Faculty of Economics and Business
Doing good or doing well Examining the link between Corporate Social Responsibility and Firm Performance 15th of July, 2018
This paper examines the relationship between Corporate Social Responsibility (CSR) and Corporate Firm Performance (CFP) in both the short- and the long run, by constructing a panel of data of more than 2700 U.S firms over a 13-year period from 1994 to 2006. This relationship is empirically tested by means of instrumental variables
and 2SLS. ROA and Tobin’s Q are used to measure short- and long-term CFP respectively. In order to proxy CSR, an index is created using data on Corporate Social
Performance (CSP) obtained from the KLD Stats database. The results show that a positive link exists between CSR and CFP in the short run for firms with high levels of
awareness. In the long run this effect was only present for firms with high reputational scores.
Master Thesis: Business Economics Number of ECTS: 15
Specialization: Managerial Economics & Strategy
Name: Thierry Belt
Email Address: thierry.belt@student.uva.nl Student Number: 10587713
Thesis Supervisor: dr. S. Dominguez-Martinez
Word Count: 21198
1
Statement of Originality
This document is written by Thierry Belt who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.
The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
2 Table of Contents 1. Introduction ... 4 2. Literature review ... 7 2.1 Theoretical Channels of CSR ... 7 2.2 Empirical Findings ... 11 2.3 Hypotheses ... 13 3. Methodology ... 15 3.1 Variable Description ... 15 3.2 Regression Set-up ... 24 3.2.1 Panel data ... 25 3.2.2 Cross-sectional model ... 28 3.2.3 Reversed Relation ... 29 4. Results ... 30 4.1 Panel Model... 31 4.2 Cross-Sectional Model ... 35 4.3 Reversed Relation ... 38 5. Discussion of Results ... 40 6. Limitations ... 43 7. Conclusion ... 45 8. References ... 47 9. Appendix ... 54
9.1 Corporate Social Performance ... 54
9.2 Variable Selection ... 56
9.3 Robustness Checks ... 58
9.3.1 Panel Model ... 58
9.3.2. Cross-sectional Model ... 62
9.3.3 Reversed Relation ... 65
9.4 Hausman specification test ... 66
9.5 Time fixed effects... 67
9.6 Dummy Variable Adjustment Method ... 68
9.7 Instrumental Variables ... 71
9.7.1 Tests of Endogeneity – Cross-section ... 71
3 Table of Graphs and Figures
Table 1 Summary Statistics ... 30
Table 2 Correlations ... 31
Table 3 Regression analyses, ROA. ... 33
Table 4 Regression analyses, Tobin's Q. ... 35
Table 5 Regression analyses, cross-section, ROA. ... 37
Table 6 Regression analyses, cross-section, Tobin’s Q ... 38
Table 7 Regression analyses, reversed relationship ... 39
Table 8 Corporate Social Performance, KLD ... 54
Table 9 Variable selection ... 57
Table 10 Robustness Check Regression analyses, ROE ... 58
Table 11 Robustness Check Regression analyses, ROS ... 59
Table 12 Robustness Check Regression analyses, Net Income ... 60
Table 13 Robustness Check Regression analyses, Market Capitalization ... 61
Table 14 Regression analyses, cross-section, ROE ... 62
Table 15 Regression analyses, cross-section, ROS ... 63
Table 16 Regression analyses, cross-section, Net Income ... 64
Table 17 Regression analyses, cross-section, Market Capitalization ... 65
Table 18 Robustness Check, Reversed Link ... 66
Table 19 Hausman Specification Test ... 67
Table 20 Time Fixed Effects ... 68
Table 21 Dummy Variable Adjustment Method, Panel Analyses ... 69
Table 22 Dummy Variable Adjustment Method, Cross-sectional Analyses. ... 70
Table 23 Tests of Endogeneity, cross-section ... 71
4 1. Introduction
Traditional economics supposes that managers should make decisions that maximize the wealth of their firm’s equity holders (Friedman, 1962). But already in 1975, Davis argued that businesses have tremendous power in social and environmental fields and that any of such power should be accompanied with social responsibility. He argued that businessmen cannot and should not make decisions depending only on the direct consequences on their businesses, as their decisions have other social implications as well. In other words, (strategic) decision making should also take the social externalities into account. This is where Corporate Social Responsibility (hereafter CSR) arises. Davis (1975) defined CSR as:
“The obligation of a business decision maker not to take actions that only improve his own – and the organization’s interests, but also the welfare of society as a whole.”
Later business and society scholars emphasized the duty of firms to the society that goes well beyond maximizing the wealth of equity holders (Swanson, 1999; Whetten et al., 2002). An overly narrow focus can lead to the management ignoring other important stakeholders, including employees, suppliers, customers, and society at large. Sometimes, the interests of these other vital stakeholders should supersede the interests of a firm's equity holders in managerial decision making, even if this reduces the present value of the firm's cash flows (Clarkson, 1995; Donaldson & Preston, 1995; Mitchell et al., 1997; Wood & Jones, 1995).
During even more recent years, it is argued that Davis’ (1975) ethical argument is not the only side of the story. A recent strand of literature suggests that maintaining a high level of CSR can be in line with a purely profit-maximizing strategy as well. Besides the possible social obligation to do so, another reason for investing in CSR might be to attract (new) consumers through image improvement. If and only if this beneficial attraction effect exceeds the costs of investing, firms already have an incentive to be involved in CSR irrespective of the ethical argument. This ―strategic CSR‖, as termed by Porter and Kramer (2006), creates a symbiotic relationship where the success of the community and the success of the firm become mutually reinforcing. In other words, a relationship between CSR and Corporate Financial Performance (hereafter CFP) might exist in which CSR leads to costs in the short run and to added value in the longer run. Porter and Kramer (2006) argue that these costs and benefits do not cancel each other out, as the relationship between corporate success and social welfare is not necessarily a zero-sum game.
Whether it is the ethical or the strategic argument, CSR-activities become increasingly part of the daily firm practices. A recent UN Global Compact-Accenture CEO study reports
5
that a vast majority of participating CEOs worldwide acknowledge CSR to be important (Hayward & Rennie, 2018).
Since the 1950s, the effect of CSR on CFP has been studied extensively. However, the results have been mixed. A vast majority of these studies originate from the 20th century when a comprehensive measure of CSR was non-existent. As Aupperle et al. (1985) already argued, the most important limitation of previous research used to be the lack of an adequate measure of CSR in order to make any empirical statements. The lack of consistency in the literature might therefore be explained by a large variety of proxies for CSR. For example, Aupperle et al. (1985) themselves used a survey to determine how socially-responsibly oriented firms were. They found no significant relationship between various measures for CFP and the outcomes of this survey. Another example of a study using its creativity due to the lack of an actual CSR measure is the study of McGuire et al., (1988). They used the ratings of corporate reputations obtained from Fortune’s magazine as a proxy for CSR. The authors did not find a relationship between CSR and subsequent CFP using various performance measures as well. However, they did find a reverse effect of previous CFP on CSR.
