Climate Impact Investing

Climate Impact Investing
Tiziano De Angelis∗
University of Turin, Collegio Carlo Alberto
Peter Tankov†
CREST – ENSAE, Institut Polytechnique de Paris
Olivier David Zerbib†‡
ISFA, CREST – ENSAE, Institut Polytechnique de Paris
June 25, 2021
Abstract
This paper shows how green investing spurs companies to mitigate their carbon emissions by
raising the cost of capital of the most carbon-intensive companies. Companies’ emissions decrease
when the proportion of green investors and their sensitivity to climate externalities increase. We
show that the impact of green investors primarily governs companies’ long-term emissions. Compa-
nies are further incentivized to reduce their emissions when green investors anticipate tighter climate
regulations and climate-related technological innovations. However, heightened uncertainty regard-
ing future climate risks alleviates green investors’ pressure on the cost of capital of companies and
pushes them to increase their emissions. We provide empirical evidence supporting our results by
focusing on United States stocks between 2004 and 2018 and using green fund holdings as a proxy
for green investors’ beliefs. When the fraction of assets managed by green investors doubles, com-
panies’ carbon intensity drops by 4.9% over one year; when climate uncertainty doubles, companies’
carbon intensity increases by 6.7% the following year.
Keywords: Climate finance; socially responsible investing; ESG; impact investing.
JEL codes: G12, G11.
∗
School of Management and Economics, Dept. ESOMAS, Univeristy of Turin, C.so Unione Sovietica 218bis, 10134
Torino, Italy; Collegio Carlo Alberto, Piazza Arbarello 8, 10122 Torino, Italy. Email: [email protected]
†
CREST – ENSAE, Institut Polytechnique de Paris, 5, avenue Henry Le Chatelier, 91120 Palaiseau, France.
Email: [email protected]
‡
Université Claude Bernard - Lyon 1, Institut de Science Financière et d’Assurances, 50 Avenue Tony Garnier,
F-69007 Lyon, France. Email: [email protected]
§
This paper previously circulated under the title ”Environmental Impact Investing.” We thank Marco Ceccarelli,
Patricia Crifo, Joost Driessen, Caroline Flammer, Jesse Grabowski, Ying Jiao, Sonia Jimenez Garces, Rüdiger Kiesel,
Frank de Jong, Lionel Melin, Martin Oehmke, Christian Robert, Bert Scholtens, Luca Taschini, Dimitri Vayanos,
as well as participants at the Bachelier Finance Society One World Seminar, the seminairs of CREST - Ecole Poly-
technique, the University of Zurich, the Universität Duisburg-Essen, the University of Edinburgh, the 2020 GRASFI
Annual Meeting and the 2020 PRI Academic Network Conference, for their valuable comments and suggestions.
We thank ISS for kindly providing us with data on shareholder proposals. We gratefully acknowledge the financial
support from the Europlace Institute of Finance. T. De Angelis was also supported by EPSRC Grant EP/R021201/1.
Electronic copy available at: https://ssrn.com/abstract=3562534

1 Introduction
Figure 1. Percentage of sustainable investments and average carbon intensity of the

AMEX, NASDAQ, and NYSE stocks. This figure presents the evolution of the proportion of
sustainable investing relative to total managed assets over time, according to the Global Sustainable
Investment Alliance (2018), as compared to the average carbon intensity of AMEX, NASDAQ
and NYSE companies provided by S&P-Trucost between 2014 and 2018. The carbon intensity
corresponds to the direct (scope 1 and 2) and indirect (upstream scope 3) greenhouse gas emissions
of the companies, expressed in tCO2e per million dollars of revenue generated.
From 2014 to 2018, sustainable investments grew from 18% to 26% of the total assets under man-
agement (AUM) in the United States (U.S.) (US SIF, 2018) while, over the same period, the average
carbon intensity of the companies listed on the National Association of Securities Dealers Auto-
mated Quotations (NASDAQ), American Stock Exchange (AMEX), and New York Stock Exchange
(NYSE) decreased from 140 tons of CO2 equivalent per million dollars of revenue (tCO2e/USDmn)
to 100 tCO2e/USDmn (Figure 1).1 The downward trend in corporate greenhouse gas intensity may
be driven by several factors, such as the reductions in the costs of green technologies, tighter cli-
mate regulations, consumer pressure for more sustainable practices, and pressure exerted by green
1
The carbon intensity of a company is defined as its emission rate relative to its revenue over one year. This
metric is expressed in terms of tons of equivalent carbon dioxide per million dollars.

investors.2 The two main channels through which green investors can have an impact on companies’
practices are portfolio screening and shareholder engagement. Through portfolio climate screening,
by underweighing or excluding the most carbon-intensive3 companies from their investment scope,
green investors increase these companies’ cost of capital (Pastor, Stambaugh, and Taylor, 2021;
Pedersen, Fitzgibbons, and Pomorski, 2021; Zerbib, 2020) and can push them to reform. We focus
on the specific channel of climate screening (referred to as green investing hereinafter) and address
the issue of impact investing by answering the following questions: does green investing push com-
panies to reduce their greenhouse gas emissions? If so, what are the factors that lead companies to
mitigate their emissions? And how do these factors affect the dynamics of companies’ emissions?
We show that the development of green investing—both in terms of the proportion of AUM
and the sensitivity to climate externalities of green investors—pushes companies to reduce their
greenhouse gas emissions by raising their cost of capital. By internalizing the negative impact of
green investors on their financial valuation, companies are incentivized to pay a price to mitigate
their emissions by adopting less carbon-intensive technologies, thereby lowering their cost of capital.
These incentives are further strengthened when investors anticipate tighter climate regulations,
climate-related technological innovations, and when they account for the negative impact of other
companies’ emissions on the company under consideration. However, in a sufficiently large or
diversified market, investors’ uncertainty regarding future climate risks reduces the incentives for
companies to mitigate their emissions.
We develop a dynamic equilibrium model populated by heterogeneous investors and companies.
We model two different groups of investors with constant absolute risk aversion (CARA investors).
Both groups determine their optimal allocation by maximizing their expected wealth at a given
terminal date, but they differ in their climate beliefs. Of the two groups of investors, one is a group
2
Green investing is a form of socially responsible investing aimed at contributing to environmental objectives,
mostly reducing greenhouse gas emissions by internalizing climate externalities.
3
We refer to carbon-intensive companies and companies with high greenhouse gas emissions interchangeably since
carbon dioxide is the main gas contributing to global warming. In the United States (U.S.), it accounted for more
than 80% of the total emissions in 2018: https://www.epa.gov/ghgemissions/overview-greenhouse-gases.

of green investors and the other one of regular investors. Green investors internalize the expected
financial impact of future climate externalities of companies in which they invest, while regular
investors do not. In the first version of our model, green investors internalize deterministic climate
externalities, which can be positive or negative. These externalities reflect a company’s exposure to
various climate transition risks, such as the rise in carbon price (Jakob and Hilaire, 2015), physical
risks, such as the deterioration of the production fleet due to an increase in the frequency and
intensity of natural disasters (Arnell and Gosling, 2016), or litigation risks (Hunter and Salzman,
2007).
These investors invest in n companies, each with a different marginal cost of reducing green-
house gas emissions (referred to as marginal abatement cost hereinafter). Corporate managers are
stock-value optimizers, who balance the benefit of mitigating greenhouse emissions, thus attract-
ing green investors, against the cost of implementing these reforms. To represent the fact that a
company reforms its environmental practices over a long period of time, at the initial date, t = 0,
each company chooses a deterministic greenhouse gas emission schedule up to a final date T to
maximize its expected discounted future market value. We allow companies to have their own cli-
mate beliefs. In addition, each company accounts for the strategies adopted by all other companies,
hence reducing the companies’ problem to a nonzero-sum game. This framework notably differs
from standard heterogeneous belief models because the choice of each company’s emission schedule
directly affects the parameter on which investors disagree—companies’ climate externalities.
We obtain a tractable formula of the equilibrium asset prices and show that they include an
externality premium. This premium increases with the future financial impact of climate external-
ities (for simplicity, referred to as climate externalities hereinafter) internalized by green investors,
which can be either positive or negative, and scales in proportion to the relative wealth of green
investors. Therefore, all else being equal, the asset price of a carbon-intensive company (also re-
ferred to as a brown company hereinafter) will be lower than that of a company with a low carbon

footprint (also referred to as a green company hereinafter). Consequently, the expected returns
increase when the climate externalities are negative and decrease when they are positive.
We characterize companies’ optimal emission schedules in a general setup and derive an explicit
solution for the case when climate externalities are measured as a decreasing quadratic function of
the company’s emissions. In equilibrium, emissions decrease with the proportion of assets managed
by green investors, their sensitivity to climate externalities (also referred to as climate sensitivity
hereinafter), as well as companies’ climate sensitivity, but increase with the marginal abatement
cost. In particular, we show that investors’ climate sensitivity mainly drives long-term emissions
whereas companies’ climate sensitivity predominantly drives short-term emissions. Therefore, when
the average climate sensitivity of investors is lower than that of the companies, optimal emissions
decrease over time. In addition, corporate emissions are a convex function of time, with the degree
of convexity increasing in the rate of time preference. We calibrate the model on the AMEX,
NASDAQ, and NYSE stocks between 2004 and 2018 using the carbon intensity of companies
as a proxy for their emissions. We then simulate emissions’ mitigation in several scenarios by
considering an electrical equipment manufacturing company. Such a company reduces its emissions
by an average of 1% per year over a 20-year period when green investments account for 25% of the
total AUM in the economy. When either green investments account for 50% of the AUM or when
green investors’ climate sensitivity doubles, this company reduces its emissions by an average of
3% per year over the same period.
We also show that tighter climate regulations as well as climate-related technological innova-
tions, when anticipated and internalized by green investors, increase the pressure on companies to
further reduce their emissions. Using previous research, we recalibrate the marginal abatement cost
and the climate sensitivities accounting for the effects of technological change and expected regu-
latory action, respectively. Our model estimates that emissions decline 2.2 times faster when green
investors anticipate regulatory tightening. We also find that emissions decline 3.6 times faster when

they anticipate climate-related technological change. Finally, when both changes are anticipated,
declines are 4 times faster. In addition, we illustrate the effect of strategic interactions between
companies by showing that when green investors internalize the negative financial impact of the
economy’s average emissions, companies are further incentivized to curb their emission schedules.
We next extend our model to the case where green investors also internalize uncertainty about
future climate externalities. Climate risks, such as a rise in carbon price or the occurrence of
natural disasters, usually have non-Gaussian fat-tailed distributions (Weitzman, 2009; Barnett,
Brock, and Hansen, 2020). Therefore, we model future climate risks internalized by green investors
as a stochastic jump process (Poisson process). We give a tractable expression of optimal portfolio
allocations, asset prices, expected returns and emission schedule in equilibrium. We show that, in a
sufficiently large or diversified market, uncertainty about future climate risks leads green investors
to lower the risk of their portfolios by reducing their tilt towards green assets, shifting portfolio
allocations away from green assets and into brown assets. Consequently, climate uncertainty reduces
the magnitude of the externality premium on expected returns. As a result, the incentive for
companies to reform is also reduced, leading them to increase their optimal emission schedule in
equilibrium.
We support our results with empirical evidence. We estimate the emission schedule equation for
the AMEX-, NASDAQ- and NYSE-listed companies in the Centre for Research in Security Prices
(CRSP) database between 2004 and 2018. We approximate the emissions of companies using
their carbon intensities from the S&P–Trucost database. As proxies for the proportion of wealth
held by green investors and their uncertainty about future climate risks, we use the proportion of
green mutual fund holdings in the investment universe and the variation of holdings across green
mutual funds, respectively, constructed from Bloomberg and FacstSet. We control for several key
variables that may have an impact on companies’ emissions through proxies for environmental policy
stringency, research and development (R&D) dedicated to renewable energy, business cycles, and

corporate engagement using the ISS shareholder proposal database. We show that the proportion
of wealth held by green investors and their confidence about future climate risks have a significant
negative impact on the companies’ carbon intensities: when the proportion of wealth held by green
investors doubles, the carbon intensity falls, on average, by 4.9% over a one-year horizon with a
95% confidence interval ranging from 4.1% to 5.6%; when climate uncertainty doubles, the carbon
intensity increases, on average, by 6.7% the following year with a 95% confidence interval ranging
from 4.1% to 9.4%.
The results of this paper are of interest to both investors and policymakers. For investors,
we identify three major implications. First, the findings suggest that investors can increase their
impact on companies by raising their environmental requirements, for example by restricting the
range of companies in which they invest or by significantly underweighing the most carbon-intensive
companies. Second, to increase the climate impact of their asset allocations, investors also have
a key role to play as shareholders in pressing companies to increase transparency about future
climate-related risks and raise their environmental standards. Third, impact investing is financially
beneficial if investors favor companies that are on a pathway towards reducing their climate foot-
prints. Investors can also benefit from financial gains by investing in green companies for which
information on their climate footprints is still poorly available.
From the public authorities’ viewpoint, the results of this paper have four implications. First,
they highlight their role in supporting the development of green investments. In particular, they
suggest policymakers should set rigorous standards for environmental impact assessments and dis-
closure to foster and increase impact investing. This is consistent with the recommendations of
the European Union High Level Expert Group on Sustainable Finance (2018) and the European
Commission (2018)’s Action Plan, which led to the recent development of a green taxonomy and
an official standard for green bonds. Second, these results emphasize the importance of access
to information regarding companies’ climate footprints, which enables green investors to internal-

ize climate externalities as accurately as possible, thereby maximizing their impact on companies.
Third, these results clarify the effect of climate regulations and their predictability on the adjust-
ment of investors’ beliefs: a tighter climate regulation is amplified by the adjustment of green
investors’ expectations, which increases their pressure on companies’ cost of capital, thereby forc-
ing them to further reduce their emissions. Fourth, our analysis shows the importance of low-cost
access to greener technological solutions as a lever for companies to mitigate their climate impact.
Specifically, industries for which green alternatives are limited, such as cement or aircraft, face a
structural barrier to which an increase in R&D is an essential response. In addition, enabling in-
vestors to better forecast and internalize the likelihood of future technological innovations increases
their impact on corporate emissions.
Related literature. This paper contributes to the emerging literature on asset pricing and im-
pact investing in sustainable finance. From an asset pricing perspective, we clarify the relationship
between the development of sustainable investing4 and asset returns. The empirical literature on
the effects of Environment, Social and Governance (ESG) integration on asset returns is mixed:
some authors highlight the negative impact of ESG performance on asset returns, while others
suggest a positive relationship or find no significant impact.5 Three recent papers by Pastor et al.
(2021), Pedersen et al. (2021) and Zerbib (2020) study this relationship using a single-period model
with investor disagreement. They show that the stock returns of the most carbon-intensive com-
panies are increased by a positive premium. We contribute to this literature by characterizing the
4
Sustainable investing can be motivated by pecuniary or non-pecuniary motives (Krüger, Sautner, and Starks,
2020). Riedl and Smeets (2017) and Hartzmark and Sussman (2020) highlight the positive effect of sustainable
preferences on sustainable fund flows. Pro-social and pro-environmental preferences also impact asset returns since
they induce an increase in the return on sin stocks (Hong and Kacperczyk, 2009), a decrease in the return on impact
funds (Barber, Morse, and Yasuda, 2021) and a decrease in the return on bonds (Baker, Bergstresser, Serafeim, and
Wurgler, 2018; Zerbib, 2019).
5
For negative impacts, see Brammer, Brooks, and Pavelin (2006), Renneboog, Ter Horst, and Zhang (2008),
Sharfman and Fernando (2008), ElGhoul, Guedhami, Kowk, and Mishra (2011), Chava (2014), Barber et al. (2021),
Bolton and Kacperczyk (2021) and Hsu, Li, and Tsou (2019). For positive impacts, see Derwall, Guenster, Bauer, and
Koedijk (2005), Statman and Glushkov (2009), Edmans (2011), Eccles, Ioannou, and Serafeim (2014), Krüger (2015)
and Statman and Glushkov (2016). Finally, Bauer, Koedijk, and Otten (2005), Galema, Plantinga, and Scholtens
(2008) and Trinks, Scholtens, Mulder, and Dam (2018) find no significant impact.

