A comparative study on the financial

performance of Green bonds and their

conventional peers

Louis William Wagner Ley∗

Master Thesis in Financial Economics

Erasmus University Rotterdam

Erasmus School of Economics

Dr. Jan Lemmen

November, 2017

Keywords: Asset Pricing, Financial Economics, Fixed Income,

Green Bonds, Financial Performance, Bond Pricing
JEL Classification: G12, G17, Q54

∗
Louis William Wagner Ley Germain Joseph Fernand (449910), Ernastrasse 20, CH-8004
Zurich, Switzerland. Phone: +352 621 50 68 91. Email: [email protected]
Acknowledgments

The author wants to thank Annemarie Arens, General Manager of the Lux-
embourg Finance Labelling Agency, and Maurice Bauer, Secretary General of
the Luxembourg Stock Exchange, for their very helpful insights, opinions and
the provision of still unpublished data and reports.

I
Non-plagiarism Statement

By submitting this thesis the author declares to have written this thesis com-
pletely by himself/herself, and not to have used sources or resources other than
the ones mentioned. All sources used, quotes and citations that were literally
taken from publications, or that were in close accordance with the meaning of
those publications, are indicated as such.

Copyright Statement
The author has copyright of this thesis, but also acknowledges the intellectual
copyright of contributions made by the thesis supervisor, which may include
important research ideas and data. Author and thesis supervisor will have
made clear agreements about issues such as confidentiality.
Electronic versions of the thesis are in principle available for inclusion in any
EUR thesis database and repository, such as the Master Thesis Repository of
the Erasmus University Rotterdam.

II
Abstract

The market of Green Bonds has seen exponential growth over the recent years
and was fuelled even more by the COP 21 agreement. However, until today
and to the best knowledge of the author, no empirical analysis on the financial
return of those instruments has been conducted. This investigation conducts
the first comparative analysis of the financial performance of Green Bonds
and their conventional peers. Based on a dataset of 359 Green Bonds and
1291 conventional bonds, the analysis is conducted over the period between
2011 and 2017 and uses an extended Fama-French model in a Fama-Macbeth
regression procedure. Green Bonds do outperform conventional ones over the
full sample period but with a low significance. In a subsample period aligned
to the “take-off” of the corporate Green Bond issuance, the outperformance
can still be confirmed but this time with a high significance. We can observe
that the significance is constantly increasing over time. This can be seen as
an important supporting argument for the investment in Green Bonds and the
fight against climate change. At the same time it implies that institutional in-
vestors are not acting against their fiduciary duties due to investment in those
sustainable debt capital products.

III
Contents
1 Introduction 1

2 Literature 6

3 Methodology 12
3.1 The Capital Asset Pricing Model . . . . . . . . . . . . . . . . . 12
3.2 An extended Capital Asset Pricing Model . . . . . . . . . . . . 12
3.3 The Fama-Macbeth Procedure . . . . . . . . . . . . . . . . . . . 15

4 Data 18
4.1 Bond Dataset . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
4.2 Green Bonds . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
4.3 Conventional Bonds . . . . . . . . . . . . . . . . . . . . . . . . . 21
4.4 Risk-free Rate . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
4.5 Fama-French Factors . . . . . . . . . . . . . . . . . . . . . . . . 22
4.6 Momentum Factor . . . . . . . . . . . . . . . . . . . . . . . . . 23
4.7 Bond Factors . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

5 Empirical Research 25
5.1 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
5.2 CAPM Regression Results . . . . . . . . . . . . . . . . . . . . . 27
5.3 Fama-French Three-Factor Model Regression Results . . . . . . 29
5.4 Carhart Four-Factor Model Regression Results . . . . . . . . . . 31
5.5 Seven-Factor Model Regression Results . . . . . . . . . . . . . . 33

6 Summary and Conclusions 37

7 Limitations and further research 40

Appendix A: Appendix 44
1.1 Seven-factor cross-sectional regression results over time (bi-annual) 44
1 Introduction

“The gentleman understands moral duty, the petty person knows about profit”
(Ebrey, 2010)

After the recent global financial crisis, the restructuring of national and
international financial regulations as well as the still increasing complexity
and interconnectedness of financial markets, this 2500 year old Confucian say-
ing is more accurate than ever before.

For a long time, the consideration of ethical values and principles has been
considered a threat to profits in the corporate world. However, these attitudes
changed recently and stakeholders ranging from consumers to investors are not
only encouraging but requiring higher levels of transparency and compliance
with ethical standards (Glomsrød and Taoyuan, 2016). Environmental, Social
and Governance criteria (ESG) are gaining more and more importance when
it comes to corporate decision-making and strategy-setting.

The financial sector, heavily criticised for its unethical behaviour in the
wake of the last financial crisis, is increasingly introducing ESG and other sus-
tainability criteria in their investment decision processes. Today, ethics “sells”
and unethical behaviour is punished by corporate image losses and shareholder
activism.
For a long time, climate change has not been a priority on the ethical investors’
agenda. Lately though, the corporate world is increasingly willing to act on
these matters. Stakeholders around the globe are requiring sustainable and
responsible business models(EY, 2016).
Moving towards a low- or even carbon-neutral world requires the adoption
of green solutions for financial capital. After the United Nations Framework
Convention on Climate Change (UNFCC) and the Kyoto Protocol, the 2015
Paris Agreement is the first treaty that brings together all nations - including
the United States of America - for collective action by an international public
legal agreement that constitutes at the same time a platform for investors to
step up against climate change. The aim is to reorientate all financial flows
away from fossil fuels and towards clean forms of development. The COP21
meeting in Paris, with the parties confirming their commitment to keep the

1
global warming well below 2°C compared to pre-industrial levels and the mo-
bilisation of USD 100 billion per year by 2020 in order to lessen the effects
of climate change in developing countries, is generally considered as a turning
point. The United Nations are also considering the tackling of climate change
as one of their 17 sustainable development goals.

In general, the political support for addressing the issue is gaining momentum
and providing clear signals for future investment patterns (Clapp et al., 2016),
despite the “change in mind” of the U.S. government, the corporate sector sees
the necessity to act on the matter. Therefore, Green Finance in general, is set
to play a crucial role in the implementation of these goals.

One can observe a strong rise in the demand of institutional investors for a
financial tool that is incorporating both, their need for an adequate short-term
portfolio risk and return balance as well as the decrease of risk of long-term
climate change (Climate Bonds Initiative, 2017). Since recently, the term of
“Green Bonds” is circulating more and more frequently in the news as being
the new financial tool that addresses the fight against climate change and does,
at the same time, have the afore-mentioned required characteristics.
The World Bank defines Green Bonds as “fixed income, liquid financial instru-
ments that are used to raise funds dedicated to climate-mitigation, adaptation,
and other environment-friendly projects” (World Bank, 2015).

Green Bonds are therefore considered to fulfill this premise and key in the
implementation and capital raising of the Nationally Determined Contribu-
tions (NDCs) to the 2015 Paris Climate agreement. They provide a suitable
and simultaneously risk-managed tool for both public and private investors.

The history of Green Bonds can be traced back to the issuance of the first
so-called Climate Awareness Bond by the European Investment Bank (EIB)
in 2007, which was quickly followed by the first Green Bond set-up by the
World Bank in 2008 (EY, 2016). The Climate Bond Initiative, an interna-
tional Non-Governmental Organisation with the goal to activate debt-capital
markets for climate finance saw the day in 2009.
While the first Green Bonds were mostly issued by development banks and
municipalities, the proportion of issuance by commercial banks and other cor-
porates is steadily increasing.

In general, any entity that can issue a conventional bond can also set up a

2
Green Bond, which means that even a “brown” company (a company invest-
ing in fossil resources) is able to issue a Green Bond in compliance with the
Green Bond Principles. Also, there is an increasing amount of divestment ini-
tiatives for projects exposed to coal and other fossile energy or non-renewable
resources by large institutions and multinational firms such as the Rockefeller
Brother Fund, the Norwegian Government Pension Fund Global, AXA, Bank
of America or Citigroup (Glomsrød and Taoyuan, 2016).

