The differing corporate social responsibility – firm value
relationship across industries
Bachelor Thesis 30-07-2013 Alwin Fafieanie 10080163
Abstract
This thesis tests if different Corporate social responsibility (CSR) – firm value relationship exist across industries. All literature up to now has modeled this relationship to be the same across all industries. Financial data from S&P 500 companies is regressed on KLD CSR ratings over a period of 21 years using fixed effects OLS estimations. Controlling for Advertising intensity, firm size and R&D intensity, it is found that different CSR – firm value relationships exist across industries. This result is robust to using a different CSR measures and is also found when testing the model using First difference estimation in stead of fixed effects. Specific relations between CSR and firm value are however not robust to testing them using first differences or using a different measure of CSR. This leads to the conclusion that different CSR – firm value relationships exist from one industry to another but that further research is needed to determine these exactly.
Table of content
1. Introduction 1.
2. Literature review 3.
2.1 Definition of CSR 3. 2.2 The relationship between CSR and firm value 4. 2.3 Differing relationships across industries 6. 3. Data and regression 7. 3.1 Sample consutruction 7.
3.2 Dependend variable: A measure of firm value 7. 3.3 Independend variable: A measure of CSR 8. 3.4 Dummy variables: industries 9.
3.5 Control variables 10.
3.6 Descriptive statistics 11.
4. Method 15.
4.1 Comparison to prior research 15.
4.2 The differing CSR – firm value relationship across industries 16. 4.3 Robustness 17.
4.4 Hypothesis 17.
5. Results 18.
5.1 Comparison to prior research 18.
5.2 The differing CSR – firm value relationship across industries 18. 5.3 Robustness 21.
6. Conclusion 25.
1. Introduction
Corporate social responsibility (CSR) plays a very important role in business today. Many companies pursue CSR efforts and even issue annual CSR reports to showcase these efforts (Servaes and Tomayo, 2012). There is a wide variety of companies participating in CSR efforts, from food & beverage industries like Unilever and Starbucks, financial services as Triodos bank and KPMG to chemistry as DSM and Akzo Nobel. It is striking that so many companies implement CSR policies nowadays when there is no clear picture of their payoffs. Milton Friedman (1970) even said pursuing CSR policies was not in the best interest of companies nor it shareholders. He pointed out that taking money from, or taxing, shareholders and redistributing it to fight problems elsewhere was not within the jurisdiction of managers, and should be left to the government. But could investing in CSR actually be in the interest of the companies’ shareholders?
Together with the growing amount of CSR efforts there has also been an increase in the amount of academic research into the topic. Margolis and Walsh (2003) present a list of 127 researches in to the CSR – firm value relationship in 2003, this number has grown since. The fascinating thing about the outcomes of all this research is that no definitive conclusions can be drawn; both positive and negative relationships between the level of CSR and companies’ financial performance have been found (Margolis and Walsh, 2003, Servaes and Tomayo, 2012, Tsoutsoura, 2004, Aguinis and Glavas, 2012). Despite the overall consensus on a small positive correlation between CSR and firm value there is still much debate in the CSR literature. Most of the positive relations found stem from models in which CSR has been homogenusly linked to firm value; a misspecified model according to stakeholder theory. Stakeholder theory says that CSR influences stakeholders who in their turn influence firm value. This results in intricate relationships between CSR and firm value, as the relationship can be influenced by many different variables. Many of these possible variables have not been uncovered yet; let alone empirically tested.
In all previous literature the CSR – firm value relationship has been modeled to be the same across all industries. Grunig (1979) showed that people have differing CSR preferences, it seems like an awkward assumption therefore that the average CSR preferences from one industry to another are equal. People who value CSR heavily might
steer clear of certain industries and concentrate in others. For instance: When looking for an employer would candidates in the banking industry appreciate CSR activities by their future employers differently than, for example, teachers? It has been empirically shown by Servaes and Tomayo (2012) that companies’ previous CSR reputations have large influence on the effects their current CSR policies have on firm value. Could it be that not only the CSR reputation of the company itself influences the CSR – firm value relationship, but that the CSR reputation of the whole industry has an influence? To uncover if there are differing relationships between CSR and firm value across industries this thesis tests empirically if different relations exist across ten industries. This is tested using two different models; one assuming a heterogeneous relation between CSR and firm value and one assuming a heterogeneous relation. To make sure the findings are robust two different measures of CSR and two different estimation methods are used. The models are tested using data of S&P 500 companies over a period of 21 years, from 1991 to 2011. As a measure of CSR efforts the KLD stats database is used. Tobin’s Q is used as a measure for firm value.
In the second chapter a better overview of the literature and indications why the CSR – firm value relationship should differ are given. In the third chapter the data and variable construction is described. In the fourth chapter the method of research is described in detail. Chapter five presents the results. Chapter six concludes this thesis.
2. Literature review 2.1 Definition of CSR
When writing about the relationship between corporate social responsibility (CSR) and firm value, first of all it is crucial to know exactly which behavior constitutes CSR. Despite of all the research that has been done into the topic there is still no consensus on a definition of CSR. In this thesis CSR will be defined as is done by the World Business Council for Sustainable Development: “CSR is the commitment of a business to contribute to sustainable economic development, working with employees, their families, the local community and society at large to improve their quality of life” as has been done in previous research (Servaes and Tomayo, 2012). This definition of CSR is a broad one, which encompasses often cited CSR concepts such as the community, human rights, the environment, diversity and employee relationships. This definition encompasses all behavior that strokes with the triple bottom line concept, initiated by Aguinis (2011) and used by others (Rupp, Wiliams &Aguilera, 2010; Rupp, 2011). In this concept a firm has three main goals it should keep in mind when operating; economic, social and environmental performance. The triple bottom line concept is widely used in business, where it is more practically know as the three P’s; People, Planet and Profit. Another model this definition strokes with is the 4 layered CSR concept from Carroll (2008) which lays out CSR as a Maslov’s pyramid; on the bottom is the CSR required by law and at the top are the more altruistic forms of CSR. This model again has much in common with the ‘virtue matrix’ as developed by Martin (2002) in which CSR is split up in four quadrants; the bottom quadrants contain all CSR activities in that are directly related to higher shareholder value. This can be actions such as conforming to laws or business standards. The upper two quadrants include actions that are not directly required by law nor directly related to creating value for the companies shareholders. One important thing stands out in all models; a clear distinction is made between CSR behavior required by law and CSR behavior not required by law. The latter kind is clearly the most interesting in this thesis, as this is the kind of CSR that can vary amongst companies thereby destroying or crating value for its shareholders.
