Do Sustainable Investment Strategies Hedge Climate Change Risks? Evidence from Germany's Carbon Tax

Finance and Economics Discussion Series Federal Reserve Board, Washington, D.C. ISSN 1936-2854 (Print) ISSN 2767-3898 (Online) Do Sustainable Investment Strategies Hedge Climate Change Risks? Evidence from Germany’s Carbon Tax Marcelo Ochoa, Matthias Paustian, and Laura Wilcox 2022-073 Please cite this paper as: Ochoa, Marcelo, Matthias Paustian, and Laura Wilcox (2022). “Do Sustainable Investment Strategies Hedge Climate Change Risks? Evidence from Germany’s Carbon Tax,” Finance and Economics Discussion Series 2022-073. Washington: Board of Governors of the Federal Reserve System, https://doi.org/10.17016/FEDS.2022.073. NOTE: Staff working papers in the Finance and Economics Discussion Series (FEDS) are preliminary materials circulated to stimulate discussion and critical comment. The analysis and conclusions set forth are those of the authors and do not indicate concurrence by other members of the research staff or the Board of Governors. References in publications to the Finance and Economics Discussion Series (other than acknowledgement) should be cleared with the author(s) to protect the tentative character of these papers.

Do Sustainable Investment Strategies Hedge Climate Change Risks? Evidence from Germany’s Carbon Tax Marcelo Ochoa Matthias Paustian Laura Wilcox * November 1, 2022 Abstract It is difficult to assess the effectiveness of investment strategies that screen companies based on environmental criteria to hedge climate change risk because physicalriskshavenotyetfullymaterializedandpoliciestocombatclimatechange are usually widely anticipated. This paper sidesteps these limitations by analyzing the stock market response to plausibly exogenous changes in expectations about the level of a carbon tax in Germany. The risk-adjusted return on two sustainable investment approaches—screening companies based on environmental scores and on firms’ carbon footprint—around the carbon tax news reveals that firms with a high environmental score did not perform any better than those with a low environmental score. In contrast, the stock price of firms with low carbon emissions increased in value relative to those with a high carbon footprint. Carbon intensity explains the cross-sectional reaction to the carbon tax news because it predicts revisions in expected profitability. *Ochoa, marcelo.ochoa@frb.gov, Division of Monetary Affairs, Federal Reserve Board. Paustian, mathias.o.paustian@frb.gov, Division of Research & Statistics, Federal Reserve Board. Wilcox, laura.f.wilcox@frb.gov, Division of Monetary Affairs, Federal Reserve Board. We appreciate comments from seminar participants at the Federal Reserve Board and the Federal Reserve Bank of San Francisco. Theanalysisandconclusionssetfortharethoseoftheauthorsanddonotindicateconcurrencebyother membersoftheresearchstaffortheBoardofGovernors.

1 Introduction There is increasing scientific consensus that, as carbon dioxide (CO ) emissions and 2 global temperatures continue their upward trajectory, the potential risks from climate change are substantial. Indeed, the scientific evidence on the potential physical risks from climate change shows that the frequency and severity of extreme weather events and natural disasters—such as heat waves, droughts, floods, wildfires—are likely to increase with CO emissions and that temperature increases can reduce eco- 2 nomic growth (see the evidence in Dell, Jones, and Olken (2012), and Colacito, Hoffmann, and Phan (2019)). Mitigating the adverse effects of climate change requires substantial cuts to greenhouse emissions, which could be achieved by implementing policies that promote the transition from carbon-intensive activities to low-emission alternatives.1 Investors are beginning to recognize that climate change, and the policy responses toit,couldposearisktotheirinvestments. Itisnosurprisethatinvestmentstrategies thatscreencompaniesbasedonsomeenvironmentalcriteriahavegrowndramatically over the past years.2 Indeed, the practice of investing in companies or funds that aim to achieve market-rate financial returns, while considering positive social or environmental impact, is gaining more popularity among institutional investors. Among assetownerssurveyedbyMorganStanley, 80%saidthattheyactivelyintegratedsustainable investing in 2019, up 10 percentage points from the 2017 survey (see, Morgan Stanley’s 2019 survey). This trend reflects both a shift in preferences toward investment opportunities that would help contain the climate crisis and a growing concern abouttherisksfromclimatechange, particularlythosefrompolicyresponsestospeed up the transition to a carbon neutral economy—namely, transition risks. There is, however, no systematic empirical evidence that investing in companies 1The most recent Intergovernmental Panel on Climate Change (IPCC) report asserts that near-term actions that limit global warming could avert the projected catastrophic damages from climate change in human systems and ecosystem. For more details, see IPCC, 2022: “Climate Change 2022: Impacts, Adaptation,andVulnerability.” ContributionofWorkingGroupIItotheSixthAssessmentReportofthe IntergovernmentalPanelonClimateChange. 2In 2020, about $17 trillion—roughly one-third of all assets under professional management in the U.S.—werebeingmanagedusingsometypeofsustainable-investmentstrategy,accordingtotheUSSIF Foundation2020ReportonUSSustainableandImpactInvestingTrends. 1

with high environmental ratings or in sustainable-themed products is a good climate risk-hedging strategy because physical risks from climate change have not yet fully materialized, and policies to combat climate change are usually widely anticipated by the time they are signed into law. In this paper, we sidestep these limitations by analyzing the performance of simple dynamic investment strategies intended to hedge climate change risks around a likely exogenous change in expectations about the level of a carbon tax. Our analyses exploit the events leading to the approval of the German carbon pricing system for the transport and buildings sectors in December of 2019. The passage of this legislation presents a unique quasi-natural experiment for studying the hedging properties of common investment strategies used by environmentally minded investors. For one thing, after a last-minute round of negotiations, the climate law set a CO price at €25 (about US$ 27) per ton of carbon, which was 2 more than double than the initially planned carbon tax of €10 (about US$ 11). German legislators also agreed to steeper increases in the carbon tax over the next years. Both of these actions came as surprises to financial markets. 3 Furthermore, the increase may have also signaled the government’s commitment to strong implementation of future climate policies in Germany.4 These qualities could lead to substantial crossfirm heterogeneity in the reaction to the carbon tax news and make this event well suited to an event study. Our analysis exploits the unexpected increase in the carbon tax to assess the performance of two dynamic investing strategies that are commonly used by investors in equity markets to hedge climate risks. First, we consider a strategy that integrates a firm’s environmental pillar of the environmental, social, and governance (ESG) score to rank and remove or underweight stocks with a low environmental score in a portfolio. The second approach relies on using information about corporate CO emissions 2 to reduce the carbon footprint of an investor’s portfolio. Krueger, Sautner, and Starks 3The increase in the carbon tax was a surprise not only to financial market participants but also to expertsinGermanclimatepolicy. See,forexample,theinterviewtotheSecretaryGeneraloftheMercator Research Institute on Global Commons and Climate Change based in Berlin: Klimaschützer haben sich gegenBremserdurchgesetzt. 4For press coverage of the event, see, for example, Bloomberg, December 16, 2019, “Germans Agree CO TaxesAren’tHighEnoughandWanttoPayMore”. 2 2

(2020)reportthat,amonginstitutionalinvestors,ESGintegrationandanalysisofcarbon footprints are among the top risk management tools used hedge against climate risks as investors consider that these variables capture a firm’s exposure to climate change risks. Using a sample of German companies listed at the Frankfurt Stock ExchangethateitherreportCO emissionsorhaveemissionsestimatedfromenergyuse, 2 we investigate the effect of the news about an agreement to raise the carbon tax on the market value of investment strategies that rely on a firm’s environmental score or CO emissions. We conduct our analyses of the stock market performance of these 2 investment strategies over the following event windows: the immediate reaction to the news about the carbon tax agreement, December 16 (Monday, 1-day window); the time between the news and the passage of the climate package, including the steeper carbon tax, in the lower house of Parliament, December 16 to December 18 (3-day window); and the time between the news and the passage of the climate package in the upper house of Parliament, December 16 to December 20 (5-day window). Our main identification assumption is that the carbon tax news was both largely unexpected and the likely the main event driving asset prices around the time when the news of the carbon tax increase was reported. We use two complementary approaches to assess the effect of unexpected news about the carbon tax. We begin by constructing portfolios ranked by either a firm’s environmental score or a firm’s CO emissions and examine the response of the risk- 2 adjustedstockreturnsontheseportfoliostothehigher-than-expectedcarbontax.5 If these dynamic investing strategies are a good hedge against climate risks, an investor would expect that, in response to the announcement of a higher-than-expected carbon tax, a portfolio of stocks with a low environmental score or with a high carbon footprint would decline in value relative to a portfolio of stocks with a high environmental score or low levels of CO emissions. 2 Our portfolio analysis shows that there is little evidence that holding a portfolio 5Using a factor pricing model, we isolate the portion of risk that cannot be fully diversified and that is explained by known risk factors likely unrelated to climate policy risk. We use estimated firm-level exposurestobroadmarketmoves(CAPMmodel)aswellastotheFamaandFrench(1995)sizeandvalue factorstocomputerisk-adjustedreturns. 3

with stocks at the top of the environmental score distribution hedges the realization of climate policy risks. We find that, around the carbon tax news, the risk-adjusted returnonaportfolioofstockswiththehighestenvironmentalrating—A-orhigher— was not statistically different from a portfolio of stocks at the bottom of the rating distribution—C+ or lower. In contrast, we document that strategies that screen companies based on their CO emissions are likely to produce a good hedge to realizations 2 of climate policy risk. We find strong evidence of a negative relationship between the stock market response to the carbon tax news and a portfolio’s CO intensity as cap- 2 tured by the ratio of CO emissions to the total value of assets. A strategy that is long 2 a portfolio with low CO intensity firms and short a portfolio with high CO intensity 2 2 firms produces a positive risk-adjusted return of around 1.3% the day of the carbon tax news, around four-fifths of a standard deviation of risk-adjusted returns realized over the same event window. Interestingly, the spread in risk-adjusted returns between low and high CO intensity portfolios remains positive and increases mod- 2 estly as we expand the event window to include the subsequent votes in Parliament demonstratingthattherelativegainsinvalueofportfolioswithalowcarbonfootprint were persistent and the results are robust to substantial variations in the event study window. Next, weusefirm-levelinformationonstockpricestoperformcross-sectionalregressionsandassesstheresponseofstockpricesalongthetwodimensions—environmental scores and CO emissions—used by environmentally minded investors to hedge cli- 2 mate risks. The advantage of cross-sectional regressions over sorting stocks into portfolios is that by using firm-level observations it provides more efficient estimates of a firm’s risk exposure. At the same time, it allows us to control for potential observable firm-level characteristics unrelated to a firm’s exposure to climate change but that could explain the moves of stock prices over the event window, including industry effects. The results show that a firm’s environmental score cannot explain the change in the firm’s market value around the carbon tax news, which calls into question the view that environmental scores capture a firm’s exposure to risks from climatechange. Theinabilityofenvironmentalscorestopredicttheresponseinequity 4

prices to the carbon tax news does not seem to stem from the potential presence of measurementerrorinscores(Berg,Kölbel,Pavlova,andRigobon,2021). Weshowthat the stock price reaction to carbon tax news is essentially uncorrelated with a firm’s environmental score for the largest German firms in our sample—which have scores measured with a higher precision than the overall sample because these firms provide high quality inputs that rating agencies use to compute the scores—and for estimates suing the scores from three additional data providers. While the stock price of firms with a high environmental score did not perform any better than the stock price of a firm with a low environmental score, we find that a firm’s CO intensity—the ratio of CO emissions to the total value of assets—is a 2 2 robust predictor of the reaction of stock prices to the carbon tax news. Our estimates suggest that a one standard deviation decline in CO intensity is associated with an 2 increase in the risk-adjusted equity value of 0.3 percentage points—around a onefifth of the standard deviation of risk-adjusted returns over this window. We also show that the estimated coefficient on CO intensity increases when we expand the 2 event window to capture the carbon tax approval in the lower and upper houses of the German Parliament, suggesting that the unexpected increase in the carbon tax also conveyed new information about the likelihood of and risk from further policies aimed at reducing carbon emissions in Germany. What explains the success of carbon intensity in explaining the cross-sectional reaction in equity values to the carbon tax news? We show that CO intensity ex- 2 plains much of the cross-sectional reaction to the carbon tax news because carbon intensity predicts revisions in expected profitability after the unexpected carbon tax news. Inparticular,usinginformationfromanalysts’earningsforecastsoverthenext two years, we find that equity analysts’ revised down significantly the earnings forecasts of firms with a high CO intensity relative to their counterparts with a lower 2 CO intensity, suggesting that the reaction of stock prices over the event window re- 2 flects, to some extent, revisions in the expected profitability of firms in response to climatepolicynews. Wealsoshowthatanalystsmarkeddowntheirforecastforlongterm growth in earnings, suggesting that the carbon tax announcement changed not 5

