See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/352836844

Sustainable Investing with ESG Rating Uncertainty

Preprint · June 2021

CITATIONS READS
0 6,723

4 authors, including:

Doron Avramov Abraham Lioui

Hebrew University of Jerusalem Edhec Business school
91 PUBLICATIONS 4,023 CITATIONS 107 PUBLICATIONS 1,115 CITATIONS

SEE PROFILE SEE PROFILE

Andrea Tarelli
Catholic University of the Sacred Heart
20 PUBLICATIONS 296 CITATIONS

SEE PROFILE

All content following this page was uploaded by Andrea Tarelli on 30 June 2021.

The user has requested enhancement of the downloaded file.

 Sustainable Investing
with ESG Rating Uncertainty*
Doron Avramov„ Si Cheng Abraham Lioui§ Andrea Tarelli¶

This Version: 26th June 2021

Abstract

This paper analyzes the asset pricing and portfolio implications of an important barrier
to sustainable investing—uncertainty about the corporate ESG profile. In equilibrium,
the market premium increases and demand for stocks declines under ESG uncertainty.
In addition, the CAPM alpha and effective beta both rise with ESG uncertainty and
the negative ESG-alpha relation weakens. Employing the standard deviation of ESG
ratings from six major providers as a proxy for ESG uncertainty, we provide supporting
evidence for the model predictions. Our findings help reconcile the mixed evidence on
the cross-sectional ESG-alpha relation and suggest that ESG uncertainty affects the
risk-return trade-off, social impact, and economic welfare.

Keywords: ESG, Rating Uncertainty, Portfolio Choice, Capital Asset Pricing Model
JEL: G11, G12, G24, M14, Q01

* We especially thank an anonymous referee for insightful comments and suggestions. We also thank
Bill Schwert (the editor), Yakov Amihud, Marcin Kacperczyk, Lubos Pastor, Lasse Heje Pedersen, Luke
Taylor, seminar participants at Catholic University of Milan, Ben-Gurion University of the Negev, Bar-Ilan
University, Bocconi University, and EDHEC Business School, and conference participants to the 2021 North
American Summer Meeting of the Econometric Society for useful comments and discussions. We are solely
responsible for any remaining errors.
„ Interdisciplinary Center (IDC), Herzliya, Israel. [email protected]
Chinese University of Hong Kong, Hong Kong. [email protected]
§
EDHEC Business School and EDHEC & Scientific Beta Research Chair, Nice, France. abra-
[email protected]
¶ Catholic University of Milan, Milan, Italy. [email protected]
 Sustainable Investing
with ESG Rating Uncertainty

This Version: 26th June 2021

Abstract

This paper analyzes the asset pricing and portfolio implications of an important barrier
to sustainable investing—uncertainty about the corporate ESG profile. In equilibrium,
the market premium increases and demand for stocks declines under ESG uncertainty.
In addition, the CAPM alpha and effective beta both rise with ESG uncertainty and
the negative ESG-alpha relation weakens. Employing the standard deviation of ESG
ratings from six major providers as a proxy for ESG uncertainty, we provide supporting
evidence for the model predictions. Our findings help reconcile the mixed evidence on
the cross-sectional ESG-alpha relation and suggest that ESG uncertainty affects the
risk-return trade-off, social impact, and economic welfare.

Keywords: ESG, Rating Uncertainty, Portfolio Choice, Capital Asset Pricing Model
JEL: G11, G12, G24, M14, Q01
1 Introduction
The global financial market has experienced exponential growth in sustainable investing,
an investment approach that considers environmental, social, and governance (ESG) factors
in portfolio selection and management. Since the launch of United Nations Principles for
Responsible Investment (PRI) in 2006, the number of signatories has grown from 734 in 2010
to 1,384 in 2015 and 3,038 in 2020, with total assets under management of US$21 trillion
in 2010, US$59 trillion in 2015, and US$103 trillion in 2020.1 In line with the increasing
concerns about global warming, BlackRock CEO Larry Fink wrote in a recent annual letter
that climate change will force businesses and investors to shift their strategies, leading to a
“fundamental reshaping of finance” and “significant reallocation of capital”.2
As the ESG objective is becoming a primary focus in asset management, the reallocation
of capital has major implications for portfolio decisions and asset pricing. However, ESG
investors often confront a substantial amount of uncertainty about the true ESG profile of
a firm. In the absence of a reliable measure of the true ESG performance, any attempt to
quantify it needs to cope with incomplete and opaque ESG data and nonstructured meth-
odologies. A meaningful illustration of uncertainty about the ESG score is the pronounced
divergence across ESG rating agencies.3 While such uncertainty could be an important bar-
rier to sustainable investing, to date, little attention has been devoted to the role of ESG
uncertainty in portfolio decisions and asset pricing.
This paper aims to fill this gap by analyzing the equilibrium implications of ESG uncer-
tainty for both the aggregate market and the cross section. To pursue this task, we consider
brown-averse agents who extract nonpecuniary benefits from holding green stocks, following
Pástor, Stambaugh, and Taylor (2021a). We first study the aggregate market through a
mean-variance setup that consists of the market portfolio and a riskless asset. Due to uncer-
tainty about the ESG profile, equities are perceived to be riskier. In addition, the demand
for equities consists of two components: (1) the usual demand when ESG preferences are
muted and (2) a demand for a pseudo-asset with a positive payoff for a green market and a
negative payoff for a brown market as well as volatility that evolves from uncertainty about
the market ESG score. Aggregating these components, we show that the overall demand for
equities falls due to ESG uncertainty, even when the market is green.
We then formulate the market premium in equilibrium. While the higher risk due to
1
See, https://www.unpri.org/pri.
2
See, https://www.blackrock.com/corporate/investor-relations/larry-fink-ceo-letter.
3
Berg et al. (2020) document that the average correlation among six major rating providers is only 0.54.
They also find that, even when the categories of attributes considered for the evaluation of a firm’s ESG
profile are fixed, raters largely disagree on the measurement of these granular characteristics.

1
ESG uncertainty essentially commands a higher market premium, there is an offsetting
force when the market is green because ESG investors extract nonpecuniary benefits from
holding green stocks. The ultimate implications of ESG preferences with uncertainty for
the market premium are thus inconclusive. When the market is green neutral, however, the
equity premium rises with ESG uncertainty. For perspective, when ESG uncertainty is not
accounted for and the market is green (green neutral), the market risk does not change, the
demand for risky assets rises (does not change), and the market premium drops (does not
change) relative to ESG indifference.
We further derive a CAPM representation where both alpha and the effective beta vary
with firm-level ESG uncertainty. The effective beta differs from the CAPM beta in the
following way. While the CAPM beta is based on the covariance and variance of actual
returns, the effective beta reflects the notion that both the market and individual stock
returns are augmented by a random ESG-based component, which is positive for a green
asset and negative otherwise. Thus, the effective beta is based on the covariance and variance
of ESG-adjusted returns. Regarding alpha, when ESG uncertainty is not accounted for, the
CAPM alpha exclusively reflects the willingness to hold green stocks due to nonpecuniary
benefits, and the ESG-alpha relation is, hence, negative.4 Accounting for ESG uncertainty,
the equilibrium alpha increases with ESG uncertainty and the ESG-alpha relation weakens.
We move on to empirically test the model implications using U.S. common stocks from
2002 to 2019. We collect ESG ratings from six major rating agencies, namely, Asset4 (Refin-
itiv), MSCI KLD, MSCI IVA, Bloomberg, Sustainalytics, and RobecoSAM. We employ the
average (standard deviation of) ESG ratings across rating agencies to proxy for the firm-level
ESG rating (ESG uncertainty). Consistent with existing studies, we confirm that there are
substantial variations across different rating providers, while the average rating correlation
is 0.48. The variations are quite persistent throughout the entire sample period.
We first examine how the ESG rating and uncertainty affect investor demand. To better
capture the demand from ESG-sensitive investors, we consider three distinct types of institu-
tions: norm-constrained institutions, hedge funds, and other institutions. Norm-constrained
institutions, such as pension funds as well as university and foundation endowments, are more
likely to make socially responsible investments compared to hedge funds or mutual funds
that are natural arbitrageurs (Hong and Kacperczyk (2009)). We first confirm that norm-
constrained institutions display preferences for greener firms. Consistent with the model
prediction, we find that in the presence of uncertainty about the ESG profile, ESG-sensitive
investors lower their demand for risky assets. For instance, among the high-ESG-rating
portfolios, norm-constrained institutions hold 22.8% of the low-uncertainty stocks while only
4
See, e.g., Heinkel et al. (2001) and Pástor, Stambaugh, and Taylor (2021a).

2
18.1% of the high-uncertainty stocks, indicating a 21% decline. The results are particu-
larly strong among high-ESG stocks, suggesting that rating uncertainty matters the most
for ESG-sensitive investors in their ESG investment. Notably, even with the growing ESG
awareness, their demand for green assets continues to diminish with rating uncertainty in
the recent decade. In addition, while hedge funds invest more in low-ESG stocks, rating
uncertainty plays a similar role in discouraging stock investment.
We next examine the cross-sectional implications of ESG uncertainty. We first sort stocks
into quintile portfolios based on their ESG uncertainty. Within each uncertainty group, we
further sort stocks into quintile portfolios according to their ESG ratings. We find that
the ESG rating is negatively associated with future performance among stocks with low
ESG uncertainty, providing empirical support for the predictions of Pástor, Stambaugh, and
Taylor (2021a), who rely on deterministic ESG scores. For instance, brown stocks outperform
green stocks by 0.59% per month in raw return and 0.40% per month in CAPM-adjusted
return. However, in the presence of ESG uncertainty, our model shows that the ESG-alpha
relation can be nonlinear and ambiguous. Indeed, we demonstrate empirically that the
negative return predictability of ESG ratings does not hold for the remaining firms. The
results are robust to adjusting returns for alternative risk factors and controlling for firm
characteristics in Fama and MacBeth (1973) regressions.
Finally, we calibrate the model for plausible values of market volatility and risk aver-
sion. The investment universe consists of a riskless asset and the market portfolio. Our
calibration considers two types of agents who observe the returns on investable assets. One
type of agents accounts for ESG preferences with uncertainty in assessing the risk-return
profile of the optimal portfolio, while the other type is ESG indifferent. Accounting for ESG
uncertainty significantly reduces the demand for the market portfolio and the certainty equi-
valent rate of return of ESG-sensitive agents. The calibration results reinforce the notion
that ESG uncertainty could negatively, and significantly so, affect the risk-return trade-off,
social impact, and economic welfare.
This paper contributes to several strands of the literature. First, we explicitly account for
uncertainty about the ESG profile in equilibrium asset pricing for both the aggregate market
and the cross section. Prior work has focused on investors’ ESG preferences (e.g., Heinkel
et al. (2001) and Pástor, Stambaugh, and Taylor (2021a)), while our model predictions and
calibration results highlight the importance of considering ESG uncertainty when analyzing
sustainable investing. Specifically, the perceived equity risk increases with ESG uncertainty,
while the demand for equity falls. ESG uncertainty also affects the market premium in
aggregate, as well as the CAPM alpha and effective beta in the cross section.
Second, we contribute to the growing literature on the cross-sectional return predictab-

3
ility of the ESG profile. Prior studies document weak return predictability of the overall
ESG rating (e.g., Pedersen et al. (2021)) and mixed evidence based on different ESG prox-
ies (e.g., Gompers et al. (2003); Hong and Kacperczyk (2009); Edmans (2011); Bolton and
Kacperczyk (2020)). Our contribution is to propose that ESG uncertainty could tilt the
ESG-performance relationship and serve as a potential mechanism to explain the opposing
findings. We show that ESG ratings are negatively associated with future performance when
there is little uncertainty and that the ESG-performance relationship could be insignificant or
positive when uncertainty increases. Thus, the sin premium (Hong and Kacperczyk (2009))
and carbon premium (Bolton and Kacperczyk (2020)) could be attributed to the notion that
sin stocks (i.e., companies involved in producing alcohol, tobacco, and gaming) and carbon
emissions are clearly defined and thus subject to minimal uncertainty among investors. On
the other hand, other ESG profiles could be more challenging to measure or rely on non-
standardized information and methodologies, thereby displaying more uncertainty and mixed
evidence on return predictability. A recent work by Pástor, Stambaugh, and Taylor (2021b)
further highlights the distinction between ex ante expected returns and ex post realized re-
turns, and shows that U.S. green stocks outperformed brown stocks during the last decade,
due to unexpectedly strong increases in environmental concerns. While our model is static in
nature and formulates expected returns, we also confirm that our findings are stronger in the
pre-2011 period. This suggests that the equilibrium outcome over longer horizons could be
even stronger than the full sample evidence we document, due to the unexpected outcomes
realized over the last decade.
To the extent that ESG uncertainty will decrease with a better understanding of a firm’s
true ESG profile, our work enriches academic and policy discussions in that context. Despite
the rapid growth in the sustainable investing and ESG data markets,5 the comparability
of ESG information remains a critical issue. Due to the lack of standards governing the
reporting of ESG information, it is not a trivial task to compare the ESG data of two different
companies (Amel-Zadeh and Serafeim (2018)). In addition, the construction of ESG ratings
is nonregulated, and methodologies can be opaque and proprietary, leading to substantial
divergence across data providers (e.g., Mackintosh (2018); Berg et al. (2020)). Our findings
imply that the lack of consistency across ESG rating agencies makes sustainable investing
riskier and hence reduces investor participation and potentially hurts economic welfare.
Our study of the equilibrium implications of ESG uncertainty owes a debt to the in-
novative setup developed by Pástor, Stambaugh, and Taylor (2021a), although our focus is
different. Pástor, Stambaugh, and Taylor (2021a) comprehensively analyze the equilibrium
5
The estimated spending on ESG data was US$617 million in 2019 and could approach US$1 billion by
2021. See, http://www.opimas.com/research/547/detail/.

4
implications of sustainable investing and conduct an analysis of welfare and social impact.
They also account for the possibility that ESG investors can disagree about a firm’s ESG
profile and analyze cases where the market is green neutral or green. Notably, in their setup,
the ESG score is certain because investors are dogmatic about their ESG perceptions and
can observe each other’s perceived ESG values. Relative to their important work, we study
the implications of uncertainty about the corporate ESG profile. In particular, the investors
in our model agree that the ESG scores are uncertain and they also agree on the underlying
distribution of the uncertain scores. The empirical proxy for uncertainty is the dispersion,
or disagreement, across raters. We show that ESG uncertainty affects the equity premium,
investor’s demand for risky assets, economic welfare, and the alpha and beta components of
stock returns.
The remainder of this paper is organized as follows. Section 2 presents the model. Section
3 describes the data and the main variables used. Section 4 empirically examines how
ESG ratings and uncertainty affect investor demand and cross-sectional return predictability.
Section 5 calibrates the model and explores its quantitative implications. The conclusion
follows in Section 6.

2 ESG and Market Equilibrium

The theory section develops the economic setup. We start with a single risky asset, i.e.,
the market portfolio, and a riskless asset. We derive the optimal portfolio and discuss the
implications of uncertainty about the ESG profile for the market premium and welfare. The
single-asset setup is then extended to consider multiple risky assets. We analyze the implic-
ations of ESG uncertainty for the demand of individual stocks, derive an asset pricing model
for the cross section of stock returns, and discuss incremental effects of ESG uncertainty on
the alpha and beta components of returns.

2.1 One Risky Asset

Consider a single-period economy in which an optimizing agent trades at time 0 and liquidates
the position at time 1. Let r̃M denote the random rate of return on the market portfolio in
excess of the riskless rate, rf , and let g̃M denote the true, but unobservable, ESG score of

5
the market portfolio.6 We model the excess market return and the ESG score as

r̃M = µM + ˜M , (1)

g̃M = µg,M + ˜g,M , (2)

where E (r̃M ) = µM is the expected market excess return, E (g̃M ) = µg,M is the expected
value of the market ESG score, and ˜M and ˜g,M are zero-mean residuals. We assume
that the residuals obey a bivariate normal distribution with σM , σg,M , and ρg,M denoting
the standard deviation of return, the standard deviation of ESG score, and the correlation
between residuals, respectively. The correlation ρg,M is assumed to be positive. In particular,
if the agent learns that the market ESG score is higher than previously thought (i.e., ˜g,M
is positive), the price that he would be willing to pay for the market will be revised upward
(positive ˜M ), while a downward price revision applies for a score lower than previously
thought.7
It is also assumed that the agent knows the joint distribution of return and the ESG
score as well as the underlying parameters. In the empirical analysis that follows, µg,M and
σg,M are proxied by the average and standard deviation of ESG ratings across six major
data vendors, respectively. From an investor’s perspective, a higher σg,M indicates more
disagreement among ESG raters and hence more uncertainty about the true ESG profile of
the market.
Following Pástor, Stambaugh, and Taylor (2021a), we consider an optimizing agent
who derives nonpecuniary benefits from holding stocks based on their ESG characteristics.
Moreover, preferences are formulated through the exponential utility (CARA) function
 
V W̃1 , x = −e−AW̃1 −BW0 xg̃M , (3)

where W̃1 = W0 (1 + rf + xr̃M ) is the terminal wealth, W0 is the initial wealth, x is the
fraction of wealth invested in the risky asset, A stands for the agent’s absolute risk aversion,
and B characterizes the nonpecuniary benefits that the agent derives from stock holdings.
Positive (negative) B indicates that the agent extracts benefits from holding green (brown)
stocks. Hence, B can be interpreted as the absolute brown aversion. In the following, we
make the sensible assumption of a nonnegative brown aversion (B ≥ 0). Slightly departing
from Pástor, Stambaugh, and Taylor (2021a), we formulate preferences for ESG to be wealth-
dependent. Then, the expression BW0 represents the relative brown aversion.
6
Consistent with static setups, we do not formulate intertemporal preferences; hence, the riskless rate is
exogenously specified.
7
We thank the referee for suggesting this avenue.

6
 Observe from equation (3) that the investment in the riskless asset does not contribute
to the portfolio’s ESG profile, as perceived by the agent. This is because the riskless asset
is implicitly assumed to be green neutral. As ESG scores are ordinal in nature, the choice
of considering the riskless asset as a reference level does not imply loss of generality. In
addition, to capture the ESG benefits and costs from investing in the market, we allow the
market portfolio to depart from green neutrality.
The agent picks x attempting to maximize the expected value of preferences in equation
(3). The first-order condition suggests that the optimal portfolio in the presence of ESG
uncertainty is given by
1 µM + bµg,M
x∗ = 2
, (4)
γ σM,U

where b = B A
, γ = AW0 stands for the relative risk aversion, and σM,U 2
= σM2
+ b2 σg,M
2
+
2bσM σg,M ρg,M is the variance of return, as perceived by the agent. Henceforth, b is referred
2 2
as brown aversion for brevity. The ex-ante market variance, σM,U , is no longer equal to σM
because, with ESG uncertainty, the risky asset is perceived to be a package of two distinct
securities. The first delivers the market excess return r̃M , while the second reflects exposure
to ESG uncertainty and yields bg̃M . The latter component can be interpreted as investing b
units in a pseudo-asset that pays g̃M per unit. As b increases, i.e., when the ratio between
brown aversion and risk aversion increases, the ESG component becomes more meaningful
2 2
in investment decisions. A sufficient condition for σM,U ≥ σM is that the brown aversion
and the correlation between market return and ESG score are nonnegative (i.e., b ≥ 0 and
ρg,M ≥ 0). As noted earlier, these conditions are likely to be satisfied.
In what follows, we consider a positive market premium (i.e., µM > 0), which is plaus-
ible in the presence of risk aversion. The brown-aversion assumption is sensible for ESG-
perceptive investors. Additionally, to distill the incremental effects of ESG uncertainty, we
consider two benchmark cases. In the first, the agent is ESG indifferent, and in the second,
preference for ESG is accounted for, while the ESG profile is known for certain. The latter
case is studied by Pástor, Stambaugh, and Taylor (2021a) in a multiple-security setup.
Equation (4) presents the optimal stock position in the presence of uncertainty about
the ESG profile. Stock investment is thus driven by the relative risk aversion, γ, and the
price of risk of the portfolio that yields r̃M + bg̃M . To give perspective on the optimal equity
demand, consider the case that incorporates ESG preferences but excludes uncertainty. Then,
the perceived volatility of the stock return is still σM . Conforming to intuition, the demand
for stocks rises as b rises and the market is green. Essentially, stocks are more attractive to
a green-loving agent.
When ESG uncertainty is accounted for, however, this intuition is no longer binding. To

7
illustrate, consider two limiting cases. In the first, b grows with no bound. The investor
µ +bµ
then avoids equities, i.e., lim γ1 Mσ2 g,M = 0. Similarly, when ESG uncertainty rises with no
b→∞ M,U
bound, the demand for stocks evaporates. Thus, both increasing brown aversion and increas-
ing uncertainty translate into increasing equity risk. In the presence of ESG uncertainty, a
brown-averse agent could substantially reduce stock investing, even when the market is green,
on average.
Moving beyond the two limiting cases, we further examine portfolio tilts in the presence
of ESG uncertainty. For that purpose, we rewrite the optimal portfolio as
!
2
∗ 1 µM 1 µg,M 1 µM + bµg,M 2
σg,M σM σg,M ρg,M
x = 2
+ b 2 − 2
b 2 + 2b 2
. (5)
γ σM γ σM γ σM σM,U σM,U

The first term on the right-hand side of equation (5) describes the benchmark case of ESG
bµ
indifference. Preferences for ESG generate the second and third terms. The term γ1 σg,M 2
M
corresponds to the second benchmark case with ESG preferences when the ESG profile is
known for certain. It suggests that as b rises, the demand for risky asset rises and portfolio
tilt intensifies. The third term purely reflects the incremental effect of ESG uncertainty.
σ2
The ratio σ2g,M stands for the contribution of ESG uncertainty to the total, ex ante, market
M,U
variance. Additionally, in the presence of a positive correlation between market return and
the ESG profile, the agent employs the market portfolio to hedge against risk evolving from
σ σ ρg,M
ESG uncertainty, as captured by the hedge ratio M σg,M 2 . Hence, the incremental effect
M,U
of ESG uncertainty on stock investing (captured by the third term) is negative.8
In addition, when the market is green neutral (i.e., µg,M = 0) and when the ESG profile
is known for certain, stock investing is unaffected relative to ESG indifference. In contrast,
when the market is green neutral and ESG uncertainty is accounted for, participation in the
equity market is discouraged, relative to both benchmark cases.
We now turn to analyzing the equilibrium implications of ESG preferences with uncer-
tainty. It is assumed that, in equilibrium, the representative agent’s wealth is fully invested
in the market portfolio. Thus, equalizing the optimal stock allocation in equation (4) to
1 yields the market premium. The market premiums for the cases of ESG indifference (I),
ESG preference with no uncertainty (N ), and ESG preference with uncertainty (U ) are given
8
In the case where µM +bµg,M is negative, the ESG uncertainty effect on stock investing goes the opposite
way. This requires the interaction of extreme brown aversion along with an extreme brown market.