With the manifestation of the Kinder, Lydenberg, Domini Research & Analytics (KLD) STATS database on Corporate Social Performance in 1991, a proxy for CSR was introduced. This database has been widely accepted and extensively used since its introduction. Currently, the KLD database is the undisputed ranking leader in the area of social performance measures. As cited from Waddock (2003):
“KLD data are currently the best available and it is the de facto research standard at the moment.”
Even after the manifestation of the KLD Stats database, the results have been mixed. McWilliams and Siegel (2000) argue that model misspecifications are the main reason for this lack of consensus. According to the authors, the omission of an important proxy for innovation results in upwards biased results for CSR. McGuire et al. (1988) and Waddock and Graves (1997) agree that model misspecification is the main determinant of spurious results. However, they argue that the relationship between CSR and CFP is subject to an endogeneity problem, namely simultaneous causality. Many studies do not address this problem, leading to biased results. The final reason for the lack of consensus within the field is the dependent variable choice. Waddock and Graves (1997) and Turban & Greening (1997), for example, used the accounting-based ROA, whereas Hull and Rothenberg (2008)
6
and Servaes and Tamayo (2013) used the marked-based Tobin’s Q. Besides the fact that these measures have a different fundament, they are also argued to consider different time frames: ROA reflects short-term performance, where Tobin’s Q reflects long-term performance. These different measures are not necessarily affected equally by CSR (McWilliams & Siegel, 2000). Also, it is argued that CSR enables firms to develop intangible assets, something that is not reflected in accounting-based measures (Gardberg & Fombrun, 2006; Hull & Rothernberg, 2008; Waddock & Graves, 1997). According to a framework of intangible resources developed by Hall (1992), intangible resources can be major contributors to business success. Therefore, it is important to include market-based measures as well.
The current research goes deeper into the aforementioned relationship between CSR and CFP, differentiating between short- and long-term effects using various performance measures, accounting- and market-based. Similar to the most closely related study of Servaes and Tamayo (2013), the role of public awareness and firm reputation will be emphasized. It is argued that firms with higher levels of Awareness and better Reputations experience a stronger positive effect of CSR on CFP. In the study of Servaes and Tamayo (2013), CSR was found to have a negative effect on firms with low levels of awareness. Besides the explicit distinction between short- and long-term measures, another novelty compared to this study is the addition of the debt-asset ratio. In order to mitigate the problem of endogeneity as mentioned by Waddock and Graves (1997) and McGuire et al. (1988), the model will be estimated using Two-Stage least squares (2SLS) regression analyses in both a cross-sectional and a panel setting. Using data for more than 2700 firms for the period 1994-2006, a final and comprehensive answer is tried to construct to the following question:
Does Corporate Social Responsibility affect Corporate Firm Performance in the short- and long run and is this relationship moderated by public awareness and firm reputation?
To answer this question, this paper is structured as follows. First of all, Section 2 provides an overview of the existing literature, including theoretical channels and empirical findings. The section will be completed with the hypotheses. Subsequently, in the third section, an extensive description of the methodology and regression analyses will be presented. The fourth section provides the results of the regression analyses, which will be discussed in more detail in Section 5. The limitations of the current research and the conclusions will be presented in sections 7 and 8 respectively. The paper will be completed with the references and the Appendix in sections 8 and 9 respectively.
7 2. Literature review
This section provides an overview of the existing literature on CSR and its relationship with CFP. The section is divided into three subsections; Theoretical Channels of CSR, Empirical Findings and Hypotheses. The first subsection provides a brief summary of the most important theorems, models, and other theoretical channels through which CSR might affect CFP. The second subsection presents the essence of the empirical findings of previous research using different measures for CSR and CFP. Section 2 will be completed with the predictions and hypotheses.
2.1 Theoretical Channels of CSR
Brammer and Millington (2008) argue that good social performance comes at the expense of good financial performance, because the resources that are devoted to social performance cannot be alternatively invested in projects that generate direct shareholder value. On the other hand, CSR can also be seen as an investment variable that has a direct negative effect on the firm’s cost structure, but an indirect positive effect through various other channels. Without doubt, CSR increases costs, but it is too short-sighted to recognize this as an unequivocal argument to forgo any CSR investments. Specifically, a typical feature of investment variables is that it results in direct costs and delayed benefits (Brammer & Millington, 2008). The concept of CSR is very broad and consists of many dimensions, which might all affect CFP separately and differently. The main channels through which CSR affects CFP are consumers, moderated by awareness; (potential) employees, and access to finance. Some of these dimensions are argued to have a positive effect, whereas others are said to have a negative effect on CFP, either directly or indirectly. These different theoretical channels will be discussed in the upcoming section.
The first channel through which CSR affects CFP is through the consumers. Hull and Rothenberg (2008) and Waddock and Graves (1997) argue that investment in CSR can lead to product differentiation. This differentiation effect is argued to result in some bargaining power for the seller and subsequently in a competitive advantage (Porter, 1989). Some characteristics of the production process or the product itself might signal to the consumer that the firm is not only self-interested, but cares about certain social issues as well (McWilliams & Siegel, 2000). By creating a corporate image of being socially responsible, firms might encourage consumers to buy their products or use their services, as consumers might perceive their engagement with this specific firm as an indirect contribution to the greater good (McWilliams & Siegel, 2000). Hillman and Keim (2001) argue that this results
8
in an attraction effect of socially conscious consumers, as there is strong evidence that large numbers of consumers attach positive values to these CSR attributes (McWilliams & Siegel, 2000). Moreover, CSR might lead to the reputation of being honest, reliable and delivering high-quality products; aspects that are often perceived as valuable for the consumer, but challenging to determine (McWilliams & Siegel, 2000). Brown & Dacin (1997) also found that consumers’ preferences for a (new) product are affected by their overall opinion about the firm itself, with the latter being positively influenced by CSR. In summary, CSR might be used tactically in order to increase CFP through an upward shift of the demand curve or a reduction in the elasticity of the firm’s demand curve (Navarro, 1988).
The 2015 Cone Communications/Ebiquity Global CSR Study has shown that a vast majority of consumers demand more from companies than just being profit maximizers. This large-scale consumer survey also reveals that consumers search for responsible products whenever possible and that they are willing to avoid irresponsible or deceiving companies. However, in order to benefit from this consumer attraction effect, firms have to make sure that the consumers are aware of their investments in CSR (McWilliams & Siegel, 2001). This implies the relationship between CSR and CFP is empowered by the presence of advertising. Others argue that CSR itself may also function as an advertisement device (Navarro, 1988; Sen & Bhattacharya, 2001; Milgrom & Roberts, 1986). This implies that firms can use their CSR in order to raise awareness for their products and services.
