Corporate Social Responsibility and the Cost of Corporate Bonds:
The impact of ESG scores on the Yield Spread
An empirical study on the US bond issuing firms
Abstract
This study aims to investigate the impact of ESG scores on the cost of corporate bonds by using a sample of 189 firms and 798 corporate bonds listed on the NYSE, NASDAQ, London Stock Exchange and the Deutsche Börse AG between the period 2014-2019. Using the yield spread as an ex ante cost of corporate bonds, evidence is found that higher overall ESG scores are negatively associated with lower yield spreads. In addition, this paper investigates the impact of the individual ESG pillar scores on the cost of corporate bonds. The results indicate that only the environmental- and social pillar are negatively associated with the yield spread, while the effect of the governance pillar is insignificant. Furthermore, this study examines the role of the investment horizon on the relationship between overall ESG rating and the yield spread. Evidence is found that the effect of ESG scores are more pronounced on short-term bonds.
Keywords: Corporate Social Responsibility, ESG score, Corporate Bonds, Yield Spread JEL Classification: C33, G12, G32, M14
Author: Frank van Esch Student number: s2681439
University of Groningen – Faculty of Economics and Business Master Thesis – Finance
Supervisor: D. Vullings, Msc
1. Introduction
In the last decades, there has been an increase in companies considering their contribution to corporate social responsibility (CSR) as part of their business and investment strategies. There are many available definitions of CSR, but one of the most commonly used definitions is stated by the European Commission of Communities (2001): ‘’CSR is a concept whereby companies integrate social and environmental concerns in their business operations and in their interaction with their stakeholders on a voluntary basis’’. Environmental, social and governance (ESG) considerations often play an important role in disclosing CSR information to investors (Cini and Ricci, 2018). ESG ratings can also be highly relevant to investors in representing risk factors next to the traditional measures of financial performance. Companies that act irresponsible towards society are prone to regulatory, reputational, and financial risk that could lead to costly penalties (Bauer and Hann, 2010). In this case, ESG ratings are able to function as a supplement risk indicator next to other financial indicators such as the credit rating (Polbennikov, Desclée, Dynkin, and Maitra, 2016). Therefore, investors believe that ESG investing is no longer a niche strategy in the fixed income market (BlackRock, 2019). Even though studies on CSR related topics are widely available in the existing financial literature, evidence on the impact of ESG investing in the fixed income markets is still limited. Most prior literature on CSR contains equity-focused research. This empirical study aims to close the gap in the existing literature by investigating the impact of ESG scores on the cost of corporate bonds in the fixed income market.
First of all, this paper will investigate the impact of the overall ESG score on the yield spread of corporate bonds. Secondly, I will investigate the impact of the individual ESG pillars (Environment; Social; Governance) on the yield spread. In addition, I will examine the role of the investment horizon on the relationship between overall ESG rating and the yield spread. In order to achieve this, I will use academic literature on corporate social responsibility, ESG ratings and fixed income. The data used for this research is obtained from the Thomson Reuters database.
2015) and bank loan credit spread (Hoepner, Oikonomou, Scholtens, and Schröder, 2016). Some papers support the finding that firms with higher CSR performance can issue bonds at a lower yield spread and improve the credit quality of specific bond issues (Cooper and Uzun, 2015; Ge and Liu, 2015; Oikonomou, Pavelin, and Brooks, 2014). However, opposing studies have made a critical note on the actual incorporation and value creation of CSR activities into the pricing of corporate bonds (Jo and Harjoto, 2012; Menz, 2010). These dissimilar viewpoints can be both supported by following the line of reasoning of the stakeholder-or shareholder theorists.
According to the existing literature, most studies in equity markets focus on the link between ESG performances and returns (Khan, 2019; Evans, and Peiris, 2010). ESG scores might reflect the ability for investors to seize potential reward opportunities. On the other hand, fixed income investors are concerned about the downside risks instead of returns. ESG scores might be relevant for investors if these scores are able to illustrate new risks that are not contained in other traditional financial measures (Polbennikov, Desclée, Dynkin, and Maitra, 2016). Firms with lower idiosyncratic risk may have lower price premiums which subsequently decreases the cost of capital (Goss and Roberts, 2011). Taking the increasing global trend of disclosing ESG performances into account (KPMG, 2020), it is even more interesting to explore the ability of ESG scores to help identify new risk factors in the fixed income market (BlackRock, 2019). This paper aims to empirically investigate the relationship between both the combined- as well as individual ESG scores and the cost of corporate bonds by addressing the following research question:
Research question: ‘’To what extent do ESG scores of bond issuing firms affect the yield spread of corporate bonds?’’
impact of overall ESG scores are more pronounced on short-term bonds. Overall, the economic significance of all the results are relatively small and the results of the robustness checks are not quantitatively different from the primary model.
The remainder of this paper is organized as follows. Section 2 highlights the relevant concepts in the existing literature and hypothesis development. Section 3 is allocated to the data and methodology. In section 4, the results will be presented and the reasoning behind the results in the context of this paper will be discussed. Section 5 provides a summary and conclusion of this study.
2. Literature review & hypothesis development
2.1 Corporate Social ResponsibilityCSR is built on the stakeholder theory (Freeman and Dmytriyev, 2017). The stakeholder theory was introduced in the publication ‘Strategic Management: A Stakeholder Approach’ by R. Edward Freeman in 1984. The stakeholder theory advocates that if organizations want to be effective, firms should consider creating value for all stakeholders, instead of only for shareholders. Many firms run their business in a way that is highly consistent with the stakeholder theory (Freeman, Wicks, and Parmar, 2004). All these firms have broadened their focus on not only the fundamental driver of making profits, but also on acting morally and ethically responsible towards its stakeholders. This approach faces the challenge of running your business ethically or improving moral performance in business, and is embedded in the daily practices of organizations where managers act responsibly towards the society and environment. According to prior research, the framework of the stakeholder theory can identify value creating opportunities (Tantalo and Priem, 2016; Jones, Harisson, and Felps, 2018). Based on this framework CSR disclosures may decrease the cost of capital for the following reasons (Ge and Liu, 2015).
and lower potential risks (Jones, 1995). This is consistent with the risk mitigation view of Goss and Roberts (2011). The risk mitigation view suggests that investors are able to assess the risk profile of firms because CSR-related disclosures can reflect the extent of risk exposure. Firms with lower idiosyncratic risk may have lower price premiums, which in turn leads to lower cost of capital.
Another explanation could be that firms with higher CSR performance face less litigation risk because companies that are more socially responsible face fewer lawsuits regarding environmental issues and social norms (Hong and Kacperczyk, 2009). In addition, Bauer and Hann (2010) suggest that costly penalties of environmental misconduct can result in higher reputational-, litigation- and financial risk. This is supported by the study of Schneider (2010). The study found that future clean-up and compliance costs of corporations can be high enough to threaten the ability to meet their debt obligations.
