Climate Risk and Capital Structure

We use new data that measure forward-looking physical climate risk at the firm level to examine the impact of climate risk on capital structure. We find that greater climate risk leads to lower leverage in the post-2015 period, i.e., after the Paris Agreement. Our results hold after controlling for firm characteristics known to determine leverage, including credit ratings. Our evidence shows that the reduction in leverage related to climate risk is shared between a demand effect (the firm’s optimal leverage decreases) and a supply effect (lenders increase the spreads when lending to firms with the greatest risk).


Introduction
Compared to the temperature in preindustrial times, the world has warmed by 1°C on average. Global temperatures are still trending upward, with an increase of 2 to 4°C expected by 2100. Climate change has dramatic effects in the forms of sea level rise and weather-related natural catastrophes, such as droughts, storms, heat waves, floods, and heavy rainfall (Stern, 2008). The consequences of climate risks for investors are difficult to assess and to hedge. 1 In 2015, Mark Carney, the former chair of the Financial Stability Board (FSB), stated that investors face potentially substantial losses from climate change consequences in terms of physical risks, liability risks, and transition risks, which may be an issue for financial stability. 2 These climate risks may lead to a reassessment of the value of a large range of firms' assets (plants, property, and equipment) and to increased operational costs, such as relocation costs and insurance costs, resulting in lower profits and reduced repayment capacity. Several recent papers emphasize that climate risk affects the pricing of stocks, bonds and real estate (Berkman et al. 2019, Bernstein et al. 2019, Painter 2020, and Seltzer et al. 2020) and a majority of institutional investors believe that climate risk are important concerns (Krueger et al. 2020).
In this paper, we use new firm-level measures to examine whether the physical climate risks faced by a firm have an impact on its capital structure. Under a Modigliani and Miller (1958) framework, climate risk should play no role. However, in the presence of market frictions, climate risk is likely to alter the tradeoff between benefits and costs of debt. We hypothesize that 2 physical climate risk may affect financial leverage via two possible channels: larger expected distress costs and higher operating costs. 3 We find strong support for the conclusion that greater climate risk leads to lower leverage in the post-2015 period, i.e., after the Paris Agreement (COP21), a historic global climate deal to limit warming to 2°C by 2100, which was signed by 195 countries in December 2015 and supported by a high degree of commitment from large firms, institutional investors and central banks. 4 The Paris Agreement raised awareness of the extent of climate risks among all stakeholders. Engle et al. (2020) note that the research on climate risk needs better data to measure firm-level climate risk exposure. In this paper, we use two comprehensive measures of physical climate risk at the firm level. We first rely on the "Climate Risk Impact Screening" (CRIS) methodology developed by a French firm, Carbone 4, with support from several institutional investors and public agencies, including the French Development Agency (AFD) and Caisse des Dépôts et Consignations (CDC). The CRIS risk rating is a forward-looking measure that captures the increase in intensity or frequency of climate-related hazards due to climate change at two time horizons, 2050 and 2100. For each firm in the MSCI World Index, climate risk grades are quantified based on climate projections from IPCC models, the geographical division of activities, country-specific vulnerabilities and industry-specific vulnerabilities.
As a second measure of climate risk, we use alternative data provided by Four Twenty Seven, a provider of data related to physical climate and environmental risks that has been part of 3 Although some firms will benefit from increased climate risk, for example, those specialized in providing services for adjusting to this risk, most will see negative effects on their earnings. In a study of the effects of climate change and weather effects on earnings for the firms in the S&P500 index, more than 90% of the mentions were negative (S&P Global (2018)). 4 Before 2015 and the Paris Agreement, despite trying for decades, the world failed to reach a global agreement on climate change due to coordination and free-riding problems.
Moody's ESG solutions since July 2019. Four Twenty Seven's models assess projected exposure to climate hazards at the facility level aggregated at the firm level. They also assess a firm's dependence on natural resources threatened by climate change. Four Twenty Seven provides a comprehensive climate risk score for each firm. Using two databases of physical climate risk measures enables us to cross-validate our results.
We begin our empirical analysis by estimating the relationship between a firm's leverage ratio and our measures of climate risk. Specifically, we regress the observed debt ratios of the firms that belong to the MSCI World Index over the period of 2010-2019 on climate risk measures for each firm in addition to several fixed effects and other control variables. We find that an increased physical climate risk reduces firms' leverage in the post-2015 period, i.e., after the Paris Agreement. We then examine the impact of subrisks on leverage and find that acute risks, i.e., the risks that lead to natural disasters (droughts, storms, heat waves, heavy rainfalls), are the main drivers of the leverage decrease. Sea level rise has also a significant negative effect on leverage. Our results are both statistically and economically significant. The patterns that we observe in our baseline tests remain after various robustness checks that involve changes in empirical specifications, variable construction methods and sampling restrictions. Furthermore, by using the 2015 Paris Agreement as a shock to the awareness of firms, bankers and investors of climate risks, we also conduct a difference-in-differences approach to compare the leverage of high climate risk firms versus low climate risk firms before and after the Paris Agreement. Our findings remain unchanged.
Climate risk could also be a component of the overall corporate credit risk; therefore, credit rating agencies (CRAs) should include it in their risk assessment, with credit ratings also reflecting climate risk. Rating agencies are increasingly aware of the need to incorporate the risks 4 and opportunities associated with environmental and climate (E&C) factors into their corporate credit ratings. 5 However, our results suggest that credit ratings do not reflect all the information related to physical climate risk, confirming that credit rating agencies are conservative in adjusting their ratings (Altman and Rijken 2004). 6 In all our tests, we control for credit ratings and find that the physical climate risk grades provide additional information that is not already embedded in credit ratings. We also find that our measures of climate risk do not impact credit ratings when controlling for the usual determinants of credit ratings. Recently, major CRAs have acquired extra-financial rating agencies, which leads to the reinforcement of their expertise in climate risk rating and could result in better recognition of climate risk in the future. 