Carbon emissions and crash risk: US evidence from firm-level data

Escola de Economia e Gestão Sara Filipa Ribeiro Rodrigues Carbon Emissions and Crash Risk: US evidence from firm-level data dezembro de 2024 Universidade do Minho Universidade do Minho Escola de Economia e Gestão Sara Filipa Ribeiro Rodrigues Carbon Emissions and Crash Risk: US evidence from firm-level data Dissertação de Mestrado Mestrado em Finanças Trabalho efetuado sob a orientação do Professor Doutor Nelson Manuel Pinho Brandão dezembro de 2024 Costa Areal i DIREITOS DE AUTOR E CONDIÇÕES DE UTILIZAÇÃO DO TRABALHO POR TERCEIROS Este é um trabalho académico que pode ser utilizado por terceiros desde que respeitadas as regras e boas práticas internacionalmente aceites, no que concerne aos direitos de autor e direitos conexos. Assim, o presente trabalho pode ser utilizado nos termos previstos na licença abaixo indicada. Caso o utilizador necessite de permissão para poder fazer um uso do trabalho em condições não previstas no licenciamento indicado, deverá contactar o autor, através do RepositóriUM da Universidade do Minho. Licenc  a concedida aos utilizadores deste trabalho Atribuico-NoComercial-SemDerivaces CC BY-NC-ND https://creativecommons.org/licenses/by-nc-nd/4.0/ ii Acknowledgments First and foremost, I would like to thank my supervisor, Professor Nelson Areal, for bringing the weight of his considerable experience and knowledge to this dissertation and for his guidance throughout this journey. A heartfelt thank you to my family, especially my parents and my brother, and close friends for their love and encouragement. A special thanks to my boyfriend for his unwavering support and stimulating discussions. His presence has been a significant source of motivation. Thank you all. iii Statement of Integrity I hereby declare having conducted this academic work with integrity. I confirm that I have not used plagiarism or any form of undue use of information or falsification of results along the process leading to its elaboration. I further declare that I have fully acknowledged the Code of Ethical Conduct of the University of Minho. iv Resumo Num mundo cada vez mais consciente das questões climáticas, compreender as implicações financeiras das emissões de carbono tornou-se crucial. À medida que investidores e governadores lidam com riscos associados a fatores ambientais, é vital examinar de que forma as emissões de carbono influenciam a estabilidade financeira e a dinâmica dos mercados. Esta dissertação investiga a relação entre as emissões de carbono e o risco de queda do preço das ações através de uma análise de 1,279 empresas listadas nos EUA, no período de 1999 a 2022. O estudo baseia-se no enquadramento teórico proposto por Jin and Myers (2006), que sugere que a assimetria de informação leva a colapsos nos preços das ações quando informações ocultas são reveladas. Partindo da premissa de que empresas com maiores emissões de carbono reportadas são mais transparentes e genuínas, o objetivo central é avaliar se estas empresas estão associadas a um menor risco de colapso. Duas medidas de risco de queda são utilizadas – Assimetria Condicional Negativa (𝑁𝐶𝑆𝐾𝐸𝑊) e Volatilidade (𝐷𝑈𝑉𝑂𝐿) – para avaliar esta relação. Os resultados revelam uma associação não significativa entre as emissões de carbono e o risco de queda, medido pela 𝑁𝐶𝑆𝐾𝐸𝑊 e pela 𝐷𝑈𝑉𝑂𝐿, após controlar para os fatores que preveem o risco de queda e para os efeitos fixos de ano e empresa. Estudos futuros deveriam expandir a análise para uma amostra mais ampla e diferentes contextos geográficos de forma a retirar comparações e resultados mais robustos. Palavras-chave: Assimetria Condicional Negativa, Assimetria de Informação, Emissões de Carbono, Risco de Queda do Preço das Ações, Volatilidade. v Abstract In today’s increasingly climate-conscious world, understanding the financial implications of carbon emissions has become crucial. As investors and policymakers grapple with the risks associated with environmental factors, it is vital to examine how carbon emissions influence financial stability and market dynamics. This dissertation investigates the relationship between carbon emissions and stock price crash risk through an analysis of 1,279 U.S. listed firms over the period from 1999 to 2022. The study is grounded in the theoretical framework proposed by Jin and Myers (2006), which suggests that information asymmetry leads to stock price crashes when hidden information is revealed. Based on the premise that companies with higher reported carbon emissions are more transparent and genuine, the main purpose is to evaluate whether these companies are associated with a lower risk of stock price collapse. Two measures of crash risk – Negative Conditional Skewness (𝑁𝐶𝑆𝐾𝐸𝑊) and Down-to-Up Volatility (𝐷𝑈𝑉𝑂𝐿) – are employed to assess this relationship. The findings reveal no statistically significant association between carbon emissions and crash risk as measured by 𝑁𝐶𝑆𝐾𝐸𝑊 and 𝐷𝑈𝑉𝑂𝐿, after controlling for predictors factors of crash risk and year, and firm fixed effects. Further studies should extend the analysis to a broader sample and different geographical contexts to enhance comparative insights. Keywords: Carbon Emissions, Down-to-Up Volatility, Information Asymmetry, Negative Conditional Skewness, Stock Price Crash Risk. vi List of Contents 1. Introduction ................................................................................................................................... 1 2. Literature Review ........................................................................................................................... 3 3. Methodology .................................................................................................................................. 5 3.1. The Sample................................................................................................................................. 5 3.2. Carbon Emissions Measure ......................................................................................................... 5 3.3. Crash Risk Measures ................................................................................................................... 5 3.4. Control Variables ......................................................................................................................... 7 3.5. Empirical Model .......................................................................................................................... 