The opposite theory of the stakeholder theory is developed by Milton Friedman (1970). Friedman famously stated that the only social responsibility and top priority of a business is to increase its profits. This position is known as the shareholder theory. Friedman argued that CSR initiatives are a waste of firm resources because these investments could otherwise be used to generate profits accruing to its shareholders (Ge and Liu, 2015). This could lead to a higher cost of public debt financing because the lower profits result in reduced interest-paying abilities and higher distress risk. This is consistent with the overinvestment view of Goss and Roberts (2011). This view put forward an alternative perspective on the agency conflict theory. Managers may overinvest in CSR at the expense of the shareholders or creditors (Barnea and Rubin, 2010). In doing so, managers can take credit when it comes to acting socially responsible, which might enhance their reputation. Overinvestments in CSR would increase distress risk, which eventually could result in higher cost of capital.
Despite the different dimensions of CSR beliefs, Mosca and Civera (2017) found that there has been a global growth within organisations of the integration of socially responsible practices, led on by a heightened interest and more intricate globalised business phenomena. The significant increase in the number of organizations issuing CSR reports is supported by the KPMG International Survey of Corporate Responsibility1. Another explanation for the increase
1 The KPMG survey of corporate social responsibility provides a detailed look at global trends of CSR
reporting and disclosures:
in responsible behaviour, is that firms who participate in CSR programs are more likely to create value in the long run (Husted and Allen, 2009). Although the global trends in CSR disclosures are significantly on the rise, opposing points of view within existing literature still exist.
According to Gholami (2011), the reciprocal dependence between businesses and society can create value in both ways due to competitive advantage opportunities and awareness. On the contrary, research conducted by McWilliams and Siegel (2002) delivers critique on CSR activities and claims that investments in CSR deplete scarce resources. However, the majority of the existing literature has drawn positive conclusions regarding CSR.
Firms undertake investments in CSR activities due to a positive impact on financial performance (Cochran and Wood, 1984; Palmer, 2012). Tsoutsoura (2004) investigated the sign of the relationship between CSR and the financial performance of firms. With a panel data set consisting of 422 companies of the S&P 500 over the years 1996-2000, the author found a positive and significant relationship between CSR performance and financial performance.
Next to financial performance, another paramount effect is the impact of CSR on firm value. Gregory, Tharyan, and Whittaker (2014) conducted their empirical research on the effect of CSR on firm value. The authors try to find the origin of firm value by separating the effects on forecasted profitability, cost of capital and long-term growth. With a sample of 3,100 firms over the period 1992-2009, the coefficients show a strong positive relationship between CSR and the market-to-book ratio, which is a proxy for firm value. The authors show that markets positively value most aspects of CSR, especially in the long run. In addition, firms with a high CSR score tend to have a higher expected growth rate in their abnormal earnings, which can be explained by a lower cost of equity.
2.2. The link between CSR and Fixed Income
As studies on CSR are becoming increasingly investigated, there remains to be a limitation in the literature about the link between CSR and corporate bonds. The majority of studies are focused on the influence of CSR on the cost of equity (Reverte, 2012; El Ghoul, Guedhami, and Kwok, 2011; Weber, 2018). The trend of non-financial disclosures, like CSR practices, have become commonplace under organizations’ stakeholders over the past two decades. Therefore, studies conducted on this growing trend in the corporate culture have obtained more attention. It is important to understand the insights about the impact of CSR on the cost of corporate debt (Huang, Hu, and Zhu, 2018; Cooper and Uzun, 2015)
Goss and Robert (2011) studied the relationship between CSR and the cost of debt through bank loans. The authors provided with the risk mitigation view an argument in favour of CSR because these initiatives should reduce risk. This view proposed that managers are able to use CSR initiatives as a tool to manage the risk profile of companies. To elaborate, with a sample of 3996 loans to US firms, the authors suggest that firms with stronger CSR performances paid a lower price premium on their bank loans.
Hoepner, Oikonomou, Scholtens, and Schröder (2016) investigated the impact of corporate- and country sustainability on the cost of bank loans by studying 470 loan agreements within the period 2005-2012. The authors found that environmental-and social performance are significantly and economically related with direct financing. A higher country sustainability performance is related to lower corporate spreads of bank loans. The results of this paper suggest that the impact of environmental and social activities cannot be understated. By distinguishing between environmental- and social dimensions, the authors found that the environmental impact is approximately two times larger than the social impact on the cost of corporate bank loans.
The study of Menz (2010) made a critical note on the question whether CSR has a positive impact on the corporate bond market. By making use of a data panel consisting of 498 bonds over 39 months with a total of 16,957 observations, the results reveal that CSR information does not directly affect the credit spread of bonds. One reason for this is that fixed income investors care more about the credit rating. Extra CSR disclosures do not seem to add valuable information in determining the bond spread because credit ratings already include to some extent CSR information.
Ge and Liu (2015) investigated if an organization’s CSR performance is associated with the cost of corporate bonds by using the yield spread as a measure of the issuers’ incremental cost of a bond. The authors collected bond specific data with 4260 observations from 2317 firms that disclosed CSR information over the period 1992-2009. In order to control for the explanatory power of CSR performance, the authors used firm and bond-level control variables such as the size of the issuer, return on assets, leverage, bond rating, maturity and more. Consistent with the stakeholder theory, Ge and Liu found that overall CSR performance is related to better credit ratings and lower yield spreads in new corporate bond issues. However, after controlling for credit ratings, the results indicate that some effects are absorbed by credit ratings. When taking this all into consideration, the results suggest that firms with better CSR performance can raise public debt at lower cost which is more pronounced for investment grade bonds. This study is very relevant for the research conducted in this paper.
2.3 Hypothesis development
Based on the previously stated literature and theories, this study strives to extend the research of the relationship between CSR and the cost of corporate bonds. Recent studies provide evidence that CSR is negatively associated with the cost of corporate bonds (Ge and Liu, 2015; Oikonomou, Pavelin, and Brooks, 2014; Stellner, Klein, and Zwergel, 2015). In order to answer the research question, this study will follow and combine the research design of two studies (Ge and Liu, 2015; Oikonomou, Pavelin, and Brooks, 2014). These two studies are recognized as leading papers on this topic by concurrent studies. Both studies collected corporate social performances, strengths and concerns data from the KLD STATS database. Alternative to these two studies, this paper will investigate the impact of CSR on the cost of debt by making use of overall- and individual ESG scores as CSR measures.
In addition, Bauer and Hann (2010) find that costly penalties of environmental misconduct can result in higher reputational-, litigation- and financial risk. Jung, Herbohn, and Clarkson (2016) suggest that lenders incorporate the awareness of carbon-related risks of firms into their lending decisions. The authors found that the cost of debt is negatively affected by firms that show their carbon-risk awareness through green investment disclosures.
Based on these findings in the existing literature, it is expected that the overall ESG scores of firms will have a negative effect on the yield spread. This expectation is of heightened interest, given the current global trend that recognizes the importance of ESG information in the valuation of assets. Therefore, the following hypothesis will be tested:
Hypothesis 1: The overall ESG score of firms is negatively associated with the yield spread of corporate bonds, ceteris paribus.