8
by

2
µIM = γσM , (6)
µN 2
M = γσM − bµg,M , (7)
2
µUM = γσM 2
− bµg,M + γ(σM,U 2
− σM ). (8)

Retaining the assumptions of a green market and a brown-averse agent, the market
premium diminishes relative to equation (6), as captured by the second term in equation
(7). This is because, as implied by Pástor, Stambaugh, and Taylor (2021a) in a multi-asset
context, an agent who extracts nonpecuniary benefits from holding green stocks is willing
to compromise on a lower risk premium relative to an ESG-indifferent agent. If the market
is green neutral, the second term disappears; hence, the equity premium is unchanged even
when ESG preferences are accounted for.
Further accounting for uncertainty in equation (8), there are two conflicting forces. On
the one hand, the agent extracts nonpecuniary benefits from holding the green market, a
force leading to diminished market premium. On the other hand, the market is perceived
to be riskier; thus, it commands a higher market premium, as formulated in the third term
of equation (8). The overall effect is inconclusive. If the market is green neutral, the equity
premium increases relative to both benchmark cases due to the increasing risk channel.
The same conflicting forces apply to the equilibrium Sharpe ratio (the slope of the capital
allocation line) when accounting for ESG uncertainty, SRU , relative to ESG indifference,
2
σM,U
U bµ
SRI . Given market return volatility, σM , it follows that SR
SR I = σM2 − γσg,M
2 . The first term
M
is greater than one and reflects the increase in perceived equity risk. The second captures
the decrease in the market premium due to the nonpecuniary benefits from ESG investing.
In the presence of ESG preferences, the market risk premium thus incorporates an ESG
incremental premium that can be defined as

µN I
M − µM = −bµg,M , (9)
µUM − µIM = γ σM,U
2 2


− σM − bµg,M . (10)

The no-uncertainty case is associated with a negative ESG incremental premium when the
market is green and the agent is brown-averse, while the incremental premium is zero when
the market is green neutral. In addition, it is evident from equation (10) that the market
premium increases with ESG uncertainty. Collectively, with ESG uncertainty, the incre-
mental premium is positive when the market is green neutral. Otherwise, with a green
market and a brown-averse agent, the sign of the incremental premium is inconclusive due

9
to the conflicting forces.
The single-security economy establishes a solid benchmark in which to comprehend the
more complex multi-asset setup to be developed later in the text. While the cross-sectional
ESG-alpha relation is negative when ESG uncertainty is not accounted for, the single-security
case provides the first clue that (1) the risk premium increases with ESG uncertainty, and
(2) the risk premium of a green stock could exceed that of a brown stock in the presence
of ESG uncertainty. Taking together, the ESG-alpha relation in the cross section can be
subject to conflicting forces.
Up to this point, we have considered a single-agent economy for ease of exposition. In
what follows, to assess the welfare implications of ESG uncertainty in the aggregate and to
study the multi-asset economy, we extend the framework to account for multiple heterogen-
eous agents. Thus, consider I agents indexed by i = 1, . . . , I, who differ in their initial wealth
Wi,0 , absolute risk aversion Ai , and absolute brown aversion Bi . Market clearing requires
W
that Ii=1 wi x∗i = 1, where wi = Wi,0
P
0
is the fraction of agent i’s initial wealth relative to
aggregate wealth. With heterogeneous agents, the market premium equivalent to equation
(8) is given by
µUM = γM σM,U
2
− bM µg,M,U , (11)
wi γi−1 bi
PI
1
where γM = PI −1 is the aggregate risk aversion, bM = i=1
−1
γM
is the aggregate
i=1 wi γi
−1
2 γM
brown aversion, σM,U = PI −1 −2 is the perceived aggregate variance, and µg,M,U =
i=1 wi γi σi,U
PI −1 −2
i=1 wi bi γi σi,U
−1 −2
bM γM σM,U
µg,M is the perceived aggregate ESG score. Online Appendix A.1 provides
details.
The changing cost of equity capital due to ESG preferences has implications for economic
welfare and social impact. For instance, Pástor, Stambaugh, and Taylor (2021a) show that
when the market is green, the lower cost of equity capital could trigger increasing capital
investment and social impact. In our setup, a green representative firm would be harmed by
the higher cost of equity capital induced by ESG uncertainty, which could trigger adverse
effects on capital investment and social impact. In the calibration experiment described in
Section 5.1, we comprehensively analyze the utility loss attributable to ESG uncertainty. We
also calibrate the market premium, as well as equity demand and welfare for two types of
agents: the first is indifferent to ESG, while the other is ESG perceptive and recognizes the
uncertainty about the sustainability profile.

10
2.2 A Multi-Asset Economy
We move on to formulate an economy populated with I optimizing agents, N risky assets,
and a riskless asset. We aim to derive an asset pricing model for the cross section of equity
returns in the presence of ESG uncertainty, while we also extend the analysis of portfolio
selection.
We model the excess returns and ESG scores on N assets as

r̃ = µr + ˜r , (12)
g̃ = µg + ˜g , (13)

where µr is an N -vector of expected excess returns and µg is an N -vector of expected

ESG scores. The residuals from both equations are assumed to obey a 2N -variate normal
distribution. The N × N covariance matrices of returns and ESG ratings are denoted by Σr
and Σg , respectively, while Σrg is the N × N cross-covariance matrix between r̃ and g̃ with
diagonal elements that are assumed to be positive.  
Similar to equation (3), the agent maximizes an exponential utility function, V W̃i,1 , Xi =
0
−e−Ai W̃i,1 −Bi Wi,0 Xi g̃ , where W̃i,1 = Wi,0 (1 + rf + Xi0 r̃) is the terminal wealth and Xi is the
N -vector of portfolio weights per investor i.
Proposition 1 describes the optimal portfolio in the presence of multiple risky assets. The
proof is in Online Appendix A.2.

Proposition 1. The optimal portfolio strategy of investor i is given by

1 −1
Xi∗ = Σ (µr + bi µg ) , (14)
γi i,U

where Σi,U = Σr + b2i Σg + 2bi Σrg is the covariance matrix of r̃ + bi g̃.

This portfolio strategy is the multi-asset version of equation (4). It suggests that in
the presence of ESG preferences, investors perceive asset excess returns to be the sum of
(1) N stock excess returns r̃ and (2) N returns on pseudo-assets yielding bi g̃. Several
implications are in order. First, infinitely brown-averse agents act as if they were infinitely
risk averse, as, in the presence of ESG uncertainty, lim Xi∗ = 0. Second, in another extreme
bi →∞
case when ESG uncertainty grows with no bound for all stocks, economic agents avoid
stocks altogether. Third, in intermediate cases, uncertainty about ESG profiles nonlinearly
intervenes in formulating the optimal portfolio, through the inverse of Σi,U , and tends to
reduce the demand for both green and brown stocks.

11
 To highlight the incremental implications of ESG uncertainty for portfolio selection, we
rewrite equation (14) as

1 −1 1
Xi∗ = Σr (µr + bi µg ) + Ψi (µr + bi µg ) , (15)
γi γi
−1
where Ψi = −Σ−1 2 −1 2 −1
r (bi Σg + 2bi Σrg ) Σr (IN + (bi Σg + 2bi Σrg ) Σr ) and IN stands for the
N × N identity matrix.
The first term of the optimal portfolio coincides with the strategy in Pástor, Stambaugh,
and Taylor (2021a) (equation (4)). The second term is exclusively attributable to ESG
preferences with uncertainty about the sustainability profile. Interestingly, in the presence
of heterogeneous agents, the ESG uncertainty term precludes fund separation because the
incremental portfolio, evolving from ESG uncertainty, is agent specific.
In particular, consider the alternative decomposition of the optimal portfolio:

λr −1 Σ−1 µr λg Σ−1r µg
Xi∗ = Γi 0 r −1 + bi Γ−1
i , (16)
γi 1 Σr µr γi 10 Σ−1
r µg

where λr = 10 Σ−1 g 0 −1 2 −1 −1
r µr , λ = 1 Σr µg , and Γi = IN + bi Σr Σg + 2bi Σr Σrg .
The decomposition shows that each optimizing agent holds three portfolios: (1) a riskless
asset, (2) the maximum Sharpe ratio portfolio in the risk-return space, and (3) the maximum
Sharpe ratio portfolio in the risk-ESG space. Note that ESG uncertainty affects the demand
for risky assets through the N × N matrix Γi , which enters both risky asset portfolios. If all
agents have the same bi , then the matrix Γi is common to all agents and, therefore, a three-
fund separation results. Otherwise, the two risky portfolios are agent specific and, hence,
fund separation does not apply in the setup of heterogeneous agents with ESG uncertainty.

2.3 CAPM with ESG Uncertainty

The next two propositions illustrate the cross-sectional asset pricing implications of ESG
preferences, first excluding and then accounting for ESG uncertainty. The proofs are in
Online Appendix A.3 and A.4.

Proposition 2. Excluding ESG uncertainty, the equilibrium expected excess returns of the
risky assets are given by
µr = βµM − bM (µg − βµg,M ) , (17)
2 2 0
where µM = γM σM − bM µg,M is the equilibrium market premium, σM = XM Σr XM is the
Σr XM 0
market return variance, β = σ2 is the N -vector of market beta, µg,M = XM µg is the
M

12
aggregate market greenness, XM = Ii=1 wi Xi is the N -vector of aggregate market positions
P

in risky assets, γM is the aggregate risk aversion, and bM is the aggregate brown aversion.
In the absence of ESG uncertainty, the expected excess return expression in equation
(17) is identical to that derived by Pástor, Stambaugh, and Taylor (2021a), with a slight
modification that the market can depart from green neutrality. Expected returns are affected
by ESG preferences through (1) the modified market premium and (2) the alpha component
that stands for excess return unexplained by βµM . Alpha depends on the effective ESG
score, i.e., the difference between the firm’s own ESG score and the market ESG score
multiplied by the stock’s beta. A numerical example is useful to illustrate. Assume a stock
with β = 1.2 and µg,M = 2. As long as the ESG score is below 2.4, the stock has a positive
alpha even when the stock is green. The threshold value 2.4 reflects zero alpha, while alpha
turns negative if the ESG score goes above the threshold. For instance, if the ESG score is
3 (2), the effective ESG score is 0.6 (−0.4), and alpha is negative (positive). Altogether, as
long as the market is not green neutral, it is not the firm’s own ESG score that dictates the
sign and magnitude of alpha. Instead, it is the effective ESG score.
In the presence of ESG preferences and certainty about the ESG profile, the beta meas-
uring exposure to total market risk, β, coincides with the CAPM beta. This is because, as
noted earlier, the perceived return on any security is equal to the sum of (1) the actual return
and (2) the pseudo-asset return that is proportional to the ESG score, while the ESG score
is nonrandom. Thus, in the absence of ESG uncertainty, the covariance and variance terms
used to define beta are unchanged. With uncertainty, the ESG score is random; hence, the
resulting beta is no longer identical to the standard CAPM beta.
As proposed by Pástor, Stambaugh, and Taylor (2021a), in the absence of ESG uncer-
tainty, equilibrium expected returns compensate for exposure to (1) the market risk factor
and (2) an ESG-based factor. When ESG uncertainty is in play, fund separation no longer
results; thus, expected returns cannot be represented through a multifactor model. Instead,
we propose a CAPM-type representation, where expected excess returns are expressed as
the sum of two components: the first reflects the exposure to the market factor, while the
second is a nonzero alpha that stands for (1) nonpecuniary benefits from ESG investing and
(2) an additional risk premium attributable to ESG uncertainty. The following proposition
explains the equilibrium expected returns with ESG uncertainty, which is the core of our
analysis.
Proposition 3. With ESG uncertainty, the equilibrium expected excess returns of the risky
assets are formulated as

µr = βµM + (βeff − β) µM − bM (µg,U − βeff µg,M,U ) , (18)

13
 2 Σr XM
where µM = γM σM,U − bM µg,M,U is the equilibrium market premium, β = σM2 the N -
ΣM,U XM −1
vector of the equilibrium CAPM beta, βeff = 2
σM,U
the N -vector of effective beta, ΣM,U =
PI −1 −1
i=1 wi γi Σi,U
PI −1 the inverse of the covariance matrix of ESG-adjusted perceived asset returns.
i=1 wi γi
BM µg
µg,U = bM is the perceived aggregate ESG scores of individual assets, where BM =
( Ii=1 wi γi−1 Σ−1
P −1
PI −1 −1 0
i,U ) i=1 wi γi bi Σi,U , and µg,M,U = XM µg,U is the perceived aggregate mar-
ket ESG score. Online Appendix A.4 displays simplified expressions for asset pricing with
ESG uncertainty assuming homogeneous agents.

The expected excess return expression in equation (18) modifies the no-uncertainty case
in equation (17) by replacing the market beta with the effective beta. Thus, it is the effective
beta that is priced in the cross section of equity returns. To give perspective on the notion
of effective beta, note that the perceived return on an arbitrary asset still consists of two
components: (1) the actual return and (2) b times the ESG score of that asset. Because ESG
scores for the market and individual assets are random, both the covariance and variance
terms, used to define beta, depart from the standard return-based counterparts. The effective
beta is based on ESG-adjusted returns. Collectively, expected excess returns on N risky
assets are formulated as the sum of three terms. The first term reflects exposure to market
risk, as in the standard CAPM. Then, the difference between the effective beta and the
market beta gives rise to the second term. The third term accounts for the uncertainty-
adjusted effective ESG scores, analogously to equation (17) but using the effective beta
instead.
To provide further intuition on the beta-pricing specification, we consider a simplified
case in which agents have homogeneous preferences (γ and b are equal across agents). The
effective beta can then be represented as

2
σM b2 σg,M
2
2bσrg,M
βeff = 2
β + 2
β g + 2
βrg , (19)
σM,U σM,U σM,U

where βg = Σσg2XM , βrg = Σσrgrg,M

XM 2
, and σM,U 2
= σM + b2 σg,M
2
+ 2bσrg,M . The effective beta
g,M
is a weighted average of (1) the CAPM beta, β, (2) the ESG uncertainty beta, βg , and
(3) the ESG-return cross-covariance beta, βrg . The ESG uncertainty beta represents the
co-movement between the asset’s own ESG uncertainty and the market ESG uncertainty.
The cross-covariance beta represents the asset’s contribution to the aggregate ESG-return
cross covariance, σrg,M . The weights in equation (19) reflect the relative contributions to
the perceived market return variance, i.e., the actual return, the ESG component, and the
cross-covariance component.
The asset’s effective beta coincides with its market beta if preferences for ESG are muted

14
(b = 0) or if the market is not subject to ESG uncertainty (σg,M = σrg,M = 0). To provide
more intuition about the dependence of the effective beta on ESG uncertainty, consider the
2
case where the covariance matrix of ESG uncertainty, Σg , is diagonal with elements σg,j
(j = 1, . . . , N ), while Σrg is diagonal with elements σrg,j . The effective beta of asset j can
be written as
σ2 b2 σg,M
2 2
Xj σg,j 2bσrg,M Xj σrg,j
βeff ,j = 2M βj + 2 2
+ 2 . (20)
σM,U σM,U σg,M σM,U σrg,M
Given positive market weights in equilibrium, Xj > 0, the effective beta increases with
2
the asset’s own ESG uncertainty, σg,j , and with the covariance between firm’s ESG and
return, σrg,j , while it does not depend on the mean ESG score. Interestingly, as long as the
aggregate ESG uncertainty is nonzero, a positive market beta asset with certain ESG profile
σ2
(σg,j = σrg,j = 0) has an effective beta equal to σ2M βj , which is lower than the market beta
M,U
βj . This is because the asset contributes to the aggregate return-based risk, but not to the
aggregate ESG uncertainty.
We next analyze alpha variation with ESG uncertainty in the case of homogeneous agents.
Combining equations (18) and (19), we show in Online Appendix A.4 that the CAPM alpha
can be expressed as
!
b2 σg,M
2
2bσrg,M
α= 2
(βg − β) + 2 (βrg − β) (µM + bM µg,M ) − bM (µg − βµg,M ) . (21)
σM,U σM,U

The second term on the right-hand side of equation (21) is identical to that in equation (17)
when ESG uncertainty is excluded. The first term represents the incremental effect of ESG
uncertainty and is further analyzed below. For ease of interpretation, we assume again that
Σg and Σrg are diagonal. Then, it follows that
! !
b2 σg,M
2 2
Xj σg,j

2bσrg,M Xj σrg,j
αj = 2 2
− βj + 2 − βj (µM + bM µg,M )
σM,U σg,M σM,U σrg,M
− bM (µg,j − βj µg,M ) . (22)

Given positive market portfolio weights in equilibrium, Xj > 0, the asset alpha increases
2
with ESG uncertainty, σg,j . Likewise, alpha increases with the asset ESG-return cross cov-
ariance, σrg,j . Additionally, in the presence of aggregate ESG uncertainty, a positive market
beta asset with zero effective ESG score (µg,j − βj µg,M = 0) and with certain ESG profile
has a negative alpha, because its effective beta in equation (20) is smaller than its market
beta, as noted earlier.
We have shown that both alpha and the effective beta rise with ESG uncertainty. The

15
analysis is based on the simplifying assumption of homogeneous brown-averse agents. Re-
laxing the homogeneity assumption, alpha and beta variations with ESG uncertainty appear
quite complex to analyze analytically. However, in the calibration developed in Section 5.2,
we consider heterogeneous agents in a two-asset economy (both brown and green) and show
that, even then, alpha and the effective beta do increase with ESG uncertainty. Below, we
provide further analytical results for the two-asset economy for ease of interpretation.

2.4 Demand and Expected Return in a Two-Asset Economy

In particular, to gain additional intuition about the demand for multiple risky assets and
their equilibrium expected returns, it is useful to consider a simplified economy consisting
of two risky assets (along with a riskless asset), both green and brown. In that economy,
expected excess returns are denoted by µr,green for the green stock and µr,brown for the brown,
2
the corresponding ESG scores are µg > 0 and −µg , the variances of the ESG scores are σg,green
2
and σg,brown , and the correlation between the scores is assumed to be zero. Asset returns are
assumed to be uncorrelated with identical variance denoted by σr2 . Finally, ESG scores are
assumed to be positively correlated with returns of the same asset, with covariances denoted
by σrg,green and σrg,brown . The expressions below follow from Propositions 1 and 3. Online
Appendix A.5 provides further details.
The two-asset optimal strategy is formulated as

∗ 1 µr,green + bi µg
Xi,green = , (23)
γi σr2 + b2i σg,green
2 + 2bi σrg,green
∗ 1 µr,brown − bi µg
Xi,brown = 2 2 2
. (24)
γi σr + bi σg,brown + 2bi σrg,brown

The optimal portfolio illustrates that, for ESG-sensitive agents (bi > 0), demand falls with
higher ESG uncertainty but rises with higher ESG scores. The notion is that when targeting
an ESG level, uncertainty about the precise ESG profile should be accounted for. As in
the single-asset setup, the effect of ESG uncertainty is amplified by the positive correlation
between return and the ESG score. For ESG-indifferent agents (bi = 0), the demand for
green and brown stocks is equal to the mean-variance demand when ESG preferences are
excluded.
We next formulate expected excess returns in equilibrium. We denote the fraction and
brown aversion of ESG-sensitive investors by wESG and bESG > 0, while the corresponding
parameters of ESG-indifferent agents are wIND = 1 − wESG and bIND = 0. Assuming that
all agents have the same relative risk aversion γ, expected excess returns on the green and

16
brown assets are formulated as
σ2
 
2 σrg,green
βgreen γσM 1 + b2ESG g,green
σr2 + 2b ESG σr2 − wESG bESG µg
µr,green = 
σ2
 , (25)
σrg,green
1 + (1 − wESG ) b2ESG g,greenσr2
+ 2b ESG σr2
 σ 2 
2 σ
βbrown γσM 1 + b2ESG g,brown
σr2
+ 2bESG rg,brown
σr2
+ wESG bESG µg
µr,brown =  σ 2  , (26)
σrg,brown
1 + (1 − wESG ) b2ESG g,brown σ 2 + 2b ESG σ 2
r r

where βgreen and βbrown are the equilibrium CAPM betas. In the limiting case where wESG = 0
or bESG = 0, all agents are ESG indifferent and equilibrium expected excess returns boil
2 2
down to βgreen γσM and βbrown γσM . In the opposite extreme, where wESG = 1, expected
return diminishes with the ESG score and rises with ESG uncertainty. The latter force can
magnify the required return to the extent that a green asset could, possibly, deliver higher
return than a brown asset.
Otherwise, in the intermediate case where both ESG-sensitive and ESG-indifferent agents
populate the economy, the expected return difference between the brown and the green
assets diminishes with ESG uncertainty. To see why, consider two assets with identical beta
(βgreen = βbrown = β) and ESG uncertainty (σg,green = σg,brown = σg and σrg,green = σrg,brown =
σrg ). The expected return gap (also the alpha gap) is given by

2wESG bESG µg
µr,brown − µr,green = 
σg2
. (27)
2 σrg
1 + (1 − wESG ) bESG σ2 + 2bESG σ2
r r

When all agents are ESG sensitive (wESG = 1 and bESG > 0), the difference in expected re-
turns is independent of ESG uncertainty and equal to 2bESG µg . Put another way, controlling
for ESG uncertainty and beta, the expected return gap between the brown and the green
assets is fixed, reflecting the nonpecuniary benefits from holding green assets. The return
gap is nonexistent when either bESG = 0 or wESG = 0, as all agents are ESG indifferent.
Otherwise, when ESG preferences are heterogeneous, the expected return gap mono-
tonically decreases with σg2 and σrg .9 This suggests that ESG uncertainty could weaken the
negative ESG-performance relation, as the asset demand of ESG-sensitive agents diminishes,
which, in turn, implies lower aggregate nonpecuniary benefits from ESG investing. In the
limit, when ESG uncertainty grows with no bound, the expected return gap between green
and brown assets is approaching zero.
9
The no-uncertainty case leads to µr,brown − µr,green = 2bM µg , where bM = wESG bESG .