9 4. Data ............................................................................................................................................10 4.1. Initial Stock Data .......................................................................................................................10 4.2. Data Filters ...............................................................................................................................10 4.2.1. Stock filters based on static information .............................................................................11 4.2.2. Stock and stockday filters based on return index information ..............................................13 4.2.3. Stock filters based on carbon emissions measure and industry classification ......................17 4.3. Financial Accounting Variables ...................................................................................................17 4.3.1. Share Turnover ..................................................................................................................19 4.3.2. Book Value of Equity ..........................................................................................................19 5. Empirical Results .........................................................................................................................20 5.1. Descriptive Statistics .................................................................................................................20 5.2. Pearson Correlation Matrix ........................................................................................................23 5.3. Findings ....................................................................................................................................26 6. Conclusions .................................................................................................................................28 References............................................................................................................................................29 Appendices ...........................................................................................................................................33 A. Data Variables Description: Datastream Datatypes ...................................................................33 B. Data Variables Description: ESG Datatypes ..............................................................................35 C. Data Variables Description: Worldscope Datatypes ...................................................................36 D. Industry Classification of Sample Firm .....................................................................................38 5 3. Methodology 3.1. The Sample My sample comprises 1,279 listed firms in the United States spanning from December 1999 to December 2022, encompassing both actively traded stocks and stocks delisted during the analysis period. The sample and time-period selection will be explored further in Section 4. Appendix D displays the breakdown of sample firms by industry, revealing a predominant presence of companies in the Business Services industry (15.95%), with Electronic Equipment (6.96%), Retail (6.33%), and Petroleum and Natural Gas (6.10%) following closely behind. 3.2. Carbon Emissions Measure To effectively measure carbon emissions, it is crucial to acquire a comprehensive understanding of the diverse categories of emissions a company may produce, along with methods to normalize this metric. This normalization is essential to ensure uniformity and comparability across the different companies within my sample. According to the United States Environmental Protection Agency (EPA), scope 1 emissions comprise direct emissions from sources under the organization’s control. Scope 2 includes indirect emissions linked to the acquisition of electricity, steam, heat, or cooling services. Conversely, scope 3 emissions extend beyond direct control or ownership, encompassing emissions from activities associated with assets not owned or controlled by the reporting entity but indirectly influenced within its value chain. Building upon existing literature and acknowledging the limited availability of data in Datastream, this study will narrow its focus to scope 1 carbon emissions, recognized as an indispensable part of corporate carbon responsibility and management. These emissions will be scaled by the firm’s net sales or revenues at the end of each year, providing insights into the carbon emissions (in tons of direct CO2 and CO2 equivalent) per US dollars of revenues or net sales for each firm (Aljughaiman et al., 2024; Qian & Schaltegger, 2017). The variable 𝐶𝐴𝑅𝐵𝑂𝑁 will serve this purpose in the empirical model, enabling to understand the relationship between the level of carbon emissions and the risk of a stock price crash. 3.3. Crash Risk Measures Following the prior literature (Chen et al., 2001; Jin & Myers, 2006), two measures of firm-specific crash risk will be employed. These measures are based on firm-specific daily returns, 𝑅𝑗,𝜏. This ensures that our crash risk measures indicate firm-specific factors instead of broad market movements (Kim et al., 2014). 𝑅𝑗,𝜏 is computed as the natural logarithmic of one plus the residual return, 𝑅𝑗,𝜏 =ln(1 + 𝜀𝑗,𝜏), of the following extended market model: 6 𝑟 𝑗,𝜏 = 𝛼𝑗 + 𝛽1,𝑗𝑟𝑚,𝜏−2 + 𝛽2,𝑗𝑟𝑚,𝜏−1 + 𝛽3,𝑗𝑟𝑚,𝜏 + 𝛽4,𝑗𝑟𝑚,𝜏+1 + 𝛽5,𝑗𝑟𝑚,𝜏+2 + 𝜀𝑗,𝜏 (1) where, 𝑟 𝑗,𝜏 = return on stock 𝑗 on day 𝜏, 𝑟𝑚,𝜏 = return on the value-weighted market portfolio 𝑚 on day 𝜏. The return on stock 𝑗 on day 𝜏, 𝑟 𝑗,𝜏, is computed based on the daily changes in the Total Return Index (RI) 1 value retrieved from Datastream. The return on the value-weighted market portfolio 𝑚 on day 𝜏, is sourced from the Kenneth R. French Library and provides the value-weighted return of all CRSP (Center for Research in Security Prices) firms incorporated in the US and listed on the NYSE, AMEX, or NASDAQ (Fama & French, 2015). The lagged and leading market terms are included to allow for non-synchronous trading as different stocks have different trading frequencies (Dimson, 1979). The residuals, 𝜀𝑗,𝜏, of the extended market model are not evenly distributed and are skewed. By computing the firm-specific daily returns, 𝑅𝑗,𝜏, one can easily identify both crashes and positive jumps in the data symmetrically as the transformation helps to even out the distribution (Hutton et al., 2009). The first measure of crash risk is the Negative Conditional Skewness, 𝑁𝐶𝑆𝐾𝐸𝑊, which confines the asymmetry of the return distribution (Hunjra et al., 2020). It is calculated by taking the negative of the third moment of each stock’s firm-specific daily returns for each year and normalizing it by the cubed standard deviation of firm-specific daily returns. Thus, for each firm j in year t, 𝑁𝐶𝑆𝐾𝐸𝑊 𝑗,𝑡 = −[𝑛(𝑛 − 1)3/2 ∑𝑅3𝑗,𝜏] /[(𝑛 − 1)(𝑛 − 2)(∑𝑅2𝑗,𝜏)3/2] (2) where, 𝑅𝑗,𝜏 = firm-specific daily return for firm 𝑗 on day 𝜏, 𝑛 = number of daily returns during year 𝑡. This measure is multiplied by -1 so that a higher value corresponds to a stock being more crash prone, i.e., having a more left-skewed distribution. 1 In Appendix A, a detailed description of RI is given. 