Existing literature argues that individual ESG pillars consolidated in the overall ESG score might have different impacts on the cost of debt (Sassen, Hinze, and Hardeck, 2016). Another study found evidence that social capital can constrain opportunistic firm behaviour and, consequently reduce the cost of bank loans (Hasan, Hoi, Wu, and Zhang, 2017). Schneider (2011) discovered evidence that environmental performance of firms can affect the bankruptcy risk and thus the cost of bonds. The author found that costs of environmental misconduct can be high enough to threaten the credit positions of firms, which would eventually affect the cost of debt. Bhojraj and Sengupta (2003) suggest that corporate governance disclosures reduce potential conflicts between bondholders and management. Governance mechanisms can monitor the actions of firms and in turn reduce firms’ default risk. Existing literature also provides evidence that the environmental and social dimensions affect the cost of debt, while the governance dimension remains unrelated (Eliwa, Aboud, and Saleh, 2019; Erragragui, 2017). Both papers found that the environmental pillar has the largest impact on the cost of debt. Hoepner, Oikonomou, Scholtens, and Schröder (2016) also found that the environmental dimension of a country’s institutional framework had a bigger effect on the cost of bank loans than the social dimension.
expectation is that each individual pillar will have a negative effect on the cost of corporate bonds. Therefore, this study will test the following hypothesis:
Hypothesis 2: The individual ESG (Environmental; Social; Governance) scores of firms are negatively associated with the yield spread of corporate bonds, ceteris paribus.
In line with Oikonomou, Pavelin, and Brooks (2014), this study will also investigate the effect of ESG scores on the yield spread by distinguishing between short- and long-term bonds. Prior research provides evidence about the positive relation between long-term institutional investment and corporate social performances (Cox, Brammer, and Millington, 2004; Graves and Waddock, 1994). The high costs of ESG investments are immediately noticeable, while reaping the benefits of investments will happen in the future (Dorfleitner, Kreuzer, and Sparrer, 2020). This holds when the benefits of the costly investments exceed the expectations in the long-term. Therefore, investing in high ESG activities would be less attractive for short-term investors because this may be harmful to short-short-term profits, and subsequently the credit positions of firms (Starks, Venkat, and Zhu, 2017). Furthermore, successful stakeholder management can result in a competitive advantage and value creation in the long-term (Hillman and Keim, 2001). Effective stakeholder management is built on long-term relationships with key stakeholders, such as customers, suppliers or employees. These rewarding relationships are difficult to duplicate in the short-term and are therefore value creating. As debt is the foremost method to raise long-term capital in the United States (Bhojraj and Sengupta, 2003), it is reasonable to explore the role of the investment horizon on the relationship between ESG scores and cost of bonds. The expectation is that the effect of overall ESG scores are more pronounced in bonds with longer maturities. Therefore, this paper will test the following hypothesis:
3. Data and Methodology
3.1 Sample SelectionIn order to investigate the relationship between CSR and cost of corporate bonds, this empirical research employs a panel dataset of corporate bonds listed on the NYSE, NASDAQ, London Stock Exchange and the Deutsche Börse AG during the period 2014-2019. These corporate bonds are only issued in US Dollars by publicly listed U.S. companies with headquarters in the United States of America. 2014 has been chosen as a starting point to exclude for the lasting effects of the financial crisis over the years (Simpson and Grossman, 2017). Another reason is that ESG investing started to become more rewarded in the market from 2014 onwards (Bennani, Guenedal, Lepetit, Mortier, Roncalli, and Sekine, 2018). 2019 has been chosen as the end point because this is generally the last year for which data is completely available. Financial institutions are removed from the existing sample, given that previous research states that financial institutions operate under different regulations and debt characteristics (Ge and Liu, 2015; van Landschoot, 2008). This incomparability with corporate issuers can cause potential outliers within the sample collection.
The data is collected from different data sources. The data of all available publicly listed US companies in combination with the respective International Securities Identification Number (ISIN) are obtained from Thomson Reuters Eikon. Data on environmental, social and governance performance are collected from Thomson Reuters ASSET4 ESG database. Data with regard to corporate bond specific characteristics are obtained from the Thomson Reuters Datastream database. Finally, the data regarding firm characteristics is retrieved from Thomson Reuters Worldscope database and the Standard Industrial Classification (SIC) code is obtained from Thomson Reuters Datastream database. A comprehensive overview of the variables and the corresponding data sources can be found in Table 1 in Appendix A.
The initial sample started with 6,265 publicly listed US companies and 23,824 US bonds. The sample size was reduced to 189 bond-issuing firms and 798 bonds after matching the ISIN of the bond issuing firms with the data of the publicly listed US companies. One reason for this reduction may be explained by Bloomberg2, who states that: ‘’the lack of consistent
2 Bloomberg LP is a major provider in daily financial news and information, including current and
historic financial data. Bloomberg Professional Services provided the following article regarding the need of high-quality data of ESG performances:
reporting standards for ESG data presents a major barrier to the increased adoption of sustainable investing’’ (Bloomberg, 2019). Another reason for the reduction is that the corporate bonds are filtered for the same characteristics (corporate bonds, fixed coupon payments, maturity, issuance date).
3.2 Variables
3.2.1. Dependent variable
Based on previous studies, the dependent variable of this empirical research is the natural logarithm of corporate yield spread. The reason behind the use of the natural logarithm will be explained in the next section. By definition, the yield spread is the difference between the corporate bond yield and the Treasury yield with comparable maturity (Ge and Liu, 2015). Prior studies use the yield spread as a proxy for the risk premium or bond performance because it is an explicit measure of the borrower’s incremental cost of a corporate bond over a comparable risk-free Treasury bond (Ge and Liu, 2015; Menz, 2010). US Treasury bonds are considered to be risk-free because of the excellent credit power of the government (Oikonomou, Brooks, and Pavelin, 2014). Economic factors can be controlled for by subtracting the Treasury yields from the corporate yields (Ge and Liu, 2015). Another study is using the z-spread as a proxy for corporate bond performance, which is defined as the basis-point spread over the spot yield curve that is needed to equal the discounted cash flows to its market price (Stellner, Klein, and Zwergel, 2015).
In order to utilize the yield spread as a measure for the cost of corporate bonds, this study uses the yield to maturity (YTM), also known as the yield to redemption for corporate bonds. This principle is functioning as the discount rate in the calculation of the bond’s present value of every single future cash flow, equalling the bond’s market price until maturity date. This is including coupon payments and capital gains or losses by buying a bond at discount or premium.
3.2.2 Explanatory variables
The explanatory variables of this empirical research are the overall ESG scores and the individual environmental-, social- and governance scores retrieved from Thomson Reuters ASSET4 ESG database3. Stellner, Klein, and Zwergel (2014) used this data source while conducting their empirical research on CSR and Eurozone corporate bonds. This database is one of the most comprehensive databases in measuring ESG performances and covers over 6,000 public companies all over the world by making use of more than 400 different ESG metrics dating back to 2002 (Thomson Reuters, 2017). Based on reported data, the ESG scores are delineated across ten categories to get a transparent and objective measure of the overall corporate performance. Analysts of Thomson Reuters aim to provide manually up to date measures for each individual firm on a continuous basis. The analysts go through standard procedures in weighting these ten categories to enumerate these performance scores within each of the three following pillars; Environmental, Social and Governance. Table 2 in Appendix A, presents an overview of the three pillars based on the ten categories and the corresponding weights.