Fisman et al. (2006) constructed a model around corporate philanthropy to investigate this signaling role of CSR. In their model, it is the quality of the provided products or services that the consumer has to be made aware of through observable CSR expenditures rather than CSR investment itself. Several assumptions are made in the construction of this model. First, unrelated CSR expenditures are assumed to be easily observable, but product quality is not. This assumption follows from the distinction between search- and experience goods. In contrast to the quality of search goods, experience goods have to be consumed before knowing the true quality (Siegel & Vitaliano, 2007). Secondly, it is argued that CSR might serve as a means to signal its trustworthiness in providing goods and services of high(er) quality to the consumer. Third, it is assumed that there are two types of firms; firms that are purely profit-motivated and firms that do also care about the externalities that result from their actions. According to the authors, it is less costly for the latter to be engaged in CSR than for the former. Since only firms that establish credibility through good CSR performance are trusted to deliver high-quality products or services, CSR can have a signalling effect on consumers (Fisman et al., 2006). Moreover, the authors mention that this
9
vertical differentiation effect is larger in more crowded industries where the provided products and services are relatively uniform. In addition, the effect is also argued to be larger in industries where the consumers attach relatively higher valuations to firms’ images (that is: in industries with high advertising intensity). Although the authors argue that the signalling role of CSR creates value for the firm, it can still occur at the expense of short-run profitability. Simply because it takes some time before the benefits are noticeable, whereas the costs are incurred immediately.
Another channel through which CSR affects CFP is through the (potential) employees. According to Turban and Greening (1997) and Greening and Turban (2000), the signalling- and the social identity theories play an important role in this conjunction. The former suggests that a firm’s social performance signals to potential future employees what it is like to work at the particular firm. Good social performance is associated with an attractive work environment, whereas poor social performance has a deterrent effect. Being perceived as an attractive place to work may develop competitive advantages, especially since the quality of the workforce has become increasingly important over the years (Greening & Turban, 2000). The latter suggests that working at a socially-responsible firm results in a higher self-image compared with working at a less-responsible firm. The authors also indicate that applicants are more likely to apply to firms with a higher level of CSP, and are more likely to accept a possible job offer from those same firms. In other words, CSP might result in a competitive advantage in the search for suitable applicants (Turban & Greening, 1997).
A fourth channel through which CSR affects CFP; CSR enhances the long-term value of the firm by lowering the idiosyncratic constraints that a firm faces in financing projects and other operations (Cheng et al., 2014) and provides better access to valuable resources (Waddock & Graves, 2007). These capital constraints as discussed by Cheng et al. (2014) consist of any market frictions that limit the firm from funding all desired investments, that is; all investments with a positive ―Net Present Value‖ (NPV). Previous studies have already shown that a firm’s capital structure choices (Hennessy & Whited, 2007) and strategic decision-making (Stein, 2003) are heavily influenced by these capital constraints, thereby directly affecting the firm’s ability to undertake certain investments. This is also supported by Lamont et al. (2001) who provide evidence that financially constrained firms perform worse in terms of stock returns compared to unconstrained firms. Furthermore, Levine (2005) argues that smaller, newer and riskier firms have to cope with capital constraints relatively more often.
10
According to Cheng et al. (2014), multiple reasons exist why firms with higher levels of CSR face lower capital constraints. First of all, remarkable CSR performance is associated with better stakeholder commitment, thereby reducing the likelihood of short-term opportunistic behaviour (Benabou & Tirole, 2010; Eccles et al., 2012). Subsequently, this results in a reduction in overall contracting costs according to Jones (1995) due to the engagement between the firm and stakeholders based on cooperation and mutual trust. Additionally, firms that are more engaged in CSR are more likely to disclose their CSR activities to the market (Dhaliwal et al., 2011). This, in turn, generates a feedback loop as mentioned by Cheng et al. (2014). First, it increases transparency about the company’s environmental and social impact. Subsequently, Cheng et al. (2014) argue that it may change the internal control system, thereby further improving the compliance with regulations and the reliability of reporting. The increased transparency combined with improved reliability of reporting, enhances the firm’s data availability in both quantitative and qualitative terms, thereby reducing the informational asymmetry. This subsequently lowers the capital constraints, that is; it makes it easier for the firm to obtain finance (Hubbard, 1997). Altogether, a firm with better CSR performance faces lower capital constraints due to lower agency costs through stakeholder commitment on the one hand and increased transparency through CSR reporting on the other hand (Cheng et al., 2014). Ultimately, these lower capital constraints allow the firm to be engaged in more value-adding investments. Using a panel dataset of 2439 firms originating from 49 countries over a period of 8 years from 2002 to 2009, Cheng et al. (2014) found support for this connection.
Similar to the consumer attraction effect, CSR might also attract socially conscious investors and thereby making it easier to obtain resources (Kapstein, 2001). In summary, CSR thus leads to reputational gains on the side of the consumers (McWilliams & Siegel, 2000), the employees (Turban & Greening, 1997), and the investors (Kapstein, 2001). Additionally, Gardberg and Fombrun (2006) argue that a positive firm image can strengthen the ties with regulators.
Lastly, financially constrained firms are more likely to diminish investments in a wide range of strategic activities (Campello et al., 2010), including investments in inventory (Carpenter et al., 1998) as well as investments in R&D activities (Himmelberg & Petersen, 1994; Hall & Lerner, 2010), and in labour hoarding during recessions (Sharpe, 1994), which significantly and adversely diminishes the capacity of the firm to grow over time. Overall, firms that invest more in CSR and as a result are less financially constrained have therefore more value-creating opportunities in the long run.
11
Moreover, Fombrun et al. (2000) argue that the investment in corporate citizenships may create the potential for additional earnings by increasing the real options that are at the firm’s disposal. In other words, it is important to invest in corporate citizenship today in order to benefit from opportunities that might emerge tomorrow. Corporate citizenship, and CSR in general, can be comparable to R&D or employee training; they are platform investments that do not create value directly but might unlock future growth opportunities (Fombrun et al., 2000).
2.2 Empirical Findings
This section provides an overview of the empirical findings of previous research. CSR is not a new phenomenon and it has been examined thoroughly in the past. However, it was not until the beginning of the 1990s that a common measure of CSR was created. A considerable part of the variation in the empirical findings can be explained by the wide range of proxies for CSR. Although the current measure is also not impeccable, its usefulness is widely agreed within the field as a vast majority of the more recent studies used similar data from the KLD Stats database.
