1
The Effect of Corporate Social Responsibility on the ROA, ROE and
the Cost of Equity: Evidence from Europe.
by
Tim Labberton
Supervisor: Dr. H. Vrolijk Co-assessor: Prof. Dr. C. L. M. Hermes
Master: International Financial Management Master: International Business Management
2
Abstract
This study examines the relationship between Corporate Social Responsibility and ROA, ROE and the cost of equity for 535 European firms between 2004 and 2012. This study uses earnings forecasts from a cross-sectional model to estimate the implied cost of capital. The analysis of this study yields several important results. I find that firms with better CSR scores have higher ROA, ROE and exhibit cheaper equity financing, especially during the last couple of years. These findings support companies who are already engaged in CSR activities and encourage companies to engage in more CSR activities. I do not find any cross-culture variation towards the awareness of CSR between Northern and Southern European countries.
JEL Classifications: G15, G30, M14
Keywords: CSR, Return on Assets, Return on Equity, Implied Cost of Capital, European firms
I.
Introduction
3
I.I. Corporate Social Responsibility and Financial Performance
The question whether Corporate Social Responsibility (CSR) positively affects the financial performance of firms is an old, but still hotly debated topic in the current literature. One of the most all-encompassing studies is the meta-analysis conducted by Margolis et al. in 2007. The authors started searching articles which are related to CSR and financial performance by manually checking the table of contents of different top journals, using keywords in databases and consultation with colleagues. After the study selection, the authors found that there were 167 studies conducted which were related to CSR and financial performance between 1972 and 2007. The authors use Return on Assets (ROA) and Return on Equity (ROE) as a proxy to measure the financial performance of firms. Of the 167 studies which are being investigated, 58 per cent report a non-significant relationship, 27 percent a positive significant relationship and 2 per cent of the studies repot a negative significant relationship between CSR and financial performance. The empirical results of their study reveal that the overall effect of CSR on the financial performance of firms is positive, but small. Wu and Shen (2013) have conducted one of the most recent studies which investigate the relation between CSR and firm financial performance. Their study is focused on the banking industry in 22 different counties. The authors conclude their study by showing that CSR is positively related with financial performance of banks in terms of ROA and ROE. These results show that there still exists no unambiguous conclusion in the literature nowadays. My study will investigate the relationship between CSR and the financial performance in firms in Europa, where the financial performance of firms is measured as the ROA and ROE. This study focuses on Europe since most of the previous studies are focused on the United States. My study tries to improve the current knowledge of CSR by focusing on European countries.
I.II. Corporate Social Responsibility and Cost of Equity
4 cost of equity of firms. The authors call for elaboration of their research by extending it to other regions in the world (El Ghoul et al., 2011). According to Salaber (2007), the perception of investors towards corporate social responsibility is dependent on the country in which they live. Differences between countries in terms of culture and religion create different perceptions of investors towards CSR. In fact, European investors could perceive CSR in a different way as American investors would. This could lead to differences in the effect of the cost of equity between European and American firms. I would like the answer the call of El Ghoul et al. (2011) and investigate the relation between CSR behavior and the cost of equity in European firms.
The costs of equity are for several reasons very important for investors. Firstly, the cost of equity represents the required rate of return for investors (El Ghoul et al., 2011). So in fact, the cost of equity plays an important role in the decision-making of long-term investments for investors. Lower cost of equity will make investments, which were previously unprofitable due to high cost of equity financing, now profitable. This will result in more profitable projects for investors. Secondly, investors use the cost of equity as a discount rate to calculate future cash flows of firms. These cash flows are used to determine the current market value of firms.
5 European equities while American investors invest most of their portfolio in American equities. Third, my research complements the current literature that is focused on CSR and cost of capital (Sharfman and Fernando, 2008; Goss and Roberts, 2011; Dhaliwal, Li, Tsang and Yang, 2011; Cheng, Ioanno and Serafeim, 2014). At this moment, there has not been done a lot of research in this area and this study will contribute to the development of this research area. Fourth, my study show the potential benefits for investors when investing in companies that score high on CSR scores. If the results of this study show that firms which have better CSR scores have higher returns on their assets, equity and exhibit cheaper equity financing it will help investors to build a portfolio of stocks which are both socially responsible and more profitable. Fifth, this research adds to a more comprehensive understanding of the effect of CSR on several firm indicators. Sixth, a model-based method is used to estimate the cost of equity instead of an analyst-based method which previous studies used. The model-based method is better proxy in estimating the implied cost of capital for firms (Hou, Van Dijk and Zhang, 2012). The next section argues how CSR can influence the return on assets, the return on equity and the cost of equity in European firms. Section III describes the sample, the regressions and the variables which are used. Section IV will present the empirical results of the hypothesis and section V will discuss these results. Eventually, section VI will conclude this research.
II.
Literature
This section develops the hypotheses on the basis of the underlying literature. First I will develop arguments why firms with better CSR activities have higher ROA and ROE than firms which are not engaged in CSR activities. Second, I argue what the effect of perceived risk of a firm and the size of the investor base has on the cost of equity of the firm. Thereafter, I argue what the effect of locations (country-effect) and the increased awareness (time-effect) of CSR of the last years have on the cost of equity of firms.
II.I Return on Assets and Return on Equity and CSR
6 performance. CSR can have three effects on the financial performance of firms: a negative association, a neutral association and a positive association. Theorists who argue that CSR has a negative association with financial performance of firms base their arguments on the fact that firms which are engaged in CSR activities have higher costs than firms which are not engaged in CSR activities. These costs could have been avoided or should be borne by others, e.g. consumers or governments. Other scholars argue that there exist no relationship between CSR and financial performance. They claim that there are too many intervening variables between CSR and financial performance. In contrast to the previous views, some theorists developed arguments how CSR can have a positive impact on financial performance. This view is also called the ‘stakeholder perspective view’. I will follow this line of reasoning and will give arguments why I expect that firms which are more engaged into CSR activities have higher ROA and ROE. First of all, firms which are more engaged in CSR activities tend to attract better job applicants, retain valuable employees and enhance the performance/productivity of their employees (Lech, 2013). Eventually, this will lead to lower turnover, recruitment and training cost for firms. Second, Waddock and Graves (1997) argue that socially responsible firms can enhance their profitability by increasing their sales to socially conscious consumers. This higher profitability can be achieved by increasing the volume of sales or by charging a higher price per unit. Third, prior research has shown that adopting better corporate governance standards lead to more transparency in the firm. More transparency will lead to less insider trading, better shareholder rights, higher disclosure levels and more transparent compensation policies. According to the results of Lech (2013), this will increase the financial performance of firms. Fourth, firms which implement stricter environmental controls could prevent heavy fines which can cost the firm millions of dollars. These heavy fines can significantly decrease the profitability of the firm. For these reasons, I argue, ceteris paribus;
Hypothesis I: Firms which score high on corporate social responsibility have higher returns on assets when compared with firms which score low on corporate social responsibility.
7
II.II Cost of Equity and CSR
This section argues what the effect of perceived risk of a firm and the size of the investor base has on the cost of equity of the firm. Thereafter, I argue what the effect of culture and religion (country-effect) and the increased awareness (time-effect) of CSR have on the cost of equity of firms.