As presented in Table 2, the overall ESG score is calculated by adding the total corresponding weights for the respective categories within each pillar. Based on the indicators of scoring, the total combined ESG score is conducted from the sum of the following weights: 34% with respect to the environmental pillar, 35.5% with respect to the social pillar and 30.5% with respect to the governance pillar. The score that can be achieved is scaled between the range from 0, which is considered to be the lowest score, to 100, which is considered to be the most outstanding score.
Next to the overall ESG score based on the three pillars, each individual pillar is also separately used as an explanatory variable to convey peculiar differentiations in the effects on corporate bond performances. The three individual pillars are also retrieved from the Thomson Reuters ASSET4 ESG database with the same practices as the overall ESG score.
Similar studies on this topic chose the KLD STATS database (Ge and Liu, 2015; Oikonomou, Pavelin, and Brooks, 2014). However, this study has deliberately chosen to use
3 The Thomson Reuters (2017) provides a report with all the insights about Thomson Reuters ASSET4
ESG Scores:
Thomson Reuters database for data collection because it is frequently updated in alignment with corporate reporting patterns. Especially ESG news and controversies are updated on a continuous basis. Furthermore, Thomson Reuters has one of the largest ESG content collection operations in the world (Refinitiv, 2020).
3.2.3. Control variables
A series of control variables are included in order to avoid the bias of omitted variables and make the empirical results more robust. The control variables are divided in bond characteristics and firm characteristics.
3.2.3.1. Bond characteristics
The bond’s years to maturity is the first bond characteristic to control for. The years to maturity are the years from the date of issuance of the bond until the date of redemption. This is a proxy for default risk because bonds with longer maturities tend to bear more risk than bonds with shorter maturities (Huang, Hu, and Zhu, 2018; Oikonomou, Pavelin, and Brooks, 2014). Bonds with a longer term structure until maturity potentially underpins higher yield spreads.
The second bond specific characteristic is the modified duration of the bond. The modified duration describes the interest rate risk of a fixed income security since it measures the sensitivity of the bond price to a change in the interest rate. Consequently, a higher value for the modified duration of a bond could translate into a larger yield spread due to the increase in the volatility of the bond’s price (Menz, 2010).
Finally, according to Ge and Liu (2015), the Moody’s4 credit rating of bonds is incorporated within the model as a control variable. Based on the bond’s issuer’s financial strength, the assigned bond rating by Moody’s reports the creditworthiness of a bond. Table 3 in Appendix A, gives a comprehensive overview of the bond ratings by Moody’s with a division in ‘grade bonds’ and ‘non-investment grade bonds’. The investment-grade bonds are ranked from ‘Baa3’, as the lowest rating, to the highest rating ‘Aaa’.
3.2.3.2. Firm characteristics
Following previously conducted studies, the firm size is incorporated as a control variable. Ge and Liu (2015) did employ the natural logarithm of an issuer’s asset at the fiscal-year-end right before the issuance date of the bond as a proxy for firm size. This study follows the method according to Oikonomou, Pavelin, and Brooks (2014), in which the market capitalization is implemented as a firm-level control variable. The market capitalization is calculated by the multiplication of the market price at the end of the year with the common shares outstanding of the borrower, and is retrieved from Thomson Reuters Datastream. The market capitalization should be negatively linked with the yield spread. Accordingly, larger firms tend to be more financially stable over time (Lee, 2009).
The second firm-level control variable is debt to equity ratio to control for the firm’s leverage. A high leverage ratio can be associated with higher default risk and can be measured by means of miscellaneous ratios. Ge and Liu (2015) employ the ratio of long-term debt divided by total assets at fiscal year-end right before the issuance of the bond as a proxy for leverage. Furthermore, Stellner, Zwergel, and Klein (2014) use the debt over capital ratio to account for the apprehensive default risk. In line with Oikonomou, Pavelin, and Brooks (2014), this study utilizes the debt to equity ratio which is calculated by dividing total debt by total equity at the end of the year. As overleveraged firms can be reprimanded for experiencing difficulties with paying its interest and principal payments, firms with a satisfactory optimal leverage ratio can reflect financial health. Hence, the aforementioned ratio is used since it could have a plausible effect on the yield spread.
Furthermore, previous studies claim that the financial performance of companies is expressed by the return on assets (ROA). The issuer’s ROA is defined by dividing net income
4 Next to Standard & Poor’s Rating Services and Fitch Ratings, Moody’s Corporation belongs to the
by total assets at year-end. It gives an indication of how efficient a business is at using its assets in generating returns. The belief is that it is negatively linked to the spread (Ge and Liu, 2015; Oikonomou, Brooks, and Pavelin, 2014; Stellner, Zwergel, and Klein, 2014). Hence, the issuer of the bond can cover its debt obligations more easily due to the higher earnings generated from its invested capital.
Lastly, previous empirical studies have shown that identical rated bonds have different risk premiums in varying industry sectors (Longstaff and Schwartz, 1995; Ge and Liu, 2015). To avoid industry-specific heterogeneity, 13 industrial classes are included as industry-specific effects. These industrial classes are defined from the firm’s two-digit standard industrial classification (SIC) codes (Cooper and Uzun, 2015; Jo and Harjoto, 2011). However, the category ‘Transportation’ has been split into three parts (transportation, communications and electric, gas & sanitary services) to clearly differentiate the various sub-classes within this specific category. An overview of the industry classification with the corresponding SIC-codes is provided by Table 4 in Appendix A.
3.3 Methodology
This research performs a panel data analysis given that the dataset contains dimensions of both time-series and cross-sectional elements. There are three possible estimation techniques to deal with this data; pooled OLS, fixed effects or random effects.
3.3.1. Pooled OLS
method to analyse this data set. This can be statistically proven by the redundant fixed effects test.
The redundant fixed effects test is an F-test whereby the null hypothesis states that all error terms are equal to zero. By running a pooled OLS on a panel data, you are likely to violate the assumption of uncorrelated error terms. Rejecting the null hypothesis means that the error terms are jointly different from zero, which gives you an indication of heterogeneity in bonds and its underlying firm- and industry characteristics. This might indicate the preference for a panel estimation technique by making use of a fixed- or random effects model over a pooled OLS regression. The results of this test will be reported in section 4.1.
3.3.2. Panel data models
If the outcome of the redundant fixed effects indicates a preference for a panel data model over a pooled OLS regression, either the fixed effects model or the random effects model can be used. A panel data model approach has multiple advantages in comparison to a simple pooled regression. The advantage of performing a panel estimation technique is that it can control for individual heterogeneity in bonds and its underlying firm- or industry characteristics by controlling for unobserved factors which vary across entities and/or over time (Brooks, 2019). In addition, the panel data approach is able to examine how variables dynamically change over time.
3.3.2.1. Fixed effects model
consideration, a fixed effects model appears to be an appropriate form of testing, as it allows for bond or industry heterogeneity.
The Wald test is performed to statistically prove the urgency of including entity and/or time fixed effects, whilst also testing for the statistically significant improvement in the fit of the model. Rejecting the null hypothesis indicates that the coefficients of the fixed effects are not simultaneously equal to zero and therefore include entity and/or time fixed effects. The results of the Wald test will be reported in section 4.1.