The year 2013 is generally seen as the tipping point of the market. By 2014,
Green Bonds compromised a third of the total corporate bond issuance, which
was equivalent to about USD 15 billion (NEPC Impact Investing Committee,
2016).
In 2016, the market for Green Bonds hit a new record with a total issuance
of USD 81 billion and an outstanding volume of USD 180 billion (Climate
Bonds Initiative, 2017).
Today, the major part of issuers and purchasers comes from Europe or
North America, while most of the projects financed are located in developing
countries.
In its outlook, New Climate Economy, a research initiative, notes that, to
stay in the limits of the 2 degree target, a USD 93 trillion of investment are
required across the global economy which translates into a huge growth market
for Green Bonds (Climate Bonds Initiative, 2016).

For a long time, the labelling process was largely unregulated and mainly sub-
ject to self-labelling. Even if reputational risks are against it, the absence of
a central regulatory authority is a potential incentive for greenwashing in the
sector.
This lack led to the development of different projects, associations, as well as
regulatory or independent opinion initiatives such as the Green Bond Princi-
ples, the Climate Bonds Initiative or different actions taken at national and
supranational legislative levels (NEPC Impact Investing Committee, 2016).
While the Green Bond Principles, launched by a group of banks and the In-
ternational Capital Markets Association (ICMA), are mainly addressing the
traceability of proceeds in terms of the issuance, disclosure and reporting pro-
cesses, the Climate Bonds Initiative is focusing more on the development of
clear and science-based criteria to define the term “green”.
Similar actions are taken by the People’s Bank of China, The National De-
velopment & Reform Commission and the Securities and Exchange Board of
India or the Luxembourg Stock Exchange with the launch of the world’s first

3
green exchange (Climate Bonds Initiative, 2016). Moreover, audit and con-
sulting firms offering third-party verification services as well as international
labelling agencies for financial instruments such as the Luxembourg Finance
Labelling Agency (LuxFLAG) are enhancing the credibility of the green bond
market (LuxFLAG, 2017).

Despite the in-depth research on the necessity and impacts of green finance
and Green Bonds in special, there is little empirical evidence of the financial
performance of these investments. Many investors are still considering the in-
strument as a “charity investment” and businesses still seem anxious about
returns of so-called green investments (EY, 2016).
Some of them are said to be willing to pay a “greenium”, a premium for being
green, that translates into lower returns but less costly funding for the issuer.
One cause for this premium could be a lack of supply for Green Bonds relative
to their demand.
However, there is little evidence for this greenium in the general market (NEPC
Impact Investing Committee, 2016).

Yet, as for any other financial instrument, the aspect that is of key impor-
tance to investors is the risk-adjusted financial return. An increasing risk
of “carbon bubbles” is challenging the profitability of “brown” investments
(Glomsrød and Taoyuan, 2016) but that does not imply that green invest-
ments are providing a relatively superior performance. “The credit metrics of
green bonds have been on-par with their traditional peers” says Samantha D.
Palm of Parnassus Investments (Parnassus Investments, 2015). Nonetheless,
until today, little academic evidence exists to support this quote.

The present study conducts the first comparative analysis of the financial per-
formance of global Green and conventional bonds. The results shall thereby
contribute to getting a 360 degree view on the financial characteristics as well
as drawing stronger statistical insights on the performance of Green Bonds.
The following investigation shall explain if the fact of a bond being green in-
creases its expected financial performance versus its traditional peers. This is
the central research question that is to be answered by this analysis.

The study is conducted with an extended Fama-French framework. By this

it takes into account different equity and bond factors that are capturing the
major part of the underlying fundamentals in the bond market (Fama and
French, 1993). These factors are the so-called Fama-French factors being a

4
market risk premium, a size and value premium as well as Carhart’s momen-
tum factor and a term structure and default probability proxy. To compare
the Green Bond to its conventional peer the GREEN -dummy variable, which
accounts for the fact of a bond being green in accordance to the Green Bond
Principles is introduced. The research is conducted with a set of 1650 bonds
and over a six and a half year time horizon. It finds that Green Bonds do
actually have superior estimated future returns. The results indicate that the
significance of the GREEN factor does increase in recent years and is more
or less aligned with the expansion of the Green Bond market. The same holds
true for the explanatory power of the used seven-factor asset pricing model.
This pattern is also similar to the ones found by other studies focusing on sus-
tainable investments but other asset classes. To the author’s knowledge this
is the first empirical study to find evidence for a hypothesis that can support
the marketing for and interest in Green Bonds even further.

The remainder of the thesis is structured as follows: the following chapter

describes the existing literature in terms of fixed income pricing as well as the
characteristics of sustainable and responsible investments that this study is
based on. Chapter 3 defines the data used to conduct the analysis; Chapter 4
sets out the methodology used; Chapter 5 presents the results of the empirical
analysis; and Chapter 6 concludes while Chapter 7 provides insights on the
limitations of this investigation and possible future research.

5
2 Literature

In general, the literature on bond pricing and the modelling of their expected
returns is far less extensive that the one focusing on equities. However, both
asset classes do have similar key factors that can be used for estimating their ex-
pected returns. The most commonly known model for calculating expected fu-
ture returns in equity markets is the single-factor Capital Asset Pricing Model
(CAPM) of William Sharpe (Sharpe, 1964) and John Lintner (Lintner, 1965).
Given its poor explanatory power, Fama and French (1992) constructed an
extended three-factor model that substantially improved the quality of the
model. Fama and French (1993) then extended their own three-factor CAPM
model used to determine excess equity returns to a five factor model that in-
corporates both, stock and bond market risk factors that could be commonly
used to model bond returns.
Besides the initially used market premium, size and value factors (Fama
and French, 1992), the authors developed a factor to proxy the risk in bond
returns arising from unexpected interest rate changes as well as one proxying
the default probability of the relative company or institution. The impor-
tance of the default factor was confirmed by Merton (1973) stating that it
is a major explanatory variable for the pricing of bonds. Gabbi and Sironi
(2005) provided more evidence and spoke up against the relevance of primary
and secondary market liquidity as a relevant factor. The utility of the Fama-
French model in the pricing of bonds was further tested and acknowledged by
Johansson and Lundgren (2012). Carhart (1997) and Grinblatt et al. (1995)
constructed a momentum factor to further increase the power of the model.
Considering this, the present investigation does opt for the inclusion of the
momentum factor into the final pricing model.

The Fama-Macbeth procedure (Fama and Macbeth, 1973) using both, time-
series and cross-sectional regressions, is considered as the reference approach
to testing asset pricing models with panel data and has therefore been adopted
in this analysis.

The market of sustainable and responsible investments (SRI), being still in

development and gaining attention within the general financial markets only
lately, explains the rather poor quantity of research in this area and especially
on the financial performance of these SRI instruments.
While Xiao et al. (2012) did not find a significant relationship between sus-

6
tainability and returns, Gil-Bazo et al. (2010) provided evidence for a better
before- and after-fee performance of US SRI fund in comparison to non-SRI
funds.
Besides these, one of the only studies comparing the financial performance
of green and conventional investments was conducted by Ibikunle and Steffen
(2015).
Despite the fact that this study focused only on equity mutual funds, this the-
sis will adopt a similar approach for evaluating and comparing the performance
of green bonds to their conventional peers. The researchers used a four-factor
extended Capital Asset Pricing Model following Carhart (1997).
For the comparative analysis, the authors included a dummy variable to their
model that is related to the class of the fund (green, conventional or black).
They found no significant difference in the performance of conventional and
green mutual funds and in their subsample covering only the recent years they
can even observe an outperformance of the green funds in comparison to the
traditional ones.
This paper is going to adopt the same method in order to distinguish the per-
formance of green and conventinal bonds.
These formerly found results for the equity market are a strong incentive to
conduct a similar study on the fixed income side and suggest the hypothesis
that green bonds do not differ from their conventional counterparts in terms
of their financial performance.
Still unpublished articles and papers on the same topic that were conducted
by commercial banks and made available to the author of this investigation do
suggest similar results.

The central hypothesis of the paper is that expected returns on green bonds
are not statistically different from those of conventional bonds. The investiga-
tion focuses the yield-to-maturity as return measure for the reasons specified
in the Data Chapter and defines a Green Bond in accordance with the Green
Bonds Principles that are explained in detail in the same chapter.
Being the first study conducting a comparative analysis on the financial per-
formance of those two bond categories, the study might potentially make a
significant contribution to the literature and provide further insights for in-
vestors to the world of sustainable and responsible investing.
The thesis is purely focusing on financial performance and does not take into
account behavioural considerations or other non-financial factors potentially
driving performance.
In addition to this, the thesis relies on the Green bonds quality check by Natixis

7
(2017) still unpublished at the current date of writing and kindly made avail-
able to the author by the Luxembourg Stock Exchange. Figures 1, 2 and 3
provide a more detailed overview on the existing research on asset and more
specifically bond pricing as well as on the financial performance of SRI invest-
ments.