An example of a study that also used data from the KLD Stats database in order to proxy for CSR is the study of Servaes and Tamayo (2013). Within the field, this study is relatively recent and the most closely aligned with the current research. The authors have studied the effect of CSR on Tobin’s Q using a large dataset on U.S. firms over a period of 15 years from 1991 to 2005. They controlled for Size, Advertising Intensity, and R&D Intensity. Interaction terms between CSR on the one hand and Advertising Intensity and Firms’ Reputation, or both, on the other hand, were included as well. Based on this model specification, they made three conclusive statements. First of all, firms with high public awareness can increase their firm value by engaging in CSR. Yet, high public awareness can also have an opposite effect if there happens to be concerns with their CSR activities. Secondly, CSR has a negative or insignificant effect on firms with low public awareness. Thirdly, if an inconsistency between the firm’s CSR efforts and its overall reputation is present, advertising has a negative effect on firm value. Moreover, the authors argue that unobservable firm characteristics exist that drive both CSR and CFP. Therefore, model specifications without firm fixed effects may therefore lead to spurious results.
Similarly, Hull and Rothenberg (2008) found a positive effect of CSR on CFP as well. CSR was proxied by data on CSP obtained from the KLD Stats database. However, this study used Return on Assets (ROA) as a measure of CFP rather than Tobin’s Q. The authors
12
included a proxy for Innovation, and controlled for Size, Risk, and Industry. Weighted averages of all independent- and control variables were calculated over a three year period from 1998 to 2000. Subsequently, the impact of these weighted averages on future performance (that is; 2001) was assessed. Using this data set, they concluded that the positive link between CSR and CFP is stronger among firms in relatively undifferentiated industries and among firms that have a low(er) level of innovation. The latter effect being explained by the fact that innovation and CSR can act as (strategic) substitutes. Moreover, the authors mentioned that a certain level of CSR can be achieved without a decrease in financial performance. In other words, investing in CSR is not necessarily a zero-sum game, and if executed properly, both the society and the firm can benefit.
Using a multi-country panel data set, Surroca et al. (2010) found no direct effect of CSR on Tobin’s Q. Only an indirect effect through the development of intangible assets was found. In order to proxy for CSR, data were obtained from Sustainalytics, a database similar to KLD Stats. The authors included measures for Innovation, Human Capital, Reputation, Culture, Physical Resources, Leverage, and Financial Resources and included controls for Size, Risk, Year, and Industry. Their main objective was to examine the indirect link between CSR and CFP via intangible resources, such as innovation, reputation, organizational culture, and human resources, operating as mediators. CSR was found to stimulate the development of intangibles, which in turn enhanced CFP (Surroca et al., 2010).
Building on previous research, Orlitzky et al. (2003) conducted a meta-analysis of 52 studies to examine the link between CSR and CFP. Altogether, this led to a total of 33,878 observations. They concluded that a positive relationship between CSR and CFP exists across industries and across study contexts.
However, McWilliams and Siegel (2000) found no significant effect of CSR on various short- and long-term performance measures. In their view, a majority of statistically significant effects found in the past are due to model misspecification, that is; the omission of R&D resulted in upwards biased estimates. Their conclusion was drawn from a panel of data covering 524 firms over a period of six years from 1991 to 1996. KLD Stats data were used to proxy for CSR and the authors controlled for the R&D Intensity, size, debt/asset ratio and advertising intensity. The underlying reason is that many firms that actively engage in CSR are likely to pursue a differentiation strategy involving innovating investments in R&D as well. Many facets of CSR result in innovations; either in terms of new products, process modifications, or both (McWilliams & Siegel, 2000). This problem mainly arises in the part of the literature that focuses on long-term CFP. As there is a wide range of literature
13
connecting R&D investment to better CFP in the long run. Knowledge accumulation and thereby innovation often leads to productivity enhancement in the long run (McWilliams & Siegel, 2000). Various other studies emphasize the importance of including crucial variables such as R&D in order to avoid spurious results (Margolis & Walsh, 2003; Orlitzky et al., 2003, Hull & Rothenberg, 2008).
The majority of previous research has either found a positive or a neutral link between CSR and CFP, either directly or indirectly. However, Luo and Bhattacharya (2006) concluded from a sample of publicly traded Fortune 500 companies that CSR can, under certain conditions, also have a negative effect on CFP. The authors emphasized the role of innovation as a moderator between CSR and CFP in their analysis, while controlling for Innovativeness capability, Consumer satisfaction and Product quality. Stock Returns and Tobin’s Q were used as measures for CFP. They found that CSR reduces customer satisfaction, and subsequently Market Capitalization, of firms with low innovative capability.
Summarized, the relationship between CSR and CFP has been studied extensively with divergent empirical findings. However, once controlled for a proxy for innovation and including firm fixed effects, fewer studies find a significant direct relationship between CSR and CFP. Nonetheless, several studies have emphasized the presence of indirect effects through intangible resources such as innovation and reputation.
2.3 Hypotheses
This section contains the predicted effects of CSR on the two performance measures. All predictions are based on the theoretical channels and empirical findings of previous research as considered in the previous two subsections.
Based on our research question, the following hypotheses will be tested:
Hypothesis 1: CSR has a negative short-term effect on CFP.
Following the reasoning in Brammer and Millington (2008), it is hypothesized that CSR has a negative effect on CFP in the short run. The main reason is the simple fact that it takes some time before the returns on investment become visible, whereas the costs of investment are directly noticeable. This phenomenon is also known as the payback period.
14
Considering the theoretical channels and empirical findings of Surroca et al. (2010) and Servaes and Tamayo (2013), it is hypothesized that CSR has a positive effect on CFP in the long run.
Hypothesis 3: The effect of CSR on CFP is positively affected by public awareness. In light of the Fisman’s (2006) signalling theory and the consumer attraction effect, it is hypothesized that awareness serves as a moderator and increases the positive effect of CSR on CFP, similar to the findings of Servaes and Tamayo (2013).
Hypothesis 4: The effect of CSR on CFP is positively affected by a firm’s R&D intensity. Following the empirical findings of Luo and Bhattacharya (2006), it is expected that the effect of CSR on CFP is stronger for firms with higher levels of innovation, measured by the R&D intensity.
Hypothesis 5: The effect of CSR on CFP is positively affected by a firm’s reputation. Considering Servaes and Tamayo’s (2013) argument of the required consistency between the firm’s reputation and the CSR efforts, it is hypothesized that higher reputational scores increase the returns to CSR.
15 3. Methodology
This section contains the methodology and the regression models used in the current research to empirically analyse the effects of CSR on the two performance measures; ROA and Tobin’s Q. The methodology section is divided into three subsections; variable selection, regression set-up and the diagnostic tests. Section 3.1 gives an overview of the selected variables, whereas section 3.2 provides an overview of the panel model and the cross-sectional model. The methodology section will be completed with the diagnostic tests, provided in section 3.3, in order to determine the exact specification of the model.
3.1 Variable Description
This section provides a detailed description of all the variables used in this study including their sources. The set of variables is similar to the one used in Servaes and Tamayo (2013), expanded with a control for the firm’s Leverage Ratio.
