CLIMATE CHANGE, AIR POLLUTION, AND CORPORATE PERFORMANCE: EVIDENCE FROM WILDFIRE SMOKE By Jing Kong A DISSERTATION Submitted to Michigan State University in partial fulfillment of the requirements for the degree of Business Administration — Accounting — Doctor of Philosophy 2022 ABSTRACT CLIMATE CHANGE, AIR POLLUTION, AND CORPORATE PERFORMANCE: EVIDENCE FROM WILDFIRE SMOKE By Jing Kong This paper examines the effect of climate change on corporate performance. Using exposure to wildfire smoke as a novel setting, I document a significant physical effect of climate change on corporate operating income. On average, a one-standard-deviation increase of wildfire smoke exposure is associated with an $18.7 million loss in operating income. The effect is strongest for firms with high R&D-to-employee ratios, with high average labor costs, or that operate in industries that are highly dependent on skilled labor, suggesting that decreased productivity among high-skilled employees drives the observed effect. I also find that the stock market does not fully incorporate the effect of wildfire smoke into stock prices until annual earnings announcements. More climate change disclosure moderates the negative market reaction around annual earnings announcements. Overall, this paper documents the less salient effects of climate change on corporate performance and asset prices. To Harlow and Sugar, for being the light of my life. And to my parents, for allowing me to not work until 32. iii ACKNOWLEDGMENTS I want to thank my dissertation chair John Jiang, who is the biggest force behind my admission to the Ph.D. program at Michigan State University five years ago. I would also like to thank Isabel Wang for her support and guidance throughout the Ph.D. program. I have benefited tremendously from their expertise, experience, and insights both personally and professionally. I am forever grateful for the time they have given so generously to help me on this journey. As always, I am grateful for my mother’s unconditional love. She tried her best to support me for all my decisions. She supplies me with what she believes is the best. Her emotional support in the past five years motivate me to finish this journey. Most importantly, I am thankful to my wife, Harlow, for her generous love and support throughout this journey. She believes in me when I doubt about myself, gives me advice when I needed, and takes care of my life every day. She is my greatest blessing in the doctoral program. iv TABLE OF CONTENTS LIST OF TABLES vi LIST OF FIGURES vii CHAPTER 1: INTRODUCTION 1 CHAPTER 2: LITERATURE REVIEW 9 2.1. Literature on Climate Change 9 2.2. Wildfires and Air Pollution 10 CHAPTER 3: DATA AND SAMPLE 12 3.1. Wildfire Smoke Data 12 3.2. Wildfire and Weather Data 12 3.3. Firm-Specific Data 13 CHAPTER 4: RESEARCH DESIGN 16 CHAPTER 5: RESULTS 18 5.1. Descriptive Statistics 18 5.2. Baseline Result 26 5.3. Wildfire Smoke and Cross-sectional Tests 29 5.3.1. Firm-level Differences 29 5.3.2. Industry-level Differences 32 5.3.3. County-level Differences 34 5.4. Effect of Wildfire Smoke on Sales and Productivity 36 5.5. Wildfire Smoke and Earnings Announcements Return 38 5.6. Disclosure of Climate Change Exposure 41 5.7. Adaptation Behaviors 44 5.8. Wildfire Smoke and Supply-chain Disruptions 47 CHAPTER 6: ROBUSTNESS TESTS 50 6.1. Firm with Dispersed Employees 51 6.2. Portfolio Returns 53 6.3. Unobservable Wildfires 55 6.4. Smoke Intensity 57 CHAPTER 7: CONCLUDING REMARKS 59 APPENDICES 60 APPENDIX A Variable Definitions 61 REFERENCES 65 v LIST OF TABLES Table 1 Sample Selection 15 Table 2 Summary Statistics 19 Table 3 Firm Performance and Effects of Wildfire Smoke 28 Table 4 Human Capital and Effects of Wildfire Smoke 31 Table 5 Labor Skill and Effects of Wildfire Smoke 33 Table 6 County Characteristics and Effects of Wildfire Smoke 35 Table 7 Impact of Wildfire Smoke on Sales and Expenses 37 Table 8 Wildfire Smoke and Earnings Announcement Return 40 Table 9 Voluntary Disclosure and Effects of Wildfire Smoke 43 Table 10 Plant-level Adaptation to Smoke Exposure 46 Table 11 Wildfire Smoke and Supply Chain 49 Table 12 Effect of Dispersed Operation 52 Table 13 Portfolio Returns 54 Table 14 Exclude States with Significant Wildfires 56 Table 15 Smoke Intensity: An Instrumental Variable Approach 58 Table A.1 Variable Definitions 62 vi LIST OF FIGURES Figure 1 County Annual Wildfire Smoke Exposure 21 Figure 2 Fire and Smoke on April 10th, 2018 23 Figure 3 Wildfire smoke and Air Quality Index (AQI) 24 Figure 4 Firm Locations 25 vii CHAPTER 1: INTRODUCTION This paper examines the physical effects of climate change on the performance of US companies. Climate science establishes a strong relationship between rising temperatures and recent increases in other extreme weather events, including wildfires, hurricanes, and floods (Abatzoglou and Williams, 2016; United Nations, 2020). The US Environmental Protection Agency (EPA) estimates that natural disasters caused by climate change will cost the US alone at least $224 billion per year by 2090 (Martinich and Crimmins, 2019). The far-reaching effects of climate change on the global economy and ecology make climate change one of the biggest threats to humanity today (United Nations, 2021). Investors and managers face three types of risk from climate change: the risk of physical damage, regulatory risk (e.g., the Cap-and-Trade Program and other greenhouse gas regulation), and technological risk (e.g., changes in demand caused by disruptive green technology, such as the electric vehicle). Institutional investors are the least concerned with the first category of risk-- the physical effects of climate change (Krueger, Sautner, and Starks, 2020). This perspective is supported by extant research, which finds only limited evidence that climate change affects corporate performance in the US. For instance, Addoum, Ng, and Ortiz-Bobea (2020) find no evidence that extreme temperature shocks affect US establishment-level sales or employee productivity, suggesting that firms in developed countries may be resilient to climate change. Huynh and Xia (2021) find that while stock returns decline significantly for firms exposed to natural disasters, the returns bounce back in the following month, suggesting natural disasters have no significant impairment on long-term corporate performance. These studies focus on the immediate (direct) effects of climate change on firm operations and ignore the potential diffuse (indirect) effects. While the immediate effects of climate change are highly salient, they are 1 relatively manageable because they are localized and predictable. For example, Entergy Corporation, a utility company based in New Orleans, relocated important business centers to Little Rock, Arkansas, and its transmission center to Jackson, Mississippi after Hurricane Katrina revised its assessments of climate risk (Sussman and Freed, 2008). Firms reliant on agricultural products, such as Anheuser-Busch InBev, have begun investing in research aimed at improving drought tolerance in key crops (CDP, 2020). Firms are less able to predict and manage their exposure to the diffuse effects of distant climate disasters. Smoke from California wildfires travels as far as the East Coast, triggering air quality alerts in New York State (Gard and Garrett, 2021). Floods pollute drinking water for weeks, affecting public health in non-flooded regions long after the waters have receded (Hrdinka et al., 2012; Miller and Hutchins, 2017). These diffuse effects of climate change receive less attention despite their broad geographic and temporal reach and significant economic implications. Firms, investors, and research that focus only on the immediate effects of climate change may underestimate the true physical risk of climate change. The resulting underinvestment in climate change mitigation and suboptimal demand for climate change disclosure threaten the capital and goods market.1 This paper utilizes wildfire smoke to document the diffuse physical effects of climate change on corporate performance and market prices. Climate change has increased the incidence and doubled area lost to wildfires since the 1980s (Abatzoglou and Williams, 2016). While the most visible harms from wildfires and other climate disasters are immediate and localized, the diffuse effects of wildfires are far-reaching and, therefore, hard to predict. Smoke significantly increases 1 In a recent survey by Deloitte, a-third of the companies are not taking any action to manage, mitigate or adapt to climate change because there is no significant pressure from shareholders or customers (Coppola, Krich and Blohmke, 2019). 2 the concentration of particulate matter many miles away from the originating wildfire, even when smoke is invisible to the human eye (Borgschult, Molitor, and Zou, 2020). While the West Coast is more prone to wildfires, wildfire smoke causes more deaths in the more heavily populated East Coast (O’Dell et al, 2021). On average, public firms in the US experience less than one day of significant wildfire risk (Griffin et al., 2020) but roughly 20 days of wildfire smoke exposure each year. Wildfire smoke, although less acute than the fire itself, affects many more firms. Smoke exposure may disrupt firm-level operations by impairing employee productivity. Fine particulate matter (PM2.5), one of six major air pollutants in wildfire smoke is a known health hazard. With a diameter of less than 2.5 micrometers, outdoor PM2.5 can easily enter indoor spaces, irritating the nose, throat, and lungs; increasing the incidence of headaches and leading to heart disease (Pope, 2000; Sorenson et al., 2003). Extant research documents that high levels of PM2.5 also reduce employee productivity and cognitive performance (Chang et al., 2016; Chen et al., 2017; La Nauze and Severnini, 2021). However, whether these effects are observable at the firm level is unclear. In thick labor markets, employees especially susceptible to wildfire smoke may be temporally replaced. Tasks that do not require specialized skills or equipment may be quickly shifted to production centers with better air quality. However, firms that invest heavily in human capital have fewer effective smoke mitigation strategies available. Investment in human capital is concentrated in knowledgeable and high-skilled employees, who are both critical to operations and difficult to replace (Barney, 1991; Peteraf, 1993). For instance, an unproductive research and development (R&D) team member may delay the launch of a new drug while an unproductive production-line worker may not observably affect factory output. In the case of persistent smoke coverage or long- lasting health effects, chronically unproductive workers may be more easily replaced on the 3 production line than on the R&D team (Farber and Hallock 2009; Blatter, Muehlemann, and Schenker 2012). Wildfire smoke exposure is, therefore, likely to influence firm performance most when it affects high-skilled and difficult-to-replace employees. My analysis relies on the National Oceanic and Atmospheric Administration’s Hazard Mapping System (HMS). The HMS uses high-resolution satellite images to track daily wildfire smoke coverage in the contiguous United States. Because wildfire smoke plumes can travel thousands of miles, this data allows me to disentangle the effect of smoke exposure from the immediate economic losses from wildfire damage. I aggregate daily smoke exposure data at the annual level to estimate the medium and long-run effects of wildfire smoke exposure in the county a firm headquarters. On average, smoke exposure significantly reduces operating performance. A one-day increase in annual smoke exposure decreases return on asset (ROA) by 2.0 basis points, which corresponds to an average of $18.7 million loss in operating income for a 16-day (one standard deviation) increase in annual smoke exposure.2 This estimation assumes that operations are concentrated near a firm’s headquarters. Using headquarters smoke coverage may lead to measurement errors for firms with geographically dispersed operations. When excluding geographically dispersed industries, I find a stronger relationship between headquarters smoke exposure and firm performance.3 I conduct several cross-sectional tests to identify the mechanisms through which wildfire smoke exposure impairs firm performance. While exposure to wildfire smoke lower employee 2 This magnitude is calculated based on the average book value of assets in a sample of US public firms. Effect on income = -0.0002*16 days*$5,842 million in assets = -$18.7 million. This magnitude is much larger than previous studies. For example, Pankratz, Bauer and Derwall (2021) find that a one standard deviation increase in the number of hot days leads to $1.8 million losses in non-US firms. 3 In the robustness test, I rely on LexisNexis Corporate Affiliation Database to identify firm-level dispersed operations. Compared with including dispersed employees, I find a larger magnitude of smoke exposure after excluding firms with dispersed employees. 