NBER WORKING PAPER SERIES

CLIMATE REGULATORY RISK AND CORPORATE BONDS

Lee H. Seltzer
Laura Starks
Qifei Zhu

Working Paper 29994

http://www.nber.org/papers/w29994

NATIONAL BUREAU OF ECONOMIC RESEARCH

1050 Massachusetts Avenue
Cambridge, MA 02138
April 2022

The authors appreciate the comments of Sudheer Chava, Pierre Chollet, John Griffin, Matthew
Gustafson, Emir Ilhan, Pedro Matos, Quentin Moreau, Greg Niehaus, Zacharias Sautner, Greg
Weitzner, and seminar participants at the Georgia Tech Institute of Technology, AFA meetings,
the CFMR Conference, the ESSEC-Amundi Chair Webinar, the PRI conference, the London
Business School, the University of Alabama, the University of Bristol, the University of
Minnesota, the University of Ottawa, the University of South Carolina, the University of St.
Gallen, the University of Technology Sydney, the University of Texas at Austin and the
University of Washington. Laura Starks has served on the boards of directors for mutual funds
and retirement annuities and has occasionally consulted for financial institutions. The views
expressed in this paper are those of the authors and do not necessarily reflect the position of the
Federal Reserve Bank of New York or the Federal Reserve System. The views expressed herein
are those of the authors and do not necessarily reflect the views of the National Bureau of
Economic Research.

NBER working papers are circulated for discussion and comment purposes. They have not been
peer-reviewed or been subject to the review by the NBER Board of Directors that accompanies
official NBER publications.

© 2022 by Lee H. Seltzer, Laura Starks, and Qifei Zhu. All rights reserved. Short sections of text,
not to exceed two paragraphs, may be quoted without explicit permission provided that full
credit, including © notice, is given to the source.
Climate Regulatory Risk and Corporate Bonds
Lee H. Seltzer, Laura Starks, and Qifei Zhu
NBER Working Paper No. 29994
April 2022
JEL No. G12,G14,G23,G28

ABSTRACT

Investor concerns about climate and other environmental regulatory risks suggest that these risks
should affect corporate bond risk assessment and pricing. We test this hypothesis and find that
firms with poor environmental profiles or high carbon footprints tend to have lower credit ratings
and higher yield spreads, particularly when their facilities are located in states with stricter
regulatory enforcement. Using the Paris Agreement as a shock to expected climate risk
regulations, we provide evidence that climate regulatory risks causally affect bond credit ratings
and yield spreads. Accordingly, the composition of institutional ownership also changes after the
Agreement.

Lee H. Seltzer Qifei Zhu

Federal Reserve Bank of New York Nanyang Technological University
33 Liberty Street [email protected]
New York, NY 10045
[email protected]

Laura Starks
McCombs School of Business
University of Texas at Austin 2110
Speedway
Austin, TX 78712
and NBER
[email protected]
1. Introduction

In recent years investors have become more concerned about the environmental and

climate risks embedded in their portfolios.1 In fact, research shows climate risk to be an

important factor in institutional investors’ portfolio decisions as well as corporate man-

agers’ leverage decisions, and it has been found to affect the tail risk of stock returns and

the pricing of stocks and municipal bonds in the cross-section (Ginglinger and Moreau,

2019; Krüger, Sautner, and Starks, 2020; Ilhan, Sautner, and Vilkov, 2021; Bolton and

Kacperczyk, 2020, 2021; Painter, 2020). Of the three components of climate risk (physi-

cal, technological and regulatory), regulatory risk is the one that investors and others in

the finance community believe has the most immediate relevance (Krüger, Sautner, and

Starks, 2020; Stroebel and Wurgler, 2021). This suggests that regulation is a major channel

through which climate and other types of environmental risks get embedded in security

prices. In particular, environmental regulatory costs can have significant effects on firms’

operating costs and cash flows (Karpoff, Lott, and Wehrly, 2005). Moreover, uncertainty

about future regulation itself poses costs to firms and their investors (Pindyck, 1993).2

In this paper we address the issue of whether and how climate and other environmental

regulatory risks affect firms’ bonds. We test for the effects using firms’ bonds because,

as pointed out by Gourio (2013), for many corporations, the bond market, rather than

the equity market, is the marginal source of finance. Further, climate and environmental

risks are fundamentally downside risks for most firms.3 Thus, effects from climate risk

may be more easily captured in firms’ bond credit ratings and yield spreads than in their

equity prices. Moreover, a firm’s regulatory conditions can amplify or attenuate these risks.

Consequently, we examine whether climate risks are reflected in firms’ bond ratings and
1
See, for example, Shultz (2017), Furtado (2017), Frick (2020) and Arnold (2020).
2
In theoretical models such as Pastor and Veronesi (2013), political uncertainty regarding climate
regulations affects asset prices. Empirically, Kaviani, Kryzanowski, Maleki, and Savor (2020) find a strong
relation between policy uncertainty and corporate credit spreads when they employ the Baker, Bloom, and
Davis (2016) economic policy uncertainty index.
3
Bai, Bali, and Wen (2019) show that, in the cross-section, downside risk is the strongest predictor of
future bond returns.

1
pricing (i.e., yield spreads), and whether regulatory risks significantly influence how much

firms’ environmental profiles affect their bond ratings and yield spreads.

Testing these hypotheses requires measuring firms’ environmental profiles as well as

their regulatory risk exposures. We capture environmental profiles in two primary ways.

First, we employ an assessment of the firm’s environmental quality by a third-party environ-

mental, social and governance (ESG) rating agency (Sustainalytics). Second, we construct

multiple measures of firms’ carbon emissions using data from CDP (formerly known as the

Carbon Disclosure Project).

We estimate firms’ regulatory risk exposures based on the stringency of the state envi-

ronmental regulatory agencies overseeing the firms’ establishments. In the United States,

significant environmental legislation exists at the federal level with rule making by the

EPA. However, state governments generally hold the primary responsibility for enforcing

these laws and the states vary widely in their enforcement practices. Further, some states

also impose additional environmental restrictions beyond those required by the EPA. To

estimate a firm’s aggregate regulatory risk exposure, we construct a firm-level regulatory

stringency measure by aggregating across the geographical locations of the firms’ estab-

lishments.4 Thus, even when two firms have objectively similar levels of environmental

quality, depending on the local regulatory conditions of their facility locations, these firms

can face different degrees of regulatory risks.

In initial analyses, we examine whether bond credit ratings and yield spreads are asso-

ciated with firms’ environmental profiles, their aggregated regulatory risk exposures, and

the interaction between the two. Employing a sample of newly-issued corporate bonds, we

document important empirical relationships. First, we find a firm’s environmental profile,

whether measured by a third-party rating or through the firm’s carbon footprint, to be

unconditionally reflected in both its bond credit rating and yield spreads. Firms with a

lower environmental score, higher level of carbon emissions, or higher carbon intensity,
4
We have also constructed our tests using the state in which the firm’s headquarters reside and the
results hold.

2
i.e., carbon emissions scaled by firm revenue, exhibit lower credit ratings and higher yield

spreads, on average. This echoes previous findings in the equity market that carbon risk

is priced into average stock returns and the tail risk of stocks (Bolton and Kacperczyk,

2020, 2021; Ilhan, Sautner, and Vilkov, 2021). Second, and more importantly, there exists

a statistically and economically significant interaction effect on credit ratings and spreads

between a firm’s environmental profile and its regulatory environment. The lower credit

ratings (and higher yield spreads) for low environmental score firms and high-emission firms

are more pronounced if they operate in states where environmental regulations are enforced

more stringently. This suggests that regulatory risk is an important channel through which

firms’ environmental profiles affect their credit risks.

Recognizing the potential endogenous relationship between firms’ environmental profiles

and market participants’ perceptions of the firms’ risks, we consider a setting in which

expectations regarding future climate regulations receive an exogenous shock, namely the

December 2015 Paris Agreement, under which world governments agreed to take actions

to limit global temperature increases. When the Agreement was announced, a natural

implication for rating agencies and bond investors to draw was that governments—including

U.S. federal and state governments—would tighten their environmental regulations related

to the mitigation of climate change.5 In fact, consistent with this presumption, at least

one rating agency adjusted their baseline scenarios to include expectations of increased

regulations after the Paris Agreement.6 Survey results also suggest that firms upwardly

revised their beliefs about future regulation intensity in their disclosure to the CDP around

the time of the Agreement (Ramadorai and Zeni, 2021). This shock implies that U.S. firms

would face greater climate regulatory risk, especially those firms more exposed to this risk
5
The fact that so many nations would sign on to the Paris Agreement does not appear to have been
foreseen far in advance of the United Nations Climate Change Conference, which began on November 30,
2015. For example, a headline in a British newspaper on November 1, 2015 stated “Why climate treaty will
be the flop of the year.” In mid-November there still existed divisions among the world’s leading countries
regarding a deal. As late as November 23, the EU’s climate and energy czar warned that an agreement was
far from certain.
6
Moody’s Environmental Services June 28, 2016 report “Moody’s to Analyse Carbon Transition Risk
Based on Emissions Reduction Scenario Consistent with Paris Agreement.”

3
because of their business activities. The importance of this event is reflected in the fact that

it is the third highest spike in the Engle, Giglio, Kelly, Lee, and Stroebel (2020) climate

change news index, which the authors constructed over the 1984 to 2017 period to capture

innovations in climate change information.7

To test the hypothesis that the Paris Agreement had greater effects on the corporate

bonds more exposed to climate regulatory risks, we employ difference-in-differences anal-

yses of firms’ credit ratings and yield spreads in the months around the Agreement date.

The treated bonds are those issued by firms that prior to the Agreement have poor en-

vironmental scores, high carbon emissions, high carbon intensities, or that belong to a

top 15 carbon-emitting industry. Using bonds issued at least twelve months prior to the

Agreement and traded during the testing period, we find that after the Agreement, bonds

from the treated firms experience an average decrease in credit ratings of 0.48 to 0.63 notch

relative to bonds from other firms. The effect on ratings is consistent regardless of which

environmental definition we employ for the treatment. These results, which control for

time-invariant firm characteristics and macroeconomic trends, support the hypothesis that

changes in climate regulatory risk affect bond credit ratings for firms with more problem-

atic environmental performance, specifically including those with more significant carbon

footprints. Further, our evidence suggests that credit rating analysts consider expected

regulatory changes when evaluating climate risk effects on firms’ default risks.

We also find that the yield spreads of treated bonds increase significantly after the

Paris Agreement, suggesting that, besides credit rating analysts, bond investors also react

to potential future regulatory changes. For example, yield spreads for bonds issued by

firms that belong to top carbon-emitting industries increased by about 38 bps. Similarly,

bonds issued by firms with high total carbon emissions, high carbon intensity, or a low

environmental score also experience a significant increase in yield spreads after the Paris

Agreement.
7
See, Figure 2, p. 1193.

4
 Given that the expected tightening of environmental regulations following the Paris

Agreement would presumably be carried out under a state-enforcement regime, we hy-

pothesize that any effects on credit ratings and yield spreads should be stronger for issuers

operating in high-enforcement states. Consequently, we conduct a triple-difference analy-

sis in which we include an indicator variable for firms operating in states with relatively

more enforcement actions. The analysis indicates that following the Paris Agreement, the

changes in credit ratings and yield spreads for environmentally problematic firms are more

pronounced if a firm’s aggregated establishment locations are in states with stricter en-

forcement of environmental regulations. We find the effects to be both economically and

statistically significant. For example, bonds issued by firms in the top 15 carbon-emitting

industries that have facilities located in the stricter states experience an additional decrease

in their credit ratings of 1.09 notch.

The bond pricing results suggest that after the Paris Agreement investors should have

reevaluated their holdings in bonds more exposed to climate risk. Substantial theoretical

and empirical evidence provides evidence that various segments of the institutional investor

population employ differing investment strategies regarding ESG risks, including climate

risks.8 Consequently, we hypothesize that after the Paris Agreement reactions should differ

between the two major institutional investor types in the corporate bond market, mutual

funds and insurance companies, primarily due to the variations in their typical investment

horizons. Using difference-in-differences analyses with a matched sample, we find after

the Paris Agreement, insurance companies, which tend toward longer investment horizons,

lowered their holdings in the treated bonds. Mutual funds, which tend toward shorter

investment horizons, either kept the same holdings or increased their holdings, depending

on the definition of problematic environmental profile.

These results of investor holding changes along with our triple-difference analyses of
8
See, for example, Heinkel et al. (2001); Pastor et al. (2021); Pedersen et al. (2021); Oehmke and Opp
(2020); Baker et al. (2020); Goldstein et al. (2021); Dyck et al. (2019); Starks et al. (2021); Ilhan et al.
(2021).

5
the credit ratings and bond yield spreads all support the hypothesis that the expected

increases in climate regulatory requirements are associated with lower credit ratings and

higher yield spreads for issuers with poor environmental profiles. A natural follow-on ques-

tion is whether the changes in ratings and yield spreads subsequently reverse when the

political environment changes and the market then expects climate regulatory require-

ments to decrease. We test this proposition through the November 2016 presidential elec-

tion, with its unexpected change to a Republican administration, and the subsequent June

2017 announcement that the U.S. would withdraw from the Paris Agreement. While these

events suggest that the treated firms’ regulatory risks could lessen, there appears to have

been more uncertainty attached to the expected regulatory outcomes of these events than

to the original Paris Agreement.9 Thus, although the outcome of the 2016 election was

unexpected (Berlinger, 2016), and consequently, a shock to the regulatory setting for envi-

ronmentally problematic firms, there existed uncertainty regarding the extent of the shock.

Nevertheless, when we examine changes in credit ratings and yield spreads after both of

these events for firms with relatively poor environmental profiles, we observe a partial re-

bound of these firms’ yield spreads after the 2016 election. We also find suggestive evidence

on a reversal in credit ratings after the withdrawal from the Paris Agreement.