dynamic of asset prices and expected returns when green investors internalize non-Gaussian climate
uncertainty. We show that, in the presence of uncertainty about future climate risks, the expected
return gap between brown and green assets narrows as green investors diversify their exposure to
mitigate their risk.
We also contribute to the emerging literature on impact investing. In a seminal study, by
constructing a single-period model in which green investors have the ability to exclude the most
polluting companies, Heinkel, Kraus, and Zechner (2001) show that such companies are pushed to
reform because exclusionary screening negatively impacts their valuations. Chowdhry, Davies, and
Waters (2018) study the optimal contracting for a company that cannot commit to social objectives
and show that impact investors must hold a large enough financial claim to incentivize the company
to internalize social externalities. Oehmke and Opp (2019) develop a general equilibrium model and
show that, in addition to regular investors, sustainable investors enable a scale increase for clean
production by internalizing social costs. Landier and Lovo (2020) also build a general equilibrium
model where markets are subject to search friction. They show that the presence of an ESG fund
forces companies to partially internalize externalities. In contrast, by analyzing venture capital
funds, Barber et al. (2021) suggest instead that the pressure exerted by sustainable investors on
companies is tied to their willingness to pay to invest in impact funds. Finally, through an asset
pricing model, Pastor et al. (2021) show that green investors produce positive social impact by
shifting investment towards green firms and making firms greener. We also contribute to the liter-
ature on impact investing from an asset pricing perspective along four different avenues. First, we
characterize the dynamics of companies’ optimal emissions and show that increases in either the
share of green investors or their climate sensitivity pushes companies to cut their emissions by af-
fecting their cost of capital. Notably, long-term emission dynamics are governed by green investors’
beliefs, while short-term dynamics are driven by companies’ beliefs. Second, when investors inter-
nalize future climate-related regulations and technological innovations, they increase their pressure

on companies’ cost of capital, thereby incentivizing them to further reduce their emissions. Third,
we show that green investors’ uncertainty about future climate risks leads companies to increase
their emissions. Finally, we provide empirical evidence supporting our results by using green fund
holdings to approximate the proportion of wealth held by green investors and their uncertainty
about future climate risks. To the best of our knowledge, this is the first paper providing empirical
evidence of the impact of ESG integration on companies’ practices.
The remainder of this paper is structured as follows. Section 2 introduces an economy populated
by greenhouse gas emitting companies and investors with heterogeneous beliefs. Section 3 describes
the equilibrium pricing equations and companies’ emission schedules when green investors inter-
nalize deterministic climate externalities. Section 4 extends the model to non-Gaussian stochastic
climate externalities. Section 5 provides empirical evidence and describes the model calibration.
Section 6 concludes the paper. Proofs of the mathematical results are presented in detail in the
Appendix.
2 A simple economy with greenhouse gas emitting companies and
heterogeneous beliefs
We develop a simple model with heterogeneous beliefs in which climate externalities are internal-
ized by green investors in a deterministic way. We introduce the dynamics of the assets available in
the market and heterogeneous beliefs about climate externalities of three types of agents—a group
of regular investors, a group of green investors and n companies. We then present the investors’
and companies’ optimization programs.
2.1 Securities market
In this section, we consider a financial market consisting of n risky stocks and a risk-free
asset, which is assumed to be free of arbitrage and complete. The risk-free asset is in perfectly

elastic supply and we assume that the risk-free rate is zero without loss of generality.6 Each stock
i ∈ {1, . . . , n} is in positive net supply of one unit and is a claim on a single liquidating dividend
DTi at horizon T . The terminal dividend of the i-th asset, DTi , is broken down into three terms:
(i) the cost of an environmental reform, decided by the company at time t = 0 and implemented
between t = 0 and t = T , with an associated greenhouse gas emission schedule; (ii) the initial
dividend forecast at time t = 0 net of the cost of the environmental reform; (iii) the entire cash flow
news sequence between t = 0 and t = T . The values of the first two terms are public information at
time t = 0 and their sum is also referred to as the initial dividend forecast. We isolate the cost of
environmental reform from the initial dividend forecast because it is a parameter of interest in our
model. This cost is known at the initial state because a company’s decision to reduce its emissions
is usually made over a long period of time. For example, the transformation of a generating fleet
by an electric utility, or the development of a line of electric vehicles by a car manufacturer are the
result of long-term decisions known to the public well in advance of their completion. Therefore,
the vector of terminal dividends for all assets, DT ∈ Rn , reads
Z T Z T
DT = ct (ψt − ψb )dt + d
|{z} + σt dBt . (1)
|0 {z } Initial dividend forecast | 0 {z }
net of cost of reform
| Cost of reform {z } Cash flow news
Initial dividend forecast
In the first term of the above expression, representing the cost of an environmental reform, ψt is the
vector of greenhouse gas emissions per unit of time of the companies at date t.7 We refer to green-
house gas emissions for simplicity, but ψ can be seen as a measure of relative emissions compared to
a level of production (e.g., carbon intensity) or a sector average (e.g., avoided emissions), or more
n is the vector of initial (or baseline) emissions,
generally as a climate rating. The constant ψb ∈ R+
and ct is a diagonal matrix with elements (cit ), which correspond to the marginal abatement costs
6
As stressed by Atmaz and Basak (2018), the interest rate can be taken as exogenous since consumption occurs
only at time T , i.e., there is no intermediate consumption.
7 n n n
Formally, ψ ∈ F ([0, T ], R+ ), where F ([0, T ], R+ ) is the set of Borel-measurable functions of [0, T ] in R+ .
10

for each company. Thus, by reducing its emissions by x, the i-th company reduces its terminal
dividend by cit x per unit of time t. The process (ψti )t∈[0,T ] , which is also referred to as the emission
schedule of the i-th company, is determined at time t = 0: it is deterministic and does not depend
on the future cash flow news so as to reflect the long-term nature of companies’ reform strategies.
The second term of the expression, d, is a constant vector corresponding to the initial dividend
forecast under a reference probability measure net of the cost of the environmental reform. In the
third term, σt dBt (t ∈ [0, T ]) is the sequence of cash flow news, where (Bs )s∈[0,T ] is a standard
n-dimensional Brownian motion defined on a probability space (Ω, F, P) equipped with a filtration
(Fs )s∈[0,T ] . We refer to P as the reference probability measure and we will later introduce other
probability measures for the companies and the investors, which reflect their different beliefs. For
each s ∈ [0, T ], σs is a deterministic, n × n, invertible matrix.
Denoting by (pt )t∈[0,T ] the equilibrium assets price process in Rn , we assume pT = DT . We also
denote the dividend forecast under the reference probability measure at t ∈ [0, T ] by
Z T Z t
Dt = E[DT |Ft ] = ct (ψt − ψb )dt + d + σs dBs , (2)
0 0
and in particular,
Z T
D0 = E[DT |F0 ] = ct (ψt − ψb )dt + d.
0
This Gaussian continuous-time specification of the dividend dynamics is consistent with previous
literature on models with heterogeneous beliefs that study investors’ reaction to good and bad
news (Veronesi, 1999), excess confidence (Scheinkman and Xiong, 2003) and extrapolation bias
(Barberis, Greenwood, Jin, and Shleifer, 2015).8 We choose a setup with Gaussian dividends
and prices because we are after explicit formulae for equilibrium prices, which companies use to
endogenously determine their prospective greenhouse gas emissions.

8
Other articles on heterogeneous beliefs adopt this same setup in discrete time such as Hong and Stein (1999),
Barberis and Shleifer (2003) and Barberis, Greenwood, Jin, and Shleifer (2018).
11

2.2 Investors’ and companies’ beliefs
The market is populated by two types of investors, regular and green, who have different expec-
tations regarding companies’ future cash flow news. Regular investors only consider the information
related to the flow of financial news, in addition to the initial dividend forecast. Therefore, under
their probability measure Pr , conditional on the information in t, they account for the past cash
Rt
flow news, 0 σs dBs , which is known at time t, but the expectation of the future cash flow news,
RT
t σs dBs , is zero as Bs is a Brownian motion. Denoting by Ert this conditional expectation,
Ert (DT ) = Dt , (3)
that is, the dividend forecast of the regular investors coincides with the dividend forecast under
the reference probability measure. From the point of view of the properties of the cash flow news,
σs dBs , there is no difference between the measures P and Pr , and we can simply assume P = Pr .
However, it should be noted that P is a technical device and, as such, we make no assumptions
about the realistic nature of this measure. The expectations of regular investors are thus not
necessarily consistent with the realized events.
In contrast, green investors internalize the financial impact of the expected climate externalities
of the companies in which they invest. These climate externalities can be negative and correspond
to several types of risks, for example: climate transition risks related to a rise in carbon price (Jakob
and Hilaire, 2015; Battiston, Mandel, Monasterolo, Schutze, and Visentin, 2017) or a changes in
consumer practices (Welsch and Kühling, 2009); the exposure of a company to physical risks,
such as the impact of natural disasters on its infrastructure (Mendelsohn, Emanuel, Chonabayashi,
and Bakkensen, 2012; Arnell and Gosling, 2016); and litigation risks related to the company’s
climate impact (Hunter and Salzman, 2007). These externalities can also be positive and reflect,
for example, a company’s pioneering environmental positioning in an economic segment or its
12

limited exposure to physical risks. In our first model specification (Sections 2 and 3), we assume
that green investors have a perfect anticipation of the future climate externalities. As a result, in
addition to accounting for the cash flow news and the initial dividend forecast as regular investors
do, green investors internalize, under their probability measure, the financial impact of future
climate externalities at date t ∈ [0, T ]. The latter is expressed by
Z T
θs (ψs )ds. (4)
t
Here, θs (ψs ) ∈ Rn is the vector of the financial impact of climate externalities (also referred to as
climate externalities) at date s ∈ [0, T ]. Naturally, we assume that θsi is a decreasing function of
ψsi so that higher emissions of the i-th company correspond to stronger negative financial impact
on the i-th asset. However, in a general case, θsi depends on the emissions of all companies because
more global emissions increases transition risks (e.g., regulations, carbon price) and physical risks
(e.g., natural disasters) for each company. This is why θsi is a function of the whole vector ψs .
Moreover, θsi is a function of time to reflect growing climate sensitivity of green investors, or their
anticipation of stronger regulatory pressure. As a consequence, green investors internalize their
climate beliefs regarding the i-th company by attributing a fundamental value to the i-th stock at
RT RT
time t that is higher (if t θsi (ψs )ds is positive) or lower (if t θsi (ψs )ds is negative) than the value
of the dividend forecast (see Equation (2)). Denoting by Egt the expectation of the green investors
conditional on the information at time t, we have
Z T
Egt (DT ) = Dt + θs (ψs )ds, (5)
t
that is, the dividend forecast of the green investors equals the forecast under the reference measure
augmented by the financial impact of future climate externalities internalized by these investors.
It is worth noticing that the variable Dt is constructed from the actual realization of the past
13

cash flow news between 0 and t, which is a known quantity to investors at time t. However, from a
probabilistic point of view, the stochastic process Dt under the probability Pg includes an additional
drift θs (ψs )ds associated with the beliefs of green investors.9
Alongside the two types of investors, we also introduce the productive sector by modeling the
views of the companies about the n assets available on the market. As in Oehmke and Opp (2019),
corporate managers (referred to as companies hereinafter) also have subjective beliefs about the
financial impact of climate externalities on the dividend dynamics of each of the n companies. We
denote by θsc (ψs ) the vector of the climate externalities internalized by all companies. Denoting by
Ect the expectation of the companies conditional on the information in t, we have
Z T
Ect (DT ) = Dt + θsc (ψs )ds. (6)
t
Similarly to Equation (5), the dividend forecast of the companies equals the forecast under the
reference measure augmented by the financial impact of future climate externalities internalized by
the companies. Here again, the variable Dt is constructed from the actual realization of the past
cash flow news between 0 and t, which is known to market participants at time t. However, from
a probabilistic point of view, the dynamic of the stochastic process Dt under the probability Pc
includes an additional drift θsc (ψs ) associated with the companies’ beliefs.10
2.3 Investors’ preferences and optimization
Regular and green investors are assumed to have CARA preferences. Subject to their budget
constraints, investors maximize the expected exponential utility of their terminal wealth11 WT ,
9
Formally, the probability measure of green investors is constructed through a change
RT >
of measure. The Radon-
1 RT 2
Nikodym density that connects the two probability measures Pg and Pr is ZT R= e 0 λs dBs − 2 0 kλs kR ds , where
−1 T t
λt := σt θt (ψt ). Under P , the dynamics of the stochastic process Dt reads Dt = 0 cs (ψs − ψb )ds + d + 0 σs dBsg +
g
Rt g g
θ (ψs )ds, where B is a brownian motion under P .
0 s
10
In the same way as presented in footnote 9, the companies’ measure is constructed through a similar change of
measure where θt (ψt ) is replaced by θtc (ψt ).
11
As Atmaz and Basak (2018) point out, investors’ preferences are based on their wealth at the terminal date
rather than on intermediate dates, which would have led to endogenizing the interest rate in equilibrium.
14

which reads
jW j
Ej (1 − e−γ T ), γ j > 0, j ∈ {r, g},
where the superscripts r and g refer to the regular and green investors, respectively, and γ j s are
their absolute risk aversions. The wealth processes follow the dynamics
Z t Z t
Wtr = wr + (Nsr )> dps , Wtg = wg + (Nsg )> dps , (7)
0 0
where Ntr and Ntg are quantities of assets held by the regular and green investors, respectively, at
time t, and prices (pt )t∈[0,T ] are determined by the market clearing condition. The initial wealth
levels of regular and green investors are denoted by wr and wg , respectively, and the symbol >
stands for the transposition operator.
In what follows, we denote by γ ∗ the global risk aversion, defined by 1

γ∗ = 1
γr + 1
γg , and set
γr γg
α= γ r +γ g and 1 − α = γ r +γ g . To simplify the interpretation of the impact of green and regular
investors’ wealth on the variables in equilibrium, and without losing generality, we assume that
green and regular investors have equal relative risk aversions; that is, γ R = γ g wg = γ r wr , where
γ R denotes the relative risk aversion. In this case, α is the proportion of the green investors’ initial
wg wr
wealth at t = 0, and 1 − α is that of the regular investors; that is, α = wg +wr and 1 − α = wg +wr .
2.4 Companies’ utility and optimization
A company’s decision to implement reforms to mitigate its climate footprint is made over a
sufficiently long time horizon. Therefore, at t = 0, the i-th company chooses its emission schedule,
(ψti )t∈[0,T ] , up to the horizon T so as to maximize its future valuation. Consequently, in our setup, we
endogenize companies’ emissions through their market values: on the one hand, investors allocate
their wealth according to companies’ emission schedules,12 thereby impacting companies’ market
values; on the other hand, companies take into account their market values to determine their
12
Green investors internalize climate externalities and all investors account for the costs of environmental reforms.
15

emission schedules. We denote by ρ the rate of time preference and by ψ −i the vector of emission
schedules of companies other than the i-th company. The market value of the i-th company’s
asset at time t is denoted by pit (ψ i , ψ −i ) to reflect its dependence on the vector ψ of all companies’
emission schedules. The companies have a linear utility and risk neutral preferences (Lambrecht
and Myers, 2017; van Binsbergen and Opp, 2019). Therefore, at t = 0, the i-th company chooses
(ψti )t∈[0,T ] so as to maximize the following objective function:
Z T
−i
i i
J (ψ , ψ ) = E c
e−ρt pit (ψ i , ψ −i )dt . (8)
0
Maximizing the sum of the market values over the entire period is consistent with Pastor et al.
(2021) as well as recent studies on Chief Executive Officers’ (CEO) compensation plans. Larcker
and Tayan (2019), for example, report that “stock-based performance awards have replaced stock
options as the most prevalent form of equity-based pay.” In addition, CEOs are generally required
to hold their companies’ stocks. Managers are therefore directly interested in the valuation of
their company’s stock price at each date, which endogenizes the financial impact of the company’s
emission schedule. This optimization program is also in line with the approach of Heinkel et al.
(2001) in the context of a multi-period model where the company’s climate impact is endogenized.
The optimal emission schedule, ψ ∗ , corresponds to a Nash equilibrium in which each company
i ∈ {1, ..., n} determines its own emissions, ψ i,∗ , in t = 0, so that
J i (ψ ∗,i , ψ ∗,−i ) ≥ J i (ψ i , ψ ∗,−i ), for all ψ i ∈ F ([0, T ], R+ ). (9)
Table 1 summarizes the preferences and optimization programs of the different players and their
interactions in the economy we model.
16

Table 1 Summary of agents’ actions. This table summarizes the optimization programs of
each agent as well as their interactions between t = 0 and t = T .
Date Agent Choose Given

At t = 0 Companies Their deterministic emission - Their expected market value between
schedule, ψ, from 0 to T 0 and T
∀t ∈ [0, T ] Regular investors Their asset allocation, N r - The observed cash flow news between 0
and t, and the expected cash flow news
between t and T
- The cost of reducing companies’ emis-
sions between 0 and T
∀t ∈ [0, T ] Green investors Their asset allocation, N g - The observed cash flow news between 0
and t, and the expected cash flow news
between t and T
- The cost of reducing companies’ emis-
sions between 0 and T
- Companies’ emission schedules be-
tween t and T
3 Equilibrium in the presence of greenhouse gas emitting compa-
nies and heterogeneous beliefs
This section presents equilibrium asset prices and returns in the model developed in Section
2. The optimal portfolio allocations of regular and green investors are found explicitly. We derive
the optimal dynamics of companies’ emissions, which we render tractable by assuming that climate
externalities are quadratic. Finally, we analyze the effects of regulatory changes, technological
changes, and companies’ interactions on companies’ optimal emission schedules.
3.1 Equilibrium stock prices and returns
In equilibrium, investors choose their allocations, which maximize their expected utility. Equi-
librium prices are determined such that the market clears. Denoting Σt = σt> σt , and letting 1 be
the vector of ones of length n, Proposition 1 gives the equilibrium prices and allocations.
17

Proposition 1. Given an emission schedule (ψt )t∈[0,T ] , asset prices in equilibrium read
Z T
pt = Dt − µs ds with µt = γ ∗ Σt 1 − αθt (ψt ), (10)
t
where −αθt (ψt ) is the externality premium. The optimal number of shares for the regular and green
investors are

1 −1 1 −1
Ntr = (1 − α) 1 − g Σt θt (ψt ) and Ntg = α 1 + r Σt θt (ψt ) , (11)
γ γ
respectively.
The different beliefs of green investors introduce an externality premium, which is an additional
drift in the price dynamics. When future climate externalities are negative (i.e., the emissions are
high), the price is adjusted downward proportionally to the fraction of the initial wealth held by the
green investors, α. Conversely, when future externalities are positive (i.e., the emissions are low),
green investors bid up the price, which is adjusted upwards. However, the cost of environmental
reform arises in the dividend forecast, Dt (see Equation (2)), and impacts the price in the opposite
direction: a reduction (increase) in emissions thus lowers (increases) the price. Therefore, the net
effect of a change in emissions on the price depends on the intensity with which green investors
internalize climate externalities (through θti ) and the cost of reform (through cit ).
The effect of heterogeneous beliefs on climate externalities can also be analyzed in terms of
expected dollar returns (referred to as expected returns hereinafter), E(dpt ) = µt dt. Since θti is a
decreasing function of ψti , expected returns increase with companies’ emissions. The externality
premium on asset returns can be positive (θti (ψ) < 0) or negative (θti (ψ) > 0). This result is sup-
ported by extensive empirical evidence, including Renneboog et al. (2008), Sharfman and Fernando
(2008), Chava (2014), Barber et al. (2021), Bolton and Kacperczyk (2021) and Hsu et al. (2019).
It is also consistent with the theoretical works of Pastor et al. (2021), Pedersen et al. (2021) and
18