4 productivity uniformly, high-skilled employees are of greater importance to firm performance. Hence, the effect should be stronger when firm performance depends more on highly paid and skilled employees. Consistent with this hypothesis, cross-sectional tests show that wildfire smoke has a stronger effect on ROA among firms with higher R&D-to-employee ratios and higher average labor costs. In the third cross-sectional test, I examine the effect of wildfire smoke on firms with different employee replacement costs. I follow Ghaly, Dang, and Stathopoulos (2017) in constructing an industrial-level labor skills index. Industries that rely on highly skilled employees incur high costs to replace current employees because labor supply is limited (Farber and Hallock 2009; Blatter et al., 2012). I find that the effect of wildfire smoke on firm performance is stronger for firms in industries that rely on highly skilled employees. I then examine whether capital market participants incorporate the effect of wildfire smoke exposure into security prices. Market participants may either directly price in wildfire smoke exposure or the resulting lower employee productivity. I find that increases in smoke exposure are associated with more negative earnings surprises, indicating financial analysts cannot accurately model the effect of wildfire smoke exposure on firm performance. Moreover, high smoke exposure is associated with a negative market reaction to annual earnings announcements.4 Together, these findings suggest that the capital market does not yet fully incorporate wildfire smoke exposure into security prices. The significant effects of wildfire smoke exposure on operating income and market reactions around earnings announcements prompt calls for additional climate change disclosure. However, it is unclear to what extent climate change disclosure will improve market efficiency. To shed light 4 In a supplement test, I assign firms into four quartiles based on smoke exposure as of the year t-1. A trading strategy of buying and holding firms in the lowest smoke exposure quartile and selling firms in the highest smoke exposure quartile in year t can earn up 6.6% abnormal annual returns. The results are presented in Table A.1. 5 on this question, I explore whether voluntary disclosure of climate change exposure mitigates the effect of wildfire smoke on firm performance and the capital market. The benefits of disclosure are twofold: first, an improved information environment (Dhaliwal et al., 2011), and incentives for managers to investigate and proactively mitigate their climate change exposure (Downar et al., 2021). I use the proportion of climate-related keywords in the earnings conference call as a proxy for climate change disclosure (Sautner et al., 2020). I find that climate-change disclosure does not change the association between smoke exposure and ROA. However, high climate change disclosure mediates the negative association between smoke exposures and market reactions around earnings announcements. This result suggests firms with high climate change disclosures provide a better information environment to investors, which alleviates the underpricing of climate change risk. This paper contributes to the accounting and finance literature on climate change by documenting that the significant effect of physical climate change on US firms is significant. Prior evidence that climate change affects corporate performance is concentrated in developing countries (Hsiang and Jina, 2014) but is limited for US firms (Addoum et al., 2020; Pankratz et al., 2021). This paper finds that the diffuse effects of climate change can significantly impair firm performance in the US. A handful of studies find that sustained climate change reduces consumer demand and disrupts supply chains (Pankratz and Shiller, 2020; Jin et al., 2021), but a direct relationship between reductions in labor productivity following climate events and firm performance is limited. Pankratz et al. (2021) find that firms compensate for temporary reductions in labor productivity following climate disasters by increasing labor hours in subsequent quarters. My paper suggests that firms 6 do not fully recover from the effect of wildfire smoke exposure on labor productivity, at least not in the short term. This paper also contributes to the literature on the economic and health consequences of air pollution. This broad literature has documented evidence of air pollution’s deleterious effects on health and labor productivity (e.g., Currie and Neidell 2005; Graff-Zivin and Neidell, 2012; Deryugina et al., 2019). Although this research establishes that air pollution impairs employee productivity, it does not necessarily lead to lower firm performance. A key challenge for measuring air pollution’s causal effect on firm performance is in part driven by economic activities. Observed associations between air pollution and firm performance may be positive because pollutants are byproducts of operations. Using wildfire smoke as an exogenous source of air pollution, I document the effect of air pollution on firm performance. Recent wildfires have captured the attention of researchers from multiple disciplines. Current studies document significant economic losses from fire damage and the health effects of wildfire smoke (Wettstein et al., 2018; Wang et al., 2021). Griffin et al. (2021) find that a firm is more likely to mention wildfire-related information when its headquarters county is at risk of exposure to wildfires. While some firms flag wildfires as a threat to firm operations, few reference wildfire smoke, which causes more persistent, unpredictable, and widespread disruptions.5 I extend this literature by considering wildfire smoke, an overlooked, real effect of wildfires, on firm performance. Finally, this paper examines the role of climate change disclosure in reducing investors’ exposure to the effects of climate change. The capital market’s failure to fully price the effects of climate change has policy implications regarding disclosure requirements. Although existing 5 For example, Marriot mentions “wildfires” as a potential operational risk in the 2020 10-K. 7 literature finds that investors price in climate regulatory risk (Seltzer et al., 2020; Bolton and Kacperczyk, 2021; Ilhan, Sautner, and Vilkov, 2021), my findings suggest that investors most likely underestimate the physical threats of climate change. When climate disclosure is poor, the average $18.7 million loss in operating income from one standard deviation increase in smoke exposure is not reflected in capital market pricing before earnings announcements. I find that climate change disclosure improves the information environment and the capital market’s ability to assess climate change risk. 8 CHAPTER 2: LITERATURE REVIEW 2.1. Literature on Climate Change The extant economics literature on climate change focuses on the effect of extreme temperatures. Studies find that extreme temperatures have negative effects on labor supply and economic output. In particular, abnormal temperature shocks decrease gross domestic product (GDP) growth and exports in developing counties (Jones and Olken, 2010; Dell, Jones, and Olken, 2012). Abnormal temperatures also affect the US economy in specific industries. For example, Fisher et al. (2012) find that climate change reduces US agricultural yields. Graff-Zivin and Neidell (2014) examine the effects of temperatures on workers’ time allocation in the US. They find that in industries with high exposure to weather conditions, employees reduce working hours when temperatures increase. However, whether these changes in employee-level productivity have observable aggregate effects on firm performance is unclear. Using US establishment-level data, Addoum et al. (2020) find no evidence that temperature shocks affect sales or profit. Their study suggests a tenuous relationship between climate change and firm performance in developed counties. Huynh and Xia (2021) find that although the capital market has a significant negative reaction to climate disasters, prices bounce back in the following month. Their findings suggest that investors overreact to disasters. Using a sample of non-US firms, Pankratz et al. (2021) find extreme temperatures reduce firms’ operating income because firms increase labor inputs to compensate for decreased productivity. The literature does not explore whether these results translate to the most developed US economy. 9 2.2. Wildfires and Air Pollution Climate change is observable in changing weather patterns and the increased incidence of extreme weather events (National Park Service, 2018; National Interagency Coordination Center, 2020). The resulting high temperatures and low precipitation have contributed to the recent increase in wildfires. These changes have led to a 55% increase in the aridity of high-risk fire zones since the 1970s and doubled the annual acreage lost to wildfires since 1984 (Abatzoglou and Williams, 2016). While wildfire damage is localized, wildfire smoke plumes travel hundreds of miles downwind. Wildfire smoke contains a mixture of pollutants, including sulfur dioxide (SO2), hazardous air pollutants (HAPs), and particle pollution. Fine particulate matter (PM2.5) is the main component of wildfire smoke and is a known health hazard (EPA, 2021). Borgschulte et al. (2020) find a significant association between smoke exposure and same-day air pollution levels. Extensive medical literature finds a positive association between wildfire smoke and cardiovascular and respiratory illnesses (e.g., Wettstein et al., 2018). Environmental economics shows that PM2.5 and other air pollutants negatively affect labor outcomes. Hanna and Oliva (2015) find that an increase in air pollution, measured by levels of SO2, leads to decreased working hours. Using a sample of wealthy households, Aragon, Miranda, and Oliva (2017) find that the negative association between air pollution and working hours is due to increased demand for childcare and eldercare. Chang et al. (2016) and Graff-Zivin and Neidell (2012) find that high levels of PM2.5 have a sizable effect on short-term labor productivity for agricultural workers. PM2.5 can enter indoor spaces through closed windows and doors, affecting office worker productivity. Chang et al. (2019) find that call center workers have lower daily output when daily PM2.5 levels are high. High levels of PM2.5 impair both physical and cognitive performance, 10 evidenced by test scores (Lavy, Ebenstein, and Roth, 2014) and reduced audit quality (Chen et al., 2017). La Nauze and Severnini (2021) further identify that the adverse effects of PM2.5 on cognition are greatest for employees under fifty and workers engaged in problem-solving tasks. While air pollution has clear adverse effects on individuals, its effect on firm performance is less straightforward because pollutants are byproducts of normal operations. Firms that pursue profit maximization tend to pollute more, and their abatement efforts largely depend on regulatory costs (Fowlie, 2010). As a result, observed associations between air pollution and firm performance may be positive. Using satellite data, Zou (2021) finds that cities with intermittent air monitoring experience significantly higher air pollution levels on unmonitored days, suggesting firms tend to increase pollution levels when the regulatory cost is low. Wildfire smoke introduces an exogenous source of air pollution that is not associated with economic activities or driven by local regulation, allowing me to examine the causal relationship between air pollution and firm performance. 11 CHAPTER 3: DATA AND SAMPLE 3.1. Wildfire Smoke Data I obtain wildfire smoke data from National Oceanic and Atmospheric Administration’s Hazard Mapping System (HMS). The HMS uses high-resolution satellite imagery from Geostationary Operational Environmental Satellite. Trained analysts process the satellite data to estimate smoke plumes over the contiguous United States each day. Although HMS data do not contain smoke height and intensity, the data’s measure of smoke plume distribution is commonly used to capture daily smoke coverage (Aguilera et al., 2021).6 Following Miller et al. (2017) and Borgschulte et al. (2020), I use the HMS smoke plume data from 2006 to 2019 to construct daily smoke exposure at the county level.7 I compare the HMS spatial data with county spatial data from the Census to determine if smoke plumes fully cover a county on a given day. I sum up smoke exposure days on an annual basis to construct a county- level smoke exposure measure. 3.2. Wildfire and Weather Data To separate smoke exposure from actual wildfire risk, I collect wildfire records from the National Fire and Aviation Management group of the US Forest Service. This dataset consolidates wildfire records from the reporting systems of federal, state, and local fire organizations between 1992 and 2015 (Short, 2017). For the remaining years during the sample period, I collect the wildfire records from Monitoring Trends in Burn Severity (MTBS) and Wildfire Fire Decision Support System (WFDSS). I further control for weather characteristics that could correlate with both wildfire smoke and firm performance. Temperature and precipitation data come from the PRISM Climate Group, the 6 HMS analysts use fire detection algorithms to differentiate wildfire smoke from clouds or normal air pollution. 