What functional form fits the CSR-CFP relationship?
Abstract
This study explores what functional form fits the relationship between corporate social responsibility (CSR) and corporate financial performance (CFP). It investigates the effect of CSR on CFP, and the effect of CFP on CSR. Linear function, quadratic function and cubic function are used to explore the CSR-CFP relationship. This study applies accounting-based (Return on Assets) and market-based (Excess Stock Return) measures of CFP, aggregate CSR (ESG), and disaggregate dimensions of CSR (ENV, SOC and CGV). Results show that we couldn’t find a functional form that best fits the CSR-CFP relationship. The measures of CSR and CFP, and the use of zero or one-year lag in independent variable appear to affect the results. We conclude that the previous findings of the linear relationship may not adequately reflect the CSR-CFP relationship.
Key words: Corporate social responsibility, corporate financial performance, relationship,
functional form, linear, non-linear
Student: Wei Sun
Student number: s3084922
Study program: MSc. International Financial Management Faculty of Economics and Business
University of Groningen
Supervisor: Prof. dr. L.J.R. Scholtens Co-Assessor: dr. R.O.S. Zaal
1. Introduction
During the last decade, corporate social responsibility (CSR) attracts significantly increasing attention (Scholtens, 2008), and after financial crisis this topic is more extensively debated (Nollet, Filis and Mitrokostas, 2016). Numerous studies examine the relationship between CSR and corporate financial performance (CFP), and the effect from CSR to CFP is investigated in most studies (Margolis and Walsh, 2001). However, those studies fail to reach a consensus. Positive relationship, negative relationship, no relationship and mixed results are all found in literature (Margolis and Walsh, 2001). Although the effect of CFP on CSR is less often investigated, positive relationship (Waddock and Graves, 1997) and negative relationship (Gonenc and Scholtens, 2017) are also found. According to Nollet et al. (2016), the non-linear relationship between CSR and CFP has seldom been studied, but it is consistent with economic intuition. Lankoski (2008) argue that an inverse U-shape could describe the effect of CSR on CFP. Barnett and Salomon (2012) find U-shaped impact of CSR on CFP. Lee and Park (2010) examine whether CSR exerts cubic influence on CFP, but they fail to find this kind of relationship. Compared with the non-linear effect of CSR on CFP, there’s a lack of theory and investigation about the non-linear effect of CFP on CSR. Therefore, this study tends to comprehensively explore the relationship between CSR and CFP. The research question could be formulated as:
What functional form best fits the relationship between CSR and CFP?
squares (OLS) regressions with country and industry fixed effects to explore the relationship between CSR and CFP. The data of 5045 companies in the period of 2006 to 2015 are used in this study. I test the hypotheses about the CSR-CFP relationship and report the results based on the functional form.
Three dimensions are embodied in this study: international, financial and management. Companies from 37 countries/regions are included in the sample. The highly international sample reflects the international dimension of this study. The differences in country characteristics are controlled by using country fixed effects. With more attention paid to CSR, it’s important for a firm to understand the CSR-CFP relationship. The investigation of the CSR-CFP relationship provides more insights into a firm’s financial management. This study may give some implications to the management team about how to benefit from the participation in socially responsible activities.
This study finds that we couldn’t find a functional form that best fits the CSR-CFP relationship. Different measures of CSR and CFP could lead to different findings of the relationship between CSR and CFP. In addition, even though we use the same indicator of CSR and CFP, the results of contemporaneous relationship and relationship with one-year lag are also different. Furthermore, we find that the relationship between CSR and CFP seems to be bidirectional. Non-linear relationship is more likely to appear in the effect of CFP on CSR.
The remaining part of this thesis is organized as follows. The second section provides relevant literature review and develops hypotheses about the CSR-CFP relationship. The third section presents data and sample. The fourth section introduces the methodology. The fifth section shows the descriptive statistics and correlations of the variables. The sixth section discusses the regression results and the last section gives the conclusion and limitations of this study.
2. Background and hypotheses
2.1 Corporate social responsibility
There are many definitions and studies of CSR in existing literature, therefore, it’s not possible to give an accurate definition of it. Using decade-by-decade categories, Carroll (1999) studies the evolutions of CSR from 1950s to 1990s and finds that the concepts and definitions of CSR have gradually developed in history. With the conduct of researches and development of theories, existing definitions of CSR could be changed and new definitions may be put forward according to previous groundwork (Carroll, 1999). A definition of CSR relating to this study is from European Commission (2001, p.6): “a concept whereby companies integrate social and environmental concerns in their business operations and in their interaction with their stakeholders on voluntary basis”. This concept highlights the social and environmental dimensions of CSR, and the two dimensions are widely studied in literature. However, except for social and environmental concerns, governance concern should be considered because it is also an important component of CSR.
Customers, investors and regulators are concerned about firms’ business ethics and have a preference for firms emphasizing entrepreneurial profit with the goal of long-term development and valuing their stakeholders (Nollet et al., 2016). Those stakeholders regard “CSR-oriented governance acts as a credible signal for long run commitment to CSR values” (Nollet et al., 2016, p404). This study considers social, environmental and governance performance as all necessary components of CSR, therefore, in order to construct a more comprehensive concept of it, it is better to add governance concern in the definition (European Commission, 2001, p.6) of CSR.
(Bénabou and Tirole, 2010). Board members and management team have desires to involve in philanthropy, but profits may not be maximized. A firm’s taking of CSR activities may depend on a specific vision, but in many cases it could be affected by the combination of different visions.
2.2 The effect of corporate social responsibility on corporate financial performance
There are many review studys about the CSR-CFP relationship (Orlitzky, Schmidt and Rynes, 2003; Margolis, Elfenbein and Walsh, 2009; Friede, Busch and Bassen, 2015). Those studies show that CSR has impact on CFP. Generally, the linear effect of CSR on CFP is widely investigated and discussed in literature, but Barnett and Salomon (2012) find that the effect of CSR on CFP is U-shaped. According to the study of McWilliams and Siegel (2000), CSR seems to have no financial impact. However, we couldn’t rule out the possibility that alternative functional forms might also fit the CSR-CFP relationship. In this part, we first discuss the linear relationship, then focus on the non-linear relationship based on functional form.
2.2.1 Linear and positive relationship
Stakeholder theory is a major theory describing the linear and positive effect of CSR on CFP. This theory dates back to Freeman (1984), highlighting the importance for a firm to develop good relationship with its stakeholders. Responsible companies are perceived to be honest in their operation, thus customers may have more interests in them, leading to better satisfaction (Martínez and Bosque, 2013). In addition, a firm’s CSR also influences its employees. Sprinkle and Maines (2010) provide evidence that some companies regard CSR as a way to improve the recruit, motivation and retention of employees. It can be assumed that companies with good CSR performance will attract more competent employees and these people will contribute to better operation. Moreover, in order to facilitate sustainable development, socially responsible activities are support by governments using different policies (Wagner, 2010).
more resources from investors in favour of CSR (Brammer and Millington, 2008). If a firm’s CSR keeps increasing, this firm will have more revenues and reduced costs, leading to better CFP. However, in existing studies there is a lack of discussion about why the relationship between CSR and CFP is linear. Many people simply assume linearity without articulating very well why the relationship is linear. The linearity of the relationship assumes that the positive effect of CSR on CFP is constant. It means that a unit increase of the costs of engaging in socially responsible activities will lead to certain increase of CFP, and the increase of CFP outweighs the increase of costs. When CSR keeps increasing, the effect of CSR on CFP won’t change. Many previous studies find linear and positive effect of CSR on CFP.
Based on stakeholder theory, Preston and O’Bannon (1997) develop social impact hypothesis. This hypothesis indicates that corporate social performance precedes financial performance, and the relationship is linear and positive. They argue that if a firm satisfies its major stakeholders, its reputation could be improved, leading to better financial performance.
