CSR DISCLOSURES: PREDISPOSITIONS AND CONSEQUENCES Charl de Villiers* The University of Waikato, New Zealand and University of Pretoria cdev@waikato.ac.nz Ana Marques Nova School of Business and Economics ana_marques@novasbe.pt January 2013 Acknowledgements: The authors appreciate the valuable feedback and suggestions provided by Dan Dhaliwal, Vic Naiker, Joana Story, Marna de Klerk, and workshop participants at the University of Waikato. Ana Marques gratefully acknowledges the financial support of Nova Forum. The authors appreciate the research assistance of Luís Araújo. *Corresponding author: Professor Charl de Villiers Department of Accounting The University of Waikato Private Bag 3105 Hamilton 3240 New Zealand Telephone: +64 21 0715 703 Email: cdev@waikato.ac.nz 0 CSR DISCLOSURES: PREDISPOSITIONS AND CONSEQUENCES Abstract Corporate social responsibility (CSR) disclosures are becoming more prevalent among firms. We investigate the predisposition of firms to disclose CSR information and find that firms are likely to disclose more in countries with: better investor protection, higher levels of democracy, more effective government services, higher quality regulations, more press freedom, and a lower commitment to environmental policies. We also find that CSR disclosures are associated with higher share prices. In addition, we find this relationship to be stronger in countries with more democracy, more government effectiveness, better regulatory quality, and more press freedom. These results imply that market participants find CSR disclosures more informative in countries where investors are in a better position to voice their concerns (through democratic mechanisms and through the media), and where opportunities to voice concerns have resulted in better regulation and more effective government implementation of regulations. Keywords: corporate social responsibility, global reporting initiative, voluntary disclosure 1 INTRODUCTION Firms increasingly disclose social and environmental information, otherwise known as Corporate Social Responsibility (CSR) information. Whereas financial disclosures are highly regulated, CSR information is mostly disclosed on a voluntary basis. Firms do this for two main reasons: 1) to conform to societal expectations and thereby ensure continued access to resources, such as capital, customer support, etc., and 2) to provide additional information that allows capital market participants to more accurately assess firms’ financial prospects and risk profiles, potentially leading to higher share prices and higher firm values. We are interested in this second reason. Voluntary disclosures in general are used by managers to pursue positive economic outcomes (Healy and Palepu [2001]), and voluntary disclosures have been shown to be associated with, among other economic measures, lower cost of capital (Francis, Nanda, and Olsson [2008]), and higher market prices (Healy, Hutton, and Palepu [1999]). Capital market participants pay specific attention to CSR disclosures as evidenced by a recent survey that shows 1) investors and analysts use CSR information, and 2) they prefer corporate disclosures as the source for such information (Radley Yeldar [2012]). Empirical evidence also links CSR disclosure with positive economic outcomes, such as reduced cost of equity capital (Dhaliwal, Li, Tsang, and Yang [2011]), and increased analyst forecast accuracy (Dhaliwal, Radhakrishnan, Tsang, and Yang [2012]). However, the prior research does not consistently report a positive relation between CSR disclosure and economic outcomes. For example, Hassel, Nilsson, and Nyquist [2005] report a negative correlation with share prices, and Gietl, Göttsche, Habisch, Roloff, and Schauer [2012] find a negative correlation with Tobin’s Q. These counterintuitive findings suggest a 2 need for further study to gain a better understanding of the conditions where CSR disclosures are linked with positive economic performance, as well as the characteristics of an appropriate measure to proxy CSR disclosures. We examine the predispositions towards, and the consequences of, CSR disclosures by the top 500 European firms during the period 2007 to 2010, using the Global Reporting Initiative (GRI) guideline’s level of CSR disclosure as our measure of CSR disclosure. We provide more information on the GRI and our measure in the next section of the paper. By way of predispositions, we identify both country level variables and firm level variables that predispose firms towards CSR disclosures. We then examine whether CSR disclosures are associated with higher share prices. Finally, we document the type of country where CSR disclosures and positive economic outcomes are more closely linked. We focus on share prices, because share prices incorporate the market’s assessment both of firms’ future cash flows and of an appropriate discount rate that reflect the risk inherent in the expected cash flows (i.e. cost of capital). Therefore, share prices provide wide-ranging market information in a single measure, incorporating more information content than cost of capital. CSR initiatives themselves and any potential benefits associated with CSR initiatives tend to be long-term in nature (De Villiers, Naiker, and Van Staden [2011]) and therefore we choose a measure (share price) that incorporates the market’s view of all future prospects. An improved understanding of the predispositions towards and consequences of CSR disclosures is of interest to capital market participants (because it can inform their investment decision making), managers (because they will be interested in any apparent advantages that emanate from voluntary CSR disclosure), regulators (because if there are opportunities for managers to act opportunistically, they may want to consider 3 implementing CSR disclosure regulation), and social and environmental activists (because our findings may assist them in forming opinions on firms’ CSR activities based on CSR disclosure). This paper represents several advances on the prior literature. Prior research has shown that managers’ incentives are shaped by the unique legal and economic conditions found in each country, which in turn, affects financial reporting practices (Ball, Robin, and Wu [2003], Burgstahler, Hail, and Leuz [2006]). There is no reason to expect CSR disclosure practices to be any different, and, in fact, some evidence is provided by the fact that more than 90% of the top 100 companies in the UK disclose CSR information, compared to only 20% in India (KPMG [2011]). Thus, in contrast with single country CSR studies like Dhaliwal et al. [2011] and Hassel et al. [2005], a multi-country approach is called for. Prior multi-country CSR studies typically use two country level measures, namely a continuous “rule of law” measure and a “code law/common law” dummy variable (e.g. Simnett, Vanstraelen, and Chua [2009], Dhaliwal et al. [2012]).1 These variables, because of their broad nature, do not facilitate an understanding of the more specific underlying mechanisms at work. We identify and use several new country level measures that provide better understandings of the multiple driving forces behind CSR disclosures, such as investor protection mechanisms, democratic institutions, press freedom, and the prioritization of environmental policy goals. We explain the causal mechanisms at work in the literature review and hypotheses development. We use these more explicatory country 1 We are aware of a working paper, Cahan, De Villiers, Jeter, Naiker, and Van Staden [2012], that use more country level disclosures in a multi country CSR study, but this study uses a different CSR disclosure measure (composed of several sub-measures), examine disclosure in a single year, and do not focus exclusively on European firms. 4 measures to document the kinds of jurisdictions (related to regulation and freedoms to voice concerns) where this link is stronger. In another advance on the prior literature, we choose a relatively homogeneous group of countries, namely European countries, in order to avoid the possibility that our results may be driven by outlier values in the country measures used. For example, additional analyses show that one of the major results of Simnett et al. [2009] (around code/common law) was entirely driven by the US firms in the sample. European countries are relatively similar in terms of factors like economic development. However, there are still differences among European countries in terms of the relative levels of investor protection, democracy, government efficiency, regulatory quality, press freedom, and emphasis on environmental policies. In this paper, we investigate how these factors influence the relationship between CSR disclosure and firm value. A final major advance on the prior literature is based on the characteristics of our CSR disclosure measure, which is superior by 1) encompassing disclosures in all media, 2) not being a simple dummy variable that ignores the actual content of CSR disclosures, but incorporating a level of disclosure, 3) providing greater variation, 4) not being based on a self-constructed disclosure index incorporating potential bias, 5) being less open to incorrect coding, and 6) being less labor intensive to collect, potentially allowing for a larger sample size.2 With our CSR disclosure measure, we are able to show that it is not 2 Firms use different disclosure media for CSR information and prior CSR studies typically focus on only one of these, e.g., annual reports (Ingram and Frazier [1980]); 10K reports (Cho and Patten [2007]); websites (Clarkson et al. [2008]); and stand-alone reports (Dhaliwal et al. [2011], Simnett, Vanstraelen, and Chua [2009]); whilst ignoring the information provided in the others. Most of these studies hand collect a disclosure score composed of several individual items disclosed in one of these media. This method entails the risks of misinterpretation and miscoding. A superior method of assessing CSR disclosures would be to use a measure that encompasses all disclosure media, without the need for hand collecting individual data items. The GRI publishes and promotes a framework and guideline for CSR disclosure. Many firms throughout the world subscribe to the GRI guideline and disclose the fact. GRI is the most popular CSR framework and is 5 only the initiation of a separate CSR report that is associated with positive market related outcomes (as shown by e.g. Dhaliwal et al. [2011]), but also that the level of CSR disclosure is associated with firm value. In terms of predisposition, we find that firms are likely to disclose more CSR information in countries with: better investor protection, higher levels of democracy, more effective government services, higher quality regulations, more press freedom, and a lower commitment to environmental policies. In addition, we find firms are likely to disclose more CSR information if they: are bigger, are more profitable, have a higher book to market ratio, have higher leverage, have older assets, have higher capital expenditure, and operate in environmentally sensitive industries. In terms of consequences, we find that higher levels of CSR disclosures are associated with higher share prices. In economic terms, ceteris paribus, the European firms in our sample have a share price that is, on average, 2.89 Euro higher for each higher GRI application level they choose to disclose at, e.g., firms using GRI guidelines and disclosing at the highest of the three GRI disclosure levels, have a share price that is on average 2.89 Euro higher than firms disclosing at the second highest GRI disclosure level. When combining country level predispositions with our consequences analysis, we find these positive share price consequences to be stronger in countries with stronger governance structures, i.e., in countries with more democracy, more government effectiveness, better regulatory quality, and more press freedom, share prices are more strongly associated with CSR disclosure levels. These results imply that market participants find CSR disclosures more informative in countries where investors are in a better position especially popular among European firms (KPMG [2008], KPMG [2011]). An additional advantage of GRI is that it allows firms to choose among three levels of compliance with GRI. 6 to voice their concerns (through democratic mechanisms and through the media), and where these opportunities to voice concerns have resulted in better regulation and more effective government implementation of regulations. LITERATURE REVIEW AND HYPOTHESES DEVELOPMENT The Global Reporting Initiative (GRI) The Global Reporting Initiative (GRI) is a non-profit organization that promotes economic, environmental and social sustainability by providing a comprehensive sustainability reporting framework in cooperation with global business, governments, civil society, labor, academic and professional institutions (GRI [2012]). According to KPMG [2011] 80% of the world’s 250 largest companies and 69% of the