Contents lists available at ScienceDirect
Journal of Financial Stability
journal homepage: www.elsevier.com/locate/jfstabil
What greenium matters in the stock market? The role of greenhouse gas
emissions and environmental disclosures☆
Lucia Alessia,b,⁎
, Elisa Ossolaa
, Roberto Panzicaa
a
European Commission, Joint Research Centre, Italy
b
CefES − Center for European Studies (Universitá degli Studi di Milano-Bicocca), Italy
a r t i c l e i n f o
Article history:
Received 4 March 2019
Received in revised form 2 November 2020
Accepted 18 November 2020
Available online 31 March 2021
JEL classification:
G01
G11
G12
Q01
Keywords:
Climate risk
Environmental disclosure
Factor models
Asset pricing
Stress test
a b s t r a c t
This study provides evidence on the existence of a negative greenium, i.e. a risk premium related to the
greenness of a firm, based on European individual stock returns. We define a priced ‘greenness and
transparency’ factor based on companies’ greenhouse gas emissions and the quality of their environmental
disclosures, and show that what is priced by the market is the combination of environmental performance
and environmental transparency. Based on this factor, we offer a tool to assess the exposure of a portfolio to
the risk associated with the low-carbon transition, and hedge against it. We estimate that in a stressed
scenario where greener and more transparent firms very much outperform brown stocks, there would be
losses at the global level, including for European large banks, should investors fail to price climatetransition
risks. These results call for the introduction of climate stress tests for systemically important
financial institutions.
© 2021 The Authors. Published by Elsevier B.V.
CC_BY_4.0
1. Introduction
Climate change is a fact, but we are not sure what the economic
costs associated with this change will be.1
By the same token, it is
difficult to estimate what the economic benefits of doing something
about it would be. In particular, it would be hard to pin down the net
present value of activities aimed at climate change adaptation and
mitigation, as well as those directed to broader environmental objectives
such as the sustainable use and protection of water and
marine resources, the transition to a circular economy, waste prevention
and recycling, pollution prevention and control, and the
protection of healthy ecosystems.2
At the same time, the consequences
of a transition to a low-carbon, resource-efficient and
circular economy, or lack thereof, are also largely uncertain. Hence,
these issues have to be addressed as aspects of long-run risk. The
contribution of this paper is to measure the added value of greener
economic activities in terms of market excess returns. To do so, we
first show that indeed, the European market prices climate risk, in
the context of a standard asset pricing model. However, we show
that not only environmental performance matter, but a combination
of greenness and environmental transparency. Second, we estimate
that the market associates a negative risk premium, i.e. a greenium,
to more environmentally friendly and transparent firms.
We argue that to identify the market price of climate risk, or
environmental risk more generally, one should broaden the analysis
https://doi.org/10.1016/j.jfs.2021.100869
1572-3089/© 2021 The Authors. Published by Elsevier B.V.
CC_BY_4.0
☆
Disclaimer: The content of this article does not necessarily reflect the official
opinion of the European Commission. Responsibility for the information and views
expressed therein lies entirely with the authors. This research did not receive any
specific grant from funding agencies in the public, commercial, or not-for-profit
sectors. We thank the co-editors and two anonymous referees for constructive criticism
and numerous suggestions which have lead to substantial improvements over
the previous versions. We thank the authors of Battiston et al. (2017) for providing us
with their data, which we use for the climate stress test. We are grateful to Ivan Faiella
for his comments on the construction of the greenness and transparency indicator.
We thank participants at a number of conferences, including CefES, IRMC, Finance for
Sustainability, CREDIT, EAEPE, as well as at meetings and workshops at the European
Central Bank, and seminars at the Universidad Autonoma de Madrid, at the University
of Pavia and at the EC Joint Research Centre, for useful suggestions.
]]]]
]]]]]]
⁎
Corresponding author at: European Commission, Joint Research Centre, Italy.
E-mail address: lucia.alessi@ec.europa.eu (L. Alessi).
1
On the uncertainty on the rate of increase in average temperatures in the longmedium
horizon, and on the effects of climate change, see e.g. Pindyck (2013).
2
These objectives are listed in the 2018 European Commission Action plan on financing
sustainable growth, which sets out an EU strategy for sustainable finance.
Journal of Financial Stability 54 (2021) 100869
to considering other dimensions than firms’ actual environmental
performance. In particular, recent papers suggest that green bond
issuers’ trustworthiness as well as the existence of a company climate
strategy do matter for investors.3
Hence, we estimate the
market price of climate risk based on a priced factor which considers
both environmental performance and environmental transparency.
In particular, we first construct portfolios characterized by a different
shade of green and a different degree of environmental
transparency. This is based on firm-level information on greenhouse
gas (GHG) or CO2 emissions, combined with a measure of the
completeness of firms’ environmental disclosures, to yield a synthetic
greenness and transparency index for each stock. Companies
which disclose a lower emission intensity, and are very transparent,
attain the highest scores and are included in a greener and more
transparent portfolio. The most straightforward example of greener
companies would be those with a large share of their turnover in
green economic sectors, e.g. renewable energy. Conversely, companies
which do not disclose information on their environmental performance
are labeled as non-transparent. Among these nontransparent
companies, those active in carbon-intensive sectors, e.g. companies
operating coal power plants, are included in a brown portfolio. The
greenness and transparency factor is constructed based on 942
companies listed on the STOXX Europe Total Market Index.
By relying on company-level disclosures and factoring-in their
transparency, we try to tackle the issue of greenwashing, which is
likely to be the reason why the literature has so far failed to find a
consensus on the existence of a priced green factor. Indeed, looking
at the actual composition of the portfolios of publicly traded investment
funds which label themselves as ‘green’ or ‘sustainable’, it
turns out that many funds are clearly less environmentally friendly
than their name would suggest. For example, a fund might indeed
limit its exposure to carbon-intensive sectors, but at the same time
mainly invest in e.g. financial stocks. Banks, insurers and other financial
institutions are admittedly directly responsible for a very
small fraction of GHG emissions, but financial institutions are
probably not what comes to mind when thinking of companies that
are at the forefront of efforts to reduce emissions. At the level of the
individual company, there could be a tendency to disclose only
partial information, emphasizing the environmental dimensions
where the firm performs best and neglecting those where the firm
does not as well. For example, a car manufacturer may report its
scope 1 GHG emissions, i.e. emissions from sources that it owns or
controls, but could have an incentive not to disclose its scope 3
emissions, which include emissions resulting from the use of the
vehicles it produces and sells. By considering the completeness of
the relevant information that firms disclose, we deal with this type
of greenwashing.
At the same time, however, one should be careful not to take
extreme views when defining ‘green’ stocks. As a matter of fact,
portfolio diversification is crucial for asset managers, and concentrating
the exposure on a small set of pure green players is not a
viable option. Our society will still need e.g. steel and cement for
quite a while, and companies’ low-carbon transition is a much
needed but gradual process. For all these reasons, a sensible approach
would be to broaden the scope of the definition of ‘green’
beyond pure players to also include firms that meet the highest level
of energy efficiency and the lowest CO2 emissions within the relevant
sector. By taking this approach, a steel manufacturer which
also uses scrap steel is greener than one that does not. By the same
token, an energy company that reduces its reliance on fossil fuels though
not having an entirely renewable energy mix - is greener
than one that is not reducing its carbon footprint. This is the approach
taken by several providers of environmental ratings, which
assess the sustainability of firms relative to their peers, and also the
one we take in this paper.
We show that in the context of a standard asset pricing model,
the portfolios we build based on firms’ environmental performance
and disclosures are associated with a statistically significant intercept,
suggesting the existence of an omitted factor. Based on this
evidence, we propose to include a greenness and transparency
factor, which we construct based on a long-short strategy involving
the greener and more transparent portfolio and the brown portfolio.
We find that the greenium, i.e. the risk premium associated with this
greenness and transparency factor, is negative and highly statistically
significant. This means that investors accept a lower remuneration
for their investments, ceteris paribus, insofar as these
investments are linked to greener and more transparent firms. We
interpret this as evidence of climate risk being viewed as significant,
with the market seeing value in investing in greener assets as a
hedging strategy towards worse environmental outcomes. Indeed, in
a scenario of heightened risks resulting from climate change, there
would be a stronger push towards more environmentally friendly
activities, with more decisive political action likely to be taken to
promote sustainable growth. Hence, companies active in green
sectors and more transparent on their environmental performance
would operate in a more favorable environment, possibly supported
by incentives, e.g. fiscal or of other nature. At the same time, the
likelihood would increase that some assets, e.g. coal, would become
stranded. In this context, forward-looking investors who base their
portfolio allocation on a broader information set than past returns,
invest in greener assets already today.
The evidence we provide on the existence of a greenium has clear
financial stability implications. Indeed, we show that the European
market as a whole does price climate risk. In this context, if an investor
does not factor in climate risk in the construction of her
portfolio, she is in fact pricing her holdings based on a misspecified
model, where the greenness and environmental transparency factor
is omitted. Should this mispricing affect the assets held by systemically
important financial institutions (SIFIs) such as large banks,
insurers and pension funds, there could be consequences in terms of
systemic risk. In particular, asset returns on their holdings could be
negatively affected by climate change via two main channels. First,
in a longer horizon perspective, more frequent and severe natural
catastrophes stemming from climate change (e.g. typhoons and
floods) could negatively affect returns on assets linked to particularly
vulnerable economic activities.4
These are the so-called physical
risks related to climate change, that we do not tackle in this paper
directly. However, it has been shown that rising temperatures have
strong adverse effects on asset valuations, as well as on key macroeconomic
aggregates and productivity (see Donadelli et al. (2017)).
Second, in a medium-term perspective, the implementation of sustainable
finance policies will imply higher costs for firms with higher
emissions, causing a generalized drop in the dividend that brown
firms will be able to pay to their shareholders. In parallel, carbonintensive
assets will increasingly become ‘stranded’ (see Campiglio
et al. (2017)). This is the so-called transition risk. These two channels
characterize a climate risk factor that investors should price. Given
the lack of data on the exposure of individual companies to physical
risks related to climate change, in this analysis we focus on transition
risks, i.e. the potential impacts of a shift to a lower carbonfootprint
economy on firms active in climate-policy-relevant sectors.
3
Looking at the green bond market, Larcker and Watts (2020) examine the yield
differential between same-issuer green and conventional bonds and find no evidence
of a greenium. However, Fatica et al., (2021) document the existence of a greenium
when green bonds are issued by more credible issuers, i.e. supranational institutions,
as well as a positive effect of external review and repeated green bond issuance.
Looking at stocks, Ramelli et al. (2019) show that following Trump’s election, climateresponsible
firms actually received a premium. 4
See Daniel et al. (2016).
L. Alessi, E. Ossola and R. Panzica Journal of Financial Stability 54 (2021) 100869
2
Based on our model, we estimate that in an extreme but plausible
scenario where greener and more transparent companies
outperform brown companies, all institutional sectors at the global
level, including e.g. governments, non-financial institutions and financial
corporations, as well as all European SIFIs, would be hit by
losses. By halving their exposure to carbon-intensive sectors and
reallocating their investments towards greener assets, investors
could somewhat reduce the loss. However, they could only avoid
losing money if they would reallocate their investments towards
greener and more transparent firms. The magnitude of the expected
losses we estimate is admittedly not breathtaking. Still, we show
that no one is in a safe place when it comes to climate risk, as the
consequences of brown asset mispricing would be widespread.