2.2.2 Linear and negative relationship
A well-known argument in favour of the negative relationship between CSR and CFP comes from Friedman (1970, p.1): “the social responsibility of business is to increase its profits”. Friedman (1970) states that the purchase of social interest reduces stockholders’ returns because this kind of behaviour consumes their money. In addition, the raising of product prices and lowering of wages mean socially responsible activities are using customers’ and employees’ money (Friedman, 1970). If CSR means the decrease of profits and damage of stockholders and stakeholders’ interests, this kind of behaviour would affect a firm’s operation.
increase their personal reputation, resulting in deterioration of profits. However, there is also a lack of discussion in previous literature about why the negative relationship is linear. Similar to the aforementioned linearity of positive relationship, the linearity of negative relationship assumes that the negative effect of CSR on CFP is constant. Namely, a unit increase of the costs of engaging in socially responsible activities will result in certain decrease of CFP. The constant negative effects won’t be influenced by the level of CSR. Some former studies find linear and negative relationship between CSR and CFP (Brammer, Brooks and Pavelin, 2006; Boyle, Higgins and Rhee, 1997)
The trade-off hypothesis (Preston and O’Bannon, 1997) is a reflection of the argument in Friedman (1970). This hypothesis also indicates that corporate social performance precedes financial performance, however, the relationship is linear and negative. The contention of this hypothesis is in line with above-mentioned reasons for the linear and negative effect of CSR on CFP.
Figure 1 shows the relationship between CSR and CFP based on social impact hypothesis and trade-off hypothesis.
Figure 1
The linear effect of CSR on CFP
Panel A is based on social impact hypothesis and panel B is based on trade-off hypothesis
2.2.3 Non-linear relationship
The debate of positive or negative effect of CSR on CFP hasn’t been resolved yet (Barnett and Salomon, 2012). In fact, this may be explained by the reason that the linear approach couldn’t provide a complete explanation for the effect of CSR on CFP (Miras-Rodríguez, Carrasco-Gallego, and Escobar-Pérez, 2015). Barnett and Salomon (2012) argue that the two kinds of effect might be right over some range. However, many previous studies only test the linear relationship without the assumption that non-linearity may also exist. Accordingly,
CFP
CSR CFP
CSR
different functional form other than linear relationship might give more insights into the relationship between CSR and CFP.
Barnett (2007, p.803) develops the concept of stakeholder influence capacity (SIC) as “the ability of a firm to identify, act on, and profit from opportunities to improve stakeholder relationships through CSR”. By continuously taking part in socially responsible activities, firms receive SIC (Barnett, 2007). Firms with different levels of SIC will have different responses from their stakeholders, leading to different financial performance from socially responsible behaviour (Barnett and Salomon, 2012). If a firm has high level of SIC, the good relationship with stakeholders will reduce its transaction costs. Therefore, this firm is more likely to be perceived as credible; if a firm has low level of SIC, it is less likely to be perceived as credible, because stakeholders regard its socially responsible activities as “self-serving” and “greenwashing”, making it hard for this firm to transfer CSR into good CFP (Barnett and Salomon, 2012).
The concept of SIC provides the motivation for the U-shaped relationship between CSR and CFP. If a firm doesn’t receive adequate SIC, it couldn’t create enough gains to offset the costs of participating in socially responsible activities (Barnett and Salomon, 2012). Thus firms with low CSR will have negative returns when CSR increases. However, if a firm keeps increasing its spending on CSR, it will accrue adequate SIC to transfer the costs of CSR into financial returns (Barnett and Salomon, 2012). The positive returns could offset increased costs, so the relationship between CSR and CFP becomes positive (Barnett and Salomon, 2012). Therefore, the CSR-CFP relationship could be considered as U-shape: before acquiring enough SIC, there is downward slope of the curve; after acquiring adequate SIC, there is upward slope of the curve (Barnett and Salomon, 2012).
Wagner and Schaltegger (2004) state that firms face a trade-off between environmental performance and financial performance. Regulation determines the minimal degree of environmental performance, which decides the win-win point. Before reaching the win-win point, better environmental performance means better CFP. After the win-win point, a firm is likely to face a trade-off between environmental performance and CFP. If a firm is under weak regulation, the best degree of environmental performance may be higher than what is required by regulation (Wagner and Schalteger, 2004).
Figure 2 shows the relationship between CSR and CSP based on Barnett and Salomon (2012), Lankoski (2008) and Wagner and Schaltegger (2004).
Figure 2
The quadratic effect of CSR on CFP
Panel A is based on Barnett and Salomon (2012) and panel B is based on Lankoski (2008) and Wagner and Schaltegger (2004)
Although Barnett and Salomon (2012) find the quadratic relationship between corporate social performance and financial performance, they state that the relationship might be described by many functional forms. In addition to linear relationship, Lee and Park (2010) investigate quadratic and cubic relationship between CSR and CFP, however, they fail to give an explanation about why quadratic and cubic relationship may exist. Actually, even though cubic relationship may exist between CSR and CFP, this functional form is rarely discussed in existing literature. And it might be possible that the relationship between CSR and CFP is polynomial.
Furthermore, in economic and business literature, cubic relationship is found between some variables. The Environmental Kuznets Curve Hypothesis often tests cubic relationship. Some studies of the EKC Hypothesis find that the cubic functional form best fits the relationship between air pollutants and GDP per capita (Grossman and Krueger, 1993; Grossman and Krueger, 1995). Amato and Amato (2007) find cubic relationship between charitable giving
CSR CSR
CFP CFP
and firm size. Cubic relationship could be regarded as the extension of linear and quadratic relationship. In addition to cubic relationship, polynomial relationship might also exist. Polynomial relationship means that the functional form incorporates more than one term of the linear term, quadratic term and cubic term. When the functional form incorporates not only cubic term, CSR still has cubic effect on CFP. Before the exploration of cubic and polynomial relationship, the cubic functional form and polynomial functional form are discussed below.
Assuming the functional form is f(x)=ax3+bx2+cx+d (a≠0). f ’(x)=3ax2+2bx+c and ∆=4b2 -12ac= 4(b2-3ac). f ’(x) represents the slope of f(x). When a≠0 and b=c=0, it means cubic relationship; When at least two coefficients of a, b and c≠0, it means polynomial relationship. If ∆˃0, it means that the sign of the slope will change two times; If ∆≤0, it means the sign of the slope won’t change, but the value of the slope will change. This functional form could lead to the following four different functional images in figure 3. The four different functional images are labelled as panel A, panel B, panel C and Panel D:
Figure 3
The cubic or polynomial effect of CSR on CFP
The functional images in the four panels are depicted according to the sign of the coefficient a and ∆
Among the four different functional images in figure 3, panel B and panel D could represent the cubic relationship (without linear and quadratic term). Because there is a lack of investigation and theory about the cubic and polynomial effect of CSR on CFP, the following explanations are based on my personal understanding.
In panel A of figure 3, at the initial stage the relationship between CSR and CFP is positive. It may be explained by the minimal degree of CSR required by regulation (Wagner and Schaltegger, 2004). When a firm’s CSR is less than what is required by regulation or the average level of its industry, this firm will have bad relationship with its stakeholders and it
a˃0, ∆˃0 CSR a˃0, ∆≤0 a˂0, ∆˃0 a˂0, ∆≤0
CSR CSR CSR
CFP CFP CFP CFP
may even receive fines. In this situation, the increase of CSR will lead to better CFP until the level of CSR reaches the win-win point (Wagner and Schaltegger, 2004). The remaining part of panel A could be explained by the reasons for the U-shaped effect of CSR on CSP. When the level of CSR reaches the win-win point, because this level just equals to what is required by regulation, this firm doesn’t accumulate enough stakeholder influence capacity (Barnett, 2007). As a result, the costs of continuously participating in socially responsible activities couldn’t be offset by the benefits it brings (Barnett and Salomon, 2012). However, if the firm continues to increase its CSR, this firm’s SIC will also increase (Barnett and Salomon, 2012). After a certain point, this firm could transfer CSR to positive financial returns again.
In panel B, the increase of CSR means better CFP, but the slope is not a constant: it decreases first, then increases. This relationship indicates that even though there is positive relationship between CSR and CFP, the effect of CSR on CFP will change with the increase of CSR. The difference between panel B and panel A is that when CSR increases, the slope in panel A will reduce to a negative value, however, the slope in panel B only reduces to zero. From my perspective, this finding could be explained by SIC (Barnett, 2007). Because different firms have different characteristics, the relationship with their stakeholders are also different. Although in some range the increase of CSR means the increase of marginal costs and the decrease of marginal revenues (Lankoski, 2008), some firms have enough SIC to transform CSR into positive financial performance (Barnett and Salomon, 2012). For these firms, though the slope will decrease, it won’t be reduced to a negative value.