Zerbib (2020), who show, through a single-period model, that expected returns increase along with
a company’s climate impact as green investors require a higher cost of capital.
The number of shares purchased by investors is also adjusted by the climate externalities.
Green investors overweigh assets with the higher positive externalities and underweigh or short
assets with the higher negative externalities. Regular investors have a symmetrical allocation
by providing liquidity to green investors. This result is consistent with optimal allocations in
disagreement models where some investors have an optimistic market view and others a pessimistic
one (Osambela, 2015; Atmaz and Basak, 2018): the risk is transferred from pessimists to optimists
who increase their holding of the asset under consideration.
3.2 Equilibrium emission schedules
At the initial date, companies choose their optimal emission schedules by maximizing their
expected market values between times 0 and T .
Proposition 2. The i-th company’s optimal emission schedule, ψ i , given a vector ψ −i of all other
companies’ emissions, is the one that maximizes for all t ∈ [0, T ]
βtc θtc,i (ψt ) + αβt θti (ψt ) + cit ψti , (12)
where
e−ρt − e−ρT 1 − e−ρt
βtc = and βt = .
1 − e−ρT 1 − e−ρT
At each time t, the i-th company maximizes the sum of three terms. The first and second terms
measure the financial benefits associated to two climate externality premia: one endogenized by
the company (θtc,i (ψt )) and the other endogenized by the green investors (αθti (ψt )), both adjusted
by suitable time factors (βtc and βt , respectively). The third one, cit ψti , accounts for the financial
benefits obtained by not reducing its emissions. The optimal emission schedule of a company is a
19

trade-off between the positive effect of reducing its emissions—especially, the positive effect on its
cost of capital through αβt θti (ψt )—and the cost of reform to achieve the target emission schedule.13
Research in environmental economics consensually suggests the use of a convex specification to
model the economic damage associated with environmental risks (Dietz and Stern (2015); Burke,
Hsiang, and Miguel (2015); Burke, Davis, and Diffenbaugh (2018)). Following Nordhaus (2014),
who argues that “damages can be reasonably well approximated by a quadratic function of tempera-
ture change”, we use a quadratic climate damage function to model the economic impact associated
with climate change. We assume that climate externalities are quadratic in carbon emissions and
consider a first simple case where the climate externalities of a given company, θti , depend only on
its own emissions, ψti .14 In this case, Proposition 2 yields a simple solution detailed in Corollary 3.
κct 2
Corollary 3. Assuming θti (x) = κ0,t − κt 2
2 x and θtc,i (x) = κc0,t − 2 x , for x ≥ 0, where κt , κct ,
κ0,t and κc0,t are positive deterministic functions of time,15 the i-th company’s optimal emission
schedule reads
cit
ψt∗,i = (13)
βtc κct + αβt κt
Emissions decrease with respect to the proportion of wealth held by green investors, α, and
when green investors’ and companies’ sensitivities to climate externalities increase. Indeed, κt
and κct measure the sensitivity with which green investors and companies, respectively, internalize
13
The companies’ optimization problem in the reduced form of (12) is obtained by writing the equilibrium price
(10) under the companies’ probability measure Pc :
Z T Z T Z t Z t Z T Z T
pt = Dt − µs ds = ct (ψt − ψb )dt + d + σs dBsc + θsc (ψs )ds − γ ∗ Σs 1ds + α θs (ψs )ds,
t 0 0 0 t t
c c
where B is a Brownian motion under the measure P . Substituting this expression for the price into Equation (8)
leads to the result of Proposition 2 after an integration by parts.
14
The financial impact of climate externalities, θti , represents the opposite of a damage function (θti decreases as
carbon emissions increase), so it is concave. Other types of climate externalities functions can be considered, such as
an exponential function (Barnett et al., 2020). However, in the case of our model, the exponential specification does
not lead to tractable solutions.
15
For simplicity we assume that κt , κct , κ0,t and κc0,t are the same for all companies but the generalization to
different constants is straightforward.
20

climate externalities at time t (also referred to as climate sensitivity). Thus, green investors can
increase their impact on companies by raising their climate sensitivity, for example by restricting
the range of companies in which they invest or by significantly underweighing the most carbon-
intensive companies. It should be noted that even if the company does not internalize climate
externalities (κct = 0), green investors’ beliefs and the threat they pose to a company’s market
value are sufficient to prompt a company to reduce its climate impact. In such a case, the optimal
emission schedule simplifies to:
cit
ψt∗,i = . (14)
αβt κt
As expected, the emissions of the i-th company decrease when the marginal abatement cost,
cit , decreases. In the special case where the marginal abatement cost is zero, the company cuts its
emissions to zero. The marginal abatement cost is a company specific factor that plays an important
role in the greening dynamics of the economy. R&D in industries where green alternatives are still
limited (e.g., cement, aviation) is therefore a key tool to support and accelerate the ecological
transition.
Irrespective of technological and regulatory changes, which will be analyzed in Section 3.3,
the emission schedule is not necessarily constant over time. The effect of investors’ beliefs, ακt ,
t
which changes with βt ' T (for small ρ), grows over time. In the long term, this effect becomes
t
prominent over the effect of companies’ beliefs, κct , which changes with βtc ' 1 − T (for small
ρ) and fades over time. This dynamic is explained by the fact that the asset prices at time t
under the companies’ probability measure depend on companies’ climate beliefs between 0 and
Rt
t ( 0 θsc (ψs )ds) and on green investors’ climate beliefs between t and T through the externality
RT
premium (α t θs (ψs )ds) as detailed in footnote 13. Therefore, by maximizing the expectation of
the price’s integral between 0 and T , companies give more weight to their beliefs at the beginning of
the period and to green investors’ beliefs at the end of the period. Consequently, when the climate
21

sensitivity of companies is lower than the climate sensitivity of the average investor, κc < ακ,16 they
emit more at the beginning of the period and less at the end of the period because the pressure
exerted by investors is more pronounced over the long run. Conversely, the emission schedule
increases over time if companies have a higher climate sensitivity than that of the average investor.
In the limiting case where companies’ climate beliefs are equal to investors’ average climate beliefs,
κc = ακ, the emission schedule is constant over time. In short, companies’ climate sensitivity, κc ,
mainly drives short-term emissions, whereas green investors’ proportion of wealth, α, and their
climate sensitivity, κ, mainly drive long-term emissions.17 This result captures two main routes
available to green investors to impact corporate decisions. First, green investors have the ability
to reduce the long-term emission target of the companies in which they invest by internalizing
climate externalities in their investment decisions. Second, they can also contribute to reducing
companies’ short-term emissions by pushing them to internalize climate externalities, for example
through shareholder engagement (Dimson, Karakaş, and Li, 2015; Broccardo, Hart, and Zingales,
2020). Aligning corporate objectives with climate issues can be achieved via incentive mechanisms
in managers’ compensation schemes as suggested by Edmans, Gabaix, Sadzik, and Sannikov (2012),
Varas (2018) and Aggarwal, Dizon-Ross, and Zucker (2020).
Figure 2 shows several optimal emission schedules for an electrical equipment company as a
function of α, κ, κc , and ρ. The calibration is detailed in Section 5.2. It is reasonable to assume

16
ακ is the climate sensitivity of the average investor because green investors’ climate sensitivity, κ, is weighted by
their proportion of wealth, α, and regular investors have zero climate sensitivity.
17
Two special cases arise: when κc is zero (that is, companies do not internalize climate externalities), emissions
tend to infinity close to time t = 0 (because βt tends to 0); similarly, when ακ is zero (that is, there are no green
investors), emissions tend to infinity close to time t = T (because βtc tends to 0). This effect is due to the fact that, in
order to obtain a tractable and interpretable solution for the emission schedule, we have opted for a constant marginal
abatement cost for each date t, which is profitable to companies when their emissions are higher than the initial level,
ψb (for example, because they are exempt from the cost of infrastructure maintenance). Therefore, companies benefit
from letting their emissions grow when there are no incentives to reduce them (in t = 0 when κc is zero, and in t = T
when ακ is zero). In the paper, we focus on this simplified model because (i) the main objective is to analyze the
impact of green investors-induced incentives on companies and (ii) the model allows us to obtain tractable formulae
that explain the pressures exerted by investors and companies on greenhouse gas emissions. However, in the Internet
Appendix, we study a less tractable version of the model in which companies have zero marginal gain from letting
their emissions grow above the initial level, ψb . In that framework the emissions’ dynamics is similar to the one
observed here but for an important difference: companies never increase their emissions above the initial level, ψb ,
irrespective of whether α, κ or κc are zero.
22

(a) ψt with different values for α (b) ψt with different values for κ
(c) ψt with different values for κc (d) ψt with different values for ρ
Figure 2. Emission schedules. This figure shows the optimal emission schedules, ψt , according
to several values of the proportion of green investors (α, sub-figure (a)), the green investors’ climate
sensitivity (κ, sub-figure (b)), the companies’ climate sensitivity (κc , sub-figure (c)), and the rate of
time preference (ρ, sub-figure (d)). The parameters are calibrated according to the values estimated
in Section 5.2: ψb = 147, α = 0.25, ρ = 0.01, κ = 3 × 10−7 , κc = 6 × 10−8 , celec = 8 × 10−6 .
23

a lag in a company’s responses to pressure from investors. To represent this delay, we use as an
example the case where the company internalizes climate externalities with slightly less sensitivity
than the average investor, κc < ακ, and thus where emissions decrease over time. With these
parameter values, when 25% of total AUM are managed by green investors, the company reduces
its emissions by 1% per year on average. This drop reaches 3% per year on average when green
investments account for 50% of total AUM, or when green investors’ climate sensitivity doubles.
The decrease in emissions is convex in time because of the dynamics of the time factors βt and
βtc : when ρ is small, the emission schedule has a hyperbolic temporal dynamic of the form 1/t.
This convexity increases with the rate of time preference, ρ, which accelerates the substitution
effect between the impact of the company’s beliefs (through βtc ) and the impact of green investors’
beliefs (through βt ) on the optimal emission schedule.18 In the present example where the average
investor is more climate sensitive than the company, κc < ακ, green investors’ pressure on long-run
emissions occurs earlier and accelerates the emissions decline, thereby increasing the convexity of
the schedule. Therefore, in cases where executives have weak incentives to reduce their companies’
climate footprints, a strong preference for the present—for example, through short-term objectives
in compensation schemes (Bolton, Scheinkman, and Xiong, 2006; Marinovic and Varas, 2019)—
might mitigate their adverse impact on the companies’ optimal emission schedules. Conversely,
a company with low climate requirements and a low preference for the present will emit more
greenhouse gases at the optimum.19
As in Pastor et al. (2021), this model extends the work of Heinkel et al. (2001) by (i) endogenizing
the climate impacts of companies and (ii) allowing them to choose among a continuum of climate
impacts, in contrast to Heinkel et al. (2001) where companies reform in a binary way (from brown to
18
The Internet Appendix (Figure IA.1.) shows the dynamics of βtc and βt for different rates of time preference, ρ.
19
If companies optimize over an infinite horizon, the economic insights remain the same. Indeed, in equilibrium, the
optimal allocations, prices and emission schedules converge to a well-defined limit when T tends to +∞. Specifically,
regarding optimal emissions, the direct consequence of an infinite horizon optimization is that the effect of green
investors’ beliefs on climate externalities, which impact long-term emissions, substitutes more slowly for the effect of
companies’ beliefs. For example, for a rate of time preference of 0.01, the two effects have the same weight after 69
c
years (β69 = β69 = 0.5).
24

green). Compared to Pastor et al. (2021), in this first approach where externalities are deterministic,
we develop a dynamic model that allows us to characterize the dynamics of companies’ climate
footprints (Section 3.2) as well as to study the dynamic impact on companies’ climate footprints
of (i) technological changes (Section 3.3), (ii) regulatory changes (Section 3.3), and (iii) interaction
effects between companies’ emissions (Section 3.4).
3.3 Technological and regulatory changes
The anticipation of technological changes and more demanding climate regulations by green
investors can further push companies to reduce their climate footprints. We develop the analysis
of this mechanism in this section.
Technological changes. Technological changes allowing companies to reduce their climate foot-
prints can take three major forms: the use of new machines that improve energy efficiency, that is,
the ratio of energy use over emissions; end-of-pipe innovations, such as carbon capture technologies
for utilities, which reduce emissions without modifying the production process; and process inno-
vations, which offer alternative production processes reducing the use of fossil fuels. Although the
effect on the marginal abatement cost curve is not unequivocal (Amir, Germain, and Van Steen-
berghe, 2008; Bauman, Lee, and Seeley, 2008),20 technological breakthroughs generally induce a
decrease in the marginal abatement cost (Milliman and Prince, 1989; Palmer, Oates, and Portney,
1995; Jaffe, Newell, and Stavins, 2002).
Mekaroonreung and Johnson (2014) estimate the effect of technological change on nitrogen
oxides (NOx ) marginal abatement costs of U.S. coal power plants in 2000–2008 by analyzing 325
boilers operating in 134 bituminous coal power plants. They find that technological change reduced
the NOx marginal cost by 28.3% in 2000–2004 and 26.5% in 2004–2008. Based on the order of
20
The decrease in the marginal abatement cost is consensual for end-of-pipe innovations; for process innovations, the
decrease in the marginal abatement cost is favored by a strong substitutability between the two factors of production
(energy and capital).
25

magnitude of their result and using the parameters calibrated in Section 5.2, we simulate the effect
of technological changes that would reduce the marginal cost of carbon intensity abatement by 5%
per year (Figure 3). Such technological changes, when anticipated by companies and investors,
push companies to multiply the pace of emissions reduction by a factor of 3.6 (from 1% to 3.6%
per year on average). Compared to the situation where no technological change is anticipated, the
carbon intensity is reduced by 40% after 10 years and by 64% after 20 years.
Figure 3. Emission schedule with technological change. This figure shows the optimal
emission schedules, ψt , without technological change (Benchmark) and with technological change
for which the marginal abatement cost decreases by 5% per year. The parameters are calibrated
according to the values estimated in Section 5.2: ψb = 147, α = 0.25, ρ = 0.01, κ = 3 × 10−7 ,
κc = 6 × 10−8 , celec = 8 × 10−6 (Benchmark).
This result not only underscores the importance of R&D, particularly in sectors where marginal
abatement costs are high, but also the need to enable agents to forecast and internalize a likely path
of future technological change. Even if, by definition, the occurrence of climate-related innovations
is unpredictable, it is possible to anticipate a vigorous dynamic of technological change when
R&D is largely supported by public and private funding; the development of renewable energy
26

infrastructures over the last 20 years as well as ongoing research on energy storage21 or carbon
capture and storage22 are insightful examples.
Regulatory changes. Tightening climate regulations can take two main forms: the introduction
of more demanding standards or an increase in the price of carbon, whether through taxes or
pollution permits. For green investors, such regulatory changes raise the future financial risks
of the brownest companies, specifically the transition risks. Therefore, when investors anticipate
regulatory tightening, given the same level of emissions, ψt , they adjust their expected financial
impact of climate externalities, θt (ψt ), by increasing their climate sensitivity, κt . However, tighter
regulations can have an additional effect: as suggested by Porter and van der Linde (1995), they also
encourage companies to innovate and may lower the marginal abatement cost. Porter’s hypothesis
has been supported by empirical evidence, such as the introduction of sulfur emission standards
in India (Sugathan, Bhangale, Kansal, and Hulke, 2018) and a carbon emissions permit trading
system in China (Xian, Wang, Wei, and Huang, 2020).
To calibrate the effect of tighter regulation on the climate sensitivity of green investors, we use
the carbon price trajectory. Although carbon prices vary from USD 1/tCO2e to USD 119/tCO2e in
different jurisdictions, the High-Level Commission on Carbon Prices estimates that carbon prices
of USD 40-80/tCO2 by 2020 and USD 50-100/tCO2 by 2030 are required to reduce emissions in
line with the temperature goals of the Paris Agreement (World Bank, 2020). Consistent with this
trajectory, we therefore consider an increase of κt by 2% per year.
Figure 4 shows the effects of tighter regulation on companies’ optimal emission schedule, using
the parameters calibrated in Section 5.2. In addition, we also show the effect of a simultaneous
policy change and technological change by including a 5% annual drop in marginal abatement cost
along with the 2% annual increase in climate sensitivity.
When investors anticipate regulatory tightening, they push companies to multiply the rate of
21
https://ec.europa.eu/energy/sites/ener/files/documents/batstorm_d10_roadmap.pdf
22
https://www.energy.gov/fe/science-innovation/carbon-capture-and-storage-research/carbon-capture-rd
27