8
9
Figure 1: Literature Overview
10
Figure 2: Literature Overview
11
Figure 3: Literature Overview
3 Methodology
3.1 The Capital Asset Pricing Model

The Capital Asset Pricing Model was developed by William Sharpe and John
Lintner (Fama and French, 2004) and is based on Markowitz’s (1991) model of
portfolio choice. Markowitz assumes investors to be risk-averse and considering
the investment outcome as a probability distribution. Two single parameters
are at the basis of the investor’s portfolio choice: expected value and standard
deviation. The utility function can be thought of in the following form:

U = f (Ew , σw )

where Ew stands for the expected future wealth and σw is the estimated stan-
dard deviation of the possible discrepancy between expected future wealth and
the actual future wealth (Sharpe, 1964).

All mean-variance efficient portfolios are therefore a combination of a risky

tangency portfolio and the risk-free asset (Fama and French, 2004).

The formula for the CAPM used in this research is:

rit = rf t + ßiM ∗ [rM t − rf t ]

∀i

where rf is the risk-free rate and rM t − rf t the weighted excess return of the
market portfolio. ßiM measures the correlation of the stock return with the
excess market portfolio return, rf is the return of a risk-free asset that is not
correlated with the market and therefore also called “zero-beta asset” (Fama
and French, 2004).

3.2 An extended Capital Asset Pricing Model

The econometric methodology used in this paper is based on the Capital Asset
Pricing Model by William Sharpe (Sharpe, 1964) and John Lintner (Lintner,
1965).
However, given the poor explanatory power of the single-factor CAPM, for

12
the equity and fixed income market, this study proceeds with the use of the
three-factor model developed by Fama and French (1993).
As the authors showed, the extended version of their model is also able to
catch the common risk factors for the bond market and thereby a good ap-
proximation of expected returns in the considered market.

The basic Fama-French three-factor model is set up as follows:

rit − rf t = αi,t + ßiM [rM t − rf t ] + ßi,SM B SM Bt + ßi,HM L HM Lt + i,t

where rM,t − rf,t is the excess market return calculated by subtracting the
risk-free rate from a global market portfolio with βi,M being the coefficient
for the market exposure , SM Bt the size premium with βi,SM B measuring the
small firm effect on the bond return and HM Lt the return difference between
a high and low book-to-market portfolio with βi,HM L accounting for the value
premium.

The size and value premium are based on the presumption that, on average,
small firms earn higher returns than their larger counterparts as a compensa-
tion for illiquidity risk. This difference in the returns between a large corporate
and small corporate portfolio is considered as a factor accounting for the size
risk which is the equivalent to a premium based on size that investors require.
The same intuition goes for the value premium that incorporates the difference
in risk between growth and value firms (Petkova, 2011).

For this comparative study, I extend the initial Fama-French three-factor model
with a momentum factor following Carhart (1997). Grinblatt et al. (1995)
found that investments based on momentum strategies performed significantly
better than those omitting this factor. The so-called Carhart four-factor model
is constructed as follows:

rit −rf t = αi,t +ßiM [rM t −rf t ]+ßi,SM B SM Bt +ßi,HM L HM Lt +ßi,M OM M OMt +i,t

with M OMt being defined as the return difference between a portfolio of 12-
months winner and 12-months looser stocks at time t, while ßi,M OM measures
the effect of the momentum strategy on the return.
Fama and French (1993) demonstrated the relevance of their stock risk factors
for the bond market. Furthermore, they extended their model with two spe-

13
cific factors relating to the fixed income market:

rit − rf t = αi,t + ßiM [rM t − rf t ] + ßi,SM B SM Bt + ßi,HM L HM Lt +

ßi,T ERM T ERMt + ßi,DEF DEFt + i,t

Unexpected interest rate changes are captured by the T ERM factor. The
term spread proxies for a deviation in the return of long term bonds from the
expected changes coming from a shift in short-term interest rates (Fama and
French, 1993).

The DEF factor captures the default spread. It thereby accounts for the
probability of default of a corporate firm. This spread between long-term gov-
ernment bonds and a portfolio of long-term corporate bonds is found to have
a high explanatory power for pricing the default premium (Fama and French,
1993).

The six-factor model used for the research of this paper then takes the form of:

rit − rf t =
αi,t + ßiM [rM t − rf t ]
+ßi,SM B SM Bt + ßi,HM L HM Lt
+ßi,M OM M OMt
+ßi,T ERM + ßi,DEF
+i,t

where T ERMi,t is the term spread of bond i at time t and DEFi,t the default
spread of bond i at time t. ßi,T ERM is measuring the term-structure impact on
the bond return while accounting for the effect of the default probability.

In order to compare the performance of green and conventional bonds with

each other, I make use of a dummy variable that is controlling for the effect of
the respective bond’s classification and determines its relative performance to
it’s counterpart. With this approach, the analysis follows Ibikunle and Steffen
(2015).

The final model that is used in the second-pass of the Fama-Macbeth pro-
cedure is then constructed as follows:

14
 rit − rf t =
αi,t + ßiM [rM t − rf t ]
+ßi,SM B SM Bt + ßi,HM L HM Lt
+ßi,M OM M OMt
+ßi,T ERM + ßi,DEF
+δi,GREEN GREENi
+i,t

where δi,GREEN is measuring the effect on the returns of a bond being green.
GREEN takes the value 1 if the bond is green and 0 otherwise.

3.3 The Fama-Macbeth Procedure

For the regression procedure, I follow Fama and French (1992) in applying the
Fama-Macbeth procedure as described in detail in Fama and Macbeth (1973).

The approach is commonly known as “two-pass cross-sectional regression”

(CSR) since it consists of a two-step procedure testing the time-series aver-
age of estimated risk premia in cross-sectional regressions (Cavenaile et al.,
2009).
The procedure firstly conducts a time-series regression to estimate betas
(first-pass regression) before making use of a cross-sectional regression (second
pass) to test the hypothesis derived from the used model (Cochrane, 2009).
The first stage of the Fama-Macbeth approach runs a seperate time-series re-
gression of the excess returns against the different considered risk factors for
every bond in the sample. These regressions are performed using the Ordinary
Least Squares (OLS) regression:

rtei − rtf = ai + ßi ft + it

t = 1, ..., T
∀i

where rtei is the return of asset i at time t, rtf the risk-free rate at time t, ft the
value of the explanatory variable at time t and ßi the coefficient of ft over time.

15
The outcome of the time-series regression are 1880 coefficients, for every bond
i and every factor included in the extended Fama-French model. Those coeffi-
cients are estimated over the considered time period of the analysis. Since the
affiliation of the bonds to the green or conventional class does not vary over
time, I only integrate the GREEN dummy for the second, cross-sectional,
pass of the procedure.

In the second step of their procedure, Fama and Macbeth run cross-sectional
regressions. Following this, I regress the excess returns of the 1880 consid-
ered bonds against the estimates which are the outcomes of the previously
conducted time-series regression. The intuition is that the estimates are now
treated as explanatory variables (Cavenaile et al., 2009). Following Cochrane
(2009), I do not use an intercept in the second pass.
For the cross-sectional regression, we observe “multiple entities” at a single
point in time t (Stock and Watson, 2012).
In this case 359 green and 1521 conventional bonds. The intention is to test
the multiple factor model across the different bonds (Cochrane, 2014).

To estimate the risk premia, I run cross-sectional regressions of:

rtei − rtf = ßi λ̂t + αi

i= 1,2, ..., N
∀t

In accordance with this approach, I estimate 1570 cross-sectional regressions to

receive 1570 lambda estimates for every observed risk factor and every point in
time t. As required by the applied procedure, I then conclude with averaging
λ̂t by:

T
1X
λ̄ = λ̂t
T t=1

with T being the number of time periods observed.

In an all explaining model, the cross-sectional disparities in expected returns

should be explained by the risk loadings with the cross-sectional R2 being close

16
to 1.