onlyanalysts’expectationsoffirms’levelofearningsbutalsotheirgrowthtrajectory. Our study is related to a recent but growing literature studying the implications of climate change risks for sustainable investing. Engle, Giglio, Kelly, Lee, and Stroebel (2020),forexample,proposesusingaportfoliothatmimicsfluctuationsinnewsabout climate change from leading news outlets to hedge climate risks. Alekseev, Giglio, Maingi,Selgrad,andStroebel(2022)useschangesinstockholdingsfrommutualfund managers experiencing unusually high temperatures to identify stocks that are exposed to climate change risks. Our empirical evidence highlights the value of using a simple measure of climate risk exposure—a firm’s carbon footprint—to dynamically hedge risk from policies intended to reduce carbon emissions, and the risk of relying on environmental scores to identify exposure to climate change. Our results are also informative for the literature studying the theoretical implications of climate risks for portfolio allocation. Heinkel, Kraus, and Zechner (2001), for example, explore the implications of exclusionary ethical investing on equity prices. More recently, Roth Tran (2019) studies the trade-offs that a philanthropic foundation faces when deciding how much to invest in a firm whose activities might be considered objectionable, including, for example, foundations concerned about climate change investing in fossil fuel stocks. In a model featuring environmentally minded investors, Pástor, Stambaugh, and Taylor (2021) and Pedersen, Fitzgibbons, and Pomorski (2021), show that stocks with a high environmental score—green stocks— have lower expected returns than those environmentally unfriendly—brown stocks. Our empirical evidence points to carbon emissions as a relevant and informative firm characteristic to test the implications of these models. Our study builds and extends the literature employing an event study methodology to evaluate the effects of changes in climate policy through the lenses of financial markets. Ramelli, Wagner, Zeckhauser, and Ziegler (2021) uses the reaction of stock prices to the results of the U.S. 2016 and 2020 Presidential elections to explore the effects of shifts in expectations about climate policy on financial markets. Similarly, Meng (2017) uses an event study methodology around the failed attempt to pass the 6

Waxman-Markey bill—a cap-and-trade climate policy—in the U.S. Senate.6 Finally, our paper contributes to the literature asking whether financial markets reflect the potential risks from climate change. The equity market is studied in, for example, Bansal, Kiku, and Ochoa (2019), Bolton and Kacperczyk (2021), Pástor, Stambaugh, and Taylor (2022); the corporate bond market is explored in Huynh and Xia (2021), Caramichael and Rapp (2022), Duan, Li, and Wen (2021); the municipal bondmarketisstudied inPainter(2020), Goldsmith-Pinkham, Gustafson, Lewis, and Schwert (2021); and the options market in Kruttli, Tran, and Watugala (2019), Ilhan, Sautner, and Vilkov (2021). Our paper complements these studies by documenting the impactsofpoliciesthatleadtheeconomytoalowcarbontransitionthroughthelenses of financial markets. The remainder of the paper is organized as follows. In section 2, we detail our empirical approach, as well as the events around the carbon tax negotiations, and describe the data. Section 3 uses the reaction of stock prices to the unexpected carbon tax increase to assess the climate-hedging properties of two dynamic investing strategies. Section 4 uses earnings forecasts to shed some light on the reasons behind the success of carbon intensity in explaining the cross-sectional reaction in equity values to the carbon tax news. Section 5 section concludes. The Appendix presents additional results to examine the robustness of our empirical evidence. 2 Methodology and Data 2.1 Sustainable Investment Strategies Investors in equity markets are increasingly turning to information about a firm’s carbon footprint as well as to environmental, social, and governance (ESG) scores to design investment strategies that allow them to hedge the physical and transition risks from climate change. Our focus is on two investment strategies that integrate 6Ourpaperalsoaddstotheliteratureusingthehigh-frequencyresponseofassetpricestoexplorethe impacts of policy changes. Cutler (1988) explores the impact of changes in tax policy, and Snowberg, Wolfers,andZitzewitz(2007)evaluatestheeffectsofelectionoutcomesontheeconomy. 7

ESG scores or carbon emissions data directly into the security selection process. First, a currently popular strategy uses the Environmental pillar of the ESG score—the environmental score—to rank companies and remove or underweight stocks with a low environmental score. The recently created S&P 500 ESG Index, for example, uses proprietary ESG scores to define the weight in the index of a large set of companies included in the original S&P 500 Index. The underlying assumption behind this strategy is that the environmental score is a good proxy for a firm’s exposure to climate risk, which allows investors build portfolios resilient to climate risks and, at the same time, produce better long-term, risk-adjusted returns.7 Second, as a growing number of companies are disclosing their greenhouse gas emissions following widely accepted standards,8 investors are increasingly using this information to trim the carbon footprint of their portfolios and produce fully or partially“decarbonized”portfoliosbyunderweightingorexcludingstockswithrelatively high carbon emissions. Indeed, Krueger et al. (2020) report that, among institutional investors, the most frequently risk management approach to hedge against climate risks relies on the analysis of firms’ carbon footprints. This approach implicitly assumesthatcorporateCO emissionscaptureafirm’sexposuretoclimatechangerisks. 2 While a company’s environmental pillar score takes into account information on CO emissions, the environmental score also reflects other areas of environmental 2 friendlinesssuchasresourceuseandgreeninnovation,whichinmanyindustrieshave a much bigger influence than CO emissions on a company’s environmental ranking.9 2 As a result, investors rely on the environmental score to assess a company’s efforts in environmental issues beyond those that are linked to current carbon emissions — for example, the environmental score may include a firm’s plans to achieve carbon neu- 7About80%ofrespondentstoMorganStanley’s2019surveyofinvestorssaidthattheyactivelyintegrate ESG factors into the investment process, and 15% are already considering doing so. The growing interest in ESG investing reflects the view that 78% of investors responding to the same survey see financialreturnpotential. 8CompaniesanddataprovidersusethestandardssetbytheGreenhouseGasProtocoltomeasureand reportcarbonemissions. 9Most industries have a weight of about one-third on emissions in the environmental score—for example,thetransportationindustryhasanemissionsweightof0.29,utilitieshaveanemissionsweight of 0.43, and the durable goods industry has a weight of 0.35. Overall, the weight on emissions in the environmental score ranges from 0.46 to 0.22, with the weight on emissions declining with the carbon intensityofindustries. 8

tralityoverthemediumterm,soitisconsidered,tosomeextent,asaforward-looking measure of a firm’s environmental friendliness. Both the environmental score and carbon emissions are used in the literature and by investors to capture a company’s exposure to climate change risks. 2.2 Expectations of Carbon Tax Changes: the Case of Germany InSeptemberof2019,Germany’sChancellorAngelaMerkelintroducedaclimatepolicy package that included a national pricing system for carbon emissions in the transport and building sectors. The package proposed a starting carbon tax of €10 (about US$11) per ton of CO in 2021 and planned a gradual increase over the next five years, reach- 2 ing to 35 euros (about US$40) per ton of CO in 2025. After the climate package was 2 introduced, climate scientists and members of Germany’s Green Party criticized the €10 carbon tax on transport and heating as too low to effectively reduce carbon emissions from these sectors. Chancellor Angela Merkel’s government, however, saw a low initial carbon price as necessary to secure public support. OnNovember15,theproposedpricingsystemforCO emissionsfromtransportand 2 heating was approved by the German lower house of Parliament, and it was expected to pass the upper house in late 2019. Other parts of the government’s climate package of which the carbon pricing scheme was an element required a mediation agreement between the two chambers due to impacts on state tax revenues. The domestic carbon pricing scheme was not reportedly under negotiation. After debating all the weekend of December 14 and 15, however, lawmakers agreed to raise the 2021 initial carbon tax from €10 per ton of carbon to €25 per ton of carbon, and a steeper increase in the carbon tax over the next five years. Monday’s headlines reported “Germans Agree CO Taxes Aren’t High Enough and Want to Pay More,” and news coverage described 2 the debate as “grueling” and a “hard-fought compromise” that “broke a parliamentary deadlock.”10 It is possible that financial markets anticipated a compromise on 10Formoredetails,seeBloomberg,December16,2019,“GermansAgreeCO TaxesAren’tHighEnough 2 and Want to Pay More,” Phys.org, December 16, 2019, “Germany Agrees to CO Pricing Deal After Gru- 2 elingDebate.” 9

the carbon tax level. For one thing, the Green Party criticized the tax of €10 per ton of carbon and climate activists felt that a €40 per ton of carbon was more appropriate. However, meeting environmental advocates’ desired €40 per ton of carbon halfway was more than double the initial carbon tax level, so the negotiations could have easily settled on a much lower carbon tax. In fact, climate policy advocates expressed surprise at the ultimate increase in the carbon level on the day the news broke. 11 The events around the passage of this legislation present a unique quasi-natural experiment for assessing the hedging properties of sustainable investment strategies. Our identifying assumption is that the carbon tax news was both largely unexpected and the main shock driving asset prices around December 16 and is therefore well suited to an event study. Moreover, setting a higher CO price—though targeted to 2 specific sectors—might have also signaled an increased likelihood of further policies aimed at reducing carbon emissions in Germany. If so, we would expect cross-firm heterogeneity in the reaction to the carbon tax news, beyond the sectors targeted by the carbon tax of the climate policy package. 2.3 Methodology We use two complementary approaches to assess the effect of unexpected news about the carbon tax on the market value of investment strategies that screen companies based on some environmental criteria as described in Section 2.1. One, we form portfolios ranked by either a firm’s environmental score or a firm’s carbon emissions, which are used by investors to proxy for a firm’s exposure to climate change risks. If these dynamic investing strategies are a good hedge against climate risks, investors wouldexpectthatinresponsetothehigher-than-expectedcarbontax,portfolioswith a low environmental score or high carbon footprint would decline in value more than 11Forexample,Dr. Knopf,theSecretaryGeneraloftheMercatorResearchInstituteonGlobalCommons and Climate Change based in Berlin, who co-authored the MCC report on options for a carbon pricing reforminGermanythatwaspresentedtotheClimateCabinetexpressed,“Yes,Iwassurprisedthatthe measures that had been decided on were improved at all. I hadn’t expected that, despite the strong protests against the original government decision on the climate package in September,” said in an interviewwhenaskediftheresultsurprisedher. See,Spiegel,December16,2019,Klimaschützerhaben sichgegenBremserdurchgesetzt. 10

those with higher environmental scores or with a lower carbon footprint. Two, using firm-level observations, we model stock returns around the carbon tax vote as a function of a firm’s environmental score—or CO emissions—and financial 2 characteristics, ARi = φ + φ E + φ(cid:48)xi +εi (1) τ 0 1 t−1 t−1 t where ARi is the risk-adjusted cumulative stock return from the day or days around τ the carbon tax vote, E is either the environmental score of firm i or the firm i’s t−1 CO emissions reported for the fiscal year before the event period τ. The vector xi 2 t−1 includes controls for the following firm’s financial characteristics reported before the event: (log) market capitalization (ln MKTCAP); (log) price-to-book value ratio (ln PRICEBOOK); profit margin (PROFIT); the volatility of stock returns over the past 12 months (RETVOL); and an indicator if a firm participates in the EU Emissions Trading System (ETS). Finally, all regressions include industry fixed-effects for the following industry groups: consumer nondurable and durable goods; manufacturing, and mining; oil, utilities, and transportation; and services, which are categorized according to the Fama-French industry classification.12 These control variables capture the exposure of firms to other macroeconomic shocks unrelated to climate risks that could bias the estimates of the slope coefficient φ . 1 Each method, sorting stocks into portfolios and cross-sectional regressions, have its own advantages and are, in turn, complementary. Portfolio sorts do not assume a linear relationship and, in situations when the relationship between returns and the firm characteristic is unknown, portfolios are robust to misspecification. In fact, portfolio sorts can be interpreted as a nonparametric cross-sectional regression (see, Cochrane, 2011). Cross-sectional regressions, however, make use of the full data providing more efficient estimates of a firm’s risk exposure. In addition, cross-sectional regressionsallowtocontrolforfirm-levelcharacteristicsthatareknowntoexplainthe exposureofstockpricestorisksthatarelikelyunrelatedtoclimatechange. Werelyon the complementary evidence from both approaches—portfolio sorts as well as cross- 12Weconstructfiveindustrygroupsgroupingthe17industrydefinitionsfromKennethFrench’swebsite. 11