7 The HMS data is available from September 2005. 12 US Department of Agriculture’s official climatological database. This database provides daily temperatures and precipitation data in the contiguous US across 4 by 4 km grids. I use the weather observations closest to each county’s centroid to construct weather conditions at the county level. I obtain wind speed and wind direction data from the North American Regional Reanalysis (NARR) daily reanalysis data. This database provides daily wind conditions from land-based weather stations. Daily wind conditions consist of vector pairs, one for the east-west wind direction (u-component) and one for the north-south wind direction (v-component). Following prior literature (e.g., Deryugina et al., 2019), I calculate wind speed as √u2 + v 2 , where u and v are the county-day-level vectors. The wind direction is calculated as follows: 270 − 𝜃 𝑖𝑓 𝑢 > 0 𝑎𝑛𝑑 𝑣 > 0 270 + 𝜃 𝑖𝑓 𝑢 > 0 𝑎𝑛𝑑 𝑣 < 0 𝑊𝐼𝑁𝐷𝐷𝐼𝑅 = { 90 + 𝜃 𝑖𝑓 𝑢 < 0 𝑎𝑛𝑑 𝑣 > 0 90 − 𝜃 𝑖𝑓 𝑢 < 0 𝑎𝑛𝑑 𝑣 < 0 180 |𝑣| Where 𝜃 = 𝐴𝑟𝑐𝑡𝑎𝑛(|𝑢|). 𝑊𝐼𝑁𝐷𝐷𝐼𝑅 indicates the wind direction, with zero corresponding to 𝜋 wind blowing from the north and higher values corresponding to clockwise movement around the compass. I average the estimated monitor-day level wind direction and speed to at the county-year level. 3.3. Firm-Specific Data Firm data comes from the 2006 - 2019 period in the CRSP-Compustat merged database. I drop firms not traded on the NYSE, AMEX, or NASDAQ and firms that operate in the highly regulated financial and utility industries. Next, I use the headquarters zip code to identify firms’ headquarters county location. Table 1 presents detailed sample selection procedures. The resulting sample contains 3,978 firms and 32,193 firm-year observations. Stock return data are obtained from the 13 CRSP database; earnings announcement dates are from the IBES database; financial statement data are from the Compustat database. I obtain firm-level voluntary climate change exposure disclosure data (CC_expo) from Sautner et al. (2020), who use a machine-learning algorithm to measure climate change disclosure from earnings conference calls. The algorithm first identifies climate change bigrams by comparing training libraries of climate-related texts and non-climate-related texts. Then, it counts the frequency of climate change-related bigrams in each conference call transcript, scaled by the total number of bigrams in that transcript.8 The authors find a significant association between CC_expo and climate-related events, such as the Paris Agreement and the Doha Climate Change Conference. I use the annual-level CC_expo in my analysis to match with the measure of wildfire smoke. In the main analysis, I assume most of the firm performance changes are due to smoke exposure in the county where a firm is headquartered.9 Measurement errors arise if a firm has establishments in different counties. In robustness tests, I obtain affiliate locations from the LexisNexis Corporate Affiliations Database. I identify all affiliation locations, including plants, units, subsidiaries, and branches, located in different counties from headquarters. This database also provides the number of employees that work in each affiliate. I use the number of affiliation employees and the total number of corporate employees to estimate the magnitude of potential measurement errors. 8 Top ten keywords by frequency are “renewable energy,” “electric vehicle,” “clean energy,” “new energy,” “wind power,” “wind energy,” “energy efficient,” “climate change,” “greenhouse gas,” “solar energy.” 9 Henderson and Ono (2008) show that a firm’s major plants and operations are typically located close to a firm’s headquarters. Pirinsky and Wang (2006) note that corporate headquarters are usually close to core business activities. 14 Table 1 Sample Selection This table presents the sample selection procedures in the main analysis sample. The sample period is from 2006 to 2019. Observations Firms Firm-year observations from 2006 through 2019 in Compustat 71,036 9,423 Less: firms not listed on Amex, NYSE, and NASDAQ (8,517) Less: observations without location information (15,535) Less: financial and utility firms (13,793) Less: observations with less than 1 million assets (101) Less: observations with insufficient data for control (897) variables Main Sample 32,193 3,978 15 CHAPTER 4: RESEARCH DESIGN Because wildfire smoke plumes travel thousands of miles, I can examine the effect of smoke exposure independent of the direct economic losses caused by wildfires using the following equation: 𝑃𝑒𝑟𝑓𝑜𝑟𝑚𝑎𝑛𝑐𝑒𝑖𝑡 = 𝛽 ∙ 𝑆𝑚𝑜𝑘𝑒𝐷𝑎𝑦𝑠𝑐𝑡 + 𝐶𝑜𝑛𝑡𝑟𝑜𝑙 𝑉𝑎𝑟𝑖𝑎𝑏𝑙𝑒𝑠 + 𝛼𝑖 + 𝛼𝑡 /𝛼𝑠𝑡 + 𝜀𝑖𝑡 (1) The primary measure of firm performance is ROA, which equals operating income standardized by total assets in year t-1. Compared with net income, operating income is subject to less distortion from accounting choices (Moliterno and Wiersema, 2007). 10 The primary measure of smoke exposure is 𝑆𝑚𝑜𝑘𝑒𝐷𝑎𝑦𝑠𝑐𝑡 , which equals the number of days a county is fully covered by wildfire smoke each year 𝑡. I use headquarters location to determine a firm’s smoke exposure. In the robustness test, I replace 𝑆𝑚𝑜𝑘𝑒𝐷𝑎𝑦𝑠𝑐𝑡 with 𝑆𝑚𝑜𝑘𝑒𝐼𝑛𝑡𝑒𝑛𝑠𝑖𝑡𝑦𝑐𝑡 , which equals the difference between air quality index (AQI) during smoke days and non-smoke days. This estimating equation includes firm fixed effects (𝛼𝑖 ) to isolate within-firm differences in the effect of smoke exposure. Including firm fixed effects also partially alleviate the concern that 𝑆𝑚𝑜𝑘𝑒𝐷𝑎𝑦𝑠𝑐𝑡 does not capture the intensity of smoke coverage. Year fixed effects (𝛼𝑡 ) or industry by year fixed effects ( 𝛼𝑠𝑡 ) account for correlations between smoke exposure and economic trends. I control for wildfire events-the number of days that a county experienced wildfires-to separate the effect of smoke exposure from wildfire damages. I also control for weather patterns that may directly affect financial performance (Addoum et al., 2020) and correlate with smoke exposure (Miller et al., 2017; Borgschult et al., 2020). The time-varying weather characteristics include annual counts of the number of days above 30oC and below 0oC, the annual precipitation level, and annual average wind conditions closest to each county’s centroid. 10 In the robustness test, I also examine the effect of smoke exposure on net profit standardized by total assets. 16 Following the literature on air pollution (e.g., Deryugina et al, 2019; Borgschult et al., 2020), I weight regressions by county population in each year so that my estimates reflect the experience of a representative employee. 17 CHAPTER 5: RESULTS 5.1. Descriptive Statistics Table 2 provides summary statistics of the main analysis sample. A firm’s exposure to wildfire smoke is not uncommon. On average, firms are fully covered by smoke plumes for days (14 weekdays) a year during the 2006-2019 sample period. Smoke exposure is closely related to wildfires. Around half of the observations located in counties experienced at least one wildfire throughout the year. Counties on average experience 30 days with some wildfire activities each year. However, the incidence of meaningful wildfire exposure is much lower-around one day per year (Griffin et al., 2021). An average firm experiences 8 days of average temperature above 30oC and 22 days of average temperature below 0oC each year. The average ROA is 8%, and the average 3-day cumulative abnormal return around earnings announcements is 0.38%. 18 Table 2 Summary Statistics This table presents the summary statistics in the analysis sample. The sample period is 2006- 2019. All firm-specific variables are winsorized at the 1st and 99th percentiles. Variable Name N Mean St Dev P25 P50 P75 Dependent Variables Operating Income/Asset (%) 32,193 7.993 20.504 5.022 11.575 17.751 Operating Expense/Asset (%) 32,193 102.049 72.667 50.444 83.350 133.008 Sale/Asset (%) 32,193 109.318 76.795 55.180 91.441 143.835 CAR [-1, 1] (%) 22,652 3.822 8.038 2.281 4.611 6.649 Earning Surp (%) 24,008 0.351 8.598 -4.148 0.300 5.009 Net Income/Asset (%) 32,367 -0.964 20.423 -2.837 3.907 8.588 Weather/Economic Characteristics SmokeDays 32,193 20 16 8 16 28 WildFire 32,193 30 80 0 1 21 GDP Percap 32,193 10.944 0.341 10.704 10.901 11.106 County Population 32,193 1,602,642 1,803,899 597,597 991,847 1,765,137 Wind Direction 32,193 202.014 27.678 191.379 208.506 218.489 Wind Speed 32,193 3.351 0.867 2.747 3.216 3.5443 Temp above 30 32,193 7.709 14.704 0.000 4.000 9.000 Temp below 0 32,193 20.326 31.791 0.000 0.000 39.000 Precip (mm) 32,193 10.944 0.341 10.70385 10.901 11.106 Firm Characteristics Asset 32,193 5,842.484 2,7753.89 171.249 710.1 2,784.349 R&D 32,193 0.061 0.114 0.000 0.006 0.073 Leverage 32,193 0.198 0.208 0.002 0.151 0.314 Sales Growth 32,193 0.128 0.402 -0.024 0.065 0.180 M/B 32,193 2.364 2.055 1.188 1.703 2.685 PPE/Asset 32,193 0.257 0.257 0.072 0.162 0.355 Tobin’s Q 32,193 2.078 1.463 1.184 1.597 2.388 R&D/Emp 17,361 0.075 0.143 0.006 0.022 0.064 Labor Cost 31,575 0.183 2.063 0.039 0.087 0.176 Labor Skills 28,205 2.618 0.529 2.249 2.583 2.998 19 Figure 1 provides summary statistics for the annual frequency and geographic distribution of smoke events in our sample. The Midwest states tend to have high smoke exposure, and the southwest states experience low smoke exposures. Smoke exposure also changes each year significantly, with a standard deviation of 16 days.11 11 Perhaps counter-intuitively, the smoke coverage in California is lower than the country average. This is due to wind conditions and is consistent with the anecdotal evidence (Saldanha, 2021). California has significantly more days of smoke exposure in 2018, when it experienced the worse fires in history. 20 Figure 1 County Annual Wildfire Smoke Exposure This figure plots the number of days a county is fully covered by wildfire smoke in selected years. The sample period is 2006-2019. Darker colors suggest more smoke days. 2006: 12.6 Days 2009: 11.5 Days 2012: 47.1 Days 2015: 26.0 Days 2018: 33.7 Days 2019: 32.2 Days 21 Figure 2 illustrates the wildfires and smoke in the contiguous US on April 10th, 2018. This plot shows that fire-related smoke can travel across the country. This feature allows us to separate the effect of wildfire smoke from direct fire damages. In figure 3, we present the coefficient of regressing air quality index (AQI) on indicators of daily smoke exposure [-10, 10] days around the day of smoke exposure. To alleviate the concern that smoke is correlated with same-county wildfires, we exclude daily observations if a county also experienced a wildfire. We find a significant arise of AQI around the smoke coverage date. Figure 4 plots the counties that are covered in the analysis sample. The sample contains 421 unique counties. Although the sample only represents a small fraction of counties, those counties are home to 68.7% of the US population, according to the 2010 Census. 22 Figure 2 Fire and Smoke on April 10th, 2018 This map plots the fire and smoke in selected US states on April 10th, 2018. The red triangles represent fire locations, and grey areas are smoke coverage. 23 Figure 3 Wildfire smoke and Air Quality Index (AQI) This figure plots the coefficients of regressing AQI on indicators of daily smoke exposure [-10, 10] days around the day of smoke exposure. The baseline AQI is 10 days before each smoke coverage date. We exclude daily observations if a county also experienced a wildfire. The regression includes county by year-month fixed effects. Standard errors are clustered at the state levels. 12 10 8 AQI 6 4 2 0 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 Days Relative to Smoke Exposure 24 Figure 4 Firm Locations This figure highlights the counties where sample firms’ headquarters locate. The sample contains 421 counties, which cover 68.7% of the US population according to the 2010 Census. 25 5.2. Baseline Result Table 3 presents the baseline effect of wildfire smoke on ROA. In column 1, I control for wildfire and weather characteristics. This column does not include further county-level GDP or firm-level control variables to avoid “over-controlling problems” in climate-economy literature (Dell, Johns, and Olken, 2014). 12 Column 2 adds controls for county-level GDP and firm characteristics. I find that more days of smoke exposure significantly reduce a firm’s financial performance at the 95% confidence interval in both columns. The economic magnitude is non- trivial. According to the coefficient estimate in column 2, a one-day increase in annual smoke exposure decreases ROA by 2.0 basis points. A one standard deviation (16 days) increase in annual smoke exposure decreases ROA by 32 basis points, corresponding to an $18.7 million loss in operating income for an average firm (based on the book value of assets). Perhaps unsurprisingly, more days of wildfires are associated with lower ROA. In column 3, I replace year fixed effects with 3-digit SIC industry by year fixed effects to alleviate the concern that the county-level smoke exposure measure captures industry-level performance trend and industry agglomeration. The estimated coefficient on smoke exposure decreases slightly but is still statistically significant. So far, I have assumed that most operating performance changes are due to smoke exposure in the county where a firm is headquartered. This assumption can lead to measurement errors if a firm has significant operations in other counties. Measurement errors in smoke exposure should result in an attenuation bias. To reduce potential measurement errors, column 4 excludes firms in industries known to have geographically dispersed workforces: retail, wholesale, and transportation industries. Compared with column 3, I find a slightly larger effect of smoke exposure on firm performances, suggesting that measurement errors bias against the estimation. In 12 “Over-controlling problems” happen when control variables are also the consequences of climate change and will bias the estimation. 