Our analysis and results contribute on a number of dimensions. First, we contribute to

the literature on the pricing of firm securities with respect to climate and environmental

risk. Our evidence that corporate bond investors demand higher interest rates from issuers

with poor environmental performance is consistent with earlier work on the higher cost of

bank loans for firms with poorer environmental performance (Chava, 2014), the relationship

between climate risk and municipal bonds (Painter, 2020), and evidence regarding carbon
9
For example, although during the campaign the winning candidate, Donald Trump, promised to loosen
environmental regulations, including a U.S. withdrawal from the Paris Agreement, whether and how these
goals would be achieved was uncertain. (See, for example, Parker and Davenport (2016)). Moreover,
even with the announcement, there still existed uncertainty given that the official withdrawal could not
occur until the day after the November 2020 presidential election and the winner of that election could
reverse the decision. After becoming the U.S. president, the winner of the 2020 election, Joseph Biden, did
indeed announce that the U.S. would abide by the Paris Agreement. See https://www.whitehouse.gov/
briefing-room/statements-releases/2021/01/20/paris-climate-agreement/

6
premia in equity markets (Bolton and Kacperczyk, 2020, 2021). Further, we provide a

mechanism through which climate and environmental risks affect security pricing: the

regulatory risks.10

Our paper also contributes to the literature on investor preferences for environmentally

friendly securities such as the work on the emerging importance of green bonds (Baker

et al., 2018; Flammer, 2021; Tang and Zhang, 2020; Zerbib, 2019), and the pricing effect of

ESG on sovereign bonds (Margaretic and Pouget, 2018; Capelle-Blancard et al., 2019).11

We contribute by showing that ratings and spreads for corporate bonds as well as the

institutional investor ownership of these bonds are affected by not only a firm’s environ-

mental activities but also their regulatory risk exposure.12 Similarly, our paper is related

to Jiraporn et al. (2014) and Amiraslani et al. (2021) in that we also examine the relation-

ship between some aspect of ESG and measures of bond risk and pricing. However, these

papers focus on the relation between firms’ CSR/ESG ratings and either credit ratings or

bond yield spreads. In contrast, our focus is on the regulatory risk aspect of the firms’

environmental qualities, and how the analyst and investor perceptions of this regulatory

risk are reflected in credit ratings and yield spreads. We also differ from Amiraslani et al.

(2021) in that we employ the Paris Agreement as an exogenous shock that could affect

bond pricing through the channel of regulatory risk, while their shock is the 2008 financial

crisis, which they hypothesize makes trust more important to investors.

Several papers use political changes regarding environmental issues to examine how

prospects for future governmental actions affect different aspects of firms’ actions, their

managers’ beliefs and investor expectations. Ramadorai and Zeni (2021) employ the CDP
10
Other research provides insights into the effects of physical climate risk such as Ginglinger and Moreau
(2019), the mixed evidence on the effects of physical climate risk on real estate values (Bernstein, Gustafson,
and Lewis, 2019; Murfin and Spiegel, 2020; Baldauf, Garlappi, and Yannelis, 2020) and the relationship
between derivative prices and climate models (Schlenker and Taylor, 2021). Huynh and Xia (2021) examine
the effect of climate news risk on corporate bond returns.
11
In addition, Fernando, Sharfman, and Uysal (2017) find that institutional investors avoid firms with
very poor environmental performance and Hong, Li, and Xu (2019) provide evidence of negative stock
return predictability of occurrence of droughts.
12
We also contribute to the literature on bond risk and pricing. For example, Bai, Bali, and Wen (2019)
and Bai, Bali, and Wen (2021) provide evidence on the relationship between bond risk and returns.

7
data of firms’ disclosures to develop a model of managers’ beliefs and actions regarding

climate regulation, particularly around the Paris Agreement and test it out-of-sample with

the U.S. withdrawal announcement. Our results regarding credit analysts and investors are

consistent with their findings regarding firms’ reactions. In terms of managerial actions,

Bartram, Hou, and Kim (2021) provide evidence that differences across states in their

regulatory risk can have real consequences due to firms’ regulatory arbitrage and Dai,

Duan, Liang, and Ng (2021) show that firms outsource their emissions due to increases

in state regulatory enforcements. Ginglinger and Moreau (2019) provide evidence that

firms reduced their leverage after the Paris Agreement driven by both their demand for

debt and lenders’ capital supply. Our paper is complementary in that we find decreased

credit ratings and higher yield spreads, suggesting that the Paris Agreement increased

costs of debt for the environmentally problematic firms that face more stringent regulatory

oversight. Our findings using corporate bonds are consistent with those of Ramelli, Wagner,

Zeckhauser, and Ziegler (2021) who examine stock market reactions to the 2016 and 2020

U.S. presidential elections, two events that reflect the changing political assessment of

environmental issues in the U.S.13

Similarly, our paper contributes to the research showing the relation between firms’

costs of debt and the liability and political uncertainty risks that they face (Gormley and

Matsa, 2011; Bradley et al., 2016; Kaviani et al., 2020; Ilhan et al., 2021). Our paper is

particularly complementary to that of Ilhan, Sautner, and Vilkov (2021), who examine the

effects of the Paris Agreement on firms’ tail risk by using out-of-the-money put options

on firms’ equity securities. They conclude that the Paris Agreement was followed by

significantly increased tail risk for the top polluting industry firms. Our study also focuses

on downside risks, but our interest is in rating agencies’ and bondholders’ perceptions and

actions, while their interest is in the equity holders’ perceptions and actions. The results
13
Our paper is also related to Child, Massoud, Schabus, and Zhou (2021) who analyze the effects of
firms associated with the Republican candidate, Li, Massa, Zhang, and Zhang (2021) who examine how a
change in environmental policy in China affected investor choices, and Kempf and Tsoutsoura (2021) who
examine how the 2016 U.S. presidential election affected credit ratings by partisan analysts.

8
between the two papers support the hypothesis that climate regulatory risk is an important

factor in the pricing of both equity and fixed income securities.

Finally, we contribute to the literature on the financial market responses to corporate

environmental news. Previous work has examined the stock market response. For example,

in examining CSR events, Krüger (2015) finds the strongest reactions occur for community

and environmental news. In addition, Karpoff, Lott, and Wehrly (2005) provide evidence

that a firm’s equity investors respond negatively to new information regarding EPA vi-

olations and that this response is tied to the expected legal penalties. Our focus is on

how credit rating analysts and bond investors respond to changes in perceptions of firms’

environmental regulatory risks.

2. Data

2.1. Sample construction

Our sample includes bonds issued by U.S. public non-financial companies over the 2009-

2017 period, which are classified as corporate debentures and corporate medium term notes

with maturities ranging from one month to 30 years.14 We obtain data on these bonds

and their issuing firms from a number of sources: Mergent FISD, Trade Reporting and

Compliance Engine (TRACE), CRSP, Compustat, Sustainalytics and CDP.

We use the Mergent FISD database for characteristics of the bonds such as offering

terms, maturity, the principal amount outstanding, and credit ratings (which originate

from Moody’s, Standard and Poor’s and Fitch). We employ the Moody’s ratings as the

primary source of credit ratings and transform the qualitative rating to a quantitative

measure by assigning each rating a numerical value, giving a 1 to the lowest rating (D) and

increasing by 1 for each notch such that the Moody’s Aaa rating (or the S&P and Fitch
14
We omit any non-standard corporate bonds such as Yankee bonds, convertible bonds, puttable bonds,
exchangeable bonds, Canadian bonds, bonds listed in foreign currency, private placements, variable rate
bonds and zero coupon bonds

9
equivalent) receives a value of 22.15 This approach has the advantage that when a credit

rating is downgraded, the representative number is lower.

Using the Mergent FISD offering terms, we define yield spread as the difference between

a bond’s offering yield and the yield of a cash flow-matched synthetic Treasury bond. In

this measure the discount rates of varying maturities derive from the U.S. Treasury yield

curve provided by Gürkaynak et al. (2007), where the yield of the synthetic Treasury bond

is inverted from its price.

We combine the Mergent FISD bond characteristics data with data on secondary market

pricing for corporate bonds from the TRACE database.16 We calculate a bond’s monthly

yield as the median yield on all trades of that security occurring on its last active-trading

day of a given month.17 When possible, we linearly interpolate yields for months with

missing yields. We then calculate trading yield spreads and the difference between a bond’s

trading yield and the yield of the Treasury bond with the same maturity in that month.

Data on characteristics of the issuing companies are obtained through the CRSP and

Compustat databases where we use the six-digit CUSIP to link companies across databases.

We drop observations for which we are unable to obtain information on either the firm’s

headquarters location or the SIC industry code.18

Our first measure of the issuing firm’s environmental profile relies on Sustainalytics En-

vironmental Scores from their ESG rating service, which during this period are based on

57 environmental indicators and range from 0-100, with a higher score indicating stronger

environmental performance. We employ the summary Environmental Score, which is cal-

culated as a weighted average of the indicators, where the weights used are industry specific
15
If Moody’s did not rate the security, we use the S&P rating and if that rating is also unavailable, we
employ the Fitch rating.
16
We adopt the procedure suggested in Dick-Nielsen (2009) to clean the TRACE data.
17
Based on the suggestion in Edwards, Harris, and Piwowar (2007), all trades that deviate from the
security daily median price by greater than 10% are dropped. Additionally, all price reversals greater than
10% are dropped.
18
Headquarters location and SIC industry code are obtained from the Compustat. Since Compustat
provides only current headquarter locations, we use historic headquarter locations provided by Gao (2020).
Only 0.2% of the bonds in the sample are dropped because of missing information.

10
and proprietary, that is, the environmental scores are industry adjusted. We merge the

corporate bond data with the Sustainalytics data at the issuer-year level using firm ticker

symbols.

We derive three additional measures of a firm’s environmental profile using firms’ carbon

emissions provided by CDP. Firms submit their carbon emissions data to CDP at the end of

June each year, covering emissions for the previous year. The data includes information on

Scope 1, Scope 2 and Scope 3 emissions, although not all firms that report to CDP provide

the Scope 2 and Scope 3 emissions. Scope 1 emissions are direct emissions produced by

the firm. Scope 2 emissions are indirect emissions from the generation of purchased energy.

Scope 3 emissions are other indirect emissions that occur due to the firm’s value chain.

We focus on Scope 1 emissions as the firm has the most direct control over this type of

emissions, and these emissions have the most precision.19 Using the Scope 1 emissions data,

we also calculate carbon intensity by dividing carbon emissions (in tons) by firm revenue

(in thousands of dollars). We employ both total carbon emissions and carbon intensity in

our tests.

Because not all firms submit their carbon emissions to CDP, we identify the highest

carbon emission industries in the sample and for the difference-in-differences tests we em-

ploy an additional measure according to a firm’s industry. Specifically, we rank industries

by total carbon emissions within our sample, and define the industries with the top 15

carbon emissions as top carbon emission industries. We employ total industry emissions

for this definition because political attention for climate regulations seems to focus on the

size of total emissions rather than the carbon intensity.20

19
In untabulated results we examine the sum of Scope 1 and Scope 2 emissions and our results are robust
to this change in variable.
20
These industries are electricity, gas and sanitary, oil and gas extraction, transportation by air,
petroleum and coal, chemical and allied products, primary metal, railroad transport, food and kindred
products, paper and allied products, motor freight transportation, metal mining, general merchandise
stores, stone, clay and glass, non-classifiable establishments, and transportation equipment.

11
2.2. Environmental regulations data

U.S. environmental policy is designed as a shared responsibility between the federal

government and the individual states—in general, federal environmental policies are es-

tablished through laws passed by Congress and rules developed by the EPA. According to

federal enforcement protocols, the individual states are authorized and expected to enforce

EPA regulations for violations within the state. Thus, for most states, state government

personnel evaluate compliance with the EPA regulations and issue enforcement actions if

they come to the conclusion that the compliance standards are not being met. In addition,

although states are allowed to create and enforce laws stricter than EPA regulations, they

are also expected to handle enforcement at least as strictly as EPA standards. Since some

states enforce regulations with the bare minimum standards and others enforce them more

stringently, this allows us to observe cross-sectional variation in regulatory standards.

We obtain EPA enforcement data from the Integrated Compliance Information System

for Federal Civil Enforcement Case Data. Employing this data we construct a measure of

state-level environmental regulatory stringency that captures compliance and enforcement

actions for the Clean Water Act (CWA), Clean Air Act (CAA) and Resource Conservation

and Recovery Act (RCRA) in a given state in a given year. Our measure, which we adopt

from the political science literature (Konisky, 2007), uses the number of enforcement ac-

tions, both informal enforcement actions (notifications of violation) and formal enforcement

actions resulting in a penalty for the firm (fines and administrative orders). We normalize

the number of enforcement actions by the total number of facilities that are subject to

EPA regulations in that state (measured in thousands), which is obtained from the Facility

Registry Services (FRS).21

Because firms often have facilities in multiple states, we adapt the state-level EPA mea-
21
If states fail to enforce regulations at the minimally acceptable level, the EPA has the op-
tion to enforce the laws themselves through their regional offices. States for which this is relevant
are detailed at https://www.epa.gov/compliance/state-review-framework-compliance-and-enforcement-
performance. Since we cannot observe whether the EPA or the state is the lead investigator on a given case,
we drop all enforcement actions occurring in the few states in which the EPA is responsible for enforcement.

12
sures to firm-specific measures to capture the regulatory environment for individual firms.

In order to determine each firm’s aggregate exposure to state-level EPA enforcement, we

use the National Establishment Time Series Database (NETs). The NETs is produced by

Wall & Associates based on the Dun & Bradstreet dollar-directory database, and provides

establishment-level information on firms, which we use to calculate each firm’s revenue

within each state in the United States. We then define the firm-level regulatory strin-

gency as the weighted-average state-level environmental regulatory stringency across all of

a firm’s establishments.22,23

X StateRevenuej,s
RegStringencyj,t = ( × EP AEnf orcementss,t ), (1)
T otalRevenuej
s∈Sj

where T otalRevenuej is total revenue by firm j in all states, StateRevenuej,s are total

revenue by firm j in state s and EP AEnf orcementss,t are total EPA enforcement actions

in state s scaled by the number of EPA-registered facilities (in thousands) in state s at

time t. Therefore, RegStringencyj,t captures firm j’s exposure to environmental regulatory

enforcement at time t across the states within which it operates.

2.3. Summary statistics

Our initial data set covers 5,548 bonds and 830 issuers contained in Mergent and

TRACE databases over the 2009-2017 sample period. After merging the data with Sus-

tainalytics, the sample size reduces to 4,235 securities from 478 firms. For the tests that

employ the CDP carbon emissions data, we have 3,368 corporate bonds, corresponding to

287 firms. In Table 1 we report the sample summary statistics. Panel A contains summary

statistics for the at-issue bonds. The average bond in the sample has a credit rating of

about 15.3 (which is a little higher than a Baa1 rating) and an offering yield spread of
22
For firms for which we cannot observe establishments in the NETs data, we use the total number of
EPA enforcement actions for the state in which the firm’s headquarters are located.
23
The NETS data we employ are available as of 2014.

13
1.84%, and an average maturity of about 10 years. The average Environmental Score for

the bond issuers is 60, above the halfway point of the 0-100 range. 48.7% of bond issues

in the sample are from top 15 emissions industries.

