Impact of Environmental Disaster Movies on Corporate Environmental and Financial Performance

Using a unique United States box office data set, we investigate the impact of environmental sentiment on corporate environmental and financial performance of the United States listed firms. The influence of mass media on public and investor sentiments is well documented in the existing literature. However, little is known about the effect of movies, although they may influence the public more than other mass media because people, regardless of age and gender, enjoy watching movies. Using the event study methodology and multivariable regression analysis, we show that the release of anthropogenic environmental disaster movie(s) creates environmental sentiment and influences corporate behaviors. Specifically, firms significantly increase their environmental performance in the subsequent year of strong environmental sentiment after the release of environmental movies. More importantly, the positive effect of corporate environmental performance on financial performance is stronger when the environmental sentiment is stronger.


Introduction
People of all ages, genders, and cultures enjoy watching movies. If a person decides to watch a movie, one must stay in a dark and quiet space (theater) for at least an hour and focus on the movie alone. According to the 2016 Theatrical Market Statistics Report by the Motion Picture Association of America, the global box office is growing annually, and in 2016, about 71% of the population of the United States (US) and Canada-approximately 246 million people-visited a movie theater at least once [1]. A 2002 parliamentary report on the British film industry by the United Kingdom (UK) House of Commons Digital, Culture, Media, and Sport Committee stated that approximately 20% of visitors in 2001 traveled to the UK because of the country's portrayal in films. Furthermore, for every dollar spent on films, the flow-on benefit to the economy is estimated to be $1.50 [2].
In India, the former Prime Minister, Nehru, asserted that the influence of films is greater than that of newspapers and books combined. He further mentioned that given movies often reflect the current society and social problems, they are a powerful vehicle for not only culture and leisure but also education and propaganda. To avoid confusion, movies and films are used interchangeably in this study. Further, given the availability and reliability issues, only movies that were released at the box office are considered.
Although movies may strongly influence people and society both economically and psychologically, empirical evidence on the social and economic impact of movies is scarce, especially in the finance literature. By contrast, studies on the role of various types of mass media such as newspaper articles and news broadcasts have been conducted extensively ever since the rise of mass media in the late 20th century. The development of mass media has significantly increased access to any new information, thereby, reducing the informational friction among people. However, people can also be easily influenced by mass media, including even fake news occasionally.
In the finance literature, numerous studies examine the relationship between media and the stock market and emphasize the importance of mass media and news coverage [3][4][5][6][7][8][9]. These studies assume that media coverage is an exogenous event, investigating the causal effect of media coverage on stock or fund prices. Meanwhile, Ahern and Sosyura [10] argue that firms may manage the mass media to influence their stock prices before and during critical corporate events, while Solomon [11] and Cahan et al. [12] demonstrate that firms intentionally manipulate their media coverage.
In addition to finance, the role of mass media is examined empirically in social and environmental literature. Hilgartner and Bosk [13] contend that mass media is one of the key "public arenas" in which social problems are framed and where they grow. Similarly, Anderson [14] argues that the news media plays a crucial role in promoting public and political attention to environmental issues. Boykoff and Boykoff [15] demonstrate that journalistic norms are shaped by politics and influential newspaper and television sources in the US misrepresented the popular scientific perspective on climate, thereby creating an information bias regarding anthropogenic climate change. Similar studies have been conducted in other countries such as Japan and China. Mikami et al. [16] examine how mass media in Japan influenced the public awareness of global environmental issues during and before the United Nations Conference on Environment and Development (UNCED), also known as the Earth Summit, in 1992, while Xu et al. [17] focus on how media coverage plays an important role in the relationship between environmental violation events and shareholder's wealth in China. In addition, Olsen, Carstensen, and Hoyen [18], and Brown and Minty [19] show that media coverage of environmental disasters has a dramatic impact on humanitarian assistance such as donations.
The effect of mass media in many different areas has also been studied extensively. Murray [20] highlights that how media reports on mass killings inspire future killers. Eggermont [21] and Vandenbosch and Eggermont [22,23] illustrate that exposure to a certain message through media such as television programs, fashion magazines, and social networking sites significantly influences adolescents' development. Wakefield, Loken, and Hornik [24] find that mass media campaigns can alter health-related behavior positively across large populations. Cohen [25] and McCombs and Shaw [26] demonstrate the power of mass media in setting the "agenda" and the audience's perception of a given issue. Andreyenkov et al. [27] and Robinson et al. [28] study the relationship between news media and adolescents' opinions about nuclear issues, arguing that students learn more from newspapers than from television news. The impact of mass and social media is also studied in politics [29][30][31][32][33].
