NO.

977 CRISK: Measuring the

SEPTEMBER 2021

REVISED
Climate Risk Exposure
MARCH 2023
of the Financial System
Hyeyoon Jung | Robert Engle | Richard Berner
CRISK: Measuring the Climate Risk Exposure of the Financial System
Hyeyoon Jung, Robert Engle, and Richard Berner
Federal Reserve Bank of New York Staff Reports, no. 977
September 2021; revised March 2023
JEL classification: Q54, C53, G20

Abstract

We develop a market-based methodology to assess banks’ resilience to climate-related risks and study the
climate-related risk exposure of large global banks. We introduce a new measure, CRISK, which is the
expected capital shortfall of a bank in a climate stress scenario. To estimate CRISK, we construct climate
risk factors and dynamically measure banks’ stock return sensitivity (that is, climate beta) to the climate
risk factor. We validate the climate risk factor empirically and the climate beta estimates by using
granular data on large U.S. banks’ loan portfolios. The measure is useful in quantifying banks’ climate-
related risk exposure through the market risk and the credit risk channels.

Key words: climate risk, financial stability, systemic risk

_________________

Jung: Federal Reserve Bank of New York (email: hyeyoon.jung@ny.frb.org). Engle, Berner: Stern
School of Business, New York University (emails: rengle@stern.nyu.edu, rberner@stern.nyu.edu). This
paper was previously circulated under the title “Climate Stress Testing.”

The authors thank Marcin Kacperczyk, Viral Acharya, Jaehoon Jung, Kristian Blickle, Joao Santos, Julian
di Giovanni, Lee Seltzer, and Linda Goldberg for valuable comments. They also thank discussants and
participants at the Volatility and Risk Institute Conference, EBA EAIA Seminar, Green Swan
Conference, IFABS Oxford Conference, Federal Reserve Bank of New York Seminar, Federal Reserve
Stress Testing Research Conference, Central Bank of Chile Workshop, MIT GCFP Conference, European
Central Bank Workshop, European Conferences of the Econometrics Community Conference,
Australasian Finance and Banking Conference, Federal Reserve Bank of Richmond Seminar, ESSEC-
Amundi Chair Webinar, Federal Reserve Cross-bank Climate Risk Community Seminar, Banque de
France Workshop, Federal Reserve Board CREST Seminar, University of Oklahoma Energy and Climate
Finance Research Conference, the Federal Reserve Bank of New York and Columbia University
Conference, the European Systemic Risk Board Workshop, System Climate Meeting, and Office of the
Comptroller of the Currency Symposium, Banco de Portugal Conference on Financial Intermediation,
Cornell University ESG Investing Research Conference, International Journal of Central Banking
Conference, European Finance Association Meeting, Central Bank Research Association Meeting, Office
of Financial Research Climate Implications for Financial Stability Conference, UN PRI Academic
Network Week, Paris December Meeting, and American Economic Association Meeting. In addition,
they thank Georgij Alekseev, Janavi Janakiraman, Oliver Hannaoui, and Yihao Zhao for excellent
research assistance.

This paper presents preliminary findings and is being distributed to economists and other interested
readers solely to stimulate discussion and elicit comments. The views expressed in this paper are those of
the author(s) and do not necessarily reflect the position of the Federal Reserve Bank of New York or the
Federal Reserve System. Any errors or omissions are the responsibility of the author(s).
To view the authors’ disclosure statements, visit
https://www.newyorkfed.org/research/staff_reports/sr977.html.
1 Introduction

Over the past three decades, the number of climate-related policies adopted globally has
increased significantly (see Exhibit 1). The risk to economic activity from changes in policies
in response to climate risks, such as carbon taxes and green subsidies, is often referred
to as transition risk. Transition risk can adversely affect the real economy through the
banking sector. For example, a shock to borrowers’ transition risk can impair their ability
to repay, which can then lead to an amplified effect on banks’ current and expected future
profits, resulting in a systemic undercapitalization of banks. It is well known that such
undercapitalization of the financial system could hamper economic growth through a decrease
in credit supply.

Exhibit 1: The number of climate-related policies across the world It covers climate-
related laws, as well as regulations promoting low carbon transitions. (Source: Climate Change
Laws of the World Data)

Despite the widespread adoption of climate policies and the importance of understanding
their effect on the banking sector, there has been little understanding of the potential impact
of climate change on the financial system due to several challenges, as noted by Bolton et al.
(2020). In fact, while the literature on systemic risk measurement (e.g., Brownlees and Engle,
2017; Acharya et al., 2016; Adrian and Brunnermeier, 2016; Allen et al., 2012) has produced

1
useful indices of systemic distress in the context of financial crises, no such measures exist
to analyze climate-related risks.
In this paper, we focus on a particular dimension of climate risk (transition risk) and seek
to answer the following question: are banks sufficiently capitalized to absorb losses during
stressful conditions due to heightened climate risk? To answer this question, we take a novel
approach to measuring the potential adverse effect of transition risk on banks’ capitalization.
Measuring the climate risk exposure of financial institutions faces several challenges.
First, analyses based on past climate events may not effectively capture the changes in the
perception of risk. For instance, market expectations may change without a direct experience
of climate change events, and asset prices today can reflect changes in future climate risk
even though the damages or impacts are decades away. Second, both the climate risk itself
and how firms, banks, and markets respond to the perceived risk change over time. Third,
the lack of reliable data sources for systematically assessing climate-related risks poses a
significant challenge. While voluntary climate-related disclosures exist, they often suffer
from incompleteness and inconsistencies in quality. 1
We develop a methodology that addresses the aforementioned challenges. We address the
first challenge by constructing climate risk factors by forming portfolios designed to decline
in value as the transition risk rises and then measuring the banks’ stock return sensitivity,
called climate beta, to the climate risk factor. We address the second challenge by estimating
the climate beta dynamically, which allows us to avoid making strong assumptions such as
a static balance sheet and time-invariant responses of firms and investors to change in the
transition risk. Our methodology addresses the third challenge, as it uses only the market
data that are consistent in quality, comparable across firms, and less susceptible to noise and
bias inherent in voluntary disclosures. The importance of these elements was also envisioned
in Bolton et al. (2020) and Brainard (2021) among others.
We create a novel measure CRISK, defined as the expected capital shortfall of a financial
1
See Brainard (2021), Financial Stability Board (2021), and European Systemic Risk Board (2020) among
others.

2
firm under a climate stress scenario. CRISK is a function of a given financial firm’s size,
leverage, and expected equity loss conditional on a climate stress event, which is calculated
using the estimated climate beta. We define a climate stress event as a shock to a given
climate risk factor, an equity portfolio designed to decline in value as climate risk rises. To
consider a sufficiently severe yet plausible stress scenario, we take the lowest one percentile
of the 6-month return distribution of a climate risk factor to calibrate the stress level. 2,3
Additionally, we introduce marginal CRISK, which isolates the effect of climate stress from
concurrent undercapitalization by subtracting CRISK under zero climate stress from CRISK.
We apply our methodology to learn about the climate risk exposure (CRISK) of large
global banks. The estimated CRISK varies depending on the severity of the scenario and the
climate risk factors. We summarize our findings using the stranded asset factor by Litterman
(n.d.) as the climate risk factor, which serves as a proxy for market expectations on future
transition risk, as fossil fuel energy firms’ assets are likely to become “stranded” along most
transition paths.
We find that the climate beta varies over time, highlighting the importance of dynamic
estimation. The climate beta and CRISK substantially increased during 2020, across all
banks in our sample. In 2020, the aggregate CRISK of the top four US banks increased
by 425 billion US dollars (USD), which corresponds to approximately 47% relative to their
market capitalization. Our decomposition analysis reveals that 40% of the CRISK increase
in 2020 was due to an increase in climate betas, and 40% was due to a decrease in equity
values. The aggregate marginal CRISK of the top four US banks reached 260 billion USD
in 2020, indicating a significant potential impact of climate stress.
These results are consistent with the following mechanism. When fossil fuel energy prices
plummeted in 2020, which would happen under a sudden and disorderly transition, “brown”
borrowers’ loans became particularly riskier with a shorter distance to default, and the banks’
2
Basel Committee on Banking Supervision (2018) describes as part of stress testing principles that a
stress testing framework should consider “scenarios that are sufficiently severe but plausible.”
3
Future work of scientific and economic analyses could suggest other approaches to calibrate the stress
level.

3
stock returns became more sensitive to the transition risk, thereby affecting banks’ climate
risk exposure. Indeed, we find evidence supporting this mechanism from the validation
exercise.
We note that we are not identifying that transition risk caused the collapse in fossil fuel
energy prices in 2020; rather, we are merely exploiting an event in which the climate factor
declined severely. Indeed, it would be ideal to estimate CRISK based on a realized transition
stress event. However, as there has not yet been a realized large-scale transition risk event, it
is impossible to exploit an empirical distributional relationship between transition risk and
its effect on bank capitalization. We can only extrapolate from events that have a similar
effect on the climate factor, building upon climate beta, which measures the sensitivity of
bank stock returns on the climate factor. One may be concerned that the climate factor
is capturing the effect of the concurrent COVID outbreak rather than the transition risk.
To address this concern, we validate the climate factors in event study analyses, where we
find that they respond to transition events that materialized (but are not sufficiently severe
for stress testing). Moreover, we find that our results are robust after controlling for the
non-energy related COVID effect.
Our framework is versatile since it can be applied to financial institutions other than
banks and can be aggregated at the economy level. To gauge the system-wide measure of
climate risk, we compute the aggregate CRISK and the aggregate marginal CRISK of 105
financial firms, including banks, broker-dealers, and insurance companies, in the US. The
aggregate CRISK of the US reached almost 500 billion USD in 2020 but declined to under
150 billion USD at the end of 2021, suggesting that climate risk does not pose an immediate
threat to the US financial system as of the end of 2021. However, the aggregate marginal
CRISK reached over 500 billion USD in 2020 and remained as high as 400 billion USD at the
end of 2021, which indicates that the effect of climate stress could potentially be substantial
in the future if banks are not sufficiently capitalized.
Our framework can also admit a wide variety of scenarios. Given that there has been

4
no consensus in terms of what constitutes sufficiently severe yet plausible scenarios in the
context of climate risk, we conduct a sensitivity analysis. For example, moving from a stress
level corresponding to the 1% quantile to less severe scenarios such as 5% quantile, 10%
quantile, and median, the peak marginal CRISK of the top four US banks in 2020 falls from
260 billion USD to 140, 120, and 10 billion USD, respectively. The results discussed so far
are based on the stranded asset factor. We find similar but slightly higher marginal CRISK
under the scenario associated with a stylized version of a carbon tax; however, we find much
lower marginal CRISK under the scenario associated with a stylized version of a carbon tax
combined with a green subsidy.
We validate our analysis using granular data on large US banks’ loan portfolios, taken
from Federal Reserve Y-14 Q (Y-14) forms. From this data set, we construct a panel of loan
portfolio climate beta by taking the loan-size-weighted average climate beta of the borrowers’
sector stock returns. We find that the constructed loan portfolio climate betas are strongly
aligned with the climate betas based only on the market data of bank stock returns and their
conditional covariance with climate risk factors, corroborating the economic validity of our
measures.4 Additionally, we find that banks’ climate betas are higher when lending more
to industries with high emissions (“brown” industries) and when the risk of loans made to
these industries is high relative to that of other industries.
Our results highlight the credit risk and market risk channels through which transition
risk affects banks’ capitalization. Our finding of a sharp rise in the probability of default of
firms in the brown industries (relative to all other industries) in 2020 when fossil fuel energy
prices collapsed, which could happen under a sudden and disorderly transition, suggests that
a shock to borrowers’ transition risk can adversely affect their ability to repay even within
a short horizon (credit risk channel). Borrowers’ credit risk can affect banks beyond the
maturity of loans because (1) banks’ lending relationships are typically persistent (e.g., Beck
4
This finding can also serve as a basis to measure the climate risk exposure of non-listed banks, as long
as data on loan composition are available. This is in the spirit of Engle and Jung (2018), who applied this
approach to non-listed banks in Latin America in the SRISK framework.

5
et al., 2018; Liberti and Sturgess, 2018; Nakashima and Takahashi, 2018) and (2) banks tend
to “specialize” by concentrating their lending disproportionately in one industry (Blickle et
al., 2021), which implies that finding lending opportunities outside the specialized industry
would likely be costly. Even if those loans are small relative to the bank’s entire balance
sheet, their rise in credit risk, within maturity or even beyond, can have an amplified effect on
the banks’ current and expected future profits and therefore the bank’s equity valuation. As
a result, a bank’s stock return sensitivity to climate risk moves in tandem with its borrowers’
exposure to climate risk (market risk channel).
We conduct a battery of exercises to verify the robustness of our estimates of bank
climate betas. First, we find that our results are robust to including additional bank stock
return factors, including interest rates, housing, and COVID. Second, our results remain
similar when we use close alternative climate and market factors. Third, we confirm that
our results are robust to various details of the estimation procedure, such as correcting for
asynchronous trading, using an annual sample instead of a full sample, or using a common
dynamic conditional beta parameter across banks to reduce estimation error. Fourth, the
results from the validation exercise hold when we use the unlevered climate beta of borrowers
in computing the loan portfolio climate beta to account for the firm leverage effects, and the
results also remain robust outside of the COVID period.

Contribution to Literature This paper contributes to the literature studying the effect
of transition risk on banks. Studies have documented that banks respond to transition
risk through the credit risk channel by adjusting loan prices and quantities. Kacperczyk
and Peydro (2021) find that high-emission firms receive less bank credit from banks that
make commitments. Chava (2014) finds that banks charge higher interest rates to firms
with environmental issues. Ivanov et al. (2021) show that banks reduce their transition
risk exposure by shortening maturities and limiting access to permanent financing for high-
emission firms. Delis et al. (2019) document that banks charge higher rates to fossil fuel

6
firms, and Laeven and Popov (2022) show that banks shift lending to high-emission sectors
in countries with laxer policies. While these papers suggest that banks respond to transition
risk, it is not clear to what extent banks could manage their risk of undercapitalization in
face of a sudden transition. This paper thus contributes to this literature by estimating
systemic climate risk, despite the means banks currently employ to mitigate climate risk.
Moreover, we incorporate not only the credit risk channel but also the market risk channel.
The current research on measuring systemic climate risk only offers measures that are
backward-looking, static, and based on deterministic transition scenarios, unlike the more
developed literature on measuring the systemic risk of financial institutions in the context
of financial crises (e.g., Brownlees and Engle, 2017; Allen et al., 2012; Adrian and Brunner-
meier, 2016; Acharya et al., 2016). Reinders et al. (2023) use Merton’s contingent claims
model to assess the impact of a carbon tax shock on the value of corporate debt and residen-
tial mortgages in the Dutch banking sector. Battiston et al. (2017) provide a network-based
approach and Nguyen et al. (2023) employ a bottom-up approach to climate stress tests.
Many regulators also have conducted climate stress tests,5 relying on the book values and
projections of realized losses of loans using confidential supervisory data. These tests typi-
cally assume that the impacts of climate risk on firms’ cash flows (and therefore the impacts
on the banking sector) only appear far in the future (e.g., in 30 years), without incorporating
the possibility that banks’ balance sheets and policies can change within such a long horizon.
In contrast, our approach incorporates market expectations, and thus yields measures that
are forward-looking, time-varying, can be estimated in real time, and requires only publicly
available data.
We modify the SRISK framework of Brownlees and Engle (2017) along three dimensions
to assess the impact of climate-related risks. First, while the SRISK uses the market return as
the only risk factor, we employ a variety of climate risk factors to design stress scenarios. We
5
Based on a survey of 53 institutions from 36 jurisdictions conducted by the Financial Stability Board
and Network for Greening the Financial System (2022), 54 climate stress tests or scenario analyses were
completed or in progress, and 12 exercises were in the planning stage.

7
also validate the climate risk factors by showing that they negatively respond to events that
are associated with a movement toward a greener economy. Second, we introduce several new
market-based metrics of climate risk exposures of financial institutions. On top of CRISK,
we also introduce marginal CRISK, which isolates the effect of climate stress from market
stress. To test for a scenario where market stress and climate stress arrive at the same time,
we introduce a compound risk metric, S&CRISK.6 This measure is useful because, when
market risk and climate risk are correlated, the CRISK alone may underestimate the risk.
Third, we validate our analysis using Y-14 data, due to the lack of realized climate stress
episodes that would allow for a direct assessment of the predictive power of CRISK.

Outline of the Paper The remainder of the paper proceeds as follows: Section 2 describes
the data. Section 3 develops a methodology to estimate the potential adverse effect of
climate transition risk on bank capitalization, and introduces measures of banks’ transition
risk exposure. Section 4 validates the measures. Section 5 presents the application of our
measures and Section 6 shows robustness results. Section 7 concludes.

2 Data

We use various data sets for analyses. We use market data for estimating climate betas and
CRISKs of large global banks in the US, the UK, Canada, Japan, and France for the sample
period from 2000 to 2021. We focus on large global banks, since they hold more than 80%
of syndicated loans made to the oil and gas industry.7 We use carbon emissions data to
construct some of the climate factors and bank-level data on financial variables and loan
portfolio composition to validate our measures.
6
It is a sum of three components: marginal SRISK, marginal CRISK, and the undercapitalization of the
bank under zero climate stress and zero market stress.
7
This is based on the syndicated loan data from LPC DealScan and Bloomberg League Table.