sectional regressions—to assess the performance of the two sustainable investment strategies around the carbon tax news to mitigate concerns that any shortcomings associated with one of these approaches are influencing our conclusions. Our analyses assess the performance of the characteristic-sorted portfolio and obtain the estimates of the cross-sectional regression (1) over the following event windows: the immediate reaction to the carbon tax news, December 16 (Monday, 1-day window);thetimebetweenthenewsandthepassageoftheclimatepackage,including the steeper carbon tax, in the lower house of Parliament, December 16 to December 18 (3-day window); and the time between the news and the passage of the climate package in the upper house of Parliament, December 16 to December 20 (5-day window). 2.4 Measuring Risk-adjusted Returns To isolate the portion of risk that cannot be fully diversified and that is explained by known risk factors—likely unrelated to climate risks—driving movements in equity prices, our empirical analysis uses a factor pricing model to obtain a measure of firmlevel risk-adjusted returns, namely, Ri = α +β(cid:48)f +ARi, (2) t i i t t where Ri is the return on stock i in excess of the risk-free rate, f is a vector of factors t t capturing aggregate risk. The coefficient β captures the exposure of firm i to aggrei gate risks embedded in f that cannot be fully diversified. The risk-adjusted return is t then the residual ARi = Ri−α −β(cid:48)f . Our empirical exercise uses two sets of factors to t t i i t obtain risk-adjusted returns. First, in the spirit of the CAPM model of Sharpe (1964) and Lintner (1965), we use the return on the market portfolio in excess of the riskfreerate asthefactor explainingexpectedreturns anddenotetheabnormalreturnsas CAPM-adjusted returns. Second, we use the three-factor model of Fama and French (1996) and let f be a vector with the excess return on the market portfolio, the return onaportfolioofsmallfirmsinexcessofthereturnonaportfoliooflargestocks(SMB, small minus big), and the return on high book-to-market stocks minus the return on 12

low book-to-market stocks (HML, high minus low). We denote the abnormal returns from this specification as Fama French-adjusted returns. Using risk-adjusted returns allow us to control for known differences in risk and characteristicsofstocksthatarelikelyunrelatedtotheexposureofstockstotransition risks from climate change. The three factors of Fama and French (1996), for example, have been shown to control for dimensions of risk observed in portfolios formed on earnings-to-price, cash flows-to-price, sales growth, and reversals, which are unlikely to be linked to risks from climate change (see, for example, Fama and French, 1995). At the same time, Bansal et al. (2019) document that high book-to-market portfolios are negatively exposed to temperature shocks, although the estimated risk premium is relatively small. This evidence suggests that controlling for the HML factor might lead to understate the “true” impact of carbon tax news on stock returns. On the other hand, climate-related risks are unlikely to be the key drivers of SMB and HML factors. Controlling for risks embedded in these variables, in turn, likely reduces the noise in stock returns unrelated to the carbon tax news, improving the identification of the impact of transition risks within the event study window. We estimate the factor model’s coefficients α and β for each firm using a 1-year i i sample of daily data between December 1, 2018 and December 1, 2019. We then use the estimated factor exposures β along with the estimated factor returns around the i event-study windows to compute risk-adjusted returns. Note that we also subtract α to control for differences in average returns over the year preceding the carbon tax i news that could emerge from a surge in demand for particular stocks. In particular, adjusting for α would likely capture the potential rise in value of stocks with high i environmental scores or low carbon emissions due to increased investor preference for environmentally friendly stocks before the carbon tax event.13 13In,Park,andMonk(2019),forexample,showsthatstocksofU.S.companieswithlowcarbonemissions outperformed overthe 2005-2015 period, earning, on average, anabnormal return of 3.5%–5.4% peryearrelativetohighCO emissionsstocks. 2 13

2.5 Data and Summary Statistics Our sample consists of German firms listed on the Frankfurt Stock Exchange that are part of the Prime Standard segment.14 For these group of firms, we collect firm-level environmental scores—the environmental sub-score of the ESG score—provided by Refinitiv. These scores are assigned as letter grades, which we convert to a numeric scale between 0 and 1 using Refinitiv’s grading rubric. We also collect the Carbon Disclosure Project climate change score, RobecoSAM environmental score (obtained from Eikon), and S&P IQ environmental score. For our analyses, we mostly rely on Refinitiv’senvironmentalscoreasitcoversalargernumberofpubliclytradedGerman firms and use the scores from the other data providers to mitigate the concern that our findings might depend heavily on the data provider we select. We also obtain CO emissions from Refinitiv, which are grouped in three different 2 categories: scope 1 emissions, which are direct emissions from the firm’s production; scope 2 emissions, which are the result from the generation of purchased energy; and scope 3 emissions, which are downstream emissions from product use from customers.15 We focus our analysis on scope 1 and scope 2 emissions because there are widely used greenhouse gas accounting standards that standardizes how corporations measure and report scope 1 and scope 2 emissions. Bolton and Kacperczyk (2021), for example, shows that the correlation of scope 1 and scope 2 CO emissions among five 2 data providers is very close to 1. In contrast, scope 3 emissions are rarely reported and their estimates, as noted in Busch, Johnson, and Pioch (2022) and Bolton and Kacperczyk (2021), are usually inconsistent across different data providers. The analyses in this paper are based on environmental scores and CO emissions reported for the fis- 2 cal year 2018, which are known to investors by the time when the carbon tax event occurred. The following financial characteristics of firms in our sample are obtained from 14ThePrimeStandardisasegmentofthestockmarketthatmeetsthehighestEuropeantransparency requirements. Firms part of this segment, for example, are required to produce quarterly financial reports, hold at least one analyst conference per year, and apply international accounting standards. The constituentsofthebroadstockmarketindexeswidelyfollowedbyinvestors,suchastheDAX,MDAXor SDAX,arePrimeStandard. 15ThesecategoriesfollowtheguidelinesintheGreenhouseGasProtocol 14

Refinitiv for the fiscal year 2018: market capitalization, book-to-market equity ratio, ratio of annual income to market equity. In addition, we collect information about a firm’s participation in the EU ETS. In a few cases, when the financial data for 2018 was missing, we use data for the 2017 fiscal year. We also collect the NAICS industry code for each firm to construct the industry dummy variables. The daily stock returns at the firm level are computed from daily individual stock prices obtained from Refinitiv. To obtain risk-adjusted stock returns, as described in Section 2.4, we use the daily return on the CDAX index16 as a proxy for the market return and the German sovereign yield on a 1-month security as a proxy for the riskfree rate, both series come from Haver Analytics. We also collect the three factors of Fama and French (1996)—MKT, SMB, and HML—for portfolios on European stocks fromKennethFrench’sDataLibrary,whichweconverttolocalcurrencyusingthespot USD/EUR exchange rate obtained from Bloomberg. We use one year of daily data prior to the carbon tax news to compute each stock’s CAPM-adjusted and Fama Frenchadjusted returns using the factor model (2). Consequently, our daily stock market data covers December 1, of 2018 to December 20, 2019. To select the sample of firms in our analyses, we begin by restricting our sample to firms that report CO scope 1 and 2 emissions or whose emissions are estimated 2 by Refinitiv using a firm’s energy use. 17 Next, we restrict the firms in our sample to those that have been traded in the stock exchange for at least one year before the carbon tax news. Finally, we drop from our sample companies whose stock prices are below €5 per share at the beginning of our sample, namely, those that are considered a penny stock. Our final sample consists of 115 unique companies. Table 1 provides summary statisticsforvariousmeasuresofcarbonemissions, firmfinancialcharacteristics, and 16TheCDAXisaGermanstockmarketindexthatcapturestheperformanceofallstockstradedonthe FrankfurtStockExchangethatareinthePrimeStandardandGeneralStandardsegment. 17Whenreportedemissionsarenotavailable,Refinitivestimatesemissionsusingoneofthreemethods: thefirm’scarbonemissionsfromthepreviousyear,emissionsestimatedfromenergyuse,orthemedian carbonemissionsforthefirm’sindustry. Weexclude54firmsfromoursamplefornotreportingemissions or having emissions estimated using the industry median emissions or previous years emissions. The firms that do not report CO emissions and instead have emissions estimated tend to be small—as 2 capturedbytheirmarketcapitalization—andhavealowenvironmentalscore. 15

stock returns around the carbon tax news. In terms of environmental variables, the carbonfootprintofthefirmsinoursample,capturedeitherbyCO emissionsorcarbon 2 intensity, shows important variation. The average firm in our sample produces about 4.1 million tons of scope 1 and 2 CO emissions, with a standard deviation of 15.5 mil- 2 liontons. Thecarbonintensityofafirm,measuredastheratioofCO emissionstothe 2 totalvalueofassets,isabout141tonspermillionofeuros,withastandarddeviationof 331 tons per million. Similarly, environmental scores of the firms in our sample show important variation. The mean environmental score is 0.61, with a standard deviation of 0.24. For a deeper look at the distribution of CO emissions and environmental scores, 2 Table 2 reports summary statistics across different industry groups. Using the industry definitions from Kenneth French’s website, we group our firms into four industry groups: consumer nondurable goods and consumer durable goods; manufacturing and mining; oil, utilities, and transportation; and services and other sectors. As shown in the table, industries in the service sector are the least carbon intensive, while those in the oil, utilities, and transportation sectors have on average the highest CO intensity. Interestingly, the table also shows that firms in the least carbon- 2 intensive industry group have the lowest mean environmental score, while the most carbon-intensiveindustrygrouphasthehighestenvironmentalscore. Thecorrelation between CO intensity and environmental scores within industry groups is negative 2 in the services; and oil, utilities, and transportation industry groups. The table also shows that even within industry, there is important variation of carbon intensity as well as environmental scores. Thestockmarketreactiontothecarbontaxnews,asshowninPanelBofTable1,is onaveragesmallbutthereisimportantcross-sectionalvariationinhowfirmsreacted to the carbon tax news. Finally, the average firm in our sample has a market value of €13 billion, with a standard deviation of €21 billion. The average price-to-book value ratio is 2.2, with an average deviation of 1.9. About 23% of the firms in our sample participate in the EU Emissions Trading System. 16

Table 1: Summary Statistics of Firm Characteristics and Stock Returns Around the Carbon Tax News Percentile Mean Std. Dev. 5th 95th Panel A: Firm Characteristics CO emissions (scope 1+2) 4,125 15,471 2 23,665 2 ln CO emissions intensity 9.87 2.34 5.56 13.68 2 Environmental score 0.61 0.24 0.25 0.98 Market capitalization (billion €) 12.7 20.7 0.5 68.5 Price-to-book ratio 2.19 1.91 0.56 5.53 Return volatility 1.86 0.66 0.94 3.05 Profit margin 24.32 49.66 2.16 150.98 ETS participation 0.2 0.4 0 1 Panel B: Stock Returns CAPM-Adjusted (Dec. 16) 0.19 1.57 -2.09 3.39 CAPM-Adjusted (Dec. 16–18) 0.34 3.03 -2.97 4.82 CAPM-Adjusted Day (Dec. 16–20) 0.12 3.41 -5.88 4.73 Fama French-Adjusted (Dec. 16) -0.22 1.62 -2.66 3.04 Fama French-Adjusted (Dec. 16–18) 0.12 3.10 -3.29 4.54 Fama French-Adjusted (Dec. 16–20) -0.37 3.49 -6.67 4.14 This table presents summary statistics of environmental and financial characteristics for the 115 firms listed on the Prime Standard segment of the German stock exchange that are included in our sample. Carbonemissions,environmentalscores,andfinancialcharacteristics,showninPanelA,correspondto fiscal year 2018. Carbon intensity is computed as the ratio of carbon emissions to the firm’s value of total assets, and normalized using the natural logarithm. Panel B presents summary statistics of riskadjusted returns for three different event windows in 2019: the day after news about the carbon tax increase was announced (Dec. 16), the 3-day period that encompasses the news and the passage of the climatepackageinthelowerhouseofParliament(Dec. 16–18),andthe5-dayperiodthatencompasses thecarbontaxnewsandthevoteinthelowerandupperhousesofParliament(Dec. 16–20). Returnsare expressedinpercentagepoints. 17