7 The second measure of crash risk is the Down-to-Up Volatility, 𝐷𝑈𝑉𝑂𝐿. For every stock 𝑗 over a fiscal-year period 𝑡, firm-specific daily returns are separated into “down” days, when returns are below the period mean, and “up” days, when returns are above the period mean. The standard deviation is computed for each of these subsamples separately and then we take the log of the ratio of the standard deviation on the down days to the standard deviation on the up days. Specifically, 𝐷𝑈𝑉𝑂𝐿 is calculated as follows: 𝐷𝑈𝑉𝑂𝐿𝑗,𝑡 =𝑙𝑛[(𝑛𝑢− 1)∑𝑅2𝑗,𝜏/(𝑛𝑑− 1)∑𝑅2𝑗,𝜏𝑈𝑝𝐷𝑜𝑤𝑛 ] (3) where, 𝑛𝑢= number of up days in year t, 𝑛𝑑= number of down days in year t. As suggested in Chen et al. (2001), this alternative measure does not involve third moments, and hence is less likely to be overly affected by extreme returns. Again, the convention is that a higher value of 𝐷𝑈𝑉𝑂𝐿 indicates greater crash risk. 3.4. Control Variables For this research, I decided to employ several control variables which researchers have previously found to have an influence on future crash risk. Chen et al. (2001) showed that trading volume, a proxy for the intensity of differences of opinion among investors, is a predictor of stock price crash risk. This is explained by the Hong and Stein (1999) model 2 which predicts that negative skewness in returns will be most pronounced around periods of heavy trading volume. Following their baseline specification, I used detrended turnover, 𝐷𝑇𝑈𝑅𝑁, to account for this impact by computing the average monthly share turnover in year 𝑡 minus the average monthly share turnover in year 𝑡 − 1 (Kim et al., 2014; Yildiz & Karan, 2020). The reason for detrending is to remove any turnover component that can be considered as a relatively stable firm characteristic, adhering to a conservative approach. Detrending is then able to capture the intensity of disagreements in the market (Murata & Hamori, 2021). Chen et al. (2001) discovered that historical returns and market-to-book ratios also play a role in predicting the risk of market crashes. These relationships are perhaps most clearly suggested by models of stochastic bubbles in which high past returns or high market-to-book ratios (glamour stocks) suggest a prolonged buildup of the bubble. Consequently, when the bubble bursts and prices revert to fundamental levels, the 2 The Hong-Stein model relies on differences of opinion among investors regarding fundamental value and the presence of short-sales constraints for some. When bearish investors with constraints are forced to sell all shares due to differences of opinion, their information is not fully incorporated into prices, creating a corner solution. As more-bullish investors exit the market, the initially bearish group may become support buyers, revealing more about their signals. This process results in hidden information coming to light during market declines, leading to negatively skewed returns. 8 decline is more pronounced. This heightened decline is the underlying reason for both being anticipated to have a higher risk of crashing. I thus control for past returns, 𝑅𝐸𝑇, calculated as the mean of firm-specific daily returns over the fiscal year times 100, and for the market-to-book ratio, 𝑀𝐵, calculated as the market value of equity divided by the book value of equity (Kim et al., 2014). Furthermore, it has been well documented in several studies (Chen et al., 2001; Harvey & Siddique, 2000) that skewness is more negative on average for large-cap firms. Specifically, Chen et al. (2001) developed a discretionary-disclosure hypothesis grounded in the notions that (i) managers tend to promptly reveal positive news but gradually release negative ones, and (ii) managers of small companies can hide bad news more easily using this approach. Hence, we control for firm size, 𝑆𝐼𝑍𝐸, calculated as the natural logarithm of the market value of equity (Kim et al., 2014; Murata & Hamori, 2021). The next control variable is stock volatility, 𝑆𝐼𝐺𝑀𝐴, as more volatile stocks are likely to be more crash prone. Foundational articles (Chen et al., 2001; Hutton et al., 2009) are based on a “volatility feedback” mechanism in which the release of good news, while positive, introduces market volatility, and bad news, in addition to its direct negative impact, is further magnified by an increased risk premium. However, some studies have found the opposite relationship between stock volatility and crash risk (Murata & Hamori, 2021). 𝑆𝐼𝐺𝑀𝐴 is calculated as the standard deviation of firm-specific daily returns over the fiscal year. In addition, I control for financial leverage, 𝐿𝐸𝑉, calculated as total long-term debt divided by total assets, and for profitability measured by return on assets, 𝑅𝑂𝐴, (Kim et al., 2014). Prior literature used leverage as a proxy for default risk but failed to find support for this proposition (Habib et al., 2018). For instance, Hutton et al. (2009) and Kim et al. (2011a, 2011b) find a negative association between leverage and crash risk, when leverage is expected to be positively correlated with bankruptcy risk (Campbell et al., 2008). According to Habib et al. (2018), one possible explanation for this unexpected finding is that highly leveraged firms might initially be undervalued by investors, reducing the likelihood of subsequent price crashes. As previously mentioned, glamour stocks are considered more crash-prone, therefore a positive relationship is expected between profitability, 𝑅𝑂𝐴, and crash risk. 9 3.5. Empirical Model Following the empirical model developed by (Kim et al., 2014), in order to examine the impact of carbon emissions on future stock price crash risk, the following regression model will be estimated: 𝐶𝑅𝐴𝑆𝐻_𝑅𝐼𝑆𝐾𝑗,𝑡 = 𝛽0+ 𝛽1(𝐶𝐴𝑅𝐵𝑂𝑁𝑗,𝑡−1)+ 𝛽2(𝐷𝑇𝑈𝑅𝑁𝑗,𝑡−1) + 𝛽3(𝑅𝐸𝑇 𝑗,𝑡−1) + 𝛽4(𝑀𝐵𝑗,𝑡−1) + 𝛽5(𝑆𝐼𝑍𝐸𝑗,𝑡−1) +𝛽6(𝑆𝐼𝐺𝑀𝐴𝑗,𝑡−1) + 𝛽7(𝐿𝐸𝑉 𝑗,𝑡−1)+ 𝛽8(𝑅𝑂𝐴𝑗,𝑡−1) + 𝛽𝑥(𝐷𝑌𝑒𝑎𝑟)+ 𝛽𝑦(𝐷𝐹𝑖𝑟𝑚) + 𝜀𝑗,𝑡 (4) In Equation 4, 𝐶𝑅𝐴𝑆𝐻_𝑅𝐼𝑆𝐾 is the dependent variable and is a proxy for 𝑁𝐶𝑆𝐾𝐸𝑊 or 𝐷𝑈𝑉𝑂𝐿. 𝐶𝐴𝑅𝐵𝑂𝑁 is the main independent variable as it contains the carbon emissions measure. All independent variables are lagged by one year so that we can test if the carbon emissions level of firm 𝑗 in year 𝑡–1, 𝐶𝐴𝑅𝐵𝑂𝑁𝑗,𝑡−1, predict crash risk in year 𝑡, 𝐶𝑅𝐴𝑆𝐻_𝑅𝐼𝑆𝐾𝑗,𝑡. The previously outlined control variables serve the purpose of accounting for firm-specific factors anticipated to influence the likelihood of future stock price crashes. Therefore, for each firm 𝑗 in year 𝑡–1, we incorporate controls for changes in trading volume, 𝐷𝑇𝑈𝑅𝑁𝑗,𝑡−1; past returns, 𝑅𝐸𝑇 𝑗,𝑡−1; market-to-book ratio, 𝑀𝐵𝑗,𝑡−1; firm size, 𝑆𝐼𝑍𝐸𝑗,𝑡−1; stock volatility, 𝑆𝐼𝐺𝑀𝐴𝑗,𝑡−1; financial leverage, 𝐿𝐸𝑉 𝑗,𝑡−1; and return on assets, 𝑅𝑂𝐴𝑗,𝑡−1, (Kim et al., 2014; Yildiz & Karan, 2020). All these firm-specific factors, including firmspecific daily returns, 𝑅𝑗,𝜏, are winsorized at the 1st and 99th percentiles, a widely used technique in the literature to mitigate the impact of outliers (Zaman et al., 2021). The regression model also includes year and firm fixed effects to account for unobserved heterogeneity. 