8
Market Data Our market-based approach only requires publicly available data. In the
construction of climate factors, we use the daily return on financial stocks, S&P 500 index,
and other ETFs, including VanEck Vectors Coal ETF (KOL), Energy Sector SPDR ETF
(XLE), and iShares Clean Energy ETF (ICLN) downloaded from Datastream. To form
industry portfolios, we use a CRSP-Compustat merged data set.

Carbon Emissions Data Some climate risk factors are constructed based on past car-
bon emissions, calculated as the sum of Scope 1 and Scope 2 emissions, downloaded from
Bloomberg. The data set includes emissions reported by firms in disclosure as well as emis-
sions reported to the carbon disclosure project. Scope 1 emissions are direct emissions from
sources controlled by or owned by the company. Scope 2 emissions are indirect emissions
associated with the purchase of electricity, steam, heat, or cooling. We use emission levels,
rather than emission intensities, since emission levels are associated with a risk premium
(Bolton and Kacperczyk, 2021, 2022). We additionally use carbon emissions data by S&P
Global Trucost to test for robustness.

Financial Variables and Loan Portfolio Data of US Banks We use data from FR
Y-14Q (Y-14) and FR Y-9C (Y-9C) to validate climate beta measures by examining the
relationship between climate beta estimates and bank loan composition as well as bank
characteristics. Y-14 provides granular data on banks’ loan holdings, and Y-9C provides
consolidated financial statement data of bank holding companies. Data from both forms are
maintained by the Federal Reserve. Y-14 is the closest data to the credit registry in the
US. Unlike commercially available databases that cover only a subset of the loan market,
Y-14 covers more than 75% of all corporate lending in the US. We use its sub-database
“Schedule H.1,” which provides granular information on all commercial and industrial loans
over 1 million USD in size for all stress-tested banks in the US at a quarterly frequency.
In the sample period between 2012:Q2 and 2021:Q4, we observe over 5 million loans for

9
21 listed banks.8 This data set is particularly useful because we can make use of data on
the borrowers’ industry, primarily classified by the North American Industry Classification
System (NAICS), and the probability of default. The probability of default variable is based
on each bank’s internal assessment and reported as part of the stress testing requirements
of the Dodd-Frank Act.9

Others To test the robustness of climate beta measures, we use an index measuring seated
diners downloaded from OpenTable and an index measuring air passengers downloaded from
the Transportation Security Administration (TSA) to proxy for the effect of COVID on the
leisure and hospitality sector.

3 Methodology

The market-based methodology to measure the transition risk exposure of financial insti-
tutions involves three steps. The first step is to build stress test scenarios by constructing
portfolios designed to respond to climate risk. The second step is to estimate the time-
varying climate betas of financial institutions using the Dynamic Conditional Beta (DCB)
model of Engle (2016). The third step is to compute CRISK, which is the expected capital
shortfall conditional on climate stress.

3.1 Climate Transition Risk Factors and Climate Stress Scenarios

Every stress test begins with designing scenarios. To build market-based climate stress
scenarios, we build upon studies on forming climate hedge portfolios (e.g., Engle et al.,
2020; Alekseev et al., 2022; De Nard et al., 2022; Litterman, n.d.). These studies construct
portfolios that are expected to rise in value as climate risk increases. We form climate risk
factors by taking a short position in such climate hedge portfolios or in the factors correlated
8
The bank-quarter panel is unbalanced.
9
This variable has also been used by Correa et al. (2022).

10
with them. While our framework is flexible such that other existing measures can be used,
market-based return factors have distinctive benefits in that they are forward-looking and
time-varying. Compared to unsigned news-based measures that mainly capture attention
to climate news, our measures can differentiate between attention to a tightening transition
policy from attention to a loosening transition policy.
We consider four climate risk factors: a stranded asset factor, an emission factor, a brown
minus green factor, and a climate efficient factor mimicking portfolio factor. Each of these
factors can be associated with stylized versions of climate transition scenarios, and all of these
factors can be easily computed on a daily basis. We further show that all of them negatively
respond to climate transition events that are associated with movements towards a greener
economy, while they respond to different types of climate transition events (section 3.1).

Stranded Asset Factor

The first factor we consider is a stranded asset factor. McGlade and Ekins (2015) find that,
globally, a third of oil reserves, half of the gas reserves, and over 80% of current coal reserves
should remain unused from 2010 to 2050 to meet the target of limiting global warming to
2 degrees Celsius. This implies that fossil fuels would likely become “stranded assets” more
quickly as economies move into a less carbon environment. Indeed, van der Ploeg and Rezai
(2020) find that the assets in the fossil fuel industries are at risk of losing market value due
to transition risk triggered by changes in renewable technology and climate policies in light
of the Paris commitments. In this sense, the return on a stranded asset portfolio is a useful
proxy measure reflecting market expectations on future transition climate risk.
The stranded asset portfolio was developed by Litterman (n.d.) and the World Wildlife
Fund, whose investment committee he chairs, takes a short position in the stranded asset
portfolio to get a climate hedge.10 The stranded asset factor is composed of a 70% long
position in VanEck Vectors Coal ETF (KOL), a 30% long position in Energy Select Sector
10
The stranded asset portfolio return acts as a proxy for the World Wildlife Fund stranded assets total
return swap.

11
SPDR ETF (XLE), and a short position in SPDR S&P 500 ETF Trust (SPY). During the
period in which VanEck Vectors Coal ETF is not available, we use the average return on
the top 4 coal companies instead. We use the performance of firms, not the performance of
commodities, to reflect the firms’ responses to a commodity shock, such as hedging.
Based on the stranded asset factor, we build a scenario. We consider a scenario where
the stranded asset factor declines by 50% over a six-month period. This is a sufficiently
severe yet plausible scenario suitable for a market-based stress test because a 50% decline
in the stranded asset factor corresponds to the left tail (1% quantile) of the past realized
return distribution. We note that this scenario may not materialize in the short run. For
instance, a high carbon tax without alternative energy can lead to an increase in energy
prices. Indeed, not only energy prices but also fossil fuel stock prices rose in 2022 due to
a reduction in supply. While it is unlikely that policymakers would implement a disruptive
policy like a high carbon tax imminently due to a lack of alternative energy in place, it
is likely that regulatory interventions will eventually be implemented to shift into a less
carbon-intensive economy (e.g., to meet the Paris agreement goal). If such implementations
were never to arrive, there would be no transition risk at all to consider, by definition. In
fact, a rapidly growing number of climate-related policies have been adopted globally (as
presented in Exhibit 1). As such measures get tighter and broader, it is plausible that
producers and consumers alike will be incentivized to reduce fossil fuel energy use and shift
to lower carbon fuels or renewable energy sources through investment or consumption. When
a tighter and/or faster than expected measure gets implemented, the value of the stranded
asset portfolio may fall sharply over a short horizon in a sufficiently severe “1% of the time”
stress event.
For the rest of the factors, we use the same approach to build scenarios. We consider
scenarios in which each factor falls substantially, corresponding to a 1% quantile of the return
distribution, over six months.

12
Emission Factor

While the stranded asset factor is intuitive, the portfolio weights are not optimized to best
reflect transition risks. Moreover, a carbon tax can have a broader effect than hurting
fossil fuel firms. To consider a stylized version of a carbon tax, we construct an emission
factor in the following steps. We first compute daily industry returns by calculating the
value-weighted stock returns of US firms in the CRSP-Compustat database.11 Industries are
classified by SIC-4 digit. Then, for each year and industry, we compute the average carbon
emissions (sum of Scope 1 and Scope 2 emissions).12 Lastly, we compute weighted average
industry returns where the weight is the carbon emissions. Because the emissions data from
Bloomberg are available only from 2010, we apply the same emission weights as 2010 for the
pre-2010 period.13

Brown Minus Green Factor

Subsidizing the production and consumption of renewable energy (“green subsidy”) is an-
other regulatory measure that can lead to a rise in transition risk. To consider a stylized
scenario with mixtures of a carbon tax and green subsidy, we construct a brown minus green
(BMG) factor. We use the emission-based factor as the brown factor and the iShares Global
Clean Energy ETF (ICLN) return as the green factor.

Climate Efficient Factor Mimicking Portfolio Factor

To consider climate stress besides stranded assets, we construct a climate-efficient factor

mimicking portfolio (CEP) factor by taking a short position in the CEP formed by De Nard
et al. (2022). The CEP portfolio is a long-only portfolio of publicly available sustainable
funds selected based on two criteria, (1) minimum variance, and (2) maximum correlation
11
We focus on ordinary common shares (share codes 10 and 11) traded on the NYSE, AMEX, and
NASDAQ (exchange codes between 1 and 3).
12
Here, we confine the sample to S&P 500 constituents following Ilhan et al. (2020) to address the time-
varying coverage of emissions data.
13
The results are robust to using emissions data from S&P Trucost.

13
with climate news after controlling for standard financial risks, the price of oil, and the
stranded assets portfolio.

Climate Factor Responses around Climate Change Events

To test whether the constructed climate risk factors capture climate transition risk, we
conduct an event study analysis. We take the list of transition climate risk events from
Barnett (2019), which goes until March 2019, and extend it to the end of 2021. This gives us
107 events, including electoral events, Intergovernmental Panel on Climate Change (IPCC)
meetings, climate-related policy events, and others. The list includes the sign of shock, where
a positive sign is associated with a movement toward a greener economy (“green” event),
such as the Paris agreement, and a negative sign is associated with a movement away from
a greener economy (“brown” event), such as the withdrawal from the Paris agreement. The
climate factor summary statistics (Table B.1), correlation table (Table B.2), and the full list
of events (Table C.1) are included in the appendix.
We use the following specification to test the climate risk factors’ responses to the tran-
sition events:
5
X
CFt = α + γn shockt−n + M KTt + εt
n=0

where CF denotes climate risk factor, either stranded asset, emission, BMG, or CEP factor.
shockt takes a value of 1 if there was a green event, a value of -1 if there was a brown event,
and a value of 0 if there was no event on the day t. We use the SPDR S&P 500 ETF for the
market return, M KT . The expected sign of γ is negative because a rise in transition risk is
associated with a positive shock and a lower value of CF . The standard errors are Newey-
West adjusted for serial correlation. Figure 1 plots the cumulative γ coefficient and it shows
that all proposed climate risk factors respond negatively to greener events, as expected.
The γ coefficients are statistically significant for the emission and the BMG factors, and
marginally significant for the stranded asset factor. The CEP factor’s insignificant response
may be due to an asymmetric response to green events versus brown events. If the market

14
tends to respond more to brown events than to green events, the CEP factor is not likely
to respond significantly to transition events because the CEP factor is designed to capture
green news, after taking out the stranded asset factor.
To address a potential concern that geopolitical risk is a confounding factor, we include
the global common volatility, COVOL, of Engle and Campos-Martins (2023) as a control
variable. We find that the γ coefficients remain close (Figure C.1). Furthermore, for robust-
ness, we take a two-step approach closer to the standard event study analysis. Specifically,
we construct non-overlapping data around the event dates and first obtain the abnormal
ˆ t , from a market model CFt = α + bM KT M KTt + εt
return on climate factor, art = CFt − CF
on a 1-year rolling window basis. Then we regress cumulative abnormal return on shock:
cart−1,t+n = α + γ shockt + εt . Based on this alternative specification, we find consistent
results (Appendix C).14

3.2 Climate Beta Estimation

Following the standard factor model approach, we model bank i’s stock return as:

rit = βitM kt M KTt + βitClimate CFt + εit (1)

where rit is the stock return of bank i, M KT denotes market return, and CF denotes climate
risk factor. We include the market factor in the model to control for confounding factors,
such as the COVID shock and aggregate demand shock, that influence both the bank stock
returns and the climate risk factor. The market beta and climate beta, in this regression,
measure the sensitivity of bank i’s return to overall market risk and to the climate risk factor,
respectively.
The expected sign of the climate beta is positive for banks that hold loans and/or financial
14
With this approach, the number of observations drops even for 1-day abnormal returns, because (1) we
estimate the market model based on the rolling-window regression and (2) we include only one observation
per 5-day window after the shock, following the standard event study approach.

15
assets that are exposed to transition risk because the banks’ loan portfolios would likely
deteriorate as transition risk rises (climate risk factor falls). The rise in credit risk, either
due to the borrower’s outright inability to repay or deterioration in the borrower’s ability to
repay, would negatively affect the banks’ current and expected future profits and therefore
the banks’ stock returns.
We use the DCB model to estimate the time-varying climate betas on a daily basis.
The GARCH-DCC model of Engle (2002, 2009, 2016) allows volatility and correlation to
vary over time. The details of estimation steps and the parameter estimates are reported
in Appendix D. For stock markets with a closing time different from that of the New York
market, we take asynchronous trading into consideration.15

3.3 CRISK, Marginal CRISK, and S&CRISK Estimation

Following the SRISK methodology in Acharya et al. (2011), Acharya et al. (2012), and
Brownlees and Engle (2017), we define CRISK as the expected capital shortfall conditional
on a systemic climate change event

CF
CRISKit = Et [CSi,t+h |Rt+1,t+h < C]
15
Consider the following specification including the lags of the independent variables:
M kt M kt Climate Climate
rit = β1it M KTt + β2it M KTt−1 + β1it CFt + β2it CFt−1 + εit

Assuming that returns are serially independent, we estimate the following two specifications separately and
sum the coefficients.
M kt Climate
rit = β1it M KTt + β1it CFt + εit
M kt Climate
rit = β2it M KTt−1 + β2it CFt−1 + εit
M kt M kt Climate Climate
The sum, β1it + β2it , is the estimate of market beta and the sum, β1it + β2it , is the estimate of
climate beta.

16
where CSit is the capital shortfall of bank i on day t. We define the capital shortfall as the
capital reserves the bank needs to hold minus the firm’s equity:

CSit = k(Dit + Wit ) − Wit

where Wit is the market value of equity and Dit is the book value of debt, and k is the
prudential ratio of equity to assets. The sum of Dit and Wit can be considered as the value
CF
of quasi assets. {Rt+1,t+h < C} is associated with a climate stress scenario. Assuming that
CF
banks’ liabilities are immune to the stress, E[Di,t+h |Rt+1,t+h < C] = Dit , CRISK for each
financial institution can be expressed as the following.16

CRISK it = k · Dit − (1 − k) · Wit · (1 − LRM ESit ) (2)

where LRM ES is the long run marginal expected shortfall, the expected firm equity multi-
period arithmetic return conditional on a systemic climate change event:

i CF
LRM ESit = −Et [Rt,t+h |Rt+1,t+h < C] (3)

Based on equations (1)– (3), CRISK can be written as17 :

CRISKit = k · Dit − (1 − k) · Wit · exp βitClimate log(1 − θ)

(4)

CRISK is higher for banks that are larger, more leveraged, and with higher climate beta.
We set the prudential capital fraction k to 8% (5.5% for European banks to account for
accounting differences) and the climate stress level θ to 50%, as discussed in subsection 3.1.
16
This is not a strong assumption given that the liabilities of banks are largely deposits, which are
relatively immune to the stress.
17
See Appendix E for the derivation.

17
Marginal CRISK We propose a measure, marginal CRISK, to capture the effect of cli-
mate stress, in isolation from the realized undercapitalization as well as the effect of market
stress. The marginal CRISK, mCRISK, is defined as the difference between CRISK and
non-stressed CRISK, where the non-stressed CRISK is simply the capital shortfall of a bank
without any climate stress (θ = 0). From equation (2),

mCRISK = (1 − k) · W · LRM ES (5)

Put differently, CRISK is the sum of the bank’s undercapitalization and the bank’s marginal
CRISK.

Systemic Climate Risk We introduce two measures to understand a system-wide climate

risk. First, we use the CRISK measure across all firms to construct a system-wide measure
of climate risk. The total amount of systemic climate risk in the financial system is measured
as
N
X
CRISKt = (CRISKit )+
i=1

where (x)+ denotes max(x, 0). We ignore the contribution of negative CRISK in computing
the aggregate CRISK because it is unlikely that the capital surplus can easily be transferred
from one institution to another, especially during the distress period. The aggregate CRISK
of an economy can be interpreted as the amount of capital injection needed for the financial
system in climate stress.
Second, we use the marginal CRISK measure across all firms to construct a system-wide
measure of exposure to climate risk, in isolation from the concurrent capitalization:

N
X
mCRISKt = mCRISKit
i=1

In order to construct a system-wide exposure measure, we do not truncate each institution’s

mCRISK at zero.

18
S&CRISK We also offer a framework to compute a compound risk, S&CRISK, based
on a value of market stress, θM kt and that of climate stress, θClimate .18 Equation 4 can be
extended to compute compound S&CRISK:

S&CRISKit = k · Dit − (1 − k) · Wit · exp βitClimate log(1 − θClimate ) + βitM kt log(1 − θM kt )

This measure is useful because when the market risk and climate risk are correlated, the
CRISK alone can underestimate the risk.

4 Validation

We validate the climate beta measure using granular data on loan holdings of large US
banks from Y-14. We link market-based climate beta estimates to banks’ loan portfolio
composition and bank characteristics. The sample includes 21 listed banks in Y-14 for the
sample period from 2012:Q2 to 2021:Q4. The bank-level variables’ summary statistics and
correlation tables are reported in Table B.4. We test two main hypotheses: (1) climate beta
reflects banks’ loan exposure to climate risk, and (2) banks with higher brown loan exposure
have higher climate beta, and climate beta is high when the risk of brown loans is high.

