressing human rights issues?
Q4: How do the companys nes for health and safety convictions compare with other companies nes over the last 3 years?
Q5: Does the Company have a system for implementing a code of ethics and, if so, how comprehensive is it?
Q6: What proportion of turnover comes from alcohol sale or production?
Our standardized method for positive attitude questions is equal to [(ith value  min value)/(max value  min value)]  4 + 1, where the ith value denotes the observed numerical value, and the min and max values are the
minimum and maximum of all obtained numerical values, respectively. For the negative attitude questions, the standardized method is equal to (1)  {[(ith value  max value)/(min value  max value)]  4 + 1}. In the second
step, we do not extend number-grade rating of zero into ve scales 15 by standardization method.

3545
3546 M.-W. Wu, C.-H. Shen / Journal of Banking & Finance 37 (2013) 35293547
Our results are robust to the subprime crisis consideration, dif-
ferent specications for the CSR dummies and more CSR classes.
Thus, CSR and FP are positively linked.
We suggest that governments should provide less support to
banks engaging in CSR activities if the activities are based on stra-
tegic motives. Governments could support banks engaging in CSR
activities with altruistic motives by providing loans to rms en-
gaged in environmental protection. Studies merely investigating
the relationship between CSR and FP tend to ignore the motive.
Future research work on the following issues may be interest-
ing. First, although our study is based on the overall aggregate
index, examining the relation between CSR and FP using the sub-
components of these indices, for example, environmental, social,
and governance, is helpful in understanding the issue further.
Second, as discussed in the Robust Section, no nonlinear relation-
ship exists between CSR and FP in our case. Future studies can
re-visit this nonlinear issue when more CSR choices are available.
Finally, the follow-up effects of the private and social costs of the
subprime nancial crisis can also be considered in future studies.
References
AL-Hussami, M., 2008. A Study of nurses job satisfaction: the relationship to
organizational commitment, perceived organizational support, transactional
leadership, transformational leadership, and level of education. European
Journal of Scientic Research 22 (2), 286295.
Aktas, N., De Bodt, E., Cousin, J.G., 2011. Do nancial markets care about SRI?
Evidence from mergers and acquisitions. Journal of Banking & Finance 35 (7),
17531761.
Barnett, M.L., Salomon, R.M., 2002, Unpacking Social Responsibility: The Curvilinear
Relationship between Social and Financial Performance. Academy of
Management Best Paper Proceedings [CD-ROM] (Sect. SIM, pp. B1B6).
Barnett, M.L., Salomon, R.M., 2006. Beyond dichotomy: the curvilinear relationship
between social responsibility and nancial performance. Strategic Management
Journal 27, 11011122.
Baron, D., 2001. Private politics, corporate social responsibility and integrated
strategy. Journal of Economics and Management Strategy 10, 745.
Baron, D., 2009. A positive theory of moral managementm, social pressure, and
corporate social performance. Journal of Economics and Management Strategy
18, 743.
Barth, J.R., Caprio, G., Levine, R., 2006. Rethinking Bank Regulation: Till Angels
Govern. Cambridge University Press, New York.
Barth, J.R., Caprio, G., Levine, R., 2008. Bank regulations are changing- but for better
or worse. Comparative Economic Studies 50, 537563.
Becchetti, L., Ciciretti, R., 2009. Corporate social responsibility and stock market
performance. Applied Financial Economics 19 (16), 12831293.
Bnabou, R., Tirole, J., 2010. Individual and corporate social responsibility.
Economica 77 (305), 119.
Beck, T., Demirg-Kunt, A., Levine, R., 1999. A New Database on Financial
Development and Structure. Policy Research Working Paper Series 2146, The
World Bank.
Beck, T., Demirg-Kunt, A., Levine, R., 2010. Financial institutions and markets
across countries and over time: the updated nancial development and
structure database. World Bank Economic Review 24 (1), 7792.
Billet, M., Flannery, M., Garnkel, J., 1995. The effect of lender identity on a
borrower rms equity return. Journal of Finance 50, 699718.
Bikker, J.A., Haaf, K., 2002a. Measures of competition and concentration in the
banking industry: a review of literature. Economic and Financial Modeling 9 (2),
5398.