17
3 Data
3.1 Data Sources
Our sample consists of all NYSE/AMEX/Nasdaq common stocks with share codes 10 or 11;
daily and monthly stock data are obtained from the Center for Research in Security Prices
(CRSP). We collect ESG rating data from six data vendors, including Asset4 (Refinitiv),
MSCI KLD, MSCI IVA, Bloomberg, Sustainalytics, and RobecoSAM. These data providers
represent the major players in the ESG rating market, and their ratings are widely used
by practitioners as well as a growing number of academic studies (e.g., Eccles and Stroehle
(2018); Berg et al. (2020); Gibson et al. (2020)).
Quarterly and annual financial statement data come from the COMPUSTAT database.
Analyst forecast data come from the Institutional Brokers’ Estimate System (I/B/E/S). We
also acquire quarterly institutional equity holdings from the Thomson-Reuters Institutional
Holdings (13F) database.10 The full sample period ranges from 2002 to 2019. Our sample
begins in 2002, as we require ESG ratings from at least two data vendors.

3.2 Main Variables

We focus on the overall ESG rating from each data provider, i.e., “ESG Combined Score”
from Asset4, “ESG Rating” from MSCI IVA, “ESG Disclosure Score” from Bloomberg, “Sus-
tainalytics Rank” from Sustainalytics, and “RobecoSAM Total Sustainability Rank” from
RobecoSAM.11 For MSCI KLD data, we construct an aggregate ESG rating by summing all
strengths and subtracting all concerns (e.g., Lins et al. (2017); Berg et al. (2020)).
ESG rating agencies may differ in sample coverage and rating scale. Panel A of Online
Appendix Table B.1 reports the number of U.S. common stocks covered by each data vendor
over time. In addition, Asset4, Bloomberg, Sustainalytics, and RobecoSAM apply a scale
from 0 to 100, MSCI IVA uses a seven-tier rating scale from the best (AAA) to the worst
(CCC), and the MSCI KLD rating ranges from −11 to +19 in our sample. Panel B further
demonstrates that requiring a common sample covered by all data vendors could significantly
reduce the sample size and shorten the sample period. Therefore, we focus on pairwise ESG
rating disagreement and then average across all rater pairs. Note that the ESG uncertainty
10
The institutional ownership data come from money managers’ quarterly 13F filings with the U.S. Se-
curities and Exchange (SEC). The database contains the positions of all institutional investment managers
with more than $100 million U.S. dollars under discretionary management. All holdings worth more than
$200,000 U.S. dollars or 10,000 shares are reported in the database.
11
Although the Bloomberg ESG disclosure score measures the extent of disclosure of ESG-related data by
a company, it is positively associated with ESG quality due to the largely voluntary nature of ESG disclosure
requirements (López-de-Silanes et al. (2020)).

18
in our model is motivated by the fundamental difficulty and lack of consensus in measuring
and interpreting the true ESG profile. The disagreement among ESG raters is largely due
to the lack of consensus on the scope and measurement of ESG performance (Berg et al.
(2020)), and as a result, investors cannot reliably observe the firm’s true ESG profile and are
exposed to uncertainty in their sustainable investment. Hence, we employ the disagreement
among ESG raters as a proxy for uncertainty about a firm’s ESG profile and label such
disagreement ESG uncertainty to be consistent with the model terminology.
Specifically, we obtain 14 rater pairs from the six data providers.12 To achieve comparab-
ility across rating agencies, we proceed as follows. For each rater pair-year, we sort all stocks
covered by both raters according to the original rating scale of the respective data provider
and calculate the percentile rank (normalized between 0 and 1) for each stock-rater pair.
Then, for each stock, we compute the pairwise rating uncertainty as the sample standard
deviation of the ranks provided by the two raters in the pair. Specifically, let gj,t,A and gj,t,B
denote the ESG rank for stock j in year t from raters A and B, respectively. The pairwise
|gj,t,A√−gj,t,B | 13
rating uncertainty is calculated as 2
. For perspective, a company that is ranked
by two data providers at the 33rd and 59th percentiles would generate a rating uncertainty
of 0.18.
Finally, we compute the firm-level ESG rating uncertainty as the average pairwise rating
uncertainty across all rater pairs. Similarly, we compute the pairwise average rank and then
average across all rater pairs to obtain the firm-level ESG rating. Notably, the pairwise
measure has the advantage of maximizing the use of available rating information while still
preserving comparability across raters, despite the difference in their sample coverage.14 In
addition, investors may not have access to all six data vendors, therefore the average pairwise
rating level and rating uncertainty provide an approximate assessment for the perceived ESG
profile and rating uncertainty among investors. As a robustness check, we also consider
12
There are 14 (instead of 15) rater pairs because MSCI KLD data are only available until 2015, while
RobecoSAM data start in 2016, as shown in Panel A of Online Appendix Table B.1.
13
r To illustrate, consider two ratings g1 and g2 . The pairwise rating uncertainty is given by
2 2
(g1 − g1 +g2
)+(g2 −
g1 +g2
)
2
2−1 = |g1√−g
2
2
2|
.
14
Relative to standard economic measures that are cardinal and can be directly compared, ESG scores
are ordinal in nature. Thus, ESG scores are sensitive to the sample coverage considered by the particular
data vendor. Indeed, as ESG rating agencies differ in their sample coverage (Panel A of Online Appendix
Table B.1), the stand-alone rank (e.g., 90th percentile) provided by one rater may not be directly comparable
with the corresponding figure from another rater, if, for instance, one rater covers, on average, more green
firms. To ensure comparability across all vendors covering a stock, a proper experiment for determining
the stock-level average ESG rating and rating uncertainty is to narrow down the focus to only those stocks
jointly covered by all vendors. This experiment, however, could considerably shrink the sample, which reflects
the coverage intersection of all vendors providing a rating for the stock. In contrast, the pairwise measure
requires only a minimal set of restrictions on common coverage, and, hence, allows us to explore the richness
in ESG ratings provided by each data vendor, while still preserving comparability across vendors.

19
alternative proxies for ESG rating (ESGALL ) and rating uncertainty (ESG UncertaintyALL )
using all ESG ratings from all raters (instead of rater pairs), without requiring common
coverage, at a given point in time. Online Appendix Table B.2 provides detailed definitions
for each variable.
Panels A and B of Online Appendix Table B.3 present the pairwise ESG uncertainty and
correlation of ESG ratings, respectively. The average correlation across all rater pairs is 0.48
and ranges from 0.25 to 0.71. MSCI KLD and MSCI IVA exhibit the lowest correlation and
the highest rating disagreement with other raters, and the average correlation is 0.38 and
0.34, respectively. On the other hand, ratings provided by Sustainalytics and RobecoSAM
are more correlated with those of other raters, and the average correlation is 0.59 and 0.56,
respectively. Our findings are largely consistent with the existing literature and echo the
growing concerns related to the lack of agreement across ESG rating agencies (e.g., Chatterji
et al. (2016); Amel-Zadeh and Serafeim (2018); Berg et al. (2020); Gibson et al. (2020)).
Panel C of Online Appendix Table B.3 reports the summary statistics for the stock-level
data used in the paper. We report the mean, standard deviation, median, and quantile distri-
bution of the annual ESG rating and ESG rating uncertainty and other stock characteristics.
The average ESG rating is 0.46, and the ESG rating uncertainty is 0.18. In addition, to
study the demand for risky assets and the cross section of equity returns, we construct 25
equity portfolios independently sorted on the ESG rating and rating uncertainty. The av-
erage ESG rating, ESG rating uncertainty, and monthly return are presented in Panel D of
Online Appendix Table B.3.
In addition, we examine the market ESG uncertainty throughout the sample period,
as well as the time trend in ESG uncertainty at the market and individual stock level.
While ESG data vendors do not provide a direct assessment for the market ESG profile, we
evaluate the value-weighted ESG score of the U.S. market by using firm-level ratings per
the different vendors. To preserve comparability across data vendors, we rely on the same
pairwise measures used at the single-stock level.15 For each stock-rater-year, we average the
percentile ranks corresponding to the specific rater across all rater pairs covering this stock.
For each rater-year, we then value-weight firms’ ESG average percentile ranks to obtain a
rater-specific market-level ESG rating. Finally, for each year, using all rater-specific market
ESG ratings, we evaluate the aggregate market-level ESG rating and rating uncertainty as
the pairwise mean and standard deviation across raters.
In Figure 1, the top graph plots the time-series of the market ESG ratings corresponding
to each data vendor, and we observe a significant dispersion across vendors. The bottom
15
In unreported analysis, we confirm that the alternative measurement method described above (ESGALL
and ESG UncertaintyALL ) provides similar results.

20
graph shows the time-series of market ESG uncertainty, as well as equal- and value-weighted
average of stock-level ESG uncertainty. Stock-level ESG uncertainty, on average, diminishes
during the first half of the sample, as the number of raters increases and their coverage
widens. Stock-level uncertainty remains stable in the second subperiod. Focusing on the
market, as ESG ratings are correlated across firms and vendors, the evidence indicates that
the market ESG uncertainty does consistently prevail throughout the entire sample period.
This further supports our intuition that ESG uncertainty could play an important role in
asset pricing.

4 Investor Demand, Stock Return, and Alpha

4.1 Investor Demand
We start with the first testable hypothesis generated from the model, i.e., investor demand
for risky assets increases with the ESG score, consistent with Pástor, Stambaugh, and Taylor
(2021a), while it diminishes with ESG rating uncertainty, as formulated in Proposition 1 and
equations (23) and (24). We rely on institutional ownership as a proxy for the demand for
ESG investment, as Pedersen et al. (2021) document that institutional investors incorporate
ESG when forming their portfolios. While retail investors could still have ESG preference, it
is highly costly to obtain and analyze the ESG information, especially when even the most
specialized raters do not agree, on average, on the firm ESG profile. Due to the complex
nature of ESG investment, retail investors often rely on financial institutions to achieve their
ESG target, thereby making institutional ownership a reasonable source to investigate the
ESG demand. For instance, Hartzmark and Sussman (2019) show that once Morningstar
published sustainability ratings for mutual funds, there was a massive shift of fund flows from
low-sustainability funds to high-sustainability ones. A recent study on Robinhood investors
also documents that retail investors do not respond to ESG disclosures (Moss et al. (2020)).
To test the model predictions based on ESG-sensitive investors, it is also critical to ac-
count for the heterogeneity among institutions, as they are subject to different social norm
pressures and apply various strategies to make socially responsible investments. For instance,
pension funds, universities, religious organizations, banks, and insurance companies are more
norm-constrained than hedge funds or mutual funds that are natural arbitrageurs (Hong and
Kacperczyk (2009)). We therefore consider three distinct groups: norm-constrained institu-
tions, hedge funds, and other institutions. Specifically, we disaggregate the 13F institutional
holdings based on institution type, including bank trust (type 1), insurance company (type
2), investment company (type 3), independent investment advisor (which includes hedge

21
funds, type 4), and others (including corporate/private pension funds, public pension funds,
university and foundation endowments, and miscellaneous, type 5), following Abarbanell
et al. (2003).16 We follow Hong and Kacperczyk (2009) to consider types 1, 2, and 5 as
norm-constrained institutions. Our data on hedge fund holdings are constructed by match-
ing the 13F institutional holdings with a manually collected list of the names of hedge fund
companies.17 The remaining institutions are mostly mutual funds.18
The analysis proceeds as follows. At the end of each year t, we independently sort stocks
into quintile portfolios based on their ESG rating and rating uncertainty to generate 25 (5×5)
portfolios. The low- (high-) ESG-rating and ESG-rating-uncertainty portfolios comprise
the bottom (top) quintile of stocks based on the ESG rating and ESG rating uncertainty,
respectively. For each type of institution, we compute the average institutional ownership in
each quarter in year t + 1 for each of the 25 portfolios, and rebalance the portfolios at the
end of year t + 1. We report the time-series averages of quarterly institutional ownership for
each of the 25 portfolios and the average difference in institutional ownership between high-
and low-ESG-rating portfolios (“HML-R”) as well as between high- and low-ESG-rating-
uncertainty portfolios (“HML-U”). The standard errors in all estimations are corrected for
autocorrelation using the Newey and West (1987) method.
We tabulate the results in Table 1, with Panel A for the stock ownership from norm-
constrained institutions, Panel B for hedge funds, and Panel C for other institutions. Several
findings are worth noting in Panel A. First, as expected, norm-constrained institutions are in
favor of greener firms. For instance, they hold 17.7% of the brown stocks (i.e., stocks in the
bottom ESG rating quintile) while they hold 23.0% of the green stocks (i.e., stocks in the top
ESG rating quintile), indicating a 30% increase. Second, the ownership gap between low-
and high-ESG-rating portfolios attenuates when rating uncertainty increases. When uncer-
tainty is low, green stocks display 5.8% higher institutional ownership than brown stocks,
while the ownership gap declines to an insignificant 0.2% when rating uncertainty is high.
More importantly, this pattern is due to a decline in the demand for green firms when ESG
uncertainty is high, and the difference is statistically significant and economically mean-
ingful. For instance, among the high-ESG-rating portfolios, norm-constrained institutions
hold 22.8% of the low-uncertainty stocks while only 18.1% of the high-uncertainty stocks,
indicating a 21% decline. In line with our working hypothesis, demand for green firms from
16
We thank Brian Bushee for making the institutional investor classification data available via his website:
https://accounting-faculty.wharton.upenn.edu/bushee/.
17
We thank Vikas Agarwal for generously sharing the data. A detailed description of the hedge fund list
is provided by Agarwal et al. (2013).
18
While mutual funds and hedge funds are increasingly subject to social norm pressures as shown by the
rapid growth of ESG investment, some may still prioritize financial returns at the cost of lower ESG standard.
However, this remains an empirical question that we directly test.

22
norm-constrained institutions diminishes with ESG rating uncertainty, suggesting that rat-
ing uncertainty matters the most for ESG-sensitive investors in their ESG investment (i.e.,
green stocks).19
Panel B reports similar statistics for hedge fund ownership. Hedge funds invest more in
brown stocks on average, e.g., they hold 15.7% of the brown stocks but hold 12.7% of the green
stocks.20 The ownership gap between low- and high-ESG rating portfolios tends to diminish
as ESG rating uncertainty rises. For high-uncertainty stocks, the ownership gap is no longer
significant. Unlike the case of norm-constrained institutions, rating uncertainty mostly affects
hedge fund holdings for brown stocks. For instance, within the lowest rating group, hedge
funds hold 15.7% of the low-uncertainty stocks but 13.0% of the high-uncertainty stocks,
indicating a 17% decline. Despite the different incentives for hedge funds to implement
sustainable investment, we continue to find that the rating uncertainty matters the most for
investors in their preferred investment universe.
As shown in Panel C, we do not find strong ESG preference among other institutions.
Conditional on the level of ESG rating, we find evidence that rating uncertainty reduces
investor demand, while the economic magnitude is much smaller than in the previously
discussed subsamples for norm-constrained institutions and hedge funds.
Overall, our findings support the model prediction that for ESG-sensitive investors, de-
mand for risky assets increases with the ESG score but diminishes with ESG rating un-
certainty. Our findings suggest that although institutional investors are likely to be more
sophisticated and have access to privileged information, the uncertainty about corporate
ESG profile remains an important barrier to their investment. This could further limit their
capacity to engage in ESG issues and improve the ESG performance of the firm (e.g., Dim-
son et al. (2015); Dyck et al. (2019); Chen et al. (2020); Krueger et al. (2020)). As more
institutions seek sustainable investing, it is likely that ESG-induced investor demand will
play an even more prominent role in the future.

4.2 Cross-Sectional Return Predictability

In line with Pástor, Stambaugh, and Taylor (2021a), our model predicts a negative rela-
tionship between the ESG rating and CAPM alpha when there is no uncertainty in ESG
19
Perhaps not surprisingly, investor demand is less affected among other ESG rating groups, as such
investment may not be entirely ESG-driven; hence, the rating uncertainty plays a lesser role in asset allocation
decisions.
20
Note that hedge funds can take both long and short positions, hence the long position per se may not
fully reflect the ESG preference of hedge funds. Unreported results examine the net hedge fund ownership,
defined as the hedge fund ownership minus the short interest, where the short interest is computed as the
number of shares held short scaled by the number of shares outstanding (Jiao et al. (2016)). The net hedge
fund ownership is 10.3% for brown stocks and 9.4% for green stocks.

23
ratings (Proposition 2). Negative return predictability stems from nonpecuniary benefits
from holding green stocks. However, the ESG-alpha relationship is less clear in the presence
of ESG uncertainty due to the conflicting forces of the uncertainty-adjusted stock beta and
ESG rating (Proposition 3).
We assess return predictability using a conventional portfolio sort. In particular, at the
end of each year t, we sort stocks into quintile portfolios based on their ESG rating uncer-
tainty. Within each rating uncertainty group, we further sort stocks into quintile portfolios
according to their ESG ratings and generate 25 (5 × 5) portfolios.21 The low- (high)-ESG-
rating and ESG-rating-uncertainty portfolios comprise the bottom (top) quintile of stocks
based on the ESG rating and ESG rating uncertainty, respectively. For each of the 25 port-
folios, we compute the value-weighted return in each month in year t + 1 and rebalance the
portfolios at the end of year t + 1. Within each quintile of portfolios sorted by ESG rating
uncertainty, we also implement the zero-cost trading strategy by taking long positions in the
bottom quintile of stocks (lowest ESG rating) and selling short stocks in the top quintile
(highest ESG rating). The payoff of the long-short investment strategy is computed as the
low (bottom quintile) minus high (top quintile) portfolio return (“LMH-R”), indicating the
return predictability of ESG ratings after controlling for rating uncertainty. We then report
the time-series averages of monthly returns for each of the 25 portfolios and the long-short
strategy.
In addition to raw portfolio returns, we report risk-adjusted returns from (1) the CAPM,
i.e., only adjusting for the market factor (MKT, defined as the excess return on the value-
weighted CRSP market index over the one-month Treasury bill rate); (2) the Fama-French-
Carhart 4-factor model (FFC) consisting of the market factor (MKT), the size factor (SMB,
defined as small minus big firm return premium), the book-to-market factor (HML, defined as
the high book-to-market minus the low book-to-market return premium) (Fama and French
(1993)), and Carhart (1997) the momentum factor (MOM, defined as the winner minus loser
return premium); and (3) the Fama-French 6-factor model (FF6) consisting of the market
factor (MKT), the size factor (SMB), the book-to-market factor (HML), the profitability
factor (RMW, defined as the robust minus weak return premium), the investment factor
(CMA, defined as the conservative minus aggressive return premium), and the momentum
factor (MOM) (Fama and French (2018)).22 The standard errors in all estimations are
corrected for autocorrelation using the Newey and West (1987) method.
Table 2 reports the results, with Panel A for raw return and Panel B for CAPM-adjusted
21
We employ a conditional sort to better control for rating uncertainty, while an independent sort yields
similar findings (Online Appendix Table B.3, Panel D3).
22
We thank Kenneth French for making the common factor returns available via his website: https:
//mba.tuck.dartmouth.edu/pages/faculty/ken.french/data_library.html.