2.2 The relationship between CSR and firm value
Aguinis and Glavas (2012) present a meta-analysis of the CSR literature within economics over the period 1970 to 2011 They find 588 papers and 102 book chapters covering the subject. 47% of these are empirical while 53% are theoretical. 43% has been written after 2005, showing the recent uptick in interest for CSR. They find that 90% of the papers focus on CSR at the institutional or organizational level.
Margolish and Walsh (2003) present a meta-analysis of the empirical CSR – firm value literature between 1971 and 2001. They find 127 papers in total of which 109 use firm value as a dependent variable of CSR. Of these 109 papers, 54 uncover positive correlations between CSR and firm value, while only 7 uncover negative relationships. 28 papers find only insignificant correlations and the remaining 20 papers find mixed results. Similar findings are presented by Peloza (2009), who reviews 128 papers; 59% uncover positive relations, 27% find mixed or neutral relations and 14% find negative relations. Orlitzky et al. (2003) review 52 papers and similarly find a positive relation between CSR and firm value. An important thing to mention is that only correlations have been uncovered; there is very little evidence to support arguments on the causality between CSR and firm value. All in all there seems to be an agreement on a small positive correlation between CSR and firm value (Aguinis and Glavas, 2012). But still there is a large body of literature that is inconclusive or finds negative relationships.
Margolis and Walsh (2003) present two possible causes of the wide variety of correlations found. The first is that no agreement exists as to what behavior exactly constitutes CSR. Different definitions or operationalizations of CSR performance could lead to different numbers being used causing different outcomes. This problem is confirmed by Peloza (2009) who adds that firm financial performance is not that clear cut either. In his review of 128 papers, 36 different metrics for CSR and 39 different metrics for financial performance have been used. The second and more important cause presented by Margolis and Walsh (2003) is that many of the models in empirical papers are misspecified. A clear example of this argument is given by McWilliams and Siegel (2000) who find that all models testing the CSR – firm value relationship, while not correcting for R&D intensity, are biased. This specific example is a classic omitted variable bias problem, but there is a more deeply rooted problem. Since most research has been done on the institutional and organizational levels there is a knowledge gap in the
micro foundations of CSR (Aguinis and Glavas, 2012). Most research done so far proposes homogenus relationships between CSR and firm value; Aguinis and Glavas (2012), for instance, found this not to be the case in only 7% of their reviewed papers. This homogenus model is misspecified as it leaves out CSR influences on stakeholders within a firm, who in their turn have an influence on firm value. This model of heterogeneous influence is called stakeholder theory. An obvious example of a stakeholder is a consumer, which purchases goods from a firm. As consumers prefer goods from companies with higher CSR standards their consumption is influenced by CSR. The altered consumer behavior in turn influences firm value. Important stakeholders that have been identified in theoretical stakeholder theory are amongst others consumers, investors and employees.
Employees have been shown to be more intrinsically motivated when companies have high CSR standards. For companies this means that they have to pay their employees less to put in the same amount of effort, leading to higher firm value caused by investment in CSR (Jones, 2010). Furthermore companies with high CSR standards are more attractive to prospective employees (Turban and Greening, 1997). Companies with higher CSR efforts are able to attract investments from altruistic investors for lower rents. CSR efforts furthermore signal good quality management to potential investors who are therefore more willing to invest in companies with higher CSR standards (Fernandez-Kranz and Santalo, 2011; Vogel, 2005). Consumers are willing to pay more for products that have been produced by companies with high CSR standards. Or if they are not willing to pay a higher price, they will at least prefer the product that has been produced by a company with high CSR standards, all else being equal (Servaes and Tomayo, 2012; Sen and Battacharya, 2001; Sen and battacharya, 2004). CSR can in this light be seen as a product attribute that will raise the demand for a product (Baron, 2001).
Besides these heterogenous relationships between CSR and firm value there are also more homogenous relationships. Saving on raw materials is one important example. CSR policies concerned with saving the environment may be focused on packing products in smaller containers; thereby creating less waste. This obviously leads to cost savings for companies who no longer have to buy the packing materials, and save on shipping and storing costs as they can fit more products in the same space.
stakeholder – firm value relationships. A good example of an empirical stakeholder theory model is Servaes and Tomayo’s (2012) model that explains firm value by CSR and advertising expenditures. Advertising expenditures here proxies consumers’ awareness of companies’ CSR actions: as companies advertise more consumers are more aware of companies and their CSR efforts. This model leads Servaes and Tomayo to conclude that CSR has a more positive relation with firm value for companies that have higher advertising intensities.
2.3 Differing relationships across industries
In all empirical research up till now the relationship between CSR and firm value has, even if assumed to work via a stakeholder, been assumed to be the same across all industries. Servaes and Tomayo empirically showed that the prior CSR reputation of a company has great influence on the effects their current CSR efforts have on firm value. For companies with bad prior CSR reputations negative CSR – firm value relationships are found whereas they are positive for companies with good prior CSR reputations. This thesis builds on this insight combined with the idea that when stakeholders evaluate a firms’ CSR reputation they would not only look at the specific firm, but also at the industry in which a firm is present. It is for instance highly likely that the CO2 emissions from an energy firm are appreciated differently by its stakeholders than those of a bank.
