David Bots
Master Controlling
Corporate Social Responsibility Performance
and
Financial Analyst Properties
A meta-analytic review
Corporate Social Responsibility Performance
and
Financial Analyst Properties
A meta-analytic review
David Bots
S3273016
Master Controlling
Faculty of Economics & Business
University of Groningen
June, 2018
Supervisor
dr. A. Bellisario
Co-assessor
prof. dr. D.M. Swagerman
Word count
10.946
Abstract
This study investigates the relationship between corporate social responsibility (CSR) performance and financial analyst properties. Conducting a meta-analysis, a quantitative method to draw conclusions out of the literature, it is the goal of this study to find a clear picture in the fragmented literature. This meta-analytic review with 39 studies resulted in five findings. First, analyst recommendations are positively related with CSR performance. Second, there is a negative relation between CSR performance and information asymmetry. Third, there is a positive relation between CSR performance and forecast accuracy. Fourth, there is a positive relation between CSR performance and forecast error. Last, there is a positive relation between analyst coverage and CSR performance.
Keywords: corporate social responsibility performance, financial analyst properties, analyst coverage,
Table of content
Abstract ... 2
Table of content ... 3
1. Introduction ... 4
2. Literature background ... 7
2.1 Corporate social responsibility performance ... 7
2.2 The role of financial analysts and corporate social responsibility performance ... 7
3. Hypotheses development ... 9 3.1 Agency theory ... 9 3.2 Legitimacy theory ... 10 3.3 Signaling theory ... 12 3.4 Stakeholder theory ... 13 4. Research methodology ... 16 4.1 Literature search ... 16
4.2 Criteria for relevance ... 17
4.3 Analysis ... 18
5. Results ... 20
5.1 Corporate social responsibility performance and analyst coverage ... 20
5.2 Corporate social responsibility performance and forecast accuracy ... 22
5.3 Corporate social responsibility performance and forecast error ... 25
5.4 Corporate social responsibility performance and information asymmetry ... 26
5.5 Corporate social responsibility performance and the recommendations of financial analysts 28 6. Discussion and conclusion ... 29
References ... 32
Appendix 1 – Characteristics of included studies ... 46
Appendix 2 – Search log ... 51
Appendix 3 – Coding document ... 61
Appendix 4 – PRISMA Checklist ... 62
1. Introduction
In academic literature there is increasingly attention to corporate social responsibility (CSR) and it is becoming increasingly complex due to the many directions in the literature (Huang & Watson, 2015). Also, the question “Does it pay to be good?” is asked more often (Barnett & Salomon, 2012; Johnson, 2003; Margolis et al., 2009). Which reflects the idea that CSR is not only good from the ideological point of view, but also from a financial perspective. Therefore, it is increasingly investigated in relation with financial performance (Margolis et al., 2009), but what is the relationship with financial analysts?
Why firms choose to operate in a social responsible manner is not clearly answered in the current literature, since fragmentation of results have led to highly conflicting evidence. Gupta et al. (2017), for example, have argued that firms participate in CSR due to internal and external factors. External factors are, for example, government regulations (Kassinis & Vafeas, 2006), activists (King, 2008), isomorphic and institutional pressures (Briscoe & Safford, 2008; Matten and Moon, 2008), and product-market competition (Pacheco & Dean, 2015). The internal triggers, as found by Gupta et al. (2017), could be determined by the ideology of employees in the company (as found in US firms). Lindgreen and Swaen (2010) have argued that companies also choose to participate in CSR because it can be good for the company (win–win).
Much research has been done to find a relation between CSR and financial performance (Griffin & Mahon, 1997; Margolis & Walsh, 2003; Orlitzky et al., 2003; Roman et al., 1999; Rowley & Berman, 2000; Ullman, 1985). Some researchers have found a positive relationship (Dhaliwal et al., 2011; Margolis et al., 2009; Plumlee et al., 2010; Porter & van der Linde, 1995). They argue that investing in CSR will result in a competitive advantage. Cheng et al. (2014) have found that when a company’s CSR performance is good, it has better access to financing. Other researchers have found CSR to be an indication of a strong commitment to stakeholders (Choi & Wang, 2009), which will reduce the risk of opportunistic behaviour or short-term orientation. This will lead to reduced agency costs (Jones, 1995; Eccles et al., 2012). In a similar vein, Roberts (2011) has found that companies with low CSR scores pay more interest on their bank loans compared to companies with high CSR scores, and El Ghoul et al. (2011) has found that companies with high CSR scores pay less for their equity. Mackey et al. (2007) have theorized that managers can maximize the market value of the company, but at the same time not have the maximum present value of future cash flows due to funding social responsible activities. Besides that, Nguyen et al. (2017) finds that CSR activities create shareholder value due to the lower cash flow risk.
On the other side of the argument, some researchers found that social behaviour has higher costs than benefits and conclude that CSR is negatively associated with a firm’s financial performance (Friedman, 1970; McWilliams & Siegel, 1997). Moore (2001) for example found a negative relationship between CSR and financial performance in the supermarket industry in the United Kingdom. This relation was confirmed by Brammer et al. (2006), who also found that firms with a high CSR score have lower returns. Masulis and Reza (2015) have further found that corporate philanthropy, a part of CSR, destroys shareholder value. Another study, performed by Krüger (2015), has found that investors react negatively against CSR when this is caused by agency problems, because this money cannot be put into other more profitable projects. Otherwise, Nelling and Webb (2009) found no relationship between CSR and the financial performance of a firm.
Financial analysts making information about the company, such as financial performance and CSR performance, readable for investors, so that investors are better in assessing information about CSR performance (Ivković and Jegadeesh, 2004). According to Luo et al. (2015), analysts have a mediation role between companies and investors. Both Luo et al. (2015) and Fieseler (2011) have interviewed analysts concerning CSR performance and found that the analysis of CSR performance and strategy are becoming mainstream.
Of the five properties of financial analysts investigated in this study, the first is the number of financial analysts following or analyst coverage . Adhikari (2016) has investigated the relationship between CSR and analyst coverage. This author found a negative relation between analyst coverage and CSR. He argues that CSR is an agency problem and that the presence of analysts cause managers to reduce spending on social projects. This coverage has also been examined by Jo and Harjoto (2014). They have found that when more analysts are following a company, CSR activities decrease. However, the strengths of CSR are not significant influenced by the number of analysts following. Hong and Kacperczyk (2009) studied ‘sin’ stocks. These are stocks for listed companies that selling alcohol, tobacco, and gaming. ‘Sin’-companies are held less by institutional investors (e.g. pension funds) and are less covered by analysts, but have higher returns than comparable not-‘sin’ companies.
The second property concerns forecast accuracy from financial analysts. The opposite of the second property is the third property— forecast error, which measures the dispersion of the forecast. Both forecast accuracy and forecast error are involved in the information environment in which financial analysts work. In recent literature, the main finding is that when more information is available, the accuracy of financial analysts will increase and the forecast error will decrease (Orens & Lybaert, 2007; Vanstraelen et al., 2003; Yu, 2010).