24
return. In the interest of brevity, we tabulate the results of FFC-adjusted return and FF6-
adjusted return in the Online Appendix Table B.4 and only discuss the main findings in
this subsection. Several findings are worth noting. First, the ESG rating is negatively
associated with future performance among stocks with low rating uncertainty, and the long-
short portfolio return is significant at 0.59% per month. Brown stocks (i.e., stocks in the
bottom ESG rating quintile) continue to outperform green stocks (i.e., stocks in the top
ESG rating quintile) after adjusting for risk exposures, i.e., the long-short portfolio yields a
CAPM-adjusted (FFC-adjusted, FF6-adjusted) return of 0.40% (0.46%, 0.50%) per month.23
Second, the negative return predictability of ESG ratings no longer holds for the remain-
ing firms and even turns positive in some cases. For perspective, we also consider a univariate
portfolio sort based on ESG ratings and report similar statistics in the column titled “All”.
The ESG rating does not predict stock returns for the full sample, which is consistent with
the existing literature showing weak return predictability of the overall ESG rating (e.g., Ped-
ersen et al. (2021)) and mixed evidence based on different ESG proxies (e.g., Gompers et al.
(2003); Hong and Kacperczyk (2009); Edmans (2011); Bolton and Kacperczyk (2020)). The
empirical evidence that ESG uncertainty can nontrivially interact with the ESG-performance
relation is also consistent with equation (27). Our results further highlight the importance
of rating uncertainty, as it not only affects investor demand but also has meaningful asset
pricing implications, i.e., the negative ESG-alpha relation only exists among stocks with low
rating uncertainty. The lack of consistency across ESG rating agencies could be a barrier
for investors who have to balance information on ESG scores and uncertainty when making
portfolio decisions.
Additionally, we consider a univariate portfolio sort based on ESG uncertainty and report
the results in the row titled “All”. Consistent with the model prediction, as shown in equation
(22), we find that when ESG uncertainty is in play at the market level, stocks with low ESG
uncertainty carry a negative and statistically significant CAPM alpha of −0.16% per month.
As shown in Online Appendix Table B.4, the result is also robust to FFC-adjusted and
FF6-adjusted returns. Furthermore, returns are increasing in ESG uncertainty, even though
the patterns are not always monotonic. For instance, the high-minus-low ESG uncertainty
portfolio (“HML-U”) shows a monthly CAPM alpha of 0.23% that is statistically significant at
the 10% level, supporting the model prediction that CAPM alpha increases with ESG rating
uncertainty. Collectively, our findings support the prediction that brown stocks outperform
green stocks only in the absence of rating uncertainty, and ESG uncertainty could tilt this
23
As our model is derived in market equilibrium, it is based on one market factor. However, the economic
magnitude and statistical significance in FFC-adjusted and FF6-adjusted returns reinforce our conclusion
that accounting for rating uncertainty can be useful even for investors who use multiple investment factors
in their portfolio decisions.

25
relationship via conflicting forces, as illustrated in Proposition 3.
As a robustness check, we perform regression analysis to further control for other firm
characteristics. Specifically, we estimate the following monthly Fama and MacBeth (1973)
regression:

Perfi,m = α0 + β1 ESGi,m−1 + β2 ESGi,m−1 × Low ESG Uncertaintyi,m−1

+ β3 Low ESG Uncertaintyi,m−1 + β40 Mi,m−1 + ei,m , (28)

where Perfi,m refers to the excess return or CAPM-adjusted return of stock i in month m,
ESGi,m−1 refers to the ESG rating, and Low ESG Uncertaintyi,m−1 refers to a dummy vari-
able that takes a value of 1 if the ESG rating uncertainty is in the bottom quintile across
all stocks in that month and 0 otherwise. The vector M stacks all other control variables,
including the Log(Size), Log(BM), 6M Momentum, Log(Illiquidity), Gross Profitability, Cor-
porate Investment, Leverage, Log(Analyst Coverage) and Analyst Dispersion. The parameter
of interest is β2 . Since the model predicts a negative ESG-performance relationship when
there is no rating uncertainty, we should see a negative value of β2 . Online Appendix Table
B.2 provides detailed definitions for each variable. We also report Newey and West (1987)
adjusted t-statistics.
We tabulate the results in Table 3, with models 1 to 4 for excess return and models
5 to 8 for CAPM-adjusted return. As expected, the ESG rating does not predict stock
returns for the full sample. More importantly, the ESG rating is negatively associated with
future stock performance when rating uncertainty is low. This relation is significant across
all regression specifications after controlling for other potential sources of uncertainty about
corporate ESG profiles and disagreement on firm fundamentals, such as analyst dispersion.
Overall, we confirm the early results in the portfolio sort and provide supporting evidence
for the ESG-augmented CAPM after considering rating uncertainty.

4.3 Additional Analysis and Robustness Checks

In the presence of a rapid growth in sustainable investing during the last decade (e.g., GSIA
(2018); PRI (2020)), we next assess how our findings evolve over time. We then conduct
robustness checks using alternative proxy for ESG rating and rating uncertainty.
We divide the full sample into two subperiods: 2003–2010 and 2011–2019, and repeat
the main analysis. Table 4 has a similar layout as Table 1, where Panels A, B and C
show the results for the norm-constrained institutions, hedge funds, and other institutions,
respectively. First, we confirm that for all three types of institutions, their preference for
green assets increases over time. Norm-constrained institutions hold 12.3% of the brown

26
stocks (i.e., stocks in the bottom ESG rating quintile) while they hold 19.2% of the green
stocks (i.e., stocks in the top ESG rating quintile) in the post-2011 period, indicating a 56%
increase. For perspective, they hold 14% more green stocks than brown stocks in the pre-
2011 period. While hedge funds invest more in brown stocks during both periods, they hold
33% less green stocks in the pre-2011 period and only 10% less green stocks in the post-2011
period. Interestingly, other institutions exhibit a shift in ESG preference over time, i.e., from
brown-loving to green-loving. They hold 7% less green stocks in the pre-2011 period while
12% more green stocks in the post-2011 period.
Second, for norm-constrained institutions, demand for green firms diminishes with ESG
rating uncertainty in both periods, while the effect is stronger in the pre-2011 period. Among
the green stocks, norm-constrained institutions hold 27.1% (19.0%) of the low-uncertainty
stocks while 19.5% (16.8%) of the high-uncertainty stocks in the pre-2011 (post-2011) period,
indicating a 28% (12%) decline. It is possible that the rising popularity in sustainable in-
vesting also incentivizes institutional investors to invest in ESG research and even create
internal ratings tools (e.g., Mooney (2019)), partially mitigating the negative effect of rat-
ing uncertainty. Overall, our findings confirm that even with the growing ESG awareness,
demand for green assets diminishes with ESG rating uncertainty for ESG-sensitive investors.
When there is no uncertainty in ESG ratings, our model predicts a negative relationship
between the ESG rating and expected CAPM alpha due to the nonpecuniary benefits from
holding green stocks. However, Pástor, Stambaugh, and Taylor (2021b,a) document that
green assets have higher realized alphas when investors’ tastes for green holdings shift un-
expectedly during the last decade. As a result, we expect our findings to be stronger in the
pre-2011 period, which provides a clean setting to analyze the equilibrium expected returns
of stocks.
Table 5 has a similar layout as Table 2, with Panel A for raw return and Panel B for
CAPM-adjusted return. As expected, the ESG rating is negatively associated with future
performance among stocks with low rating uncertainty in the pre-2011 period, yielding a
significant long-short portfolio return (“LMH-R”) of 1.12% (t-stat=3.06) per month and
CAPM-adjusted return of 0.96% (t-stat=2.81) per month. Consistent with our model pre-
diction, the negative ESG-CAPM alpha relationship does not hold for the remaining firms.
A univariate portfolio sort based on ESG uncertainty further confirms that CAPM alpha
increases with ESG rating uncertainty, i.e., the high-minus-low ESG uncertainty portfolio
(“HML-U”) shows a monthly CAPM alpha of 0.42% (t-stat=2.04) in the pre-2011 period.
In contrast, we do not find a negative return predictability of ESG ratings across all ESG-
rating-uncertainty portfolios, nor a positive ESG uncertainty-CAPM alpha relationship in
the post-2011 period. Our findings in both subperiods remain unchanged for FFC-adjusted

27
return and FF6-adjusted return, as reported in Online Appendix Table B.5. Note that our
results should not be interpreted as ESG rating uncertainty no longer matters in the future.
Instead, the equilibrium outcome over longer horizons could be even stronger than the full
sample evidence we document, due to the unexpected outcomes realized over the last decade.
Next, we conduct robustness checks by using alternative definition of ESG rating and
rating uncertainty. Specifically, for each rater-year, we sort all stocks covered by this rater
according to the original rating scale and calculate the percentile rank (normalized between
0 and 1) for each stock. The firm-level ESG rating is defined as the average rank across
all raters (labelled as ESGALL ), and the ESG rating uncertainty is defined as the standard
deviation of the ranks provided by all raters (labelled as ESG UncertaintyALL ). As noted
earlier, this method can entail some bias due to the lack of comparability across vendors.
We repeat our main analysis using the alternative proxy for ESG rating and rating
uncertainty. Online Appendix Table B.6 has the same layout as Table 1, where Panels
A, B and C show the results for the norm-constrained institutions, hedge funds, and other
institutions, respectively. The results confirm that norm-constrained institutions have strong
preference for green assets in general, while display lower demand for green firms when ESG
uncertainty is high. For instance, among the high-ESG-rating portfolios, norm-constrained
institutions hold 23.4% of the low-uncertainty stocks while only 15.5% of the high-uncertainty
stocks, indicating a 33% decline. As a result, green stocks no longer attract more norm-
constrained institutional investors than brown stocks when rating uncertainty is high.
Online Appendix Table B.7 has a similar layout as Table 2, with Panel A for raw return,
Panel B for CAPM-adjusted return, Panel C for FFC-adjusted return, and Panel D for
FF6-adjusted return. Our findings are largely consistent with the model prediction that the
ESG rating is negatively associated with future performance among stocks with low rating
uncertainty. The long-short portfolio return (FFC-adjusted return, FF6-adjusted return) is
significant at 0.52% (0.35%, 0.35%) per month. While the CAPM-adjusted return is not
statistically significant, the magnitude is sizable at 0.31% per month. Unreported results
show that the long-short portfolio yields a return of 1.05% per month and a CAPM-adjusted
(FFC-adjusted, FF6-adjusted) return of 0.87% (0.75%, 0.73%) per month in the pre-2011
period, all statistically significant at the 5% or 1% level.
Online Appendix Table B.8 presents the same Fama and MacBeth (1973) regression
results as in Table 3. We confirm that ESG rating is negatively associated with CAPM-
adjusted return when rating uncertainty is low, after controlling for other firm characteristics.
In short, our main results are robust to the alternative definition of ESG rating and rating
uncertainty.

28
5 Calibration
As final experiments, we calibrate the model to study the general equilibrium implications of
ESG rating uncertainty for the market premium, the cross section of stock returns, economic
welfare, and equity demand. Following Pástor, Stambaugh, and Taylor (2021a), we consider
ESG-indifferent (IND) and ESG-sensitive (ESG) agents. The former group does not derive
utility from ESG externalities (i.e., bIND = 0), while the utility of the latter positively depends
on the market ESG score and negatively depends on rating uncertainty, through bESG > 0.
Specifically, we assume that 20% of the agents have ESG preferences, while the remaining
fraction consists of ESG-indifferent agents. Hence, ESG-sensitive agents are not the vast
majority in the economy, yet they account for a substantial fraction.24
The ESG parameters, bESG , µg,M , σg,M , ρg,M , and the stock level counterparts of µg,M ,
σg,M , and ρg,M are unknown. In the data section above, we describe ways to map ESG
ratings into scores for individual securities, and the market-level ESG rating follows through
aggregation. The resulting quantities are not on the scale of equity returns and are ordinal
in nature. In particular, a higher ESG rating indicates a greener stock, while a higher
standard deviation among raters amounts to greater ESG uncertainty. Thus, stock-level and
market-level ratings, as well as measures of rating uncertainty, can comfortably be used to
assess the model implications through cross-sectional regressions and portfolio sorts. In the
calibration experiments that follow, we choose ESG parameters that conform to payoffs on
pseudo-assets, as formulated in the theory section.25 Further details are provided below.

5.1 Market Premium, Welfare, and Equity Demand

The analysis for the aggregate market is based on an economy that consists of the market
portfolio and a riskless asset (in zero net supply). The market volatility parameter employed
in the calibration is σM = 15.19%, which is the annual estimate from monthly U.S. market
returns, spanning the period July 1963 through December 2019. Then, employing the sample
estimate for the equity premium (6.5%), we obtain γ = 2.81, following equation (6). Two
remarks are in order. First, while our sample for individual stocks starts in 2002, due to
limited data for ESG ratings, the possibility to use longer return histories from the aggregate,
to sharpen estimates, builds on Pástor and Stambaugh (2002). In addition, expected market
return is endogenous in our setup, while the sample estimate is used to set the risk aversion
24
In unreported results, we confirm that an increasing fraction of ESG-sensitive investors leads to stronger
implications of ESG uncertainty.
25
In our model, the g = 0 case reflects green neutrality. Having this reference point, all the model
implications are invariant to a multiplicative scaling of ESG ratings and rating uncertainty, as long as the
brown aversion parameter is also scaled such that the pseudo return, bg, remains unchanged.

29
parameter.
We evaluate the equilibrium market premium on the basis of equation (11) for the
multiple-agent case. The market demand and the certainty equivalent return from invest-
ment differ across agent types. In particular, based on equation (4), the optimal market
µ +b µ
demand for agent i is x∗i = γ1i M σ2i g,M , where σi,U
2
= σM2
+ b2i σg,M
2
+ 2bi σM σg,M ρg,M . In
i,U
addition, as derived in Online Appendix A.6, the certainty equivalent excess return for agent
  2
µ +b µ
i is given by CE i = 2γ1 i M σi,U
i g,M
. Both the market demand and the certainty equivalent
return increase in the perceived market premium and diminish in the perceived market vari-
ance. For ESG-sensitive agents, the perceived certainty equivalent return increases with the
market ESG score, while the perceived variance rises with ESG uncertainty and the correla-
tion between the market ESG score and market return. The effect of ESG rating uncertainty
is stronger for higher values of bi and ρg,M .
To make the analysis sufficiently comprehensive, we run calibration experiments for mul-
tiple scenarios. First, we consider both green-neutral (µg,M = 0) and green (µg,M = 0.01)
markets. The ESG implications of the former case are exclusively attributed to ESG uncer-
tainty. The latter case involves the two conflicting forces noted earlier, i.e., the nonpecuniary
benefits from holding the green market versus aversion to ESG uncertainty. For ESG-sensitive
agents, we consider two values for brown aversion, namely, bESG is equal to 1 or 2. When the
market is green, both cases generate ESG return of 1% and 2% per year, respectively. When
the market is green neutral, brown aversion is not mapped into the incremental expected re-
turn. We also consider two values for the correlation between ESG and market return, ρg,M ,
namely, 0 and 0.5. The zero-correlation is a benchmark case that reflects the lower bound
on the implications of ESG uncertainty. The positive correlation is sensible, as described in
the theory section. Finally, the market ESG uncertainty, σg,M , ranges between 0 and 0.04.26
Panel A of Figure 2 describes the green-neutral market case, with solid lines representing
the case ρg,M = 0 and dashed lines corresponding to ρg,M = 0.5. The limiting case of bESG = 0
represents the departure point, where all agents are indifferent to the market ESG profile.
In that case, it follows that (i) the equilibrium market premium equals the ESG-indifference
2
value, γσM = 6.50%, regardless of the level of ESG uncertainty, (ii) both agent types hold
the market portfolio (x∗ESG = x∗IND = 1), and (iii) the agents perceive the same certainty
σ2
equivalent excess return (CE ESG = CE IND = γ 2M = 3.25%).
26
Empirically, the magnitude of ESG uncertainty is comparable to the scale of differences in ESG scores.
For instance, considering the summary statistics of our dataset (Online Appendix Table B.3, Panel C), the
quartile deviation of ESG ratings is 0.14. The values of ESG uncertainty are of the same order of magnitude
of differences in ESG scores: the median ESG uncertainty is 0.16, while the 90th percentile is 0.33. Similarly,
for calibration, we consider values of ESG uncertainty that conform to ESG levels: a green (brown) asset
has a mean ESG score of 0.01 (−0.01), and ESG uncertainty is of the order of 0.01 and multiples.

30
 When bESG > 0, the ESG agents are sensitive to the market rating uncertainty. Then,
2 2 27
the perceived market variance σM,U is higher than σM . This force leads to an increasing
equilibrium market premium, and more so for higher values of bESG , σg,M , and ρg,M .
As a result, the two types of agents have different certainty equivalent return and demand
for the market portfolio. On the one hand, the IND agents are not sensitive to ESG un-
2 2
certainty (σIND,U = σM ). Thus, they benefit from the higher equilibrium market premium,
which translates into a higher certainty equivalent return and a levered position in the mar-
ket portfolio (x∗IND > 1). On the other hand, the ESG agents are more sensitive to ESG
2 2
uncertainty than the aggregate market (σESG,U > σM,U ). Thus, their certainty equivalent
return and their demand for the market portfolio decline with increasing values of bESG , σg,M ,
and ρg,M .
We next quantitatively assess the economic cost of ESG uncertainty, as perceived by
ESG agents. The cost is represented by a diminishing certainty equivalent return relative
to σg,M = 0. When ρg,M = 0 and ESG uncertainty σg,M is set to 0.02 (0.04), the utility
loss is 0.03% (0.13%) per year for bESG = 1 and 0.13% (0.47%) for bESG = 2. Considering
ρg,M = 0.5 instead, the corresponding figures are 0.26% (0.55%) for bESG = 1 and 0.55%
(1.08%) for bESG = 2. The calibrated utility loss accounts for a nontrivial proportion of
the overall certainty equivalent excess return when compared to the benchmark case of no
uncertainty, i.e., 3.25%. Therefore, from the perspective of ESG agents, ESG uncertainty
leads to a significant utility loss.
When the market is green neutral, preferences for ESG essentially hurt welfare because
the only effect that comes into play is aversion to ESG uncertainty. Departing from a green-
neutral market, the nonpecuniary benefits from holding green stocks intervene, and more so
for higher values of brown aversion and market ESG score.
Panel B of Figure 2 describes the green-market case, with solid lines corresponding to
ρg,M = 0 and dashed lines to ρg,M = 0.5. In the absence of ESG uncertainty (σg,M = 0)
and when bESG > 0, the equilibrium market premium diminishes with bESG . This translates
into a lower certainty equivalent return and market demand for IND agents, who confront
a lower market premium but do not extract nonpecuniary benefits from holding the green
market. In contrast, ESG agents extract nonpecuniary benefits from the positive market
ESG tilt, which leads to a higher certainty equivalent return and higher market demand for
increasing values of bESG .
As the parameter σg,M captures the trade-off between the two conflicting forces of ESG
27 2
As we derive in Online Appendix A.1, the perceived aggregate market variance, σM,U , is a harmonic
2 2
weighted average of the market variances perceived by the agents, which in our example are σM,IND = σM
2 2 2 2 2 2 2
and σM,ESG = σM + bESG σg,M + 2bESG σM σg,M ρg,M . It follows that σM,IND < σM,U < σM,ESG .

31
preferences, we derive a break-even value of σg,M when the utility loss of ESG uncertainty
entirely offsets the benefits from holding green stocks. When ρg,M = 0 and bESG is 1 (2), the
welfare benefits of a green market perceived by ESG agents vanish, due to ESG uncertainty,
for σg,M = 9.9% (7.2%), well above reasonable values. However, a positive correlation
between market return and ESG rating amplifies the effects of ESG uncertainty. When
ρg,M = 0.5 and bESG is 1 (2), the threshold σg,M is much lower at 4.9% (4.3%).
The market premium is also subject to the two conflicting forces, i.e., the negative ESG
premium due to the green market versus the positive contribution due to ESG uncertainty.
When ρg,M = 0 and bESG is 1 (2), the two forces are equal for σg,M = 6.0% (4.2%), while if
ρg,M = 0.5 and bESG is 1 (2), the threshold σg,M is at 2.1% (1.9%).
Overall, we reinforce the notion that ESG uncertainty increases the market premium,
as well as reduces the economic welfare for ESG-sensitive investors and discourages their
participation in the stock market.

5.2 Cross Section of Expected Returns, Alpha, and Effective Beta

We next calibrate the cross section of expected return, the CAPM alpha, and the effective
beta in equilibrium, all of which are formulated in Section 2.3.
To distill cross-sectional implications of ESG uncertainty, we focus on the green-neutral
market described in Section 5.1. At the stock level, we consider green and brown assets,
with mean ESG scores µg,green = 0.01 and µg,brown = −0.01. Thus, for the green asset,
ESG agents perceive an incremental ESG return equal to 1% per year for bESG = 1 and 2%
per year for bESG = 2. The corresponding return figures are negative for the brown asset.
It is assumed that βgreen = βbrown = 1, and the idiosyncratic annualized return volatility
is 20% for both assets. As σM = 15.19%, the total stock return volatility is 25.12%.28
We consider a positive correlation between return and ESG score for each asset, setting
ρg,M = ρrg,green = ρrg,brown = 0.5. The off-diagonal elements in Σg and Σrg are assumed to
be zero.
Figure 3 illustrates how the expected excess return, the CAPM alpha, and the effective
beta vary with ESG uncertainty for green and brown assets (σg,green and σg,brown ). The solid
lines represent the green asset while dashed lines represent the brown asset. We consider a
market-wide ESG uncertainty, σg,M , equal to 0.01 for the left graphs and 0.02 for the right
graphs. Starting from the benchmark case of ESG indifference (bESG = 0), the expected
28 2 2 2 2
The total return variance of the green asset, σgreen , is given by βgreen σM +σid,green , where σid,green is the
idiosyncratic volatility. For βgreen = 1, σM = 15.19%, and σid,green = 20%, it follows that σgreen = 25.12%.
2 2 2
The same applies to σbrown . The covariance between returns is βgreen βbrown σM = (15.19%) , corresponding
to a correlation ρgreen,brown = 36.59%.