7 In the traditional empirical capital structure literature, debt supply frictions are not observed, and the firms' characteristics are the main determinants of leverage. In this framework, the observed reduction in leverage would entirely result from firms becoming aware of their climate risks and lowering their leverage. To adjust their leverage, in addition to lowering their demand for debt, high climate risk firms can increase shareholders' equity. We find that, after 2015, high climate risk firms increase their net equity offerings, suggesting that at least a fraction of the reduction in leverage results from a demand effect. Another way to examine the demand side is to review firms' CSR performance. As Engle et al. (2020) underline, CSR expenses may act as a hedge against physical and regulatory risks. Our results related to the impact of climate 5 For example, Standard and Poor's (S&P) examined 9,000 updates between July 2015 and August 2017 to gage how these factors have featured in S&P Global Ratings' corporate credit analysis. E&C factors were an important consideration in the analysis of 717 cases and a driver for rating changes in 106 cases. Interestingly, of the examples that have an environmental or climate factor that was key to a rating change in the S&P analysis, most are linked to physical climate risks. See this report from S&P Global Ratings. 6 Some anecdotal evidences point in this direction: this article by Fitch ; this article by a former Moody's senior vice president. See also this article on municipal bonds. 7 For example, S&P acquired Trucost, a provider of carbon and environmental data and risk analysis (2016), and Robecom SAM (2019), a European ESG rating agencies, and besides Four Twenty Seven, Moody's acquired Vigeo-Eiris, a global leader in ESG data (2019). 5 risk on leverage remain unchanged after considering CSR scores, which suggests that physical climate risk is an additional risk besides the environmental issues that nonfinancial rating agencies usually rate. Furthermore, we find that the reduction in leverage is mainly observed for firms with low CSR performance, suggesting that high CSR firms are likely to take proactive actions to handle their climate risk rather than decrease their leverage.
On the supply side, bondholders and bankers may be willing to reduce their exposure to climate risks by limiting the amount of debt that they lend to high climate risk firms or by increasing the cost of debt for these firms. To test whether a supply effect occurs, we use loanlevel data to examine interest rates charged on bank loans and bonds issues. We find that greater climate risk implies higher spreads on both bank loans and bond issues in the post-2015 period.
Overall, our findings suggest that the reduction in leverage related to climate risk is shared between a demand effect, whereby firms lower their demand for debt or issue more equity, and a supply effect, whereby bankers and bondholders increase the interest rate that they charge to high climate risk firms.
Our paper contributes to several lines of research. First, this research is related to the literature on physical climate risk and its impacts on firms and investors. The macroeconomic literature provides a great deal of evidence of global warming and extreme natural events that affect agricultural output, industrial output, energy demand, labor productivity, health, conflict, political stability and economic growth. 8 Evidence on a microeconomic level gives rise to a recently growing literature. For example, Barrot and Sauvagnat (2016) examine the impact of natural disasters on sales growth and find that disasters negatively affect both the sales growth of directly exposed firms and their largest customers. Addoum et al. (2019), Hugon and Law (2019), 6 and Pankratz et al. (2019) observe that abnormal temperature negatively impacts firms' earnings. Bansal et al. (2016) establish that long-term temperature shifts have a significant negative effect on equity valuations, Berkman et al. (2019) use a firm-specific climate risk measure based on textual analysis and find that firm value is negatively related to climate risk, whereas Kruttli et al.
(2019) observe that the uncertainty surrounding natural disasters is priced in option and stock prices. 9 Bernstein et al. (2019) find that coastal properties exposed to projected sea level rise (SLR) sell at an approximately 7% discount relative to otherwise similar properties. This SLR exposure discount is primarily driven by properties unlikely to be inundated for over half a century, which suggests that this discount is due to investors pricing  This result emphasizes how climate risk discounts asset values and potentially reduces their pledgeability, which, in turn, may be part of the explanation of the leverage reduction that we document in our study. 11 Second, our paper is also related to the literature that examines the impact of operating costs on firms' financial leverage. Physical climate risks may increase operating costs (climate resilience expenses, costs related to operational disruptions, supply chain changes, insurance premiums), which could lead to a substitution effect between operating and financial leverage.
Several authors examine various types of operating costs and risks and find a negative relationship between operating leverage and financial leverage. Petersen (1994) Chen et al. (2011) argue that the presence of labor unions reduces operating flexibility and underline that "the concept of operating leverage in its nature is forward looking". In our paper, we rely on a forward-looking climate risk measure to proxy for increased operating costs and find that after 2015, the risk related to climate change, even if not yet materialized, leads to a reduction in the leverage of the world's largest firms. In this respect, our approach differs from that of Elnahas et al. (2018), who investigated the impact of natural disasters on the leverage of US firms and observed that firms adjust their leverage once disaster risks materialize.
Third, our research also contributes to the literature on the impact of climate risks on credit risks. Painter (2020) examines municipal bonds and finds that counties more likely to be affected by climate change pay more in underwriting fees and initial yields. Several recent papers find that physical or transition climate risks increase bond spreads (Seltzer et al. 2020, Huynh and Xia 2020) as well as bank spreads (Delis et al. 2019, Anginer et al. 2020. These risks also lead to a decrease in the amount of loans granted (Faiella and Natoli, 2019). Our results not only confirm these supply side effects but also underline that they occur mainly after 2015. Several other articles also find post-2015 effects. For example, Zerbib (2019) finds negative yield premiums for green bonds after May 2016, andSeltzer et al. (2020) provide evidence of a causal relation between climate regulatory risks and bond yield spreads after the 2015 Paris Agreement. Overall, there is currently a strong set of results that emphasize the tangible effects of the rising awareness of bankers and institutional investors regarding climate risks in the post-2015 period. 8 Fourth, our paper is also related to the literature that examines the impact of CSR issues on capital structure and the cost of capital. Chava (2014) finds that firms that cause environmental externalities (e.g., toxic waste) have higher equity and debt costs. Chang et al. (2018) observe that firms with greater environmental liabilities maintain lower financial leverage ratios, with a lower fraction of bank debt in total debt. According to Sharfman and Fernando (2008), improved environmental risk management is associated with lower capital costs and allows for more leverage, whereas El Ghoul et al. (2011) report that firms with better CSR scores exhibit more inexpensive equity financing. Lins et al. (2017) find that during the [2008][2009] financial crisis, high CSR firms were able to raise more debt. Amiraslani et al. (2019) show that these firms benefited from lower spreads, better credit ratings and longer maturities, suggesting that social capital can mitigate the agency costs of debt. Our finding that the reduction in leverage with climate risk occurs mainly for low CSR firms suggests that high CSR firms are probably aware of their climate risks earlier than low CSR firms and are likely to face them in a more proactive manner.
The rest of the paper is structured as follows. In section 2, we present the institutional context of climate risk and our hypotheses. In section 3, we present our climate risk measures and our dataset. We analyze our empirical results in section 4, and section 5 concludes.