The fourth property is the information asymmetry. When a company discloses more about its performance, information asymmetry decreases. Information asymmetry is the gap between what is known at the company and what is known by financial analysts. There is not a clear measure for information asymmetry; therefore, researchers use different proxies (e.g., share price volatility, bid-ask spreads) to measure it (Cormier et al., 2015). High information asymmetry means that a company’s actions or performance cannot be properly assessed (Cui et al., 2018). This is especially the case with CSR performance, which is complex to measure (Luo et al., 2015, Surroca et al., 2010).
The final property is the recommendations from financial analysts. These recommendations influence investors, the stock-price, and trading volume (Francis & Soffer, 1997; Ioannou & Serafeim, 2015; Moreton & Zenger, 2005; Stickel, 1995; Womack, 1996). Interviews in the studies of Luo et al. (2015) and Fieseler (2011) have revealed that analysts take CSR performance into account when they issue their recommendations. A study from Ioannou and Serafeim (2015) has linked CSR performance and analysts’ recommendations. They have investigated the relationship between CSR ratings and the assessments of sell-side analysts. They found that in the 1990s , analysts were pessimistic about firms with high CSR ratings, but became more and more optimistic over time. Above that, this study has revealed that high-status analysts (for example working at Goldman Sachs) are the first to be optimistic about high CSR ratings.
What is the relationship between CSR Performance and financial analyst properties?
This gap meant to be filled is also urged by Ioannou and Serafeim (2015, p.1076). They say that ‘more needs to be done to understand the more nuanced mechanisms at work regarding how CSR categories and policies are perceived and evaluated by analysts as well as other social actors within public equity markets’. Zhao and Murrell (2016), as well, are curious about stakeholders’ responses to CSR performance.
In order to answer the research question, I perform an effect-size meta-analysis which synthesizes existing literature (Pomeroy & Thornton, 2008). This quantitative research method is appropriate for providing insights into the reasons for conflicting findings in the literature (Pomeroy & Thornton, 2008). Besides that, this method can offer a clear picture of a complex and highly varied literature.
Results from the 39 studies that are used in this meta-analysis offer new insights into the field of corporate social responsibility. First, I find a positive relationship between CSR performance and analyst coverage. Second, there is a positive relation between CSR performance and forecast accuracy. Third, there is a positive relation between CSR performance and forecast error. Fourth, there is a negative relation between CSR performance and information asymmetry. Lastly, there is a positive relation between CSR performance and the recommendations of financial analysts. I discussed that financial analysts are becoming more positive about CSR and that CSR performance improves the information environment around a company.
This thesis is structured as follows. In the next section I give my hypotheses. Thereafter I discuss the chosen research method and give the results of my research. In the last chapter, I discuss the findings and give the main limitations, implications, and future research opportunities.
2. Literature background
In this chapter, I give a short description about what the literature tells about corporate social responsibility performance and how it is measured as a prelude to the hypothesis development.
2.1 Corporate social responsibility performance
Corporate social responsibility is of growing interest in the academic literature and for practitioners. According to Basu and Palazzo (2008), three lines of CSR inquiry are present in the academic literature. The first addresses stakeholders. In this, CSR is seen as a response to the demands of stakeholders. The second area is based on CSR performance, where CSR activities are monitored and measured according to the standard of expectations. This area of research concerns the question ‘What is business expected to be or to do to be considered a good corporate citizen?’ (Carroll, 1998, p. 1; Wood, 1991). The third area is about companies’ motivation to behave in a socially responsible way. Motivation can be extrinsic: corporate reputation (Fombrun, 2005), avoiding legal punishments (Parker, 2002), avoiding risk (Fombrun et al., 2000; Husted, 2005), or generating loyalty (Bhattacharya & Sen, 2004; Sen & Bhattacharya, 2001). Alternatively, motivation can be intrinsic, mainly present in psychological concepts (Bowie, 1999; Donaldson & Dunfee, 1994). Aguinis and Glavas (2012) synthesized the literature and found three reasons why firms engage CSR. First, organizations expect that they will financially benefit from this behaviour. Second, the organization’s values concern the social behaviour of the company. Third, organizations believe that behaving in a ‘good’ manner will result in positive outcomes for the organization, such as improved quality, improved operational efficiency, and attractiveness to investors.
Accumulating different perspectives from the literature (e.g. Wartick and Cochran, 1985) , Wood (1991) tried to define CSR performance. According to Wood (1991, p. 693) CSR performance is “a business organization’s configuration of principles of social responsibility, processes of social responsiveness, and policies, programs, and observable outcomes as they relate to the firm’s societal relationships”. However, this definition is lacking because there is no word about the company’s environmental performance, despite this being measured in corporate social performance (CSP)-ratings (Chatterji et al., 2009). Dahlsrud (2008) noticed this and investigated how CSR is defined by practitioners and academics. He found five dimensions: the stakeholder dimension, social dimension, economic dimension, voluntariness dimension, and environmental dimension. Using the literature review by Carroll (1999), he argued that the environmental dimension had not been included in the definitions in the past, and therefore it was often not included in the definition later either.
Measuring these dimensions of CSR performance in terms of strengths and weaknesses is often done by rating agencies such as Kinder, Lydenberg, and Domini (KLD) Research and Analytics. They explain CSR activities from the past and future goals for their clients’ knowledge. Therefore, they try to reduce the information asymmetry for their clients (Chatterji et al., 2009).
2.2 The role of financial analysts and corporate social responsibility performance
Financial analysts create analyst reports to inform investors. They are intermediaries in the capital market. Their analyst reports contain private financial information (from annual reports, press releases, and public conservations with the management, Soltes, 2014) and analyse future earnings and cash flows
an important role in providing an in-depth understanding of CSP. In addition, the information that KLD and Thomson Reuter (Asset4) provide with their CSP ratings must be analysed (Fombrun et al., 2000). According to Stout (2012), a company’s valuable assets are not merely its shares, but also its stakes in the community, economy, and planet. This broad definition of an asset makes it even more difficult for financial analysts to interpret CSR performance (Luo et al., 2015). With this role in the information environment, they can reduce the information asymmetry by solving two problems. First they can help capital providers to evaluate their investment opportunities by reducing biased information that companies give, the valuation problem. Second they can reduce agency problems by monitoring. These problems occur when there is a separation between the ownership (shareholders) and the control of a company (management, Beyer et al., 2010).
Financial analysts also provide buy or sell recommendations about companies. Evidence demonstrates that the behaviour of investors and the public is significant influenced by analysts’ reports. Their reports also affect the price of shares and trading volume (Francis & Soffer, 1997; Ioannou & Serafeim, 2015; Moreton & Zenger, 2005; Stickel, 1995; Womack, 1996). Womack (1996), for example, found that positive (added-to-buy) recommendations can increase stock price by 5%. Negative (added-to-sell) recommendations ensure that price goes down by 11% on average. This effect takes place immediately, but also endures in the months thereafter.
Jegadeesh et al. (2004) have, however, some criticisms of the work of financial analysts. They disagree that analysts have the ability to downgrade stock quickly when there are some other negative investment signals. In addition, they find that analysts’ recommendations can be driven by incentives that are misaligned with information that is available about a company. The researchers suggest that there are some incentives that ensure that growth stocks have better recommendations than they actually should. This perspective is supported by the findings of Jackson (2005), who argues that while analysts want to give accurate recommendations for the sake of their reputations, they also want to give optimistic forecasts so that they earn money via trading commissions when their clients buy stocks.
