Corporate social responsibility, country sustainability as moderator, and credit ratings:

Corporate social responsibility, country

sustainability as moderator, and credit ratings:

Testing the relationship using disaggregated scores for Eurozone firms

A.H.J. Deten

S2760959

a.h.j.deten@student.rug.nl

13-06-2016

MSc Finance

Faculty of Economics and Business

University of Groningen

MSc International Financial Management

Faculty of Economics and Business

University of Groningen

MSc Business and Economics

Department of Business Studies

Uppsala University

Supervisor: Dr. S.G. Ursu

Abstract

In this study the impact of corporate social responsibility, measured by ESG scores, on the Fitch credit ratings is examined. Ordered logistic regressions are used as the main regression method. Additionally, pooled OLS and fixed effects regressions are performed on an alternative measure of the Fitch credit ratings. Evidence is found for a positive impact of the ESG score on the credit ratings of Eurozone firms. Looking at the impact of the individual dimensions of ESG, it appears that the social dimension is the key driver in the CSR-credit ratings relationship. The other two dimensions, environmental and corporate governance, provide only weak evidence. The potential moderating role of country sustainability is investigated by dividing the sample into above average scoring countries and below average scoring countries on sustainability. For the firms from above average scoring countries similar results as for the total sample are found, but the impact and the significance is now even stronger. In contrast, no evidence is found for a significant relationship for the firms from countries that score below average on sustainability. In addition to the linear models, this research also investigated potential non-linear relationships. Only strong evidence is found for an inverse U-shaped relationship, but only for the environmental dimension, and only in the countries that score below average on sustainability.

INTRODUCTION

This paper is focusing on the effects of corporate social responsibility (CSR) on the credit ratings of firms. By looking at the individual dimensions of CSR, this paper contributes to existing literature by identifying what the impacts of the individual dimensions of CSR are on credit ratings, while taking into account the moderating role of sustainability at the country level.

In order to do so, it is first of all important to make clear what is meant by CSR. In this paper the definition of Bénabou and Tirole is used. They state that CSR is about sacrificing profits in the social interest, whereby the firm “must go beyond its legal and contractual obligations, on a voluntary basis.” (Bénabou and Tirole, p.2, 2010) CSR is costly, and according to the classical shareholder view, firms should not engage in CSR if this does not increase firm value. However, the growing pressure from the society and the support for the stakeholder view has led to firms that engage more and more in CSR projects. Recently, a new view has emerged that has led to a convergence of the shareholder- and stakeholder view, the so-called enlightened shareholder value, where it is argued that the creation of shareholder value can also go together with serving the interest of other stakeholders. According to this view, investments in CSR activities should be made by the firm if it creates value for the shareholders and all other stakeholders (Fatemi, Fooladi, Tehranian, 2015). Investments in CSR can for example go together with serving the interest of other stakeholders when CSR is able to generate long-term shareholder wealth or when it is able to reduce risks for example. Therefore, an important question is whether these investments in CSR projects are good or bad for firms. The umbrella term for the dependent variables in most of the existing literature is corporate financial performance (CFP). Although this research is specifically focused on credit ratings, one specific element of CFP, other ways to test the relationship between CSR and CFP are also

discussed in this paper to get a clear understanding about the effects of CSR.

The existing literature shows inconsistent results about the relationship between CSR and CFP, varying from positive to negative or even neutral or mixed relationships. This inconsistency becomes very clear in the review of Peloza (2009) for example. The fact that there is an enormous amount of different measures of CFP used in the existing literature could be one of the reasons for the inconsistent results, which is also indicated by Peloza. In contrast to most of the existing literature, this paper is focusing on intermediate cost-based measures instead of the widely used end state measures. The major advantage of using intermediate cost-based measures is that they make it easier to track and measure cost-based effects of CSR.

One of these cost-based measures is the cost of capital of a firm, consisting of the cost of equity and the cost of debt. In this thesis, the focus will be especially on the cost of debt, which will be measured by looking at credit ratings of the firm. The credit rating is an evaluation of the credit risk of a firm. When the firm receives a low credit rating by the rating agency, it indicates that the likelihood of default is relatively high. In this way the credit rating is linked to the cost of debt, because the providers of credit will have to ask a premium or a so-called ‘spread’ above the risk-free rate when providing money to a firm with a low credit rating, which in turn leads to relatively high cost of debt compared to firms with high credit ratings and lower credit spreads. Ultimately, the higher cost of debt will result in a lower firm value.

countries that score high on sustainability than in countries that score low on sustainability. This question is addressed by taking into account the level of sustainability of the country of residence of the particular firms in our sample.

GOAL & RELEVANCE

CSR is becoming more and more important nowadays. According to Bénabou and Tirole (2010), this trend can be explained by the following four factors: “(i) social responsibility is likely to be a normal good; (ii) information about companies’ practices throughout the world has become much more accessible and quick to travel; (iii) the scope of environmental and social externalities exerted by multinationals in less developed, more laxly regulated countries is likely to have expanded in pace with globalization; (iv) the long-run cost of atmospheric pollution (e.g. global warming), or at least the public’s awareness of it, has risen significantly” (Bénabou and Tirole, p.1-2, 2010). Although the recognition of the importance of CSR, the academic world has not yet come to a consistent conclusion about the effects of CSR on CFP. Another issue with the existing literature on CSR and CFP is that most literature is focused on finding a relationship between CSR and end state measures of CFP, such as return on assets, return on equity and stock returns. The paper by Stellner et al. (2015) is one of the few papers that focuses on the relationship between CSR and credit risk, an intermediate cost-based measure of CFP. Stellner et al. showed that high CSR results in better credit ratings and lower z-spreads in countries with above average sovereign CSR scores. However, the paper by Stellner et al. used equal-weighted ESG (environmental, social, and governance) scores in order to investigate the relationship. In order to get a better understanding of the impacts of CSR, this paper investigates the impact of the individual dimensions of CSR (environmental, social, and governance) on the credit ratings of Eurozone firms. The

importance of investigating the individual dimensions of CSR is supported by the recent paper of Nollet et al. (2016), which showed that, in their research on the relationship between CSR and financial performance, especially the governance dimension is a key driver affecting the CSR-CFP relationship.

The main goal of this paper will therefore be to contribute to the existing knowledge about the CSR-credit rating relationship by identifying what the impacts of the individual dimensions of CSR are on credit ratings of Eurozone firms, while taking into account the possible moderating effect of country-level sustainability.

This paper will also be relevant for practitioners, because it is of high importance for them to know whether or not, in what circumstances, in which areas they should invest in CSR, and what the effects are on their credit ratings, the impact of their credit rating on the cost of debt, and ultimately the impact on the firm value.

RESEARCH QUESTIONS

This research will focus on one central research question, which will be answered by elaborating a number of sub-questions.

CENTRAL RESEARCH QUESTION

The central research question in this paper will be: “What are the effects of corporate social responsibility on the credit ratings of Eurozone firms?”

SUB-QUESTIONS

The central research question will be answered through a number of sub-questions, which are as follows:

SQ1. “How does corporate social responsibility affect corporate credit ratings in the Eurozone?”

SQ3. “Are firms with high CSR scores rewarded with higher credit ratings if their country of residence scores above average on sustainability?”

SQ4. “Are firms with high CSR scores punished with lower credit ratings if their country of residence scores below average on sustainability?”

SQ5. “Is the relationship between corporate social responsibility and credit ratings U-shaped?”

The major method used to answer these questions is ordered logistic regression, and the final dataset that is used consist of 141 firms from 11 Eurozone countries, with data from the years 2006 till 2014. More details about the methods and the data are provided in the methodology section.

In the next section the literature review is performed, after which the hypotheses are developed. This section is followed by the methodology in which the methods and the data are described. The methodology is followed by the results section in which the results and additional robustness checks are described. The final sections of this thesis are the conclusion, the appendices and finally the references.