Effect of climate risk on leverage
Static tradeoff theory, pecking order, and market timing are the three preeminent theories of capital structure. The static tradeoff theory suggests firms choose their capital structure to 9 balance the benefits (corporate tax savings) and the costs (bankruptcy costs, agency costs) of debt financing and manage their leverage towards a target (see, for example, Bradley et al. 1984, Fischer et al. 1989, Leland 1994, Flannery and Rangan 2006. The pecking order theory predicts a financing hierarchy in which firms use internal funds first, then debt, and issue equity only as a last resort due to adverse selection costs of issuing equity (Myers and Majluf 1984). Finally, the market timing hypothesis posits that firms issue equity when they perceive the relative cost of equity is low and issue debt otherwise (Baker and Wurgler 2002). All these models involve tradeoffs between costs and benefits but differ in their assessment of which market frictions are the most relevant. There are a large number of empirical studies, often aimed at providing support for one of these theories. Overall, although results vary over time and depend on the type of sample selected and the methodology that is used, the evidence suggests that firms borrow more when they are subject to lower debt issuance costs, higher corporate taxes, lower bankruptcy costs, a higher liquidation value of assets and lower operating costs and earnings volatility. 12 To assess the impact of climate risk on corporate leverage, we focus on two variables: operating costs and bankruptcy costs.
First, firms exposed to physical climate risks will incur climate resilience expenses, costs related to operational disruptions, production adjustments, supply chain changes and increased insurance premiums. The increase in insurance premiums is a major factor in the rise in operating costs. Climate risk is becoming increasingly difficult to insure due to its systemic nature and to the exit of a number of insurers from this market (Born and Viscusi 2006). A growing number of insurers are considering not renewing insurance contracts for clients or sectors most at risk, 10 leading to reduced competition and higher prices. 13 These effects are already materializing for firms subject to major natural disasters, such as those operating in areas highly exposed to hurricanes. However, our climate risk measure is forward-looking, which means that firms' awareness of this risk may lead them to re-estimate their future operating costs. Managers may therefore reduce financial leverage to counter the concern that higher operating costs could increase the probability of default. The value of a firm's assets may be reduced if they are located in areas that are subject to significant climatic risks. The impairment may be related to direct asset destruction by an extreme climatic event or to a reduction of asset value due to their exposure to future climate risks (for example, seashore property exposed to a sea level rise). In addition, a loss in the assets' market value may also result from the inability to sell these assets to an acquirer due to the increased climate risks. 14 Insurance companies can partly mitigate the first type of costs (asset destruction by extreme events) but do not cover the second type.
The traditional hypothesis in the empirical capital structure literature is that the observed level of debt equals the firm's demand level, which means that there is no supply friction. Firm characteristics are then the main determinants of leverage. Therefore, our first hypothesis is that firms with greater climate risk exposure will reassess their operating costs and distress costs, which should lead them to reduce their leverage compared to firms with low exposure to climate risk. To adjust their leverage, high climate risk firms may decrease their demand for debt or issue new shares.

Climate risk and leverage: is there a supply effect?
Supply-side factors are likely to be important in explaining capital structure (Faulkender and Petersen 2006). There may be climate effects related to the debt supply if lenders become increasingly aware of climate risk and subject firms to more stringent regulations and disclosure requirements. The climate risk effects can occur directly through a quantity channel if lenders are willing to lend less to firms exposed to higher climate risk or indirectly through a price channel if lenders are increasing the cost of debt for high climate risk firms. To verify this last effect, we conduct empirical tests by using loan-level data, specifically, bank loans on the one hand and bond issues on the other hand. Therefore, our second hypothesis is that climate risk should increase the cost of debt.