Bikker, J.A., Haaf, K., 2002b. Competition, concentration and their relationship: an
empirical analysis of the banking industry. Journal of Banking & Finance 26 (11),
21912214.
Bikker, J.A., Spierdijk, L., 2010. Measuring and explaining competition in the
nancial sector. Journal of Applied Business and Economics 11 (1), 1142.
Bonin, J.P., Hasan, I., Wachtel, P., 2005a. Privatization matters: bank efciency in
transition countries. Journal of Banking and Finance 29 (89), 21552178.
Bonin, J.P., Hasan, I., Wachtel, P., 2005b. Bank performance, efciency and
ownership in transition countries. Journal of Banking & Finance 29 (1), 3153.
Booth, A., 1993. Private sector training and graduate earnings. The Review of
Economics and Statistics 75 (1), 164170.
Bourguignon, F., Fournier, M., Gurgand, M., 2007. Selection bias corrections based
on the multinomial logit model: Monte Carlo comparisons. Journal of Economic
Surveys 21 (1), 174205.
Brammer, S., Brooks, C., Pavelin, S., 2006. Corporate social performance and stock
returns: UK evidence from disaggregate measures. Financial Management, 97
116. Autumn.
Brammer, S., Pavelin, S., 2004. Building a good reputation. European Management
Journal 22 (6), 704713.
Brilius, P., 2010. Dynamic model of dependencies between economic crisis and
corporate social responsibility contribution to sustainable development.
Sustainable Development 15, 422429.
Brine, M., Brown, R., Hackett, G., 2007. Corporate social responsibility and nancial
performance in the Australian context. Economic Round-up, 4758. Autumn.
Brnn, P.S., Vrioni, A.B., 2001. Corporate social responsibility and cause-related
marketing: an overview. International Journal of Advertising 20, 207222.
Bushman, R.M., Wittenberg-Moerman, R., 2012. The Role of Bank Reputation in
Certifying Future Performance Implications of Borrowers Accounting
Numbers. Journal of Accounting Research 50 (4), 883930.
Cellier, A., Chollet, P., 2011. The Impact of Corporate Social Responsibility Rating
Announcements on European Stock Prices. AFFI International Conference,
Montpellier, 11th to 13th May.
Cerutti, E., Ariccia, G.D., MartnezPera, M.S., 2007. How banks go abroad: Branches
or subsidiaries? Journal of Banking & Finance 31 (6), 16691692.
Chaney, P., Jeter, D., Shivakumar, L., 2004. Self selection of auditors and audit pricing
in private rms. Accounting Review 79, 5172.
Chatterji, A.K., Levine, D.I., 2008, Imitate or Differentiate? Evaluating the Validity of
Corporate Social Responsibility Ratings. Working Paper Series, 1-55.
Chatterji, A.K., Levine, D.I., Toffel, M.W., 2010. How well do social ratings actually
measure corporate social responsibility? Journal of Economics & Management
Strategy 18 (1), 125169.
Chemmanur, T., Fulghieri, P., 1994. Investment bank reputation, information
production, and nancial intermediation. Journal of Finance 49, 5779.
Chih, H.L., Chih, H.H., Chen, T.Y., 2010. On the determinants of corporate social
responsibility: international evidence on the nancial industry. Journal of
Business Ethics 93 (1), 115135.
Chih, H.L., Shen, C.H., Kang, F.C., 2008. Corporate social responsibility, investor
protection, and earnings management: some international evidence. Journal of
Business Ethics 79, 179198.
Clatworthy, M.A., Makepeace, G.H., Peel, M.J., 2009. Selection bias and the Big Four
premium: new evidence using Heckman and matching models. Accounting and
Business Research 39 (2), 139166.
Coffey, B., Fryxell, G.E., 1991. Institutional ownership of stock and dimensions of
corporate social performance: an empirical examination. Journal of Business
Ethics 10, 437444.
Cox, P., Brammer, S., Millington, A., 2004. An empirical examination of institutional
investor preferences for corporate social performance. Journal of Business
Ethics 52, 2743.
Cuesta-Gonzlez, M., Muoz-Torres, M., Fernndez-Izquierdo, M.., 2006. Analysis
of social performance in the Spanish nancial industry through public data: a
proposal. Journal of Business Ethics 69 (3), 289304.