32
excess return for both assets is equal to the market premium, 6.50%, while the alpha is zero
and the effective beta coincides with the unit market beta.
Considering ESG-sensitive agents (bESG > 0), the positive ESG score of a green asset is
associated with lower expected return and alpha in equilibrium, as in Pástor, Stambaugh,
and Taylor (2021a). The effect is stronger for larger values of bESG . In addition, expected
return rises with ESG uncertainty. Thus, in the presence of the conflicting forces of ESG
score (negative effect on alpha) and ESG uncertainty (positive effect on alpha), a green asset
with high ESG uncertainty could have higher expected return and alpha than a brown asset
with low ESG uncertainty. For instance, when σg,M = 0.01, σg,green = 0.10, and bESG = 1
(bESG = 2), the green asset displays an expected excess return of 6.78% (7.09%) and an
alpha of 0.20% (0.42%). To compare, when the ESG profile of the brown asset is known for
certain, its expected excess return is 6.70% (6.90%) and alpha is 0.11% (0.23%).
The σg,green = 0 case merits further analysis. The zero-uncertainty asset does not contrib-
ute to the aggregate ESG uncertainty; thus, its effective beta is lower than the unit market
beta, per equation (19), and the effect is stronger when brown aversion and market-wide
ESG uncertainty are higher. For instance, when σg,M = 0.01, the effective beta is 0.987
(0.974) for bESG = 1 (bESG = 2). When σg,M = 0.02, the effective beta is 0.974 (0.950) for
bESG = 1 (bESG = 2). The diminished effective beta relative to the market beta induces a
negative contribution to alpha and expected return.
As demonstrated in equation (20), the effective beta does not depend on the mean ESG
score. Consequently, green and brown assets have the same effective beta for identical levels
of ESG uncertainty. The effective beta increases with ESG uncertainty and can rise above
the unit market beta, and the effect is stronger for higher values of brown aversion.
Finally, as long as the green and the brown assets have the same ESG uncertainty, the
performance difference between brown and green assets (both expected return and alpha)
diminishes with increasing ESG uncertainty. Consider, for instance, σg,M = 0.01. As the
ESG uncertainty increases from 0 to 0.10, the difference in expected return (µr,brown −µr,green )
decreases from 0.40% to 0.23% when bESG = 1, and from 0.80% to 0.29% when bESG = 2.
Similar patterns apply to alpha. Such calibration results follow from equation (27).
The overall evidence from the calibration indicates that ESG uncertainty has meaningful
implications for expected return, alpha, and effective beta. Notably, both alpha and the
effective beta increase with ESG uncertainty. Moreover, the alpha gap between brown and
green assets diminishes with ESG uncertainty.

33
6 Conclusion
We comprehensively analyze the equilibrium implications of ESG rating uncertainty for port-
folio choice and asset pricing. Starting with the market portfolio as the single risky asset, we
show that rating uncertainty leads to higher perceived market risk, higher market premium,
and lower investor demand. Next, we consider multiple risky assets and heterogeneous eco-
nomic agents and derive an ESG-augmented CAPM for the cross section of stock returns. In
particular, we propose that ESG uncertainty could tilt the ESG-CAPM alpha relationship
and serve as a potential channel to explain the mixed evidence in prior studies.
We collect ESG ratings from six major rating agencies and employ the standard deviation
of the ratings to proxy for ESG uncertainty. We empirically test the model implications
and provide supporting evidence. First, ESG rating uncertainty reduces investor demand
for stocks, especially for ESG-sensitive investors (i.e., norm-constrained institutions) in their
ESG investment (i.e., green stocks). Second, brown stocks outperform green stocks only when
rating uncertainty is low, and the negative return predictability of ESG ratings does not hold
for the remaining firms. We then calibrate the model to assess its quantitative implications in
the presence of rating uncertainty. The analysis reinforces the notion that ESG uncertainty
could affect investors’ demand, the risk-return trade-off, and reduce economic welfare for
ESG-sensitive agents.
Our findings echo the growing concerns regarding the lack of consistency of ESG in-
formation disclosure and ratings provided by different rating agencies. In the presence of
rating uncertainty, investors are less likely to make ESG investments and actively engage
in corporate ESG issues. This could increase the cost of capital for green firms and further
limit their capacity to make socially responsible investments and generate real social impact.
As the amount of sustainable investing is expected to keep growing, the overall impact will
become even more striking. Viewed from this perspective, our results provide a conservative
assessment of rating uncertainty.
Our paper also suggests avenues for future research. While existing work studying equi-
librium with ESG focuses on a single-period environment, it would be natural to extend
ESG equilibrium to multiperiod dynamic setups. Then, the market ESG can display time
variation, which would give rise to an incremental asset pricing factor. It would also be
instructive to account for investors’ learning about the ESG profile of a firm. These and
other topics in dynamic asset pricing are left for future research.

34
References
Abarbanell, J. S., B. J. Bushee, and J. Smith Raedy (2003). Institutional investor preferences
and price pressure: The case of corporate spin-offs. The Journal of Business 76 (2), 233–
261.

Agarwal, V., V. Fos, and W. Jiang (2013). Inferring reporting-related biases in hedge fund
databases from hedge fund equity holdings. Management Science 59 (6), 1271–1289.

Amel-Zadeh, A. and G. Serafeim (2018). Why and how investors use ESG information:
Evidence from a global survey. Financial Analysts Journal 74 (3), 87–103.

Amihud, Y. (2002). Illiquidity and stock returns: Cross-section and time-series effects.
Journal of Financial Markets 5 (1), 31–56.

Back, K. (2010). Asset pricing and portfolio choice theory. Oxford University Press.

Berg, F., J. F. Kölbel, and R. Rigobon (2020). Aggregate confusion: The divergence of ESG
ratings. Available at SSRN 3438533 .

Bhandari, L. C. (1988). Debt/equity ratio and expected stock returns: Empirical evidence.
The Journal of Finance 43 (2), 507–528.

Bolton, P. and M. T. Kacperczyk (2020). Do investors care about carbon risk? Journal of
Financial Economics, forthcoming.

Carhart, M. M. (1997). On persistence in mutual fund performance. The Journal of Fin-

ance 52 (1), 57–82.

Chatterji, A. K., R. Durand, D. I. Levine, and S. Touboul (2016). Do ratings of firms

converge? Implications for managers, investors and strategy researchers. Strategic Man-
agement Journal 37 (8), 1597–1614.

Chen, T., H. Dong, and C. Lin (2020). Institutional shareholders and corporate social
responsibility. Journal of Financial Economics 135 (2), 483–504.

Dimson, E., O. Karakaş, and X. Li (2015). Active ownership. The Review of Financial
Studies 28 (12), 3225–3268.

Dyck, A., K. V. Lins, L. Roth, and H. F. Wagner (2019). Do institutional investors drive
corporate social responsibility? International evidence. Journal of Financial Econom-
ics 131 (3), 693–714.

35
Eccles, R. G. and J. C. Stroehle (2018). Exploring social origins in the construction of ESG
measures. Available at SSRN 3212685 .

Edmans, A. (2011). Does the stock market fully value intangibles? Employee satisfaction
and equity prices. Journal of Financial Economics 101 (3), 621–640.

Fama, E. F. and K. R. French (1993). Common risk factors in the returns on stocks and
bonds. Journal of Financial Economics 33 (1), 3–56.

Fama, E. F. and K. R. French (2015). A five-factor asset pricing model. Journal of Financial
Economics 116 (1), 1–22.

Fama, E. F. and K. R. French (2018). Choosing factors. Journal of Financial Econom-

ics 128 (2), 234–252.

Fama, E. F. and J. D. MacBeth (1973). Risk, return, and equilibrium: Empirical tests.
Journal of Political Economy 81 (3), 607–636.

Gibson, R., P. Krueger, N. Riand, and P. S. Schmidt (2020). ESG rating disagreement and
stock returns. Available at SSRN 3433728 .

Gompers, P., J. Ishii, and A. Metrick (2003). Corporate governance and equity prices. The
Quarterly Journal of Economics 118 (1), 107–156.

GSIA (2018). Global sustainable investment review. URL: http://www.gsi-

alliance.org/trends-report-2018/ .

Hartzmark, S. M. and A. B. Sussman (2019). Do investors value sustainability? A natural

experiment examining ranking and fund flows. The Journal of Finance 74 (6), 2789–2837.

Heinkel, R., A. Kraus, and J. Zechner (2001). The effect of green investment on corporate
behavior. Journal of Financial and Quantitative Analysis 36 (4), 431–449.

Hong, H. and M. Kacperczyk (2009). The price of sin: The effects of social norms on markets.
Journal of Financial Economics 93 (1), 15–36.

Jegadeesh, N. and S. Titman (1993). Returns to buying winners and selling losers: Implica-
tions for stock market efficiency. The Journal of Finance 56 (2), 699–720.

Jiao, Y., M. Massa, and H. Zhang (2016). Short selling meets hedge fund 13F: An anatomy
of informed demand. Journal of Financial Economics 122 (3), 544–567.

36
Krueger, P., Z. Sautner, and L. T. Starks (2020). The importance of climate risks for
institutional investors. The Review of Financial Studies 33 (3), 1067–1111.

Lins, K. V., H. Servaes, and A. Tamayo (2017). Social capital, trust, and firm performance:
The value of corporate social responsibility during the financial crisis. The Journal of
Finance 72 (4), 1785–1824.

López-de-Silanes, F., J. A. McCahery, and P. C. Pudschedl (2020). ESG performance and

disclosure: A cross-country analysis. Singapore Journal of Legal Studies, 217–241.

Mackintosh, J. (2018). Is Tesla or Exxon more sustainable? It depends whom you ask. Wall
Street Journal. URL: https://on.wsj.com/2MQCC4m.

Mooney, A. (2019). ESG investing sparks race in tech and hiring at asset managers. Financial
Times. URL: https://www.ft.com/content/247f4034-4280-318a-9900-87608a575ede.

Moss, A., J. P. Naughton, and C. Wang (2020). The irrelevance of ESG disclosure to retail
investors: Evidence from Robinhood. Available at SSRN 3604847 .

Newey, W. K. and K. D. West (1987). A simple positive-definite heteroskedasticity and

autocorrelation consistent covariance matrix. Econometrica 55, 703–708.

Novy-Marx, R. (2013). The other side of value: The gross profitability premium. Journal of
Financial Economics 108 (1), 1–28.

Pástor, L. and R. F. Stambaugh (2002). Investing in equity mutual funds. Journal of

Financial Economics 63 (3), 351–380.

Pástor, L., R. F. Stambaugh, and L. A. Taylor (2021a). Sustainable investing in equilibrium.

Journal of Financial Economics, forthcoming.

Pástor, L., R. F. Stambaugh, and L. A. Taylor (2021b). Dissecting green returns. Available
at SSRN 3864502 .

Pedersen, L. H., S. Fitzgibbons, and L. Pomorski (2021). Responsible investing: The ESG-
efficient frontier. Journal of Financial Economics, forthcoming.

PRI (2020). Principles for responsible investment. URL: ht-

tps://www.unpri.org/download?ac=10948 .

Titman, S., K. C. J. Wei, and F. Xie (2004). Capital investments and stock returns. Journal
of Financial and Quantitative Analysis 39 (4), 677–700.

37
 Table 1: Institutional Ownership of Portfolios Sorted by ESG Rating and Uncertainty
At the end of year t, stocks are independently sorted into quintiles according to their ESG ratings and ESG rating uncertainty to generate
25 (5 × 5) portfolios. The low- (high)-ESG-rating and ESG-rating-uncertainty portfolios comprise the bottom (top) quintile of stocks
based on the ESG rating and ESG rating uncertainty, respectively. For each of the 25 portfolios, we compute the average institutional
ownership in each quarter in year t + 1 and rebalance the portfolios at the end of year t + 1. Panel A reports the time-series averages of
quarterly institutional ownership of norm-constrained institutions for each of the 25 portfolios and the average difference in institutional
ownership between high- and low-ESG-rating portfolios (“HML-R”), as well as between high- and low-ESG-rating-uncertainty portfolios
(“HML-U”). Panels B and C report similar statistics for average ownership of hedge funds and other institutions, respectively. Online
Appendix Table B.2 provides detailed definitions for each variable. Newey-West adjusted t-statistics are shown in parentheses. Numbers
with “*”, “**”, and “***” are significant at the 10%, 5%, and 1% levels, respectively.

Panel A: Norm-Constrained Institutions

ESG Rating ESG Uncertainty

Low 2 3 4 High HML-U t-stat All

Low 0.170 0.183 0.187 0.178 0.179 0.009 (0.80) 0.177

2 0.185 0.192 0.207 0.209 0.184 -0.001 (-0.23) 0.195
3 0.189 0.215 0.210 0.212 0.191 0.002 (0.40) 0.200
4 0.211 0.211 0.211 0.215 0.211 0.000 (0.04) 0.211
High 0.228 0.236 0.238 0.225 0.181 -0.047*** (-2.73) 0.230

HML-R 0.058*** 0.053*** 0.050*** 0.047*** 0.002 0.053***

(10.21) (12.00) (8.33) (8.51) (0.08) (11.39)

Panel B: Hedge Funds

ESG Rating ESG Uncertainty

Low 2 3 4 High HML-U t-stat All

Low 0.157 0.157 0.160 0.156 0.130 -0.027*** (-3.70) 0.157

2 0.143 0.147 0.155 0.153 0.149 0.006 (1.31) 0.149
3 0.153 0.144 0.144 0.149 0.153 -0.000 (-0.08) 0.150
4 0.148 0.144 0.140 0.142 0.141 -0.006* (-1.96) 0.142
High 0.127 0.124 0.128 0.128 0.119 -0.008 (-1.33) 0.127

HML-R -0.031*** -0.033*** -0.032*** -0.029*** -0.011 -0.030***

(-6.14) (-8.15) (-6.30) (-5.57) (-1.25) (-8.06)

Panel C: Other Institutions

ESG Rating ESG Uncertainty

Low 2 3 4 High HML-U t-stat All

Low 0.347 0.367 0.357 0.363 0.317 -0.030** (-2.57) 0.356

2 0.343 0.374 0.387 0.390 0.354 0.010 (1.43) 0.370
3 0.370 0.373 0.371 0.384 0.360 -0.011 (-1.66) 0.368
4 0.382 0.375 0.378 0.369 0.360 -0.022*** (-3.25) 0.370
High 0.363 0.368 0.363 0.357 0.328 -0.035 (-1.63) 0.363

HML-R 0.016 0.001 0.006 -0.005 0.011 0.007

(1.28) (0.13) (0.59) (-0.37) (0.35) (0.71)

38
 Table 2: Performance of Portfolios Sorted by ESG Rating and Uncertainty
At the end of year t, stocks are first sorted into quintiles according to their ESG rating uncertainty. Within each ESG rating
uncertainty group, stocks are further sorted into quintiles according to their ESG ratings to generate 25 (5×5) portfolios. The low-
(high)-ESG-rating and ESG-rating-uncertainty portfolios comprise the bottom (top) quintile of stocks based on the ESG rating and
ESG rating uncertainty, respectively. For each of the 25 portfolios, we compute the value-weighted return in each month in year t+1 and
rebalance the portfolios at the end of year t + 1. Panel A reports the time-series averages of monthly returns for each of the 25 portfolios,
as well as for the investment strategy of going long (short) the low- (high)-ESG-rating stocks (“LMH-R”). The column “All” reports
similar statistics for portfolios sorted by ESG ratings only. The row “All” reports returns for portfolios sorted by ESG uncertainty
only, as well as the investment strategy of going long (short) the high (low) ESG-uncertainty stocks (“HML-U”). In Panel B, portfolio
returns are further adjusted by the CAPM. Online Appendix Table B.2 provides detailed definitions for each variable. Newey-West ad-
justed t-statistics are shown in parentheses. Numbers with “*”, “**”, and “***” are significant at the 10%, 5%, and 1% levels, respectively.

Panel A: Return Panel B: CAPM-Adjusted Return

ESG Rating ESG Uncertainty ESG Uncertainty

Low 2 3 4 High All Low 2 3 4 High All

Low 1.235*** 1.113*** 0.767** 0.875** 0.760** 0.923** 0.168 0.064 -0.311* -0.141 -0.101 -0.101
(2.95) (2.99) (1.98) (2.30) (2.32) (2.58) (0.93) (0.40) (-1.82) (-0.89) (-0.58) (-0.84)
2 1.245*** 1.026*** 1.093*** 1.043*** 1.095*** 0.963*** 0.187 0.076 0.115 0.042 0.151 -0.008
(3.36) (2.84) (3.30) (2.74) (2.91) (2.85) (1.16) (0.38) (0.77) (0.29) (0.77) (-0.07)
3 1.096*** 0.965*** 1.050*** 1.104*** 0.949*** 1.021*** 0.040 -0.031 0.002 0.064 0.079 0.053
(2.69) (2.83) (2.86) (2.89) (3.15) (3.11) (0.23) (-0.20) (0.02) (0.46) (0.42) (0.64)
4 0.730** 0.695* 1.105*** 1.019*** 0.990*** 1.017*** -0.192 -0.389*** 0.108 0.040 0.006 0.095
(2.09) (1.81) (2.90) (2.96) (2.68) (3.42) (-1.24) (-3.28) (0.55) (0.34) (0.03) (1.32)
High 0.642* 0.842** 0.855*** 1.184*** 0.854*** 0.805** -0.230* -0.063 -0.012 0.245* -0.001 -0.095
(1.97) (2.53) (3.06) (3.62) (2.81) (2.57) (-1.95) (-0.55) (-0.10) (1.83) (-0.01) (-1.49)

LMH-R 0.594*** 0.271 -0.088 -0.309 -0.094 0.118 0.398* 0.128 -0.299 -0.387* -0.100 -0.006
(2.72) (1.30) (-0.39) (-1.43) (-0.42) (0.78) (1.86) (0.58) (-1.25) (-1.75) (-0.42) (-0.04)

ESG Rating ESG Uncertainty ESG Uncertainty

Low 2 3 4 High HML-U Low 2 3 4 High HML-U

All 0.753** 0.875*** 0.935*** 1.083*** 0.940*** 0.187 -0.155** -0.090 -0.003 0.120* 0.071 0.226*
(2.31) (2.61) (3.07) (3.28) (3.29) (1.40) (-1.98) (-1.20) (-0.04) (1.72) (0.84) (1.67)

39
 Table 3: ESG Rating, Uncertainty, and Stock Returns
This table presents the results of the following monthly Fama-MacBeth regressions, as well as their corresponding Newey-West adjusted
t-statistics:

Perfi,m = α0 + β1 ESGi,m−1 + β2 ESGi,m−1 × Low ESG Uncertaintyi,m−1 + β3 Low ESG Uncertaintyi,m−1 + β40 Mi,m−1 + ei,m ,

where Perfi,m refers to the excess return (models 1 to 4) or CAPM-adjusted return (models 5 to 8) of stock i in month m, ESGi,m−1
refers to the ESG rating, Low ESG Uncertaintyi,m−1 refers to a dummy variable that takes a value of 1 if the ESG rating uncertainty
is in the bottom quintile across all stocks in that month and 0 otherwise. The vector M stacks all other control variables, including
the Log(Size), Log(BM), 6M Momentum, Log(Illiquidity), Gross Profitability, Corporate Investment, Leverage, Log(Analyst Coverage)
and Analyst Dispersion. Online Appendix Table B.2 provides detailed definitions for each variable. Numbers with “*”, “**”, and “***”
are significant at the 10%, 5%, and 1% levels, respectively.

Stock Returns Regressed on Lagged ESG Rating and Uncertainty

Excess Return CAPM-Adjusted Return

Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7 Model 8

ESG 0.002 0.098 0.062 0.199 0.042 0.139 0.162 0.301

(0.01) (0.65) (0.33) (1.03) (0.23) (0.91) (0.77) (1.65)
ESG × Low ESG Uncertainty -0.163* -0.223* -0.254** -0.312**
(-1.91) (-1.75) (-2.26) (-2.36)
Low ESG Uncertainty 0.114* 0.109 0.125** 0.114
(1.86) (1.38) (2.20) (1.61)
Log(Size) -0.100 -0.036 -0.101 -0.038 -0.044 0.111 -0.044 0.111
(-1.28) (-0.27) (-1.30) (-0.29) (-0.59) (0.77) (-0.60) (0.77)
Log(BM) 0.001 0.009 -0.001 0.008 -0.021 0.019 -0.024 0.017
(0.01) (0.14) (-0.01) (0.12) (-0.19) (0.18) (-0.21) (0.17)
6M Momentum 0.336 0.188 0.335 0.194 0.275 0.105 0.276 0.111
(0.70) (0.40) (0.69) (0.42) (0.50) (0.20) (0.50) (0.21)
Log(Illiquidity) 0.056 0.056 0.103** 0.103**
(1.00) (1.03) (2.17) (2.15)
Gross Profitability 0.178 0.180 0.355* 0.359*
(0.99) (1.00) (1.83) (1.85)
Corporate Investment 0.037 0.037 -0.005 -0.007
(0.49) (0.50) (-0.08) (-0.09)
Leverage -0.037 -0.037 -0.034 -0.034
(-0.78) (-0.79) (-0.73) (-0.73)
Log(Analyst Coverage) -0.019 -0.019 -0.174 -0.175
(-0.15) (-0.14) (-1.40) (-1.41)
Analyst Dispersion -0.536*** -0.539*** -0.828*** -0.831***
(-2.67) (-2.71) (-4.37) (-4.37)
Constant 2.309* 1.800 2.281* 1.775 0.591 -0.555 0.533 -0.614
(1.71) (1.09) (1.70) (1.09) (0.46) (-0.31) (0.42) (-0.34)

Obs 283,671 254,873 283,671 254,873 272,728 245,451 272,728 245,451

R-squared 0.045 0.080 0.048 0.082 0.043 0.076 0.045 0.078

40
 Table 4: Institutional Ownership of Portfolios Sorted by ESG Rating and Uncertainty: Subsample Analysis
At the end of year t, stocks are independently sorted into quintiles according to their ESG ratings and ESG rating uncertainty to generate 25 (5×5) portfolios. The low- (high)-ESG-rating
and ESG-rating-uncertainty portfolios comprise the bottom (top) quintile of stocks based on the ESG rating and ESG rating uncertainty, respectively. For each of the 25 portfolios, we
compute the average institutional ownership in each quarter in year t + 1 and rebalance the portfolios at the end of year t + 1. Panel A reports the time-series averages of quarterly
institutional ownership of norm-constrained institutions for each of the 25 portfolios and the average difference in institutional ownership between high- and low-ESG-rating portfolios
(“HML-R”), as well as between high- and low-ESG-rating-uncertainty portfolios (“HML-U”). We divide the full sample into two subperiods, and report results for 2003–2010 on the left
and 2011–2019 on the right. Panels B and C report similar statistics for average ownership of hedge funds and other institutions, respectively. Online Appendix Table B.2 provides
detailed definitions for each variable. Newey-West adjusted t-statistics are shown in parentheses. Numbers with “*”, “**”, and “***” are significant at the 10%, 5%, and 1% levels,
respectively.