Grunig (1979) showed that stakeholders all have different CSR expectations of firms. These preferences might lead people to take part in certain industries while avoiding others. It could for instance be very thinkable that the average teacher is more easily intrinsically motivated than the average banker. It would in this case be a weird assumption to have the same relationship between CSR and firm value as the school has to pursue less CSR efforts in order to reap the same intrinsic motivation benefits. To test if the effects of CSR on firm value indeed differ across industries, the CSR – firm value relationships is estimated in 10 different industries. As no prior research has been done into differences across industries first a homogenus CSR-firm value model is tested. Servaes and Tomayo’s (2012) however showed that advertising intensity may be of influence on the CSR-firm value relationship. To control for this effect advertising intensity is added as a regressor in two of the four models tested.
3. Variable construction and data description 3.1. Sample construction
In this thesis data from two databases has been combined into one dataset. Data from the KLD stats database is used to measure companies’ CSR efforts. The KLD stats database contains the CSR scores given by KLD Research & Analitics, inc. (KLD, 2008): an investment research firm, specializing in environmental, social and governance data. KLD has been scoring companies’ CSR performance on several strengths and weaknesses since 1991. The KLD stats database has been widely used to measure CSR performance in the academic literature (Margolish and Walsh, 2003). Data from the KLD stats database was merged with financial data and SIC ticker codes from the Compustat database: a database administered by Standard & Poor’s Capital IQ containing financial data for all listed companies in the United States. The methods used to construct the variables and regressions are deducted largely from Servaes and Tomayo (2012). Only data for the S&P 500 constituents is used. This data is used for all years it is available; from 1991 to 2011.
3.2 Dependent variable: A measure of firm value
As a measure of financial performance Tobin’s Q is used. Tobin’s Q is the market value of a firm divided by the replacement value of the assets (Breinard and Tobin, 1968). It was introduced in 1968 and has since been widely used in academic literature as a financial performance measure. The advantage of using Tobin’s Q over performance measures such as sales or profitability is that Tobin’s Q is based on the market value. The market value of a company is equal to the present value of all expected future cash flows. This makes Tobin’s Q a long-term financial performance measure as both present and future cash flows are incorporated. Furthermore the market value has been adjusted for risk as the future cash flows are discounted at the required rate of return, making it a more reliable measure as it is less prone to short term shocks (Breinard and Tobin, 1968). As CSR is expected to have long run financial benefits taking a short-term measure of performance would fail to capture the full benefits of CSR (Waddock and Graves, 1997). Tobin’s Q is preferred over market value as it is less prone to shocks from selling new shares to the market; and thereby rising the market value. Tobin’s Q only rises if companies actually outperform markets by rising their market value faste than the
replacement value of their assets.
To prevent problems with outliers Servaes and Tomayo (2012) are followed by winsorising the data at the 99th and 1st percentile. Winsorising is a technique to deal with outliers by setting the lowest (highest) 1% observations equal to the 1st (99th) percentile.
3.3 Independent variable: A measure of CSR
KLD tracks the CSR performance of companies in seven categories: community, diversity, employee relations, environment, human rights, product and corporate governance. Within each category KLD scores the performance of companies on several CSR strengths and concerns. In total there are approximately 80 strengths and concerns although this number has changed over time. In 1995 there were for instance six strengths and three concerns within the community category. In 2001 this number had changed to seven strengths and four concerns. Once a year companies are rated on each potential strength and weakness; they can be rated negative or neutral for each concern and positive or neutral for each strength. In 1995 a company could therefore have a maximum of six positive and three negative ratings in the community category. Besides these qualitative categories KLD tracks the involvement of companies in six controversial industries causing CSR concerns; alcohol, gambling, firearms, military, nuclear and tobacco. For involvement in each of these industries a company can be rated negative; a positive rating in these industries is not possible.
This thesis employs the same measure of CSR as Servaes and Tomayo (2012). Corporate governance is concerned with the degree to which shareholders can exert power over the management of a company. Servaes and Tomayo (2012) argue that this is not related to the social objectives of a company and should therefore not be part of a CSR measure. Aguinis and Glavas (2012) argue that better corporate governance lowers investors’ risk of getting a bad return on their investment; thereby rising the market value of a company with good corporate governance. As the market value is already taken into account this would lead to double counting the effects of corporate governance. Corporate governance is therefore excluded from the measure of CSR. A higher CSR score in the product category can be achieved by manufacturing products of higher quality or products that are more innovative than competitors’ products (KLD, 2008). Servaes and Tomayo (2012) argue that these choices don’t have to be related to the social
objectives of companies, the product category is therefore excluded from the measure of CSR. The same argument goes for the industry categories, which are likewise not included in the measure of CSR.
This narrow measure of CSR may however not encompass all CSR activities, as companies may consider CSR arguments when choosing in which industries to participate or what products to produce. To make sure that the results are robust to taking these possible CSR actions into consideration a second, wide, measure of CSR is made which includes the product and industry categories as well.
As the total number of strengths and concerns has changed over time it is impossible to simply compare the number of positive and negative scores over different years. To make comparison possible the method described by Servaes and Tomayo (2012) is followed. For each year the number of positive ratings within each category is divided by the total number of possible strengths likewise the number of negative ratings is divided by the total possible number of concerns for each year. This leads to two scores that range from 0 to 1 for each category, one measuring the concerns and one measuring the strengths. To create an indicator of CSR for each category the concern score is deducted from the strength score leading to a CSR score that ranges from -1 to +1. The six industry concern categories are added up and then divided by six (the total number of industry concerns); creating a industry concern category score that can range from -1 to 0 (there are no positive industry scores).
To compute the narrow measure of CSR the scores for the community, diversity, employee relations, diversity and human rights categories are added up; creating a CSR measure that ranges from -5 to +5. The broad CSR measure is constructed in the same way; the scores for the community, diversity, employee relations, diversity, human rights, product and industry categories are added up. This leads to a broad measure of CSR that ranges from -7 to +6.