II.II.I Perceived risk of a firm
Previous research finds that investors perceive socially responsible firms less risky compared to firms which behave socially irresponsible (Frederick, 1995; Starks, 2009). Orlitzky and Benjamin (2001) empirically find that firms which score higher on CSR ratings have lower financial risks than firms which score low on CSR ratings. Boutin-Dufresne and Savaria (2004) research whether CSR practices could reduce the financial risk of Canadian firms. They conclude their study by showing that firms which adapt CSR practices could reduce the overall business risk of the firm. Husted (2005) argues that real options theory could use CSR practices as a risk-management tool to mitigate against future business risks. Hong and Kacperczyk (2009) study ‘sin’ stocks and their returns in the stock market. ‘Sin’ stocks are defined as the stock of firms which are involved in producing alcohol, gaming and tobacco. The authors find that these ‘sin’ stocks face greater litigation risk. In a more practical way, firms which score low on CSR ratings are more prone to future lawsuits than firms which score high on CSR ratings. For example, air polluters and manufactures of cigarettes are prone to future lawsuits which can cost them millions of dollars. Summarizing, firms which score higher on CSR rankings have lower financial, business and litigations risks. Investors perceive these firms therefore as less risky and have consequently lower cost of equity than firms which score low on CSR.
II.II.II Size of the investor base
Merton developed the capital market equilibrium model in 1987. In his model he shows that larger investor base1 will result in lower cost of equity and a higher market value for firms. Heinkel, Kraus and Zechner (2001) argue in the similar direction as Merton did. Heinkel et al. (2001) reason, in their equilibrium model, that firms which score low on CSR are held by
1
8 fewer investors since some investors avoid firms with low CSR scores. According to the authors, this smaller investor base will result in less risk diversification possibilities and consequently to higher cost of equity financing.
El Ghoul et al. (2011) empirically find in their study that firms with better CSR scores exhibit cheaper equity financing than firms with lower CSR scores. The authors argue that this is due to the fact that high CSR firms tend to have a bigger investor base due to the preferences of investors and to less information asymmetry. First, regarding to the preference of investors, Consolandi, Innocenti and Vercelli, (2009) investigate the behavior of investors towards CSR in a laboratory setting. The authors find that stocks with higher ethical standards attract 30 per cent more investors than stocks which lower ethical standards. The study of Hong and Kacperczyk (2009) on ´sin´ stocks shows that norm-constrained institutional investors include less ´sin´ stocks in their portfolio. So, these stocks, which score low on CSR, have a smaller investor base than other stocks. Furthermore, the research of Alniacik, Alniacik and Genc (2011) also find that firms with more CSR activities are associated with more investors. Second, recent research has shown that information asymmetry is higher for low CSR firms. Gelb and Strawser (2001) find that there exist a positive relationship between financial disclosure and CSR. Hong and Kacperczyk (2009) argue that ‘sin’ stocks receive less attention from analysts and media. Due to this lack of attention, investors are less interested in these stocks. Additionally, Dahliwal et al. (2011) empirically show that high CSR firms disclose more information than firms which score low on CSR. Diamond and Verracchia (1991) find that reducing information asymmetry2 will lower the cost of equity by attracting more and larger investors. In summary, firms which score high on CSR have a bigger investor base due to preferences of investors and to less information asymmetry. This larger investor base will lead to lower cost of equity by providing more diversification possibilities for investors. Therefore, I hypothesize that:
Hypothesis III: Firms which score high on corporate social responsibility have lower cost of equity when compared with firms which score low on corporate social responsibility.
9
II.IV Role of Culture and Religion
Salaber (2007) investigates the return of sin stocks – alcohol, gambling and tobacco- in Europe. The author made a distinction between Protestant and Catholic countries in Europe. She argues that Protestant investors demand higher returns on alcohol, gambling and tobacco stocks than Catholic investors. According to Salaber (2007), this is caused due to aversion of sin stocks by Protestant investors. She categories Austria, Belgium, Czech Republic, France, Greece, Hungary, Ireland, Italy, The Netherlands, Poland, Portugal, Spain and Switzerland as Catholic countries. Denmark, Finland, Germany, Sweden and the United Kingdom are categorized as Protestant countries. Eventually, she finds that sin stocks are outperforming other stocks in Protestant countries; this was not the case in Catholic countries. Salaber (2007) argues that this evidence shows that Protestant investors are more sin averse as Catholics investors. As a consequence, Protestants investors require a higher risk premium for social irresponsible firms which cause higher cost of equity for firms. Albareda, Lozano and Ysa (2007) analyze CSR policies in the EU-15 countries. The authors come up with four different models that call for governmental action in Europe. They categorize Denmark, Finland, Sweden and The Netherlands as a group. The Anglo-Saxon countries are categorized as a group. Germany, Austria, Belgium and Luxembourg are categorizes as group three. The last group exists of the Southern European counties (France, Spain, Portugal, Greece and Italy). Steurer, Martinuzzi and Margula (2012) make exactly the same distinction between European countries when they investigate public policies on CSR in Europe.
10 behave in such a manner through exerting their influence or through proxy fights3. This result is broadly consistent with the research of Salaber (2007). Both studies find that investors from Southern European countries care less about CSR activities than investors from North and West European countries. The investors from North and West Europe value CSR activities more than investors from South Europe. This difference in awareness towards CSR is probably caused by differences in political systems, financial systems, cultural systems and education and labor systems. As argued by Heinkel et al. (2001), when a firm attracts more investors their cost of equity will fall. Consequently, the cost of equity will decline faster for North and West European firms, when they try to improve their CSR score (because of the higher awareness towards CSR in these countries, which will lead to a larger investor base and eventually lower cost of equity financing), compared to firms from Southern Europe. Therefore, I hypothesize that:
Hypothesis IV: The relationship between CSR index score and the cost of equity is more negative for North and West European than it is negative for Southern Europe firms.
II.V Social Awareness Effect (Time-Effect)
As argued earlier, CSR is becoming more and more important in the last couple of decades. Governments are promoting the use of CSR activities all over the world (Albareda et al. 2008; Matten and Moon, 2008). Governments from different European countries have initiated several programs to increase the attention towards CSR activities. A few examples of these initiatives are the creation of knowledge centers, the financial assistance for companies that are implementing CSR practices, development of international certification systems, tax incentives and campaigns to promote sustainable consumption. However, this is only a small part of the total amount of programs which have been implemented the last couple of years. Furthermore, mutual funds and individual investors apply socially responsible screens when buying new stocks to build a portfolio (Bassen et al., 2006; Kempf and Osthoff, 2007). Moreover, the Social Investment Forum calculated that ten percent of all investments are screened to some kind of criteria that assess whether investments are social responsible.
3
11 Eurosif (2014) shows that the total amount of social responsible investing have grown rapidly the last couple of years. From 2011 to 2013 it is estimated that the total amount of social responsible investing have grown between the 20 and 35 per cent. This trend is not only visible in Europe. The total amount of social responsible investing is also growing in countries like the United States, Canada and Japan. Furthermore, the research of El Ghoul et al. (2011) showed that the inverse relation4 between CSR and cost of equity is more significant in recent years. These developments show that investors invest more on the CSR-based criteria the last couple of years. Investors’ awareness about CSR has significantly risen the last decade and investors perceive CSR more important than ever before. As a result of this, the investor base has risen for companies which are more engaged in CSR activities and therefore, I hypothesize that:
Hypothesis V: There exist a stronger and a more inverse relationship between CSR index score and the cost of equity in the last couple of years.