Panel A: Norm-Constrained Institutions

2003–2010 2011–2019
ESG Rating ESG Uncertainty ESG Uncertainty
Low 2 3 4 High HML-U t-stat All Low 2 3 4 High HML-U t-stat All
Low 0.234 0.239 0.243 0.238 0.262 0.028 (1.59) 0.239 0.114 0.133 0.137 0.124 0.105 -0.009* (-1.77) 0.123
2 0.238 0.250 0.257 0.264 0.244 0.006 (0.76) 0.251 0.138 0.140 0.163 0.160 0.130 -0.008 (-0.93) 0.145
3 0.244 0.266 0.258 0.261 0.251 0.007 (1.09) 0.255 0.140 0.170 0.168 0.167 0.137 -0.003 (-0.54) 0.151
4 0.262 0.265 0.255 0.265 0.269 0.007 (1.06) 0.264 0.166 0.163 0.172 0.170 0.161 -0.006 (-0.71) 0.165
High 0.271 0.276 0.277 0.266 0.195 -0.076** (-2.41) 0.273 0.190 0.200 0.203 0.188 0.168 -0.022*** (-3.65) 0.192
HML-R 0.037*** 0.037*** 0.034*** 0.028*** -0.067 0.034*** 0.076*** 0.068*** 0.065*** 0.064*** 0.063*** 0.069***
(8.58) (8.60) (6.87) (7.18) (-1.67) (12.78) (20.93) (21.05) (10.36) (11.55) (10.22) (27.30)
Panel B: Hedge Funds

41
2003–2010 2011–2019
ESG Rating ESG Uncertainty ESG Uncertainty
Low 2 3 4 High HML-U t-stat All Low 2 3 4 High HML-U t-stat All
Low 0.133 0.128 0.136 0.127 0.097 -0.036*** (-4.08) 0.129 0.179 0.183 0.181 0.182 0.160 -0.019* (-1.97) 0.181
2 0.117 0.110 0.115 0.119 0.107 -0.010** (-2.39) 0.113 0.166 0.180 0.189 0.184 0.187 0.021*** (5.08) 0.182
3 0.104 0.114 0.115 0.106 0.112 0.008 (1.55) 0.111 0.196 0.170 0.169 0.187 0.189 -0.008 (-1.65) 0.184
4 0.093 0.100 0.096 0.098 0.091 -0.002 (-1.00) 0.096 0.196 0.183 0.178 0.182 0.186 -0.009* (-1.81) 0.183
High 0.086 0.087 0.090 0.088 0.066 -0.020** (-2.13) 0.088 0.163 0.157 0.162 0.162 0.166 0.003 (0.78) 0.162
HML-R -0.047*** -0.041*** -0.046*** -0.039*** -0.030*** -0.042*** -0.016*** -0.026*** -0.019*** -0.019** 0.006 -0.019***
(-13.62) (-7.73) (-7.20) (-13.21) (-3.17) (-12.15) (-3.08) (-6.31) (-5.25) (-2.56) (0.51) (-5.68)
Panel C: Other Institutions
2003–2010 2011–2019
ESG Rating ESG Uncertainty ESG Uncertainty
Low 2 3 4 High HML-U t-stat All Low 2 3 4 High HML-U t-stat All
Low 0.385 0.389 0.384 0.414 0.356 -0.029 (-1.54) 0.390 0.314 0.347 0.333 0.317 0.283 -0.031** (-2.42) 0.327
2 0.387 0.409 0.401 0.407 0.379 -0.008 (-0.97) 0.396 0.304 0.343 0.374 0.374 0.331 0.027*** (3.11) 0.347
3 0.394 0.385 0.377 0.384 0.363 -0.031*** (-6.31) 0.376 0.349 0.362 0.365 0.385 0.356 0.007 (1.02) 0.361
4 0.375 0.383 0.388 0.367 0.351 -0.024** (-2.64) 0.370 0.388 0.369 0.368 0.372 0.369 -0.019** (-2.19) 0.370
High 0.357 0.367 0.361 0.353 0.291 -0.066 (-1.57) 0.360 0.369 0.370 0.365 0.361 0.361 -0.008 (-1.09) 0.366
HML-R -0.028** -0.023 -0.023** -0.061*** -0.065 -0.029*** 0.055*** 0.023*** 0.032*** 0.044*** 0.078*** 0.039***
(-2.13) (-1.67) (-2.23) (-5.46) (-1.36) (-3.34) (10.32) (3.35) (3.62) (4.56) (5.44) (9.84)
Table 5: Performance of Portfolios Sorted by ESG Rating and Uncertainty: Subsample Analysis
At the end of year t, stocks are first sorted into quintiles according to their ESG rating uncertainty. Within each ESG rating uncertainty
group, stocks are further sorted into quintiles according to their ESG ratings to generate 25 (5×5) portfolios. The low- (high)-ESG-
rating and ESG-rating-uncertainty portfolios comprise the bottom (top) quintile of stocks based on the ESG rating and ESG rating
uncertainty, respectively. For each of the 25 portfolios, we compute the value-weighted return in each month in year t + 1 and rebalance
the portfolios at the end of year t + 1. Panel A reports the time-series averages of monthly returns for each of the 25 portfolios, as well
as for the investment strategy of going long (short) the low- (high)-ESG-rating stocks (“LMH-R”). The column “All” reports similar
statistics for portfolios sorted by ESG ratings only. The row “All” reports returns for portfolios sorted by ESG uncertainty only, as well
as the investment strategy of going long (short) the high (low) ESG-uncertainty stocks (“HML-U”). We divide the full sample into two
subperiods, and report results for 2003–2010 on the left and 2011–2019 on the right. In Panel B, portfolio returns are further adjusted
by the CAPM. Online Appendix Table B.2 provides detailed definitions for each variable. Newey-West adjusted t-statistics are shown
in parentheses. Numbers with “*”, “**”, and “***” are significant at the 10%, 5%, and 1% levels, respectively.

Panel A: Return

2003–2010 2011–2019

ESG Rating ESG Uncertainty ESG Uncertainty

Low 2 3 4 High All Low 2 3 4 High All

Low 1.427* 0.845 0.528 0.949 0.667 0.773 1.065*** 1.351*** 0.980** 0.809** 0.842** 1.056***
(1.86) (1.35) (0.77) (1.43) (1.23) (1.23) (2.93) (3.25) (2.52) (2.00) (2.26) (2.92)
2 1.235* 0.973 0.955* 0.984 0.902 0.957 1.254*** 1.073*** 1.215*** 1.096*** 1.266*** 0.968***
(1.83) (1.44) (1.75) (1.53) (1.34) (1.64) (3.61) (3.42) (3.19) (2.68) (3.55) (2.79)
3 0.944 1.014* 0.919 1.157* 0.879* 0.764 1.231*** 0.921** 1.166*** 1.057*** 1.011*** 1.249***
(1.26) (1.74) (1.43) (1.74) (1.70) (1.33) (3.53) (2.55) (3.20) (2.80) (2.94) (3.83)
4 0.497 0.502 0.928 0.763 1.108* 0.976* 0.937** 0.868** 1.262*** 1.247*** 0.884* 1.054***
(0.86) (0.73) (1.29) (1.22) (1.91) (1.87) (2.52) (2.40) (4.15) (4.43) (1.92) (3.62)
High 0.309 0.346 0.524 1.205** 0.619 0.420 0.937*** 1.283*** 1.150*** 1.166*** 1.062*** 1.147***
(0.52) (0.57) (1.08) (2.05) (1.18) (0.75) (3.36) (5.14) (3.98) (3.78) (3.38) (4.26)

LMH-R 1.119*** 0.499* 0.004 -0.256 0.048 0.353 0.127 0.068 -0.170 -0.357 -0.220 -0.091
(3.06) (1.78) (0.01) (-0.74) (0.12) (1.45) (0.59) (0.23) (-0.70) (-1.22) (-0.87) (-0.50)

ESG Rating ESG Uncertainty ESG Uncertainty

Low 2 3 4 High HML-U Low 2 3 4 High HML-U

All 0.482 0.533 0.666 1.011* 0.832* 0.350 0.994*** 1.180*** 1.174*** 1.146*** 1.037*** 0.043
(0.81) (0.87) (1.25) (1.70) (1.71) (1.51) (3.50) (4.41) (3.97) (3.92) (3.35) (0.31)
Panel B: CAPM-Adjusted Return

2003–2010 2011–2019

ESG Rating ESG Uncertainty ESG Uncertainty

Low 2 3 4 High All Low 2 3 4 High All

Low 0.568** 0.058 -0.284 0.162 0.011 -0.006 -0.147 0.022 -0.376* -0.433** -0.262 -0.224
(1.99) (0.27) (-0.91) (0.68) (0.04) (-0.03) (-0.67) (0.10) (-1.95) (-2.14) (-1.17) (-1.52)
2 0.397** 0.172 0.236 0.222 0.175 0.218 0.023 0.086 -0.083 -0.162 0.098 -0.259**
(2.00) (0.56) (0.98) (1.18) (0.55) (1.45) (0.09) (0.32) (-0.54) (-0.79) (0.43) (-2.06)
3 0.088 0.259 0.133 0.358* 0.237 0.034 0.061 -0.344* -0.169 -0.228 -0.158 0.019
(0.32) (1.14) (0.79) (1.89) (0.81) (0.25) (0.27) (-1.89) (-1.09) (-1.14) (-0.71) (0.17)
4 -0.192 -0.348* 0.109 -0.049 0.381* 0.243** -0.264 -0.408*** 0.178 0.204 -0.418 -0.035
(-0.86) (-1.70) (0.26) (-0.25) (1.82) (2.34) (-1.22) (-2.94) (1.49) (1.27) (-1.23) (-0.37)
High -0.391** -0.403** -0.159 0.461* -0.030 -0.294*** -0.070 0.310*** 0.110 0.052 -0.037 0.093*
(-2.06) (-2.29) (-0.80) (1.98) (-0.12) (-2.92) (-0.56) (3.02) (1.02) (0.32) (-0.23) (1.83)

LMH-R 0.959*** 0.460 -0.126 -0.299 0.041 0.289 -0.077 -0.288 -0.486* -0.486 -0.225 -0.317*
(2.81) (1.60) (-0.30) (-0.85) (0.10) (1.12) (-0.31) (-1.06) (-1.88) (-1.62) (-0.81) (-1.74)

ESG Rating ESG Uncertainty ESG Uncertainty

Low 2 3 4 High HML-U Low 2 3 4 High HML-U

All -0.238* -0.255** -0.068 0.243* 0.181 0.419** -0.077 0.117* 0.048 0.027 -0.107 -0.029
(-1.87) (-2.35) (-0.45) (1.93) (1.33) (2.04) (-0.95) (1.74) (0.79) (0.32) (-1.07) (-0.19)

42
 Figure 1: Market ESG Ratings and ESG Uncertainty, Average Stock-Level ESG Uncertainty
The top graph shows the time-series of the market ESG score obtained from each data vendor, as well as the mean ESG rating across
data vendors. The bottom graph shows the time-series of market ESG uncertainty, as well as equal- and value-weighted average of
stock-level ESG uncertainty. Section 3.2 provides details on the construction of the variables.

0.8

0.6

0.4

0.2

0
2004 2006 2008 2010 2012 2014 2016 2018

0.3

0.25

0.2

0.15

0.1

0.05

0
2004 2006 2008 2010 2012 2014 2016 2018

43
 Figure 2: Equilibrium Equity Premium, Certainty Equivalent Return, and Market Demand
This figure shows the equilibrium market premium (µM ), the certainty equivalent excess return for ESG-sensitive (CE ESG ) and ESG-
indifferent (CE IND ) agents, the optimal market participation (x∗ESG and x∗IND ), and their variation with the market ESG uncertainty,
σg,M . The relative risk aversion, γ, is 2.81, and the market volatility, σM , is 15.19%. ESG-sensitive agents represent a fraction of
wESG = 20% of the population and have a brown aversion bESG = {0, 1, 2}. ESG-indifferent agents represent wIND = 80% of the
population and have a brown aversion bIND = 0. The correlation between the market return and the ESG score, ρg,M , is 0 (solid lines)
or 0.5 (dashed lines). Panel A focuses on a green-neutral market (µg,M = 0), while Panel B describes a green market (µg,M = 0.01).

Panel A: Green-Neutral Market Panel B: Green Market

7 7

6.5 6.5

6 6
0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04 0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04

5 5

4 4

3 3

2 2
0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04 0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04

4 4

3.5 3.5

3 3

2.5 2.5
0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04 0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04

120 120

100 100

80 80

60 60
0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04 0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04

110 110

105 105

100 100

95 95

0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04 0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04

44
 Figure 3: Two-Asset Pricing Equilibrium: Expected Stock Return, Alpha, and Effective Beta
Considering the green-neutral market described in Figure 2 Panel A, for green (solid lines) and brown (dashed lines) assets, this figure
displays the equilibrium expected excess stock return, (µr,green and µr,brown ), the CAPM alpha, (αgreen and αbrown ), the effective beta,
(βeff ,green and βeff ,brown ), and their variation with ESG uncertainty, σg,green , σg,brown . The mean ESG scores of the two assets are
µg,green = 0.01 and µg,brown = −0.01. The market betas of the two assets are βgreen = βbrown = 1, while their idiosyncratic return
volatility is equal to 0.2. The correlation between return and the same-asset ESG score is ρg,M = ρrg,green = ρrg,brown = 0.5. The
graphs on the left describe a market-wide ESG uncertainty that is equal to σg,M = 0.01, while the right plots display results for
σg,M = 0.02.

7.5 7.5

7 7

6.5 6.5

6 6
0 0.02 0.04 0.06 0.08 0.1 0 0.02 0.04 0.06 0.08 0.1

0.5 0.5

0 0

-0.5 -0.5

0 0.02 0.04 0.06 0.08 0.1 0 0.02 0.04 0.06 0.08 0.1

1.1 1.1

1.05 1.05

1 1

0.95 0.95

0 0.02 0.04 0.06 0.08 0.1 0 0.02 0.04 0.06 0.08 0.1

45
 Sustainable Investing
with ESG Rating Uncertainty
Online Appendix

A Proofs and Derivations

In all derivations that follow, the expectation operators are taken under the joint distribution
of returns and ESG ratings.

A.1 Equilibrium Equity Premium in Multiple-Agent One-Asset Economy

Based on equation (4), the optimal market demand for agent i is

1 µM + bi µg,M
x∗i = 2
, (A.1)
γi σi,U

2 2
where σi,U = σM 2
+ b2i σg,M + 2bi σM σg,M ρg,M . Aggregating across agents, we impose market
PI
clearing by setting i=1 wi x∗i = 1. Thus,

I I
X 1 µM X 1 bi µg,M
wi 2
+ wi 2
= 1. (A.2)
i=1
γi σi,U i=1
γi σi,U

Finally, we solve for the equilibrium equity premium:

2
µM = γM σM,U − bM µg,M,U , (A.3)

where

1
γM = PI 1
, (A.4)
i=1 wi γi
PI −1
2 i=1 wi γi
σM,U = PI −1 1
, (A.5)
i=1 wi γi σi,U
2

A-1
 PI
wi γi−1 bi
bM = Pi=1
I −1
, (A.6)
i=1 wi γi
PI
i=1 wi γi σbi2
i,U
µg,M,U = bM
µg,M . (A.7)
2
γM σM,U

A.2 Proof of Proposition 1: Optimal Portfolio under ESG Uncertainty

The expected utility of agent i can be written as
h  i h i
−Ai W̃i,1 −Bi Wi,0 Xi0 g̃
E V W̃i,1 , Xi = E −e
h i
= E −e−Ai Wi,0 (1+rf +Xi r̃)−Bi Wi,0 Xi g̃
0 0

h 0
i
= −e−γi (1+rf ) E e−γi Xi (r̃+bi g̃) , (A.8)

Bi
where γi = Ai Wi,0 and bi = Ai
. Note that

r̃ + bi g̃ ∼ N (µr + bi µg , Σi,U ) , (A.9)

where Σi,U = Σr + b2i Σg + 2bΣrg . Then, the expected utility in (A.8) can be developed
analytically
γi2
h  i 0 0
E V W̃i,1 , Xi = −e−γi (1+rf ) e−γi Xi (µr +bi µg )+ 2 Xi Σi,U Xi . (A.10)

The investor chooses the optimal portfolio weights by maximizing the expression

γi2 0
γi Xi0 (µr + bi µg ) − X Σi,U Xi . (A.11)
2 i

The first-order condition is

0 = −γi (µr + bi µg ) + γi2 Σi,U Xi . (A.12)

The optimal portfolio that solves the above equation is then

1 −1
Xi∗ = Σ (µr + bi µg ) . (A.13)
γi i,U

Then, we express the inverse of the covariance matrix of the return bundle as

−1
Σ−1 2
i,U = Σr + bi Σg + 2bi Σrg = Σ−1
r + Ψi , (A.14)

A-2
where

−1
Ψi = Σ−1 −1 −1
b2i Σg + 2bi Σrg Σ−1 IN + b2i Σg + 2bi Σrg Σ−1



i,U − Σr = −Σr r r . (A.15)

The optimal strategy can finally be written as

1 −1 1
Xi∗ = Σr (µr + bi µg ) + Ψi (µr + bi µg ) . (A.16)
γi γi

A.3 Proof of Proposition 2: Expected Returns without ESG Uncertainty

As the riskless asset is in zero net supply, the market portfolio consists exclusively of risky
assets and is given by

I I
X Wi,0 ∗ X 1 −1
XM = X = wi Σr (µr + bi µg ) , (A.17)
i=1
WM,0 i i=1
γi

Wi,0
where wi = WM,0
and µr has to be determined in equilibrium. Then, it follows that

I
! I
!
Σ r XM 2 X 1 X bi
2
σM = wi µr + wi µg , (A.18)
σM i=1
γi i=1
γi

which entails that

Σr XM 2
µr = 2
γM σM − bM µg , (A.19)
σM
P −1 PI −1
I −1 i=1 wi γi bi
where γM = i=1 wi γi and b M = P I −1 . The expected excess return of the
i=1 wi γi
market portfolio is
0 2 0
µ M = XM µr = γM σM − b M XM µg , (A.20)

which yields
0 2
µ M + b M XM µg = γM σM , (A.21)

and, finally, combining (A.19) and (A.21) leads to

Σ r XM Σ r XM 0
µr = 2
µ M + bM 2
XM µg − bM µg . (A.22)
σM σM

Σr XM
Recalling that β = σM2 is the N -vector of market betas, the result follows.