Table 2: Carbon Intensity and Environmental Score by Industry CarbonIntensity Env. Score Mean Std. Dev. Mean Std. Dev. Corr. Count Services & Other 7.93 2.39 0.54 0.25 -0.23 37 Consumer 10.47 1.42 0.67 0.25 0.05 25 Manufacturing & mining 10.71 1.61 0.61 0.23 0.24 42 Oil, Utilities, & Transport 11.86 2.02 0.73 0.22 -0.36 11 Thistablepresentssummarystatisticsofthecarbonintensityandtheenvironmentalscorebyindustry for the 115 firms listed on the Prime Standard segment of the German stock exchange that are included inoursample. Carbonintensityiscomputedastheratioofcarbonemissionstothefirm’svalueoftotal assets,andnormalizedusingthenaturallogarithm. Thedatacorrespondtofiscalyear2018. 3 The Effect of the Carbon Tax News on Market Values of Sustainable Investment Strategies 3.1 Evidence from Portfolio Sorts Portfoliosortsareanimportantandpopulartoolinempiricalfinancetoevaluateasset pricing models and to identify potential profitable investment strategies. It is conventional practice, for example, to form portfolios ranked by some characteristic and test whether expected returns vary systematically with such characteristic.18 We followthispopularpracticeandformportfoliosaccordingtoafirm’sexposuretoclimate change risks. We consider two proxy variables for a firm’s exposure to climate change risks that are popular among environmentally minded investors: a firm’s environmental score, and a firm’s CO footprint. If these two dynamic investing strategies 2 are a good hedge against climate risks, investors would expect that in response to the announcement of a higher-than-expected carbon tax, portfolios with a low environmental score or a high carbon footprint would decline in value more than those with higher environmental scores or with a lower carbon footprint. To test this hypothesis, Table 3 presents CAPM-adjusted returns (Panel A) and Fama French-adjusted returns (Panel B) on three portfolios around the carbon tax 18TheCAPM,forexample,impliesapositiverelationshipbetweenexpectedreturnsandmarketbetas. 18

news. Firms are grouped into each portfolio based on their environmental score: the constituents of the A-rated portfolio have a score of A- or higher, the B-rated portfolio is comprised of firms with a score between B- and B+, and the stocks in the C-rated & below portfolio have scores below C+. Table 3 also reports the spread in returns between the portfolios with the highest and lowest environmental scores, which corresponds to a strategy that is long stocks of highly-rated firms and short firms at the bottom of the environmental score distribution. Finally, we report, in brackets, its associated 90 percent confidence interval. Each panel presents the portfolio returns for three different event windows: the day after news about the carbon tax increase was announced (Dec. 16), the 3-day period that encompasses the news and the passage of the climate package in the lower house of Parliament (Dec. 16–18), and the 5-dayperiodthatencompassesthecarbontaxnewsandthevoteinthelowerandupper houses of Parliament (Dec. 16–20). The last rows of Table 3 report the following characteristics for each portfolio: number of stocks, the average (log) CO emissions, 2 CO intensity(ratioofCO emissionstototalassets),theaverageenvironmentalscore, 2 2 and the average (log) market value. Our sort on environmental scores in Table 3 provides little evidence that holding a portfolio with stocks at the top of the environmental score distribution hedges the realization of climate policy risks. In particular, the portfolio with the A-rated stocks does not seem to have experienced higher returns relative to the portfolio holding stocks with low environmental scores around the December 16 event. In fact, the Crated & below portfolio seems to outperform after the carbon tax news (Dec. 16) and this superior performance as measured by the return spread between high and low environmentally rated portfolios is statistically significant for both measures of riskadjusted returns. The estimated return spread suggests that an investor long the Arated portfolio and short the C-rated&below would have experienced a negative return around 0.6%. Over longer event windows, the return spread between the A-rated and C-rated&belowportfolioisnotstatisticallysignificantforbothCAPM-adjustedreturns or Fama French-adjusted returns. Table 4 examines the stock market performance around the carbon tax news of 19

Table 3: Cumulative Risk-Adjusted Returns on Portfolios Sorted on a Firm’s Environmental Score Around the Carbon Tax News Panel A: CAPM-adjusted A-rated B-rated C-rated & below A-rated minus C-rated (1) (2) (3) (1) − (3) Dec. 16 -0.01 -0.15 0.54 -0.55 [ -1.05, -0.07] Dec. 16–18 0.69 -0.01 0.31 0.38 [ -0.81, 1.17] Dec. 16–20 0.57 -0.38 0.11 0.46 [ -0.64, 1.63] Panel B: Fama French-adjusted A-rated B-rated C-rated and below A-rated minus C-rated (1) (2) (3) (1) − (3) Dec 16 -0.45 -0.60 0.16 -0.61 [ -1.16, -0.10] Dec 16–18 0.31 -0.16 0.17 0.14 [ -0.86, 1.18] Dec 16–20 0.04 -0.91 -0.33 0.37 [ -0.80, 1.53] Portfolio Characteristics A-rated B-rated C-rated and below No. of Stocks 35 30 50 ln Market Value 25.72 23.57 22.70 ln CO emissions 16.17 14.77 13.28 2 CO intensity 11.17 11.93 11.31 2 Environmental score 0.91 0.66 0.38 This table presents CAPM and Fama French-adjusted equally-weighted returns on three portfolios around the carbon tax news sorted according to environmental scores. Portfolio returns are presented for the day the carbon tax increase is announced (December 16, 2019), as well as the three and fivedayperiodbeginningwiththecarbontaxnews. Thelastcolumnreportsthereturnspreadbetweenthe portfolios with the highest and lowest environmental scores and, in brackets, its associated 90 percent confidence interval. The confidence intervals are obtained using a bootstrap methodology with 1000 samples. 20

the following three portfolios sorted on CO emissions intensity of a firm, which we 2 defineastheratioofCO emissionstothetotalvalueofassets: Lowemitters(firmsin 2 the bottom 30%), Neutral (firms in the middle 40%), and High emitters (firms in the upper 30%). Our sort of stocks on CO emissions intensity produces a strong nega- 2 tive relationship between the stock market response to the carbon tax news (Dec. 16) and the carbon footprint of the portfolio. For one thing, the portfolio of low emitters increases in value the day when the unexpected increase in the carbon tax is announced, while the portfolio of high emitters declines in value. The estimated return spread suggests that a strategy long low carbon intensity firms and short high carbon intensity firms leads to a positive return of around 1.3% the day of the carbon day news, which is highly statistically significant and represents around four-fifths of a standard deviation of risk-adjusted returns over this event window. Moreover, as shown in Figure 1, the return spread between low and high carbon intensive portfolios remains positive and increases modestly as we expand the event window. The results are generally supportive of the effectiveness of investment strategies that use a firm’s carbon intensity to screen firms in search for investment opportunities that hedge risks from policies to combat climate change. As shown in the Appendix, our conclusions are robust to reasonable variations to the way we construct the climate change risk-hedging portfolios. First, we compute value-weighted risk-adjusted returns to alleviate the concern that our results are driven by small firms with volatile returns. Second, we use environmental scores from other prominent data providers as an alternative proxy for a firm’s climate risk exposure. Third, we form portfolios using alternative proxy variables for a firm’s carbon footprint; namely, CO emissions, and the ratio of CO emissions to the total 2 2 market value. 3.2 Evidence from Cross-Sectional Regressions We now assess our claim that companies with a higher environmental score or a smaller carbon footprint should outperform in response to the carbon tax news using 21

Table 4: Cumulative Risk-Adjusted Return on Portfolios Sorted on CO Emissions In- 2 tensity Around the Carbon Tax News Panel A: CAPM-adjusted Low emitters Neutral High emitters Low minus High Dec 16 0.88 0.09 -0.34 1.22 [ 0.36, 1.73] Dec 16–18 1.59 -0.76 0.58 1.01 [ -0.12, 2.21] Dec 16–20 1.57 -0.76 -0.12 1.70 [ 0.34, 3.02] Panel B: Fama French-adjusted Low emitters Neutral High emitters Low minus High Dec 16 0.54 -0.39 -0.73 1.27 [ 0.40, 1.75] Dec 16–18 1.43 -0.91 0.22 1.21 [ -0.03, 2.15] Dec 16–20 1.16 -1.28 -0.66 1.82 [ 0.41, 3.22] Portfolio Characteristics Low Mid High No. of Stocks 34 46 35 ln Market Value 23.20 23.42 23.09 ln CO emissions 11.51 13.22 16.36 2 CO intensity 6.79 9.74 13.31 2 Environmental score 0.58 0.59 0.67 This table presents CAPM and Fama French-adjusted equally-weighted returns on three portfolios around the carbon tax news sorted according to carbon emissions intensity. Portfolio returns are presented for the day the carbon tax increase is announced (December 16, 2019), as well as the three and five-dayperiodbeginningwiththecarbontaxnews. Thelastcolumnreportsthereturnspreadbetween the portfolios with the highest and lowest emissions intensity and, in brackets, its associated 90 percentconfidenceinterval. Theconfidenceintervalsareobtainedusingabootstrapmethodologywith1000 samples. 22

3.5 Low minus High Carbon Emissions 3 2.5 2 1.5 1 0.5 0 Dec 16 Dec 17 Dec 18 Dec 19 Dec 20 Figure 1: Fama French-adjusted Cumulative Returns Around the Carbon Tax News and CO Emissions Intensity 2 This figure shows the spread between cumulative returns on portfolios with lowest and highest carbon emissions intensity around the carbon tax news. Returns are Fama French-adjusted and are shown for eachdayintheeventwindowbeginningDecember16andendingDecember20. 23

a cross-sectional regression model (see equation (1)). We begin by estimating the variation in the response of firms’ stock returns to the carbon tax news with firms’ environmental scores. To ease the interpretation of the coefficient estimate on the environmental score (ENVSCORE), we standardize this variable such that the cross-sectional standard deviation of the environmental score is equal to 1. We estimate the cross-sectional regression using the risk-adjusted returns around the carbon tax news and we control for the following firm’s financial characteristics: (log) market capitalization (ln MKTCAP); (log) price-to-book value ratio (ln PRICEBOOK); profit margin (PROFIT); the volatility of stock returns over the past 12 months (RETVOL); and an indicator if a firm participates in the EU Emissions TradingSystem(ETS).Wealsoincludeindustryfixed-effectsforthefollowingindustry groups: consumer nondurable and durable goods; manufacturing, and mining; oil, utilities, and transportation; and services. We also include the firm’s exposure to the SMB and HML factors as captured by their estimated betas (BSMB, BHML) when we use as dependent variable the CAPM-adjusted returns. ShowninTable5,theestimationresultsprovidenoevidencethatfirmswithahigh environmental score performed any better than those with a low environmental score around the carbon tax news. The model performs poorly in two ways. First, environmentalscoresdonotexplaintheimmediateresponseofstockpricestotheunexpected increase in the carbon tax. Column 1 of Panels A and B displays the estimates of the cross-sectional regression for the day news of the last-minute agreement to increase thecarbontaxbroke(Dec. 16). Estimatesofthecoefficientontheenvironmentalscore show that one cannot reject the hypothesis that differences in the initial reaction of stock prices to the carbon tax news are unrelated to a firm’s environmental score for both CAPM- and Fama French-adjusted returns. Second, estimates of the response of stock prices over the 5-day period that encompasses the carbon tax news and the vote inthelowerandupperhousesofParliament—theresultsincolumns2and3—provide mixedevidenceofapositiverelationshipbetweenthestockpricereactionandtheenvironmentalscore. WhiletheCAPM-andFama-adjustedreturnsarepositivelyrelated to a firm’s environmental score, a one-tailed hypothesis test suggests that in most 24

cases we cannot reject the null hypothesis that the coefficient on the environmental score is negative or zero under conventional confidence levels. The lack of strong evidence of a positive relationship between the reaction to the carbon tax news and the environmental score might reflect the fact that environmental scores are measured with noise. Berg et al. (2021) argues that regression estimates of stock returns on ESG scores are biased toward zero and the bias increases with the noise in the estimated ESG measure, which is also likely true for the environmental component of ESG scores.19 Thus, to reduce the amount of measurement error in the environmental scores, we restrict our sample to the constituents of the DAX index, which are the largest firms traded in the Frankfurt Stock Exchange; it stands to reasonthattheenvironmentalperformanceofthesefirmsislikelymeasuredwithhigher precision because these firms provide high quality inputs that rating agencies use to compute the scores.20 In addition, we also estimate the cross-sectional regressions using the environmental scores from four different rating agencies: Refinitiv, Carbon Disclosure Project, RobecoSAM, and S&P Global, which provide alternative and complementary information about a firm’s climate risk. Table 6 presents the estimates from this exercise. Column 1 presents the estimated coefficients using the immediate stock market reaction to the carbon tax news (Dec. 16), and column 2 reports the cross-sectional regression estimates using the 5-day cumulative return (Dec. 16–20). Panel A and Panel B present the results using CAPM- and Fama French-adjusted returns, respectively.21 We continue to find little support of a positive relationship between the immediate stock market reaction and the environmental score across all the rating agencies in our sample. Most notably, the coefficient on the environmental score, shown in column 2, is economically and statistically smaller than the baseline estimates using the full sample. The insensitivity of our baseline results using the full sample to restricting the sample to the 19Bergetal.(2021)proposesrelyingoncomplementaryratings. Inparticular,Bergetal.(2021)develops an instrumental variable approach to overcome the problem of measurement error. The instrument is constructedfromenvironmentalscoresfromseveraldataproviders. 20The German firms listed in the DAX, for example, produce corporate social responsibility reports, whichareakeyinputtoESGratingagencies. 21These regressions do not include industry fixed effects because of the small sample size in each industrycategory. 25