Climate risks: why is 2015 a key year for climate risk awareness?
Although the United Nations Framework Convention on Climate Change (UNFCCC), which was adopted in 1992, establishes the general legal framework for international climate change action, it was not until 1997 that countries agreed on quantified emissions limits for developed countries for the first commitment period of the Kyoto Protocol (2008)(2009)(2010)(2011)(2012).
However, as Andersson et al. (2016b) underline, these top-down rules imposed on businesses by governments resulted in little progress in the field of climate change mitigation. In contrast, 2015 was a pivotal year in considering climate change, as economic actors decided to take up the issue.
In 2015, the FSB established the Task Force on Climate-related Financial Disclosures (TCFD). Furthermore, the Paris Agreement, which was signed in December 2015, applies for the first time to all countries, including major developing countries with large emissions, such as India and China. 15 In advance of the Paris Climate Agreement, several private initiatives involving businesses declared their collective support for an effective climate change agreement to be reached at COP21. 16 One of the core aims of the Paris Agreement is to make all financial flows consistent with a pathway towards low emissions and climate-resilient development. The Agreement sends a strong signal that all finance, both public and private, needs to be directed towards the climate challenge. Several initiatives have since been developed to increase investors and central banks' awareness of the climate risks to which they are exposed. 17 Between 2013 and 2017, the number of subnational and national-level policy and regulatory measures more than doubled (from 139 to 300), 18 with a substantial rise in system-level initiatives (finance regulations and guidelines and national level roadmaps for green finance). In 2016, China adopted the "Guidelines for establishing a green financial system". In the same year, the European Union established the High-Level Expert Group on Sustainable Finance (HLEG), which led in 2018 to the European Commission's "Action Plan on Financing Sustainable Growth" including regulations on the establishment of a taxonomy to facilitate green investments 13 not only on disclosures by institutional investors and asset managers but also on carbon-related benchmarks. Furthermore, according to its Climate Change Action Plan 2016-2020, the World Bank pledged to invest $29 billion annually to fight against climate change, where $13 billion comes from the private sector.
To the extent that the many recent climate change initiatives have increased the attention of firms, investors and central banks to climate risk, we assume that the effects of climate risks on capital structure will mainly materialize in the period after 2015.

Physical climate risk measures
To tackle the issue of better disclosure of climate risk, the FSB, on the request of G20 countries, established the TCFD in 2015. In 2017, the task force issued a set of recommendations that outlined two major categories of climate-related risk: transition risks (related to carbon and mitigation issues) and physical risks (related to impacts and adaptation issues). Several data providers (for example, RepRisk and MSCI ESG) focus on transition risk and CSR in relation to environmental externalities. However, these data providers do not provide data on the exposure of firms to the physical consequences of climate change. The assessment of climate risk at the firm level is not easy because it depends on both geographical factors and vulnerability factors specific to the firm's activity.
In this paper, we use two measures of climate risk. The first is the CRIS methodology, which was developed by the French firm Carbone 4 in cooperation with several financial institutions. 19 The CRIS measures aim at assessing the climate-related physical risks that face firms and their 19 More information is available here. 14 business units in the future by breaking down the firm's activity into geographical and industrial segments and by assessing the future climate risk for each country-industry pair. Thus, rather than focusing on externalities produced at the firm level, CRIS data capture externalities incurred at the firm level. Each climate risk rating is a function of location-specific climate hazards and sector-specific vulnerabilities. Industry information comes from the GICS, ICB and NAICS codes. The geographical division of activities is based on sales, tangible assets, or a combination of both, depending on the low, high or medium capital intensity of the sector to which the firm belongs. Geographical information depends on the granularity of the information disclosed by the firms. Six of the seven largest countries (Brazil, Canada, China, India, Russia, and the US) are further divided into 4 subcountries. At its broadest level, climate risk is measured through an index that aggregates 7 hazards: 4 of these are acute (extreme) hazards, i.e., heatwaves, rainfall extremes, drought, and storms, and 3 are chronic hazards, i.e., increases in average temperature, changes in rainfall patterns and sea level rise.
The CRIS measures are split into two time horizons (2050 and 2100) and three intensity cases (low, medium, high), which reflect the Intergovernmental Panel on Climate Change (IPCC) scenarios and are formally named Representative Concentration Pathways (RCPs). 20 The CRIS risk rating does not capture the absolute risk from future climate or weather but does capture the increased risk due to the increase in the intensity or frequency of the climate-related hazards in the future due to global warming compared to historical reference average hazards. Final ratings are attributed on a scale of 0 to 99, and when the rating is higher, the risk is greater. As the rating 15 scale is relative, a low rating does not necessarily imply low risk in absolute terms but rather means that the risk is in the lower part of the gradient in relative terms. For a firm with multiple business segments (various sectors in various countries), for each hazard, the risk rating is based on the weighted arithmetic mean of all the risk ratings calculated for each of the firm's business segments for this same hazard. The weighting is proportional to the breakdown of the firm's revenue or fixed assets (if capital intensive) in its various segments. For each hazard, the risk rating of a specific sector in a specific country is a combination of the hazard rating of the country and the vulnerability rating of the sector.
In this paper, for the sake of clarity, we use a unique CRIS rating that corresponds to the 2050 horizon and medium intensity risk. This horizon seems distant, as the majority of bond issues have a maturity of less than 30 years, but the reader should keep in mind that climate risk will gradually materialize over the coming years. As Krueger et al. (2020) show in their survey on climate risk, most institutional investors believe that climate risks will materialize within the next few years. The CRIS ratings cover the sphere of the MSCI World Index for 2016.
The second measure of climate risk that we use is provided by Four Twenty Seven. 21 Each firm is scored on three components of physical climate risk: operations risk (70%), supply chain risk (15%) and market risk (15%). A firm's operations risk is based on its facility-level exposure to hurricanes & typhoons, sea level rise, floods, extreme heat and water stress. Supply chain risk is based on the risk in countries that export commodities to the firm and a firm's reliance on climate-sensitive resources such as water, land and energy, based on its industry. Market risk is based on countries of sales and weather sensitivity for market risk. Scores consider projected climate impacts in the 2030-2040 time period under a single RCP scenario, RCP 8.5 (the worst 21 See here for more information. 16 scenario). The number of observations available for the Four Twenty Seven scores is slightly smaller than that for the CRIS scores, as all MSCI firms are not yet graded.
As climate risk scores are determined based on a 2050 horizon (CRIS) or 2030-2040 Twenty Seven), we assume that this risk remains stable over the period studied (2010-2019) and that the firm's activities and locations do not undergo major changes over the period, which is the hypothesis adopted by the two rating companies (see Appendix B for a detailed comparison of these measures).
After excluding financial firms and observations with missing data (see below), we are left with 1,212 firms. In Table 1

