• No results found

What does it cost to be green?

N/A
N/A
Protected

Academic year: 2021

Share "What does it cost to be green?"

Copied!
62
0
0

Loading.... (view fulltext now)

Full text

(1)

What does it cost to be green?

An empirical investigation of the European

green bond market

Master’s Thesis 30 credits

Department of Business Studies

Uppsala University

Spring Semester of 2020

Date of Submission: 2020-06-02

Anton Pettersson

Gustaf Söderström

(2)
(3)

Abstract

The green bond market offers investors the opportunity to take an explicit focus on sustainable investment projects. However, it is yet to be determined whether this novel asset class offers attractive yields compared to non-green bonds. To address this question, we study European green bonds and how they diverge from conventional bonds in terms of yields. Using a dataset of 88 matched pairs of European green bonds between 2015 and 2019, we document a significant negative green bond premium of -12 bps on average in the secondary market. The green bond premium is defined as the yield differential between a green and a conventional bond while controlling for liquidity. The results suggest that European investors accept a lower financial return in exchange for receiving non-pecuniary benefits and thus challenging the assumptions of classical asset pricing models. Furthermore, we use a matching method and two-step regression to control for liquidity and identify the determinants of the green bond premium. The results show that the negative green bond premium is less pronounced for lower-rated bonds. Moreover, we find support for variations in the green bond premium across different business sectors. Government-related green bonds experience a greater negative green bond premium than green bonds related to financials and industrial corporates.

(4)

Acknowledgement

We want to express our gratitude to our supervisor Derya Vural-Meijer at the Department of Accounting for her inspiration, guidance and support throughout the writing process. We also want to thank Kajsa Gejhammar and Sara Jolind, Master’s students in the Accounting & Finance Program, for their thoughtful suggestions during the writing process. Finally, we want to express our gratitude to our families and closest friends for providing us with encouragement and advice during the work of this thesis.

(5)

I. Table of contents

1. Introduction 1

1.1. Background 1

1.2. Research Problem 3

2. Theoretical Background and Literature Review 6

2.1. Bond Pricing Theory 6

2.2. The Capital Asset Pricing Model 7

2.3. Green Bond Standards 8

2.4. Previous Research 9

2.4.1. Sustainable Responsible Investments 9

2.4.2. Green bonds 10

2.5. Hypothesis formulation 14

3. Data and Methodology 16

3.1. Methodological approach 16

3.1.1. Panel Data Regression 16

3.2. Data description 17

3.2.1. Green bonds 17

3.2.2. Conventional bonds 19

3.3. Matching Procedure 20

3.3.1. Matching green bonds with conventional bonds 20

3.3.2. The estimation of the green bond premium 21

3.3.3. Determinants of the green premium 22

3.4. Statistical tests 26 3.4.1. Heteroskedasticity 26 3.4.2. Serial Correlation 26 3.4.3. Normality 26 3.4.4. Multicollinearity 26 3.5. Limitations 27

4. Results and Analysis 28

4.1. Descriptive Statistics 28

4.2. The Green Bond Premium 29

4.2.1. Hypothesis 1 - The existence of a green bond premium 29

4.3. Factors explaining the green bond premium 33

4.3.1. Hypothesis 2 - Credit ratings 35

4.3.2. Hypothesis 3 - Business sectors 37

5. Discussion 39

5.1. The Green Bond Premium 39

5.2. A note on liquidity 41

5.3. The future of the green bond market 42

6. Conclusion 44

References 45

(6)

II. List of figures and tables

List of figures

Figure 1. Yearly issuance of green bonds 2

Figure 2. The Average Green Bond Premium Over Time 30

List of tables

Table 1. Summary of key studies on green bonds 13

Table 2. The green bond sample 20

Table 3. Summary of matching criterias 21

Table 4. Variable summary 25

Table 5. Descriptive Statistics 27

Table 6. The green bond premium 29

Table 7. Distribution of the green bond premium 30

Table 8. One Sample T-test of green bond premium 31

(7)

III. List of abbreviations

CAPM Capital Asset Pricing Model

CBI Climate Bond Initiative

CBS Climate Bond Standards

CSP Corporate Social Performance

CSR Corporate Social Responsibility ESG Environmental, Social and Governance

GBP Green Bond Principles

ICMA International Capital Market Association SRI Socially Responsible Investment

(8)

1. Introduction

1.1. Background

An increasingly urgent concern today is climate change and its impact on the environment and society in the future (Moss et al., 2010). Widespread uncertainty rests in future of climate change, but it stands clear that the implications of climate change will depend on humankind responses through economics, technology and politics (ibid). The climate concern exacerbates the need for financing to facilitate the transition into a low-carbon economy. According to the World Bank (2018), investments of approximately USD 90 trillion by the year 2030 is required to produce and develop new infrastructure and reduce carbon-emission.

The notable gap between supplied funding and required financing is one of the significant hurdles for climate action (Karpf & Mandel, 2018). This necessitates substantial investments and, consequently, private-sector investments into low-carbon infrastructure to be scaled up significantly (Della Croce, 2011). Given this, the need to align capital markets and sustainability has perhaps never been greater. As important suppliers of capital, financial investors are essential due to their capacity to mobilize the amount of capital required for investments of this size (Zerbib, 2019). Besides, investors’ increased demand for pro-environmental attributes upturns their investments in ethical assets while excluding sin-stocks from the investment portfolio (ibid.). Generally, studies confirm that environmental, social and governance (ESG) criteria have a positive impact on corporate financial performance and evidence suggests that this effect is even higher for bonds than equities (Friede et al., 2015). Against this background, the emergence of new financial instruments focusing on sustainable investments, with green bonds being the most prominent, is emphasized as an essential and critical funding vehicle to bridge the financing gap by encouraging portfolio allocation towards sustainable investments (Karpf & Mandel, 2018).

The International Capital Market Association (ICMA) has defined green bonds in their Green Bond Principles (GBP) as: “Green Bonds are any type of bond instrument where the proceeds will be exclusively applied to finance or refinance, in part or in full, new and/or existing eligible Green Projects” (ICMA, 2018). Since its introduction in 2008, green bonds have emerged as an alternative for conventional financing, in various sectors, issued by governments, on a national level and the local level (municipalities), corporates, and international organizations (ibid). Similar to traditional fixed income securities, firms issue green bonds to finance their investments (Tang & Zhang, 2018).

(9)

largest sectors are financials and corporates, which combined accounts for 50% of the green bond market issuance, with Europe and North America being the primary geographical drivers for the continued development of green bonds (SEB, 2019).

Figure 1. Yearly issuance of green bonds

The intentions of investors to invest in a green bond can be categorized into non-financial and financial motives. Non-financial motives include ethical concerns, where investors derive utility from positive environmental benefits and by addressing the long-term risks associated with climate change (Hong & Kacperczyk, 2009; Tang & Zhang, 2018). Evidence suggests that investors are willing to pay for the non-pecuniary attributes of investments (Bauer & Smeets, 2015; Barber et al., 2019; Zerbib, 2019). The financial motives, on the other hand, are associated with the expectation of lower risk (Inderst et al., 2012; Kruger, 2015), diversification (Nanayakkara & Colombage, 2019) or better financial performance (Friede et al., 2015). Additionally, green bonds provide issuers access to a previously unapproachable market of investors, to whom environmental concerns and long-termism are crucial (Flammer, 2020).