LITERATURE REVIEW

There is a lot of research on the CSR-CFP relationship available, however, there is no consistent result. Some papers indicate that there is a positive relationship between CSR activities and firm performance (63%), other papers indicate that there is a negative relationship (15%), and there are even papers that show a neutral or mixed relationship (22%) (Peloza, 2009). One explanation for the inconsistency is, according to Peloza, that the majority of the papers are measuring end state measures of CFP, like ROA, ROE and stock returns. This makes it harder to identify the effects of CSR, because CSR projects “tend to be lost in the noise of hundreds or thousands of other firm initiatives that are unrelated to social performance.” (Peloza, p.1524, 2009)

(annual excess stock returns), and they found only significant results for the accounting-based measures. Nollet et al. expect that this could be due to CSR policies that change only little over time having too little impact on excess stock returns. Another explanation could be that share prices are affected by a lot of other factors. Fatemedi, Fooladi and Tehranian (2015) also suggest a non-linear relationship. They expect that certain levels of CSR expenditure can lead to positive effects on the firm value. Furthermore, they believe that this relationship can be industry and firm dependent.

We have now discussed some of the different impacts of CSR on CFP found in the literature, an important question is of course why CSR will be able to improve CFP, if it does so? Saeidi et al. (2015) tested three particular factors that could influence the relationship between CSR and firm performance, namely sustainable competitive advantage, reputation, and customer satisfaction. Saeidi et al. showed that CSR can increase firm performance through improving all three factors mentioned above. Gregory, Tharyan, and Whittaker (2014) identified three value-creating sources, namely profitability, long-term growth and the cost of capital. They conclude that most of the aspects of CSR are valued positively by markets. Especially CSR performance associated with better long-term growth prospects has effects on firm valuation, and the lower cost of equity capital to less extent. Fatemedi, Fooladi and Tehranian (2015) developed a model to analyse the effects of CSR engagement on firm value. They conclude that probability of survival can be increased by CSR expenditures and that the cost of capital can be reduced. They also give examples of positive effects of CSR, for example “the ability to secure a more loyal customer base, hire and retain more dedicated workforce, avoid the costs associated with adverse actions by labour unions, consumer-advocacy groups, or governmental agencies empowered to monitor its activities” (Fatemedi, Fooladi, and Tehranian, p.190, 2015). Ho (2010) also states that serving the interest of stakeholders can be seen as a way

of generating long-term shareholder wealth and improving portfolio- and firm-level risk assessment. By looking at the reasons why CSR has an impact on CFP, we shift our focus towards more intermediate measures, like for example the cost of capital. If CSR is able to decrease the cost of capital for example, this will ultimately result in better CFP and higher values of firms. We will now continue with discussing papers that focus on those intermediate measures of CFP and how those measures are influenced by CSR.

One way of looking at intermediate measure of CFP is to investigate how firm risk is affected by CSR. Jo and Na (2012) found that CSR inversely affects firm risk, in this paper measured by the standard deviation of returns. They also concluded that this effect is more significant for controversial industries. Godfrey et al. (2009) performed an event study in order to investigate the risk management hypothesis. They concluded that activities in CSR can indeed work like an ‘insurance-like’ benefit, however, this effect is only found for the CSR activities aimed at the society, so-called institutional CSR activities. The benefit is not found for CSR activities aimed at trading partners, so-called technical CSR activities. Oikonomou et al. (2012) measured risk with the beta, downside risk measures (downside beta), and utility measures. Their conclusion was that CSR is weakly negative related to systematic firm risk. Another important finding of Oikonomou et al. is that corporate social irresponsibility is strongly positively related to financial risk.

capital. Are high performing firms on CSR able to reduce their capital constraints? And is this relationship influenced by other factors? One element of the cost of capital is the cost of equity, which is investigated by a number of articles. For example by El Ghoul et al. (2011), who found that the relationship is negative, where better CSR leads to lower cost of equity, higher valuation and lower risk. Related to CSR is reputation, the independent variable used in the article by Cao et al. (2014), who found that companies with a better reputation have lower cost of equity capital. However, the research by El Ghoul et al. and Cao et al. are both based on U.S. firms, the effects worldwide might be different, as suggested by Feng et al. (2015), who found that especially in Europe and North America CSR is able to reduce the cost of equity, but no such evidence could be found for most of the Asian countries.

Another way to look at the cost of capital is to focus on the cost of debt. One of the papers that focused on this topic is the paper by Goss and Roberts (2011), who looked at the effects of CSR on the cost of bank loans. Goss and Roberts made their hypotheses based on two developed views. The first one is the risk mitigation view, which states that companies with high scores on CSR have more favourable risk profiles, compared to identical firms with lower scores on CSR. The opposite view is the overinvestment view, which sees investments as a waste of scarce resources. The results of their research show that lenders provide firms with modest incentives to participate in CSR. Lenders demand higher yield spreads from borrowers with the worst records in CSR.

Another way to measure cost of debt is to look at credit ratings. As discussed in the introduction, credit ratings are an evaluation of the credit risk of a firm, and therefore credit ratings influence the cost of debt, which ultimately affects the firm value. The most used credit ratings in the literature are from S&P. Based on U.S. samples, the papers by Attig et al. (2013), Jiraporn et al. (2014), and Oikonomou et al. (2014) found that high scores on CSR results in lower yield spreads and higher

credit ratings, and bad performance is penalized with higher yield spreads and lower credit ratings. In addition, Jiraporn et al. (2014) concludes that, by looking at geographic effects, the CSR policy of a particular firm is influenced by the CSR policies of surrounding firms, which is due to market segmentation, investor clientele, local competition and social interactions (Jiraporn et al., 2014).

As discussed earlier in this section, Nollet et al. (2016) concluded that especially the governance dimension of CSR is a key driver in the CSR-CFP relationship. It is therefore interesting to look at studies that especially focus on the relationship between the corporate governance of a company and the cost of debt, measured by the credit rating of a company. This is done by both the paper by Alali et al. (2012) and the paper by Ashbaugh et al. (2006). Alali et al. found that higher scores on corporate governance are indeed associated with higher credit ratings. Another very interesting finding of this paper is that only firms with the highest scores on corporate governance are associated with an increase in credit ratings. If we look back to earlier in this section, this sounds logical if we take into account the idea of accruing SIC over time, introduced by Barnett (2007), and related to this concept the suggested U-shaped relationship found by Barnett and Salomon (2012). The paper by Ashbaugh et al. (2006) also identified a number of effects of corporate governance on credit ratings. Ashbaugh et al. (2006) showed that credit ratings are negatively associated with the number of blockholders and CEO power on the board. Furthermore, credit ratings are positively related to weaker shareholder rights, overall board independence, board stock ownership and board expertise.

credit ratings and lower credit spreads. However, Stellner et al. also have taken into account the results by Jiraporn et al. (2014), who suggested that CSR policy is influenced by the environment of the firms. They did so by using a dummy variable based on a country-level sustainability score. The results of this next regression showed that high CSR results in better credit ratings and lower z-spreads in countries with above average sovereign CSR scores, which is in line with the risk mitigation view of Goss and Roberts (2011) discussed earlier. Important here is that the relative ESG score of a firm matches the relative ESG score of a country, only then a firm will benefit from it by better ratings and lower spreads. This means that firms with high ESG scores in countries with weak scores on ESG are not rewarded with lower spreads and better ratings, which indicates that for these countries the investments in CSR are seen as a waste of scarce resources, in line with the overinvestment view (Goss and Roberts, 2011).

Based on the findings above, the following hypotheses are developed:

Hypothesis 1: “Higher scores on corporate social responsibility are rewarded with higher credit ratings.” Hypothesis 2: “Higher scores on corporate social responsibility in countries with above average ESG scores is rewarded with higher credit ratings.”

Hypothesis 3: “Higher scores on corporate social responsibility in countries with below average ESG scores is punished with lower credit ratings.”

Hypothesis 4: "Governance is the key driver affecting the CSR-credit ratings relationship.”

Hypothesis 5: “The relationship between CSR and credit ratings is non-linear and U-shaped.”

METHODOLOGY AND DATA

SAMPLE

The sample will consist of the Eurozone publicly listed firms that are available in the Thomson Reuters Asset4 database. The sample period will be from 2006-2014. This nine-year period consists of the most recent data available in the relatively young database of Thomson Reuters. The ESG ratings are on a yearly basis, therefore an annual sample interval will be used.