Figure I: Conceptual model which graphically visualize all developed hypotheses.
III. Data and Methodology
Several databases have to be used to research the relation between CSR and ROA, ROE and the cost of equity. These databases are used in order to obtain data about CSR scores, the ROA/ROE, cost of equity and control variables. The ORBIS database will provide the
4 An inverse relation is defined as a relation between two variables in which an increase in the value of one variable (X) results in a decrease in the value of the other variable (Y).
CSR Rating
Return on Assets (+)
Religious and cultural effects
Time-Effect
Return on Equity (+)
12 necessary ISIN codes. These ISIN codes are necessary to obtain data from DataStream, a database which is compiled by Thomson Reuters. Firms have to fulfil several criteria in order to be included in the sample. First of all, the firm has to be active and publicly listed. In addition, firms with less than $100 million annual operating turnover are deleted to remove small firms for which no CSR data is available. This eventually led to a total sample of 2,699 firms in Europe. However, in order to answer hypothesis IV, the sample needs to be split up into two sub-samples. One sample for Western and Northern European countries and one sample for Southern European countries.
Austria, Belgium, Denmark, Finland, Germany, Ireland, Luxembourg, Netherlands, Norway, Sweden, Switzerland and the United Kingdom are classified as Western or Northern European countries. Cyprus, France, Greece, Italy, Portugal, and Spain are classified as Southern European countries. The first step is to remove firms for which no CSR data is available, this yielded in removing a total of 1803 (!) firms. After this first step, the total sample consisted out of 896 firms. The second step is to remove firms for which more than one year of CSR data was not available. Eventually, removing these missing values lead to a total sample of 535 firms for Europe. Using these 535 firms over 9 years (2004-2012) accumulates to a total of 4,462 firm observations (after removing firms with missing observations) in order to examine the relation between CSR and ROA/ROE and the cost of equity.
Table I
Sample breakdown by countries
North/West-Europe N % South-Europe N % Austria 6 1.12 Cyprus 1 0.19 Belgium 15 2.80 France 56 10.47 Denmark 17 3.18 Greece 10 1.87 Finland 16 2.99 Italy 27 5.05 Germany 42 7.85 Portugal 4 0.75 Ireland 13 2.43 Spain 29 5.42
Luxembourg 3 0.56 Total South-Europe 127 23.74
Netherlands 20 3.74
Norway 14 2.62
Sweden 37 6.92
Switzerland 34 6.36
United Kingdom 191 35.70
13 Table I shows the sample composition by countries. The table shows that the North and Western-European countries are overrepresented in this sample with a total percentage of 76.26 per cent. South-Europa yields a percentage of 23.74 of the total European sample. Furthermore, the United Kingdom provides 191 firms while Cyprus only has one firm in the sample. This difference underscores the importance to control for country differences, more on this subject later on.
Table II summarizes the sample composition by the NACE Rev.2 industry classification system, which is a code that is assigned by the European Union to particular class of economic activity5, and by years. Some industries are heavily overrepresented in the sample. Manufacturing and financial and insurance activities together represent more than half of the total sample (52.15). However, there are a few industries which are not represented in
5
For a detailed explanation and the actual NACE Rev. codes I refer you to:
http://ec.europa.eu/eurostat/ramon/nomenclatures/index.cfm?TargetUrl=LST_NOM_DTL&StrNom=NACE_REV2
Table II
Sample breakdown by industry and year
Industry N % Year N %
Agriculture, forestry and fishing 1 0.19 2004 451 10.11
Mining and quarrying 25 4.67 2005 496 11.12
Manufacturing 189 35.33 2006 503 11.27
Electricity, gas, steam and air conditioning supply 15 2.80 2007 510 11.43
Water supply; sewerage; waste management 7 1.31 2008 508 11.39
Construction 28 5.23 2009 500 11.21
Wholesale and retail trade of (motor)-vehicles 34 6.36 2010 502 11.25
Transporting and storage 20 3.74 2011 499 11.18
Accommodation and food service activities 13 2.43 2012 493 11.05
Information and communication 44 8.22
Financial and insurance activities 90 16.82 Total 4,462 100.00
Real estate activities 23 4.30
Professional, scientific and technical activities 21 3.93 Administrative and support service activities 21 3.93
Public administration and defence 0 0.00
Education 0 0.00
Human health and social work activities 1 0.19
Arts, entertainment and recreation 2 0.37
Other services activities 1 0.19
Activities of households as employers 0 0.00
Activities of extraterritorial organisations and bodies 0 0.00
14 the sample of European firms. I will, later on, explain what the effect of these industrial differences can have on the ROA, ROE and the cost of equity and how this study will control for these differences. Table II shows that there are a total of 4,462 observations between 2004 and 2012.
III.I CSR Scores
ASSET4 provides a CSR rating for individual firms and is established in 2003. This database is compiled by Thomson Reuters and covers more than 4,000 companies worldwide. Thomson Reuters is an independent firm and is the world’s leading source of intelligent information for businesses and professionals. Professional investors use ASSET4 to define and construct portfolios with responsible shares. Furthermore, corporate executives use this database to benchmark the position of their company towards other firms in the market. ASSET4 offers the most comprehensive Environmental, Social and Governmental (ESG) database available for European firms.
15 which are filled in by the experts of ASSET4. This is visually represented in Figure I. All these key performance indicators will eventually lead to an overall score for each individual firm. Appendix II show a few measurements (of the total of 1,305) how ASSET4 measures CSR and how this data is obtained.
Figure II. Indicators of CSR score in ASSET4 Database
There are some important differences between my study and the study of El Ghoul et al. (2011) in measuring CSR-index scores. KLD uses strengths and concerns for different subcategories to calculate binary CSR scores. The different subcategories consist out of: community, corporate governance, diversity, employee relations, environment, human rights, product and controversial business issues. Each of these eight subcategories is further subdivided. For every subcategory a firm can have several strengths and concerns. If the company have strength or concern in one of these subdivided categories this is indicated with a 1. If the company did not have a strength or concern in that issue, this is indicated with a 06. For example, a firm can have 4 strengths and 3 concerns in the subcategory community. All these scores lead to an overall score for each individual firm.
The measurement of CSR scores is evidently different between KLD and ASSET4. Chatterji, Durand, Levine and Touboul (2014) empirically show that KLD, a CSR rater for firms based in the United States, puts 71 per cent of its subcategories in the social issues domain. ASSET4,
6
16 the European counterpart of KLD, puts 47 per cent of the subcategories on social issues. This difference shows that KLD puts more weight on social issues than ASSET4. On the other hand, ASSET4 puts more weight on issues relating to employees as KLD does (Chatterji et al., 2014). Chen, Srinidhi, Tsang, and Yu (2012) research whether CSR affects audit fees and audit opinions. The authors empirically find, using KLD, that firms with superior CSR performance are charged lower fees by auditors. Chen et al. (2012) use the ASSET4 database as a robustness check to see if their findings yield the same results when using this database. The authors find consistent results using the ASSET4 database. This evidence shows that, despite the differences between KLD and ASSET4, the results reached to the same conclusion.
Table III reports the summary statistics (mean, minimum, 25th percentile, median, 75th, maximum and standard deviation of the average CSR score for each firm in year t. The average CSR score consist out of economic performance, environmental performance, social performance and corporate governance performance.