Corporate Social Responsibility
During the 80s and 90s of the previous century, CSR has been studied extensively. However, a well-defined measure of this concept was non-existent at the time. Therefore, in 1991, a comprehensive index was introduced; a measure on Corporate Social Performance (CSP) provided by Kinder, Lydenberg, Domini Analytics, Inc. (KLD) Social Ratings database. KLD have specialized in evaluating corporate social performance using firms’ strengths and weaknesses along various dimensions: Alcohol, Community, Corporate Governance, Diversity, Employee Relations, Environmental Performance, Firearms, Gaming, Human Rights, Military, Nuclear, Product, and Tobacco. In Appendix 9.1, an overview of these different (sub)dimensions determining the overall CSR-score can be found. The database, made available by Wharton Research Data Services (WRDS) on Compustat, contains data of a wide range of publicly traded U.S and non-U.S. companies. The dataset was initiated in 1991 containing data on the firms from the S&P 500 Index. Over the years, firms from the Russell 1000- and Russell 2000 indices have been included in the dataset. In recent years, the dataset is expanded with non-U.S firms. The companies have been examined by experienced research analysts ever since, and over the years more companies have been added to the database (MCSI, 2015).
The measure, constructed by Kinder, Lydenberg and Domini Research & Analytics, was initially created to provide information for investors who wanted to take social factors
16
into considerations for their investment decisions (McWilliams & Siegel, 2000), but has developed to be one of the most used data sets in CSR-related studies. The quality of the data is not solely depending on the quality of company reports, as the data are derived from a wide variety of sources, with company sources being only a part of it (Waddock, 2003; Kinder, 2006). Where possible, quantitative data are used (for example in charitable giving), but in certain cases, judgment is necessary (for example in freedom of expression and censorship). However, KLD staff members meet on a weekly basis to assure that ratings are created consistently from year-to-year and from company-to-company (Waddock & Graves, 1997).
Following Servaes and Tamayo (2013), the following dimensions are considered in the construction of the CSR index: Community Relations, Diversity, Employee relations, Environmental Performance, and Human Rights. All dimensions are assigned an equal weight in determining the overall CSR index. The dataset provides an annual assessment of the several performance indices. The KLD Stats database contains data on a substantial amount of US and non-US firms from 1991 until 20131. The database contains data on approximately 650 firms covering the complete timespan. For a subset of this timespan, the database contains data on several thousands of firms.
In order to derive a proxy for CSR from the KLD data, a similar procedure as in Servaes and Tamayo (2013) is executed. Alcohol, Gambling, Military, Nuclear Power, and Tobacco are excluded from the analysis, as these are controversial exclusionary dimensions which only include concerns. In the current research, only the inclusive dimensions; Community, Diversity, Employee Relations, Environment, and Human Rights that include both strengths and concerns are considered in the construction of this measure. The two remaining inclusive dimension that are reported by KLD; Product and Corporate Governance are not included in the measure, as these are not argued to be components of CSR (Jo & Harjoto, 2011; Servaes & Tamayo, 2013). For each of the five residuary dimensions, KLD Stats contains data on the number of strengths and concerns for each company. A high overall number of strengths accompanied by a low overall number of concerns indicates a high level of CSR. The number of strengths and concerns has been modified over time.2 In order to make comparisons over time, the number of strengths for each firm-year within each dimension is divided by the maximum number of strengths in each category year. The concerns are scaled in like manner. Subsequently, the concerns index
1 June 2018
2 To some extent, changes in the CSR score may be the result of the addition of new strengths or
concerns, rather than an actual change of the firms’ behavior. However, all dimensions also include ―other strengths‖ or ―other concerns‖ to moderately control for yet unendorsed factors.
17
is subtracted from the strengths index to obtain the overall index within that dimension. Such an index, ranging from -1 to +1, is created for each dimension separately. Thereafter, the five individual indices are combined into one overall CSR index ranging from -5 to +5 (Servaes & Tamayo, 2013). ∑ [ ]
Where subscript i refers to the five dimensions as introduced above.
To deal with the earlier mentioned problem of endogeneity, instrumental variables will be used. In line with previous literature (Cheng et al., 2014; Kim et al., 2014; Jiraporn et al., 2014), CSR will be instrumented by the industry’s average CSR index (excluding the data on the focal firm). It is argued that a firm’s level of CSR is systematically influenced by the level of CSR of other firms within the same industry. For an instrument to be valid it is required to be correlated with the substituted explanatory variable and uncorrelated with the residual. In other words, the instrument should be both relevant and exogenous.
18 Figure 1
CSR, Historical Development
Note: the vertical lines denote the years of expansion. After the years 2000 and 2002 the
coverage universe has expanded from 650 firms to approximately 1100 and 2400 firms respectively.
Datasource: KLD Stats Database
Figure 1 shows the development of the average level of CSR over time. From the beginning of the sample period until 1997 the average level of CSR grew steadily before it remained constant for a few years. From the year 2001, the average level of CSR decreased to zero again in 2003. Afterward, it rose steadily until the end of the sample period. Two reasons might exist for the sharp decline in average CSR after the year 2000. First of all, the September 11 attacks in 2001 had a significant impact on the United States’ economy and the world economy in general. It is likely that this affected the levels of CSR as well. Secondly, after the years 2000 and 2002, the coverage universe of the KLD Stats database has been expanded with approximately 2000 firms in total. Therefore, the dashed line gives a ―cleaner‖ overview, as it only included the firms that were included before 2000 as well. A substantial decline is still noticeable after the period of the September 11 attacks. However, the decline in average CSR is considerably smaller keeping the number of firms more or less constant.
-0.2 -0.15 -0.1 -0.05 0 0.05 0.1 0.15 Full S&P
19
CFP
Companies formulate all kinds of goals: social-, environmental- but mostly economic. The economic goals that firms often strive for vary from maximum market share to consumer satisfaction and from simply surviving to being the leader in advanced technology. Yet in the end, the ultimate goal of private firms is almost always to maximize medium- to long-term profits. In order to compare the actual firm performance to the predetermined goals, various performance measures are available. Each of these measures is created to inform the stakeholders in their decision-making. However, as Behn (2003) argues, different stakeholders have different purposes. Managers wish to measure the firm’s performance in order to monitor, set new goals, or to allocate resources, governments might use the measures to evaluate the effectiveness of past policies or to introduce new policies, whereas investors are mainly interested in the development of the shareholder value (Behn, 2003). In order to increase the usefulness to the different stakeholders, multiple measures of CFP are reported or created on a regular basis. Commonly used accounting-based measures of CFP are Net Income, ROA, ROE, and ROS, whereas Tobin’s Q and Market Capitalization are commonly used market-based measures of CFP.