26 column 5, I exclude firms headquartered in California, a state known for its widespread wildfire risk. . I find the baseline result is robust to excluding California.13 One drawback of this measure is the lack of smoke intensity. For example, the state of California has fewer days of smoke exposure than the country average, but the smoke intensity should be much higher due to its wildfire activities. The inclusion of firm fixed effects partially alleviates the measurement error because the coefficient estimates capture within-firm year-over- year differences. In section 6.3, I further addressed this issue by utilizing the daily air quality index (AQI) from EPA as a proxy for smoke intensity. 13 The results are also robust to excluding firms that headquartered in states with significant wildfires. Section 6.2 provides more deatailed discussions. 27 Table 3 Firm Performance Effects of Wildfire Smoke This table presents the effect of wildfire smoke on firm performance. The dependent variable is operating income before depreciation divided by total assets in year t-1. The interested independent variable is SmokeDays, which equals the number of days the county where a firm headquartered are covered by wildfire smoke. Columns 1 and 2 include firm fixed effects and year fixed effects. Columns 3 and 4 include firm fixed effects and industry by year fixed effects, and Column 4 excludes firms operating in geographically dispersed industries. Column 5 excludes firms operating in California. Regressions are weighted by county-level population each year. Robust standard errors that are clustered at the state level are reported in the parentheses. ***, **, and * indicate statistical significance at the 1%, 5%, and 10% levels using two-tailed tests, respectively. (1) (2) (3) (4) (5) Operating Income/Asset (%) Exclude Exclude Disperse Ind California SmokeDays -0.0202** -0.0202*** -0.0182*** -0.0201*** -0.0319*** (0.0077) (0.0045) (0.0057) (0.0066) (0.0103) WildFire -0.0040*** -0.0036*** -0.0017*** -0.0025*** (0.0009) (0.0008) (0.0006) (0.0007) Wind Speed 1.4826 0.9756 0.6620 0.8393 1.2726 (0.9333) (0.7883) (0.5606) (0.7630) (1.0046) Wind Direction -0.0134 -0.0067 -0.0020 -0.0010 -0.0138 (0.0110) (0.0074) (0.0094) (0.0111) (0.0123) Temp above 30 0.0119 -0.0056 0.0113 0.0125 -0.0118 (0.0243) (0.0180) (0.0114) (0.0100) (0.0180) Temp below 0 0.0060 -0.0046 0.0040 0.0014 -0.0110 (0.0194) (0.0130) (0.0087) (0.0090) (0.0161) Precip -0.0015** -0.0006 0.0003 0.0005 -0.0012*** (0.0006) (0.0005) (0.0005) (0.0006) (0.0004) GDP Percap 10.9416* 7.9946* 8.3183* 12.1952*** (5.4631) (4.3658) (4.7759) (3.7332) (3.1962) (3.4291) (4.1115) (3.1921) Observations 32,386 31,743 31,017 26,242 26,847 Adjusted R-squared 0.7591 0.7978 0.8142 0.8125 0.7857 Firm, Ind by Firm, Ind by Firm, Year Firm, Year Firm, Year Fixed Effect Year Year 28 5.3. Wildfire Smoke and Cross-sectional Tests I hypothesize that wildfire smoke leads to lower firm performance through reducing labor productivity. While higher levels of PM2.5 brought by wildfire smoke reduce productivity for both indoor and outdoor employees (Chang et al., 2016; 2019), high-skilled employees are of greater importance to firm performance because they are key components of a firm’s competitive advantage (Barney, 1991; Lengnick-Hall, 1992; Peteraf, 1993; Crook et al., 2011). Furthermore, high-skilled employees are more costly to replace due to limited labor supply (Ghaly et al., 2017). If the effect of wildfire smoke on firm performance is through reducing labor productivity, firms that rely more on high-skilled employees and have less flexibility to adjust their labor demand should suffer higher losses from wildfire smoke. 5.3.1. Firm-level Differences I conduct three cross-sectional tests to examine this hypothesis. High-skilled employees are of greater importance to firms with a higher emphasis on innovation (Zingales, 2000). In the first cross-sectional test, I follow the previous literature and use R&D investment per employee to proxy firms’ emphasis on innovation (Hill and Snell, 1988; Kor, 2006). A small decrease in R&D employees’ performance is likely to lead to a significant decrease in firm performance. Columns 1 and 2 of Table 4 present the estimation results. I restrict the sample to firms with positive R&D investment and further divide firms into two subsamples. The effect of wildfire smoke is much stronger in firms with higher R&D to employee ratios. In this subsample, one standard deviation (16 days) increase in annual smoke exposure decreases ROA by 58 basis points. Wildfire smoke has a smaller magnitude and insignificant effect on firms with lower reliance on human capital.14 14 Following Bhojraj, Sengupta, and Zhang (2017), I use z-statistics to test equality of two coefficients. The formula 𝛽1 −𝛽2 is √𝑆𝐸12 +𝑆𝐸22 29 In untabulated tests, I interact the R&D-to-employee ratio with wildfire smoke exposure. I find that the estimated coefficient on the interaction term is significantly negatively associated with ROA, suggesting that a firm is subject to higher impairment from wildfire smoke when its reliance on innovation increases. An alternative explanation for this cross-sectional test is that in firms with a high R&D-to-employee ratio, employees are more likely to work near headquarters. If so, firms in the subsample with high R&D-to-employee ratios may be subject to lower measurement errors in the wildfire smoke exposure variable. To rule out this alternative explanation, I focus on a two- digit SIC industry with the highest R&D to employee ratio in my sample (two-digit SIC = 28). I run model (1) with this subsample, and I find that the 𝑆𝑚𝑜𝑘𝑒𝐷𝑎𝑦𝑠 × 𝑅&𝐷/𝐸𝑚𝑝 interaction term is still significantly negative. Overall, these results suggest that even in the industry with the highest reliance on R&D, the effect of smoke exposure on firm performance is stronger when a firm’s R&D-to-employee ratio is high. Skilled labor, however, exists outside R&D teams. As skilled employees are likely to demand higher pay, I examine the heterogeneous effects of wildfire smoke among firms categorized by labor costs per employee. I use selling, general & administrative expense (SG&A) divided by the number of employees as a proxy for a firm’s average labor costs. I expect that firms with high labor costs are likely to suffer more from wildfire smoke than those with low labor costs. Columns 3 and 4 of Table 4 present this cross-sectional test. The effect of wildfire smoke is concentrated among firms with higher average labor costs. In this subsample, one standard deviation (16 days) increase in annual smoke exposure decreases ROA by 40 basis points. On the other hand, firms with lower average labor costs have reductions in ROA from wildfire smoke exposure that are smaller magnitudes and statistically insignificant. 30 Table 4 Human Capital and Effects of Wildfire Smoke This table presents the effect of wildfire smoke on firm performance with different levels of human capital importance. The dependent variable is operating income before depreciation divided by total assets in year t-1. The proxy for human capital importance is the R&D expenditure per employee, and the sample is restricted to firms with R&D expenditures. In columns 3-4, the proxy for human capital importance is labor cost. All columns include firm fixed effects and year fixed effects. Regressions are weighted by county-level population each year. Robust standard errors that are clustered at the state level are reported in the parentheses. ***, **, and * indicate statistical significance at the 1%, 5%, and 10% levels using two-tailed tests, respectively. (1) (2) (3) (4) Operating Income/Asset (%) Above-Median Below-Median Above-Median Below-Median R&D/Emp R&D/Emp Labor Cost Labor Cost SmokeDays -0.0344*** -0.0020 -0.0252*** -0.0077 (0.0124) (0.0068) (-0.0079) (0.0065) WildFire -0.0335** -0.0471*** -0.0057*** -0.0019* (0.0156) (0.0129) (0.0015) (0.0010) Wind Speed 1.0114 2.0473** 2.1196** -0.3962 (0.7654) (0.8670) (0.8982) (0.7233) Wind Direction 0.0644** -0.0344* 0.0145 -0.0186 (0.0251) (0.0188) (0.0151) (0.0118) Temp above 30 0.0295 -0.0180 -0.0026 -0.0136 (0.0284) (0.0207) (0.0204) (0.0150) Temp below 0 0.0205 -0.0215 0.0012 -0.0101 (0.0190) (0.0141) (0.0181) (0.0081) GDP Percap 0.9437 11.3316 8.8029 9.6897** (11.7091) (6.8616) (7.3151) (4.4284) Precip 0.0011 -0.0011*** -0.0012 -0.0001 (0.0010) (0.0004) (0.0012) (0.0003) Z test of Difference -2.2910** -1.7106* between Coefficients Observations 8,831 8,690 15,681 15,314 Adjusted R-squared 0.8254 0.7543 0.8040 0.7174 Fixed Effect Firm, Year Firm, Year Firm, Year Firm, Year 31 5.3.2. Industry-level Differences The previous two tests examine heterogeneity in the effect of wildfire smoke caused by firm- level characteristics. In the third cross-sectional test, I examine the effect of wildfire smoke on firm performance in industries with different levels of labor replacement costs. I follow Ghaly et 𝐸 al. (2017) in constructing industry reliance on skilled labor as ∑𝑜𝑗=1( 𝐸𝑗𝑖 ∗ 𝑍𝑗 ), where 𝑍𝑗 is required 𝑖 occupation skill-level identified by O*NET. 𝐸𝑗𝑖 is the number of employees in four-digit NAICS industry 𝑖 working in occupation 𝑗 , 𝐸𝑖 is the total number of employees in industry 𝑖, and O is the total number of occupations in industry 𝑖. Industries with higher dependence on skilled labor incur higher costs to replace current employees due to limited labor supply (Farber and Hallock 2009; Blatter, Muehlemann, and Schenker 2012). Therefore, if wildfire smoke affects firm performance by decreasing employee productivity, the observed effect should be stronger when a firm has higher labor replacement costs. In Table 5, I separate the sample into two subsamples based on a firm’s labor replacement costs. Consistent with my hypothesis, I find the effect of wildfire smoke is concentrated among firms with high labor replacement costs. In this subsample, a one standard deviation (16-day) increase in annual smoke exposure decreases ROA by 52 basis points. 32 Table 5 Labor Skill and Effects of Wildfire Smoke This table presents the effect of wildfire smoke on firm performance with different levels of labor skills. The dependent variable is operating income before depreciation divided by total assets in year t-1. The measure for labor skill follows Ghaly et al. (2017). Columns 1 includes the subsample with above-median labor skills, and columns 2 includes the sample with below-median labor skills. Both columns include firm fixed effects and year fixed effects. Regressions are weighted by county-level population each year. Robust standard errors that are clustered at the state level are reported in the parentheses. ***, **, and * indicate statistical significance at the 1%, 5%, and 10% levels using two-tailed tests, respectively. (1) (2) Operating Income/Asset (%) Above-Median Labor Below-Median Labor Skill Skill SmokeDays -0.0307** -0.0032 (0.0132) (0.0090) WildFire -0.0427*** -0.0053 (0.0130) (0.0104) Wind Speed 1.3088** -0.4717 (0.6155) (1.2779) Wind Direction 0.0218 -0.0080 (0.0200) (0.0214) Temp above 30 0.0079 -0.0167 (0.0168) (0.0197) Temp below 0 -0.0146 -0.0017 (0.0148) (0.0128) GDP Percap 15.9184** 6.7598 (6.4737) (4.7215) Precip -0.0016** -0.0004 (0.0006) (0.0004) Z test of Difference 1.7213* between Coefficients Observations 14,191 14,014 Adjusted R-squared 0.8198 0.7742 Fixed Effect Firm, Year Firm, Year 33 5.3.3. County-level Differences According to Figure 1, the Midwest region is subject to more days of smoke coverage than the west coast and the east coast due to wind conditions. It is possible the the negative association between smoke coverage and firm performance is also concerntrated in the Midwest. To further disentangle the differential effect of smoke exposure, I examine the effect of smoke coverage on firm performance with respect to different levels of county-level economic characteristics: population and income per capita. Table 6 presents the effect. I find that the effect of smoke coverage on operating income is statistically significant in counties with above-median population and income per capita, and statistically insignificant in counties with below-median population and income per capita. This result is consistent with Borgschulte et al (2021), who find that the adverse effect of air pollution on labor market is more severe in counties with above-median urbanization rate and home value. This finding also alleviates the concern that the negative association between smoke coverage and firm performance is a result of weaker economic conditions in the Midwest region. 