4.1 Climate Beta and Loan Portfolio Climate Beta

First, we test whether the market valuation of banks’ exposure to climate risk factors, proxied
by climate beta, reflects banks’ loan portfolio composition. To test this, we construct a panel
of loan portfolio climate beta by computing the weighted average climate beta for each bank
where the weight is the loan size and each loan is assigned the climate beta of the respective
18
We note that this metric does not model the tail dependence. While it is certainly possible that a large
climate stress would be more damaging in a recession than in a period of strong growth, calibrating the tail
dependence requires an equilibrium model, given that there has been no such event realized in the past.

19
industry:
X
Loan Portfolio Climate Beta = wj βjClimate
j∈J

where the weight, wj is the proportion of C&I loans made to the respective industry j.
βjClimate denotes the climate beta of industry j, and it is computed as the value-weighted
average climate beta of firms in each 3-digit NAICS industry.19 The industry climate betas
are computed based on all listed firms in the US. While they are based on the listed firms,
we incorporate all firms including non-listed firms in the Y-14 by applying the same industry
climate beta for non-listed firms in the respective industry. This is a benefit of focusing on
the industry level rather than the firm-level composition of banks’ loan portfolios.
Consistent with the hypothesis, Figure 2 shows that the market-based climate beta and
the loan portfolio climate beta are strongly aligned, after controlling for the time-fixed ef-
fect and the bank-fixed effect. We formally test this hypothesis with the following OLS
specification:

βitClimate = α + b · Loan P ortf olio Climate Betait + BankControlsit + δi + γt + εit (6)

The dependent variable, βitClimate , is bank i’s time-averaged daily climate beta during
the quarter-end month. Bank control variables include: log assets, leverage, return on
assets (ROA), loans/assets, deposits/assets, book/market, loan loss reserves/loans, non-
interest income/net income, and market beta. Table 1 shows the result. Columns (2)–(4)
include bank control variables, and Columns (3) and (4) add bank fixed effects to control
for unobservable time-invariant bank characteristics. Column (4) adds year fixed effects to
control for any potential trends. Standard errors are clustered at the bank level. Consistent
with the hypothesis, we find that b is positive and significant across specifications.
This relationship remains strong when the loan portfolio climate beta is computed based
19
For some banks and periods, the borrowers’ industries are classified primarily based on the SIC code
instead of NAICS code. For these cases, we compute the value-weighted average climate beta of firms in each
3-digit SIC industry. We drop observations (bank-quarter level) if industry classification by SIC or NAICS
is not available.

20
on firms’ unlevered climate beta, to account for the leverage effect of the firms (Table 2).
Moreover, this result is not driven by observations during the COVID period. When we
confine the sample to the pre-COVID period, before 2020, the significance and the magnitude
of the coefficients remain similar (Table 3).

4.2 Climate Beta, Brown Loan Exposure, and Brown Loan Prob-

ability of Default

The previous analysis characterizes loans based on their industry climate beta. Another
way to characterize loans in the context of transition climate risk is by using the borrowers’
carbon emissions data. In this subsection, we test whether banks lending more to industries
with high carbon emissions have higher climate betas. We define a brown loan as a C&I
loan made to a firm in the top 30 SIC 4-digit industries by the sum of Scope 1 and Scope
2 emissions.20 The emissions of the top 30 industries cover about 88% of the total reported
emissions.
We further hypothesize that climate betas are higher during the time period when the
risk of brown loans is high. Figure 3 shows that during the first two quarters of 2020, the
size-weighted average probability of default increased for firms in brown industries as well as
non-brown industries; however, that for the firms in brown industries increased much more
sharply.
We test formally whether a higher exposure to brown loans and a higher risk of brown
loans are associated with a higher climate beta, using the following OLS specification:

βitClimate = a + bBrownLoanShare · Brown Loan Shareit

+ bBrownLoanP D · Brown P D Spreadt + Bank Controlsit + δi + γt + εit (7)

The dependent variable, βitClimate , is bank i’s time-averaged daily climate beta during the
20
We use the industry rankings by emissions from Ilhan et al. (2020), and extend it to 2020.

21
quarter-end month t. Brown Loan Shareit is defined as bank i’s brown loans scaled by
total loans, for quarter t. Brown P D Spreadt is defined as the spread between the size-
weighted average probability of default of firms in the 30 brown industries and that of firms
in all other industries, and it captures the time-series variation in the risk of brown loans
relative to non-brown loans.21 We use the same bank control variables as the previous
analysis. The sample period for this analysis is from 2014:Q4 to 2021:Q4, since the data on
the probability of default are mostly available from 2014:Q4. Standard errors are clustered
at the bank level.
Table 4 presents the results. Consistent with the hypothesis, the coefficient on the
Brown P D Spreadt is positive and significant across specifications.22 In addition, the
coefficients on Brown Loan Shareit are positive and significant. These results suggest that
both exposure and risk of brown loans explain variations in climate beta. In untabulated
results, we find that the results are robust to using the emission intensity rankings, where
emission intensity is emission divided by the market capitalization of the firm.
These results shed light on channels– the credit risk channel and the market risk channel–
through which transition risk affects banks’ capitalization. A shock to borrowers’ transition
risk can adversely affect their ability to repay even within a short horizon (credit risk chan-
nel), as evidenced by the sharp rise in the probability of default of the brown borrowers
(relative to all other industries) in 2020 when fossil fuel energy prices collapsed, which would
happen under a sudden and disorderly transition. Borrowers’ credit risk can affect banks
beyond the maturity of loans because (1) banks’ lending relationships are typically persis-
tent (e.g., Beck et al., 2018; Liberti and Sturgess, 2018; Nakashima and Takahashi, 2018)
and (2) banks tend to “specialize” by concentrating their lending disproportionately in one
industry (Blickle et al., 2021), which implies that finding lending opportunities outside the
specialized industry would likely be costly. Even if those loans are small relative to the
21
The probability of default is weighted by the log asset of the obligor. The results are robust when they
are equally weighted.
22
We omit the coefficient on Brown P D Spreadt in specification (4) as we include year-fixed effects.

22
bank’s entire balance sheet, their rise in credit risk, within maturity or even beyond, can
have an amplified effect on the bank’s current and expected future profits and therefore the
bank’s equity valuation. As a result, a bank’s stock return sensitivity to climate risk moves
in tandem with its borrowers’ exposure to climate risk (market risk channel), as evidenced
by the strong alignment of climate beta and loan portfolio climate beta.
The results also imply that the empirical model of equation (6) provides a potential
framework to estimate the climate beta of non-listed banks. While it is not possible to
estimate the market-based climate beta of non-listed banks, they can be approximated by
using balance-sheet information along with granular information on loan composition.

5 Applications

In this section, we apply the methodology to large global banks in the US, the UK, Canada,
Japan, and France for the sample period from 2000 to 2021. For the main application, we
focus on large global banks, since they hold more than 80% of syndicated loans made to the oil
and gas industry.23 In the systemic climate risk analysis, we analyze the metrics aggregated
across large financial firms, including banks, broker-dealers, and insurance companies.24 We
first show the results based on the stranded asset factor and then present the results based
on other factors in subsection 5.6.

5.1 Climate Beta

Figure 4 presents the 6-month moving average climate betas of the 10 largest US banks in
the scenario using the stranded asset factor. They show that climate betas vary over time,
suggesting that it is important to estimate the betas dynamically. Climate betas of banks
started from zero in early 2000, fell slightly below zero during the beginning of the global
23
This is based on the syndicated loan data from LPC DealScan and Bloomberg League Table.
24
The real-time measures for all major financial firms across the world are published on the V-Lab website
(https://vlab.stern.nyu.edu/climate) on a regular basis.

23
financial crisis, and spiked in 2020. We find that this pattern was common for banks in other
countries as well (Appendix F). The climate betas during 2020 are statistically significant,
based on the full sample OLS regression results (Appendix IA.A) and the rolling-window-
based OLS regression results (Appendix IA.B). In the validation exercise in section 4, we
show that a high climate beta is associated with a bank asset portfolio’s high exposure to
industries with high climate betas or industries with high carbon emissions, as well as those
industries’ probability of default. While those results are based on US banks, it is likely
that the climate beta of other countries also increased in 2020 because the loans they made
to brown industries became riskier as the demand for fossil fuel energy fell following the
common COVID shock. On the other hand, the proximity of climate betas to zero could be
related to the non-linearity in the climate beta as a function of the return on the stranded
asset factor. That is, we expect that the values of bank stocks are relatively insensitive to
fluctuations in the stock prices of oil and gas firms as long as those firms are sufficiently far
from default.

5.2 CRISK

Figure 5 presents the estimated CRISKs of the top 10 largest US banks in the scenario using
the stranded asset factor. Since CRISK is the expected capital shortfall, a negative CRISK
indicates that the bank holds a capital surplus. The reason why the estimated CRISKs
are often negative until 2019 is likely related to the non-linear relationship between climate
beta and the stranded asset factor. A bank will not have a capital shortfall if its climate
beta is small and will therefore have a negative CRISK. In contrast, the CRISKs increased
substantially across countries in 2020 (Appendix G).
Since CRISK is a function of climate beta, as well as a function of the size and leverage of
a bank, the ranking of CRISKs can differ from that of climate beta estimates. For instance,
in December 2020, climate betas of the top 10 US banks declined to below 0.5; however,
CRISKs of some banks (e.g., the bank anonymized as “C”) were substantial, as high as 100

24
billion USD. To put this magnitude into context, the SRISK of bank “C” was 110 billion USD
in December 2020. This suggests that the bank’s expected capital shortfall in the climate
stress scenario is close to the magnitude of the expected capital shortfall in a potential future
financial crisis.25
We see high CRISKs during the global financial crisis and the European financial crisis
because when banks were undercapitalized, they are vulnerable to both overall market risk
and climate risk. To isolate the effect of climate stress from the effect of market stress, we
analyze marginal CRISK in subsection 5.4.

5.3 CRISK Decomposition

To better understand what drives the substantial increase in CRISK in 2020, we decompose
CRISK into three components based on equation (2):

dCRISK = k ∆D} −(1 − k)(1 − LRM ES) · ∆W + (1 − k) · W · ∆LRM ES

| ·{z | {z } | {z }
(8)
dDEBT dEQU IT Y dRISK

The first component, dDEBT = k · ∆D, is the contribution of the firm’s debt to CRISK.
CRISK increases as the firm takes on more debt. The second component, dEQU IT Y =
−(1 − k)(1 − LRM ES) · ∆W , is the effect of the firm’s equity on CRISK.26 CRISK increases
as the firm’s market capitalization deteriorates. The third component, dRISK = (1 − k) ·
W · ∆LRM ES, is the contribution of an increase in climate beta to CRISK.27
Table 5 decomposes the change in CRISK of the top 10 US banks during the year 2020
into three components. For the top 4 banks, the equity deterioration and the risk (due to
climate beta) each contributed about 40% to the increase in CRISK during 2020. On average
across the banks, equity deterioration contributed 48% and the risk contributed 30% to the
25
Brownlees and Engle (2017) show that precrisis SRISK predicts the capital injections carried out by
the Federal Reserve Banks during the crisis.
26
Here, LRM ES represents the average value of LRM ESt and LRM ESt+1 . In the LRM ES calculation,
we use the monthly average climate beta to reduce the volatility of climate beta.
27
Here, W represents the average value of Wt and Wt+1 .

25
change in CRISK during 2020. We find similar results for the UK banks (Table H.1). For
banks in Canada, France, and Japan, where the increase in CRISK was relatively small, we
find that the debt deterioration was the primary component and the risk due to climate beta
contributed to about a third of the increase in CRISK during 2020 (Appendix H).

5.4 Marginal CRISK

Figure 6 plots the marginal CRISKs (mCRISK) of the top 10 US banks, in the scenario
using the stranded asset factor. It shows that the mCRISKs opened up before 2020, and
reached 45–90 billion USD for the top four US banks at the end of 2020. The top four
banks’ aggregate mCRISK is approximately 260 billion USD. These correspond to roughly
28% of their equity. This suggests that the effect of climate stress in 2020 would have been
economically substantial. In contrast, during the global financial crisis or the European
financial crisis, the mCRISKs were close to zero, differentiating the latest peak in CRISK
from the earlier two peaks in Figure 5. Interestingly, the mCRISKs remain high even after
fossil fuel energy prices rebound to their pre-2020 level in late 2021. In other countries, we
find that the mCRISKs of some banks increased during 2020, although they are much lower
than those of the US banks mainly because they are smaller than the US banks (Appendix I).

5.5 Systemic Climate Risk

We aggregate CRISK and aggregate marginal CRISK across large financial firms, including
banks, broker-dealers, and insurance companies. To focus on large financial firms, we analyze
all financial firms with a higher than 25th percentile market capitalization in each country
as of the end of 2019. This sample includes 105 firms in the US, 34 firms in the UK, 50 firms
in Japan, 24 firms in France, and 18 firms in Canada. The full list of tickers and company
names for each country is reported in Appendix J.
Figure 7 plots the aggregate CRISK, stacked by country. The aggregate CRISK of
the sample firms reached almost 2 trillion USD in November 2020. This amount can be

26
interpreted as the total amount of capital injection needed in climate stress, after taking the
concurrent capitalization of financial institutions into account. Therefore, a high aggregate
CRISK can be due to a high aggregate marginal CRISK, a concurrent undercapitalization
of financial firms, or both.
Figure 8 reports the aggregate marginal CRISK by country. This measure takes out the
effect of concurrent capitalization, and therefore, we interpret this measure as a system-wide
exposure to climate risk. The aggregate marginal CRISK in the US was substantial in 2020,
reaching over 500 billion USD, while it was not as high in other countries. It is useful to
monitor both the aggregate CRISK and the aggregate marginal CRISK. For instance, the
low aggregate marginal CRISK of Japan suggests that its high aggregate CRISK in the
recent period was due to undercapitalization.
Figure 9 plots the US financial firms’ marginal CRISK aggregated by industry group.
The total aggregate CRISK of the US financial system was substantial during the global
financial crisis, the European financial crisis, and 2020–2021. However, at the end of 2021,
the total aggregate CRISK of the US was lower than 150 billion USD, which suggests that
climate risk does not seem to pose a substantial threat to the US financial system. During
times of stress, CRISK was concentrated in the banking sector. We compute the Herfindahl
index associated with the CRISK shares to measure the degree of systemic climate risk
concentration in the system. The CRISK share is defined as

CRISKit
CRISK% = if CRISKit > 0
CRISKt

We construct the index for each month, and we find that the index mostly stayed above 0.1
from January 2009 to December 2021 when the aggregate CRISK was non-negligible. This
suggests that CRISK is concentrated among a relatively small number of financial firms.
Figure 10 plots the aggregate marginal CRISK across the financial industry group. The
peak of the marginal CRISK of banks was over 400 billion USD, while that of broker-dealers

27
and insurance companies was about 80 billion USD each. Based on this measure, we find
that the climate risk exposure of all financial industry groups increased during 2019-2020.

5.6 More Scenarios

The results discussed so far have been based on the scenario that the stranded asset fac-
tor falls by 50% over 6 months. Our framework, however, can go considerably further by
employing a wide variety of scenarios.

Severity of Scenario Given that there has been no consensus in terms of what consti-
tutes sufficiently severe yet plausible scenarios in the context of climate risk, we conduct a
sensitivity analysis. Figure 11 plots the aggregate marginal CRISK of the top 4 US banks
with respect to the severity of the scenario. Moving from the stress level corresponding to
the 1% quantile to less severe levels corresponding to a 5% quantile and a 10% quantile,
the peak marginal CRISK of the top four US banks in 2020 falls from 260 billion USD to
140 and 120 billion USD, respectively. If we do not use a tail scenario, where a stress level
corresponds to the median of the stranded asset factor, the peak marginal CRISK of the top
four US banks in 2020 is only about 10 billion USD.

Various Transition Scenarios The same set of measures can be computed based on other
factors constructed in subsection 3.1, motivated by various stylized versions of transition
scenarios. We highlight the key findings here and report the full results in Appendix K.
Based on the emission factor, which can be associated with a carbon tax, we find that the
marginal CRISKs are slightly higher than using the baseline stranded asset portfolio return.
The aggregate marginal CRISK of the top four US banks was about 270 billion USD at the
end of 2020. This is likely because the emission-based factor incorporates non-coal firms
with high emissions. Based on the BMG factor, which is motivated by a mixture of a carbon
tax and a green subsidy, the marginal CRISKs are lower; the top four US banks’ marginal
CRISKs ranged between 10 and 30 billion USD in 2020, suggesting that a green subsidy can

28
partially offset the potential negative effect of a carbon tax on bank stock returns. The CEP
factor is designed to assess the effect of climate stress besides the stranded asset factor. The
marginal CRISKs based on the CEP factor are lower by 30 billion USD, which suggests that
the effect of climate stress besides the stranded asset factor is relatively low. The climate
beta and marginal CRISK plots for the three scenarios are reported in Appendix K.

Compound Risk Scenarios We also apply the compound risk framework. We consider a
scenario where the market stress and the climate stress are severe at the same time. Specifi-
cally, we calibrate the market stress level (θM kt ) to 40% and the climate stress level (θClimate )
to 50%. Each level corresponds to the 1% quantile of the 6-month return distribution of the
market factor and that of the climate factor, respectively. This is the scenario that was
realized during the global financial crisis and, therefore, can be considered a sufficiently se-
vere yet plausible scenario. Figure 12 and Figure 13 show the S&CRISK and the marginal
S&CRISK of the top ten US banks. The aggregate marginal S&CRISK of the top four US
banks reached approximately 590 billion USD at the end of 2021.