The average issuing firm’s carbon emissions is 6.68 million tons, but the median is only

0.44. The average carbon intensity is 0.32 tons of emissions per $1,000 in revenue, but the

median is 0.01. These statistics reflect the fact that both carbon emissions and carbon

intensity are highly positively skewed. Finally, the average Regj,t for sample bond issuers

is 0.71, indicating that, on average, 0.71 facility receives an enforcement action for every

1,000 facilities located in the same states as the issuer.

3. Credit risk, environmental profile, and regulatory

stringency

3.1. Regression specifications

We first examine the relationship between bond credit risks and the issuing firms’

environmental profiles, and whether that relationship is heightened by the firms’ exposures

to differing regulatory risks across states. In this set of analyses, we employ bond credit

ratings and offering yield spreads as separate dependent variables that capture credit risk.

The key independent variables in these regressions are firms’ environmental profiles, the

level of regulatory enforcement intensity each firm faces, and the interactions between these

two variables. The specifications for bond i issued by firm j at time t are as follows:

Ratingit = β1 EnvP rofjt−1 + β2 Regjt−1 + β3 EnvP rofjt−1 × Regjt−1 + β4 Xjt−1 + F E + it ,

(2)

Spreadit = β1 EnvP rofjt−1 + β2 Regjt−1 + β3 EnvP rofjt−1 × Regjt−1 + β4 Xjt−1 + β5 Zit

+ F E + it , (3)

14
where EnvP rofjt−1 is firm j ’s environmental profile at time t−1, which we proxy for using

the Sustainalytics Environmental Score, the firm’s total carbon emissions (in millions of

tons), or the firm’s carbon intensity (tons of emissions divided by revenue in thousands of

dollars). Regjt−1 is the regulatory stringency for firm j at time t−1, proxied by the revenue-

weighted average state EPA enforcement intensity. Xjt−1 are firm j’s characteristics at

time t − 1 and Zit are bond i ’s characteristics at time t. Firm characteristics include book

leverage, pre-tax interest coverage, the natural log of total assets, cash-to-assets ratio,

profitability, tangibility of assets, annual stock returns and standard deviation of stock

returns. In the regression on ratings, we control for a firm’s weighted-average maturity of

all outstanding bonds at time t. In the regression on yields, we additionally control for bond

characteristics such as principal amount, time to maturity, and an indicator for callable

bonds. For all specifications, time fixed effects are used to control for macroeconomic

trends. In some specifications, industry fixed effects are also included to control for time-

invariant industry characteristics.

In Equations (2) and (3), the primary coefficient of interest is β3 , which captures the

interaction effect between firms’ environmental profiles and their regulatory conditions.

Based on our hypothesis that firms with poor environmental scores tend to have higher

regulatory risk exposure, we expect β3 to be positive when credit ratings are the dependent

variable. That is, conditional on a certain level of regulation, higher environmental scores

imply the firm should be subject to less regulatory risk, and therefore higher credit ratings.

Alternatively, when we employ the carbon emission measures in Equation (2), we expect

β3 to be negative since under conditions of stricter regulatory enforcement, higher carbon

emissions imply more regulatory risk exposure, and therefore, lower credit ratings.

When we employ the bond’s offering yield spread as the dependent variable in Equation

(3), we expect the opposite signs on β3 because yield spreads should be decreasing in

regulatory risk exposure for higher environmental profile firms and increasing for high

carbon emission firms.

15
 We are also interested in the coefficient of EnvP rof , β1 , as it captures the unconditional

effect of a firm’s environmental profile on its credit risk. Recent studies have shown that

carbon risk seems to command a positive risk premium in the equity market (Bolton and

Kacperczyk, 2020, 2021). If bond investors also care about carbon risk, we would expect

credit ratings and yield spreads to differ across issuers with different environmental profiles,

even for issuers exposed to average levels of regulatory risk.

In this set of analyses, we focus on at-issue bonds to better capture the relation between

environmental regulatory risk exposure and firms’ costs of debt because the offering spreads

reflect the costs of issuing debt. Additionally, at-issue bond spreads are less noisy than

trading yields, particularly given the general illiquidity of the secondary bond market.

3.2. Results

In Table 2 Columns (1) through (3), we report the results for the regressions in which

credit ratings are the dependent variables and we use time fixed effects. In column (1),

using the firms’ Sustainalytics environmental scores, we find that bonds issued by firms

with higher environmental scores tend to have higher ratings. In particular, an increase

in a firm’s environmental score of one point is associated with a statistically significant

0.027 notch increase in credit rating for firm-years with an average regulatory stringency.24

Importantly, the interaction term between a firm’s environmental score and the weighted-

average regulatory stringency the firm faces is positive, and both economically and statis-

tically significant. When an increase in a firm’s environmental score of one is combined

with a one standard deviation increase in the firm’s regulatory stringency, ratings increase

by 0.047 notches (0.027 + 0.020). The results suggest that a firm’s environmental profile

affects its credit rating particularly through the channel of regulatory risks.

In the next two columns we employ the firm’s carbon emissions in tons (Column (2)) and

carbon intensity (Column (3)). Although we do not find a statistically significant relation
24
The standard deviation of the environmental score is 14.1.

16
employing the absolute value of carbon emissions, when we use the relative measure, carbon

intensity, we find a strong negative effect on the firm’s credit rating. If carbon intensity

increases by one (ton per $1,000 revenue), a firm’s credit rating decreases by 0.514 notch.

This result suggests that carbon risk is priced in corporate bond spreads unconditionally.

Moreover, when combined with a one standard deviation increase in Reg, the same increase

in carbon intensity is associated with a 0.797 notch decrease in credit ratings (−0.514 −

0.283).

We include industry fixed effects in the regression-specifications in columns (4) through

(6) of Table 2 in order to examine whether the relationship between firm environmental

profile and credit risk is also present within a given industry. Column (4), in which we

use the firm’s environmental score, shows results similar to column (1), an increase in a

firm’s environmental score of one is associated with a 0.014 notch increase in their bonds’

credit ratings. Moreover, when combined with a one standard deviation increase in firms’

regulatory stringency, the same magnitude increase in environmental score is associated

with an even larger increase in credit ratings: a 0.033 notch increase (0.014 + 0.019).

In Column (5) using industry fixed effects, we now find a statistically significant positive

relation between a firm’s total carbon emissions and its credit ratings. In particular, a one

million ton increase in a firm’s credit ratings is associated with a 0.014 notch decrease in its

bond ratings. Moreover, if a firm operates in states with a one standard deviation increase

in regulatory stringency, the same amount of additional carbon emissions is associated

with a 0.035 notch decrease (−0.014 − 0.021) in credit ratings. Column (6) displays results

using carbon intensity. An increase in carbon intensity of one (ton per $1,000 revenue) is

associated with a 0.231 decrease in credit ratings. When combined with a one standard

deviation increase in Reg, it is instead associated with a 0.559 decrease (−0.231 − 0.328)

in credit ratings.25

The results in Table 2 imply that credit rating agencies consider regulatory risk when
25
In order to save space, we do not report the results for the case when a firm is a member of a top 15
carbon-emitting industry as the coefficient is not significantly different from zero for these tests.

17
evaluating how environmental concerns affect bond risk, which is consistent with rating

agencies’ policies. According to methodology published in 2018, credit rating analysts at

Moody’s consider both direct environmental implications and regulatory costs when eval-

uating ESG effects on credit ratings. Specifically, they state that they consider regulation

more closely because forecasting is easier (Moody’s, 2018). These statements are consis-

tent with our finding that the effects of detrimental environmental activities on bond credit

ratings are sensitive to the strictness of states’ EPA regulation enforcement.

In Table 3, we present the regression results for the relationship between firms’ offering

bond spreads and their environmental risk exposures, where the regressions in columns

(1) through (3) include results using only time fixed effects and columns (4) through (6)

include both time and industry fixed effects. The results in column (1) indicate that a one

unit increase in a firm’s environmental score is associated with a 0.9 bp decrease in their

bonds’ offering yield spreads, holding the regulatory stringency a firm faces at the average

level. Additionally, when a firm operates in states with a one standard deviation increase

in EPA stringency, the same increase in environmental score is instead associated with

a 1.4 bps decrease (−0.9 − 0.5) in offering yield spreads. Considering that the standard

deviation of environmental score is 14.1, this effect is economically large. Our finding that

firms with higher environmental scores have lower yield spreads is consistent with Chava

(2014) who concludes that firms with higher environmental scores pay lower interest rates

on their bank loans. Both our results and that of Chava (2014) imply that such firms face

lower risks, which is an effect widely believed by many ESG investors.

The results in column (2) show that a one million ton increase in firm carbon emissions

is associated with a 0.7 bp increase in yield spreads. When combined with a one standard

deviation increase in Reg, the increase in emissions is associated with a 1.4 bps increase

(0.7 + 0.7). Column (3) shows similar results when using carbon intensity as a carbon

intensity increase of one (ton per $1,000 revenue) is associated with a 16.2 bps increase

in bond spreads. These results are consistent with the argument that issuers with higher

18
carbon emissions face higher costs in raising capital.

In columns (4) through (6), we report broadly similar results when employing both time

and industry fixed effects. All three columns show that when combined with an increase

in environmental scores, carbon emissions or carbon intensity, the offering yield spreads

become higher. For example, column (5) show that a one million ton increase in carbon

emissions, is associated with 0.9 bp higher spread for firms located in states where the EPA

enforcement stringency is one standard deviation higher than average.

The results in Tables 2 and 3 show that bonds from firms with poor environmental

performance tend to have both lower credit ratings and higher yield spreads. These results

from the corporate bond markets are consistent with findings that carbon risk is priced

into equity markets (Bolton and Kacperczyk, 2020, 2021). Importantly, these findings

highlight that the effect is particularly pronounced when issuing firms have establishments

in states with more stringent environmental regulation enforcement. The results imply that

these firms face a higher probability of regulatory costs such as fines or possibly reputation

losses, which in turn increases their credit risk.26 The results, which are also consistent

with previous research showing the greater negative consequences for firms that pollute

under stricter regulatory regimes, imply strictness in regulation forces firms to internalize

the costs of pollution (Greenstone, 2002).

The results are also informative about the channel through which environmental reg-

ulations affect credit ratings and yield spreads. In particular, these findings show that

environmental regulatory risk is related to credit ratings and yields spreads both across-

industry and within-industry, which suggests that for the high carbon emission sectors,

firms that operate in states with stricter regulatory enforcement likely bear even more

credit risk than their counterparts in less stringent states.

26
These results are consistent with the legal cost evidence provided by Karpoff et al. (2005). Although
the authors of this article conclude there exist no reputational losses for EPA violations during their sam-
ple period, investors, including institutional investors, have become much more concerned about firms’
environmental activities over the approximately two decades since their sample ended.

19
4. The Paris Agreement shock

A firm’s environmental profile and its regulatory conditions may be jointly determined,

thus, creating potential endogeneity issues. For example, state governments could impose

stricter environmental regulations because the economic conditions in the state are favor-

able; these favorable economic conditions might in turn attract high carbon emission firms

to locate there. To mitigate such endogeneity concerns, we exploit a shock that increases

the climate regulatory risks faced by firms, while changing neither the performance nor the

environmental profiles of these firms.

The setting we use in our research design is the passage of the Paris Agreement, an-

nounced on December 12, 2015. The Paris Agreement had the primary goal of limiting

global temperature rise in this century to 1.5 degrees Celsius above pre-industrial levels.

To achieve this goal, the Agreement calls for the signing countries to submit national action

plans to reduce emissions with sufficient speed to achieve the goal. Such plans imply the

development of expectations that more stringent environmental regulations are likely to

be imposed in the future, since the national action plans would need to include regulatory

responses to induce firms to help achieve the climate goal. We hypothesize that the Paris

Agreement shock resulted in firms with poor environmental practices or high carbon foot-

prints facing greater climate regulatory risks relative to other firms. The increased risks

should then be reflected in changes in firms’ bond credit ratings and spreads.

To test this hypothesis, we conduct difference-in-differences analyses so as to compare

changes in the credit ratings and yield spreads of bonds from firms with problematic envi-

ronmental profiles versus those from other firms, both before and after the Paris Agreement.

4.1. Descriptive evidence of changes in bond credit ratings and spreads

Before conducting formal difference-in-differences analyses, we visually inspect changes

in the average credit ratings and spreads for bonds issued by firms in high carbon-emitting

20
industries and by firms with different environmental scores. Figure 1(a) displays the av-

erage credit ratings for each of the top 15 carbon-emitting industries before and after the

December 2015 Paris Agreement. It should be noted that unconditionally, there exists sig-

nificant variation in the credit ratings across industries. Issuers in some industries such as

petroleum and coal products or motor freight transportation tend to be more creditworthy,

on average, while other high carbon emission industries, such as stone, clay and glass or

metal mining, appear to be less creditworthy, on average.

Overall, the figure demonstrates observable differences in average industry credit ratings

across the two periods, which supports the hypothesis that an association exists between the

Paris Agreement event and a ratings decrease for firms in high carbon emission industries.

In particular, some of the high-carbon emitting industries whose ratings are most affected

by the announcement of the Paris Agreement (such as primary metal and metal mining)

are relatively less sensitive to oil prices, suggesting that the effect is unlikely driven by any

concurrent changes in oil prices (an issue we examine in more depth later in the paper).

Figure 1(b) shows changes in yield spreads for bonds issued by firms in the top 15

carbon-emitting industries before and after the Paris Agreement. As in the case of the

credit ratings, substantial differences exist across industries in the magnitude of the yield

spreads and their changes. Nonetheless, in most cases, large increases in spreads occurred

after the Agreement with the largest increases in primary metals, water transport, oil and

gas extraction and metal mining industries.

In Figures 2(a) and 2(b), we illustrate average credit ratings and yield spreads by

environmental score bins for the periods before and after the Paris Agreement. We first sort

bonds by their issuer’s December 2014 Sustainalytics Environmental Score into eight groups

and compare credit ratings and bond spreads across the environmental score bins.27 Figure

2(a) contains the distribution of the average credit ratings according to environmental score

bin from the lowest environmental scores on the left and the highest environmental scores
27
Note the Sustaintalytics data does not include any United States firms with environmental scores
below 20.

21
on the right. The figure clearly demonstrates that, unconditionally, bond issues from

firms with lower environmental scores tend to have lower average credit ratings, consistent

with evidence in Chava (2014) regarding bank loans. More importantly, after the Paris

Agreement, the average credit ratings decrease substantially for bonds from firms with

lower environmental scores, whereas there appears to be little effect for bonds from firms

with higher environmental scores.