3.4 Dummy variables: industries
The industries used are those specified by the Standardized Industrial Classification (SIC) codes at industry level; the code for each company has been downloaded from the Compustat database. The codes are made into ten dummies are for each company, being 1 if a company belongs to a specific industry and 0 if it does not. The 10 industries at SIC
industry level are: Energy, Materials, Industrials, Consumer Discretionary, Consumer Staples, Health Care, Financials, Information Technology, Telecommunication Services and Utilities.
3.5 Control variables
To avoided missing variable bias three control variables are added. Larger firms tend to have more CSR activities as they are exposed to more stakeholders demanding firms to be active in CSR (Russo and Fouts, 1997). Furthermore larger firms tend to have lower Tobin’s Q ratios as they have less investment opportunities. To control for these effects a proxy for the size of a firm is added to the regression; the natural logarithm of the assets. Investment in R&D has been shown to be an important determinant of financial performance while it is thought of to be closely related to investment in CSR (McWilliams and Siegel, 2000). Controlling for R&D spending is therefore crucial when examining the relationship between CSR and firm value. To control for investment in R&D, R&D intensity is added as a control variable. R&D intensity is defined as R&D spending over sales. As R&D spending is not required by the SEC to be disclosed R&D spending is missing for 2703 observations. These observations are dropped from the dataset as it is not sure how high their actual R&D spending has been.
Servaes and Tomayo (2012) empirically showed that advertising intensity is of influence on the relationship between CSR and firm value. As the average advertising intensity varies from industry to industry correcting for the advertising intensity is especially important in this thesis. Advertising spending is controlled for by adding the variable advertising intensity, which is defined as advertising spending over sales. Advertising spending is not required to be disclosed by the SEC and is therefore missing for a large part of the observations. As it is not sure how high the advertising spending of these companies all 4435 companies with missing observations are dropped. In order to avoid problems with outliers both the variable for R&D intensity and advertising intensity are winsorized at the 1st and 99th percentile; once again following Servaes and Tomayo (2012).
3.6 Descriptive statistics
After merging the two databases and removing observations for which data was missing there is a total of 1981 observations over 21 years. Due to changing ticker codes which were updated in the Compustat database but not in the KLD stats database the merger of the databases was less successful for earlier years. For 1991 there are only 70 companies in the database; this number grows steadily to 141 for the year 2011. There are 258 unique companies, which on average stay 7.7 years in the dataset.
The observations are not evenly distributed over the ten different industries; there are 12 observations for the energy industry while there are 575 observations for the consumer discretionary industry. There are no observations for the utilities industry; none of the companies has disclosed both R&D and marketing spending data. The average company in the sample has assets of 5645 million U.S. dollar; a Tobin’s Q of 4.611; a narrow measure of CSR is 0,255; a wide CSR measure of 0,287; an advertising intensity of 3.37% and a R&D intensity of 6.08%. Please see table 1 on the next page for the exact amount of observations and averages for each industry.
Table 2 on the next page presents further descriptive statistics on all of the variables used in the regression analysis. As models with firm and time fixed effects are used, all time and firm invariant data is dropped from the regression. If firms do not alter their levels of CSR significantly over time this would lead to all CSR data being lost and the models having little explanatory power. To make sure this is not the case the variation of the CSR measures within and between firms is presented in table 2 as well. The between firm variation for the narrow measure of CSR is .447 and the within firm variation is .338. This shows that some of the variation in the CSR measure is dropped when using fixed effects; but that the variation coefficient is still large enough to justify using models with firm and time fixed effects. The same argument goes for the wide CSR measure which has a between firm variation of .558 and a within firm variation of .444.
Table 1: averages per industry
Variable indusry
Number of observations Assets
(in million $) Tobin’s Q CSR Narrow CSR Wide
Advertising intensity (In % of sales) R&D intensity (In % of sales) Energy 12 17922 5.241 -0.983 -1.018 0.0663 0.156 Materials 80 3385 3.191 -0.131 -0.270 1.997 2.993 Industrials 150 33660 3.203 0.0534 0.140 1.575 2.571 Consumer Discetionary 574 3266 3.901 0.0951 0.1707 3.583 1.216 Consumer Staples 373 2987 6.188 0.4164 0.3850 6.726 1.097 Health Care 226 7906 5.843 0.3793 0.1694 3.823 12.830 Financials 10 1071 4.422 -0.575 -0.150 1.371 1.240 Information Technology 541 7855 4.506 0.4523 0.5984 2.159 13.647 Telecommunication Services 15 1305 6.562 0.2887 -0.0112 1.918 6.074 Utilities 0 Total 1981 5644 4.611 0.255 0.287 0.337 6.079
Table 2: summary statistic
N = 1,981
Variable Variation Mean Std. Dev. Min Max
Assets in $1000 4231243 6853547 6035 8.60e+07 Tobin Q 4.619376 3.895309 .4027708 22.39622 CSR narrow overall .0666776 .5436472 -2.202381 3.5 between .4473981 -1.892857 2.363095 within .3376682 -1.564842 2.411916 CSR wide overall .1836661 .7004223 -2.726191 3.764286 between .5576908 -1.892857 2.113095 within .4443284 -2.166334 2.986274 Advertising Intensity .0355398 .0367363 0 .1658015 R&D Intensity .0587655 .0731488 0 .341828
Table 3 presents the correlation matrix for the used variables. A large correlation (0.264) is seen between the narrow measure of CSR and firms assets, for the wide measure of CSR this correlation is lower (0.125). This correlation does not trade over to Tobin’s Q where lower correlations are seen. Very high correlations are unsurprisingly seen between the narrow and wide CSR measures. A correlation of 0.1815 between R&D intensity and the narrow measure of CSR seem to confirm the arguments of McWilliams and Siegel that R&D intensity is an important factor to control for. A similar correlation (0.1688) is observed between advertising intensity and the narrow measure of CSR; thereby confirming Sevaes and Tomayo’s argument that advertising intensity is an important control variable.