The cross-sectional R2 is calculated by:

PN ¯2
η̂
R2 = 1 − PN i=1 i 2
(ȳ − ȳ¯)
i=1 i

with ηi being the regression error of the cross-sectional regression.

The adjusted R2 adjusts for the number of independent variables in the model
and therefore only increases if an additional predictor increases the predictive
power of the model by calculating:

2 (1−R2 )∗(N −1)

Radj. =1− N −p−1
,

with N being the total sample size and p being the number of predictors used
in the model.

17
4 Data

This section outlines the data collection process for the two bond categories
(conventional and Green) as well as the market benchmarks and the Fama-
French factors used for the empirical analysis. Furthermore, the data section
documents the detailed construction of the TERM and DEF factors and pro-
vides in-depth information on the procedures for data quality assurance.

Since most of the Green bond buyers are institutional investors such as pension
funds with a long-term investment approach, this research is based on a hold-
to-maturity approach. It therefore focuses on yield-to-maturity to proxy the
return of conventional and Green bonds in the extended Fama-French frame-
work. By doing so, the paper takes into account the current coupon income,
any gains or losses in capital realised by holding the bond until its maturity
as well as the timing of the cash-flows. As such, this investigation follows
Houweling et al. (2005) that pronounced a “distinct advantage” over the use
of realised returns: While the latter is looking at the outcome of a stochastic
process, the former is a market expectation of the bond’s return to maturity.
The paper thereby also follows the literature on bond liquidity. This measure
assumes however, that all coupon payments can be reinvested at the computed
Yield-to-Maturity.

4.1 Bond Dataset

The initial bond dataset consists of a total of 2480 bonds issued between June
2007 and July 2017, a ten-year period. All data was provided by Bloomberg
and converted to U.S. Dollar to ensure comparability that could otherwise be
negatively influenced by changes in exchange and inflation rates. In order to
assure the quality and robustness of the data, the initial sample is further
cleaned and filtered for possible bias drivers as described later in this section.

18
4.2 Green Bonds

The analysis starts with an initial set of 1062 Green bonds. Since there is still
no universally agreed on definition of a Green bond, the sample of this paper
is compiled using the “Green Bond” tag in the Bloomberg database. This ex-
cludes all so-called Green bonds which are not clearly classified as such by the
issuer or have not been provided with the needed supporting information. In
general, Bloomberg’s working definition is adopted from the afore-mentioned
Green Bonds Principles and reads as follows: “Labelled green bonds are fixed
income instruments for which the proceeds will be applied towards projects or
activities that promote climate change mitigation or adaptation or other en-
vironmental sustainability purposes”(Bloomberg New Energy Finance, 2016).

While the first Green bond was issued as a “climate awareness bond” by the
European Investment Bank in 2007, the real growth in the market started only
in 2011 when supranationals, governments and especially corporates started
issuing a multitude of green bonds. Table I provides an overview of the amount
of green bonds issued categorised into years and industry sectors.
Considering this, and in order to present a significant study, the present
sample is revised to the period between January 4th, 2011 and June 30th,
2017. Moreover, all data is converted to USD data by the provider in order
to eliminate potential currency bias. After applying the different filters and
manipulations, the final sample consists of 359 Green bonds. The exact filters
are showed in the Figure 4 below:

Figure 4. Green Bond Selection

19
 Table I. Green Bond Issuance
Green Bond Issucance by Industry Sector and Year
Year
Industry Group
2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017
Consumer Discretionary 0 0 0 0 0 0 0 4 4 6 6
Energy 0 0 0 0 0 0 0 8 154 16 12
Financials 0 0 0 0 0 0 12 41 40 81 83

20
Government 1 1 4 55 30 19 25 61 99 90 54
Health Care 0 0 0 0 0 0 0 0 0 0 1
Industrials 0 0 0 0 0 1 0 5 9 9 18
Materials 0 0 0 0 0 0 0 0 0 5 4
Technology 0 0 0 0 0 0 0 1 0 2 1
Utilities 0 0 0 0 0 0 1 13 15 32 39
Total 1 1 4 55 30 20 38 133 321 241 218
4.3 Conventional Bonds

The conventional bonds sample is selected from the global bond universe pro-
vided by the Bloomberg database. The sample is then filtered for the time
period between 1st July, 2007 to 30th June, 2017. To be coherent with the
Green bonds sample, the paper only includes bonds whose use of proceeds are
labelled with the “investment” or “project finance” tag by the data provider.
Also, all non-investment-grade bonds are excluded since for a bond to be com-
pliant with the Green Bond Principles and included in this investigation’s sam-
ple it needs to have an investment-grade ranking. This investigation further
excludes all bonds issued in a currency other than EUR or USD since these two
currencies represent the major part of the global green bond issuance. With
a mean maturity of 4.16 and 4.12 years respectively and a mean duration of
4.12 and 3.55 years the Green and conventional bonds sets are well-suited for
a comparative study as can also be seen in Table II. Following this selection
process, the conventional bonds sample consists of 1521 bonds. By revising
the time period coherently with the procedure adopted for the Green bonds
sample, for the period between January 4th, 2011 and June 30th, 1291 con-
ventionals stay in our final sample that is used for the further research in this
study. The exact filters are showed in the Figure 5 below:

Figure 5. Conventional Bond Selection

21
4.4 Risk-free Rate

This paper follows the the approach of Fama and French (1993) and considers
the one-month T-Bill rate of the United States as risk-free. This is also coher-
ent with the USD data for the bonds sample. The rates are downloaded on a
daily basis from the CRSP data server.

4.5 Fama-French Factors

The market, size and value premium, generally considered as the Fama-French
factors are gathered from the CRSP database.
In their research on the cross-section of expected stock returns, Fama and
French (Fama and French, 1993) find that the ratio of book equity to market
equity (BE/ME) captures most of the cross-section of average stock returns.
Therefore, when rationally priced stocks are presumed, systematic differences
in average returns can be traced back to differences in risk (Fama and French,
1992).

Moreover, the authors find evidence that the book-to-market equity ratio can
substantially improve the model for the prediction of stock returns. They also
find a similar pattern when considering firm size as a further explanatory vari-
able in the model (Fama and French, 1993). The size and value premia are
constructed from the 6 value-weighted portfolios formed on size and book-to-
market ratio which are available on Kenneth French’s personal website (French,
2015). The authors find that those factors do not only have an explanatory
power for the stock market but are common risk factors for stocks and bonds
(Fama and French, 1993).

The market premium is the difference between the market return of the and
the risk-free rate.

The size premium, SMB (Small Minus Big), is the average return on the three
small portfolios minus the average return on the three big portfolios,

22
 SM B =
1/3(SmallV alue + SmallN eutral + SmallGrowth)
−1/3(BigV alue + BigN eutral + BigGrowth)

The value premium, HM L (High Minus Low), is the average return on the
two value portfolios minus the average return on the two growth portfolios,

HM L =
1/2(SmallV alue + BigV alue)
−1/2(SmallGrowth + BigGrowth)
(French, 2015)

In Fama and French (1993) ”Common Risk Factors in the Returns on Stocks
and Bonds”, a complete description of the factor returns is provided (Fama
and French, 1993) (French, 2015).

4.6 Momentum Factor

The momentum factor is provided by the CRSP database in accordance with
Grinblatt et al. (1995) on a daily basis.

4.7 Bond Factors

The bond factors are constructed according to Fama and French (1993).
The T ERM factor is defined as “the difference between the monthly long-
term government bond return and the one-month Treasury bill rate” (Fama
and French, 1993).
For the long-term government bond return this paper uses the ten-year con-
stant maturity rate provided by the FRED database of the Federal Reserve
Bank of St. Louis on a daily basis.
The risk-free rate is again the one-month T-Bill rate provided by the CRSP.

Eugene Fama and Kenneth French define the DEF factor as “the difference
between the return on a market portfolio of long-term corporate bonds and

23
the long-term government bond return” (Fama and French, 1993).
This paper uses the Bloomberg Barclays Long US Corporate Total Return In-
dex from the Bloomberg database as proxy for the long-term corporate bond
returns. The long-term government bond return is constructed in the same
way as for the T ERM factor.
Both factors are constructed with daily data.