The Asset4 database consists of around 5000 firms globally. The first step in selecting the firms is to limit the sample to firms from Eurozone countries. There are in total 423 firms from Eurozone countries available in the Thomson Reuters Asset4 database. The next step is to limit the sample to firms from Eurozone countries that do have Fitch credit ratings available in the database. There are 279 firms from Eurozone countries of which no Fitch credit ratings are available, which means that we are left with a sample of 144 firms with Fitch credit ratings. After eliminating 3 more firms with insufficient control variables available, the sample contains 141 firms. The sample of 141 firms in combination with the nine-year period results in 1269 firm-year observations. After eliminating 141 firm-year observations due to missing data, the final total sample that is used in this research contains 1128 firm-year observations.

credit ratings available. We continue with the twelve Eurozone countries that were in the Eurozone in the beginning of our sample period.

The next step is to divide the sample into two groups. In contrast to Stellner et al. (2015), who used the Bloomberg country ESG ratings in their research, this thesis uses the RobecoSAM Country Sustainability Ranking of April 2015 in order to split the sample. The RobecoSAM Country Sustainability Ranking is based on 17 ESG indicators, and is used in practice by investors to compare countries and to make investment decisions on the basis of the sustainability ranking of the countries (Duyvesteyn et al., 2015). Unfortunately no historical rankings were available to me, but I do not expect major impacts on the results by using the April 2015 ranking, because I expect country sustainability rankings to be relatively stable in a period of approximately ten years. This expectation is supported by the findings of Phillis et al. (2011). Phillis et al. found that economic development, a factor that is not expected to change very rapidly, plays a very important role in the overall sustainability rankings, which is illustrated by the high rankings of European countries and countries like Canada, Australia and New Zealand.

The first group is the group of 6 countries that score above average on country sustainability, hereafter called the ‘high-country sample’, which are Ireland, Germany, Austria, Luxembourg, Finland, and the Netherlands. Unfortunately no firms from Luxembourg meet the requirements of having sufficient data, therefore we are left with firms from 5 countries in the high-country sample. The distribution among the high-country sample is as follow: 4 firms from Ireland, 28 firms from Germany, 4 firms from Austria, 5 firms from Finland, and 13 firms from the Netherlands. The high-country sample contains in total 54 firms.

The next group is the group of six countries that score below average on country sustainability, hereafter called the ‘low-country sample’, which are France, Belgium, Spain, Italy, Portugal, and Greece. The distribution

among the low-country sample is as follow: 36 firms from France, 5 firms from Belgium, 17 firms from Spain, 20 firms from Italy, 4 firms from Portugal, and 5 firms from Greece. The low-country sample contains in total 87 firms.

INDEPENDENT VARIABLE

In this study the ratings from the Thomson Reuters Asset4 database are used as a proxy for CSR engagement of global firms. The database comprises of around 5000 global publicly listed firms, over 500 measures and with data available up to 2002. The firms are rated based on environmental, social and governance issues (ESG). The equal weighted average of the ESG scores are used as a proxy for CSR. Furthermore, this paper will also run regressions on each individual score of the ESG in order to check whether the results are equal for all the three aspects of ESG, or if one of the scores has a different relationship with credit ratings. The firms in the database are given raw scores on environmental, social and governance practices. In order to increase the comparability, the raw scores are converted into ratings, so-called z-scores, which results in rankings varying from 0 to 100. This means that the medians and means of the raw scores are converted to a score of around 50. The lowest possible z-score is 0, indicating that the firm performs the worst on that ESG factor compared to all other firms. The highest score is 100, indicating that the firm performs the best on that ESG factor compared to all other firms. The more the score of a firm is above 50, the better the firms is performing compared to the median and median, and the other way around, the more the score is below 50, the more worse a firm is performing compared to the median and mean.

DEPENDENT VARIABLE

Thomson Reuters Asset4 database. There are two measures of the Fitch ratings available in this database. The first measure is the Fitch credit rating value, which is on a 1-24 scale (AAA-DDD). The other measure is a z-score of the Fitch rating, which means that the credit ratings of the different firms in the database are transformed into z-scores, similar to the way in which the ESG scores are calculated. This means that both the value of the credit rating and the z-score are on an ordinal scale. The z-score will be referred to as the Fitch credit score in the rest of this paper. Although the credit rating value is the most used measure in the literature and also the main dependent variable in this research, the credit rating score is also used in this paper to get more robust results, because it provides us with more differentiation. This differentiation arises because the z-score compares firms relatively to other firms in the same year. For example, if the credit rating values (1-24 scale) of the overall sample decline because of the crisis, then a credit rating value of 20 in 2008 results in a better credit rating score than a credit rating value of 20 in 2006 for example. In this way, firms that score relatively high in difficult times like the crisis, are rewarded with relatively higher credit rating scores in these times, while their credit rating value might be constant over the whole sample period. Using both the credit rating value and the credit rating score therefore provides us with a more thorough understanding of the CSR-credit ratings relationship.

Table 1. Means per year of main variables

FITS FITV ESG ENV SOC CGV

Mean 2006 60.058 18.357 73.433 74.377 78.652 51.969 Mean 2007 59.730 18.274 76.312 77.153 81.077 53.503 Mean 2008 61.885 18.123 76.492 77.303 81.682 53.397 Mean 2009 61.987 17.846 79.370 79.477 82.825 59.524 Mean 2010 62.437 17.719 80.207 80.588 82.778 63.087 Mean 2011 57.554 17.022 79.416 81.510 82.363 64.013 Mean 2012 53.883 16.439 81.176 81.688 82.023 62.439 Mean 2013 53.150 16.341 80.318 82.056 82.622 61.667 Mean 2014 53.020 16.358 79.637 82.207 82.863 58.761 The variables are FITS = Fitch credit rating score, FITV = Fitch credit rating value, ESG = weighted average ESG score, ENV = environmental score, SOC = social score, CGV = corporate governance score. All data is obtained from the Asset4 database.

CONTROL VARIABLES

This paper controls for a number of control variables. All control variables are obtained from Datastream. First, firm-specific control variables are taken into account, which are revenues, debt-to-total assets, the interest expenditure, the EBIT margin, CAPEX divided by revenue, return on invested capital (ROIC), and price volatility. The first control variable, revenue, controls for the size of the firms, which is in line with various papers on the CSR-CFP relationship (e.g. Stellner et al. (2015), Nollet et al. (2015)). Revenue is expected to have a positive relationship with credit ratings, the bigger the firm the better it should cope with fluctuations affecting the financial stability of the firm. Revenues are measured in billions. In line with Stellner et al. and Nollet et al., the revenues are not converted into logarithms. Later on in this section we will see that the firms in our sample are around twice as big, but do show very similar standard deviations as in the paper by Stellner et al. If we had found much higher standard deviations in the revenues, we should have thought about converting the revenues into logarithms because of the higher differences between the highest revenues and the bulk of the data.

The leverage of the firm is also widely used as control variable (e.g. Stellner et al. (2015), Nollet et al. (2015), Oikonomou et al. (2014), and Jiraporn et al. (2014)), and is measured by debt-to-total assets in this paper. Leverage is expected to be negatively related to credit ratings, the higher the leverage, the higher the risk, the lower the credit ratings. Interest expenditure is also taken into account, in line with Stellner et al., and is measured by the EBIT divided by the total interest expense ratio, which also measures leverage. We should expect that firms with lower EBIT over interest expenses have higher credit risk and therefore lower credit ratings.

profitability should be rewarded with higher credit ratings.

Higher CAPEX divided by revenue should result in higher credit ratings, because financially healthy firms are more likely to invest more capital into projects (Stellner et al., 2015).

The last firm-specific control variable used in this paper is equity volatility, where higher equity volatility is expected to result in lower credit ratings as indicated by Stellner et al. (2015).

Furthermore, this study controls for a number of factors that reflect the state of the economy in Europe. These factors are in line with Stellner et al. (2015) and are returns on the EURO STOXX 50, the German slope, German risk free rate, and the VDAX. The EURO STOXX 50 is an index with 50 stocks from the twelve Eurozone countries of our analysis. The German risk free rate is measured by the three-month German sovereign bond yield. The German slope represents the difference between the ten-year German sovereign bond yield minus the three-month German sovereign bond yield. These three measures indicate the state of the economy and are therefore expected to be positively related to the credit ratings. The VDAX represents the German volatility index, and since higher volatility leads to higher risks, the returns on the VDAX are expected to be negatively related to credit ratings. (Stellner et al., 2015). Three of the four European control variables are German. Germany is chosen because it is one of the highest rated countries and it is for example often used as indicator of the risk free rate in the Eurozone.