3. Hypotheses development
In this chapter I discuss how the properties of financial analysts are related to corporate social responsibility performance. Besides this, I give several hypotheses as answers to the research question about the role of financial analysts with CSR performance. As Deegan (2002, p. 288) has said, “However, reflecting the fact that we do not have an “accepted” theory for social and environmental accounting, there is much variation in the theoretical perspectives being adopted”. Since no current theory relates specifically to CSR, I build upon several others: agency theory, stakeholder theory, legitimacy theory, and signalling theory (based on Pérez, 2015). Properties related to financial analysts considered in this section are analyst coverage, forecast accuracy, forecast error, information asymmetry, and analyst recommendations. Figure 3.1 overviews the five hypotheses that develop from considering these properties and theories.
3.1 Agency theory
The first direction in which CSR performance can be placed is agency theory. This theory says that managers must act according to the interests of shareholders (Jensen & Meckling, 1976). Based on agency theory, Milton Friedman said in the New York Times in 1970 that “the social responsibility of the firm is to increase its profits” and noted further that a company must make as much money for the shareholders as possible (Friedman, 1970). According to Friedman (1970), costs of CSR are agency costs.
This statement by Friedman (1970) is examined by Adhikari (2016). He studied analyst coverage and CSR. Adhikari (2016) argues that analyst coverage has a negative causal effect on CSR appearing not immediately, but two years after the fact. With his findings, he supports agency theory’s idea that monitoring a firm result in decreased discretionary spending by managers, including on costs involved with CSR. To explain the negative correlation, he found that when the ownership of a CEO in an organization declines, together with an imperfect monitoring mechanism by analysts (lower analyst coverage), CSR spending can increase (Cheng et al., 2014). A second explanation, from Irani and Oesch (2016), is that less analyst coverage will result in more discretionary spending. This is because of the less focus on earnings management and therefore more CSR expenses.
In addition, Masulis and Reza (2015) have investigated if corporate philanthropy, a part of CSR performance, is beneficial for the company’s revenue or for the shareholder’s wealth. They argue that corporate giving is a conflict of interest between shareholders and the organization’s managers. First, they found that the CEO-affiliated charity contributions decline if the CEO’s financial interests are more in line with the interests of the shareholders. Second, they found a negative cumulative abnormal return at firms that disclose their charity awards. This exceeds the nominal value of announced charitable award programs. With as consequence, that the market capitalizes future contributions into the share price. Finally, they found that organizations’ donations to foundations increased when the CEO had connections with the charity foundation or when there was a weak corporate governance. Masulis and Reza (2015) concluded that corporate philanthropy is indeed an agency problem and it destroys shareholder value.
problems and not by solving operational issues. This is because investors do not want CSR policies in the company, because they see it as a cost.
Financial analysts play an important role in reducing the agency costs. Increasing the number of financial analysts (analyst coverage) helps to monitor managers and therefore protect the shareholders’ interests (Chen et al., 2015; Jung et al., 2012; and Yu, 2008). Chen et al. (2015) found clear evidence that when there is less analyst coverage, the chance is higher that the manager of the company uses the company’s assets for his own interests and not the company’s. Other research has argued that analysts put too much pressure on a company’s management, causing managers to only focus on short-term performance. This can be at the expense of spending on social welfare (He & Tian, 2013). Alternatively, as has been found by Shi et al. (2017), financial analysts can bring so much pressure that the managers’ intrinsic motivation to serve the shareholders’ interests will disappear. This has as consequence a greater risk of financial fraud or illegal activities (Barsky, 2008; Mishina et al., 2010).
Concluding from this, there is evidence that higher analyst coverage helps monitor managers and therefore reduce agency costs (e.g., CSR). Above that, high coverage could lead to more pressure and therefore short-term thinking which will reduce social behaviour. Based on this, I hypothesize that analyst coverage is negatively related with CSR performance. The higher the analyst coverage, the lower the corporate social behaviour. Thus,
Hypothesis 1: Analyst coverage is negatively related with Corporate social responsibility performance.
3.2 Legitimacy theory
The second theory that can play a role in explaining CSR performance is legitimacy theory. Applying this theory, a company would disclose information about CSR performance to earn a socially responsible image. With legitimacy theory, a company can legitimate its behaviour to its stakeholders (Branco & Rodrigues, 2006). Legitimacy theory is built upon the idea that a company exists because of society. In return, society expects that the company will fulfil its expectations, values, and norms (Deegan, 2002). Societal expectations are not the same for every industry; some industries have more reasons to legitimate their social behaviour. For example, research in the banking sector found that banks with a higher visibility among consumers exhibit greater concern to improve their social image through disclosing their CSR performance (Branco and Rodrigues, 2006).
According to Lindblom (1994), a company can use CSR to gain legitimacy in the following ways (Magness, 2006):
To correct the public view about the company’s performance in the case of a legitimacy gap. To change the public’s expectations to a more realistic perception of the company’s
responsibilities.
To show how the company has improved its performance, if the company has not been meeting the expectations.
To make sure that good performance is more emphasized than poor performance.
In legitimacy theory, there are opposite views about CSR performance and CSR disclosure. As Gray et al. (1995, p.47) have noted about the CSR literature, “...there is little about CSR which is not contestable – and contested”. On one hand, researchers have argued that companies that perform well on CSR are
2008). On the other hand, researchers have argued that companies want to disclose more CSR to improve the public perception (Cho & Patten, 2007; De Villiers and van Staden, 2006).
Besides needing financial information, a company also has to take care of its legitimacy in the community. According to Cormier and Magnan (2015), there is a tension between how these two factors work together. Cormier and Magnan (2015) found that disclosures of environmental performance are useful to financial analysts and that these disclosures help to improve (directly and indirectly) the information environment, because financial analysts are able to recognize inaccuracies in the disclosure based on the underlying CSR performance. Ultimately, more information will lead to a higher forecast accuracy, less uncertainty, and therefore fewer/lower forecast errors. This further affects stakeholder perception of the company’s legitimacy (Bhat et al., 2006).
Lanis and Richardson (2013) used legitimacy theory to compare tax-aggressive companies (companies that try to pay as little tax as possible through different constructions and have therefore low CSR performance) and non-tax-aggressive companies in Australia. They found that tax-aggressive companies disclose more information about their corporate social behaviour than the non-tax-aggressive. The reason is that companies want to alleviate potential negative concerns of the public; therefore, they move focus to positive points and the ways they are meeting expectations and they give less attention to their tax-aggressive policies. Or as has been found by Patten (2002), there is a significant relation between levels of irresponsible behaviour (such as toxics released) and levels of environmental disclosure. High polluting companies, for example, disclose this performance indication because they face pressure from their social and political environment. Consistent with Patten (2002), Cormier et al. (2011) have found that CSR performance has a direct effect on the disclosure of CSR: companies that are more polluting disclose more about their CSR performance than low polluting companies. Information asymmetry is reduced by this disclosing. This means that the market has more information available. This information could be available in annual reports, investor relations, or other kinds of disclosures. These kinds of communications from the company are an important source through which analysts can improve forecast accuracy (Lang & Lundholm, 1996).