A-3
A.4 Proof of Proposition 3: Expected Returns with ESG Uncertainty
Market clearing implies that, as the riskless asset is in zero net supply, the market portfolio
consists exclusively of risky assets and is given by

I I I
X X 1 −1 X 1
XM = wi Xi∗ = wi Σi,U µr + wi bi Σ−1
i,U µg , (A.23)
i=1 i=1
γi i=1
γi

where µr has to be determined in equilibrium. We introduce the following notation:

I
X
−1
γM = wi γi−1 , (A.24)
i=1
PI −1 −1
i=1 wi γi Σi,U
Σ−1
M,U = PI −1
, (A.25)
i=1 wi γi
I
X
BM = γM ΣM,U wi γi−1 bi Σ−1
i,U . (A.26)
i=1

We then have
1 −1 1 −1
XM = ΣM,U µr + Σ BM µg , (A.27)
γM γM M,U
and hence
Σ r XM 2 1 1
2
σM = Σr Σ−1
M,U µr + Σr Σ−1
M,U BM µg , (A.28)
σM γM γM
where ΣrσX
2
M
is the vector of equilibrium market betas, β. Therefore, solving for the vector
M
of expected excess returns µr :

Σ r XM 2
µr = γM ΣM,U Σ−1
r 2
σM − BM µg . (A.29)
σM

In addition, we aggregate to obtain the market expected excess return µM :

0 0 0
µ M = XM µr = γM XM ΣM,U XM − XM BM µg . (A.30)

The vector of expected excess asset returns is then given by

 
ΣM,U XM ΣM,U XM 0
µr = 0 µM − BM µg − 0 X BM µg . (A.31)
XM ΣM,U XM XM ΣM,U XM M

A-4
From (A.30) and (A.31), we can therefore write that the equilibrium market and stock
expected excess returns are

2
µM = γM σM,U − bM µg,M,U , (A.32)
µr = βeff µM − bM (µg,U − βeff µg,M,U ) , (A.33)

where
I
X
−1
γM = wi γi−1 , (A.34)
i=1
PI −1
i=1 wi γi bi
bM = PI −1
, (A.35)
i=1 wi γi
PI −1 −1
i=1 wi γi Σi,U
Σ−1
M,U = PI −1
, (A.36)
i=1 wi γi
I
!−1 I
X X
BM = wi γi−1 Σ−1
i,U wi γi−1 bi Σ−1
i,U , (A.37)
i=1 i=1
2 0
σM,U = XM ΣM,U XM , (A.38)
0
µg,M,U = XM µg,U , (A.39)
1
µg,U = BM µg , (A.40)
bM
ΣM,U XM
βeff = 0 . (A.41)
XM ΣM,U XM

Under the hypothesis that agents are homogeneous in preferences (γi = γ and bi = b, ∀i),
several simplifications are possible, leading to

γM = γ, (A.42)
bM = b, (A.43)
ΣM,U = Σr + b2 Σg + 2bΣrg , (A.44)
BM = bIN , (A.45)
2 0 0 0
σM,U = XM Σ X +b2 XM Σ X +2b XM Σrg XM , (A.46)
| {zr M} | {zg M} | {z }
2
σM 2 σrg,M
σg,M

0
µg,M,U = XM µg , (A.47)
µg,U = µg , (A.48)
(Σr + b2 Σg + 2bΣrg ) XM 2
σM b2 σg,M
2
2bσrg,M
βeff = 2 2
= 2
β + 2
β g + 2
βrg , (A.49)
σM + b2 σg,M + 2bσrg,M σM,U σM,U σM,U

A-5
 Σr XM Σg XM Σrg XM
where β = σM2 , βg = 2
σg,M
, and βrg = σrg,M
. The difference βeff − β is therefore

b2 σg,M
2
2bσrg,M
βeff − β = 2
(βg − β) + 2 (βrg − β) . (A.50)
σM,U σM,U

Recalling equation (18), the N -vector of alpha can be expressed as

α = µr − βµM = (βeff − β) (µM + bM µg,M ) − bM (µg − βµg,M )

!
b2 σg,M
2
2bσrg,M
= 2
(βg − β) + 2 (βrg − β) (µM + bM µg,M ) − bM (µg − βµg,M ) . (A.51)
σM,U σM,U

A.5 Derivations for Two-Asset Economy

In the two-risky-asset case, we consider that ESG rating uncertainty is in play and use
equation (14) to derive the optimal portfolio. Denoting by I2 the 2 × 2 identity matrix, we
assume that
   
2
σg,green 0 σ rg,green 0
Σr = σr2 I2 , Σg =  , Σrg =  . (A.52)
2
0 σg,brown 0 σrg,brown

In this case, we have

Σi,U = Σr + b2i Σg + 2bi Σrg

 
2 2 2
σr + bi σg,green + 2bi σrg,green 0
= , (A.53)
2 2 2
0 σr + bi σg,brown + 2bi σrg,brown

and then  
1
σr2 +b2i σg,green
2 +2bi σrg,green
0
Σ−1
i,U =
 . (A.54)
1
0 σr2 +b2i σg,brown
2 +2bi σrg,brown

We also assume that a green firm has a mean ESG score µg > 0, while the brown firm has a
mean score −µg . We can then write the optimal portfolio strategy as

∗ 1 µr,green + bi µg
Xi,green = 2 2 2
, (A.55)
γi σr + bi σg,green + 2bi σrg,green
∗ 1 µr,brown − bi µg
Xi,brown = 2 2
. (A.56)
γi σr2 + bi σg,brown + 2bi σrg,brown

A-6
Notice that, for σg,green , σg,brown > 0 and σrg,green , σrg,brown ≥ 0,

∗ 1 µr,green + bi µg
lim Xi,green = lim 2
= 0, (A.57)
bi →∞ bi →∞ γi σr2 + bi σg,green
2 + 2bi σrg,green
∗ 1 µr,brown − bi µg
lim Xi,brown = lim = 0. (A.58)
bi →∞ γi σr + bi σ 2
2 2
g,brown + 2bi σrg,brown
bi →∞

We now attempt to determine the equilibrium expected excess returns of the two risky
assets. We consider that there are two categories of agents, ESG and IND, whose fractions
of total wealth are wESG and wIND . The first category has a brown aversion bESG , while
bIND = 0. The relative risk aversion is γ for all agents. Recalling (A.29), the equilibrium
excess returns are µr = γM ΣM,U Σ−1 2
r βσM − BM µg , where:

γM = γ, (A.59)
 
σr2 +b2ESG σg,green
2 +2bESG σrg,green
 2 2
 1+(1−wESG ) ESG σg,green
b 2bESG σrg,green
 0
+

 2
σr 2 σr 
ΣM,U = σr2 +b2ESG σbrown
2 +2bESG σrg,brown
, (A.60)
0
 
!
b2 σ2
 2bESG σrg,brown
ESG g,brown
1+(1−wESG ) 2 + 2
σr σr
 
wESG bESG

b2 σ2 2bESG σrg,green
 0
 1+(1−wESG ) ESG g,green +

 2
σr 2
σr 
BM = wESG bESG
. (A.61)
0
 
!
b2 σ2
 2b σ
ESG g,brown
1+(1−wESG ) 2 + ESG rg,brown
2
σr σr

Consequently, the two assets have expected excess returns

σ2
 
2 σrg,green
βgreen γσM 1 + b2ESG g,green
σr2
+ 2b ESG σr2
− wESG bESG µg
µr,green = 
σ2
 , (A.62)
σrg,green
1 + (1 − wESG ) b2ESG g,greenσr2
+ 2b ESG σr2
 σ 2 
2 σ
βbrown γσM 1 + b2ESG g,brown
σr2
+ 2bESG rg,brown
σr2
+ wESG bESG µg
µr,brown =  σ2
 . (A.63)
σrg,brown
1 + (1 − wESG ) b2ESG g,brownσ2
+ 2b ESG σ2
r r

Equation (27) is obtained taking the difference between (A.63) and (A.62) for βgreen = βbrown ,
2 2
σg,green = σg,brown = σg2 , and σrg,green = σrg,brown = σrg .

A-7
A.6 Welfare in a One-Asset Economy
In the single-asset setup, the expected utility for optimizing agent i is given by
h  i h ∗
i
E V W̃i,1 , x∗i = −E e−AW̃i,1 −BWi,0 xi g̃M . (A.64)

Considering the optimal market demand in (A.1) in the presence of ESG uncertainty, the
value function can be evaluated as
"  2 #
1 µM +bi µg,M
h  i −γi (1+rf )−γi 2γi σi,U
E V W̃i,1 , x∗i = −e . (A.65)

Following Back (2010),1 the term in square brackets represents the investor’s certainty equi-
valent excess rate of return CE i .

B Supplementary Material
To save space, we tabulate the supplementary material in this section and discuss our main
findings in the paper.

1
See equations 2.17 and 2.18 on page 39.

A-8
 Table B.1: Number of Stocks Over Time
Panel A reports the number of stocks covered by each data vendor on a year-by-year basis. Panel B reports the number of stocks
simultaneously covered by N data vendors on a year-by-year basis, where N ranges between 1 and 5.

Panel A: Number of Stocks Covered By Each Data Vendor

Year Asset4 MSCI KLD MSCI IVA Bloomberg Sustainalytics RobecoSAM

2002 398 1,055 0 0 0 0

2003 400 2,805 0 0 0 0
2004 535 2,851 0 0 0 0
2005 600 2,687 0 125 0 0
2006 606 2,655 528 209 0 0
2007 620 2,566 609 709 0 0
2008 789 2,580 600 984 0 0
2009 892 2,598 599 1,065 0 0
2010 915 2,630 551 1,957 0 0
2011 912 2,472 537 2,077 0 0
2012 895 2,418 2,253 2,149 0 0
2013 890 2,125 2,388 2,242 0 0
2014 885 2,098 2,328 2,380 413 0
2015 1,436 2,124 2,282 2,514 441 0
2016 2,083 0 2,255 2,530 460 419
2017 2,218 0 2,139 2,658 452 616
2018 2,178 0 2,104 2,794 473 818

Panel B: Number of Stocks Covered By Multiple Data Vendors

Year N =1 N =2 N =3 N =4 N =5 N ≥2

2002 677 388 0 0 0 388

2003 2409 398 0 0 0 398
2004 2324 531 0 0 0 531
2005 2199 518 59 0 0 577
2006 2069 241 349 100 0 690
2007 1756 380 264 299 0 943
2008 1579 505 320 351 0 1,176
2009 1601 487 373 365 0 1,225
2010 1240 1,093 385 368 0 1,846
2011 1136 1,109 392 367 0 1,868
2012 631 702 1,060 625 0 2,387
2013 741 591 1,038 652 0 2,281
2014 781 586 1,030 289 381 2,286
2015 851 341 811 669 431 2,252
2016 797 645 1,119 87 391 2,242
2017 781 512 1,140 162 442 2,256
2018 817 425 1,042 336 446 2,249

A-9
 Table B.2: Variable Definitions

Variables Definitions
Panel A: ESG Rating Measures
ESG We collect ESG rating data from six data vendors: Asset4 (Refinitiv), MSCI KLD, MSCI IVA, Bloomberg,
Sustainalytics, and RobecoSAM. For each rater pair-year, we sort all stocks covered by both raters according
to the original rating scale of the respective data provider and calculate the percentile rank (normalized
between 0 and 1) for each stock-rater pair. Then for each stock, we compute the pairwise average rating
as the average rank across the two raters in the pair. Finally, we compute the firm-level ESG rating as the
average pairwise rank across all rater pairs.
ESG Uncertainty For each rater pair-year, we sort all stocks covered by both raters according to the original rating scale of
the respective data provider and calculate the percentile rank (normalized between 0 and 1) for each stock-
rater pair. Then, for each stock, we compute the pairwise rating uncertainty as the standard deviation of
the ranks provided by the two raters in the pair. Finally, we compute the firm-level ESG rating uncertainty
as the average pairwise rating uncertainty across all rater pairs.
ESGALL For each rater-year, we sort all stocks covered by this rater according to the original rating scale and
calculate the percentile rank (normalized between 0 and 1) for each stock. We compute the firm-level ESG
rating as the average rank across all raters.
ESG UncertaintyALL For each rater-year, we sort all stocks covered by this rater according to the original rating scale and
calculate the percentile rank (normalized between 0 and 1) for each stock. We compute the firm-level ESG
rating uncertainty as the standard deviation of the ranks provided by all raters.
Panel B: Other Stock Characteristics
Excess Return Stock return minus the one-month Treasury bill rate in a given month.
CAPM-Adjusted Return Stock excess return minus the product of a stock’s beta and excess return on the market in a given month.
The excess return on the market is computed as the CRSP value-weighted index return minus the one-
month Treasury bill rate. The beta of the stock is estimated in a five-year rolling window.
P
IO The institutional ownership in a given quarter q, computed as follows: IO i,q = f SHR i,f,q /SHROUT i,q ,
where SHR i,f,q refers to the number of shares of stock i held by institution f in quarter q, and SHROUT i,q
refers to the shares outstanding at the same time. For each stock, we separately compute the ownership
of three types of institutions: norm-constrained institutions, hedge funds, and other institutions. Specific-
ally, we disaggregate the 13F institutional holdings based on institution type, namely, bank trust (type
1), insurance company (type 2), investment company (type 3), independent investment advisor (which in-
cludes hedge funds, type 4), and others (including corporate/private pension funds, public pension funds,
university and foundation endowments, and miscellaneous, type 5), following Abarbanell et al. (2003). We
consider types 1, 2, and 5 as norm-constrained institutions (Hong and Kacperczyk (2009)) and manually
collect a list of names of hedge fund companies (Agarwal et al. (2013)). The remaining institutions are
classified as other institutions, i.e., types 3 and 4, excluding hedge funds.
Log(Size) The logarithm of stock market capitalization, computed as the number of common shares outstanding
times the share price as reported in CRSP.
Log(BM) The logarithm of the book-to-market ratio of a stock, where the book-to-market ratio is computed as the
book value of equity divided by market capitalization at fiscal year-end, following Fama and French (2015).
6M Momentum Formation period return for six-month momentum in a given month m, computed as the cumulative return
from month m − 6 to month m − 1, following Jegadeesh and Titman (1993).
Log(Illiquidity) The logarithm of stock illiquidity. Stock illiquidity in a given month m is computed as follows: ILLIQ i,m =
8
P

d∈m Ri,d,m /VOLD i,d,m /Di,m × 10 , where Ri,d,m refers to the percentage return of stock i in day
d of month m, VOLD i,d,m refers to the dollar trading volume at the same time, and Di,m is the number
of trading days for stock i in month m, following Amihud (2002).
Gross Profitability Gross profitability in a given year t, computed as follows: GP i,t = (REVT i,t − COGS i,t ) /ASSET i,t ,
where REVT i,t refers to the total revenue (COMPUSTAT annual item REVT) of stock i in year t, COGS i,t
refers to the cost of goods sold (item COGS), and ASSET i,t is the total assets (item AT), following Novy-
Marx (2013).
Corporate Investment Corporate investment in a given quarter q, computed as follows: CI i,q = PPE i,q −
(PPE i,q−1 + PPE i,q−2 + PPE i,q−3 ) /3, where PPE i,q refers to the ratio of change in net property, plant,
and equipment (COMPUSTAT quarterly item PPENTQ) divided by sales (item SALEQ) of stock i in
quarter q. If SALEQ is 0 or negative, then replace SALEQ with 0.01, following Titman et al. (2004).
Leverage Total liabilities (COMPUSTAT annual item LT) divided by market capitalization at fiscal year-end, fol-
lowing Bhandari (1988).
Log(Analyst Coverage) The logarithm of the number of analysts following the firm as reported in I/B/E/S in each quarter.
Analyst Dispersion The standard deviation of analysts’ earnings (earnings per share, EPS) forecasts divided by the absolute
value of the median earnings forecast as reported in I/B/E/S in each quarter.

A - 10
 Table B.3: Summary Statistics
We collect ESG rating data from six data vendors: Asset4 (Refinitiv), MSCI KLD, MSCI IVA, Bloomberg, Sustainalytics, and Robe-
coSAM. Panel A reports the average ESG rating uncertainty for each rater pair. For each rater pair-year, we sort all stocks covered by
both raters according to the original rating scale of the respective data provider and calculate the percentile rank (normalized between
0 and 1) for each stock-rater pair. Then, for each stock, we compute the pairwise rating uncertainty as the standard deviation of the
ranks provided by the two raters in the pair. Finally, we compute the average ESG rating uncertainty across all stocks for each rater
pair-year and average them over time. In Panel B, we compute the correlation in the percentile ranks for each rater pair-year and
then average them over time. Panel C presents the summary statistics for the stock-level data used in the paper. We report the mean,
standard deviation, median, and quantile distribution of the annual ESG rating and ESG rating uncertainty, monthly stock performance,
quarterly institutional ownership, and other annual and monthly stock characteristics. Panel D presents the summary statistics for the
portfolio-level data used in the paper. At the end of year t, stocks are independently sorted into quintiles according to their ESG ratings
and ESG rating uncertainty to generate 25 (5 × 5) portfolios. For each of the 25 portfolios, we compute the value-weighted ESG rating
and ESG rating uncertainty in year t + 1, and the value-weighted return in each month in year t + 1. We rebalance the portfolios at the
end of year t + 1. We report the time-series averages of annual ESG rating and ESG rating uncertainty, and monthly return for each of
the 25 portfolios. Our sample period ranges from 2002 to 2019. Table B.2 provides detailed definitions for each variable.

Panel A: Pairwise ESG Rating Uncertainty

Asset4 MSCI KLD MSCI IVA Bloomberg Sustainalytics RobecoSAM

Asset4 - 0.185 0.185 0.134 0.144 0.149

MSCI KLD 0.185 - 0.180 0.183 0.151 -
MSCI IVA 0.185 0.180 - 0.195 0.171 0.181
Bloomberg 0.134 0.183 0.195 - 0.133 0.138
Sustainalytics 0.144 0.151 0.171 0.133 - 0.119
RobecoSAM 0.149 - 0.181 0.138 0.119 -

Panel B: Pairwise ESG Rating Correlation

Asset4 MSCI KLD MSCI IVA Bloomberg Sustainalytics RobecoSAM

Asset4 - 0.321 0.326 0.639 0.595 0.547

MSCI KLD 0.321 - 0.349 0.310 0.547 -
MSCI IVA 0.326 0.349 - 0.253 0.411 0.353
Bloomberg 0.639 0.310 0.253 - 0.677 0.645
Sustainalytics 0.595 0.547 0.411 0.677 - 0.707
RobecoSAM 0.547 - 0.353 0.645 0.707 -

Panel C: Quantile Distribution of Stock Characteristics

Mean Std.Dev. Quantile Distribution

10% 25% Median 75% 90%

ESG 0.461 0.202 0.219 0.310 0.437 0.595 0.753

ESG Uncertainty 0.180 0.112 0.051 0.097 0.162 0.246 0.330
ESGALL 0.490 0.206 0.239 0.337 0.472 0.630 0.788
ESG UncertaintyALL 0.207 0.124 0.051 0.110 0.195 0.291 0.373
Return 1.049 11.171 -11.257 -4.627 1.005 6.443 12.964
Excess Return 0.980 11.174 -11.330 -4.698 0.936 6.373 12.897
CAPM-Adjusted Return -0.195 9.796 -10.840 -5.028 -0.231 4.410 10.128
Norm-Constrained IO 0.182 0.097 0.051 0.117 0.189 0.243 0.295
Hedge Fund IO 0.173 0.110 0.047 0.101 0.159 0.227 0.310
Other IO 0.381 0.168 0.132 0.289 0.401 0.494 0.577
Log(Size) 14.726 1.608 12.703 13.575 14.669 15.792 16.890
Log(BM) -0.772 0.808 -1.819 -1.243 -0.688 -0.213 0.149
6M Momentum 0.054 0.264 -0.235 -0.085 0.045 0.175 0.333
Log(Illiquidity) -7.119 2.079 -9.698 -8.647 -7.278 -5.736 -4.303
Gross Profitability 0.313 0.303 0.037 0.109 0.273 0.460 0.696
Corporate Investment 0.159 7.035 -0.077 -0.018 0.000 0.018 0.092
Leverage 1.596 3.222 0.090 0.227 0.546 1.378 4.558
Log(Analyst Coverage) 2.175 0.815 1.099 1.609 2.303 2.773 3.135
Analyst Dispersion 0.121 0.365 0.007 0.014 0.030 0.079 0.224

A - 11
 Table B.3 (continued)

Panel D: Portfolio Characteristics Sorted by ESG Rating and Uncertainty

ESG Rating ESG Uncertainty

Low 2 3 4 High

Panel D1: ESG Rating

Low 0.252 0.259 0.277 0.287 0.305

2 0.368 0.386 0.384 0.390 0.373
3 0.474 0.479 0.467 0.487 0.483
4 0.575 0.589 0.600 0.603 0.597
High 0.848 0.811 0.753 0.732 0.702

Panel D2: ESG Uncertainty

Low 0.135 0.146 0.179 0.215 0.275

2 0.159 0.152 0.172 0.207 0.317
3 0.131 0.156 0.179 0.207 0.293
4 0.139 0.136 0.172 0.208 0.295
High 0.093 0.114 0.160 0.193 0.260

Panel D3: Return

Low 1.117 1.149 0.733 1.003 0.916

2 1.353 1.133 0.976 1.061 0.831
3 1.009 0.855 0.951 1.101 1.050
4 0.857 0.856 1.169 1.147 0.916
High 0.664 0.784 0.840 1.158 1.026

A - 12
 Table B.4: Risk-Adjusted Performance of Portfolios Sorted by ESG Rating and Uncertainty
At the end of year t, stocks are first sorted into quintiles according to their ESG rating uncertainty. Within each ESG rating uncertainty
group, stocks are further sorted into quintiles according to their ESG ratings to generate 25 (5×5) portfolios. The low- (high)-ESG-
rating and ESG-rating-uncertainty portfolios comprise the bottom (top) quintile of stocks based on the ESG rating and ESG rating
uncertainty, respectively. For each of the 25 portfolios, we compute the value-weighted return in each month in year t + 1 and rebalance
the portfolios at the end of year t + 1. Panel A reports the time-series averages of monthly Fama-French-Carhart 4-factor-adjusted
returns (FFC) for each of the 25 portfolios, as well as for the investment strategy of going long (short) the low- (high)-ESG-rating stocks
(“LMH-R”). The column “All” reports similar statistics for portfolios sorted by ESG ratings only. The row “All” reports returns for
portfolios sorted by ESG uncertainty only, as well as the investment strategy of going long (short) the high (low) ESG-uncertainty stocks
(“HML-U”). In Panel B, portfolio returns are adjusted by the Fama-French 6-factor model (FF6). Table B.2 provides detailed definitions
for each variable. Newey-West adjusted t-statistics are shown in parentheses. Numbers with “*”, “**”, and “***” are significant at the
10%, 5%, and 1% levels, respectively.