24
5 Empirical Research
5.1 Summary

Table II provides the summary statistics of the Green and conventional bond
samples. After the previously mentioned data cleansing and filtering, the
Green bond sample consists of 359 bonds while the set of conventional bonds
is made of 1291 fixed income contracts. The overall sample then consists of
1650 different bonds.
Over the considered time period between January 4th, 2011 and June 30th,
2017, the average yield-to-maturity of the conventional bonds amounts to
4.12%. The descriptive statistics suggest that the Green bonds sample, with
an average yield-to-maturity of 4.16% seems to outperform the conventional
sample by about 0.04% on average. Conventionals do show a standard de-
viation of 4.81% while the one of green bonds amounts to 10.52%. Figure 6
summarises the average amount of Green and conventional bond issuances per
year while Table III provides basic descriptive statistics for the factor loadings.

Table II.
Summary Statistics of Green and conventional bond samples

Bond Class M ean Y tM M ean M od. Duration Std. Dev. # of Bonds

Green 4.162198 4.12427 10.52012 359

Conventional 4.11899 3.55318 4.808956 1291

Total 1650

The table provides summary statistics on both Green and conventional bond samples used
in this paper after applying the different data cleansing and filtering operations mentioned
earlier in the study. The considered time period is January 4th, 2011 until June 29th, 2017.
The average yield-to-maturity is calculated using daily data and constructing the equally
wheighted mean of all 1650 bonds over the sample period. The same procedure is applied
for the calculation of the standard deviation. All figures are in U.S. Dollars.

25
 Figure 6.

Table III.
Descriptive statistics for independent variables used in seven-factor
model

M arket SM B HM L M OM T ERM DEF

Mean 0.00055 -0.00006 0.00002 0.00005 0.02251 0.05610

The table provides basic descriptive statistics for the market risk premium (M arket), the
size and value premia (SM B, HM L), the momentum (M OM ) factor as well as term and
default premia (T ERM , DEF ) used in the different models of this research in order to
estimate bond returns. The are calculated by averaging the data provided by the CRSP and
the T ERM and DEF factors constructed as explained in detail in Section 4.7.

26
5.2 CAPM Regression Results

Table IV shows the CAPM results for the study using the Kenneth R. French
factor provided by the Center for Research in Security Prices as market bench-
mark. The overall international investment orientation of the considered bonds
supports the choice of this factor which follows Ibikunle and Steffen (2015).
The results show an outperformance of Green bonds over the market bench-
mark in comparison to conventional ones over the full sample period from
2011-2017. However, we can observe a rather low adjusted R2 of 0.1475 that
is consistent with the literature of Fama and French (1993) and Lettau and
Ludvigson (2001). This would support the hypothesis, that the GREEN co-
efficient is so high because it incorporates other effects not taken into account
by the model. We are probably facing an unobserved variable bias and ex-
tend the initial CAPM with further predictor values to increase its power as
demonstrated in the literature (Fama and French, 1993).

27
 Table IV.
Fama-Macbeth Regressions: λ̄j Coefficient
Average estimates on betas in cross-sectional regression following
Fama Macbeth Two-Step Procedure

Asset Pricing Model:

λ̄RM λ̄GREEN R2 2
Radj.

CAPM 0.8828 4.3821* 0.1549 0.1475

(10.52) (6.49)

Observations 1570

The table presents λ̄j estimates for the CAPM-based regressions. The Fama-Macbeth Ap-
proach is used with excess yield-to-maturities of 359 Green and 1291 conventional bonds.
The λ̄j estimates result from averaging the results of the cross-sectional regressions (second
pass) of the Fama-Macbeth procedure and are estimates for the betas of the factor in the
column heading. RM is the market proxy collected from the Center for Research in Se-
curity Prices that is used to approximate the equity market performance and measure the
risk-adjusted returns of the two bond classes. λ̄RM measures the effect of RM .
The standard errors are shown in parentheses. * , **
and ***
are corresponding to a statistic
significance at the 10 %, 5 % and 1 % level respectively.
The R2 and Radj.
2
are calculated by averaging the R2 and Radj.
2
of all cross-sectional re-
gressions run in the Fama-Macbeth procedure. The adjusted R2 adjusts for the number of
independent variables in the model and therefore only increases if an additional predictor
increases the predictive power of the model.

28
5.3 Fama-French Three-Factor Model Regression Re-
sults

Fama and French (1993) prove the inferior explanatory power of the Capital
Asset Pricing Model and suggest an extension that should enhance the ca-
pability of predicting equity returns. For the bond sample that this paper
is analysing, adding the size and value premia as explanatory variable does
2
add to the quality of the model but only very little with an Radj. of 0.1811
2
compared to an Radj. of 0.1475 for the CAPM as it is shown in Table V. In
2
their research Fama and French (1993) find a high Radj. . The difference can
still be consistent with their findings since their three-factor model is focusing
on stocks and not bonds. This research is going to test their five-factor model
including the bond factor later on.

29
 Table V.
Fama-Macbeth Regressions: λ̄j Coefficient
Average estimates on betas in cross-sectional regression following
Fama Macbeth Two-Step Procedure

Asset Pricing Model:

λ̄RM λ̄SM B λ̄HM L λ̄GREEN R2 2
Radj.

FF3F 0.5993 0.2283 0.0105 4.3768* 0.1958 0.1811

(4.22) (5.11) (2.98) (7.00)

Observations 1570

The table presents λ̄j estimates for the Fama-French three-factor Model-based regressions.
The Fama-Macbeth Approach is used with excess yield-to-maturities of 359 Green and 1291
conventional bonds. The λ̄j estimates result from averaging the results of the cross-sectional
regressions (second pass) of the Fama-Macbeth procedure and are estimates for the betas
of the factor in the column heading. RM is the market proxy collected from the Center
for Research in Security Prices that is used to approximate the equity market performance
and measure the risk-adjusted returns of the two bond classes. λ̄RM measures the effect of
RM . SM B is the size premium accounting for the size anomaly while HM L is the value
premium that proxies for the value anomaly. Both factors are retrieved from the CRSP
database. λ̄SM B and λ̄HM L measure the effect of SM B and HM L respectively.
The standard errors are shown in parentheses. * , **
and ***
are corresponding to a statistic
significance at the 10 %, 5 % and 1 % level respectively.
The R2 and Radj.
2
are calculated by averaging the R2 and Radj.
2
of all cross-sectional re-
gressions run in the Fama-Macbeth procedure. The adjusted R2 adjusts for the number of
independent variables in the model and therefore only increases if an additional predictor
increases the predictive power of the model.

30
5.4 Carhart Four-Factor Model Regression Results

Grinblatt et al. (1995) find a superior performance for equity investments based
on momentum strategies. A similar reasoning could be made for including the
momentum factor to the bond pricing model of this paper. Table VI shows
that the results of the Fama-Macbeth procedure for this model do provide
an superior adjusted R2 of 0.2486 that leads to support the afore-mentioned
argument. The GREEN coefficient is contiously decreasing with the addi-
tion of new factors to the initial model which supports the hypothesis of the
omitted variable bias. All four-factors of this model do however relate to the
equity market which could be is another explanation of the, still, rather poor
performance of the model. The following seven-factor model is going take into
account two bond factors and provides support for this argument.

31
 Table VI.
Fama-Macbeth Regressions: λ̄j Coefficient
Average estimates on betas in cross-sectional regression following
Fama Macbeth Two-Step Procedure

Asset Pricing Model:

λ̄RM λ̄SM B λ̄HM L λ̄M OM λ̄GREEN

Carhart four-factor Model 0.7440 0.0133 −0.1543 −0.3649 3.6451*

(5.49) (−2.76) (−2.32) (−12.16) (5.18)

R2 0.2666

2
Radj. 0.2486

Observations 1570

The table presents λ̄j estimates for the Fama-French three-factor Model-based regressions.
The Fama-Macbeth Approach is used with excess yield-to-maturities of 359 Green and 1291
conventional bonds. The λ̄j estimates result from averaging the results of the cross-sectional
regressions (second pass) of the Fama-Macbeth procedure and are estimates for the betas
of the factor in the column heading. RM is the market proxy collected from the Center
for Research in Security Prices that is used to approximate the equity market performance
and measure the risk-adjusted returns of the two bond classes. λ̄RM measures the effect of
RM . SM B is the size premium accounting for the size anomaly while HM L is the value
premium that proxies for the value anomaly and M OM proxies momentum in the market
to take account of the momentum anomaly. The three factors are retrieved from the CRSP
database. λ̄SM B , λ̄HM L and λ̄M OM measure the effect of SM B, HM L and M OM respec-
tively.
The standard errors are shown in parentheses. * , **
and ***
are corresponding to a statistic
significance at the 10 %, 5 % and 1 % level respectively.
The R2 and Radj.
2
are calculated by averaging the R2 and Radj.
2
of all cross-sectional re-
gressions run in the Fama-Macbeth procedure. The adjusted R2 adjusts for the number of
independent variables in the model and therefore only increases if an additional predictor
increases the predictive power of the model.