10 4. Data The Refinitiv Eikon Datastream platform serves as the primary source for data in this dissertation. In subsection 4.1, detailed insights will be provided regarding the initial stock universe, while in subsection 4.2 I will present the subsequent application of filters to arrive at the final sample. Adhering to the guidelines outlined by Landis and Skouras (2021), particularly addressing static and return index information, was imperative to enhance data accuracy and mitigate survivorship bias. These guidelines, tailored for the US market, ensure a more nuanced approach to working with Datastream data, thereby contributing to the overall quality of the results. Finally, subsection 4.3 will delve into a comprehensive presentation of all firmrelated information. 4.1. Initial Stock Data To create the initial stock universe, while Landis and Skouras (2021) advocated for an alternative approach, which involved extracting information on all equities traded across every stock exchange within each country using TDS’ Navigator GUI, I decided not to follow this method due to its time-consuming nature and limitations in data availability from the Datastream platform. Instead, I opted by relying on market-specific constituent lists provided by Datastream and Worldscope, an approach commonly employed by research teams working with data from these sources, as it is more efficient and aligns with established practices in the field (Karolyi et al., 2012; Schmidt et al., 2019). Accordingly, different constituent lists were extracted and merged, including: (i) the FUSALL* lists, comprising all currently active stocks, (ii) the DEADUS* lists, consisting of delisted stocks, and (iii) the WSUS* lists provided by Worldscope. Worth mentioning that the stock universe should include not only stocks that were active at the time of the study, but also stocks that were inactive due to mergers, acquisitions, or failures, but show a portion of the price history over the entire analysis period (Landis & Skouras, 2021). Employing this method yielded a dataset comprising 116,834 stocks. 4.2. Data Filters In this subsection, I will provide a comprehensive overview of the filtering process employed to address data challenges within the initial sample. The filtering strategies encompass two main categories: stock filters, which involve excluding the entire historical record of an instrument, and stockday filters, which target specific days for individual stocks. These filters are subcategorized based on the type of information they use, including static data and return indexes. Considering the time-consuming nature of applying all filters provided by Landis and Skouras (2021), I opted to select those identified as most essential when working with Datastream data and most adaptable to my research. The authors provide users with the flexibility to apply their filters in the manner they deem 11 appropriate, emphasizing that their work aims to establish best practices for working with these data but acknowledging that there is no “source of truth”. This approach underscores the importance of exercising judgement in filter selection and implementation. At the end of this subsection, I illustrate the need of removing specific stocks based on carbon emissions data availability and industry classification. 4.2.1. Stock filters based on static information For each stock on each constituent list, the following variables were retrieved (Oliveira, 2023): Type of Instrument (TYPE), Datastream Code (DSCD), Base Date (BDATE), Expanded Security Name (ENAME), Datastream Exchange Mnemonic (EXMNEM), Geographical Classification (GEOGN), ISIN (ISIN), Primary Indicator Flag (ISINID), Code Local (LOC), Currency (PCUR) and Security Type Code (TRAC). In Appendix A, the description of these variables is given. Table 1Error! Reference source not found. presents a description of the filters, that will be further explained, organized in the order in which they are applied in this Dissertation. Firstly, I focused exclusively on instruments classified as equities, ensuring that all observations represented equity securities (where the datatype TYPE should be “EQ”). This filter is pertinent because constituent lists often include various types of securities, such as American Depository Receipts (ADR), Closed-end Funds (CF), and Global Depository Receipts (GDR), which could otherwise introduce noise or inaccuracies into the analysis. Next, I narrowed down the dataset to stocks that are traded on US exchanges (EXMNEM). According to Landis and Skouras (2021), some researchers prefer to exclude stocks listed on secondary exchanges because they assume these stocks are likely to be small or traded over-the-counter (OTC). However, using this approach with Datastream data can lead to a problem as this database only provides information on the current exchange where a stock is listed. So, if researchers exclude secondary exchanges to avoid issues related to small stocks, they might also exclude stocks that got moved from main stock exchanges due to poor performance, which can create bias in the sample. Following Oliveira (2023), the US has 12 exchanges plus the OTC markets: NYSE (NYS), NYSE MKT (ASE), NYSE ARCA (XC), NASDAQ (NAS), NASDAQ/NMS (NMS), OTC Bulletin (XBQ), Non-Nasdaq OTC (OTC), Boston (BOS), Chicago (MID), Pacific (PSE), Philadelphia (PHL), and BARS (E1). Therefore, I will only consider stocks traded on these exchanges. To refine the selection further, I included only instruments that could be classified as common stocks based on both the TRAC and ENAME variables. As noted by Landis and Skouras (2021), the classification of a stock as common remains consistent over time, thus using this filter helps prevent survivorship bias. Column 3 of Table 1 provides information of the valid security type codes and country-specific text strings that should 12 not be present in the extended names of stocks. Additionally, stocks with expanded names containing country-specific local listing identifiers were excluded, indicating potential listings in other countries. This additional filter aids in identifying cross-listed stocks and ensures their exclusion from the sample. Moreover, I ensured that the dataset only consisted of companies domiciled in the United States (GEOGN) and securities traded in the US dollar (PCUR). To prevent any duplicates from affecting my analysis, I removed any redundant entries based on both LOC and DSCD codes. The datatype LOC represents a code assigned to an instrument by the exchange where it is traded. When Datastream designates a stock as primary (datatype ISIND equals “P”), the authors noticed that all other stocks with the same local code are not common stocks. Therefore, it is recommended to exclude any stocks with a non-unique local code and whose ISINID is not “P”, as long as there is at least one