Table III shows the descriptive statistics of the ASSET4 data. The average CSR score of the four sub-pillars (as shown in table II) is taken to provide a clear overview of the developments in the area of CSR. As can been seen in table III; the mean rose from 59.32 in 2004 to 73.06 in 2012. This evidence shows that firms are putting more effort into CSR activities in the last couple of year. Moreover, the first quantile rose proportionally stronger than the mean of the ASSET4 scores. This shows that more and more companies commit to some kind of CSR activities.
Table III
Descriptive statistics for CSR Data
Year Mean Min Q1 Median Q3 Max STD
17
III.II Return on Assets and Return on Equity
In line with previous studies, ROA and ROE are used as a proxy of financial performance of firms. Both proxies are accounting-based measures of financial returns. ROA indicates how profitable the firm’s assets are in generating revenue, this proxy is displayed as a percentage: 𝑅𝑒𝑡𝑢𝑟𝑛 𝑜𝑛 𝐴𝑠𝑠𝑒𝑡𝑠 =𝑇𝑜𝑡𝑎𝑙 𝑎𝑠𝑠𝑒𝑡𝑠𝑁𝑒𝑡 𝐼𝑛𝑐𝑜𝑚𝑒. The ROE measures how the much profit the firm generates with the amount of money that the shareholders have invested in the firm. This proxy is also displayed as a percentage: 𝑅𝑒𝑡𝑢𝑟𝑛 𝑜𝑛 𝐸𝑞𝑢𝑖𝑡𝑦 =𝑆ℎ𝑎𝑟𝑒ℎ𝑜𝑙𝑑𝑒𝑟𝑁𝑒𝑡 𝐼𝑛𝑐𝑜𝑚𝑒′𝑠 𝐸𝑞𝑢𝑖𝑡𝑦. Both of these accounting-based measures will be retrieved from the DataStream database of Thomson Reuters.
Panel A of this table reports the summary statistics (mean, minimum, 25th percentile, median, 75th, maximum and standard deviation of the return on assets and return on equity variable between 2004-2012. Panel A represents ROA, which is measured as net income divided by total assets. Panel B represents the return on equity variable which is measured as net income divided by shareholder’s equity.
Table IV
Descriptive statistics for ROA and ROE
Panel A: ROA
Year Mean Min Q1 Median Q3 Max STD
2004 6.80 -32.11 3.29 6.07 9.23 50.41 6.91 2005 8.04 -14.74 3.80 6.91 10.54 106.82 8.17 2006 8.92 -18.01 4.86 7.77 11.86 38.13 7.01 2007 9.15 -19.21 4.54 7.79 11.85 62.35 7.91 2008 5.41 -55.89 1.75 5.41 9.63 46.19 9.19 2009 4.08 -38.72 0.82 3.62 7.28 47.23 7.70 2010 6.69 -33.03 2.81 5.43 9.11 37.61 6.52 2011 5.87 -23.39 2.09 5.41 8.47 66.95 7.09 2012 5.67 -23.13 1.67 4.71 8.58 100.83 8,39 2004-2012 6.74 -55.89 2.60 5.95 9.78 106.82 7.87 Panel B: ROE
Year Mean Min Q1 Median Q3 Max STD
18 Table IV shows that the ROA and ROE percentages fell when the global financial crises broke out in 2008. However, in 2010 the mean of ROE rose to 34.14. This rapid increase, with respect to 2009, is probably caused by the outlier of 7206.45. The influence of these outliers on the dataset will be reduced by winsorising all the variables. This will be further elaborated in the methodology section. Also noteworthy, the 25th percentile dropped much harder than the 75th percentile after the crisis broke out. Companies that were underperforming before the crisis were hit harder by that crisis than companies which performed better before the crisis broke out.
III.III Cost of Equity
Numerous scholars have tried to estimate the ex-ante cost of equity by considering current stock prices and forecasts of analysts. In the past decade several new models are developed to estimate the cost of equity. However, some of the earlier models provided poor proxies in estimating the cost of equity. Fama and French (1997) studied their previous three-factor model from 1993 and came to the conclusion that their model did not gave accurate estimations of the cost of equity. Hou, Van Dijk and Zhang (2012) argue that ‘most of the prior studies rely on ex post realized returns to measure ex ante expected returns (p.505).’ Numerous authors have tried to overcome this deficiency by introducing the concept of the implied cost of capital. Hou, Van Dijk and Zhang (2012) define the implied cost of capital (ICC) as: ‘the discount rate that the market uses to discount the expected cash flow of the firm. The greatest advantage of this new concept is that it does not rely on noisy realized returns (p.505)’. The ICC analyst-based method uses the value of stock prices and the forecasts of cash flows (estimated by analysts) to calculate cost of equity in firms.
However, there is growing criticism in the current literature which argues that the performance of analyst-based ICC as a proxy for expected returns is not fully satisfactory (Hou et al., 2012). Previous studies show that analyst-based ICC is not a trustworthy proxy for estimating the expected returns of firms. Furthermore, analysts have a habit of being overly optimistic when estimating forecasts. Another issue regarding the analyst-based ICC method is coverage. Data for estimating the ICC, according to the analyst-based method, is only available after the late 1970s and smaller firms are heavily underrepresented.
19 will follow the research of Hou et al. (2012) and use the model-based method to estimate the cost of equity for European firms. The model-based method makes it possible to estimate the ICC for smaller firms and requires fewer accounting variables. Furthermore, the time-coverage is much larger for the model-based method than for the analyst-based method. Hou et al. (2012) empirically find that their model-based method signifies a better proxy of future earnings than the analyst-based method. This study will use an extension of the cross-sectional profitability model. The cross-sectional profitability model is used to forecast earnings of individual firms instead of using the forecasts of analysts. These forecasts are needed to solve the equation, which estimate the cost of equity that is shown hereafter. In line with Hou et al. (2012) the following pooled cross-sectional regression will be used to calculate forecasts:
𝐸𝑖,𝑡+ 𝜏 = 𝛼0+ 𝛼1 𝐴𝑖,𝑡 + 𝛼2 𝐷𝑖,𝑡 + 𝛼3 𝐷𝐷𝑖,𝑡 + 𝛼4 𝐸𝑖,𝑡 + 𝛼5 𝑁𝑒𝑔𝐸𝑖,𝑡 + 𝛼6 𝐴𝐶𝑖,𝑡 + 𝜀𝑖,𝑡+ 𝜏 (I)
where 𝐸𝑖,𝑡+ 𝜏 are the earnings of firm i in year 𝑡 + 𝜏 (τ = 1 to 3 years, ), 𝐴𝑖,𝑡 are total assets
of firm i¸ 𝐷𝑖,𝑡 is the dividend payment, 𝐷𝐷𝑖,𝑡 is a dummy variable that equals 1 for firms
which pay dividend and equals 0 if firms did not pay dividends, 𝑁𝑒𝑔𝐸𝑖,𝑡 is a dummy variable
that equals 1 for firms which negative cash flows and equals 0 for firms with positive cash flows, 𝐴𝐶𝑖,𝑡 are accruals of firm I (calculated as the difference between earnings and cash
flow from operations). Following Hou et al. (2012) earnings and level variables are winsorized each year at the 1st and 99th percentile. For each firm (i) in the total sample the earnings forecasts are computed up to three years (t) into the feature by multiplying the coefficients, from the pooled cross-sectional regression, with the independent variables from equation I. The estimated forecasts are thereafter matched to the market equity of each firm (i) at the end of June of year t.