32
5.5 Seven-Factor Model Regression Results

Table VII shows the results for the Fama-Macbeth estimation of the seven-
factor model as specified in Section 3.2. With a coefficient of 1.6047 the
GREEN dummy still remains positive and suggests that Green bonds are
outperforming conventional ones over the sample period. Thereby the initial
hypothesis, stating that the performance of Green bonds is inferior to the one
of their conventional peers can be rejected. The adjusted R2 of the seven-factor
model does increase in comparison to the Carhart four-factor model but is how-
ever not comparable to the one found in Fama and French (1993). By referring
to the Green bond issuance stated in Table I we observe a very low number of
Green bond issuances over the first years of the research. Furthermore the first
Green bonds are mostly issued by supranational and governmental institutions
whereas the real growth of the market only starts with corporate Green bond
issuances. Considering this and as a further robustness check, a subperiod
coinciding with the “take-off” of the corporate green bond issuances in 2015 is
constructed. The results from this test can be seen in Table VIII.
One can observe that the constructed seven-factor model is able to explain
much of the underlyings for the movements in bond yield-to-maturities in the
subperiod. The substantial increase of the adjusted R2 in the subperiod com-
pared to the whole sample period supports the argument that the explanatory
power of the model increases with the “take-off” of the green bond market.
At the same time, the adjusted R2 of 0.6319 in the subperiod leads to the
conclusion that there are still other, unobserved factors such as inflation or
issuer-specific risk factors that are influencing the outcome and that have to
be identified and observed in order to further improve the model although the
results obtained here are consistent with those of Fama and French (1993).
Again, the substantial factor loading of the GREEN dummy provides em-
pirical evidence of the outperformance of Green bonds versus their conventional
peers.
The pattern of the findings of this research is very similar to the one Ibikunle
and Steffen (2015) find for the green equity mutual fund market: They find
that the performance of green funds progressively improves over time and sug-
gest that this could be driven by “the transition from a fossil fuel age into
an emission-constrained one”. The same reasoning could be adopted for the
findings of this paper on behalf of the Green bond market even if this analysis
does not provide a definite proof for this pattern.
If one analyses the single regression results of the cross-sectional regressions,

33
one notifies the consistent increase of the adjusted R2 as well as the fact that
2
after April 2017, the Radj. is consistently above 0.79 as it can be seen in Figure
8 and in the appendix. From this point, it would be interesting to conduct
the same research again at a later point in time and analyse if this pattern is
continuing.

34
 Table VII.
Fama-Macbeth Regressions: λ̄j Coefficient
Average estimates on betas in cross-sectional regression following
Fama Macbeth Two-Step Procedure

Model:
λ̄RM λ̄SM B λ̄HM L λ̄M OM λ̄T ERM λ̄DEF λ̄GREEN

7-Factor Model 0.5906 0.2450 0.3187 −0.6559 0.0241* −0.5780 1.6047

(3.95) (0.85) (3.48) (−10.02) (4.28) (−7.75) (3.22)
R2 0.3164

2
Radj. 0.2917

Observations 1570

The table presents λ̄j estimates for the Fama-French three-factor Model-based regressions.
The Fama-Macbeth Approach is used with excess yield-to-maturities of 359 Green and 1291
conventional bonds. The λ̄j estimates result from averaging the results of the cross-sectional
regressions (second pass) of the Fama-Macbeth procedure and are estimates for the betas
of the factor in the column heading. RM is the market proxy collected from the Center
for Research in Security Prices that is used to approximate the equity market performance
and measure the risk-adjusted returns of the two bond classes. λ̄RM measures the effect of
RM . SM B is the size premium accounting for the size anomaly while HM L is the value
premium that proxies for the value anomaly and M OM proxies momentum in the market
to take account of the momentum anomaly. The three factors are retrieved from the CRSP
database. λ̄SM B , λ̄HM L and λ̄M OM measure the effect of SM B, HM L and M OM respec-
tively. T ERM accounts for the long-term bonds deviation from expected returns caused
by interest rate shifts. It is calculated from the difference of a long-term government bond
return and the one-month T-Bill. DEF proxies the default probability change due to eco-
nomic condition changes and is calculated as the difference between a portfolio of long-term
corporate bonds and the long-term government bond return as in Fama and French (1993).
Again, λ̄T ERM and λ̄DEF account for the effect of both factors respectively.
The standard errors are shown in parentheses. * , **
and ***
are corresponding to a statistic
significance at the 10 %, 5 % and 1 % level respectively.
The R2 and Radj.
2
are calculated by averaging the R2 and Radj.
2
of all cross-sectional re-
gressions run in the Fama-Macbeth procedure. The adjusted R2 adjusts for the number of
independent variables in the model and therefore only increases if an additional predictor
increases the predictive power of the model by calculating:
2 (1−R2 )∗(N −1)
Radj. = 1− N −p−1 , with N being the total sample size and p being the number of
predictors used in the model.

35
 Table VIII.
Fama-Macbeth Regressions: λ̄j Coefficient
Average estimates on betas in cross-sectional regression following
Fama Macbeth Two-Step Procedure from 25/11/2015 - 29/06/2017

Model:
λ̄RM λ̄SM B λ̄HM L λ̄M OM λ̄T ERM λ̄DEF λ̄GREEN

7-Factor Model 0.7510*** 0.2958*** 0.0936** −0.4123*** 0.0003 −0.0.0257 3.7113***

(21.10) (9.06) (2.63) (−8.48) (0.26) (1.15) (13.80)

R2 0.64319

2
Radj. 0.6336

Observations 1570

The table presents λ̄j estimates for the Fama-French three-factor Model-based regressions.
The Fama-Macbeth Approach is used with excess yield-to-maturities of 359 Green and 1291
conventional bonds. The λ̄j estimates result from averaging the results of the cross-sectional
regressions (second pass) of the Fama-Macbeth procedure and are estimates for the betas
of the factor in the column heading. RM is the market proxy collected from the Center
for Research in Security Prices (CRSP) that is used to approximate the equity market per-
formance and measure the risk-adjusted returns of the two bond classes.λ̄RM measures the
effect of RM . SM B is the size premium accounting for the size anomaly while HM L is
the value premium that proxies for the value anomaly and M OM proxies momentum in the
market to take account of the momentum anomaly. The three factors are retrieved from the
CRSP database. λ̄SM B , λ̄HM L and λ̄M OM measure the effect of SM B, HM L and M OM
respectively. T ERM accounts for the long-term bonds deviation from expected returns
caused by interest rate shifts. It is calculated from the difference of a long-term government
bond return and the one-month T-Bill. DEF proxies the default probability change due to
economic condition changes and is calculated as the difference between a portfolio of long-
term corporate bonds and the long-term government bond return as in Fama and French
(1993). Again, λ̄T ERM and λ̄DEF account for the effect of both factors respectively.
The standard errors are shown in parentheses. * , ** and *** are corresponding to a statistic
significance at the 10 %, 5 % and 1 % level respectively.
The R2 and Radj.
2
are calculated by averaging the R2 and Radj.
2
of all cross-sectional re-
gressions run in the Fama-Macbeth procedure. The adjusted R2 adjusts for the number of
independent variables in the model and therefore only increases if an additional predictor
increases the predictive power of the model by calculating:
2 (1−R2 )∗(N −1)
Radj. = 1− N −p−1 , with N being the total sample size and p being the number of
predictors used in the model.

36
6 Summary and Conclusions

Especially in the follow-up to the COP 21 agenda and the UN 17 sustainable

development goals, the sustainable and responsible investment market sees in-
creasing inflows and gets more and more attention even from investors without
a specific SRI focus. Green bonds channeling funds to environmentally friendly
projects are one instrument to implement those investments. Evidence for this
increase can be found in the huge increase of Green bond issuances in the re-
cent year and months as can be seen in Figure 7 below. However, even though
the market of Green bonds doubled in terms of amounts issued from 2016 to
2017, the global Green bond market still consists of less than one percent of
the overall bond market. Despite more and more investors being obliged to
invest sustainably by either the regulator or the client demand, the ultimate
criteria for the invest in a certain asset still remains its financial return.