3.3.2.2. Random effects model
The second panel data model is the random effects model. The reasoning behind the random effects model is that, contrary to the fixed effects model, the variation across entities is assumed to be random and uncorrelated with the explanatory variables (Brooks, 2019). This model can be used if there is reason to assume that differences across bonds have some influence on the yield spread. The advantage of using a random effects model is that it can obtain, in contrast to the fixed effects model, an actual estimate for the industry effect since time-invariant variables do not fully cancel out. However, a major drawback of the random effects model is that it has the strict assumption of endogeneity, which states that there should be no correlation between the error term and independent variables. In practice it is difficult to meet this assumption, therefore the fixed effects model is more convincing to use most of the time.
3.3.2.3 Fixed effects vs. random effects
report the eventual model specification supported by the results of the previously mentioned tests.
3.4 Empirical models
This section is allocated to the empirical models that are conducted within this study, in order to answer the hypotheses of section 2.3. The first regression model tests for the combined effect of ESG scores on the yield spread of corporate bonds. This model will be further explained below:
!"!" = $ + ' (")!"#$+ *+!"#$+ ,-!"#$+ .! + /"+ 0!" (1)
According to the first regression formula, the subscript i is the cross-sectional dimension in bonds. In addition, the subscript t-1 is the time-series dimension in years. These subscripts have been added due to the use of a panel data set. The dependent variable, !"!" is the
corporate bond yield spread of a bond i at time t. The spread is calculated by subtracting the Treasury bond yield from the corporate bond yield with the same characteristics. Additionally, the $ on the right hand side of the equation is the intercept term.
The test-and control variables are lagged by one period to partially control for endogeneity and the bias of existing reverse causality between independent variables and the yield spread (Oikonomou, Pavelin, and Brooks, 2014). As this paper is interested in the effect on the yield spread, lagging the right-hand side variables by one year helps to ensure that the ESG scores, bond characteristics and firm characteristics have been determined before the yield. By doing this, the yield spread can only be affected by these variables and not the other way around. (")!"#$ is the test variable as it is a measure for the combined ESG score of each
company i at time t-1. Furthermore, +!"#$ is a vector containing the bond characteristic for
each firm i at time t-1 as control variables. * is a vector of parameters that shows how control variables are affecting the yield spread. Subsequently, -!"#$ is a vector that indicates the firm
characteristics as control variables with , as a vector of parameters. The bond and firm characteristics are outlined in more detail in the sections 3.2.3.1 and 3.2.3.2 . .! represents
the industry effects and is encapsulating all of the variables that affect !"!" across entities by
results, /" are the time effects which allows intercepts to vary over time and not
cross-sectionally. This can be done to eliminate yearly economic-wide trends. As mentioned in the previous section, the need to control for unobserved elements across entities and the need to over time depends on test results of section 4.1. Lastly, 0!" is the error term.
The next data regression models tests for the relationship between the yield spread and each individual score of the three pillars; Environment, Social and Governance.
!"!" = $ + '$(12!"#$+ *+!"#$+ ,-!"#$+ .!+ /"+ 0!"#$
(2)
!"!" = $ + '$"34!"#$+ *+!"#$+ ,-!"#$+ .!+ /"+ 0!"#$
(3)
!"!" = $ + '$)32!"#$+ *+!"#$+ ,-!"#$+ .!+ /"+ 0!"#$ (4)
The parameters '$ denote the linear impact of the individual environmental, social and
governance test variables on the yield spread in models 2, 3 and 4. These models will perform separate tests for the effect of each individual pillar on the yield spread. Again, the subscripts i and t-1 in all models describe the cross-sectional and time dimensions in bonds and years. Equivalent to model 1, the test-and control variables are lagged by one period in all models for the same reason. The vectors of the parameters and control variables that contain the bond and firm characteristics are the same in all models as in equation 1. Another similarity to equation 1 is that .! and /" are added into all models to control for the allowance to vary
across entities and to control over time. Finally, 0!"#$ denotes the error term in all models.
3.5 Robustness tests
It is necessary to investigate whether any limitations exist in this panel data set, so that the coefficient estimates could validly be conducted.
3.5.1. Multicollinearity
of the coefficients on other variables to change. In this case, the statistical power of the model would be difficult to interpret. But, a small degree of correlation between independent variables is normal and will almost always occur. It will only become a problem when these variables are very highly correlated. The employment of a fixed effects model instead of a pooled OLS can reduce this problem of high correlation. In comparison to a pooled OLS, the advantage of using a fixed effects test is that the combination of cross-sectional and time series data can mitigate problems of multicollinearity due to the additional variation across cross-sections. The Variance Inflation Factor (VIF) test and Pearson’s correlation matrix between all variables are employed to measure the extent of multicollinearity. The VIF provides an estimate to what extent the variance of a parameter estimate increases because the explanatory variables are correlated. The value of VIF>10 is used as a threshold to indicate whether multicollinearity is sufficiently large to be a cause for concern. The result of this test will be discussed in section 4.1.
3.5.2. Heteroskedasticity
The Breusch-Pagan and White test is performed to detect heteroscedasticity. This concept refers to the circumstance in which the variance of the errors are not constant. The existence of heteroscedasticity is a violation of one of the Gauss-Markov requirements, which assumes that the variance of the errors are homoscedastic. Heteroscedastic errors result in wrong standard errors and therefore misleading inferences (Brooks, 2019). If the form of heteroscedasticity is present, robust standard errors are used to correct for the consequences of misleading t-statistics due to wrong standard errors. The technique of using White’s robust standard errors allows the employment of standard error estimates that have been modified to account for heteroscedasticity, so that more evidence would be required against the null hypothesis before it would be rejected (Brooks, 2019).
3.5.3. Autocorrelation
Cluster-robust standard errors, also called heteroskedasticity autocorrelation (HAC) standard errors, can be added to the model to control for any presence of autocorrelation.
3.6. Robustness checks
Robustness checks are necessary to verify the robustness and plausibility of the coefficients by adding or removing regressors (Lu and White, 2014). In doing so, any differences in the magnitudes or signs of the coefficients will be examined. This research will add three more variables to the core regression model of the overall ESG score as robustness check; interest coverage ratio, market-to-book ratio and the current ratio. Furthermore, the independent variables are lagged for two and three more years as additional robustness checks. Lastly, the pillars will be measured jointly in one regression model as an additional robustness check for the individual ESG pillars. By doing this, possible changes in the robustness of the results are studied.
3.7 Descriptive Statistics
Table 1 summarizes the descriptive statistics of the variables conducted in this study. The final sample consists of 189 bond-issuing firms and 798 corporate bonds listed on the NYSE, NASDAQ, London Stock Exchange and the Deutsche Börse AG during the period 2014-2019. Each bond can generate multiple observations for each year it was active within the period 2014-2019. This results in a maximum of 3,211 observations. However, as can be seen in Table 1, this study deals with an unbalanced panel dataset because many variables have different observations.