In addition, country dummies are included as control variables, which is also in line with Stellner et al. (2015).

MODELS

Because the dependent variable in this research, the Fitch credit rating value, is on an ordinal scale, it is not appropriate to perform OLS. As indicated in the book of

Brooks (2014), especially with credit ratings the most appropriate regression model is ordered logistic (also called ordered logit). In order to get robust results, the ordered logistic regressions are performed with Huber-White standard errors.

A panel dataset is used, where the model is

FITVi,t = β1ESGi,t-n + β2REVi,t-n + β3EBIi,t-n + β4LEVi,t-n + β5CAPXi,t-n + β6ROICi,t-n + β7INTXi,t-n + β8VOLi,t-n + β9EURXi,t-n + β10RISFi,t-n + β11GSLi,t-n + β12VDAXi,t-n + β13ATi,t-n + β14BEi,t-n + β15DEi,t-n + β16FIi,t-n + β17FRi,t-n + β18GRi,t-n + β19IEi,t-n + β20ITi,t-n + β21NLi,t-n + β22PTi,t-n + εi,t, (1) where FITV is the Fitch credit rating value for firm i at time t. ESG is either the weighted average ESG score, or one of the individual ESG scores. REV represents the control variable revenue, EBI controls for the EBIT margin, LEV controls for leverage by the debt-to-total assets ratio, CAPX controls for the capital expenditure ratio, ROIC controls for the return on invested capital, INTX controls for the interest expense ratio, and VOL controls for price volatility. EURX controls for the EURO STOXX 50 return, RISF controls for the German risk free rate, GSL controls for the German slope, and the last control variable is the VDAX which controls for the VDAX index. The other control variables represent country dummies, where AT represents Austria, BE is Belgium, DE is Germany, FI is Finland, FR is France, GR is Greece, IE is Ireland, IT is Italy, NL is the Netherlands and PT is Portugal. The error term is represented by εi,t.

In order to test whether the relationship between CSR and credit ratings is non-linear and u-shaped, a quadratic term, ESG², is added to the model explained above, as showed in the following equation:

β18FRi,t-n + β19GRi,t-n + β20IEi,t-n + β21ITi,t-n + β22NLi,t-n + β23PTi,t-n + εi,t, (2) Where the abbreviations are the same as in equation 1.

Table 2. Descriptive statistics

The sample period runs from 2006-2014. The variables are FITS = Fitch credit rating score, FITV = Fitch credit rating value, ESG = weighted average ESG score, ENV = environmental score, SOC = social score, CGV = corporate governance score, REV = revenue, EBI = EBIT-margin, LEV = leverage, CAPX = capital expenditure divided by revenues, ROIC = return on invested, INTX = interest expenditure ratio, VOL = equity volatility, EURX = return on the EURO STOXX 50, RISF = 3-month German risk-free rate, GSL = German slope, VDAX = return on the VDAX. All data is obtained from the Asset4 database.

In line with for example Oikonomou et al. (2014), winsorization at a 1% level is applied for all control

variables to deal with outliers that potentially affect the results.

The descriptive statistics of all the variables are shown in table 2. This table shows us for example that weighted average ESG score, the environmental score and the social score have very high medians of around 90 with means of around 80, while the corporate governance variable shows a median and mean of around 60, just like the Fitch credit rating score. In table 1 we can see that the means of the ESG dimensions all follow an increasing trend over the years, while the means of Fitch credit ratings decline between 2006 and 2014.

Table 3 shows the correlation matrix. Not surprisingly, the individual components of ESG are very correlated with each other and with the weighted average ESG score, especially this is true for the environmental and social score, and to a less extent for the corporate governance score. Other variables that show relatively high correlations are the variables that control for the state of the economy (EURX, RISF, GSL, and VDAX), which is also not surprising of course. In the following section, the results of the regressions are shown.

Table 3. Correlation matrix

The sample period runs from 2006-2014. The variables are FITS = Fitch credit rating score, FITV = Fitch credit rating value, ESG = weighted average ESG score, ENV = environmental score, SOC = social score, CGV = corporate governance score, REV = revenue, EBI = EBIT-margin, LEV = leverage, CAPX = capital expenditure divided by revenues, ROIC = return on invested, INTX = interest expenditure ratio, VOL = equity volatility, EURX = return on the EURO STOXX 50, RISF = 3-month German risk-free rate, GSL = German slope, VDAX = return on the VDAX. All data is obtained from the Asset4 database.

Mean Median Max. Min. Std.Dev. N FITS 57.804 60.950 98.400 0.020 26.996 1128 FITV 17.323 18.000 23.000 7.000 2.837 1128 ESG 79.180 90.060 97.900 2.530 24.014 1128 ENV 79.946 90.425 96.740 8.680 23.738 1128 SOC 82.293 91.570 98.530 3.590 21.793 1128 CGV 59.859 63.415 96.650 1.980 24.235 1128 REV 30.515 15.027 180.287 0.282 34.738 1128 EBI 0.127 0.110 0.714 -0.695 0.168 1128 LEV 30.498 30.210 65.793 2.371 15.156 1128 CAPX 0.075 0.042 0.897 0.000 0.122 1128 ROIC 5.995 5.730 30.392 -15.807 6.640 1128 INTX 5.399 3.501 45.921 -12.546 7.908 1128 VOL 26.457 24.490 62.542 13.100 9.046 1128 EURX -0.011 0.106 0.216 -0.528 0.217 1128 RISF 1.232 0.546 3.897 -0.018 1.514 1128 GSL 1.489 1.500 2.672 0.346 0.775 1128 VDAX 0.024 0.181 0.601 -0.551 0.396 1128

FITS FITV ESG ENV SOC CGV REV EBI LEV CAPX ROIC INTX VOL EURX RISF GSL VDAX

Table 4. Ordered logistic regressions on the CSR-credit ratings relationship (full sample)

Independent variable ESG ENV SOC CGV ESG ENV SOC CGV

Dependent variable Credit rating I Credit rating II Credit rating III Credit rating IV Credit rating V Credit rating VI Credit rating VII Credit rating VIII ESG 0.0070** 0.0054** 0.0123*** 0.0056* 0.0086*** 0.0056* 0.0135*** 0.0065** (0.0028) (0.0028) (0.0031) (0.0029) (0.0031) (0.0031) (0.0035) (0.0031) Revenue 0.0312*** 0.0317*** 0.0306*** 0.0318*** 0.0304*** 0.0314*** 0.0300*** 0.0314*** (0.0022) (0.0022) (0.0022) (0.0022) (0.0025) (0.0025) (0.0024) (0.0024) EBIT-margin 6.7188*** 6.7777*** 6.7440*** 6.6578*** 7.5528*** 7.5903*** 7.6216*** 7.4282*** (0.7843) (0.7787) (0.7960) (0.7748) (0.9984) (0.9865) (1.0285) (0.9608) Leverage -0.0418*** -0.0418*** -0.0434*** -0.0414*** -0.0460*** -0.0460*** -0.0476*** -0.0454*** (0.0055) (0.0055) (0.0055) (0.0055) (0.0059) (0.0059) (0.0059) (0.0059) CAPEX/Revenue -1.4091*** -1.3853** -1.4252** -1.4045*** -1.5443** -1.4754** -1.6066** -1.4538** (0.5421) (0.5384) (0.5647) (0.5270) (0.7553) (0.7450) (0.7831) (0.7030) ROIC -0.0696*** -0.0672*** -0.0713*** -0.0671*** -0.0673*** -0.0640*** -0.0682*** -0.0638*** (0.0123) (0.0122) (0.0122) (0.0122) (0.0128) (0.0128) (0.0127) (0.0127) Interest expenditure 0.0100 0.0086 0.0122 0.0101 0.0112 0.0091 0.0131 0.0109 (0.0077) (0.0077) (0.0079) (0.0078) (0.0093) (0.0093) (0.0096) (0.0093) Volatility -0.0828*** -0.0826*** -0.0825*** -0.0849*** -0.0796*** -0.0793*** -0.0795*** -0.0818*** (0.0110) (0.0110) (0.0110) (0.0110) (0.0116) (0.0116) (0.0114) (0.0115) EURO STOXX 50 -0.5127 -0.5402 -0.5210 -0.5176 0.0281 -0.0010 0.0197 0.0253 (0.3306) (0.3314) (0.3310) (0.3292) (0.3506) (0.3524) (0.3507) (0.3495) Risk free rate 0.6664*** 0.6623*** 0.6675*** 0.6655*** 0.6895*** 0.6803*** 0.6826*** 0.6915***