20 36.1036). Total assets are, contrary to the findings of Hou et al. (2012), not significantly related to future earnings. Forecasted earnings are in line with Hou et al. (2012) significantly positively related to dividend payments and firms with lower accruals have a tendency of having higher future earnings. The negative earning dummy variable is significant for all three horizons. The pooled cross-sectional profitability model captures a large part of the variance in future earnings. The adjusted 𝑅2 is 0.76 for one-year ahead forecast, 0.70 for two-year ahead forecast and 0.64 for the three-year ahead forecast.
Panel A of this table reports the summary statistics (mean, minimum, 25th percentile, median, 75th, maximum and standard deviation) of the variables which were used in the pooled cross-sectional regression. All variables except 𝐷𝐷𝑡 and𝑁𝐸𝐺 𝐸𝑡, are expressed in millions of dollars. Panel B of table IV presents the coefficients and
their t-statistics from the pooled cross-sectional regression estimated during 2004-2012. 𝐸𝑡+1 , 𝐸𝑡+2 and 𝐸𝑡+3
are the one, two and three-year ahead earning forecast. 𝐴𝑡 are total assets, 𝐷𝑡 is dividend payment, 𝐷𝐷𝑡 is a
dummy variable (1 for dividend payers, 0 otherwise), 𝐸𝑡 are earnings (income before extraordinary items),
𝑁𝑒𝑔𝐸𝑡 is dummy variable (1 for firms with negative earnings, 0 otherwise and 𝐴𝐶𝑡 represents accruals. ). The
value between the parentheses reports the t-value. .* Significant at 10% level, ** significant at 5% level, ***
significant at 1% level.
Succeeding and elaborating El Ghoul et al. (2011), this study will make use of the Gebhardt, Lee and Swaminathan (2001) model-based methods to estimate the ex-ante cost of equity. Gebhardt et al. (2001) propose an alternative technique to estimate the cost of capital for firms. The authors develop the following equation to estimate the implied cost of capital:
Table V
Cross-sectional earnings regression 2004-2012
Panel A
Variable Mean Min Q1 Median Q3 Max STD
At 49,542.36 124.31 1,626.55 4,976.57 20,124.01 1,250,809.43 163,266.53 Dt 364.09 0.00 20.07 76.48 261.64 5,477.67 857.04 DDt 0.88 0.00 1.00 1.00 1.00 1.00 0.32 Et 714.30 -2,470.81 48.96 175.96 577.01 10,281.05 1,709.03 NegEt 0.11 0.00 0 0.00 0.00 1.00 0.31 Act -862.78 -18,520.15 -594.42 -120.89 -7.10 3,289.51 2,780.48 Panel B
Variable Intercept At Dt DDt Et Neg Et Act Adj. R²
21 𝑀𝑡 = 𝐵𝑡+ ∑ (𝐸𝑡 [(𝑅𝑂𝐸𝑡+𝑘− 𝑅) 𝑥 𝐵𝑡+𝑘−1 ] (𝐼 + 𝑅)𝐾 + 𝐸𝑡 [(𝑅𝑂𝐸𝑡+12− 𝑅) 𝑥 𝐵𝑡+11] 𝑅 𝑥 (1 + 𝑅)11 ) 11 𝑘 =1
where 𝑀𝑡 is the total market equity in year t¸ R is the implied cost of capital for firms in year
t, 𝐵𝑡 is book value of firm in year t, 𝐸𝑡 [ ] denotes the market expectations based on the
information which is available in year t, and (𝑅𝑂𝐸𝑡+𝑘− 𝑅) 𝑥 𝐵𝑡+𝑘−1 is the residual income
in year t +k, defined as the difference between the return on book equity and the ICC multiplied by the book equity in the previous year. In line with Gebhardt et al. (2001), I estimate the expected ROE in years t +1 to t+3 using the model-based earnings forecasts.
Table VI
Descriptive statistics for Cost of Equity (GLS)
Year Mean Min Q1 Median Q3 Max STD
2004 8.3417 1.7928 5.9743 7.7779 10.0259 35.9837 3.7288 2005 8.1848 1.6736 5.7112 7.5254 10.1339 30.7429 3.5554 2006 7.6539 0.9839 5.3658 7.1607 9.4676 44.8009 3.7434 2007 7.1428 0.9470 4.8021 6.5885 8.5729 30.8707 3.5019 2008 9.1501 1.4351 6.2430 8.4451 11.2884 43.5258 4.6108 2009 10.3522 0.2572 6.9580 9.0679 12.4201 47.8557 5.5391 2010 9,9069 1.7787 6.8453 9.0428 12.0081 44.9239 4.6083 2011 9.1355 2.0765 6.0965 8.0211 10.8200 74.0340 5.7227 2012 11.1202 1.0000 7.0132 9.3048 13.0561 73.1960 7.3931 2004-2006 8.0495 0.9839 5.7092 7.4541 9.8490 44.8009 3.6874 2007-2009 8.8716 0.2572 5.8334 7.9277 10.7708 47.8557 4.8055 2010-2012 10.0496 1.0000 6.5560 8.7325 11.9034 74.0340 6.0642 2004-2012 8.9989 0.2572 6.0220 8.0540 10.7438 74.0340 5.0256
Table VI represents the summary statistics (mean, minimum, 25th percentile, median, 75th, maximum and standard deviation of the cost of equity premium estimate (GLS estimate) for each firm in year t. The GLS method to estimate the cost of equity is computed for each firm at the end of June of each year using end-of-June market prices and model-based earnings forecasts for up to three years into the future.
22 the lowest implied equity premium estimates of all the five ICC estimation methods (0.99 per cent lower than the average of the five ICC estimation methods). Concluding, the GLS cost of equity estimates equals the previous work of El Ghoul et al. (2011) and Hou et al. (2012).
III.IV Control variables
In the multivariate model several control variables are added in order to assess the relationship between CSR scores and ROA, ROE and the cost of equity. Prior studies are followed to decide which control variables needs to be included in the multivariate model (Margolis et al., 2007; El Ghoul et al., 2011). My research includes the same control variables as Margolis et al. (2007) and El Ghoul et al. (2011) did in their research.
In line with Margolis et al. (2007) and El Ghoul et al. (2001) the following control variables are included in the regression:
Size Leverage Firm Riskiness Industry-effects Book-to-Market ratio Beta Financial Development
23 firm size is positively (negative coefficient) associated with the cost of equity. Firms which are larger attract more attention from the media and reduce the information asymmetry. Prior studies find that leverage has a positive effect on return on equity. Following Waddock and Graves (1997) and Margolis et al. (2007) a proxy is taken for firm riskiness. Firms which can be seen as more risky are expected to have lower return on assets and equity. The industry-effects variable will take the differences between the industries into account. Some industries may have higher average return on assets, return on equity and cost of equity. Book-to-market ratio is positively related with cost of equity because higher book-to-market firms earn higher ex post returns (Fama and French, 1992). Sharpe (1964) find firms’ beta is positively associated with the expected return (Sharpe, 1964). In addition, firms which are active in countries with better developed financial markets are expected to exhibit cheaper equity financing than firms which are active in countries with less developed financial markets (King and Levine (1993), La Porta et al. (1997) and Wurgler (2000)).
SIZE, is measured as the natural logarithm of the total assets.
LEV, leverage is measured as the ratio of total debt to market value of the equity.