6 Robustness Tests

We conduct several tests to ensure that our results are robust to including additional bank
stock return factors, using close alternative climate factors, and taking alternative estimation
procedures.
One may be concerned about missing important factors that explain bank stock returns.
Since banks manage a portfolio of interest-rate-related products, we test whether our results
are robust to including interest-rate factors. Following Gandhi and Lustig (2015), we consider
a long-term government bond factor (LTG) and a credit factor (CRD). We use excess return
on the long-term US government bond index for the long-term interest rate factor and
excess return on the investment-grade corporate bond index for the credit factor. To test
how these factors affect the climate beta estimates, we first regress each bank stock return

29
rit on LT Gt and CRDt , and then regress the residual on M KTt and CFt . In Figure L.7,
we plot the coefficient on CFt , and it shows that the climate beta estimates based on the
baseline specification (1) are robust to including the interest-rate factors. We find that the
results are also robust to including the housing factor measured by the return on a bond fund
specializing in government mortgage-backed securities (Appendix L and Appendix IA.C).
One may be concerned about the COVID related factor being a confounding factor.
For instance, the restaurant, travel, and entertainment industries were hit hard during the
COVID pandemic, but they may not be the industries most affected by climate change. To
address this concern, we construct the COVID industry factor by taking the value-weighted
return on stocks that belong to the NAICS 3-digit industries most affected by COVID,
selected by Fahlenbrach et al. (2021). We exclude five industries that are in the top 20 by
emissions in 2020 because carbon-intensive sectors are likely to be most affected by climate
change.28 We first regress bank stock return on a COVID industry factor. Then, we regress
the residual from the first step on M KT and CF and plot the coefficient on CF using a
1-year rolling window regression. We find that our results remain similar after including
the COVID industry factor (Figure L.10). For a limited sample period, we use an index
measuring seated diners from OpenTable, and an index measuring air passengers from the
TSA as non-transition-related COVID proxy variables and we find that our results are robust
(Appendix IA.C).
We do not include the HML factor of Fama and French (1993), because it is not clear
that the HML is exogenous in the context of our model. Pástor et al. (2022) find that value
stocks tend to be brown and growth stocks green and their two-factor model with a market
factor and a green factor explains much of the recent underperformance of value stocks. In
addition, we find that the HML factor is significant only in the post-GFC period, and this is
likely due to changes in the regulatory framework following the GFC. This also suggests that
the correlation between bank stock returns and the HML factor is potentially an endogenous
28
The excluded SIC industry codes are 211, 486, 483, 481, and 324.

30
outcome of the GFC. Instead, we include banks’ book-to-market ratio as an independent
variable to explain variation in climate beta. Table 4 displays the results of the analysis.
We find that the book-to-market ratio is significant in explaining climate beta in columns
(1)–(3); however, it becomes insignificant when we control for year fixed effects in column
(4).
We test for robustness to using close alternative climate risk factors. One could be worried
that normalizing the stranded asset portfolio by market return could confound our results.
However, we find that using a non-hedged stranded asset portfolio, 0.3XLE + 0.7KOL,
instead of 0.3XLE + 0.7KOL − SP Y leads to consistent results. Moreover, using the MSCI
All Country World Index (ACWI) instead of SPY yields similar results, since they are highly
correlated.29
We corroborate that the results are not driven by a certain detail of our estimation
procedure. First, we find that the procedure to adjust for the time zone difference makes
a small difference. When the asynchronous trading is not corrected, the betas are slightly
smaller in absolute value. Second, we tested whether our results are sensitive to a choice
of the sample window. When betas are dynamically estimated based on an annual sample
(by calendar year) instead of the full sample, the results remain consistent. Based on the
annual sample, some extreme returns are picked up by time variation in the intercept; for
instance, betas are slightly less negative during the early global financial crisis. Third, one
might be worried that the dynamic parameters that govern the speed of adjustment of the
correlations through the dynamic conditional correlation estimation may be too noisy and
introduce errors for some banks. To test this, we took a two-step approach, where each
bank’s DCB parameter is estimated in the first step and the median DCB parameter is used
to estimate the betas in the second step. We find that this makes almost no difference. We
further confirm that our DCB estimation results are consistent with the rolling-window OLS
estimation results.
29
Using a common market factor across countries, for instance, ACWI, facilitates cross-country compar-
isons; however, a country-specific market factor may not be fully incorporated.

31
7 Conclusion

We develop a market-based methodology to assess the resilience of financial institutions

to climate-related risks. The procedure involves three steps. The first step is to measure
the climate risk factor. The second step is to estimate the time-varying climate betas of
financial institutions. The third step is to compute the CRISKs, the capital shortfall of
financial institutions in a climate stress scenario.
We empirically validate the climate risk factors in event study analyses, by documenting
that they negatively respond to transition events associated with movement toward a less
carbon-intensive economy. We validate the climate beta measure by comparing it with
banks’ loan portfolio composition, using Y-14 data. We find that climate beta reflects the
loan portfolio holdings of banks, and a higher climate beta is associated with higher loan
exposure to brown industries and higher risk of brown loans, corroborating the economic
validity of our measure.
We use the methodology to study the climate risks of large global banks in the US, UK,
Canada, Japan, and France. Based on a sufficiently severe yet plausible scenario in which
stranded assets sharply fall in value over a short horizon, we document a substantial rise in
climate betas and CRISKs across banks during 2020. Combined with the results from the
validation exercise, our findings are consistent with the following mechanism. When fossil
fuel energy prices collapsed to zero, which would happen under a sudden and disorderly
transition, “brown” borrowers’ loans became riskier relative to other loans, and banks’ stock
returns became more sensitive to the transition risk, thereby affecting banks’ climate risk
exposure.
There are multiple directions for future research. Our analysis is based on the climate risk
factors that are closer in spirit to transition climate risk. While physical risk may already
be embedded in the climate risk factors we measure in this paper, it might be interesting to
isolate the contribution of physical risk from that of transition risk by constructing a common
physical risk factor directly tied to the damages following extreme weather events. However,

32
this is beyond the scope of this paper, as it would involve identifying market expectations
on a systemic component of physical risk. Another interesting question concerns modeling
the interaction between market stress and climate stress, and we leave it for future research.

33
References
Acharya, Viral V., Christian T. Brownlees, Farhang Farazmand, and Matthew
Richardson, “Measuring Systemic Risk,” Regulating Wall Street: The Dodd-Frank Act
and the New Architecture of Global Finance, chapter 4, 2011.

, Lasse H. Pedersen, Thomas Philippon, and Matthew Richardson, “Measuring

Systemic Risk,” The Review of Financial Studies, 10 2016, 30 (1), 2–47.

, Robert F. Engle, and Matthew Richardson, “Capital Shortfall: A New Approach

to Ranking and Regulating Systemic Risks,” American Economic Review: Papers and
Proceedings, 2012.

Adrian, Tobias and Markus K. Brunnermeier, “CoVaR,” American Economic Review,

July 2016, 106 (7), 1705–41.

Alekseev, Georgij, Stefano Giglio, Quinn Maingi, Julia Selgrad, and Johannes
Stroebel, “A quantity-based approach to constructing climate risk hedge portfolios,”
Working Paper, 2022.

Allen, Linda, Turan G. Bali, and Yi Tang, “Does Systemic Risk in the Financial Sector
Predict Future Economic Downturns?,” The Review of Financial Studies, 2012, 25 (10),
3000–3036.

Barnett, Michael, “A RunonOil? The Implications of Climate Policy and Stranded Assets
Risk,” Working Paper, 2019.

Basel Committee on Banking Supervision, “Stress testing principles,” Bank for Inter-
national Settlements, 2018.

Battiston, Stefano, Antoine Mandel, Irene Monasterolo, Franziska Schütze, and

Gabriele Visentin, “A Climate Stress-Test of the Financial System,” Nature Climate
Change, 2017.

Beck, Thorsten, Hans Degryse, Ralph De Haas, and Neeltje Van Horen, “When
arm’s length is too far: Relationship banking over the credit cycle,” Journal of Financial
Economics, 2018, 127 (1), 174–196.

Blickle, Kristian S., Cecilia Parlatore, and Anthony Saunders, “Specialization in

Banking,” New York Fed Staff Report, 2021, 53 (967).

Bolton, Patrick and Marcin T. Kacperczyk, “Do Investors Care about Carbon Risk?,”
Journal of Financial Economics, 2021.

and , “Global pricing of carbon-transition risk,” Journal of Finance, 2022.

, Morgan Despres, Luiz Awazu Pereira da Silva, Frédéric Samama, and Romain
Svartzman, The green swan, Bank for International Settlements, 2020.

34
Brainard, Lael, “Building Climate Scenario Analysis on the Foundations of Economic
Research,” 10 2021. Speech by Governor Lael Brainard at the 2021 Federal Reserve Stress
Testing Research Conference, Federal Reserve Bank of Boston, Boston, Massachusetts (via
webcast).

Brownlees, Christian T. and Robert F. Engle, “SRISK: A Conditional Capital Short-

fall Index for Systemic Risk Measurement,” Review of Financial Studies, 2017.

Chava, Sudheer, “Environmental Externalities and Cost of Capital,” Management Science,

2014, 60 (9), 2223–2247.

Correa, Ricardo, Ai He, Christoph Herpfer, and Ugur Lel, “The Rising Tide Lifts
Some Interest Rates: Climate Change, Natural Disasters and Loan Pricing,” Working
Paper, 2022.

De Nard, Gianluca, Robert Engle, and Bryan Kelly, “Hedging Climate Change Risk:
An Efficient Factor Mimicking Portfolio Approach,” Working Paper, 2022.

Delis, Manthos D, Kathrin De Greiff, and Steven Ongena, “Being stranded with
fossil fuel reserves? Climate policy risk and the pricing of bank loans,” Climate Policy
Risk and the Pricing of Bank loans (September 10, 2019). EBRD Working Paper, 2019,
(231).

Engle, Robert F., “Dynamic conditional correlation: A simple class of multivariate gen-
eralized autoregressive conditional heteroscedasticity models,” Journal of Business and
Economic Statistics, 2002.

, “Anticipating correlations: A new paradigm for risk management,” Princeton, NJ:

Princeton University Press, 2009.

, “Dynamic Conditional Beta,” Journal of Financial Econometrics, 2016.

and Hyeyoon Jung, “Estimating SRISK for Latin America,” Working Paper, 2018.

and Susana Campos-Martins, “What are the events that shake our world? Measuring
and hedging global COVOL,” Journal of Financial Economics, 2023, 147 (1), 221–242.

Engle, Robert F, Stefano Giglio, Bryan Kelly, Heebum Lee, and Johannes
Stroebel, “Hedging Climate Change News,” The Review of Financial Studies, 02 2020,
33 (3), 1184–1216.

European Systemic Risk Board, “Positively green: Measuring climate change risks to
financial stability,” Report, 2020.

Fahlenbrach, Rüdiger, Kevin Rageth, and René M Stulz, “How Valuable Is Financial
Flexibility when Revenue Stops? Evidence from the COVID-19 Crisis,” The Review of
Financial Studies, 2021.

Fama, Eugene F. and Kenneth R. French, “Common risk factors in the returns on

35
 stocks and bonds,” Journal of Financial Economics, 1993, 33 (1), 3–56.

Financial Stability Board, “The Availability of Data with Which to Monitor and Assess
Climate-Related Risks to Financial Stability,” Report, 2021.

and Network for Greening the Financial System, “Climate Scenario Analysis by
Jurisdictions: Initial findings and lessons,” Technical Report, Financial Stability Board,
Network for Greening the Financial System 2022.

Gandhi, Priyank and Hanno Lustig, “Size Anomalies in U.S. Bank Stock Returns,”
The Journal of Finance, 2015.

Ilhan, Emirhan, Zacharias Sautner, and Grigory Vilkov, “Carbon Tail Risk,” The
Review of Financial Studies, 06 2020, 34 (3), 1540–1571.

Ivanov, Ivan, Mathias S. Kruttli, and Sumudu W. Watugala, “Banking on Carbon:

Corporate Lending and Cap-and-Trade Policy,” Working Paper, 2021.

Kacperczyk, Marcin T. and Jose-Luis Peydro, “Carbon Emissions and the Bank-
Lending Channel,” Working Paper, 2021.

Laeven, Luc and Alexander Popov, “Carbon Taxes and the Geography of Fossil Lend-
ing,” November 2022.

Liberti, José Marı́a and Jason Sturgess, “The Anatomy of a Credit Supply Shock:
Evidence from an Internal Credit Market,” Journal of Financial and Quantitative Analysis,
2018, 53 (2), 547–579.

Litterman, Robert, “Selling stranded assets: Profit, protection, and prosper-

ity,” https://www.intentionalendowments.org/selling_stranded_assets_profit_
protection_and_prosperity

McGlade, C. and P Ekins, “The geographical distribution of fossil fuels unused when
limiting global warming to 2 °C.,” Nature 517, 187–190, 2015.

Nakashima, Kiyotaka and Koji Takahashi, “The real effects of bank-driven termination
of relationships: Evidence from loan-level matched data,” Journal of Financial Stability,
2018, 39 (C), 46–65.

Nguyen, Quyen, Ivan Diaz-Rainey, Duminda Kuruppuarachchi, Matthew Mc-

Carten, and Eric K.M. Tan, “Climate transition risk in U.S. loan portfolios: Are all
banks the same?,” International Review of Financial Analysis, 2023, 85, 102401.

Pástor, Ľuboš, Robert F. Stambaugh, and Lucian A. Taylor, “Dissecting green

returns,” Journal of Financial Economics, 2022, 146 (2), 403–424.

Reinders, Henk Jan, Dirk Schoenmaker, and Mathijs van Dijk, “A finance approach
to climate stress testing,” Journal of International Money and Finance, 2023, 131, 102797.

36
van der Ploeg, Frederick and Armon Rezai, “Stranded Assets in the Transition to a
Carbon-Free Economy,” Annual Review of Resource Economics, 2020, 12 (1), 281–298.

37
Figure 3: Average Probability of Default: Brown Firms vs. Non-brown Firms The
log-asset-weighted average probability of default of firms in brown industries and that of firms in
non-brown industries, based Y-14 from 2014:Q4 to 2021:Q4.

Figure 4: Climate Beta of US Banks The sample banks are the top 10 large US banks by the
average total assets in 2019. The sample period is from June 2000 to December 2021.

40
Figure 5: CRISK of US Banks The sample banks are the top 10 large US banks by the average
total assets in 2019. The sample period is from June 2000 to December 2021.

41
Figure 6: Marginal CRISK of US Banks The sample banks are the top 10 large US banks by
the average total assets in 2019. Marginal CRISK is difference between the stressed CRISK
and
non-stressed CRISK. The stressed CRISK is computed as: k·D−(1−k)·exp β Climate log(1 − θ) ·W
and the non-stressed CRISK is computed as: k · D − (1 − k) · W where k is prudential capital ratio,
D is debt, and W is market equity of each bank. The marginal CRISK values are truncated at
zero. The sample period is from June 2000 to December 2021.

42
Figure 11: Sensitivity Analysis The figure plots the aggregate marginal CRISK of the top 4
US banks across different severity of the scenario. The stranded asset factor is used. The scenario
with 1% quantile is the most severe and the scenario with 50% quantile (median) is the least severe.

Figure 12: S&CRISK of US Banks The sample banks are the top 10 large US banks by the
average total assets in 2019. The sample period is from June 2000 to Dec 2021.

47
Figure 13: Marginal S&CRISK of US Banks The sample banks are the top 10 large US banks
by the average total assets in 2019. The sample period is from June 2000 to Dec 2021.

48
Tables

(1) (2) (3) (4)

Climate Beta Climate Beta Climate Beta Climate Beta
Loan Portfolio Climate Beta 1.743∗∗∗ 1.630∗∗∗ 1.289∗∗∗ 0.833∗∗∗
(8.94) (9.27) (6.20) (3.28)

Log Assets 0.0141 0.400∗∗∗ 0.0927

(0.97) (5.64) (1.29)

Leverage 1.746 0.0513 -0.925

(1.54) (0.03) (-0.78)

ROA 7.997∗∗∗ 5.527∗∗∗ 2.236∗∗

(5.36) (4.61) (2.47)

Loans/Assets -0.0207 -0.192 -0.305

(-0.21) (-0.50) (-1.06)

Deposits/Assets 0.363∗∗∗ 0.395 -0.185

(3.72) (1.20) (-0.54)

Loan Loss Reserves/Loans -3.605∗ 3.597∗∗∗ 2.526

(-1.74) (3.47) (1.48)

Non-interest Income/Net Income 0.00186 0.00199 0.00225

(1.20) (1.42) (1.63)

Market Beta 0.172∗∗∗ 0.127∗∗∗ 0.0250

(5.16) (6.41) (1.32)

Book/Market 0.137∗∗∗ 0.0982∗∗∗ 0.00158

(3.35) (3.24) (0.04)
N 696 696 696 696
Bank Controls N Y Y Y
Bank FE N N Y Y
Year FE N N N Y
Adj R2 0.372 0.453 0.581 0.685
t statistics in parentheses
∗
p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

Table 1: Bank Climate Beta and Loan Portfolio Climate Beta Quarterly data from 2012:Q2
to 2021:Q4 for listed US banks in Y-14. Standard errors are clustered by banks.