Moreover, our analysis is limited to equity holdings, that for some
investors are not as relevant as other types of exposures. In a
stressed scenario, however, losses would almost certainly be recorded
also on the bond portfolio and notably, on banks’ loan exposure.
Finally, we use a simple model to compute losses, based on
the marginal expected shortfall. This approach does not factor in
losses resulting from second-round effects, like fire sales, which
could magnify first-round losses. Taking all this into account, we
conclude that a climate or climate-policy shock could have serious
implications in terms of financial stability, especially if coupled with
shocks of other nature. Hence, we argue that a climate stress test is
warranted for systemically important institutions to monitor their
resilience to climate change. The greenness and transparency factor
we construct could indeed be used by investors, to hedge against
climate risk, and by supervisors, to measure SIFIs exposure to this
risk. Notice that looking ahead, we can only expect greater policy
pressure to reducing carbon emissions and moving to a sustainable
development path.5
The paper is structured as follows. In the next section, we provide
an overview of the relevant literature. In Section 3, we present our
synthetic greenness and transparency indicator at the level of the
individual company. Section 4 outlines the asset pricing theoretical
framework. In Section 5, we present the results of the empirical
application. First, we focus on portfolios by estimating the standard
asset pricing models and defining the greenness and transparency
factor. Then, we estimate our proposed model on individual stocks.
In Section 6 we carry out a battery of robustness checks. Section 7
tests the performance of the equity portfolios of global institutional
sectors and European SIFIs in a climate-stressed scenario. Section 8
concludes.
2. Related literature
This paper stands at the crossroad of sustainable finance, asset
pricing and financial stability. The sustainable finance literature has
so far mostly focused on corporate performance, starting from the
seminal work by Bragdon and Marlin (1972). They asked the fundamental
question, whether there would be a reward for a company’s
virtue. Trying to answer this question, Porter (1991), Gore
(1993), and Porter and van der Linde (1995) argue that improving a
company environmental performance can lead to a better economic
or financial performance, not necessarily accompanied by an increase
in costs. Ambec and Lanoie (2008) review several empirical
works showing that improvements in the environmental performance
of a firm tend to be associated with improvements in its
economic or financial performance, owing to potential revenue increases
and/or cost cuts. More recently, Hoepner et al. (2018) show
that engagement on sustainability issues can benefit shareholders’
by reducing firms’ downside risks.
Despite increasingly available evidence on the performance of
green or sustainable corporates, however, no consensus has yet been
reached in the asset pricing literature about the performance of
green assets, or on environmental risk being a priced macro factor.
Evidence based on a large number of studies on the performance of
sustainable investment funds compared with conventional peers
(e.g. Statman, (2000); Renneboog et al. (2007); Seitz, (2010)) is
mixed. For example, Hartzmark and Sussman (2019) find that sustainability
is viewed as positively predicting future performance;
however, they do not find evidence of outperformance of ‘high
sustainability’ investment funds vs ‘low sustainability’ ones. Trinks
et al. (2018) show that divesting from fossil fuels does not impair
portfolio performance. Bernardini et al. (2019) focus on European
electric utilities and show that there was a significant low-carbon
premium during the years in which the decarbonization process
accelerated. Moreover, they show that a portfolio strategy focused
on low-carbon companies would have delivered higher returns after
2012. On the contrary, Bolton and Kacperczyk (2020) find that stocks
of companies with higher CO2 emission intensity earn higher returns.
Other analyses based on publicly traded environmental
portfolios find that green stocks are, on average, underperforming
the market. This finding would indicate that investors are willing to
earn comparatively less on these assets because they are hedging an
environmental long-run risk. Finally, recent papers attempt to build
climate risk hedging portfolios (see Engle et al. (2020), Choi et al.
(2020), Hong et al. (2019), Alok et al. (2020), Goergen et al. (2019),
Monasterolo and De Angelis (2020)); however, none of these works
goes all the way to quantifying the associated risk premium.
Finally, the financial stability literature has started to put forward
the idea of ‘climate stress tests’ on the exposures of financial institutions
(see Battiston et al. (2017) and Battiston and Monasterolo
(2018)), as well as to develop climate stress-test methodologies for
e.g. loan portfolios (see Monasterolo et al. (2018)). Central Banks and
international institutions, starting with the seminal speech by
Carney (2015), have also emphasized on different occasions that
climate change could affect systemic risk. In particular, Gros et al.
(2016) distinguishes between a benign scenario, with a gradual
transition to a low-carbon economy, and an adverse scenario, where
the transition occurs more abruptly. In both cases, there could be
financial stability consequences: with a too slow transition, the Paris
Agreement goal would be missed and the catastrophic consequences
of climate change would become unavoidable.6
A too quick transition,
on the other hand, would imply a sudden repricing of brown
assets. We provide evidence that these concerns are shared by the
market.
3. Synthetic greenness and transparency indicator
Different indicators are available to assess a company’s commitment
to the environment. However, identifying synthetic proxies
for firms’ environmental performance is not obvious and no consensus
has yet been reached in the literature on this issue (see
Oikonomou et al. (2012)). The main source of information in this
respect are firms’ environmental disclosures, which are typically
published by companies in their annual reports, or in separate Corporate
Social Responsibility or Sustainability reports, as well as in
dedicated ESG releases or Corporate Governance reports. Based on
these disclosures, investors distinguish companies that are doing
green business from firms that are not, or that are not transparent in
5
Andersson et al. (2016) shows that divestment in higher emission stocks entails a
cost, which investors are more likely to bear the stronger the perception of a serious
commitment on the side of policymakers towards fighting climate change.
6
The objective of the Paris Agreement, signed in 2015, is to keep a global temperature
rise this century well below 2 degrees Celsius above pre-industrial levels and
to pursue efforts to limit the temperature increase even further to 1.5 degrees Celsius.
L. Alessi, E. Ossola and R. Panzica Journal of Financial Stability 54 (2021) 100869
3
this respect. However, a single indicator might not be sufficient to
ensure a careful assessment of a company’s environmental performance.
Hence, we look at both of the following two dimensions: (i)
environmental transparency, related to the quality of firms’ environmental
disclosures, and (ii) GHG emissions.
As a proxy for the environmental transparency of a company we
use the Bloomberg Environmental disclosure score, which we refer
to as E score.7
This is an index quantifying the completeness of a
firm’s disclosure on its impact on the environment. The E score
embraces several environmental aspects of a firm’s business. In
particular, it looks at how transparent a company is with respect to
its carbon emissions, air and water pollution, its commitment to the
protection of biodiversity, and its waste management, among others.
The weighted E score is normalized to range from zero for companies
that do not disclose environmental data to 100 for those which
disclose detailed information for each pillar. The score is also tailored
for industry sectors, and each component is weighted based on
its importance. In particular, GHG emission disclosure is attached
the highest weight.
We use the E score as a measure of the transparency of a firm
with respect to its environmental sustainability commitment, and
assume that higher commitment is associated with higher transparency.
In other words, based on the E score, we make a first selection
among firms that are transparent, at least to some extent,
about their environmental performance, and firms that are not. We
do not claim that firms that do not disclose information on environmental
issues are necessarily ecologically destructive. However,
we find it legitimate to label them as non-green, as their environmental
commitment appears weaker compared to firms that do
disclose (quantitative) information on this pillar. Moreover, Marquis
et al. (2016) show that more environmentally friendly firms tend to
disclose more because they know that their environmental performance
is generally better than the one of their peers, while more
environmentally damaging firms are indeed less likely to engage in
non-mandatory disclosure, in particular in those countries where
they are more exposed to scrutiny and norms.
To build a comprehensive index of a company’s environmental
performance, we combine this transparency measure with quantitative
disclosure on emissions. In particular, we consider the total
GHG emission intensity, i.e. the total amount on GHG emissions
normalized by revenues. If this is not available, we take the total
carbon dioxide (CO2) emitted, weighted by revenues. The synthetic
greenness and transparency indicator Gi,y of company i at year y is
defined as the weighted average of the two components, as follows:
= +G K E(1 ) , with [0, 1],i y i y i y, , , (1)
where Ki,y is the inverse of the ranking of firm i in term of emission
intensity, and Ei,y is the ranking of firm i in term of E score.8
The
parameter γ controls for the relative importance of the two components
of the index. We set γ = 0.5 as benchmark case, and show
results for γ = 0.2 and γ = 0.8 as robustness checks in Section 6.9
More
transparent companies, for a given level of intensity of GHG or CO2
emissions, are associated with larger values of the indicator Gi,y.
Firms that attain a lower emission intensity, for a given level of
transparency, are also associated with larger values of Gi,y. Fig. 1
shows the number of companies in our sample which did some
environmental disclosure, from 2005 to 2017 (yellow bar). The figure
also reports the number of firms that disclosed their emission
intensity in a given year (gray bar). The number of companies reporting
on their environmental performance has exhibited an upward
trend in the last ten years, reaching around 700 EUROSTOXX
companies in 2017, i.e. more than half of the sample.
The indicators we use to build the synthetic greenness and
transparency indicator have some limitations. First, the analysis
provided in Section 5 includes data from 2005, while the Bloomberg
ESG application was launched only in 2009. This potentially introduces
look-ahead bias (see, e.g., Derwall et al. (2005)). However,
this bias is mitigated by the availability of firms’ environmental
disclosures, including on GHG emissions, also prior to 2009. In other
words, market participants could start using the Bloomberg E score
as such only as of 2009, but the information this index is based on
was already available to the market.10
A second issue relates to the self-reported nature of GHG emissions,
as well as of other environmental disclosures. As a matter of
fact, the EU Non-Financial Reporting Directive (NFRD) only requires
auditors to verify the publication of non-financial disclosures by
relevant firms, but there are no assessment and verification requirements
on the content of non-financial disclosures.11
Based on
this, we should emphasize that those firms we identify as greener
based on their environmental disclosures may actually be less green
than they claim. Still, our asset pricing analysis should not be affected
by this problem, insofar as investors base their decisions only
on publicly available information as we do.
Related to the issue of self-reporting is that of self-selection, due
to the fact that unless firms are subject to the NFRD, they are free to
choose whether to disclose environmental information. Those that
are subject to the NFRD, can choose what to disclose in particular.
Typically, self-selection may introduce a bias in the analysis.
However, in our case self-selection should work in the right direction,
as greener firms may have a natural incentive to disclose more
as they have less to hide, and in our framework, firms that are
characterized by a lower emission intensity and better environmental
disclosures are correctly identified as greener and more
transparent. At the same time, more environmentally damaging
firms may also have an incentive to do voluntary disclosure, as they
are subject to greater external pressure (see Marquis et al. (2016)).
While it’s true that in our framework these firms are treated as
transparent, at least to some extent, they will rank very low based on
their emission intensity if they are in fact engaged into highly polluting
activities. Still, we miss information on those firms that do not
disclose.
Finally, ESG ratings may differ quite substantially across different
data providers, as shown by a growing stream of empirical literature,
including, among others, Chatterji et al. (2016), Escrig-Olmedo et al.
(2019), and Berg et al. (2020). This latter paper, for example, shows
that there is much more disagreement among raters on environmental
ratings compared to credit ratings, but still less compared to
social and governance ratings. To partly account for the limited reliability
of environmental ratings and scores, we use the E score in
combination with quantitative data on carbon emissions. We also
show that using only one or the other component of the greenness
and transparency indicator does not yield meaningful results (see
Section 6).