The left U-shape in panel C could be explained by the reasons for the U-shaped relationship between CSR and CFP (Barnett and Salomon, 2012). However, in my view, even if a firm may acquire enough SIC after a certain point to transform CSR into better CFP, the positive effect of CSR on SIC will decrease. When CSR reaches a certain level, SIC will increase to its highest value. Thereafter, SIC won’t increase with increased CSR. Accordingly, the costs of high level of CSR will outweigh the revenues, leading to negative financial performance. This explains the negative relationship between CSR and CFP after the U-shape in panel C.
leading to increased revenues. Although the revenues are increasing and the costs are decreasing, the increased revenues still couldn’t offset the costs of CSR. Then the revenues may equal to the costs at a certain point. After that point, SIC is not likely to increase, thus the benefits of improved relationship with stakeholders will gradually disappear. If the costs continue to increase, the gap between costs and revenues will be widened.
According to aforementioned discussion, following hypotheses could be formulated in regard to the effect of CSR on CFP:
H1: CSR affects CFP, and the relationship is linear and positive. H2: CSR affects CFP, and the relationship is linear and negative. H3a: CSR affects CFP, and the relationship is quadratic.
H3b: CSR affects CFP, and the relationship is cubic. H3c: CSR affects CFP, and the relationship is polynomial.
In all cases the null hypothesis could be formulated as: H0: There is no relationship between CSR and CFP.
2.3 The effect of corporate financial performance on corporate social responsibility
Unlike the effect of CSR on CFP, the effect of CFP on CSR is less often investigated in literature. Scholtens (2008) tests the interaction between CSR and CFP in two directions using lagged OLS and Granger causation, and the results show that CFP is more likely to be the cause of CSR. However, there’s a lack of theories and investigation of the non-linear effect of CFP on CSR. Studies about this kind of effect mainly focus on linear relationship. Therefore, in addition to linear functional form, the investigation of other functional forms could provide more insights into the effect of CFP on CSR.
2.3.1 Linear and positive relationship
effect of CFP on CSR is linear. Many people simply assume linearity without articulating very well why the relationship is linear. The linearity of the relationship assumes that the positive effect of CFP on CSR is constant. When CFP keeps increasing, the effect of CFP on CSR won’t change.
Preston and O’Bannon (1997) put forward available funds hypothesis to support the causal relationship from financial performance to social performance. According to Preston and O’Bannon (1997), even if firms want to take part in more socially responsible activities and have good corporate citizenship, those behaviours rely on available resources. Socially responsible behaviours will spend a firm’s money. If a firm doesn’t have good profitability, it will not have enough supports for CSR. For this reason, good CFP is the premise of good CSR.
2.3.2 Linear and negative relationship
Compared with the positive effect of CFP on CSR, the negative effect is infrequently studied in literature. Preston and O’Bannon (1997) propose managerial opportunism hypothesis to explain this kind of relationship, and they believe this hypothesis is original in their paper. This hypothesis relates to agency theory, highlighting the different interests of managers from those of shareholders and stakeholders.
Managerial opportunism hypothesis (Preston and O’Bannon, 1997) argues that the linkage between managers’ compensation and financial performance could give rise to negative relationship. If financial performance is good, managers are more likely to cut down the expenditures for participating in socially responsible activities in order to increase their private interests; if financial performance is poor, managers may seek to mask this dissatisfying result by taking part in more socially responsible activities (Preston and O’Bannon, 1997). As a result, good CFP may be an indicator of bad CSR.
role, and this kind of power leads to the same decrease of CSR with a unit increase of CFP. It fails to consider other factors that may influence the relationship.
Figure 4 shows the relationship between CFP and CSR based on available funds hypothesis and managerial opportunism hypothesis.
Figure 4
The linear effect of CFP on CSR
Panel A is based on available funds hypothesis and panel B is based on managerial opportunism hypothesis
2.3.3 Non-linear relationship
Existing literature has a lack of investigation of the non-linear influence of CFP on CSR. Some studies simply test the linear relationship, ignoring the possibility that non-linear relationship may also exist. Thus the discussion of the non-linear effect from CFP to CSR in this study is mainly the exploration of the potential relationship. From my perspective, the motivation for the non-linear effect of CFP on CSR is that when CFP increases, the effect of CFP on CSR may not be constant, because a single factor brought by CFP that affects CSR is not likely to be at dominance with different level of CFP. Given the fact that available funds hypothesis and managerial opportunism hypothesis (Preston and O’Bannon, 1997) may be two most well-known hypotheses about the effect of CFP on CSR, following analysis is based on the combination of available funds hypothesis and managerial opportunism hypothesis (Preston and O’Bannon, 1997).
Similar to the effect of CSR on CFP, the effect of CFP on CSR may also be quadratic. As mentioned above, different CFP may result in different levels of CSR. However, it remains unclear what factor will be at dominance with different CFP. Accordingly, the quadratic effect of CFP on CSR may be U-shaped or inverse U-shaped because there might be different situation about the effect of CFP on CSR. The different situations are discussed below.
CFP CFP
CSR CSR
The left part of the U-shape could be explained by managerial opportunism hypothesis (Preston and O’Bannon, 1997). When CFP gets better, managers may reduce a firm’s participation in socially responsible activities with the purpose of increasing a firm’s short-term profits, thus they could receive more private gains (Preston and O’Bannon, 1997). However, from my perspective, with CFP consistently getting better, the negative relationship won’t last forever. Because the high profitability of a firm will attract public attention, placing this firm in a situation where it is under much pressure to take more CSR. This firm has to show its good corporate citizenship and maintain good relationship with its stakeholders. In this case, even though managers still have their private interests, they couldn’t reduce a firm’s CSR anymore and have to increase it as CFP getting better. Thus it is expected that when CFP reaches a certain point, better CFP will lead to the increase of CSR. The right part of the U-shape could be explained by available funds hypothesis (Preston and O’Bannon, 1997).
When it comes to the inverse U-shape, available funds hypothesis and managerial opportunism hypothesis could also be the explanation for it, but the situation is different. At first, when CFP gets better, a firm will have more resources to invest in CSR (Preston and O’Bannon, 1997), thus the left part shows positive relationship between CFP and CSR. In my opinion, perhaps the firm has already established good relationship with its stakeholders through the positive relationship between CFP and CSR in the left part of the inverse U-shape. As CFP continuously getting better, the firm will spend more money on broadening its business, so stakeholders may be more tolerant for the decrease of CSR. In addition, managers are more likely to cut down the money for socially responsible activities for their private interests (Preston and O’Bannon, 1997). Thus it is expected that when CFP reaches a point, there will be negative relationship between CFP and CSR.
Figure 5
The quadratic effect of CFP on CSR
Panel A shows the U-shaped effect of CFP on CSR and panel B shows the inverse U-shaped effect of CFP on CSR
However, from my perspective, the U-shape or inverse U-shape may capture only part of the effect of CFP on CSR. When studying the effect of CSR on CFP, we have already discussed the cubic or polynomial effect of CSR on CFP. Therefore, there’s reason to believe that linear or quadratic effect of CFP on CSR may not be enough to present all potential relationship between CFP and CSR. Above discussion of the U-shaped or inverse U-shaped relationship provides the basis for the cubic or polynomial effect of CFP on CSR. In previous studies, the cubic or polynomial effect of CFP on CSR is rarely studied. Before discussing the cubic and polynomial relationship, the potential four functional images are depicted in figure 6. All of them could represent the polynomial relationship, and the second and the fourth could also describe the cubic relationship.
Figure 6
The cubic or polynomial effect of CFP on CSR
The functional images in the four panels are depicted according to the sign of the coefficient a and ∆
The left inverse U-shape in panel A could be explained by the reason for the inverse U-shape in quadratic relationship. The change in the right part is an extension of the inverse U-shape. In my view, the explanation for the change is that when CFP reaches a certain level, a firm gets enough resources for its development, and the remaining resources could be used for its participation in socially responsible activities. Stakeholders may be aware of this kind of
CFP CFP
CSR CSR
CFP CFP CFP CFP
CSR CSR CSR CSR
Panel A Panel B
a˃0, ∆˃0 a˃0, ∆≤0 a˂0, ∆˃0 a˂0, ∆≤0
situation, thus at this stage, the firm is under more public scrutiny and social pressure to have more CSR, and it needs to enhance a good corporate image and have better reputation. Therefore, manager’s behaviour is more regulated. They couldn’t decrease CSR even with better CFP. Managerial opportunism hypothesis (Preston and O’Bannon, 1997) is inadequate to describe the relationship after certain point. As a result, there’s positive relationship between CFP and CSR after the inverse U-shape in panel A.