Table 5: Stock Returns Response to the Carbon Tax News and Environmental Scores PanelA:CAPM-adjustedReturn PanelB:FamaFrench-adjustedReturn (1) (2) (3) (1) (2) (3) Dec. 16 Dec. 16–18 Dec. 16–20 Dec. 16 Dec. 16–18 Dec. 16–20 ENVSCORE 0.017 0.879* 1.149** -0.068 0.701 0.826* (0.174) (0.509) (0.463) (0.174) (0.469) (0.463) ln PRICEBOOK 0.580** 0.814 0.903* 0.426* 0.954** 1.245*** (0.234) (0.525) (0.542) (0.215) (0.415) (0.439) ln MKTCAP -0.302* -0.868** -0.951*** -0.218* -0.657** -0.560** (0.159) (0.350) (0.313) (0.118) (0.269) (0.264) RETVOL -0.338 -0.011 -0.292 -0.496* -0.472 -1.204** (0.254) (0.340) (0.433) (0.282) (0.346) (0.494) PROFIT 0.005* 0.007 0.012** 0.006* 0.007* 0.012** (0.003) (0.005) (0.005) (0.003) (0.004) (0.005) ETS 0.569** 0.768 0.984 0.591** 0.901 1.168* (0.267) (0.527) (0.619) (0.288) (0.571) (0.679) R2 0.19 0.10 0.27 0.18 0.10 0.22 Obs. 115 115 115 115 115 115 Thistablepresentstheestimatedcoefficientsfromthefollowingcross-sectionalregression: ARi = φ + φ Ei + φ(cid:48)xi +εi τ 0 1 t−1 t−1 t wherethedependentvariableisfirmi’sstockmarketreturnaroundthecarbontaxnews,CAPM-adjusted (Panel A) and Fama French-adjusted (Panel B). The key explanatory variable is a firm’s environmental score (ENVSCORE), which we standardize to have a cross-sectional standard deviation equal to 1. The vectorxi includescontrolsforthefollowingfirm’sfinancialcharacteristicsreportedbeforetheevent: t−1 (log)marketcapitalization(lnMKTCAP);(log)price-to-bookvalueratio(lnPRICEBOOK);profitmargin (PROFIT); the volatility of stock returns over the past 12 months (RETVOL); and an indicator if a firm participatesintheEUEmissionsTradingSystem(ETS).Weincludethefirm’sexposuretotheSMBand HML factors as captured by their estimated betas (BSMB, BHML) when we use as dependent variable theCAPM-adjustedreturns. Allregressionsincludeindustryfixed-effectsconstructedusingtheFama- French industry classification. The sample includes German publicly traded firms that are part of the PrimeStandardsegmentandthateitherreporttheircarbonemissionsorhaveemissionsestimatedfrom energy use. Standard errors robust to heteroskedasticity are shown in parentheses. *** p<0.01; ** p<0.05;*p<0.1. 26

largest German firms and to using scores from different data providers strengthens our confidence in our conclusion that environmental scores do not explain the observed reaction in asset valuations to the carbon tax news. Moreover, measurement error in environmental scores does not seem to be the key explanation for the lack of correlation between environmental scores and the stock market performance of firms around the carbon tax news since environmental scores of the largest companies are likely measured with better precision relative to the rest of the sample. We now turn to examining the ability of a firm’s CO intensity to predict the stock 2 returnreactiontothecarbontaxnews. Theestimatesincolumn1ofTable7showthat, contrary to the results for the environmental score, a firm’s carbon emissions intensityexplainstheinitialstockpricereactiontothecarbontaxnewsforbothCAPM-and Fama French-adjusted returns. The estimated coefficient on CO intensity suggests 2 that a one standard deviation decline in CO intensity is associated with an increase in 2 market value of 0.3 percentage points, which is equivalent to a one-fifth of the standard deviation of risk-adjusted returns over this window. As the horizon over which theregressionisestimatedincreases,shownincolumns2and3,theeconomicimportanceofcarbonemissionsintensityrises,withtheestimatedcoefficientsuggestingan increase in risk-adjusted returns of 0.8 percentage points over a five-day window— around a one-third of a standard deviation of the variation of risk-adjusted returns overthiseventwindow—fromadeclineinaonestandarddeviationofafirm’scarbon emissions intensity. The increase in the coefficient on CO intensity as we expanded 2 the event window to capture the votes in the lower and upper houses of the German Parliament suggests that the political agreement to increase the carbon tax (on Dec. 18 and Dec. 20) probably also conveyed new information about the likelihood of and risk from further policies aimed at reducing carbon emissions in Germany. InTable8,weassessthesensitivityofourresultstotwovariationsinthesampleby estimating the cross-sectional regression. First, restricting our sample to the largest publicly traded firms in Germany, and, second, by excluding firms “targeted” by the carbontax(firmsintheoil, utilities, andtransportationsector). Shownincolumn1of panelsAandB,thecoefficientestimateonCO emissionsintensityforasampleofthe 2 27

sweN xaT nobraC eht ot smriF namreG tsegraL eht fo esnopseR snruteR kcotS :6 elbaT nruteRdetsujda-MPAC:AlenaP labolGP&S MASoceboR PDC vitinifeR )2( )1( )2( )1( )2( )1( )2( )1( 02–61 .ceD 61 .ceD 02–61 .ceD 61 .ceD 02–61 .ceD 61 .ceD 02–61 .ceD 61 .ceD 672.0- 080.0- 502.0- 310.0- **545.2- **407.0- 472.0 223.0- EROCSVNE )251.1( )173.0( )108.0( )962.0( )108.0( )552.0( )986.0( )043.0( 740.1- 370.0- 340.1- 040.0 073.0 606.0 385.0- 992.0- KOOBECIRP nl )353.1( )645.0( )013.1( )594.0( )400.1( )224.0( )265.1( )596.0( 632.0- 111.0- 172.0- 060.0- 472.0- 760.0- 705.0- 040.0 PACTKM nl )035.1( )744.0( )299.0( )913.0( )149.0( )762.0( )317.0( )242.0( 063.1- 129.0- 700.1- 912.0- *889.1- ***052.1- 146.0- 583.0- LOVTER )954.1( )935.0( )499.0( )993.0( )512.1( )993.0( )707.0( )482.0( 500.0- 300.0 100.0- 600.0 960.0 900.0 210.0 200.0 TIFORP )840.0( )510.0( )430.0( )310.0( )520.0( )010.0( )120.0( )900.0( 493.0 332.0 944.0 483.0 264.1 906.0 346.0 853.0 STE )218.0( )413.0( )748.0( )483.0( )090.1( )493.0( )706.0( )332.0( 32.0 91.0 32.0 81.0 74.0 34.0 32.0 22.0 2R 62 62 72 72 52 52 82 82 .sbO 28

).tnoc( sweN xaT nobraC ot smriF namreG tsegraL eht fo esnopseR snruteR kcotS :6 elbaT nruteRdetsujda-hcnerFamaF:BlenaP labolGP&S MASoceboR PDC vitinifeR )2( )1( )2( )1( )2( )1( )2( )1( 02–61 .ceD 61 .ceD 02–61 .ceD 61 .ceD 02–61 .ceD 61 .ceD 02–61 .ceD 61 .ceD 722.0- 620.0 970.0- 170.0 **698.1- **017.0- 148.0 366.0- EROCSVNE )129.0( )543.0( )657.0( )292.0( )628.0( )062.0( )700.1( )784.0( 773.0 2790.0- 787.0 152.0 488.0 960.0 771.1 960.0- KOOBECIRP nl )112.1( )764.0( )420.1( )293.0( )077.0( )713.0( )422.1( )505.0( 156.0- 322.0- 757.0- 251.0- 260.0- 411.0 098.0- 401.0- PACTKM nl )941.1( )104.0( )758.0( )823.0( )209.0( )392.0( )276.0( )442.0( 226.1- 678.0- 393.1- 000.0 855.3- **953.1- ***430.1- 103.0- LOVTER )734.1( )785.0( )859.0( )934.0( )823.1( )874.0( )488.0( )383.0( 727.0 710.0 230.0 **220.0 200.0 700.0 *240.0 010.0 TIFORP )330.0( )010.0( )430.0( )110.0( )820.0( )800.0( )420.0( )410.0( 159.0 283.0 019.0 526.0 652.1 895.0 040.1 804.0 STE )217.0( )423.0( )428.0( )973.0( )560.1( )104.0( )286.0( )913.0( 81.0 71.0 61.0 11.0 34.0 04.0 81.0 31.0 2R 62 62 72 72 52 52 82 82 .sbO :noissergerlanoitces-ssorcgniwollofehtmorfstneiciffeocdetamitseehtstneserpelbatsihT iε+ ix(cid:48)φ + iE φ + φ = iRA t 1−t 1−t 1 0 τ detsujda-hcnerFamaFdna)AlenaP(detsujda-MPAC,swenxatnobracehtdnuoranrutertekramkcotss’imrifsielbairavtnednepedehterehw dradnats lanoitces-ssorc a evah ot ezidradnats ew hcihw ,)EROCSVNE( erocs latnemnorivne s’mrif a si elbairav yrotanalpxe yek ehT .)B lenaP( tekram )gol( :tneve eht erofeb detroper scitsiretcarahc laicnanif s’mrif gniwollof eht rof slortnoc sedulcni ix rotcev ehT .1 ot lauqe noitaived 1−t tsapehtrevosnruterkcotsfoytilitaloveht;)TIFORP(nigramtiforp;)KOOBECIRPnl(oitareulavkoob-ot-ecirp)gol(;)PACTKMnl(noitazilatipac eht ot erusopxe s’mrif eht edulcni eW .)STE( metsyS gnidarT snoissimE UE eht ni setapicitrap mrif a fi rotacidni na dna ;)LOVTER( shtnom 21 ehT .snruterdetsujda-MPACehtelbairavtnednepedsaesuewnehw)LMHB,BMSB(satebdetamitseriehtybderutpacsasrotcafLMHdnaBMS ;10.0<p*** .sesehtnerapninwohserayticitsadeksoretehottsuborsrorredradnatS .xedniXADehtfostneutitsnoceratahtsmrifsedulcnielpmas .1.0<p*;50.0<p** 29

largest firms in Germany is quite similar to the estimate using our baseline sample, whichsuggeststhatthereportedandestimatedcarbonemissionsinoursamplearenot prone to measurement error. This is not very surprising given that firms and rating agencies follow the guidelines in the Greenhouse Gas Protocol to report emissions. The coefficient estimate on CO intensity excluding the firms in the oil, utilities, and 2 transportation sector, shown in column 2, is also very close to the estimates from the baseline model, lending support to the hypothesis that the carbon tax news likely led to a reassessment of the government’s commitment to strong implementation of future policies to curb carbon emissions. The empirical evidence presented so far demonstrates that a firm’s carbon emissionsintensityisarobustpredictorofafirm’sstockmarketreactiontothecarbontax news, while environmental scores are not. One explanation for the lack of predictive power of environmental scoresmaybe a potential adjustment in asset pricesin anticipationofanincreaseinthecarbontaxtheweekpoliticiansannouncedthattheywere going to negotiate some aspects of the climate package. To assess this, we run the cross-sectional regression for the days following the announcement of the mediation process (Dec. 9 and Dec. 9–11). As shown in Table 9, the response of stock prices is unrelated to CO emissions intensity, which suggests that the carbon tax increase 2 was likely unexpected. Importantly, environmental scores are not related to the stock return reaction over that period, confirming that our initial finding is unlikely due to markets having already priced in the increase in the carbon tax the days before the announcement. 4 Carbon Tax News and Earnings Forecasts Havingshownthatcarbonintensitydoesabetterjobinexplainingthecross-sectional reaction in equity values to the carbon tax news than environmental scores, we now ask what lies behind the success of an investment strategy that uses a firm’s carbon footprint to hedge the unexpected increase in the carbon tax. This section demonstrates that carbon intensity explains much of the cross-sectional reaction to the car- 30