49
 (1) (2) (3) (4)
Climate Beta Climate Beta Climate Beta Climate Beta
Loan Portfolio Climate Beta (Unlevered) 2.957∗∗∗ 2.491∗∗∗ 2.189∗∗∗ 1.231∗∗
(4.68) (4.74) (4.71) (2.70)

Log Assets 0.0164 0.464∗∗∗ 0.0755

(0.90) (5.29) (0.95)

Leverage 3.147∗∗∗ 0.534 -1.080

(2.85) (0.33) (-0.94)

ROA 8.588∗∗∗ 5.110∗∗∗ 1.881∗

(5.39) (3.66) (2.07)

Loans/Assets -0.0530 -0.557 -0.461

(-0.50) (-1.34) (-1.58)

Deposits/Assets 0.474∗∗∗ 0.901∗∗ -0.118

(2.96) (2.27) (-0.34)

Loan Loss Reserves/Loans -0.419 5.666∗∗∗ 2.989

(-0.17) (4.60) (1.70)

Non-interest Income/Net Income 0.00253 0.00211 0.00237

(1.39) (1.44) (1.72)

Market Beta 0.184∗∗∗ 0.113∗∗∗ 0.0168

(4.94) (5.74) (0.87)

Book/Market 0.168∗∗∗ 0.125∗∗∗ -0.00827

(3.18) (3.47) (-0.23)
N 696 696 696 696
Bank Controls N Y Y Y
Bank FE N N Y Y
Year FE N N N Y
Adj R2 0.191 0.326 0.550 0.678
t statistics in parentheses
∗
p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

Table 2: Bank Climate Beta and Loan Portfolio Climate Beta (Unlevered) Quarterly
data from 2012:Q2 to 2021:Q4 for listed US banks in Y-14. Standard errors are clustered by banks.
All variables are defined in Table A.1.

50
 (1) (2) (3) (4)
Climate Beta Climate Beta Climate Beta Climate Beta
Loan Portfolio Climate Beta 0.267 0.257 0.755∗∗∗ 0.733∗∗∗
(1.35) (1.68) (3.25) (2.92)

Log Assets 0.0124∗∗ 0.308∗∗∗ 0.143∗∗

(2.27) (4.03) (2.34)

Leverage 0.291 -0.600 -0.694

(0.46) (-0.33) (-0.57)

ROA 2.872∗∗ 2.702∗ -0.553

(2.56) (2.04) (-0.72)

Loans/Assets 0.0813∗ 0.613 0.0709

(1.88) (1.42) (0.21)

Deposits/Assets 0.0987∗∗∗ -0.176 -0.166

(3.11) (-0.45) (-0.50)

Loan Loss Reserves/Loans -5.764∗∗∗ 0.652 2.512

(-3.51) (0.30) (1.07)

Non-interest Income/Net Income -0.000379 -0.00111 -0.0000352

(-0.34) (-0.85) (-0.04)

Market Beta 0.212∗∗∗ 0.216∗∗∗ 0.0961∗∗

(5.58) (4.41) (2.77)

Book/Market 0.0921∗∗∗ 0.0564 -0.0843∗

(4.37) (1.43) (-1.86)
N 557 557 557 557
Bank Controls N Y Y Y
Bank FE N N Y Y
Year FE N N N Y
Adj R2 0.00564 0.114 0.145 0.400
t statistics in parentheses
∗
p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

Table 3: Bank Climate Beta and Loan Portfolio Climate Beta (Pre-COVID) Quarterly
data from 2012:Q2 to 2019:Q4 for listed US banks in Y-14. Standard errors are clustered by banks.
All variables are defined in Table A.1.

51
 (1) (2) (3) (4)
Climate Beta Climate Beta Climate Beta Climate Beta
Brown Loan Share 2.140∗∗∗ 1.466∗∗∗ 1.187∗ 0.691∗∗
(4.57) (3.02) (1.91) (2.51)

Brown PD Spread 8.265∗∗∗ 6.621∗∗∗ 4.262∗∗∗

(10.52) (8.15) (6.14)

Log Assets -0.0110 0.398∗∗∗ 0.00845

(-0.89) (2.95) (0.10)

Leverage 3.792∗∗∗ -1.155 -3.242∗∗

(5.53) (-0.83) (-2.36)

ROA 9.462∗∗∗ 5.426∗∗∗ 5.355∗∗∗

(4.27) (3.30) (4.04)

Loans/Assets -0.189∗∗ -1.437∗∗∗ -0.563

(-2.28) (-4.10) (-1.59)

Deposits/Assets 0.475∗∗∗ 1.185∗ 0.278

(3.25) (2.07) (0.77)

Book/Market 0.288∗∗∗ 0.265∗∗∗ 0.00286

(5.77) (7.78) (0.06)

Loan Loss Reserves/Loans 6.989∗∗∗ 7.158∗∗∗ 2.386

(4.01) (3.00) (1.42)

Non-interest Income/Net Income 0.00218 0.00254 0.00345

(1.06) (1.38) (1.64)

Market Beta -0.0359 -0.0335 -0.0606∗∗

(-1.06) (-1.42) (-2.80)
N 521 521 521 521
Bank Controls N Y Y Y
Bank FE N N Y Y
Year FE N N N Y
Adj R2 0.224 0.417 0.555 0.702
t statistics in parentheses
∗
p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

Table 4: Climate Beta, Brown Loan Share, and Brown-Nonbrown PD Spread The
dependent variable, βitClimate is bank i’s time-averaged daily climate beta during quarter-end month.
Brown Loan Shareit is bank i’s loan exposure to the top 30 industries with highest emissions in
quarter t. Brown P D Spread is the spread between the size-weighted average probability of default
of firms in the top 30 brown industries and that of firms not in the 30 brown industries. Bank
control variables include log assets, leverage, ROA, loans/assets, deposits/assets, book/market, loan
loss reserves/loans, non-interest income/net income, market beta. Standard errors are clustered at
bank level. The sample period is from 2014:Q4 to 2021:Q4, as the probability of default data are
mostly available from 2014:Q4. All variables are defined in Table A.1.

52
 Ticker CRISK(t-1) CRISK(t) dCRISK dDEBT dEQUITY dRISK
F:US −144.06 −7.85 136.21 37.63 35.4 63.17
J:US −42.83 71 113.84 −0.84 69.26 45.42
A:US −50.2 45.23 95.43 24.63 35.03 35.77
C:US 13.26 93.32 80.07 17.49 29.92 32.65
I:US −41.02 −4.43 36.59 4.13 16.04 16.43
H:US −25.94 −7.78 18.15 3.8 4.83 9.52
B:US −6.8 7.69 14.49 4.11 5.85 4.53
D:US −9.98 1.69 11.67 3.25 0.21 8.21
E:US 11.38 22.54 11.16 9.9 −6.63 7.9
G:US 4.38 −6.1 −10.48 3.65 −27.75 13.62
Top 4 425.55 78.91 169.62 177.01

Table 5: CRISK Decomposition (US Banks) CRISK(t) is the bank’s CRISK at the end of
2020, and CRISK(t − 1) is CRISK at the end of year 2019. dCRISK= CRISK(t)-CRISK(t − 1)
is the change in CRISK during 2020. dDEBT is the contribution of the firm’s debt to CRISK.
dEQUITY is the contribution of the firm’s equity position on CRISK. dRISK is the contribution
of increase in volatility or correlation to CRISK. All amounts are in billions USD. Top 4 banks
include F:US, J:US, A:US, and C:US.

53
Appendix
A Variable Definitions

Variable Definition
Log Assets Log of total assets
Leverage Liabilities/Assets
ROA Return on assets; Net Income/Assets
Loans/Assets Loans (gross)/Assets
Deposits/Assets Deposits/Assets
BTM Book to market; Book Value of Equity/ Market Capitalization
Loan Loss Reserves/Loans Loan Loss Reserves/Loans (gross)
Non-interest Income/Net Income Non-interest Income/Net Income
Market Beta Average market beta over the quarter-end months
(March, June, September, December)
Climate Beta Average climate beta over the quarter-end months
(March, June, September, December); Climate beta is the bank’s
stock return sensitivity to the stranded asset factor.
Loan Portfolio Climate Beta Loan-size-weighted industry climate beta; Climate beta for each
3-digit NAICS industry is the value-weighted average climate beta
of firms in the industry. The climate beta of each firm is the firm’s
stock return sensitivity to the stranded asset factor.
Loan Portfolio Climate Beta Loan-size-weighted unlevered industry climate beta; Climate beta
(Unlevered) for each 3-digit NAICS industry is the value-weighted average
unlevered climate beta of firms in the industry. The climate beta of
each firm is the firm’s stock return sensitivity to the stranded asset factor.
Brown Loan Share Share of loans made to the borrowers in the top 30 SIC 4-digit
industries by the sum of scope 1 and scope 2 carbon emissions.
Brown PD Spread The spread between the size-weighted average probability of
default of firms in the brown industries and that of firms not in
the brown industries. The brown industries are defined as the top
30 SIC 4-digit industries by sum of scope 1 and scope 2 carbon
emissions. The probability of default measures are based on
each bank’s internal assessment and reported to Y-14 as part of the
Dodd-Frank Act stress testing requirements.

Table A.1: Variable Definitions

54
B Summary Statistics

Mean St.Dev. Min Max 25th percentile 75th percentile Count

Stranded -0.00 0.01 -0.13 0.09 -0.01 0.01 5536
Emission 0.00 0.01 -0.12 0.14 -0.00 0.01 5536
BMG 0.00 0.01 -0.11 0.11 -0.01 0.01 3404
CEP -0.00 0.01 -0.10 0.14 -0.01 0.00 5158
SPY 0.00 0.01 -0.12 0.14 -0.00 0.01 5536
COVOL 0.60 0.30 0.02 2.83 0.41 0.73 5431

Table B.1: Factors Summary Statistics The sample is daily from 2000 to 2021. Stranded,
Emission, BMG, CEP each denotes stranded asset factor, emission factor, brown minus green factor,
and climate efficient factor mimicking portfolio factor.

Stranded Emission BMG CEP SPY COVOL

Stranded 1.00
Emission 0.38 1.00
BMG -0.24 -0.16 1.00
CEP -0.28 -0.79 0.34 1.00
SPY 0.10 0.89 -0.20 -0.79 1.00
COVOL -0.05 -0.01 0.06 0.02 -0.01 1.00

Table B.2: Factors Correlations The sample is daily from 2000 to 2021. Stranded, Emission,
BMG, CEP each denotes stranded asset factor, emission factor, brown minus green factor, and
climate efficient factor mimicking portfolio factor.

55
 Mean St.Dev. 25th percentile 75th percentile Count
Log Assets 19.65 1.20 18.64 20.66 696
Leverage 0.89 0.02 0.88 0.90 696
ROA 0.01 0.00 0.00 0.01 696
Loans/Assets 0.49 0.22 0.31 0.66 696
Deposits/Assets 0.68 0.16 0.66 0.78 696
Book/Market 1.00 0.35 0.75 1.18 696
Loan Loss Reserves/Loans 0.01 0.01 0.01 0.02 696
Non-interest Income/Net Income 2.32 3.97 1.40 3.07 696
Market Beta 1.04 0.23 0.88 1.16 696
Climate Beta 0.12 0.24 -0.03 0.23 696
Loan Portfolio Climate Beta 0.11 0.08 0.05 0.14 696
Loan Portfolio Climate Beta (Unlevered) 0.05 0.04 0.02 0.06 696
Brown Loan Share 0.03 0.02 0.02 0.04 696
Brown PD Spread 0.01 0.01 -0.00 0.02 521

Table B.3: Bank-level Data Summary Statistics Quarterly data from 2012 Q2 to 2021 Q4
for listed US banks in Y-14. The first eight variables (from Log Assets to Non-interest Income
Ratio) are from FR Y-9C. All variables are defined in Table A.1.

56
 (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14)

(1) Log Assets 1.00

(2) Leverage 0.26 1.00
(3) ROA -0.03 -0.14 1.00
(4) Loans/Assets -0.48 -0.68 0.16 1.00
(5) Deposits/Assets -0.64 -0.24 0.08 0.53 1.00
(6) Book/Market 0.20 -0.28 -0.34 0.05 -0.25 1.00
(7) Loan Loss Reserves/Loans 0.19 -0.36 -0.00 0.41 0.02 0.44 1.00
(8) Non-interest Income/Net Income 0.10 0.18 -0.13 -0.21 -0.11 0.15 -0.05 1.00
(9) Market Beta 0.18 0.12 -0.19 -0.26 -0.25 0.40 0.15 0.01 1.00
(10) Climate Beta 0.09 0.14 -0.07 -0.07 0.05 0.28 0.21 0.12 0.29 1.00
(11) Loan Portfolio Climate Beta 0.16 0.06 -0.15 0.03 -0.03 0.35 0.46 0.10 0.21 0.61 1.00
(12) Loan Portfolio Climate Beta (Unlevered) 0.14 -0.07 -0.17 0.10 -0.09 0.41 0.48 0.08 0.20 0.44 0.93 1.00
(13) Brown Loan Share 0.00 0.13 0.02 0.07 -0.02 0.08 0.11 -0.05 0.11 0.29 0.27 0.21 1.00
(14) Brown PD Spread 0.08 0.10 -0.19 -0.04 0.03 0.10 0.19 0.08 0.28 0.42 0.27 0.13 0.14 1.00

Table B.4: Bank-level Data Correlations Quarterly data from 2012 Q2 to 2021 Q4 for listed US banks in Y-14. The first nine
57

variables (from Log Assets to Non-interest Income Ratio) are from FR Y-9C. All variables are defined in Table A.1.
 Table C.1: List of Shock Events, extended from Barnett (2019)
Date Event Shock Source Type
11/7/2000 George W. Bush Elected POTUS -1 U.S. Presidential Elections election
11/25/2000 COP 6, The Hague, Netherlands 1 IPCC ipcc
3/28/2001 President George W. Bush withdraws from the Kyoto negotiations -1 Wikipedia policy
7/27/2001 COP 6, Bonn, Germany 1 IPCC ipcc
9/29/2001 IPCC Third assessment report 1 IPCC ipcc
11/10/2001 COP 7, Marrakech, Morocco 1 IPCC ipcc
5/13/2002 Farm Security and Rural Investment Act 1 Wikipedia policy
11/1/2002 COP 8, New Delhi, India 1 IPCC ipcc
2/6/2003 President Bush Unveils the Hydrogen Fuel Initiative 1 ProCon.org policy
2/27/2003 Plans Announced to Build World’s First Zero Emissions Coal Power 1 ProCon.org policy
Plant
12/12/2003 COP 9, Milan, Italy 1 IPCC ipcc
11/2/2004 George W. Bush Elected POTUS -1 U.S. Presidential Elections election
59

12/17/2004 COP 10, Buenos Aires, Argentina 1 IPCC ipcc

1/1/2005 EU Emissions Trading Scheme is launched, the first such scheme 1 Wikipedia/IPCC policy
2/16/2005 Kyoto Protocol comes into force (not including the US or Australia) 1 Wikipedia/IPCC policy
7/8/2005 3lat GS summit discusses climate change, relatively little progress 1 Wikipedia misc
made
8/8/2005 Energy Policy Act 1 Wikipedia policy
11/9/2005 US House Prevents Drilling for Oil in the Arcetic National Wildlife 1 ProCon.org policy
Refuge
12/9/2005 COP 11/CMP 1, Montreal, Canada 1 Wikipedia/IPCC ipcc
1/1/2006 IPCC’s Clean Development Mechanism Opens 1 IPCC ipcc
10/30/2006 The Stern Review is published 1 Wikipedia misc
11/17/2006 COP 12/CMP 2, Nairobi, Kenya 1 IPCC ipcc
2/16/2007 February 2007 Washington Declaration 1 IPCC ipcc
6/7/2007 33rd G8 summit 1 IPCC ipcc
7/31/2007 2007 UN General Assembly plenary debate 1 IPCC ipcc
 8/3/2007 September 2007 Washington conference 1 IPCC ipcc
8/31/2007 2007 Vienna Climate Change Talks and Agreement 1 IPCC ipcc
9/24/2007 September 2007 United Nations High-Level-Event 1 IPCC ipcc
11/17/2007 IPCC Fourth assessment report 1 IPCC/ProCon.org ipcc
12/17/2007 COP 13/CMP 3, Bali, Indonesia 1 IPCC ipcc
12/19/2007 Energy Independence and Security Act 1 Wikipedia policy
1/1/2008 IPCC’s Joint Implementation Mechanism Starts 1 IPCC ipcc
1/30/2008 First Commercial Cellulosic Ethanol Plant Goes Into Production 1 ProCon.org misc
5/22/2008 Food, Conservation, and Energy Act 1 Wikipedia policy
10/7/2008 National Biofuel Action Plan Unveiled 1 ProCon.org policy
11/4/2008 Barack Obama Elected POTUS 1 U.S. Presidential Elections election
12/12/2008 COP 14/CMP 4, Poznan, Poland 1 IPCC ipcc
12/22/2008 Worst Coal Ash Spill in US History in Kingston, Tennessee 1 ProCon.org misc
2/17/2009 ARRA (2009) Contains Funding for Renewable Energy 1 ProCon.org/Wikipedia policy
4/22/2009 First Framework for Wind Energy Development on the US Outer 1 ProCon.org policy
60