34 Table 6 County Characteristics and Effects of Wildfire Smoke This table presents the effect of wildfire smoke on firm performance in counties with different size of populations and income. The dependent variable is operating income before depreciation divided by total assets in year t-1. Columns 1 includes the subsample with above-median county population, and columns 2 includes the sample with below-median county population. Columns 3 includes the subsample with above-median income per capita, and columns 4 includes the sample with below-median income per capita. All columns include firm fixed effects and year fixed effects. Regressions are weighted by county-level population each year. Robust standard errors that are clustered at the state level are reported in the parentheses. ***, **, and * indicate statistical significance at the 1%, 5%, and 10% levels using two-tailed tests, respectively. (1) (2) (3) (4) Operating Income/Asset (%) Above-Median Below-Median Above-Median Below-Median Population Population Income Income SmokeDays -0.0194*** -0.0045 -0.0113* -0.0146 (0.0059) (0.0153) (0.0059) (0.0134) WildFire -0.0241* -0.0047 0.0135 -0.0006 (0.0127) (0.0183) (0.0119) (0.0150) Wind Speed 0.9939 1.1011 2.9895*** 1.1560 (1.0128) (0.6704) (0.6181) (1.2492) Wind Direction -0.0011 -0.0023 -0.0443*** -0.0038 (0.0093) (0.0158) (0.0162) (0.0139) Temp above 30 -0.0085 -0.0026 -0.0184 -0.0216 (0.0213) (0.0247) (0.0376) (0.0204) Temp below 0 -0.0064 0.0035 0.0067 -0.0170 (0.0147) (0.0063) (0.0059) (0.0184) GDP Percap 14.1577* 3.1432 5.6517*** 7.5806 (7.7532) (2.4856) (1.8741) (6.7335) Precip -0.0008 0.0000 -0.0010 -0.0004 (0.0006) (0.0006) (0.0007) (0.0005) Z test of Difference 0.9086 0.2254 between Coefficients Observations 15,674 16,083 15,969 15,517 Adjusted R-squared 0.7955 0.8064 0.8374 0.7747 Fixed Effect Firm, Year Firm, Year Firm, Year Firm, Year 35 5.4. Effect of Wildfire Smoke on Sales and Productivity Human capital is a key source of competitive advantage. A small percentage of input loss from idiosyncratic shocks can significantly decrease sales without lowering demand (Barrot and Sauvagnat, 2016). Using a sample of non-US firms, Pankratz et al. (2021) find that lower employee productivity from high temperatures leads to decreased sales revenue. If wildfire smoke leads to lower employee input, lower sale revenue should be an important portion of decreased operating income. Alternatively, firms may be able to compensate for reductions in employee productivity by increasing labor inputs. Such mitigation strategies would be observed as increases in total operating expense without significant changes to sales revenue. Table 7 presents the effects of wildfire smoke on sales revenue and operating expense. I find that more days of smoke exposure lead to lower sales revenue instead of higher operating costs. Specifically, a one standard deviation (16-day) increase in annual smoke exposure decreases 𝑆𝑎𝑙𝑒𝑠 𝑅𝑒𝑣𝑒𝑛𝑢𝑒/𝐴𝑠𝑠𝑒𝑡 by 59 to 81 basis points, corresponding to a $34 to $47 million loss in sales revenue for a firm with an average book value of assets. Columns 3 and 4 of Table 6 show that operating expense decreases insignificantly, suggesting firms do not significantly increase labor input to compensate for the lost productivity in the short run. 36 Table 7 Impact of Wildfire Smoke on Sales and Expenses This table presents the effect of wildfire smoke on sales and expenses. In columns 1-2, the dependent variable is Sale/Asset, which equals sales revenue divided by asset in year t-1. In columns 3-4, the dependent variable is operating expense, which equals total operating expense divided by asset in year t-1. The interested independent variable is SmokeDays, which equals the number of days the county where a firm headquartered are covered by wildfire smoke. Columns 1 and 3 include firm fixed effects and year fixed effects, columns 2 and 4 include firm fixed effects and industry by year fixed effects. Regressions are weighted by county-level population each year. Robust standard errors that are clustered at the state level are reported in the parentheses. ***, **, and * indicate statistical significance at the 1%, 5%, and 10% levels using two-tailed tests, respectively. (1) (2) (3) (4) Sale/Asset (%) Operating Expense/Asset (%) SmokeDays -0.0369** -0.0508** -0.0173 -0.0288 (0.0174) (0.0224) (0.0164) (0.0205) WildFire -0.0100*** -0.0050 -0.0058*** -0.0026 (0.0024) (0.0043) (0.0021) (0.0041) Wind Speed 0.4148 -2.3487* -0.3935 -2.7164** (1.6733) (1.2677) (1.5440) (1.2880) Wind Direction -0.0030 0.0450 0.0006 0.0482 (0.0257) (0.0286) (0.0268) (0.0299) Temp above 30 -0.0293 0.0342 -0.0225 0.0277 (0.0325) (0.0323) (0.0157) (0.0292) Temp below 0 0.0023 0.0058 0.0092 0.0014 (0.0315) (0.0359) (0.0222) (0.0305) GDP Percap 26.4051** 15.8621*** 11.7600** 6.7097 (10.7259) (5.6345) (4.7625) (4.6648) Precip -0.0002 0.0014 0.0007 0.0014 (0.0013) (0.0014) (0.0006) (0.0009) Observations 31,755 31,026 31,755 31,026 Adjusted R-squared 0.8724 0.9015 0.8753 0.9015 Firm, Ind by Firm, Ind by Fixed Effect Firm, Year Year Firm, Year Year 37 5.5. Wildfire Smoke and Earnings Announcements Return In this section, I examine whether smoke exposure predicts the unexpected component of firm performance. Despite significantly increasing particulate matter concentration, most wildfire smoke events can only be seen from satellites and are invisible to human eyes (Beitler, 2006; Borgschulte et al., 2020). The effect of smoke exposure on firm performance is less salient than the effect of wildfire damage and, therefore, may surprise market participants. Alternatively, if the market is efficient, market participants may incorporate publicly available wildfire smoke data into prices. In Table 8, column 1 examines whether financial analysts have taken climate-related information into earnings forecasts. Following the literature on analyst forecast accuracy, I calculate the earnings surprise as the difference between actual earnings per share and average forecasted earnings per share divided by the month-end share price before the annual earnings announcement. I find that wildfire smoke exposure is associated with negative earnings surprises. This result suggests that although sell-side analysts integrate sustainability scores into stock recommendations (Eccles, Krzus, and Serafeim 2011), they may underestimate the threats of climate change on firm performance. Analysts do not fully consider information on smoke exposure. A natural question is whether investors anticipate the effect of smoke exposure on firm performance. If the market fully understands the effect of smoke exposure on corporate performance, there should be little information content related to earnings around earnings announcements. In column 2 of Table 8, I examine the association between smoke exposure and [-1, 1] day cumulative abnormal returns (CAR) around the annual earnings announcement date. I find a both statistically and economically significant association between smoke exposure and CAR. According to the coefficient estimate 38 in column 2, a one standard deviation (16-day) increase in annual smoke exposure is associated with a 21-basis point decrease in [-1, 1] day CAR. These findings suggest that market participants do not fully anticipate the diffuse effects of physical climate risk. 39 Table 8 Wildfire Smoke and Earnings Announcement Return This table presents the effect of wildfire smoke on earnings announcement return. The sample is restricted to observations with a stock price of at least five dollars. In column 1, the dependent variable is the difference between actual earning and mean analyst forecast earnings divided by the month-end stock price before the earnings announcement. In column 2, the dependent variable is 3-day cumulative abnormal returns around the annual earnings announcement date. The interested independent variable is SmokeDays, which equals the number of days the county where a firm headquartered are covered by wildfire smoke. Both columns include firm fixed effects and year fixed effects. Regressions are weighted by county-level population each year. Robust standard errors that are clustered at the state level are reported in the parentheses. ***, **, and * indicate statistical significance at the 1%, 5%, and 10% levels using two-tailed tests, respectively. (1) (2) Earning Surp CAR [-1, 1] (%) SmokeDays -0.0189*** -0.0134** (0.0053) (0.0063) WildFire -0.0020*** -0.0034** (0.0007) (0.0016) Wind Speed 0.9849 0.1175 (0.7249) (0.5103) Wind Direction 0.0185 0.0194 (0.0132) (0.0132) Temp above 30 -0.0199 -0.0004 (0.0136) (0.0055) Temp below 0 -0.0085 0.0127** (0.0085) (0.0051) GDP Percap 10.3087* 2.6422** (5.3994) (1.1663) Precip -0.0002 -0.0010*** (0.0002) (0.0003) Observations 22,010 23,263 Adjusted R-squared 0.4614 0.0439 Fixed Effect Firm, Year Firm, Year 40 5.6. Disclosure of Climate Change Exposure Accounting theory suggests disclosure improves market efficiency (Gao, 2008). Climate change disclosure may moderate the impairment of wildfire smoke exposure on operating income and market returns. I provide insight into this discussion by examining whether firms and the financial market benefit from voluntary climate-change disclosure. The literature finds that voluntary disclosure of non-financial information is associated with positive feedback from the market. For example, firms enjoy a lower cost of equity after voluntarily disclosing their corporate social responsibility (CSR) activities (Dhaliwal et al., 2011). The benefits of issuing voluntary disclosure include improving the information environment and better internal control (Cheng, Ioannou, and Serafeim, 2013). I first examine the moderating role of climate-change disclosure on the adverse effect of wildfire smoke exposure on operating income. My proxy of climate change disclosure comes from Sautner et al. (2020), who apply a machine-learning approach to extract firm-level climate change exposure (CC_expo) from conference call transcripts. The analysis is restricted to the sample of firms with non-missing climate change exposure during the annual earnings conference calls. I divide the sample into subsamples of firms based on the climate change disclosure in year t-1. Columns 1 and 2 of Table 9 present the effect of smoke exposure on firms’ operating income with different levels of voluntary disclosure. I find no evidence that the levels of voluntary disclosure of climate change risk are associated with operating loss from wildfire smoke. This result alleviates the concerns of self-selection bias: Firms with higher CC_expo may have lower wildfire smoke exposure. I then examine whether climate change disclosure is associated with less negative feedback from the financial market. Columns 3 and 4 of Table 9 presents the effects of smoke exposure on 41 market reactions with different levels of voluntary disclosure. I find that the negative association between the market reaction to earnings announcements and smoke exposure is only significant among firms with low climate change disclosure. This result suggests that more voluntary climate change disclosure increases firms’ information environmental, and leads to better capital market efficiency. 42 Table 9 Voluntary Disclosure and Effects of Wildfire Smoke This table presents the effect of wildfire smoke on firm performance and the stock market with different levels of voluntary climate change disclosure. In columns 1 and 2, the dependent variable is operating income before depreciation divided by total assets in year t-1. In columns 3 and 4, the dependent variable is CAR [-1, 1] around the annual earnings announcement date. The proxy for climate change disclosure is the proportion of climate-exposure keywords from Sautner et al. (2020) in year t-1, and the sample is restricted to firms with climate change disclosure. Columns 1 and 3 include the subsample with above-median climate change disclosure, and columns 2 and 4 include the sample with below-median climate change disclosure. All columns include firm fixed effects and year fixed effects. Regressions are weighted by county-level population each year. Robust standard errors that are clustered at the state level are reported in the parentheses. ***, **, and * indicate statistical significance at the 1%, 5%, and 10% levels using two-tailed tests, respectively. (1) (2) (3) (4) Operating Income/Asset (%) CAR [-1, 1] (%) High Disclosure Low Disclosure High Disclosure t- Low Disclosure t-1 t-1 1 t-1 SmokeDays -0.0278** -0.0211*** 0.0073 -0.0322** (0.0137) (0.0056) (0.0074) (0.0140) WildFire -0.0041** -0.0016 -0.0058*** -0.0033*** (0.0019) (0.0011) (0.0020) (0.0011) Wind Speed 1.0965 -0.1753 0.5182 -0.4855 (0.7441) (0.7729) (0.6088) (0.8685) Wind Direction -0.0099 -0.0118 -0.0037 -0.0469*** (0.0095) (0.0231) (0.0136) (0.0162) Temp above 30 0.0078 -0.0162 -0.0101 0.0278* (0.0145) (0.0249) (0.0102) (0.0160) Temp below 0 -0.0062 -0.0098 0.0086** 0.0137 (0.0129) (0.0135) (0.0042) (0.0103) GDP Percap 8.3576** 8.1881 2.4117 2.0331 (3.2939) (5.9472) (1.8602) (2.7069) Precip -0.0003 -0.0013** -0.0000 -0.0013*** (0.0004) (0.0005) (0.0004) (0.0005) Z test of Difference -0.4527 2.4944** Observations between Coefficients 10,685 10,672 9,891 9,998 Adjusted R-squared 0.8107 0.8143 0.0826 0.0703 Fixed Effect Firm, Year Firm, Year Firm, Year Firm, Year 43 5.7. Adaptation Behaviors In response to increasing climate-change risk, firms are pursuing mitigation strategies. For example, firms are adjusting supply chains to reduce exposure to heat waves (Pankratz and Schiller, 2020). When long-term abnormal temperatures decrease local demand, firms are likely to close establishments (Jin et al., 2021). Relative to temperature change, smoke exposure is less salient and predictable. I expect firms to engage in soft adaptation, such as outsourcing production across facilities, rather than permanent changes. To identify firms’ adaptation behavior, I examine the association between smoke exposure and plant-level production activities. I collect plant-level information from the Toxic Release Inventory (TRI) database. The database covers manufacturing plants (SIC from 2000 to 3999) with more than ten full-time employees. Instead of using headquarters smoke exposure, I use each plant’s county location to determine its annual smoke exposure. I use two measures as proxies for a plant’s production activities. Following Akey and Appel (2021), I use the production ratio reported in the TRI database to construct the first production measure. The TRI reports the change in production associated with each chemical. For example, if a chemical is generated in producing 𝑄𝑢𝑎𝑛𝑡𝑖𝑡𝑦 𝑃𝑟𝑜𝑑𝑢𝑐𝑒𝑑𝑝,𝑡 product p, the TRI will report a production ratio at year t for the chemical as 𝑄𝑢𝑎𝑛𝑡𝑖𝑡𝑦 𝑃𝑟𝑜𝑑𝑢𝑐𝑒𝑑 . 