RISK, is measured as a proxy of long-term debt divided by total assets.
BTM is calculated as the value of the firm divided by the market value of the firm.
BETA, estimated by Thomson Reuters in DataStream.
The NACE industry standard classification system will be used to control for industry effects in determining the relationship between CSR and cost of equity.
As a proxy for the general level of financial development I will use the total size of the equity market (market capitalization) of a country and divide that by its Gross Domestic Product (GDP). The natural logarithm will be taken of this mathematical fraction. This will lead to the following fraction:
𝑀𝐶
𝐺𝐷𝑃= LN = (
Stock Market Capitalization
GDP (+1)) , in line with Wurgler (2000) one point is
added to this total and then the logarithm of this total is taken. The data which control for the financial development of countries is publicly available at several databases of the World Bank.
24 maximum of 6.6741. However, the mean of 0.4617 is consistent with the mean of 0.47 of El Ghoul et al. (2011). Also noteworthy is the high maximum at the leverage variable (119.0110), this resulted in a higher leverage mean in comparison with El Ghoul et al. (2011). In order to check for these outliers the all the variables will be winsorized at the 1st and 99th percentile. Nonetheless, the beta and the total assets are both in line with the previous work of El Ghoul et al. (2011).
Table VII reports in panel B the Pearson correlation coefficients between all the regression variables. First of all, higher CSR scores are associated with lower equity premiums (-0.0165). Furthermore, the table shows that the relation between the explanatory variables (leverage, firm riskiness, financial development, book-to-market ratio and beta) and the dependent variable (rGLS) are all in line with El Ghoul et al. (2011).
Table VII
Panel A. Descriptive statistics for control variables
Mean Min Q1 Median Q3 Max STD
Beta 1.0254 -0.7300 0.6700 1.0000 1.3300 3.1700 0.5312 Book-to-Market 0.4617 -0.0024 0.0077 0.3317 0.6916 6.6741 0.5826 Financial Development 0.6342 0.0840 0.4383 0.6595 0.8002 1.3416 0.2413 Firm Riskiness 0.1968 0.0000 0.0770 0.1752 0.2873 0.9362 0.1516 Leverage 1.2610 0.0000 0.1484 0.3598 0.8363 119.0110 3.9397 Total Assets 9.0175 2.5592 7.7057 8.7683 10.1625 15.1413 1.8171
Panel B. Pearson correlation coefficients between regression variables
Variable rGLS ROA ROE CSR Score Beta BTM FD FR Leverage TA
rGLS x x X X x X x x x x Return on Assets -0.2387 x X X x X x x x x Return on Equity -0.0680 0.1819 X X x X x x x x CSR Score -0.0165 -0.0307 0.0163 X x X x x x x Beta 0.0880 -0.1296 -0.0464 0.1118 x X x x x x Book-to-Market 0.3739 -0.2107 -0.0634 -0.0150 0.1977 X x x x x Financial Development -0.1397 0.1901 0.0437 0.0116 -0.0581 -0.3819 x x x x Firm Riskiness 0.0967 -0.1629 0.0091 -0.0310 -0.0113 0.0343 -0.0941 x x x Leverage 0.3297 -0.1958 -0.0398 -0.0131 0.1262 0.3454 -0.1313 0.1800 x x Total Assets 0.2046 -0.2789 -0.0442 0.4035 0.2301 0.3285 -0.2220 0.0053 0.3277 x
Panel A of table VII represents the summary statistics (mean, minimum, 25th percentile, median, 75th, maximum and standard deviation of the control variables which are used to research the relationship between ROA, ROE and the cost of equity for European firms. Panel B of table VII shows the Pearson correlation coefficients between all the regression variables. rGLS represents the estimated cost of equity premium.
25 Furthermore, the table shows that the relation between the explanatory variables (leverage, firm riskiness, financial development, book-to-market ratio and beta) and the dependent variable (rGLS) are all in line with El Ghoul et al. (2011). A striking point is that total assets move in the opposite direction as expected toward the dependent variables (cost of equity, ROA and ROA). However, total assets shows the expected relation with CSR score (0.4035) book-to-market ratio (0.3285) and leverage (0.3277). The relation between the other two dependent variables (ROA and ROE) and the explanatory variables (CSR score, firm riskiness, leverage, total assets) are almost all in the expected direction. As can been seen in panel B of table VII, leverage has a negative correlation with ROA (-0.1958) and ROE (-0.0398). Furthermore, also in line with Waddock and Graves (1997), total assets have a negative correlation with the dependent variables (-0.2798 and -0.0442). Noteworthy, the correlation coefficient between CSR score and ROA/ROE are not in the same direction. The correlation coefficient with CSR and ROA is negative (-0.0307), while the correlation coefficient with CSR and ROE is positive (0.0163). I do not find such high correlations that multicollinearity can be seen as a serious problem in upcoming regressions.7
III.V Methodology
Both a univariate analysis and a multivariate analysis will be used in order to examine the relationship between CSR and ROA, ROE and the cost of equity. Following El Ghoul et al. (2011), I first will perform a univariate test to answer the question to what extent CSR influences firms’ financial performance. In this univariate test, the cost of equity of firms with a below-median CSR score is compared against the cost of equity of firms which have score that is higher than the median CSR score. The results of this univariate test will be displayed in the section ‘’empirical results’’.
After the univariate test is performed, a multivariate regression will be performed. In this regression the ROA, ROE and the cost of equity will be regressed on the CSR score and some control variables. The multivariate model includes the control variables that are discussed before:
ROA/ROE = c(1) +c(2)*CSR score + c(3)*SIZE + c(4)*Firm Risk + c(5) Industry-effects
7 I use, following many authors, a rule of thumb for deciding whether there is multicollinearity. When r.>0.7
26 Cost of Equity = c(1) +c(2)*CSR score + c(3)*BETA +c(4)*SIZE +c(5)*BTM + c(6)*LEV + c(7)*
𝑀𝐶
𝐺𝐷𝑃 + c(8) Industry-effects
My model almost completely uses the same methodology as the research of Margolis et al. (2007) and El Ghoul et al. (2011). However, there are some important differences which cannot be ignored. First, El Ghoul el al. (2011) use a different CSR concept (KLD database) while this research use the ASSET4 database. Second, El Ghoul et al. (2011) use an analyst-based method to estimate the ICC of American firms while this research uses a model-analyst-based method to estimate the ICC of European firms. Nonetheless, the methodology and control variables are in line with the previous work of Margolis et al. (2007) and El Ghoul et al. (2011).
IV. Empirical Results
This section discusses the univariate analysis first. After the results of the univariate analysis are interpreted, this section will continue with presenting the multivariate model and the results of this model.