(10)

1.2. Research Problem

Regarding their financial structure, green bonds are identical to conventional bonds in the sense that the issuer of the bond owes debt towards the holder of the bond and is required to pay the bondholder interest (coupon payments) and repay the principal at maturity (Zerbib, 2019). Thus, following the modern portfolio theory by Markowitz (1952) and the capital asset pricing model (CAPM) of Sharpe (1964) and Lintner (1965), there should not exist any pricing difference between a green labelled bond and their non-green counterpart given the same financial characteristics, ceteris paribus. Hence, by extending the work by Markowitz (1952), an investment strategy consisting of both green bonds and conventional bonds should not alter the standard risk-return optimization. However, the majority of the research that isolates the effect of non-financial drivers on market prices indicates pricing differences between green bonds and conventional bonds (e.g. Ehlers & Packer, 2017; Baker et al. 2018; Nanayakkara & Colombage, 2019; Gianfrate & Peri, 2019; Zerbib, 2019). Nevertheless, the existing research is ambiguous in determining to what extent non-pecuniary motives explain this yield differential, which has become known as a green bond premium (Zerbib, 2019). Also, prior research shows that the green bond premium varies depending on determinants such as business sector, credit rating, issue amount, currency, maturity, seniority etc. (e.g. Hachenberg & Schiereck, 2018; Zerbib, 2019; Gianfrate and Peri (2019)). Further, Europe has received limited attention in prior research, despite being the largest green bond market (CBI, 2020a). Green bonds should attract increased demand from investors to whom environmental and sustainability factors are essential (Karpf & Mandel, 2018; Bachelet et al., 2019). Since European investors, in general, are more sustainable oriented (Schroders, 2019), the demand for green bonds should be high in Europe. Higher demand for a bond tends to imply a lower liquidity risk, generating lower yield (Amihud & Mendelson, 1986).

(11)

or higher yields. Nanayakkara and Colombage (2019) find that the green label incentivizes issuers to raise funds through green bonds while providing investors with an opportunity to diversify the returns on their investments. However, some inconsistency is evident, and the studies addressing the issue of sustainability advocates that both investors and firms stand to profit from assets with robust sustainability metrics (e.g. Ehlers & Packer, 2017; Hachenberg & Schiereck, 2018; Nanayakkara & Colombage, 2019; Zerbib, 2019).

The empirical findings from previous research cast doubt on the underlying assumptions of the traditional asset pricing models. By integrating into investors’ utility function an appetite for specific types of assets, Fama and French (2007) argue that the preferences of investors could be an essential aspect of asset pricing. Thus, any differences in asset pricing can, instead of being attributed to expectations of future cash flows, depend on investor preferences for non-pecuniary motives. Sustainable investing is based on convictions generating additional non-pecuniary compensations, and not necessarily optimizing financial rewards (Renneboog et al., 2008; Ibikunle & Steffen, 2017). In theory, for mean-variance maximizers, and with the financial features being the same, green bonds should attract equal demand as conventional bonds, for which environmental aspects are not the deciding factor (Baker et al., 2018; Karpf & Mandel, 2018). Baker et al. (2018) hypothesize, by using a modified CAPM, that lower expected returns follow higher environmental scores. This contradicts the belief of strictly return-focused investors, proposing that investors can achieve utility in non-pecuniary factors such as sustainability-related factors. Furthermore, in support of the arguments by Fama and French (2007), the appetite for such factors may modify equilibrium prices and provides new insights into the classical theory of asset pricing and investor-optimization.

Green bonds foster the awareness for the hurdles of climate change and demonstrate the possibility for financial investors to fund climate-smart investments through liquid investments without sacrificing financial returns (World Bank 2019). Thus, green bonds are not constructed to have a premium nor carry more risk than conventional bonds. In particular, as a critical factor in mitigating climate challenges, it is of interest to observe the relative returns by investing in green bonds compared to non-green bonds, and whether or not investors pay a premium price. According to Hachenberg and Schiereck (2018), the future success of green bonds in terms of aligning capital markets and sustainability is dependent on its pricing and performance. Therefore, this thesis examines whether a green bond premium exists in the European secondary bond market and identify possible determinants of that premium. The reason to focus on the European bond market, aside from it being the largest green bond market (CBI, 2020a), and European investors are highly sustainable oriented (Schroders, 2019), is that policymakers in Europe have put a strong focus on reorienting capital flows towards sustainable investments (EU, 2018). This suggests that we can expect a continued increase in demand for green bonds in Europe in the near future.

(12)

traditional asset pricing models motivates this study. Considering that and all the above, this thesis aims to provide answers to the following research questions:

How does the pricing compare between green bonds and conventional bonds on the European secondary bond market?

What could be possible determinants of a pricing difference between green bonds and conventional bonds?

In order to answer these research questions, we use a matching method similar to Zerbib (2019) to assess the yield of an equivalent synthetic conventional bond for each green bond. A counterfactual conventional bond is built from the same issuer, with identical bond characteristics, as well as a limited difference in issue size and maturity. Through a fixed-effects panel regression, we extract the green premium while controlling for residual differences in liquidity between the green bond and the conventional bond. To identify possible determinants of the green bond premium, the green bond premium is explained through a cross-sectional regression specified based on the specific determinants of the bonds. Our analysis is performed on a sample of 88 European green bonds on the secondary market. The results show that a negative green bond premium of -12 bps on average exists 1 for European green bonds in the secondary market between 2015 and 2019. Further, we show that credit rating and business sector are significant drivers of the green bond premium. More specifically, we find that the negative premium is more pronounced for government-related bonds and less pronounced for green bonds with lower credit rating.

We contribute to the existing literature on green bonds in several ways. First, we provide clear evidence of a green bond premium in the European secondary bond market, and we identify determinants that can explain variations in the green bond premium, namely, credit ratings and business sectors. Second, we focus solely on the European bond market, which, to our best knowledge, only Gianfrate and Peri (2019) have done before us, and they limit their sample to EUR-denominated bonds and do not control for liquidity. Thus, our work contributes to the existing literature on sustainable investments by identifying and analyzing the effect of pro-environmental preferences on the European secondary bond market. Third, we use more recent bond data up till December 2019 to capture any changes in the green bond market, considering the exponential growth in cumulative issuance for green bonds during recent years (see figure 1).

The remainder of this thesis is structured in the following way; Chapter 2 outlines the determinants of bond pricing and review the current literature on sustainable finance and green bonds. Chapter 3, derives the regression models and presents the matching method. Chapter 4 presents the results, while chapter 5 and chapter 6 conclude the paper.