Credit ratings
Credit ratings at the issuer level are obtained from Thomson-Reuters. This variable is based on the S&P Long-term Issuer Rating when available. If this rating is not available, we rely on Moody's Long-term Issuer Rating, and we rely on Fitch's Long-term Issuer Default Rating if both previous measures are missing. Similar to Baghai et al. (2014), we linearize these ratings from 1 to 20. Investment grade ratings are coded between 11 and 20, whereas high yield ratings are coded between 1 and 10. Missing ratings are coded 0.
Of our firm-year observations, 67% are rated and therefore have potential access to public debt markets, which reflects the fact that the sample comprises the world's largest listed firms that belong to the MSCI World Index. The average credit rating is 12.28 (median 12), which corresponds to an S&P grade of BBB.

Financial and accounting data
The financial and accounting data are from Compustat North America and Compustat Global. We first matched the firms covered by the CRIS grades with the data available in Compustat for fiscal years 2010 to 2017, which yields 11,836 firm-year observations. By relying on 2-digit SIC codes, we excluded SIC codes 60 to 69, as financial firms are subject to special regulations concerning their capital structure. Missing values for long-term debt, EBIT, R&D expenses and issuer ratings were set to zero. This assumption is noncritical, as only 71 observations have missing values of long-term debt. Missing ages were set to 1 to use the natural logarithm. We have 3 additional observations with missing EBIT. We excluded the observations with missing values of operating expenses and the observations for which we were unable to compute Tobin's Q. Therefore, we were left with 9,138 firm-year observations that cover 1,212 firms. These figures are sound as on the one hand, 1,604 firms are covered by CRIS, and on the other hand, the MSCI World Index covers approximately 1,600 firms, with 16.33% of them belonging to the financial sector. 22 We extended our database to 2019 when the data became available. In total, our database covers 11,367 firm-year observations for 1,212 firms.
Our main measure of leverage for firm i in year t is a book leverage variable, which is defined as follows: 22 See here for more details.
where is the amount of long-term debt that exceeds a maturity of one year, and is the book value of total assets. We exclude the debt in current liabilities because of the long-term nature of climate risks.
Similarly, we define the market leverage for firm i in year t as follows: if the firm is covered by Compustat North America; and if the firm is covered by Compustat Global.
All the variables computed from Compustat are winsorized at the 1% level to prevent the effect of potential outliers. Country fixed effects are based on headquarter locations, and industry fixed effects are based on the 2 digit SIC codes.
In Table 1

Bank loan and bond issuance data
We obtain bank loan data by using Dealscan and focus on loans with maturities greater than 3 years and amount greater than $100 million. We use the item Margin(Bps) as our measure of the cost of the loan. Therefore, we exclude the observations for which this item is unavailable.
We also exclude the observations for which at least one of the independent variables used in our regressions is unavailable. This bank-loan level dataset is then matched with the data described in For these reasons, and as our dataset covers the world's largest firms, our $100 million cut-off seems to be appropriate to gauge whether the decrease in leverage could come from a supply effect. In Dealscan, interest rates charged on bank loans are expressed in terms of basis points added to a reference rate (spreads). To draw a parallel between bank loans and bond issuances, we match our bond data from Thomson-Reuters

Leverage and climate risk
The descriptive statistics show that firms with high climate risk are less highly leveraged.
It may be that firms with high climate risk are also the firms that find debt less valuable.
However, as these firms are larger and have more tangible assets, the theory predicts that they should demand more debt, which suggests that they are not in a situation in which they would attach less value to debt. Based on the literature of capital structure determinants, we regress the firm's leverage on a set of firm characteristics, including credit ratings and climate risk measures.
Clustering effects could bias the statistical significance of the results because of firm leverage persistence over time. Thus, in estimating our regressions, we apply the procedures described in Petersen (2009) to adjust the standard errors for clustering by firm. Our baseline regression is as follows: (1) refers to our measure of long-term debt, either or , represents the value of the overall climate change risk exposure of a firm, is a vector of controls that have been shown to affect the level of debt holdings and is a vector of fixed effects. is also interacted with , a dummy equal to one after 2015, to take into account the Paris Agreement effect. For these regressions, the equation is as follows: (2)