ytisnetnI nobraC dna sweN xaT nobraC ot esnopseR snruteR kcotS :7 elbaT nruteRdetsujda-hcnerFamaF:BlenaP nruteRdetsujda-MPAC:AlenaP )3( )2( )1( )3( )2( )1( 02–61 .ceD 81–61 .ceD 61 .ceD 02–61 .ceD 81–61 .ceD 61 .ceD **703.0- **872.0- *331.0- *503.0- *472.0- **541.0- YTISNETNI OC 2 )941.0( )411.0( )960.0( )551.0( )611.0( )560.0( ***468.2 **609.0 **874.0 127.0 496.0 ***146.0 KOOBECIRP nl )314.0( )873.0( )312.0( )825.0( )944.0( )822.0( *682.0- ***493.0- ***372.0- *093.0- ***944.0- ***133.0- PACTKM nl )271.0( )841.0( )290.0( )112.0( )461.0( )021.0-( *399.0- 092.0- *584.0- 531.0- 611.0 213.0- LOVTER )515.0( )073.0( )482.0( )564.0( )163.0( )952.0( ***310.0 **800.0 *600.0 **210.0 *700.0 *600.0 TIFORP )500.0( )400.0( )300.0( )500.0( )400.0( )300.0( ***379.1 ***316.1 **287.0 ***669.1 ***485.1 ***508.0 STE )027.0( )116.0( )703.0( )446.0( )775.0( )882.0( 22.0 01.0 02.0 52.0 90.0 22.0 2R 511 511 511 511 511 511 .sbO :noissergerlanoitces-ssorcgniwollofehtmorfstneiciffeocdetamitseehtstneserpelbatsihT iε+ ix(cid:48)φ + iE φ + φ = iRA t 1−t 1−t 1 0 τ detsujda-hcnerFamaFdna)AlenaP(detsujda-MPAC,swenxatnobracehtdnuoranrutertekramkcotss’imrifsielbairavtnednepedehterehw snoissime OCfooitarehtfogolehtsaderusaem)YTISNETNI OC(ytisnetnisnoissimenobracs’mrifasielbairavyrotanalpxeyekehT.)BlenaP( 2 2 tekram )gol( :tneve eht erofeb detroper scitsiretcarahc laicnanif s’mrif gniwollof eht rof slortnoc sedulcni ix rotcev ehT .stessa latot ot 1−t tsapehtrevosnruterkcotsfoytilitaloveht;)TIFORP(nigramtiforp;)KOOBECIRPnl(oitareulavkoob-ot-ecirp)gol(;)PACTKMnl(noitazilatipac eht ot erusopxe s’mrif eht edulcni eW .)STE( metsyS gnidarT snoissimE UE eht ni setapicitrap mrif a fi rotacidni na dna ;)LOVTER( shtnom 21 llA .snruter detsujda-MPAC eht elbairav tnedneped sa esu ew nehw )LMHB ,BMSB( sateb detamitse rieht yb derutpac sa srotcaf LMH dna BMS ylcilbup namreG sedulcni elpmas ehT .noitacifissalc yrtsudni hcnerF-amaF eht gnisu detcurtsnoc stceffe-dexif yrtsudni edulcni snoisserger ygrenemorfdetamitsesnoissimeevahrosnoissimenobracriehttroperrehtietahtdnatnemgesdradnatSemirPehtfotraperatahtsmrifdedart .1.0<p*;50.0<p**;10.0<p*** .sesehtnerapninwohserayticitsadeksoretehottsuborsrorredradnatS .esu 31

Table 8: Sensitivity of the Relationship Between Stock Returns Response to Carbon Tax News and CO Intensity 2 PanelA:CAPM-adjustedReturn PanelB:FamaFrench-adjustedReturn (1) (2) (1) (2) DAX firms Ex. Fossil Fuels DAX firms Ex. Fossil Fuels CO INTENSITY -0.207** -0.163** -0.227*** -0.150** 2 (0.079) (0.071) (0.069) (0.074) ln PRICEBOOK 0.433 0.691** 0.387 0.506** (0.363) (0.277) (0.244) (0.239) ln MKTCAP -0.349 -0.359*** -0.396* -0.301*** (0.243) (0.126) (0.230) (0.097) RETVOL -0.098 -0.415 -0.073 -0.582* (0.235) (0.288) (0.264) (0.304) PROFIT 0.000 0.006* 0.003 0.006* (0.011) (0.003) (0.012) (0.003) ETS 0.757** 0.813** 0.787** 0.805** (0.329) (0.311) (0.360) (0.341) R2 0.41 0.22 0.36 0.21 Observations 28 104 28 104 Thistablepresentstheestimatedcoefficientsfromthefollowingcross-sectionalregression: ARi = φ + φ Ei + φ(cid:48)xi +εi τ 0 1 t−1 t−1 t wherethedependentvariableisfirmi’sstockmarketreturnaroundthecarbontaxnews,CAPM-adjusted (Panel A) and Fama French-adjusted (Panel B). The key explanatory variable is a firm’s carbon emissions intensity (CO INTENSITY) measured as the log of the ratio of CO emissions to total assets. The 2 2 vectorxi includescontrolsforthefollowingfirm’sfinancialcharacteristicsreportedbeforetheevent: t−1 (log)marketcapitalization(lnMKTCAP);(log)price-to-bookvalueratio(lnPRICEBOOK);profitmargin (PROFIT); the volatility of stock returns over the past 12 months (RETVOL); and an indicator if a firm participatesintheEUEmissionsTradingSystem(ETS).Weincludethefirm’sexposuretotheSMBand HML factors as captured by their estimated betas (BSMB, BHML) when we use as dependent variable the CAPM-adjusted returns. The sample in column 1 includes German publicly traded firms that are part of the DAX; the sample in column 2 excludes firms in the oil, utilities, and transportation sector. Regressions in column 2 include industry fixed-effects constructed using the Fama-French industry classification. Standard errors robust to heteroskedasticity are shown in parentheses. *** p<0.01; ** p<0.05;*p<0.1. 32

sweN xaT nobraC fo tnemecnuonnA erofeB noitcaeR tekraM kcotS laitnetoP ot ytivitisneS :9 elbaT nruteRdetsujda-hcnerFamaF:BlenaP nruteRdetsujda-MPAC:AlenaP )4( )3( )2( )1( )4( )3( )2( )1( 21–9 .ceD 9 .ceD 21–9 .ceD 9 .ceD 21–9 .ceD 9 .ceD 21–9 .ceD 9 .ceD 801.0 680.0 821.0 390.0 YTISNETNI OC 2 )990.0 ( )360.0 ( )890.0 ( )560.0 ( 782.0- 602.0- 404.0- 171.0- EROCSVNE )533.0 ( )391.0 ( )892.0 ( )671.0 ( 377.0- *272.0- *357.0- 062.0- ***341.1- ***116.0- ***980.1- ***706.0- KOOBECIRP nl )224.0 ( )442.0 ( )544.0 ( )842.0 ( )434.0 ( )682.0 ( )244.0 ( )472.0 ( 961.0- 901.0 *872.0- 230.0 080.0 370.0 301.0- 600.0 PACTKM nl )142.0 ( )131.0 ( )561.0 ( )790.0 ( )762.0 ( )561.0 ( )602.0 ( )331.0 ( *218.0- 670.0 *688.0- 220.0 954.0- 692.0 735.0- 752.0 LOVTER )674.0 ( )872.0 ( )405.0 ( )182.0 ( )344.0 ( )662.0 ( )744.0 ( )652.0 ( 000.0 000.0- 000.0- 000.0- 200.0- 000.0- 200.0- 100.0- TIFORP )300.0 ( )100.0 ( )300.0 ( )100.0 ( )300.0 ( )100.0 ( )300.0 ( )100.0 ( 811.0- 780.0 104.0- 031.0- 560.0 850.0 013.0- 161.0- STE )176.0 ( )123.0 ( )556.0 ( )363.0 ( )156.0 ( )472.0 ( )636.0 ( )213.0 ( 511.0 720.0 511.0 720.0 111.0 490.0 111.0 490.0 2R 511 511 511 511 511 511 511 511 snoitavresbO :noissergerlanoitces-ssorcgniwollofehtmorfstneiciffeocdetamitseehtstneserpelbatsihT iε+ ix(cid:48)φ + iE φ + φ = iRA t 1−t 1−t 1 0 τ detsujda-hcnerFamaFdna)AlenaP(detsujda-MPAC,swenxatnobracehtdnuoranrutertekramkcotss’imrifsielbairavtnednepedehterehw snoissime OCfooitarehtfogolehtsaderusaem)YTISNETNI OC(ytisnetnisnoissimenobracs’mrifasielbairavyrotanalpxeyekehT.)BlenaP( 2 2 tekram )gol( :tneve eht erofeb detroper scitsiretcarahc laicnanif s’mrif gniwollof eht rof slortnoc sedulcni ix rotcev ehT .stessa latot ot 1−t tsapehtrevosnruterkcotsfoytilitaloveht;)TIFORP(nigramtiforp;)KOOBECIRPnl(oitareulavkoob-ot-ecirp)gol(;)PACTKMnl(noitazilatipac eht ot erusopxe s’mrif eht edulcni eW .)STE( metsyS gnidarT snoissimE UE eht ni setapicitrap mrif a fi rotacidni na dna ;)LOVTER( shtnom 21 llA .snruter detsujda-MPAC eht elbairav tnedneped sa esu ew nehw )LMHB ,BMSB( sateb detamitse rieht yb derutpac sa srotcaf LMH dna BMS ylcilbup namreG sedulcni elpmas ehT .noitacifissalc yrtsudni hcnerF-amaF eht gnisu detcurtsnoc stceffe-dexif yrtsudni edulcni snoisserger ygrenemorfdetamitsesnoissimeevahrosnoissimenobracriehttroperrehtietahtdnatnemgesdradnatSemirPehtfotraperatahtsmrifdedart .1.0<p*;50.0<p**;10.0<p*** .sesehtnerapninwohserayticitsadeksoretehottsuborsrorredradnatS .esu 33

bontaxnewsbecausecarbonintensitypredictsrevisionsinexpectedprofitabilityafter the unexpected carbon tax news while environmental scores do not. Wecollectdataonanalysts’forecastsfromtheInstitutionalBrokers’EstimateSystem(I/B/E/S)providedbyRefinitiv.22 Ourempiricalanalysisusesforecastsofearnings per share over the next two years, and forecasts of long-term growth in earnings that represent annual growth over the next three to five years. We use monthly consensus forecasts reported a month before the carbon tax news (November 13, 2019) and a month after the event (January 16, 2020). Using information of forecasts for earnings per share over the next years, we compute the revision in earnings forecasts over the nexttwoyearsasthechangeinearningsforecastsbetweenamonthbeforethecarbon tax news and a month after the tax news. The two-year forecast window covers the year before and after the implementation of the carbon tax signed into law on December 2019. Similarly, we define revisions in earnings long-term growth forecasts as the change in forecasts over the same period. Table 10 shows summary statistics for three measures of earnings forecast revisions: the revision of forecasts of earnings as percentage of the initial earnings forecast, the revision of forecasts of earnings per share expressed in cents of €, and the revision of forecasts of annual growth in earnings over the next three to four years. Onaverage,forecastsofearningsoverthenexttwoyearsweremarkeddownabout2% relativetotheinitialforecast(or8cents)betweenamonthbeforeandafterthecarbon tax news, with a standard deviation of 14%, suggesting that there is important variationinforecastsrevisions. Forecastsofearningsoverthenextyearweremarkeddown about 2.5% on average relative to the initial forecast (or about 9.5 cents), with a standard deviation of about 20%. Median earnings forecast revisions tend to be smaller in magnitude than mean earnings forecast revisions: the median forecast over the next two years was marked down 0.68%, and the median forecast over the next year was marked down about half a percent. The difference between median and average fore- 22The I/B/E/S is a unique service that gathers and compiles estimates made by stock analysts on the future earnings for publicly traded companies of interest to portfolio managers and institutional investors. Refinitiv collects data for 22,000 active companies in 100 countries, and sourced from over 18,000analysts. 34