Panel B shows that CFP exerts positive influence on CSR, but the slope changes. Unlike panel A, there won’t be any negative relationship in panel B. In my opinion, at the first stage, the improvement of CFP offers the firm resources. With more resources, the investment in improving operation may be attached great importance for better development opportunities. And managers may put fewer resources for CSR for the fear that the decrease in profitability will affect their private interests (Preston and O’Bannon, 1997). Consequently, though CSR will increase according to available funds hypothesis (Preston and O’Bannon, 1997) and the intention to maintain good corporate image and reputation, CSR will increase at a decreasing pace. At the second stage, when CFP keeps getting better, because the firm already have enough resources for development, more available resources could be used for CSR (Preston and O’Bannon, 1997), guaranteeing the increasing slope of the curve.
There is a U-shape in the left part of panel C. From my perspective, this U-shape could be explained by the reasons for the U-shape in quadratic relationship. However, with the continuing improvement of CFP, a firm may focus on broadening its business. For example, when the firm has developed to certain level, it may have more foreign direct investments and corporate merger and acquisition. These activities consume much money, thus in order to provide enough resources, the firm even needs to cut down CSR expenditures. In my opinion, for those reasons stakeholders might be more tolerant for the decrease of CSR. Moreover, in order to make CFP look better, according to managerial opportunism hypothesis (Preston and O’Bannon, 1997), managers may also deliberately decrease CSR. Above reasons explain why after the U-shape, better CFP results in the decrease of CSR.
O’Bannon, 1997). However, the decrease of CSR couldn’t be so evident otherwise stakeholders will be more likely to notice it. As a result, CSR decreases at a reduced speed. When CFP reaches a certain level, the firm will be faced with many development opportunities and the firm is likely to make it known to public. As long as CSR is not lower than the minimum level required by regulation, stakeholders will accept the decrease of CSR. They expect that when this firm goes through this development stage, it will have enough resources to take its CSR.
After the discussion of how CFP could influence CSR, following hypotheses could be formulated:
H4: CFP affects CSR, and the relationship is linear and positive. H5: CFP affects CSR, and the relationship is linear and negative. H6a: CFP affects CSR, and the relationship is quadratic.
H6b: CFP affects CSR, and the relationship is cubic. H6c: CFP affects CSR, and the relationship is polynomial.
In all cases the null hypothesis is formulated as: H0: There is no relationship between CSR and CFP.
2.4 The lead-lag effect in the CSR-CFP relationship
2.5 The contribution of this study
Previous studies provide inconsistent findings about the relationship between CSR and CFP. So far, I’ve seen little focus on the functional form of the CSR-CFP relationship. This study tends to comprehensively explore the relationship between CSR and CFP based on the functional form. First, the relationship is investigated based on linear function, quadratic function and cubic function. The investigation of the quadratic effect of CFP on CSR, and any cubic and polynomial relationship between CSR and CFP is of a purely exploratory nature. Second, different measures of CSR and CFP could give more insights into the relationship between certain indicators. Third, the study of both contemporaneous relationship and relationship with one year lag tends to find out whether the relationship changes over time.
3. Data and sample
In this section, we first present the data used in this study. Then the variable measurement is introduced. The variable measurement relates to how we measure the independent variables, dependent variables and control variables. In order to facilitate the reader’s understanding of this section, we put the general form of the model here: CFP=CSR+CSR2+CSR3+Control variables; CSR=CFP+CFP2+CFP3+Control variables.
3.1 Data
The data in this study is obtained from DataStream for the period of 2006-2015. The selection of time is based on data availability of CSR (Gonenc and Scholtens, 2017). At the time of collecting data, there are many missing values of CSR before 2006 and in 2016, thus the year before 2006 and the year of 2016 are excluded. Moreover, this length of time is relatively long compared with some existing studies (Nollet et al., 2016; Callan and Thomas, 2009), and it is more likely to avoid some accidental factors and get a thorough analysis using this ten-year period.
score and -1 indicates weakness) (Barnett and Salomon, 2012). Actually KLD data is not proper in this study for the following reasons. First, McWilliams and Siegel (2000) argue that KLD rating system relies heavily on negative screens. This study tends to investigate the CSR-CFP relationship based on functional form, therefore, the focus is not on negative screens of CSR, but on CSR and its different dimensions. Second, many studies use KLD data to measure CSR by aggregating the scores of “key stakeholder attributes” and “controversial business activities” (Barnett and Salomon, 2012; Lee and Park, 2010). But Callan and Thomas (2009) separately test the effect of KLD’s “controversial business issues” on CFP and find that a firm’s participation in “controversial business issues” has much stronger effect on the firm’s returns compared with “key stakeholder attributes”. So the two kinds of indicators may cause biased measurement of CSR (Callan and Thomas, 2009). As a result, it remains unclear how to have a better measure of aggregate CSR if we use KLD data. Third, in KLD data the measure of CSR is binary, thus it fails to differentiate the levels of magnitude of CSR (Lee and Park, 2010). This kind of CSR measurement may influence the accuracy of the findings of functional form, because the numerical values of CSR are limited to certain integers.
In this study, Asset4 ESG database is used to measure CSR and its provider is Thomson Reuters. Asset4 ESG database is a comprehensive database with CSR information of more than 4300 companies. Asset4 ESG data is better than KLD data in this study because Asset4 ESG data doesn't pay much attention to negative aspect of CSR and it emphasizes environmental pillar, social pillar and corporate governance pillar (ESG) of CSR. Accordingly, the measurement of CSR in DataStream better coheres with the definition of CSR in this study. Another advantage of Asset4 ESG data is that the scores of environmental pillar, social pillar and corporate governance pillar range from 0 to100 (not limited to integers), therefore it offers a thorough and accurate measurement of CSR. This measurement is better than the binary measurement of CSR in KLD data. In addition, the data of CFP is also provided by Thomson Reuters. The same data provider for CSR and CFP will limit matching errors compared with combing different data sources (Gonenc and Scholtens, 2017). In fact, more and more recent studies use data provided by Asset4 ESG database to measure CSR (Gonenc and Scholtens, 2017; Miras-Rodríguez et al., 2015; Cheng, Ioannou and Serafeim, 2014).
fixed effect is used. It shows this sentence “Near singular matrix”. After searching for some solutions, I think this problem might be caused by the inclusion of many country dummies. Therefore, in my opinion, it might be solved if companies from some countries are deleted. Then I examine how many firm-year-observations are included in each country/region. I find that there is not valid firm-year-observation for companies from six countries (Argentina, Bahrain, Kuwait, Morocco, Oman and Qatar), and the valid firm-year-observations for companies form other six countries (Czech Republic, Portugal, Egypt, United Arab Emirates, Hungary, and Thailand) are 3, 3, 4, 5, 6, 6 respectively. For the remaining firms from other countries/regions, the smallest firm-year-observations are 12. Consequently, I decide to delete firms from countries with small firm-year-observations. This is based on the ranking of the number of valid firm-year-observations. At last, I delete firms from the aforementioned twelve countries, among which the largest valid firm-year-observations are 6. From my perspective, the deletion of companies from the twelve countries won’t exert much influence on the results because of their very limited firm-year-observations. After deleting 12 countries, Eviews could produce the results. The final sample consists of companies from 37 countries/regions. Those countries/regions are listed in appendix A.
The sample is unbalanced and observations with missing values for firm-year-observations are excluded. When Return on Assets is used as dependent variable, 12620 firm-year-observations are included in contemporaneous relationship when we test CSR-CFP relationship. When Excess Stock Return is used as dependent variable, 12272 firm-year-observations are included in contemporaneous relationship when we test the CSR-CFP relationship. For monetary variables, their values are converted into US dollars by historical exchange rate in order to provide better comparability. I observe outliers because some observations fail to fit in with the remaining data’s pattern in many variables (Brooks, 2014). Brooks (2014) mentions that serious effect can be caused by outliers. Therefore, all variables are winsorized at 1% level to avoid the possible negative effects of outliers.
3.2 Variable measurements
3.2.1 Independent variable: Corporate social responsibility
instead of required daily activities. In fact, the study of CSR should focus on its different aspects. Except for social dimension, environmental and governance dimension should also be taken into consideration. According to Nollet et al. (2016), it is better to have disaggregate measure for CSR because CSR is a concept with different dimensions, and the effect of one dimension could offset the effect of another dimension. This kind of view is in line with Callan and Thomas (2009), who also argue that the influences of different components of CSR are different. Accordingly, in order to have a better investigation of the effects of CSR on CFP, it’s better to disaggregate CSR into different components. In this study, corporate social responsibility is measured in two ways: aggregate CSR and disaggregate CSR.