Table 1. Descriptive statistics
N Mean SD Min Med Max Skewness Kurt-
osis Winsor- -ized Dependent variable: LN_YS 3,084 0.274 0.794 -3.996 0.296 1.807 -2.140 12.514 0.010 Independent variables: ESG 2,948 59.379 18.632 1.540 63.540 92.960 - 0.599 2.584 - ENV 2,949 54.778 26.342 0.000 59.500 98.530 - 0.538 2.195 - SOC 2,949 60.943 21.047 4.800 64.720 97.790 - 0.384 2.282 - GOV 2,950 61.460 21.480 0.250 64.910 98.450 - 0.601 2.628 -
Bond control variables:
MATURITY 3,211 14.512 10.212 0.000 10.000 60.000 0.931 3.149 -
MOD_DUR 3,084 9.257 4.803 0.033 7.421 20.857 0.433 1.817 -
MARKET_VALUE 3,084 741.800 538.070 7.379 568.634 2381.724 1.334 4.419 0.025
CRED_RATE 2,867 14.41577 3.010 1.000 15.000 21.000 - 0.680 4.983 -
Firm control variables:
LN_MARKCAP 3,211 17.268 1.436 13.57967 17.220 20.436 - 0.1889 2.674 0.010
DEBT_EQ 3,210 126.239 153.272 - 284.12 109.75 512.110 -0.050 5.410 0.050
ROA 3,211 7.203 5.761 - 11.530 6.490 26.170 0.3915 5.073 0.010
Note: this table represents an overview of the descriptive statistics of key variables.
According to Table 1, the yield spread is transformed into a natural logarithm to adjust for the positive skewness. The same transformation holds for the variable of the firm’s market capitalization.
By looking at Table 1, the overall- and individual ESG scores have a large deviation between the minimum, median and maximum value. Subsequently, the mean ranges between 54 and 62 with large standard deviations. This put forward the existence of large distinctions between the combined- and individual scores of firms on the ESG metrics in the dataset.
an approximately left-skewed distribution because of the long left tail. This is supported with a value of -0.599 for the skewness of the overall ESG score in Table 1. The histograms of the individual scores on the ESG metrics can be found in Appendix B. All these figures show a similar left-skewed distribution. Again, this can be supported by all the negative values regarding the skewness of these explanatory variables (Environment; Social; Governance). A possible explanation for this distribution is that firms can voluntarily disclose CSR activities. Therefore, firms are more likely to disclose CSR activities if they score well on these metrics (Dhaliwal, Li, Tsang, and Yang, 2011).
Figure 1. Histogram of the overall ESG scores
According to the bond control variables, the average maturity of a bond is 14.5 years with a minimum- and maximum value of less than one year (0) and 60 years respectively. The market value of bonds ranges from a value 7.379 thousand to 2381.724 thousand with a mean of 741.800 thousand and standard deviation of 538.070 thousand. Hence, this gives a good indication about the wide range in the nominal amounts issued of corporate bonds. The observations of credit rating are much lower in comparison to other variables in the sample. The reason for this is that the sample contains 67 bonds with withdrawn-or missing ratings. All firm control variables are ratios with exception of the variable for market capitalization.
0-10 10-20 20-30 30-40 40-50 50-60 60-70 70-80 80-90 90-100 Fr eq uen cy 0 100 200 300 400 500 600 700 800
3.4.1. Sample population of industries and maturity period
Since this study has to control for the industry effects, Table 2 provides a comprehensive overview of the amount of bonds issued by firms and their corresponding industry. As can be seen in Table 3, the manufacturing sector is the largest bond issuing industry, with 39.47%. The electricity, gas & sanitary services sector is the second largest bond issuing industry within this sample, with 16.15%. This is in line with the statistics of the SIFMA quarterly research reports5 of US fixed income markets over the past years.
Table 2. Sample population of the industries (SIC-code two digit)
Industry N (Bonds) % of total amount
Agriculture, Forestry & Fishing 2 0.250
Mining 28 3.509
Construction 7 0.877
Manufacturing 315 39.474
Transportation 70 8.631
Communications 70 8.631
Electric, Gas & Sanitary Services 131 16.153
Wholesale Trade 16 1.973
Retail Trade 85 10.481
Finance, Insurance & Real Estate 9 1.110
Services 65 8.015
Total 798 100
5The US fixed income markets are the largest in the world. The SIFMA Research Quarterlies contain
Table 3. Sample population of the maturity periods
Period (in years) N (Bonds) % of total amount
1-15 501 62.78
³ 15 297 37.22
In order to answer the third hypothesis, the sample will be split into two periods. Table 3 shows the amount of bonds with their corresponding maturity periods to gain further insight about the long term effects of ESG scores on the cost of corporate bonds.
4. Results and discussion
4.1 Model specificationAs mentioned in the previous chapter, it is necessary to test for the model specification when conducting empirical research. According to the outcome of the redundant fixed effect test, this study observes heterogeneity in the model and indicates a preference for the use of a panel estimation technique.
Next, the Durbin-Wu-Hausman test is conducted to check the suitability of using fixed-or random effects. The result of the test indicates a rejection of the null hypothesis. Thereffixed-ore the fixed effects model is suitable.
As the fixed effects model is preferred, the Wald test is performed to check the necessity of including entity and/or time fixed effects. Based on the outcome, the null hypothesis is rejected in both cases which indicates that entity- and time fixed effects significantly improve the fit of the model. As the sample contains multiple bonds per firm in different industries over time, there is a high probability of cross-sectional dependence. In this case, clustering in the cross-sectional dimension is necessary to adjust for residuals that are correlated across bonds. Furthermore, the fixed effects model cannot incorporate both bond-and industry fixed effects into the model. Different characteristics, such as industry characteristics, would be implicitly captured by the intercept with bond fixed effects. Therefore, I will use an OLS regression with bond-level clustered standard errors to mitigate potential autocorrelation and heteroscedasticity. In addition, I will use industry and time indicators to control for any differences across industries and over time.
correct for the consequences of misleading t-statistics due to wrong standard errors. Next to heteroscedasticity, the outcome of the Wooldridge test puts forward that the sample also contains a pattern of autocorrelation. Hence, clusteringbonds is effective to account for autocorrelation in the residuals.
This is in line with existing literature (Ge and Liu, 2015; Cooper and Uzun, 2015). These studies use an OLS regression with firm-level clustered- and robust standard errors to mitigate potential autocorrelation and heteroskedasticity. In addition, both studies use industry and year indicators to control for any differences across industries and over time.
Furthermore, as mentioned in the previous section, it is imperative to check for multicollinearity. Table 1 and 2 in Appendix C present the results of the VIF test for multicollinearity and demonstrate, with a mean VIF 4.29 and 3.97, that variables of interest do not show very high values of VIF, with the exception of the variables maturity and modified duration. In line with Oikonomou, Brooks, and Pavelin (2014), both variables will be kept in the model due to the economic intuition behind each of these explanatory measures. Subsequently, the use of both variables in the model will be checked by dropping iteratively one of the two collinear variables at each time. These results indicate that the robustness of the results are unaffected in terms of each magnitude, statistical significance and sign of the coefficient.
Another important feature is the correlation matrix represented by Table 3 in Appendix C. The correlation matrix contains all correlations between the included variables of this study. As shown in Table 3, the yield spread has a negative correlation with the combined- and all individual ESG scores at the 1% significance level, ceteris paribus. It is important to note that these outcomes are not corrected for differences between industries over time. Overall, Table 3 in Appendix C exhibits no significantly high correlations between explanatory variables with the modified duration and maturity, and the overall as well as individual ESG scores. The correlation between the overall-and individual ESG scores does not raise any new issues because each independent variable is used in a different regression model. All individual pillars will also be tested in one regression model to check for any differences.
industry and year indicators are included to account for differences across industries and over time. This applies to every model in order to answer the hypotheses.