(0.0642) (0.0643) (0.0641) (0.0650) (0.0755) (0.0754) (0.0755) (0.0771) German slope 0.8407*** 0.8397*** 0.8374*** 0.8263*** 0.6193*** 0.6149*** 0.6052*** 0.6095*** (0.1274) (0.1273) (0.1273) (0.1270) (0.1395) (0.1391) (0.1396) (0.1393) VDAX -0.5649*** -0.5808*** -0.5817*** -0.5813*** -0.7665*** -0.7798*** -0.7817*** -0.7842*** (0.2108) (0.2106) (0.2113) (0.2110) (0.2133) (0.2132) (0.2135) (0.2137) Observations 1128 1128 1128 1128 1011 1011 1011 1011 LR statistic 850.2939*** 847.8297*** 861.0291*** 847.6623*** 778.5681*** 773.3669*** 787.9846*** 774.3687***

Country dummies Yes Yes Yes Yes Yes Yes Yes Yes

The sample period runs from 2006-2014. All data is obtained from the Asset4 database. This table represents the coefficients and the standards errors from contemporaneous ordered logistic regressions with Huber-White standard errors (I till IV) and one-year lagged ordered logistic regressions with Huber-White standard errors (V till VIII). The Fitch credit rating value is used as dependent variable. The sample used in this table consists of all eleven countries. Country dummies are included in all the models, but are not reported for reason of readability. *,**,*** indicate significance at the ten percent, five percent, and one percent level, respectively.

RESULTS

The first regressions performed in this research are the ordered logistic regressions with the Fitch credit rating value as the dependent variable. Table 4 shows the outcomes of the regressions on the full sample, table 5 is about the firms from the top half countries, and table 6 shows the firms from the bottom part countries. Important here is that the models I till IV in these three tables represent ordered logistic regressions on the values of the Fitch credit rating while using contemporaneous independent variables, while the independent variables are lagged with one year, in order to minimize problems related to potential endogeneity, in the models V till VIII in these three tables.

Table 4 shows that all the coefficients of the independent variables representing the ESG score are positive, which

Table 5. Ordered logistic regressions on the CSR-credit ratings relationship (high-country sample)

Independent variable ESG ENV SOC CGV ESG ENV SOC CGV

Dependent variable Credit rating I Credit rating II Credit rating III Credit rating IV Credit rating V Credit rating VI Credit rating VII Credit rating VIII ESG 0.0197*** 0.0125** 0.0222*** 0.0125*** 0.0179*** 0.0111** 0.0211*** 0.0114** (0.0048) (0.0051) (0.0050) (0.0048) (0.0049) (0.0054) (0.00540 (0.0051) Revenue 0.0230*** 0.0244*** 0.0238*** 0.0254*** 0.0246*** 0.0262*** 0.0255*** 0.0272*** (0.0024) (0.0025) (0.0025) (0.0024) (0.0030) (0.0031) (0.0031) (0.0030) EBIT-margin 5.1559*** 5.3123*** 5.2872*** 4.9236*** 6.4678*** 6.6179*** 6.6298*** 6.1480*** (1.9056) (1.9249) (1.9747) (1.8178) (2.3056) (2.3313) (2.3442) (2.2798) Leverage -0.0590*** -0.0571*** -0.0621*** -0.0568*** -0.0624*** -0.0610*** -0.0658*** -0.0600*** (0.0085) (0.0083) (0.0087) (0.0079) (0.0094) (0.00910 (0.0096) (0.0087) CAPEX/Revenue -4.5158** -4.8392** -4.2770** -5.1404** -6.4654*** -6.8088*** -6.2389*** -6.9730*** (2.1028) (2.1487) (2.1663) (2.0084) (1.9628) (1.9696) (1.9923) (1.9135) ROIC -0.0368** -0.0305* -0.0366** -0.0333* -0.0417** -0.0361* -0.0406** -0.0366* (0.0176) (0.0174) (0.0178) (0.0173) (0.0193) (0.0193) (0.0193) (0.0193) Interest expenditure 0.0045 0.0007 0.0073 0.0027 0.0026 -0.0010 0.0041 -0.0010 (0.0118) (0.0116) (0.0125) (0.0115) (0.0168) (0.0164) (0.0173) (0.0161) Volatility -0.0673*** -0.0691*** -0.0698*** -0.0743*** -0.0683*** -0.0695*** -0.0707*** -0.0740*** (0.0158) (0.0154) (0.0150) (0.0152) (0.0158) (0.0157) (0.0152) (0.0153) EURO STOXX 50 -0.3522 -0.4446 -0.3715 -0.3627 -0.0235 -0.0888 -0.0036 -0.0432 (0.5838) (0.5873) (0.5834) (0.5864) (0.6023) (0.6080) (0.6010) (0.6034)

Risk free rate 0.5574*** 0.5310*** 0.5421*** 0.5349*** 0.5357*** 0.5133*** 0.5126*** 0.5161***

(0.1068) (0.1067) (0.1054) (0.1060) (0.1179) (0.1174) (0.1172) (0.1173) German slope 0.6490*** 0.6396*** 0.6266*** 0.6015*** 0.5687*** 0.5648*** 0.5350** 0.5321** (0.2060) (0.2026) (0.2032) (0.2003) (0.2185) (0.2150) (0.2178) (0.2140) VDAX -0.4338 -0.4683 -0.4908 -0.4841 -0.5506 -0.5716* -0.5772* -0.5959* (0.3302) (0.3249) (0.3335) (0.3253) (0.3377) (0.3344) (0.3380) (0.3377) Observations 442 442 442 442 398 398 398 398 LR statistic 327.1510*** 315.9315*** 330.8763*** 314.5783*** 307.0633*** 297.8592*** 310.9787*** 296.8425***

Country dummies Yes Yes Yes Yes Yes Yes Yes Yes

The sample period runs from 2006-2014. All data is obtained from the Asset4 database. This table represents the coefficients and the standards errors from contemporaneous ordered logistic regressions with Huber-White standard errors (I till IV) and one-year lagged ordered logistic regressions with Huber-White standard errors (V till VIII). The Fitch credit rating value is used as dependent variable. The sample used in this table consists of the five countries that score above average on sustainability. Country dummies are included in all the models, but are not reported for reason of readability. *,**,*** indicate significance at the ten percent, five percent, and one percent level, respectively.

significance at the 5% level. The results indicate that CSR has a positive impact on credit ratings, which is in line with the risk mitigation view by Goss and Roberts (2011). Based on these results from table 4, evidence is found that supports the first hypothesis. For all four measures of ESG, but especially for the social dimension and the weighted average ESG score, higher scores on corporate social responsibility are rewarded with higher credit ratings for the full sample.

The signs of the coefficients of the control variables revenues, EBIT margin, leverage, interest expenditure, volatility, risk free rate, German slope and the return on the VDAX are as expected by theory. However, the signs of the coefficients CAPEX/Revenue and ROIC are the opposite of what theory expects. This is surprising because high ROIC and a high CAPEX divided by the revenues are seen as a sign of a financial healthy firm,

which should be rewarded with higher credit ratings, however this cannot be seen in these results. The return on the EURO STOXX 50 shows mixed results in the contemporaneous and the one-year lagged regressions. Because the return of the EURO STOXX 50 is based on 50 stocks from the Eurozone, we expected to see a positive relationship with the credit rating, surprisingly this is not the case in 5 out of the 8 models. Important is that most of the control variables show (highly) significant coefficients, which contributes to the quality of the model that is estimated.