Figure 4. Emission schedule with regulatory change. This figure shows the optimal emission
schedules, ψt , without regulatory change (Benchmark) and with regulatory change for which (i)
green investors’ climate sensitivity, κt , increases by 2% per year, and (ii) green investors’ climate
sensitivity, κt , increases by 2% per year and the marginal abatement cost, celec , decreases by 5%
per year. The parameters are calibrated according to the values estimated in Section 5.2: ψb = 147,
α = 0.25, ρ = 0.01, κ = 3 × 10−7 (Benchmark), κc = 6 × 10−8 , celec = 8 × 10−6 (Benchmark).
28

emissions reduction by a factor of 2.2 (2.2% per year on average). This rate is multiplied by
four when investors and companies account for technological changes. Compared to the situation
without regulatory changes, the carbon intensity decreases by 11% (47%) after 10 years and by 33%
(76%) after 20 years without (with) technological change. Here again, the anticipation of future
regulatory tightening encourages companies to launch projects that emit less greenhouse gases and
to adopt a more ambitious emission schedule, partly as a result of the increased pressure exerted
by investors on their cost of capital. Therefore, by announcing plans for more stringent climate
standards or the future carbon price trajectory early enough, governments are sending a signal not
only to companies but also to investors, which further strengthens the climate impact they have
on companies.
3.4 Interaction effects
In Corollary 3 we have made the simplifying assumption that investors and companies internalize
the financial impact of the i-th company’s climate externalities as a function of its own emissions
only, θti (ψti ). However, a company can be financially impacted by the emissions of other companies
in the economy. For example, the risk of tightening climate regulations is likely to be greater if
the companies in the same geographical area have a large climate footprint. Proposition 4 shows
that when agents internalize the negative impact of the economy’s average emissions through an
interaction effect, companies decrease their optimal emission schedules compared to the case with
no such interaction.
Proposition 4. Let ψt∗,i = cit (βtc κct + αβt κt )−1 be the i-th company’s optimal emission schedule
Pn ∗,j
without interaction effect (Corollary 3), ψ̄t∗ = 1
n j=1 ψt the average optimal emissions of the n
companies without interaction effect, and ε ≥ 0 an elasticity parameter such that
κt h i 2 i κct h i 2 i
θti (ψ) = κ0,t − (ψt ) + ε ψti ψ̄t and θtc,i (ψ) = κc0,t − (ψt ) + ε ψti ψ̄t ,
2 2
29

1 Pn j
where ψ̄t = n j=1 ψt , and κ0,t , κc0,t , κt and κct are deterministic functions of time. Then, we have
the following results:
(i) Companies lower their optimal emissions compared to the case without interaction effect (that
is, when ε = 0), and the i-th company’s optimal emission schedule reads
2n2

ε
ψt∗,ε,i = ψt∗,i − ψt∗,i + ψ̄t∗ . (15)
2n + ε 2n + (n + 1)ε
(ii) When the number of companies is large and much larger than the elasticity parameter, i.e.,
n ε ≥ 0, the optimal emission schedule is approximated by
ε
ψt∗,ε,i ' ψt∗,i − ψ̄ ∗ . (16)
ε+2 t
In the above proposition, we show that when investors and companies internalize the negative
financial impact of the economy’s average emissions, companies are further incentivized to curb their
optimal emission schedules. Let us take a simple example where a company has an optimal carbon
intensity of 150 tCO2e per million dollars of revenue generated (tCO2e/USDmn), while the average
optimal carbon intensity of the n = 1000 companies in the economy is 100 tCO2e/USDmn in the
absence of interaction effect. If the agents internalize the impact of the economy’s average emissions
with an elasticity of ε = 0.5, the optimal emissions decrease to 130 tCO2e/USDmn, resulting in a
13% reduction. The downward adjustment of a company’s emission schedule in an economy with
ε
a sufficiently large number of companies is proportional to ε+2 times the average emissions of the
economy. Therefore, in the particular case where companies give equal weight to their emissions
and to the average emissions of the economy (ε = 1), they reduce their optimal emission schedules
by one third of the average emissions of the economy. As expected, in the absence of interaction
effect (ε = 0), the optimal emission schedule is maximal and it equals ψt∗,i = cit (βtc κct + αβt κt )−1 .
30

4 Equilibrium with climate uncertainty
We extend the model presented in Section 2 to the case where the climate externalities are
internalized by green investors as a non-Gaussian stochastic process. We characterize the opti-
mal emission schedule under the new setup, and we show that uncertainty about future climate
externalities reduces the incentive for companies to lower their emissions.
4.1 Climate uncertainty
The internalization of deterministic climate externalities is an imperfect approach. Barnett
et al. (2020) note that “given historical evidence alone it is likely to be challenging to extrapolate
climate impacts on a world scale to ranges which many economies have yet to experience. Both
richer dynamics and alternative nonlinearities may well be essential features of the damages that we
experience in the future due to global warming.” Indeed, climate risks are characterized by fat tails
(Weitzman, 2009, 2011) and abrupt changes beyond tipping points (Alley, Marotzke, Nordhaus,
Overpeck, Peteet, Pielke Jr., Pierrehumbert, Rhines, Stocker, and Wallace, 2003; Lontzek, Cai,
Judd, and Lenton, 2015; Cai, Judd, Lenton, Lontzek, and Narita, 2015) that will severely impact
the world economy (Dietz, 2011).
We therefore extend our model to the case where green investors internalize uncertainty about
climate-related financial risks. As climate-related financial risks are not Gaussian and occur in jerks
and turns, we describe the arrival of these risks by a Poisson process. On the same filtered prob-
ability space, (Ω, F, (Ft )t≥0 , P), we define a time-homogeneous Poisson process N := (Nt )t∈[0,T ]
(counter of the shocks), whose intensity is denoted by λ. To ensure comparability with the deter-
ministic case, we make the average effect of jumps in climate externalities equal to the average effect
of deterministic linear externalities over a given period: we denote by Ntλ := λ−1 Nt the normalized
version of N , so that E[Ntλ ] = t for any λ, i.e., t shocks occur on average over a period of length
t. Therefore, the only factor governing uncertainty is the intensity parameter, λ, which modulates
31

the strength of climate uncertainty. When λ is small, climate uncertainty is strong: the shocks are
rare but large; when λ is large, climate uncertainty is low: the shocks are frequent but small.
4.2 Investors’ and companies’ beliefs
As before, we assume that regular investors do not internalize the financial impact of climate
externalities. Therefore, according to regular investors beliefs, the vectors of terminal dividends,
DT , and dividend forecast, Ert (DT ), are still given by Equations (1) and (3), respectively. However,
in contrast to Sections 2 and 3, green investors internalize climate externalities by taking into
account their uncertainty. Although the expression of the expected terminal dividends, Egt (DT ),
given in Equation (6) continues to hold, it offers an average picture of green investors’ beliefs and,
therefore, does not allow us to specify the dynamics of jumps introduced in this section. To account
properly for the jump process, we need to slightly modify the definition of the terminal dividends
as perceived by the green investors. Specifically, green investors assume that the vector of terminal
dividends contains an additional factor that depends on the climate externalities of the companies,
θs (ψ s ), and the Poisson process, N λ :
Z T Z T Z T
DT = ct (ψt − ψb )dt + d + σt dBt + θs (ψs )dNsλ , (17)
0 0 0
where now B is a standard Brownian motion under all probability measures that we consider (that
is, under the reference measure P = Pr and under the measures of the green investors Pg and of
the companies Pc ).
This framework generalizes the first model where climate externalities were deterministic. In-
deed, when the intensity, λ, increases, the number of shocks increases and their size decreases; in
the limiting case where λ tends to +∞, the uncertainty disappears, and we recover the setting of
Section 2:
Z T Z T
lim θs (ψs )dNsλ = θs (ψs )ds.
λ→∞ 0 0
32

Like in the case where the externalities were deterministic, the variable Dt is constructed from
the actual realization of the past cash flow news between 0 and t, which is known at time t. However,
from a probabilistic point of view, the dynamic of the stochastic process Dt under green investors’
beliefs (probability measure Pg ) includes a jump process corresponding to the internalization of the
uncertainty about future climate externalities:
Z T Z t Z t
Dt = ct (ψt − ψb )dt + d + σt dBt + θs (ψs )dNsλ .
0 0 0
Similarly, under companies’ beliefs (probability measure Pc ) , we define the dynamics of the
terminal dividend using θsc (ψs ) as
Z T Z T Z T
DT = ct (ψt − ψb )dt + d + σt dBt + θsc (ψs )dNsλ ,
0 0 0
and the dynamic of the stochastic process Dt as
Z T Z t Z t
Dt = ct (ψt − ψb )dt + d + σt dBt + θsc (ψs )dNsλ .
0 0 0
4.3 Equilibrium stock prices and returns
The optimization framework and notation remain similar to those of the first model. The
equilibrium price process is denoted by (pt )t∈[0,T ] , and it is assumed that pT = DT . In equilibrium,
investors choose their allocation to maximize the expected exponential utility of their terminal
wealth (Equations (7)), and equilibrium prices are determined by the market clearing condition.
Proposition 5 gives the equilibrium price and allocations.
Proposition 5. Given an emission schedule (ψt )t∈[0,T ] , the asset price in equilibrium reads
Z T
pt = Dt − µs ds with µt = γ ∗ Σt 1 − αyλ θt (ψt ), (18)
t
33

and the optimal number of shares for the regular and green investors are

yλ yλ −1
Ntr,λ = (1 − α) 1 − g Σ−1 θt (ψt ) and Ntg,λ = α 1 + r Σt θt (ψt ) , (19)
γ t γ
respectively, where yλ > 0 is obtained by solving the one-dimensional equation
αγ g

> yλ > −1
yλ = exp − 1 θt (ψt ) + r θt (ψt ) Σt θt (ψt ) . (20)
λ γ
When green investors internalize uncertainty about climate externalities, an additional factor,
yλ , arises in the equilibrium allocations, Ntr,λ and Ntg,λ , the equilibrium price, pt , and in particular
the expected return, µt dt. Notice that yλ is determined at each time t by Equation (20). Thus,
it should be denoted yλ (t). We choose the simpler notation yλ as no confusion shall arise. The
equation for yλ admits a unique solution.23 Proposition 6 explains the effect of yλ on the optimal
allocations and expected returns depending on the properties of green investors’ optimal portfolio
without uncertainty, Ntg .

Proposition 6. Fix an emission schedule ψt . Recall that Ntg = α 1 + 1 −1
γ r Σt θt (ψt )
is the optimal

portfolio of green investors with deterministic climate externalities, and Ntg,λ = α 1 + yγλr Σ−1
t θt (ψt )
is the optimal portfolio of green investors when the uncertainty level of climate externalities equals
λ. Then,
(i) When the level of climate uncertainty increases, green investors decrease the market risk of
their portfolio, i.e., λ 7→ (Ntg,λ )> Σt Ntg,λ is increasing in λ.
(ii) If the green investors’ portfolio in the absence of climate uncertainty has positive climate ex-
ternalities, meaning that θt (ψt )> Ntg > 0, then, the function λ 7→ yλ is positive, monotonically
increasing and satisfies lim yλ = 1. Consequently, for a given level of emission schedule, ψt ,
λ→∞
the increase in climate uncertainty (λ decreases)

23
Existence and uniqueness of yλ are given as part of the proof of the next proposition.
34

– pushes green investors to reallocate their wealth towards brown assets;
– decreases (increases) the expected returns, µt dt, of the brown (green) companies.
(iii) If the green investors’ portfolio in the absence of climate uncertainty has negative climate ex-
ternalities, meaning that θt (ψt )> Ntg < 0, then, the function λ 7→ yλ is positive, monotonically
decreasing and satisfies lim yλ = 1. Consequently, for a given level of emission schedule, ψt ,
λ→∞
the increase in climate uncertainty (λ decreases)
– pushes green investors to reallocate their wealth towards green assets;
– increases (decreases) the expected returns, µt dt, of the brown (green) companies.
(iv) If the green investors’ portfolio in the absence of climate uncertainty has neutral climate
externalities, meaning that θt (ψt )> Ntg = 0, then yλ = 1 for all λ > 0. Consequently, green
investors’ portfolio and companies’ expected returns do not depend on the level of climate
uncertainty.
Uncertainty about future climate externalities is an additional source of risk for green investors
who, as a result, reduce the overall risk of their portfolio. However, the adjustment of their
asset allocation depends on the climate externalities of their optimal portfolio without uncertainty,
θt (ψt )> Ntg . Lemma 1 further elaborates on the case in which the green investors’ optimal portfolio
without climate uncertainty is green, that is, θt (ψt )> Ntg > 0.
Lemma 1. Fix an emission schedule ψt . Assume that the sum of the companies’ climate ex-
ternalities is not too negative, such that 1> θt (ψt ) > − γ1r θt (ψt )> Σ−1
t θt (ψt ). Then, the green in-
vestors’ portfolio in the absence of climate uncertainty has positive climate externalities, that is,
θt (ψt )> Ntg > 0.
The situation described by Lemma 1 is the most common.24 It requires that the sum of the
climate externalities of the companies in the market is greater than − γ1r θt (ψt )> Σ−1
t θt (ψt ), which is
24
The proof of the lemma is an immediate consequence of the expression for Ntg (see Equation (11)).
35

always negative. Typically, this occurs when the market is not excessively composed of brown assets,
for example, when it is sufficiently diversified in terms of companies’ climate externalities, that is,
when the market is formed by companies with positive and negative θti (ψt ). Moreover, the condition
is more easily verified for large markets because the quadratic correction, 1 > −1
γ r θt (ψt ) Σt θt (ψt ),
becomes larger. Pastor et al. (2021) assume that the ESG externalities of the market portfolio are
zero—here 1> θt (ψt ) = 0—which is a sufficient condition for Lemma 1 to be valid. Furthermore,
Zerbib (2020) validates the Pastor et al. (2021) hypothesis by estimating the climate externalities
of the market portfolio in the U.S. between 2007 and 2019 at a value close to zero.
In the main case described by Lemma 1 where the market is not excessively brown, green
investors are able to build a green portfolio in the absence of uncertainty. Therefore, yλ ∈ [0, 1[,
which means that uncertainty about future climate externalities pushes green investors to lower the
risk of their portfolio by reducing their exposure towards green assets. The size of this effect scales
with the degree of uncertainty: the larger the uncertainty, the more green investors decrease their
allocation in green assets and increase their allocation in brown assets. Consequently, for a given
level of emission schedule, ψt , climate uncertainty reduces the effect of the externality premium,
−αyλ θt (ψt ), on expected returns, which lowers the cost of capital of brown companies and increases
the cost of capital of green companies.
In less common cases, green investors fail to construct a green optimal portfolio in the absence
of uncertainty. This situation is possible, for example, when all or much of the economy is brown.
Therefore, yλ > 1, which indicates that the increase in climate uncertainty pushes green investors
to invest more in green assets and divest from brown assets to diversify their allocation and mitigate
their risk. Consequently, for a given level of emission schedule, ψt , climate uncertainty amplifies
the effect of the externality premium, −αyλ θt (ψt ), on expected returns, which reduces the cost of
capital of green companies and increases that of brown companies.
The following section presents the companies’ optimal emission schedules when they account
36

for the climate uncertainty internalized by green investors.
4.4 Equilibrium emission schedule
Companies fix their emission schedules at the initial date by maximizing their future market
values as in the deterministic case (Equation (8)). The situation here is more complex, however,
because of the appearance of the new parameter yλ in the price vector (see Proposition 5). Indeed,
for a fixed value of yλ , the optimal emissions schedule of the i-th company is the one that maximizes
for all t ∈ [0, T ] the expression:
βtc θtc,i (ψt ) + αβt yλ θti (ψt ) + cit ψti , (21)
by the same arguments as the ones used in Proposition 2 for deterministic externalities. However,
this shows that the equilibrium emissions schedule depends on the choice of yλ , which will affect
the investors’ portfolio allocations and asset prices in Proposition 5, creating a feedback effect. In
order to maintain tractability in our model we make an assumption of a large market that allows
us to partially decouple the optimization problems for investors and companies. Then we derive
equilibrium emission schedules in the next proposition.25
Proposition 7. Let the functions of climate externalities, θti and θtc,i , be defined as in Corollary 3.
Assuming that the market is large enough so that the aggregated climate externalities of the market
are not strongly affected by a change in the climate externalities of each company, the optimal
emission schedule for the i-th company is
cit
ψti,∗ (yλ∗ ) = , (22)
αβt κt yλ∗ + βtc κct
25
If the market is large enough, we can assume that the aggregate quantities are not strongly affected by each
individual company, and therefore, the i-th company may maximize its optimization functional by assuming that yλ is
constant. This assumption is standard, for example, in mean-field game optimization problems. At the mathematical
level, our assumption says that each company solves (21) for a fixed value of yλ , thus obtaining ψti (yλ ). Plugging
this expression back into (20), we arrive at a fixed point equation for yλ . This equation is still one-dimensional but
different from Equation (20) due to the additional dependence of ψt on yλ .
37

where yλ∗ is a solution of the fixed-point equation
αγ g >

αy > −1
y = exp − 1 θt (ψt (y)) + r θt (ψt (y)) Σt θt (ψt (y)) . (23)
λ γ
As climate uncertainty adds a multiplicative factor yλ to the externality premium, it also adds
a multiplicative factor yλ∗ to the company’s optimization program (Equation (12)) on the factor
driven by green investors’ beliefs, αβt yλ∗ θti (ψt ). As a result, the optimal emission schedule in the
presence of uncertainty (Equation (22)) is adjusted by yλ∗ in its denominator. Proposition 8 clarifies
the behavior of yλ∗ depending on λ.
Proposition 8. Under the same assumptions as in Proposition 7, the following holds true:
(i) Existence: For all λ ∈ (0, ∞), Equation (23) admits at least one positive solution.
(ii) Uniqueness: If, for all i ∈ {1, . . . , n}, and all t ∈ [0, T ],
n
X j j γr
inf (Σ−1
t )ij y θt (ψt (y)) ≥ − (24)
y>0 α
j=1
the solution of Equation (23) is unique. In particular, in the asymptotic regime where uncer-
tainty is low (λ → ∞), the solution is unique.
(iii) Assume that the condition (ii) is satisfied and denote by yλ∗ the unique solution of Equation
(23). Let ψt∗,0,i = cit (βtc κct + αβt κt )−1 be the equilibrium emission schedule without climate