In the current research two main variables will be used to measure short- and long-term CFP. These are the short-long-term accounting-based ROA and the long-long-term market-based Tobin’s Q:
The Return on Assets is a measure indicating the profitability of a firm relative to its total assets. For each firm-year data is collected on Net Income and Total Assets from Compustat in order to calculate the ROA. Net Income is defined as the difference between revenues and gains on the one hand and expenses and losses on the other3. A general advantage of using Compustat is that the data are standardized such that both cross-sectional and longitudinal comparisons can be made. In the absence of any standardization, the data are likely to show excessive variability as a result of divergent reporting standards.
( )
20
Tobin’s Q can be seen as the Market Capitalization of installed capital over the replacement costs of capital. The former being defined as the sum of Assets and the multiplication of Common Shares outstanding and the month-end price, subtracted by Common Equity4. The ratio is calculated manually using data from Compustat. Simply put, Tobin’s Q denotes the extent to which a stock is under- or overvalued. A Tobin’s Q above unity implies that the stock market assigns a value to the firm that exceeds the book value. More generally, investors are confident that the firm is able to generate revenues and earnings growth in the (near) future. This confidence is lacking when the Tobin’s Q is less than unity; the stock market assigns a value to the firm that is lower than the book value.
Figure 2
ROA, Historical Developments
Datasource: Compustat
4 Common Equity is defined as total Shareholders’ Equity minus the Treasury Stock.
-0.02 -0.01 0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006
21 Figure 3
Tobin’s Q, Historical Developments
Datasource: Compustat
Figure 2 and 3 show the historical developments of ROA and Tobin’s Q respectively. Both graphs show a similar pattern to Figure 1. Apart from a few up- and downward deviations, ROA remained relatively stable from 1994 to 2000 as can be seen in Figure 2. However, at the time of the September 11 attacks in 2001, the average levels of ROA plummeted. From 2003 and onwards, ROA increased again but did not yet reach it 2000 peak. Tobin’s Q, on the contrary, rose steadily from 1994 until 2000 as can be seen in Figure 3. After the September 11 attacks, Figure 3 shows a dip, returning the average levels of Tobin’s Q back to its 1994 value. Subsequently, the economy improved and Tobin’s Q rose slightly until the end of the sample period. Interestingly, both CFP-measures follow more or less the same pattern. However, the downward movement is more extreme for the average levels of ROA. This can be a result of the earlier mentioned difference in the time period that both measures represent. The effects in the short run are somewhat levelled out in the long run as indicated by Figure 2 and 3 respectively.
Moreover, three additional short-term accounting-based measures, ROE and ROS, and Net Income, and one long-term market-based measure, Market Capitalization, are included as robustness checks to check whether the conclusions drawn from the empirical findings depend on the choice of variables:
1.5 1.7 1.9 2.1 2.3 2.5 2.7 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006
22
Similar to the ROA, the Return on Equity is a relative profitability measure. The ROE relates Net Income to the Shareholders’ Equity rather than to its total assets. The relevant data are obtained from Compustat for each firm-year.
Return on Sales, on the other hand, relates the Earnings before Interest and Taxes (EBIT) to the firm’s Total Sales. Also for the ROS, data are obtained from Compustat for each firm-year.
( ) ( )
Net Income is defined as the difference between revenues and gains on the one hand and expenses and losses on the other hand. In the regression analyses, the growth of Net Income will be used. Annual data on Net Income for each firm are obtained from Compustat.
Market Capitalization is defined as the number of shares outstanding multiplied by the price per share. Data on both components are obtained from Compustat for each firm. This performance measure contains the consumer valuations for all future periods discounted to the current date. The log of Market Capitalization is used in order to stabilize the variance and transform the data to normality.
Size
Consistent with the literature a control for Size is included in the model as well (Servaes & Tamayo, 2013; McWilliams & Siegel, 2000). Larger firms tend to be older and have lower investment opportunities than younger firms (Servaes & Tamayo, 2013). Servaes and Tamayo (2013) used both the logarithm of employees and the logarithm of assets as a proxy
23
for size. Both proxies yielded similar results. On the other hand, Orlitzky (2001) argues that Size is positively related to CFP mainly due to net economies of scale and greater control of resources. Since assets are also in the denominator of both ROA and Tobin’s Q, it has an inevitable negative effect on the dependent variables. Therefore, the logarithm of employees will be used as the starting point within the model. Data for each firm-year are likewise obtained from Compustat.
Following the reasoning and empirical findings of Servaes and Tamayo (2013), it is hypothesized that Size has a negative effect on long-term CFP as measured by Tobin’s Q. On the other hand, the arguments raised in Orlitzky (2001) seem to be more related to costs and revenues included in the ROA. Therefore, it is hypothesized that Size has a positive effect on short-term CFP as measured by the ROA.
Public Awareness
Public awareness will be proxied by the advertising intensity of firms: advertising expenses over sales. There are two viewpoints regarding the role of advertising. Firstly, Schuler and Cording (2006) argue that firms (or other parties) disclose their CSR activities through advertising. This view implies that awareness only affects CSR strengths as firms are not likely to disclose CSR concerns voluntarily. Servaes and Tamayo (2013) on the other hand, argue that a higher advertising intensity, and thereby more public awareness, increases the likelihood that the public finds out about all the elements of CSR, both strengths and concerns. Companies that are accompanied by a higher level of public awareness will encounter more problems withholding negative information. Servaes and Tamayo (2013) have already proven that firms with a higher advertising intensity also receive more press coverage, suggesting that advertising increases awareness about the company. Therefore, the second viewpoint seems more plausible. In other words, it is hypothesized that advertising strengthens the relationship between CSR and CFP as consumers are more aware of the CSR practices of firms with higher advertising intensities (Servaes & Tamayo, 2013).
For a majority of the firms, data on advertising expenses are not available at all times. Consistent with the literature, advertising expenses were set equal to zero if missing (Servaes & Tamayo, 2013; Fee et al., 2008; Hale & Santos, 2009; Masulis et al., 2009). In general, missing values cannot simply be replaced with any non-missing value, as this might lead to biased results (Allison, 2003). However, given the circumstances, this method is perceived as a relatively safe solution, since firms are only obliged to report expenses on advertising if it exceeds a certain threshold (Hale & Santos, 2009). In other words, the missing values are
24
very small by definition. Nonetheless, a dummy variable is included in the model that is equal to one whenever a missing value is replaced by zero. This missing data indicator or the dummy variable adjustment method is explained in more detail in Appendix 9.6.
R&D Intensity
Similar to investing in CSR, investing in R&D and thereby in innovation can be seen as a way of differentiating for the firm, resulting in better CFP-scores (McWilliams & Siegel, 2000). R&D Intensity is measured as the R&D expenses over sales, similar to Servaes and Tamayo (2013) and McWilliams and Siegel (2000). Missing data on the R&D expenses were treated in a similar manner as the advertising expenses; it was set equal to zero and a dummy variable was included to indicate the replaced values (Servaes & Tamayo, 2013)5. Also, as Hull and Rothenberg demonstrated, the effect of CSR on CFP can be moderated by innovation. Therefore, it is hypothesized that R&D has a direct positive effect on firm performance and serves as a moderator between CSR and CFP as well.