Finally, among non-transparent firms, we further select a subsample
that we label ‘brown’. These are companies which are mainly
active in sectors characterized by a comparatively higher level of
carbon emissions. Information on sectoral emissions in Europe is
7
MSCI, RobecoSAM, Sustainalytics and Asset4 also provide E scores based on different
methodologies.
8
The greenness and transparency indicator Gi,y takes only positive values with a
theoretical maximum equal to N, where N is the number of firms with an E-score
attached at year y. The empirical maximum is 611.5, reached in 2016.
9
Figs. 2 and 3 in the Appendix plot the greenness and transparency indicator, as
well as its components, for two representative companies.
10
Performing the analysis from January 2010 to August 2018 yields empirical results
in line with the ones provided in Section 5.
11
See Directive 2014/95/EU. On this point, Short and Toffel (2008) document that
firms tend to disclose more when they are immune from prosecution for self-disclosed
violations.
L. Alessi, E. Ossola and R. Panzica Journal of Financial Stability 54 (2021) 100869
4
provided by Eurostat, at the NACE-2 digit level.12
Table 11 in the
Appendix lists the economic sectors responsible for the largest
amount of carbon emissions in the EU. Overall, these sectors account
for 85% of the total GHG emissions in the EU over the 2008–2017
period. Table 12 in the Appendix lists the companies that are included
in the brown portfolio in 2017.
4. Linear factor model
We assume an approximate factor structure for excess returns
combined with the absence of arbitrage opportunities to obtain
asset pricing restrictions. As the greenness and transparency indicator
defined in Eq. (1) is only available for a relatively short
sample, we opt for a time-invariant model, which assumes that the
exposition of an asset i to each observable factor does not evolve
over time. We acknowledge that a model that accounts for time
variation in parameters and hence in risk premia would be best
suited in this context, owing to the fact that awareness on climate
issues has increased over time. However, a time-varying model for
the excess returns could only be estimated on a much longer time
series and a much larger cross-section than ours. Indeed, it would
imply introducing functional specifications for the coefficients,
which would result in an incidental parameter problem.
Let us define the excess return on asset i = 1, . . . , n at time
t = 1, 2, . . . T as Ri,t = ri,t−rf,t, where ri,t is the log-return and rf,t is the
risk-free return. We assume that the excess return Ri,t satisfies the
following linear factor model:
= + +
=
R a b f ,i t i
k
K
i k t k i t,
1
, , ,
(2)
where ft,k is the k-th observable factor, with k = 1, . . . , K. The error
term εi,t is s.t. =E[ ] 0i t t, 1F , and =Cov f[ , ] 0i t t t, 1F , where t 1F
is the lagged information set. The approximate factor structure
holds for the variance-covariance of the error terms, i.e.
= =Cov[ [ , ]]t n i t j t t i j n, , , , 1 , 1,...,F with bounded largest eigenvalue
(see, e.g., Chamberlain and Rothschild (1983)). The following
parameter restriction holds.13
=
=
a b ,i
k
K
i k k
1
,
(3)
where νk is a parameter defined for each k-th factor. The asset
pricing restriction in Eq. (3) can be rewritten as the usual linear
relation between expected excess returns and risk premia:
=
=
E R b[ ] .i t
k
K
i k k,
1
,
(4)
Based on Eqs. (3) and (4), the time-invariant risk premium associated
to each k-th factor is the following:
= +E f[ ] .k t k k, (5)
From Eq. (5), the risk premium λk is the sum of the expected
return on the factor, which can be estimated as its first moment, plus
the parameter νk, defined in the asset pricing restriction (3). Risk
premia measure how much investors are willing to pay to hedge the
systematic risk captured by the observable factors. When the factors
are asset returns themselves (i.e. factors are tradable) and are assumed
to be priced by the same model in (2), the risk premia are
equal to the factor means (see, e.g., Jagannathan and Wang (2002)).
However, if factors are non tradable, the parameter νk is non zero.
Following Gagliardini et al. (2016), we do not assume a priori that
the factors ft,k are tradable. Hence, we allow for the existence of
market imperfections, such as transactions costs due to rebalancing
and short selling, which are captured by ν (see, e.g., Cremers
et al. (2012)).
Our baseline factor models are summarized in the table below.
Model Reference Abbreviation Factors K
Four-factor
model
Carhart (1997) CAR fm,t, fsmb,t, fhml,t, fmom,t 4
Three-factor
model
Fama and
French (1993)
3FF fm,t, fsmb,t, fhml,t 3
Capital Asset
Pricing Model
Sharpe (1964);
Lintner (1965)
CAPM fm,t 1
The factors that are included in the models are the following: (1)
fm,t is the market factor, defined as the excess return on the European
value-weighted market portfolio over the risk free rate; (2) fsmb,t is
the size factor, defined as the average return on small caps minus the
average return on big caps; (3) fhml,t is the book-to-market factor,
Fig. 1. Total number of companies for which E score (yellow bar) and emission intensity (gray bar) are available.
12
Source: https://ec.europa.eu/eurostat/data/database.
13
We refer to Gagliardini et al. (2016) for theoretical results and proofs.
L. Alessi, E. Ossola and R. Panzica Journal of Financial Stability 54 (2021) 100869
5
defined as the average return on the value portfolio (i.e. stocks that
have market value that is small relative to the book value) minus the
average return on the growth portfolio; (4) fmom,t is the momentum
factor, defined as the average of the returns for the winner portfolio,
based on past returns, minus the average of the returns for the loser
portfolio. Fama and French (2015) propose a five-factor model including
the three Fama-French factors plus profitability and investment
factors. However, we do not consider the five-factor model as
four factors are enough to explain excess returns in a time-invariant
specification, as shown by Gagliardini et al. (2019). By analogy with
the fm,t factor, which is constructed based on the T-bill, we proxy the
risk free rate with the 30-day T-bill beginning-of-month yield. The
time series of European factors and the risk free rate are available on
Kenneth French’s website.
5. Empirical analysis
In this section, we first compare greener and more transparent
portfolio and the brown portfolio based on the models introduced in
the previous section. Then, we propose an observable greenness and
transparency factor defined as the difference between the returns on
the greener and more transparent portfolio and those on the brown
portfolio. Finally, we estimate the greenium, i.e. the risk premium
associated to the greenness and transparency factor, using a set of
European individual stocks. Our sample spans from January 2006 to
August 2018, covering all individual stocks included in the STOXX
Europe Total Market Index (TMI) on August 2018.14
The STOXX
Europe TMI covers approximately 95% of the free float market capitalization
across 17 European countries.15
In principle, we could
estimate our model on all 20 K European listed firms. However,
enlarging the sample would only marginally increase its coverage in
terms of market capitalization, while it may jeopardize the results
owing to the quality of the information that we would feed into the
model. Indeed, the reliability of the data we use for our application
crucially depends on the quality of environmental disclosures of
European firms. On this matter, the NFRD imposes mandatory disclosures
only on larger firms (with more than 500 employees). To be
on the safe side, we construct our greenness and transparency factor
based on a sample which is more reliable in terms of data quality,
insofar as it is based on a market index, and still representative of
the market as a whole. We present an application on a much larger
sample in Section 6.
As in Fama and French (2008), we exclude financial firms (i.e.
companies classified in sectors with NACE code K or L). The final
dataset comprises n = 942 stocks. Stock returns and stock market
capitalization data are sourced from Bloomberg. The panel is unbalanced,
i.e. asset returns are not available for all firms at all dates.
To account for publication lags, in each given year we use environmental
disclosures for the previous reference year.
5.1. Portfolio analysis and the greenness and transparency factor
As described in Section 3, we distinguish between transparent
and non-transparent companies. The former belong to set T while
the latter belong to set cT . At each month t, we define the returns on
the transparent and non-transparent portfolios, i.e. r˜t and r˜t
c
, respectively,
as follows:
= =r w r r w r˜ , and ˜ ,t
i
i i t t
c
i
i i t, ,
cT T
where the weight is defined as =w MC MCi i t t i t, , , with MCi,t being
the market capitalization of stock i at month t. Focussing on transparent
firms, we study the returns on different portfolios characterized
by different shades of green and degrees of transparency.
In particular, we build portfolios of returns r˜t
q
corresponding to the
quintiles q = 1, . . . , 5 of the distribution of the greenness and
transparency indicator considering only firms belonging to T .16
The portfolio built on the fifth quintile includes top-ranked firms
in terms of environmental performance and disclosures and is labeled
‘greener and more transparent’ portfolio. The returns on this
portfolio are indicated as r˜t
g
.17
Focussing on non-transparent firms, we build the brown portfolio
by including companies in cT which are active in one or more
of the industries characterized by the highest emissions, as described
in Section 3. The returns on this portfolio are indicated as r˜t
b
.
Table 14 in the Appendix reports descriptive statistics for firms included
in portfolios characterized by various shades of green and
transparency. It shows that the various portfolios are comparable in
terms of average size of the companies and firms’ leverage, while
firms in the non-transparent and brown portfolio tend to have a
slightly better RoA compared to greener and more transparent firms.
Table 1 reports descriptive statistics for the returns on various
portfolios, namely the one including all transparent firms R˜, the
greener and more transparent portfolio R˜g
, the portfolio including all
non-transparent firms R˜c
, and the brown portfolio R˜b
. With respect
to the relative performance of the various types of firms, and looking
at the mean return, the non-transparent portfolio has outperformed
the others, followed by the brown and the transparent. A sounder
way to assess the performance of a portfolio is the Sharpe ratio,
which relates the mean performance to the standard deviation of the
returns on a portfolio. In terms of Sharpe ratio, the non-transparent
portfolio still outperforms the others, which have a similarly better
performance than the market. Neither the mean return nor the
Sharpe ratio are monotone in greenness and transparency, which is
explained by the fact that the environmental characterization of a
portfolio is only one of the determinants of its performance (see
below). Finally, the distribution of returns for all the portfolios is
characterized by excess kurtosis and negative skewness.
Taking a closer look at transparent firms, Table 15 in the Appendix
reports descriptive statistics for quintile portfolios 1–5, with
portfolio R˜5
being the same as portfolio R˜g
, i.e. the top green and
transparent portfolio, while portfolio R˜1
includes comparatively
higher emitting and less transparent firms, among those that do
some environmental disclosure. Also in this case, the average return
decreases as the level of the greenness and transparency indicator
increases, though not monotonically (see Ciciretti et al. (2020) for a
similar result based on an ESG characterization of firms). The same
result holds when looking at the Sharpe ratio.
We investigate the drivers of the excess returns for the portfolios
described above by considering the reference models described in
14
This allows us to avoid survivor bias in the data.
15
These are Austria, Belgium, Czech Republic, Denmark, Finland, France, Germany,
Ireland, Italy, Luxembourg, the Netherlands, Norway, Portugal, Spain, Sweden,
Switzerland, and the United Kingdom.
16
In the asset pricing literature there is no univocal choice on which percentiles of
the distribution one should use for the construction of the relevant portfolios. For
example, Ang et al. (2006) use quintiles, Baker and Wurgler (2006) use deciles, and
Ahern (2013) use terciles. In general, the choice of the relevant percentiles is the result
of a trade-off between the need to maximize heterogeneity between the two portfolios,
and the need to ensure a large enough portfolio size. In our case, the top decile
selects only a small number of stocks (as low as 13 in 2005), while terciles fail to
identify the top firms in terms of environmental performance and transparency. Robustness
checks using deciles and terciles are presented in Table 13 in the Appendix.