stock with this local code that does have ISINID = “P”. After applying these filters, the final sample consisted of 29,968 observations. Table 1: Filters based on static information. This table outlines all the filters applied to the static variables listed in the first column. It explains the rationale for each filter and specifies the accepted values used in the sample construction. Variable Purpose Accepted Values Type of Instrument (TYPE) To guarantee that all observations are Equities. ‘EQ’. DataStream Exchange Mnemonic (EXMNEM) To exclude stocks that are not traded on US exchanges. ‘NAS’, ‘NYS’, ‘OTC’, ‘ASE’, ‘XSQ’, ‘XBQ’, ‘NMS’, ‘BOS’, ‘MID’, ‘PSE’, ‘PHL’, ‘E1’. Security Type Code (TRAC) To exclude stocks that cannot be classified as common stocks. ‘ORD’, ‘ORDSUBR’, ‘FULLPAID’, ‘UNKNOWN’, ‘UNKNOW’, ‘KNOW’, ‘NA’. Expanded Security Name (ENAME) To exclude stocks identifiable as non-common stocks based on specific text strings within their extended names. Do not contain: 'TRUST','REPR','RIGHT','SERIES','NV','IV TST','REAL ESTATE INVESTMENT','REALTY','RLTY','ROYALTY INVESTMENT','ASSET INVESTMENT','CAPITAL INVESTMENT','ASSET MANAGEMENT','CAPITAL MANAGEMENT','INVESTMENT MANAGEMENT','VENTURE CAPITAL','FINANCIAL SHBI','PROPERTY INVESTORS','INCOME PROPERTY','UNITS','UNIT','LIMITED PARTENERSHIP','FUND','EQUITY PARTNERS','LIMITED VOTING','SUB VOTING','TIER ONE SUB','VARIABLE 13 VOTING','NON VOTINGREIT','RESIDENTIAL','R E I T','BENEFICIAL','BENEFICIARY','BENEFIT INTEREST','BEN INTEREST','SH BEN INT','WARRANT','WRTS','L P','L P INTEREST','LP UT','HOLDINGS LP','PARTENERS UNIT','PART INT','UNIT PARTENERSHIP','UNIT LIMITED','MORTGAGE','REAL ESTATE','CERTIFICATE','NO PAR VALUE','HOLDING UNIT','BACKED','ST MIN','CORTS','TORPS','TOPRS','SECURITIES TRUPS','QUIPS','STRATS HIGH YIELD','TOTAL RETURN','DIVERSIFIED HOLDINGS','(SICAV)','DEPOSITARY', 'DEPOSITOR','RECEIPT','REP & SHARES','GLOBAL SHARES','ADR','GDR','EXPD.','EXPIRED','DUPLI CATE','CONVERTIBLE','CNVRT.','CONVRT.','EX CH.','DEBANTURE','(DEB)','NIL PAID','STRUCTURED ASSET','CALLABLE','FLOATING RATE','ADJUSTABLE','REDEEMABLE','PAIRED CTF','CONSOLIDATED','INSURED','CAPITAL SHARES','DEBT STRATEGIES','LIQUIDATING','LIQUID UNIT','L UNIT','- LASD','ACQUISITION','CAP UNIT','INCOME UNIT','PREFERRED','(NYS)','(NAS)','(ASE)','(OT C)','(XSQ)','(XQB)'. Geographical Classification of company (GEOGN) To exclude stocks that are not domiciliated in the US. ‘UNITED STATES’. Currency (PCUR) To remove all securities traded in a currency other than the US dollar. ‘U$’. Code Local (LOC) and Primary Indicator Flag (ISINID) To remove duplicated observations and ensure the exclusion of any non-common stocks from the sample. No duplicated codes. If duplicates are found, remove them only if ISINID is not equal to ‘P’, provided that at least one stock with this LOC has ISINID equal to ‘P’. Datastream Code (DSCD) To remove duplicated observations. No duplicated codes. 4.2.2. Stock and stockday filters based on return index information For the 29,968 stocks previously found, I extracted the Total Return Index (datatype RI) daily from December 1999 to December 2022. The choice of this timeframe is underpinned by two principal considerations. 14 Firstly, data reliability for US stocks significantly improves after December 1984, a point emphasized by Landis and Skouras (2021) for ensuring dataset robustness. However, due to limitations in Datastream and the emergence of carbon emissions disclosure from companies in the late 1990s and early 2000s, I opted to begin data extraction from December 1999. Secondly, to include the most recent and pertinent information, I extended the dataset to cover data up until the end of 2022. For the data extraction process, I used the Datastream’s DPL function, ensuring precision by retaining up to six decimal points, which is the maximum available number of decimal places in the database. It is important to note that Datastream does not provide direct returns but instead offers a return index that monitors the hypothetical value of an investment with reinvested cash flows, notably dividends. Consequently, even small values in this index can significantly affect return calculations due to rounding issues, emphasizing the importance of this step. For certain stocks, the return index is unavailable for any date, and as is standard in the literature, these instruments were consequently excluded. For the first filtering step, I obtained information regarding each stock’s status (datatype ESTAT) and its corresponding delisting date (datatype TIME). When a company’s stock is delisted, it means it is no longer actively traded on a public exchange. Initially, certain stocks were found to have been delisted prior to December 1999, and as a result, they were promptly excluded from the sample. For stocks that underwent delisting after this date, it is important to recognize that there can be a delay between the last date for which trading data is available and the official date when the stock is marked as delisted in the database. Consequently, many stocks exhibit constant return indexes toward the conclusion of their trading history even when the series have been truncated at their delisting dates. To address this concern, the first step involved the removal of data related to stocks showing constant values beyond their delisting dates. Additionally, a comparison was made between the date of the last non-zero return and the delisting date. If a difference of more than 10 days was identified, the tenth and subsequent daily observations were excluded; otherwise, the delisting date retrieved from Datastream was used as the end date of the series. Stocks with constant return indexes throughout the entire series were also excluded to ensure data reliability 3 . In the second phase of the filtering process, I started by computing the daily percent change in the Total Return Index for each day 𝜏, or daily return (%), within the time series, taking into account the individual stock’s initial and end dates. 3 Please note that I have applied the condition of excluding stocks with constant return indexes to both dead and active companies in my sample. This ensured that subsequent filters were applied smoothly, avoiding any potential issues. 