Figure 2(b) displays the average yield spreads for bonds across the environmental score

bins. Again, the figure shows dramatic changes around the Paris Agreement for issues

from firms with environmental scores below 60, and very little change for those with envi-

ronmental scores above 60. Since our sample median environmental score is 60, corporate

bond issues with above-median environmental scores tend to have minimal changes in bond

spreads after the Paris Agreement while those with below-median scores have increases in

their spreads. These figures are consistent with our hypothesis that the Paris Agreement

led to perceptions of increased regulatory risk resulting in lower credit ratings and higher

bond yield spreads, on average, for firms with low environmental scores. We next conduct

more formal tests of whether the Paris Agreement announcement had causal effects on the

corporate bond market.

4.2. Tests for changes in credit ratings around the Paris Agreement

We first test changes in bond credit ratings in the two-year period around the December

2015 Paris Agreement through the following difference-in-differences regression:

Ratingit = β1 EnvP rofj × Af terP arist + γi + κt + it , (4)

where Af terP arist is an indicator variable for the months starting in December 2015 and

continuing through the following 12 months. We include time fixed effects, κt , and in some

specifications, security fixed effects, γi . Since the Paris Agreement is a time-series shock,

22
our sample in these tests consists of bonds issued before the Paris Agreement in order to

capture changes in ratings influenced by the Agreement.28 In constructing our test sample,

we include a pre-event period of twelve months prior to the Agreement and a post-event

period of twelve months following the Agreement. That is, the testing period runs from

December 2014 through November 2016.

We employ four measures of a firm’s environmental profile. First, we use an indicator

variable equal to one if a firm has a below-median environmental score in December 2014.

Second, we use an indicator variable for whether a firm is in one of the top 15 carbon-

emitting industries. Third, we use an indicator variable equal to one if a firm is in the

top-quartile in terms of firm-level total carbon emissions in 2014. Finally, we use an

indicator variable equal to one if a firm is in the top-quartile in terms of firm-level carbon

intensity in 2014.29

In Equation (4) β1 captures the change in bond risk assessments around the Paris

Agreement for a firm with a poor environmental profile relative to other firms, controlling

for time-invariant bond characteristics and for macroeconomic trends that affect all bond

issues. We cluster standard errors at the issuer-level to account for correlated error terms

within firm. However, the results are robust to clustering at the industry or state-level as

well.

Table 4 reports the results of the difference-in-differences regressions. Column (1) shows

that after the Paris Agreement, bond ratings decreased by 0.58 notch for bonds issued by

firms with below-median environmental scores relative to other firms. Based on this result,

the Paris Agreement led to an economically significant decrease in bond ratings for less

environmentally friendly firms relative to firms that are more environmentally friendly.

In columns (2), (3) and (4), we use the CDP data to examine whether bonds issued by
28
Specifically, the sample includes bonds issued at least one year before the Paris Agreement, i.e., in
December 2014 or earlier. We also require that the bond does not mature before the end of the sample
period.
29
Below-median environmental score is based on the median environmental score for the firm-level dis-
tribution of Sustainalytics scores in December 2014, which is 55, results are similar if we instead use the
sample median based on the full sample.

23
firms exposed to climate risk through their carbon emissions experienced ratings decreases.

The results show that after the Paris Agreement, relative to other firms’ bonds, the bond

ratings decreased by 0.48 notch for issuing firms that belong to high emission industries,

by 0.55 notch for bonds issued by firms in the top-quartile of carbon emissions and by

0.63 notch for bonds with top-quartile carbon intensity. Regardless of the measure used,

corporate bond ratings decreased after the Paris Agreement for bonds issued by firms

exposed to more climate risk relative to bonds from other issuers. Moreover, as shown

by the results using the environmental score in the first column, bond ratings decreased

after the Paris Agreement for bonds issued by firms exposed to environmental risk more

generally.

To check the parallel trends assumption, we examine the dynamics of the treatment

effects in relation to the Paris Agreement event. Specifically, we construct a series of tests

to examine the time series of differences between ratings for firms in the treatment and the

control groups. To do so we run the following regressions:

11
βk [1(t = k) × EnvP rofj ] + γi + κt + it ,
X
Ratingit = (5)
k=−11

where 1(t = k) are indicators for periods that are k months after (or before) the Paris
Agreement. The time indicator variable for the first month in our sample period (December

2014) is excluded, so all treatment effects are relative to December 2014.

Figure 3 shows the results for these regressions when the outcome variable is the bond

credit rating. The figure shows the treatment effect for each month in our sample, allowing

us to examine the effects before and after the Agreement. Panel (a) displays the treatment

effects over time for bonds issued by firms with a below-median environmental score. The

solid line and dots indicate the coefficient estimates, and the dashed lines represent bands of

a 95% confidence interval around these estimates. We find no significant differences in the

treatment effect in the entire period before the Agreement, indicating the parallel trends

24
assumption appears to hold. In contrast, after the Agreement, the treated firms’ bonds

have significantly lower credit ratings, consistent with the results reported in Table 4.

Figures 3 (b), (c) and (d) illustrate the results for Equation (5) when the treated

firms are in a top 15 carbon emission industry, produced top-quartile emissions or have

top-quartile carbon intensity, respectively. All of the figures show credit ratings indistin-

guishable from other firms’ bonds before the Paris Agreement, and significantly lower credit

ratings for the treatment bonds after the Paris Agreement. These charts together provide

strong evidence that the parallel trends assumption is likely satisfied in addition to showing

clearly observable changes in ratings for the treated bonds after the Paris Agreement.

The findings in this section imply a direct consequence of the Paris Agreement for firms

with problematic environmental profiles. In particular, they provide evidence that credit

rating agency analysts appear concerned about future regulatory changes when evaluating

the effects of environmental risk on a bond’s default risk.

4.3. Tests for changes in bond yield spreads around the Paris Agreement

To test for changes in bond yield spreads around the Paris Agreement we use the

following regression:

Spreadit = β1 EnvP rofj × Af terP arist + γi + κtp + it , (6)

where we measure a firm’s environmental profile using the same four measures as in Equa-

tion (4). Instead of employing a time fixed effect, we include a matched-pair-by-time fixed

effect κtp , where the matching procedure is described in more detail below. As a result,

this test can be interpreted as comparing the change in spread for a treated security to its

matched control security after the Paris Agreement, controlling for time-invariant security

characteristics.

To better control for noise in spreads and to compare bonds with similar creditworthi-

25
ness, we conduct a one-to-one Mahalanobis matching with replacement. The purpose of

this matching approach is to identify and match every treated bond to the most similar

control bond according to various covariates.30 This distance is calculated as of year-end

2014 using the bond’s credit rating, the bond principal outstanding, the bond’s time to

maturity, and the firm’s equity oil beta.

We believe it is particularly important to match on oil beta in order to alleviate a

potential concern that changes in bond pricing may be driven by concurrent movements in

the oil market, particularly given the volatile changes in oil prices over this period.31 We

use the following model to calculate firms’ equity oil betas:

Rit = α + βmarket M ktRett + βoil OilRett , (7)

where M ktRet is proxied by the CRSP value-weighted index and OilRett is the monthly

return on Brent Crude Oil for month t. We calculate this value for each firm in our sample

for which we observe 36 months or more of stock price data before November 2015.

We construct four matched samples, one for each of our environmental measures, the

below-median environmental score, the top carbon emission industry, the top emission

quartile and the top carbon intensity quartile treatments, respectively. In Table 5 we

report the summary statistics for all matched samples (as of the matching date). Pan-

els A, B, C, and D report statistics for the control and treated groups matched on the

below-median environmental score indicator, the high emissions industry indicator, the

top-quartile emissions indicator and the top-quartile carbon intensity indicator, respec-

tively. The last column of each panel provides difference-in-means tests between the treated

and control groups. As the differences between these two groups are generally statistically
30
We use a caliper of 1, meaning if for a given treatment firm there does not exist a control firm whose
Mahalanobis distance is 1 or less, we drop it from the sample. We further address the potential bias in
continuous variable matching using the methods proposed by Abadie and Imbens (2006).
31
Generally speaking, oil price volatility is seen as negatively predicting economic growth and aggregate
equity prices, especially for the oil sector (Gao et al., 2021).

26
insignificant and economically small, it is reasonable to conclude the treated and control

groups are observationally similar.

Table 6 reports the results from the difference-in-differences regression in which bond

spread is the dependent variable. The effects of the Paris Agreement on the treated firms’

spreads are both economically large and statistically significant. Column (1) indicates that

after the Paris Agreement bond spreads increased by 39.4 bps for bonds issued by firms

with below-median environmental scores relative to other firms. Columns (2) through (4)

display results using the carbon emissions measures. Relative to other firms, corporate

bond spreads increased by 38.6 bps, 30.1 bps, and 34.7 bps for bonds issued by firms in

industries with high carbon emissions, by firms with top-quartile carbon emissions and by

firms with top-quartile carbon intensity, respectively.32 These results provide evidence that

regardless of the specific firm environmental profile measure used, after the Paris Agreement

corporate bond spreads increased for bonds issued by firms with poor environmental profiles

relative to other firms.

We further provide visual evidence for the parallel trend assumption by running the

following dynamic difference-in-differences regression:

11
βk [1(t = k) × EnvP rofj ] + γi + κtp + it .
X
Spreadit = (8)
k=−11

The excluded period is December 2014. Additionally, we use security and matched-

pair-by-time fixed effects.

Figure 4(a) illustrates the changes in bond spreads around the Paris Agreement using

a below-median environmental score indicator as the treatment. Prior to the Paris Agree-

ment, there does not appear to be a substantial differential increase in bond spreads for

issuers with low environmental scores relative to other issuers. Additionally, right after

the announcement of the Paris Agreement, there exists a significant and sizable increase in
32
In untabulated results we also conducted the analysis using the full unmatched sample, and the results
not only continue to hold, but they are also larger in magnitude.

27
spreads for bonds issued by firms with below-median environmental scores relative to other

bonds. Similar patterns are observed when examining high carbon emission industries in

Figure 4(b), high carbon emission firms in Figure 4(c), and high carbon intensity firms in

Figure 4(d). These results provide some assurance that the parallel trends assumption for

our difference-in-differences specifications are likely to be satisfied.

The initial increase in yield spreads for firms with low environmental profiles some-

what subsides in later months. One explanation for the partial reversals is the uncertainty

surrounding what policy changes would ultimately result from the Agreement. This uncer-

tainty arose because of the 2016 presidential election campaign and outcome as well as the

subsequent June 2017 announcement that the U.S. would withdraw from the Agreement.

Thus, although the Paris Agreement appears to have resulted in a persistent increase in

credit ratings for treated firms as shown in Figure 3, Figure 4 suggest that the change

in investors’ expectations of increased climate change regulation in the United States as

shown by the change in spread may have been short-lived for investors. We examine this

possibility more thoroughly in section 6.

The substantial increases in bond spreads for firms with poor environmental profiles

support the hypothesis that investors develop expectations that these firms would soon need

to abide by new regulations, which increases their climate regulatory risk. The increase in

climate regulatory risk leads to increases in bond spreads and thus, these firms’ cost of debt

rises relative to that of more environmentally friendly firms. These results are consistent

with other research showing that environmental policies are related to firms’ costs of debt

(Chava, 2014), that the Paris Agreement increased perceptions of downside risk (Ilhan

et al., 2021), that the Paris Agreement changed firms’ leverage (Ginglinger and Moreau,

2019) and that firms revised their beliefs over the effects of climate regulation upward,

sharply increasing their carbon abatement over the year from 2016 to 2017 (Ramadorai

and Zeni, 2021). The result is also consistent with previous literature showing that firms’

cost of debt increases with political uncertainty risk and increased liability risk (Bradley

28
et al., 2016; Gormley and Matsa, 2011).

4.4. Triple-difference tests around the Paris Agreement

While the Paris Agreement increased the prospect of future environmental regulatory

risks, we expect its effects to differ across companies in part due to variations across state

governments in their enforcement of environmental regulations. In a scenario in which the

U.S. government imposes new environmental regulation at the federal level, we hypoth-

esize that firms located in high-enforcement states would be more affected because the

regulatory environments in these states would be more likely to impose stricter regula-

tory requirements. To examine this hypothesis, we conduct a triple-difference regression

in which we include an indicator variable for firms with stricter regulatory environments.

To define the stricter regulatory environments, we sort firms by Reg, which is calculated

as firm’s revenue-weighted average environmental regulatory strigency, from 2012 through

2015 (the four years leading up to the Paris Agreement). Firms with a top-quartile Reg

are defined as high regulatory enforcement firms.

Using these definitions, we run the following analyses:

Ratingit = γi + κt + β1 Af terP arist × EnvP rofj + β2 Af terP arist × HighRegj (9)

+ β3 EnvP rofj × HighRegj + β4 Af terP arist × EnvP rofj × HighRegj + it ,

Spreadit = γi + κtp + β1 Af terP arist × EnvP rofj + β2 Af terP arist × HighRegj (10)

+ β3 EnvP rofj × HighRegj + β4 Af terP arist × EnvP rofj × HighRegj + it ,

where Af terP arist is an indicator variable for the months starting in December 2015 and

continuing through the following 12 months and HighRegj is an indicator variable for high

regulatory enforcement firms.

29
 The primary parameter of interest, β4 , captures the effects of the Paris Agreement for

firms with poor environmental profiles that operate in states with strict regulatory enforce-

ment relative to firms that operate in less stringent states. If after the Paris Agreement

firms with poor environmental profiles become more exposed to climate regulatory risks in

states where any potential new regulations are expected to be enforced more strictly, we

expect β4 to be negative in the credit rating regressions and positive in the yield spread

regressions. Such a result would suggest that regulatory risk is the channel through which

the Paris Agreement affects bond credit ratings and spreads. We again use the below-

median environmental score, the high carbon emissions industry, the top-quartile firm

carbon emissions and the top-quartile firm carbon intensity indicators to define treated

firms.

Table 7 provides the results of the triple-difference regressions where the dependent

variable is credit rating. The main parameter of interest is the coefficient for the triple-

difference estimator Af terP arist × EnvP rofj × HighRegj . In Column (1) in which the

environmental measure is the below-median environmental score indicator, the results show

that after the Paris Agreement, relative to the firms located in low regulatory stringency

states, firms with low environmental scores located in strict regulatory states experienced

credit rating decreases of an additional 0.99 notch. Columns (2) through (4) display results

using the carbon emissions data and show that after the Paris Agreement, if an issuing

company is located in a high regulatory enforcement state, bond ratings decrease by an

additional 1.09 notch, 1.37 notch, and 1.39 notch for bonds issued by firms in high carbon

emission industries, the top-quartile of carbon emissions and the top-quartile of carbon

intensity, respectively. Overall, the results imply that the decrease in corporate bond

ratings following the Paris Agreement is driven by firms with operations in states that

have stricter regulatory enforcement, supporting the hypothesis that climate regulatory

risk is an important determinant of corporate bond ratings.