A company’s CSR performance and CSR disclosure are closely related. Disclosing a firm’s social or environmental performance will generally lead to greater forecast accuracy, as environmental and social disclosure give information to the market that will be factored into the value of the shares (Aerts et al., 2008; Barth & McNichols, 1994; Cormier et al., 2009; Li & McConomy, 1999). However, this improved accuracy is relatively more reliable in continental Europe than in North America (Aerts et al., 2008).
Thus, based on the evidence here presented, companies that are acting responsible or irresponsible, disclose more about their CSR activities. This will result in higher forecast accuracy because financial analysts, as part of the market, can distinguish environmental friendly firms and ‘greenwashers’ (environmental wrongdoers, Du, 2015) from each other. I therefore hypothesize that:
Hypothesis 2: Forecast accuracy is positively related with Corporate social responsibility performance.
If the outcome of CSR performance or behaviour is visible, then it is factual legitimacy. However, performance is not always visible, and therefore it is not easy to understand. According to Hunter and
legitimated by its stakeholders when it provides them extensive information about its environmental performance (Alrazi et al., 2015). A company also secures its legitimacy when it conforms to the expectations of its stakeholders (Bansal & Clelland, 2004). When more information is available about the company and its CSR performance, analysts’ forecast error will decrease. Lang and Lundholm (1996) have found that the dispersion among financial analysts decreases when more information is disclosed. Rounding the forecast is an example when analysts have too little information. Consequently, this results in greater forecast error (Herrmann & Thomas, 2005). They studied analysts who do not have the resources, time, or money to gather all the information themselves and have found that as a result, analysts rounded their forecasts.
Concluding from this, when more is known about a company’s performance (including its CSR performance), analysts make fewer/lower forecasting errors. Thus, based on the reasoning from hypothesis 2, that more information about CSR performance leads to more accuracy, I propose the following:
Hypothesis 3: Forecast error is negatively related with Corporate social responsibility performance.
3.3 Signaling theory
Signaling (or signalling) theory explains and solves the information asymmetry in markets. This theory argues that this information gap can be reduced through providing more information signals to others. To have an advantage, a signal must not be easily copied by companies that are performing badly (Morris, 1987). The incentive to reduce information asymmetry resembles agency theory. Morris (1987) has argued that signaling theory and agency theory have considerable overlap. Both pertain to monitoring and agency costs. Signaling theory aims to reduce monitoring and agency costs by sending signals. This could be done by reporting more on their CSR Performance.
Based on signaling theory, Ross (1977) and Spence (1973) have explained that high-performance companies send signals with financial information to the market. In response to these signals, the market makes a distinction between well performing companies and poorly performing companies. Sending these voluntary signals of private information reduces information asymmetry and ensures that a company’s performance is presented in its best light (Khlifi & Bouri, 2010). CSR performance is positively associated with future financial performance, according to signaling theory. However, this positive association is not due to economic factors such as ROA (return on assets), but the effects of CSR expenditures (Lys et al., 2015).
Based on the ideas from Morris (1987), Ross (1977), and Spence (1973) have observed that companies can distinguish themselves through voluntary disclosure about their performance. Besides that, Marston and Polei (2004, p. 293) have argued that voluntary disclosure could lead to “a more efficient evaluation of the future prospects of the firm by the capital markets”. This efficient evaluation can be achieved through an improved information environment by reducing the information asymmetry in the market.
Multiple researchers have studied how a company can attract potential stakeholders by sending signals about its CSR performance. Sending signals about CSR performance provides more information in the market and reduces the information asymmetry between potential stakeholders and the company (Sen et al. 2006). Mahoney et al. (2013) have found evidence that American companies with better corporate social performance publish standalone reports about CSR in their company. Mahoney et al. (2013) have confirmed that companies that are performing well in CSR disclose information about their CSR
Turban and Greening (1997) and Greening and Turban (2000) based their research upon the signaling theory and social identity theory. Their goal was to understand if CSR is a driver in potential employees’ attraction to an organization. Sen et al. (2006) have found that stakeholders who know about a company’s CSR performance have a more positive association with that company. Additionally, they have found that there is a greater intent to purchase products, seek employment, and invest in the company than when stakeholders are not familiar with the organization’s CSR activities. Turban and Greening (1997), Greening and Turban (2000), and Sen et al. (2006) have given evidence that companies want to reach potential stakeholders -such as employees, investors, and financial analysts- more effectively by sending information about their social behaviour. This signaling results in a better information environment and a decrease of the information asymmetry (Cui et al., 2018).
In addition, Albinger and Freeman (2000) have used the signaling theory to investigate if CSR performance influences the attractiveness of job seekers. Their study finds CSR performance not only a fulfilment of social responsibilities, but also as a competitive advantage. This is confirmed by Turban and Greening (1997), who have stated that companies with high CSR performance have better reputations. Each of these stakeholder groups will judge the signals that a company send with its CSR Performance in for example their annual report (Fombrun & Shanley, 1990).
One important signal is a company’s inclusion in the Dow Jones Sustainability Index (DJSI). Robinson et al. (2011) have found that when a company is added to this index, the share price will sustainably increase. Likely the benefits of being included on the DJSI outweigh the costs of it (e.g. additional reporting requirements). However, Cho et al. (2012) have concluded that being listed in the DJSI is negatively related with environmental performance because of the more extensive disclosure levels of the bad performing companies. Thus, a company has a better reputation when it is listed, but being listed does not indicate high environmental performance. However, inclusion in a sustainable index, such as the DJSI, provides information about the company and therefore reduce information asymmetry as firms must publicize information regarding their CSR performance at extensive disclosure levels (Cho et al., 2012; Doh et al.,2010).
Thus, summarizing the evidence here discussed, companies send signals to their stakeholders. When a company is performing well on CSR, there are more signals. This reduces information asymmetry between the company and stakeholders, among which are financial analysts. Thus, I hypothesize the following:
Hypothesis 4: Information asymmetry is negatively related with corporate social responsibility performance.
3.4 Stakeholder theory
A theory frequently used to explore CSR is stakeholder theory (Egels-Zanden & Sanberg, 2010). Stakeholder theory considers which people or which organizations around a company are important. Without these stakeholders the company cannot continue (Clarkson, 1995; Hillman & Keim, 2001). There are two kinds of stakeholders: primary stakeholders, such as capital suppliers (e.g., shareholders), employees, costumers, community residents, and the natural environment (Clarkson, 1995). The secondary stakeholders are the people or organizations who have influence on companies, but are not involved in direct transactions with the company (e.g., the media or special interest groups). According
understand the stakeholder environment and identify who has the most impact on the continuity of the company (Bhattacharya & Korschun, 2008). Financial analysts are one of the stakeholders in the framework. Financial analysts are secondary stakeholders because they are not directly involved in transactions, but provide information to investors, who are directly involved. These analysts influence the judgement of investors significantly (Bercel, 1994; Dhaliwal et al., 2012; Nichols, 1989; Schipper, 1991; Walther, 1997).