Figure 7.

With the financial performance of Green bonds being a rather unchartered

territory in terms of empirical research, this study aims to make a substantial
contribution to the field. To the knowledge of the author, this study is the
first to analyse the financial performance of green bonds in comparison to their
conventional peers based on a forward-looking asset pricing model. Indeed the
study tests the performance with various models and creates a GREEN factor
in order to account for a potential effect on the financial return due to the fact
that the considered bond is green.

The key contributions of the analysis are as follows: (1) it is, to the best
knowledge of the author, the first empirical studies to analyse the financial

37
performance of Green bonds compared to their traditional counterparts, (2)
it is testing an extended Fama-French model in a fixed income environment
and (3) finds that the fact of a bond being green does positively influence its
expected financial performance.

As Figure 8 shows, the explanatory power of the constructed model does signifi-
cantly increase towards the end of the financial year of 2015 which overlaps with
the “take-off” of corporate Green bond issuances as Table I demonstrates. The
bi-annual cross-sectional regression results that can be found in the Appendix
do provide further evidence for this as well as for the increasing significance of
the GREEN coefficient over time.

Figure 8. Adjusted R squared over time

As Figure 9 shows, the GREEN coefficient is negative in the beginning of the

sample period but increases continually until it gets positive towards the end of
2011. During the first half of the sample period, the factor loading of GREEN
is quite volatile but this volatility seems to evade when the amount of corpo-
rate green bond issuances starts its “take-off” in 2015. The overall average
(λ̄GREEN ) throughout the whole sample period of 1.6047 is clearly speaking
for an outperformance of the Green bonds in comparison to the traditional
one. In our subsample, the effect of the bond being green seems to have an

38
even bigger effect.

Figure 9. Seven-factor model estimations for λ̂GREEN over time

All in all, the findings of this research do not only provide a new insight on
the fundamentals of the Green bond market but they are at the same time
providing a key element for the development and marketing of the Green bond
market. With the argument of Green bonds being more of a charity investment
than a financially interesting instrument wiped out, the financing of the fight
against climate change should get further attention and support.

39
7 Limitations and further research

Even if the author has put substantial effort into the representativeness of the
data, the findings are limited by the data and only applicable for the consid-
ered time period. It is still possible that the estimated positive effect of a bond
being green is at least partly due to variables unobserved in this study such as
the state of the technology, the innovativeness and general positioning of the
issuer. This asks for the inclusion of further factors enhancing the explanatory
power of the model and providing further insights into the underlying funda-
mentals of the Green bond market.

Also, considering the fact the Green bond market was born only ten years
ago and the real “take-off” only started in 2015 with a substantial amount
of issuances coming from the corporate market, one should conduct the same
analysis again at a later point in time with a more mature market that will
enable the researcher to make statements based on a longer sample period and
a bigger sample of Green bonds that can make up for the unbalanced sample
that this research had to rely on.
Furthermore, he thesis is purely focusing on financial performance and does
not take behavioural considerations or other non-financial factors potentially
driving performance into account.

A further approach to analyse the performance while accounting for issuer

specific risks would be to adopt a two-by-two approach comparing Green and
conventional bonds coming from the same issuer. Those are considered hav-
ing the same inherent risk characteristics and the difference in performance
should purely came from the “greeness”. Also, the inclusion of a black bonds
class accounting for bonds whose proceeds are specifically targeted to fossil
and non-renewable energy projects could be of further interest to the industry.
Inflation and liquidity would be two further factors that could be integrated
into the pricing model to potentially further increase its explanatory power.
In general, future research and time is due in order to avoid premature state-
ments and tell wether the hypothesis holds true.

40
References
Bloomberg New Energy Finance (2016). Guide to green bonds on the
bloomberg terminal.

Carhart, M. M. (1997). On persistence in mutual fund performance. Journal

of Finance, 52(1), 57–82.

Cavenaile, L., Dubois, D., and Hlávka, J. (2009). Cross-Sectional Tests of

Fama and French Three-Factor Model and Fama-MacBeth Method. Working
Paper Ecole de gestion de l’Université de Liège.

Clapp, C., Alfsen, K. H., Lund, H. F., and Pillay, K. (2016). Green bonds
and environmental integrity: Insights from cicero second opinions. Techni-
cal report, CICERO - Center for International Climate and Environmental
Research.

Climate Bonds Initiative (2016). Bonds and climate change - the state of the
market in 2016. Technical report, Climate Bonds Initiative.

Climate Bonds Initiative (2017). Green bonds policy: Highlights from 2016.
Technical report, Climate Bonds Initiative.

Cochrane, J. H. (2009). Asset Pricing. Princeton University Press, Princeton.

Cochrane, J. H. (2014). Empirical Methods notes. http://faculty.chicagobooth.

edu/john.cochrane/teaching/35150 advanced investments/week5a notes.
pdf. Last accessed on 04.07.2017.

Ebrey, P. B. (2010). The cambridge illustrated history of china. Technical

report, Princeton University Press, Cambridge.

EY (2016). Green bonds - a fresh look at financing green projects. Technical

report, EY.

Fama, E. F. and French, K. R. (1992). The Cross-Section of Expected Stock

Returns. Journal of Finance, XLVII(2), 427–465.

Fama, E. F. and French, K. R. (1993). Common risk factors in the returns on

stocks and bonds. Journal of Financial Economics, 33(1), 3–56.

Fama, E. F. and French, K. R. (2004). The capital asset pricing model: Theory
and evidence. Journal of Economic Perspectives, 18(3), 25–46.

Fama, E. F. and Macbeth, J. D. (1973). Risk, return and equilibrium: Empir-

ical tests. Journal of Political Economy, 81(3).

41
French, K. R. (2015). Kenneth French’s Website.
http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/data library.html.
Last accessed on 07.05.2017.

Gabbi, G. and Sironi, A. (2005). Which factors affect corporate bonds pricing?
empirical evidence from eurobonds primary market spreads. The European
Journal of Finance, 11(1), 59–74.

Gil-Bazo, J., Ruiz-Verdú, P., and Santos, A. A. P. (2010). The performance of

socially responsible mutual funds: The role of fees and management compa-
nies. Journal of Business Ethics, 94(2), 243–263.

Glomsrød, S. and Taoyuan, W. (2016). Working paper 2016:01 business as

unusual - the implications of fossil divestment and green bonds for financial
flows, economic growth and engery market. Technical report, CICERO -
Center for International Climate and Environmental Research.

Grinblatt, M., Titman, S., and Wermers, R. (1995). Momentum investment

strategies, portfolio performance, and herding: A study of mutual fund be-
havior. The American Economic Review, 85(5), 1088–1105.

Houweling, P., Mentink, A., and Vorst, T. (2005). Comparing possible proxies
of corporate bond liquidity. Journal of Banking and Finance, 29(6), 1331–
1358.

Ibikunle, G. and Steffen, T. (2015). European mutual fund performance: A

comparative analysis with their conventional and black peers. Journal of
Business Ethics, 145(2), 337–355.

Johansson, D. and Lundgren, T. (2012). A study of corporate bond returns -

using Sharpe-Lintner CAPM and Fama-French. Master’s thesis, Stockholm
School of Economics.

Lettau, M. and Ludvigson, S. (2001). Resurrecting the (C)CAPM: A Cross-

Sectional Test When Risk Premia Are Time-Varying. Journal of Political
Economy, 109(6), 1238–1287.

Lintner, J. (1965). The Valuation of Risk Assets and the Selection of Risky
Investments in Stock Portfolios and Capital Budgets. The Review of Eco-
nomics and Statistics, 47(1), 13–37.

LuxFLAG (2017). About us. http://www.luxflag.org/about/who-

we-are.html&sid=894bcca948bf0a18e0eb87c4ce53d6ca. Last accessed on
22.09.2017.

42
Merton, R. C. (1973). An intertemporal capital asset pricing model. Econo-
metrica, 41(5), 867–887.

Natixis (2017). Green bonds 3.0: Quality check. Technical report, Natixis
Corporate and Investment Banking.