IV.I Univariate Model
27 equal. The outcomes of the F-test determine if a T-test with or without equal variances have to be used. 8 Table VIII Univariate Test N rGLS ROA ROE Panel A. Means 2004-2006 CSR Score > median (1) 725 0.0824 7.4139 19.2764 CSR Score < median (2) 725 0.0782 8.3003 20.4305 Difference (1)-(2) 0.0042 -0.8864 -1.1542 t-Statistic 2.2864** 2.5993*** 1.1861 2007-2009 CSR Score > median (1) 759 0.0900 6.1768 16.2802 CSR Score < median (2) 759 0.0865 6.5294 14.8830 Difference (1)-(2) 0.0035 -0.3527 1.3972 t-Statistic 1.5342 0.9532 1.2427 2010-2012 CSR Score > median (1) 747 0.0961 6.1054 15.5267 CSR Score < median (2) 747 0.1005 5.9343 12.6623 Difference (1)-(2) -0.0043 0.1711 2.8643 t-Statistic 1.7329* 0.5082 2.6240*** Panel B. Medians 2004-2006 CSR Score > median (1) 725 0.0779 6.7200 17.6500 CSR Score < median (2) 725 0.0722 7.0800 18.2200 Difference (1)-(2) 0.0056 -0.3600 -0.5700 z-Statistic 2.2880** 1.1869 1.1869 2007-2009 CSR Score > median (1) 759 0.0808 5.4100 15.3950 CSR Score < median (2) 759 0.0782 6.0000 14.2150 Difference (1)-(2) 0.0026 -0.5900 1.1800 z-Statistic 1.5352 0.9537 1.2436 2010-2012 CSR Score > median (1) 747 0.0866 5.3800 12.9700 CSR Score < median (2) 747 0.0878 4.9200 11.1800 Difference (1)-(2) -0.0012 0.4600 1.7900 z-Statistic 1.7479* 0.5126 2.6467***
Panel A of table VIII shows the mean-comparison test while panel B shows the median-comparison tests. The total sample consists out of 4,462 firm-year observations between 2004-2012. rGLS represents the implied cost capital which is obtained from the model of Gebhardt, Lee and Swaminathan. * Significant at 10% level, ** significant at 5% level, *** significant at 1% level.
The univariate results of table VIII shows mixed results regarding the effects of CSR on the cost of equity, ROA and ROE for European firms between 2004 and 2012. Firms with better
8
28 CSR ratings exhibit significant higher cost of equity financing during 2004-2006, this contradict hypothesis III. However, as can be seen in table VIII, during 2010-2012 firms with better CSR ratings exhibit significant lower cost of equity financing. These preliminary results show that that CSR is becoming more important for firms the last couple of years.
Furthermore, we can see in table VIII that CSR have mixed results on the ROA and ROE. The results are only (negatively) significant between 2004 and 2006 for ROA. However, there is a trend visible in table VIII. The difference between the mean and median decreases during 2007-2009 for the ROA and ROE. In fact, the difference between firms with high ROE and CSR scores became positive between 2010 and 2012. CSR had a positive effect on the ROA and ROE between 2010 and 2012. However, this effect was only significant at 1 per cent level for ROE. This emphasizes again that CSR is rapidly gaining attention the last couple of years in Europe. The next section will continue with the multivariate model to answer all the hypotheses which were hypothesized in section II.
IV.II Multivariate Model
29
The first assumption requires that the average value of the errors is zero for the variables.
The second assumption states that the variance of the errors is constant and finite over all values.
The third assumptions requires the all the errors are statistically independent of one another.
The fourth assumption states that there is no relationship between the error and the corresponding x
The last condition is met when 𝜀𝑖 is normally distributed (Brooks, 2008)
Three serious problems could arise when one of the assumptions of the CLRM is violated. The coefficient estimates could be biased, the standard errors could be biased and the distribution that where assumed for the test statistics are inappropriate (Brooks, 2008: p.130).
The first assumption of the model is never violated when a constant term is added in the equation.
The second assumption is also known as the assumption of homoscedasticity. There is heteroskedasticity when the errors of the variance are not constant and finite over all values. To test this assumption, White’s test is used. However, EViews 8 did not support the White test for panel data, so I had to restructure the data to test if there is heteroskedasticity in the data. The result of the White test can be seen in the Appendix IV. These results show that there exist heteroskedasticity in this dataset (the p-value of the F-statistic and the scaled explained SS are lower than 0.05). In order to overcome the violation of the second assumption heteroskedasticity-robust standard errors are used in EViews.
30 The results show that null hypothesis of no autocorrelations should be rejected and that this dataset does suffer from autocorrelation (see Appendix IV).
The fourth assumptions holds when the Xt are non-stochastic. The OLS estimator is consistent and unbiased in the presence of stochastic regressions (Brooks, 2008: p.160).
The last assumption tests the normality assumption. The Bera-Jarque test is used to see if this assumption holds. The results of all these on all individual variables are not reported. However, none of the variables were normally distributed (the p-value were not bigger than 0.05).
Summarizing, three of the five CLRM do not hold. Due to the existence of heteroskedasticity standard errors could be inappropriate and hence any inferences make could be misleading. Cluster-robust standard errors are used to overcome the heteroskedasticity in the data, also known as White periods. A lagged dependent variable is added to overcome the autocorrelation in the dataset. The results show that de Durban-Watson test statistics has increased to ± 2.00. This evidence shows that there is no more evidence of autocorrelation in the regressions. Furthermore, the violation of the normality assumption is will not affect the results due to the large dataset. All of the five CLRM assumptions have now been satisfied and OLS can be used.
31 variables, firm size and firm riskiness, have a negative significant impact effect on ROA at 1 per cent significance level in all models. The different models in panel A demonstrate that hypothesis I can be accepted. Firms which score high on corporate social responsibility have higher returns on assets when compared with firms which score low on corporate social responsibility.
Table IX
Multivariate model
CSR Score CSR 04-06 CSR 07-09 CSR 10-12 CSR EcoP CSR EnvP CSR SP CSR CGP
(1) (2) (3) (4) (5) (6) (7) (8) Panel: A ROA CSR 0.0152*** 0.0132** 0.0060 0.0179*** 0.0229*** 0.0046 0.0101** 0.0097** (3.6630) (2.3332) (0.9991) (3.0053) (7.0980) (1.1371) (2.1704) (2.4143) Size -0.5607*** -0.7213*** -0.3508*** -0.5628*** -0.6043*** -0.4735*** -0.5240*** -0.4576*** (-7.1317) (-6.3193) (-3.1674) (-5.1612) (-8.5733) (-5.9038) (-6.2464) (-6.7274) Firm Riskiness -4.0573*** -4.5899*** -5.1236*** -2.9335*** -3.6697*** -4.2400*** -4.2228*** -4.2737*** (-5.5454) (-3.7392) (-4.9595) (-2.5938) (-5.0822) (-5.7114) (-5.7934) (-5.8482) Intercept 7.0546*** 10.4435*** 5.8376*** 7.4503*** 7.1111*** 7.2546*** 7.3659*** 6.7023*** (8.6503) (7.0911) (4.8925) (6.6853) (8.8350) (8.8727) (9.0404) (7.8749)
Year effect Yes No No No Yes Yes Yes Yes
Industry effect Yes Yes Yes Yes Yes Yes Yes Yes
N 3941 940 1507 1494 3941 3941 3941 3941 Adjusted R² 0.44 0.52 0.38 0.41 0.44 0.44 0.44 0.44 Panel: A ROE CSR 0.0510*** 0.0239 0.0257 0.0687*** 0.0641*** 0.0169 0.0334*** 0.0387*** (4.2565) (1.3266) (1.3462) (3.7707) (6.4187) (1.4067) (2.7019) (3.2508) Size -1.1277*** -1.2254*** -0.5265*** -1.4128*** -1.1711*** -0.8522*** -1.0050*** -0.7992*** (-5.7558) (-4.1815) (-1.8521) (-4.6175) (-6.5797) (-4.2495) (-4.9178) (-4.6323) Firm Riskiness -3.8411 -2.2970 -1.1403*** 2.2860 -2.8484 -4.4716** -4.4278* -4.5638* (-1.4963) (-0.5560) (-2.8997) (0.5692) (-1.1247) (-1.7273) (-1.7308) (-1.7788) Intercept 3.6946** 18.7537*** -12.2240*** 14.9783*** 3.8444** 4.4664** 4.8260** 2.1755 (1.7785) (4.5689) (-3.9193) (5.0463) (1.8897) (2.1476) (2.3466) (0.9574)
Year effect Yes No No No Yes Yes Yes Yes
Industry effect Yes Yes Yes Yes Yes Yes Yes Yes
N 3941 940 1507 1494 3941 3941 3941 3941
Adjusted R² 0.40 0.33 0.37 0.41 0.40 0.40 0.40 0.40
32 Model 1 in panel B shows that CSR has a significant positive effect on ROE at 1 per cent significant level. During 2004-2006 and 2007-2009 the effect of CSR on ROE is positive but not significant. However, the last period (2010-2012) shows that CSR has a high positive effect on the ROE at one percent significance level (coefficient of 0.0687). Models 5 till 8 split the CSR activities into previously discussed categories. In line with the previous results, all (except environmental performance) have a positive significant effect on ROE during 2004-2012. Firm size has a negative impact on the ROE in all models at 1 per cent significance level. However, firm riskiness shows mixed results. This control variables is negative along all the models (except model 4), but not significant along all these models. Looking at all the models in panel B I can accept Hypothesis II and conclude that firms which score high on corporate social responsibility have higher returns on equity when compared with firms which score low on corporate social responsibility. It should be noted that this is particularly true for the last couple of years.