∗,0
uncertainty (given by (13)), and Ntg = α 1 + γ1r Σ−1 t θ t (ψt ) be the green investors’ corre-
sponding optimal allocation without climate uncertainty (given by (11)).
– If the green investors’ portfolio in the absence of climate uncertainty has positive climate
externalities, meaning that θt (ψt∗,0 )> Ntg > 0, then, the function λ 7→ yλ∗ is positive,
monotonically increasing and satisfies lim yλ∗ = 1. Consequently, an increase in climate

λ→∞
uncertainty (λ decreases) increases the companies’ optimal emission schedule, ψt∗ , given
by (22).
38

– If the green investors’ portfolio in the absence of climate uncertainty has negative climate
externalities, meaning that θt (ψt∗,0 )> Ntg < 0, then, the function λ 7→ yλ∗ is positive,
monotonically decreasing and satisfies lim yλ∗ = 1. Consequently, an increase in climate

λ→∞
uncertainty (λ decreases) decreases the companies’ optimal emission schedule, ψt∗ , given
by (22).
– If the green investors’ portfolio in the absence of climate uncertainty has neutral climate
externalities, meaning that θt (ψt∗,0 )> Ntg = 0, then, the function λ 7→ yλ∗ is constant and
equal to 1. Consequently, climate uncertainty does not affect the companies’ optimal
emission schedule, ψt∗ , given by (22).
In the most common case described by Lemma 1 where green investors have a green optimal
portfolio in the absence of climate uncertainty, θt (ψt∗,0 )> Ntg > 0, increasing uncertainty (λ de-
creases) decreases yλ below 1. This effect dampens the contribution of green investors’ beliefs,
αβt yλ θti (ψt ), in the companies’ optimization program (Equation (12)). Indeed, by mitigating the
externality premium on expected returns (Proposition 6), green investors reduce the incentive for
companies to decrease their emissions at unchanged abatement cost, cit ψti . Therefore, all companies
increase their optimal emission schedules in the presence of climate uncertainty compared to the
situation without uncertainty.
In the case where green investors have a brown optimal portfolio in the absence of climate uncer-
tainty, θt (ψt∗,0 )> Ntg < 0, increasing climate uncertainty (λ decreases) increases yλ above 1 and thus
amplifies the contribution of green investors’ beliefs, αβt yλ θti (ψt ), in the companies’ optimization
program. By strengthening the incentive for companies to reduce their emissions, green investors
push companies to reduce their emissions compared to the situation without uncertainty.
Figure 5 shows optimal emission schedules according to different levels of uncertainty: the cases
where information on climate risks (or the materialization of climate risks) becomes available on
average annually, every five years, every ten years, and every twenty years are displayed. When
39

green investors internalize these levels of uncertainty, companies increase their emissions by 0.4%,
2%, 4%, and 8.5%, respectively, over a 20-year horizon compared to the case where climate risks
are perfectly known.
Figure 5. Emission schedule with uncertainty. This figure shows the optimal emission
schedules with uncertainty of an electrical equipment company with an initial carbon intensity
of ψbelec = 147 tCO2e/USDmn. This company operates in a market with two companies: the
second company is a coal company with an initial carbon intensity of ψbcoal = 555 tCO2e/USDmn.
The correlation between the assets of these two companies is 50%. We consider different levels of
uncertainty through λ. The parameters are calibrated according to the values estimated in Section
5.2: α = 0.25, ρ = 0.01, κ = 3 × 10−7 , κ0 = 0.047, κc = 6 × 10−8 , celec = 8 × 10−6 , γr = 0.1.
This result underscores the value of increasing the transparency of companies’ climate im-
pacts as well as improving the forecasting of climate-related financial risks. It also emphasizes the
importance of predictability of public policies in favor of climate transition, notably, the carbon
price upward trajectory. Transparency and predictability are key pillars for a better integration
of climate-related financial risks in green investors’ asset allocation, which provides incentives for
companies to better internalize their climate externalities and thus reduce their climate footprints
more rapidly.
40

5 Empirical evidence
In this section, we provide empirical evidence of the dynamics of companies’ emissions in the
presence of uncertainty about future climate risks (Equation (22)). We also calibrate the parameters
of interest on U.S. stocks between 2004 and 2018 using green fund holdings.
5.1 Emission schedule
Two equilibrium equations can be tested. The first one gives the expected returns in the
presence of green investors (Equation (18)). Zerbib (2020) estimates such equilibrium equation
without uncertainty. In the Internet Appendix, we extend his analysis to the case where green
investors internalize uncertainty about future climate risks by estimating the expected returns
equation (Equation (18)) with Gaussian returns on U.S. stocks between 2004 and 2018. We confirm
the negative effect of the externality premium on asset returns and show that the yearly premia
range from −1.4% for the greenest industry to +0.15% for the brownest industry. The second
equilibrium equation to be tested, and the one which is presented in this section, is the emission
schedule with uncertainty about future climate risks (Equation (22)).
Equation (22) gives the companies’ optimal emission schedule as a function of the proportion
of green investors at the optimization date, α, their climate sensitivity, κ, their uncertainty about
future climate risks, yλ∗ , the marginal abatement cost of the i-th company, ci , the climate sensitivity
of the companies, κc , and the time factors, βt and βtc . For the purpose of the estimation, we assume
that κ, κc , and ci are constant over time. Assuming that the companies have a one-year optimization
horizon,26 the time factors are reduced to β1 = 1 and β1c = 0. Taking the natural logarithm of the
equilibrium equation between t and t + 1, Equation (22) is rewritten as follows:
i ∗
log(ψt+1 ) = log(ci ) − log(κ) − log(αt ) − log(yλ,t+1 ). (25)
26
We estimate a sequence of models with a one-year time horizon. The results are robust to the use of a two-year
horizon.
41

5.1.1 Variables and proxies
Emissions, ψt . As a proxy for the companies’ emissions, we use the carbon intensity, which is
the environmental metric most used by investors (Gibson, Krüger, Riand, and Schmidt, 2019).
Provided by S&P–Trucost, the carbon intensity of the i-th company during year t is defined as
the amount of greenhouse gases emitted by that company during that year divided by its annual
revenue, expressed in tons of CO2 equivalent per million dollars of revenues (tCO2e/USDmn). For
our sample of 2868 companies, between 2004 and 2018, the carbon intensity ranged between 2
tCO2e/USDmn and 23402 tCO2e/USDmn with an average value of 348 tCO2e/USDmn.
Since the proportion of wealth held by green investors, αt , as well as their perception of climate
uncertainty, through yλ∗ , are unobservable, it is necessary to approximate these variables.
Proxy for the proportion of wealth held by green investors, α̃t . To construct a proxy
for the proportion of wealth held by green investors, we use Bloomberg to identify the 348 green
funds with the following features: (i) the asset management mandate includes climate guidelines
(“climate change,” “clean energy,” and “environmentally friendly”); the investment asset classes
are defined as “equity,” “mixed allocation,” and “alternative;” (iii) the geographical investment
scope includes the U.S. Via the data provider FactSet, we retrieve the entire asset holding history
of each of these funds on a quarterly basis (March, June, September, and December)27 and we
focus on the U.S. common stocks (share type codes 10 et 11) listed on the NYSE, AMEX and
NASDAQ (exchanges codes 1, 2 and 3) in the CRSP database. By dividing the total market value
of these assets by the total market capitalization at each quarter qt , we define the proxy for the
green investors’ wealth proportion as:
Market value of U.S. stocks in green funds holdings in qt

α̃qt = . (26)
Total market capitalization of U.S. stocks in qt
27
Given that the list of green funds is not historically available, we acknowledge that the proposed approach may
introduce a survivorship bias. However, given the massive and steady increase in green investments, it is very likely
that the number of closed green funds is limited compared to the number of green funds still in operation.
42

Since the estimation is carried out on an yearly basis, we define α̃t in year t as the average of α̃qt
over the four quarters of the year. The proxy α̃t grows from 0.03% to 0.10% between 2004 and
2018. By defining α̃t in this way, we make the implicit assumption that the proportion of wealth
held by all green investors—mainly composed of pension funds and insurers—grows proportionally
to that of the listed green mutual funds, of which we know the holding history.
Proxy for climate uncertainty, ỹt . Here, the challenge is to construct a proxy for climate
uncertainty perceived by green investors on a monthly basis. Since λ 7→ yλ∗ is an increasing function
with values in [0, 1[ when the market is sufficiently large or diversified, we refer to yλ∗ as a variable
governing uncertainty. We approximate the uncertainty about future climate risks as the degree
of variability in asset holdings by green funds. More precisely, let αi,j,qt be the proportion of asset
i among the U.S. stocks of fund j in quarter qt . For each asset i, in year t, we define mi,qt and
σi,qt as the mean and standard deviation of αi,j,qt among all funds j ∈ {1, ..., 348}. A measure of
the uncertainty perceived by green investors in year t is the average ratio σi,qt /mi,qt among all n
available assets over the four quarters of year t:
n
1 X 1 X σi,qt
ut = . (27)
4 n mi,qt
qt ∈{Mar.t ,Jun.t ,Sep.t ,Dec.t } i=1
∗ is an increasing function of the degree of certainty about future climate risks in a

As yλ,t
sufficiently large or diversified market, we consider the variable −ut . Given that the range of −ut
∗ is [0, 1], we center −u in 0.5 by defining ỹ , the proxy for y ∗ , as
is [−0.935, −0.49] and that of yλ,t t t λ,t
ỹt = −ut − E(−ut ) + 0.5, (28)
which ranges between 0.39 for the year with the highest climate uncertainty and 0.62 for the year
with the lowest climate uncertainty. By constructing this proxy from listed green mutual funds, we
43

make the implicit assumption that the perception of climate risk uncertainty by all green investors
is close to the one perceived by the managers of these funds.
Of course, we acknowledge that the heterogeneity of green fund holdings may be explained by
many other factors. However, the dynamic of this heterogeneity captured by ỹt should largely
be driven by the dynamic of the heterogeneity of green funds’ common practices, especially their
disagreement about future climate risks. In addition, the low correlation between ỹt and the VIX,
which reflects U.S. stock market financial uncertainty (Bekaert, Hoerova, and Duca, 2013), mitigates
the concerns that our proxy substantially captures financial uncertainty. Figure 6 shows that the
proxy for uncertainty about climate-related financial risks decreases (ỹt increases) in the run-up to
the Paris Agreement and increases (ỹt decreases) after Donald Trump’s election.
Figure 6. Proxies α̃t and ỹt . This figure depicts the dynamics of α̃t and ỹt from 2004 to 2018.
Control variables. Given the various drivers that may impact a company’s carbon emissions,
it is crucial to carry out the estimation by controlling for several key variables. First, the strin-
gency of the U.S. environmental policy, which is likely to push companies to mitigate their climate
44

footprints, is approximated on an annual basis between 2004 and 2015 by the OECD composite
Environmental Policy Stringency indicator in the U.S. (Botta and Kozluk, 2014; Galeotti, Salini,
and Verdolini, 2020), denoted by EP St . Second, R&D funding dedicated to renewable energy,
which impacts technological change and may affect greenhouse gas emissions, is approximated by
the total public energy R&D budget in the U.S. dedicated to renewable energy sources, calculated
by the IEA (Galeotti et al., 2020), expressed in billions dollars, and denoted by RDt .28 Third,
business cycles, which may affect the ambition of and the means implemented by companies to
reduce their climate footprints, are approximated by the U.S. GDP expressed in trillion dollars and
denoted by GDPt . Finally, the pressure exerted by investors as part of their shareholder engage-
ment may lead companies to become greener irrespective of the pressure exerted on their cost of
capital through investors’ asset allocations. To approximate shareholder engagement, we use the
ISS shareholder proposal database and define the dummy variable SPti , which is 1 if shareholder
proposals containing the words “climate,” “environment,” “emission,” “carbon,” or “fossil” have
been submitted to company i in year t, and 0 otherwise.29
The descriptive statistics and the correlation matrix for all these variables are available in the
Internet Appendix.30
5.1.2 Estimation
Specification. In the equilibrium equation (25), emissions at t+1 are a function of the proportion
of wealth held by green investors at t and the climate uncertainty at t + 1. In the specification
we estimate, we use the proxy for climate uncertainty and the control variables at time t (rather
than t + 1) for two reasons. First, this specification helps to avoid any simultaneity bias.31 Second,
28
https://www.iea.org/reports/energy-technology-rdd-budgets-2020
29
We look for these keyword in the title of the proposals. We do not only focus on voted proposals, but on all
submitted proposals, because even when a proposal is not voted on, it sends an official signal to the company, which
is incentivized to reform. The estimates are robust to the use of voted proposals.
30
The whole database used for the estimation can be accessed on https://drive.google.com/file/d/
1votURWAiVKRHRp8HmkhnjF-qUTqsg2fj/view?usp=sharing.
31
For example, regarding the climate uncertainty proxy, the asset holdings of green investors depend on the current
emissions of the companies, which may lead to a simultaneity bias if ỹt+1 is used as an independent variable.
45

it is reasonable to assume that the effects of regulatory, technological, or shareholder pressure on
corporate emissions are not immediate. In addition, given the high correlation between the proxies
for regulatory pressure, EP St , technological changes, RDt , and business cycles, GDPt , they are
considered separately in the estimations; except for the specification including GDPt , for which
the maximum variance inflation factor (VIF) reaches 6.4, the VIFs of all estimations (reported in
Table 2) are lower than 4.7.
Let us denote firm fixed effects by f i and the set of control variables by Xti . Assuming that
the proportion of wealth held by green investors and the climate uncertainty are increasing linear
functions of their proxies, we estimate the following specification:
i
log(ψt+1 ) = f i + βα log(α̃t ) + βy log(ỹt ) + Xti + it , (29)
where it stands for the error term.
Estimation. The specification is estimated using a Within regression with Newey West standard
errors on a unbalanced panel of 2868 firms between 2004 and 2018.32 Table 2 presents the results
of the estimation.
As predicted by the theory, log(α̃) and log(ỹ) are significant and their loadings are negative,
whether considered independently (specifications (1) and (2)) or jointly without controls (specifi-
cation (3)) and with controls (specifications (4) to (11)). Controls for regulatory and technological
pressures or business cycles (specifications (4) to (6)) do not significantly impact companies’ emis-
sions. However, shareholder engagement has a significant effect which, in a non-intuitive way,
positively impacts emissions (specifications (7), (8), and (10)). Nevertheless, when the variable is
lagged by 2 years (specifications (9) and (11)), it is no longer significant. A likely explanation for
32
The consistency of the estimator using a fixed effect estimation is validated through a Hausman test. We confirm
the presence of an unobserved heterogeneous effect via Breusch-Pagan, F-, Honda, and Wooldridge tests. We also
show the presence of heteroscedasticity through a Breusch-Pagan test, and serial correlation through Breusch-Godfrey
Wooldridge, Durbin Watson, and AR(1) Wooldridge tests.
46

this effect is the delay in the impact of the proposals submitted to the shareholders’ meetings. In
addition, most of the proposals relate to climate reporting requests that are not directly intended
to reduce emissions the following year.
In all specifications of the model with and without controls, the loadings of the proxy for the
green investors’ proportion of wealth are approximately -0.08 and those of the proxy for climate
uncertainty are approximately -0.1. However, even if we control for regulatory and technological
pressures that may impact corporate emissions, the estimates are not equal to -1 for three main
reasons: in the theoretical sections of this paper, we construct a model including several assumptions
that may not accurately reflect the complexity of the economy, we estimate the equilibrium equation
over a one-year horizon (T = 1), and use proxies for the independent variables of interest that are
unobservable. Therefore, this section is for illustrative purposes and shows that the loadings are
indeed negative and of an acceptable order of magnitude. Focusing on specification (8), we can
estimate the impact of increasing green investors’ proportion of wealth and climate uncertainty on
companies’ carbon intensities. When the proxy for the percentage of green assets, α̃, doubles, the
carbon intensity, ψ, drops, on average, by 4.9% the following year with a 95% confidence interval
ranging from 4.1% to 5.6%. In addition, when the degree of climate uncertainty doubles (i.e., ỹ
decreases by 50%), the carbon intensity increases, on average, by 6.7% the following year with a
95% confidence interval ranging from 4.1% to 9.4%.33
5.2 Calibration
We choose the rate of time preference, ρ, equal to 0.01 (Gollier, 2002; Gollier and Weitzman,
2010). We estimate the share of assets managed taking into account climate criteria, α, at 25%
(US SIF, 2018).
We estimate κ and κ0 by using the estimates in Zerbib (2020) of the externality premium of the
33 ψ2 −ψ1
Denoting by ψ1 the current emissions and ψ2 the emissions when the percentage of green assets doubles, ψ1
=
−0.072 log(2)
e − 1 = −0.049. Similarly, denoting by ψ2 the emissions when the degree of climate uncertainty doubles
(i.e., ỹ decreases by 50%), ψ2ψ−ψ
1
1
= e−0.094 log(0.5) − 1 = +0.067.
47