Table 10: Summary Statistics of Analyst Earnings Forecast Revisions Mean Median Std. Dev. 1 year (cents of €) -9.45 -2.00 24.80 1 year (pct. change) -2.48 -0.55 19.79 2 years (cents of €) -8.30 -1.50 22.87 2 years (pct. change) -2.20 -0.68 13.65 Long-term growth -0.24 0.00 2.98 This table presents summary statistics of analyst earnings forecast revisions from firms listed on the PrimeStandardsegmentoftheGermanstockexchangewithstockpricesabove€5thatreporttheircarbonemissions. Earningsforecastrevisionsaredefinedasthedifferencebetweenthemonthlyconsensus forecast on November 13, 2019 and January 16, 2020. Earnings forecasts are from the I/B/E/S system providedbyRefinitiv. casts revisions is driven by a few large downward revisions, which we drop from the sample in our econometric analysis below. On average, forecasts of long-term growth were marked down between a month before and after the carbon tax news. The median forecast revision to long term growth, however, was zero. There is important variation in revisions to forecasts of annual long-term growth in earnings, as this is importantforobtainingpreciseestimatesoftheeffectofcarbonintensityonearnings revisions. Tobegin,weexaminetheabilityofafirm’scarbonintensitytopredictanalysts’revisionsofearningsforecastsaroundtheunexpectedcarbontaxincreasebyestimating the following econometric model, F [e ] − F [e ] τ+δ i τ−δ i = φ +φ CO INTENSITY +φ(cid:48)x +ε (3) F [e ] 0 1 2 i,t i,t i,t τ−δ i where F [e ] − F [e ] is the change in forecasts of earnings made a month before τ+δ i τ−δ i the tax news, τ−δ, and a month after the tax news, τ+δ, F [e ] is the forecast before τ−δ i the carbon tax news. The key explanatory variable is a firm’s (log) carbon emissions intensity as captured by the ratio of CO emissions to the total value of assets 2 (CO INTENSITY). To control for other factors that may affect the analysts’s forecasts 2 revisions, the vector x includes controls for size as captures by a firm’s market capi,t 35

italization (ln MKTCAP), price-to-book value per share (ln PRICEBOOK), profit margin (PROFIT), the volatility of stock returns over the past 12 months (RETVOL), the return onequityoverthepast12months(MOMENTUM).Tocontrolforindustryvariation,we include a dummy variable for firm’s that participate in the EU ETS (ETS), and industry dummy variables. As in the stock return cross-sectional regressions, the financial characteristics of a firm are from fiscal year t = 2018, which are known to analysts when they made their forecasts. Table11presentstheresultsfromourlinearregressionmodel(3). Column1presents the estimates of a regression of revisions in earnings over the next two years onto CO intensity. The estimated coefficient on carbon intensity suggests that a one stan- 2 darddeviationincreaseincarbonintensityleadstoastatisticallysignificantdownward revisionofearningsforecastsof2.2percentagepointsaroundtheunexpectedincrease in the carbon tax, or close to one-fifth of a standard deviation of earnings revisions overthathorizon. Column2reportsestimatesoftherelationshipbetweenrevisionsof forecasts of long-term growth in earnings and CO intensity. The regression estimate 2 suggests that analysts marked down their forecast for long-term growth in earnings by0.63percentagepointsaroundthecarbontaxnews, aone-sixthstandarddeviation of growth in earnings forecast revisions over this period. Our estimates highlight that the carbon tax announcement changed not only analysts’ expectations of firms’ level of earnings but also their growth trajectory over the next three to five years. Since the carbon tax was signed into law at the end of 2019 and was scheduled to take effect in 2021, earnings forecasts across two years would capture both the year between announcement and implementation as well as the first year the carbon tax wasineffect. Revisionstooneyearearningsforecasts,however,provideinsightabout changes in analysts’ outlook on earnings the year between the announcement of the carbon tax and its implementation. Accordingly, column 3 of Table 11 present results using one-year earnings forecast revisions. The estimated coefficient on carbon intensity defined suggests that a one standard deviation increase in carbon intensity leads to a downward revision in the earnings forecast of 3.7 percentage points, or about a fifth of a standard deviation. Interestingly, earnings of carbon intensive firms 36

are expected to decline relative to their less carbon intensive counterparts before the carbontaxgoesintoeffect,suggestingthattheresponseinassetpricesreflectsfactors beyond the cost of the carbon tax. Next, we explore whether environmental scores can also predict earnings forecast revisionsaroundthecarbontaxnews. Asshownincolumns1and2ofTable12,wefind that the relationship between a firm’s environmental score and forecasts revisions of both the level of earnings and the growth of earnings over the next several years is statistically insignificant. Therefore, our findings that carbon intensity predicts the revisions in earnings forecast around the announcement of the carbon tax news and that environmental scores are unrelated to analysts’ earnings revisions are consistent withthefindingspresentedintheprevioussectionandmakeitunlikelythatthemain conclusions are due to chance. 37

Table 11: Earnings Forecast Response to Carbon Tax News and CO Intensity 2 2-year earnings Long-term growth 1-year earnings CO INTENSITY -0.930** -0.279* -1.652** 2 (0.406) (0.149) (0.707) lnPRICEBOOK -0.364 0.032 1.539 (1.816) (0.188) (2.894) lnMKTCAP -1.260** 0.217 -2.660* (0.538) (0.175) (1.335) MOMENTUM 0.083*** -0.005 0.158*** (0.022) (0.014) (0.050) RETVOL -6.911*** 0.281 -2.831 (2.325) (0.433) (3.194) PROFIT -0.025 0.004* 0.011 (0.0166) (0.002) (0.023) ETS -0.917 -1.855* 9.173* (3.677) (1.106) (5.341) R2 0.34 0.24 0.20 Observations 110 76 104 Thistablepresentstheestimatedcoefficientsfromcross-sectionalregressions. Thedependentvariables areforecastrevisionasapercentchangeinearningsoverthenexttwoyears,long-termgrowthinearnings,andearningsoverthenextyear. Thekeyexplanatoryvariableisafirm’sCO intensitymeasuredas 2 thelogoftheCO 2emissionstototalassetsratio. Thecontrolvariablesare: (log)marketcapitalization 2 (lnMKTCAP);(log)price-to-bookvalueratio(lnPRICEBOOK);profitmargin(PROFIT);stockreturnover the past 12 months (MOMENTUM); the volatility of stock returns over the past 12 months (RETVOL); andanindicatorifafirmparticipatesintheEUEmissionsTradingSystem(ETS).Allregressionsinclude industry fixed-effects constructed using the Fama-French industry classification. The sample includes German publicly traded firms that are part of the Prime Standard segment and that either report their carbon emissions have emissions are estimated from energy use. Observations with a Cook distance greaterthan3aredropped. Standarderrorsrobusttoheteroskedasticityareshowninparentheses. *** p<0.01;**p<0.05;*p<0.1. 38

Table 12: Earnings Forecast Response to Carbon Tax News and Environmental Score 2-year earnings Long-term growth 1-year earnings ENVSCORE 0.045 0.237 0.156 (0.988) (0.170) (2.029) lnPRICEBOOK 1.708** 0.102 0.080 (0.790) (0.178) (3.386) lnMKTCAP -0.582 -0.231* -2.712 (0.516) (0.134) (1.763) MOMENTUM 0.073*** -0.011* 0.163*** (0.018) (0.006) (0.046) RETVOL -3.493** -0.584* -7.581** (1.432) (0.308) (3.131) PROFIT -0.003 0.002 0.001 (0.009) (0.002) (0.024) ETS 2.257 -0.209 2.348 (1.670) (0.458) (6.485) R2 0.28 0.20 0.23 Observations 109 80 108 Thistablepresentstheestimatedcoefficientsfromcross-sectionalregressions. Thedependentvariables are forecast revision as a percent change in earnings over the next two years, long-term growth in earnings, and earnings over the next year. The key explanatory variables are a firm’s environmental score. The control variables are: (log) market capitalization (ln MKTCAP); (log) price-to-book value ratio(lnPRICEBOOK);profitmargin(PROFIT);stockreturnoverthepast12months(MOMENTUM);the volatilityofstockreturnsoverthepast12months(RETVOL);andanindicatorifafirmparticipatesinthe EUEmissionsTradingSystem(ETS).Allregressionsincludeindustryfixed-effectsconstructedusingthe Fama-Frenchindustryclassification. ThesampleincludesGermanpubliclytradedfirmsthatarepartof thePrimeStandardsegmentandthateitherreporttheircarbonemissionshaveemissionsareestimated fromenergyuse. ObservationswithaCookdistancegreaterthan3aredropped. Standarderrorsrobust toheteroskedasticityareshowninparentheses. ***p<0.01;**p<0.05;*p<0.1. 5 Concluding Remarks In this paper, we assess the ability to hedge transition risks from climate change of two investment strategies that integrate ESG scores or carbon emissions data directly into the security selection process. We conduct our analysis around the approval of the German carbon tax for the transport and buildings sectors in December of 2019 to exploit the plausibly unexpected increase in the carbon tax. We show that while the stock price of firms with a high environmental score did not perform any better than thestockpriceofafirmwithalowenvironmentalscore,thecarbonintensityofafirm 39

is a robust predictor of the reaction of stock prices to the unexpected increase in the carbon tax. We demonstrate that carbon intensity does a better job in explaining the cross-sectional reaction in equity values to the carbon tax news than environmental scores because carbon intensity predicts revisions in expected profitability after the unexpected carbon tax news while environmental scores do not. Importantly, investors marked down not only the level of earnings but also their forecasts for the growth in earnings over the next several years for carbon intensive firms relative to their less carbon intensive counterparts. All in all, our empirical evidence suggests that investors should look through environmental ratings and consider a firm’s carbon footprint to hedge climate change risks. Our results speak to the ongoing debates on metrics to assess a firm’s exposure to climate change risks. Investors and regulatory agencies are increasingly concerned about the risks climate change could pose to businesses’ operations and financial health. Recent debates over disclosure of information about a firm’s resilience or exposure to climate risks—including those from policies to combat climate change, namely, transition risks—are contentious.23 The ability of a firm’s carbon emissions to predict the change in a firm’s market value in response to a higher carbon tax highlights the benefits of having high-quality and reliable measures of firm-level carbon emissions. Our results indicate that disclosing Scope 1 and Scope 2 carbon emissions would help investors and regulators have a simple and comparable metric thatcapturesafirm’sexposuretoclimatepolicyrisk. Inpractice, thismetricprovides investors, financial advisers and asset managers information about the exposure of their portfolios and financial products to climate policy risks.24 While there is strong evidence that a firm’s carbon emissions intensity is a robust predictor of a firm’s stock market reaction to the carbon tax news in Germany, the generalization of our findings to other climate transition risks, both physical and 23The SEC https://www.reuters.com/legal/legalindustry/will-secs-proposed-climate-riskdisclosure-rules-survive-supreme-court-scrutiny-2022-08-05/. 24For example, new regulations from the Securities and Exchange Commission may change the landscape of using firm carbon footprints to select stocks. Beginning in fiscal year 2023 or 2024, firms will berequiredtoreporttheircarbonemissionsandclimate-relatedrisks. Formoredetails,seeWashington Post,March21,2022,“TheSECProposedaLandmarkClimateDisclosureRule. Here’sWhattoKnow.” 40

transition risks, is unclear. In particular, the equity market reaction might be different to future policies aiming at speeding the transition to a zero emissions economy because investors, for example, might have more information about the potential effects of climate change on the economy or become more attuned to risks from climate change. Understanding the predictive content of carbon emissions for other climate risksandquantifyingtheirpotentialeffectsasinvestorsshifttheirexpectationsabout the potential effects of climate change represent a fruitful area for future research. 41