Dam, L., Koetter, M., Scholtens, B., 2009. Why Do Firms Do Good? Evidence from
Managerial Efciency. In: 4th CORE Conference Working Paper Series, 1.
El Ghoul, S., Guedhami, O., Kwok, C.C.Y., Mishra, D.R., 2011. Does corporate social
responsibility affect the cost of capital? Journal of Banking & Finance 35 (9),
23882408.
Fernndez-Kranzl, D., Juan Santal, J., 2010. When necessity becomes a virtue: the
effect of product market competition on corporate social responsibility. Journal
of Economics & Management Strategy 19 (2), 453487.
Fisman, R., Heal, G., Nair, V., 2005. Corporate Social Responsibility: Doing Well by
Doing Good?, Working Paper, Wharton School, University of Pennsylvania.
Fisman, R., Heal, G., Nair, V., 2006. A Model of Corporate Philanthropy. Working
Paper. Wharton School, University of Pennsylvania.
Fombrun, C.J., Shanley, M., 1990. What is in a name? Reputation building and
corporate strategy. Academy of Management Journal 33 (2), 233259.
Frankental, P., 2001. Corporate social responsibilitya PR invention? Corporate
Communications 6 (1), 1823.
Freeman, R.E., 1984. Strategic Management: A stakeholder approach. Pitman, Boston.
Freixas, X., Rochet, J.C., 1998. Microeconomics of Banking. MIT Press.
Friedman, M., 1962. Copitalism and Freedom. University of Chicago Press, Chicago,
IL.
Friedman, M., 1970. A Friedman Doctrine:Social responsibility of business is to
increase its prots. The New York Times Magazine, Sep. 13, 3233, and 123
125.
Fry, E.L., Hock, R.J., 1976. Who claims corporate responsibility? The biggest and the
worst. Business and Society Review 18, 6265.
Garcia-Castro, R., Ario, M.A., Canela, M.A., 2010. Does social performance really
lead to nancial performance? Accounting for endogeneity. Journal of Business
Ethics 92 (1), 107126.
Gil, F., Rico, R., Alcover, C.M., Barrasa, A., 2005. Change-oriented leadership,
satisfaction and performance in work groups: effects of team climate and
group potency. Journal of Managerial Psychology 20 (3/4), 312328.
Greene, W., 2003. Econometric Analysis, fth ed. Prentice Hall, New Jersey.
Greene, W., 2012. Econometric Analysis, seventh ed. Prentice Hall, New Jersey.
Grifn, J.J., Mahon, J.F., 1997. The corporate social performance and corporate
nancial performance debate: twenty-ve years of incomparable research.
Business and Society 36, 531.
Goddard, J., Molyneux, P., Wilson, J.O.S., 2004. The protability of european banks: a
cross-sectional and dynamic panel analysis. The Manchester School 72 (3), 363
381.
Goss, A., Roberts, G.S., 2011. The impact of corporate social responsibility on the
cost of bank loans. Journal of Banking & Finance 35 (7), 17941810.
Hamilton, B.H., Nickerson, J.A., 2003. Correcting for endogeneity in strategic
management research. Strategic Organization 1 (1), 5178.
 M.-W. Wu, C.-H. Shen / Journal of Banking & Finance 37 (2013) 35293547
Harjoto, M.A., Jo, H., 2008. Corporate social responsibility and operating
performance. Journal of Academy of Business and Economics 18 (1), 5971.
Harjoto, M.A., Jo, H., 2011. Corporate governance and CSR Nexus. Journal of Business
Ethics 100, 4567.
Heal, G., 2005. Corporate social responsibility: an economic and nancial
framework. The Geneva Papers 30, 384409.
Heckman, J.J., 1978. Dummy endogenous variables in a simultaneous equation
system. Econometrics 46 (6), 931959.
Heitmueller, A., 2004. Job mobility in Britain: are the scots different? evidence from
the BHPS. Scottish Journal of Political Economy 51 (3), 329358.
Henderson, D., 2002. Misguided Virtue: False Notions of Corporate Social
Responsibility. Institute of Economic Affairs, London.