top 100 countries in the 34 countries their report covers adhere to GRI reporting guidelines. Indeed, KPMG [2011] refers to GRI as the global de facto CSR disclosure standard. CSR disclosure is high among European firms, especially in the countries that represent a large share of our sample of Euro top 500 firms. Table 1 shows that eight countries account for 76% of our firm-year observations and these countries all rate highly in the KPMG [2011] report in terms of quality of CSR disclosure, with the percentage of the top 100 firms in each of the 8 countries disclosing CSR information being 100% in the UK, 94% in France, 62% in Germany, 64% in Switzerland, 72% in Sweden, 82% in the Netherlands, 88% in Spain, and 74% in Italy (KPMG [2011]). This shows a very high likelihood that firms in our sample disclose CSR information; a high likelihood that they 7 use the GRI guidelines; and because of the high uptake of CSR disclosures, a need to use a disclosure measure that indicates the level of CSR disclosure.3 Shadewitz and Niskala [2010], De Klerk and De Villiers [2012], and De Klerk, De Villiers, and Van Staden [2012] use a GRI-based disclosure measure to examine the value relevance of CSR disclosures in, respectively, Finland, South Africa, and the UK. We elaborate on these studies in the “consequences” section below. The only multi-country study we are aware of that use a GRI-based disclosure measure, uses a Tobin’s Q analysis to assess firm value (Gietl et al. [2012]). However, this study has two major shortcomings, namely that it ignores country differences (and only include firm level control variables), and that the CSR disclosure measure used in this study considers a combination of the GRI level of disclosure and whether the GRI level is validated by a third party. This combination is not helpful, as the relationship between these two different aspects and any economic consequences are potentially in opposite directions, as will become clear in our hypotheses development and results.4 Predispositions of CSR disclosure: Country level and firm level expectations Prior multi-country research tends to use a code law/common law country-level dummy variable in an attempt to explain cross-country variation in their variable of interest (e.g., Simnett et al. [2009], Dhaliwal et al. [2012]). They argue that common law countries 3 Note that the percentage of firms that disclose CSR is high, because the KPMG [2011] study includes all firms disclosing some form of social or environmental information. The percentage of firms disclosing CSR using the GRI guidelines is much lower, because it only includes firms that disclose information on a range of social and environmental issues, and disclose this information in accordance with GRI guidelines. 4 The potentially confounding disclosure measure used in Gietl et al. [2012] could explain their main finding, namely that disclosure at the GRI A level, validated by a third party, is associated with a lower Tobin’s Q. The study reports that the relation with the similarly defined GRI B and GRI C levels of validated disclosure is not significantly related to Tobin’s Q. 8 (e.g. the US and the UK) are more investor oriented compared to code law countries (e.g. France) that tend to be more stakeholder oriented (Simnett et al. [2009]). A stakeholder orientation suggests a greater emphasis on corporate responsibilities towards all stakeholders (in contrast with an emphasis on investor needs) and therefore a greater propensity to provide CSR information. These social expectations can drive firm behavior. However, given the countries represented in our sample, using a code/common law dummy variable is not adequate. This is because only the UK has a common law system, while all the other countries use code law. The problem is compounded by the fact that the UK happens to be one of the countries where CSR disclosure is particularly popular, whereas following the argument in the prior research would lead you to believe that a common law country would be shareholder oriented and therefore firms in common law countries would be likely to disclose less CSR information. An alternative country level measure, used by both Simnett et al. [2009] and Dhaliwal et al. [2012], is the World Bank’s ‘rule of law’ index. This index provides an indication of the confidence citizens have in the ability of the country’s laws, law enforcement, and judicial system, to ensure a fair outcome for all. The index is also an indication of the likelihood of crime and violence. It is argued that firms will be motivated to disclose more CSR information in relatively law abiding countries or, alternatively, that investors and other stakeholders are able to demand higher levels of CSR disclosure in these countries (Simnett et al. [2009]). Therefore, we expect higher levels of CSR disclosure in more law abiding countries. However, the World Bank rule of law index provides a measure of the overall rule of law within a country, including issues not directly related to corporate affairs and investor protection measures. We use a different, more sophisticated measure dealing with the rule of law in the matters of specific interest to firms, namely an anti-self-dealing 9 score, based on Djankov, La Porta, Lopes-de-Silanes, and Shleifer [2008]. This country level index considers the incidence of (1) vote by mail; (2) shares not blocked or deposited; (3) cumulative voting; (4) oppressed minority; (5) pre-emptive rights; and (6) capital to call meeting. In the same annual document where the World Bank publishes their ‘rule of law’ index, they also publish several other country level measures. Three of these measures seem to be ideal predictors of the propensity to voluntarily disclose CSR information, namely ‘voice and accountability’, a measure of democracy and freedoms; ‘government effectiveness’, a measure of the quality of public services; and ‘regulatory quality’, a measure of soundness of policy to protect investors and others. As far as we are aware, no prior published CSR study use these World Bank measures.5 We expect CSR disclosures to be higher in countries where these indexes are higher, i.e. where citizens can influence by having democracy and freedoms; where public services, such as investor protection services, are more effective; and where regulation provides more protection. Prior research has shown the ability of the press to influence CSR disclosures (e.g. Clarkson, Li, Richardson, and Vasvari [2008], Brown and Deegan [1998]). These were single country studies. In our cross-country setting where media coverage shows variation, the firm-level measures these prior studies employ would not be appropriate. However, we are able to control for the general level of press freedom associated with each country in our sample. We identify the Reporters without Borders organization’s ‘freedom of press’ index as an appropriate measure. The aim of Reporters without Borders is to promote press freedom and reporter safety throughout the world by drawing attention to countries and incidents where violations of press freedom occur. In line with prior research that generally 5 A conference paper, Cahan et al. [2012] also uses these World Bank measures. 10 find that higher levels of press coverage correlate with higher levels of CSR disclosure we expect firms in countries with higher levels of press freedom to disclose more CSR information. In recent years, sustainability, environmental issues, climate change, and greenhouse gas emissions have featured high on the public agenda in many countries. We use a country level measure that measures the importance of this item, namely Yale Law School’s environmental performance index, which emphasizes law, policy and scientific issues. Firms in countries that show a greater commitment to an environmental agenda could be more likely to disclose more CSR information to reflect the local social concerns and to respond to higher levels of stakeholder pressure to provide information. However, in more environmentally committed countries, CSR related issues are more likely to lead to negative financial consequences and managers may fear that additional CSR disclosure will play into the hand of potential litigants. For example, the 2010 Gulf of Mexico oil spill cost BP billions of dollars. Fear of litigation would make managers less likely to disclose CSR information. Based on these opposing arguments, we form no prior expectation regarding the direction of the association between this particular environmental commitment variable and the level of CSR disclosure. The above discussion regarding predisposition to CSR disclosure at the country level leads to the following hypothesis, stated in alternate form: H1: Firms are likely to disclose higher levels of CSR information in countries with less corporate self-dealing, more democratic rights for citizens, more effective government administration, better regulatory quality, more press freedom, and less/more commitment to an environmental agenda. 11 Prior CSR research has consistently found size to be an important predictor of CSR disclosure levels (e.g. De Villiers and Van Staden [2011]). Other firm level characteristics that the prior research typically used to control for the level of CSR disclosures are profitability, whether the firm needed additional finance from capital markets, book to market value of equity, leverage, riskiness measured by volatility of share price, newness of the firms’ assets, capital expenditure, and whether the firm operates in an environmentally sensitive industry (Clarkson et al. [2008], Cho and Patten [2007], De Villiers and Van Staden [2011]). In addition, in our multi-country setting, firms with international dealings are more likely to be exposed to pressure to disclose CSR information from an international audience. Therefore, we control for the percentage of sales outside of the country where the firm has its headquarters. Bigger firms and more profitable firms have more resources to devote to CSR disclosures. Big, profitable firms are also more likely to attract public and regulatory attention and are thus more likely to use CSR disclosure to appease activists and regulators. Firms that need to access the capital markets for additional funding are more likely to disclose CSR information in order to allay fears that there may be potential liability associated with CSR issues. Firms with a higher book to market value of equity ratio are in a situation of relatively low growth and such firms are more likely to attempt to find competitive advantage in areas such as additional CSR disclosure. Firms with higher leverage are more likely to come under scrutiny from the providers of external finance and are therefore likely to disclose more CSR information to satisfy their demands (e.g., Leftwich, Watts, and Zimmerman [1981]). Firms with a greater percentage of cross-border sales are more likely to face diverse CSR demands and are therefore likely to disclose more 12 CSR information. We are not certain of the direction of the correlation to expect between CSR disclosure and Volat and thus do not form an ex ante expectation. Firms with older assets do not have access to the latest, cleanest technologies, increasing the likelihood that they will experience pressure and incentivizing them to disclose more CSR information to explain their operations. The same line of argument can be applied to firms with higher capital expenditure and those operating in environmentally sensitive industries, because such firms will be pressured into explaining their operations, via CSR disclosures. Firms in environmentally sensitive industries are more likely to come under pressure to provide transparency and explanations through CSR disclosures. Given the importance of these firm level measures in predisposing firms to disclose CSR information, we control for them in our analyses. Consequences of CSR disclosure: Share price, firm value, other Prior studies link better CSR performance (not assessing CSR disclosure) with benefits that could lead to improved financial performance, e.g. increased sales (Lev, Petrovits, and Radhakrishnan [2010]), attracting better quality employees (Edmans [2011]), and more favorable treatment by regulators and policy makers (Brown, Helland, and Smith [2006]). In a meta-analysis of 52 studies, Orlitzky, Schmidt, and Rynes [2003] find evidence of a positive relationship between CSR performance (not disclosure) and financial performance. Survey results show that market participants prefer to obtain CSR information from corporate sources, i.e. CSR disclosures (Radley Yeldar [2012]). Therefore, a positive link between CSR disclosure and financial performance is plausible. 