Table 6. Ordered logistic regressions on the CSR-credit ratings relationship (low-country sample)

Independent variable ESG ENV SOC CGV ESG ENV SOC CGV

Dependent variable Credit rating I Credit rating II Credit rating III Credit rating IV Credit rating V Credit rating VI Credit rating VII Credit rating VIII ESG -0.0008 0.0002 0.0053 0.0005 0.0026 0.0013 0.0073* 0.0032 (0.0035) (0.0034) (0.00390 (0.0036) (0.0037) (0.0036) (0.0041) (0.00390 Revenue 0.0382*** 0.0379*** 0.0368*** 0.0378*** 0.0361*** 0.0366*** 0.0352*** 0.0359*** (0.0034) (0.0033) (0.0034) (0.0034) (0.0036) (0.0036) (0.0036) (0.0037) EBIT-margin 7.1771*** 7.1916*** 7.2613*** 7.1789*** 8.1003*** 8.0836*** 8.2090*** 8.0209*** (0.9363) (0.9462) (0.9422) (0.9367) (1.1731) (1.1773) (1.1785) (1.15370 Leverage -0.0282*** -0.0283*** -0.0292*** -0.0282*** -0.0336*** -0.0336*** -0.0346*** -0.0333*** (0.0075) (0.0075) (0.0076) (0.0074) (0.0078) (0.0079) (0.0079) (0.0078) CAPEX/Revenue -0.7381 -0.7692 -0.9067* -0.7746 -0.8202 -0.7693 -0.9635 -0.8067 (0.5525) (0.5526) (0.5454) (0.5394) (0.7036) (0.7011) (0.7052) (0.6818) ROIC -0.1413*** -0.1423*** -0.1466*** -0.1423*** -0.1138*** -0.1121*** -0.1170*** -0.1120*** (0.0257) (0.0255) (0.0258) (0.0258) (0.0224) (0.0221) (0.0225) (0.0222) Interest expenditure 0.0261* 0.0266* 0.0287** 0.0269* 0.0188 0.0179 0.0206 0.0202 (0.0140) (0.0140) (0.0140) (0.0145) (0.0140) (0.0140) (0.0141) (0.0144) Volatility -0.0941*** -0.0942*** -0.0942*** -0.0943*** -0.0875*** -0.0873*** -0.0872*** -0.0881*** (0.0157) (0.0157) (0.0159) (0.0158) (0.0168) (0.0167) (0.0169) (0.0168) EURO STOXX 50 -0.5877 -0.5842 -0.5841 -0.5811 0.2168 0.2081 0.2115 0.2237 (0.4117) (0.4113) (0.4114) (0.4115) (0.4403) (0.4408) (0.4409) (0.4396)

Risk free rate 0.8149*** 0.8166*** 0.8214*** 0.8175*** 0.8490*** 0.8469*** 0.8488*** 0.8549***

(0.0899) (0.0902) (0.0902) (0.0909) (0.1071) (0.1074) (0.1071) (0.1084) German slope 1.0192*** 1.0188*** 1.0171*** 1.0176*** 0.6814*** 0.6813*** 0.6748*** 0.6783*** (0.1706) (0.1706) (0.1705) (0.1707) (0.1884) (0.1883) (0.1884) (0.1884) VDAX -0.6125** -0.6133** -0.6201** -0.6133** -0.8462*** -0.8502*** -0.8537*** -0.8517*** (0.2785) (0.2790) (0.2787) (0.2789) (0.2782) (0.2784) (0.2781) (0.2786) Observations 686 686 686 686 613 613 613 613 LR statistic 584.3521*** 584.3102*** 586.0276*** 584.3297*** 528.4615*** 528.0900*** 530.9550*** 528.7230***

Country dummies Yes Yes Yes Yes Yes Yes Yes Yes

The sample period runs from 2006-2014. All data is obtained from the Asset4 database. This table represents the coefficients and the standards errors from contemporaneous ordered logistic regressions with Huber-White standard errors (I till IV) and one-year lagged ordered logistic regressions with Huber-White standard errors (V till VIII). The Fitch credit rating value is used as dependent variable. The sample used in this table consists of the six countries that score below average on sustainability. Country dummies are included in all the models, but are not reported for reason of readability. *,**,*** indicate significance at the ten percent, five percent, and one percent level, respectively.

again the highest impact on the credit ratings. In the regressions on the full sample we saw that the environmental and the corporate governance dimension showed weakly significant results. In the regressions for the firms from countries that score above average on sustainability we see that these dimensions become more significant and more positive than before, with significance at the 5% level for the environmental dimension in both the contemporaneous and the one-year lagged regressions, and significance at the 1% level in the contemporaneous regression and at the 5% level for the one-year lagged regression for the corporate governance dimension. The results from table 5 also indicate a positive effect of CSR on credit ratings, which is again in line with the risk mitigation view. These outcomes support the second hypothesis that higher scores on corporate social responsibility, in all the ESG

dimensions, are rewarded with higher credit ratings in countries with above average ESG scores.

Table 7. Ordered logistic regressions on the quadratic CSR-credit ratings relationship (full sample)

Independent variable ESG ENV SOC CGV ESG ENV SOC CGV

Dependent variable Credit rating I Credit rating II Credit rating III Credit rating IV Credit rating V Credit rating VI Credit rating VII Credit rating VIII ESG 0.0263* 0.0589*** 0.0418** 0.0103 0.0160 0.0559*** 0.0294 0.0117 (0.0154) (0.0165) (0.01770 (0.0104) 0.0171 0.0183 0.0196 0.0108 ESG² -0.0002 -0.0005*** -0.0002* 0.0000 -0.0001 -0.0004*** -0.0001 0.0000 (0.0001) (0.0001) (0.0001) (0.0001) 0.0001 0.0002 0.0002 0.0001 Revenue 0.0315*** 0.0331*** 0.0308*** 0.0318*** 0.0305*** 0.0328*** 0.0301*** 0.0313*** (0.0022) (0.0022) (0.0022) (0.0022) 0.0024 0.0024 0.0024 0.0024 EBIT-Margin 6.7084*** 6.6274*** 6.6324*** 6.6756*** 7.5534*** 7.5009*** 7.5693*** 7.4464*** (0.7893) (0.7831) (0.8167) (0.7762) 1.0048 1.0098 1.0499 0.9654 Leverage -0.0415*** -0.0415*** -0.0432*** -0.0413*** -0.0458*** -0.0454*** -0.0474*** -0.0453*** (0.0055) (0.0055) (0.0054) (0.0055) 0.0060 0.0059 0.0060 0.0059 CAPEX/Revenue -1.3527** -1.2989** -1.2859** -1.4100*** -1.5350** -1.4457* -1.5467** -1.4589** (0.5489) (0.5404) (0.5748) (0.5305) 0.7577 0.7440 0.7750 0.7113 ROIC -0.0694*** -0.0698*** -0.0724*** -0.0674*** -0.0673*** -0.0659*** -0.0685*** -0.0642*** (0.0124) (0.0125) (0.0121) (0.0123) 0.0128 0.0130 0.0127 0.0128 Interest expenditure 0.0091 0.0084 0.0134* 0.0099 0.0109 0.0081 0.0135 0.0109 (0.0077) (0.0077) (0.0080) (0.0079) 0.0094 0.0091 0.0096 0.0093 Volatility -0.0838*** -0.0794*** -0.0849*** -0.0849*** -0.0800*** -0.0761*** -0.0808*** -0.0820*** (0.0112) (0.0110) (0.0111) (0.0110) 0.0118 0.0116 0.0116 0.0116 EURO STOXX 50 -0.5607* -0.5451 -0.5831* -0.5239 0.0121 -0.0092 -0.0121 0.0214 (0.3358) (0.3384) (0.3360) (0.3303) 0.3555 0.3589 0.3553 0.3505

Risk free rate 0.6621*** 0.6619*** 0.6654*** 0.6656*** 0.6867*** 0.6827*** 0.6815*** 0.6903***

(0.0648) (0.0648) (0.0647) (0.0650) 0.0764 0.0751 0.0758 0.0773 German slope 0.8470*** 0.8476*** 0.8502*** 0.8294*** 0.6194*** 0.6267*** 0.6119*** 0.6104*** (0.1278) (0.1263) (0.1284) (0.1277) 0.1395 0.1374 0.1404 0.1393 VDAX -0.5761*** -0.5782*** -0.5890*** -0.5792*** -0.7692*** -0.7731*** -0.7841*** -0.7812*** (0.2104) (0.2099) (0.2118) (0.2112) 0.2130 0.2130 0.2137 0.2140 Observations 1128 1128 1128 1128 1011 1011 1011 1011 LR statistic 852.4809*** 861.9555*** 865.2033*** 847.8789*** 778.8626*** 784.4356*** 789.0937*** 774.6138***

Country dummies Yes Yes Yes Yes Yes Yes Yes Yes

The sample period runs from 2006-2014. All data is obtained from the Asset4 database. This table represents the coefficients and the standards errors from contemporaneous ordered logistic regressions with Huber-White standard errors (I till IV) and one-year lagged ordered logistic regressions with Huber-White standard errors (V till VIII). The Fitch credit rating value is used as dependent variable. The sample used in this table consists of all eleven countries. Country dummies are included in all the models, but are not reported for reason of readability. *,**,*** indicate significance at the ten percent, five percent, and one percent level, respectively.

relationship in this sample, which shows that there is indeed a difference in the CSR-credit rating relationship in the high- and the low-country sample. Where we did see a significant positive impact of CSR on credit ratings in the high-country sample, no significant impact at all can be found in the low-country sample. CSR does not seem to have any impact on the credit ratings of the firms in the countries that score below average on sustainability.