Asset4 ESG database has 4 pillars to measure the overall performance of a firm’s CSR. They are environmental pillar, social pillar, corporate governance pillar and economic pillar. These pillars consist of over 250 key performance indicators. The economic pillar is left out because the dependent variable (ROA and Excess Stock Return) is financial performance. Therefore, the use of economic pillar as one component of independent variable will affect the results.
Cheng et al. (2014) indicate that there’s a lack of theoretical guidance in regard to the weighting of each measure in order to construct the aggregate CSR score. They follow previous studies to construct an aggregate CSR score by giving equal weights to the three pillars: environmental, social and corporate governance. In this study, we follow Cheng et al. (2014) to take the average of environmental pillar, social pillar and corporate governance pillar to construct aggregate CSR (ESG). In addition, in order to observe the influence of different CSR dimension, environmental pillar (ENV), social pillar (SOC) and corporate governance pillar (CGV) are used separately to examine how they affect CFP.
3.2.2 Dependent variable: Corporate financial performance
Wagner (2010) indicates that stock market-based measures are better than accounting-based measures because accounting-based measures are easier to be influenced by managerial discretion when using accounting rules or by times of high inflation. He also states that accounting-based measure is based on past data, but stock market-based measure is based on expected future performance. However, the meta-analysis of Orlitzky et al. (2003) finds that corporate social performance is more related to accounting-based measures of CFP compared with market-based measures. Nollet et al. (2016) argue that firms’ revenues could benefit from CSR activities, but this positive effect may not exist in excess stock return, for the reason that many other factors could influence stock returns simultaneously. Findings of the relationship between CSR and CFP could be different depending on which measure of CFP is applied.
In this study, ROA is defined as:
𝑅𝑂𝐴 =𝑁𝑒𝑡 𝐼𝑛𝑐𝑜𝑚𝑒 − 𝐵𝑜𝑡𝑡𝑜𝑚 𝐿𝑖𝑛𝑒 + (𝐼𝑛𝑡𝑒𝑟𝑒𝑠𝑡 𝐸𝑥𝑝𝑒𝑛𝑠𝑒 𝑜𝑛 𝐷𝑒𝑏𝑡 − 𝐼𝑛𝑡𝑒𝑟𝑒𝑠𝑡 𝐶𝑎𝑝𝑖𝑡𝑎𝑙𝑖𝑧𝑒𝑑) × (1 − 𝑇𝑎𝑥 𝑅𝑎𝑡𝑒) 𝐴𝑣𝑒𝑟𝑎𝑔𝑒 𝑜𝑓 𝐿𝑎𝑠𝑡 𝑌𝑒𝑎𝑟′𝑠 𝑎𝑛𝑑 𝐶𝑢𝑟𝑟𝑒𝑛𝑡 𝑌𝑒𝑎𝑟′𝑠 𝑇𝑜𝑡𝑎𝑙 𝐴𝑠𝑠𝑒𝑡𝑠
Excess stock return is defined as:
𝐸𝑆𝑅 = 𝐿𝑛 𝑅𝐼𝑡 𝑅𝐼𝑡−1
− 𝐿𝑛 𝑅𝑀𝑡 𝑅𝑀𝑡−1
where RI refers to return index, and RM refers to local index. The code of the local market index used for each country/region in DataStream could be found in appendix B.
3.2.3 Control variables
(1) Firm size
CSR and CFP use firm size as control variables (Waddock and Graves, 1997; Elsayed and Paton, 2009; Wagner, 2010). In this study, firm size is measured by the natural logarithm of a firm’s total assets in US dollar.
𝐹𝑖𝑟𝑚 𝑠𝑖𝑧𝑒 = 𝐿𝑛 (𝑇𝑜𝑡𝑎𝑙 𝐴𝑠𝑠𝑒𝑡𝑠)
(2) Leverage
Barnett and Salomon (2012) indicate that the behaviour of manager is influenced by debt. They point out that managers could be incentivized by debt to make decisions that are best for the firm, however, they may also have less opportunities to conduct business because of decreased managerial latitude. In this situation, a firm’s profit would be negatively influenced. Furthermore, high level of leverage will put a firm at risk and result in less capital for it to take part in socially responsible activities. Leverage is measured by the ratio of a firm’s total debt divided by its total assets. Park and Lee (2010) and Nollet et al. (2016) use leverage as control variable.
𝐿𝑒𝑣𝑒𝑟𝑎𝑔𝑒 = 𝑇𝑜𝑡𝑎𝑙 𝑑𝑒𝑏𝑡 𝑇𝑜𝑡𝑎𝑙 𝑎𝑠𝑠𝑒𝑡𝑠
(3) Sales growth
The meta-analysis of Capon, Farley and Hoenig (1990) finds that sales growth has positive effects on CFP not only at firm level, but also at industry level. It could be expected that sales growth lead to more profits, thus a firm would have more capital to invest in socially responsible activities. Sales growth is used as a control variable in the study of Wagner (2010). In this study, sales growth is measured by the ratio of the difference between current year’s net sales and last year’s net sales divided by last year’s net sales.
Sales growth =𝐶𝑢𝑟𝑟𝑒𝑛𝑡 𝑦𝑒𝑎𝑟
′𝑠 𝑛𝑒𝑡 𝑠𝑎𝑙𝑒𝑠 − 𝑙𝑎𝑠𝑡 𝑦𝑒𝑎𝑟′𝑠 𝑛𝑒𝑡 𝑠𝑎𝑙𝑒𝑠 𝐿𝑎𝑠𝑡 𝑦𝑒𝑎𝑟′𝑠 𝑛𝑒𝑡 𝑠𝑎𝑙𝑒𝑠
(4) Networking capital
capital management might facilitate a firm’s taking of CSR because of better profitability. To study the relationship between environmental and financial performance, Gonenc and Scholtens (2017) use networking capital as a control variable. Following Gonenc and Scholtens (2017), networking capital is measured as the ratio of the difference between current assets and current liabilities divided by the book value of total assets.
Networking capital =𝐶𝑢𝑟𝑟𝑒𝑛𝑡 𝑎𝑠𝑠𝑒𝑡𝑠 − 𝑐𝑢𝑟𝑟𝑒𝑛𝑡 𝑙𝑖𝑎𝑏𝑖𝑙𝑖𝑡𝑖𝑒𝑠 𝑇𝑜𝑡𝑎𝑙 𝑎𝑠𝑠𝑒𝑡𝑠
(5) R&D
McWilliams and Siegel (2000) regard R&D as a kind of investment in “technical capital”, which can be seen as knowledge enhancement, resulting in product and process innovation. In this way, a firm’s productivity could be improved. They also cite the view of Griliches (1979) to emphasize the importance of R&D on long-run economic performance. The result of McWilliams and Siegel (2000) provides evidence that there is high correlation between CSR and R&D. They argue that it is difficult to investigate the influence of CSR on CFP without adding R&D as control variable. Thus the positive effect of R&D on CFP and CSR is expected. R&D is measured by the ratio of R&D expenditure divided by total assets. A Large number of CSR studies use R&D as control variables (McWilliams and Siegel, 2000; Barnett and Salomon, 2012; Nollet et al, 2016). Because many firms don’t have R&D information, the inclusion of this control variable reduces valid firm-year-observations.
R&D =𝑅&𝐷 𝑒𝑥𝑝𝑒𝑛𝑑𝑖𝑡𝑢𝑟𝑒 𝑇𝑜𝑡𝑎𝑙 𝑎𝑠𝑠𝑒𝑡𝑠
(6) Country
(7) Industry
Industry factors are important for firms because those factors give an explanation to the differences in profitability (McWilliams and Siegel, 2000). The level of input in R&D, which drives product innovation, varies from industries to industries. Combining product characteristics, those factors result in different marginal profit. Social and environmental regulation are not the same across industries, leading to different public concern and stakeholder expectations, thus different industries are under different pressure to conduct socially responsible activities. Therefore, industry is also expected to influence CFP and CSR. Industry dummies are used in many previous studies (Waddock and Graves, 1997; Callan and Thomas, 2009; Wagner, 2010). This study uses 4-digit SIC code to classify industries and this produces 9 industry dummies.