(13)

2. Theoretical Background and Literature Review

2.1. Bond Pricing Theory

To isolate the determinants of the green bond premium in empirical research, it is essential to specify the determinants of bond pricing. Bond prices are determined primarily by three main drivers, according to Merton (1974). First, bond prices are benchmarked against the rate of return for a risk-free asset (ibid.). Second, the characteristics of the bond indenture, such as maturity, call terms, coupon rate, sinking fund, seniority impact the pricing of the security (ibid.). Third, is the probability of default by the bond issuer (ibid.). Corporate bonds, unlike default-free government bonds, incorporate credit risk and bond prices are partly determined by the probability of default of the bond issuer (ibid.). Some corporate bonds may default and, in order to compensate investors for the expected loss from default, they require a higher yield (Fons, 1987; Elton et al., 2001).

However, researchers have identified that spreads on corporate bonds tend to be more extensive than what is explained by Merton’s (1974) three drivers, this is what has become known as “the credit spread puzzle”. See Collin-Dufresne et al. (2001), Collin-Dufresne et al. (2002), Driessen (2005) and Covitz and Downing (2007) for an in-depth discussion around the credit spread puzzle. There are several plausible explanations for the credit spread puzzle, where most researchers argue that bond liquidity is a possible solution (Houweling et al., 2005; Lin et al., 2011; Helwege et al., 2014).

Liquidity is a primary attribute, and a decisive element of financial instruments and investors demand compensation, in the form of a premium, for illiquid securities (Amihud & Mendelson, 1986; Li et al., 2009; Bao et al., 2011). Boudoukh and Whitelaw (1993) and Vayanos (1998) find that illiquidity disrupts the trading frequency and transaction costs induce liquidity discrepancies between assets, where higher transactions costs decrease liquidity. Furthermore, as investors frequently cannot hedge their risk against liquidity, they require an ex-ante premium to trade illiquid securities (Chen et al., 2007). As a result, bonds encountering higher liquidity risk realize higher yields (Amihud & Mendelson, 1986). Taking the bond’s cash flows as given, the bond yield is a promised yield where illiquid bonds imply higher bond yields, causing the yield spread to increase (ibid.).

(14)

The excess demand of green bonds causes the green bond market to become skewed, making the green bond market more liquid for green bondholders, but less so for prospective investors not in possession of a green bond (Wulandari et al., 2018). The liquidity for green bonds is of particular concern due to the supply shortage and relatively low issuance volume, and making the liquidity effect not as straightforward (ibid.). Two main reasons can explain such deficit: (1) the obscurity of green bonds providing financial incentives for investors (Wulandari et al. 2018; Hempstead & Kuchtyak, 2019; Zerbib, 2019); and (2) a lacking classification system for green bonds that are following market-based frameworks (Cochu et al., 2016; Wulandari et al., 2018).

2.2. The Capital Asset Pricing Model

Asset pricing models share the assumption that investors know the true joint distribution of asset payoffs (Sharpe, 1964; Lintner, 1965; Merton, 1973; Breeden, 1979). Fama and French (2007) argue that this assumption is unrealistic. Investors receive noisy signals about future asset returns and update their view based on these signals (ibid.). Another assumption in asset pricing models is that investors are only concerned about the portfolio return; that is, investment assets are not consumption goods (ibid.). The violations are plentiful. Cohen (2009) suggests that loyalty and investor preferences can cause investors to hold more of a particular security than what is justified based on payoff characteristics. For instance, some investors may have a preference for holding growth stocks (Daniel & Titman, 1997) or focus on sustainable assets (Geczy et al., 2005).

A pivotal question is whether sustainable investments provide value to investors exceeding the expected risk and return relationship. According to the assumptions of the standard asset pricing models, such as the capital asset pricing model (CAPM) of Sharpe (1964) and Lintner (1965) and the arbitrage pricing theory by Ross (1976), any pricing discrepancies of assets with identical characteristics of the bonds besides the green factor should be non-existing. In other words, investors in green bonds and conventional bonds should expect identical returns, ceteris paribus. Nevertheless, researchers have presented theoretical models suggesting that investors are willing to exchange financial benefits for environmental attributes (e.g., Heinkel & Kraus, 2001; Geczy et al., 2005; Fama & French, 2007; Friedman & Heinle, 2016; Baker et al., 2018).

(15)

(ibid.). Investors not having preferences for tastes combines the risk-free asset with the mean-variance efficient tangency portfolio of risky assets, as presented by Markowitz (1952). On the other hand, investors having taste preferences will not hold the optimized tangency portfolio due to the influence of these preferences (Fama & French, 2007). The disagreement between investors moves asset prices away from the equilibrium predicted by CAPM (Fama & French, 2007). The price effect of investors’ tastes for assets as consumption goods can be substantial when their positions significantly deviate from the market portfolio (ibid). Related to the work by Fama and French (2007), Zerbib (2020), empirically shows that the taste effect on asset returns takes form through a systematic taste premium. Notably, Petajisto (2009), argues that an exogenous change in demand for an individual asset has a limited impact on prices in a CAPM framework.

Although the final impact on bond pricing is ambiguous, the taste preferences of investors appear to constitute additional, yet unexplained, factors in determining bond prices (Fama & French, 2007; Zerbib, 2020). If the assumptions of CAPM holds on the European bond market, then no green premium should exist in the market since the model assume that investors are mean-variance optimizers. Thus, implying that factors such as sustainability should not have any impact on asset pricing.

2.3. Green Bond Standards

The issuance of green bonds is an unregulated area, and each country and issuer can decide on the criteria making the green bond classified as green (Larcker & Watts, 2020). For investors, standards provide a screening tool and regulation reduces the burden of investors having to make their subjective judgments during their due diligence (Hartzmark and Sussman, 2019). For issuers, on the other hand, certification serves as a voluntary initiative allowing them to demonstrate the integrity of their green bonds (ibid.). Certified green bonds can send a credible signal to investors that the issuer is actively engaged in environmental practice (Flammer, 2020).

However, to this date, no universal framework has been established to ensure compliance, but several initiatives have been undertaken to create a framework certifying the “greenness” of green bonds. Bachelet et al. (2019) argue that this makes the legitimacy of the green bond subject to asymmetric information originating from the bond’s “invisible” features. Today two primary standards are covering green bonds: Green Bond Principles (GBP) and Climate Bond Standards (CBS). The GBP, developed by the International Capital Market Association (ICMA), is the most extensively used standard for green bonds (ICMA, 2018). The GBP emphasize the accuracy, integrity, and transparency of the information reported to stakeholders (ibid.). Also, the GBP framework offers potential issuers guidance concerning the issuance of a green bond, how to ensure transparency, and how to certify the integrity of the green bond label (ibid.).

(16)

principles contained in GBP, creating a flexible, robust and effective certification system. However, compared to GBP, CBS is more rigorous and robust by including specific requirements such as a higher level of detail and requirements imposed on the issuers by CBS to ensure alignment. The stricter requirements concern eligibility of the proposed asset, the management and usage of proceeds, and non-financial documentation and reporting. As such, the two standards do not compete with each other (CBI, 2019a). However, critique has been raised that green bond certification still could allow for issuers of green bonds to present misleading claims about their environmental practices and therefore serve as a form of greenwashing (Flammer, 2020).