𝑝,𝑡−1 I calculate the cumulative production quantity by setting the production quantity as one in the first year that a plant reports its chemical production ratio and multiplying forward each year using the following equation: 𝑃𝑟𝑜𝑑𝑢𝑐𝑡𝑖𝑜𝑛 𝑄𝑢𝑎𝑛𝑡𝑖𝑡𝑦𝑝,𝑡 = ∏𝑡𝑡≠1 1 × ( 𝑃𝑟𝑜𝑑𝑢𝑐𝑡𝑖𝑜𝑛 𝑅𝑎𝑡𝑖𝑜𝑝,𝑡 ) For plants that emit multiple chemicals, I take the average of all cumulative production to get a plant-year measure. The second measure of production is total toxic waste emitted by a plant. 44 Table 10 presents the effect of plant-level smoke exposure on production activities. On average, I find no effect of annual smoke exposure on plant-level production activities. However, a plant’s smoke exposure leads to reduced production activities after experiencing a continuously high level of smoke exposure, measured by the quartile of smoke exposure in the past five years. These results suggest that firms learn from the long-term experience of smoke exposure. 45 Table 10 Plant-level Adaptation to Smoke Exposure This table presents the effect of wildfire smoke on plant-level production activities. In column 1, the dependent variable is the average quantity production from each chemical. In column 2, the dependent variable is the logarithm of total pollutant emissions. The interested independent variable is SmokeDays, which equals the number of days the county where a plant is covered by wildfire smoke. SmokePre5 Q4 equals 1 if a plant experienced 4th quartile of smoke days from year t-5 to year t-1. Both columns include plant fixed effects and year fixed effects. Regressions are weighted by the plant’s county-level population each year. Robust standard errors that are clustered at the plant state level are reported in the parentheses. ***, **, and * indicate statistical significance at the 1%, 5%, and 10% levels using two-tailed tests, respectively. (1) (2) Production Quantity Log (Total Emissions) Plant SmokeDays 0.0001 -0.0008 (0.0002) (0.0007) Plant SmokeDays *SmokePre5 Q4 -0.0014* -0.0039*** (0.0007) (0.0012) SmokePre5 Q4 0.0392 0.1793*** (0.0306) (0.0435) Wind Speed-Facility 0.0321 0.0592 (0.0264) (0.1027) Wind Direction-Facility 0.0003 0.0005 (0.0003) (0.0014) WildFire-Facility -0.0000* -0.0000 (0.0000) (0.0000) Temp above 30-Facility 0.0002 -0.0007 (0.0001) (0.0006) Temp below 0-Facility -0.0002 -0.0014** (0.0002) (0.0006) GDP Percap-Facility -0.0007 0.0147 (0.0140) (0.0329) Precip-Facility 0.1341 0.6554* (0.1432) (0.3878) Observations 64,456 56,207 Adjusted R-squared 0.1206 0.9158 Fixed Effect Plant, Year Plant, Year 46 5.8. Wildfire smoke and Supply-chain Disruptions The cross-sectional tests suggest that wildfire smoke exposure affects a firm’s operating performance by lowering productivity among skilled workers. It is possible that reduced consumer demand when exposed to wildfire smoke also contributes to the observed effects. Public firms’ customers may be located far from supplier headquarters, but significant correlations may exist between a firm’s smoke exposure and its customers’ smoke exposure. I use three measures to identify the effect of smoke exposure on customer demand. The first measure captures major customers’ smoke exposure using customers identified by Compustat’s segment data. I use counties where major customers are headquartered to measure customer smoke exposure. Columns 1 and 2 of Table 11 present the results. Column 2 excludes observations if major customers operate in geographically dispersed industries. I find no significant association between major customers’ smoke exposure and a firm’s operating performance. The second measure follows Srinivasan, Lilien, and Rangaswamy (2002) and uses the four- digit SIC code to determine whether a firm belongs to the business-to-business (B2B) or business- to-consumer (B2C) industry. If a firm’s smoke exposure captures its customer exposure, the effects on operating performance should be stronger in B2C industries (Jin et al., 2021). Columns 3 and 4 of Table 11 present the results. I find no significantly different effect of smoke exposure between firms in the B2B industry or B2C industry. The third measure relies on vertical integration data from Frésard, Hoberg, and Phillips (2019). This data utilizes text-based product information to identify a firm’s potential suppliers and customers. I construct aggregated customer (supplier) smoke exposure by summing each potential customer (supplier) smoke exposure weighted by relatedness scores. Columns 5 and 6 of Table 11 include potential customer (supplier) smoke exposure in equation (1). While I find potential 47 customer (supplier) smoke exposure leads to lower operating performance, the coefficient estimates of a firm’s smoke exposure change little. This evidence that supply-chain disruptions from wildfire smoke lead to lower operating performance supplements Pankratz and Shiller (2020), who find that operating performance decrease when suppliers experience abnormal temperatures and floods. In all columns, focal firms’ wildfire smoke exposure remains significantly associated with operating performance. Therefore, I conclude that my estimate of the effect of wildfire smoke exposure on firm performance is mostly driven by changes in employee productivity. 48 Table 11 Wildfire Smoke and Supply Chain This table presents the effect of wildfire smoke on operating income through supply-chain disruptions. The dependent variable is operating income before depreciation divided by total assets in year t-1. The interested independent variable is SmokeDays, which equals the number of days the county where a firm headquartered are covered by wildfire smoke. Columns 1 and 2 include the smoke exposure days of major customers in the segment reporting. Column 2 excludes observations if customers operate in dispersed industries. Columns 3 and 4 include smoke days interact with an indicator of B2B or B/C industries. Columns 5 and 6 include smoke exposures of potential customers and suppliers. Weather characteristics and firm characteristics are included in all regressions. All columns include firm fixed effects and year fixed effects. Regressions are weighted by county-level population each year. Robust standard errors that are clustered at the state level are reported in the parentheses. ***, **, and * indicate statistical significance at the 1%, 5%, and 10% levels using two-tailed tests, respectively. (1) (2) (3) (4) (5) (6) Operating Income/Asset (%) Exclude Dispersed Customers SmokeDays -0.0198** -0.0232** -0.0168** -0.0223*** -0.0172*** -0.0176*** (0.0092) (0.0098) (0.0066) (0.0073) (0.0041) (0.0043) Customer SmokeDays 0.0143 0.0205 (0.0108) (0.0130) SmokeDays*B2B -0.0035 (0.0080) SmokeDays*B2C 0.0069 (0.0102) Potential Customer SmokeDays -0.0084*** (0.0027) Potential Supplier SmokeDays -0.0062** (0.0026) Customer ROA 3.6407 -0.8086 (2.5999) (2.8374) 49 Table 11 (cont’d) Potential Customer ROA 3.9708*** (0.8778) Potential Supplier ROA 2.8006*** (0.8429) Observations 31,653 28,742 30,977 30,977 31,196 31,196 Adjusted R-squared 0.8033 0.8102 0.7977 0.7977 0.7980 0.7979 Weather Control Yes Yes Yes Yes Yes Yes Firm Control Yes Yes Yes Yes Yes Yes Fixed Effects Firm, Year Firm, Year Firm, Year Firm, Year Firm, Year Firm, Year 50 CHAPTER 6: ROBUSTNESS TESTS 6.1. Firms with Dispersed Employees In this section, I perform additional analyses to identify the potential measurement error biases. First, I examine the size of firm-specific measurement errors due to dispersed employees. I identify the locations of plants, units, subsidiaries, and branches from the LexisNexis corporate affiliation database. I consider a firm to have dispersed employees if at least one affiliation has more than 15% of employees relative to the corporate number of employees and locates in different counties as the headquarters. Table 12 presents the results. Column 1 includes all matched observations and column 2 excludes firms with dispersed employees. Consistent with the baseline results, I find a larger magnitude of smoke exposure on operating income when excluding firms with dispersed employees. Columns 3 and 4 replace operating income with net income divided by the book value of assets at year t-1 as an alternative performance measure. I find that smoke exposure also leads to decreased net income. 51 Table 12 Effect of Dispersed Operation This table presents the robustness test of the effect of wildfire smoke on firm performance. The sample contains all affiliations (plants, units, subsidiaries, and branches) in the LexisNexis corporate affiliation database. Dispersed Employee is an indicator variable that equals 1 if any affiliation has 15% or more employees of the parent firms, and 0 otherwise. The interested independent variable is SmokeDays, which equals the number of days the county where a firm headquartered are covered by wildfire smoke. In columns 1-2, the dependent variable is operating income before depreciation divided by total assets in year t-1. In columns 3-4, the dependent variable is adjusted net income divided by total assets in year t-1. All columns include firm fixed effects and year fixed effects. Regressions are weighted by county-level population each year. Robust standard errors that are clustered at the state level are reported in the parentheses. ***, **, and * indicate statistical significance at the 1%, 5%, and 10% levels using two-tailed tests, respectively. (1) (2) (3) (4) Operating Income/Asset (%) Net Income/Asset (%) Exclude Exclude Full Sample Full Sample Dispersed Dispersed Employee Employee SmokeDays -0.0212** -0.0238*** -0.0232** -0.0247** (0.0092) (0.0085) (0.0096) (0.0105) Subsidiary SmokeDays -0.0020 0.0052* -0.0024 0.0039 (0.0051) (0.0030) (0.0055) (0.0051) WildFire -0.0015*** -0.0016* -0.0016*** -0.0010* (0.0003) (0.0009) (0.0004) (0.0006) Wind Speed 0.9684** 0.8843** 1.6069** 1.9749** (0.4386) (0.3821) (0.6357) (0.8049) Wind Direction 0.0094 -0.0126 0.0034 -0.0158 (0.0137) (0.0121) (0.0142) (0.0213) Temp above 30 -0.0168 -0.0159 -0.0168** -0.0233*** (0.0117) (0.0130) (0.0082) (0.0082) Temp below 0 -0.0174 -0.0182** -0.0098 -0.0126 (0.0112) (0.0083) (0.0119) (0.0092) GDP Percap 12.2306*** 10.6583*** 12.7701*** 11.6513*** (4.5402) (3.4945) (3.4717) (3.0759) Precip -0.0001 -0.0003 -0.0004 -0.0008* (0.0004) (0.0003) (0.0005) (0.0004) (1.8213) (3.3265) (3.2794) (4.3102) Observations 13,863 9,933 14,162 10,152 Adjusted R-squared 0.7352 0.7528 0.5952 0.5927 Fixed Effect Firm, Year Firm, Year Firm, Year Firm, Year 52 6.2. Portfolio Returns While I find that higher smoke exposure is associated with greater earnings surprise and negative market reactions, it is possible that there is no long-term effects of smoke exposure on the stock return. I then explore whether the diffuse effects of climate change have long-term market effects by exploring the association between wildfire smoke exposure and long-term buy-and-hold abnormal returns (BHAR). I assign firms into four quartiles based on the number of days a firm experiences smoke coverage as of year t-1. Panel A of Table 13 presents my findings. On average, a trading strategy of buying and holding firms in the lowest smoke exposure quartile and selling firms in the highest smoke exposure quartile in year t earns 6.6% abnormal annual returns over 12 months. In Panel B, I regress BHAR-3 month to BHAR-24 month on year t-1 smoke coverage and control for firm fixed effects. The results are consistent with Panel A. 53 Table 13 Portfolio Returns This table presents the average portfolio buy-and-hold abnormal return (BHAR) for up to 24 months. The portfolio is sorted by quartile of smoke days in the previous