Leverage
Similar to McWilliams and Siegel (2000), Waddock and Graves (1997), and Inoue and Lee (2011) a measure will be included to control for risk as denoted by leverage. Leverage is calculated by the debt/asset ratio. Waddock and Graves (1997) argue that firms with higher leverage ratios may behave differently than firms with lower leverage in terms of CSR investment, because of different levels of risk involved in CSR investment. In line with the findings from a meta-analysis by Capon et al. (1990), it is hypothesized that high levels of leverage affect CFP negatively.
3.2 Regression Set-up
The goal of the current research is to analyse the effects of CSR on CFP in both the short- and the long run. Various authors emphasize the importance of the inclusion of a proxy for firm reputation in the regression specification. Frequently, firm reputation is proxied by Fortune’s yearly rankings on ―America’s Most Admired Companies” (Servaes & Tamayo, 2013; Surroca et al., 2010; Schuler & Cording, 2006; Du et al., 2010). However, these rankings are only freely available in 2006 and from 2014 and onwards. Therefore, in order to avoid a severe loss of observations or the omission of important effectors, it is chosen to
5 Similar to the reporting requirements for Advertisement Expenses, firms are only obliged to report their R&D Expenses if it exceeds a certain threshold.
25
construct a main panel model using a large sample and complement this with a cross-sectional model using a smaller sample. The former, which will be described more in detail in section 3.2.1, contains data for all firms during the complete time period, but does not include a proxy for Firm Reputation. The complemental cross-sectional model, which will be described in section 3.2.2, on the contrary, does include a proxy for Firm Reputation, but only contains data for a smaller set of firms for one year.
3.2.1 Panel data
By means of a panel of data consisting of more than 2700 U.S. firms from the S&P 500, Russell 1000 and Russell 3000 indices, the relationship between CSR and CFP in both the short- and the long between 1994 and 2006 is examined.6
These effects are estimated with the use of multiple 2SLS panel data regressions including instrumental variables, firm –and year fixed effects. According to Servaes and Tamayo (2013), firm fixed effects are highly necessary in these models but are often neglected. Firms from the S&P 500 are included in the panel for the complete time period. From the years 2001 and 2003 and onwards, the KLD STATS database also contains data on firms in the Russell 1000 and Russell 3000 indices respectively. ROA and Tobin’s Q are used as the dependent variables to proxy CFP in the short- and long run respectively. Ideally, data on CSR is used. However, data on CSR and thus a perfect measure is non-existent at the moment of writing. Therefore, data on CSP from the KLD Stats database will be used to construct an index approximating CSR.
There are multiple reasons for the choice of panel regressions. First of all, compared to conventional cross-sectional or time series regressions, panel data regressions result in a larger number of observations and reduced collinearity among independent variables (Hsiao, 2014). Additionally, panel data analysis has the advantage that it leads to more efficient estimates as it combines both the cross-sectional and the time-series dimension. Moreover, every company within the sample might have some characteristics that do not evolve over time but do vary across companies. Similarly, business cycles might have an effect on CFP that differs from year-to-year but not from company-to-company. A third advantage of panel regressions is that it is possible to control for these firm- and time fixed effects (Hsiao, 2014). However, the downside is that the slopes and intercepts provided by panel analyses are
6 Data are also available for the period 2007-2013. However, in that period a big exogenous shock
occurred, which cannot be controlled from, namely the global financial crisis of 2007-2008. In order to prevent this major event from affecting the results, the time period has been shortened.
26
pooled averages along two dimensions, rather than along the cross-sectional or longitudinal dimension alone (Hsiao, 2014).
Moreover, this study builds on Servaes and Tamayo (2013), using the KLD Stats database and a similar set of control variables. Different from Servaes and Tamayo (2013), the current research makes a distinction between the short-term and long-term effects by including both ROA and Tobin’s Q as dependent variables. Servaes and Tamayo (2013) only considered the effects of CSR on CFP using Tobin’s Q. Furthermore, this study will make use of instrumental variables in order to mitigate the problem of the endogeneity. As Waddock and Graves (1997) argue, CSR and CFP are synergistic, meaning that CSR might both be a consequence and a predictor of CFP. Finally, this study extends the time period with an additional year to 2006. The study will be extended by a cross-sectional analysis, which will be discussed in more detail in section 3.2.2.
In the study of Aupperle et al. (1985), multiple perspectives concerning the relationship between CSR and CFP are cited. The authors mention that CSR and CFP might be negatively related, due to the fact that investment in CSR leads to costs that could otherwise have been avoided or borne by others, such as (local) governments. Moskowitz (1972) acknowledges the presence of these costs. However, he argues that these are rather small and would therefore not lead to a severe reduction in CFP. The final perspective, retrieved from Cornell and Shapiro (1987), on the other hand, implies that the costs involved with CSR are significant, but are offset by a reduction in the firm’s other costs. In summary, three different perspectives with different, but not necessarily conflicting conclusions. It might be true that the relationship between CSR and CFP satisfies all three viewpoints; a negative relationship between CSR and short-term accounting-based ROA, and a neutral or positive relationship between CSR and long-term market-based Tobin’s Q. Therefore, the model will be estimated using two different measures for CFP.
McGuire et al. (1988) argue that there is an endogeneity problem involved with CSR. They found that CSR, measured by Fortune magazine’s ratings of corporate reputation, is more closely related to past CFP than to subsequent CFP. In other words, CFP predicts CSR rather than the other way around. The more recent study of Waddock and Graves (1997) endorses these findings. They describe this problem as a virtuous circle; CSP and CFP are found to be synergistic, meaning that either variable is both a predictor and a consequence. Their findings are based on an empirical model including data on CSP from the KLD STATS database and various financial performance measures, such as ROA, ROE, and ROS. Cheng et al. (2014) also consider CSR as a ―luxury good‖, implying that poor performing firms are less
27
likely to engage in CSR investments. For this reason, the relationship between CSP and CFP will be empirically examined using the instrumental variable described in the previous section.
Following Servaes and Tamayo (2013), four control variables are used: Size, Public Awareness, and R&D Intensity. They are all argued to be important predictors for CFP according to the literature and are therefore included in the model. A novelty in the current research, compared with Servaes and Tamayo (2013), is the inclusion of a measure of Leverage as an additional explanatory variable, as this is argued to be a non-negligible determinant (Waddock & Graves, 1997; McGuire et al., 1988; Surroca et al., 2010; Hull & Rothenberg, 2008). CFP is the dependent variable measured by either ROA or Tobin’s Q, whereas CSR and the interactions term including CSR are the variables of interest.
In order to draw any conclusions concerning the effects of CSR on CFP, the following 2SLS panel data model will be estimated. Multiple regressions with two different dependent variables will be conducted to test for any potential differences in the short- and long run:
In order to instrument for the endogenous variable, , the following two steps are conducted:
(1) is regressed on the instrument, and the remaining exogenous regressors in order to obtain ̂ .