(8) Year
Year dummies are also included because from macro level, the macro-economic condition is different in different year, affecting a firm’s state of operation. From micro level, individual firm may also have their own opportunities for development or encounter difficulties in production and operation in different year. Moreover, a firm may also exhibit different level of CSR in different year. Lee and Park (2010) and Wagner (2010) use time dummies in their studies.
4. Methodology
In this section, we first introduce the method of reducing multicollinearity, after that we introduce the use of one-year lag and the test of unit root. At last we present the regression models.
4.1 Reducing multicollinearity
the difference between CSR and its mean value. After that the difference is squared and cubed respectively to represent quadratic and cubic relationship. Dalal and Zickar (2012) argue that mean-centering could lower nonessential collinearity and ill-conditioning of the data without changing the fit of regression models.
This study uses the process of centering to reduce multicollinearity problem. If a variable has a prefix “ADJ”, it means this variable is under the process of centering. We use the process of certering for CSR (ESG, ENV, SOC, and CGV) and CFP (ROA and ESR). For example, ADJESG represents the difference between ESG and its mean value, ADJESG2 and ADJESG3 mean that the difference is raised to second and third functions. For every regression result, VIF test is conducted to test the degree of multicollinearity. The results of VIF test are in appendix G and H.
4.2 The use of one-year lag
This study tests the CSR-CFP relationship with zero lag and one-year lag in independent variables seperately. In all cases there is no lag in dependent variable. Waddock and Graves (1997) find that last year’s financial performance positively affects this year’s social performance, and this year’s social performance positively affects next year’s financial performance. In fact, many studies also test the CSR-CFP relationship with one year lag. Thus the use of one-year lag in this study could not only have better comparison with contemporaneous relationship, but also with some former studies. It provides insights into the lead-lag effects in the CSR-CFP relationship.
4.3 The test of unit root
Before doing regression analysis, it is necessary to test whether unit root exists. If a variable has a unit root, it is not stationary and the use of this variable will lead to spurious regression. As a result, even regression result indicates significant relationship, it is not reliable. Therefore, it is necessary to make sure that all variables used in regression analysis are stationary.
is in appendix C. It shows that for all variables, the null hypothesis that the variable has a unit root is rejected. Therefore, the stationary condition is satisfied and we could use the variables in regression models.
4.4 Regression models
Ordinary least squares (OLS) regression method is used in this study to run the unbalanced panel data. This method is widely used in existing studies to test the relationship between CSR and CFP. Following regression models are used to test the hypotheses.
Model 1, model 2, model 3, model 4 and model 5 test the effect of CSR on CFP, but they focus on different kinds of CSR. Model 1 examines the effect of aggregate CSR, model 2, model 3 and model 4 tests the effects of environmental pillar, social pillar and corporate governance pillar on financial performance respectively, and model 5 puts all three pillars together to examine their effects. At first, for all models, linear term is included to test hypothesis 1 and 2, then quadratic term is included to test hypothesis 3a and hypothesis 3c, after that cubic term is included to test hypothesis 3b and hypothesis 3c.
𝐴𝐷𝐽𝐹𝑃𝑖,𝑡= 𝛼 + 𝛽1𝐴𝐷𝐽𝐸𝑆𝐺𝑖,𝑡−𝑗+ 𝛽2𝐴𝐷𝐽𝐸𝑆𝐺𝑖,𝑡−𝑗2 + 𝛽3𝐴𝐷𝐽𝐸𝑆𝐺𝑖,𝑡−𝑗3 + 𝛽4𝐴𝐷𝐽𝐹𝑃𝑖,𝑡−1+ 𝛽5𝑆𝐼𝑍𝐸𝑖,𝑡+ 𝛽6𝑁𝑆𝐺𝑖,𝑡+ 𝛽7𝑁𝐶𝑖,𝑡+ 𝛽8𝑅&𝐷𝑖,𝑡+ 𝛽9𝑅𝐼𝑆𝐾𝑖,𝑡+ 𝛾𝐼𝑁𝐷𝑖+ 𝛿𝐶𝑂𝑈𝑖+ 𝜀𝑖,𝑡 (1) 𝐴𝐷𝐽𝐹𝑃𝑖,𝑡= 𝛼 + 𝛽1𝐴𝐷𝐽𝐸𝑁𝑉𝑖,𝑡−𝑗+ 𝛽2𝐴𝐷𝐽𝐸𝑁𝑉𝑖,𝑡−𝑗2 + 𝛽3𝐴𝐷𝐽𝐸𝑁𝑉𝑖,𝑡−𝑗3 + 𝛽4𝐴𝐷𝐽𝐹𝑃𝑖,𝑡−1+ 𝛽5𝑆𝐼𝑍𝐸𝑖,𝑡+ 𝛽6𝑁𝑆𝐺𝑖,𝑡+ 𝛽7𝑁𝐶𝑖,𝑡+ 𝛽8𝑅&𝐷𝑖,𝑡+ 𝛽9𝑅𝐼𝑆𝐾𝑖,𝑡+ 𝛾𝐼𝑁𝐷𝑖+ 𝛿𝐶𝑂𝑈𝑖+ 𝜀𝑖,𝑡 (2) 𝐴𝐷𝐽𝐹𝑃𝑖,𝑡= 𝛼 + 𝛽1𝐴𝐷𝐽𝑆𝑂𝐶𝑖,𝑡−𝑗+ 𝛽2𝐴𝐷𝐽𝑆𝑂𝐶𝑖,𝑡−𝑗2 + 𝛽3𝐴𝐷𝐽𝑆𝑂𝐶𝑖,𝑡−𝑗3 + 𝛽4𝐴𝐷𝐽𝐹𝑃𝑖,𝑡−1+ 𝛽5𝑆𝐼𝑍𝐸𝑖,𝑡+ 𝛽6𝑁𝑆𝐺𝑖,𝑡+ 𝛽7𝑁𝐶𝑖,𝑡+ 𝛽8𝑅&𝐷𝑖,𝑡+ 𝛽9𝑅𝐼𝑆𝐾𝑖,𝑡+ 𝛾𝐼𝑁𝐷𝑖+ 𝛿𝐶𝑂𝑈𝑖+ 𝜀𝑖,𝑡 (3) 𝐴𝐷𝐽𝐹𝑃𝑖,𝑡= 𝛼 + 𝛽1𝐴𝐷𝐽𝐶𝐺𝑉𝑖,𝑡−𝑗+ 𝛽2𝐴𝐷𝐽𝐶𝐺𝑉𝑖,𝑡−𝑗2 + 𝛽3𝐴𝐷𝐽𝐶𝐺𝑉𝑖,𝑡−𝑗3 + 𝛽4𝐴𝐷𝐽𝐹𝑃𝑖,𝑡−1+ 𝛽5𝑆𝐼𝑍𝐸𝑖,𝑡+ 𝛽6𝑁𝑆𝐺𝑖,𝑡+ 𝛽7𝑁𝐶𝑖,𝑡+ 𝛽8𝑅&𝐷𝑖,𝑡+ 𝛽9𝑅𝐼𝑆𝐾𝑖,𝑡+ 𝛾𝐼𝑁𝐷𝑖+ 𝛿𝐶𝑂𝑈𝑖+ 𝜀𝑖,𝑡 (4) 𝐴𝐷𝐽𝐹𝑃𝑖,𝑡= 𝛼 + 𝛽1𝐴𝐷𝐽𝐸𝑁𝑉𝑖,𝑡−𝑗+ 𝛽2𝐴𝐷𝐽𝐸𝑁𝑉𝑖,𝑡−𝑗2 + 𝛽3𝐴𝐷𝐽𝐸𝑁𝑉𝑖,𝑡−𝑗3 + 𝛽4𝐴𝐷𝐽𝑆𝑂𝐶𝑖,𝑡−𝑗+ 𝛽5𝐴𝐷𝐽𝑆𝑂𝐶𝑖,𝑡−𝑗2 + 𝛽6𝐴𝐷𝐽𝑆𝑂𝐶𝑖,𝑡−𝑗3 + 𝛽7𝐴𝐷𝐽𝐶𝐺𝑉𝑖,𝑡−𝑗+ 𝛽8𝐴𝐷𝐽𝐶𝐺𝑉𝑖,𝑡−𝑗2 + 𝛽9𝐴𝐷𝐽𝐶𝐺𝑉𝑖,𝑡−𝑗3 + 𝛽10𝐴𝐷𝐽𝐹𝑃𝑖,𝑡−1+ 𝛽11𝑆𝐼𝑍𝐸𝑖,𝑡+ 𝛽12𝑁𝑆𝐺𝑖,𝑡+ 𝛽13𝑁𝐶𝑖,𝑡+ 𝛽14𝑅&𝐷𝑖,𝑡+ 𝛽15𝑅𝐼𝑆𝐾𝑖,𝑡+ 𝛾𝐼𝑁𝐷𝑖+ 𝛿𝐶𝑂𝑈𝑖+ 𝜀𝑖,𝑡 (5)
ADJFP i,t-1 is included in order to control within-firm persistence in CFP (Barnett and Salomon, 2012). SIZE i,t is the firm size of firm i at time t; NSG i,t is the net sales growth of firm i at time t; NCi,t is thenetworking capital of firm i at time t; R&D i,t is the research and development expenditure of firm i at time t; RISK i,t is the leverage of firm i at time t; INDi is the industry of firm i and COUi is the country that firm i belongs to. 𝜀𝑖,𝑡 is the error term of firm i at time t. If j equals to 0, it means we test contemporaneous relationship; if j equals to 1, it means we test relationship with one-year lag.