13 Two recent articles focus on standalone CSR reports and their financial consequences, namely Dhaliwal et al. [2011] and Dhaliwal et al. [2012]. Dhaliwal et al. [2011] examine the effect of publishing a standalone CSR report on the cost of equity capital and vice versa. They find that firms with a high cost of equity capital are more likely to publish a standalone CSR report for the first time. Furthermore, they find that such a first-time publication leads to a reduction in the lagged cost of equity capital for firms with superior CSR performance. The authors use a CSR disclosure dummy to indicate the presence of a standalone CSR report without considering the level (or amount) of disclosure. Dhaliwal et al. [2012] find that firms in stakeholder (code law) countries are more likely to publish a standalone CSR report. They also find that analysts in stakeholder (code law) countries find CSR disclosures more informative, i.e. standalone CSR reports are associated with greater analyst forecast accuracy. These findings can be interpreted to indicate that there is a greater expectation of, a greater provision of, and a greater practical use for CSR disclosures in stakeholder (code law) countries. We build on these two papers by further examining the voluntary CSR disclosure phenomenon, but with a stronger CSR disclosure measure that allows for more variation. In addition, we extend the investigation of consequences by focusing on share prices instead of cost of equity capital (e.g. Dhaliwal et al. [2011]) or analyst forecast accuracy (e.g. Dhaliwal et al. [2012]). Prior research on the value relevance of CSR disclosure using an Olhson [1995] type model examines different countries and provides inconsistent results. Some of the studies focus on one aspect of CSR, namely environmental issues. One example is Hassel, Nilsson, and Nyquist [2005] who find published environmental performance information (not firm voluntary disclosure) to be associated with lower market values in Sweden. They explain 14 this by stating that bad environmental performance may reveal the need for remedial spending. Moneva and Cuellar [2009] find financial environmental disclosure by Spanish companies to be value relevant, and associated with higher share prices. However, according to Moneva and Cuellar [2009], non-financial environmental disclosure has no value relevance. Two working papers, Clarkson, Fang, Li, and Richardson [2010] and Plumlee, Brown, Hayes, and Marshall [2010], find voluntary environmental disclosure by US companies to be value relevant in such a way that more disclosure correlate with positive economic outcomes. Shadewitz and Niskala [2010], De Klerk and De Villiers [2012], and De Klerk et al. [2012] examine the value relevance of all CSR disclosures (not just environmental disclosures). All three studies use a GRI guideline based measure of CSR disclosure. The three studies all find CSR disclosure to be value relevant in the positive direction. Shadewitz and Niskala [2010] examine Finnish firms, De Klerk and De Villiers [2012] study South African firms, and De Klerk et al. [2012] investigate UK firms. Shadewitz and Niskala [2010] use a simple GRI dummy variable, whereas the other two studies use a dummy and a complicated ordinal measure that assumes that third party validation of the GRI application level provides added value. Neither of these GRI-based measures extracts the optimum amount of information. The dummy variable, because it ignores the levels of disclosure. The levels combined with validation measure, because of the potentially confounding effect of combining both. In our study, we use an ordinal variable that represent the GRI application level and a separate dummy variable to indicate whether the application level was independently validated. Ignoring for the moment the shortcomings in disclosure measure of these three studies, they all suggest that CSR disclosures are value relevant (associated with higher share prices) and that the combined effect of CSR disclosure with financial accounting 15 information provides a better explanation of market attributes than financial accounting information alone. Overall, CSR disclosures have been found to be associated with higher share prices in some countries, but also with lower share prices. Therefore, there is still a need to further examine this relation. CSR disclosures and how markets interpret them are influenced by country specific issues, such as local laws that affect the relationships between CSR issues and direct economic consequences. Our arguments above lead to the following hypothesis, stated in alternate form: H2: Firms that disclose higher levels of CSR information are likely to have higher share prices. We are also interested in examining the effect of country governance measures on the relationship between CSR disclosure and firm value. Stronger governance mechanisms could reduce the ability of managers to disclose CSR information opportunistically, making CSR disclosures more credible, and thus increasing the weight capital market participants put on these disclosures. This would suggest a stronger link between CSR disclosures and firm value in stronger governance countries. However, financial disclosures themselves may be more opaque in weaker governance countries. Dhaliwal et al. [2012] find the positive relation between CSR disclosure and better analyst forecast accuracy to be stronger in countries with greater financial opaqueness. They ascribe this finding to the fact that opaque financial disclosure increases the value of alternative sources of information, such as CSR disclosures, for financial market participants to form a complete picture of a firm’s prospects. If these arguments hold, this would suggest a stronger link between CSR disclosure and share price in countries with weaker governance. To investigate this relation 16 we add country measures and interaction terms to our basic model. In this exploratory analysis we focus on the most promising country level measures, as indicated by the results of the tests of our first hypothesis. METHOD Sample We begin our sample selection with the 500 largest European firms, as defined by the Financial Times in its 2010 rankings. We next exclude 84 financial firms, due to their unique financial ratio characteristics, rendering comparison with firms in other industries senseless. We also drop 23 firms, for which we cannot find financial data on Datastream. This leaves us with an initial sample of 393 firms. Our first step in hand-collecting the level of GRI information disclosed by firms is to examine the CSR reports of these firms in each of four years, namely 2007 to 2010. Not all firms have such a report. When such a report does not exist, we examine the annual report of the firm. With these documents we first establish whether these firms disclose a GRI score in that year. As a second step, when a GRI score is disclosed, we record the level of the firm’s GRI compliance (A, B, or C) and whether the GRI compliance level was independently validated in that year. From our potential sample of 1,572 observations (393 firms * 4 years) we lose some observations due to missing data on some of the variables collected from Datastream. We keep the number of observations stable at 1,227 for all analyses related to the predispositions for CSR disclosure. However, for the consequences analyses, we allow the 17 number of observations to vary from estimation to estimation, in order to maximize the number of observations used in each analysis. Research design: predispositions In order to test hypothesis 1, which analyzes the predispositions for CSR disclosure, we estimate the following general model: CSR Disclosure = α0 + α1Country Institutional Variables + α2Firm Level Variables + ε (1) Given that our CSR disclosure measure is not continuous, but indicates levels of disclosure, we use an ordered logit model. Annex 1 provides a summary of the variables included to facilitate referencing. CSR disclosure measure. We use firms’ GRI disclosure level as a proxy for their CSR disclosure. Under the GRI G3 guidelines, in force during the 2007-2010 period, firms can choose to disclose a limited number of CSR items for a C-rated GRI application level, increase their disclosures for a B-rated level, or disclose the full ambit of GRI guideline CSR disclosures for an A-rated level. We convert this GRI application level to GRI_Score as follows: A=3, B=2, C=1. We attribute a score of 0 to all firms that do not follow GRI. Country institutional variables. We use six country level variables that generally represent the level of investor protection, level of individual freedom of expression, and societal concern with CSR issues. 18 We follow Djankov et al. [2008] by using their anti-self-dealing measure, an index that reflects a specific part of corporate governance and considers (1) vote by mail; (2) shares not blocked or deposited; (3) cumulative voting; (4) oppressed minority; (5) preemptive rights; and (6) capital to call meeting. This measure, LAW, can range between 0 and 5, with higher numbers representing less self-dealing or better governance. We also use three World Bank measures, namely (1) citizens’ ability to select a government and voice their concerns (Voice), (2) public service and policy quality and effectiveness (Gov_Eff), and (3) sound regulatory quality (Reg_Qual).6 The values of these measures range between -2.5 to 2.5, where higher values reflect higher citizen participation (Voice), more effective governments (Gov_Eff), and higher regulatory quality (Reg_Qual). Reporters Without Borders publish an index reflecting the degree of freedom of journalists and the media in more than 170 countries.7 Values range from 0 to 112.5, with lower values reflecting higher freedom of the press. Therefore, lower values of this measure, Press, indicate higher level of press freedom. Our final measure is the environmental performance index of created by The Yale Center for Environmental Law and Policy and the Center for International Earth Science Information Network at Columbia University. This index covers more than 150 countries and is released every second year (Env_Perf).8 Values can range between 0 to 100, higher values indicating countries that strongly pursue environmental policy goals. Firm level variables. One of the most consistent determinants of environmental disclosure is industry. We use an environmentally sensitive industry dummy variable (Ind_Sens), which identifies the forestry, metals mining, coal mining, oil and gas 6 See http://info.worldbank.org/governance/wgi/mc_countries.asp See http://en.rsf.org/ 8 See http://epi.yale.edu/; we used the 2008 data for 2007 and the mean of 2008 and 2010 for 2009. 7 19 exploration, paper and pulp, chemicals, pharmaceuticals and plastics, iron and steel, electricity, gas and waste water industries as specified in De Villiers et al. [2011].9 In addition, we identify from the prior literature, nine firm level variables to control for the level of CSR disclosures (e.g., Clarkson et al. [2008], De Villiers and Van Staden [2011]). These variables are size (Size), profitability (ROA), need of additional finance (Fin), bookto-market value of equity (B_M), leverage (Lev), level of international trade (Internat), share price volatility (Volat), age of assets (New), and capital expenditure (Capex). We have already discussed the expected direction of the association between these variables and CSR disclosure in the literature review section. Research design: consequences We examine the influence of CSR disclosure level on share prices (hypothesis 2) with the use of a modified Ohlson [1995] research design, where share price is assumed to be determined by earnings, book value, the variables of interest, and control variables.10 Our model is as follows: Share_Price = β0 + β1EPS + β2BV_pS + β3GRI_Score + β4GRI_Validation + β5Size + β6Ind_Sens + year dummies + country dummies + ε (2) Where GRI_Score is coded the same as before, namely 3 for firms with a GRI application level of A, 2 for B, 1 for C, and 0 for firms that do not follow GRI; and where 9 Sensitive industries are defined as in De Villiers et al. [2011] as those with SIC codes between 800-899 (Forestry), 1000-1099 (Metal Mining), 1200-1399 (Coal Mining and Oil and Gas Exploration), 2600-2699 (Paper and Pulp Mills), 2800-3099 (Chemicals, Pharmaceutical and Plastics Manufacturing), 3300-3399 (Iron and Steel Manufacturing), and 4900-4999 (Electricity, Gas and Waste Water). 10 Note that earnings and book value replace dividends in valuation, according to the clean surplus relation. 