Continental Shelf Announced

5/5/2009 President Obama Issues Presidential Directive to USDA to Expand 1 ProCon.org policy
Access to Biofuels
5/27/2009 US Announces Funding in Recovery Act Funding for Solar and 1 ProCon.org policy
Geothermal Energy Development
6/26/2009 U3 House of Representatives passes the American Clean Energy 1 Wikipedia policy
and Security Act (Waxman)
9/22/2009 September 2009 United Nations Secretary General’s Summit on 1 IPCC ipcc
Climate Change
10/27/2009 US Invests $3.4 Billion to Modernize Energy Grid 1 ProCon.org policy
12/18/2009 COP 15/CMP 5, Copenhagen, Dennmark 1 IPCC ipcc
4/20/2010 BP Oil Rig Explodes & Causes Largest Oil Spill in US History 1 ProCon.org misc
12/10/2010 COP 16/CMP 6, Cancún, Mexico 1 IPCC ipcc
3/11/2011 Earthquake off Coast of Japan Damages Six Powerplants at 1 ProCon.org misc
Fukushima
 9/1/2011 Solar Power Company Solyndra Declares Bankruptcy -1 ProCon.org misc
11/22/2011 California cap-and-trade passed 1 Misc policy
12/9/2011 COP 17/CMP 7, Durban, South Africa 1 IPCC ipcc
2/9/2012 US Nuclear Regulatory Commission (NRC) Approves New Nuclear 1 ProCon.org policy
Power Plants
3/27/2012 EPA Announces First Clean Air Act Standard for Carbon Pollution 1 ProCon.org policy
from New Power Plants
4/17/2012 EPA Issues First Ever Clean Air Rules for Natural Gas Produced 1 ProCon.org policy
by Fracking
11/6/2012 Barack Obama Elected POTUS 1 U.S. Presidential Elections election
12/7/2012 COP 18/CMP 8, Doha, Qatar 1 IPCC ipcc
1/1/2013 California cap-and-trade effective 1 Misc policy
6/25/2013 President Obama Releases His Climate Action Plan 1 ProCon.org policy
9/20/2013 EPA Issues New Proposed Rule to Cut Greenhouse Gas Emissions 1 ProCon.org policy
from Power Plants
61

9/27/2013 IPCC Releases lat Part of Fifrth Assesment Report, Working Group 1 IPCC ipcc
1
11/23/2013 COP 19/CMP 9, Warsaw, Poland 1 IPCC ipcc
2/13/2014 Ivanpah, the World’s Largest Concentrated Solar Power Generation 1 ProCon.org misc
Plant, Goes Online
3/31/2014 IPCC Releases 1st Part of Fifth Assessment Report, Working 1 IPCC ipcc
Group 2
4/14/2014 IPCC Releases 3rd Part of Fifth Assessment Report, Working 1 IPCC ipcc
Group 3
5/9/2014 President Obama Announces Solar Power Commitments and Ex- 1 ProCon.org policy
ecutive Actions
6/2/2014 EPA Proposes First Ever Rules to Reduce Carbon Emissions from 1 ProCon.org policy
Existing Power Plants
9/22/2014 Rockefellers and over 800 Global Investors Announce Fossil Fuel 1 ProCon.org misc
Divestment
9/23/2014 Climate Summit 2014 1 IPCC ipcc
 11/1/2014 IPCC Fifth assessment report 1 IPCC ipcc
12/12/2014 COP 20/CMP 10, Lima, Peru 1 IPCC ipcc
1/1/2015 California cap-and-trade effective for fuel suppliers 1 Misc policy
8/3/2015 President Obama Announces Clean Power Plan 1 ProCon.org policy
9/29/2015 Carney Speech 1 Misc misc
10/23/2015 Clean Power Plan Finalized 1 ProCon.org policy
12/12/2015 COP 21/CMP 11, Paris, France 1 Wikipedia/IPCC ipcc
12/22/2015 Clean Power Plan Becomes Active 1 ProCon.org ipcc
11/8/2016 Donald Trump Elected POTUS -1 U.S. Presidential Elections election
11/18/2016 COP 22/CMP 12/CMA 1, Marrakech, Morocco 1 IPCC ipcc
3/28/2017 President Trump Signs Executive Order to Begin Reversal of Pres- -1 ProCon.org policy
ident Obama’ Clean Power Plan
6/1/2017 President Donald Trump withdraws the United States from the -1 Wikipedia policy
Paris Agreement
7/31/2017 Two Nuclear Power Reactors in South Carolina Abandoned Before -1 ProCon.org misc
62

Construction Completed
11/17/2017 COP 23, Bonn, Germany 1 IPCC ipcc
12/12/2017 One Planet Summit 1 IPCC ipcc
12/22/2017 Tax Bill Opens Arctic National Wildlife Refuge for Oil Drilling -1 ProCon.org policy
5/9/2018 Solar Power to Be Required on All New California Homes by 2020 1 ProCon.org policy
10/8/2018 Special Global Warming 1.5 Degree Celsius Report by IPCC Re- 1 IPCC misc
leased
12/14/2018 Katowice Climate Package adopted by Governments at COP 24, 1 IPCC policy
Katowice, Poland
3/22/2019 New Mexico Commits to 100% Renewable Energy for Electricity 1 ProCon.org policy
by 2050
12/2/2019 COP 25, Madrid, Spain 1 IPCC ipcc
3/31/2020 EPA Lowers Fuel Economy Standards -1 ProCon.org policy
4/1/2020 Big Banks Refuse Funds for Some Fossil Fuel Projects 1 ProCon.org misc
4/15/2020 Oil and Electricity Demands Drop during COVID-19 Pandemic 1 ProCon.org misc
 9/23/2020 California to Ban New Gas-Powered Cars by 2035 1 ProCon.org policy
11/3/2020 Biden Election 1 Elections election
12/9/2020 New York Says Employee Pension Fund Will Divest from Oil and 1 ProCon.org policy
Gas Companies if Not Aligned with Paris Agreement
12/15/2020 Fed joins NFGS 1 Misc misc
1/20/2021 Joe Biden signs executive order for the United States to rejoin the 1 Wikipedia policy
Paris Agreement
3/29/2021 Biden Administration Announces Offshore Wind Initiative 1 ProCon.org policy
4/22/2021 Biden Administration Pledges to Cut Greenhouse Gas Emissions 1 ProCon.org policy
by 50%, to 52%, by 2030
4/30/2021 Indian Nuclear Plant to Close -1 ProCon.org misc
5/11/2021 US Approves First Major American Offshore Wind Project 1 ProCon.org policy
5/18/2021 International Energy Agency Calls for No New Fossil Fuel Projects 1 ProCon.org misc
8/7/2021 IPCC Sixth Assessment Report predicting 1.5 in Warming -1 Wikipedia misc
9/21/2021 China Announces End to Building Coal-Burning Power Plants 1 ProCon.org policy
63

Abroad
11/9/2021 Major Automakers and Countries Pledge to Phase Out Gas- 1 ProCon.org policy
Powered Cars
11/10/2021 COP 26, Edinburgh, Scotland 1 Misc ipcc
12/15/2021 New York City to Ban New Natural Gas Connections 1 ProCon.org policy
 (1) (2) (3) (4) (5) (6)
ar(t-1,t) car(t-1,t+1) car(t-1,t+2) car(t-1,t+3) car(t-1,t+4) car(t-1,t+5)
shock -0.00144 -0.00233 -0.00380* -0.00174 -0.00113 -0.000367
(-1.14) (-1.41) (-1.85) (-0.68) (-0.41) (-0.12)

Constant -0.000108 -0.000196 -0.000370 -0.000312 -0.000517 -0.000784

(-0.57) (-0.50) (-0.63) (-0.40) (-0.53) (-0.68)
N 4828 2466 1677 1282 1048 892
Adj R2 0.0000705 0.000231 0.00106 -0.000460 -0.000819 -0.00111
Stranded
∗ ∗∗ ∗∗∗
p < 0.1, p < 0.05, p < 0.01

Table C.2: Responses of Climate Factor (Stranded) to Transition-related Climate

Events The list of events is from Barnett, extended to 2021. Total of 107 events are included.
shock takes a value of 1 if the event is associated with a movement toward a greener economy
(e.g., Paris Agreement) and it takes value of -1 if the event is associated with a movement away
from a greener economy (e.g., withdrawal from the Paris Agreement). The regressions are on a
non-overlapping data. Standard errors are Newey-West adjusted. Abnormal return ar is based on
the market model: rt = α + β spy spyt + εt , estimated on a 1-year rolling window basis: art = rt − rˆt

(1) (2) (3) (4) (5) (6)

ar(t-1,t) car(t-1,t+1) car(t-1,t+2) car(t-1,t+3) car(t-1,t+4) car(t-1,t+5)
shock -0.00139 -0.00230 -0.00359* -0.00123 -0.00105 -0.000567
(-1.07) (-1.35) (-1.76) (-0.49) (-0.37) (-0.18)

Constant -0.0000997 -0.000188 -0.000344 -0.000304 -0.000469 -0.000725

(-0.52) (-0.48) (-0.58) (-0.39) (-0.48) (-0.64)
N 4733 2418 1644 1258 1028 875
Adj R2 0.0000536 0.000216 0.000903 -0.000630 -0.000855 -0.00111
Stranded
∗ ∗∗ ∗∗∗
p < 0.1, p < 0.05, p < 0.01

Table C.3: Responses of Climate Factor (Stranded) to Transition-related Climate

Events after Controlling for COVOL The list of events is from Barnett, extended to 2021.
Total of 107 events are included. shock takes a value of 1 if the event is associated with a movement
toward a greener economy (e.g., Paris Agreement) and it takes a value of -1 if the event is associated
with a movement away from a greener economy (e.g., withdrawal from the Paris Agreement). The
regressions are on non-overlapping data. Standard errors are Newey-West adjusted. Abnormal
return ar is based on a two-factor model: rt = α + β spy spyt + β covol covolt + εt , estimated on a
1-year rolling window basis; art = rt − rˆt

64
 (1) (2) (3) (4) (5) (6)
ar(t-1,t) car(t-1,t+1) car(t-1,t+2) car(t-1,t+3) car(t-1,t+4) car(t-1,t+5)
shock -0.00101** -0.00179** -0.00191** -0.00210** -0.00110 -0.000863
(-2.25) (-2.44) (-2.38) (-2.26) (-1.11) (-0.85)

Constant -0.0000571 -0.000101 -0.000230 -0.000254 -0.000364 -0.000532

(-0.82) (-0.73) (-1.11) (-0.87) (-1.05) (-1.37)
N 4828 2466 1677 1282 1048 892
Adj R2 0.000731 0.00228 0.00271 0.00258 0.0000283 -0.000456
Emission
∗ ∗∗ ∗∗∗
p < 0.1, p < 0.05, p < 0.01

Table C.4: Responses of Climate Factor (Emission) to Transition-related Climate

Events The list of events is from Barnett, extended to 2021. Total of 107 events are included.
shock takes a value of 1 if the event is associated with a movement toward a greener economy
(e.g., Paris Agreement) and it takes a value of -1 if the event is associated with a movement away
from a greener economy (e.g., withdrawal from the Paris Agreement). The regressions are on non-
overlapping data. Standard errors are Newey-West adjusted. Abnormal return ar is based on the
market model: rt = α + β spy spyt + εt , estimated on a 1-year rolling window basis: art = rt − rˆt

(1) (2) (3) (4) (5) (6)

ar(t-1,t) car(t-1,t+1) car(t-1,t+2) car(t-1,t+3) car(t-1,t+4) car(t-1,t+5)
shock -0.00102** -0.00179** -0.00185** -0.00198** -0.00103 -0.00103
(-2.26) (-2.39) (-2.28) (-2.15) (-1.03) (-1.00)

Constant -0.0000296 -0.0000545 -0.000157 -0.000159 -0.000265 -0.000403

(-0.44) (-0.40) (-0.76) (-0.56) (-0.77) (-1.06)
N 4733 2418 1644 1258 1028 875
Adj R2 0.000785 0.00241 0.00268 0.00236 -0.0000790 -0.000136
Emission
∗ ∗∗ ∗∗∗
p < 0.1, p < 0.05, p < 0.01

Table C.5: Responses of Climate Factor (Emission) to Transition-related Climate

Events after Controlling for COVOL The list of events is from Barnett, extended to 2021.
Total of 107 events are included. shock takes a value of 1 if the event is associated with a movement
toward a greener economy (e.g., Paris Agreement) and it takes a value of -1 if the event is associated
with a movement away from a greener economy (e.g., withdrawal from the Paris Agreement). The
regressions are on non-overlapping data. Standard errors are Newey-West adjusted. Abnormal
return ar is based on a two-factor model: rt = α + β spy spyt + β covol covolt + εt , estimated on a
1-year rolling window basis: art = rt − rˆt

65
 (1) (2) (3) (4) (5) (6)
ar(t-1,t) car(t-1,t+1) car(t-1,t+2) car(t-1,t+3) car(t-1,t+4) car(t-1,t+5)
shock -0.00363** -0.00810*** -0.0130*** -0.0117*** -0.0114*** -0.0107**
(-2.14) (-3.38) (-4.12) (-3.49) (-2.86) (-2.30)

Constant 0.0000418 0.0000782 0.000169 0.0000220 0.000315 0.000407

(0.19) (0.17) (0.24) (0.02) (0.28) (0.29)
N 2884 1474 1004 766 630 535
Adj R2 0.00192 0.00938 0.0240 0.0175 0.0172 0.0125
BMG
∗ ∗∗ ∗∗∗
p < 0.1, p < 0.05, p < 0.01

Table C.6: Responses of Climate Factor (Brown minus Green) to Transition-related

Climate Events The list of events is from Barnett, extended to 2021. Total of 107 events are
included. shock takes a value of 1 if the event is associated with a movement toward a greener
economy (e.g., Paris Agreement) and it takes a value of -1 if the event is associated with a movement
away from a greener economy (e.g., withdrawal from the Paris Agreement). The regressions are on
non-overlapping data. Standard errors are Newey-West adjusted. Abnormal return ar is based on
the market model: rt = α + β spy spyt + εt , estimated on a 1-year rolling window basis: art = rt − rˆt

(1) (2) (3) (4) (5) (6)

ar(t-1,t) car(t-1,t+1) car(t-1,t+2) car(t-1,t+3) car(t-1,t+4) car(t-1,t+5)
shock -0.00370** -0.00819*** -0.0130*** -0.0117*** -0.0114*** -0.0107**
(-2.19) (-3.42) (-4.08) (-3.45) (-2.81) (-2.30)

Constant 0.000124 0.000245 0.000404 0.000371 0.000706 0.000865

(0.57) (0.54) (0.58) (0.40) (0.63) (0.61)
N 2884 1474 1004 766 630 535
Adj R2 0.00203 0.00970 0.0242 0.0177 0.0173 0.0128
BMG
∗ ∗∗ ∗∗∗
p < 0.1, p < 0.05, p < 0.01

Table C.7: Responses of Climate Factor (Brown minus Green) to Transition-related

Climate Events after Controlling for COVOL The list of events is from Barnett, extended
to 2021. Total of 107 events are included. shock takes a value of 1 if the event is associated
with a movement toward a greener economy (e.g., Paris Agreement) and it takes a value of -1 if
the event is associated with a movement away from a greener economy (e.g., withdrawal from the
Paris Agreement). The regressions are on non-overlapping data. Standard errors are Newey-West
adjusted. Abnormal return ar is based on a two-factor model: rt = α + β spy spyt + β covol covolt + εt ,
estimated on a 1-year rolling window basis; art = rt − rˆt

66
 (1) (2) (3) (4) (5) (6)
ar(t-1,t) car(t-1,t+1) car(t-1,t+2) car(t-1,t+3) car(t-1,t+4) car(t-1,t+5)
shock 0.000741 -0.000101 -0.000602 -0.00151 -0.00128 -0.00126
(0.83) (-0.08) (-0.35) (-0.81) (-0.60) (-0.53)

Constant 0.0000105 0.000105 0.000286 0.0000994 0.000410 0.000436

(0.09) (0.44) (0.76) (0.20) (0.69) (0.60)
N 4480 2289 1557 1192 973 829
Adj R2 -0.0000276 -0.000434 -0.000526 -0.000145 -0.000519 -0.000763
CEP
∗ ∗∗ ∗∗∗
p < 0.1, p < 0.05, p < 0.01

Table C.8: Responses of Climate Factor (CEP) to Transition-related Climate Events

The list of events is from Barnett, extended to 2021. Total of 107 events are included. shock
takes a value of 1 if the event is associated with a movement toward a greener economy (e.g.,
Paris Agreement) and it takes a value of -1 if the event is associated with a movement away from
a greener economy (e.g., withdrawal from the Paris Agreement). The regressions are on non-
overlapping data. Standard errors are Newey-West adjusted. Abnormal return ar is based on the
market model: rt = α + β spy spyt + εt , estimated on a 1-year rolling window basis: art = rt − rˆt

(1) (2) (3) (4) (5) (6)

ar(t-1,t) car(t-1,t+1) car(t-1,t+2) car(t-1,t+3) car(t-1,t+4) car(t-1,t+5)
shock 0.000704 -0.000150 -0.000629 -0.00146 -0.00116 -0.00111
(0.80) (-0.12) (-0.37) (-0.80) (-0.55) (-0.47)

Constant -0.00000564 0.0000719 0.000227 0.0000231 0.000297 0.000306

(-0.05) (0.30) (0.61) (0.05) (0.51) (0.43)
N 4480 2289 1557 1192 973 829
Adj R2 -0.0000444 -0.000429 -0.000513 -0.000172 -0.000600 -0.000857
CEP
∗ ∗∗ ∗∗∗
p < 0.1, p < 0.05, p < 0.01

Table C.9: Responses of Climate Factor (CEP) to Transition-related Climate Events

after Controlling for COVOL The list of events is from Barnett, extended to 2021. Total of
107 events are included. shock takes a value of 1 if the event is associated with a movement toward
a greener economy (e.g., Paris Agreement) and it takes a value of -1 if the event is associated
with a movement away from a greener economy (e.g., withdrawal from the Paris Agreement). The
regressions are on non-overlapping data. Standard errors are Newey-West adjusted. Abnormal
return ar is based on a two-factor model: rt = α + β spy spyt + β covol covolt + εt , estimated on a
1-year rolling window basis: art = rt − rˆt

67
D DCB Model Estimation
rit = log(1 + Rit ), rmt = log(1 + Rmt ), rct = log(1 + Rct )
Conditional on the information set Ft−1 , the return triple has a distribution D with zero
mean and time-varying covariance:
    
rit σit2 ρimt σit σmt ρict σit σct
rmt  Ft−1 ∼ D 0, Ht = ρimt σit σmt 2
σmt ρmct σmt σct 
2
rct ρict σit σct ρmct σmt σct σct

We use a GJR-GARCH volatility model and DCC correlation model. The GJR-GARCH
model for volatility dynamics are:
−
σit2 = ωV i + αV i rit−1
2 2
+ γV i rit−1 Ii,t−1 2
+ βV i σit−1 , (9)
2 2 2 − 2
σmt = ωV m + αV m rmt−1 + γV m rmt−1 Im,t−1 + βV m σmt−1 , (10)
2 2 2 − 2
σct = ωV c + αV c rct−1 + γV c rct−1 Ic,t−1 + βV c σct−1 (11)

where Iit− = 1 if rit < 0, Imt

−
= 1 if rmt < 0, and Ict− = 1 if rct < 0.