year. For each firm i, 𝐵𝐻𝐴𝑅𝑖(𝑡,𝑇) = ∏𝑇𝑡=0(1 + 𝑅𝑖𝑡 ) − ∏𝑇𝑡=0(1 + 𝐸[𝑅𝑖𝑡 |𝑋𝑡 ]) , where 𝑅𝑖𝑡 is monthly return, and 𝐸[𝑅𝑖𝑡 |𝑋𝑡 ] is value-weighted monthly market return. The holding period starts from the month after the fiscal year-end. Panel A presents the equal-weighted portfolio BHAR, and panel B presents regression results of different time-horizon BHAR. All regressions include firm fixed effects. Robust standard errors that are clustered at the state level are reported in the parentheses. ***, **, and * indicate statistical significance at the 1%, 5%, and 10% levels using two-tailed tests, respectively. Panel A: Trading Portfolio Smoke Days 3 Month 6 Month 9 Month 12 Month 18 Month 24 Month Quartile 1 0.007 0.024 0.065 0.102 0.053 0.102 2 -0.007 0.003 0.032 0.048 0.044 0.102 3 -0.006 -0.012 -0.001 0.021 0.048 0.100 4 0.001 -0.002 0.009 0.036 0.039 0.089 Quartile 1 - Quartile 0.005 0.026*** 0.056*** 0.066*** 0.014 0.014 4 Panel B: Regression Result (1) (2) (3) (4) (5) (6) 𝐵𝐻𝐴𝑅𝑖(𝑡,𝑇) 3 Month 6 Month 9 Month 12 Month 18 Month 24 Month - - SmokeDays t-1 -0.0004 -0.0016*** -0.0034*** -0.0034*** 0.0021*** 0.0035*** (0.0003) (0.0004) (0.0005) (0.0006) (0.0006) (0.0007) - - Leverage -0.0977*** -0.1717 -0.2119** -0.3286** 0.3931*** 0.5968*** (0.0293) (0.1066) (0.1040) (0.1375) (0.0583) (0.0692) M/B 0.0269*** 0.0556*** 0.0919*** 0.1350*** 0.1931*** 0.2519*** (0.0026) (0.0036) (0.0057) (0.0052) (0.0073) (0.0121) Log(MVEQ) 0.0472*** 0.0629*** 0.0767*** 0.1029*** 0.1528*** 0.2151*** (0.0045) (0.0058) (0.0065) (0.0095) (0.0114) (0.0168) Observations 33,634 33,634 33,634 33,634 32,773 31,912 Adjusted R-squared 0.0922 0.0808 0.0920 0.1097 0.2235 0.2118 54 6.3. Unobservable Wildfires While all regressions control for the number of days that a firm’s headquarters experienced wildfires, unobservable wildfires in nearby counties could associate with both wildfire smoke and corporate performance. To rule out this alternative explanation, I exclude firms from analysis if they are headquartered in states that experienced significant wildfires. Significant wildfires are defined as wildfires that caused at least $10 million in property damages according to NOAA. The analysis results are presented in Table 14. Column 1 includes firm and year fixed effects. Column 2 further includes industry by year fixed effects to control for unobservable industry trends. For both models, I find the coefficient estimate of wildfire smoke is similar to the baseline tests. This result alleviates the concern that the negative association between wildfire smoke and corporate performance is driven by unobservable wildfire damages. 55 Table 14 Exclude States with Significant Wildfires This table presents the robustness test of the effect of wildfire smoke on firm performance. State- years with significant wildfires are excluded from the sample. Significant wildfires are wildfires that caused more than $10 million property damage. The interested independent variable is SmokeDays, which equals the number of days the county where a firm headquartered are covered by wildfire smoke. Columns 1 includes firm fixed effects and year fixed effects, and columns 2 includes firm fixed effects and industry by year fixed effects. Regressions are weighted by county- level population each year. Robust standard errors that are clustered at the state level are reported in the parentheses. ***, **, and * indicate statistical significance at the 1%, 5%, and 10% levels using two-tailed tests, respectively. (1) (2) Operating Income/Asset (%) SmokeDays -0.0208*** -0.0187** (0.0075) (0.0077) WildFire -0.0335* -0.0240* (0.0191) (0.0131) Wind Speed 0.9154 0.8792 (0.8462) (0.8093) Wind Direction -0.0290* -0.0254 (0.0150) (0.0205) Temp above 30 -0.0072 0.0114 (0.0153) (0.0089) Temp below 0 -0.0116 -0.0018 (0.0137) (0.0090) Precip 10.7151** 8.0342** (4.1145) (3.7498) GDP Percap -0.0006 0.0004 (0.0005) (0.0005) Observations 25,375 24,659 Adjusted R-squared 0.7891 0.8046 Fixed Effect Firm, Year Firm, Ind by Year 56 6.4. Smoke Intensity In the main analysis, the measure of smoke exposure is the number of days that a firm’s headquarters is fully covered by smoke plumes. One drawback of this measure is that it does not capture the smoke intensity. To overcome this drawback, I obtain daily AQI data from the EPA and compare the differences in AQI between smoke days and non-smoke days. I first calculate the average monthly AQI in each county during smoke days (e.g., June 2018), and non-smoke days in the previous five years (e.g., June of 2013 - 2017). I then compare monthly AQI between smoke days in each month and non-smoke days in the same month over the previous five years. If the AQI on smoke days is lower than on non-smoke days, I code smoke intensity as zero. Otherwise, smoke intensity is the difference between the two average monthly AQIs. Finally, I aggregate the monthly smoke intensity into an annual measure by averaging the monthly smoke intensity. This measurement of smoke intensity assumes that wildfire smoke contributes to all increases in AQI during smoke days. However, the value of AQI is calculated from observed levels of several air pollutants (PM2.5, PM10, Ozone, and SO2). The main pollutant in wildfire smoke is PM2.5. To better capture the effect of wildfire smoke, I follow Deryugina et al., (2019) in using the instrumental variables (IV) approach. After controlling for other wildfire activities and other weather characteristics, I argue that a county’s wind speed and wind direction are unrelated to corporate performance except through their influence on wildfire smoke intensity. The estimate of two-stage least squares is presented in Table 15. I find that wind speed significantly predicts smoke intensity (F = 4.10). Moreover, higher smoke intensity leads to significantly lower corporate performance: A one-standard-deviation increase in smoke intensity (12.5) leads to a 73-basis point loss of annual ROA. In columns 3 and 4, I scale smoke intensity by the number of smoke days, and I find similar estimation results. 57 Table 15 Smoke Intensity: An Instrumental Variable Approach This table presents the effect of wildfire smoke on firm performance using the instrumental variable approach. The instrumental variables are wind speed and wind direction following Deryugina et al., (2019). In column 1, the instrumented variables are Smoke Intensity, which equals the annual average differences in air quality index between smoke days and non-smoke days. In column 3, the instrumented variable is Smoke Intensity/Smoke Days, which equals the smoke intensity scaled by the number of smoke days. All columns include firm fixed effects and year fixed effects. Regressions are weighted by county-level population each year. Robust standard errors that are clustered at the state level are reported in the parentheses. ***, **, and * indicate statistical significance at the 1%, 5%, and 10% levels using two-tailed tests, respectively. (1) (2) (3) (4) Smoke Operating Smoke Operating Intensity Income/Asset Intensity/ Income/Asset First Stage Second (%)Stage Smoke Days First Stage Second (%)Stage Wind Speed -13.2548** -4.2424*** (5.8540) (1.1813) Wind Direction 0.1416 0.0346** (0.1054) (0.0130) F-test (p-value) 4.10 21.56 (< Smoke Intensity (0.0231) -0.0583* 0.001) (0.0310) Smoke Intensity/Smoke Days -0.1920* (0.1033) WildFire -0.0017 -0.0010 (0.0011) (0.0009) Temp above 30 -0.0018 -0.0021 (0.0111) (0.0126) Temp below 0 0.0068 0.0046 (0.0085) (0.0087) GDP Percap 5.6741*** 4.6675** (2.0119) (2.3109) Precip -0.0002 -0.0000 (0.0004) (0.0005) (2.4445) (2.4770) Under-identification test (Kleibergen-Paap rk LM statistic) (p-value) 5.303 (0.0705) 8.284 (0.0159) Observations 30,384 30,384 30,384 30,384 Adjusted R-squared 0.1844 0.1851 Fixed Effects Firm, Year Firm, Year Firm, Year Firm, Year 58 CHAPTER 7: CONCLUDING REMARKS This paper examines the effects of climate change on US corporate performance. Unlike prior studies that use abnormal temperatures or localized natural disasters as proxies for climate change, this paper proposes wildfire smoke as a new measure of climate change. The effects of wildfire smoke are far-reaching and hard to predict. I find wildfire smoke exposure leads to decreased corporate performance. Wildfire smoke exposure has a stronger adverse effect on firm performance among firms most reliant on human capital. Although extant studies find that economies in developed countries are resistant to extreme temperatures, high reliance on human capital makes these economies more vulnerable to wildfire smoke. Moreover, smoke events are not as salient as temperature changes and natural disasters, making it hard for market participants to price its effect on corporate performance absent adequate disclosure. Finally, this paper explores whether firms and investors benefit from climate-change disclosure. Using firm’s voluntary climate-change disclosure in conference calls, I find that higher climate- change disclosure moderates the negative association between wildfire smoke exposure and market reaction to earnings announcements. This association may incentivize investors to require more climate-change disclosure and policy implication for regulators. 59 APPENDICES 60 APPENDIX A Variable Definitions 61 Table A.1 Variable Definitions This table presents the definition of all variables in this paper. I thank Srinivasan, Lilien, and Rangaswamy (2002) and Frésard, Hoberg, and Phillips (2019) for providing their data. All other data is publicly available/available through subscription. Variable Definitions Operating Operating income before depreciation divided by the total book value of Income/Asset assets in year t-1 Operating Total operating expense divided by the total book value of assets in year t- Expense/Asset 1 Sales/Asset Total sales revenue divided by the total book value of assets in year t-1 Three-day cumulative abnormal return around annual earnings CAR [-1, 1] announcement date using Fama-French 3-factor model. The difference between (Actual EPS - Forecasted EPS)/Share Price. The Earning Surp share price in month-end before the annual earnings announcements. Net Income/Asset Net income divided by the total book value of assets in year t-1 SmokeDays The number of days that a county is fully covered by wildfire smoke The data is provided by Sautner, Van Lent, Vilkov, and Zhang (2020), who construct firm-level voluntary disclosure of climate change exposure Climate Change from earnings conference calls. The measures equal the frequency of Disclosure climate change bigrams, scaled by the total number of bigrams in the transcript of each year. WildFire The number of days that a county has wildfires each year GDP Percap The natural logarithm of county-level GDP per capita in each year The average wind direction based on a county’s centroid coordinate, with Wind Direction zero corresponding to wind blowing from the north and higher values corresponding to compass directions clockwise. Wind Speed The average wind speed based on a county’s centroid The number of days a county’s centroid has an average temperature of Temp above 30 30oC or above The number of days a county’s centroid has an average temperature of Temp below 0 0oC or below Precip Annual precipitation measured at a county’s centroid 62 Table A.1 (cont’d) R&D R&D expenditures divided by the total book value of assets Leverage Firm leverage, total liability divided by the total book value of assets Difference between current year sales revenue and previous year sales Sales Growth revenue divided by previous year sales revenue The market value of equity and book value of liability divided by book M/B value of assets The net value of property, plant, and equipment divided by the total book PPE/Asset value of assets in year t-1 Book value of liabilities plus the market value of equity divided book Tobin’s Q value of assets R&D expenditures (in millions) per employee; this measure is restricted to R&D/Emp firms with positive R&D expenditures. Labor Cost SG&A expenditures (in millions) exclude R&D expenses per employee 𝐸 The industry labor skill index for industry i is calculated as ∑𝑜𝑗=1( 𝐸𝑗𝑖 ∗ 𝑍𝑗 ), 𝑖 Labor Skills where 𝐸𝑗𝑖 is the number of employees in an industry i working in occupation j, 𝐸𝑖 is the total number of employees in industry i, and O is the total number of occupations in the industry i The natural logarithm of total toxic releases (measured in pounds) from Log (Total Emission) TRI for each plant The accumulated production following Akey and Appel (2021). Production Quantity Accumulated production= ∏𝑡𝑡≠1( 𝑃𝑟𝑜𝑑𝑢𝑐𝑡𝑖𝑜𝑛 𝑅𝑎𝑡𝑖𝑜𝑝,𝑡 ), missing ratios are replaced by 1. The quartile of smoke days that a plant experienced from year t-5 to year SmokePre5 Q t-1. SmokePre5 Q4 equals 1 if a plant experienced 4th quartile of smoke days from year t-5 to year t-1, and 0 otherwise. The number of days a major customer’s headquarters county is covered by Customer SmokeDays wildfire smoke. Major customers data come from segment reporting. Business-to-Business (Business-to-consumer) industry indicator that B2B (B2C) Industry follows Srinivasan, Lilien, and Rangaswamy (2002) The number of days a firm’s potential customers and suppliers are Potential Customer covered by wildfire smoke, weighted by relatedness scores. Potential (Supplier) customers (suppliers) are identified by the text-based vertical integration SmokeDays data from Frésard, Hoberg, and Phillips (2019). 