20 GRI_Validation is a dummy variable coded 1 for firms that use an external party to validate or certify the firm’s GRI application level. We expect GRI_Score to be positively and significantly related to Share_Price. Managers choose to get third party validation of the firm’s GRI application level on a voluntary basis. Voluntary CSR assurance is obtained to increase the credibility of CSR disclosures and is usually obtained when there are preexisting credibility issues (Simnett et al. [2009]). In this case, investors can observe the disclosure level themselves and thus external validation of the GRI application level may have a limited effect on investors’ valuation decisions. If such an effect exists, there are two possible scenarios namely that both CSR disclosures and GRI_Validation are positively priced, or alternatively that GRI_Validation is chosen only by managers of firms with negative CSR issues and thus GRI_Validation could be negatively priced. Therefore, we form no prior expectation for the direction of the relationship between GRI_Validation and share prices. In the Ohlson [1995] model it is assumed that market price per share is positively associated with book value per share as well as with the value of earnings per share, as these are indicative of future dividends. Therefore, we expect to find positive coefficients for these two independent variables. Aboody, Barth, and Kasznik [2004] add analyst growth forecast to their modified Ohlson [1995] model; Hann, Lu, and Subramanyam [2007] add growth in sales, R&D expense and number of employees. In following these prior examples of adding additional controls appropriate to the issue under investigation, we identify two additional firm level control variables, namely size and industry. These controls are appropriate, because the relationship between CSR disclosure and firm value may be quite different for firms in sensitive industries, compared to other industries; and the same applies to larger firms that are potentially more exposed to political cost, due to 21 their increased visibility. We include year dummies to control for any effects specific to a particular year. Finally, because we know from our ‘predispositions’ discussion that country specific issues influence CSR disclosure, we suspect that country specific effects may be at work in the relationship between CSR disclosure and market values. Therefore, we include country dummies to account for country sources of heterogeneity and we follow Aggarwal, Erel, Stulz, and Williamson [2009] in their multi-country approach by clustering residuals at the country level. FINDINGS Descriptive statistics Table 1 shows descriptive statistics, with Panel A providing disclosure information per country, Panel B providing information regarding the frequencies of our CSR disclosure measures, and Panel C providing information for all variables used. Panel A shows that our sample covers 22 European countries. Nineteen of these countries have at least 12 observations. Most of our observations represent, in order, UK firms (246), French (202), German (149), Swiss (92), and Swedish firms (79). This panel also shows that the use of GRI is widespread among large European firms and not confined to a limited number of countries, e.g. note that the mean for GRI_Score is 1.0 or above for eight of the countries in our sample. Panel B shows that nearly one third of the firms in our sample disclose that they use GRI guidelines for their CSR disclosures, with more than 16% following the GRI guidelines fully (A application level), over 11% following them to the B 22 application level, and more than 3% following GRI to the C application level.11 Almost 79% of the firms that follow GRI get external validation that they comply with the relevant GRI application level. Panel C shows descriptive statistics for all variables, e.g., the firms in our sample have a mean return on assets of 7.1% and a mean book to market value of equity of 55.4%. Ind_Sens shows that about one third of our firm-year observations are from firms in environmentally sensitive industries. <<<Table 1 about here>>> Correlations Table 2 shows the correlations between our dependent variable, country variables, and firm level variables. Panel A reveals that a number of country variables are highly correlated, especially among the three measures derived from the World Bank (Voice, Gov_Eff, and Reg_Qual) and the press freedom measure, Press. Law and Env_Perf are correlated with other country level measures in the range from 0.3 to 0.34. These high correlations suggest the need for care in regression specification. Except between country level variables, none of the correlations among the variables used in equation 1 (predispositions) are above 0.4. <<<Table 2 about here>>> 11 The fact that about a third of the firms in our sample use GRI may seem low, but note that Gietl et al. [2012], using a sample of Eurostoxx 600 firms over the same time period (2007-2010), report that only 240 of 1686 firms, or 14%, use GRI. 23 Predispositions Table 3 shows the result of ordered logit regression estimations. This model is more appropriate for our analysis (than correlations) because our dependent variable, GRI_Score, is not a continuous measure, but provides discrete levels of CSR disclosure based on the number of GRI categories disclosed by firms. Panel A shows the results of ordered logit estimations with only one independent variable each, i.e. each of our country-level and firm-level measures in turn. As can be expected, these regressions yield qualitatively similar results to the correlation table results. The reason we show these, is that given the nature of the dependent variable, an ordered logit regression is the theoretically correct method. The results show Law, and Gov_Eff to be significant at the 10% level; Env_Perf at the 5% level; and the other three country level measures, Voice, Reg_Qual, and Press to be significant at the 1% level. These findings should be interpreted with caution, as these are univariate results without proper control for issues such as firm size, industry, etc. The direction of the correlations indicates that higher levels of CSR disclosure are associated with more investor protection (anti-self-dealing), higher levels of democracy, more efficient government, better regulatory quality, and more press freedom. These associations are consistent with our expectations. Note that the country measure we had no prior expectation for, Env_Perf, is negatively correlated with CSR disclosure level, suggesting a lower level of disclosure among firms in countries that demonstrate a higher level of commitment to the environment through policy and laws. This can be explained by firms shying away from CSR disclosure in countries with a higher environmental commitment, because of a fear that such commitment increases the likelihood of disclosure related litigation and liability. 24 Still referring to Table 3, Panel A, all our firm level control variables show significant correlations with CSR disclosure levels (GRI_Score) on the basis of individual variable ordered logit regressions, except profitability (ROA). New is significant at the 10% level; Fin, Lev, Internat, Volat, and Capex are significant at the 5% level; and the rest at the 1% level, i.e. Size, B_M, and Ind_Sens. It is important to note that these results do not include proper controls for all relevant variables. This may explain why ROA is not statistically associated with CSR_Score. The rest of the significant firm level variables are all in the predicted direction. The fact that these variables are significant highlights the need to control for these factors in the regression analyses to follow. We know from the correlation statistics in Table 2 that some of the country variables we use are highly correlated and this raises the issue of multicollinearity should we use these variables in the same model. If multicollinearity is present among variables used together in the same equation, we may not be able to rely on the coefficients or significance levels estimated to draw conclusions regarding each individual variable. To test whether multicollinearity is a concern, we estimate an ordinary least squares regression that includes all our country level variables, as well as all our firm level variables, to inspect the variance inflation factors (VIFs). Two of the country level variables have VIFs above the standard benchmark of 10, indicating the presence of multicollinearity issues. We deal with this problem in two ways. We first estimate separate ordered logit models for each country level variable including, in each case, all the firm level control variables. Next, we use principal component analysis on the country level variables to construct components that can be used together in a regression. 25 Table 3, Panel B shows the results of ordered logit estimations for each country level measure in turn combined with all the firm level measures. For example, the first column shows that Law is positively and significantly related to GRI_Score at the 5% level when firm level variables are included. In similar fashion, Voice, Gov_Eff, Reg_Qual, and Press, are related to GRI_Score at the 1% level of significance and in the expected directions. Env_Perf is negatively related to GRI_Score at the 5% level. The directions of these associations are all consistent with the correlation table results (Table 2) and with the individual ordered logit results (Table 3, Panel A). Our results suggest that firms disclose higher levels of GRI information in countries with better investor protection (less anti-selfdealing), a greater ability to choose your own government, greater government efficiency, better regulatory quality, more press freedom, and less government commitment to environmental policy and law. These results provide evidence in support of our first hypothesis. The only one of these six measures mentioned in hypothesis 1 we did not form a prior expectation on the direction for, deals with countries’ environmental commitment. You may recall that we argued that firms in countries committed to an environmental agenda could, on the one hand, be expected to come under pressure to disclose more CSR information, but on the other hand, management could fear the likelihood of increased environment-related litigation and liability. Fear of litigation appears to be a stronger motivation, resulting in less CSR disclosure in countries that demonstrate a higher commitment to environmental issues. Turning to firm level variables, a scan of all six regression results shows that most of these variables are significantly associated with GRI_Score. The exceptions, i.e., the variables which are that are not significantly associated with the dependent variable, are 26 Fin, Internat, and Volat. It is not uncommon for some control variables to be found to be not significant, e.g., Clarkson et al. [2008] found Volat to be not significantly associated with CSR disclosure, as did De Villiers and Van Staden [2011] with Fin. Neither of these two studies had Internat, because they included only US firms. It may be that in our European setting, where countries are close to each other and barriers to international trade are low, that international trade is not a very important determinant of CSR disclosures. However, we could not predict that result ex ante. The results in Table 3, Panel B can therefore be taken to suggest that firms are likely to disclose higher levels of CSR information if they: are larger, are more profitable, have high book to market firms, are more leveraged, use older equipment, spend more on capital, and operate in environmentally sensitive industries. However, we do not find support for an association between CSR disclosures and firms anticipating the need for additional finance, being exposed to international trade, or share price volatility. <<<Table 3 about here>>> We now turn to our second method of dealing with the multicollinearity issue due to the high correlation between the country level measures. This time we consider all six measures at the same time to learn more about the way that country variables interact to provide an indication of the predisposition of different firms to provide different levels of CSR disclosure. We follow Li [2010] by doing a principle component analysis of all variables together and including the resulting factors into a regression model. Table 4 provides information about this analysis. We use the first two components, because they have Eigenvalues above 1 (Panel A). Panel B shows the factor loadings and reveals that component 1 is determined mostly by Voice, Gov_Eff, Reg_Qual, and Press. Given that 27 Voice and Press both relate to citizens’ and investors’ ability to voice their concerns, and that Gov_Eff and Reg_Qual both relate to the resultant regulation and the implementation of these regulations, we label it Country_Voice&Reg; whereas component 2 is determined mostly by Law and thus we call it Country_Law. When these components are used in the ordered logit regression, Country_Voice&Reg is highly significant in the predicted direction, as shown in Panel C. Country_Law is marginally significantly related to GRI_Score (at the 10% level). All the firm level variables correlate in the same directions as before and the same three (Fin, Volat, and Internat) are not significant. Table 4, Panel C therefore provides additional evidence in support of the country level hypothesis (H1) (relating to Law, Voice, Gov_Eff, Reg_Qual, and Press); and our firm level expectations (relating to size, profitability, book to market, leverage, age of equipment, capital expenditure, and sensitivity of industry). <<<Table 4 about here>>> We have now established that each of our six country level variables, as well as the combined variable, Country_Voice&Reg, are significantly associated with the level of CSR disclosures. Most of our firm level variables are also significantly associated with CSR disclosure. We now proceed to examine whether CSR disclosures are associated with positive economic outcomes. We will do this first in general terms and then examine whether any positive economic outcomes are more pronounced in certain country types as measured by our country measures, individually and combined. 