Table 8 reports the results for the triple-difference tests when bond spreads are the

30
dependent variables. Column (1) displays results using the below-median environmental

score indicator. Bond spreads increase by an additional 91.1 bps for firms with poor

environmental profiles and located in stricter regulatory enforcement states, as compared

with poor-environmental firms located in less strict states. Column (2) displays results

for the high carbon emission industry indicator and the results are similar. Bond spreads

increase by an additional 70 bps for bonds issued by firms in high carbon emission industries

if the firm is located in stricter regulatory enforcement states. These results imply that

changes in corporate bond spreads for firms with poor environmental profiles after the

Paris Agreement are mostly driven by regulatory risk. However, the results in columns (3)

and (4), which display results using the firm emissions and carbon intensity measures, are

not statistically significant.

The triple-difference results indicate that most of the effect of the Paris Agreement

on firms’ cost of debt arises through the regulatory cost channel, which suggests that

both credit rating analysts and bond investors believe that the Paris Agreement would

have greater effects on the cost of debt for issuers located in high-regulation states. These

results are consistent with the hypothesis that bond market participants expected the Paris

Agreement to lead to increased regulations for environmentally problematic firms and that

the new regulations would most likely be enforced through the state governmental agencies.

These results are also consistent with previous research documenting that firms face costs

due to environmental regulations e.g., Karpoff et al. (2005).

5. Changes in institutional investor bond ownership

around the Paris Agreement

The bond pricing results suggest that after the Paris Agreement, investors reevaluated

their corporate bond holdings more exposed to climate risk. A number of recent theory

papers argue that green and brown investors view their investments from different per-

31
spectives (e.g., Heinkel, Kraus, and Zechner, 2001; Pastor, Stambaugh, and Taylor, 2021;

Oehmke and Opp, 2020; Baker, Hollifield, and Osambela, 2020; Pedersen, Fitzgibbons,

and Pomorski, 2021; Baker, Hollifield, and Osambela, 2020; Goldstein, Kopytov, Lin, and

Xiang, 2021). In addition, empirical work shows relationships between certain types of

institutional investors and their CSR or carbon risk equity portfolio holdings (Dyck, Lins,

Roth, and Wagner, 2019; Ilhan, Krueger, Sautner, and Starks, 2021). Further, Starks,

Venkat, and Zhu (2021) provide evidence that institutional investors with longer-term

horizons have stronger preferences to invest in firms with higher ESG profiles.

Accordingly, we distinguish two classes of major investors in the corporate bond market

that have been argued to have different investment horizons: insurance companies and

mutual funds due to the differences in their investment strategies. In particular, insurance

companies tend to hold their bonds to maturity, while mutual funds tend to trade more

frequently and hence have a much shorter horizon (Massa, Yasuda, and Zhang, 2013).

As long-term investors have been shown to care more about firms’ environmental profiles

(Starks, Venkat, and Zhu, 2021), we posit that insurance companies are more likely to

reduce their holdings of corporate bonds issued by firms with poor environmental profiles

after the Paris Agreement. Further, these changes should be relevant to the bond pricing

changes we find because insurance companies collectively hold around 25-30% of corporate

bonds and mutual funds hold around 15% of outstanding bonds.

We conduct difference-in-differences analyses using eight quarterly snapshots of insti-

tutional portfolio holdings around the Paris Agreement (from the fourth quarter of 2014 to

the fourth quarter of 2016). The data consists of institutional investor holdings obtained

from Refinitiv eMAXX (formerly Lipper eMAXX). Each quarter, we sum up individual

bond holdings of (1) all institutional investors included in the eMAXX reporting entities,

(2) all mutual funds, and (3) all insurance companies, where we scale each of the investor’s

bond holdings by the outstanding amount of the particular bond issue. Each treated bond

is matched to a control bond using bond characteristics: issue principal size, credit rat-

32
ing, time to maturity and the oil beta of the firm’s equity. We then regress the particular

institutional ownership variable (all institutional investors, mutual funds or insurance com-

panies) on an indicator variable indicating quarters after the Paris Agreement, an indicator

variable indicating issuers with low environmental profiles and the interaction between the

two variables:

Ownershipit = β1 T reatedi × Af terP arist + β2 T reatedi + BondControl + κt + it . (11)

We define T reatedi bonds in the same four way by assigning an indicator variable

equal to one if the issuing firm has a below-median environmental score in December 2014,

is in a top carbon emissions industry, is in the top-quartile in terms of firm-level total

carbon emissions in 2014, or is in the top-quartile in terms of firm-level carbon intensity

in 2014. The bond-level control variables (BondControl) include issuance amount, years

to maturity, and bond credit rating.

Table 9 shows the results of the difference-in-differences analyses for the changes in total

institutional investor ownership, mutual fund ownership, and insurance company ownership

around the Paris Agreement for the treated bonds. When we define the treated bonds as

bonds issued by firms belonging to the top emission industries, we find that the total

institutional ownership of the treated bonds declines by 1.24 percentage points relative

to control bonds that are matched on bond characteristics. When we further decompose

the change in institutional ownership into ownership ratios by mutual funds and insurance

companies, we find that the reduction of institutional ownership of high emissions industry

bonds is entirely driven by insurance companies, whose ownership drops by 1.21 percentage

points. In contrast, the ownership of these bonds by mutual funds, which typically have a

relatively shorter investment horizon, is virtually unchanged around the Paris Agreement.

The other regression results reported in Table 9 are based on alternative definitions

of the treated bonds. A consistent pattern emerges: the total institutional ownership

33
either declines (in the case of below-median environmental scores) or does not change (in

the case of top-quartile carbon emissions and carbon intensity). However, the ownership

held by insurance companies consistently decreases by around one percentage point, while

the ownership held by mutual funds consistently increases. These analyses suggest that

the Paris Agreement resulted in a transfer of ownership from relatively long-term bond

investors (insurance companies) to investors with typically shorter horizons (mutual funds),

which is consistent with the argument that environmental and climate risks are likely

to materialize in the future and investors have different considerations based on their

investment horizon.33

6. Reversals around the 2016 U.S. Presidential election

and the subsequent announcement of U.S. withdrawal

from the Paris Agreement

The empirical results in the previous sections support the hypotheses that environmen-

tal regulatory risks are an important component of bond credit rating assessments and the

pricing of the bonds. The results also support the hypothesis that market participants view

bonds issued by firms with poor environmental performance as becoming relatively more

risky after the Paris Agreement due to changed expectations regarding regulations. How-

ever, these expectations may have again changed due to subsequent U.S. political events,

namely the November 2016 U.S. Presidential election and the June 2017 announcement

that the U.S. would withdraw from the Paris Agreement.

After the Paris Agreement, one of the candidates in the U.S. presidential primaries,
33
The reduction of insurance company bond ownership of high-emission firms is not driven by the
occurrence of bond credit rating downgrades after the Paris Agreement. In untabulated tests, we drop all
bonds that experienced a credit rating downgrade during the 12 months following the Paris Agreement.
The results for the remaining bonds in the sample remain robust with a significant reduction of insurance
company ownership.

34
Donald Trump, actively campaigned on withdrawal from the Paris Agreement.34 Moreover,

when Mr. Trump was elected President of the United States on November 8, 2016, many,

including the media and betting markets appeared to view the result as a surprise.35

Consequently, the 2016 election provides a setting in which bond market participants may

have re-evaluated the likelihood of severe new climate regulation. In addition to examining

the responses in the bond market to the election result, we also analyze whether changes

occurred upon the announcement of an actual policy change, which occurred on June 1,

2017, when the U.S. President announced that the country would withdraw from the Paris

Agreement.

An important complicating factor in interpreting the market responses to these events

is that any withdrawal from the Paris Agreement could not occur before November 4,

2020.36 Therefore, we may not observe discernible responses in credit ratings, given the

documented conservatism of credit rating agencies.37 Anecdotal evidence suggests this to

be the case.38 We also note that the climate news index of (Engle et al., 2020) has a spike

in June 2017, but it is much lower than that of many other events over the authors’ sample

period.

A further complicating factor in the market response to the withdrawal announcement is

the degree to which the individual states would impose their own climate change regulation,
34
The BBC News had the headline “Donald Trump would ‘cancel’ Paris climate deal.” BBC News. May
27, 2016; See also Parker and Davenport (2016).
35
On November 9, 2016, The Wall Street Journal ran an article with the headline “Trump’s Surprise
Ends Election, Begins Uncertainty for Markets.” Similarly, CNN had an article wire with the headline,
“Trump’s Victory in U.S. Election, The Latest in a Year of Shocks.” Further, the website PredictIt in-
dicated a very sharp spike on election day, going from $0.22 on November 7 to $0.98 on November 9
(https://www.predictit.org/markets/detail/1234/Who-will-win-the-2016-US-presidential-election).
36
The Paris Agreement entered into force on November 4, 2016. According to Article 28 of the Paris
Agreement a party to the agreement may withdraw by written notification after three years and the with-
drawal will take effect one year after the written notification “or on such later date as may be specified in
the notification of withdrawal.”
37
Evidence shows that credit rating analysts tend to change ratings only with sufficient certainty regard-
ing changes in issuer default risk (Altman and Rijken, 2004; Gredil et al., 2019; Löffler, 2005). Consequently,
bond credit ratings are typically far less volatile than bond spreads.
38
In the February 16, 2017 Moody’s Report “Shift in US Climate Policy Would Not Stall Global Efforts
to Reduce Emissions,” the agency notes that as of time of publication, it is too early to tell exactly what
climate policy will be reversed.

35
even without federal regulation. In fact, upon the June 2017 announcement of the pending

U.S. withdrawal from the Paris Agreement, the governors of Washington, New York and

California declared that they were forming an alliance of states committed to upholding

the Paris accord.39

To test the extent of any changes in ratings and spreads that occurred due to changed

expectations around the election and the withdrawal accouncements, we conduct the fol-

lowing regressions:

Ratingit = β1 EnvP rofj × Eventt + γi + κt + it , (12)

Spreadit = β1 EnvP rofj × Eventt + γi + κtp + it , (13)

where Eventt is an indicator variable equal to 1 for observations after the November 2016

election or the June 1, 2017 withdrawal announcement. γi is the security fixed effects,

κt is the time fixed effects and κtp is the pair-time fixed effects. In each of these tests,

our pre- and post-event periods are set to be 6 months (as opposed to the 12 months we

used previously) so that the Paris Agreement announcement (or the 2016 election) is not

included in the pre-event period. We limit the sample to bonds issued before December

2014 (i.e. those also used in the Paris Agreement test) and that are outstanding through

the full sample period, so that we are estimating whether there exist any reversals of

the changes observed following the Paris Agreement. In these specifications, β1 can be

interpreted as the change in bond credit rating or yield spread following the respective

event for firms with poor environmental profiles relative to those from other firms.

The results for the changes in credit ratings after the 2016 election are displayed in

Panel A of Table 10 in which columns (1) - (4) display results using the low environmental

score indicator, the high carbon emissions industry indicator, the top-quartile firm carbon

emissions indicator and the high carbon intensity indicator. In all four specifications the
39
See Tabuchi and Fountain (2017). This alliance has grown into a bipartisan coalition of 25 states
representing 55 percent of the U.S. population (http://www.usclimatealliance.org/)

36
result is statistically indistinguishable from zero, thus, following the 2016 election out-

come, the rating agencies did not meaningfully change ratings for bond issuers with poor

environmental profiles.

Panel B reports the regression results after the June 2017 withdrawal announcement.

There exists strong evidence that relative credit ratings improved for the environmentally

problematic firms after the withdrawal announcement. In particular, column (1) shows

that corporate bond ratings increased by 0.096 notch for bonds issued by firms with below-

median environmental scores after the withdrawal announcement relative to other firms.

We observe a similar increase in corporate bond ratings when examining firms with any of

the high carbon emission measures. These effects can be considered partial reversals in the

increase in corporate bond spreads after the Paris Agreement. The partial reversals are

consistent with the argument that ratings agencies are only comfortable adjusting ratings

after a clear and definitive change in climate policy.

In contrast, the evidence on bond yield spreads suggests that bond investors reacted

more swiftly to these two events, as shown in Table 11, which displays the effects of the 2016

presidential election and the subsequent withdrawal announcement on bond spreads. Panel

A provides the results for the changes in bond yield spreads following the 2016 election.

Unlike the results on credit ratings, for all four treatments we find a substantial decrease

in bond spreads following the 2016 election. In particular, Column (1) shows that after the

2016 election, bond spreads decreased by 18.8 bps for firms with low environmental scores,

representing a reversal of slightly under one-half of the earlier increase in bond spreads

after the Paris Agreement. Similarly, in columns (2) through (4) we find similar results

using the high carbon emissions measures to define the treatment. Each of these effects

represents between a one-third and one-half reversal of the effect documented in Table 6.

The evidence in Panel B of Table 11 shows no significant changes in bond spreads for

firms with poor environmental profiles following the withdrawal announcement in June

2017. This result on bond spreads may be due to the response following the election.

37
 Overall, the findings for the election and withdrawal events are consistent with ratings

agencies and bond investors viewing firms with poor environmental profiles as less risky

when the prospect of future environmental regulations is lessened. The results also sug-

gest that bond investors reacted to the potential policy change affecting issuers with poor

environmental profiles more swiftly (after the 2016 election), but that credit rating ana-

lysts waited for a definitive change in policy before adjusting ratings for climate regulatory

risk, which is consistent with prior research on the conservatism of credit rating analysts

(Altman and Rijken, 2004; Gredil et al., 2019; Löffler, 2005).

7. Conclusions

Environmental risks, including climate risks, have been receiving more focused atten-

tion from financial market participants. In this study, we provide empirical evidence that

suggests uncertainty about future regulatory actions can motivate bond market partici-

pants to respond to firms’ environmental performance, and particularly, changes in firms’

exposures to climate risks.

We present empirical results suggesting that having poor environmental performance,

including having a more significant carbon footprint, is associated in general with lower

credit ratings and higher bond yield spreads, particularly for firms located in states with

stricter environmental regulations. We also provide evidence of a causal component to

these results by examining bond credit ratings and yield spreads for environmentally poor

firms after a shock to their regulatory risk. We find that the December 2015 Paris Agree-

ment appears to have had increased the regulatory risk for firms that are in high emissions

industries or have poor environmental performance in general, resulting in negative con-

sequences. More importantly, these effects on bond ratings and yields are observed to be

stronger in states that enforce regulation more strictly, suggesting that they are stronger

because potential new regulations were expected to be enforced more strictly.