Table 3: correltion matrix
N = 1,981
asset CSR narrow CSR wide Tobin’s Q Advertising
intensity R&D intensity asset 1.0000 CSR narrow 0.2635 1.0000 CSR wide 0.1249 0.8268 1.0000 Tobin’s Q -0.0584 0.1092 0.0595 1.0000 Advertising intensity -0.0030 0.1688 0.0719 0.2110 1.0000 R&D intensity 0.0300 0.1815 0.1419 0.0671 -0.0895 1.0000
4. Method
To identify empirically if different relationships between CSR and firm value exist several regressions will be run on the panel dataset that was created. All the regression models will be estimated by OLS using all the possible observations in the dataset. To control for possible missing variable bias the models will be estimated using fixed effects regressions. By using fixed effect regressions all time invariant omitted variables are controlled for by taking firm dummies and year dummies into account in each regression. As many of the companies are in the dataset multiple times (7.7 times on average) the standard errors of these observations are likely correlated. To correct for this correlation amongst standard errors, the standard errors have been clustered per firm. Furthermore all standard errors are corrected for heteroscedasticity. Several regressions will be ran; first a comparison with the existing literature will be made. Secondly the differing relationship between industries will tried to be uncovered, and finally the robustness of the used models will be tested.
4.1 Comparing to prior research
First a regression to uncover the overall relationship between CSR and firm value will be run. In this model the relation between CSR and firm vlaue will be corrected only for size, R&D intensity and advertising intensity as proposed by McWilliams and Siegel (2000), and Servaes and Tomayo (2012). The model looks like:
Model one:
Firm valueit = CSRit + control variablesit
As mentioned earlier it was empirically shown by Servaes and Tomayo (2012) that the interaction between the advertising intensity and CSR is an important influence on the CSR-firm value relationship. To correct for this influence the interacting variable Advertising intensity * CSR narrow is added to model. This model is specified as:
Model two:
4.2 The differing CSR – firm value relationship across industries
To test if the models specified above exhibit different relationships across industries an interaction is made between the CSR and a industry dummies. The first model is now specified as:
Model three:
Firm valueit = CSRit * industry dummies + control variablesit
To statistically test if the estimates for different industries are equal to each other a F-test will be conducted. The test is specified as:
Test one:
H0: All estimates for industry * CSR are equal
H1: Not all estimates are equal
The second model controlling for Advertising intensity is tested across industries in the same way. This leads to the following model:
Model four:
Firm valueit = CSR * industry dummiesit + Advertising intensityit * CSRit * inudstry
dummiesit + control variablesit
In order to test the hypothesis that different CSR-firm value relationships exist from industry to industry two more tests will be conducted. The first test will test if the CSR-firm value relationship is different across industries, it is specified as:
Test two:
H0: All estimates for industry * CSR are equal
The second test will test if the interaction between CSR and advertising intensity differs across industries, it is specified as:
Test three:
H0: All estimates for advertising intensity * industry * CSR are equal
H1: Not all estimates are equal
4.3 robustness
To test if the estimates are correct they are compared to estimates from first difference models. Again the standard errors will be clustered and corrected for heteroscedastiscity, no fixed effects are used in the first difference regressions time. If the fixed-effect models are correctly specified the estimates should look very similar to the estimates found using the first difference analyses.
To see the influence of removing the companies with no data for R&D and advertising intensity from the dataset, the models are tested again using a different treatment of the missing observations. This time when R&D or advertising spending hasn’t been disclosed it is set to 0. Two dummy variables are added which are 1 if a company hasn’t disclosed it’s R&D or advertising spending respectively, and 0 if it has. These dummy variables are to correct for the bias created if the R&D or advertising spending of the companies for which it has been set to 0 was actually higher.
4.4 Hypothesis
It is expected that for both the homogenous and heterogeneous models differing relations between CSR and firm value are present. Servaes and Tomayo (2012) showed empirically that the CSR reputation of a company is of influence on the effects CSR has on firm value. The same could be true for the CSR reputation of a industry. Grunig (1979) showed that people have differing CSR reputations it could be that people with comparable CSR preferences concentrate in one industry and avoid others, leading to different CSR-firm value relations in industries.
5. Results
5.1 Comparison to prior research
The regression estimates of the first model are specified in table four below.
Table 4
N = 1,981
Controlled for size, R&D intensity and advertising intensity.
Firm and year fixed effects used, Robust standard errors have been clustered and are represented in parentheses
*** p<0.01, ** p<0.05, * p<0
In model one using the narrow measure of CSR the insignificant negative estimate for CSR does not agree with the overall consensus in the literature that there is a small positive correlation between CSR and firm value. It does however confirm McWilliams and Siegel’s (2000) and Servaes and Tomayo’s (2012) findings that, after controlling for size, advertising intensity and R&D intensity, CSR seems to have no significant influence on firm value. When the wide measure of CSR is used the negative estimate is comparable to narrow measure but significant.
The estimates of the model two are more significant than for the first model and a correlation between CSR and firm value is clearly found in these estimates. Both the estimates for CSR and the interacting term of CSR and advertising intensity are significant at the 1% level. The relatively large positive estimate clearly highlights the importance of the advertising intensity influence on the CSR–firm value relationship confirm the findings of Servaes and Tomayo (2012).
5.2 The differing CSR – firm value relationship across industries
The estimates of model three and four are presented in table 5 on the next page. The first column presents the estimates for model three using the narrow measure of CSR; the second column presents the estimates using the wide measure of CSR. The third column Dependent variable: Tobin’s Q Model 1 Narrow Model 1 Wide Model 2 Narrow Model 2 Wide CSR -0.327 (0.223) -0.331** (0.162) -0.797*** (0.292) -0.580*** (0.192) CSR * Advertising intensity 12.30*** (4.663) 10.15*** (3.750) R-squared 0.068 0.070 0.076 0.077
Table 5
N = 1,981
Controlled for size, R&D intensity and advertising intensity.