Based on stakeholder theory, there is a positive relationship between the financial performance of a company and CSR, because pleasing the stakeholders implies that the financial performance of a company will increase (Donaldson & Preston, 1995; Jones, 1995; Orlitzky et al., 2003). This implies that financial analysts will give a positive recommendation about the company.
Brower and Mahajan (2013) have conducted a study about CSR performance in the light of stakeholder theory. They found that companies could broaden their CSR performance. The drivers behind this were sensitivity and responsiveness to stakeholders’ demands, the diversity of these demands, the nature of stakeholder scrutiny in the cases of large firms, and the risk that stakeholders acts (branding strategy of the company). Brower and Mahajan (2013) argue that one valuable approach to manage relationships with stakeholders is to invest in a CSR performance strategy. There are four arguments for investing in a CSR strategy. First, CSR performance is an important way to market the company to stakeholders (Hoeffler et al., 2010). Second, CSR performance assessment is built upon the needs of the diverse range of stakeholders (Ruf et al., 2001). Third, the visible consequences of CSR performance and the way processes are performed are the basis on which stakeholders judge a company’s overall performance (Wood, 1991). Concluding from this, Brower and Mahajan (2013) have argued that engaging in CSR gives a signal to stakeholders that is difficult for competitors to copy. CSR development will ultimately result in building a common ground and identity alignment with stakeholders on common values (Meyer & Rowan, 1977; Brickson, 2007; Janney & Gove, 2011). Based on these four arguments, CSR is a way to build a competitive advantage and better future performance. The ultimate outcome will be positive recommendations from financial analysts.
Chang et al. (2014) have studied analysts’ recommendations in Taiwan. They found evidence, after controlling for self-selection bias, that CSR-positive companies have higher recommendation scores (most of which are ‘HOLD’) and receive more attention, resulting in more recommendations. They argue that companies engaging in CSR are more recommendable than others and that this is even more the case in industries with high growth. In addition to this quantitative study, Luo et al. (2015) have performed a qualitative study of analysts’ interpretation of CSR performance. They found that financial analysts have more attention for CSR performance. As a sell-side analyst explains in an interview: “Even if financial analysis suggests a stock is undervalued, we do not issue a buy recommendation if the firm is likely to receive negative CSR reports” (Luo et al., 2015, p.125). In an interview with Fieseler (2011, p.138), another analyst argues that it is difficult to see the financial results of CSR performance immediately. However, the analyst believes that this behaviour adds value in the long-term for a company. Ioannou and Serafeim (2015) have concluded that analysts are growing positive about CSR. These researchers investigated how sell-side analysts perceived CSR and noted a theoretical shift from agency theory towards stakeholder theory. They argue that the shift from agency theory to stakeholder theory affects the proactive CSR activities that interact with the company’s stakeholders. This ensures that risk will be mitigated and firm value will increase.
CSR Perfromance Analyst coverage Information asymmetry Recommen-dations Forecast error Forecast accuracy
Figure 3.1 Overview of hypotheses regarding CSR performance.
therefore competitive advantage. This advantage could lead to better financial performance, which will positively affect the recommendations of financial analysts. Thus, the last hypothesis is:
Hypothesis 5: Analyst’ recommendations are positively related with Corporate social responsibility performance.
Figure 3.1 gives a clear view of all the hypotheses. Per property can be seen what the hypothesized relationship is.
4. Research methodology
In this chapter, I describe the methodology of my research and how I use it to answer the research question. I use for this a meta-analysis; therefore, I describe the literature search, the criteria for relevance, and my method of analysing the numbers.
In this study, I retrieve my answer to the research question through a meta-analysis. This is a combination of a literature review and a quantitative research. With this technique I collect papers that describe the relationship between CSR performance and financial analyst properties. However, performing a meta-analysis has two problems in the field of accounting and auditing (Pomeroy & Thornton, 2008). The first problem is that there are no clear definitions of the variables. By conducting an intensive literature search and making use of papers such as Wood (1991) is this problem reduced. The second problem is the existence of publication bias and researcher incentives. Publication bias refers to the fact that negative results are generally not published. This has an impact on the representativeness of published papers (Peterson, 1989). Only publishing innovative and significant research to increase their reputation is also covered by the second problem. When papers are not able to do that, they are often not published (Burgstahler, 1987; Whitley, 2000). A result is that published papers have high statistical power, but low effective levels (Pomeroy & Thornton, 2008). To reduce this problem, tests are being done to see if the publication bias has influence on the outcomes. Despite these difficulties, Pomeroy and Thornton (2008) have concluded that meta-analyses conducted in the accounting field can generate helpful insights. A meta-analysis is helpful, for example, for answering complex questions and explaining deep-seated phenomena (Khlif & Chalmers, 2015).
4.1 Literature search
My research on CSR performance and financial analysts includes 39 papers, mostly from the United States, from 2007 to 2018 (see Tables A1.1, A1.2, and A1.3 in Appendix 1 for the characteristics). My literature search was based on the recommendations of PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-analysis, Moher et al., 2009; Liberati et al., 2009). The PRISMA guidelines consists of a checklist of 27 items (Moher et al., 2009, p. 4, see also Appendix 4) and a flow-chart with four stages: Identification, Screening, Eligibility, and Inclusion (Moher et al., 2009, p.3). PRISMA offers handles to select the right papers.
I selected the papers in several ways. First, I found papers through a search on Google Scholar. I also used SSRN in the beginning, but discovered that Google Scholar searches this same database. I also used SmartCat, but this search engine did not produce the desired results. The keywords used to find variables for CSR performance were (corporate) environmental performance, (corporate) social performance, ESG performance, sustainability/sustainable performance, KLD, and Asset4. To find a relationship with analysts I used the following keywords: analyst coverage, forecast accuracy, forecast error, analyst recommendations and information asymmetry (a detailed search plan is in Appendix 2). These keywords were chosen based on the literature in the literature background. Based on Pomeroy and Thornton’s (2008) findings about the unclear definition, multiple search terms were used so that as many relevant studies as possible were found. Second, I also used the references in the articles to find papers that describe the relationship between CSR performance and financial analyst properties. Last, my supervisor and a colleague-student, who are both experienced in the field of CSR, gave usable papers.
In addition to the 32 published papers collected in this search, there are also seven unpublished or working papers (Table 4.2). They can add value, but they are generally not peer-reviewed (Pomeroy & Thornton, 2008). Published studies are peer-reviewed and represent therefore a higher quality, but can convey publication bias (Hay et al., 2006). Unpublished papers could have errors, but it is expected that errors will be cancelled out when more papers are included. Moreover, including more papers outweighs the disadvantages of using papers with limitations (Hay et al., 2006).
4.2 Criteria for relevance
For a paper to have been relevant for the meta-analysis, it had to have the following characteristics. First, the papers must describe (in English) a quantitative relation between CSR performance and financial analyst properties. Second, the studies must have a correlation matrix between these two. At the beginning of this research, studies without a correlation matrix were also included and emails were sent to the authors with the question if they could send a correlation matrix. However, it took too long to receive answers, and some authors did not reply at all. Ultimately, the paper by Adhikari (2016) is the only paper to be included in this study without a published correlation matrix, but the author sent a matrix via email. When I encountered a relevant study, I coded it in a specially made document with all relevant information, such as the number of observations and correlation coefficients (see Appendix 3).