The correlation of the volatility-adjusted returns eit = rit /σit , emt = rmt /σmt , and ect =
rct /σct is:
   
ϵit 1 ρimt ρict
Cor ϵmt  = Rt = ρimt 1 ρmct  = diag(Qimct )−1/2 Qimct diag(Qimct )−1/2
ϵct ρict ρmct 1

The DCC model specifies the dynamics of the pseudo-correlation matrix Qimct as:
   ′
eit eit
Qimct = (1 − αCi − βCi )Si + αCi emt
  emt  + βCi Qimct−1 (12)
ect ect

where Sit is the unconditional correlation matrix of adjusted returns.

The market beta βitM kt and the climate beta βitClimate are:
   −1  
βitM kt σmt2
ρmct σmt σct ρimt σit σmt
= (13)
βitClimate ρmct σmt σct 2
σct ρict σit σct

Estimation procedure is as follows:

1. For each bank i = 1 · · · N , estimate GARCH parameters and DCC parameters.

2. Take the median DCC parameters, αC̄ = median(αCi ) and βC̄ = median(βCi ).

3. Compute βitM kt and βitClimate based on the median DCC parameters, αC̄ and βC̄ , and

68
 the volatility parameters.30

Here are the estimated parameters for the top 10 US banks.

Bank alpha alphaSE gamma gammaSE beta betaSE

A:US 0.0349 0.0147 0.0836 0.0197 0.9189 0.0223
B:US 0.0278 0.0186 0.1102 0.0257 0.9038 0.0252
C:US 0.0406 0.0094 0.0981 0.0164 0.9065 0.0136
D:US 0.0452 0.0147 0.1012 0.0324 0.8939 0.0182
E:US 0.0304 0.0111 0.0632 0.0195 0.9299 0.0165
F:US 0.0318 0.0101 0.1227 0.021 0.8994 0.0178
G:US 0.0289 0.0091 0.1014 0.0165 0.9137 0.0135
H:US 0.0543 0.0165 0.1566 0.0514 0.8536 0.0403
I:US 0.0349 0.012 0.105 0.0162 0.9078 0.0167
J:US 0.0326 0.0143 0.1036 0.028 0.9114 0.0264

Table D.1: Volatility Parameters

Bank alpha alphaSE beta betaSE

A:US 0.0301 0.0055 0.962 0.0077
B:US 0.0229 0.0079 0.9711 0.0115
C:US 0.0259 0.004 0.9676 0.0056
D:US 0.0129 0.0093 0.983 0.0132
E:US 0.0236 0.0036 0.9712 0.005
F:US 0.0289 0.0037 0.9636 0.0052
G:US 0.0219 0.0039 0.9724 0.0055
H:US 0.0269 0.0061 0.9655 0.0087
I:US 0.0255 0.0038 0.9696 0.005
J:US 0.0289 0.0045 0.9657 0.0057

Table D.2: DCC Parameters

30
The results are robust to using an individual bank’s DCC parameters instead of the median DCC
parameters.

69
E CRISK Derivation
 CF M kt 
i Pt+h Pt+h
1 − LRM ESit = Et 1 + Rt+1,t+h CF − 1 = −θ, M kt − 1 = 0
Pt+1 Pt+1
" h
! #
CF M kt
X
i P t+h P t+h
= Et exp rt+j CF
− 1 = −θ, M kt − 1 = 0
j=1
Pt+1 Pt+1
" h
! #
CF M kt
X P t+h P
= Et exp M kt M kt
βi,t+j Climate CF
rt+j + βi,t+j rt+j + εi,t+j CF
− 1 = −θ, t+h
M kt
−1=0
j=1
P t+1 P t+1
" ! !! #
M kt CF CF M kt
M kt
Pt+1,t+j CF
Pt+1,t+j Pt+h Pt+h
= Et exp βit log M kt
+ βit log CF CF
− 1 = −θ, M kt − 1 = 0
Pt+1 Pt+1 Pt+1 Pt+1
= exp βitClimate log(1 − θ)

Therefore,
i CF
CRISKit = kDit − (1 − k)Wit {1 + Et [Rt+1,t+h |Rt+1,t+h < C]}
| {z }
1−LRM ESit

= kDit − (1 − k)Wit exp βitClimate log(1 − θ)

70
F Climate Betas of Non-US Banks

Figure F.1: Climate Betas of UK Banks The sample banks are the top 5 largest UK banks
by average total assets in 2019. The sample period is from June 2000 to December 2021.

71
Figure F.2: Climate Betas of Canadian Banks The sample banks are the top 6 largest
Canadian banks by average total assets in 2019. The sample period is from June 2000 to December
2021.

72
Figure F.3: Climate Betas of Japanese Banks The sample banks are the top 3 largest
Japanese banks by average total assets in 2019. The sample period is from June 2000 to December
2021.

Figure F.4: Climate Betas of French Banks The sample banks are the top 3 largest French
banks by average total assets in 2019. The sample period is from June 2000 to December 2021.

73
G CRISKs of Non-US Banks

Figure G.5: CRISKs of UK Banks The sample banks are the top 5 largest UK banks by
average total assets in 2019. The sample period is from June 2000 to December 2021.

74
Figure G.6: CRISKs of Canadian Banks The sample banks are the top 6 largest Canadian
banks by average total assets in 2019. The sample period is from June 2000 to December 2021.

Figure G.7: CRISKs of Japanese Banks The sample banks are the top 3 largest Japanese
banks by average total assets in 2019. The sample period is from June 2000 to December 2021.

75
Figure G.8: CRISKs of French Banks The sample banks are the top 3 largest French banks
by average total assets in 2019. The sample period is from June 2000 to December 2021.

H CRISK Decomposition of Non-US banks

Bank CRISK(t-1) CRISK(t) dCRISK dDEBT dEQUITY dRISK

A:LN 56.52 80.38 23.86 13.39 3.48 7
B:LN 17.72 93.4 75.68 21.75 33.8 20.12
C:LN 17.74 42.28 24.54 1.88 11.54 11.12
D:LN 26.28 39.77 13.5 3.59 5.83 4.07
E:LN 16.84 27.76 10.92 3.64 5.78 1.5

Table H.1: CRISK Decomposition (UK) CRISK(t) is the bank’s CRISK at the end of 2020,
and CRISK(t − 1) is CRISK at the end of year 2019. dCRISK= CRISK(t)-CRISK(t − 1) is the
change in CRISK during 2020. dDEBT is the contribution of the firm’s debt to CRISK. dEQUITY
is the contribution of the firm’s equity position on CRISK. dRISK is the contribution of an increase
in climate beta to CRISK. All amounts are in billions USD.

76
 Bank CRISK(t-1) CRISK(t) dCRISK dDEBT dEQUITY dRISK
A:CN 11.91 25.02 13.11 7.37 0.5 5.24
B:CN 5.91 22.94 17.03 5.6 2.53 8.9
C:CN 12.69 16.34 3.64 7.09 −0.62 −2.82
D:CN −0.07 3.73 3.8 2.58 −0.26 1.47
E:CN −6.55 8.83 15.38 15.62 −2.36 2.12
F:CN 7.31 29.46 22.15 16.42 −0.06 5.79

Table H.2: CRISK Decomposition (Canada) CRISK(t) is CRISK at the end of 2020, and
CRISK(t-1) is CRISK at the end of year 2019. dCRISK= CRISK(t)-CRISK(t-1) is the change in
CRISK during 2020. dDEBT is the contribution of the firm’s debt to CRISK. dEQUITY is the
contribution of the firm’s equity position on CRISK. dRISK is the contribution of an increase in
climate beta to CRISK.

Bank CRISK(t-1) CRISK(t) dCRISK dDEBT dEQUITY dRISK

A:JP 160.14 186.56 26.41 9.42 9.52 7.48
B:JP 101.19 126.27 25.08 11.27 5.92 7.89
C:JP 107.84 125.43 17.59 5.19 5.39 7.01

Table H.3: CRISK Decomposition (Japan) CRISK(t) is CRISK at the end of 2020, and
CRISK(t-1) is CRISK at the end of year 2019. dCRISK= CRISK(t)-CRISK(t-1) is the change in
CRISK during 2020. dDEBT is the contribution of the firm’s debt to CRISK. dEQUITY is the
contribution of the firm’s equity position on CRISK. dRISK is the contribution of an increase in
climate beta to CRISK.

Bank CRISK(t-1) CRISK(t) dCRISK dDEBT dEQUITY dRISK

A:FP 71.02 105 33.97 19.67 3.08 11.23
B:FP 66.98 127.6 60.62 37.71 5.06 17.85
C:FP 59.19 82.59 23.41 10.22 7.01 6.17

Table H.4: CRISK Decomposition (France) CRISK(t) is CRISK at the end of 2020, and
CRISK(t-1) is CRISK at the end of year 2019. dCRISK= CRISK(t)-CRISK(t-1) is the change in
CRISK during 2020. dDEBT is the contribution of the firm’s debt to CRISK. dEQUITY is the
contribution of the firm’s equity position on CRISK. dRISK is the contribution of an increase in
climate beta to CRISK.

77
I Marginal CRISKs of Non-US Banks

Figure I.1: Marginal CRISKs: UK The sample banks are the top 5 largest UK banks by
average total assets in 2019. Marginal CRISK is the difference between the stressed CRISK
and
non-stressed CRISK. The stressed CRISK is computed as: kD − (1 − k) exp β Climate log(1 − θ) W
and the non-stressed CRISK is computed as: kD − (1 − k)W where k is prudential capital ratio, D
is debt, and W is market equity of each bank. The marginal CRISK values are truncated at zero.
The sample period is from June 2000 to December 2021.

78
Figure I.2: Marginal CRISKs: Canada The sample banks are the top 6 largest Canadian
banks by average total assets in 2019. The marginal CRISK values are truncated at zero. The
sample period is from June 2000 to December 2021.

79
Figure I.3: Marginal CRISKs: Japan The sample banks are the top 3 largest Japanese banks
by average total assets in 2019. The marginal CRISK values are truncated at zero. The sample
period is from June 2000 to December 2021.

80
Figure I.4: Marginal CRISKs: France The sample banks are the top 3 largest French banks
by average total assets in 2019. The marginal CRISK values are truncated at zero. The sample
period is from June 2000 to December 2021.

J Full List of Financial Firms

Canada
Ticker Company Name Ticker Company Name
BMO Bank of Montreal BNS Bank of Nova Scotia
CIX CI Financial Corp CM Canadian Imperial Bank of Commerce
FFH Fairfax Financial Holdings Ltd FNV Franco-Nevada Corp
GWO Great-West Lifeco Inc IAG iA Financial Corp Inc
IFC Intact Financial Corp IGM IGM Financial Inc
MFC Manulife Financial Corp NA National Bank of Canada
ONEX Onex Corp POW Power Corp of Canada
RY Royal Bank of Canada SLF Sun Life Financial Inc
TD Toronto-Dominion Bank X TMX Group Ltd

Table J.1: Canadian Financial Firms

81
 Japan
Ticker Company Name Ticker Company Name
3231 Nomura Real Estate Holdings Inc 7167 Mebuki Financial Group Inc
7180 Kyushu Financial Group Inc 7181 Japan Post Insurance Co Ltd
7182 Japan Post Bank Co Ltd 7186 Concordia Financial Group Ltd
7327 Daishi Hokuetsu Financial Group Inc 8253 Credit Saison Co Ltd
8303 Shinsei Bank Ltd 8304 Aozora Bank Ltd
8306 Mitsubishi UFJ Financial Group Inc 8308 Resona Holdings Inc
8309 Sumitomo Mitsui Trust Holdings Inc 8316 Sumitomo Mitsui Financial Group Inc
8331 Chiba Bank Ltd 8334 Gunma Bank Ltd
8341 77 Bank Ltd 8354 Fukuoka Financial Group Inc
8355 Shizuoka Bank Ltd 8359 Hachijuni Bank Ltd
8366 Shiga Bank Ltd 8369 Bank of Kyoto Ltd
8370 Kiyo Bank Ltd 8377 Hokuhoku Financial Group Inc
8379 Hiroshima Bank Ltd 8382 Chugoku Bank Ltd
8385 Iyo Bank Ltd 8410 Seven Bank Ltd
8411 Mizuho Financial Group Inc 8418 Yamaguchi Financial Group Inc
8421 Shinkin Central Bank 8439 Tokyo Century Corp
8473 SBI Holdings Inc 8570 AEON Financial Service Co Ltd
8572 Acom Co Ltd 8591 ORIX Corp
8593 Mitsubishi HC Capital Inc 8601 Daiwa Securities Group Inc
8604 Nomura Holdings Inc 8628 Matsui Securities Co Ltd
8630 Sompo Holdings Inc 8725 MS&AD Insurance Group Holdings Inc
8750 Dai-ichi Life Holdings Inc 8766 Tokio Marine Holdings Inc
8795 T&D Holdings Inc 8801 Mitsui Fudosan Co Ltd
8802 Mitsubishi Estate Co Ltd 8804 Tokyo Tatemono Co Ltd
8830 Sumitomo Realty & Development Co Ltd 8905 Aeon Mall Co Ltd

Table J.2: Japanese Financial Firms

France
Ticker Company Name Ticker Company Name
ACA Credit Agricole SA ALTA Altarea SCA
BNP BNP Paribas SA COFA Coface SA
CAF Caisse Regionale de Credit Agricole CNF Caisse Regionale de Credit Agricole
Mutuel de Paris et d’Ile-de-France Mutuel Nord de France
COV Covivio COVH Covivio Hotels SACA
CRAV Credit Agricole Atlantique Vendee CRSU Credit Agricole Sud Rhone Alpes
CS AXA SA FLY Societe Fonciere Lyonnaise SA
GFC Gecina SA GLE Societe Generale SA
ICAD ICADE LI Klepierre
MERY Mercialys SA MF Wendel SA
NXI Nexity SA ODET Compagnie de L’Odet SA
PEUG Peugeot Invest RF Eurazeo SA
ROTH Rothschild & Co SCR SCOR SE

Table J.3: French Financial Firms

82
 United Kingdom
Ticker Company Name Ticker Company Name
ABDN Abrdn Plc ADM Admiral Group PLC
ASHM Ashmore Group PLC BARC Barclays PLC
BLND British Land Co PLC BYG Big Yellow Group PLC
CAPC Capital & Counties Properties PLC CBG Close Brothers Group PLC
DLG Direct Line Insurance Group PLC DLN Derwent London PLC
GPE Great Portland Estates PLC GRI Grainger PLC
HMSO Hammerson PLC HSBA HSBC Holdings PLC
ICP Intermediate Capital Group PLC IGG IG Group Holdings PLC
III 3i Group PLC JUP Jupiter Fund Management PLC
LAND Land Securities Group PLC LGEN Legal & General Group PLC
LLOY Lloyds Banking Group PLC LSEG London Stock Exchange Group PLC
NWG Natwest Group PLC PHNX Phoenix Group Holdings
PRU Prudential PLC SDR Schroders PLC
SGRO Segro PLC SHB Shaftesbury PLC
STAN Standard Chartered PLC STJ St James’s Place PLC
SVS Savills PLC TCAP TP ICAP Group PLC
UTG UNITE Group PLC VMUK Virgin Money UK PLC