Panel A: FFC-Adjusted Return Panel B: FF6-Adjusted Return

ESG Rating ESG Uncertainty ESG Uncertainty

Low 2 3 4 High All Low 2 3 4 High All

Low 0.214 0.054 -0.329* -0.115 -0.113 -0.091 0.251 0.091 -0.327* -0.155 -0.030 -0.092
(1.28) (0.37) (-1.91) (-0.76) (-0.64) (-0.76) (1.49) (0.65) (-1.88) (-1.02) (-0.15) (-0.78)
2 0.209 0.099 0.095 0.062 0.140 -0.005 0.189 0.193 0.019 0.055 0.131 -0.001
(1.37) (0.49) (0.69) (0.44) (0.70) (-0.04) (1.15) (1.03) (0.13) (0.37) (0.65) (-0.01)
3 0.111 0.006 0.018 0.090 0.048 0.051 0.113 0.031 0.043 0.171 -0.009 0.052
(0.65) (0.05) (0.16) (0.65) (0.26) (0.63) (0.69) (0.22) (0.36) (1.24) (-0.05) (0.61)
4 -0.215 -0.344*** 0.172 0.093 0.042 0.124* -0.179 -0.301*** 0.145 0.094 0.119 0.117
(-1.41) (-2.94) (0.92) (0.77) (0.20) (1.71) (-1.24) (-2.62) (0.77) (0.77) (0.58) (1.60)
High -0.246** -0.041 0.012 0.304** -0.012 -0.090 -0.250** -0.017 -0.049 0.297** -0.084 -0.094*
(-2.13) (-0.39) (0.10) (2.25) (-0.09) (-1.61) (-2.13) (-0.16) (-0.39) (2.04) (-0.65) (-1.71)

LMH-R 0.459** 0.095 -0.341 -0.419* -0.101 -0.002 0.501** 0.109 -0.277 -0.452** 0.054 0.003
(2.30) (0.49) (-1.42) (-1.96) (-0.42) (-0.01) (2.36) (0.61) (-1.14) (-2.02) (0.22) (0.02)

ESG Rating ESG Uncertainty ESG Uncertainty

Low 2 3 4 High HML-U Low 2 3 4 High HML-U

All -0.162** -0.064 0.013 0.163** 0.056 0.218 -0.154** -0.027 -0.045 0.157** 0.035 0.189
(-2.15) (-0.89) (0.17) (2.44) (0.66) (1.65) (-2.02) (-0.38) (-0.58) (2.38) (0.41) (1.42)

A - 13
Table B.5: Risk-Adjusted Performance of Portfolios Sorted by ESG Rating and Uncertainty:
Subsample Analysis
At the end of year t, stocks are first sorted into quintiles according to their ESG rating uncertainty. Within each ESG rating uncertainty
group, stocks are further sorted into quintiles according to their ESG ratings to generate 25 (5×5) portfolios. The low- (high)-ESG-
rating and ESG-rating-uncertainty portfolios comprise the bottom (top) quintile of stocks based on the ESG rating and ESG rating
uncertainty, respectively. For each of the 25 portfolios, we compute the value-weighted return in each month in year t + 1 and rebalance
the portfolios at the end of year t + 1. Panel A reports the time-series averages of monthly Fama-French-Carhart 4-factor-adjusted
returns (FFC) for each of the 25 portfolios, as well as for the investment strategy of going long (short) the low- (high)-ESG-rating
stocks (“LMH-R”). The column “All” reports similar statistics for portfolios sorted by ESG ratings only. The row “All” reports returns
for portfolios sorted by ESG uncertainty only, as well as the investment strategy of going long (short) the high (low) ESG-uncertainty
stocks (“HML-U”). We divide the full sample into two subperiods, and report results for 2003–2010 on the left and 2011–2019 on the
right. In Panel B, portfolio returns are adjusted by the Fama-French 6-factor model (FF6). Table B.2 provides detailed definitions for
each variable. Newey-West adjusted t-statistics are shown in parentheses. Numbers with “*”, “**”, and “***” are significant at the 10%,
5%, and 1% levels, respectively.

Panel A: FFC-Adjusted Return

2003–2010 2011–2019
ESG Rating ESG Uncertainty ESG Uncertainty
Low 2 3 4 High All Low 2 3 4 High All
Low 0.459* -0.020 -0.256 0.101 0.018 -0.035 0.124 0.132 -0.198 -0.210 -0.203 -0.026
(1.69) (-0.10) (-0.86) (0.40) (0.07) (-0.19) (0.67) (0.75) (-1.03) (-1.15) (-0.76) (-0.22)
2 0.327 0.148 0.213 0.141 0.288 0.221 0.138 0.335 0.059 0.046 0.245 -0.118
(1.51) (0.50) (0.87) (0.70) (0.86) (1.53) (0.77) (1.26) (0.41) (0.24) (1.17) (-1.04)
3 -0.054 0.131 0.111 0.265 0.341 0.046 0.358* -0.214 -0.050 -0.150 -0.075 0.061
(-0.22) (0.62) (0.61) (1.28) (1.06) (0.32) (1.68) (-1.23) (-0.47) (-0.79) (-0.35) (0.73)
4 -0.122 -0.254 0.164 -0.071 0.402* 0.269*** -0.220 -0.243* 0.171 0.264 -0.281 0.034
(-0.54) (-1.27) (0.44) (-0.37) (1.86) (2.84) (-1.02) (-1.89) (1.60) (1.58) (-0.80) (0.33)
High -0.366* -0.397** -0.081 0.488** 0.071 -0.229** -0.094 0.259*** 0.051 -0.010 0.057 0.056
(-1.83) (-2.56) (-0.47) (2.54) (0.26) (-2.47) (-0.74) (2.96) (0.49) (-0.06) (0.40) (1.35)
LMH-R 0.825** 0.377 -0.175 -0.387 -0.052 0.194 0.218 -0.127 -0.249 -0.200 -0.260 -0.082
(2.41) (1.38) (-0.49) (-1.22) (-0.12) (0.77) (1.02) (-0.63) (-1.02) (-0.78) (-0.81) (-0.62)
ESG Rating ESG Uncertainty ESG Uncertainty
Low 2 3 4 High HML-U Low 2 3 4 High HML-U
All -0.220* -0.238** -0.021 0.219* 0.294** 0.513*** -0.066 0.155** 0.038 0.029 -0.013 0.053
(-1.71) (-2.61) (-0.15) (1.90) (2.60) (2.69) (-0.79) (2.40) (0.63) (0.36) (-0.14) (0.33)
Panel B: FF6-Adjusted Return
2003–2010 2011–2019
ESG Rating ESG Uncertainty ESG Uncertainty
Low 2 3 4 High All Low 2 3 4 High All
Low 0.480 -0.114 -0.301 0.030 0.100 -0.109 0.163 0.232 -0.165 -0.223 -0.126 0.013
(1.63) (-0.62) (-1.02) (0.12) (0.34) (-0.62) (0.92) (1.45) (-0.87) (-1.17) (-0.45) (0.12)
2 0.205 0.304 -0.086 0.128 0.203 0.161 0.155 0.374 0.125 0.056 0.267 -0.075
(0.90) (1.07) (-0.37) (0.65) (0.59) (1.12) (0.80) (1.49) (0.86) (0.28) (1.23) (-0.65)
3 -0.078 0.085 0.057 0.320 0.254 -0.019 0.325 -0.170 0.005 -0.061 -0.094 0.090
(-0.29) (0.38) (0.30) (1.41) (0.77) (-0.13) (1.65) (-0.97) (0.05) (-0.34) (-0.47) (1.02)
4 -0.108 -0.160 -0.079 -0.056 0.469** 0.251*** -0.189 -0.251** 0.197* 0.224 -0.194 0.030
(-0.44) (-0.85) (-0.21) (-0.28) (2.14) (2.96) (-0.97) (-1.98) (1.91) (1.34) (-0.61) (0.28)
High -0.291 -0.340** -0.160 0.503** -0.061 -0.210** -0.136 0.239** -0.011 -0.035 0.060 0.030
(-1.39) (-2.05) (-0.87) (2.34) (-0.24) (-2.18) (-1.08) (2.57) (-0.11) (-0.23) (0.43) (0.73)
LMH-R 0.770* 0.227 -0.141 -0.473 0.160 0.101 0.299 -0.006 -0.153 -0.187 -0.186 -0.018
(1.98) (0.91) (-0.39) (-1.43) (0.38) (0.41) (1.41) (-0.03) (-0.66) (-0.70) (-0.56) (-0.14)
ESG Rating ESG Uncertainty ESG Uncertainty
Low 2 3 4 High HML-U Low 2 3 4 High HML-U
All -0.165 -0.178* -0.162 0.223** 0.233* 0.398* -0.086 0.154** 0.014 0.012 0.013 0.099
(-1.22) (-1.81) (-1.32) (2.01) (1.91) (1.99) (-1.03) (2.30) (0.22) (0.16) (0.14) (0.65)

A - 14
Table B.6: Institutional Ownership of Portfolios Sorted by ESG Rating and Uncertainty Based on
Alternative Definition
At the end of year t, stocks are independently sorted into quintiles according to their ESG ratings (i.e., ESGALL ) and ESG rating
uncertainty (i.e., ESG UncertaintyALL ) to generate 25 (5 × 5) portfolios. The low- (high)-ESG-rating and ESG-rating-uncertainty
portfolios comprise the bottom (top) quintile of stocks based on the ESG rating and ESG rating uncertainty, respectively. For each of
the 25 portfolios, we compute the average institutional ownership in each quarter in year t + 1 and rebalance the portfolios at the end
of year t + 1. Panel A reports the time-series averages of quarterly institutional ownership of norm-constrained institutions for each of
the 25 portfolios and the average difference in institutional ownership between high- and low-ESG-rating portfolios (“HML-R”), as well
as between high- and low-ESG-rating-uncertainty portfolios (“HML-U”). Panels B and C report similar statistics for average ownership
of hedge funds and other institutions, respectively. Table B.2 provides detailed definitions for each variable. Newey-West adjusted
t-statistics are shown in parentheses. Numbers with “*”, “**”, and “***” are significant at the 10%, 5%, and 1% levels, respectively.

Panel A: Norm-Constrained Institutions

ESG RatingALL ESG UncertaintyALL

Low 2 3 4 High HML-U t-stat All

Low 0.166 0.180 0.180 0.184 0.149 -0.016 (-0.64) 0.176

2 0.183 0.189 0.193 0.209 0.192 0.008* (1.68) 0.192
3 0.194 0.203 0.206 0.206 0.198 0.004 (0.64) 0.201
4 0.181 0.208 0.210 0.219 0.218 0.038*** (4.20) 0.212
High 0.234 0.234 0.230 0.232 0.155 -0.078*** (-2.74) 0.232

HML-R 0.068*** 0.054*** 0.049*** 0.048*** 0.006 0.056***

(11.99) (10.74) (8.53) (8.94) (0.28) (11.95)

Panel B: Hedge Funds

ESG RatingALL ESG UncertaintyALL

Low 2 3 4 High HML-U t-stat All

Low 0.157 0.156 0.160 0.162 0.127 -0.029** (-2.51) 0.158

2 0.138 0.149 0.153 0.152 0.148 0.010** (2.29) 0.148
3 0.157 0.144 0.147 0.145 0.154 -0.003 (-0.45) 0.150
4 0.135 0.143 0.141 0.140 0.139 0.004 (0.45) 0.141
High 0.127 0.125 0.129 0.125 0.102 -0.025*** (-3.83) 0.127

HML-R -0.030*** -0.031*** -0.031*** -0.038*** -0.025*** -0.032***

(-5.54) (-8.78) (-6.35) (-6.27) (-3.24) (-8.09)

Panel C: Other Institutions

ESG RatingALL ESG UncertaintyALL

Low 2 3 4 High HML-U t-stat All

Low 0.345 0.360 0.359 0.362 0.287 -0.057 (-1.42) 0.355

2 0.345 0.379 0.375 0.388 0.365 0.021** (2.03) 0.370
3 0.372 0.360 0.376 0.377 0.361 -0.011 (-1.46) 0.367
4 0.339 0.375 0.371 0.374 0.368 0.029 (1.41) 0.371
High 0.370 0.361 0.367 0.356 0.278 -0.091*** (-2.71) 0.365

HML-R 0.025* 0.001 0.008 -0.006 -0.009 0.009

(1.95) (0.07) (0.91) (-0.56) (-0.31) (0.97)

A - 15
Table B.7: Performance of Portfolios Sorted by ESG Rating and Uncertainty Based on Alternative
Definition
At the end of year t, stocks are first sorted into quintiles according to their ESG rating uncertainty (i.e., ESG UncertaintyALL ). Within
each ESG rating uncertainty group, stocks are further sorted into quintiles according to their ESG ratings (i.e., ESGALL ) to generate 25
(5×5) portfolios. The low- (high)-ESG-rating and ESG-rating-uncertainty portfolios comprise the bottom (top) quintile of stocks based
on the ESG rating and ESG rating uncertainty, respectively. For each of the 25 portfolios, we compute the value-weighted return in each
month in year t + 1 and rebalance the portfolios at the end of year t + 1. Panel A reports the time-series averages of monthly returns
for each of the 25 portfolios, as well as for the investment strategy of going long (short) the low- (high)-ESG-rating stocks (“LMH-R”).
The column “All” reports similar statistics for portfolios sorted by ESG ratings only. The row “All” reports returns for portfolios sorted
by ESG uncertainty only, as well as the investment strategy of going long (short) the high (low) ESG-uncertainty stocks (“HML-U”). In
Panel B portfolio returns are further adjusted by the CAPM, in Panel C by the Fama-French-Carhart 4-factor model (FFC), in Panel
D by the Fama-French 6-factor model (FF6). Table B.2 provides detailed definitions for each variable. Newey-West adjusted t-statistics
are shown in parentheses. Numbers with “*”, “**”, and “***” are significant at the 10%, 5%, and 1% levels, respectively.

Panel A: Return Panel B: CAPM-Adjusted Return

ESG RatingALL ESG UncertaintyALL ESG UncertaintyALL

Low 2 3 4 High All Low 2 3 4 High All

Low 1.219*** 0.997*** 1.178*** 0.848** 0.731* 0.995*** 0.134 -0.027 0.155 -0.142 -0.276 -0.049
(3.03) (2.84) (3.29) (2.16) (1.94) (2.70) (0.81) (-0.16) (0.76) (-0.77) (-1.34) (-0.37)
2 1.079*** 1.026*** 1.164*** 0.993*** 0.794** 0.846** 0.045 0.006 0.193 -0.009 -0.163 -0.147
(2.89) (2.69) (3.37) (2.68) (2.01) (2.46) (0.33) (0.03) (1.17) (-0.06) (-0.87) (-1.32)
3 1.009*** 1.028*** 1.168*** 1.165*** 1.090*** 1.019*** 0.007 0.005 0.144 0.197 0.216 0.079
(2.63) (2.83) (3.17) (3.51) (3.34) (3.12) (0.05) (0.03) (0.91) (1.62) (1.35) (0.93)
4 1.055*** 0.543 1.183*** 0.893** 1.001*** 1.045*** 0.115 -0.429*** 0.197 -0.082 0.050 0.118*
(3.13) (1.54) (3.19) (2.56) (2.91) (3.48) (0.80) (-3.07) (1.12) (-0.68) (0.31) (1.77)
High 0.697** 0.774** 0.872*** 1.146*** 1.030*** 0.816*** -0.177 -0.125 -0.050 0.198 0.203 -0.085
(2.08) (2.41) (2.83) (3.56) (3.62) (2.63) (-1.44) (-1.10) (-0.44) (1.48) (1.57) (-1.45)

LMH-R 0.522** 0.223 0.306 -0.298 -0.299 0.179 0.311 0.098 0.204 -0.340 -0.480** 0.037
(2.32) (1.08) (1.34) (-1.23) (-1.29) (1.12) (1.40) (0.42) (0.82) (-1.27) (-1.99) (0.21)

ESG RatingALL ESG UncertaintyALL ESG UncertaintyALL

Low 2 3 4 High HML-U Low 2 3 4 High HML-U

All 0.830** 0.789** 1.035*** 1.001*** 0.954*** 0.124 -0.072 -0.150* 0.071 0.049 0.079 0.150
(2.51) (2.46) (3.28) (3.18) (3.24) (0.90) (-0.83) (-1.94) (0.93) (0.75) (0.92) (1.04)
Panel C: FFC-Adjusted Return Panel D: FF6-Adjusted Return

ESG RatingALL ESG UncertaintyALL ESG UncertaintyALL

Low 2 3 4 High All Low 2 3 4 High All

Low 0.154 -0.007 0.165 -0.154 -0.288 -0.037 0.175 -0.007 0.179 -0.211 -0.333 -0.040
(1.00) (-0.05) (0.87) (-0.82) (-1.38) (-0.29) (1.14) (-0.05) (0.95) (-1.14) (-1.59) (-0.33)
2 0.075 0.030 0.162 0.024 -0.152 -0.143 0.123 0.103 0.203 -0.010 -0.106 -0.132
(0.54) (0.15) (1.03) (0.16) (-0.78) (-1.25) (0.92) (0.57) (1.22) (-0.07) (-0.53) (-1.12)
3 0.032 0.038 0.130 0.221* 0.189 0.083 0.046 0.073 0.124 0.254** 0.190 0.057
(0.24) (0.23) (0.86) (1.93) (1.21) (1.02) (0.32) (0.44) (0.83) (2.15) (1.21) (0.64)
4 0.136 -0.411*** 0.261 -0.038 0.087 0.140** 0.106 -0.371** 0.255 -0.060 0.127 0.155**
(0.98) (-2.84) (1.48) (-0.31) (0.53) (2.06) (0.73) (-2.49) (1.42) (-0.46) (0.75) (2.26)
High -0.200* -0.087 -0.033 0.260* 0.172 -0.080 -0.180 -0.064 -0.085 0.209 0.119 -0.086*
(-1.76) (-0.81) (-0.29) (1.91) (1.30) (-1.55) (-1.58) (-0.60) (-0.74) (1.38) (0.96) (-1.66)

LMH-R 0.354* 0.080 0.198 -0.414 -0.460* 0.043 0.354* 0.057 0.264 -0.420 -0.452** 0.046
(1.75) (0.40) (0.88) (-1.50) (-1.89) (0.27) (1.74) (0.28) (1.20) (-1.44) (-1.99) (0.30)

ESG RatingALL ESG UncertaintyALL ESG UncertaintyALL

Low 2 3 4 High HML-U Low 2 3 4 High HML-U

All -0.080 -0.125* 0.089 0.083 0.065 0.145 -0.062 -0.100 0.052 0.026 0.046 0.108
(-0.98) (-1.72) (1.14) (1.28) (0.76) (1.04) (-0.75) (-1.41) (0.68) (0.39) (0.52) (0.75)

A - 16
 Table B.8: ESG Rating, Uncertainty, and Stock Returns: Alternative Definition for ESG Rating
and Uncertainty
This table presents the results of the following monthly Fama-MacBeth regressions, as well as their corresponding Newey-West adjusted
t-statistics:

Perfi,m = α0 + β1 ESGi,m−1 + β2 ESGi,m−1 × Low ESG Uncertaintyi,m−1 + β3 Low ESG Uncertaintyi,m−1 + β40 Mi,m−1 + ei,m ,

where Perfi,m refers to the excess return (models 1 to 4) or CAPM-adjusted return (models 5 to 8) of stock i in month m, ESGi,m−1
refers to the ESG rating measured by ESGALL , Low ESG Uncertaintyi,m−1 refers to a dummy variable that takes a value of 1 if the
ESG rating uncertainty measured by ESG UncertaintyALL is in the bottom quintile across all stocks in that month and 0 otherwise.
The vector M stacks all other control variables, including the Log(Size), Log(BM), 6M Momentum, Log(Illiquidity), Gross Profitability,
Corporate Investment, Leverage, Log(Analyst Coverage) and Analyst Dispersion. Table B.2 provides detailed definitions for each
variable. Numbers with “*”, “**”, and “***” are significant at the 10%, 5%, and 1% levels, respectively.

Stock Returns Regressed on Lagged ESG Rating and Uncertainty

Excess Return CAPM-Adjusted Return

Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7 Model 8

ESGALL 0.002 0.110 0.003 0.137 0.048 0.161 0.110 0.247

(0.02) (0.81) (0.02) (0.80) (0.29) (1.14) (0.59) (1.48)
ESGALL × Low ESG UncertaintyALL -0.071 -0.138 -0.221* -0.275**
(-0.60) (-0.95) (-1.74) (-2.23)
Low ESG UncertaintyALL 0.117 0.128 0.175 0.174*
(1.38) (1.22) (1.62) (1.67)
Log(Size) -0.100 -0.038 -0.101 -0.036 -0.045 0.109 -0.043 0.111
(-1.31) (-0.28) (-1.29) (-0.27) (-0.59) (0.73) (-0.56) (0.74)
Log(BM) 0.000 0.009 -0.001 0.010 -0.022 0.019 -0.024 0.018
(0.01) (0.19) (-0.01) (0.21) (-0.20) (0.18) (-0.22) (0.18)
6M Momentum 0.334 0.188 0.328 0.188 0.274 0.106 0.272 0.108
(0.79) (0.43) (0.77) (0.43) (0.59) (0.23) (0.58) (0.23)
Log(Illiquidity) 0.056 0.058 0.103** 0.103**
(1.04) (1.08) (2.20) (2.21)
Gross Profitability 0.179 0.189 0.355* 0.363*
(1.05) (1.13) (1.90) (1.94)
Corporate Investment 0.036 0.034 -0.007 -0.010
(0.65) (0.61) (-0.12) (-0.16)
Leverage -0.037 -0.036 -0.035 -0.034
(-0.80) (-0.80) (-0.76) (-0.76)
Log(Analyst Coverage) -0.020 -0.023 -0.174 -0.178
(-0.16) (-0.19) (-1.45) (-1.49)
Analyst Dispersion -0.536*** -0.545*** -0.828*** -0.834***
(-2.74) (-2.75) (-4.32) (-4.24)
Constant 2.308* 1.812 2.299* 1.788 0.596 -0.535 0.530 -0.597
(1.79) (1.10) (1.72) (1.07) (0.45) (-0.28) (0.39) (-0.30)

Obs 283,671 254,873 283,671 254,873 272,728 245,451 272,728 245,451

R-squared 0.046 0.080 0.048 0.082 0.043 0.076 0.045 0.078

A - 17
View publication stats