NEPC Impact Investing Committee (2016). What a caia member should know
- an introduction to green bonds. Technical report, NEPC Impact Investing
Committee.

Parnassus Investments (2015). Parnassus digest - march 2015 - green

bonds. https://www.parnassus.com/our-firm/highlight/153. Last accessed
on 15.05.2017.

Petkova, R. (2011). Financial Economics, The Cross-Section of Stock Returns

and the Fama-French Three Factor Model, pages 361–374. Springer New
York, http://dx.doi.org/10.1007/978-1-4419-7701-4 21.

Sharpe, W. F. (1964). Capital asset prices: A theory of market equilibrium

under conditions of risk. Journal of Finance, 19(3), 425–442.

Stock, J. H. and Watson, M. M. (2012). Introduction to Econometrics. Pearson

Education Limited, 3rd edition.

World Bank (2015). Green bonds attract private sector climate fi-
nance. http://www.worldbank.org/en/topic/climatechange/brief/green-
bonds-climate-finance. Last accessed on 22.09.2017.

Xiao, Y., Faff, R., Gharghori, P., and Lee, D. (2012). An empirical study of the
world price of sustainability. Journal of Business Ethics, 114(2), 297–310.

43
Appendix A: Appendix
1.1 Seven-factor cross-sectional regression results over
time (bi-annual)

Table IX. 04.01.2011

Estimate Std. Error t value Pr(>|t|)
RM 0.3003 2.1924 0.14 0.8914
SMB 0.5517 1.5290 0.36 0.7194
HML 1.1330 2.0342 0.56 0.5794
MOM -0.3918 1.9567 -0.20 0.8419
TERM 0.0560 0.0499 1.12 0.2656
DEF -0.6869 0.9761 -0.70 0.4840
GREEN -1.9926 6.6623 -0.30 0.7658

Table X. 01.07.2011
Estimate Std. Error t value Pr(>|t|)
RM 0.8730 1.4099 0.62 0.5374
SMB 0.7344 1.1456 0.64 0.5231
HML 0.7147 0.7084 1.01 0.3157
MOM -0.4897 1.1514 -0.43 0.6716
TERM 0.0514 0.0359 1.43 0.1556
DEF -0.6818 0.7750 -0.88 0.3814
GREEN -0.6310 5.6007 -0.11 0.9106

Table XI. 03.01.2012

Estimate Std. Error t value Pr(>|t|)
RM 1.0283 1.1294 0.91 0.3647
SMB 0.7667 1.0338 0.74 0.4600
HML 0.6195 0.5987 1.03 0.3033
MOM -0.8903 0.9395 -0.95 0.3455
TERM 0.0354 0.0298 1.19 0.2366
DEF -0.3843 0.6430 -0.60 0.5513
GREEN 0.1296 4.3821 0.03 0.9765

44
 Table XII. 02.07.2012
Estimate Std. Error t value Pr(>|t|)
RM 0.6432 0.8524 0.75 0.4518
SMB 0.2781 0.8269 0.34 0.7372
HML 0.6222 0.5038 1.24 0.2190
MOM -0.8875 0.6540 -1.36 0.1771
TERM 0.0427* 0.0231 1.85 0.0663
DEF -0.5826 0.4888 -1.19 0.2354
GREEN 1.3167 3.2441 0.41 0.6855

Table XIII. 02.01.2013

Estimate Std. Error t value Pr(>|t|)
RM 0.4330 0.4866 0.89 0.3747
SMB 0.5679 0.5335 1.06 0.2886
HML 0.2337 0.3358 0.70 0.4874
MOM -0.7719* 0.4173 -1.85 0.0660
TERM 0.0237* 0.0141 1.68 0.0955
DEF -0.4955* 0.2888 -1.72 0.0880
GREEN 2.2718* 2.2616 1.00 0.3165

Table XIV. 01.07.2013

Estimate Std. Error t value Pr(>|t|)
RM 0.6026* 0.3087 1.95 0.0520
SMB 0.3470 0.2999 1.16 0.2482
HML 0.1097 0.2204 0.50 0.6192
MOM -0.8488*** 0.2133 -3.98 0.0001
TERM 0.0267*** 0.0101 2.64 0.0088
DEF -0.5953*** 0.1763 -3.38 0.0008
GREEN 3.4154** 1.6120 2.12 0.0351

45
 Table XV. 02.01.2014
Estimate Std. Error t value Pr(>|t|)
RM 0.6041*** 0.1983 3.05 0.0025
SMB -0.0106 0.1863 -0.06 0.9545
HML 0.2175 0.1470 1.48 0.1397
MOM -0.9205*** 0.1388 -6.63 0.0000
TERM 0.0256*** 0.0059 4.32 0.0000
DEF -0.6377*** 0.1070 -5.96 0.0000
GREEN 2.7866 *** 0.9616 2.90 0.0039

Table XVI. 01.07.2014

Estimate Std. Error t value Pr(>|t|)
RM 0.5807*** 0.1426 4.07 0.0001
SMB -0.1417 0.0902 -1.57 0.1164
HML 0.1907** 0.0867 2.20 0.0283
MOM -0.7661*** 0.1054 -7.27 0.0000
TERM 0.0224*** 0.0041 5.51 0.0000
DEF -0.6706*** 0.0796 -8.42 0.0000
GREEN 2.1212*** 0.6321 3.36 0.0008

Table XVII. 02.01.2015

Estimate Std. Error t value Pr(>|t|)
RM 0.6127*** 0.1006 6.09 0.0000
SMB -0.0969** 0.0458 -2.12 0.0346
HML 0.2905*** 0.0446 6.51 0.0000
MOM -0.4710*** 0.0431 -10.93 0.0000
TERM 0.0267*** 0.0026 10.34 0.0000
DEF -0.8002*** 0.0581 -13.76 0.0000
GREEN 1.4294*** 0.4527 3.16 0.0016

46
 Table XVIII. 01.07.2015
Estimate Std. Error t value Pr(>|t|)
RM 0.5770*** 0.0883 6.54 0.0000
SMB -0.1580*** 0.0393 -4.02 0.0001
HML 0.2078*** 0.0361 5.76 0.0000
MOM -0.3960*** 0.0333 -11.91 0.0000
TERM 0.0200*** 0.0019 10.34 0.0000
DEF -0.7003*** 0.0454 -15.42 0.0000
GREEN 1.5016*** 0.3616 4.15 0.0000

Table XIX. 04.01.2016

Estimate Std. Error t value Pr(>|t|)
RM 0.4275*** 0.0715 5.98 0.0000
SMB -0.0082*** 0.0347 -0.24 0.8134
HML 0.1999*** 0.0307 6.51 0.0000
MOM -0.6212*** 0.0293 -21.22 0.0000
TERM 0.0091*** 0.0015 6.00 0.0000
GREEN 1.8151*** 0.3030 5.99 0.0000

Table XX. 01.07.2016

Estimate Std. Error t value Pr(>|t|)
RM 0.5041*** 0.0639 7.89 0.0000
SMB 0.1056*** 0.0316 3.34 0.0009
HML 0.2159*** 0.0267 8.07 0.0000
MOM -0.6543*** 0.0274 -23.84 0.0000
TERM 0.0021 0.0013 1.56 0.1195
DEF -0.4156*** 0.0333 -12.49 0.0000
GREEN 1.9645*** 0.2537 7.74 0.0000

47
 Table XXI. 03.01.2017
Estimate Std. Error t value Pr(>|t|)
RM 0.5331*** 0.0637 8.37 0.0000
SMB 0.1514*** 0.0317 4.77 0.0000
HML 0.2801*** 0.0267 10.50 0.0000
MOM -0.7511*** 0.0277 -27.13 0.0000
TERM 0.0094*** 0.0013 7.03 0.0000
DEF -0.4680*** 0.0334 -13.99 0.0000
GREEN 2.0552*** 0.2521 8.15 0.0000

Table XXII. 29.06.2017

Estimate Std. Error t value Pr(>|t|)
RM 0.5114*** 0.0664 7.71 0.0000
SMB 0.5998*** 0.0331 18.14 0.0000
HML 0.0943*** 0.0278 3.39 0.0007
MOM -0.8242*** 0.0288 -28.57 0.0000
TERM 0.0018 0.0014 1.32 0.1866
DEF -0.3307*** 0.0349 -9.49 0.0000
GREEN 2.2564*** 0.2626 8.59 0.0000

48