̂
(2) The fitted values of ̂ derived from equation (1) are plugged into the original linear regression equation:
̂ ̂ ̂
28
Where represents one of the two performance indicators of firm i on time t, represents the overall index of CSR, represents the public awareness, represents the firm fixed effects, represents the time fixed effects, and is a composite error term that is uncorrelated with ̂ and the remaining regressors. Moreover, is a vector of control variables consisting of Size, Leverage, and R&D. The panel contains more than 2700 separate longitudinal data series covering an equal number of U.S. firms from different industries over the period 1994-2006.
3.2.2 Cross-sectional model
The panel model will be extended with a cross-sectional model containing data for 167 firms for the year 2006. In contrast to the panel model, the cross-sectional model also includes a measure of Firm Reputation. Firm Reputation is proxied by Fortune’s ranking on “America’s Most Admired Companies”. Fortune’s Magazine ranks firms along various dimensions on a yearly basis in order to determine their reputational score. Due to the fact that this data is only freely accessible for a limited number of years, the inclusion of a proxy for Firm Reputation restricts the time period to only one year.
The main reason for the inclusion of this proxy is the argument raised by Schuler and Cording (2006). They argue that there should be a congruency between the firm’s reputation and its actions in order for consumers to respond positively to advertising. When a firm has a good reputation, the effect of information disclosure through advertising is argued to be stronger than for firms with a poor prior reputation (Du et al., 2010). Trust and credibility are keywords in this relationship. Du et al. (2010) argue that stakeholders rely on the firm’s reputation to interpret ambiguous information, such as CSR activities. Moreover, Barnett (2007) argues that the response of consumers to CSR is path-dependent; a certain activity may lead to positive returns for one firm, but negative for another, depending on the firm’s reputation. This finding is supported by Servaes and Tamayo (2013), and therefore, it is hypothesized that better CFP increases value adding capacities of CSR.
29
are related to CFP, thereby leading to a so-called halo effect. In order to remove this effect, the residual of the model described below is employed as the clean reputation score7:
A procedure, similar to the panel model, will be followed, leading to the following final equation, using fitted values of CSR:
̂ ̂ ̂ ̂ ̂
The only difference, besides the time-frame, is the addition of Firm Reputation, denoted by , as a separate regressor, in the interaction with CSR, and in the three-way interaction with Awareness and CSR.
3.2.3 Reversed Relation
To examine whether CFP also affects CSR, the panel model is also estimated with reversed roles of CFP and CSR. The following panel model is estimated:
7
This procedure, similar to Servaes and Tamayo (2013), is a slightly adapted version of the procedure followed in Brown and Perry (1994):
30 4. Results
This section provides an overview of the primary empirical findings. The section is divided into three subsections containing the empirical results of the general panel model, the 2006 cross-sectional analysis including a proxy for firm reputation, and a model in which the roles of the independent and dependent variables have been switched. Robustness checks and specification tests are not presented in these sections but can be found in Appendix 9
Table 1 provides a summary of the data used in the current research. The table shows the total amount observations, the mean, the standard deviation, the minimum- and maximum observed value for each variable. The correlations between all the variables used in this study are shown in Table 2.
Table 1 Summary Statistics
Variable Obs. Mean St. Dev. Min Max
CSR 10410 -0.088 0.358 -1.702 2.071
ROA 10410 0.026 0.174 -5.511 2.170
Tobin 10388 2.116 1.575 0.581 35.881
ROE 10388 0.014 0.323 -17.255 3.277
Log Market Value 10388 7.429 1.530 -3.906 3.073
Log Net Income 8689 4.463 1.709 -3.576 10.584
R&Da 10410 0.401 10.288 -3.910 953.470 R&Db 5613 0.744 14.002 -3.910 953.470 Leverage 10410 1.124 31.341 -470.178 3096.639 Size 10410 1.149 1.920 -6.215 7.496 Awarenessa 10410 0.012 0.055 0 3.324 Awarenessb 4187 0.029 0.084 0 3.324 ROS 10410 -0.720 25.795 -2305.480 19.878 Reputation 167 6.964 0.640 5.220 8.600 Clean Reputation 167 0.074 0.042 -0.042 0.224
Note: both Awareness and R&D are displayed twice: superscript “a” denotes the data as used in the regressions, whereas superscript “b” denotes the original data as obtained from Compustat. Data are summarized for the period 1994-2006 for all variables except for Reputation. Data on this variable is only summarized for the year 2006.
31 Table 2
Correlations CSR Tobin’s
Q
ROA ROE ROS Net Income
Market Value
Log Employees
Aware R&D Leverage
CSR 1.00 Tobin’s Q 0.07 1.00 ROA 0.05 0.02 1.00 ROE -0.001 -0.002 0.40 1.00 ROS 0.00 -0.02 0.09 0.02 1.00 Net Income 0.23 -0.003 0.18 0.24 0.08 1.00 Market Value 0.25 0.09 0.19 0.12 0.02 0.90 1.00 Log Employees 0.18 -0.15 0.19 0.04 0.04 0.68 0.71 1.00 Awareness 0.04 0.12 -0.08 -0.02 -0.01 0.01 -0.01 -0.02 1.00 R&D -0.002 0.03 -0.10 -0.02 -0.90 -0.11 -0.03 -0.05 0.01 1.00 Leverage 0.004 -0.01 -0.01 -0.01 0.001 0.03 -0.002 -0.002 -0.003 -0.001 1.00 Reputation 0.20 0.80 0.94 -0.03 0.28 0.15 0.15 0.05 0.17 0.42 -0.42 4.1 Panel Model
This section provides two tables containing the empirical findings of the general panel model. ROA and Tobin’s Q are used as the dependent variables in Table 3 and 4 respectively. For each dependent variable, the model is estimated six times. The first specification includes all variables only as separate regressors, the second specification also includes the interaction term between CSR and Awareness, whereas the third specification can be described as the full interaction model. Each of these model specifications is estimated with- and without firm fixed effects. Time fixed effects were used in all regressions8.
The regressions results using ROA as the dependent variable are displayed in Table 3 below. The results provided in the table followed from a sample of 10410 observations divided over 2738 different U.S. companies. The basic model without firm fixed effects shows a positive and significant coefficient for Size at the 1% significance level. The coefficients for Awareness, Leverage, and R&D were all found to be negative. However, only Awareness and R&D were found to be statistically significant, both at the 1% significance level. CSR was also found to differ statistically and significantly from zero at the 1% significance level, implying a negative effect on ROA. When the model was extended with the interaction term between CSR and Awareness, the results remained more or less the same. At the 1% significance level, Size was still found to be positive and statistically significant, whereas Awareness and R&D were found to be negative and statistically significant.