21 the negative impact of the Russian-Ukraine war on the global financial market which began in February 2022 (Assaf et al., 2023). As for the 𝐶𝐴𝑅𝐵𝑂𝑁𝑡−1, it shows relatively stable values over the years, indicating a gradual increase in reporting and potentially in emissions themselves, especially after 2015. This can be linked to the reporting obligations established by the Paris Agreement in December 2015. Under its Enhanced Transparency Framework, all Parties are mandated to report their greenhouse gas (GHG) balance biannually and to monitor the progress of individual countries in achieving their mitigation targets (Perugini et al., 2021). Nevertheless, a report developed by As You Sow 7 in 2022, evaluated the effort of 55 major US corporations in reducing GHG emissions in alignment with the Paris Agreement’s target of limiting global temperature rise to 1.5 degrees Celsius, aiming for net zero emissions by 2050. They found that most companies still lack comprehensive 1.5 degree-aligned GHG reduction goals and are not making sufficient progress toward net zero emissions. Moreover, even though most companies report Scope 1 emissions, disclosure of carbon offsets is often unclear, which underscores the critical need of exercising caution while looking at carbon emissions figures over time. Moreover, there does not appear to be any specific pattern between the lagged carbon emissions and crash risk variables. Table 4: Mean values for crash risk and lagged carbon emissions measures (2000-2022). This table presents the average values of crash risk, 𝑁𝐶𝑆𝐾𝐸𝑊𝑡 and 𝐷𝑈𝑉𝑂𝐿𝑡, and lagged carbon emissions, 𝐶𝐴𝑅𝐵𝑂𝑁𝑡−1, measures from 2000 to 2022. Refer to Section 3 for variables definition. 𝐘𝐄𝐀𝐑𝐭 𝑵𝑪𝑺𝑲𝑬𝑾𝒕 𝑫𝑼𝑽𝑶𝑳𝒕 𝑪𝑨𝑹𝑩𝑶𝑵𝒕−𝟏 2000 -0.2261 0.0028 - 2001 -0.1430 0.0018 - 2002 -0.0125 0.0016 - 2003 -0.1627 0.0007 0.2062 2004 -0.1923 0.0005 0.1496 2005 -0.1801 0.0004 0.1608 2006 -0.1777 0.0003 0.1357 2007 -0.1932 0.0005 0.1240 2008 0.0042 0.0016 0.1294 2009 -0.2204 0.0013 0.1417 2010 -0.1881 0.0004 0.1777 7 As You Sow is a non-profit organization that promotes environmental and social corporate responsibility through shareholder advocacy, coalition building, and innovative legal strategies. It engages in a variety of initiatives aimed at encouraging companies to adopt more sustainable and socially responsible practices. These initiatives include advocating for corporate transparency, reducing GHG emissions, promoting fair labor practices, and minimizing environmental impacts. 22 2011 -0.1036 0.0004 0.1467 2012 -0.0989 0.0003 0.1289 2013 -0.1080 0.0002 0.1291 2014 -0.0657 0.0003 0.1427 2015 -0.0042 0.0004 0.1426 2016 -0.0332 0.0007 0.1614 2017 0.0978 0.0003 0.1786 2018 0.1318 0.0006 0.1912 2019 0.0836 0.0004 0.1915 2020 -0.0719 0.0018 0.1804 2021 -0.0772 0.0007 0.2048 2022 0.0395 0.0006 0.1601 Table 5 presents the descriptive statistics of the sample. Building on the insights provided earlier, the mean values of crash risk measures, 𝑁𝐶𝑆𝐾𝐸𝑊𝑡 and 𝐷𝑈𝑉𝑂𝐿𝑡, are -0.0724 and 0.0008, respectively. These values suggest that, on average, the sample firms have more right-skewed firm-specific daily returns and slightly higher volatility in firm-specific returns on down days compared to up days, although the difference is small. The medians (-0.0557 for 𝑁𝐶𝑆𝐾𝐸𝑊𝑡 and 0.0001 for 𝐷𝑈𝑉𝑂𝐿𝑡) are slightly lower than the respective means, indicating the presence of a few relatively high values that raise the mean, while most data points are clustered towards the lower end of the range. For 𝑁𝐶𝑆𝐾𝐸𝑊𝑡, the estimates are similar to those found by Chen et al. (2001) and Yildiz and Karan (2020) but differ from Kim et al. (2014) and Murata and Hamori (2021). For 𝐷𝑈𝑉𝑂𝐿𝑡, the mean is very close to zero, which is lower than the average reported in the existing literature, likely due to differences in sample periods and sample construction. The average value of the total carbon emission per US dollars of revenues or net sales is 0.1682. The average change in monthly trading volume (as a percentage of shares outstanding) is 0.0214. Furthermore, the average firm in the sample has a firm-specific daily return of 13.27%, a market capitalization of $3.24 billion 8 , a market-to-book ratio of 0.1650, a daily return volatility of 0.0209, a leverage of 0.2643, and a return on assets of 0.0490. 𝑅𝐸𝑇𝑡−1 and 𝑆𝐼𝑍𝐸𝑡−1 have the highest standard deviations, suggesting (i) high variability in firm-specific daily returns, and (ii) diverse set of firms in the sample from small-cap to largecap companies. 8 To interpret the mean value of the 𝑆𝐼𝑍𝐸 variable in terms of actual market value, one must reverse the log transformation by exponentiating the mean value. 23 Table 5: Summary of descriptive statistics. This table presents the descriptive statistics for the variables used in the dissertation. The data comprises a sample of 1,279 US-listed stocks from 1999 to 2022, including both active stocks and those that have been delisted but retain a portion of their price history throughout the analysis period. Financials (SIC codes 6000-6999) and utilities (SIC codes 4900-4949) are excluded, based on the Fama-French 48 industry classifications. Refer to Section 3 for variable definitions. VARIABLES MEAN MEDIAN STANDARD DEVIATION 25TH PERCENTILE 75TH PERCENTILE 𝑫𝑼𝑽𝑶𝑳𝒕 0.0008 0.0001 0.0030 0.0000 0.0002 𝑵𝑪𝑺𝑲𝑬𝑾𝒕 -0.0724 -0.0557 0.5432 -0.3805 0.2596 𝑪𝑨𝑹𝑩𝑶𝑵𝒕−𝟏 0.1682 0.0132 0.4194 0.0026 0.0848 𝑫𝑻𝑼𝑹𝑵𝒕−𝟏 0.0214 0.0077 0.5948 -0.2087 0.2344 𝑹𝑬𝑻𝒕−𝟏 0.1327 -0.0211 0.8778 -0.0956 0.0543 𝑴𝑩𝒕−𝟏 0.1650 0.1137 0.1662 0.0629 0.2030 𝑺𝑰𝒁𝑬𝒕−𝟏 8.0842 8.0577 1.7676 6.9209 9.2774 𝑺𝑰𝑮𝑴𝑨𝒕−𝟏 0.0209 0.0173 0.0128 0.0124 0.0250 𝑳𝑬𝑽𝒕−𝟏 0.2643 0.2376 0.1855 0.1323 0.3646 𝑹𝑶𝑨𝒕−𝟏 0.0490 0.0636 0.1199 0.0248 0.1035 5.2. Pearson Correlation Matrix Table 6 reports the results of the Pearson correlation matrix. The highest correlations are observed between 𝐷𝑈𝑉𝑂𝐿 and 𝑆𝐼𝐺𝑀𝐴 (0.73), and 𝐷𝑈𝑉𝑂𝐿 and 𝑅𝐸𝑇 (0.58), indicating that stocks with greater volatility and higher past returns are positively correlated with increased crash risk, as measured by 𝐷𝑈𝑉𝑂𝐿. Conversely, there is a negative correlation between 𝑁𝐶𝑆𝐾𝐸𝑊 and 𝑆𝐼𝐺𝑀𝐴 (-0.22), and 𝑁𝐶𝑆𝐾𝐸𝑊 and 𝑅𝐸𝑇 (-0.43) within my sample. Additionally, there is a strong negative correlation between 𝑆𝐼𝐺𝑀𝐴 and both 𝑆𝐼𝑍𝐸 (-0.53) and 𝑅𝑂𝐴 (-0.42), indicating that smaller firms and those with lower profitability tend to experience higher volatility in firm-specific daily returns. Furthermore, the positive and statistically significant correlation between 𝐶𝐴𝑅𝐵𝑂𝑁 and 𝐷𝑈𝑉𝑂𝐿 indicates a weak positive relationship (0.1356), suggesting that firms with higher carbon emissions are associated with slightly higher down-to-up volatility. Conversely, the negative and statistically significant correlation between 𝐶𝐴𝑅𝐵𝑂𝑁 and 𝑁𝐶𝑆𝐾𝐸𝑊 reveals a very weak negative relationship, in contrast to the positive relationship found with 𝐷𝑈𝑉𝑂𝐿, indicating that higher carbon emissions might be associated with slightly 24 less negative skewness in stock returns. However, given the weak nature of these relationships, drawing robust conclusions is challenging. 