17
Building portfolios based on the percentiles of the distribution of some relevant
firm characteristic is standard in the asset pricing literature. It should be emphasized,
however, that this is a relative approach to the classification of firms. In particular, a
firm may cease to be included in the ‘greenest and more transparent’ portfolio if other
firms reduce their emission intensity or improve their environmental disclosures, or if
greener and/or more transparent firms are included in the sample, even if its environmental
performance and transparency are unchanged in absolute terms.
L. Alessi, E. Ossola and R. Panzica Journal of Financial Stability 54 (2021) 100869
6
Section 4. In particular, Table 2 reports the estimated factor loadings
for the Cahart model (CAR), the three-factor Fama-French model
(3FF), and the CAPM. Results are reported for the various portfolios
(see also Table 16 in the Appendix). Overall, results are in line with
the literature with respect to the market, size, value, and momentum
factors, indicating that the portfolios we analyze are rather standard
with respect to these dimensions. In particular, the estimated factor
loading for the market factor bˆm is positive and significant across all
models and portfolios. However, for the transparent as well as for
the greener and more transparent portfolios, the exposition to the
market factor is lower compared to the non-transparent and brown
portfolios. This means that more transparent and greener firms tend
to be less correlated with the market compared to more opaque and
browner firms. The performance of the various portfolios can also be
explained by the different loadings on the other factors. In particular,
the exposition with respect to the size factor, bˆsmb, enters with a
negative sign for the transparent and greener and more transparent
portfolios, on the one hand, and a positive sign for the non-transparent
and brown portfolios, on the other hand. This suggests that
greener and transparent firms correlate more with bigger firms,
while non-transparent and brown firms correlate more with smaller
firms. Indeed, based on Table 14 in the Appendix, firms in the
greener and more transparent portfolio exhibit a slightly larger
mean size as measured by total assets. As for the value factor bˆhml,
the estimated loading is always negative and significant, except for
the brown portfolio for which it is negative but non significant.
Negative loadings on the value factor might mean that the portfolios
include a comparatively larger share of firms with a lower book-tomarket
value. Considering the Carhart model, the coefficient on the
momentum factor is not significant, except for the transparent
portfolio. Looking at the explanatory power of the various models
with respect to the different portfolios, the adjusted R-squared is
lower for the brown portfolio based on all the models. Finally, the
intercept is positive and significant for all portfolios and models,
suggesting the existence of an omitted factor.
We define the greenness and transparency factor as the difference
between the monthly returns on the greener and more transparent
portfolio and those of the brown portfolio. Formally:
=f r r˜ ˜ .g t t
g
t
b
, (6)
Table 3 reports the descriptive statistics of the Fama-French observable
factors, the momentum and the greenness and transparency
factor. The table includes also the cross-correlation structure
among the factors. The greenness and transparency factor is comparable
to the other observable factors in terms of mean, standard
deviation, kurtosis and skewness. It is also generally only mildly
correlated with the other factors.
5.2. Asset pricing analysis
In this section we investigate whether the greenness and transparency
factor defined in Eq. (6) affects the cross-section of
European stock returns. Further, we test whether investors accept
lower (higher) compensation for holding environmentally friendly
stocks by searching for a negative (positive) risk premium, i.e. a
greenium. The excess return Ri,t follows the model in Eqs. (2) and (3).
In particular, we consider the same linear factor models used in the
previous section, adding the greenness and transparency factor fg
among the observable factors as follows:
Model Factors K
CAR + G fm,t, fsmb,t, fhml,t, fmom,t, fg,t 5
3FF + G fm,t, fsmb,t, fhml,t, fg,t 4
CAPM + G fm,t, fg,t 2
The risk premium associated with the greenness and transparency
factor is defined as follows:
= +E f[ ] .g g t g, (7)
In order to estimate the risk premia for the observable factors
using individual stocks, we follow the estimation procedure proposed
in Gagliardini et al. (2016). This procedure allows to deal with
unbalanced panels, hence allowing to estimate the model on individual
stocks rather than portfolios, and involves the following
steps. First, we estimate the linear factor model by using the Ordinary
Least Square (OLS) estimator. Second, we use the fitted residuals
to test whether the model is correctly specified. In particular,
we compute the diagnostic criterion proposed in Gagliardini et al.
(2019), which checks whether the error terms share at least one
Table 1
Descriptive statistics for the returns from January 2006 to August 2018 on various
portfolios, namely the transparent R˜, greener and more transparent R˜g
, nontransparent
R˜c
and brown R˜b
portfolios.
Portfolio Mean Std Kurt Skew Sharpe t-stat
R˜ 1.102 0.497 3.744 -0.391 0.204 2.522
R˜g 0.943 0.502 4.097 -0.593 0.188 2.315
R˜c 1.732 0.586 5.210 -0.632 0.296 3.643
R˜b 1.425 0.638 6.985 -0.909 0.224 2.754
Note: The table reports the monthly mean and standard deviation (Std), kurtosis
(Kurt) and skewness (Skew), the Sharpe ratio, and the t-stat for the null hypothesis
that the mean return is zero.
Table 2
Estimates of linear factor models on portfolio excess returns. The table gathers results
for transparent, green, non-transparent and brown portfolios considering the following
linear models: four-factor Carhart model (CAR), three-factor Fama-French
model (3FF) and the CAPM. Statistical significance at the 1% (***), 5% (**) and 10% (*)
levels, and the adjusted R-squared (Radj
2 ).
Portfolio R˜ Green R˜c Brown
CAR model
aˆ 0.005*** 0.004*** 0.011*** 0.007***
bˆm
0.953*** 0.945*** 1.061*** 1.112***
bˆsmb
-0.208*** -0.261*** 0.476*** 0.702***
bˆhml
-0.176*** -0.194*** -0.144** -0.141
bˆmom
0.056*** 0.046 -0.028 0.029
Radj
2 0.979 0.947 0.940 0.864
3FF model
aˆ 0.005*** 0.005*** 0.011*** 0.008***
bˆm
0.944*** 0.938*** 1.065*** 1.107***
bˆsmb
-0.213*** -0.264*** 0.478*** 0.700***
bˆhml
-0.212*** -0.224*** -0.126** -0.159
Radj
2 0.978 0.947 0.940 0.865
CAPM
aˆ 0.006*** 0.005*** 0.012*** 0.009***
bˆm
0.899*** 0.891*** 1.032*** 1.063***
Radj
2 0.966 0.931 0.916 0.822
Table 3
Descriptive statistics of the three Fama-French factors, momentum and greenness and
transparency factors. The table reports annualized mean return, standard deviation,
kurtosis and skewness, as well as the factor correlation matrix.
Factor Mean Std Kurt Skewn fm fsmb fhml fmom
fm 6.035 1.885 4.690 -0.642 1
fsmb 1.671 0.641 3.195 -0.129 -0.034 1
fhml -1.378 0.788 3.582 0.519 0.533 -0.062 1
fmom 9.398 1.313 19.610 -2.546 -0.439 -0.009 -0.506 1
fg -4.350 1.291 4.563 0.103 -0.224 -0.483 -0.206 0.268
L. Alessi, E. Ossola and R. Panzica Journal of Financial Stability 54 (2021) 100869
7
unobserved common factor. Based on our sample, the criterion does
not detect any common factor for the residuals, suggesting the validity
of the factor structure.18
Third, we compute the cross-sectional
estimator νk from Eq. (3) by Weighted Least Squares (WLS). Finally,
the estimate of the risk premium ˆk for each factor is given by the
sum of the expected return on the factor E[fk,t] and the estimate of ˆk.
Table 4 shows the estimated risk premia attached to the factors,
including the greenium, as well as the estimates for νk. Looking at
the first two columns of the table, almost all risk premia are significant
across the board, and have the expected signs. In particular,
the estimated risk premia for the market, size, value and momentum
factors are comparable with the results in Gagliardini et al. (2016)
and Chaieb et al. (2020). The estimated greenium is negative and
significant at the 1% level in all cases. A negative greenium indicates
that investors accept lower compensation, ceteris paribus, to hold
assets that correlate positively with the greenness and transparency
factor, i.e. greener and more environmentally transparent assets. The
mainstream interpretation of this result is based on the assumption
that investors only care about their portfolios’ future payoffs. Hence,
if they accept a lower remuneration to hold a certain type of assets,
this must be because by doing so they are hedging some risks. In this
specific case, holding greener and more transparent stocks constitutes
a hedging strategy against climate risk. In the case of more
environmentally friendly and transparent stocks, however, other
considerations may also play a role. In particular, as suggested
by Fama and French (2007), investors decisions may also be driven
by some ‘taste for assets’, unrelated to expected returns. In this
light, an emerging ‘taste for green’ in the market could also explain
our result.
The last two columns of Table 4 refer to ˆk. Focussing on the
greenness and transparency factor, ˆg is always negative and significant
at the 1% level for the CAR+G, and at the 5% for the 3FF+G and CAPM+G
models. For this component of the risk premium, the literature has
proposed an interpretation linked to market imperfections (see, e.g.
Daniel and Titman, (1997); Haugen and Baker, (1996)). With reference
to the greenium, our hypothesis is that νg could capture alternative
preferences of market participants, for example reflecting alternative
expectations on future states of the economy (see Black and McMillan,
(2006)). In other words, some of the information that market participants
have may not be fully captured based only on past returns. In this
context, the difference between the investors’ larger information set
and the smaller, backward-looking information set on which the model
is estimated could be reflected in νk. This may be true in particular in the
case of green and brown assets, in which case future perspectives may
play a comparatively more important role than for other categories of
assets.
In order to analyze the relative importance of environmental
disclosures and environmental performance, in Table 5 we report
results for the greenium, as defined in Eq. (7), obtained by tuning the
parameter γ, i.e. the weight attached to the emission intensity.19
In
particular, by imposing γ > 0.5, we construct an indicator where the
quality of a firm’s environmental disclosures has a lower weight
compared to its emission intensity. This version of the indicator
attaches a larger weight to hard data and quantitative information,
and a smaller weight to a transparency score which may also rely on
descriptive statements and high level disclosures. Conversely, γ < 0.5
attaches a lower weight to quantitative information and a higher
weight to the overall quality and completeness of environmental
disclosures. We also investigate the extreme cases for which γ = 0
and γ = 1, i.e. the cases for which the indicator Gi,y is based on the E
score Ei,y only, and on the emission intensity Ki,y only, respectively.
The first main result is that the greenium is negative and significant
at the 1% confidence level also for γ = 0.2 and γ = 0.8, showing
that our findings are not dependent on giving the same weight to the
two components of the greenness and transparency factor. In
broader terms, our approach is still valid if we assume that the
market gives more weight to emissions than to transparency, or the
other way around. However, the second important finding is that
both components matter. Indeed, the greenium is not, or only mildly,
significantly different from zero when γ assumes extreme values,
thereby neglecting one of the two components. This indicates that
only the combination of emission intensity and disclosure quality is
systematically priced by the market. In particular, an indicator based
on emissions only (γ = 1) is associated with a risk premium not significantly
different from zero, suggesting that investors do not look
at emissions only. At the same time, with γ = 0, i.e. when only the
completeness of environmental disclosures matters, the greenium is
still negative and significant, but only at the 10% level. As for ν, which
is linked to the existence of market frictions, the estimates ˆg are in
line with the results in Table 4 for γ = 0.2, 0.8, 1, while ˆg with γ = 0 is
not significantly different from zero.