2.4. Previous Research

2.4.1. Sustainable Responsible Investments

Since the 1970s, more than 2000 studies have been published around environmental, social, and governmental (ESG) factors and its impact on financial investments (Friede et al., 2015). Close to 90% of these studies find a positive relationship between ESG and corporate financial performance (ibid). Menz (2010) argue that the interest in corporate social responsibility (CSR) has increased significantly in recent years, mainly driven by firms realizing the social necessity of developing holistic objectives. Particular for socially responsible investments (SRI) is that both social objectives and financial targets are pursued. In the global market, estimates have shown that SRI has reached USD 22.89 trillion; however, only a relatively small part of that constitutes of green bonds (Hachenberg & Schiereck, 2018). With the increasing popularity of SRI, evidence advocates that SRI investors are as keen as conventional investors to receive a financial return (Statman, 2000; Ibikunle & Steffen, 2017). Engaging in CSR-related activities often provide economic benefits (Sun & Cui, 2014), and is related to value maximization (Renneboog et al., 2008). Also, research confirms that corporate bonds with high ESG ratings tend to outperform their lower ESG rating counterparts and have tighter spreads (Polbennikov et al., 2016). However, the majority of research argues that SRI mainly provides non-financial compensation (see Renneboog et al., 2008).

(17)

However, not all studies find a positive relation between SRI and financial performance. Magnanelli and Izzo (2017) use a global sample to show that corporate social performance (CSP) increases the cost of debt, meaning that CSR may harm the performance of the firm. Additionally, Menz (2010) finds that socially responsible firms experience higher risk premium than non-socially responsible firms when issuing corporate bonds. However, Menz (2010) expected that socially responsible firms should have lower risk premia since they are often regarded as economically more successful and less risky, suggesting that CSR has not yet been incorporated in the pricing of bonds. Studies are identifying that high CSP lead to decreased levels of idiosyncratic risks (Lee & Faff, 2009).

In terms of credit rating, both Oikonomou et al. (2014) and Jiraporn et al. (2014) examine the influence of corporate social performance on corporate bond ratings and bond yields and show that corporate social performance is associated with higher ratings and lower yields. By extension, Sun and Cui, (2014) find that CSR has a significant impact on the reduction of a company’s default risk. Bauer and Hann (2014) find that bonds with better environmental performance are associated with lower credit risk, thus lowering bond yields. Stellner et al. (2015) find weak statistical support for a negative relationship between CSP and credit ratings, but argue that superior CSP is regarded as risk-reducing.

2.4.2. Green bonds

The majority of the previous studies analyzing pricing differences between green and conventional bonds identifies a yield difference for green bonds (e.g. Ehlers & Packer, 2017; Baker et al., 2018; Karpf & Mandel, 2018; Zerbib, 2019). The magnitude and sign of these findings differ depending on sample, methodology and institutional setting. In the green bond literature, yield differences are calculated, by creating a synthetic bond (e.g. Zerbib, 2019), by comparing credit spreads (e.g. Ehlers & Packer, 2017) or i-spreads (Hachenberg &2 3 Schiereck, 2018), in order to capture any pricing difference between green and conventional bonds.

Studying the credit spreads of 21 green and conventional bonds issued on the primary market in the U.S. and Euro area, between the years 2014-2017, Ehlers and Packer (2017) examine the pricing differences by the same issuer while minimizing the time between issuances. If a large enough proportion of investors is willing to pay a premium for green bonds, this should be reflected in the issuing price of such bonds (ibid.). Their result indicates a green bond premium of -18 bps. Ehlers and Packer (2017) argue that the high demand for green bonds, relative to supply, influence pricing. Also, researchers have identified factors such as supply and demand to be irrelevant for credit risk, but have a significant influence on bond yields (Greenwood & Vayanos, 2014).

Baker et al. (2018) analyze a large sample of green municipal and conventional bonds in the U.S. primary market for the investment period 2010 to 2016. Controlling for issue 2 ​A credit spread is the difference in yield between two bonds with similar maturity but different credit quality.

(18)

amount as a proxy for liquidity, they find evidence of a -7 bps after-tax yield premium, supporting the findings by Ehlers and Packer (2017). This premium would be considerable enough to justify the higher costs encountered by issuers at the time of green bond issuance (Baker et al., 2018). Further, Baker et al. (2018) suggest that only a subset of investors are willing to sacrifice return for holding a green-labelled bond, finding a high degree of ownership by long-term investors compared to conventional bonds. This is in congruence with Flammer (2020) who argues that green bonds attract buy-and-hold investors that emphasizes long-termism and environmental factors. The longer investment period could have an impact on green bonds being priced differently than conventional bonds (Petitt et al., 2015).

There are also studies identifying limited support for a yield premium for green bonds in the global market (Tang & Zhang, 2018). When only considering country effects, the authors find a premium of 6.9 bps. However, when adjusting for firm fixed effects, and examining yield spread differences of the same issuers, no statistical relationship is established. Also, Larcker and Watts (2020) examine the U.S. municipalities on the primary and secondary bond market. By matching 640 green bonds, the authors’ primary result is that the green premium is equal to zero.

In terms of the European bond market, Gianfrate and Peri (2019) use 121 EUR-denominated green bonds issued between 2013 and 2017 to perform an analysis on the primary and secondary bond market. By adopting a propensity score matching technique, Gianfrate and Peri (2019) find a green bond premium of about -20bps for the primary market and about -5 bps for the secondary market. Also, they find that these results are stronger for corporate bonds, suggesting that the private sector benefits more from issuing green bonds. Gianfrate and Peri (2019) denote the lower yields of green bonds as a consequence of strong demand for these financial products, reflecting the willingness of investors to fund green projects. However, Gianfrate and Peri (2019) did not control for liquidity; thus, a liquidity bias could be inherent in these results.

Studying the green bond premium and the effects of liquidity of a global sample in the secondary market, Zerbib (2019) evaluates the yield spread between 110 green and conventional bonds. After controlling for liquidity differences, evidence indicates a green bond premium of -2 bps. The green bond premium varies between different business sectors and is more prominent for financial institutions (ibid). Zerbib (2019) also emphasize the low bearing of investors’ pro-environmental preferences on bond prices, which does not signify a disincentive for investors to care for the growth of the green bond market.

Similar to Zerbib (2019), Preclaw and Bakshi (2015) and Bos et al. (2018) support a green bond premium. Nanayakkara and Colombage (2019) find that green bonds trade at a premium by providing to investors a lower risk investment, through diversification. Accordingly, the lower yields and the increasing supply of green bonds decreases the demand from investors, unless, investors derive utility from non-pecuniary factors, counterbalancing for the lower yields (ibid.).

(19)

conventional bonds. However, between 2015 and 2016, they find evidence indicating the opposite. This is explained by the increased demand for green bonds and an improvement in the credit ratings (Karpf & Mandel, 2018). Moreover, failing to find any significant yield differences, Hachenberg and Schiereck (2018) use a global dataset of 63 green bonds issued between 2015 and 2016 and finds indications that green bonds trade at 1.18 bps lower yields than their non-green counterparts.