21
Our results are presented in Table 2 for book leverage and in Table 3 for market leverage.
In Tables 2 and 3, CRIS data are used to measure climate risk in Panel A, whereas regressions in Panel B use Four Twenty Seven data. Our findings confirm the previous work on capital structure. Firms with more tangible assets, as measured by a firm's property, plant and equipment to total asset ratio, have a higher debt ratio. In contrast, intangible assets, as measured as research and development expenses scaled by total assets, reduce a firm's leverage. More profitable firms (EBIT/total assets) and firms with a higher proportion of operational expenses are less leveraged.
Furthermore, by including country-industry fixed effects and year fixed effects ( Table 2, columns 1 to 4), we can completely control for any determinant of leverage that is constant within a year or a pair industry-country. Thus, we control for any specific industry structure or regulation in a country.
Controlling for these fundamental differences between firms, we find that increased physical climate risk reduces leverage for the whole period when using CRIS climate scores (Panel A, column 1). This result is not confirmed when using Four Twenty Seven data to measure climate risk (Panel B, column 1). The year 2015 was a pivotal year for considering climate risk that resulted from the Paris Agreement (COP21) and the implementation of the TCFD. Therefore, we also examine whether the impact of climate risk on leverage changed after 2015 by interacting our climate risk measure with a dummy variable equal to one for the post-2015 period. We find that the climate risk effect on leverage materializes mainly after 2015: a one standard deviation increase in climate risk reduces debt by 1.53% (-0.00141*10.833) with the CRIS score (Panel A, column 2) or 1.43% (-0.00108*13.225) with the Four Twenty Seven score (Panel B, column 2).
This effect is economically significant, as it represents 7.01% of the leverage (CRIS scores) and 6.55% of the leverage (Four Twenty Seven scores). To graphically plot the effect of climate risk on leverage over time, we regress leverage on the interaction of climate risk with year dummies, after controlling for traditional determinants of leverage as well as firm and year fixed effects.
The coefficients of the interaction variables are significantly different from zero as of the year 2015, confirming that this year is pivotal in the consideration of climate risk (Figure 1).
Climate risk could also be a component of the overall corporate credit risk. Graham and Harvey (2001) find that for CFOs, credit ratings are their second highest concern when determining their capital structure. If credit ratings already reflect climate risk, adding climate risk variables would not provide any additional information to the determinants of leverage. To verify that our climate risks measures are not mere proxies for credit risk, we add a variable that linearizes the credit ratings from 1 (D) to 20 (AAA) for firms that benefit from a rating and is zero otherwise. We find that firms with more favorable ratings have more long-term debt than firms that are poorly rated ( Our findings may also result from a reverse causality between the credit rating and leverage. To address this potential problem, we use an instrumental variables approach. In the first stage, 23 we estimate the endogenous variable (CreditRating) as a function of the exogenous variable in the second stage plus an additional instrument. Our credit rating variable instrument is based on its means for groups by year/sector/country. This instrument is correlated with our credit rating variable, although it is unlikely that the debt level of a given firm will depend on the average rating of the sector for a given year and country once fixed effects are considered. Our results are confirmed, and the magnitude of the coefficients of the climate risk measure remains similar for book leverage (Table 2, Panels A and B, columns 6 and 7).
When market values are considered ( These findings suggest that our climate risk measures provide an additional risk factor that has an impact on leverage after 2015 and that is not already included in the credit risk ratings.
After the strong signals sent to all participants in the financial system in 2015 regarding the necessity to develop climate-related disclosures and better understand their exposure to climaterelated risks, both managers and investors became more aware of climate risks, which, in turn, can explain the reduction in leverage that we observe. Overall, our results in a difference-in-differences setting are consistent with the findings highlighted in Tables 2 and 3.

Acute risks and chronic risks
CRIS climate risk ratings combine information on the following seven direct climate hazards: three chronic hazards (increases in average temperature, changes in rainfall patterns, and sea level rise) and four acute hazards (heat waves, droughts, rainfall extremes, and storms). For each hazard, the rating is based on the analysis of information on the magnitude, duration and frequency of the hazard (particularly relevant for acute hazards). To build a rating of 0 to 99 for each climate variable and each country, the relative changes are first extracted in the future time horizons compared to the historic reference period and then normalized across all scenarios and time horizons. These direct hazards are associated with information on the risk-aggravating context to capture indirect hazards. For example, the impact of heavy rainfall is larger when the proportion of high slopes in the area is high because of increased landslide risks, and extreme droughts lead not only to water scarcity but also to wildfires.
We examine the impact of each of these 7 climate subrisks on the leverage of firms. In equations (1) and (2), the overall climate risk variable is replaced by subrisk measures. Since the risk variables by category are normalized, their values are of the same magnitude as the overall rating. Therefore, the regression coefficients reflect the relative impact of the risk variables on debt but not the weight of each risk in the total risk to explain the climate impact on debt.
The results in These results emphasize that the impact of aggregate climate risk on leverage is primarily because of the potential increase in the risks of extreme events on the 2030-2050 horizon.

Credit ratings and climate risk
We have seen in previous tests that the climate risk rating provides additional information compared to the credit rating to explain a firm's leverage after 2015. In this paragraph, we intend to explore the relationship between credit risk and climate risk in more detail.
refers to our linearized credit rating variable, represents the overall risk exposure of a firm, is a vector of controls that have been shown to affect the level of credit ratings, and is a vector of fixed effects. is also interacted with , which is a dummy equal to one after 2015.  Baghai et al. 2014) and for country-industry fixed effects (as business risk varies across sectors and the sovereign rating represents in almost all cases a ceiling for the private sector). Alternatively, we use a firm fixed effect regression to control for time invariant firm characteristics. As the results in Table 6 indicate, the coefficient of our climate risk variable is not significantly different from zero, either before or after 2015, whether using CRIS scores or Four Twenty Seven scores, which suggests that credit ratings do not reflect physical climate risk specific to the firm beyond the headquarter country climate risk that is captured by the country-industry dummies.
Accordingly, physical climate risk as measured by the CRIS or Four Twenty Seven ratings does not seem to be reflected in the credit ratings issued by the rating agencies, at least over the period that we examine.

Climate risk and leverage: demand or supply effect?
The observed level of debt is a function of a firm's demand for debt: the empirical capital structure literature traditionally assumes that in the absence of frictions, firms can borrow up to their optimum leverage, which depends on their characteristics. However, the reduction in leverage that we observe in the post-2015 period may also be the result of supply factors.