Table 2 Estimation of the emission schedule. This table presents the estimates of the emission schedule specification (Equation
i ) = f i + β log(α̃ ) + β log(ỹ ) + X i + i . In the equation, ψ i
(22)): log(ψt+1 α t y t t t t+1 is the carbon intensity of the i-th industry at time t + 1
provided by S&P–Trucost and defined as the greenhouse gas emissions emitted by the companies including scope 1, scope 2 and upstream
scope 3 expressed in tCO2e per million dollars of revenue generated, α̃t is the proxy for the proportion of green AUM in t defined in
Equation (26), ỹ is the proxy for the climate uncertainty defined in Equation (28), Xti is the set of control variables including the proxies for
the environmental policy stringency, EP St , the total public energy R&D budget dedicated to renewable energy sources, RDt , the business
cycles, GDPt , and the shareholder proposals focusing on climate issues, SPti . Finally, f i is the firm fixed effect and i,t is the error term.
The equation is estimated using a Within fixed effect regression with Newey West standard errors. The standard errors are reported in
brackets.
i
Dependent variable: log(ψt+1 )
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11)
log(α̃t ) −0.076∗∗∗ −0.077∗∗∗ −0.071∗∗∗ −0.078∗∗∗ −0.084∗∗∗ −0.078∗∗∗ −0.072∗∗∗ −0.072∗∗∗ −0.079∗∗∗ −0.079∗∗∗
(0.004) (0.004) (0.006) (0.004) (0.008) (0.004) (0.006) (0.006) (0.004) (0.004)
log(ỹt ) −0.068∗∗∗ −0.104∗∗∗ −0.096∗∗∗ −0.103∗∗∗ −0.118∗∗∗ −0.102∗∗∗ −0.094∗∗∗ −0.102∗∗∗ −0.102∗∗∗ −0.107∗∗∗
(0.018) (0.018) (0.018) (0.018) (0.021) (0.018) (0.018) (0.018) (0.018) (0.018)
48
EP St −0.009 −0.008 −0.007
(0.008) (0.008) (0.008)
RDt 0.004 0.004 0.004

(0.003) (0.003) (0.003)
GDPt 0.003
(0.003)
SPti 0.046∗∗ 0.059∗∗∗ 0.046∗∗

(0.020) (0.022) (0.020)
i
SPt−1 0.022 0.026

(0.021) (0.020)
Firm FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Observations 16,229 16,229 16,229 11,526 16,229 16,229 16,229 11,526 11,245 16,229 15,773
R2 0.981 0.981 0.981 0.986 0.981 0.981 0.981 0.986 0.986 0.981 0.981
Adjusted R2 0.981 0.981 0.981 0.986 0.981 0.981 0.981 0.986 0.986 0.981 0.981
Maximum VIF n.a. n.a. 1 4.66 1.10 6.39 1 4.66 4.66 1.10 1.10
∗ ∗∗ ∗∗∗
Note: p<0.1; p<0.05; p<0.01
electrical equipment (-1.11%) and the coal (+0.12%) industries in the U.S. between 2013 and 2019:
knowing that this premium is equal to −αθ(ψ) in the present paper, with θ(ψ) = κ0 − κ2 ψ 2 , and
that the average carbon intensity of the electrical equipment and coal industries are ψ elec = 147
tCO2e/USDmn and ψ coal = 555 tCO2e/USDmn, respectively, we get κ = 3 × 10−7 and κ0 =
0.047. Considering companies’ responsiveness to internalize climate externalities following market
pressure, we choose the companies’ climate sensitivity slightly lower than that of the average investor
sensitivity (ακ = 0.25κ): κc = 0.2κ = 5.5 × 10−8 .
We calibrate the marginal abatement costs for an electrical equipment company, which has
an average carbon intensity of 147 tCO2/USDmn, and a coal company, which has an average
carbon intensity of 555 tCO2/USDmn. To do so, we use the equilibrium equation of the emission
schedule without climate uncertainty (Equation (13)) and we assume that the initial emissions of
the companies (ψbelec = 147 tCO2e/USDmn and ψbcoal =555 tCO2e/USDmn) are adjusted to the
optimal level: ψb = ψ0∗ . Therefore, celec = 8 × 10−6 and ccoal = 3 × 10−5 .
Finally, to calibrate the parameters needed for the simulations of the model with climate uncer-
tainty, we assume that the correlation between the two assets is 50% and we take regular investors’
γr
absolute risk aversion, γr , equal to 0.1 (Barberis et al. (2015)). Thus, as α = (γr +γg ) = 0.25,
γg = 0.3. In the simulations, we consider different levels of intensities, λ, and find yλ∗ as the
solution of the fixed point Equation (23).
Table 3 summarizes the calibrated parameters.
6 Conclusion
In this paper we show how green investing impacts companies’ practices by increasing their cost
of capital. Companies are pushed to internalize their climate externalities and thereby reduce their
greenhouse gas emissions. Green investors’ impact is further strengthened when they anticipate
tighter climate regulations, technological advances, and when they account for the negative financial
49

Table 3 Calibrated parameters. This table gives the values of the parameters calibrated
based on the estimates in this section and used for the simulations presented in Figures 2, 3, 4, and
5.
Parameter Value
α 0.25
ρ 0.01
κ 3 × 10−7
κ0 0.047
κc 5.5 × 10−8
ψbelec 147
ψbcoal 555
celec 8 × 10−6
ccoal 3 × 10−5
Cor(elec, coal) 0.5
γr 0.1
γg 0.3
impact of the economy’s average emissions. However, uncertainty about climate risks pushes green
investors to diversify their asset allocation, thereby reducing the incentive for companies to mitigate
their climate footprints.
The results of this paper suggest that investors can increase their impact on companies by rais-
ing their environmental requirements as well as by pressing companies to increase transparency and
their environmental standards. In addition, impact investing is financially beneficial if investors fa-
vor companies that are on a pathway towards reducing their climate footprints or green companies
for which information on their climate footprints is still poorly available. From the viewpoint of
public authorities, this study emphasizes the importance of developing a regulatory framework that
supports the development of green investing and encourages the transparency of information on
companies’ climate footprints. These actions are naturally compatible with the strengthening of cli-
mate regulation and support for climate-related technological innovation, which, when anticipated
by green investors, enhance the pressure the latter exert on companies to cut their emissions.
Impact investing may go beyond climate screening, for example, by favoring brown companies
that are inclined to green up quickly or small green companies that would benefit from financial
50

support to grow (Green and Roth, 2020; Heeb and Kölbel, 2020). Future research could analyze
the impact of these new forms of investment on corporate practices, including their ability to
further reduce the aggregate emissions of an economy. The impact of climate screening could also
be empirically compared to that of shareholder engagement, which Broccardo et al. (2020) find
more effective in reducing the environmental footprint of companies. A third line of research could
introduce the ability for companies to reform dynamically in response to the stochastic dynamics
of cash flow news and investors’ portfolio allocations.
51

References
Aggarwal, Shilpa, Rebecca Dizon-Ross, and Ariel D. Zucker, 2020, Incentivizing Behavioral Change:
The Role of Time Preferences, NBER Working Paper .
Alley, R. B., J. Marotzke, W. D. Nordhaus, J. T. Overpeck, D. M. Peteet, R. A. Pielke Jr., R. T.

Pierrehumbert, P. B. Rhines, L. D. Stocker, T. F.Talley, and J. M. Wallace, 2003, Abrupt Climate
Change, Science 299, 2005–2010.
Amir, Rabah, Marc Germain, and Vincent Van Steenberghe, 2008, On the Impact of Innovation
on the Marginal Abatement Cost Curve, Journal of Public Economic Theory 10, 985–1010.
Arnell, Nigel W., and Simon N. Gosling, 2016, The impacts of climate change on river flood risk
at the global scale, Climatic Change 134, 387–401.
Atmaz, Adem, and Suleyman Basak, 2018, Belief Dispersion in the Stock Market, Journal of
Finance 73, 1225–1279.
Baker, M., D. Bergstresser, G. Serafeim, and J. Wurgler, 2018, Financing the Response to Climate
Change: The Pricing and Ownership of U.S. Green Bonds , Working Paper NBER.
Barber, Brad M., Adair Morse, and Ayako Yasuda, 2021, Impact investing, Journal of Financial
Economics 139, 162–185.
Barberis, N., and A. Shleifer, 2003, Style Investing, Journal of Financial Economics 68, 161–199.
Barberis, Nicholas, Robin Greenwood, Lawrence Jin, and Andrei Shleifer, 2015, X-CAPM: An
extrapolative capital asset pricing model, Journal of Financial Economics 115, 1–24.
Barberis, Nicholas, Robin Greenwood, Lawrence Jin, and Andrei Shleifer, 2018, Extrapolation and
bubbles, Journal of Financial Economics 129, 203–227.
Barnett, Michael, William Brock, and Lars Peter Hansen, 2020, Pricing Uncertainty Induced by
Climate Change, Review of Financial Studies 33, 1024–1066.
Battiston, Stefano, Antoine Mandel, Irene Monasterolo, Franziska Schutze, and Gabriele Visentin,
2017, A climate stress-test of the financial system, Nature Climate Change 7, 283–288.
Bauer, R., K. Koedijk, and R. Otten, 2005, International evidence on ethical mutual fund perfor-
mance and investment style, Journal of Banking and Finance 29, 1751–1767.
Bauman, Yoram, Myunghun Lee, and Karl Seeley, 2008, Does Technological Innovation Really
Reduce Marginal Abatement Costs? Some Theory, Algebraic Evidence, and Policy Implications,
Environmental and Resource Economics 40, 507–527.
Bekaert, Geert, Marie Hoerova, and Marco Lo Duca, 2013, Risk, Uncertainty and Monetary Policy,
Journal of Monetary Policy 60, 771–788.
52

Bolton, Patrick, and Marcin T. Kacperczyk, 2021, Do Investors Care about Carbon Risk?, Journal
of Financial Economics, Forthcoming.
Bolton, Patrick, José Scheinkman, and Wei Xiong, 2006, Executive Compensation and Short-
Termist Behaviour in Speculative Markets, Review of Economics Studies 73, 577–610.
Botta, Enrico, and Tomasz Kozluk, 2014, Measuring environmental policy stringency in oecd coun-
tries: A composite index approach, OECD Economics Department Working Papers 1177, OECD
Publishing.
Brammer, Stephen, Chris Brooks, and Stephen Pavelin, 2006, Corporate Social Performance and
Stock Returns: UK Evidence from Disaggregate Measures, Financial management 35, 97–116.
Broccardo, Eleonora, Oliver Hart, and Luigi Zingales, 2020, Exit vs. Voice, NBER Working Paper
.
Burke, M., W. M. Davis, and N. S. Diffenbaugh, 2018, Large Potential Reduction in Economic
Damages Under UN Mitigation Targets, Nature 557, 549–553.
Burke, M., S. M. Hsiang, and E. M. Miguel, 2015, Global Non-linear Effect of Temperature on
Economic Production, Nature 527, 235–239.
Cai, Yongyang, Kenneth L. Judd, Timothy M. Lenton, Thomas S. Lontzek, and D. Narita, 2015,
Environmental tipping points significantly affect the cost-benefit assessment of climate policies,
Proceedings of the National Academy of Sciences of the United States of America 112, 4606–4611.
Carhart, M. M., 1997, On Persistence in Mutual Fund Performance, Journal of Finance 52, 57–82.
Chava, S., 2014, Environmental externalities and cost of capital, Management Science 60, 2223–
2247.
Chowdhry, Bhagwan, Shaun William Davies, and Brian Waters, 2018, Investing for Impact, Review
of Financial Studies 32, 864–904.
Derwall, J., N. Guenster, R. Bauer, and K. Koedijk, 2005, The eco-efficency premium puzzle,
Financial Analysts Journal 61, 51–63.
Dietz, Simon, 2011, High impact, low probability? An empirical analysis of risk in the economics
of climate change, Climatic Change 108, 519–541.
Dietz, Simon, and Nicholas Stern, 2015, Endogenous Growth, Convexity of Damage and Climate
Risk: How Nordhaus’ Framework Supports Deep Cuts in Carbon Emissions , The Economic
Journal 125, 574–620.
Dimson, Elroy, Oğuzhan Karakaş, and Xi Li, 2015, Active ownership, Review of Financial Studies
28, 3225–3268.
53

Eccles, Robert G., Ioannis Ioannou, and George Serafeim, 2014, The Impact of Corporate Sustain-
ability on Organizational Processes and Performance, Management Science 60, 2835–2857.
Edmans, Alex, 2011, Does the stock market fully value intangibles? Employee satisfaction and
equity prices, Journal of Financial Economics 101, 621–640.
Edmans, Alex, Xavier Gabaix, Tomasz Sadzik, and Yuliy Sannikov, 2012, Dynamic CEO compen-
sation, Journal of Finance 67, 1603–1648.
ElGhoul, S., O. Guedhami, C. C.Y. Kowk, and D. R. Mishra, 2011, Does corporate social respon-
sibility affect the cost of capital?, Journal of Banking and Finance 35, 2388–2406.
European Commission, 2018, Action plan: Financing Sustainable Growth, Report.
European Union High Level Expert Group on Sustainable Finance, 2018, Final Report, 2018.
Fama, Eugene F., and Keneth R. French, 1993, Common risk factors in the returns on stocks and
bonds, Journal of Financial Economics 33, 3–56.
Galema, R, A. Plantinga, and B. Scholtens, 2008, The stocks at stake: Return and risk in socially
responsible investment, Journal of Banking and Finance 32, 2646–2654.
Galeotti, Marzio, Silvia Salini, and Elena Verdolini, 2020, Measuring environmental policy strin-
gency: Approaches, validity, and impact on environmental innovation and energy efficiency,
Energy Policy 136.
Gibson, Rajna, Philipp Krüger, Nadine Riand, and Peter S. Schmidt, 2019, ESG rating disagree-
ment and stock returns, Working paper, SSRN.
Global Sustainable Investment Alliance, 2018, 2018 Global Sustainable Investment Review, Tech-
nical report, Global Sustainable Investment Alliance.
Gollier, Christian, 2002, Discounting and uncertain future, Journal of Public Economics 85, 149–
166.
Gollier, Christian, and Martin Weitzman, 2010, How should the distant future be discounted when
discount rates are uncertain?, Economic Letters 107, 350–353.
Green, Daniel, and Benjamin Roth, 2020, The allocation of socially responsible capital, Working
Paper SSRN.
Hartzmark, Samuel, and Abigail Sussman, 2020, Do Investors Value Sustainability? A Natural
Experiment Examining Ranking and Fund Flows, Journal of Finance 74, 2789–2837.
Heeb, Florian, and Julian Kölbel, 2020, The investor’s guide to impact, Technical report, University
of Zurich, Center for Sustainable Finance & Private Wealth.
54

Heinkel, R., A. Kraus, and J. Zechner, 2001, The effect of green investment on corporate behaviour,
Journal of Financial and Quantitative Analysis 36, 377–389.
Hong, H., and M. Kacperczyk, 2009, The price of sin: the effects of social norms on markets,
Journal of Financial Economics 93, 15–36.
Hong, Harrison, and Jeremy C. Stein, 1999, A Unified Theory of Underreaction, Momentum Trad-
ing, and Overreaction in Asset Markets, Journal of Finance 54, 2143–2184.
Hsu, Po-Hsuan, Kai Li, and Chi-Yang Tsou, 2019, The Pollution Premium, Working paper, SSRN.
Hunter, David, and James Salzman, 2007, Negligence in the Air: The Duty of Care in Climate
Change Litigation, University of Pennsylvania Law Review 155, 1741–1794.
Jaffe, Adam, Richard Newell, and Robert Stavins, 2002, Environmental Policy and Technological
Change, Environmental and Resource Economics 22, 41–70.
Jakob, Michael, and Jérôme Hilaire, 2015, Unburnable fossil-fuel reserves, Nature 517, 150–152.
Krüger, Philipp, 2015, Corporate goodness and shareholder wealth, Journal of Financial Economics
115, 304–329.
Krüger, Philipp, Zacharias Sautner, and Laura Starks, 2020, The Importance of Climate Risk for
Institutional Investors, Review of Financial Studies, 33, 1067–1111.
Lambrecht, Bart M., and Stewart C. Myers, 2017, The Dynamics of Investment, Payout and Debt,
Review of Financial Studies 30, 3759–3800.
Landier, Augustin, and Stefano Lovo, 2020, ESG Investing: How to Optimize Impact?, Working
Paper, SSRN.
Larcker, David F., and Brian Tayan, 2019, CEO Compensation: Data Spotlight, Stanford GSB
Corporate Governance Research Initiative .
Lontzek, Thomas S., Yongyang Cai, Kenneth L. Judd, and Timothy M. Lenton, 2015, Stochastic
integrated assessment of climate tipping points indicates the need for strict climate policy, Nature
Climate Change 5, 441–444.
Marinovic, Ivan, and Felipe Varas, 2019, CEO Horizon, Optimal Pay Duration, and the Escalation
of Short-Termism, Journal of Finance 74, 2011–2053.
Mekaroonreung, Maethee, and Andrew L. Johnson, 2014, A nonparametric method to estimate a

technical change effect on marginal abatement costs of U.S. coal power plants, Energy Economics
46, 45–55.
Mendelsohn, Robert, Kerry Emanuel, Shun Chonabayashi, and Laura Bakkensen, 2012, The impact
of climate change on global tropical cyclone damage, Nature Climate Change 2, 205–209.
55