Table 4
Estimated annualized premia ˆk and cross-sectional parameters ˆk for the factors in
the Carhart model and for the greenness and transparency factor. The confidence
intervals are reported at the 99% level. ***, ** and * denote significance at 1%, 5% and
10% levels, respectively.
CAR + G model
ˆm
10.659** ˆm 4.625***
(−4.913, 26.232) (3.876, 5.373)
ˆsmb
3.326** ˆsmb 1.655***
(−1.354, 8.006) (0.682, 2.627)
ˆhml
-4.582* ˆhml -3.203***
(−10.723, 1.560) (−4.510,−1.896)
ˆmom
8.986** ˆmom -0.412
( −1.463, 19.436) (−3.117, 2.293)
ˆg
-9.860*** ˆg -4.076***
(−17.017, −2.702) (−6.221, −1.931)
3FF + G model
ˆm
10.534* ˆm 4.499***
(−5.038, 26.106) (3.766, 5.231)
ˆsmb
2.634 ˆsmb 0.963***
(−2.046, 7.314) (0.007, 1.918)
ˆhml
-5.903** ˆhml -4.525***
(−12.045, 0.238) (−5.812, −3.238)
ˆg
-7.545*** ˆg -1.781**
(−14.722, −0.407) (−3.886, 0.325)
CAPM + G
ˆm
11.137* ˆm 5.102***
(−4.435, 26.708) (4.397, 5.807)
ˆg
-7.282*** ˆg -1.498**
(−14.440, −0.125) (−3.360, 0.364)
Table 5
Estimated annualized greenium ˆg and ˆg associated with the greenness and transparency
factor, based on the CAR+G model including five factors overall. Results are
reported for γ = 0, 0.2, 0.8, 1. ***, ** and * denote significance at 1%, 5% and 10%,
respectively.
γ = 0 γ = 0.2 γ = 0.5 γ = 0.8 γ = 1
ˆg
-5.461* -8.290*** -9.860*** -9.853*** -0.152
ˆg 0.574 -2.860*** -4.076*** -5.554*** -1.477**
18
For example, for the Carhart model plus the greenness and transparency factor,
the difference between the largest eigenvalue of the empirical cross-sectional covariance
matrix of the residuals ˆi t, and the penalization term is negative, pointing to
the absence of omitted factors.
19
We only report results for the risk premium and the cross-sectional parameter ν
associated with the greenness and transparency factor. Estimation results for the
other observable factors are available upon request. The results for these factors are in
line with the ones in Table 4.
L. Alessi, E. Ossola and R. Panzica Journal of Financial Stability 54 (2021) 100869
8
6. Robustness checks
In this section we provide a battery of robustness checks that are
summarized in Table 6. The exercises are combinations of the following
three dimensions: (i) the definition of the greenness and
transparency indicator (i.e. ‘functional form’); (ii) the reference
sample of individual stocks; and (iii) the definition of the greenness
and transparency factor.
The first set of exercises (Rob. 1.1−Rob 2.3) relate to the greenness
and transparency indicator as defined in Eq. (1). The second set of
exercises (Rob. 3.1−Rob. 4.3) involves a different specification for the
greenness and transparency indicator. In particular, we propose an
alternative functional form, where the two components of the indicator
are related to each other in the form of a ratio. In this specification,
we use the emission intensity and the E score as such,
while the benchmark specification involves the rankings based on
these two indicators. The alternative definition is thus as follows:
= =G
E
K
E
Sales
Emissions
*
*
*
* ,i y
i y
i y
i y
i y
,
,
,
,
, (8)
where E*i y, is the E score and K*i y, is the ratio of total GHG or CO2
emissions over sales. Unlike the indicator in Eq. (1), the indicator
based on Eq. (8) suffers from a non-linearity issue. In particular, a
variation in the denominator corresponds to a more than proportional
change in the greenness and transparency indicator. This
nonlinearity yields an unstable pattern for the greenness and
transparency indicator over time for some companies.
In robustness checks Rob.2.1–2.3 and Rob. 4.1–4.3, we expand our
sample to include all listed European companies which do some environmental
disclosure, i.e. have an E score larger than zero, and those
that do no disclosure and belong to brown sectors, as defined in
Section 3. By doing so, we more than double the size of the sample,
bringing it to 2154 stocks. However, as discussed in Section 5, enlarging
the sample to include mid and small caps may affect the quality
of the environmental information used to construct the greenness
and transparency indicator. For this reason, in this exercise we still
use the greenness and transparency factor as constructed on the
smaller, more reliable sample. In particular, the greener and more
transparent portfolio and the brown portfolio include the same firms
as described in Section 3, hence the factor is the same as in the
benchmark exercise described in Section 5. However, this factor is
used in Rob.2.1–2.3 and Rob. 4.1–4.3 to price a larger set of assets.
Formally, referring to Eq. (2), this set of robustness checks involves the
same factors ft,k as in the benchmark case, but a larger number of
stocks N > n.
Additionally, we check the robustness of the results with respect
to the definition of the greenness and transparency factor. In the
benchmark case (see Section 5.2) as well as in Rob. 2.1, 3.1 and 4.1,
we build a portfolio of brown firms selecting the ones belonging to
the highest emitting NACE economic sectors, among those that do no
environmental disclosure. However, one could argue that the NACE
classification is in some cases unsuitable for sustainability analysis.
Hence, we build an alternative greenness and transparency factor fg t,
1
based on the returns on the greener and more transparent portfolio
r˜t
g
, on the one hand, and those of the portfolio including all nontransparent
firms r˜t
c
, on the other. Formally,
=f r r˜ ˜ .g t t
g
t
c
,
1
Rob 1.1, 2.2, 3.2, and 4.2 perform the estimation of risk premia using
the greenness and transparency factor fg t,
1
.
Finally, we test yet another specification for the greenness and
transparency factor, only based on transparent firms. In particular,
we construct the greenness and transparency factor fg t,
2
as the difference
between the returns on the greener and more transparent
firms and the firms that do some environmental disclosure, but only
attain lower levels of greenness and transparency. The former set of
firms correspond to the greener and more transparent portfolio as
defined in Section 5, i.e. the one including firms in the top quintile of
the distribution of the greenness and transparency indicator. The
latter set of firms correspond to those in the lower quintile of the
distribution of the greenness and transparency indicator. Formally,
the greenness and transparency factor is constructed as follows:
=f r r˜ ˜ .g t t
g
t,
2 1
Also in this case, we test both specifications of the greenness and
transparency indicator on the two samples (see Rob. 1.2, 2.3, 3.3
and 4.3).
Table 6
Overview of robustness checks.
Functional form γ Size sample Factor
Rob. 1.1 Eq. (1) 0, 0.2, 0.5, 0.8, 1 946 fg t,
1
Rob. 1.2 Eq. (1) 0, 0.2, 0.5, 0.8, 1 946 fg t,
2
Rob. 2.1 Eq. (1) 0, 0.2, 0.5 0.8, 1 2154 fg,t
Rob. 2.2 Eq. (1) 0, 0.2, 0.5, 0.8, 1 2154 fg t,
1
Rob. 2.3 Eq. (1) 0, 0.2, 0.5, 0.8, 1 2154 fg t,
2
Rob. 3.1 Eq. (8) n.a. 946 fg,t
Rob. 3.2 Eq. (8) n.a. 946 fg t,
1
Rob. 3.3 Eq. (8) n.a. 946 fg t,
2
Rob. 4.1 Eq. (8) n.a. 2154 fg,t
Rob. 4.2 Eq. (8) n.a. 2154 fg t,
1
Rob. 4.3 Eq. (8) n.a. 2154 fg t,
2
Table 7
Results for robustness checks Rob. 1.1−Rob 2.3. The estimated annualized greenium ˆg
and the estimated ˆg are computed from the CAR + G model. The greenness and
transparency factor is computed based the indicator Gi,y, with γ = 0, 0.2, 0.5, 0.8, 1 and
is defined using the alternative specifications fg,t, fg t,
1
and fg t,
2
. Results in the upper
part of the table are based on the benchmark sample comprising 946 European individual
stocks, while results in the lower part of the table are based on a larger
sample comprising 2154 stocks. ***, ** and * denote significance at 1%, 5% and 10%
levels, respectively.
γ = 0 γ = 0.2 γ = 0.5 γ = 0.8 γ = 1
n = 946
Rob. 1.1: greenness and transparency factor fg t,
1
ˆg
1.308 -4.611** -6.238*** -7.199*** -2.757
ˆg 6.138*** 4.509*** 3.235*** 0.821 0.808
Rob. 1.2: greenness and transparency factor fg t,
2
ˆg
-3.222 -1.888 -4.746** -6.648*** 2.853
ˆg 1.534** -1.759*** -3.280*** -6.770*** 1.828***
n = 2154
Rob. 2.1: greenness and transparency factor fg,t
ˆg
-5.056* -3.699 -5.156* -4.260 3.159
ˆg 0.979 1.731** 0.628 0.071 1.833***
Rob. 2.2: greenness and transparency factor fg t,
1
ˆg
3.209 -3.448 -4.531** -5.127** -3.023
ˆg 8.039*** 5.671*** 4.942*** 2.893*** 0.542
Rob. 2.3: greenness and transparency factor fg t,
2
ˆg
-5.569** -2.781* -3.434* -4.364** 0.990
ˆg -0.813 -2.652*** -1.968*** -4.486*** -0.035
L. Alessi, E. Ossola and R. Panzica Journal of Financial Stability 54 (2021) 100869
9
Table 7 reports the estimated greenium ˆg and the estimated
coefficient ˆg for Rob. 1.1 - Rob. 2.3.20
The results are based on a
five-factor model including the factors in the Carhart model and the
greenness and transparency factor.
Rob. 1.1 and 1.2 use the alternative definitions of the greenness
and transparency factor. Building the factor by considering all nontransparent
firms instead of only brown firms (Rob. 1.1) still yields a
negative and significant greenium for all intermediate values of γ.
Building the factor by considering only transparent firms (Rob. 1.2)
yields a negative and significant greenium for γ = 0.5 and γ = 0.8, at
the 5% and 1% confidence level, respectively. Notice that the variation
in the size of the greenium in Rob. 1.1 and 1.2 compared to the
results in Table 5 is purely mechanical, due to the smaller expected
value of fg t,
1
and fg t,
2
compared to E[ fg,t].
Looking at the lower part of Table 7, Rob. 2.1–2.3 estimate the
model on the larger sample of European stocks. Considering the
benchmark definition for the factor and with γ = 0.5, the greenium
remains negative and significant at the 10% level. Considering the
two alternative definitions for the factor, the greenium remains
negative and significant at the 5% and 10% level for Rob. 2.2 and Rob.
2.3, respectively. Varying γ on this larger sample yield a negative
greenium in some instances, but results are generally less strong in
terms of significance. Notably, the specifications based on γ = 1, i.e.
using only on emission data, are associated with a non-significant
greenium on the larger sample as well.