38
 Our results have important implications for how firms’ environmental profiles are re-

lated to market participants’ assessments of their corporate bonds’ risks and values. The

results suggest that credit rating analysts and bond investors are concerned with issuers’

environmental profiles because of potential regulatory costs. Thus, if bond investors expect

an issuer to be punished for poor environmental performance, they are more likely to price

those costs into the firms’ bonds.

39
References
Abadie, Alberto, and Guido W Imbens, 2006, Large sample properties of matching esti-
mators for average treatment effects, Econometrica 74, 235–267.

Altman, Edward I, and Herbert A Rijken, 2004, How rating agencies achieve rating stabil-
ity, Journal of Banking & Finance 28, 2679–2714.

Amiraslani, Hami, Karl V Lins, Henri Servaes, and Ane Tamayo, 2021, Trust, social cap-
ital, and the bond market benefits of ESG performance, Review of Accounting Studies
Forthcoming.

Arnold, Martin, 2020, ECB to consider using climate risk to steer bond purchases, Financial
Times.

Bai, Jennie, Turan G Bali, and Quan Wen, 2019, Common risk factors in the cross-section
of corporate bond returns, Journal of Financial Economics 131, 619–642.

Bai, Jennie, Turan G. Bali, and Quan Wen, 2021, Is there a risk-return tradeoff in the
corporate bond market? Time-series and cross-sectional evidence, Journal of Financial
Economics 142, 1017–1037.

Baker, Malcolm, Daniel Bergstresser, George Serafeim, and Jeffrey Wurgler, 2018, Financ-
ing the response to climate change: The pricing and ownership of US green bonds, NBER
Working paper 25194.

Baker, Scott R, Nicholas Bloom, and Steven J Davis, 2016, Measuring economic policy
uncertainty, The Quarterly Journal of Economics 131, 1593–1636.

Baker, Steven D., Burton Hollifield, and Emilio Osambela, 2020, Asset prices and portfolios
with externalities, Unpublished working paper .

Baldauf, Markus, Lorenzo Garlappi, and Constantine Yannelis, 2020, Does climate change
affect real estate prices? Only if you believe in it, The Review of Financial Studies 33,
1256–1295.

Bartram, Söhnke M, Kewei Hou, and Sehoon Kim, 2021, Real effects of climate policy:
Financial constraints and spillovers, Journal of Financial Economics.

Berlinger, Joshua, 2016, Trump’s victory in the US election the latest in a year of shocks,
CNN Wire.

Bernstein, Asaf, Matthew T Gustafson, and Ryan Lewis, 2019, Disaster on the horizon:
The price effect of sea level rise, Journal of Financial Economics 134, 285–272.

Bolton, Patrick, and Marcin Kacperczyk, 2020, Do investors care about carbon risk?,
Journal of Financial Economics.

40
Bolton, Patrick, and Marcin Kacperczyk, 2021, Global pricing of carbon-transition risk,
Working Paper .

Bradley, Daniel, Christos Pantzalis, and Xiaojing Yuan, 2016, Policy risk, corporate polit-
ical strategies, and the cost of debt, Journal of Corporate Finance 40, 254–275.

Capelle-Blancard, Gunther, Patricia Crifo, Marc-Arthur Diaye, Rim Oueghlissi, and Bert
Scholtens, 2019, Sovereign bond yield spreads and sustainability: An empirical analysis
of OECD countries, Journal of Banking & Finance 98, 156–169.

Chava, Sudheer, 2014, Environmental externalities and cost of capital, Management Sci-
ence 60, 2223–2247.

Child, Travers Barclay, Nadia Massoud, Mario Schabus, and Yifan Zhou, 2021, Surprise
election for trump connections, Journal of Financial Economics 140, 676–697.

Dai, Rui, Rui Duan, Hao Liang, and Lilian Ng, 2021, Outsourcing climate change, ECGI
working paper .

Dick-Nielsen, Jens, 2009, Liquidity biases in TRACE, The Journal of Fixed Income 19,
43–55.

Dyck, Alexander, Karl V. Lins, Lukas Roth, and Hannes F. Wagner, 2019, Do institu-
tional investors drive corporate social responsibility? International evidence, Journal of
Financial Economics 131, 693–714.

Edwards, Amy K, Lawrence E Harris, and Michael S Piwowar, 2007, Corporate bond
market transaction costs and transparency, The Journal of Finance 62, 1421–1451.

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

Fernando, Chitru S, Mark P Sharfman, and Vahap B Uysal, 2017, Corporate environmental
policy and shareholder value: Following the smart money, Journal of Financial and
Quantitative Analysis 52, 2023–2051.

Flammer, Caroline, 2021, Corporate green bonds, Journal of Financial Economics 142,
499–516.

Frick, Fiona, 2020, Pandemic has led to tipping point in understanding risk, Financial
Times.

Furtado, Lindsey, 2017, Why activists are cheerleaders for corporate social responsibility,
Financial Times.

Gao, Lin, Steffen Hitzemann, Ivan Shaliastovich, and Lai Xu, 2021, Oil volatility risk,
Journal of Financial Economics .

41
Gao, Mingze, 2020, Firm Historical Headquarter State from SEC 10K/Q Filings, Technical
report.

Ginglinger, Edith, and Quentin Moreau, 2019, Climate risk and capital structure, Unpub-
lished working paper .

Goldstein, Itay, Alexander Kopytov, Shen Lin, and Haotian Xiang, 2021, On ESG investing:
Heterogeneous preferences, information, and asset prices, Unpublished working paper .

Gormley, Todd A, and David A Matsa, 2011, Growing out of trouble? Corporate responses
to liability risk, The Review of Financial Studies 24, 2781–2821.

Gourio, Francois, 2013, Credit risk and disaster risk, American Economic Journal: Macroe-
conomics 5, 1–34.

Gredil, Oleg, Nishad Kapadia, and Jung Hoon Lee, 2019, Do rating agencies deserve some
credit? Evidence from transitory shocks to credit risk, working paper .

Greenstone, Michael, 2002, The impacts of environmental regulations on industrial activ-

ity: Evidence from the 1970 and 1977 Clean Air Act amendments and the census of
manufactures, Journal of Political Economy 110, 1175–1219.

Gürkaynak, Refet S, Brian Sack, and Jonathan H Wright, 2007, The US Treasury yield
curve: 1961 to the present, Journal of Monetary Economics 54, 2291–2304.

Heinkel, Robert, Alan Kraus, and Josef Zechner, 2001, The effect of green investment on
corporate behavior.

Hong, Harrison, Frank Weikai Li, and Jiangmin Xu, 2019, Climate risks and market effi-
ciency, Journal of Econometrics 208, 265–281.

Huynh, Thanh D, and Ying Xia, 2021, Climate change news risk and corporate bond
returns, Journal of Financial and Quantitative Analysis 56, 1985–2009.

Ilhan, Emir, Philipp Krueger, Zacharias Sautner, and Laura Starks, 2021, Climate risk
disclosure and institutional investors, Unpublished working paper .

Ilhan, Emirhan, Zacharias Sautner, and Grigory Vilkov, 2021, Carbon tail risk, Review of
Financial Studies 34, 1540–1571.

Jiraporn, Pornsit, Napatsorn Jiraporn, Adisak Boeprasert, and Kiyoung Chang, 2014, Does
corporate social responsibility (CSR) improve credit ratings? Evidence from geographic
identification, Financial Management 43, 505–531.

Karpoff, Jonathan M, John R Lott, Jr, and Eric W Wehrly, 2005, The reputational penal-
ties for environmental violations: Empirical evidence, The Journal of Law and Economics
48, 653–675.

42
Kaviani, Mahsa S., Lawrence Kryzanowski, Hosein Maleki, and Pavel Savor, 2020, Policy
uncertainty and corporate credit spreads, Journal of Financial Economics .

Kempf, Elisabeth, and Margarita Tsoutsoura, 2021, Partisan professionals: Evidence from
credit rating analysts, Journal of Finance 76, 2805–2856.

Konisky, David M, 2007, Regulatory competition and environmental enforcement: Is there

a race to the bottom?, American Journal of Political Science 51, 853–872.

Krüger, Philipp, 2015, Corporate goodness and shareholder wealth, Journal of Financial
Economics 115, 304–329.

Krüger, Philipp, Zacharias Sautner, and Laura T Starks, 2020, The importance of climate
risks for institutional investors, Review of Financial Studies 33, 1067–1111.

Li, Jennifer (Jie), Massimo Massa, Hong Zhang, and Jian Zhang, 2021, Air pollution,
behavioral bias, and the disposition effect in China, Journal of Financial Economics
142, 641–673.

Löffler, Gunter, 2005, Avoiding the rating bounce: Why rating agencies are slow to react
to new information, Journal of Economic Behavior & Organization 56, 365–381.

Margaretic, Paula, and Sébastien Pouget, 2018, Sovereign bond spreads and extra-financial
performance: An empirical analysis of emerging markets, International Review of Eco-
nomics & Finance 58, 340–355.

Massa, Massimo, Ayako Yasuda, and Lei Zhang, 2013, Supply uncertainty of the bond
investor base and the leverage of the firm, Journal of Financial Economics 110, 185–
214.

Moody’s, 2018, General principles for assessing environmental, social and governance risks.

Murfin, Justin, and Matthew Spiegel, 2020, Is the risk of sea level rise capitalized in
residential real estate, Review of Financial Studies 33, 1217–1255.

Oehmke, Martin, and Marcus Opp, 2020, A theory of responsible investment, Unpublished
working paper .

Painter, Marcus, 2020, An inconvenient cost: The effects of climate change on municipal
bonds, Journal of Financial Economics 135, 468–482.

Parker, Ashley, and Coral Davenport, 2016, Donald Trump’s energy plan: More fossil fuels
and fewer rules, The New York Times.

Pastor, Lubos, Robert F. Stambaugh, and Lucian A. Taylor, 2021, Sustainable investing
in equilibrium, Journal of Financial Economics 142, 550–571.

Pastor, Lubos, and Pietro Veronesi, 2013, Political uncertainty and risk premia, Journal
of Financial Economics 110, 520 – 545.

43
Pedersen, Lasse Heje, Shaun Fitzgibbons, and Lukasz Pomorski, 2021, Responsible invest-
ing: The ESG-efficient frontier, Journal of Financial Economics 142, 572–597.

Pindyck, Robert S, 1993, Investments of uncertain cost, Journal of Financial Economics

34, 53–76.

Ramadorai, Tarun, and Federica Zeni, 2021, Climate regulation and emissions abatement:
Theory and evidence from firms’ disclosures, ECGI Working Paper .

Ramelli, Stefano, Alexander F Wagner, Richard J Zeckhauser, and Alexandre Ziegler, 2021,
Investor rewards to climate responsibility: Stock-price responses to the opposite shocks
of the 2016 and 2020 U.S. elections, Review of Corporate Finance Studies Forthcoming.

Schlenker, Wolfram, and Charles A. Taylor, 2021, Market expectations of a warming cli-
mate, Journal of Financial Economics 142, 627–640.

Shultz, Abby, 2017, Fund of information: Bonds now face scrutiny on sustainability, Dow
Jones Newswire, June 17, 2017 .

Starks, Laura T, Parth Venkat, and Qifei Zhu, 2021, Corporate ESG profiles and investor
horizons, Unpublished working paper .

Stroebel, Johannes, and Jeffrey Wurgler, 2021, What do you think about climate finance?,
Journal of Financial Economics 142, 487–498.

Tabuchi, Hiroko, and Henry Fountain, 2017, Bucking Trump, these cities, states and com-
panies commit to Paris Accord, The New York Times.

Tang, Dragon Yongjun, and Yupu Zhang, 2020, Do shareholders benefit from green bonds?,
Journal of Corporate Finance 61, 101427.

Zerbib, Olivier David, 2019, The effect of pro-environmental preferences on bond prices:
Evidence from green bonds, Journal of Banking & Finance 98, 39–60.

44
Fig. 1. Credit ratings and yield spreads of high carbon emission industries’ bonds before
and after the Paris Agreement.
This figure displays equal-weighted average ratings and spreads for each of the top 15
carbon-emitting industries, before and after the Paris Agreement, where the pre-period
runs from December 2014 through November 2015 and the post period runs from December
2015 through November 2016. A numerical rating of 1 corresponds to a D rating, a rating
of 5 to a Caa2 rating, a rating of 10 to a Ba3 rating, a rating of 15 to a Baa1 rating a
rating of 20 to a Aa2 rating and a rating of 22 to a Aaa rating.
(a) Ratings

(b) Spreads

45
Fig. 2. Ratings and spreads by environmental scores before and after the Paris Agreement.
This figure displays equal-weighted average ratings and spreads for firms divided by their
levels of Sustainalytics Environmental scores, before and after the Paris Agreement, where
the pre-period runs from December 2014 through November 2015 and the post period runs
from December 2015 through November 2016. A numerical rating of 1 corresponds to a D
rating, a rating of 5 to a Caa2 rating, a rating of 10 to a Ba3 rating, a rating of 15 to a
Baa1 rating a rating of 20 to a Aa2 rating and a rating of 22 to a Aaa rating.
(a) Ratings

(b) Spread

46
Fig. 3. Bond credit ratings around the Paris Agreement announcement.
This figure P
plots the coefficients from the following regression equation:
Ratingit = 11 k=−11 βk [1(t = k) × EnvP rofj ] + γi + κt + it .
EnvP rofj is equal to one for treated observations, where the treatment is defined al-
ternatively as a below-median environmental score, being in the top 15 carbon-emitting
industries, being in the top-quartile of CDP emissions, or being in the top-quartile of CDP
carbon intensity (tons of emissions divided by revenue in $1,000). Control observations are
all other securities. γi , κtp are security and matched-pair-by-time fixed effects. Pre-period
runs from December 2014 through November 2015 and post-period runs December 2015
through November 2016. The chart includes all interaction terms except for December
2014, so the regression coefficient can be interpreted as the effect of being a low envi-
ronmental profile firm on bond credit ratings in each period relative to December 2014.
(Higher numerical scores indicate better credit ratings.)
(a) EnvP rofj = BelowM edEnvj (b) EnvP rofj = HighEmissionsIndj

(c) EnvP rofj = T opQtEmissionsj (d) EnvP rofj = T opQtCarbonInt.j

47
Fig. 4. Yield spreads around the Paris Agreement announcement.
This figure plots
P the coefficients from the following regression equation:
Spreadit = 11 k=−11 βk [1(t = k) ∗ EnvP rofj ] + γi + κtp + it .
EnvP rofj is equal to one for treated observations, where the treatment is defined al-
ternatively as a below-median environmental score, being in the top 15 carbon-emitting
industries, being in the top-quartile of CDP emissions, or being in the top-quartile of CDP
carbon intensity (tons of emissions divided by revenues in $1,000). Control observations are
selected using a one-to-one nearest neighbor matching with replacement by Mahalanobis
matching with replacement procedure on rating, time to maturity, issue principal outstand-
ing and oil beta as of year-end 2014. γi , κtp are security and matched-pair-by-time fixed
effects. Pre-period runs from December 2014 through November 2015 and post-period runs
December 2015 through November 2016. The chart includes all interaction terms except
for December 2014, so the regression coefficient can be interpreted as the effect of being a
low environmental profile firm on bond yield spreads in each period relative to December
2014.
(a) EnvP rofj = BelowM edEnvj (b) EnvP rofj = HighEmissionsIndj

(c) EnvP rofj = T opQtEmissionsj (d) EnvP rofj = T opQtCarbonInt.j

48
 Table 1: Summary statistics.

This table reports the summary statistics for the at issue bond sample with a sample period of 2009
through 2017. Trading yield spread, yield, profitability, leverage, annual returns, ln(total assets), and
cash/assets are winsorized at the 1% and 99% levels. The ratings variable is assigned such that a higher
number indicates a better rating. A numerical rating of 1 corresponds to a D rating, a rating of 5 to a
Caa2 rating, a rating of 10 to a Ba3 rating, a rating of 15 to a Baa1 rating a rating of 20 to a Aa2 rating
and a rating of 22 to a Aaa rating. Reg stringency is measured as the firm’s regulatory stringency
determined as the revenue-weighted average number of EPA penalties issued in a given year divided by
the number of facilities (in thousands) in that state for the states the firm operates in. When this is not
available, the number of EPA penalties in the state the firm’s headquarters are located in divided by the
number of plants regulated by the EPA in that state (in thousands). Top 15 emissions industries are
defined as the top 15 carbon emissions industries based on carbon emissions using the CDP data.