In terms of the demand and supply dynamics of the green bond market, Wulandari et al. (2018) argue that the green bond premium is a consequence of the lack of supply, driving up bond prices. The shortage of green bonds’ supply and the low issuance volumes makes it possible for issuers to offer green bonds at lower interest rates, relative to conventional bonds. The thin market that follows causes liquidity to become relevant (Della Cruse et al., 2011). Hence, the excess demand and shortage of supply in the green bond market may trigger a liquidity premium, rather than a green bond premium (Wulandari et al., 2018). Surprisingly, Wulandari et al. (2018) find that green bonds are, on average, more liquid compared to conventional bonds, but that liquidity risk on yield spread has become negligible in recent years. Also, Wulandari et al. (2018) find that higher green bond liquidity causes the yield spread between green bonds and conventional bonds to increase. This would leave green bonds as a less attractive investment compared to conventional bonds, since the liquidity of the bond and makes the investment unattractive to the investor and costly for the issuer (Wulandari et al., 2018; Bachelet et al., 2019). However, the increasing maturity of the green bond market and the substantial upsurge in green bond issuance would justify a reduction in premium over time as the impact of liquidity risk decreases (Wulandari et al., 2018).

Comparable to a liquidity premium, the green bond premium reduces with an improvement of the credit quality (Zerbib, 2019). Interestingly, when segmenting green and conventional bonds in terms of credit rating, the green bond premium is more evident for riskier borrowers (Ehlers & Packer, 2017). In contrast, Hachenberg and Schiereck (2018), fails to provide significant evidence that differences in pricing are related to different credit ratings. In line with the credit-spread puzzle, Huang and Huang (2012) find that credit risk accounts for less than 25% of the yield spread measured as corporate bond yield over treasury bond yield. However, non-related to credit rating, Cochu et al. (2016) argue that due to the intention of green bonds to fund new and less mature innovations, green bonds may be perceived as riskier compared to conventional bonds.

Furthermore, in general, issuers of green bonds are typically highly rated, with only a modest portion below investment grade (Ehlers & Packer, 2017; Zerbib, 2019). Also, Larcker and Watts (2020) find that green bonds have higher credit quality. This indicates that a bond with better environmental performance is associated with lower credit risk, thus increasing its valuation.

(20)

Table 1. Summary of key studies on green bonds

Zerbib (2019) Hachenberg &

Schiereck (2018) Baker et al. (2018) Nanayakkara & Colombage (2019)

GBP Aligned Yes Yes No Not Disclosed

Scope Global Global US Corporate and

Municipal bonds with Bloomberg green flag.

Global

Primary/Secondary Market

Secondary Secondary Primary Primary

Time Period Jul. 2013- Dec. 2017 Oct. 2015 - Mar.

2016 2010-2016 2016-2017

Number of Bonds 101 63 2083 82

Method Matching + Panel

Regressions Matching + Panel Regression +

non-parametric tests

Pooled OLS Panel Regression +

Option-Adjusted-S preads

Liquidity Control Bid-Ask Spread Issue amount Issue Amount Time Since

Issuance

Strict Maturity Control

Yes Yes Yes Yes

Yield Premium -2 bps Not significantly

different from zero. -7 bps -63 bps

Karpf & Mandel (2018)

Ehlers & Packer (2017)

Gianfrate & Peri (2019)

GBP Aligned No Yes Yes

Scope U.S. Municipal bonds

with Bloomberg green flag.

Euro and U.S. Euro market and

Euro-denominated Bonds

Primary/Secondary

Market Secondary Primary Primary & Secondary

Time Period 2010-2016 2014-2017 2013-2017

Number of Bonds 1880 21 121

Method Oaxaca-Blinder

Composition

Comparison Propensity score

matching approach

Liquidity Control No. of transactions within the past 30 days.

No No

Strict Maturity Control

Yes Yes Not disclosed

Yield Premium +7.8 bps -18 bps -20 (-5) bps in primary

(21)

2.5. Hypothesis formulation

A careful review of the literature on green bonds has led us to the conclusion that existing research is ambiguous in terms of a green bond premium on the European secondary bond market. A premium on the secondary bond market is a concern for investors only. The green bond premium represents the additional yield that is demanded by investors to invest in green bonds (Zerbib, 2019). If the green bond premium is negative, this demonstrates the presence of a green bond preference, and investors are willing to receive lower yield from investing in green bonds compared to non-green bonds. A negative green bond premium could be evident if non-pecuniary or pecuniary motives drive investors. On the other hand, if the green bond premium is positive, no green preference exists, and investors demand a higher yield to hold green bonds.

Understanding the arguments of investors’ utility functions is essential, given the increased demand for sustainability factors for investors (Larcker & Watts, 2020). Following the traditional asset pricing models, risk and return are the critical factors for investment decisions (Sharpe, 1964; Lintner, 1964). However, investors with non-pecuniary motives care about other factors than just financial returns, and thereby derive utility by investing in assets aligned with their tastes (Fama and French, 2007). Also, Fama and French (2007) and Zerbib (2020), argues that a preference or taste for sustainability modifies equilibrium prices. From the utility perspective, the idea that investors value securities beyond the expected risk-return attributes is not novel (Baker et al., 2018; Larcker & Watts, 2020). However, to empirically isolate this effect has proven challenging.

(22)

demand and justifying a pricing difference between green and non-green European bonds. Therefore we test for the following hypothesis:

Hypothesis 1: Green bonds trade at a premium on the European secondary bond market compared to conventional bonds.

Researchers have identified possible determinants of the green bond premium. One examined factor is credit ratings and its effect on the green bond premium. Zerbib (2019) finds that the credit rating significantly affects the negative green bond premium, in that it is more pronounced for bonds with lower credit ratings. On the opposite, it suggests that green bonds with good credit quality experiences less green bond premium. Furthermore, Karpf and Mandel (2018) also support that the green bond premium is correlated with credit quality. They find that an increase in credit quality decrease the green bond premium, and even can turn it positive, which means that green bonds on average trade tighter than similar conventional bonds with identical credit rating (ibid.). Also, Ehlers and Packer (2017) show that the green bond premium is more evident for riskier borrowers. Hachenberg and Schiereck (2018) and Gianfrate and Peri (2019) also attempt to find support for differences in green bond premium based on different credit ratings, although their findings are not statistically significant. Based on this, we formulate the following hypothesis:

Hypothesis 2: Credit rating affects the green bond premium on the European secondary bond market.

In terms of differences between various business sectors, Zerbib (2019) finds that green bonds in the utility sector carried a higher green bond premium compared to green bonds within the financial industry. Also, Zerbib (2019) finds evidence supporting that, given the same credit rating, financial bonds and government-related bonds differ in the level of green bond premia. Hachenberg and Schiereck (2018) find that government green bonds trade wider than financial and corporate green bonds. Furthermore, Bachelet et al. (2019) argue that industries differ in terms of their liquidity, which, according to the liquidity literature (e.g. Chen et al., 2007) impact a bond’s yield. Bachelet et al. (2019) further concluded that the green bond premium is driven by informational asymmetries and is more evident in industries associated with higher credit risk. Thus, we predict that the green bond preference is dependent on what sector the bond is associated with, and the underlying credit risk associated with that sector. Based on this, we formulate the following hypothesis:

Hypothesis 3: The green bond premium differs across business sectors on the European secondary bond market.