Focussing on the columns corresponding to extreme values for γ,
i.e. γ = 0 and γ = 1, the greenium is not, or only mildly, significantly
different from zero in most exercises. This confirms that only the
combination of emission intensity and disclosure quality is generally
priced by the market. In no case an indicator based on emissions
only (γ = 1) is associated with a significant risk premium. When only
the completeness of environmental disclosures matters, the
greenium is still negative and significant in two cases, but only at the
5% level at best. This finding confirms the results in Table 5. Moreover,
estimates of ν remain significant in most cases and tend to lose
significance when the distribution of the factor loadings bˆi for the
greenness and transparency factor is characterized by a larger
standard deviation.
Finally, Table 8 shows results using the greenness and transparency
indicator defined in Eq. (9). The more unstable behavior of this
version of the indicator does not affect the results. The greenium
remains negative and highly significant with the benchmark factor
fg,t as well as with fg t,
1
and fg t,
2
(Rob 3.1–3.3). The greenium is still
negative on the larger sample considering all the three alternatives
for the construction of the factor. However, significance is lower in
Rob. 4.2 and Rob. 4.3, while the estimate in Rob. 4.1 is not significant.
Overall, the greenium remains negative and significant across the
majority of the several robustness checks. It tends to loses statistical
significance mostly in extreme cases, when we only consider one of
the two components of the synthetic greenness and transparency
indicator.
7. Climate stress test on actual holdings
Based on the estimates derived in the previous section, we carry
out a climate stress test on actual investors’ equity holdings. We
consider the various institutional sectors at the global level, as well
as European SIFIs in particular. The aim of a climate stress test is to
measure the exposure of investors to climate risk, in a scenario
where more stringent sustainability-oriented policies are
progressively implemented, with increasing pressure on comparatively
more carbon-intensive firms and sectors. In such a scenario,
the expected returns on greener stocks increase, as more sustainable
firms are able to distribute higher dividends, while the price of
brown stocks drops for the same reason. Notably, one of the first
areas where policy pressure may increase, and is in fact already
increasing in Europe, is that of environmental disclosures. Against
this background, firms that have already implemented suitable nonfinancial
accounting procedures and adopted more advanced environmental
reporting standards will be better off once the non-financial
disclosure regulation becomes more stringent. In other
words, the expected return on stocks of more environmentally
sustainable and transparent firms conditional to the implementation
of sustainability policies increases. Formally, this implies that the
return on the greenness and transparency factor in Eq. (6), which is
positively correlated with returns on greener and more transparent
stocks, increases.
We test the resilience of investors to climate risk by borrowing
data on equity exposures and the classification of economic sectors
into climate-policy-relevant sectors from Battiston et al. (2017).
Following the indication provided by the authors as supplementary
information in Table 3, we group individual stocks (see
Section 5) according to their associated NACE code. In particular,
we classify stocks in the following economic sectors: fossil fuels,
energy intensive activities, housing, utilities, transport, finance
and other. Table 1 in Battiston et al. (2017) provides aggregate
holdings into climate-policy-relevant sectors, as of 2015, for the
following institutional sectors: Individuals, Governments (GOV),
Non-Financial Companies (NFCs), Other Credit Institutions (OCIs),
Other Financial Services (OFSs), as well as the institutional
financial sectors as defined in the ESA classification, i.e. Banks,
Investment Funds (IFs), and Insurance and Pension Funds (IPFs).
Battiston et al. (2017) also classify equity holdings of individual
financial institutions by climate-policy-relevant sectors, obtaining
the share of their portfolio invested into each of these sectors.
Based on their data, we focus on European SIFIs, as identified by
the Financial Stability Board.
The equity portfolio of an investor j at time t is defined as
follows:
=
=
r r ,j t t,
1
7
,
(9)
where ωκ corresponds to the equity exposure to the climatepolicy-relevant
sector κ and rκ,t is the monthly average value
weighted portfolio return of sector κ.
Table 8
Results for robustness checks Rob. 3.1−Rob 4.3. The estimated annualized greenium ˆg
and the estimated ˆg are computed from the CAR + G model. The greenness and
transparency factor is computed based the indicator G*i y,
. The greenness and transparency
factor is defined using the alternative specifications fg,t, fg t,
1
and fg t,
2
. Results
in the upper part of the table are based on the benchmark sample comprising 946
European individual stocks, while results in the lower part of the table are based on a
larger sample comprising 2154 stocks. *, ** and *** denote significance at 10%, 5%, and
1% levels, respectively.
Factor fg,t fg t,
1
fg t,
2
n = 946 Rob. 3.1 Rob. 3.2 Rob. 3.3
ˆg
-8.873*** -6.217*** -5.692***
ˆg -4.313*** 2.031*** -5.303***
n = 2154 Rob. 4.1 Rob. 4.2 Rob. 4.3
ˆg
-4.330 -5.240** -3.992*
ˆg 0.229 3.008*** -3.603***
20
We only report results for the risk premium and the cross-sectional parameter ν
corresponding to the greenness and transparency factor. Estimation results for the
other observable factors are available upon request. The results for these factors are in
line with the ones for the benchmark case described in Section 5.2.
L. Alessi, E. Ossola and R. Panzica Journal of Financial Stability 54 (2021) 100869
10
For each institutional sector and individual bank j, we compute
the marginal expected shortfall (MES) introduced by Acharya et al.,
(2017). The MES is defined as the expected equity loss conditional on
a particular factor return taking a loss greater than Γ. In this application
we estimate the expected equity loss conditional on the
greenness and transparency factor return defined in Eq. (6) realizing
a gain greater than Γ, i.e. a scenario where greener and more
transparent stocks outperform brown stocks by more than a particular
threshold. Hence, we can write the MES as follows:
= < = >MES E r f E r f[ ] [ ].j t j t g t j t g t, , , , ,
We compute the MES considering the following three cases,
which are defined in terms of portfolio allocation:
• Baseline Case: the investors’ portfolio allocation is defined as in
Eq. (9) and reflects the actual allocation of institutional sectors
and financial institutions as in Battiston et al. (2017). The portfolio
share invested in each of the stocks included in on our
sample is derived accordingly.
• Case 1: the investors’ portfolio allocation is defined as
= + ++
=r r r rj t j t j t j t,
1
2 ,1 1,
1
2 ,1 2
7
, , , where the exposure to the
fossil fuel sector, characterized by the highest emissions, is reduced
by 50% compared to the baseline. At the same time, we
assume that investments are reallocated to greener and more
transparent stocks, defined as the stocks with a positive exposition
to the greenness and transparency factor.
• Case 2: the investors’ portfolio allocation is defined as
= =
+
r rj t j t, 1
7
, , , i.e. only greener and more transparent stocks, as
defined above, are included in the portfolio.
In all three cases, the MESj,t is computed w.r.t. the event fg,t > q0.95,
where q0.95 indicates the 95th
percentile of the distribution of the
greenness and transparency factor. This corresponds to an extreme,
but still plausible scenario.
Tables 9 and 10 report MES results for the institutional sectors at
the global level and for European SIFIs, respectively. The MES is
expressed both as percentage loss and in billions of US dollars.
Looking at Table 9, the average MES at the global level in the baseline
scenario, i.e. given the actual portfolio allocation in 2015, is estimated
to be −1.5%. The very limited variation across institutional
sectors and institutions indicates that no one would be immune. A
loss of −1.5% on equity portfolios globally corresponds to USD 387
bn. For comparison, this figure is close to the total disbursements
under the Troubled Asset Relief Program (TARP), through which the
US Government purchased or insured troubled assets between 2008
and 2014. Table 9 also shows what would happen should a global
portfolio reallocation take place. The figures obtained under Scenario
1 indicate that halving the exposure to carbon-intensive activities
would reduce the MES only marginally. Losses would be
avoided only under Scenario 2, characterized by a radical portfolio
reallocation.
Turning to Table 10, we estimate a loss of −1.6%, with individual
banks recording losses of up to −2.2% on their equity portfolio. An
average loss of −1.6% for European SIFIs corresponds to almost USD 7
bn. These figures are admittedly not breathtaking, and one might
argue that losses of this magnitude would be unlike to trigger a financial
crisis. However, this exercise only focuses on equity exposures,
and only on first-round losses. Actually, one should consider
that a scenario where greener assets very much outperform brown
ones would be rooted into a deep transformation of the economy as
a whole. The low-carbon transition, if implemented in a sudden and
disorderly manner, would be associated with stranded assets and
non-performing loans, on top of losses on the stock market, which
would also very much weigh on bank’s profits. In fact, only a
comparatively small fraction of the overall exposure of banks to
carbon-intensive economic sectors is due to banks’ equity holdings.
Moreover, the stress-testing literature has shown that second-round
effects, such as contagion dynamics, the devaluation of counterparties’
debt obligations, as well as the price impact of fire sales, may
considerably amplify first-round losses. In particular, works based on
network models show that second-round losses are comparable in
magnitude to first round losses (see Battiston et al. (2017) and
references therein). Having said that, our data is not granular enough
to estimate the amount of second-round losses that could add up to
first-round impacts in a scenario like the one we are considering.
Moreover, the MES-based stress-testing tool we propose is specifically
targeted to assess climate risk in the equity portfolio. However,
it’s worth recalling that the global financial crisis, culminating in
trillions of losses in world GDP, was triggered by writedowns on the
value of loans and securitized assets due to the US subprime crisis,
which for many banks amounted to just few billions.
8. Conclusions and further research
Based on European stocks, we provide evidence of the existence
of a pricing factor linked to climate risk and find that the greenium,
i.e. the associated risk premium, is negative and highly statistically
significant. This is a novel result in the asset pricing literature. We
obtain this result because we take greenness and environmental
transparency seriously, unlike studies based on publicly traded funds
or indices, which often market themselves as greener than they
actually are. To attenuate the problem of greenwashing, we construct
an index of greenness and environmental transparency at the
individual company level, which takes into account both the GHG
emission intensity of a company and the quality of its environmental
disclosure. Based on this index we identify the greener and more
transparent companies, while we select brown companies among
those that do no disclosure, and operate in brown sectors. These two
Table 9
The table reports the MES in percentage terms and in billions of dollars, for the three scenarios, for global institutional sectors. The MES is computed conditional to the event
fg,t > q0.95.
MES (%) MES (Bn $)
Baseline Scenario 1 Scenario 2 Baseline Scenario 1 Scenario 2
OCIs -1.592 -1.511 0.113 -8.236 -7.821 0.584
Governments -1.411 -1.259 -0.085 -8.169 -7.286 -0.493
Individuals -1.433 -1.383 0.245 -37.270 -35.964 6.375
Banks -1.495 -1.411 0.062 -40.864 -38.553 1.686
IPFs -1.434 -1.339 0.096 -46.529 -43.460 3.119
OFSs -1.447 -1.376 0.200 -50.261 -47.791 6.931
Non-Financial Companies -1.462 -1.355 0.095 -68.476 -63.444 4.469
Investment Funds -1.404 -1.323 0.211 -127.646 -120.310 19.194
Average and Total -1.460 -1.370 0.117 -387.451 -364.630 41.866
L. Alessi, E. Ossola and R. Panzica Journal of Financial Stability 54 (2021) 100869
11
portfolios are used to define the greenness and transparency factor.
The negative sign attached to the greenium indicates that investors
buy stacks of greener and more transparent firms accepting a ceteris
paribus lower return, as a hedging strategy to reduce their exposure
to climate risk. We show that alternative definitions of ‘green factor’
which only consider GHG emissions or transparency do not work as
well, if at all, in the pricing model, implying that investors take into
account both firms’ GHG emissions and the quality of their environmental
disclosures. Further research is warranted to identify
other dimensions that may be relevant in the definition of a
greenium in a stock price framework, as well as the investigation of a
possibly time-varying greenium and its drivers.