Variable Observations Mean Median Std. Dev.

Credit rating 1,940 15.312 15.000 2.828

Offering spread 1,940 1.835 1.481 1.273
Firm-weighted average maturity 1,940 9.414 9.274 3.336
Environmental score 1,940 59.960 60.000 14.050
Reg Stringency 1,940 0.714 0.446 0.950
Top 15 emissions industry 1,940 0.487 0.000 0.500
Emissions (millions of ton) 1,312 6.680 0.438 19.665
Carbon intensity (ton per $1,000 revenue) 1,312 0.319 0.014 0.997
Ln(1 + Principal) 1,940 13.384 13.305 0.602
Time to maturity 1,940 9.969 10.000 7.289
Callable 1,940 0.970 1.000 0.170
Leverage 1,940 0.287 0.274 0.147
Pre-tax interest coverage 1,940 19.303 11.760 23.433
Ln(Total assets) 1,940 10.197 10.234 1.259
Cash/assets 1,940 0.118 0.072 0.131
Profitability 1,940 0.222 0.168 0.185
Tangibility 1,940 0.302 0.188 0.256
Annual stock returns 1,940 15.599 13.648 24.203
Ln(Standard deviation of returns) 1,940 2.939 2.902 0.415

49
 Table 2: Credit ratings and regulatory stringency

This table displays results from the following panel regression:

Ratingit = β1 EnvP rofjt−1 + β2 Regjt−1 + β3 EnvP rofjt−1 × Regjt−1 + β4 Xjt−1 + F E + it .

All observations are at-issue bonds. Environmental scores, leverage, ln(total assets), profitability,
annualized stock return, and the standard deviation of stock returns are winsorized at the 1% and 99%
levels and are measured at the end of the previous year. Reg stringency, defined as the revenue-weighted
average number of EPA penalties in a given year divided by the number of facilities in that state (for the
states in which the firm operates), is also standardized by mean and scaled by standard deviation. *, **
and *** indicate 10%, 5% and 1% significance, respectively. Fixed effects are indicated in each column.
Standard errors, which are clustered at the firm level, are shown in parentheses.

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

Environmental Score × Reg 0.020** 0.019***
(0.008) (0.007)
Emissions × Reg -0.021 -0.021***
(0.013) (0.008)
Carbon Intensity × Reg -0.283** -0.328***
(0.120) (0.103)
Environmental Score 0.027*** 0.014*
(0.008) (0.007)
Emissions -0.013 -0.014**
(0.010) (0.006)
Carbon Intensity -0.514*** -0.231*
(0.141) (0.119)
Reg -1.031** 0.180 0.140 -0.983*** 0.127 0.114
(0.428) (0.117) (0.111) (0.372) (0.082) (0.082)
Firm Weighted Average Maturity 0.016 -0.026 -0.035 0.041 -0.004 -0.003
(0.031) (0.038) (0.038) (0.026) (0.029) (0.030)
Leverage -1.978* -0.394 -0.221 -2.146*** -1.312 -1.299
(1.050) (1.272) (1.215) (0.812) (1.027) (1.035)
Pre-tax interest coverage 0.029*** 0.037*** 0.035*** 0.021*** 0.017*** 0.016***
(0.005) (0.005) (0.005) (0.004) (0.004) (0.004)
Ln(Total Assets) 0.946*** 1.085*** 1.033*** 1.059*** 1.201*** 1.191***
(0.117) (0.133) (0.130) (0.111) (0.131) (0.129)
Cash/Assets 3.904*** 5.178*** 5.324*** 2.501** 3.099*** 3.163***
(1.234) (1.407) (1.439) (1.113) (1.131) (1.135)
Profitability 0.776 1.187 0.688 0.442 3.716*** 3.647***
(0.662) (0.890) (0.826) (0.773) (1.266) (1.263)
Tangibility -0.258 0.496 1.100* 1.679* 1.991* 2.006*
(0.450) (0.588) (0.570) (0.875) (1.192) (1.182)
Annual Stock Returns -0.006** -0.005 -0.004 -0.005** -0.001 -0.001
(0.003) (0.003) (0.003) (0.002) (0.003) (0.003)
Ln(Standard Deviation Returns) -1.785*** -1.957*** -2.100*** -1.665*** -1.533*** -1.533***
(0.246) (0.334) (0.323) (0.209) (0.286) (0.286)
Time Fixed Effects Y Y Y Y Y Y
Industry Fixed Effects N N N Y Y Y
Adj. R2 0.587 0.555 0.574 0.546 0.526 0.527
Obs 1,940 1,312 1,312 1,938 1,309 1,309

50
 Table 3: Offering spreads and regulatory stringency.

This table displays results from the following panel regression:

Spreadit = β1 EnvP rofjt−1 + β2 Regjt−1 + β3 EnvP rofjt−1 × Regjt−1 + β4 Xjt−1 + β5 Zit + F E + it .

All observations are at-issue bonds. Environmental scores, coupon rate, leverage, ln(total assets),
profitability, annualized stock return, and the standard deviation of stock returns are winsorized at the
1% and 99% levels. Firm characteristics are measured as of the end of the previous year. Reg stringency,
defined as the revenue-weighted average number of EPA penalties in a given year divided by the number
of facilities in that state (for the states in which the firm has facilities), is also standardized by mean and
scaled by standard deviation. *, ** and *** indicate 10%, 5% and 1% significance, respectively. Fixed
effects are indicated in each column. Standard errors, which are clustered at the firm level, are shown in
parentheses.

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

Environmental Score × Reg -0.005* -0.006**
(0.003) (0.003)
Emissions × Reg 0.007* 0.009**
(0.004) (0.004)
Carbon Intensity × Reg 0.073 0.101
(0.070) (0.069)
Environmental Score -0.009*** -0.004
(0.003) (0.003)
Emissions 0.007*** 0.003
(0.003) (0.003)
Carbon Intensity 0.162*** 0.088*
(0.052) (0.049)
Reg 0.225 -0.078** -0.065* 0.333** -0.062** -0.054*
(0.164) (0.039) (0.036) (0.155) (0.029) (0.029)
Ln(1 + Principal) 0.254*** 0.189*** 0.216*** 0.217*** 0.179*** 0.184***
(0.058) (0.057) (0.056) (0.045) (0.045) (0.045)
Time to Maturity 0.090*** 0.093*** 0.093*** 0.091*** 0.094*** 0.094***
(0.003) (0.003) (0.003) (0.002) (0.002) (0.002)
Callable 0.391* 0.296 0.275 0.058 0.033 0.022
(0.225) (0.183) (0.180) (0.155) (0.131) (0.131)
Leverage 1.193*** 0.520 0.435 1.004*** 0.582* 0.586*
(0.416) (0.413) (0.391) (0.331) (0.331) (0.333)
Pre-tax interest coverage -0.007*** -0.007*** -0.006*** -0.006*** -0.005*** -0.005***
(0.001) (0.002) (0.001) (0.001) (0.001) (0.001)
Ln(Total Assets) -0.250*** -0.190*** -0.177*** -0.297*** -0.236*** -0.237***
(0.036) (0.036) (0.034) (0.043) (0.042) (0.042)
Cash/Assets -0.473** -0.564** -0.635*** -0.184 -0.361 -0.368
(0.237) (0.231) (0.227) (0.244) (0.258) (0.262)
Profitability -0.246 -0.211 -0.081 -0.322 -0.715 -0.685
(0.202) (0.151) (0.139) (0.374) (0.476) (0.474)
Tangibility 0.157 -0.291* -0.366** 0.197 -0.530 -0.561*
(0.151) (0.167) (0.153) (0.356) (0.329) (0.334)
Annual Stock Returns -0.003** -0.003** -0.003*** -0.003** -0.003*** -0.003***
(0.001) (0.001) (0.001) (0.001) (0.001) (0.001)
Ln(Standard Deviation Returns) 1.045*** 0.850*** 0.878*** 0.927*** 0.675*** 0.677***
(0.104) (0.125) (0.123) (0.094) (0.123) (0.122)
Time Fixed Effects Y Y Y Y Y Y
Industry Fixed Effects N N N Y Y Y
Adj. R2 0.596 0.672 0.678 0.601 0.715 0.715
Obs 1,940 1,312 1,312 1,938 1,309 1,309

51
 Table 4: Effects of the Paris Agreement on credit ratings.

This table displays results from the following regression:

Ratingit = β1 EnvP rofj × Af terP arist + γi + κt + it

Env.P rofj is alternatively one of the following: an indicator variable equal to one if the issuing firm has a
below median environmental score in December 2014 (Low Environmental Score), an indicator variable
equal to one if the firm is in one of the top 15 carbon-emitting industries (High Emission Industry), an
indicator variable equal to one if the firm is in the top quartile of carbon emissions in 2014 (Top Quartile
Emissions), or an indicator variable equal to one if the firm is in the top quartile of carbon intensity
(defined as tons of emissions per $1,000 in revenue) in 2014 (Top Quartile Carbon Intensity).
Af terP arist is an indicator variable equal to one if the observation occurs in December 2015 or later.
γi , κt are security and time fixed effects. Sample runs from December 2014 through November 2016. *, **
and *** indicate 10%, 5% and 1% significance, respectively. Standard errors, which are clustered at the
firm level, are shown in parentheses.

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

After Paris × Low Environmental Score -0.580***
(0.165)
After Paris × High Emissions Industry -0.482***
(0.127)
After Paris × Top-Quartile Emissions -0.551***
(0.187)
After Paris × Top-Quartile Carbon Intensity -0.627***
(0.193)
Time FE Y Y Y Y
Security FE Y Y Y Y
Adj. R2 0.052 0.040 0.068 0.083
Obs 33,336 33,336 23,184 23,184

52
Table 5: Summary statistics – matched sample for yield spreads around the Paris Agree-
ment.

This table shows summary statistics as of December 2014 (one year before the Paris Agreement)for the
sample matched on the alternative environment variables. The environment variables are defined
alternatively as one of the following: the firm has a below median environmental score (Panel A), the firm
is in a top 15 carbon emissions industry (Panel B), the firm is in the top quartile of carbon emissions in
2014 (Panel C) and the firm is in the top quartile of carbon intensity in 2014, measured as tons of
emissions divided by firm revenue in thousands of dollars (Panel D). The matched sample is formed by
using one-to-one nearest neighbor Mahalanobis matching of treated bond issues to control bond issues by
oil beta, issue principal outstanding, time to maturity and credit rating as of year-end 2014. Spread,
profitability, leverage, tangibility, and the ln(Total Assets) are winsorized at the 1% and 99% levels. The
ratings variable is assigned such that a higher number indicates a better rating. *, ** and *** indicate
10%, ** 5% and *** 1% significance respectively.

Group Sample Control Diff Mean

Variable Obs Mean St. Dev. Obs Mean St. Dev.
Panel A: Below Median Environmental Score
Security-Level Variables
Credit Rating 448 13.493 2.47 448 13.714 2.331 -0.221
Spread 448 2.011 1.283 448 2.004 1.388 0.007
Time to Maturity 448 9.46 7.447 448 9.541 7.406 -0.081
Firm Level Variables
Profitability 129 0.233 0.196 110 0.25 0.191 -0.017
Oil Beta 129 0.011 0.032 110 0.001 0.032 0.01**

Panel B: High Carbon Emissions Industry

Security Level Variables
Credit Rating 646 15.265 2.761 646 15.317 2.804 -0.052
Spread 646 1.565 1.244 646 1.38 0.927 0.185***
Time to Maturity 646 11.131 8.902 646 11.219 8.924 -0.088
Firm Level Variables
Profitability 135 0.187 0.139 114 0.288 0.22 -0.101***
Oil Beta 135 0.006 0.033 114 0.003 0.031 0.003

Panel C: Top Quartile Carbon Emissions

Security Level Variables
Credit Rating 252 15.905 2.47 252 15.813 2.33 0.092
Spread 252 1.292 0.787 252 1.268 0.917 0.024
Time to Maturity 252 11.937 9.504 252 12.049 9.521 -0.112
Firm Level Variables
Profitability 39 0.161 0.121 64 0.225 0.188 -0.064*
Oil Beta 39 0.004 0.032 64 -0.001 0.029 0.005

Panel D: Top Quartile Carbon Intensity

Security Level Variables
Credit Rating 217 15.212 2.165 217 15.263 2.182 -0.051
Spread 217 1.413 0.83 217 1.311 0.907 0.102
Time to Maturity 217 11.791 9.412 217 11.818 9.493 -0.027
Firm Level Variables
Profitability 38 0.134 0.09 64 0.228 0.165 -0.094***
Oil Beta 38 0.002 0.029 64 -0.004 0.025 0.006

53
 Table 6: Effects of the Paris Agreement on yield spreads.

This table displays results from the following regression:

Spreadit = β1 EnvP rofj × Af terP arist + γi + κtp + it

Env.P rofj is alternatively one of the following: an indicator variable equal to one if the issuing firm has a
below-median environmental score (Low Environmental Score) in December 2014, an indicator variable
equal to one if the firm is in one of the top 15 carbon-emitting industries (High Emission Industry), an
indicator variable equal to one if the firm is in the top quartile of carbon emissions in 2014 (Top Quartile
Emission), or an indicator variable equal to one if the firm is in the top quartile of carbon intensity
(defined as tons of emissions per $1,000 in revenue) in 2014 (Top Quartile Carbon Intensity).
Af terP arist is an indicator variable equal to one if the observation occurs in December 2015 or later.
γi , κtp are security and matched-pair-by-time fixed effects. The sample is formed by using one-to-one
nearest neighbor Mahalanobis matching of treated bond issues to control bond issues by oil beta, issue
principal outstanding, time to maturity and credit rating as of year-end 2014. The sample period includes
observations from December 2014 through November 2016. *, ** and *** indicate 10%, 5% and 1%
significance, respectively. Standard errors, which are clustered at the firm level, are shown in parentheses.