25 COLUMN1 𝑫𝑼𝑽𝑶𝑳 𝑵𝑪𝑺𝑲𝑬𝑾 𝑪𝑨𝑹𝑩𝑶𝑵 𝑫𝑻𝑼𝑹𝑵 𝑹𝑬𝑻 𝑴𝑩 𝑺𝑰𝒁𝑬 𝑺𝑰𝑮𝑴𝑨 𝑳𝑬𝑽 𝑹𝑶𝑨 𝑫𝑼𝑽𝑶𝑳 1.0000 𝑵𝑪𝑺𝑲𝑬𝑾 -0.2128*** 1.0000 𝑪𝑨𝑹𝑩𝑶𝑵 0.1356*** -0.0407*** 1.0000 𝑫𝑻𝑼𝑹𝑵 0.1062*** 0.0250*** 0.0314*** 1.0000 𝑹𝑬𝑻 0.5764*** -0.4294*** 0.0363*** -0.0171** 1.0000 𝑴𝑩 0.0300*** -0.0719*** -0.1902*** 0.0059 0.0525*** 1.0000 𝑺𝑰𝒁𝑬 -0.3222*** 0.1142*** -0.1875*** -0.0268*** -0.1399*** 0.2407*** 1.0000 𝑺𝑰𝑮𝑴𝑨 0.7272*** -0.2182*** 0.2176*** 0.1624*** 0.3701*** -0.0161** -0.5325*** 1.0000 𝑳𝑬𝑽 0.0400*** 0.0059 0.0968*** 0.0216*** 0.0060 -0.1026*** -0.0050 0.0567*** 1.0000 𝑹𝑶𝑨 -0.3061*** 0.0518*** -0.1198*** -0.0160** -0.0734*** 0.1162*** 0.3108*** -0.4177*** -0.0729*** 1.0000 Table 6: Pearson Correlation Matrix. This table presents the Pearson correlation coefficients for the variables used in this dissertation. The data comprises a sample of 1,279 US-listed stocks from 2000 to 2022, including both active stocks and those that have been delisted but retain a portion of their price history throughout the analysis period. Financials (SIC codes 6000-6999) and utilities (SIC codes 4900-4949) are excluded, based on the Fama-French 48 industry classifications. Refer to Section 3 for variable definitions. Note: *, **, and *** indicate significance at the 10%, 5% and 1% levels, respectively. 26 5.3. Findings Table 7 presents the OLS estimation results, with 𝑁𝐶𝑆𝐾𝐸𝑊𝑡 as the dependent variable in Panel A, and 𝐷𝑈𝑉𝑂𝐿𝑡 as the dependent variable in Panel B. Both regressions include fixed effects at the year and firm levels. Examining the results presented in Panel A, the model shows a negative association between the level of carbon emissions per US dollars of revenues or net sales, 𝐶𝐴𝑅𝐵𝑂𝑁𝑡−1, and the risk of a stock price crash in the upcoming year, measured by 𝑁𝐶𝑆𝐾𝐸𝑊𝑡. This relationship seems to be aligned with the initial expectation that being more genuine and transparent in reporting would associate with lower crash risk. However, this relationship did not achieve statistical significance, meaning that there is insufficient evidence to conclude a meaningful association between these variables within the context of the study. The only variable found to be statistically significant at the 1% level is 𝑆𝐼𝑍𝐸𝑡−1 with a coefficient of 0.3091. This positive and significant coefficient suggests that larger firms are more likely to experience higher 𝑁𝐶𝑆𝐾𝐸𝑊𝑡, indicating an increased risk of a stock price crash. This finding supports the discretionarydisclosure hypothesis proposed by Chen et al. (2001), which posits that large-cap firms tend to exhibit more negative skewness on average. This relationship is further corroborated by recent studies (Hunjra et al., 2020; Kim et al., 2014). Turning to the results in Panel B, we observe a positive association between the level of carbon emissions per US dollars of revenues or net sales, 𝐶𝐴𝑅𝐵𝑂𝑁𝑡−1, and the risk of a stock price crash in the upcoming year, 𝐷𝑈𝑉𝑂𝐿𝑡. However, this association also lacks statistical significance. Despite this, the positive relationship appears counterintuitive to my analysis. This discrepancy suggests that 𝐷𝑈𝑉𝑂𝐿𝑡 might be capturing a different dimension of carbon emissions over time. This indicates a need to investigate whether the relationship between emissions and crash risk is more complex than initially assumed. Additionally, the model identifies 𝑅𝐸𝑇𝑡−1 (0.0009***) and 𝑀𝐵𝑡−1 (0.0009*) as predictors of one-year ahead crash risk, with both showing positive and statistically significant relationships at the 1% and 10% levels, respectively. These associations seem to be aligned with the theory behind models of stochastic bubbles in which high past returns and high market-to-book ratios (glamour stocks) are associated with higher risk of stock price crash (Chen et al., 2001). Volatility, 𝑆𝐼𝐺𝑀𝐴𝑡−1 (0.1475***) and financial leverage, 𝐿𝐸𝑉𝑡−1, (-0.0005*) are also found to have a statistically significant impact on the risk of a stock price crash. These results are also in line with the literature, which suggests that high volatility and lower leveraged firms are linked to increased crash risk (Chen et al., 2001; Hutton et al., 2009; Kim et al., 2011a, 2011b). 27 Table 7: Carbon Emissions on Crash Risk - Regression Results. This table presents the regression results of the effect of carbon emissions on firm-level stock price crash risk. The standard errors clustered at the firm and year levels are reported in parentheses. Refer to Section 3 for variable definitions. Note: *, **, and *** indicate significance at the 10%, 5% and 1% levels, respectively. DEPENDENT VAR. = 𝑵𝑪𝑺𝑲𝑬𝑾𝒕 PANEL A 𝑫𝑼𝑽𝑶𝑳𝒕 PANEL B 𝑪𝑨𝑹𝑩𝑶𝑵𝒕−𝟏 -0.0633 (0.0606) 0.0005 (0.0003) 𝑫𝑻𝑼𝑹𝑵𝒕−𝟏 -0.0254 (0.0191) 0.0001 (0.0001) 𝑹𝑬𝑻𝒕−𝟏 0.0226 (0.0019) 0.0009*** (0.0003) 𝑴𝑩𝒕−𝟏 0.2876 (0.1841) 0.0009* (0.0005) 𝑺𝑰𝒁𝑬𝒕−𝟏 0.3091*** (0.0416) -0.0001 (0.0001) 𝑺𝑰𝑮𝑴𝑨𝒕−𝟏 -1.9829 (2.4056) 0.1475*** (0.0308) 𝑳𝑬𝑽𝒕−𝟏 0.0126 (0.1038) -0.0005* (0.0003) 𝑹𝑶𝑨𝒕−𝟏 -0.1247 (0.1881) -0.0001 (0.0007) CONSTANT -0.0701 (0.0961) -0.0021** (0.0006) YEAR FE Yes Yes COMPANY FE Yes Yes OBSERVATIONS 5581 5581 R-SQUARED 0.302 0.751 28 6. Conclusions In this Dissertation, the aim was to examine the effects of carbon emissions on the one-year ahead crash risk for a sample of 1,279 US-listed firms. The research was based on the model developed by Jin and Myers (2006), which argues that stock price crashes result from information asymmetry. When a certain threshold is reached, all the hidden information comes simultaneously, leading to the collapse in stock prices. Two measures of crash risk were employed: the Negative Conditional Skewness, 𝑁𝐶𝑆𝐾𝐸𝑊, which captures the asymmetry of the return distribution, and the Down-to-Up Volatility, 𝐷𝑈𝑉𝑂𝐿, which assesses the asymmetry in stock return volatility between periods of positive and negative returns relative to a stock’s average return. The findings indicate no statistically significant relationship between carbon emissions, 𝐶𝐴𝑅𝐵𝑂𝑁𝑡−1, and one-year ahead crash risk measured by both 𝑁𝐶𝑆𝐾𝐸𝑊𝑡 and 𝐷𝑈𝑉𝑂𝐿𝑡. Regarding the variables of control, despite some findings lacking statistical significance, the analysis identified statistically significant relationships between crash risk and past returns, 𝑅𝐸𝑇𝑡−1, volatility, 𝑆𝐼𝐺𝑀𝐴𝑡−1, firm size, 𝑆𝐼𝑍𝐸𝑡−1, market-to-book ratio, 𝑀𝐵𝑡−1, and financial leverage, 𝐿𝐸𝑉𝑡−1, in line with the literature. It is important to acknowledge that this study encountered several data availability challenges, particularly regarding carbon emissions. Following the methodology of Landis and Skouras (2021), significant time and effort were required to compile an accurate sample. Due to these constraints, direct comparisons with existing literature may not be straightforward. Most prior studies used different databases that provided easier access to more comprehensive datasets, which were not available for this analysis. These limitations underscore the need for improved data accessibility and consistency in future research to facilitate more robust comparisons and conclusions. For future research, it would be valuable to conduct further investigation into the various measures used to compute crash risk and explore how they might capture different information during the same event or time period. Expanding the analysis to a larger sample of firms and including companies from diverse countries could provide more meaningful comparisons and a richer understanding of the relationship between crash risk and carbon emissions in different contexts. Additionally, exploring whether companies with higher carbon emissions experience lower information asymmetry compared to those with lower emissions could offer valuable insights into how this characteristic influence crash risk dynamics. This further investigation may reveal nuances in how carbon emissions impact information transparency and, consequently, stock price stability. 