Milliman, Scott R, and Raymond Prince, 1989, Firm incentives to promote technological change in
pollution control, Journal of Environmental Economics and Management 17, 247–265.
Newey, Whitney K., and Kenneth D. West, 1987, A simple, positive semi-definite, heteroskedasticity
and autocorrelation consistent covariance matrix, Econometrica 55, 703–708.
Nordhaus, William, 2014, Estimates of the social cost of carbon: Concepts and results from the
dice-2013r model and alternative approaches, Journal of the Association of Environmental and
Resource Economists 1, 273–312.
Oehmke, Martin, and Marcus Opp, 2019, A Theory of Socially Responsible Investment, Working
Paper, SSRN.
Osambela, Emilio, 2015, Differences of Opinion, Endogenous Liquidity, and Asset Prices, Review
of Financial Studies 28, 1914–1959.
Palmer, Karen, Wallace E. Oates, and Paul R. Portney, 1995, Tightening Environmental Standards:
The Benefit-Cost or the No-Cost Paradigm?, Journal of Economic Perspectives 9, 119–132.
Pastor, Lubos, Robert F. Stambaugh, and Lucian A. Taylor, 2021, Sustainable Investing in Equi-
librium, Journal of Financial Economics, Forthcoming.
Pedersen, Lasse Heje, Shaun Fitzgibbons, and Lukasz Pomorski, 2021, Responsible Investing: The
ESG-Efficient Frontier, Journal of Financial Economics, Forthcoming.
Porter, Michael E., and Claas van der Linde, 1995, Toward a New Conception of the Environment-
Competitiveness Relationship, Journal of Economic Perspectives 9, 97–118.
Renneboog, L., J. Ter Horst, and C. Zhang, 2008, The price of ethics and stakeholder governance:
The performance of socially responsible mutual funds, Journal of Corporate Finance 14, 302–322.
Riedl, Arno, and Paul Smeets, 2017, Why Do Investors Hold Socially Responsible Mutual Funds?,
Journal of Finance 72, 2505–2550.
Scheinkman, Jose A., and Wei Xiong, 2003, Overconfidence and Speculative Bubbles, Journal of
Political Economy 111, 1183–1219.
Sharfman, M.P., and C.S. Fernando, 2008, Environmental risk management and the cost of capital,
Strategic Management Journal 29, 569–592.
Statman, Meir, and Denys Glushkov, 2009, The Wages of Social Responsibility, Financial Analysts
Journal 65, 33–46.
Statman, Meir, and Denys Glushkov, 2016, Classifying and Measuring the Performance of Socially
Responsible Mutual Funds, Journal of Portfolio Management 42, 140–151.
56

Sugathan, Anish, Ritesh Bhangale, Vishal Kansal, and Unmil Hulke, 2018, How can Indian power
plants cost-effectively meet the new sulfur emission standards? Policy evaluation using marginal
abatement cost-curves, Energy Policy 121, 124–137.
Trinks, Arjan, Bert Scholtens, Machiel Mulder, and Lammertjan Dam, 2018, Fossil Fuel Divestment
and Portfolio Performance, Ecological Economics 146.
US SIF, 2018, Report on US Sustainable, Responsible and Impact Investing Trends, Technical
report, The Forum for Sustainable and Responsible Investment.
van Binsbergen, Jules H., and Christian C. Opp, 2019, Real Anomalies, Journal of Finance 74,
1659–1706.
Varas, Felipe, 2018, Managerial Short-Termism, Turnover Policy, and the Dynamics of Incentives,
Review of Financial Studies 31, 3409–3451.
Veronesi, Pietro, 1999, Stock Market Overreaction to Bad News in Good Times: A Rational
Expectations Equilibrium Model, Review of Financial Studies 12, 975–1007.
Weitzman, Martin, 2009, On Modeling and Interpreting the Economics of Catastrophic Climate
Change, Review of Economic Studies 91, 1–19.
Weitzman, Martin, 2011, Fat-Tailed Uncertainty in the Economics of Catastrophic Climate Change,
Review of Environmental Economics and Policy 5, 275–292.
Welsch, Heinz, and Jan Kühling, 2009, Determinants of pro-environmental consumption: The role
of reference groups and routine behavior, Ecological Economics 69, 166–176.
World Bank, 2020, State and Trends of Carbon Pricing 2020, Technical report.
Xian, Yujiao, Ke Wang, Yi-Ming Wei, and Zhimin Huang, 2020, Opportunity and marginal abate-
ment cost savings from China’s pilot carbon emissions permit trading system: Simulating evi-
dence from the industrial sectors, Journal of Environmental Management 271.
Zerbib, Olivier David, 2019, The effect of pro-environmental preferences on bond prices: Evidence
from green bonds, Journal of Banking and Finance 98, 39–60.
Zerbib, Olivier David, 2020, A Sustainable Capital Asset Pricing Model (S-CAPM): Evidence from
green investing and sin stock exclusion, Working Paper, SSRN.
57

Appendix: Proofs
In this appendix we collect proofs and some supporting mathematical materials, needed to
justify rigorously our claims.
6.1 Proof of Proposition 1
Since the market is assumed to be free of arbitrage and complete, there exists a unique state
price density ξT , i.e., a positive FT -measurable integrable random variable such that the market
price at time t of every contingent claim with terminal value XT , satisfying E[ξT |XT |] < ∞, is given
by
ξt−1 E[ξT XT |Ft ], (30)
where ξt := E[ξT |Ft ] = Et [ξT ]. In particular, since the interest rate is zero, E[ξT ] = 1. It is worth
recalling that P = Pr and that (Bt )t∈[0,T ] is a Brownian motion under this measure.
The optimization problems of the two investors read:
−γ g WTg
−γ r W r h i
r g
min
r
E e T , min
g
E Z T e , (31)
WT ∈AT WT ∈AT
subject to the budget constraints
E[ξT WTr ] = wr , E[ξT WTg ] = wg , (32)
where wr > 0 and wg > 0 are the initial wealth of the regular and green investor, respectively. Both
investors use the real-world probability measure for pricing but every investor uses her subjective
58

measure for computing the utility function. Here we consider admissible controls from the class
AT := {X ∈ FT : Er [ξT |X|] < ∞}
and denote by ZT the Radon-Nikodym density that connects the two probability measures Pg and
Pr . More precisely, recalling (3) and (5), we have
RT 1
RT
λ>
s dBs − 2 kλs k2 ds
ZT = e 0 0 , (33)
where we set λt := σt−1 θ(ψt ), to simplify the notation, and k · k is the Euclidean norm in Rn .
The optimization problem is over the set of all admissible contingent claims, but we shall see
later that the optimal claims will be attainable. Moreover, we assume that
Er [ξT | log ξT |] < ∞ and Er [ξT | log ZT |]. (34)
This assumption will be checked a posteriori for the equilibrium state price density.
By the standard Lagrange multiplier argument, the solutions to problems (31)-(32) are given
by

1 1 1 ξT 1 r ξT
WTr r
= w − r log ξT + r Er [ξT log ξT ] , WTg g
= w − g log + g E ξT log . (35)
γ γ γ ZT γ ZT
The equilibrium state price density ξT is found from the market clearing condition
WTr + WTg = 1> DT + K,
where K is a constant that allows the market to clear since the bond supply is endogenous. Recall
that the interest rate and the initial wealth are exogenous.
59

Substituting the formulas for WTr and WTg , yields
γ∗

∗ >
ξT = c exp −γ 1 DT + g log ZT
γ
1 1 1
for some constant c, where we recall γ∗ = γr + γg . Note that since DT and log ZT are Gaussian,
our a priori assumptions (34) are satisfied.
We can now use the fact that Er [ξT ] = 1 to conclude that:
∗

exp −γ ∗ 1> DT + γγ g log ZT
ξT = h ∗
i .
Er exp −γ ∗ 1> DT + γγ g log ZT
Substituting the explicit formulae for DT and ZT (see (1) and (33)) and using that
T
γ∗ >
Z
∗ >
−γ 1 σt + g λt dBt
0 γ
is normally distributed with zero mean and variance
2
T
γ∗
Z
−γ 1 σt + g λ>
∗ >
dt,
0 γ t
because (σt )t∈[0,T ] and (λt )t∈[0,T ] are deterministic, we have:
Z ·
γ∗ >

∗ >
ξT = E −γ 1 σt + g λt dBt . (36)
0 γ T
Here E denotes the stochastic exponential, i.e., for any adapted square integrable process X ∈ Rn ,
Z · Z t Z t
1 2
E Xs dBs = exp Xs dBs − kXs k ds .
0 t 0 2 0
From (36) and (33) we can easily verify that (34) holds, since (σt ) and (λt ) are deterministic.
60

Using the no-arbitrage pricing rule (30), the vector of equilibrium prices is then given by
Z t Z T
pt = ξt−1 Ert [ξT DT ] = D0 + σs dBs + EQ
t σs dBs ,
0 t
where Q is the risk-neutral measure defined by
dQ
= ξT .
dPr FT
Under Q, the process

γ∗
Z t
et = Bt −
B −γ ∗ σs> 1 + g λs ds
0 γ
is a standard Brownian motion. Hence, the equilibrium prices are computed as follows.
pt = ξt−1 Ert [ξT DT ] (37)

γ∗
Z t Z T
∗ >
= D0 + σs dBs + σs −γ σs 1 + g λs ds
0 t γ
Z T
= Dt + {−γ ∗ Σs 1 + αθs (ψs )} ds,
t
with
t
γr
Z
Dt = D0 + σs dBs , Σt = σt σt> , θt (ψt ) = σt λt , and α = .
0 γr + γg
This completes the proof of (10).
Next we determine the number of shares that each investor holds in her portfolio. The values
of the investors’ portfolios are determined through the no-arbitrage pricing rule (30). In particular,
we have
Wtr = ξt−1 Ert [ξT WTr ]

1 ξT ξT 1 ξT ξT ξT
= wr − r Ert log + log ξt + r Er ξt log + ξt log ξt ,
γ ξt ξt γ ξt ξt ξt
61

by simple algebraic manipulations. Then, using that ξT /ξt is independent of Ft (hence of ξt ) and
that Er [ξt ] = Er [ξT ] = Ert [ξT /ξt ] = 1 we obtain the wealth at time t of the regular investor
1 1
Wtr = wr − r
log ξt + r Er [ξt log ξt ] . (38)
γ γ
By construction Wtr = EQ [WTr |Ft ], hence it is a Q-martingale. Moreover, by (38) we see that the
only stochastic term in the dynamics of (Wtr ) is −1/γ r log ξt . Then, using
Z ·
γ∗ >

∗ >
ξt = E −γ 1 σs + g λs dBs ,
0 γ t
we can conclude that, under the measure Q, the process (Wtr ) has martingale dynamics
Z t
> 1 >
Wtr r
= w + (1 − α) 1 σs − g λs dB
es .
0 γ
The price derived in (37), on the other hand, has martingale dynamics under the measure Q given
by
Z t
pt = p0 + σs dB
es ,
0
where
Z T
p 0 = D0 + (−γ ∗ Σs 1 + αθs (ψs )) ds.
0
It follows that the optimal claim for the investor is replicable by a self-financing portfolio whose
value can be written as follows:
Z t
1 >
Wtr = wr + (1 − α) 1> σ s −
λ
g s
σs−1 dps
0 γ
Z t
1
= wr + (1 − α) 1> − g θs (ψs )> Σ−1
s dps .
0 γ
We conclude that the vector of quantities of shares held by the regular investor at time t is
62

given by

1 −1
Ntr = (1 − α) 1 − g Σt θt (ψt ) ,
γ
while that of the green investor is given by

1 −1
Ntg = α 1 + r Σt θt (ψt ) .
γ
The latter can be obtained by the former and the market clearing condition. Alternatively, the
risk-neutral pricing principle and calculations analogous to the ones above allow us to deduce that
Z t
1
Wtg = ξt−1 Ert [ξT WTg ] =w +α g > > −1
1 + r θt (ψs ) Σs dps
0 γ
from the formula in (35). Hence, the expression of Ntg follows.
Recalling (6), the measure Pc has density with respect to the measure Pr given by
RT RT
(λcs )> dWs − 12 kλcs k2 ds
ZTc = e 0 0 ,
where λct := σt−1 θc (ψt ).
Using (37) and Girsanov theorem, the vector of expected equilibrium prices under the measure
Pc reads
Z T Z t Z T Z T
c ∗
E (pt ) = d + ct (ψt − ψb )dt + θtc (ψs )ds +α θt (ψs ) − γ Σs 1 ds.
0 0 t t
63

Then, the profit function of the i-th company reads
Z T Z T Z t Z T Z T
i i −i −ρt ∗
J (ψ , ψ ) = e d+ cis (ψsi − ψbi )ds + θsc,i (ψs )ds +α θsi (ψs )ds −γ [Σs 1]i ds dt,
0 0 0 t t
where [Σs 1]i is the i-th coordinate of the vector Σs 1.
Maximizing J i (ψ i , ψ −i ) over ψ i is equivalent to maximizing
Z T Z T Z t Z T
J˜i (ψ i , ψ −i ) = e −ρt
cis ψsi ds + θsc,i (ψs )ds +α θsi (ψs )ds dt.
0 0 0 t
Applying integration by parts to the integral with respect to ‘dt’ we have
T
e−ρt − e−ρT c,i 1 − e−ρt i −ρT

i1−e
Z
J˜i (ψ i , ψ −i ) = θt (ψt ) + α θt (ψt ) + ct i
ψt dt. (39)
0 ρ ρ ρ
The problem reduces to maximizing the integrand above along the entire trajectory of (ψti )t∈[0,T ] .
That is
e−ρt − e−ρT c,i 1 − e−ρt i 1 − e−ρT i

max θt (ψt ) + α θt (ψt ) + cit ψt ,
ψti ρ ρ ρ
and the claim follows (see (12)).
Recall the optimization problem (12). Let us denote by Fi the function that the i-th company
needs to maximize:
Fi (ψti , ψt−i ) := β c (t)θtc,i (ψt ) + αβ(t)θti (ψt ) + cit ψti
64

Since ψ 7→ Fi (ψ, ϕ) is concave for each ϕ, it is enough to impose first order conditions:
β c (t)∂ψi θtc,i (ψt ) + αβ(t)∂ψi θti (ψt ) + cit = 0
for all i = 1, 2, . . . n. This leads to
 
n
ε
ψt∗,ε,i = ψt∗,i − ψt∗,ε,j + ψt∗,ε,i  ,
X
 (40)
2n
j=1
where ψ ∗,i := cit [β c (t)κc + αβ(t)κ]−1 is the solution for ε = 0 (i.e., without interaction).
Taking sums over i = 1, 2, . . . n on both sides of the equation, we have
n n n n
ε X ∗,ε,i ε X ∗,ε,i
ψt∗,ε,i = ψt∗,i −
X X
ψt − ψt ,
2 2n
i=1 i=1 i=1 i=1
which gives,
n n
2n
ψt∗,ε,i
X X ∗,i
= ψt .
2n + (n + 1)ε
i=1 i=1
Pn ∗,ε,i
Substituting i=1 ψt back into Equation (40), we get
2n2

ε
ψt∗,ε,i = ψt∗,i − ψt∗,i + ∗
ψ̄ ,
2n + ε 2n + (n + 1)ε t
Pn ∗,j
where ψ̄t∗ = 1
n j=1 ψt . The term in brackets in the expression above is positive, hence proving
(i).
In addition, it is clear that
2n2

ε ε
ψt∗,i + ψ̄ ∗ ∼ ψ̄ ∗ , (41)
2n + ε 2n + (n + 1)ε t n→+∞ ε+2 t
65

which proves (ii).
6.4 Proof of Proposition 5 and 6
The standard approach to the problem, via dynamic programming, requires us to introduce the
value processes for the two agents:
Vtr = minr Ert [exp (−γ r WTr )] , Vtg = ming Egt exp −γ g WTg ,

N ∈At,T N ∈At,T
where, for t ≤ T and j ∈ {r, g}, we define
Ajt,T := {(Nsλ )t≤s≤T : N λ is Rn -valued, (Fs )t≤s≤T -adapted and Pj -square integrable}
and Pj -square integrable means

Z T
j
E |Ntλ |2 dt < +∞.
0
Moreover, we assume that the equilibrium price has the following dynamics.
Z t Z t Z t
pt = p0 + µs ds + σs dBs + θs (ψs )dNsλ (42)
0 0 0
under the probability Pg of the green investors and
Z t Z t
pt = p0 + µs ds + σs dBs (43)
0 0
under the probability Pr of the regular investors, where µ is deterministic and must be found in
equilibrium. We shall show a posteriori that an equilibrium price process of this form can indeed
be found.
66