4.2. The effect of the overall ESG score
The first hypothesis concerns the impact of the overall ESG score on the yield spread. To get the bigger scheme of this study, Table 4 represents the estimated coefficients with the robust standard errors in the parentheses. The goodness-of-fit of the models is illustrated by the adjusted 5& and the stars above coefficients indicate the level of significance.
Model 1 represents the final results of the main model, which answers the first hypothesis. In line with prior research, I find a negative relationship between the overall ESG score and the corporate yield spread that is statistically significant at the 1% level, ceteris paribus (Polbennikov, Desclée, Dynkin, and Maitra, 2016). Hence, firms with higher ESG scores face lower cost of corporate bonds. This result supports the first hypothesis and is consistent with the stakeholder theory of Freeman (1984). From a stakeholder perspective, one possible explanation for this could be that ESG-related disclosures can reduce the information asymmetry, and therefore, the cost of bonds. This is consistent with the risk mitigation view of Goss and Roberts (2011). The risk mitigation view suggests that investors are able to assess the risk profile of bond issuing firms because ESG-related disclosures can reflect the extent of risk exposure. Hence, the increase in information can result in a reduction of adverse selection and moral hazard, which in turn leads to lower cost of capital. Another explanation could be that firms with higher ESG scores face less litigation risk because companies that are more socially responsible are less likely to be faced with lawsuits regarding environmental issues and social norms (Hong and Kacperczyk, 2009).
Although the sign and statistical significance of the results clear, it is not apparent to make any inference about the economic significance of the coefficients. As mentioned before, the yield spread has been transformed into a natural logarithm to account for its positive skewness. Therefore, it is hard to interpret the coefficients in comparison to the standard outcome of a normal regression. To deal with this, the change in percentage of a natural logarithm has to be calculated by taking the exponent of the coefficient minus one (Brooks, 2014).
ESG score with 10 points. This increase would cause the yield spread to decrease with 0.0342%. Hence, a decrease of around 3 basis points. Based on the results of prior research (Goss and Roberts, 2011; Oikonomou, Pavelin, and Brooks, 2014), the economic significance of this test result is rather small. Given this result, I assume that a firm’s overall ESG score is a second-order determinant of the yield spread.
Furthermore, all bond-and firm control variables have the expected signs and are statistically significant except for the debt to equity ratio. As expected, the credit rating, market capitalization and return on assets are associated with lower yield spreads. The maturity, modified duration and market value all have the expected positive relationship with the yield spread. The majority of these results are in line with similar studies on this topic (Ge and Liu, 2015; Oikonomou, Pavelin, and Brooks, 2014). The adjusted 5& of the model is 0.6437
which means that the model explains around 64,37% of the variance in the yield spread by the variables on the right hand side of the model.
4.3 The effect of the individual ESG scores
Table 5 represents the test results of the individual ESG pillars (Environmental; Social; Governance). Model 1 shows the results with the environmental score as the explanatory variable. Next, model 2 shows the results with the social score as the explanatory variable. In addition, model 3 shows the results with governance as the explanatory variable. Model 4 shows the results of measuring the pillars jointly in one regression model and will be discussed in section 4.5 as an additional robustness check. Table 5 shows the coefficients of the corresponding variables and the cluster-robust standard errors within the parentheses. All four models have the natural logarithm of the yield spread as the dependent variable.
study of Schneider (2011). The study found that future clean-up and compliance costs of corporations can be high enough to threaten the ability to meet their debt obligations.
Table 4. Results of the overall ESG score
Model 1 Variables LN_YS ESGt-1 -0.0032*** (0.0009) MATURITYt-1 0.0131** (0.0061) MOD_DURt-1 0.0282* (0.0157) MARKET_VALUEt-1 0.0001*** (0.0000) CRED_RATEt-1 -0.1340*** (0.0079) LN_MARKCAPt-1 -0.0715*** (0.0146) DEBT_EQt-1 -0.0001 (0.0000) ROAt-1 -0.0110*** (0.0030) Constant 3.4114*** (0.2018) Observations 2,013 Adj. R-squared 0.6408
Industry indicators Yes
Year indicators Yes
Note: The table presents the OLS regression results with (bond-level) clustered- and robust standard
Furthermore, I find a statistically significant relationship between the social pillar and the yield spread. Model 2 shows that a one point increase of the social pillar causes the yield spread to decrease with 0.0035%, ceteris paribus. This result shows that the social pillar has a stronger impact on the yield spread in comparison to the environmental pillar. This can be supported with evidence from existing literature (Jo and Harjoto, 2011). The authors found that internal social enhancements within a firm, such as product quality, workforce diversity, and working conditions, enhances the value of the firm more than other CSR dimensions. A higher firm value would indicate a better credit rating and thus lower cost of debt (Hull, 2020). A high social performance in the form of a management’s commitment towards maintaining a good reputation, protecting public health and respecting business ethics can be effective in reducing idiosyncratic risk (Sassen, Hinze, and Hardeck, 2016). In addition, Hasan, Hoi, Wu, and Zhang (2017) found that social capital can constrain opportunistic firm behaviour and, consequently reduce the cost of bank loans.
Table 5. Results of the individual ESG pillars
Model 1 Model 2 Model 3 Model 4
Variables LN_YS LN_YS LN_YS LN_YS
ENVt-1 -0.0024*** - - -0.0012 (0.0007) (0.0008) SOCt-1 - -0.0035*** - -0.0027** (0.0009) (0.0011) GOVt-1 - - -0.0006 0.0006 (0.0005) (0.0006) MATURITYt-1 0.0129** 0.0127** 0.0129** 0.0126** (0.0061) (0.0061) (0.0062) (0.0061) MOD_DURt-1 0.0290* 0.0290* 0.0287* 0.0292* (0.0157) (0.0156) (0.0162) (0.0156) MARKET_VALUEt-1 0.0001*** 0.0001*** 0.0001*** 0.0001*** (0.0000) (0.0000) (0.0000) (0.0000) CRED_RATEt-1 -0.1341*** -0.1337 -0.1357*** -0.1335*** (0.0079) (0.0079) (0.0081) (0.0079) LN_MARKCAPt-1 -0.0684*** -0.0690*** -0.0889*** -0.0641*** (0.0153) (0.0149) (0.0140) (0.0155) DEBT_EQt-1 -0.0001 -0.0000 -0.0001 -0.0001 (0.0001) (0.0001) (0.0001) (0.0001) ROAt-1 -0.0115*** -0.0104*** -0.0104*** -0.0108*** (0.0030) (0.0029) (0.0029) (0.0029) Constant 3.3037*** 3.3718*** 3.5877*** 3.2737*** (0.2104) (0.2034) (0.2072) (0.2156) Observations 2,013 2,013 2,013 2,013 Adj. R-squared 0.6413 0.6425 0.6357 0.6432
Industry indicators Yes Yes Yes Yes
Year indicators Yes Yes Yes Yes
Note: The table presents the OLS regression results with (bond-level) clustered- and robust standard
4.4 The impact of ESG scores and the length of the investment horizon
In order to answer the third hypothesis, Table 1 in Appendix D presents the results of the different lengths in the investment horizon. The hypothesis was that the effect of the overall ESG score on the yield spread would be more pronounced in the long term. In line with Oikonomou, Pavelin, and Brooks (2014), the maturities of bonds are used as a proxy for the investment horizon. The short-term bonds have a maturity length of 1-15 years. The long-term bonds have a maturity length of 15 years and older. Prior research provides evidence about the positive relation between long-term investment and corporate social performances (Cox, Brammer, and Millington, 2004; Graves and Waddock, 1994). However, as can be seen from Table 1 in Appendix D, the results show the opposite.