29 References Adhikari, A., & Zhou, H. (2021). Voluntary disclosure and information asymmetry: do investors in US capital markets care about carbon emission? Sustainability Accounting, Management and Policy Journal , 13(1) , 195-220. Aljughaiman, A. A., Cao, N. D., & Albarrak, M. S. (2024). The impact of greenhouse gas emission on corporate’s tail risk. Journal of Sustainable Finance & Investment , 14 (1) , 68-85. Assaf, R., Gupta, D., & Kumar, R. (2023). The price of war: Effect of the Russia-Ukraine war on the global financial market. The Journal of Economic Asymmetries , 28 , e00328. Bistline, J., Blanford, G., Brown, M., Burtraw, D., Domeshek, M., Farbes, J.,…Jones, R. (2023). Emissions and energy impacts of the Inflation Reduction Act. Science , 380 (6652), 1324-1327. Bolton, P., & Kacperczyk, M. (2021). Do investors care about carbon risk? Journal of financial economics , 142 (2), 517-549. Borghei, Z., Leung, P., & Guthrie, J. (2018). Does voluntary greenhouse gas emissions disclosure reduce information asymmetry? Australian evidence. Afro-Asian Journal of Finance and Accounting , 8 (2), 123-147. Campbell, J. Y., Hilscher, J., & Szilagyi, J. (2008). In search of distress risk. The Journal of finance , 63 (6), 2899-2939. Capasso, G., Gianfrate, G., & Spinelli, M. (2020). Climate change and credit risk. Journal of Cleaner Production , 266 , 121634. Caragnano, A., Mariani, M., Pizzutilo, F., & Zito, M. (2020). Is it worth reducing GHG emissions? Exploring the effect on the cost of debt financing. Journal of Environmental Management , 270 , 110860. Chen, J., Hong, H., & Stein, J. C. (2001). Forecasting crashes: trading volume, past returns, and conditional skewness in stock prices. Journal of Financial Economics , 61 (3), 345-345-381. Christoffersen, P. (2012). Elements of financial risk management (2nd ed.). Academic press. da Silva, P. P. (2022). Crash risk and ESG disclosure. Borsa Istanbul Review , 22(4) , 794-811. Dai, P.-F., Xiong, X., Liu, Z., Huynh, T. L. D., & Sun, J. (2021). Preventing crash in stock market: The role of economic policy uncertainty during COVID-19. Financial Innovation , 7 , 1-15. 30 Delbeke, J., Runge-Metzger, A., Slingenberg, Y., & Werksman, J. (2019). The paris agreement. In Towards a climate-neutral Europe (pp. 24-45). Routledge. Dimson, E. (1979). Risk measurement when shares are subject to infrequent trading. Journal of financial economics , 7 (2), 197-226. Fama, E. F., & French, K. R. (2006). Profitability, investment and average returns. Journal of financial economics , 82 (3), 491-518. Fama, E. F., & French, K. R. (2015). A five-factor asset pricing model. Journal of financial economics , 116 (1), 1-22. Feng, J., Goodell, J. W., & Shen, D. (2022). ESG rating and stock price crash risk: Evidence from China. Finance Research Letters , 46 , 102476. Habib, A., Hasan, M. M., & Jiang, H. (2018). Stock price crash risk: review of the empirical literature. Accounting & Finance , 58 , 211-251. Harvey, C. R., & Siddique, A. (2000). Conditional skewness in asset pricing tests. The Journal of finance , 55 (3), 1263-1295. Hong, H., & Stein, J. C. (1999). Differences of opinion, rational arbitrage and market crashes . National bureau of economic research Cambridge, Mass., USA. Hunjra, A. I., Mehmood, R., & Tayachi, T. (2020). How do corporate social responsibility and corporate governance affect stock price crash risk? Journal of Risk and Financial Management , 13 (2), 30. Hutton, A. P., Marcus, A. J., & Tehranian, H. (2009). Opaque financial reports, R2, and crash risk. Journal of financial Economics , 94 (1), 67-86. In, S. Y., & Schumacher, K. (2021). Carbonwashing: A New Type of Carbon Data-Related ESG Greenwashing. Available at SSRN . Jin, L., & Myers, S. C. (2006). R2 around the world: New theory and new tests. Journal of Financial Economics , 79 (2), 257-257-292. Kabir, M. N., Rahman, S., Rahman, M. A., & Anwar, M. (2021). Carbon emissions and default risk: International evidence from firm-level data. Economic Modelling , 103 , 105617. 37 WC03351 Total Liabilities All shortand long-term obligations expected to be satisfied by the company. Time Series WC04101 Deferred Income Taxes & Investment Tax Credit The increase or decrease in the deferred tax liability from one year to the next resulting from timing differences in recognition of revenues and expenses for tax and financial reporting purposes. Time Series WC03451 Preferred Stock A claim prior to the common shareholders on the earnings of a company and on the assets in the event of liquidation. Time Series WC03251 Long-Term Debt All interest-bearing financial obligations, excluding amounts due within one year. It is shown net of premium or discount. Time Series WC08326 Return On Assets (Net Income – Bottom Line + ((Interest Expense on Debt-Interest Capitalized) * (1-Tax Rate))) / Average of Last Year's and Current Year’s Total Assets Time Series 38 D. Industry Classification of Sample Firm This Appendix presents a breakdown of the 1,279 US-listed stocks by industry, classified according to the Fama-French 48 industry categories described in section 4. Accordingly, financials (SIC codes 6000-6999) and utilities (SIC codes 4900-4949) are excluded from this classification. Industry Classification Number of Firms % of the Total 34 Business Services 204 15,95% 36 Electronic Equipment 89 6,96% 42 Retail 81 6,33% 30 Petroleum and Natural Gas 78 6,10% 40 Transportation 60 4,69% 21 Machinery 60 4,69% 14 Chemicals 57 4,46% 13 Pharmaceutical Products 48 3,75% 41 Wholesale 44 3,44% 23 Automobiles and Trucks 42 3,28% 43 Restaurants, Hotels, Motels 38 2,97% 12 Medical Equipment 37 2,89% 35 Computers 34 2,66% 32 Communication 34 2,66% 37 Measuring and Control Equipment 32 2,50% 2 Food Products 31 2,42% 17 Construction Materials 27 2,11% 9 Consumer Goods 25 1,95% 19 Steel Works Etc 25 1,95% 18 Construction 23 1,80% 38 Business Supplies 20 1,56% 22 Electrical Equipment 19 1,49% 10 Apparel 15 1,17% 28 Non-Metallic and Industrial Metal Mining 14 1,09% 33 Personal Services 14 1,09% 4 Beer & Liquor 13 1,02% 24 Aircraft 13 1,02% 15 Rubber and Plastic Products 12 0,94% 6 Recreation 10 0,78% 8 Printing and Publishing 10 0,78% 39 39 Shipping Containers 10 0,78% 11 Healthcare 9 0,70% 7 Entertainment 7 0,55% 48 Other 7 0,55% 3 Candy & Soda 6 0,47% 29 Coal 6 0,47% 1 Agriculture 4 0,31% 5 Tobacco Products 4 0,31% 20 Fabricated Products 4 0,31% 27 Precious Metals 4 0,31% 16 Textiles 4 0,31% 25 Shipbuilding, Railroad Equipment 3 0,23% 26 Defense 2 0,16% Total 1279 100,00%