Firm and year fixed effects used, Robust standard errors have been clustered and are represented in parentheses *** p<0.01, ** p<0.05, * p<0.
Dependent variable: Tobin’s Q Model 3
Narrow Model 3 Wide Model 4 Narrow Model 4 Wide Energy * CSR -7.078 (5.545) -3.455 (3.661) -6.388** (3.020) -4.083** (1.638) Materials * CSR 0.624*** (0.185) 0.277*** (0.100) 0.186 (0.264) -0.440 (0.381) Industrials * CSR 0.0461 (0.816) 0.205 (0.225) -1.041 (0.697) -0.0466 (0.600) Consumer discretionary * CSR -0.393 (0.342) -0.372 (0.252) -0.503 (0.474) -0.340 (0.359) Consumer staples * CSR 0.772** (0.379) 0.362 (0.319) 0.540 (0.453) 0.172 (0.374) Healthcare * CSR -0.411 (0.613) -0.973* (0.494) -2.959** (1.249) -0.903 (0.761) Financials * CSR -0.160 (0.790) 0.256*** (0.0814) -0.500 (0.432) 1.190*** (0.126) Information Technology * CSR -1.126** (0.478) -0.873** (0.401) -1.192** (0.599) -1.106** (0.522) Telecommunication services * CSR -0.509*** (0.181) -0.251 (0.154) -1.645 (1.685) 3.343 (3.217)
Utilities * CSR Missing Missing Missing Missing
Energy * AI * CSR -950.3 (3,431) 769.8 (2,488) Materials * AI * CSR 20.32* (12.02) 44.96** (22.11) Industrials * AI * CSR 70.58* (39.79) 17.96 (32.98) Consumer discretionary * AI * CSR 2.895 (6.185) -0.990 (5.821) Consumer staples * AI * CSR 3.355 (3.843) 3.347 (3.120) Healthcare * AI * CSR 63.69** (27.63) -1.804 (17.83) Financials * AI * CSR 54.73 (56.90) -75.90*** (12.60) Information Technology * AI * CSR 2.891 (17.68) 11.00 (15.17) Telecommunication services * AI * CSR 63.57 (75.79) -207.7 (149.5)
Utilities * AI * CSR Missing Missing
presents the results for model four using the narrow measure of CSR; column four presents the results using the wide measure of CSR.
When looking at model three the most interesting is that the estimates for the interacting variables between industry and CSR are varying in both size and sign for different industries, hinting at possible differing relationships across them. Notable is that the estimate for the financials industry differs in sign between the estimates using the wide and narrow measure for CSR. Four industries have significant estimates for each model. The interacting variable for the information technology (IT) industry is significantly negative for both the narrow and wide measures of CSR at the 5% level. The estimate on the materials industry is significantly positive at the 1% level for both the narrow and wide measures of CSR. For the narrow measure of CSR the telecommunications industry yields a significantly negative estimate at the 1% level. The consumer stapels industry yields a significant positive estimate at the 5%. Using the wide measure of CSR the estimate for the financials industry is significantly positive.
The divergence in estimates across industries seems to confirm the expectations that differing relations indeed exist across industries. Test one statistically confirmed the outcomes as the 0-hypothesis was rejected with a p value of 0.0001.
In model four once again the most important finding is that the estimates for different industries vary; with some of them being significant. The differences between the models estimated using the narrow and wide measures of CSR are larger than in model three; with now three industry–CSR variables having an estimate with a different sign. This divergence could hint at misspecifications in the models, which will be discussed further in the robustness paragraph of this chapter.
The CSR estimates for the energy and IT industries are significantly negative at the 5% level for both the narrow and wide measures of CSR. The ICT industry therefore has negative estimates for both model three and four while using either the narrow or wide measure of CSR. The CSR estimate for the healthcare industry is significantly negative at the 5% level using the narrow measure of CSR. The CSR estimate for the financials industry is again positive at the 5% using the wide measure of CSR.
The estimates for the CSR-advertising intensity interacting variables are nearly all positive when using the narrow measure of CSR, but not so when using the wide measure
of CSR. The differences between using the narrow and wide measure of CSR are rather large, once again hinting at a bias in the models, which will be explored further in the robustness section of this chapter. Only the materials industry has significant (positive) estimates for both measures, at 10% and 5% significance levels for the narrow and wide measure of CSR respectively. The CSR-advertising intensity estimates for the industrials and materials are significantly positive at a 10% significance level when using the narrow measure of CSR. The estimate for the CSR-advertising intensity estimate of the financials industry is significantly negative at the 1% level when the wide measure of CSR is used. It is notable that in the financials industry the estimates differ a lot between using the narrow and wide measures of CSR; this could be due to the low amount of observations for this industry.
Test two confirmed the suspicion that different CSR-firm value relationsips are present across industries by rejecting the 0-hypothesis at the 1% level with a p-value of 0.0068. The influence of advertising intensity however does not seem to differ across industries as test three did not reject the 0-hypothesis with a p-value of 0.1061.
All in all it is difficult to identify relationships between CSR and firm value for different industries that are robust to tests using a different measure of CSR.. The results from all three tests performed confirm however that it is highly unlikely that the same CSR – firm value relationships exist from one industry to another. This leads to the conclusion that there are different CSR-firm value relationships across industries.
5.3 Robustness
The large differences in estimates found between the models tested with the closely related narrow and wide measures of CSR, hint at something not being right in these estimations. Table 6 on the next page presents the estimates generated using the first differences estimation method for model three and four. For comparison the estimates found using fixed effects are included in the tables. The first difference models presented are tested using the narrow measure of CSR only, using the wide measure of CSR gave similar results and presenting them would unnecessarily confuse the reader with too much information. Model one and two have been tested using first differences as well; the findings from these test are comparable to model three and four and therefore not shown.