To be able to test the fifth hypothesis, a quadratic is added to the regressions models that were performed before, as explained in the equation 2 in the methodology and data section. Table 7 shows the outcomes of the

Table 8. Ordered logistic regressions on the quadratic CSR-credit ratings relationship (high-country sample)

Independent variable ESG ENV SOC CGV ESG ENV SOC CGV

Dependent variable Credit rating I Credit rating II Credit rating III Credit rating IV Credit rating V Credit rating VI Credit rating VII Credit rating VIII ESG -0.0309 0.0189 0.0104 0.0157 -0.0436* 0.0051 -0.0043 0.0243 (0.0254) (0.03100 (0.02550 (0.0153) (0.0264) (0.0328) (0.0271) (0.0149) ESG² 0.0004** -0.0001 0.0001 0.0000 0.0005** 0.0001 0.0002 -0.0001 (0.0002) (0.0002) (0.0002) (0.0001) (0.0002) (0.0003) (0.0002) (0.0001) Revenue 0.0227*** 0.0245*** 0.0238*** 0.0254*** 0.0243*** 0.0261*** 0.0254*** 0.0272*** (0.0023) (0.0025) (0.0025) (0.0024) (0.0030) (0.0031) (0.0031) (0.0030) EBIT-Margin 5.1791*** 5.3068*** 5.3119*** 4.9218*** 6.5175*** 6.6135*** 6.6896*** 6.1331*** (1.9312) (1.9264) (1.9933) (1.8153) (2.3547) (2.3303) (2.3691) (2.2633) Leverage -0.0590*** -0.0570*** -0.0616*** -0.0566*** -0.0630*** -0.0612*** -0.0648*** -0.0588*** (0.0082) (0.0083) (0.0086) (0.0079) (0.0091) (0.0091) (0.0097) (0.0089) CAPEX/Revenue -5.6044** -4.7332** -4.5573** -5.1391** -7.7623*** -6.9086*** -6.8618*** -6.9535*** (2.1962) (2.2543) (2.2448) (1.9967) (2.1273) (2.0633) (2.1084) (1.8742) ROIC -0.0383** -0.0306* -0.0364** -0.0330* -0.0416** -0.0361* -0.0405** -0.0359* (0.0176) (0.0174) (0.0178) (0.0174) (0.0194) (0.0193) (0.0194) (0.0194) Interest expenditure 0.0037 0.0010 0.0060 0.0026 -0.0002 -0.0011 0.0017 -0.0006 (0.0121) (0.0117) (0.0128) (0.0116) (0.0171) (0.0164) (0.0176) (0.0162) Volatility -0.0681*** -0.0680*** -0.0688*** -0.0742*** -0.0680*** -0.0704*** -0.0686*** -0.0743*** (0.0153) (0.0161) (0.0155) (0.0153) (0.0151) (0.0161) (0.0155) (0.0155) EURO STOXX 50 -0.3159 -0.4369 -0.3412 -0.3658 -0.0255 -0.0947 0.0643 -0.0477 (0.5785) (0.5901) (0.5851) (0.5858) (0.5974) (0.6064) (0.6047) (0.6031)

Risk free rate 0.5538*** 0.5318*** 0.5416*** 0.5357*** 0.5375*** 0.5123*** 0.5113*** 0.5179***

(0.1052) (0.1072) (0.1053) (0.1064) (0.1175) (0.1176) (0.1167) (0.1174) German slope 0.6385*** 0.6382*** 0.6204*** 0.6035*** 0.5748*** 0.5653*** 0.5217** 0.5346** (0.2074) (0.2028) (0.2047) (0.2017) (0.2195) (0.2150) (0.2180) (0.2143) VDAX -0.4078 -0.4677 -0.4778 -0.4837 -0.5355 -0.5721* -0.5510 -0.5967* (0.3298) (0.3252) (0.3327) (0.3252) (0.3398) (0.3339) (0.3361) (0.3373) Observations 442 442 442 442 398 398 398 398 LR statistic 332.2928*** 316.0018*** 331.1699*** 314.6209*** 314.3446*** 297.9192*** 312.2693*** 297.511***

Country dummies Yes Yes Yes Yes Yes Yes Yes Yes

The sample period runs from 2006-2014. All data is obtained from the Asset4 database. This table represents the coefficients and the standards errors from contemporaneous ordered logistic regressions with Huber-White standard errors (I till IV) and one-year lagged ordered logistic regressions with Huber-White standard errors (V till VIII). The Fitch credit rating value is used as dependent variable. The sample used in this table consists of the five countries that score above average on sustainability. Country dummies are included in all the models, but are not reported for reason of readability. *,**,*** indicate significance at the ten percent, five percent, and one percent level, respectively.

social dimension only shows significance in the contemporaneous regression, but does not pass the test to control for endogeneity problems by lagging the independent variables. The conclusion can be made that only the environmental dimension shows results indicating an inverse U-shaped relationship with credit ratings. Surprisingly this inverse U-shaped relationship for the environmental dimension is in contrast with the papers of for example Barnett (2007), Barnett and Salomon (2012) and Nollet et al. (2016). Based on the earlier explained concept of accruing stakeholder influence capacity these papers suggested and found exactly the opposite, namely a U-shaped relationship. After investigating the quadratic relationship for the

high- and the low-country sample separately, a further discussion will take place on the reason for the contrast with the existing literature mentioned above.

Table 9. Ordered logistic regressions on the quadratic CSR-credit ratings relationship (low-country sample)

Independent variable ESG ENV SOC CGV ESG ENV SOC CGV

Dependent variable Credit rating I Credit rating II Credit rating III Credit rating IV Credit rating V Credit rating VI Credit rating VII Credit rating VIII ESG 0.0610*** 0.0936*** 0.0750*** 0.0107 0.0401** 0.0810*** 0.0464* 0.0060 (0.01800) (0.0200) (0.0231) (0.0136) (0.0200) (0.0224) (0.0240) (0.0148) ESG² -0.0006*** -0.0008*** -0.0006*** -0.0001 -0.0003* -0.0007*** -0.0003 0.0000 (0.0002) (0.0002) (0.0002) (0.0001) (0.0002) (0.0002) (0.0002) (0.0001) Revenue 0.0399*** 0.0405*** 0.0375*** 0.0378*** 0.0370*** 0.0388*** 0.0355*** 0.0359*** (0.0033) (0.0033) (0.0034) (0.0034) (0.0036) (0.0035) (0.0036) (0.0037) EBIT-Margin 7.3819*** 7.0898*** 7.2204*** 7.2621*** 8.2082*** 8.0078*** 8.1633*** 8.0479*** (0.9241) (0.8937) (0.9320) (0.9410) (1.1650) (1.1413) (1.1845) (1.1722) Leverage -0.0275*** -0.0306*** -0.0273*** -0.0286*** -0.0328*** -0.0348*** -0.0332*** -0.0334*** (0.0074) (0.0071) (0.0075) (0.0074) (0.0079) (0.0076) (0.0081) (0.0078) CAPEX/Revenue -0.8537 -0.8269 -0.8841 -0.8014 -0.9063 -0.8491 -0.9430 -0.8167 (0.5563) (0.5410) (0.5505) (0.5419) (0.7092) (0.6889) (0.7105) (0.6919) ROIC -0.1449*** -0.1526*** -0.1545*** -0.1450*** -0.1163*** -0.1183*** -0.1210*** -0.1127*** (0.0254) (0.0253) (0.0253) (0.0266) (0.0222) (0.0219) (0.0222) (0.0225) Interest expenditure 0.0212 0.0222 0.0287** 0.0263* 0.0162 0.0138 0.0207 0.0201 (0.0143) (0.0137) (0.0137) (0.0146) (0.0143) (0.0137) (0.0141) (0.0145) Volatility -0.1027*** -0.0990*** -0.1013*** -0.0950*** -0.0925*** -0.0894*** -0.0911*** -0.0883*** (0.0165) (0.0148) (0.0160) (0.0160) (0.0172) (0.0158) (0.0170) (0.0169) EURO STOXX 50 -0.8337* -0.7147* -0.7371* -0.5973 0.0834 0.0938 0.1447 0.2209 (0.4284) (0.4242) (0.4214) (0.4134) (0.4549) (0.4518) (0.4463) (0.4415)