(23)

3. Data and Methodology

3.1. Methodological approach

The primary empirical methodology used in the sustainable investments literature to evaluate bond spreads concerns structural models, that is suitable regressions on appropriate model specifications (e.g. Chen et al., 2007; Zerbib, 2019). Related to the bond market, and since bonds have different characteristics explaining its yield, an analysis of matched pairs is applied to circumvent the problem of heterogeneity among bonds (Hachenberg & Schiereck, 2018; Zerbib, 2019). Similar to Zerbib (2019), each green bond is matched with the two closest possible conventional bonds, given a set of restrictions presented in section 3.2.2. The properties defining bond yield must be identical, except for the one property whose effect we are interested in evaluating, i.e. the green label of the bond. The bond yield spreads are calculated by aligning a green and a non-green yield. This approach is beneficial due to its straightforwardness but is best used in combination with regression analysis, as illustrated in this study.

The method in this paper aims towards restricting liquidity bias and dependence on exogenous variables by creating a database of matched bonds to evaluate yield spreads. In addition to liquidity bias, any maturity bias is eliminated by creating a synthetical bond, from two conventional bonds, matching the maturity of the green bond (Zerbib, 2019). The purpose of the regressions is to recognize a green bond premium and identify possible determinants of the green bond premium (i.e. to test hypothesis 2 and hypothesis 3). The comparison of the yield of a green bond and that of a synthetic conventional bond enables the isolation of the pro-environmental characteristic on bond prices (Zerbib, 2019). Inspired by the work of Baker et al. (2018), Wulandari et al. (2018) and Zerbib (2019), this thesis employs a panel data methodology. The variable construction is closely related to Helwege et al. (2014) and Zerbib (2019) in terms of controlling for the liquidity effects on bond spreads. The regression analysis involves two steps, where the first step captures the existence of the green bond premium, whereas the second step aims to analyze possible determinants of the green bond premium.

3.1.1. Panel Data Regression

Panel data analysis has established a solid ground in the financial literature to examine the determinants of price differentials between green bonds and conventional bonds (e.g. Baker et al., 2018; Hachenberg & Schiereck, 2018; Nanayakkara & Colombage, 2019; Zerbib, 2019). A panel regression allows for the analysis of the same cross-sectional unit over time (Gujarati & Porter, 2009).

(24)

Panel data account for such heterogeneity by allowing for bond-specific variables (ibid.). Through a panel regression on matched bonds where the bonds’ attributes are equal except for the greenness of the bond, this study avoids two biases in a cross-sectional methodology on yield spreads; an omitted variable bias, since all price determinants are identical for both bonds, and a bias to assign greater weights to securities with a longer price data (Zerbib, 2019).

This research design allows for intertemporal stability in the regressors and control variables. Practically, this implies that variables representing determinants for bond pricing, e.g. sector classification and credit rating, have a low year-to-year variation within each bond. The fixed-effect model compares bond with themselves across the sample period by generating a within-issuer estimator (Gujarati & Porter, 2009). Therefore, the fixed-effects regression allows us to isolate the bond-specific time-invariant unobserved effect without using information about other bonds (ibid.). This is in contrast to the OLS regression that compares each bond with the overall sample (ibid.).

Furthermore, as different bonds are issued on different dates, the panel is unbalanced. Lastly, regressions with robust standard errors are carried out to address the concern of serial correlation and heteroskedasticity. By using robust standard errors, dependency across observations and heteroscedasticity are both controlled for (ibid.).

3.2. Data description

3.2.1. Green bonds

The first step in the data collection process considers constructing a green bond database containing bond-specific information. Our data are drawn from two data sources, the Bloomberg Fixed Income Database and Thomson Reuters Datastream. The reason we use two different data sources is that we want to use Bloomberg’s definition of green bonds, but we have limited access to the Bloomberg terminal. Therefore the remaining data is collected from Thomson Reuters Datastream. All green bonds are drawn from the Bloomberg terminal, to ensure consistency, where a green bond needs to be aligned with the Green Bond Principles (GBP) in order to be classified as green.

(25)

Additionally, only a small proportion of green bonds are issued before 2015 and thus not representative for the liquid and diverse green bond market that has emerged in recent years. Further, we use the country of domicile to determine if a bond is a European bond or not, similar to Gianfrate and Peri (2019). However, in contrast to Gianfrate and Peri (2019), we put no restrictions on the currencies to be allowed in the sample, to enable an increase in the sample size. Moreover, these bonds are only issued on the domestic market or the Eurobond market to ensure compliance with the European classification.

This dataset incorporates a variety of different types of bonds, such as corporate, supranational, municipal, government and covered bonds. These bonds also contain different seniority, different credit ratings, as well as different coupon types and maturity features. In line with Wulandari et al. (2018) and Zerbib (2019), the sample is restricted to straight and fixed-rate coupon bonds to simplify and avoid bias in the yield calculations. Also, bonds with maturity features such as callability and putability or are perpetual are excluded. These additional restrictions leave us with 360 European green bonds to examine.

For each green bond, data is collected concerning issuer, issue date, issue year, maturity date, coupon, coupon type, country of domicile, currency, amount issued, seniority, bond ratings (initial ratings provided by either Fitch, Moody's, S&P) and sector. The issue amount in the local currency is recalculated into USD for all bonds, using the FX rates corresponding to the issue date of each bond, in order to make bonds comparable with each other. The choice of currency for the issue amount is irrelevant to the empirical analysis, but it ensures that the issue amounts of the bonds are converted into one uniform scale. Using USD as the currency for the conversion is in accordance with other studies examining more than one currency (e.g. Ehlers and Packer, 2017; Hachenberg et al., 2018; Bachelet et al., 2019; Zerbib, 2019). The sector classification is following the Thomson Reuters Business Classification (TRBC), which is an industry classification of global companies (Refinitiv, 2020). Bonds that do not have any initial rating from either Fitch, Moody's or Standard & Poor's are removed in order to control for credit risk.

(26)

Adjusting according to these requirements leaves us with a dataset of 186 green bonds. As is shown in section 3.3.1., the matching procedure is successful for 88 of the 186 green bonds in the sample. Therefore the final sample used in the regression is 88 green bonds. In Table 2 below, we summarize the sample selection criteria and the number of green bonds lost for each step.

Table 2. The green bond sample

Initial sample of European green bonds issued between 2014-12-31 and 2018-12-31

461

Less non-straight and non-fixed-rate coupon bonds

-101 360

Less expired bonds and bonds with missing data

-174 186

Successful matches in the matching

procedure* (see 3.3.1.) -98 88

Final sample 88 green bonds

*A note about the matching procedure. (See section 3.3.1 for more details). In our sample of 186 green bonds, 88 green bonds lack conventional bonds to be compared with, and an additional 10 green bonds have incomplete pricing data for the time period used between 2015 and 2019.