Below is an overview of how the search was performed according to the PRISMA stages of Moher et al. (2009) and Liberati et al. (2009) (Figure 4.1):
Stage 1, Identification. From different papers (such as Wood, 1991) keywords emerged that could identify the papers in the databases. These keywords were used systematically so that the search could be replicated (see Appendix 2).
Stage 2, Screening. The title, abstract and the name of the journal were screened to see if there was information that I could use for the meta-analysis. Per search only the first five pages of the search were screened. This is due to the limited time I have and after the first five pages of the search the relevance of the items was not high anymore, because these papers were no longer about the relationship I was looking for (CSR performance – financial analyst properties). Some papers were excluded because the relationship was not in the text itself, but mentioned only in the references.
Stage 3, Eligibility. The articles that remained after stage two were still not yet usable for the meta-analysis. At this point, I examined in detail whether there was a relation between CSR performance and financial analysts. Besides that, I checked for the presence of a correlation matrix with both parts. Reasons for exclusion per search were not having the right variables, no correlation matrix in the study, the paper being already included in another search, and the variables were named in references, the theoretical background, or in other non-relevant parts.
Stage 4, Inclusion. The last stage revealed the number of approved articles that were used for the meta-analysis.
Figure 4.1 Flowchart of the PRISMA process.
4.3 Analysis
There are two ways of performing a meta-analysis. First, the vote-counting method. This method is still widely used today, but has validity issues. It has some characteristics of a summary. The statistical numbers (such as the estimated partial regression coefficient) of the study are ordered in categories of positive, negative, and unknown outcomes. When one category has the most ‘votes’ (results), it represents the hypothesized relation (Hedges et al., 1994). The vote-counting method does not correct for sampling and measurement error, and therefore it can draw false inferences. Using an effect-size meta-analysis will reduce these errors (Hedges and Olkin, 1980; Hunter and Schmidt 1990; Orlitzky et al., 2003). I followed the HOMA procedure to reduce these errors (Hedges & Olkin, 1985). The average correlation numbers of each study were recalculated into Fischer’s z-score, because this z-score will faster tend to normality and therefore I could use the statistics for normal distribution (Fisher, 1921; Van Rhee et al., 2015). The calculations were done with Fisher’s z and the result was recalculated into the average correlation coefficient (Borenstein et al., 2009). Figure 4.2 depicts this process graphically.
Other than the option for vote-counting or effect-size, there is another option in performing the meta-analysis: the fixed effects method or the random effects method. In this study, the random effects method is used, because I assumed high variance in the sample (Van Rhee et al., 2015). Besides, if variance was ultimately not high, the random effects model would automatically converge into the fixed effects model. 187.466 10 n/a 2.128 770 1.332 1.293 39
* Google already filters the duplicates.
Identification
# of records identified through database searching
# of additional records identified through other sources
Screening
# of records after duplicates removed *
# of records screened # of records excluded
Eligibility
# of full-text articles assessed for eligibility
# of full-text articles excluded, with reasons
Inclusion
# of studies included in quantitative synthesis
Figure 4.2 Graphical representation of calculation combined effect size. (Borenstein et al., 2009, p. 42).
The input of the selected papers is filled into a correlation table using the Meta-Essentials workbook from the Erasmus Research Institute of Management (Van Rhee et al., 2015) and the Comprehensive Meta-Analysis software. To verify validity, the outcomes of both software programs were compared. In the end the Meta-Essentials workbook was used because of its usability and the license restrictions on the Comprehensive Meta-Analysis software.
The method by Van Rhee et al. (2015) also takes into account the publication bias. As noted earlier, publication bias could mean that relevant studies exist that have not been published and are therefore not included in my analysis. To adjust the combined effect size or to measure if there is a publication bias, Van Rhee et al. (2005) has given some models. All of these models have disadvantages because of underlying assumptions (Borenstein et al., 2009). To reduce these disadvantages, I performed three of these tests and compared the outcomes. The first was the funnel plot with the ‘trim and fill’ method. This model calculates the adjusted combined effect size and shows in the graph where ‘unknown’ studies would be. The second method was Rosenthal’s fail-safe N test, or the file drawer analysis (Rosenthal, 1979). This method calculates how many studies are needed in the sample to make it nonsignificant and puts this number next to a rule of thumb. The underlying assumption is that only the studies with the highest effect-sizes are published. The calculations in this model have several drawbacks, such as not calculating the current feasible summary effect and then computing the p-value (Borenstein et al., 2009). Despite the issues, this model is well-known and common in the meta-analysis literature; I therefore include it in my publication bias analysis. The third method was Fisher’s fail-safe N test. This test is based on combined significance. The test calculates how many studies are required to bring the test to non-significance (Rothstein et al., 2005).
5. Results
In this chapter, I describe the results from the meta-analysis. Table 5.1 presents a summary of the results from the five relationships. For each meta-analysis was chosen for the random effects model and performed in the Meta-Essentials software. The meta-analysis tested the relationship between corporate social responsibility performance and analyst coverage (5.1), forecast accuracy (5.2), forecast error (5.3), information asymmetry (5.4), and the recommendations of financial analysts (5.5).
Table 5.1 Meta-analysis results for the relationship between CSR performance and financial analysts
Note: k = number of effect sizes/correlations; N = total observations; Mean = mean of effect sizes/correlations; 95% CI = 95 % confidence interval; Q(p) = Cochran’s homogeneity test statistic (p = probability of Q); I2 = percentage heterogeneity in sample; Failsafe-N test: Rosenthal’s – Fisher’s. * = Study is not homogeneous, therefore the failsafe-N test is not interpretable.
5.1 Corporate social responsibility performance and analyst coverage
Hypothesis 1 predicts that CSR performance is negatively related with analyst coverage. To investigate this relation, I used 26 studies to meta-analyse this relationship (Table 5.2). In total there were 27 variables, because the study of Manning (2017) uses two different variables of analyst coverage. The descriptions of the variables used in all of the studies are not entirely the same, but they broadly measure the same information: how many financial analysts are following the company.
With these 27 items and a total of 661.737 observations, the combined effect size is 0,1691. Concluding from this, there is a positive relation between CSR performance and analyst coverage (mean = 0,1691; 95% CI: 0,1146-0,2226). This can be seen in the forest plot (Figure 5.2); most of the studies have a positive correlation (Table 5.2).
The heterogeneity (I2) shows a high value of 99,54%, which supports the fragmentation idea behind this
study. This means that there is considerable variance in the research findings (Figure 5.3). Because of this high heterogeneity, I could not interpret the fail-safe N tests from Rosenthal and Fisher.
Figure 5.2 Forest plot of analyst coverage.
The blue dots are the effect sizes per study. The lines are the 95% confidence intervals. The green dot is the combined effect size.