28 Consequences In order to assess whether the CSR disclosure level of a firm has value relevance we use an Ohlson [1995] type model in a per share specification with additional control variables – see Table 5. We further include year dummies and country dummies and show the results after removing outliers.12 Country dummies control for any country effects here, because we do not (yet) include any country variables in the model. The disclosure measures we include as independent variables are GRI_Score and GRI_Validation. Table 5, Panel A shows the coefficient for GRI_Score to be positive and significant at the 1% level. This indicates that CSR disclosures are value relevant to investors in such a way that firms with higher levels of CSR disclosure are associated with higher share prices after controlling for several accounting measures. The coefficient for GRI_Validation is negative and significant, at the 5% level. This suggests that firms that have their GRI disclosure level validated by a third party are associated with lower share prices. Managers will not knowingly indulge in a voluntary activity that lowers share prices. A plausible explanation for this result is that firms with pre-existing credibility issues are more likely to seek external validation. In addition, we perform an F-test, which indicates GRI_Score and GRI_Validation are not statistically equal to zero (p = 0.05), showing these variables add explanatory value to the equation. <<<Table 5 about here>>> 12 Following Belsley, Kuh, and Welsch [2004] we remove the observations where either Rstudent or Dfitts have an absolute value in excess of 2. This results in the elimination of 16 observations (1% of the initial sample of 1,418 observations) in the basic model in Panel A. 29 There is potentially a self-selection bias when we include our disclosure measure as an independent variable, as firms can choose whether they disclose that they use the GRI for CSR disclosure. We therefore perform the Heckman procedure to re-estimate our base model, using all the determinants (or predispositions) used in Table 3 to estimate an indicator variable coded as one when a GRI score is disclosed and zero otherwise, in order to assess whether we have a self-selection problem. Self-selection does not appear to be a significant concern, because lambda of the Mills ratio has a p-value of 0.09, being only marginally significant. Nevertheless, following Chiburis and Lokshin [2007], we next estimate a Heckman procedure where the selection model is an ordered probit (oHeckman), allowing us to consider all four possible levels of our GRI_Score variable. We find that the four estimated lambdas are not statistically significant (with p-values of 0.362, 0.662, 0.594, and 0.147). Therefore, we do not use a two-step estimation when obtaining our next set of results. Consequences by country governance strength The results in the first column of Table 5, Panel A show that higher levels of CSR disclosures are associated with higher share prices. This analysis includes appropriate control variables, including country dummies to account for any country effects. However, we are interested in knowing whether there are more general country effects at work. If this is the case, we can report in which types of countries it is more likely for positive economic outcomes to be related to higher levels of CSR disclosure. 30 We examine the influence of country variables on economic consequences of CSR disclosure by adding the composite country variable we calculated in Table 4 and that was highly significant in the Table 4 regression, Country_Voice&Reg, as well as an interaction term with GRI_Score to the basic consequences equation, i.e. equation 2. The variable of interest is the interaction term, GRI* Country_Voice&Reg. We expect CSR disclosures to be more likely to lead to positive economic outcomes in countries with a higher Country_Voice&Reg score, i.e. countries with more democracy, more effective government, better regulation, and more press freedom. The results are shown in the final column of Table 5, Panel A. The interaction term is significant, showing that share prices and CSR disclosures are more positively correlated in countries with better governance as measured by Country_Voice&Reg. Note that GRI_Score remains significant, although at a lower level. To confirm whether this result holds for each of the main constituent parts of the composite measure, Country_Voice&Reg, we estimate four separate regressions where we substitute Voice, Gov_Eff, Reg_Qual, and Press for Country_Voice&Reg in turn. The results, in Table 5, Panel B are in each case consistent with our results and interpretation above therein that the interaction terms between GRI_Score and the four country variables are in each case significant in the predicted direction. GRI_Score, not the variable of interest here, losses significance in most of these estimations. To summarize, higher levels of CSR disclosure appear to be more likely to be rewarded with positive economic outcomes in countries with higher levels of democracy, more effective government, better quality regulation and more press freedom. 31 Additional tests Table 6 shows that our result in the basic model, i.e. that more CSR disclosure (as measured by GRI_Score) is associated with higher share prices, hold in three of the four years in our sample, namely 2007, 2009, and 2010. During 2008, the turbulence in the financial markets as a consequence of the so-called global financial crisis may have influenced the general relationship we identify in our main analysis (Table 5). <<<Table 6 about here>>> Our measure of CSR disclosure is based on companies’ disclosing the level to which they follow the GRI guidelines. Although GRI is the most popular CSR guideline, especially among European firms (KPMG [2011]), there is a possibility that some firms in our sample disclose high levels of CSR information without indicating that they follow GRI guidelines. We would have coded such firms’ GRI_Score as “0”, thereby reducing the likelihood of getting our predicted results. We next perform tests to assess whether our results were diminished by this decision. Specifically, we ignore all firm-years where GRI_Score equals zero and re-estimate all the equations reported in both panels of Table 5. The results, shown in Table 7, are all consistent with the main results in Table 5, providing additional evidence in support of our hypotheses and findings that higher levels of CSR disclosure is associated with higher share prices; and that this association is stronger in countries with more democratic institutions, more effective governments, better quality regulations, and more press freedom. We also confirm that our results hold when we use a simplified modified Ohlson [1995] model specification. Specifically, we estimate the two equations in Table 5, but 32 without the additional control variables (Size and Ind_Sens). Focusing on the variables of interest only, the untabulated results are that the estimated coefficient for GRI_Score is positive and significant at the 5% level in the first equation and GRI*Country_Voice&Reg is positive and significant at the 5% level in the second equation. We also use this simplified model to re-estimate all the models reported in the rest of Table 5, as well as Tables 6 and 7. The variable of interest is significant and in the predicted direction in each case, except for the sample with only the observations from 2010, where GRI_Score loses significance. <<<Tables 7 and 8 about here>>> In order to test for the existence of positive economic consequences to the disclosure of GRI, we perform a Tobin’s Q analysis, testing whether CSR disclosures are associated with higher firm values measured by Tobin’s Q (market value to book value of firm assets). We follow Aggarwal et al. [2009] in their multi-country approach, clustering errors by country. We further draw on Jiao [2011], Coles, Daniel, and Naveen [2008], and Roll, Schwartz, and Subrahmanyam [2009] in identifying control variables (i.e. Size, ROA, Capex, Long term debt, and Internat) and add our CSR disclosure measures (GRI_Score and GRI_Validation). We also include year and country dummies. Outliers were removed by following the same procedure as in the main analyses.13 The results in the first column of Table 8 show that the coefficient for GRI_Score is positive and highly significant, and the coefficient for GRI_Validation is not statistically different from zero. These results are indicative that firms with higher levels of CSR disclosure are associated with higher market 13 Following Belsley et al. [2004] we remove the observations where either Rstudent or Dfitts have an absolute value in excess of 2. This results in the elimination of 43 observations (3% of the initial sample of 1,324 observations) in the basic model. 33 values of assets (scaled by book values) and getting these GRI disclosure levels validated by a third party do not affect firm value measured in this way. You may recall that Gietl et al. [2012] perform Tobin’s Q analyses using a CSR disclosure measure that includes both the GRI level and whether it was validated by a third party. Our results in Table 8 demonstrate the superiority of our disclosure measures, i.e. keeping the GRI disclosure level separate from an indicator of whether the GRI level was validated by a third party. In the second column of Table 8, the results of the expanded model including Country_Voice&Reg and its interaction GRI_Score are similar to the results in Table 5, i.e. the interaction term is significant (p-value=0.017), suggesting that the association between the market to book ratio and CSR disclosure is higher in countries with a combination of more democracy, more effective government, better quality regulation, and more press freedom. GRI_Score and Country_Voice&Reg are not statistically significant in this equation. CONCLUSION We examine the CSR disclosures of the top 500 European firms during a recent four year period (2007-2010). We hand collect the GRI application level firms use to disclose CSR as a measure of the extent of their CSR disclosure. About one third of the firms in our sample disclose their GRI application level. We find evidence to support our hypothesis that there is a higher likelihood to disclose higher levels of CSR among firms in countries with: greater investor protection measures, higher levels of democracy, more government effectiveness, higher quality 34 regulations, more press freedom, and a lesser commitment to environmental policies. At the firm level, we find that firms are more likely to disclose higher levels of CSR if they are larger, more profitable, have higher book to market ratios, are more highly leveraged, have older assets, spend more on capital, and operate in environmentally sensitive industries. We also find evidence that higher levels of CSR disclosure are associated with higher share prices. As such, CSR disclosures can be said to embody information that is value relevant to investors. We find strong evidence to support the view that this positive economic association is more pronounced in countries with higher levels of democracy, government effectiveness, regulatory quality and press freedom. This implies that market participants find CSR disclosures more informative in countries where investors are in a better position to voice their concerns (through democratic mechanisms and through the media), and where opportunities to voice concerns have resulted in better regulation and more effective government implementation of regulations. Our results may be limited by the fact that some firms who do not disclose a GRI application level may also provide a relatively high level of CSR disclosures. However, if this was the case, then our results would be weaker. We also repeat our tests with the subsample of firms that disclose a GRI application level and find similar results. As another way of measuring positive economic consequences, we show that higher levels of CSR disclosure are associated with higher Tobin’s Q and find preliminary evidence to suggest that this association is stronger in countries with more democracy, more government effectiveness, higher regulatory quality and press freedom. We suggest a full Tobin’s Q analysis as an avenue for future CSR research. 