In the last part of the paper we use our model to price investors’
equity holdings. We find that their current portfolio allocation exposes
them to non-negligible losses in a severe but plausible scenario,
where greener and more transparent firms outperform brown
firms. We estimate that direct losses could amount to 1.5% of the
global equity exposure, and to USD 7 bn for European SIFIs overall.
By adding up second-round effects and losses on loans and other
assets, this figure could rapidly increase. We calculate that halving
investors’ exposure to carbon-intensive activities would decrease
losses only marginally, while only a radical portfolio reallocation
towards greener assets would ensure resilience. Based on our results,
and considering that we only focus on equity exposures, we
argue in favor of introducing climate stress tests for systemic financial
institutions to make sure that climate risk is suitably priced.
The approach we propose to assess climate risk in equity portfolios
is based on a simple and well-known methodology, the marginal
expected shortfall, and could easily become part of a broader stresstesting
exercise.
Given that the awareness of investors towards climate-related
issues has clearly increased in recent years, it would make sense to
estimate a model with a time-varying risk premium. As discussed,
this presents challenges in terms of estimation, and will be the object
of future research. Another interesting avenue relates to the
drivers of the greenium, and in particular to the hypothesis that its
negative sign could be driven by an increasing ‘taste for green’
among investors.
Acknowledgements
This paper is published as part of the Special Issue “Climate risks
and financial stability,” which was co-edited by Stefano Battiston
(University of Zurich and Ca' Foscari Univ. of Venice), Yannis
Dafermos (SOAS University of London), and Irene Monasterolo
(Vienna University of Economics and Business) and was kindly
supported by the Joint Research Centre (JRC) of the European
Commission, the Journal of Financial Stability, and the Center for
Research in Contemporary Finance and the Gabelli School of
Business at Fordham University. We thank the co-editors and two
anonymous referees for constructive criticism and numerous suggestions
which have lead to substantial improvements over the
previous versions. We thank the authors of Battiston et al. (2017) for
providing us with their data, which we use for the climate stress test.
We are grateful to Ivan Faiella for his comments on the construction
of the greenness and transparency indicator. We thank participants
at a number of conferences, including CefES, IRMC, Finance for
Sustainability, CREDIT, EAEPE, as well as at meetings and workshops
at the European Central Bank, and seminars at the Universidad
Autonoma de Madrid, at the University of Pavia and at the EC Joint
Research Centre, for useful suggestions. Disclaimer: The content of
this article does not necessarily reflect the official opinion of the
European Commission. Responsibility for the information and views
expressed therein lies entirely with the authors. This research did
not receive any specific grant from funding agencies in the public,
commercial, or not-for-profit sectors.
Appendix: Additional tables and figures
Figs. 2 and 3 show the time-series of the greenness and transparency indicator Gi,y (top panels) for two representative companies, as well as
their ranking in terms of E score and emission intensity over time. The bottom panels plot the firms’ E score over time, as well as their emission
intensity. NIBE Industrier AB develops solutions for smart heating and intelligent control in industry and infrastructure. International Airlines
Group is the largest airline group globally. The greenness and transparency indicator for the two companies differs by three orders of magnitude.
For NIBE Industrier AB, the quality and quantity of disclosures have improved over time, together with a slightly decreasing emission
intensity, both resulting in an increasing value for the greenness and transparency indicator over time. International Airlines Group attained an
E score in 2011 which was comparable to that of NIBE Industrier AB in 2009. However, International Airlines Group’s disclosures only marginally
improved over time, if at all. With respect to emissions, they are obviously incomparably larger for airlines than for many other
companies.
Table 10
The table reports the MES in percentage terms and in billions of dollars, for the three scenarios, for European SIFIs. The MES is computed conditional to the event fg,t > q0.95.
MES (%) MES (Bn $)
Baseline Scenario 1 Scenario 2 Baseline Scenario 1 Scenario 2
DEUTSCHE BANK AG via its funds -1.455 -1.321 -0.032 -2.348 -2.131 -0.052
BPCE SA via its funds -1.590 -1.539 0.112 -2.325 -2.251 0.164
BNP PARIBAS via its funds -1.621 -1.518 -0.141 -1.090 -1.021 -0.095
UNICREDIT SPA via its funds -1.482 -1.415 0.145 -0.438 -0.418 0.043
BARCLAYS PLC via its funds -1.512 -1.394 -0.079 -0.572 -0.528 -0.030
CREDIT SUISSE GROUP AG via its funds -1.420 -1.325 0.158 -1.300 -1.212 0.145
BANCO SANTANDER SA -1.912 -1.904 -0.486 -0.155 -0.154 -0.039
UBS GROUP AG via its funds -1.432 -1.314 0.097 -2.604 -2.390 0.176
ING BANK NV -2.225 -2.049 -1.120 -0.042 -0.039 -0.021
SOCIETE GENERALE GESTION -1.571 -1.496 0.088 -0.771 -0.734 0.043
Average and Total -1.647 -1.552 -0.167 -6.971 -6.496 0.222
L. Alessi, E. Ossola and R. Panzica Journal of Financial Stability 54 (2021) 100869
12
Fig. 2. Greenness and transparency indicator GNIBE,y, E score and emissions intensity in ranked and row values, of the NIBE Industrier AB.
L. Alessi, E. Ossola and R. Panzica Journal of Financial Stability 54 (2021) 100869
13
Fig. 3. Greenness and transparency indicator GIntAir,y, E score and emissions intensity in ranked and raw values, of the International Airlines Group.
L. Alessi, E. Ossola and R. Panzica Journal of Financial Stability 54 (2021) 100869
14
Table 11
List of highest-emitting economic sectors in our sample. The last column reports the share of emissions associated to each sector.
Description NACE Rev 2 Emissions (%)
Code
Mining of coal and lignite B05
Extraction of crude petroleum and natural gas B06 2.08%
Mining of metal ores B07
Mining support service activities B09
Manufacture of coke and refined petroleum products C19 3.55%
Manufacture of chemicals and chemical products C20 3.77%
Manufacture of rubber and plastic products C22 10.23%
Manufacture of other non-metallic mineral products C23
Manufacture of basic metals C24 9.00%
Manufacture of fabricated metal products, except machinery and equipment C25
Electricity, gas, steam and air conditioning supply D35 28.82%
Sewerage E37
Waste collection, treatment and disposal activities; materials recovery E38 4.04%
Remediation activities and other waste management services E39
Land transport and transport via pipelines H49 4.98%
Water transport H50 2.96%
Air transport H51 3.53%
Total 72.96%
Note: Sector A01 “Crop and animal production, hunting and related service activities” is responsible for 11.91% of global emissions. However, there are no companies
belonging to this sector in our sample.
Source: Eurostat
Table 12
List of brown companies in 2017.
Company name NACE code Company name NACE code
Lubelski Wegiel Bogdanka SA B05 Ibstock PLC C23
Genel Energy Plc B06 Rhi Magnesita NV C23
Norwegian Energy Co ASA B06 Bossard Holding AG C25
BHP Billiton PLC B07 Beijer Alma AB C25
Grupa Kety SA B07 Indus Holding AG C25
Stalprodukt SA B07 Boryszew SA C25
Alumetal SA B07 Mennica Polska SA C25
Elkem ASA B07 SFS Group AG C25
Northern Drilling Ltd B09 Fiskars OYJ Abp C25
Bonheur ASA B09 Elektrobudowa SA D35
Borr Drilling Ltd B09 Kogeneracja D35
Shelf Drilling Ltd B09 BKW AG D35
Odfjell Drilling Ltd B09 Arendals Fossekompani A/S D35
EMS-Chemie Holding AG C20 Nobina AB H49
Ciech SA C20 PKP Cargo SA H49
Tikkurila Oyj C20 Dfds A/S H50
Tessenderlo Group SA C20 Fjord1 ASA H50
Recticel SA C22 Ocean Yield ASA H50
Sanok Rubber Co SA C22 Frontline Ltd/Bermuda H50
Forbo Holding AG C22 Wizz Air Holdings Plc H51
Vidrala SA C23
L. Alessi, E. Ossola and R. Panzica Journal of Financial Stability 54 (2021) 100869
15
Table 13
Estimated annualized risk premia ˆk and the cross-sectional estimator ˆk for the factors in the Carhart model and the greenness and transparency factor fg,t as
in Eq. (6). The factor fg,t is built based on the deciles and terciles of the distribution of the greenness and transparency indicator. Confidence intervals are
reported at the 99% level. ***, ** and * denote significance at 1%, 5% and 10% levels, respectively.
Panel A: greener and more transparent portfolio includes top decile firms
ˆm
10.592* ˆm 4.557***
(−4.982, 26.164) (3.805, 5.309)
ˆsmb
3.306* ˆsmb 1.635***
(−1.375, 7.986) (0.65, 2.62)
ˆhml
-4.805** ˆhml -3.427***
(−10.947, 1.337) (−4.745, −2.109)
ˆmom
9.203** ˆmom -0.196
(−1.247, 19.653) (−2.881, 2.491)
ˆg
-10.479*** ˆg -4.948***
(−17.298, −3.661) (−7.187, −2.708)
Panel B: greener and more transparent portfolio includes top tercile firms
ˆm
10.669* ˆm 4.634***
(−4.904, 26.241) (3.886, 5.383)
ˆsmb
3.471* ˆsmb 1.8***
(−1.21, 8.151) (0.828, 2.772)
ˆhml
-4.635* ˆhml -3.257***
(−10.776, 1.507) (−4.56, −1.953)
ˆmom
8.246** ˆmom -1.152
(−2.204, 18.696) (−3.968, 1.664)
ˆg
-8.612*** ˆg -3.331***
(−15.787, −1.437) (−5.428, −1.234)
Table 14
Companies’ fundamentals (year 2017). The table reports descriptive statistics for size (measured by the log of total assets), leverage and RoA, considering companies included in
the various portfolios.
Portfolio Size Leverage RoA
Mean Std Mean Std Mean Std
R˜1 9.683 1.013 25.933 16.168 6.619 5.790
R˜2 9.368 1.099 23.027 15.349 6.474 6.825
R˜3 9.440 1.258 22.208 16.330 6.685 6.741
R˜4 9.496 1.201 22.019 14.405 6.977 7.577
R˜g 9.513 1.184 23.466 15.666 5.526 6.585
R˜ 9.473 1.158 23.196 15.365 6.627 6.922
R˜c 9.368 1.201 24.018 16.372 7.187 8.208
R˜b 9.433 1.164 24.888 17.461 7.163 8.077
Table 15
Descriptive statistics for portfolios including transparent firms. The table reports the mean and standard deviation (Std), kurtosis (Kurt) and skewness (Skew), the Sharpe ratio,
t-stat for the null hypothesis that the mean return is zero. Portfolio R˜5
is the top green and transparent.
Portfolio Mean Std Kurt Skew Sharpe t-stat
R˜1
1.065 0.503 3.579 -0.099 0.212 2.611
R˜2
1.160 0.547 5.257 -0.684 0.212 2.617
R˜3
0.786 0.530 4.175 -0.391 0.148 1.827
R˜4
0.920 0.489 3.445 -0.232 0.188 2.317
R˜5 0.943 0.502 4.097 -0.593 0.188 2.315
L. Alessi, E. Ossola and R. Panzica Journal of Financial Stability 54 (2021) 100869
16
References
Ahern, K.R., 2013, Network centrality and the cross section of stock returns. USC
Marshall Working Paper.