Table J.4: UK Financial Firms

83
 United States
Ticker Company Name Ticker Company Name
AFG American Financial Group Inc AFL Aflac Inc
AIG American International Group Inc AIZ Assurant Inc
AJG Arthur J Gallagher & Co AL Air Lease Corp
ALL Allstate Corp ALLY Ally Financial Inc
AMP Ameriprise Financial Inc AON Aon PLC
APO Apollo Global Management Inc ARCC Ares Capital Corp
AXP American Express Co BAC Bank of America Corp
BEN Franklin Resources Inc BK Bank of New York Mellon Corp
BLK BlackRock Inc BOKF BOK Financial Corp
BPOP Popular Inc BRO Brown & Brown Inc
BX Blackstone Inc C Citigroup Inc
CACC Credit Acceptance Corp CBOE CBOE Global Markets Inc
CBRE CBRE Group Inc CBSH Commerce Bancshares Inc
CFG Citizens Financial Group Inc CFR Cullen/Frost Bankers Inc
CI Cigna Corp CINF Cincinnati Financial Corp
CMA Comerica Inc CME CME Group Inc
CNA CNA Financial Corp COF Capital One Financial Corp
DFS Discover Financial Services EFX Equifax Inc
ERIE Erie Indemnity Co EWBC East West Bancorp Inc
FAF First American Financial Corp FCNCA First Citizens BancShares Inc
FHN First Horizon Corp FITB Fifth Third Bancorp
FNF Fidelity National Financial Inc FRC First Republic Bank
GL Globe Life Inc GS Goldman Sachs Group Inc
HBAN Huntington Bancshares Inc HHC Howard Hughes Corp
HIG Hartford Financial Services Group Inc HUM Humana Inc
ICE Intercontinental Exchange Inc IVZ Invesco Ltd
JEF Jefferies Financial Group Inc JLL Jones Lang LaSalle Inc
JPM JPMorgan Chase & Co KEY KeyCorp
KKR KKR & Co Inc KMPR Kemper Corp
L Loews Corp LNC Lincoln National Corp
LPLA LPL Financial Holdings Inc MA MasterCard Inc
MCO Moody’s Corp MET MetLife Inc
MKL Markel Corp MMC Marsh & McLennan Cos Inc
MS Morgan Stanley MSCI MSCI Inc
MTB M&T Bank Corp NDAQ Nasdaq Inc
NTRS Northern Trust Corp NYCB New York Community Bancorp Inc
ORI Old Republic International Corp PB Prosperity Bancshares Inc
PFG Principal Financial Group Inc PGR Progressive Corp
PNC PNC Financial Services Group Inc PRI Primerica Inc
PRU Prudential Financial Inc RF Regions Financial Corp
RGA Reinsurance Group of America Inc RJF Raymond James Financial Inc
SBNY Signature Bank/New York NY SCHW Charles Schwab Corp
SEIC SEI Investments Co SIVB SVB Financial Group
SNV Synovus Financial Corp STT State Street Corp
TFC Truist Financial Corp TFSL TFS Financial Corp
THG Hanover Insurance Group Inc TROW T Rowe Price Group Inc
TRV Travelers Cos Inc UNH UnitedHealth Group Inc
UNM Unum Group USB US Bancorp
V Visa Inc VOYA Voya Financial Inc
WAL Western Alliance Bancorp WBS Webster Financial Corp
WFC Wells Fargo & Co WRB WR Berkley Corp
WTW Willis Towers Watson PLC WU Western Union Co
ZION Zions Bancorporation

Table J.5: US Financial Firms

84
K More Scenarios
K.1 Emission Factor

Figure K.1: Climate Betas based on emission-cased factor. The sample banks are the top
10 largest US banks by average total assets in 2019. The emission-based factor is constructed
by weighting emissions across industries and weighting stock returns by market value within each
industry. The sample period is from June 2000 to December 2021.

85
Figure K.2: Marginal CRISKs based on emission-cased factor. The sample banks are the
top 10 largest US banks by average total assets in 2019. The emission-based factor is constructed
by weighting emissions across industries and weighting stock returns by market value within each
industry. The sample period is from June 2000 to December 2021.

86
K.2 Brown Minus Green Factor

Figure K.3: Climate Betas based on Brown minus Green factor. The sample banks are the top
10 largest US banks by average total assets in 2019. We use the emission-based factor as brown
factor and the iShares Global Clean Energy ETF return as green factor. The sample period is from
June 2008 to December 2021.

87
Figure K.4: Marginal CRISKs based on Brown minus Green factor. The sample banks are
the top 10 largest US banks by average total assets in 2019. We use the emission-based factor as
brown factor and the iShares Global Clean Energy ETF return as green factor. The sample period
is from June 2008 to December 2021.

88
K.3 Climate Efficient Factor Mimicking Portfolio Factor

Figure K.5: Climate Betas based on climate efficient factor mimicking portfolio factor. The
sample banks are the top 10 largest US banks by average total assets in 2019. The sample period
is from July 2001 to December 2021.

89
Figure K.6: Marginal CRISKs based on climate efficient factor mimicking portfolio factor.
The sample banks are the top 10 largest US banks by average total assets in 2019. The sample
period is from July 2001 to December 2021.

90
L Robustness Tests

Figure L.7: Climate Beta after Controlling for LTG and CRD The sample banks are the
top 10 largest US banks by average total assets in 2019. First, we regress bank stock return on
LTG and CRD. Second, we regress the residual from the first step on MKT and CF and plot the
coefficient on CF using 252-day rolling window regression. LTG is log daily return on long-term
US government bond index. CRD is log daily return on investment-grade corporate bond index
and can be downloaded from Bloomberg. The sample period is from June 2000 to December 2021.

91
Figure L.8: Climate Beta after Controlling for HOUSE The sample banks are the top 10
largest US banks by average total assets in 2019. First, we regress bank stock return on HOUSE.
Second, we regress the residual from the first step on MKT and CF and plot the coefficient on CF
using 252-day rolling window regression. HOUSE is the log daily return on a bond fund specializing
in government mortgage-backed securities (VFIJX). The sample period is from February 2001 to
December 2021.

92
Figure L.9: Climate Beta after Controlling for LTG, CRD, and HOUSE The sample
banks are the top 10 largest US banks by average total assets in 2019. First, we regress bank stock
return on HOUSE, LTG, and CRD. Second, we regress the residual from the first step on MKT
and CF and plot the coefficient on CF using 252-day rolling window regression. HOUSE is the log
daily return on a bond fund specializing in government mortgage-backed securities (VFIJX). LTG
is log daily return on long-term US government bond index and CRD is the log daily return on
investment-grade corporate bond index. The sample period is from February 2001 to December
2021.

93
Figure L.10: Climate Beta after Controlling for COVID Industry Factor The sample
banks are the top 10 largest US banks by average total assets in 2019. First, we regress bank stock
return on a COVID industry factor. The COVID industry factor is a value-weighted return on stocks
that belong to the NAICS 3-digit industries most affected by COVID(selected by Fahlenbrach et al.
(2021)). We exclude five industries that are in the top 20 by emissions in 2020. Second, we regress
the residual from the first step on MKT and CF and plot the coefficient on CF using 252-day rolling
window regression. The sample period is from January 2001 to December 2021.

94
 Internet Appendix to
“CRISK: A Market-based Climate Stress Test”

IA.A Fixed Beta Estimation

For each firm i we estimate the following OLS specification:

rit = α + βiM kt M KTi + βiClimate CFi + ϵit

M KT denotes return on market and SPY is used. For CF , the stranded asset factor is
used. The full sample period is 06/02/2000 - 12/31/2021 and the post-crisis sample period
is 01/01/2010 - 12/31/2021. Standard errors are Newey-West adjusted with an optimally
selected number of lags. We focus on the top 10 banks by average total assets in the year
2019.

US Banks

Bank CF tstatCF MKT tstatMKT CONS tstatCONS Rsq N

A:US 0.12 3.03 1.53 19.61 -0.0002 -0.77 0.45 5431
B:US 0.07 2.13 1.32 27.26 -0.0002 -1.22 0.5 5431
C:US 0.11 2.94 1.66 20.58 -0.0007 -2.4 0.46 5431
D:US 0.03 0.79 1.57 24.91 -0.0002 -0.74 0.42 5431
E:US 0.02 0.56 1.35 31.33 0 -0.23 0.53 5431
F:US -0.02 -0.48 1.46 19.83 -0.0001 -0.55 0.54 5431
G:US -0.01 -0.17 1.82 19.41 -0.0004 -1.65 0.55 5431
H:US 0.03 0.93 1.24 15.78 0 0.04 0.42 5431
I:US 0 0.01 1.14 19.36 0 -0.13 0.43 5431
J:US 0.08 2.31 1.27 17.06 -0.0001 -0.43 0.43 5431

Table IA.A.1: Large banks, SPY, Stranded Asset Factor

IA.1
 Bank CF tstatCF MKT tstatMKT CONS tstatCONS Rsq N
A:US 0.27 7.3 1.44 27.63 -0.0002 -0.57 0.54 3021
B:US 0.17 6.04 1.14 33.24 -0.0002 -0.77 0.54 3021
C:US 0.33 8.86 1.5 33.9 -0.0003 -1.23 0.6 3021
D:US 0.2 4.23 1.36 24.72 -0.0001 -0.34 0.51 3021
E:US 0.19 6.67 1.23 38.36 -0.0002 -0.87 0.56 3021
F:US 0.21 6.81 1.24 44.42 0 0.12 0.61 3021
G:US 0.26 8.12 1.51 34.79 -0.0002 -0.62 0.59 3021
H:US 0.15 4.53 1.2 32.47 0 -0.03 0.56 3021
I:US 0.13 3.84 1.13 29.97 -0.0001 -0.61 0.56 3021
J:US 0.17 4.67 1.25 28.22 -0.0003 -1.11 0.55 3021

Table IA.A.2: Large banks, SPY, Stranded Asset Factor, Post-Crisis

Non-US Banks
To account for non-synchronous trading, we include a lagged value of each explanatory
variable:
rit = α + β1i M KTt + β2i M KTt−1 + γ1i CFt + γ2i CFt−1 + ϵit
We report the bias-adjusted coefficients β1i + β2i (labeled as MKT), γ1it + γ2it (labeled
as CF) and their t-statistics below.

Bank CF tstatCF MKT tstatMKT CONS tstatCONS Rsq N

A:LN 0.25 4.45 1.61 20.41 -0.0004 -1.08 0.24 5335
B:LN 0.15 4.74 0.97 22.08 -0.0001 -0.65 0.28 5335
C:LN 0.2 3.79 1.33 13.21 -0.0005 -1.49 0.18 5335
D:LN 0.26 3.83 1.48 15.22 -0.0005 -1.35 0.2 5335
E:LN 0.28 5.59 1.32 16.43 -0.0002 -0.87 0.25 5335
A:CN 0.15 4.4 0.97 18.53 0.0002 1.31 0.39 5317
B:CN 0.21 6.9 0.97 20.17 0.0002 1.54 0.39 5317
C:CN 0.14 4.17 1.05 19.23 0.0001 0.35 0.44 5317
D:CN 0.17 4.79 0.96 15.45 0.0004 1.95 0.34 5317
E:CN 0.18 6.01 0.96 19.41 0.0003 2 0.42 5317
F:CN 0.15 5.29 1 24.8 0.0002 1.43 0.43 5317
A:JP 0.14 3.41 0.74 13.18 -0.0002 -0.82 0.11 4909
B:JP 0.18 3.5 0.81 14.33 -0.0002 -0.79 0.13 4514
C:JP 0.17 3.06 0.75 12.62 -0.0001 -0.36 0.1 4452
A:FP 0.27 4.26 1.45 19.76 -0.0002 -0.83 0.26 5000
B:FP 0.22 5.07 1.37 18.14 -0.0001 -0.36 0.27 5378
C:FP 0.22 3.95 1.59 21.61 -0.0003 -1.01 0.28 5378

Table IA.A.3: Large banks, SPY, Stranded Asset Factor

IA.2
 Bank CF tstatCF MKT tstatMKT CONS tstatCONS Rsq N
A:LN 0.49 8.07 1.64 15.8 -0.0006 -1.6 0.32 2967
B:LN 0.31 7.44 0.87 17.01 -0.0003 -1.4 0.3 2967
C:LN 0.34 5.62 1.43 14.93 -0.0005 -1.44 0.26 2967
D:LN 0.38 6.24 1.46 16.33 -0.0006 -1.36 0.24 2967
E:LN 0.48 8.43 1.19 19.73 -0.0006 -1.94 0.28 2967
A:CN 0.31 10.86 0.98 12.09 0.0001 0.36 0.51 2958
B:CN 0.36 10.93 0.94 15.31 0 0.02 0.51 2958
C:CN 0.29 9.8 0.95 10.7 0 -0.21 0.52 2958
D:CN 0.32 9.28 1 10.28 0.0001 0.55 0.45 2958
E:CN 0.28 10.36 0.92 21.78 0 0.25 0.51 2958
F:CN 0.29 10.58 0.92 18.04 0.0001 0.64 0.53 2958
A:JP 0.25 5.34 0.76 14.45 -0.0002 -0.64 0.14 2838
B:JP 0.24 5.64 0.72 14.41 -0.0002 -0.55 0.14 2838
C:JP 0.17 3.83 0.64 12.48 -0.0003 -1.03 0.11 2838
A:FP 0.49 7.68 1.56 15.41 -0.0005 -1.26 0.31 2995
B:FP 0.43 6.73 1.52 16.95 -0.0005 -1.41 0.33 2995
C:FP 0.49 6.8 1.78 16.5 -0.0008 -1.82 0.34 2995

Table IA.A.4: Large banks, SPY, Stranded Asset Factor, Post-Crisis

IA.B Rolling Window Beta Estimation

This section presents climate beta estimates based on 252-day rolling window regressions.

IA.3
IA.B.1 US Banks

Figure IA.B.1: Climate Beta of US Banks based on 252-day rolling window regression from
June 2000 to December 2021. The sample banks are the top 10 largest US banks by average total
assets in 2019.

IA.4
Figure IA.B.2: Market Beta of US Banks based on 252-day rolling window regression from
June 2000 to December 2021. The sample banks are the top 10 largest US banks by average total
assets in 2019.

IA.5
IA.B.2 UK Banks

Figure IA.B.3: Climate Beta of UK Banks based on 252-day rolling window regression from
June 2000 to December 2021. The sample banks are the top 5 largest UK banks by average total
assets in 2019.

Figure IA.B.4: Market Beta of UK Banks based on 252-day rolling window regression from
June 2000 to December 2021. The sample banks are the top 5 largest UK banks by average total
assets in 2019.

IA.6
IA.B.3 Canadian Banks

Figure IA.B.5: Climate Beta of Canadian Banks based on 252-day rolling window regression
from June 2000 to December 2021. The sample banks are the top 6 largest Canadian banks by
average total assets in 2019.

IA.7
Figure IA.B.6: Market Beta of Canadian Banks based on 252-day rolling window regression
from June 2000 to December 2021. The sample banks are the top 6 largest Canadian banks by
average total assets in 2019.

IA.B.4 Japanese Banks

Figure IA.B.7: Climate Beta of Japanese Banks based on 252-day rolling window regression
from June 2000 to December 2021. The sample banks are the top 3 largest Japanese banks by
average total assets in 2019.

IA.8
Figure IA.B.8: Market Beta of Japanese Banks based on 252-day rolling window regression
from June 2000 to December 2021. The sample banks are the top 3 largest Japanese banks by
average total assets in 2019.

IA.B.5 French Banks

Figure IA.B.9: Climate Beta of French Banks based on 252-day rolling window regression
from June 2000 to December 2021. The sample banks are the top 3 largest French banks by average
total assets in 2019.

IA.9
Figure IA.B.10: Market Beta of French Banks based on 252-day rolling window regression
from June 2000 to December 2021. The sample banks are the top 3 largest French banks by average
total assets in 2019.

IA.C Additional Robustness Results

Figure IA.C.1: Climate Beta after Controlling for the number of seated diners The
sample banks are the top 10 largest US banks by average total assets in 2019. First, we regress bank
stock return on DINER. Second, we regress the residual from the first step on MKT and CF and
plot the coefficient on CF using 252-day rolling window regression. DINER is the daily percentage
change of the number of seated diners on same day of the same week in 2020-22 compared to the
same day of the same week in 2019 (pre-pandemic). The sample period is from February 19, 2020
to December 31, 2021. DINER data is from OpenTable.

IA.10
Figure IA.C.2: Climate Beta after Controlling for the number of air passengers The
sample banks are the top 10 largest US banks by average total assets in 2019. First, we regress
bank stock return on PASS. Second, we regress the residual from the first step on MKT and CF and
plot the coefficient on CF using 252-day rolling window regression. PASS is the daily percentage
change of the number of passengers on same day of the same week in 2020-22 compared to the
same day of the same week in 2019 (pre-pandemic). The sample period is from January 3, 2020 to
December 31, 2021. PASS data is from TSA.

IA.11
Figure IA.C.3: Climate Beta after Controlling for number of seated diners and air
passengers The sample banks are the top 10 largest US banks by average total assets in 2019.
First, we regress bank stock return on DINER and PASS. Second, we regress the residual from the
first step on MKT and CF and plot the coefficient on CF using 252-day rolling window regression.
DINER is the daily percentage change of the number of seated diners on same day of the same
week in 2020-22 compared to the same day of the same week in 2019 (pre-pandemic). PASS is
the daily percentage change of the number of passengers on same day of the same week in 2020-22
compared to the same day of the same week in 2019 (pre-pandemic). The sample period is from
February 19, 2020 to December 31, 2021. DINER data is from OpenTable and PASS data is from
TSA.

IA.12