35 Annex 1 – Variable definitions Disclosure Variables GRI_Score Coded 3 for firms disclosing GRI at the A level, 2 for GRI B level disclosure, 1 for GRI C level disclosure, 0 for not reporting using the GRI disclosure framework GRI_Validation Indicator variable coded one when GRI score is validated by an independent firm and zero otherwise, in cases where there is a GRI score Country-level variables Law Anti-self-dealing score, based on Djankov et al. [2008]. The index considers (1) vote by mail; (2) shares not blocked or deposited; (3) cumulative voting; (4) oppressed minority; (5) pre-emptive rights; and (6) capital to call meeting Voice Voice and accountability, measured by Worldbank, reported in Kaufman et al. [2010] Gov_Eff Government effectiveness, measured by Worldbank, reported in Kaufman et al. [2010] Reg_Qual Regulatory quality, measured by Worldbank, reported in Kaufman et al. [2010] Press Freedom of the press, as measured by Reporters without borders Env_Perf Environmental performance index, Yale Centre for environmental law and policy Firm-level variables Size Size, measured as logarithm of total assets ROA Net income (before extraordinary items and preferred dividends) / beginning total assets Fin Financing, measured as (sales less purchases) of common and preferred shares plus change in long term debt B_M Book to market value of equity ratio Lev Leverage ratio, calculated as total debt divided by total assets Internat International trade, calculated as the percentage of sales made in countries other than the firms’ headquarters firm Volat Share price volatility New Firms’ asset newness, measured as net property plant and equipment (PPE)/ gross PPE Capex Capital expenditure, measured as capital expenditures / sales Ind_Sens Industry sensitivity, an indicator variables coded one for firms operating in environmentally sensitive industries, and zero otherwise. Sensitive industries are identified by the SIC codes described in De Villiers et al. [2011] – see footnote for SIC codes Share_Price Closing market value per share, three months after fiscal year end EPS Earnings per share BV_pS Book value per share, taken at the end of the fiscal year 36 REFERENCES ABOODY, D., M.E. BARTH, and R. KASZNIK. ‘SFAS No. 123 Stock-Based Compensation Expense and Equity Market Values.’ The Accounting Review 79 (2) (2004): 251-275. AGGARWAL, R., I. EREL, R. STULZ, and R. WILLIAMSON. ‘Differences in Governance Practices between U.S. and Foreign Firms: Measurement, Causes, and Consequences.’ The Review of Financial Studies 22(8) (2009): 3131-3169. BALL, R., A. ROBIN, and J. WU. ‘Incentives versus standards: properties of accounting income in four East Asian countries.’ Journal of Accounting and Economics 36 (2003): 235–270. BELSLEY, D., E. KUH, and R. WELSCH. ‘Regression diagnostics: Identifying influential data and sources of collinearity.’ Wiley, 2004. BROWN, N., and C. DEEGAN. ‘The public disclosure of environmental performance information-a dual test of media agenda setting theory and legitimacy theory.’ Accounting and Business Research 37(1) (1998): 21-41. BROWN, W., E. HELLAND, and J. SMITH. ‘Corporate philanthropic practices.’ Journal of Corporate Finance 12 (5) (2006): 855–877. BURGSTAHLER, D., L. HAIL, and C. LEUZ. ‘The importance of reporting incentives: Earnings management in European private and public firms.’ The Accounting Review 81 (2006): 983–1017. CAHAN, S., C. DE VILLIERS, D. JETER, N. NAIKER, and C. VAN STADEN. ‘Fact or Fiction: Are CSR Disclosures Credible? Cross-Country Evidence.’ Unpublished conference paper, 2012. AFAANZ Conference, Melbourne, Jul. CHIBURIS, R., and M. LOKSHIN. ‘Maximum likelihood and two-step estimation of an ordered-probit selection model.’ The Stata Journal 7 (2007): 167-182. CHO, C.H., and D.M. PATTEN. ‘The role of environmental disclosures as tools of legitimacy: a research note.’ Accounting, Organizations and Society 32(7–8) (2007): 639–647. 37 CLARKSON, P. M., Y. LI, G. D. RICHARDSON, and F. P. VASVARI. ‘Revisiting the relation between environmental performance and environmental disclosure: An empirical analysis.’ Accounting, Organizations and Society 33 (2008): 303-327. CLARKSON, P., X. FANG, Y. LI, and G. D. RICHARDSON. ‘The relevance of environmental disclosures for investors and other stakeholder groups: are such disclosures incrementally informative?’ Available at http://ssrn.com/abstract=1687475. COLES, J. L., N. D. DANIEL, and L. NAVEEN. ‘Boards: Does one size fit all?’ Journal of Financial Economics 87 (2008): 329-356. DE KLERK, M., and C. DE VILLIERS. ‘The Value Relevance of Corporate Responsibility Reporting by South African Companies.’ Meditari Accountancy Research 20(1) (2012): 21-38. DE KLERK, M., C. DE VILLIERS, and C. VAN STADEN. ‘The value relevance of Corporate Social Responsibility disclosure in the UK.’ Unpublished conference paper, 2012. AFAANZ Conference, Melbourne, July. DE VILLIERS, C., V. NAIKER, and C. VAN STADEN. ‘The Effect of Board Characteristics on Firm Environmental Performance.’ Journal of Management 37(6) (2011): 1636-1663. DE VILLIERS, C. and C. J. VAN STADEN. ‘Where firms choose to disclose voluntary environmental information.’ Journal of Accounting and Public Policy 30 (2011): 504 – 525. DHALIWAL, D. S., O. Z. LI, A. H. TSANG, and Y. G. YANG. ‘Voluntary Non-Financial Disclosure and the Cost of Equity Capital: The Case of Corporate Social Responsibility.’ The Accounting Review 86 (2011): 59-100. DHALIWAL, D. S., S. RADHAKRISHNAN, A. TSANG, and Y. G. YANG. ‘Nonfinancial Disclosure and Analyst Forecast Accuracy: International Evidence on Corporate Social Responsibility Disclosure.’ The Accounting Review 87(3) (2012): 723-759. DJANKOV, S., R. LA PORTA, F. LOPEZ-DE-SILANES, and A. SHLEIFER. ‘The law and economics of self-dealing.’ Journal of Financial Economics 88 (2008): 430-465. 38 EDMANS, A. ‘Does the Stock Market Fully Value Intangibles? Employee Satisfaction and Equity Prices.’ Journal of Financial Economics 101 (3) (2011): 621–640. FRANCIS, J, D. NANDA, and P. OLSSON. ‘Voluntary disclosure, earnings quality, and cost of capital.’ Journal of Accounting Research 46 (1) (2008): 53–99. GRI. ‘About GRI.’ (2012): Available at https://www.globalreporting.org/Information/about-gri/Pages/default.aspx accessed on 28/08/2012. GIETL, S., M. GÖTTSCHE, A. HABISCH, M. ROLOFF, and M. SCHAUER. ‘Does CSR reporting destroy firm value? Empirical evidence on GRI-aligned European firms.’ Financial Accounting eJournal 4 (148) (2012): Available at http://papers.ssrn.com/sol3/papers.cfm?abstract_id=2183601. HANN, R.N., Y.Y. LU, and K.R. SUBRAMANYAM. ‘Uniformity versus Flexibility: Evidence from Pricing of the Pension Obligation.’ The Accounting Review 82 (1) (2007): 107-137. HASSEL, L., H. NILSSON, and S. NYQUIST. ‘The value relevance of environmental performance.’ European Accounting Review 14(1) (2005): 41-61. HEALY, P., A. HUTTON, and K. PALEPU. ‘Stock performance and intermediation changes surrounding sustained increases in disclosure.’ Contemporary Accounting Research 16 (3) (1999): 485-520. HEALY, P. M., and K. G. PALEPU. ‘Information asymmetry, corporate disclosure, and the capital markets: A review of the empirical disclosure literature.’ Journal of Accounting and Economics 31 (2001): 405–440. INGRAM, R. W., and K. B. FRAZIER. ‘Environmental performance and corporate disclosure.’ Journal of Accounting Research 18 (1980): 614-622. JIAO, Y. ‘Corporate disclosure, market valuation, and firm performance.’ Financial Management (2011): 647-676. KPMG. ‘International survey of corporate responsibility reporting.’ KPMG International, 2001. Available at http://www.kpmg.com/Global/ IssuesAndInsights/ArticlesAndPublications/Pages/Sustainability-corporateresponsibility-reporting-2008.aspx. 39 KPMG. ‘International survey of corporate responsibility reporting 2011.’ KPMG International, 2011. Available at http://www.kpmg.com/PT/pt/IssuesAndInsights/Documents/corporateresponsibility2011.pdf. LEFTWICH, R.W., R.L. WATTS, and J. ZIMMERMAN. ‘Voluntary corporate disclosure: The case of interim reporting.’ Journal of Accounting Research 19 (1981): 50 – 77. LEV, B., C. PETROVITS, and S. RADHAKRISHNAN. ‘Is doing good good for you? Yes, charitable contributions enhance revenue growth.’ Strategic Management Journal 31(2) (2010): 182–200. LI, X. ‘The impacts of product market competition on the quantity and quality of voluntary disclosures.’ Review of Accounting Studies 5 (2010): 663–711. MONEVA, J. M., and B. CUELLAR. ‘The value relevance of financial and non-financial environmental reporting.’ Environmental and Resource Economics 44 (2009): 441456. OHLSON, J. A. ‘Earnings, book values and dividends in equity valuation.’ Contemporary Accounting Research 11(2) (1995): 661-687. ORLITZKY, M., SCHMIDT, F.L. and S.L. RYNES. ‘Corporate Social and Financial Performance: A Meta-analysis.’ Organization Studies 24(3) (2003): 403–441. PLUMLEE, M., D. BROWN, R. HAYES, and R.S. MARSHALL. ‘Voluntary environmental disclosure quality and firm value: Further evidence.’ Unpublished paper, 2010. Available at http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1744114. RADLEY YELDAR. ‘The value of extrafinancial disclosure: What investors and analysts said. Report commissioned by Accounting for Sustainability and the Global Reporting Initiative.’ Radley Yeldar, 2012. Available at https://www.globalreporting.org/resourcelibrary/The-value-of-extra-financialdisclosure.pdf ROLL, R., E. SCHWARTZ, and A. SUBRAHMANYAM. ‘Options trading activity and firm valuation.’ Journal of Financial Economics 94 (2009): 345-360. 40 SCHADEWITZ, H. and M. NISKALA. ‘Communication via responsibility reporting and its effect on firm value in Finland.’ Corporate Social Responsibility and Environmental Management 17 (2010): 96-106. SIMNETT, R., A. VANSTRAELEN, and W. F. CHUA. ‘Assurance on Sustainability Reports: An International Comparison.’ The Accounting Review 84 (2009): 937-967. 41 TABLE 1 Descriptive statistics Panel A – Descriptive statistics for CSR disclosure measures by country Country Frequency Percent of total Mean of GRI_Score Mean of GRI_Validation Austria 13 1.06 1.85 1.00 Belgium 32 2.61 0.56 0.75 Czech Republic 4 0.33 0.00 0.00 Denmark 40 3.26 0.45 0.67 Finland 40 3.26 1.23 1.00 France 202 16.46 0.29 0.79 Germany 149 12.14 0.85 0.73 Greece 13 1.06 0.85 0.40 Hungary 12 0.98 1.00 1.00 Ireland 24 1.96 0.50 1.00 Italy 48 3.91 1.27 1.00 Luxemburg 3 0.24 0.33 1.00 Netherlands 65 5.30 1.02 0.69 Norway 21 1.71 0.90 0.88 Poland 4 0.33 1.00 1.00 Portugal 21 1.71 1.38 1.00 Russia 43 3.50 0.21 0.60 Spain 54 4.40 1.70 1.00 Sweden 79 6.44 0.85 0.48 Switzerland 92 7.50 0.91 0.62 Turkey 22 1.79 0.05 0.00 United Kingdom 246 20.05 0.66 0.84 1,227 100.00 Total 0.75 0.79 GRI_Score measures the level of corporate social responsibility disclosure and takes a value of 3 for firms disclosing GRI at the A level, 2 for GRI B level disclosure, 1 for GRI C level disclosure, 0 for not reporting using the GRI disclosure framework. GRI_Validation is a dummy variable coded 1 when the GRI disclosure level is validated by an independent firm, and zero otherwise. 42 Panel B – Descriptive statistics for CSR disclosure frequencies GRI_Score GRI_Validation Variable value Frequency (%) Frequency (%) 0 845 (68.87) 81 (21.20) 1 42 (3.42) 301 (78.80) 2 138 (11.25) 3 202 (16.46) Total 1,227 (100) 382 (100) GRI_Score measures the level of corporate social responsibility disclosure and takes a value of 3 for firms disclosing GRI at the A level, 2 for GRI B level disclosure, 1 for GRI C level disclosure, 0 for not reporting using the GRI disclosure framework. GRI_Validation is a dummy variable coded 1 when the GRI disclosure level is validated by an independent firm, and zero otherwise. 