Table 6
Controlled for size, R&D intensity and advertising intensity.
Firm and year fixed effects used, Robust standard errors have been clustered and are represented in parentheses *** p<0.01, ** p<0.05, * p<0.1
Dependent variable: Tobin’s Q Model 3
Fixed effects Model 3 First differences Model 4, Fixed effects Model 4, first Differences Energy * CSR -7.078 (5.545) -8.317 (7.424) -6.388** (3.020) -2.869 (2.258) Materials * CSR 0.624*** (0.185) -0.365 (0.270) 0.186 (0.264) -0.339 (0.698) Industrials * CSR 0.0461 (0.816) -0.496** (0.221) -1.041 (0.697) -0.613** (0.271) Consumer discretionary * CSR -0.393 (0.342) -0.250 (0.214) -0.503 (0.474) -0.481 (0.301) Consumer staples * CSR 0.772** (0.379) 0.219* (0.125) 0.540 (0.453) 0.0742 (0.153) Healthcare * CSR -0.411 (0.613) -0.461 (0.315) -2.959** (1.249) -1.114 (0.852) Financials * CSR -0.160 (0.790) -0.325*** (0.0788) -0.500 (0.432) -2.271*** (0.261) Information Technology * CSR -1.126** (0.478) -0.0207 (0.253) -1.192** (0.599) 0.605* (0.325) Telecommunication services * CSR -0.509*** (0.181) -0.674*** (0.109) -1.645 (1.685) -3.744 (3.220)
Utilities * CSR Missing 0 (0) Missing 0 (0)
Energy * AI * CSR -950.3 (3,431) -7,323 (5,749) Materials * AI * CSR 20.32* (12.02) -1.290 (22.73) Industrials * AI * CSR 70.58* (39.79) 6.441 (12.71) Consumer discretionary * AI * CSR 2.895 (6.185) 5.758 (5.300) Consumer staples * AI * CSR 3.355 (3.843) 2.463 (1.765) Healthcare * AI * CSR 63.69** (27.63) 15.29 (19.49) Financials * AI * CSR 54.73 (56.90) 283.5*** (51.46) Information Technology * AI * CSR 2.891 (17.68) -30.70* (16.32) Telecommunication services * AI * CSR 63.57 (75.79) 153.4 (146.2) Utilities * AI * CSR Missing 0 (0) Observations 1,981 1,690 1,981 1,690
Number of unique companies 258 237 258 237
In model three the differences between fixed effects estimation and first difference estimation are quite large. Only the telecommunications service industry is significantly negative for both estimation methods. All other industry estimates are either insignificant using one estimation method or insignificant using both. The materials and industrials industries give estimates with different signs for the different estimation methods.
In model four again large differences are found between the two estimation methods. There are no estimates that are significantly of the same sign. The estimate for the telecommunication services industry is even significantly negative using fixed effects and significantly positive (at 10% level) when using first differences.
To see if the estimates per industry are significantly different from each other when using first differences again tests one two and three are performed on the first difference estimations. All the 0-hypotheses are rejected with a p-value of 0.0001. This leads to the conclusion that when using first difference estimation there are different CSR-firm value relationships across industries.
The sizes and signs on the variables found are not completely comparable to the results presented by Servaes and Tomayo (2012). This difference in estimates is due to the deletion of the observations with missing advertising spending, instead of setting these to 0. When setting the variables to 0, much more similar estimates to Servaes and Tomayo’s (2012) estimates are found. When excluding the companies with missing advertising intensity observations the significance of the ‘Advertising intensity * CSR narrow’ variable gets remarkably higher. Thereby showing that Servaes and Tomayo’s (2012) model underestimates the effect of the crucial interacting variable ‘Advertising intensity * CSR narrow’. Setting the advertising intensities to 0 is odd as it is not certain that they actually are 0, as they are simply not required by the SEC to be disclosed (Fernandez-Kranz and Santalo, 2010). When performing the tests used to see if different relationships are present among industries the 0-hypothesis are once again rejected; thereby confirming that different relations indeed exist across industries indifferent of how the missing advertising and R&D intensities are treaded. This outcome also shows that the bias that could have been created by not randomly selecting companies to keep in the dataset has no influence on the conclusion that relationships differ across industries.
The difficulty in identifying different relationships between CSR and firm value could be due to model misspecification. It is possible that the specified models find
differing firm value relationships from industry to industry that are caused by omitted variables that are present to differing degrees in different industries. As still much is unknown about the ways through which CSR influences firm value there are still many possible variables that have been omitted from the regressions performed. It is also possible that the CSR – firm value relationship is not strictly linear or that a lagged model represents the relationship better as CSR is thought to have a long term influence on firm value. This bias should be corrected for by using a market value based performance measure, the market may be irrational and undervalue CSR efforts as so much is still unknown about the effects on firm value.
Another possible cause of the large differences found is that sequiential exogenity does not hold in the used dataset. It is required that sequential exogenity holds for OLS estimation using fixed effects to produce correct estimates. Meaning that the current period error terms may not be correlated with past or current observations. If this requirement is not met, using fixed effects leads to inaccurate estimations (Heckman and Learner, 2007). Comparing estimates from first differences to fixed effects estimations is a way of seeing if the data is meeting the sequential exogenity requirement. If the model is correctly specified large differences between the two estimation methods signals that the requirement is likely not met, this could be the case in this dataset. OLS using fixed effects may therefore not be the right statistical technique to uncover correlations between CSR and firm value in this case. It could just be a problem of the particular dataset used in this thesis. But it could also be that the CSR – firm value relationship is not testable using this kind of estimation technique. If CSR indeed has a long term influence on firm value as suggested it could be that CSR levels in certain years are correlated with error terms in later years as the CSR-firm value relation is tested from year to year. Future research could identify if sequential exogenity holds when testing the CSR firm value relationship.