3.2.2. Conventional bonds

In order to evaluate the yield difference between green bonds and conventional bonds, we need to construct a database of conventional bonds. Therefore, the next step in the data collection process is to find matching conventional bonds for each green bond in the dataset. We search for active conventional bonds from the same issuer as each green bond, having identical characteristics except for liquidity. Thus, they all have the same coupon type, issue amount, issue date, sector, currency, maturity, collateral, credit rating, the market of issue and seniority. This is in similarity to Hachenberg and Schiereck (2018) and Zerbib (2019). Ideally, a green bond should be matched with a conventional bond with same issue date and maturity date. However, due to the difficulty of identifying such bonds, we limit the search by only allowing conventional bonds to be issued within six years of the issue date of the green bond. The maturity date of conventional bonds should be within two years before or after the maturity date of the green bond.

Furthermore, since liquidity is an essential factor of bond pricing (Chen et al., 2007; Hachenberg & Schiereck, 2018; Zerbib, 2019) the issue amount of each conventional bond is limited to either four times the green bond or one quarter than that of the green bond (Zerbib, 2019). These limitations on issue date and issue amount will indirectly control for liquidity bias when estimating the green bond premium (ibid.). Also, in a similar manner as for green bonds, price data are collected for conventional bonds as described in section 3.2.1.

(27)

Table 3. Summary of matching criterias

Characteristics Matching criteria

Issuer Exact match

Issue Date +/- 6 years

Maturity Date +/- 2 years

Credit Rating Exact match

Sector Exact match

Currency Exact match

Market of Issue Exact match

Seniority Exact match

Collateral Exact match

Issue Amount 25% to 400% of the green bond.

3.3. Matching Procedure

3.3.1. Matching green bonds with conventional bonds

A crucial concern when quantifying the green bond premium or any other sustainability metric is not being able to isolate the effect from other factors affecting bond yields (Baker et al., 2018; Zerbib, 2019). Failing to control for bond price determinants makes the presence of a green bond premium challenging to detect (Zerbib, 2019). A matching method is thus applied to circumvent this problem, based on the specifications outlined in section 3.2.2. Following this method, each green bond is matched with two similar conventional bonds. In doing so, it is possible to examine the difference in the yield of these bonds (ibid.). Researchers have used the matching method in various applications, for example, to assess the cost of liquidity (Helwege et al., 2014) and to compare the returns of ethical funds with conventional funds (Renneboog et al., 2008).

In the matching procedure, a bond is only allowed to be used once (Helwege et al., 2014; Zerbib, 2019). If a conventional bond fulfils the matching criteria, the bonds are added to the sample of bond triplets and excluded from the set of bonds available for the next draw (Helwege et al., 2014). If no matched conventional bond is identified, the green bond is excluded from the sample.

(28)

synthetic bond is constructed with the same maturity as the green bond whose yield is linearly interpolated or extrapolated at the green bond’s maturity date.

In practical terms, for each triplet of bonds, the daily yield of the synthetical conventional bond is determined by the linear relationship between the two conventional bonds at the time of maturity of the green bond. By letting ​a ​be the intercept and ​b be the slope parameter of the linear function passing through (MaturityCB1​, y

CB1​) and (Maturity​CB2​,

yCB2​), the yield of the synthetic conventional bond is derived through the following equation:

a Maturity y

︿

i, t

SB = + GBi× b (1)

with y︿i, tSB reflecting the daily yield of the synthetic bond and MaturityGB representing the maturity, measured in years, for the green bond. Letting yi, tGB and y︿ be the daily yield of

i, t

SB

the green bond and the synthetic bond, respectively, ​defined as the yield over appropriate risk-free interest rate, in time t, the daily yield spread is measured as:

y y

Δ︿i, t = GBi, t − y︿ i, t

SB (2)

This process is repeated for all green bonds in the sample. Following this matching procedure, the initial dataset of 186 green bonds is reduced by 98 green bonds. This can be explained by the high requirements for the matching process and the presence of nisch issuers of green bonds in our sample. First, 88 green bonds were excluded due to lacking comparable conventional bonds to be matched with; then another 10 green bonds were lost due to missing price data on either green bonds or conventional bonds. Therefore, we end with a final sample of 88 green bonds matched with corresponding synthetical bonds from 51 issuers and 46,205 daily observations of y .Δ︿i, t

3.3.2. The estimation of the green bond premium

The initial step of the empirical methodology entails the isolation of the green bond premium (hypothesis 1). Since the matched bond pairs are set up to be as identical as possible, the assumption is that the only difference, besides the green factor, is liquidity (Zerbib, 2019). Therefore, it is still necessary to control for the residual difference in liquidity between the conventional bonds and the green bond, as liquidity has a profound impact on the pricing of a bond (Chen et al., 2007). Thus, following Zerbib (2019), a regression model is constructed where differences in daily yield spreads are regressed against the daily residual difference in liquidity between the green and conventional bonds.

(29)

, with being the error term

y Greenium ΔLiquidity ε

Δ︿i, t = i+ β1 i, t + i, t εi, t (3)

with yΔ︿i, t being the daily yield spread defined in eq. (2). The constant term, i.e. ​Greenium​i in eq. (3), captures the average green bond premium in the secondary market, and if negative illustrates the average investor’s willingness to receive a lower yield on the green bond. The second term (ΔLiquidity )i, t in eq. (3) captures the residual difference in liquidity, defined as:

Liquidity iquidity iquidity

Δ i, t = L i, tGB− L

i, t

SB (4)

The selection of liquidity estimators is restricted due to the infrequent trading of bonds (Helwege et al., 2014). In this study, bond-related liquidity effects are estimated by calculating the bid-ask spread, similar to Helwege et al. (2014) and Zerbib (2019). Other liquidity measures such as the Amihud measure (Amihud, 2002) need intraday trading volumes which are not accessible for some green bonds. Additionally, liquidity controls could also involve the issue amount. The logic behind the size is that a larger issue amount causes the investor base to increase, and thus lowering the search costs (Zerbib, 2019). However, any variable that is constant over time is not a suitable liquidity estimator (ibid.). Thus, the issue amount is considered in the form of a control variable. Instead, as the bond data is low-frequency data, the percent quoted bid-ask spread act as the proxy for liquidity. Various researchers show that this is an unbiased low-frequency liquidity estimate (Chen et al., 2007; Fong et al., 2017). Thus, the liquidity estimator is expressed as follows : 4

Liquidity ΔBA BA A

Δ i, t = i, t = GBi, t − B SBi, t (5)

Similar to the yield spread of the synthetic bond, the bid-ask spread of the synthetic bond is a distance-weighted average of the bid-ask spread based on the conventional bonds’ maturity in relation to the maturity of the green bond:

A BA BA B SBi, t = d1 d + d1 2 i, t CB1 + d2 d + d1 2 i, t CB2

where d1 = ||MaturityGB− MaturityCB1||and d2 = ||MaturityGB− MaturityCB2||.