Model 6, model 7, model 8 and model 9 tests the influence of CFP on CSR. Model 6 examines the effect of CFP (ADJROA and ADJESR) on aggregate CSR; model 7 examines the effect of CFP on environmental pillar; model 8 examines the effect of CFP on social pillar; model 9 examines the effect of CFP on corporate governance pillar.
𝐴𝐷𝐽𝐸𝑆𝐺𝑖,𝑡= 𝛼 + 𝛽1𝐴𝐷𝐽𝐹𝑃𝑖,𝑡−𝑗+ 𝛽2𝐴𝐷𝐽𝐹𝑃𝑖,𝑡−𝑗2 + 𝛽3𝐴𝐷𝐽𝐹𝑃𝑖,𝑡−𝑗3 + 𝛽4𝐴𝐷𝐽𝐸𝑆𝐺𝑖,𝑡−1+ 𝛽5𝑆𝐼𝑍𝐸𝑖,𝑡+ 𝛽6𝑁𝑆𝐺𝑖,𝑡+ 𝛽7𝑁𝐶𝑖,𝑡+ 𝛽8𝑅&𝐷𝑖,𝑡+ 𝛽9𝑅𝐼𝑆𝐾𝑖,𝑡+ 𝛾𝐼𝑁𝐷𝑖+ 𝛿𝐶𝑂𝑈𝑖+ 𝜀𝑖,𝑡 (6) 𝐴𝐷𝐽𝐸𝑁𝑉𝑖,𝑡= 𝛼 + 𝛽1𝐴𝐷𝐽𝐹𝑃𝑖,𝑡−𝑗+ 𝛽2𝐴𝐷𝐽𝐹𝑃𝑖,𝑡−𝑗2 + 𝛽3𝐴𝐷𝐽𝐹𝑃𝑖,𝑡−𝑗3 + 𝛽4𝐴𝐷𝐽𝐸𝑁𝑉𝑖,𝑡−1+ 𝛽5𝑆𝐼𝑍𝐸𝑖,𝑡+ 𝛽6𝑁𝑆𝐺𝑖,𝑡+ 𝛽7𝑁𝐶𝑖,𝑡+ 𝛽8𝑅&𝐷𝑖,𝑡+ 𝛽9𝑅𝐼𝑆𝐾𝑖,𝑡+ 𝛾𝐼𝑁𝐷𝑖+ 𝛿𝐶𝑂𝑈𝑖+ 𝜀𝑖,𝑡 (7) 𝐴𝐷𝐽𝑆𝑂𝐶𝑖,𝑡= 𝛼 + 𝛽1𝐴𝐷𝐽𝐹𝑃𝑖,𝑡−𝑗+ 𝛽2𝐴𝐷𝐽𝐹𝑃𝑖,𝑡−𝑗2 + 𝛽3𝐴𝐷𝐽𝐹𝑃𝑖,𝑡−𝑗3 + 𝛽4𝐴𝐷𝐽𝑆𝑂𝐶𝑖,𝑡−1+ 𝛽5𝑆𝐼𝑍𝐸𝑖,𝑡+ 𝛽6𝑁𝑆𝐺𝑖,𝑡+ 𝛽7𝑁𝐶𝑖,𝑡+ 𝛽8𝑅&𝐷𝑖,𝑡+ 𝛽9𝑅𝐼𝑆𝐾𝑖,𝑡+ 𝛾𝐼𝑁𝐷𝑖+ 𝛿𝐶𝑂𝑈𝑖+ 𝜀𝑖,𝑡 (8) 𝐴𝐷𝐽𝐶𝐺𝑉𝑖,𝑡= 𝛼 + 𝛽1𝐴𝐷𝐽𝐹𝑃𝑖,𝑡−𝑗+ 𝛽2𝐴𝐷𝐽𝐹𝑃𝑖,𝑡−𝑗2 + 𝛽3𝐴𝐷𝐽𝐹𝑃𝑖,𝑡−𝑗3 + 𝛽4𝐴𝐷𝐽𝐶𝐺𝑉𝑖,𝑡−1+ 𝛽5𝑆𝐼𝑍𝐸𝑖,,𝑡+ 𝛽6𝑁𝑆𝐺𝑖,𝑡+ 𝛽7𝑁𝐶𝑖,𝑡+ 𝛽8𝑅&𝐷𝑖,𝑡+ 𝛽9𝑅𝐼𝑆𝐾𝑖,𝑡+ 𝛾𝐼𝑁𝐷𝑖+ 𝛿𝐶𝑂𝑈𝑖+ 𝜀𝑖,𝑡 (9)
For all models, linear term is added to to test H4 and H5, quadratic term is added to test H6a and H6c, and cubic term is added to test H6b and H6c. The regression results are in appendix E and F.
5. Descriptive statistics and correlations 5.1 Descriptive statistics
N Mean Median Std. Dev. Minimum Maximum Skewness Kurtosis ESG 31813 50.496 50.020 24.936 6.303 93.093 -0.008 1.832 ENV 31813 50.254 48.240 32.092 8.450 97.310 0.077 1.370 SOC 31813 50.386 49.810 31.250 4.550 96.770 0.023 1.500 CGV 31813 50.861 56.630 30.307 1.800 95.380 -0.230 1.629 ADJESG 31813 0.000 -0.476 24.936 -44.193 42.597 -0.008 1.832 ADJENV 31813 0.000 -2.014 32.092 -41.804 47.056 0.077 1.370 ADJSOC 31813 0.000 -0.576 31.250 -45.836 46.384 0.023 1.500 ADJCGV 31813 0.000 5.769 30.307 -49.061 44.519 -0.230 1.629 ADJESG2 31813 621.763 434.293 567.175 0.000 1953.025 0.750 2.332 ADJENV2 31813 1029.881 1155.016 626.208 0.000 2214.223 -0.268 1.702 ADJSOC2 31813 976.527 970.679 690.687 0.000 2151.501 0.071 1.618 ADJCGV2 31813 918.499 801.378 728.258 0.000 2407.019 0.428 1.963 ADJESG3 31813 -125.922 -0.108 31262.150 -86310.110 77292.220 -0.193 4.069 ADJENV3 31813 2549.752 -8.175 47241.170 -73058.050 104191.500 0.259 1.984 ADJSOC3 31813 715.127 -0.191 48494.590 -96296.910 99795.800 0.054 2.476 ADJCGV3 31813 -6406.216 191.962 48586.420 -118091.700 88231.790 -0.573 2.971 ROA 45811 0.055 0.053 0.102 -0.436 0.357 -1.351 10.295 ESR 42592 0.022 0.034 0.362 -1.207 1.046 -0.344 4.672 ADJROA 45811 0.000 -0.002 0.102 -0.491 0.302 -1.351 10.295 ADJESR 42592 0.000 0.012 0.362 -1.229 1.024 -0.344 4.672 ADJROA2 45811 0.010 0.001 0.032 0.000 0.241 5.460 36.287 ADJESR2 42592 0.131 0.036 0.251 0.000 1.511 3.366 15.482 ADJROA3 45811 -0.001 0.000 0.014 -0.118 0.028 -6.749 54.096 ADJESR3 42592 -0.016 0.000 0.283 -1.858 1.075 -2.968 26.059 SIZE 47279 15.116 15.077 1.879 10.107 20.091 0.037 3.282 NSG 46260 0.153 0.072 0.469 -0.616 3.298 4.139 26.090 NC 36993 0.121 0.117 0.278 -1.440 0.753 -1.943 13.318 R&D 20037 0.037 0.013 0.063 0.000 0.371 3.108 14.189 RISK 47213 0.242 0.220 0.194 0.000 0.837 0.757 3.152
Table 1 Descriptive statistics
Note: ESG=the aggregated CSR scores; ENV, SOC and CGV=environmental, social and corporate governance scores; ADJESG, ADJENV, ADJSOC and ADJCGV= ESG-mean (ESG), ENV-mean (ENV), SOC-mean (SOC) and CGV-mean (CGV); ROA=(Net income-Bottom Line+((Interest Expense on Debt-Interest Capitalized)*(1-Tax Rate)))/Average of Last year’s and Current Year’s Total Assets; ESR=Excess Stock Return=Ln(Total Return Index(t)/Total Return Index(t-1))-Ln(Local Market Index(t)/Local Market Index(t-1)); ADJROA and
5.2 Correlations
The correlation matrix is reported in appendix D. The correlation between disaggregate CSR (ADJENV, ADJSOC and ADJCGV) and aggregate CSR (ADJESG) is high (0.826, 0.897 and 0.613 respectively). It is normal because ADJESG is calculated by taking the average of ADJENV, ADJSOC and ADJCG. Among disaggregate CSR, the correlation between ADJSOC and ADJENV is high (0.803), but the correlations between ADJSOC and ADJCGV, and ADJENV and ADJCGV are relatively low (0.298 and 0.128 respectively). It shows that there is strong correlation between ADJSOC and ADJENV, but ADJCGV doesn’t have strong correlation with ADJSOC and ADJENV. Therefore, the result of regression model 5 might be not reliable because it contains ADJENV, ADJSOC and ADJCGV in the same model. The correlation matrix also shows that the correlation between firm size and ADJENV, and firm size and ADJSOC are larger than 0.5 (0.518 and 0.511 respectively). If this study only uses aggregate CSR as independent variable, this kind of correlation will not be observed. However, given the fact that firm size is likely to affect financial performance and it is a widely used control variable in testing the relationship between CSR and CFP, and the correlations (0.518 and 0.511) are only slightly larger than 0.5, firm size is still used as control variable in this study.