Table 5.2 Studies with a relationship between CSR performance and analyst coverage
# Study name Correlation Number of subjects CI Lower limit CI Upper limit Weight
1 Ioannou & Serafeim (2015) 0,3120 16.064 0,2980 0,3259 3,94%
2 Jo & Harjoto (2014) 0,0628 11.058 0,0442 0,0813 3,93% 3 Adhikari (2016) 0,1717 19.830 0,1582 0,1852 3,94% 4 Zhang et al. (2015) 0,2417 10.555 0,2237 0,2596 3,92% 5 Manning (2017) -0,0903 152 -0,2472 0,0712 2,42% 6 Manning (2017) 0,2847 87 0,0757 0,4697 1,86% 7 Cormier et al. (2011) -0,0200 137 -0,1886 0,1497 2,32%
11 Du et al. (2016) 0,1568 3.023 0,1218 0,1914 3,84% 12 Lopatta et al. (2016) 0,0589 31.495 0,0479 0,0699 3,95% 13 Du et al. (2016)b -0,1775 1.598 -0,2246 -0,1296 3,74% 14 Hoi et al. (2016) 0,1825 19.389 0,1689 0,1961 3,94% 15 Holbrook (2014) 0,0600 6.531 0,0358 0,0841 3,90% 16 Griffin et al. (2016) 0,1250 48.656 0,1162 0,1337 3,95% 17 Cui et al. (2018) 0,1486 12.123 0,1311 0,1660 3,93% 18 Lee (2017) 0,2850 5.578 0,2607 0,3089 3,89% 19 Jo & Harjoto (2011) 0,3500 12.527 0,3345 0,3653 3,93% 20 Harjoto & Jo (2015) 0,0900 9.259 0,0698 0,1102 3,92% 21 Jo et al. (2015) 0,2040 5.072 0,1775 0,2302 3,89% 22 Lee (2014) 0,3344 49.804 0,3266 0,3422 3,95% 23 Hawn et al. (2017) 0,1027 46.542 0,0937 0,1117 3,95%
24 Hasan & Habib (2017) 0,2075 25.417 0,1957 0,2192 3,94%
25 Gao & Zhang (2015) 0,0100 10.755 -0,0089 0,0289 3,92%
26 Zhan (2015) 0,2500 277.573 0,2465 0,2535 3,96%
27 Chollet (2018) 0,0725 23.194 0,0597 0,0853 3,94%
28 Total 0.1691 661.737 0.1146 0.2226 100%
CI = Confidence interval
5.2 Corporate social responsibility performance and forecast accuracy
Supporting hypothesis 2, I found a positive relationship between CSR performance and forecast accuracy. The combined effect size is 0,009 and the 95% confidence interval is between -0,0021 and 0,0202 (Figure 5.5 and Table 5.6). This meta-analysis includes only four studies, a minimum according to Borenstein et al. (2017). However, because the included studies are homogenous (there is little
Figure 5.3 Funnel plot of analyst coverage: studies that exhibit the variance in the sample.
already makes sense with two studies (Borenstein et al., 2009). In total 70.291 observations are in this sample (Table 5.5). Figure 5.5 demonstrates that all included studies have a positive effect size. The study from Cormier et al. (2015) has a weight of only 0,24% because of the low number of observations in the study compared to other studies (Table 5.5).
As said before, in contrast to the first relationship between CSR performance and analyst coverage is this relationship homogenous. Therefore, there is low variance between these studies (Figure 5.6). Due to this homogenous sample, there is the possibility to see if there is a publication bias. According to Van Rhee et al. (2015, p.19) is calculating a publication bias only possible with a homogenous sample, and even then it “should be interpreted with much caution”. The first way to determine if there is possible publication bias is through a funnel plot (Duval & Tweedie, 2000). Using the trim and fill method from Duval and Tweedie (2000), the adjusted combined effect size is 0,0076, and the 95% confidence interval is between -0,0033 and 0,0185 after putting in two extra studies on the left side of the mean. There are differences with the observed funnel plot. The adjusted combined effect size is smaller (0,0076 < 0,009), and the 95% confidence interval is larger (-0,033 – 0,0185 < -0,0021 – 0,0202). Thus, I am 95% confident that the effect size of the studies is in a broader range between -0,033 and 0,0185 in the adjusted situation. Figure 5.4 demonstrates that the adjusted combined effect size is lower than the observed combined effect size and depicts where the imputed ‘missing’ studies are.
A second way to determine publication bias is the fail-safe N test. This test does not provide an adjusted combined effect size, but could be helpful in discovering a publication bias. According to Rosenthal’s fail-safe N test, nine missing studies with a z-value of zero are required to make the overall sample insignificant (Table 5.3). Additionally, Rosenthal gives a rule of thumb, the ad-hoc rule: when the outcome of the fail-safe N test is below 5k + 10, then there is a chance that there is a publication bias
Table 5.3 Outcome of Rosenthal's fail-safe N test
According to the second fail-safe N test from Fisher (Fisher, 1921), 21 missing studies added with a p-value of 0,5 are required to make the sample insignificant (Table 5.4). The Fisher fail-safe N test has no rule of thumb to conclude whether there is a publication bias. However, 21 studies is a relatively small number. The outcome of the Fisher fail-safe N test is in line with the results of the funnel plot and Rosenthal’s fail-safe N test. I can therefore say that there is a possible publication bias
Figure 5.5 Forest plot of forecast accuracy.
The blue dots are the effect sizes per study. The lines are the 95% confidence intervals. The green dot is the combined effect size.
Table 5.5 Studies with a relationship between CSR performance and forecast accuracy
# Study name Correlation Number of subjects CI Lower limit CI Upper limit Weight 1 Cormier et al. (2015) 0,0950 172 -0,0565 0,2422 0,24% 2 Holbrook (2014) 0,0219 6.531 -0,0024 0,0461 9,21% 3 Griffin et al. (2016) 0,0067 48.656 -0,0022 0,0156 68,61%
4 Lin & Chiang (2016) 0,0100 15.562 -0,0057 0,0257 21,94%
5 Total 0,0090 70.921 -0,0021 0,0202 100% CI = Confidence interval Fa i l s a fe‐N 21 p(Chi‐square test) 0,007 Fisher
5.3 Corporate social responsibility performance and forecast error
In my third hypothesis, I argued, based on legitimacy theory, that there is a negative relationship between CSR performance and forecast error. The results of the meta-analysis, however, reveal a positive relationship. The combined effect size is 0,0094. The left side of the 95% confidence interval range is negative: -0,071; the right side is + 0,09. To investigate this relation, 11 studies and 109.103 observations were used (Table 5.6). Despite the positive combined effect size, Figure 5.7 and Table 5.6 show that nine of the studies report a negative correlation. So, if I have used the vote-counting method, this will result in a negative correlation. The 11 studies have a relatively high variance of 99,22%, making a publication analysis impossible (Figure 5.7).
Figure 5.7 Forest plot of forecast error
The blue dots are the effect sizes per study. The lines are the 95% confidence intervals. The green dot is
Figure 5.6 Funnel plot of forecast accuracy: studies depicting the variance in the sample.