Ioannou, I., Serafeim, G., 2010. The Impact of Corporate Social Responsibility on
Investment Recommendations. Best Paper Proceedings, Academy of
Management 2010 Annual Meeting (Montreal Canada). Social Issues in
Management (SIM) Division.
Ireland, C.J., Lennox, C.S., 2002. The large audit rm fee premium: a case of
selectivity bias? Journal of Accounting, Auditing and Finance 17 (1), 7391.
James, H.S., 2006. Self-selection bias in business ethics research. Business Ethics
Quarterly 16 (4), 559578.
Jehn, K.A., Northcraft, G.B., Neale, M.A., 1999. Why differences make a difference: a
eld study of diversity, conict, and performance in workgroups.
Administrative Science Quarterly 44 (4), 741764.
Jensen, M.C., 2002. Value maximization, stakeholder theory, and the corporate
objective function. Business Ethics Quarterly 12 (2), 235256.
Kaufman, D., Kraay, A., Mastruzzi, M., 2009. Governance Matters VIII: Aggregate and
Individual Governance Indicators for 19962008. World Bank Policy Research
Working Paper, No.4978.
Karaibrahimoglu, Y.Z., 2010. Corporate social responsibility in times of nancial
Crisis. African Journal of Business Management 4, 382389.
Kay, J., 1993. Foundations of Corporate Success. Oxford University Press, Oxford.
King, R.G., Levine, R., 1993. Finance and growth: Schumpeter might be right. The
Quarterly Journal of Economics 108 (3), 717737.
Kim, M., Kristiansen, E.G., Bent, V., 2005. Endogenous product differentiation in
credit markets: what do borrowers pay for? Journal of Banking & Finance 29 (3),
681699.
Lee, L.F., 1982. Health and wage: a simultaneous equation model with multiple
discrete indicators. International Economic Review 23, 199221.
Lee, L.F., 1983. Generalized econometric models with selectivity. Econometrica 51
(2), 507512.
Levine, R., 2005. Finance and growth: theory and evidence. In: Philippe Aghion,
Steven Durlauf (Eds.), Handbook of Economic Growth, rst ed., vol. 1, pp. 865
934 (Chapter 12).
Li, K., Prabhala, N.R., 2007. Self-selection models in corporate nance. In: Eckbo,
Espen.B. (Ed.), Handbook of Corporate Finance I. Elesevier, New York, pp. 3786.
Magableh, I.K., Kharabsheh, R., Al-Zubi, K.A., 2011. Determinants and impact of
training: the case of SMEs in Jordan. International Journal of Economics and
Finance 3 (5), 104116.
Margolis, J.D., Walsh, J.P., 2003. Misery loves companies: rethinking social
initiatives by business. Administrative Science Quarterly 48 (2), 263305.
Margolis, J., Elfenbein, H., Walsh, J., 2007, Does it pay to be good? A Meta-analysis
and Redirection of Research on the Relationship between Corporate social and
Financial Performance. Harvard Business School Working Paper.
Margolis, J., Elfenbein, H., Walsh, J., 2009. Does it Pay to be Good. . . and does it
Matter? A Meta-Analysis of the Relationship between Corporate Social and
Financial Performance. Available at SSRN: <http://ssrn.com/abstract=1866371>
or <http://dx.doi.org/10.2139/ssrn.1866371>.
McWilliams, A., Siegel, D., 1997. Event studies in management research: Theoretical
and empirical issues. Academy of Management Journal 40, 626657.
McWilliams, A., Siegel, D., 2000. Corporate social responsibility and nancial
performance: correlation or misspecication? Strategic Management Journal 21
(5), 603609.
McWilliams, A., Siegel, D., 2001. Corporate social responsibility: a theory of the rm
perspective. Academy of Management Review 26 (1), 117127.
Missbach, A., 2004. The Equator Principles: drawing the line for socially responsible
banks? An interim review from an NGO perspective. Development 47 (3), 78
84.
Orchard-Webb, J., 2008. The Equator Principles? corporate? greenwash? or
improving global environmental standards? Environmental Law &
Management 20 (4), 184186.
3547
Petersen, M.A., 2009. Estimating standard errors in nance panel data sets:
comparing approaches. Review of Financial Studies 22 (1), 435480.