6. Regression results
The criteria for choosing the regression model that describes the relationship between the dependent variables and the independent variable are first based on the value of VIF. If in a regression model the VIF of a variable is larger than 10, the result of this regression model is not taken into consideration. The result of VIF test is in appendix G and H. Given the fact that the values of adjusted R2 (F-statistic) are close to each other in different models that test the relationship between the same variables when the lag length of the independent variable is same, we first look at the model in which the highest term is cubic term. If none of the coefficient of linear term, quadratic term and cubic term is significant, then we look at the model in which the highest term is quadratic term; if neither of the coefficient of the quadratic term nor the linear term is significant, we look at the model that only has linear term; if the coefficient of linear term is not significant, we conclude that there’s no effect of the independent variable on the dependent variable.
6.1 Contemporaneous Relationship 6.1.1The effect of CSR on CFP
Table 2
The effect of CSR on CFP
This table summarizes the results of the effect of CSR on CFP when the dependent variable or the independent variable has zero lag length. It is based on the regression results using the process of centering in appendix E1, E2 and E3. DV is the dependent variable and IV is the independent variable. For the reason of brevity, ADJ is not added before DV and IV. Functional form describes the relationship between IV and DV in detail (the effects of control variables and constant are not reported here). *, **, and *** show that the significant level of the coefficient is at 10%, 5% and 1% respectively.
DV IV Relationship Functional Form
ROAt ESGt Cubic (1.02E-07**)×ESGt
3
ESRt ESGt Quadratic (1.24E-05**)×ESGt2
ROAt ENVt Linear (9.31E-05***)×ENVt
ROAt SOCt Polynomial (1.66E-07***)×SOCt
3
+(-9.87E-05*)×SOCt ROAt CGVt Polynomial (1.82E-06**)×CGVt2+(1.68E-04***)×CGVt
ESRt ENVt Linear (-3.26E-04***)×ENVt
ESRt SOCt Quadratic (1.30E-05***)×SOCt
2
ESRt CGVt Linear (-2.94E-04*)×CGVt
findings exhibit linear relationship between CSR and CFP. It shows that in contemporaneous relationship, we could find more non-linear relationship between CSR and CFP using different indicators for DV and IV. However, many previous studies fail to pay attention to the non-linear relationship between CSR and CFP.
When it comes to the functional form, we find that most of the coefficients are significant at 5% and 1% level. In general, it shows that the effect of CSR on CFP is significant in contemporaneous relationship. In each functional form, the size of the coefficient is relatively small, even when the functional form only contains linear term. This may be partly explained by the fact that the absolute value of the dependent variable (ROA and ESR) is much smaller than the absolute value of the independent variable (ESG, ENV, SOC and CGV). Of course we couldn’t ignore the effect of control variable, thus the absolute value is only one kind of possibility. Another explanation for the small coefficient is that when the functional form contains the quadratic or cubic term, the effects of CSR will be amplified. As a result, the absolute value of the product will be much larger than the coefficient. As for the coefficients of the linear term, overall, they are a little larger than the coefficients of the non-linear term. In my opinion, the larger coefficient of the linear term compensates for the low amplifying effect of the linear term. The finding of the relatively small coefficients of CSR is in line with the meta-analysis of Margolis et al. (2009). They argue that the overall effect of CSR on CFP is positive but small. But according to the analysis above, except for the small effect of CSR on CFP, the absolute value of dependent variable and independent variable, and the amplifying effect of non-linear term might also be the reasons for the relatively small coefficients.
insignificant or negative financial growth, when a firm’s ESG reaches a certain level, a unit increase of ESG will bring relatively higher increase of CFP. This result provides motivation for a firm to increase its aggregate CSR. If a firm is too shortsighted to ignore the importance of CSR, it may fail to take advantage of CSR.
When disaggregate CSR is used to test the effects of ENV, SOC and CGV on CFP, we observe that the effects of those different dimensions of CSR are different, and the effects of any dimension of CSR are also different from the effects of aggregate CSR. This finding
confirms the argumentation of Callan and Thomas (2009), who indicate that the influences of
different components of CSR are different. When ROA is used as dependent variable, table 2 shows that ENV has linear effect on ROA, however, SOC and CGV have non-linear effect on ROA. When ESR is used as dependent variable, ENV and CGV have linear effect on ESR, but SOC has non-linear effect on ESR. In addition, the sign of the coefficients of linear term may also be different.
6.1.2 The effect of CFP on CSR
Table 3
The effect of CFP on CSR
This table summarizes the results of the effect of CFP on CSR when the dependent variable or the independent variable has zero lag length. It is based on the regression results using the process of centering in appendix E4, E5, E6 and E7. DV is the dependent variable and IV is the independent variable. For the reason of brevity, ADJ is not added before DV and IV. Functional form describes the relationship between IV and DV in detail (the effects of control variables and constant are not reported here). *, **, and *** show that the significant level of the coefficient is at 10%, 5% and 1% respectively.
DV IV Relationship Functional Form
ESGt ROAt Polynomial -39.765**×ROAt3-19.286***×ROAt2+4.405***×ROAt
ESGt ESRt Polynomial -0.956**×ESRt2+0.937***×ESRt
ENVt ROAt Quadratic -17.676**×ROAt2
ENVt ESRt No Relationship NA
SOCt ROAt Polynomial -77.902***×ROAt3-26.843***×ROAt2+9.038***×ROAt
SOCt ESRt Polynomial -1.595***×ESRt2+1.377***×ESRt
CGVt ROAt Polynomial -15.034**×ROAt2+3.473**×ROAt
CGVt ESRt Polynomial -1.020**×ESRt2+0.992**×ESRt
This table tests the following hypotheses: H4, H5, H6a, H6b and H6c. H4 and H5 are not supported in all cases, but H6a and H6c are not rejected if certain variables are used for DV and IV. When ENV is used as dependent variable and ESR is used as the independent variable, H0 is not rejected, indicating that there’s no relationship between the two variables. Although we couldn’t find pure cubic relationship which supports H6b, cubic term exists in some polynomial relationship. Among the eight findings, six findings show polynomial relationship, one finding indicates quadratic relationship and the remaining finding reveals that there’s no relationship between the DV and IV. Therefore, polynomial relationship is at dominance.