Acharya, V., Pederson, L., Philippon, T., Richardson, M., 2017. Measuring systemic risk.
Rev. Financ. Stud. 30 (1), 2–47.
Alok, S., Kumar, N., Wermers, R., 2020. Do fund managers misestimate climatic disaster
risk. Rev. Financ. Stud. 33 (3), 1146–1183.
Ambec, S., Lanoie, P., 2008. Does it pay to be green? A systematic overview. Acad.
Manag. Perspect. 22, 45–62.
Andersson, M., Bolton, P., Samama, F., 2016. Hedging climate risk. Financ. Anal. J. 72
(3), 13–32.
Ang, A., Hodrick, R., Xing, Y., Zhang, X., 2006. The cross-section of volatility and expected
returns. J. Financ. 61, 259–299.
Baker, M., Wurgler, J., 2006. Investor sentiment and the cross-section of stock returns.
J. Financ. 4, 1645–1680.
Battiston, S., Mandel, A., Monasterolo, I., Schütze, F., Visentin, G., 2017. A climate
stress-test of the financial system. Nat. Clim. Change 7 (4), 283–288.
Battiston, S. and Monasterolo, I., 2018, A carbon risk assessment of central banks’
portfolios under 2∘
C aligned climate scenarios. Working Paper.
Berg, F., Koelbel, J., and Rigobon, R., 2020, Aggregate confusion: the divergence of ESG
ratings. Working Paper.
Bernardini, E., Giampaolo, J.D., Faiella, I., Poli, R., 2019. The impact of carbon risk on
stock returns: evidence from the European electric utilities. J. Sustain. Financ.
Invest. 0 (0), 1–26.
Black, A., McMillan, D., 2006. Asymmetric risk premium in value and growth stocks.
Int. Rev. Financ. Anal. 15 (3), 237–246.
Bolton, P. and Kacperczyk, M., 2020, Do investors care about carbon risk? Working
Paper.
Bragdon, J.H., Marlin, J., 1972. Is pollution profitable? Risk Manag. 19 (4), 9–18.
Campiglio, E., Godin, A., and Kemp-Benedict, E., 2017, Networks of stranded assets: a
case for a balance sheet approach. AFD Research Papers.
Carhart, M., 1997. On persistence of mutual fund performance. J. Financ. 52 (1), 57–82.
Carney, M., 2015, Breaking the tragedy of the horizon–climate change and financial
stability. Speech given at Lloydas of London, September, 29.
Chaieb, I., Langlois, H., and Scaillet, O., 2020, Factors and risk premia in individual
international stock returns. Forthcoming in Journal of Financial Economics.
Chamberlain, G., Rothschild, M., 1983. Arbitrage, factor structure, and mean-variance
analysis on large asset markets. Econometrica 51 (5), 1281–1304.
Chatterji, A.K., Durand, R., Levine, D.I., Touboul, S., 2016. Do ratings of firms converge?
Implications for managers, investors and strategy researchers. Strateg. Manag. J.
37 (8), 1597–1614.
Choi, D., Gao, Z., Jiang, W., 2020. Attention to global warming. Rev. Financ. Stud. 33 (3),
1112–1145.
Ciciretti, R., Dalo, A., and Dam, L., 2020, The contributions of betas versus characteristics
to the ESG premium. Working Paper.
Cremers, M., Petajisto, A., Zitzewitz, E., 2012. Should benchmark indices have alpha?
Revisiting performance evaluation. Crit. Financ. Rev. 2, 1–48.
Daniel, K.D., Litterman, R.B., and Wagner, G., 2016, Applying asset pricing theory to
calibrate the price of climate risk. Technical report, National Bureau of Economic
Research.
Daniel, K.D., Titman, S., 1997. Evidence on the characteristics of cross sectional variation
in stock returns. J. Financ. 53, 1–33.
Derwall, J., Guenster, N., Bauer, R., Koedijk, K., 2005. The eco-efficiency premium
puzzle. Financ. Anal. J. 61 (2), 51–63.
Donadelli, M., Jüppner, M., Riedel, M., Schlag, C., 2017. Temperature shocks and welfare
costs. J. Econ. Dyn. Control 82, 331–355.
Engle, R.F., Giglio, S., Kelly, B., Lee, H., Stroebel, J., 2020. Hedging climate change news.
Rev. Financ. Stud. 33 (3), 1184–1216.
Escrig-Olmedo, E., Fernandez-Izquierdo, M., Ferrero-Ferrero, I., Rivera-Lirio, J., MuñozTorres,
M., 2019. Rating the raters: evaluating how ESG rating agencies integrate
sustainability principles. Sustainability 11, 915.
Fama, E.F., French, K.R., 1993. Common risk factors in the returns on stocks and bonds.
J. Financ. Econ. 33 (1), 3–56.
Fama, E.F., French, K.R., 2007. Disagreement, tastes, and asset prices. J. Financ. Econ. 83
(3), 667–689.
Fama, E.F., French, K.R., 2008. Dissecting anomalies. J. Financ. 63 (4), 1653–1678.
Fama, E.F., French, K.R., 2015. A five-factor asset pricing model. J. Financ. Econ. 116 (1), 1–22.
Fatica, S., Panzica, R., Rancan, M., 2021. The pricing of green bonds: are financial institutions
special? Forthcoming in Journal of Financial Stability.
Gagliardini, P., Ossola, E., Scaillet, O., 2016. Time-varying risk premium in large crosssectional
equity datasets. Econometrica 84 (3), 985–1046.
Gagliardini, P., Ossola, E., Scaillet, O., 2019. A diagnostic criterion for approximate
factor structure. J. Econ. 212 (2), 503–521.
Goergen, M., Jacob, A., Nerlinger, M., Riordan, R., Rohleder, M., and Wilkens, M., 2019,
Carbon risk. Technical report, University of Augsburg, Queenas University.
Gore, A., 1993. Earth in the Balance: Ecology and the Human Spirit. Penguin, New
York.
Gros, D., Lane, P., Langfield, S., Matikainen, S., Pagano, M., Schoenmaker, D., and
Suarez, J., 2016, Too late, too sudden: transition to a low-carbon economy and
systemic risk. Technical report, European Systemic Risk Board.
Hartzmark, S.M., Sussman, A.B., 2019. Do investors value sustainability? A natural
experiment examining ranking and fund flows. J. Financ. 74 (6), 2789–2837.
Haugen, R., Baker, N., 1996. Commonality in the determinants of expected stock returns.
J. Financ. Econ. 41, 401–439.
Hoepner, A.G.F., Oikonomou, I., Sautner, Z., Starks, L.T., and Zhou, X., 2018, ESG
shareholder engagement and downside risk. AFA 2018 paper.
Hong, H., Li, F.W., Xu, J., 2019. Climate risks and market efficiency. J. Econ. 208 (1),
265–281.
Jagannathan, R., Wang, Z., 2002. Empirical evaluation of asset-pricing models: a
comparison of the SDF and beta methods. J. Financ. 57 (5), 2337–2367.
Larcker, D.F., Watts, E., 2020. Where’ s the Greenium? Journal of Accounting and
Economics 2020, 69.2–3.
Table 16
Estimates of linear factor models on various portfolios including transparent firms, with an increasing degree of greenness and transparency from 1 to 4. Results are
reported for the following models: four-factor Carhart model (CAR), three-factor Fama-French model (3FF) and the CAPM. ***, ** and * denote significance at 1%, 5%
and 10% levels, respectively.
Portfolio
R˜1
R˜2
R˜3
R˜4
CAR model
aˆ 0.005*** 0.007*** 0.003*** 0.004***
bˆm
0.925*** 1.004** 0.991*** 0.931***
bˆsmb
-0.119 -0.052 -0.202*** -0.255***
bˆhml
-0.103 -0.289*** -0.317*** -0.205***
bˆmom
0.144*** -0.057 -0.050 0.086***
Radj
2 0.878 0.916 0.941 0.938
3FF model
aˆ 0.006*** 0.006*** 0.003*** 0.005***
bˆm
0.902*** 1.014*** 0.999*** 0.916***
bˆsmb
-0.131 -0.048 -0.198*** -0.262***
bˆhml
-0.194** -0.252*** -0.285*** -0.260***
Radj
2 0.87 0.915 0.940 0.935
CAPM
aˆ 0.006*** 0.007*** 0.003** 0.005***
bˆm
0.860*** 0.958*** 0.938*** 0.861***
Radj
2 0.864 0.908 0.926 0.917
L. Alessi, E. Ossola and R. Panzica Journal of Financial Stability 54 (2021) 100869
17
Lintner, J., 1965. The valuation of risk assets and the selection of risky investments in
stock portfolios and capital budgets. Rev. Econ. Stat. 47 (1), 13–37.
Marquis, C., Toffel, M.W., Zhou, Y., 2016. Scrutiny, norms, and selective disclosure: a
global study of greenwashing. Organ. Sci. 27 (2), 483–504.
Monasterolo, I., De Angelis, L., 2020. Blind to carbon risk? an analysis of stock market
reaction to the Paris Agreement. Ecol. Econ. 170, 106571.
Monasterolo, I., Zheng, J.I., Battiston, S., 2018. Climate transition risk and development
finance: a carbon risk assessment of chinaas overseas energy portfolios. China
World Econ. 26 (6), 116–142.
Oikonomou, I., Brooks, C., Pavelin, S., 2012. The impact of corporate social performance
on financial risk and utility: a longitudinal analysis. Financ. Manag. 41 (2),
483–515.
Pindyck, R.S., 2013. Climate change policy: what do the models tell us? J. Econ. Lit. 51
(3), 860–872.
Porter, M., 1991. America’s green strategy. Sci. Am. 264 (4), 168.
Porter, M., van der Linde, C., 1995. Toward a new conception of the environmentcompetitiveness
relationship. J. Econ. Perspect. 9 (4), 97–118.
Ramelli, S., Wagner, A., Zeckhauser, R., and Ziegler, A., 2019, Investor rewards to climate
responsibility: Evidence from the 2016 climate policy shock.Swiss Finance
Institute Research, 18–63, National Bureau of Economic Research, Inc.
Renneboog, L., terHorst, J.R., and Zhang, C., 2007, Socially responsible investments:
methodology, risk exposure and performance. TILEC Discussion Paper No.
2007–013.
Seitz, J., 2010. Nachhaltige investments: eine empirisch-vergleichende analyse der
performance ethisch-nachhaltiger investmentfonds in europa. Diplomica Verlag,
Hamburg.
Sharpe, W., 1964. Capital asset prices: a theory of market equilibrium under conditions
of risk. J. Financ. 19 (3), 425–442.
Short, J.L., Toffel, M.W., 2008. Coerced confessions: self-policing in the shadow of the
regulator. J. Law, Econ., Organ. 24 (1), 45–71.
Statman, M., 2000. Socially responsible mutual funds (corrected). Financ. Anal. J. 56
(3), 30–39.
Trinks, A., Scholtens, B., Mulder, M., Dam, L., 2018. Fossil fuel divestment and portfolio
performance. Ecol. Econ. 146, 740–748.
L. Alessi, E. Ossola and R. Panzica Journal of Financial Stability 54 (2021) 100869
18