(6)

3.3.3. Determinants of the green premium

After examining the existence of the green bond premium in the first regression, the aim is now to analyze possible determinants of the variation in the green bond premium. Through an OLS-regression, the model specification carefully considers the characteristics through which

4 ​For each day a bond’s bid-ask spread is calculated as the ask quote (​P-ask​) minus the bid quote (​P-bid​) divided by the average of the bid quote (​P-bid) ​and ask quote

(30)

the bonds differ. To answer hypothesis 2 and hypothesis 3, we construct three models where the variables under consideration are credit rating and industry sector.

It is argued by most researchers (e.g. Hachenberg et al., 2018; Nanayakkara & Colombage, 2019; Zerbib, 2019) that the green bond premium varies with bond risk. Therefore, bond pricing characteristics are considered in the model alongside the testable variables to control for additional variation in the dependent variable. The control variables are​Currency, Issue amount, Maturity, Seniority​. ​Currency is a dummy variable, taking on the value one if the bond issue is USD or EUR denominated and zero otherwise, in similarity with Hachenberg and Schiereck (2018). The expectation is that investors regard green bonds issued in USD or EUR as more favourable since these currencies are more liquid and have less risk of suffering from currency devaluation than smaller currencies (Nanayakkara & Colombage, 2019).The​Issue Amount in USD is logarithmized due to the presence of outliers and skewness, in order to decrease heteroskedasticity, in accordance with Zerbib (2019). Maturity is calculated as the number of days remaining of the bond as of 2019-12-31, in line with Hachenberg and Schiereck (2018) and Zerbib (2019). ​Seniority refers to the order of repayment to bondholders in the event of bankruptcy and is coded from 1 to 4, where 1 indicates the highest priority and 4 the lowest. Firm-specific variables such as market capitalization, interest rate coverage ratio, leverage are not considered since both the green and conventional bonds are issued by the same issuer.

To test hypothesis 2, and the possible impact of the credit rating on the green bond premium, a dummy variable capturing credit rating ( ​Credit Rating​) is included in the model specification. All the bonds in the sample have a credit rating from the rating agencies Moody’s, Fitch or S&P. In order to make these comparable, we compute a standardized rating scale in accordance with Norden and Weber (2004), Friewald et al. (2012), and Hachenberg and Schiereck (2018). The credit rating scales are standardized by assigning integer numbers to their respective credit rating scale, based on a scale of 1-17, where 1 represents the highest credit rating from each rating agency and 17 the lowest (see Table A3 for details). The ratings from the three rating agencies are then converted into one common rating scale representing the highest rating assigned by the three agencies (Norden & Weber, 2004; Friewald et al., 2012; Hachenberg & Schiereck, 2018). That is, for each bond (green or conventional), we assign one credit rating that is based on the rating provided by the three rating agencies. In order to calculate the common credit rating scale, a minimum of one credit rating is required. This leads us to the following model specification (a):

CreditRating Currency Greenium

︿

i= α0+ Σn−1i=1α1 i + α2 i +

IssueAmount Maturity Seniority

α3 i+ α4 i+ α5 i+ εi (7)

(31)

Industrials​. However, since we only have two bonds in the category ​Governments​, they are included in the group ​Supranational​, which is equivalent to government-related bonds, to ensure sample representativeness. This is in-line with Hachenberg and Schiereck (2018) who categorize all supranational organizations as government-related bonds. Furthermore, in this model specification (b), ​Supranationals is the reference group. The ​Sector variable is constructed as a dummy variable to test for sector fixed effects. This leads us to the following model specification (b):

Sector Currency Greenium

︿

i= α0+ Σi=1n−1α1 i + α2 i +

IssueAmount Maturity Seniority

α3 i+ α4 i+ α5 i+ εi (8)

In addition, we will test the full model specification (c) in which we include both Sector and ​Credit Ratings together with the control variables for robustness control. Model specification (c) is given below:

CreditRating Sector Currency

Greenium

︿

i= α0+ Σn−1i=1α1 i+ Σn−1i=1α2 i + α3 i +

IssueAmount Maturity Seniority

α4 i + α5 i+ α6 i+ εi (9)

(32)

Table 4. Variable summary

Variable Type Units Description

Yield difference in bps Δy )( ︿i, t

Quantitative Continuous Calculated as the difference in ask yield between a green bond and the synthetic bond. The ask yield is calculated as the spread between a bond’s ask yield and risk-free rate. Data source: Thomson Reuters Datastream.

Greenium Quantitative Continuous Greenium is the sum of the constant and the error term when performing the first regression eq (3):

,

y Greenium ΔLiquidity ε

Δ︿i, t= i+ β1 i, t + i, t

Liquidity

Δ Quantitative Continuous Difference in bid-ask spread between a green bond and the synthetic

bond.

BA Quantitative Continuous The ask price (​P-ask​) minus the bid price (​P-bid​) divided by the

average (spread) of both prices. Data from Thomson Reuters Datastream.

P-ask Quantitative Continuous Clean bond ask quote, from Thomson Reuters Datastream.

P-bid Quantitative Continuous Clean bond bid quote, from Thomson Reuters Datastream.

Credit Rating Quantitative Dummy The bond’s credit rating, converted into a standardized credit rating scale by assigning integer numbers to different credit ratings, based on a scale of 1-17. Data from Thomson Reuters Eikon & Bloomberg.

Sector Qualitative Dummy Thomson Reuters Business Classification with minor adjustments. Collected from Thomson Reuters Datastream.

Currency Qualitative Dummy The currency of the bond issue. In our sample we have EUR, USD, NOK, SEK, DKK, GBP, CAD and CHF. The variable currency is a dummy variable, taking on the value one if the bond issue is in EUR or USD and zero otherwise. Data collected from Thomson Reuters Eikon & Bloomberg.

Maturity Quantitative Continuous The maturity of the bond on December 31, 2019, measured in years.

IssueAmount Quantitative Continuous The issue amount converted to USD using the FX-rate as of issue date for each bond. FX-rates collected from Thomson Reuters Datastream. This variable is also logarithmized, due to skewness.

References

Related documents

The theory is therefore part of an interpretive turn in organization studies, where effort is dedicated not to increasing the efficiency or effectiveness of

46 Konkreta exempel skulle kunna vara främjandeinsatser för affärsänglar/affärsängelnätverk, skapa arenor där aktörer från utbuds- och efterfrågesidan kan mötas eller

In the corporate bond sample, the yield spread and difference in liquidity are different from zero and negative with a 95% confidence interval indicating that the bonds in

Fredrik Svinhufvud, Chairman, Vindkraft Ukraina Discussant: Chloé Le Coq, Assistant Professor, SITE Date: Thursday, November 21, 9.15 -11.45. Place: Stockholm School of

Industrial Emissions Directive, supplemented by horizontal legislation (e.g., Framework Directives on Waste and Water, Emissions Trading System, etc) and guidance on operating

Although this is a preliminary study but the identified determinants that complementing green thinking into lean are valuable in the sense of manufacturers could focus

European SMEs indicates to have a higher degree of impact on the relationship between social and environmental dimension of CSR and financial performance, proved by

This will be examined through two case studies of Uganda and Mozambique on the local impacts of carbon forestry on employment and income, access to land, and food security..