4.5. Additional robustness checks
This section discusses the additional robustness checks to examine the behaviour of the core estimate coefficients of the regression model. In doing so, I try to mitigate concerns of reverse causality or other types of endogeneity. Similar ESG related studies suggest that endogeneity and the omitted variable bias might weaken the relationship between ESG and the cost of bonds (Eliwa, Aboud, and Saley, 2019; Ge and Liu, 2015). For instance, it could be possible that other omitted variables may be driving the results. Following prior related studies, a recurring example is that firms with a good financial performance are more likely to invest in CSR activities and are associated with lower cost of debt (Ge and Liu, 2015). Overall, the results of the robustness checks are not quantitatively different from the primary model.
Table 2 in Appendix D presents three regression models. Model 1, 2 and 3 present the test results after adding the interest coverage ratio, market to book ratio and current ratio to the primary model. These results can be compared with Table 4 of the primary model in section 4.2. The majority of the variables remain stable except for the sign and significance of the debt to equity ratio in model 2 and 3. This change is consistent with my expectation of the debt to equity ratio because the coefficient is now a positive and statistically significant.
In line with Stellner, Klein, and Zwergel (2015), the independent variables are lagged by two-and three more years as additional robustness checks. The results can be found in Table 3 in Appendix D. All the signs of the coefficients remain stable in both models. Only the statistical significance has changed for most of the coefficients. Therefore, the majority of the results survived the additional robustness checks.
5. Conclusion
This study examined the relationship between CSR and cost of corporate bonds. Only a limited number of studies investigated the relationship between CSR and fixed income securities. This empirical study aimed to add to the existing literature by investigating the impact of overall- and individual ESG scores on the yield spread of corporate bonds. The sample consists of 189 bond issuing firms and 798 corporate bonds listed on the NYSE, NASDAQ, London Stock Exchange and the Deutsche Börse AG during the period 2014-2019. I have created multiple hypotheses to answer my main research question.
First of all, the impact of ESG scores can be interpreted from a shareholder- or stakeholder perspective. The shareholder theory suggests that investments in ESG activities are a waste of resources because it can increase a firm’s distress risk. In contrast, the stakeholder theory suggests that ESG activities can be beneficial for investors because it can reduce the information asymmetry and thus the cost of bonds.
In line with the first hypothesis, I find a negative and statistical significant relationship between the overall ESG score and the yield spread. Firms with a better ESG performance are able to issue bonds at lower costs. This result supports the stakeholders theoretic framework. ESG-related disclosures can reduce the information asymmetry, and therefore, the cost of bonds. This might indicate that ESG metrics can help identify new risk factors for investors. However, the impact of the result is so small, namely a reduction of 3 basis points, that it can only be a second-order determinant of the yield spread.
In contradiction to the third hypothesis, the effect of ESG scores seem to have a stronger impact on the yield spread of bonds with short-term maturities. It could potentially be that investors are more short-term oriented and perceive firms with high ESG scores to be more risky in the long term. In addition, investors might expect that the future benefits of ESG investments will not live up to their expectation. This could explain the bigger reduction in the yield spread of bonds with shorter-dated maturities.
This research contributes to the existing research of the relationship between CSR and the cost of debt. It supports the research that does find a statistically negative association between ESG scores and the cost of corporate bonds. Given the increasing global trend of CSR, this may be useful for banks, institutional investors or other types of fixed income investors in providing an alternative way to interpret risk with non-financial disclosures.
References
Aman, H., Nguyen, P., 2013. Does good governance matter to debtholders? Evidence from the credit ratings of Japanese firms. Research in International Business and Finance 29, 14-34.
Attig, N., El Ghoul, S., Guedhami, O., Suh, J., 2013. Corporate social responsibility and credit ratings. Journal of Business Ethics 117, 679-694.
Barnea, A., Rubin, A., 2010. Corporate social responsibility as a conflict between shareholders. Journal of Business Ethics 97, 71-86.
Bauer, R., & Hann, D., 2010. Corporate environmental management and credit risk. Unpublished working paper. Maastricht University
Bennani, L., Le Guenedal, T., Lepetit, F., Ly, L., Mortier, V., Roncalli, T., Sekine, T., 2018. How ESG Investing Has Impacted the Asset Pricing in the Equity Market. SSRN. 1-19.
Bhojraj, S., Sengupta, P., 2003. Effect of corporate governance on bond ratings and yields: The role of institutional investors and outside directors. The Journal of Business 76, 455-475.
BlackRock, 2019. Sustainability: the bond that endures. Retrieved From
https://www.blackrock.com/us/individual/literature/whitepaper/bii-sustainable-investing-bonds-november-2019.pdf
Bloomberg, 2019. As demand for ESG investing grows, so too does the need for high-quality data. Retrieved from
https://www.bloomberg.com/professional/blog/demand-esg-investing-grows-need-high-quality-data/
Brooks, C., 2019. Introductory Econometrics for Finance. Cambridge University Press
Cochran, P.L., & Wood, R.A., 1984. Corporate social responsibility and financial performance. Academy of Management Journal 27, 42-56.
Commission of the European Communities, 2003. Report from the Commission on European governance. Office for Official Publications of the EC. Luxembourg.
Cooper, E. W., Uzun, H. 2015. Corporate Social Responsibility and the Cost of Debt. Journal of Accounting & Finance 15. 11-29.
Cox, P., Brammer, S., Millington, A., 2004. An empirical examination of institutional investor preferences for corporate social performance. Journal of Business Ethics 52, 27-43.
Crifo, P., Diaye, M.A., Oueghlissi, R., 2017. The effect of countries’ ESG ratings on their sovereign borrowing costs. The Quarterly Review of Economics and Finance 66, 13-20.
Dhaliwal, D.S., Li, O.Z., Tsang, A., Yang, Y.G., 2011. Voluntary nonfinancial disclosure and the cost of equity capital: The initiation of corporate social responsibility reporting. The Accounting Review 86, 59-100.
Dorfleitner, G., Kreuzer, C., Sparrer, C., 2020. ESG controversies and controversial ESG: about silent saints and small sinners. Journal of Asset Management 21, 393-412.
El Ghoul, S., Guedhami, O., Kwok, C.C., Mishra, D.R., 2011. Does corporate social responsibility affect the cost of capital?. Journal of Banking & Finance 35, 2388-2406.
Erragragui, E., 2018. Do creditors price firms’ environmental, social and governance risks?. Research in International Business and Finance 45, 197-207.