Renneboog, L., Horst, J.T., Zhang, C., 2008. Socially responsible investments:
institutional aspects, performance and investor behavior. Journal of Banking
and Finance 32 (9), 17231742.
Ross, D., 2010. The Dominant Bank Effect: how high lender reputation affects the
information content and terms of bank loans. The Review of Financial Studies
23 (7), 27302756.
Rose, P.S., Hudgins, S., 2010. Bank Management & Financial Services, eighth ed.
McGraw- Hill Irwin, New York.
Rubin, D.B., 1986. Statistical matching using le concatenation with adjusted
weights and multiple imputations. Journal of Business and Economic Statistics
4, 8794.
Saunders, A., Cornett, M.M., 2010. Financial Institutions Management: A Risk
Approach, seventh ed. McGraw-Hill.
Scholtens, B., 2009. Corporate social responsibility in the international banking
industry. Journal of Business Ethics 86 (2), 159175.
Scholtens, B., Dam, L., 2007. Banking on the equator: are banks that adopted the
equator principles different from non-adopters? World Development 35 (8),
13071328.
Schumpeter, J.A., 1912. The Theory of Economic Development (Redvers Opie, Trans.,
1934). Harvard University Press, Cambridge, MA.
Sen, S., Bhattacharya, C.B., 2001. Does doing good always lead to doing better?
Consumer reactions to corporate social responsibility. Journal of Marketing
Research 38 (2), 225243.
Shaver, J.M., 1998. Accounting for Endogeneity when assessing strategy
performance: does entry mode choice affect FDI survival? Management
Science 44 (4), 571585.
Shen, C.H., 2002. Credit rationing for bad companies in bad years: evidence from
bank loan transaction data. International Journal of Finance and Economics 7,
261278.
Shen, C.H., Chang, Y.H., 2006. Do regulations affect banking performance?
Government governance may matter. Contemporary Economic Policy 24 (1),
92105.
Shen, C.H., Chang, Y., 2009. Ambition versus conscience, does corporate social
responsibility pay off? The application of matching methods. Journal of
Business Ethics 88 (1), 133153.
Shen, C.H., Chang, Y., 2012-a. Corporate Social Responsibility, Financial Performance
and Selection Bias-Evidence from TWSE-listed Banks of Taiwan. In: Barth, J.R.,
Lin, C., Wihlborg, C. (Eds.), Research Handbook on International Banking and
Governance. Edward Elgar Publishing (Chapter 28).
Shen, C.H., Chang, Y., 2012-b. To Join or Not to Join?-Do Banks That Are In a
Financial-Holding-Company Perform Better Than Banks That Are Not.
Contemporary Economic Policy 30 (1), 113128.
Shen, C.H., Lee, C.C., 2005. Same nancial development yet different economic
growthand why? Journal of Money, Credit, and Banking 38, 19071944.
Shen, C.H., Lin, C.Y., 2012. Why government banks underperform: a political
interference view. Journal of Financial Intermediation 21, 181202.
Simpson, W., Kohers, T., 2002. The link between social and nancial performance:
evidence from the banking industry. Journal of Business Ethics 35, 97109.
Trotta, A., Iannuzzi, A., Cavallaro, G., DellAtti, S., 2011. Banking reputation and CSR:
a stakeholder value approach. In: Gummesson, E., Mele, C., Polese, F. (Eds.), The
Naples Forum on Service 2011 Service Dominant Logic. Network & System
Theory and Service Science: Integrating Three Perspectives for a New Service
Agenda. Giannini Editore, Napoli. ISBN: 978-88-7431-525-3.
Vance, S., 1975. Are socially responsible rms good investment risk? Management
Review 64 (8), 1824.
Waddock, S.A., Graves, S.B., 1997. The corporate social performance-nancial
performance link. Strategic Management Journal 18 (4), 303319.
Watchman, P., Freshelds, P., Deringer, B., 2006. Banks, business & human rights.
Butterworths Journal of International Banking and Financial Law, 4650.
Wu, M.W., Shen, C.H., 2011. The elusive effect of bank size on prots. The Service
Industries Journal 31 (11), 17031724.
White, H., 1980. A heteroskedastic-consistent covariance matrix estimator and a
direct test of heteroskedasticity. Econometrica 48, 817838.