No significant industry estimates are found that are robust to either using a different measure of CSR or using a different estimation method. This leads to the final conclusion that no definite conclusions can be drawn on the sings and sizes of the firm value CSR relationships in specific industries. It is however shown that different CSR firm value relationships exist. This result is robust to testing it using a different measure of CSR and using a different estimation method.
6. Conclusion
There is still much unknown about the effects CSR policies have on firm value. Given the large investments of companies in CSR however, they seem to have already decided that the relationship must be positive. The remarkable thing is that companies from all sorts of industries are investing in CSR. This thesis has tried to uncover if these investments are wise for all of them, or maybe just for some. While at the same time filling a gap in the academic literature on the CSR – firm value relationship. Never before has there been tried to assess if different CSR – firm value relationships exist across industries.
By testing two different CSR – firm value models, one for a homogenous CSR – firm value relation and one focusing on consumers as stakeholders, different correlations were indeed uncovered across industries. When a homogenus CSR – firm value relation was assumed, differing relations across industries were found. After controlling for firm advertising intensity it was again shown that different CSR - firm value relations exist across industries. This conclusion is robust to using both different measures of CSR and different estimation methods. The found correleations between CSR and firm value in specific industries were however not robust to using different CSR measures or estimation methods. Leading to the conclusion that the models used are likely biased. Causes of the bias can be; omitted variable bias, non-random sampling or the not holding of the sequential exogenity requirement. As however differing CSR-firm value relationships are found in all models it is concluded that different relationships do exist between industries. Future research could look more into the ways in which CSR works through stakeholders and how these stakeholders differ between industries.
7. Bibliography
Aguinis, H., 2011. “ Organization responsibility: doing good and doing well.” In S. Zedeck (Ed.), APA handbook of industrial and organizational psychology (vol. 3): 855-879. Washington, DC: American psychological association.
Aguinis, H. and A. Glavas, 2012. “What we know and don’t know about corporate social responsibility: a review and research agenda.” Journal of management, march 1 2012.
Baron, D. P., 2001. “Private politics, corporate social responsibility, and integrated strategy.” Journal of economics & management strategy 10: 7-45.
Breinard, C. and J. Tobin, 1964. “Pitfalls in econometric model building.” The American economic review, 2: 99 - 122.
Bhattacharya, C.B. and S. Sen, 2001. “Does doing good always lead to doing better? Consumer reactions to corporate social responsibility.” Journal of marketing research 38: 225-243.
Bhattacharya, C.B. and S. Sen, 2004. “Doing better at doing good; when, why, and how consumers respond to corporate social initiatives.” California management review 47: 9-24.
Carroll, A. B., 2008. “A history of corporate social responsibility: concepts and practices.” In A. Crane, A. McWilliams, D. Matten, J. Moon, and D. S. Siegel: The Oxford handbook of corporate social responsibility: 19-46. New York: Oxford university press.
Fernandez-Kranz, D. and J. Santalo, 2011. “When necessity becomes a virtue: the effect of product market competition on corporate social responsibility.” Journal of economics & management strategy 19: 453-487
York times magazine, September 13: 32-33
Grunig, J. E., 1979. “A new measure of public opininons on corporate social responsibility.” Acadamy of management Journal, 22: 738-764.
Heckman, J. J. and E. E. Learner, 2007. Handbook of econometrics 2: Volume 6A. Elsevier.
Jones, D. A., 2010. “ Does serving the community also serve the company? Using organizational identification and social exchange theories to understand employee responses to a volunteerism programme.” Journal of occupational and organizational psychology, 83: 857-878
KLD Stats, 2008. “Getting started with KLD Stats and rating definitions” Boston: KLD research & analytics inc. accesed 22-07-2013 at:
http://cdnete.lib.ncku.edu.tw/93cdnet/english/lib/Getting_Started_With_KLD_STATS.pdf.
Margolish, J. D. and J. P. Walsh, 2003. “Misery loves companies: rethinking social initiatives by business.” Administrive science quarterly 48: 268-305.
Martin, R. 2002. “The virtue matrix: calculating the return on corporate responsibility” Harvard business review, march 2012.
McWilliams A. and D. Siegel, 2000. “corporate social responsibility and financial performance: correlation or misspecification?” Strategic management journal, 21: 603-609.
Orlitzky, M., F. L. Schmidt, and S. L. Rynes, 2003. “Corporate social and financial performance: a meta analysis.” Organization studies, 24: 403-441.
Peloza, J., 2009. “the challenge of measuring financial impacts form investments in coporate social performance.” Journal of Management, 35: 1518-1541.
Rupp, D. E., C. A. Williams and R. V. Aguilera, 2010. “Increasing corporate social responsibility through stakeholder value internalization (and the catalyzing effect of new governance): an application of organizational justice, self-detirmination and social influence theories. In M. Schminke (Ed.), Managerial ethics: mangaging the psychology of morality: 69-88. New York: Routledge.
Rupp, D. E., 2011. “ An employee-centered model of organizational justice and social responsibility.” Organizational psychology review, 1: 72-94.
Russo, M. V. and P. A. Fouts, 1997. “A resource-based perspective on corporate environmental performance and profitability.” Academy of management journal, 40: 534-559.
Servaes, H., and A. Tomayo (2012) “The impact of corporate social responsibility on firm value: the role of customer awareness” Management science, forthcoming.
Tsoutsoura, M., 2004. “Corporate social responsibility and financial performance” Working paper series, Haas, University of California, Berkley.
Turban, D. B., and D. W. Greening (1997) “Corporate social performance and organizational attractiveness to prospective employees.” Acadamy of Management Journal, 40: 658-672.
Vogel, D., 2005. The market for virtue. Washington, D.C.: Brookings institution press.
Waddock, S. A. and S. B. Graves, 1997. “The corporate social performance-financial performance link. Strategic management journal, 18: 303-31