Climate risk and leverage: the demand effect
To adjust their leverage to climate risk, firms can reduce their demand for debt in line with the variation of their characteristics or issue new equity. We first examine whether firms subject to higher climate risk increase their net equity issuance (equity offerings minus repurchases). Table 7 presents our results. In columns (1) and (2), we use the CRIS climate risk score, and in columns (3) and (4), we use the Four Twenty Seven score. Columns (1) and (3) include country-industry and year fixed effects, and columns (2) and (4)

29
An alternative way to examine the demand side is to focus on CSR performance. We first check whether our measure of climate risk is not a mere proxy for a more general CSR assessment. In Table 8, Panel A, we verify that our results remain unaffected after controlling for various CSR indicators. The regressions in columns (1) and (3) use the general CSR score given by the MSCI IVA ratings. The regressions in columns (2) and (4)  On the other hand, CSR expenses may allow firms to adapt their activities to climate risk and decrease operational risk. A reduction in operating leverage may be an alternative to a reduction in leverage. In Table 8, Panel B, we construct subsamples based on the values of the CSR variables. Columns 1 and 2 report the regressions conducted on firms with an above-median overall CSR score and firms with below or equal to the median overall CSR score, respectively.
Only low CSR firms significantly reduce their leverage after 2015. In regressions (3) and (4), we split our sample between firms included and firms not included on CDP's A list. Firms on the A list have had a smaller decrease in their leverage ratio post-2015 compared to firms not on the A list. In columns (5) to (8), regressions are presented using the Four Twenty Seven score, and the results are similar. All differences between high CSR and low CSR firms are significant at the 1% level (except between columns (3) and (4), significant at the 10% level). Taken together, these results are consistent with the view that firms with better CSR scores are more likely to take proactive actions to hedge their climate risk, thereby reducing the need for a decrease in their debt ratio.

Climate risk and leverage: the supply effect
To test whether supply factors are involved, we examine loan-level data that cover bond issues on the one hand and bank loans on the other hand. If a supply effect exists, the reluctance to finance high climate risk firms should materialize as higher spreads.

Climate risk and public debt markets
We first focus on the impact of physical climate risks on the cost of bonds. With the benchmark spread at issue as our measure of the cost of borrowing, we find a post-2015 rise in interest rates in bond markets. Columns (1) to (6) in Table 9 report the results. The effect is concentrated in high-risk firms. We find that post-2015, a one standard deviation increase in climate risk generates a 5.98 basis point increase (1.087*5.505=5.99, with 5.505 being the standard deviation of the CRIS indicator within the high-risk group) in the spread at issue in the high-risk group when using CRIS scores (column 2) and a 9.83 basis point increase when using Four Twenty Seven scores (column 5). In both cases, we do not find any significant effect within the low-risk group, and the difference in the coefficients is significant between the two risk subgroups when using CRIS scores. All of our specifications include fixed effects to account for the number of loans to the firm on the same date, loan purpose and secured/unsecured status. We further include firm fixed effects and year fixed effects.
Our findings indicate a moderate, albeit significant, impact of physical climate risks on public debt cost in the post-2015 period. Table 10 reports the results for bank loans. Similar to bonds, the effect of climate risk in the post-2015 period is concentrated in high-risk firms. For these firms, the effect of a one standard deviation increase in climate risk, as measured by CRIS scores, is 23.56 basis points (Table 10, column 2). We do not find any significant effect within the low-risk group (column 3), and the difference in the coefficients between the two risk subgroups is significant. When using Four Twenty Seven scores, the coefficient of our climate risk measure for high-risk firms is positive but insignificant.

Climate risk and bank loans
Overall our findings suggest that physical climate risks affect debt supply by increasing the cost of debt for high climate risk firms.

Robustness checks
We conduct several robustness checks. To consider the possibility of time effects that are specific to certain industries, we re-estimate our basic regressions (Table 2), including country and industry-year fixed effects, and the results remain unchanged. Similarly, the institutional characteristics of countries may evolve over time in different ways, but the patterns of the results are qualitatively similar when including country-year fixed effects. We also rerun our regressions, including several dummy variables for each level of credit rating rather than our linearized variable, and our results remain similar. Our results are also qualitatively unchanged when using country, industry and year fixed effects and clustering at the country-industry level, instead of using country-industry fixed effects and clustering at the firm level.
As an alternative to our 2050 horizon CRIS climate risk rating, we also use the 2100 horizon rating and low/high intensity risks, and the results are qualitatively unchanged, although the coefficients of the variables change slightly depending on the chosen combination.
We also verify that the results are robust to the exclusion of firms threatened by transition   (1) and (3) use CRIS scores and columns (2) and (4) use Four Twenty Seven scores. Our results that physical climate risk reduces leverage after 2015 remains similar. We thus rule out the possibility that our findings account for transition risks rather than physical risks. Moreover, we also verify that these results are not driven by some particular industries, as they remain qualitatively unchanged after the exclusion of the 5 or 10 most represented in-sample industries or after elimination of the 5 most represented industries in each of the 2 risk-level groups.
Sautner et al. (2020) propose a method that identifies firm-level climate change exposure to climate change. They use transcripts of earnings conference calls by listed firms to build firm- year climate change measures. One of their measures is related to the exposure of each firm to physical climate risk. We use these data to create a dummy equal to one when climate risk is mentioned in earnings conference calls of a firm in a given year. We rerun our difference-indifferences test (Table 4) and replace the dummy variable "high climate risk" with this new dummy variable. Our results remain similar.

Conclusion
In this paper, we analyze the impact of the climate risk rating on firms' leverage. We use new forward-looking measures for physical climate risk at the firm level. Our work builds on the capital structure and climate risk literature. We find that firms exposed to greater climate risk are less leveraged in the post-2015 period, i.e., after the Paris Agreement (COP21) and the call from the Financial Stability Board for standard measures and disclosures of climate risks. We also show that the reduction in debt related to climate risk is shared between a demand effect and a supply effect. On the one hand, we find that, after 2015, increased climate risk lowers financial leverage and increases net equity issuance. The reduction in leverage is mainly observed for firms with low CSR performance, suggesting that high CSR firms are likely to take proactive actions to handle their climate risk rather than decrease their leverage. On the other hand, we find that the reduction in debt related to climate risk is at least partly due to a supply effect, as lenders charge higher interest rates to high climate risk firms. Overall, our results suggest that over the recent period, climate risk has become an important factor in understanding the capital structure of firms.