A Robustness of Evidence Based on Portfolio Sorts Thissectionpresentsevidencethattheconclusionsbasedonportfoliosortsarerobust toseveralvariationstothewayweconstructtheportfolios. First,assuggestedinFama and French (2008), we construct value-weighted instead of equally-weighted portfoliostocheckthatourconclusionsarenotsensitivetovolatilereturnsthatcharacterize smallfirms. AsshowninA.1, consistentwithourevidencebasedonequally-weighted portfolios, the return on a portfolio comprised of A-rated firms does not perform any better that of a portfolio with firms that have a score C&below. Similarly, the results in A.2 confirm our finding that firms with low CO intensity increase in value and 2 firms with a high CO intensity decrease in value in response to the carbon tax news. 2 The return spread for value-weighted portfolios sorted on carbon intensity is positive andstatisticallysignificantonthedaythecarbontaxincreasewasannouncedforboth CAPM-adjusted and Fama French-adjusted returns. In Table A.3 we explore if our evidence that environmental scores do not capture a firm’s exposure to transition risk from climate change is independent on the provider of environmental scores. In particular, we use environmental scores provided by the CarbonDisclosureProject,RobecoSAM,andS&PIQ.Portfoliossortedonenvironmental scoresfromallthreeoftheseprovidersyieldresultsthatareconsistentwithourresults in Section 3.1. The return spread between the A-rated portfolio and C-rated & below portfolio for all three scores is not statistically distinguishable from zero, for both CAPM-adjusted and Fama French-adjusted returns. In the same spirit, we explore if our conclusion that CO intensity is a good pre- 2 dictor of a firm’s exposure to climate risk is robust to alternative definitions of carbon emissions intensity. Table A.4 shows that the log of CO emissions and CO intensity 2 2 defined as the ratio of CO emissions to the market value of a firm the year before the 2 carbon tax news are both negatively related to the change in value of the portfolios in response to the unexpected increase in the carbon tax. 42

Table A.1: Risk-Adjusted Returns on Value-Weighted Portfolios Sorted on Environmental Score On December 16 A-rated B-rated C-rated & below A-rated minus C-rated CAPM-adjusted -0.17 0.21 0.57 -0.74 [ -1.22, -0.23] Fama French-adjusted -0.59 -0.23 0.23 -0.82 [ -1.34, -0.28] Portfolio Characteristics A-rated B-rated C-rated & below No. of Stocks 35 30 50 ln CO emissions 16.17 14.77 13.28 2 CO intensity 11.17 11.93 11.31 2 Refinitiv E-score 0.91 0.66 0.38 ln Market value 25.72 23.57 22.70 ln Total assets 24.15 22.76 21.83 ThistablepresentsCAPMandFamaFrench-adjustedvalue-weightedreturnsonthreeportfoliosaround the carbon tax news sorted according to environmental scores. Portfolio returns are presented for the daythecarbontaxincreaseisannounced(December16,2019). Thelastcolumnreportsthereturnspread betweentheportfolioswiththehighestandlowestenvironmentalscoresand,inbrackets,itsassociated 90 percent confidence interval. The confidence intervals are obtained using a bootstrap methodology with1000samples. 43

Table A.2: Risk-Adjusted Returns on Value-Weighted Portfolios Sorted on Carbon Emissions Intensity On December 16 Low emitters Neutral High emitters Low minus High CAPM-adjusted 0.51 -0.22 -0.27 0.78 ( 0.32) ( 0.31) ( 0.18) ( 0.36) [ 0.06, 1.26] Fama French-adjusted 0.11 -0.64 -0.71 0.81 ( 0.30) ( 0.30) ( 0.19) ( 0.36) [ 0.15, 1.33] Portfolio Characteristics Low Mid High No. of Stocks 34 46 35 ln CO emissions 11.51 13.22 16.36 2 CO intensity 6.79 9.74 13.31 2 Refinitiv E-score 0.58 0.59 0.67 ln Market Value 23.20 23.42 23.09 ln Total assets 25.44 24.20 23.77 ThistablepresentsCAPMandFamaFrench-adjustedvalue-weightedreturnsonthreeportfoliosaround the carbon tax news sorted according to carbon emissions intensity. Portfolio returns are presented for the day the carbon tax increase is announced (December 16, 2019). The last column reports the return spread between the portfolios with the highest and lowest emissions intensity and, in brackets, its associated 90 percent confidence interval. The confidence intervals are obtained using a bootstrap methodologywith1000samples. 44

Table A.3: Risk-Adjusted Returns on Portfolios Sorted on Environmental Score On December 16 A-rated B-rated C-rated & below A-rated minus C-rated Panel A: Carbon Disclosure Project CAPM-adjusted -0.07 0.21 0.25 -0.33 ( 0.24) ( 0.25) ( 0.43) ( 0.50) [ -1.15, 0.51] Fama French-adjusted -0.50 -0.24 -0.11 -0.39 ( 0.24) ( 0.28) ( 0.44) ( 0.50) [ -1.25, 0.37] Panel B: RobecoSAM CAPM-adjusted -0.09 -0.09 -0.38 0.29 ( 0.24) ( 0.24) ( 0.49) ( 0.55) [ -0.58, 1.22] Fama French-adjusted -0.51 -0.57 -0.90 0.39 ( 0.26) ( 0.24) ( 0.56) ( 0.60) [ -0.56, 1.42] Panel C: S&P CAPM-adjusted -0.29 0.20 0.34 -0.63 ( 0.26) ( 0.26) ( 0.60) ( 0.65) [ -1.62, 0.52] Fama French-adjusted -0.72 -0.23 -0.12 -0.60 ( 0.27) ( 0.27) ( 0.62) ( 0.68) [ -1.55, 0.69] ThistablepresentsCAPMandFamaFrench-adjustedvalue-weightedreturnsonthreeportfoliosaround the carbon tax news sorted according to environmental scores. Portfolio returns are presented for the daythecarbontaxincreaseisannounced(December16,2019). Thelastcolumnreportsthereturnspread betweentheportfolioswiththehighestandlowestenvironmentalscoresand,inbrackets,itsassociated 90 percent confidence interval. The confidence intervals are obtained using a bootstrap methodology with1000samples. 45

TableA.4: Risk-AdjustedReturnsonPortfoliosSortedonCarbonEmissionsandEmissions Intensity On December 16 Low emitters Neutral High emitters Low minus High Panel A: Carbon emissions CAPM-adjusted 0.75 0.06 -0.16 0.92 ( 0.35) ( 0.25) ( 0.17) ( 0.39) [ 0.26, 1.57] Fama French-adjusted 0.33 -0.36 -0.56 0.89 ( 0.36) ( 0.24) ( 0.17) ( 0.40) [ 0.20, 1.52] Panel B: Carbon emissions intensity (market value) CAPM-adjusted 0.39 0.50 -0.40 0.79 ( 0.26) ( 0.26) ( 0.26) ( 0.38) [ 0.18, 1.38] Fama French-adjusted -0.07 0.09 -0.77 0.70 ( 0.29) ( 0.27) ( 0.25) ( 0.37) [ 0.04, 1.28] This table presents CAPM and Fama French-adjusted equally-weighted returns on three portfolios aroundthecarbontaxnewssortedaccordingtocarbonemissionsandcarbonemissionsintensitycomputed as the ratio of carbon emissions to market value. Portfolio returns are presented for the day the carbon tax increase is announced (December 16, 2019). The last column reports the return spread between the portfolios with the highest and lowest emissions and, in brackets, its associated 90 percent confidence interval. The confidence intervals are obtained using a bootstrap methodology with 1000 samples. 46

References Alekseev, G., S. Giglio, Q. Maingi, J. Selgrad, and J. Stroebel. 2022. A quantity-based approach to constructing climate risk hedge portfolios. WorkingPaper . Bansal, R., D. Kiku, and M. Ochoa. 2019. Climate change risk. FederalReserveBankof SanFranciscoWorkingPaper . Berg, F., J. F. Kölbel, A. Pavlova, and R. Rigobon. 2021. ESG Confusion and Stock Returns: Tackling the Problem of Noise. AvailableatSSRN3941514 . Bolton, P., and M. Kacperczyk. 2021. Do Investors Care About Carbon Risk? Journalof FinancialEconomics 142:517–549. Busch, T., M. Johnson, and T. Pioch. 2022. Corporate Carbon Performance Data: Quo Vadis? JournalofIndustrialEcology 26:350–363. Caramichael, J., and A. C. Rapp. 2022. The Green Corporate Bond Issuance Premium. InternationalFinanceDiscussionPaper . Carbon Disclosure Project. CDP Worldwide. URL https://www.cdp.net/en. Cochrane, J. H. 2011. Presidential Address: Discount Rates. The Journal of Finance 66:1047–1108. Colacito, R., B. Hoffmann, and T. Phan. 2019. Temperature and Growth: A Panel Analysis of the United States. JournalofMoney,CreditandBanking 51:313–368. Cutler, D. M. 1988. Tax reform and the stock market: An asset price approach. The AmericanEconomicReview 78:1107–1117. Dell,M.,B.F.Jones,andB.A.Olken.2012. TemperatureShocksandEconomicGrowth: Evidence from the Last Half Century. American Economic Journal: Macroeconomics 4:66–95. Duan, T., F. W. Li, and Q. Wen. 2021. Is carbon risk priced in the cross-section of corporate bond returns? AvailableatSSRN3709572 . 47

Engle, R. F., S. Giglio, B. Kelly, H. Lee, and J. Stroebel. 2020. Hedging climate change news. TheReviewofFinancialStudies 33:1184–1216. Fama, E. F., and K. R. French. 1995. Size and Book-to-Market Factors in Earnings and Returns. TheJournalofFinance 50:131–155. Fama,E.F.,andK.R.French.1996. MultifactorExplanationsofAssetPricingAnomalies. TheJournalofFinance 51:55–84. Fama, E. F., and K. R. French. 2008. Dissecting anomalies. The Journal of Finance 63:1653–1678. Goldsmith-Pinkham, P. S., M. Gustafson, R. Lewis, and M. Schwert. 2021. Sea level rise exposure and municipal bond yields. Jacobs Levy Equity Management Center for QuantitativeFinancialResearchPaper . Heinkel,R.,A.Kraus,andJ.Zechner.2001.Theeffectofgreeninvestmentoncorporate behavior. Journaloffinancialandquantitativeanalysis 36:431–449. Huynh, T. D., and Y. Xia. 2021. Climate change news risk and corporate bond returns. JournalofFinancialandQuantitativeAnalysis 56:1985–2009. Ilhan, E., Z. Sautner, and G. Vilkov. 2021. Carbon tail risk. The Review of Financial Studies 34:1540–1571. In, S.Y., K.Y.Park, andA.Monk.2019. Is"BeingGreen"RewardedintheMarket? An Empirical Investigation of Decarbonization Risk and Stock Returns. StanfordGlobal ProjectCenterWorkingPaper 46. Krueger, P., Z. Sautner, and L. T. Starks. 2020. The Importance of Climate Risks for Institutional Investors. TheReviewofFinancialStudies 33:1067–1111. Kruttli, M. S., B. R. Tran, and S. W. Watugala. 2019. Pricing Poseidon: Extreme Weather Uncertainty and Firm Return Dynamics. Finance and Economics Discussion Series,BoardofGovernorsoftheFederalReserveSystem . 48

Lintner, J. 1965. Security Prices, Risk, and Maximal Gains from Diversification. The JournalofFinance 20:587–615. Meng, K. C. 2017. Using a free permit rule to forecast the marginal abatement cost of proposed climate policy. AmericanEconomicReview 107:748–84. Painter, M. 2020. An inconvenient cost: The effects of climate change on municipal bonds. JournalofFinancialEconomics 135:468–482. Pástor, L., R. F. Stambaugh, and L. A. Taylor. 2021. Sustainable investing in equilibrium. JournalofFinancialEconomics 142:550–571. Pástor, L., R. F. Stambaugh, and L. A. Taylor. 2022. Dissecting green returns. Journal ofFinancialEconomics 146:403–424. Pedersen, L. H., S. Fitzgibbons, and L. Pomorski. 2021. Responsible investing: The ESG-efficient frontier. JournalofFinancialEconomics 142:572–597. Ramelli, S., A. F. Wagner, R. J. Zeckhauser, and A. Ziegler. 2021. Investor rewards to climateresponsibility: Stock-priceresponsestothe oppositeshocksofthe2016 and 2020 US elections. TheReviewofCorporateFinanceStudies 10:748–787. Refinitiv.2022. EikonPremiumKioskswithDatastreamforOffice. URLhttps://eikon. refinitiv.com/. Roth Tran, B. 2019. Divest, disregard, or double down? philanthropic endowment investments in objectionable firms. AmericanEconomicReview: Insights 1:241–56. Sharpe, W. F. 1964. Capital Asset Prices: A Theory of Market Equilibrium Under Conditions of Risk. TheJournalofFinance 19:425–442. Snowberg, E., J. Wolfers, and E. Zitzewitz. 2007. Partisan impacts on the economy: evidence from prediction markets and close elections. The Quarterly Journal of Economics 122:807–829. S&P Global. Capital IQ Pro Platform. URL https://www.capitaliq.spglobal.com/. 49