43 Panel C – Descriptive statistics for country and firm-level variables Variable Mean Std. Dev. Min Max Law 3.480 1.023 1.000 5.000 Voice 1.243 0.488 -0.986 1.618 Gov_Eff 1.481 0.555 -0.425 2.338 Reg_Qual 1.430 0.465 -0.453 1.924 Press 7.796 10.479 0.000 60.880 81.682 7.806 58.100 95.500 Size 16.144 1.273 12.585 19.385 ROA 0.071 0.075 -0.339 0.631 Fin (millions) 0.414 2.891 0.000 45.800 B_M 0.554 0.442 -0.454 4.687 Lev 0.264 0.161 0.0000 1.197 Internat 0.572 0.311 0.0000 1.536 Volat 26.760 8.488 11.990 73.880 New 0.512 0.153 0.135 1.000 Capex 0.102 0.203 0.000 3.786 Ind_Sens 0.334 0.472 0.000 1.000 60.602 366.757 0.020 7,810.670 3.968 34.025 -224.977 813.341 34.596 280.892 -0.566 5,581.547 Country level measures Env_Perf Firm level measures Share_Price EPS BV_pS N = 1,227, the number of observations with no missing data for all variables in this table. Law is an anti-self-dealing score, based on Djankov et al. [2008]. The index considers (1) vote by mail; (2) shares not blocked or deposited; (3) cumulative voting; (4) oppressed minority; (5) pre-emptive rights; and (6) capital to call meeting. Voice, Gov_Eff, and Reg_Qual are the following Worldbank measures as reported by Kaufman et al. [2010]: Voice and Accountability, Government effectiveness, and Regulatory quality. Press is freedom of the press, as measured by Reporters without borders. Env_Perf is the environmental performance index of the Yale Centre for environmental law and policy. Size is measured as logarithm of total assets. ROA is net income (beipd) / beginning total assets. Fin is financing, measured as (sales less purchases) of common and preferred shares plus change in long term debt. B_M is book to market value of equity ratio. Lev is the leverage ratio, calculated as total debt divided by total assets. Internat is a measure of international trade, calculated as the percentage of sales made in countries other than the firms’ headquarters firm. Volat is share price volatility. New is firms’ asset newness, measured as net property plant and equipment (PPE)/ gross PPE. Capex is capital expenditure, measured as capital expenditures / sales. Ind_Sens is industry sensitivity, an indicator variables coded one for firms operating in environmentally sensitive industries, and zero otherwise. Sensitive industries are identified by the SIC codes described in De Villiers et al. [2011]. Share_Price is the closing market value per share, three months after fiscal year end. EPS is earnings per share. BV_pS is book value per share, taken at the end of the fiscal year. 44 TABLE 2 Correlation tables Panel A – Correlations between GRI_Score and country variables GRI_Score GRI_Score Law Voice Gov_Eff Reg_Qual Press Env_Perf 1 Law 0.044 1 Voice 0.104 -0.044 1 Gov_Eff 0.029 -0.038 0.875 1 Reg_Qual 0.069 0.136 0.916 0.902 1 Press -0.119 0.060 -0.965 -0.822 -0.886 1 Env_Perf -0.059 0.046 0.383 0.443 0.315 -0.352 1 The correlations in bold are statistically significant at the 5% level. GRI_Score measures the level of corporate social responsibility disclosure and takes a value of 3 for firms disclosing GRI at the A level, 2 for GRI B level disclosure, 1 for GRI C level disclosure, 0 for not reporting using the GRI disclosure framework. Law is an anti-self-dealing score, based on Djankov et al. [2008]. The index considers (1) vote by mail; (2) shares not blocked or deposited; (3) cumulative voting; (4) oppressed minority; (5) pre-emptive rights; and (6) capital to call meeting. Voice, Gov_Eff, and Reg_Qual are the following Worldbank measures as reported by Kaufman et al. [2010]: Voice and Accountability, Government effectiveness, and Regulatory quality. Press is freedom of the press, as measured by Reporters without borders. Env_Perf is the environmental performance index of the Yale Centre for environmental law and policy. 45 Panel B – Correlations between GRI_Score and firm-level variables GRI_Score Size ROA FIN B_M LEV Internat Volat New Size 0.305 1 ROA -0.032 -0.303 1 FIN 0.078 0.187 -0.005 1 B_M 0.094 0.248 -0.289 0.009 1 LEV 0.072 0.219 -0.332 0.131 -0.009 Internat 0.056 0.121 -0.039 0.006 -0.071 -0.161 Volat -0.066 -0.187 -0.118 -0.071 0.224 -0.123 0.063 1 New -0.046 0.023 0.028 0.093 0.104 0.178 -0.134 0.088 Capex 0.056 0.013 -0.061 0.043 0.021 0.224 -0.088 -0.014 0.391 1 Ind_Sens 0.170 0.128 0.000 0.101 0.128 -0.068 0.044 0.044 0.165 0.082 1 Share_Price -0.033 0.034 0.019 -0.012 0.021 -0.002 -0.003 -0.010 0.022 0.003 -0.011 1 EPS -0.023 0.041 0.047 -0.019 0.010 -0.000 0.014 -0.013 0.016 0.015 -0.013 0.851 1 BV_pS -0.017 0.072 -0.023 -0.009 0.062 0.015 -0.012 0.001 0.037 0.011 -0.026 0.939 0.771 Capex Ind_Sens Price EPS BV_pS 1 1 1 1 The correlations in bold are statistically significant at the 5% level. GRI_Score measures the level of corporate social responsibility disclosure and takes a value of 3 for firms disclosing GRI at the A level, 2 for GRI B level disclosure, 1 for GRI C level disclosure, 0 for not reporting using the GRI disclosure framework. Size is measured as logarithm of total assets. ROA is net income (beipd) / beginning total assets. Fin is financing, measured as (sales less purchases) of common and preferred shares plus change in long term debt. B_M is book to market value of equity ratio. Lev is the leverage ratio, calculated as total debt divided by total assets. Internat is a measure of international trade, calculated as the percentage of sales made in countries other than the firms’ headquarters firm. Volat is share price volatility. New is firms’ asset newness, measured as net property plant and equipment (PPE)/ gross PPE. Capex is capital expenditure, measured as capital expenditures / sales. Ind_Sens is industry sensitivity, an indicator variables coded one for firms operating in environmentally sensitive. 46 industries, and zero otherwise. Sensitive industries are identified by the SIC codes described in De Villiers et al. [2011]. Share_Price is the closing market value per share, three months after fiscal year end. EPS is earnings per share. BV_pS is book value per share, taken at the end of the fiscal year. 47 TABLE 3 Predisposition towards CSR disclosures (Ordered Logit models) Panel A – Individual tests of significance of each country-level and firm-level variable using an ordered logit model GRI_Score Country-level measures Firm-level measures Law Expected sign + Voice + Gov_Eff + Reg_Qual + Press - Env_Perf ? Size + ROA + Fin + B_M + Lev + Internat + Volat ? New - Capex + Ind_Sens + Coef (P-value) 0.083 (0.082) Coef (P-value) Coef (P-value) Coef (P-value) Coef (P-value) Coef (P-value) Coef (P-value) Coef (P-value) Coef (P-value) Coef (P-value) 0.593 (0.000) 0.156 (0.082) 0.371 (0.005) -0.032 (0.000) -0.016 (0.029) 0.503 (0.000) -0.754 (0.177) 0.000 (0.016) 0.419 (0.001) 0.840 (0.013) 0.428 (0.015) -0.016 (0.036) -0.657 (0.053) 0.510 (0.025) 0.755 (0.000) 48 N = 1,227. The table reports the results of ordered logit regressions for each variable individually with GRI_Score as the dependent variable. P-values are one-tailed for variables with predicted signs. GRI_Score is the dependent variable and measures the level of corporate social responsibility disclosure and takes a value of 3 for firms disclosing GRI at the A level, 2 for GRI B level disclosure, 1 for GRI C level disclosure, 0 for not reporting using the GRI disclosure framework. Law is an anti-self-dealing score, based on Djankov et al. [2008]. The index considers (1) vote by mail; (2) shares not blocked or deposited; (3) cumulative voting; (4) oppressed minority; (5) pre-emptive rights; and (6) capital to call meeting. Voice, Gov_Eff, and Reg_Qual are the following Worldbank measures as reported by Kaufman et al. [2010]: Voice and Accountability, Government effectiveness, and Regulatory quality. Press is freedom of the press, as measured by Reporters without borders. Env_Perf is the environmental performance index of the Yale Centre for environmental law and policy. Size is measured as logarithm of total assets. ROA is net income (beipd) / beginning total assets. Fin is financing, measured as (sales less purchases) of common and preferred shares plus change in long term debt. B_M is book to market value of equity ratio. Lev is the leverage ratio, calculated as total debt divided by total assets. Internat is a measure of international trade, calculated as the percentage of sales made in countries other than the firms’ headquarters firm. Volat is share price volatility. New is firms’ asset newness, measured as net property plant and equipment (PPE)/ gross PPE. Capex is capital expenditure, measured as capital expenditures / sales. Ind_Sens is industry sensitivity, an indicator variables coded one for firms operating in environmentally sensitive industries, and zero otherwise. Sensitive industries are identified by the SIC codes described in De Villiers et al. [2011]. 49 Panel B – Testing each country-level variable together with all firm-level variables using ordered logit models GRI_Score Expected sign Law Coef. 0.137 P-value 0.020 Voice Coef. P-value Gov_Eff Coef. P-value Reg_Qual Coef. P-value Press Coef. P-value Law + Voice + Gov_Eff + Reg_Qual + Press - Env_Perf ? Size + 0.499 0.000 0.527 0.000 0.521 0.000 0.530 0.000 0.543 ROA + 3.174 0.001 3.989 0.000 3.504 0.001 3.754 0.000 Fin + 0.000 0.450 0.000 0.500 0.000 0.478 0.000 B_M + 0.291 0.026 0.403 0.005 0.314 0.018 Lev + 1.003 0.017 1.104 0.010 1.121 Internat + 0.261 0.118 -0.010 0.483 Volat ? 0.001 0.864 0.009 New - -2.254 0.000 Capex + 0.920 Ind_Sens + 0.735 0.893 Env_Perf Coef. P-value 0.000 0.361 0.004 0.713 0.000 -0.048 0.000 -0.016 0.049 0.000 0.500 0.000 4.059 0.000 3.155 0.001 0.466 0.000 0.469 0.000 0.457 0.387 0.006 0.433 0.004 0.216 0.079 0.009 1.075 0.012 1.081 0.012 0.992 0.018 0.120 0.300 -0.019 0.467 -0.038 0.435 0.314 0.076 0.316 0.003 0.706 0.005 0.553 0.008 0.359 -0.001 0.866 -1.618 0.001 -1.821 0.000 -1.910 0.000 -1.575 0.001 -2.058 0.000 0.002 0.847 0.003 0.934 0.002 0.900 0.002 0.841 0.003 0.936 0.002 0.000 0.810 0.000 0.761 0.000 0.787 0.000 0.789 0.000 0.720 0.000 N 1,227 1,227 1,227 1,227 1,227 1,227 Prob > chi2 0.000 0.000 0.000 0.000 0.000 0.000 The table reports the results of ordered logit regressions with errors clustered by country. P-values are one-tailed for variables with predicted signs. GRI_Score measures the level of corporate social responsibility disclosure and takes a value of 3 for firms disclosing GRI at the A level, 2 for GRI B level disclosure, 1 for GRI C level disclosure, 0 for not reporting using the GRI disclosure framework. Law is an anti-self-dealing score, based on Djankov et al. [2008]. The index considers (1) vote by mail; (2) shares not blocked or deposited; (3) cumulative voting; (4) oppressed minority; (5) pre-emptive rights; and (6) capital to call meeting. Voice, Gov_Eff, and Reg_Qual are the following Worldbank measures as reported by Kaufman et al. [2010]: Voice and Accountability, Government effectiveness, and Regulatory quality. Press 50 is freedom of the press, as measured by Reporters without borders. Env_Perf is the environmental performance index of the Yale Centre for environmental law and policy. Size is measured as logarithm of total assets. ROA is net income (beipd) / beginning total assets. Fin is financing, measured as (sales less purchases) of common and preferred shares plus change in long term debt. B_M is book to market value of equity ratio. Lev is the leverage ratio, calculated as total debt divided by total assets. Internat is a measure of international trade, calculated as the percentage of sales made in countries other than the firms’ headquarters firm. Volat is share price volatility. New is firms’ asset newness, measured as net property plant and equipment (PPE)/ gross PPE. Capex is capital expenditure, measured as capital expenditures / sales. Ind_Sens is industry sensitivity, an indicator variables coded one for firms operating in environmentally sensitive industries, and zero otherwise. Sensitive industries are identified by the SIC codes described in De Villiers et al. [2011]. 51 TABLE 4 Principal components analysis of predisposition towards CSR disclosures Panel A – Eigenvalues of the correlation matrix Principal Variance explained Cumulative (%) variance (%) Eigenvalue components PC1 3.877 64.62 64.62 PC2 1.033 17.22 81.83 PC3 0.816 13.60 95.43 PC4 0.196 3.26 98.69 PC5 0.050 0.83 99.52 PC6 0.029 0.48 100.00 N=1,227 Panel B – Weights of the two country factors Variable PC1 PC2 Law 0.003 0.974 0.019 Voice 0.494 -0.068 0.050 Gov_Eff 0.478 -0.039 0.114 Reg_Qual 0.483 0.112 0.083 -0.481 0.090 0.093 0.251 0.154 0.731 Press Env_Perf Unexplained Law is an anti-self-dealing score, based on Djankov et al. [2008]. The index considers (1) vote by mail; (2) shares not blocked or deposited; (3) cumulative voting; (4) oppressed minority; (5) pre-emptive rights; and (6) capital to call meeting. Voice, Gov_Eff, and Reg_Qual are the following Worldbank measures as reported by Kaufman et al. [2010]: Voice and Accountability, Government effectiveness, and Regulatory quality. Press is freedom of the press, as measured by Reporters without borders. Env_Perf is the environmental performance index of the Yale Centre for environmental law and policy. 52 Panel C – Ordered logit with 2 country factors and firm-level variables GRI_Score Expected sign Coef. P-value Country_Voice&Reg + 0.166 0.000 Country_Law + 0.092 0.086 Size + 0.528 0.000 ROA + 3.740 0.000 Fin + 0.000 0.477 B_M + 0.405 0.005 Lev + 1.094 0.011 Internat + 0.013 0.478 Volat ? 0.007 0.419 New - -1.904 0.000 Capex + 0.885 0.003 Ind_Sens + 0.789 0.000 / cut1 9.783 / cut2 9.970 / cut 3 10.730 N 1,227 Prob > chi2 0.000 The table reports the results of an ordered logit regression with errors clustered by country P-values are one-tailed for variables with predicted signs. GRI_Score measures the level of corporate social responsibility disclosure and takes a value of 3 for firms disclosing GRI at the A level, 2 for GRI B level disclosure, 1 for GRI C level disclosure, 0 for not reporting using the GRI disclosure framework. Country_Voice&Reg is PC1 calculated in Table 4, Panel B. Country_Law is PC2 calculated in Table 4, Panel B. Size is measured as logarithm of total assets. ROA is net income (beipd) / beginning total assets. Fin is financing, measured as (sales less purchases) of common and preferred shares plus change in long term debt. B_M is book to market value of equity ratio. Lev is the leverage ratio, calculated as total debt divided by total assets. Internat is a measure of international trade, calculated as the percentage of sales made in countries other than the firms’ headquarters firm. Volat is share price volatility. New is firms’ asset newness, measured as net property plant and equipment (PPE)/ gross PPE. Capex is capital expenditure, measured as capital expenditures / sales. Ind_Sens is industry sensitivity, an indicator variables coded one for firms operating in environmentally sensitive industries, and zero otherwise. Sensitive industries are identified by the SIC codes described in De Villiers et al. [2011]. 