Table X reports the multivariate model which regress the cost of equity premium on different CSR proxies and control variables. In all 10 models, the dependent variable is the cost of equity (estimated by the estimation method of Gebhardt et al. (2001)). The numbers are all multiplied by 100. These numbers can be directly interpreted as a percentage. The explanatory variables include the four CSR proxies, six control variables and year and industry effects.
The first model (basic regression) examines the impact of CSR on the cost of equity while controlling for six control variables, year and industry effects. The model shows that CSR has a negative effect on the cost of equity for European firms. However, this result is not significant. All control variables are in the expected direction (except financial development) and significant. On basis of this model I can neither accept nor reject Hypothesis III which stated that firms which score high on corporate social responsibility have lower cost of equity when compared with firms which score low on corporate social responsibility.
33 equity channel, in South-Europe than in North Europe. Hypothesis IV must be rejected on basis of these results.
The following models examine whether there exist a stronger and a more inverse relationship between CSR score and the cost of equity in the last couple of years. I expect this relationship due to the growth and increasing awareness of corporate socially responsibility in the last couple of years. The full sample is split up into three time-periods: 2004-2006 (model 4), 2007-2009 (model 5) and 2010-2012 (model 6). During 2004-2006 (model 4) I find that the CSR coefficient is negative but statistically insignificant. One period later (model 5), the CSR coefficient became positive but is still statically insignificant. The last sub-period, 2010-2012 (model 6), show that the CSR coefficient is statistically significant at the 10 per cent level. The results of three sub-periods indicate that the inverse relation between CSR and the cost of equity is more significant in later years. Based on these results I accept Hypothesis V.
The rest of the models in table X examine the relation between the cost of equity and the sub-categories of the average CSR score. These sub-categories consist of CSR economic performance (CSR EcoP), CSR environmental performance (CSR EnvP), CSR social performance (CSR SP) and CSR corporate governance performance (CSR CGP). The coefficient of model 7 is negative statistically significant at the one per cent. Model 8 and 9 examines the effect of CSR environmental and social performance on the cost of equity. Both coefficients are negative (-0.0018 per cent and -0.0032 per cent), but both are statistically insignificant. Model 10 shows a very remarkable result. The coefficient of CSR corporate governance is positive and significant at the one per cent level.
34
Table X
Multivariate model
CSR Score North-Europe South-Europe CSR 04-06 CSR 07-09 CSR 10-12 CSR EcoP CSR EnvP CSR SP CSR CGP
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) CSR -0.0018 -0.0006 -0.0042 -0.0005 0.0051 -0.0064* -0.0639*** -0.0018 -0.0032 0.0102*** (-0.9131) (-0.2706) (-0.9029) (-0.1904) (1.4623) (-1.6889) (-3.8955) (-0.8877) (-1.5947) (4.3947) Beta 0.2873** 0.3051** -0.0262 -0.1151 -0.0915 0.8460*** 0.2784** 0.2863** 0.2950** 0.2308* (2.2222) (2.0112) (-0.1081) (-0.9193) (-0.5066) (3.6260) (2.1722) (2.2091) (2.2760) (1.7752) Size -0.0566 -0.0585 0.0986 0.0940** -0.1635*** -0.1085* -0.0143 -0.0573 -0.0434 -0.1153*** (-1.4484) (-1.2303) (1.1917) (1.7149) (-2.6037) (-1.7352) (-0.3916) (-1.4555) (-1.0890) (-3.2430) BTM 1.3261*** 1.1096*** 2.4587*** 1.4510*** 1.1357*** 1.6041*** 1.3232*** 1.3330*** 1.3233*** 1.4788*** (8.5853) (6.0544) (5.8597) (4.9692) (5.5274) (6.5944) (8.7354) (2.2091) (8.6323) (9.0877) Leverage 0.2175*** 0.2437*** 0.1587*** -0.0363 0.2389*** 0.3061*** 0.2056*** 0.2184*** 0.2176*** 0.2249*** (6.2081) (4.6109) (3.0565) (-0.6033) (4.2032) (6.2592) (5.9269) (6.2462) (6.2123) (6.4706) Financial Development 0.4983** 0.4100 -1.1141 0.4565 -1.6508*** 0.9636* 0.5434** 0.4850** 0.4869** 0.2250 (2.0765) (1.6392) (-1.5559) (1.5836) (-5.0338) (2.3807) (2.2912) (2.0326) (2.0437) (0.9366) Intercept 2.6275*** 2.2944*** 2.5040*** 0.2912 6.0516*** -0.7700 2.5916*** 2.6376*** 2.5849*** 2.4044*** (6.5315) (4.9121) (3.0115) (0.6535) (11.1335) (-1.2085) (6.4775) (6.5515) (6.3984) (5.9451)
Year effects Yes Yes Yes No No No Yes Yes Yes Yes
Industry effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
N 3941 3019 880 940 1507 1494 3941 3941 3941 3941
Adjusted R² 0.63 0.60 0.75 0.70 0.50 0.64 0.63 0.63 0.63 0.63
35 firm size has a positive coefficient (model 3 and model 4 in table X). The variables which control for financial development show mixed results. I expected a negative coefficient along all models. Only two of the ten models show a negative coefficient for financial development. Five coefficients report a positive significant coefficient.
Summarizing, the results of table X show that firms with better CSR activities exhibit lower equity financing cost in the period 2010-2012. I interpret this as evidence which suggest that the awareness of investors toward CSR practices have increased the last couple of years. The pre-argued difference between North and South-Europe is not reflected in table X. There is no evidence that the market prices CSR activities more in Northern and Western European firms than in Southern European firms. There exists no cross-culture variation in the relationship between corporate social responsibility and cost of equity financing.