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

After Paris × Low Environmental Score 0.394***
(0.138)
After Paris × High Emissions Industry 0.386***
(0.103)
After Paris × Top-Quartile Emissions 0.301***
(0.096)
After Paris × Top-Quartile Carbon Intensity 0.347***
(0.088)
Pair-Time FE Y Y Y Y
Security FE Y Y Y Y
Adj. R2 0.015 0.023 0.029 0.051
Obs 21,504 31,008 12,096 10,416

54
Table 7: Regulatory stringency and the effects of the Paris Agreement on credit ratings.

This table displays results from the following regression:

Ratingit = γi + κt + β1 Af terP arist × EnvP rofj + β2 Af terP arist × HighRegj

+ β3 EnvP rofj × HighRegj + β4 Af terP arist × EnvP rofj × HighRegj + it .

EnvP rofj is alternatively either an indicator variable equal to one if the issuer has a below-median
environmental score (Low Environmental Score) in December 2014, an indicator variable equal to one if
the issuer is in one of the top 15 carbon-emitting industries (High Emission Industry), an indicator
variable equal to one if a firm is in the top quartile of carbon emissions in 2014 (Top Quartile Emission),
or an indicator variable equal to one if a firm is in the top quartile of carbon intensity (defined as tons of
emissions per $1,000 in revenue) in 2014 (Top Quartile Carbon Intensity). Af terP arist is an indicator
variable equal to one if the observation occurs in December 2015 or later. HighRegj is equal to one if the
firm is in the top-quartile of exposure to EPA penalties from 2012 through 2015. γi , κt are security and
time fixed effects. Sample is from December 2014 until November 2016. *, ** and *** indicate 10%, 5%
and 1% significance, respectively. Standard errors, which are clustered at the firm level, are shown in
parentheses.

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

After Paris × Low Environmental Score × High Reg -0.990***
(0.355)
After Paris × High Emissions Industry × High Reg -1.094***
(0.319)
After Paris × Top-Quartile Emissions × High Reg -1.371***
(0.446)
After Paris × Top-Quartile Carbon Intensity × High Reg -1.385***
(0.411)
After Paris × Low Environmental Score -0.100
(0.115)
After Paris × High Emissions Industry -0.111
(0.082)
After Paris × Top-Quartile Emissions -0.153
(0.099)
After Paris × Top-Quartile Carbon Intensity -0.157
(0.102)
After Paris × High Reg -0.121 -0.009 0.037 0.112
(0.134) (0.126) (0.121) (0.112)
Time FE Y Y Y Y
Security FE Y Y Y Y
Adj. R2 0.126 0.134 0.182 0.189
Obs 33,336 33,336 23,184 23,184

55
 Table 8: Regulatory stringency and effects of the Paris Agreement on yield spreads.

This table displays results from the following regression:

Spreadit = γi + κtp + β1 Af terP arist × EnvP rofj + β2 Af terP arist × HighRegj

+ β3 EnvP rofj × HighRegj + β4 Af terP arist × EnvP rofj × HighRegj + it .

EnvP rofj is alternatively one of the following: an indicator variable equal to one if the issuing firm has a
below median environmental score (Low Environmental Score) in December 2014, an indicator variable
equal to one if the firm is in one of the top 15 carbon-emitting industries (High Emission Industry), an
indicator variable equal to one if the firm is in the top quartile of carbon emissions in 2014 (Top Quartile
Emission), or an indicator variable equal to one if the firm is in the top quartile of carbon intensity
(defined as tons of emissions per $1,000 in revenue) in 2014 (Top Quartile Carbon Intensity).
Af terP arist is an indicator variable equal to one if the observation occurs in December 2015 or later.
HighRegj is equal to one if the firm is in the top-quartile of exposure to EPA penalties from 2012
through 2015. γi , κtp are security and matched-pair-by-time fixed effects. The sample is formed by using
one-to-one nearest neighbor Mahalanobis matching of treated bond issues to control bond issues by oil
beta, issue principal outstanding, time to maturity, and credit rating as of year-end 2014. The sample
period runs from December 2014 until November 2016. *, ** and *** indicate 10%, 5% and 1%
significance respectively. Standard errors, which are clustered at the firm level, are shown in parentheses.

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

After Paris × Low Environmental Score × High Reg 0.911**
(0.425)
After Paris × High Emissions Industry × High Reg 0.700**
(0.336)
After Paris × Top-Quartile Emissions × High Reg 0.199
(0.359)
After Paris × Top-Quartile Carbon Intensity × High Reg 0.264
(0.240)
After Paris × Low Environmental Score 0.033
(0.103)
After Paris × High Emissions Industry 0.146**
(0.070)
After Paris × Top-Quartile Emissions 0.181*
(0.096)
After Paris × Top-Quartile Carbon Intensity 0.227**
(0.111)
After Paris × High Reg -0.169 0.083 0.340 0.125
(0.320) (0.160) (0.248) (0.127)
Pair-Time FE Y Y Y Y
Security FE Y Y Y Y
Adj. R2 0.029 0.044 0.050 0.066
Obs 21,504 31,008 12,096 10,416

56
Table 9: Changes in institutional investor bond ownership around the Paris Agreement.

This table reports changes in institutional investor ownership of corporate bonds around the signing of
the Paris Agreement. Quarterly observations cover the fourth quarter of 2014 through the fourth quarter
of 2016. The periods after the fourth quarter of 2015 constitute the Post Paris Agreement periods.
Treated bonds are defined in four ways: (1) the issuer company has a below-median Sustainalytics
environmental score as of December 2014, (2) the issuer company belongs to a high emissions industry
(one of the top 15 most carbon-emitting industries), (3) the issuer company has a top-quartile carbon
emission level as of 2014 (per CDP disclosure), or (4) the issuer company has a top-quartile carbon
intensity (carbon emissions scaled by revenues) as of 2014. Control bonds are one-to-one matched to
treated bonds based on issue principal size, credit rating, bond time to maturity and the firm’s equity oil
beta. Standard errors are two-way clustered at the bond and quarter level and shown in parentheses. *,
**, and *** indicate statistical significance at the 10%, 5%, and 1% levels.

Treated bond defined by: Below-median firm environmental score High emission industries
Ownership (%) by All institutions Mutual funds Insurance firms All institutions Mutual funds Insurance firms
(1) (2) (3) (4) (5) (6)
Treated bonds * Post Paris Agreement -0.426∗ 0.265∗∗∗ -0.675∗∗ -1.237∗∗∗ -0.0333 -1.209∗∗∗
(0.221) (0.0231) (0.218) (0.173) (0.0512) (0.215)

Treated bonds 4.759∗∗∗ 2.921∗∗ 1.802 0.297 0.780 -0.516

(1.338) (1.001) (1.672) (1.083) (0.826) (1.376)

Ln(Issue amount) -3.633∗∗ 2.965∗∗ -6.747∗∗∗ -7.633∗∗∗ 1.148 -8.812∗∗∗

(1.210) (0.983) (1.475) (1.007) (0.687) (1.241)

Years to maturity -0.107 -0.258∗∗∗ 0.160 -0.00141 -0.214∗∗∗ 0.222∗∗

(0.0886) (0.0510) (0.103) (0.0710) (0.0288) (0.0789)

Credit rating (numerical) 0.195 -1.458∗∗∗ 1.667∗∗∗ -0.554∗∗∗ -1.110∗∗∗ 0.563∗∗

(0.182) (0.270) (0.328) (0.153) (0.183) (0.190)
Observations 7640 7640 7640 11082 11082 11082
Adjusted R2 0.075 0.219 0.140 0.112 0.191 0.094
Time FE Y Y Y Y Y Y

Treated bond defined by: Top-quartile firm carbon emission Top-quartile firm carbon intensity
Ownership (%) by All institutions Mutual funds Insurance firms All institutions Mutual funds Insurance firms
(1) (2) (3) (4) (5) (6)
Treated bonds * Post Paris Agreement -0.278 0.741∗∗∗ -1.022∗∗∗ -0.308 0.883∗∗∗ -1.230∗∗
(0.198) (0.112) (0.231) (0.332) (0.111) (0.388)

Treated bonds 2.218 0.137 2.060 1.618 0.431 1.192

(1.586) (1.110) (1.851) (1.479) (1.233) (1.864)

Ln(Issue amount) -8.282∗∗∗ 0.813 -9.118∗∗∗ -6.491∗∗∗ 0.535 -7.045∗∗∗

(1.369) (0.871) (1.578) (1.740) (0.992) (1.994)

Years to maturity -0.100 -0.204∗∗∗ 0.114 -0.0929 -0.187∗∗∗ 0.103

(0.0907) (0.0379) (0.0969) (0.0947) (0.0366) (0.102)

Credit rating (numerical) 0.131 -0.444∗∗ 0.571∗∗ 0.235 -0.549∗∗ 0.780∗∗

(0.218) (0.142) (0.240) (0.191) (0.204) (0.259)
Observations 4375 4375 4375 3742 3742 3742
Adjusted R2 0.110 0.113 0.105 0.088 0.110 0.093
Time FE Y Y Y Y Y Y

57
Table 10: Credit ratings and events indicating potential U.S. withdrawal from the Paris
Agreement

This table displays results from the following regression:

Ratingit = β1 EnvP rofj × Eventt + γi + κt + it .

Env.P rofj is alternatively one of the following: an indicator variable equal to one if the issuer has a
below median environmental score (Low Environmental Score), an indicator variable equal to one if the
issuer is in one of the top 15 carbon-emitting industries (High Emission Industry), an indicator variable
equal to one if a firm is in the top quartile of carbon emissions in 2015 (Top Quartile Emission), or an
indicator variable equal to one if a firm is in the top quartile of carbon intensity (defined as tons of
emissions per $1,000 in revenue) in 2015 (Top Quartile Carbon Intensity). Af terEventt is an indicator
variable equal to one if the observation occurs within six months following the 2016 election (Panel A) or
following the June 2017 announcement of the U.S. intention to withdraw from the Paris Agreement (Panel
B). γi , κt are security and time fixed effects. The sample period runs from May 2016 until April 2017 for
results on the 2016 presidential election and December 2016 until November 2017 for the results on the
Paris Agreement withdrawal announcement. *, ** and *** indicate 10%, 5% and 1% significance
respectively. Standard errors, which are clustered at the firm level, are shown in parentheses.

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

Panel A – 2016 Election
After Election × Low Environmental Score 0.046
(0.039)
After Election × High Emissions Industry 0.014
(0.035)
After Election × Top-Quartile Emissions -0.033
(0.038)
After Election × Top-Quartile Carbon Intensity -0.056
(0.045)
Adj. R2 0.003 0.000 0.002 0.005
Obs 15,036 15,036 10,704 10,704

Panel B – June Withdrawal Announcement

After Withdraw × Low Environmental Score 0.096**
(0.042)
After Withdraw × High Emissions Industry 0.082**
(0.036)
After Withdraw × Top-Quartile Emissions 0.084**
(0.038)
After Withdraw × Top-Quartile Carbon Intensity 0.076*
(0.042)
Adj. R2 0.014 0.011 0.012 0.009
Obs 13,584 13,584 9,624 9,624
Time FE Y Y Y Y
Security FE Y Y Y Y

58
Table 11: Yield spreads and events indicating potential U.S. withdrawal from the Paris
Agreement

This table displays results from the following regression:

Spreadit = β1 EnvP rofj × Eventt + γi + κtp + it .

Env.P rofj is alternatively one of the following: an indicator variable equal to one if the issuer has a below
median environmental score (Low Environmental Score), an indicator variable equal to one if the issuer is
in one of the top 15 carbon-emitting industries (High Emission Industry), an indicator variable equal to
one if a firm is in the top quartile of carbon emissions in 2015 (Top Quartile Emission), or an indicator
variable equal to one if a firm is in the top quartile of carbon intensity (defined as tons of emissions per
$1,000 in revenue) in 2015 (Top Quartile Carbon Intensity). Af terEventt is an indicator variable equal to
one if the observation occurs within six months following the 2016 election (Panel A) or following the
June 2017 announcement of the U.S. intention to withdraw from the Paris Agreement (Panel B). γi , κtp
are security and matched-pair-by-time fixed effects. The sample is formed by using one-to-one nearest
neighbor Mahalanobis matching of treated bond issues to control bond issues by oil beta, issue principal
outstanding, time to maturity and credit rating as of year-end 2014. The sample periods include May 2016
through April 2017 for tests on the 2016 presidential election and December 2016 through November 2017
for the tests on the Paris Agreement withdrawal announcement. *, ** and *** indicate 10%, 5% and 1%
significance respectively. Standard errors, which are clustered at the firm level, are shown in parentheses.

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

Panel A – 2016 Election
After Election × Low Environmental Score -0.188*
(0.098)
After Election × High Emissions Industry -0.193**
(0.081)
After Election × Top-Quartile Emissions -0.098**
(0.047)
After Election × Top-Quartile Carbon Intensity -0.143***
(0.045)
Adj. R2 0.018 0.026 0.024 0.050
Obs 8,448 13,152 5,328 4,632

Panel B – June Withdrawal Announcement

After Withdraw × Low Environmental Score 0.025
(0.085)
After Withdraw × High Emissions Industry 0.015
(0.029)
After Withdraw × Top-Quartile Emissions -0.022
(0.033)
After Withdraw × Top-Quartile Carbon Intensity -0.024
(0.029)
Adj. R2 -0.000 -0.000 0.000 0.001
Obs 8,496 12,996 5,004 4,320
Time-by-Pair FE Y Y Y Y
Security FE Y Y Y Y

59