53 TABLE 5 Consequences of CSR disclosures (Share price valuation models) Panel A – Basic model and including Country_Voice&Reg Market price per share Expected Basic model With Country_Voice&Reg sign Coef. P-value Coef. P-value EPS + 1.550 0.000 1.574 0.000 BV_pS + 0.790 0.000 0.793 0.000 GRI_Score + 2.889 0.009 3.664 0.072 GRI_Validation ? -7.312 0.027 -12.871 0.072 Country_Voice&Reg + 1.138 0.110 GRI*Country_Voice&Reg + 0.843 0.026 Size ? -4.104 0.004 -3.129 0.010 Ind_Sens ? 5.558 0.102 5.533 0.072 77.871 0.002 65.974 0.005 Constant Dummies for years Included Dummies for countries Included Included Adjusted R2 0.984 0.980 N 1,521 1,516 The table reports the results of ordinary least squares regressions with errors clustered by country. P-values are one-tailed for variables with predicted signs. The dependent variable is the market price per share, Share_Price. EPS is earnings per share. BV_pS is book value per share, taken at the end of the fiscal year. GRI_Score measures the level of corporate social responsibility disclosure and takes a value of 3 for firms disclosing GRI at the A level, 2 for GRI B level disclosure, 1 for GRI C level disclosure, 0 for not reporting using the GRI disclosure framework. GRI_Validation is a dummy variable coded 1 when the GRI disclosure level is validated by an independent firm, and zero otherwise. Country_Voice&Reg is PC1 calculated in Table 4, Panel B and represent mainly Voice, Gov_Eff, Reg_Qual and Press. Voice, Gov_Eff, and Reg_Qual are the following Worldbank measures as reported by Kaufman et al. [2010]: Voice and Accountability, Government effectiveness, and Regulatory quality. Press is freedom of the press, as measured by Reporters without borders. GRI*Country_Voice&Reg is an interaction term between GRI_Score and Country_Voice&Reg. Size is measured as logarithm of total assets. Ind_Sens is industry sensitivity, an indicator variables coded one for firms operating in environmentally sensitive industries, and zero otherwise. Sensitive industries are identified by the SIC codes described in De Villiers et al. [2011]. 54 Panel B – Share price analysis with Country_Voice&Reg decomposed into four main parts Voice Market price per Gov_Eff Reg_Qual Press Expected Coef. P-value Coef. P-value P-value Coef. P-value + 1.574 0.000 1.575 0.000 1.575 0.000 1.575 0.000 BV_pS + 0.793 0.000 0.792 0.000 0.793 0.000 0.793 0.000 GRI_Score + 0.859 0.350 1.742 0.243 0.348 0.454 4.829 0.055 GRI_Validation ? -13.432 0.072 -12.651 0.075 -13.663 0.082 -13.717 0.076 Voice + 4.330 0.082 GRI * Voice + 2.465 0.059 Gov_Eff + 5.102 0.086 GRI * Gov_Eff + 1.476 0.066 Reg_Qual + 2.251 0.264 GRI * Reg_Qual + 2.719 0.016 Press - -0.158 0.108 GRI * Press - -0.103 0.043 Size ? -3.250 0.008 -3.145 0.008 -3.147 0.007 -3.179 0.007 Ind_Sens ? 5.808 0.028 5.695 0.060 4.911 0.082 5.396 0.066 62.369 0.002 58.801 0.001 63.202 0.001 67.950 0.004 share sign EPS Constant Dummies for years Coef. Included Included Included Included Adj. R2 0.980 0.980 0.980 0.980 N 1,521 1,521 1,521 1,520 The table reports the results of ordinary least squares regressions with errors clustered by country. P-values are one-tailed for variables with predicted signs. The dependent variable is the market price per share, Share_Price. EPS is earnings per share. BV_pS is book value per share, taken at the end of the fiscal year. GRI_Score measures the level of corporate social responsibility disclosure and takes a value of 3 for firms disclosing GRI at the A level, 2 for GRI B level disclosure, 1 for GRI C level disclosure, 0 for not reporting using the GRI disclosure framework. GRI_Validation is a dummy variable coded 1 when the GRI disclosure level is validated by an independent firm, and zero otherwise. Voice, Gov_Eff, and Reg_Qual are the following Worldbank measures as reported by Kaufman et al. [2010]: Voice and Accountability, Government effectiveness, and Regulatory quality. Press is freedom of the press, as measured by Reporters without borders. GRI * denotes interaction terms between GRI_Score and the country level measure indicated. Size is measured as logarithm of total assets. Ind_Sens is industry sensitivity, an indicator variables coded one for firms operating in environmentally sensitive industries, and zero otherwise. Sensitive industries are identified by the SIC codes described in De Villiers et al. [2011]. 55 TABLE 6 Consequences of CSR disclosures (Basic model decomposed by year) 2007 Market price per share 2008 2009 2010 Expected Coef. P-value Coef. P-value Coef. P-value Coef. P-value sign EPS + 0.703 0.294 3.340 0.000 4.875 0.000 5.964 0.006 BV_pS + 1.635 0.000 0.649 0.000 0.962 0.000 0.139 0.332 GRI_Score + 4.142 0.018 -0.089 0.472 4.165 0.020 2.215 0.065 GRI_Validation ? -10.004 0.107 -1.641 0.506 -10.543 0.005 -2.777 0.329 Size ? -5.093 0.011 -2.178 0.004 -4.889 0.010 -4.366 0.002 Ind_Sens ? 3.983 0.338 3.552 0.050 3.998 0.257 3.599 0.113 79.074 0.011 18.399 0.126 86.392 0.000 77.467 0.001 Constant Dummies for Included Included Included Included 0.996 0.869 0.989 0.995 371 380 375 379 countries Adj. R2 N The table reports the results of ordinary least squares regressions with errors clustered by country. P-values are one-tailed for variables with predicted signs. The dependent variable is the market price per share, Share_Price. EPS is earnings per share. BV_pS is book value per share, taken at the end of the fiscal year. GRI_Score measures the level of corporate social responsibility disclosure and takes a value of 3 for firms disclosing GRI at the A level, 2 for GRI B level disclosure, 1 for GRI C level disclosure, 0 for not reporting using the GRI disclosure framework. GRI_Validation is a dummy variable coded 1 when the GRI disclosure level is validated by an independent firm, and zero otherwise. Size is measured as logarithm of total assets. Ind_Sens is industry sensitivity, an indicator variables coded one for firms operating in environmentally sensitive industries, and zero otherwise. Sensitive industries are identified by the SIC codes described in De Villiers et al. [2011]. 56 TABLE 7 Consequences of CSR disclosures (Share price valuation models excluding firms with no GRI_Score) Panel A – Basic model and including Country_Voice&Reg (Sample excluding firms with no GRI_Score) Basic model Market price per share With Country_Voice&Reg Expected Coef. P-value Coef. P-value sign EPS + 3.465 0.001 2.786 0.001 BV_pS + 0.504 0.001 0.605 0.000 GRI_Score + 3.518 0.034 3.042 0.021 GRI_Validation ? -5.990 0.018 -7.921 0.017 Country_Voice&Reg + -0.934 0.152 GRI*Country_Voice&Reg + 1.116 0.025 Size ? -1.999 0.018 -0.759 0.351 Ind_Sens ? 3.630 0.248 3.586 0.099 35.372 0.004 23.148 0.058 Constant Dummies for years Included Dummies for countries Included Adjusted R2 N Included 0.998 0.998 431 430 The table reports the results of ordinary least squares regressions with errors clustered by country. P-values are one-tailed for variables with predicted signs. The dependent variable is the market price per share, Share_Price. EPS is earnings per share. BV_pS is book value per share, taken at the end of the fiscal year. GRI_Score measures the level of corporate social responsibility disclosure and takes a value of 3 for firms disclosing GRI at the A level, 2 for GRI B level disclosure, 1 for GRI C level disclosure, 0 for not reporting using the GRI disclosure framework. GRI_Validation is a dummy variable coded 1 when the GRI disclosure level is validated by an independent firm, and zero otherwise. Country_Voice&Reg is PC1 calculated in Table 4, Panel B and represent mainly Voice, Gov_Eff, Reg_Qual and Press. Voice, Gov_Eff, and Reg_Qual are the following Worldbank measures as reported by Kaufman et al. [2010]: Voice and Accountability, Government effectiveness, and Regulatory quality. Press is freedom of the press, as measured by Reporters without borders. GRI*Country_Voice&Reg is an interaction term between GRI_Score and Country_Voice&Reg. Size is measured as logarithm of total assets. Ind_Sens is industry sensitivity, an indicator variables coded one for firms operating in environmentally sensitive industries, and zero otherwise. Sensitive industries are identified by the SIC codes described in De Villiers et al. [2011]. 57 Panel B – Interactions with Country_Voice&Reg decomposed into four main parts (Sample excl. firms with no GRI_Score) Voice Market price per share Expected sign Coef. Pvalue EPS + 2.771 0.001 BV_pS + 0.607 GRI_Score + GRI_Validation Gov_Eff Reg_Qual Press Pvalue Coef. Pvalue Coef. P-value 2.835 0.000 2.497 0.000 2.478 0.000 0.000 0.598 0.000 0.647 0.000 0.650 0.000 -1.804 0.240 -1.915 0.145 -3.922 0.122 3.695 0.016 ? -8.105 0.016 -7.563 0.021 -8.038 0.019 -8.219 0.018 Voice + -2.374 0.239 GRI * Voice + 3.797 0.050 Gov_Eff + -2.611 0.205 GRI * Gov_Eff + 3.470 0.013 Reg_Qual + -5.152 0.135 GRI * Reg_Qual + 4.728 0.016 Press - 0.057 0.289 GRI * Press - -0.125 0.033 Size ? -0.822 0.319 -0.706 0.377 -1.033 0.252 -0.958 0.294 Ind_Sens ? 3.738 0.093 3.177 0.128 2.674 0.121 3.136 0.093 27.523 0.030 25.928 0.020 26.260 0.098 17.055 0.265 Constant Dummies for years Adj. R2 N Coef. Included Included Included Included 0.998 0.998 0.998 0.998 430 430 428 428 The table reports the results of ordinary least squares regressions with errors clustered by country. P-values are one-tailed for variables with predicted signs. The dependent variable is the market price per share, Share_Price. EPS is earnings per share. BV_pS is book value per share, taken at the end of the fiscal year. GRI_Score measures the level of corporate social responsibility disclosure and takes a value of 3 for firms disclosing GRI at the A level, 2 for GRI B level disclosure, 1 for GRI C level disclosure, 0 for not reporting using the GRI disclosure framework. GRI_Validation is a dummy variable coded 1 when the GRI disclosure level is validated by an independent firm, and zero otherwise. Voice, Gov_Eff, and Reg_Qual are the following Worldbank measures as reported by Kaufman et al. [2010]: Voice and Accountability, Government effectiveness, and Regulatory quality. Press is freedom of the press, as measured by Reporters without borders. GRI * denotes interaction terms between GRI_Score and the country level measure indicated. Size is measured as logarithm of total assets. Ind_Sens is industry sensitivity, an indicator variables coded one for firms operating in environmentally sensitive industries, and zero otherwise. Sensitive industries are identified by the SIC codes described in De Villiers et al. [2011]. 58 TABLE 8 Consequences of CSR disclosures (Alternative model – Tobin’s Q) Basic model With Country_Voice&Reg Tobin’s Q Expected sign Coef. P-value Coef. P-value GRI_Score + 0.045 0.013 0.027 0.125 GRI_Validation ? -0.079 0.321 -0.104 0.290 Country_Voice&Reg + 0.032 0.006 GRI* Country_Voice&Reg + 0.014 0.017 Size ? -0.217 0.000 -0.201 0.000 ROA ? 5.435 0.000 5.521 0.000 Capex ? -0.206 0.016 -0.085 0.113 LT_debt ? 0.128 0.463 0.135 0.445 Internat ? -0.028 0.426 0.003 0.939 3.879 0.000 4.371 0.000 Constant Dummies for years Included Dummies for countries Included Included R2 0.595 0.556 N 1,281 1,275 The table reports the results of ordinary least squares regressions with errors clustered by country. P-values are one-tailed for variables with predicted signs. The dependent variable is Tobin’s Q. GRI_Score measures the level of corporate social responsibility disclosure and takes a value of 3 for firms disclosing GRI at the A level, 2 for GRI B level disclosure, 1 for GRI C level disclosure, 0 for not reporting using the GRI disclosure framework. GRI_Validation is a dummy variable coded 1 when the GRI disclosure level is validated by an independent firm, and zero otherwise. Country_Voice&Reg is PC1 calculated in Table 4, Panel B and represent mainly Voice, Gov_Eff, Reg_Qual and Press. Voice, Gov_Eff, and Reg_Qual are the following Worldbank measures as reported by Kaufman et al. [2010]: Voice and Accountability, Government effectiveness, and Regulatory quality. Press is freedom of the press, as measured by Reporters without borders. GRI*Country_Voice&Reg is an interaction term between GRI_Score and Country_Voice&Reg. Size is measured as logarithm of total assets. ROA is net income (beipd) / beginning total assets. Capex is capital expenditure, measured as capital expenditures / sales. LT_debt is long term debt scaled by total assets. Internat is a measure of international trade, calculated as the percentage of sales made in countries other than the firms’ headquarters firm. 59