Risk free rate 0.7928*** 0.8065*** 0.8165*** 0.8169*** 0.8284*** 0.8418*** 0.8459*** 0.8538***

(0.0906) (0.0903) (0.0908) (0.0909) (0.1083) (0.1054) (0.1071) (0.1092) German slope 1.0474*** 1.0612*** 1.0483*** 1.0238*** 0.6888*** 0.7254*** 0.6929*** 0.6789*** (0.1719) (0.1667) (0.1716) (0.1714) (0.1889) (0.1830) (0.1892) (0.1884) VDAX -0.6640** -0.6170** -0.6264** -0.6092** -0.8650*** -0.8453*** -0.8455*** -0.8493*** (0.2786) (0.2756) (0.2805) (0.2788) (0.2777) (0.2762) (0.2799) (0.2796) Observations 686 686 686 686 613 613 613 613 LR statistic 597.4888*** 607.0675*** 596.2186*** 584.8528*** 532.814*** 542.8577*** 533.9077*** 528.7594***

Country dummies Yes Yes Yes Yes Yes Yes Yes Yes

The sample period runs from 2006-2014. All data is obtained from the Asset4 database. This table represents the coefficients and the standards errors from contemporaneous ordered logistic regressions with Huber-White standard errors (I till IV) and one-year lagged ordered logistic regressions with Huber-White standard errors (V till VIII). The Fitch credit rating value is used as dependent variable. The sample used in this table consists of the six countries that score below average on sustainability. Country dummies are included in all the models, but are not reported for reason of readability. *,**,*** indicate significance at the ten percent, five percent, and one percent level, respectively.

the contemporaneous model, and only weakly significant in the one-year lagged model. These results for the weighted average ESG score can be partly explained by the concept of accruing stakeholder influence capacity, where firms are able to profit relatively more and more if they increase their investments in CSR. However, this theory also expects a U-shaped relationship, where it is very costly and thus negative for the firm to invest lower amounts in CSR, which is not proven by the results of table 8.

In conclusion, no evidence is found for a quadratic relationship between the individual components of ESG and credit ratings in the sample with countries that score

above average on sustainability. However, there is evidence for the weighted average ESG score that supports partly the theory of accruing stakeholder influence capacity and the associated U-shaped relationship.

only weak significance in the one-year lagged regression, while the social and the corporate governance dimension show insignificant results here. In conclusion, the environmental dimension shows strong evidence for an inverse U-shaped relationship in the low-country sample, while the weighted average ESG score shows only weak evidence for such a relationship.

Taking the results of the tables 7, 8 and 9 together, the findings for the inverse U-shaped relationship for the environmental score in the total sample are mainly driven by the low-country sample. In table 6, the linear regression on the low-country sample, we saw already that the results were insignificant, not supporting either the risk mitigation view or the overinvestment view. The results after adding the quadratic indicate that for the environmental dimension this could be due to an inverse U-shaped relationship where the risk mitigation view is supported for lower levels of CSR (below a certain optimal point), and where the overinvestment view becomes dominant at the higher levels of CSR scores (above the optimal point).

With regards to the fourth hypothesis, which states that the corporate governance dimension is the key driver in the CSR-credit rating relationship, we have to conclude that we found no evidence that supports this hypothesis. Although the corporate governance dimension shows a significant relationship with credit ratings, especially in the linear regressions in the high-country sample, the social dimension shows in each of these cases higher impacts on credit ratings and also higher levels of significance, therefore the fourth hypothesis is rejected.

ROBUSTNESS CHECKS

The dependent variable used in the regressions so far has been the Fitch credit rating value. In this section the other measure, the Fitch credit rating score, will be used as dependent variable. Since this score is on a 0-100 scale it provides us with more differentiation. Investigating this

alternative dependent variable while using other regression models provides us with more robustness. In order to choose the correct panel regression model for the Fitch credit rating scores, the Hausman test and the likelihood ratio redundant fixed effects test are performed. The conclusion can be made that the fixed effect model is more preferable than the random effects model with the data used in this paper, and therefore the model with both cross-sectional and time-fixed effects is the most appropriate regression model in this research. The time-fixed effects will also exclude potential effects of the financial crisis of 2007-2008 on the results. In addition to the fixed effects model, pooled OLS is also performed, because ESG scores are not expected to show large changes over time. The White period standard errors and covariance estimates are applied in the pooled OLS, which are robust to arbitrary within cross-section residual correlation.

In Appendix A the results of the pooled OLS regressions and the regressions with cross-sectional and time-fixed effects are shown for the full sample. The results look similar to the results of the ordered logistic regressions. Again the social dimension appears to have to highest positive impact on the credit ratings. Compared to the ordered logistic regressions, the significance levels appear to be a bit lower for all four dimensions. This means that the environmental and the governance dimension, which were weakly significant in the ordered logistic regressions, become insignificant now.

OLS and fixed effects regression. The corporate governance dimension only shows a significant and positive relationship in the fixed effect regression and an insignificant positive relationship in the pooled OLS regression. The environmental dimension shows in both the regressions an insignificant positive relationship. The results of the pooled OLS and fixed effects regressions for the low-country sample are very similar to the results of the ordered logistic regressions. Appendix C shows that all the dimensions show insignificant relationship in both types of regressions.

In order to control for potential endogeneity problems, the pooled OLS and fixed effect regressions are also performed with one-year lagged independent variables. The results are very much the same as for the contemporaneous regressions, therefore no changes in the earlier conclusions have to be made.

In summary the results from the pooled OLS and fixed effects regressions showed very similar trends as for the ordered logistic regressions. However, in all the cases the level of significance was lower in the pooled OLS and fixed effects regression than in the ordered logistic regression. Although the Fitch credit rating score provides some more differentiation because it is on a 0-100 instead and it compares firms relative to each other in each of the years, this differentiation might have been too weak to be really different from the typical ordinal scale of 1-24 of the Fitch credit rating value. Therefore it is not surprising that the pooled OLS and the fixed effects regressions provided weaker significance levels than the ordered logistic regressions that are specially designed for ordinal scales such as credit ratings. Nevertheless, finding similar trends in the results while using an alternative dependent variable and two alternative types of regressions adds to the robustness of the results found earlier.

Next to the additional types of regressions, this paper also added a crisis dummy in all the ordered logistic regressions performed in this study. Although potential

effects of the crisis on the results are already taken into account by using the time-fixed effects regressions, these potential effects are even further excluded by implementing a crisis dummy. This crisis dummy takes a value of 1 in the crisis years 2007 and 2008, and 0 otherwise. No significant differences from the earlier results can be found by adding the crisis dummy, therefore we can conclude that we do not have to change the earlier conclusions.

CONLUSION

In this study the impact of corporate social responsibility on credit ratings is examined, using a sample of 141 firms from Eurozone countries. For the full sample evidence is found for a positive impact of the weighted average ESG score on the credit rating of these firms. This is in line with the risk mitigation view, which states that firms with high scores on corporate social responsibility have more favourable risk profiles, compared to identical firms with lower scores on CSR (Goss and Roberts, 2011). If then the impact of the individual dimensions is taken into account, it appears that especially the social dimension contributes to this relationship by having a strongly significant and relatively high positive impact on credit ratings. The other two dimensions, environmental and corporate governance, only provide weakly significant and a less positive impact on credit ratings.