Following a well-known ansatz we expect
Vtr = exp (−γ r Wtr + Qrt ) , Vtg = exp (−γ g Wtg + Qgt ) ,
where Qr and Qg are absolutely continuous deterministic processes with
dQrt = qtr dt and dQgt = qtg dt.
Applying the Itô’s formula for jump processes to V g under the green investor measure yields
(γ g )2

g
dVtg = Vt−
g
−γ g dWtg + qtg dt + d[W g ]ct + (e−γ∆Wt − 1 + γ∆Wtg )
2
(γ g )2 g,λ >

g,λ >
g g g,λ > g c g,λ −γ(Nt− ) ∆pt g g,λ >
= Vt− −γ (Nt ) dpt + qt dt + (Nt ) d[p]t Nt + (e − 1 + γ (Nt− ) ∆pt )
2
(γ g )2 g,λ >
g

g g g,λ > g > g,λ − γλ (Ntr,λ )> θt (ψt )
= Vt− −γ (Nt ) µt + qt + (Nt ) σt σt Nt + λ(e − 1) dt + Mt ,
2
where (Mt ) is a Pg -martingale on [0, T ] and [W g ]c is the continuous part of the quadratic variation
of the process W g . Since V g must be a martingale along the trajectory of the optimal process
(Ntg,λ ) and a submartingale along every trajectory, we conclude that the drift term in ‘dVtg ’ must
be non-negative and
(γ g )2 λ >
g

g λ > g λ − γλ (Ntλ )> θt (ψt )
min −γ (Nt ) µt + qt + (Nt ) Σt Nt + λ(e − 1) = 0 (44)
Ntλ 2
for each t ∈ [0, T ]. Since Σ is nondegenerate, the function to be minimized is strictly convex and
coercive (i.e., it tends to +∞ as kNtλ k → ∞), thus the unique minimum is always attained. With
a slight abuse of notation, we denote the minimizer of (44) (which does not depend on qt ) by Ntg,λ ,
as this will be the number of assets held by the green investors. By imposing first order conditions
67

we have that Ntg,λ must be the unique solution of
γg
(Ntλ )> θt (ψt )
µt − γ g Σt Ntλ + e− λ θt (ψt ) = 0.
By the same logic, the regular investors use the measure Pr to compute the dynamic ‘dVtr ’ and
find the optimal quantity of assets. In particular, the optimal quantity Ntr,λ is the minimizer of
(γ r )2 λ >

min −γ r
(Ntλ )> µt + qtr + (Nt ) Σt Ntλ = 0.
Ntλ 2
Since the regular investors do not take into account the climate uncertainty, there is no jump term
in this formula, and the minimizer is given explicitly by
1 −1
Ntr,λ = Σ µt .
γr t
The market clearing condition therefore allows to compute (µ, N r,λ , N g,λ ) by solving the fol-
lowing system of equations:
1 −1
Ntr,λ = Σ µt ;
γr t
γg
(Ntg,λ )> θt (ψt )
µt − γ g Σt Ntg,λ + e− λ θt (ψt ) = 0; (45)
Ntr,λ + Ntg,λ = 1.
Substituting µt from the second equation into the first one, allows to eliminate it, obtaining the
following equation:
γg
γ r Σt (1 − Ntg,λ ) − γ g Σt Ntg,λ + e− λ θt (ψt ) = 0. (46)
The left-hand side of this equation coincides with the gradient of the strictly convex, differentiable
68

and coercive function
γr + γg > λ γg >
f (N ) := −γ r 1> Σt N + N Σt N + g e− λ N θt (ψt ) ,
2 γ
γg
which proves existence and uniqueness of the solution of (46). Let us write yλ = e− λ
in (46) and solve for Ntg,λ to obtain the explicit expression

yλ −1
Ntg,λ = α 1 + r Σt θt (ψt ) .
γ
Plugging this back into (46) we find yλ by solving the one-dimensional equation
g n
y
o
− αγ 1> θt (ψt )+ γλr θt (ψt )> Σ−1
t θt (ψt )
yλ = e λ
.
Finally the expression for p0 is obtained from the condition pT = DT .
First we show existence and uniqueness of the solution yλ to Eq. (20). Writing
y
g(y) := α(1> θt (ψt ) + θt (ψt )> Σ−1
t θt (ψt )),
γr
Equation (20) becomes
−λ log y = γ g g(y), (47)
where it is also worth noticing that g(1) = θt (ψt )> Ntg . Since g is increasing and linear and the
function y 7→ −λ log y is strictly decreasing, continuous and maps (0, ∞) onto R, there exists one
and only one positive solution of (20).
Next we show the monotonicity of the map λ 7→ yλ stated in (ii)–(iv). If g(1) > 0, the solution
belongs to the interval (0, 1), and it is easy to see that it is monotonically increasing in λ. If
69

g(1) = 1, the solution is constant and equal to 1. If g(1) < 0, the solution satisfies yλ > 1 and it is
easy to see that it is decreasing in λ.
To study the asymptotic behavior of yλ as λ → ∞, we expand the expression on the right hand
side of (20) using Taylor up to the first order in λ−1 . That gives
αγ g

yλ > −1
yλ = 1 − >
1 θt (ψt ) + r θt (ψt ) Σt θt (ψt ) + O(λ−2 ).
λ γ
Then we substitute yλ = y0 + y1 λ−1 on both sides of the expression above and, equating terms of
the same order in λ−1 , we find y0 = 1 and

g > 1 > −1
y1 = −αγ 1 θt (ψt ) + r θt (ψt ) Σt θt (ψt ) .
γ
Since yλ = y0 + y1 λ−1 it is now immediate to obtain limλ→∞ yλ = 1 in all cases.
It remains to prove (i). Since ψt is fixed, for simplicity we omit it from some of the formulae
below. A straightforward computation yields the following expression for (Ntg,λ )> Σt Ntg,λ :
1/2 2α2 α2
(Ntg,λ )> Σt Ntg,λ = kΣt Ntg,λ k2 = α2 kNtg k2 + (yλ − 1)θ >
t 1 + (y 2 − 1)θt> Σ−1
t θt .
γr (γ r )2 λ
Let
2 1
f (y) := r
(y − 1)θt> 1 + r 2 (y 2 − 1)θt> Σ−1
t θt
γ (γ )
and notice that f is a quadratic function with f 0 (y) = 2/(γ r α)g(y). From the arguments above,
if g(1) > 0 we have λ 7→ yλ ∈ (0, 1) and increasing so that by Equation (47), g(y λ ) > 0, and
therefore λ 7→ f (y λ ) is increasing as well. If, on the other hand g(1) < 0, then λ 7→ yλ ∈ (1, ∞) and
decreasing so that by Equation (47), g(y λ ) < 0, and therefore λ 7→ f (y λ ) is increasing once again.
The claim holds trivially in the remaining case g(1) = 0.
70

6.5 Proof of Propositions 7 and 8
Proof of Proposition 7. For each y > 0 the derivation of Eq. (22) follows immediately from (21),
as in Corollary 3. Next we prove solvability of Eq. (23) and some properties of its solution.
Proof of Proposition 8. Writing
αy
g ∗ (y) := 1> θt (ψt (y)) + θt (ψt (y))> Σ−1
t θt (ψt (y)),
γr
Equation (23) becomes
−λ log y = αγ g g ∗ (y). (48)
Unlike g(y), used in the proof of Proposition 6, g ∗ (y) is a nonlinear and possibly non-monotonic
function of y, which is more difficult to study. Nevertheless, it is clear that g ∗ (0) < ∞ and
ακ20 > −1
g ∗ (y) ∼ γr 1 Σt 1 y as y → ∞, which implies that Equation (23) admits a solution on (0, ∞).
Then (i) holds.
To prove uniqueness (that is (ii)) it is sufficient to show that (24) implies that g ∗ is monotonic
increasing. It is immediate to check
n
d > X d i i d i
1 θt (ψt (y)) = θ (ψ (y)) ψ (y) ,
dy dψ t t dy t
i=1
and

d αy α αy d
r
θt (ψt (y)) Σt θt (ψt (y)) = r θt (ψt (y))> Σ−1
> −1
t θt (ψt (y)) + r θt (ψt (y))> Σ−1
t θt (ψt (y))
dy γ γ γ dy
αy d
≥ θt (ψt (y))> Σ−1
t θt (ψt (y)).
γ r dy
71

The latter expression can be expanded as
n
d > −1
X
−1
j j d i i d i
θt (ψt (y)) Σt θt (ψt (y)) = Σt ij θt (ψt (y)) θ (ψ (y)) ψ (y) .
dy dψ t t dy t
i,j=1
Combining the calculations above we obtain
 
n n
d ∗ X α X
−1
j j d i i d i
g (y) ≥ 1 + y Σt ij θt (ψt (y)) θ (ψ (y)) ψ (y) .
dy γr dψ t t dy t
i=1 j=1
Recalling the expressions for ψt (y) and θt (ψ) (see (22) and Corollary 3, respectively), we immedi-
ately see that

d i i d i
θ (ψ (y)) ψ (y) ≥ 0, for all indexes i.
dψ t t dy t
Then the condition (24) is sufficient to guarantee that g ∗ is monotonic increasing and the solution
to (48) is unique.
All the statements in (iii) are shown using the same arguments of proof as in Proposition 6,
using that g ∗ is increasing and that ψt∗ (1) = ψt∗,0 (1).
72

Internet Appendix for
”Climate Impact Investing”
This document provides additional figures and tables:
– the dynamics of the time factors β and β c as functions of several values of the rate of time
preference, ρ (Figure IA.1);
– the analysis of an alternative specification of the marginal abatement cost presented in Foot-
note 17 of the paper: the new marginal abatement cost function is shown in Figure IA.2 and
the optimal emission schedules are shown in Figure IA.3;
– descriptive statistics (Table IA.1) and a correlation matrix (Table IA.2);
– the estimation of the expected returns in equilibrium under the model with climate uncer-
tainty (Table IA.3).
1−e−ρt
Figure IA.1. Dynamics of βt and βtc . This figure depicts the dynamics of βt = 1−e−ρT
and
e−ρt −e−ρT
βtc = 1−e−ρT
over time for different rates of time preference, ρ ∈ {0.01, 0.1, 0.2}.
Internet Appendix page 1

Figure IA.2. Alternative marginal abatement cost. This figure shows the shape of the
marginal abatement cost function used in the alternative analysis presented in Footnote 17. This
marginal cost function is constant and positive when emissions decrease below the initial emissions
level, ψb ; in contrast, the marginal gain is zero (the cost is zero) when emissions increase above the
cb
initial level, ψb . The function of the marginal abatement cost is ψ 7→ c(ψt ) = 1+exp(1000(ψ t −ψb ))
,
where cb is the constant marginal abatement used in the main model. In this framework, the
numerically simulated optimal emission schedules are presented in Figure IA.3 of the Internet
Appendix.
Table IA.1 Summary statistics. This table provides the summary statistics on the depen-
dent and independent variables in the estimation of specification (29) giving the emission schedule
in equilibrium between December 2004 and December 2018. The statistics relate to the carbon
intensity, ψt ; the proxies for the proportion of wealth held by green investors and their climate un-
certainty, α̃t and ỹt , respectively; the OECD proxy for the environmental policy stringency, EP St ;
the R&D budget in the U.S. dedicated to renewable energy sources (expressed in billion dollars);
the GDP, GDPt ; and the dummy variable, SPti , which is 1 if shareholder proposals containing the
words ”climate,” ”environment,” ”emission,” ”carbon,” or ”fossil” have been submitted to firm i
in year t, and 0 otherwise. The statistics presented are the means, medians, standard deviations,
minima, and maxima of the variables of interest for 2868 companies listed on the NYSE, AMEX
and NASDAQ common stocks between December 31, 2004, and December 31, 2018.
ψt α̃t ỹt EP St RDt GDPt SPt

Mean 348 0.0005 0.5 2.21 0.94 15.51 0.006
Median 72 0.0005 0.51 2.47 0.8 14.99 0
Std. Dev. 1155 0.0003 0.07 0.68 0.73 2.55 0.076
Min. 2 0 0.39 1.05 0.3 11.46 0
Max. 23402 0.001 0.62 3.17 3 20.58 1

(a) ψt with different values for α (b) ψt with different values for κc
Figure IA.3. Emission schedules with the marginal abatement cost depicted in Figure
IA.2 of this Internet Appendix. This figure shows the optimal emission schedules, ψt , according
to several values of the proportion of green investors (α, sub-figure (a)) and the companies’ climate
sensitivity (κc , sub-figure (b)) using the marginal abatement cost depicted in Figure IA.2 of this
Internet Appendix and presented in Footnote 17 of the paper. The parameters are calibrated
according to the values estimated in Section 5.2: ψb = 147, α = 0.25, ρ = 0.01, κ = 3.10−7 ,
κc = 6.10−8 , cb = 8.10−6 .
Table IA.2 Correlation matrix. This table reports the correlation matrix between the in-
dependent variables involved in specification 29: the proxies for the proportion of wealth held by
green investors and their climate uncertainty, α̃t and ỹt , respectively; the OECD proxy for the
environmental policy stringency, EP St ; the R&D budget in the U.S. dedicated to renewable energy
sources (expressed in billion dollars); the GDP, GDPt ; and the dummy variable, SPti , which is 1
if shareholder proposals containing the words ”climate,” ”environment,” ”emission,” ”carbon,” or
”fossil” have been submitted to firm i in year t, and 0 otherwise.
log(α̃t ) log(ỹt ) EP St RDt GDPt

log(ỹt ) -0.31
EP St 0.92 -0.29
RDt 0.48 -0.17 0.72
GDPt 0.89 -0.07 0.79 0.19
SPt 0.05 -0.02 0.04 0.02 0.05

Table IA.3 Estimation of the asset pricing equation with climate uncertainty. This
table presents the estimates of the asset pricing Equation (18) written in a framework with Gaussian
returns (see Zerbib (2020)) : E(ri ) = ι + βmkt Cov(ri , rm ) + βpremium ỹ α̃θ̃i (ψ i ). The estimation is
performed using value-weighted monthly returns in excess of the 1-month T-Bill for the 48 SIC
industry-sorted portfolios between December 31, 2004, and December 31, 2018. ri is the value-
weighted excess return on industry i (i = 1, ..., n), rm is the market excess return, α̃ is the proxy
for the proportion of wealth held by green investors, ỹ is the proxy for the uncertainty about climate
risks, and θ̃i (ψ i ) is the proxy for the climate externalities of industry Ii constructed as in Zerbib
(2020). This specification is compared with two other specifications: (i) we add to our model the
beta of the Carhart (1997) momentum factor and betas of the Fama and French (1993) size and
value factors denoted by SMB, HML, and MOM, respectively; (ii) we add to specification (i) the
proxy for the unexpected shifts in beliefs ∆θ̃i (ψ i ) defined as θ̃ti (ψti ) − θ̃t−1
i (ψ i ) (see Zerbib (2020)).
t−1
First, the variables are estimated by industry in a 3-year rolling window at monthly intervals. In the
second step, a cross-sectional regression is performed by month on all the industries. The estimated
parameter is the average value of the estimates obtained on all months during the period. t-values,
estimated following Newey and West (1987) with three lags, are reported in round brackets. The
last column reports the average OLS adjusted-R2 and the GLS R2 on the row underneath. The
95% confidence intervals are shown in square brackets.
Constant βmkt βpremium SMB HML MOM βshift Adj OLS/GLS R2

Estimation of the market and externality premia separately
Estimate 0.0132 -0.7872 0.05 [0.03,0.07]
t-value (10.91) (-1.17) 0.07 [0.05,0.09]
Estimate 0.013 -0.427 -0.01 [-0.02,-0.01]
t-value (13.01) (-2.87) 0.01 [0.01,0.01]
Main estimation
Estimate 0.0134 -0.7753 -0.4143 0.04 [0.02,0.05]
t-value (11.15) (-1.14) (-3.37) 0.08 [0.06,0.09]
Main estimation with SMB, HML and MOM betas
Estimate 0.013 0.2884 -0.8329 0.000 0.0002 -0.0001 0.23 [0.19,0.26]
t-value (11.63) (0.41) (-5.85) (-0.1) (1.23) (-0.88) 0.31 [0.28,0.34]
Main estimation with SMB, HML and MOM betas, and control for unexpected shifts in beliefs
Estimate 0.0131 0.1601 -0.4376 0.000 0.0001 -0.0001 0.0119 0.22 [0.19,0.26]
t-value (11.52) (0.23) (-2.14) (-0.17) (1.04) (-0.86) (3.53) 0.32 [0.29,0.35]

Climate Impact Investing

Uploaded by

Copyright:

Available Formats

Climate Impact Investing

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Climate Impact Investing

Uploaded by

Copyright:

Available Formats

Climate Impact Investing

June 25, 2021

Electronic copy available at: https://ssrn.com/abstract=3562534

Figure 1. Percentage of sustainable investments and average carbon intensity of the

Electronic copy available at: https://ssrn.com/abstract=3562534

companies’ emissions on the company under consideration. However, in a sufficiently large or

companies to mitigate their emissions.

We develop a dynamic equilibrium model populated by heterogeneous investors and companies.

Electronic copy available at: https://ssrn.com/abstract=3562534

directly affects the parameter on which investors disagree—companies’ climate externalities.

Electronic copy available at: https://ssrn.com/abstract=3562534

3% per year over the same period.

Electronic copy available at: https://ssrn.com/abstract=3562534

Electronic copy available at: https://ssrn.com/abstract=3562534

from 4.1% to 9.4%.

information on their climate footprints is still poorly available.

Electronic copy available at: https://ssrn.com/abstract=3562534

their impact on corporate emissions.

Electronic copy available at: https://ssrn.com/abstract=3562534

mitigate their risk.

We also contribute to the emerging literature on impact investing. In a seminal study, by

forces companies to partially internalize externalities. In contrast, by analyzing venture capital

Electronic copy available at: https://ssrn.com/abstract=3562534

evidence of the impact of ESG integration on companies’ practices.

2 A simple economy with greenhouse gas emitting companies and

and companies’ optimization programs.

2.1 Securities market

Electronic copy available at: https://ssrn.com/abstract=3562534

the vector of terminal dividends for all assets, DT ∈ Rn , reads

Electronic copy available at: https://ssrn.com/abstract=3562534

each s ∈ [0, T ], σs is a deterministic, n × n, invertible matrix.

endogenously determine their prospective greenhouse gas emissions.

Electronic copy available at: https://ssrn.com/abstract=3562534

Ert (DT ) = Dt , (3)

necessarily consistent with the realized events.

for example, a company’s pioneering environmental positioning in an economic segment or its

Electronic copy available at: https://ssrn.com/abstract=3562534

climate externalities at date t ∈ [0, T ]. The latter is expressed by

anticipation of stronger regulatory pressure. As a consequence, green investors internalize their

conditional on the information at time t, we have

Electronic copy available at: https://ssrn.com/abstract=3562534

drift θs (ψs )ds associated with the beliefs of green investors.9

Ect the expectation of the companies conditional on the information in t, we have

2.3 Investors’ preferences and optimization

Electronic copy available at: https://ssrn.com/abstract=3562534

stands for the transposition operator.

In what follows, we denote by γ ∗ the global risk aversion, defined by 1

2.4 Companies’ utility and optimization

Electronic copy available at: https://ssrn.com/abstract=3562534

(ψti )t∈[0,T ] so as to maximize the following objective function:

i ∈ {1, ..., n} determines its own emissions, ψ i,∗ , in t = 0, so that

J i (ψ ∗,i , ψ ∗,−i ) ≥ J i (ψ i , ψ ∗,−i ), for all ψ i ∈ F ([0, T ], R+ ). (9)

interactions in the economy we model.

Electronic copy available at: https://ssrn.com/abstract=3562534

Date Agent Choose Given

3 Equilibrium in the presence of greenhouse gas emitting compa-

nies and heterogeneous beliefs

changes, and companies’ interactions on companies’ optimal emission schedules.

3.1 Equilibrium stock prices and returns

Electronic copy available at: https://ssrn.com/abstract=3562534

n ε ≥ 0, the optimal emission schedule is approximated by