Table 5.6 Studies with the relationship between CSR performance and forecast error
# Study name Correlation Number of subjects CI Lower limit CI Upper limit Weight
1 Ioannou & Serafeim (2015) -0,0150 16.064 -0,0305 0,0005 9,37%
2 Balabat et al. (2012) -0,1972 208 -0,3253 -0,0620 6,95% 3 El Ghoul et al. (2011) 0,2900 12.915 0,2741 0,3057 9,36% 4 Lopatta et al. (2016) -0,0267 31.495 -0,0377 -0,0157 9,39% 5 Holbrook (2014) 0,1156 6.531 0,0916 0,1395 9,31% 6 Schulz (2017) -0,0380 2.287 -0,0789 0,0030 9,12% 7 Lee (2017) -0,0550 5.578 -0,0811 -0,0288 9,29% 8 Harjoto & Jo (2015) -0,0233 9.259 -0,0437 -0,0029 9,34% 9 Bouslah et al. (2011) 0,0481 4.132 0,0176 0,0785 9,25%
10 Lin & Chiang (2016) -0,0363 15.562 -0,0520 -0,0206 9,37%
11 Jo et al. (2015) -0,0200 5.072 -0,0475 0,0075 9,28%
12 Total 0,0094 109.103 -0,0715 0,0901 100%
CI = Confidence interval
5.4 Corporate social responsibility performance and information asymmetry
Hypothesis 4 predicts a negative relationship between CSR performance and information asymmetry. This relation is confirmed by the outcomes of the meta-analysis performed on six studies and 33.125 observations (Table 5.7). The combined effect size is -0,0651 with a wide 95% confidence interval between -0,1888 and 0,0607. Figure 5.9 graphically depicts three negative effect sizes and three positive effect sizes. This relationship exhibits in Figure 5.10 considerable variance in the sample with a heterogeneity level of 98,1%, making a publication bias impossible.
Figure 5.8 Funnel plot of forecast error: studies depicting the variance in the sample.
Figure 5.9 Forest plot of information asymmetry.
The blue dots are the effect sizes per study. The lines are the 95% confidence intervals. The green dot is the combined effect size.
Table 5.7 Studies with a relationship between CSR performance and information asymmetry
# Study name Correlation Number of subjects CI Lower limit CI Upper limit Weight 1 Cormier et al. (2011) -0,0100 137 -0,1789 0,1595 13,05%
2 Diebecker & Sommer (2017) -0,2226 2.999 -0,2564 -0,1883 19,75%
3 Cho et al. (2013) -0,1603 17.555 -0,1747 -0,1459 20,15%
4 De Villiers & Van Staden (2011) 0,0075 120 -0,1738 0,1883 12,41%
5 Clarkson et al. (2007) 0,0750 191 -0,0686 0,2156 14,53%
6 Cui et al. (2016) 0,0079 12.123 -0,0099 0,0257 20,11%
7 Total -0,0651 33.125 -0,1888 0,0607 100%
CI = Confidence interval
Figure 5.10 Funnel plot of information asymmetry: studies depicting the variance in the sample.
5.5 Corporate social responsibility performance and the recommendations of financial analysts
The last hypothesis, developed on stakeholder theory, posits a positive relationship between the recommendations of financial analysts and CSR performance. Performed on six studies and 64.661 observations, I find that there is indeed a positive relation between both variables. As can be seen in Figure 5.11, the combined effect size is 0,0013, and the 95% confidence interval is between -0,046 and 0,0485. This sample shows also a high percentage of heterogeneity, making a publication bias impossible (93,47%, Figure 5.12).The blue dots are the effect sizes per study. The lines are the 95% confidence intervals. The green dot is the combined effect size.
Table 5.8 Studies with a relationship between CSR performance and recommendations of financial analysts
# Study name Correlation Number of subjects CI Lower limit CI Upper limit Weight
1 Ioannou & Serafeim (2015) -0,0420 16.064 -0,0574 -0,0266 21,71%
2 El Ghoul et al. (2011) 0,0400 12.915 0,0228 0,0572 21,50%
3 Chang et al. (2014) 0,0503 546 -0,0340 0,1338 10,17%
4 Alazzani & Fitri (2012) 0,1010 114 -0,0865 0,2816 3,24%
5 Lee et al. (2016) 0,0102 11.828 -0,0079 0,0282 21,41%
6 Chollet (2018) -0,0400 23.194 -0,0528 -0,0271 21,98%
7 Total 0,0013 64.661 -0,0460 0,0485 100%
CI = Confidence interval
6. Discussion and conclusion
This study contributes to the literature of corporate social responsibility in five ways by investigating the relationship between financial analyst properties and CSR performance through quantitively aggregating existing empirical studies. As confirmed in the heterogeneity analysis there is variance among the sample studies. Suggesting that the CSR literature is fragmented and that a meta-analysis could be helpful in finding more clarity. This resulted in five findings.
First, stakeholder theory gave indications that analyst recommendations would be positively related with CSR performance. This is confirmed by the results from the meta-analysis and is in line with the findings from Ioannou and Serafeim (2015). According to stakeholder theory, stakeholders, such as financial analysts, are important. Cooperating with them, will lead to better financial performance and therefore more favourable recommendations. In a qualitative study, Luo et al. (2015) have found that analysts pay more attention to CSR behaviour in a company and that investors generally follow the advice of these analysts. Analysts are therefore an important stakeholder around the company. Moreover, the analysis suggests that financial analysts are optimistic about social responsible behaviour within firms.
Second, the findings provide support for the negative relation between information asymmetry and CSR performance. This is in line with recent literature based on the signaling theory about CSR and financial analysts. Signaling literature stated that information asymmetry decreases when companies send signals towards stakeholders, because of the better information environment created (Cui et al., 2018). Thus, this meta-analysis shows evidence for the role of CSR performance in reducing the information asymmetry.
Third, the study refines the understanding of the relationship between forecast accuracy and CSR performance. My study confirms the legitimacy theory by finding that there is a positive relation. By acting in a social responsible way, companies try to get legitimacy by their stakeholders among which financial analysts (Cormier & Magnan, 2015). To get this, they provide more information to financial analysts. As a consequence, the financial analysts make better forecasts. In this particular relationship there may exist a publication bias. However, according to the funnel plot analysis, the results will not change into a negative relationship. All taken together, evidence is found that there is a positive relationship between forecast accuracy and CSR performance.
Fourth and related to the third contribution, I considered whether CSR performance could reduce the forecast error. Findings from the meta-analysis do not match the arguments that the forecast error will decrease in the legitimacy theory (Herrmann & Thomas, 2005). This is interesting, because forecast error is the opposite of forecast accuracy and my meta-analysis found that forecast accuracy will increase. I would therefore expect that forecast error will decrease. A possible reason for the positive relation is that companies want to disguise or perk up their CSR performance, which is called Greenwashing (Du, 2015). As a consequence, financial analysts are working with bad information (garbage in = garbage out). This is also noted by Hooghiemstra (2000). He used the study from Lindblom (1994), to show how companies could communicate to external pressures about their social and environmental behaviour without changing their behaviour. Future research should be undertaken to explore why there is a positive relation between CSR performance and forecast accuracy and also a positive relation between CSR performance and forecast error. As mentioned, greenwashing could be a