63 Table A.1 (cont’d) An indicator that equals to 1one if any subsidiary or plant that located in Dispersed Employee different counties from headquarters and has 15% or more employees relative to total corporate employees 64 REFERENCES 65 REFERENCES Abatzoglou, J.T. and Williams, A.P., 2016. Impact of anthropogenic climate change on wildfire across western US forests. Proceedings of the National Academy of Sciences, 113(42), pp.11770-11775. Addoum, J.M., Ng, D.T. and Ortiz-Bobea, A., 2020. Temperature shocks and establishment sales. The Review of Financial Studies, 33(3), pp.1331-1366. Aguilera, R., Corringham, T., Gershunov, A. and Benmarhnia, T., 2021. Wildfire smoke impacts respiratory health more than fine particles from other sources: observational evidence from Southern California. Nature Communications, 12(1), pp.1-8. Akey, P. and Appel, I., 2021. The limits of limited liability: Evidence from industrial pollution. The Journal of Finance, 76(1), pp.5-55. Aragón, F.M., Miranda, J.J. and Oliva, P., 2017. Particulate matter and labor supply: The role of caregiving and non-linearities. Journal of Environmental Economics and Management, 86, pp.295-309. Barney, J., 1991. Firm resources and sustained competitive advantage. Journal of Management, 17(1), pp.99-120. Barrot, J.N. and Sauvagnat, J., 2016. Input specificity and the propagation of idiosyncratic shocks in production networks. The Quarterly Journal of Economics, 131(3), pp.1543-1592. Beitler, J. 2006. Tracking Nature’s Contribution to Pollution. NASA Earth Observatory. NASA, 17 Oct. Available at http://earthobservatory.nasa.gov/Features/ContributionPollution/ Bhojraj, S., Sengupta, P. and Zhang, S., 2017. Takeover defenses: Entrenchment and efficiency. Journal of Accounting and Economics, 63(1), pp.142-160. Blatter, M., Muehlemann, S. and Schenker, S., 2012. The costs of hiring skilled workers. European Economic Review, 56(1), pp.20-35. Bolton, P. and Kacperczyk, M., 2020. Do investors care about carbon risk? (No. w26968). National Bureau of Economic Research. Borgschulte, M., Molitor, D. and Zou, E., 2020. Air pollution and the labor market: Evidence from wildfire smoke. Working Paper. Carozzi, F. and Roth, S., 2020. Dirty density: Air quality and the density of American cities. CDP, 2020.Anheuser Busch InBev - Climate Change. Available at https://www.ab- inbev.com/content/dam/universaltemplate/ab-inbev/sustainability/home- 66 page/AB%20InBev%20CDP%20Climate%20Response%202020.pdf (Accessed: 11 November 2021). Chang, T., Graff-Zivin, J., Gross, T. and Neidell, M., 2016. Particulate pollution and the productivity of pear packers. American Economic Journal: Economic Policy, 8(3), pp.141- 69. Chang, T.Y., Graff-Zivin, J., Gross, T. and Neidell, M., 2019. The effect of pollution on worker productivity: evidence from call center workers in China. American Economic Journal: Applied Economics, 11(1), pp.151-72. Chen, F., Peng, X. and Zeng, J., 2017. Does Severe Air Pollution Affect Audit Judgment?: Evidence From China. Evidence From China. Available at SSRN 1538658. Cheng, B., Ioannou, I. and Serafeim, G., 2014. Corporate social responsibility and access to finance. Strategic management journal, 35(1), pp.1-23. Coppola, M., Krick, T. and Blohmke, J., 2019. Feeling the heat. Companies are under pressure to act on climate change and need to do more. The Deloitte sustainability services. Crook, T.R., Todd, S.Y., Combs, J.G., Woehr, D.J. and Ketchen Jr, D.J., 2011. Does human capital matter? A meta-analysis of the relationship between human capital and firm performance. Journal of Applied Psychology, 96(3), p.443. Currie, J. and Neidell, M., 2005. Air pollution and infant health: what can we learn from California's recent experience?. The Quarterly Journal of Economics, 120(3), pp.1003-1030. Dell, M., Jones, B.F. and Olken, B.A., 2012. Temperature shocks and economic growth: Evidence from the last half century. American Economic Journal: Macroeconomics, 4(3), pp.66-95. Dell, M., Jones, B.F. and Olken, B.A., 2014. What do we learn from the weather? The new climate-economy literature. Journal of Economic Literature, 52(3), pp.740-98. Deryugina, T., Heutel, G., Miller, N.H., Molitor, D. and Reif, J., 2019. The mortality and medical costs of air pollution: Evidence from changes in wind direction. American Economic Review, 109(12), pp.4178-4219. Dhaliwal, D.S., Li, O.Z., Tsang, A. and Yang, Y.G., 2011. Voluntary nonfinancial disclosure and the cost of equity capital: The initiation of corporate social responsibility reporting. The accounting review, 86(1), pp.59-100. Downar, B., Ernstberger, J., Reichelstein, S., Schwenen, S. and Zaklan, A., 2021. The impact of carbon disclosure mandates on emissions and financial operating performance. Review of Accounting Studies, 26(3), pp.1137-1175. Eccles, R.G., Serafeim, G. and Krzus, M.P., 2011. Market interest in nonfinancial information. Journal of Applied Corporate Finance, 23(4), pp.113-127. 67 EPA, 2021. EPA Challenge Seeks Innovative Solutions for Reducing Indoor Air Pollution from Wildfires. Available at https://www.epa.gov/newsreleases/epa-challenge-seeks-innovative- solutions-reducing-indoor-air-pollution-wildfires (Accessed: 20 March 2021) Farber, H.S. and Hallock, K.F., 2009. The changing relationship between job loss announcements and stock prices: 1970–1999. Labour Economics, 16(1), pp.1-11. Fisher, A.C., Hanemann, W.M., Roberts, M.J. and Schlenker, W., 2012. The economic impacts of climate change: evidence from agricultural output and random fluctuations in weather: comment. American Economic Review, 102(7), pp.3749-60. Fowlie, M., 2010. Emissions trading, electricity restructuring, and investment in pollution abatement. American Economic Review, 100(3), pp.837-69. Frésard, L., Hoberg, G. and Phillips, G.M., 2020. Innovation activities and integration through vertical acquisitions. The Review of Financial Studies, 33(7), pp.2937-2976. Gao, P., 2008. Keynesian beauty contest, accounting disclosure, and market efficiency. Journal of Accounting Research, 46(4), pp.785-807. Gard, H., and Garrett, M., 2021 Wildfire smoke from the West's massive blazes stretches all the way to the East Coast. CNN, 20 July. Available at https://www.cnn.com/2021/07/20/weather/wildfire-smoke-haze-visible-across-us/index.html (Accessed: 23 July 2021). Ghaly, M., Anh Dang, V. and Stathopoulos, K., 2017. Cash holdings and labor heterogeneity: the role of skilled labor. The Review of Financial Studies, 30(10), pp.3636-3668. Graff-Zivin, J. and Neidell, M., 2012. The impact of pollution on worker productivity. American Economic Review, 102(7), pp.3652-73. Graff-Zivin, J. and Neidell, M., 2014. Temperature and the allocation of time: Implications for climate change. Journal of Labor Economics, 32(1), pp.1-26. Griffin, P., Jiang, Y. and Sun, E., 2020. Do Firms Experiencing More Wildfires Disclose More Wildfire Information in Their 10-Ks?. Working Paper. Hanna, R. and Oliva, P., 2015. The effect of pollution on labor supply: Evidence from a natural experiment in Mexico City. Journal of Public Economics, 122, pp.68-79. Henderson, J.V. and Ono, Y., 2008. Where do manufacturing firms locate their headquarters?. Journal of Urban Economics, 63(2), pp.431-450. Hill, C.W. and Snell, S.A., 1988. External control, corporate strategy, and firm performance in research‐intensive industries. Strategic Management Journal, 9(6), pp.577-590. Hrdinka, T., Novický, O., Hanslík, E. and Rieder, M., 2012. Possible impacts of floods and droughts on water quality. Journal of Hydro-environment Research, 6(2), pp.145-150. 68 Hsiang, S.M. and Jina, A.S., 2014. The causal effect of environmental catastrophe on long-run economic growth: Evidence from 6,700 cyclones (No. w20352). National Bureau of Economic Research. Huynh, T. and Xia, Y., 2021. Panic Selling When Disaster Strikes: Evidence in the Bond and Stock Markets. Management Science, Forthcoming. Ilhan, E., Sautner, Z. and Vilkov, G., 2021. Carbon tail risk. The Review of Financial Studies, 34(3), pp.1540-1571. Jin, Z., Li, F.W., Lin, Y. and Zhang, Z., 2021. Do firms adapt to climate change? Evidence from establishment-level data. SSRN. Jones, B.F. and Olken, B.A., 2010. Climate shocks and exports. American Economic Review, 100(2), pp.454-59. Kor, Y.Y., 2006. Direct and interaction effects of top management team and board compositions on R&D investment strategy. Strategic Management Journal, 27(11), pp.1081-1099. Krueger, P., Sautner, Z. and Starks, L.T., 2020. The importance of climate risks for institutional investors. The Review of Financial Studies, 33(3), pp.1067-1111. La Nauze, A. and Severnini, E.R., 2021. Air Pollution and Adult Cognition: Evidence from Brain Training (No. w28785). National Bureau of Economic Research. Lavy, V., Ebenstein, A. and Roth, S., 2014. The impact of short term exposure to ambient air pollution on cognitive performance and human capital formation (No. w20648). National Bureau of Economic Research. Lengnick-Hall, C.A., 1992. Innovation and competitive advantage: What we know and what we need to learn. Journal of Management, 18(2), pp.399-429. Martinich, J. and Crimmins, A., 2019. Climate damages and adaptation potential across diverse sectors of the United States. Nature Climate Change, 9(5), pp.397-404. Miller, J.D. and Hutchins, M., 2017. The impacts of urbanisation and climate change on urban flooding and urban water quality: A review of the evidence concerning the United Kingdom. Journal of Hydrology: Regional Studies, 12, pp.345-362. Miller, N., Molitor, D. and Zou, E., 2017. Blowing Smoke: Health Impacts of Wildfire Plume Dynamics. Working paper. Moliterno, T.P. and Wiersema, M.F., 2007. Firm performance, rent appropriation, and the strategic resource divestment capability. Strategic Management Journal, 28(11), pp.1065- 1087. 69 National Interagency Coordination Center, 2020. Total Wildland Fires and Acres (1983-2020). Available at https://www.nifc.gov/fire-information/statistics/wildfires (Accessed: 23 March 2021). National Park Service, 2018. Wildfire Causes and Evaluations. Available at https://www.nps.gov/articles/wildfire-causes-and-evaluation.htm (Accessed: 23 March 2021). O’Dell, K., Bilsback, K., Ford, B., Martenies, S.E., Magzamen, S., Fischer, E.V. and Pierce, J.R., 2021. Estimated mortality and morbidity attributable to smoke plumes in the United States: Not just a western US problem. GeoHealth, 5(9), p.e2021GH000457. Pankratz, N., and Schiller, C.M., 2020. Climate change and adaptation in global supply-chain networks. SSRN. Pankratz, N., Bauer, R. and Derwall, J., 2021. Climate change, firm performance, and investor surprises. Working Paper. Peteraf, M.A., 1993. The cornerstones of competitive advantage: a resource‐based view. Strategic Management Journal, 14(3), pp.179-191. Pirinsky, C. and Wang, Q., 2006. Does corporate headquarters location matter for stock returns?. The Journal of Finance, 61(4), pp.1991-2015. Pope, C.A., 2000. Epidemiology of fine particulate air pollution and human health: biologic mechanisms and who's at risk?. Environmental Health Perspectives, 108(suppl 4), pp.713- 723. Saldanha, A., 2021. Dangerous Air: We Mapped The Rise InWildfire Smoke Across America. Here’s How We Did It. Available at https://insideclimatenews.org/news/28092021/dangerous-air-data-map/ (Accessed: 26 November 2021) Sautner, Z., van Lent, L., Vilkov, G. and Zhang, R., 2020. Firm-level climate change exposure. Seltzer, L., Starks, L.T. and Zhu, Q., 2020. Climate regulatory risks and corporate bonds. SSRN. Short, K.C., 2017. Spatial wildfire occurrence data for the United States, 1992-2015. Sørensen, M., Autrup, H., Hertel, O., Wallin, H., Knudsen, L.E. and Loft, S., 2003. Personal exposure to PM2. 5 and biomarkers of DNA damage. Cancer Epidemiology and Prevention Biomarkers, 12(3), pp.191-196. Srinivasan, R., Lilien, G.L. and Rangaswamy, A., 2002. Technological opportunism and radical technology adoption: An application to e-business. Journal of Marketing, 66(3), pp.47-60. Sussman, F.G. and Freed, J.R., 2008. Adapting to climate change: A business approach (p. 41). Arlington, VA: Pew Center on Global Climate Change. 70 United Nations. 2020. Staggering’ rise in climate emergencies in last 20 years, new disaster research shows. Available at https://news.un.org/en/story/2020/10/1075142 (Accessed: 15 April 2021). Wang, D., Guan, D., Zhu, S., Mac Kinnon, M., Geng, G., Zhang, Q., Zheng, H., Lei, T., Shao, S., Gong, P. and Davis, S.J., 2021. Economic footprint of California wildfires in 2018. Nature Sustainability, 4(3), pp.252-260. Wettstein, Z.S., Hoshiko, S., Fahimi, J., Harrison, R.J., Cascio, W.E. and Rappold, A.G., 2018. Cardiovascular and cerebrovascular emergency department visits associated with wildfire smoke exposure in California in 2015. Journal of the American Heart Association, 7(8), p.e007492. Zingales, L., 2000. In search of new foundations. The Journal of Finance, 55(4), pp.1623-1653. Zou, E., 2021. Unwatched Pollution: The Effect of Intermittent Monitoring on Air Quality. American Economic Review, 111 (7), pp.2101-26. 71