Figure 1. Leverage and climate risk around the Paris Agreement
This figure plots the effect of climate risk on leverage over time using the following regression: where represents the effect of climate risk on leverage over time (with 2010 as the reference year), is a vector of control variables (EBIT, Log Age, TobinQ, OpEx, R&DExp, Log TotAssets, PPE, CreditRating), and accounts for firm fixed effects and year fixed effects. In (a), climate risk is measured by CRIS. In (b), climate risk is measured by Four Twenty Seven. Bands corresponding to 99% confidence intervals based on standard errors clustered by company are included.  Table 1 Descriptive statistics.
This table reports summary statistics. Panel A presents the descriptive statistics for the CRIS climate variables. Each firm of the panel is covered by eight CRIS climate grades (an overall rating and seven subrisk ratings). Panel B presents the descriptive statistics for the Four Twenty Seven climate variables. Each firm of the panel is covered by five Four Twenty Seven climate grades (an overall rating and four subrisk ratings). In Panel C, descriptive statistics of various firm-year characteristics are reported for the full sample. In Panel D, descriptive statistics of firm-year characteristics are disaggregated between low climate risk (<40 th percentile) and high climate risk (>60 th percentile) observations. In Panel E, descriptive statistics of bank loan and borrower characteristics are reported for the full sample. In Panel F, descriptive statistics of bond and bond issuer characteristics are reported for the full sample. All Compustat, Thomson-Reuters, Dealscan and Bloomberg variables Table 2 Climate risk and long-term debt: book leverage.
This table presents difference-in-differences estimates for the leverage before and after the year 2015. All regressions report estimates using as independent variables the interaction between Post2015 and a dummy variable equal to 1 if the climate risk indicator is above the 60 th percentile and 0 if the climate risk indicator is below the 40 th percentile. The dependent variable is BookLev in Columns (1) and (2), and MarketLev in Columns (3) and (4). The regressions are conducted on all firm-year observations except those between the 40 th and the 60 th percentiles of the climate risk indicator. All regressions include firm and year fixed effects. Appendix A presents variable definitions. The sample comprises all firms in the MSCI World index from 2010 to 2019, excluding financial firms (SIC . Standard errors are clustered at the firm level. T-statistics are reported in parentheses. ***, **, and * indicate significance at the 1%, 5% and 10% levels, respectively.
This table presents OLS estimates of the effects of overall climate risk on the cost of bond loans, using Spread as the dependent variable. Columns (1) to (3) report estimates using the CRIS measure of climate risk. Column (4) to (6) report estimates using the Four Twenty Seven measure of climate risk. Regressions (1) and (4) are conducted on the total sample. Regressions (2) and (5) cover the high risk companies with a climate risk rating above the 60 th percentile, and regressions (3) and (6) cover the low risk companies with a climate risk rating below the 40 th percentile. All regressions include firm, loan characteristics (seniority, number of loans to the company on the same date, loan purpose, and secured/unsecured status), and year fixed effects. Appendix A presents variable definitions. The total sample comprises all vanilla fixed-coupon bond loans over $100 million with a maturity of more than 3 years granted to firms in the MSCI World index from 2010 to 2019, excluding financial firms (SIC . Standard errors are clustered at the firm level. T-statistics are reported in parentheses. ***, **, and * indicate significance at the 1%, 5% and 10% levels, respectively. (1)  Climate risk and cost of bank loans.
This table presents estimates of the effects of overall climate risk on the cost of bank loans using Spread as the dependent variable. Columns (1) to (3) report estimates using the CRIS measure of climate risk. Column (4) to (6) report estimates using the Four Twenty Seven measure of climate risk. Regressions (1) and (4) are conducted on the total sample. Regressions (2) and (5) cover the high risk companies with a climate risk rating above the 60 th percentile, and regressions (3) and (6) cover the low risk companies with a climate risk rating below the 40 th percentile. All regressions include firm, loan characteristics (loan and repayment types, seniority, number of loans to the company on the same date, loan purpose, and secured/unsecured status), and year fixed effects. Appendix A presents variable definitions. The total sample comprises all bank loans over $100 million with a maturity of more than 3 years granted to firms in the MSCI World index from 2010 to 2019, excluding financial firms (SIC . Standard errors are clustered at the firm level. T-statistics are reported in parentheses. ***, **, and * indicate significance at the 1%, 5% and 10% levels, respectively.

Appendix B. Description of the CRIS and Four Twenty Seven methodologies
CRIS Four Twenty Seven General overview CRIS ratings capture the increase in risk due to the increase in intensity or frequency of the climate-related hazards in the future due to global warming. They do not capture the absolute risk from future climate or weather. The historical reference period is .
Scores range from 0 to 99. The higher the score, the higher the risk. Each company receives one rating, with the assumption that a company's climate exposure is stable over a few years.
Four Twenty Seven ratings capture both historical risks and the increase in intensity or frequency of the climate-related hazards in the future.
Scores range from 0 to 100. The higher the score, the higher the risk. Each company receives one rating, with the assumption that a company's climate exposure is stable over a few years. Scoring principles For each company, CRIS first identifies the geographic and sectoral breakdown of the activities (fixed assets or revenues depending on the sectoral capital intensity). Then, CRIS assigns a rating by combining climate projections for the relevant locations with sectoral and sovereign vulnerability assessments.

Risks covered
Companies are scored on their Operations Risks by aggregating the climate hazards of each of their sites. Scores at the site level depend on the location of the site and on the company's activity. Supply chain and market risks depend on industry and country factors.