Prior studies, regardless of their research objectives, consistently point out the importance of mass media and media coverage through news broadcasts and newspaper articles and concur with mass media's substantial influence on public sentiment. By contrast, the impact of movies at the box office, an important form of mass media, has not been examined in-depth previously. Movies may have comparative advantages over television news and newspapers because movies attract and encompass all age-groups, whereas television news and newspaper articles focus on a particular age group. Moreover, the number of people regularly watching television news has declined [30] while the movie industry is steadily growing. Considering the significant influence of other mass media types on public sentiment, popular movies with a certain message may also affect and mold the sentiment on a particular subject, which, in this study, is the environment.
Environmental protection from both individual and corporate perspectives has been emphasized in this era because global warming and climate change are the biggest problems that humanity is currently facing. For example, Cai and Ye [34] and Halliru et al. [35] examine the relationship between economic growth and environmental protection, and possible factors in the relationship such as environmental regulation and human capital. In addition, many studies in various fields emphasize the importance of environmental responsibility or sustainability [36][37][38][39].
This study sheds new light on the unexplored impacts of movies on society and corporations, thereby contributing to the existing literature on mass media. Furthermore, we suggest a new factor that might affect public sentiment toward the environment and corporate environmental performance (CEP). In this study, we focus on the impact of anthropogenic environmental disaster movies, which are, by definition, environmental disasters triggered by human actions (e.g., global warming and climate change). The list of anthropogenic environmental disaster movies considered for this study is presented in the Appendix A. One of the sample movies is "The Day After Tomorrow", a 2004 American science fiction disaster movie depicting the catastrophic climate change to a new ice age caused by global warming. The movie was a success at the US and international box offices, ranking sixth at the US box office in 2004 with more than 30 million people watching it in the US alone.
Natural disasters, not caused by humans, or disasters not related to the environment have not been considered in this study. For example, the natural disaster movie "San Andreas", which depicts devastation in Los Angeles and the San Francisco area due to earthquakes in the San Andreas Fault, are not included. No clear evidence exists of an earthquake being triggered by global warming or human actions. Further, apocalyptic disaster films such as "Resident Evil" and "World War Z" are not counted as anthropogenic environmental disaster movies in this study because these movies are too unrealistic, although the disasters portrayed in the movies resulted from human actions. Economic disaster movies such as "The Big Short", "Inside Job", and "Too Big to Fail" are not included as well because these movies are not related to environment, but rather they are related to economy.
In addition, documentary films about environmental issues are included in the sample only if they were released at the box office. If not screened in theaters, then documentary films are not considered as anthropogenic environmental disaster movies in this study because it is hard to estimate their performance and to predict their impacts on public sentiment. For brevity, the term "environmental movies" used henceforth in this study implies anthropogenic environmental disaster movies.
The remainder of the study is organized as follows. Section 2 reviews the related literature, outlines the empirical implications, and develops the main hypotheses. Section 3 describes the data sources and research methods. Section 4 discusses the empirical results, and Section 5 summarizes and concludes the paper.

The Role of Media Coverage in Public Sentiment
The crucial role of mass media in conveying information to the public and the financial market is well documented. Klibanoff, Lamont, and Wizman [3] test if salient countryspecific news affects the reaction of closed-end country fund prices to asset value and find that prices respond slowly or quickly depending on news coverage. For example, prices react much faster when the news appears on the front page of The New York Times. Tetlock [4] utilizes daily linguistic content from The Wall Street Journal to construct a measure of media content that corresponds to either negative investor sentiment or risk aversion. Further, he studies the interactions between the media content and stock market activity and finds that high media pessimism puts downward pressure on stock prices, and unusually high or low pessimism results in high trading volume. However, the results seem to be temporary, and pessimistic media content does not convey negative fundamental information but a noise that affects the behavior of individual investors. Continuing the linguistic analysis of Tetlock [4], Tetlock et al. [5] extend the analysis by adding a continuous intraday news source, the Dow Jones News Service, and find that the fraction of negative words used in the news stories does predict low future earnings and stock returns. In other words, linguistic media content reflects hard-to-quantify aspects of firm fundamentals, which investors quickly capture, and the predictability of earnings and return is the strongest for news stories that focus on the fundamentals. In addition, Fang and Peress [6] highlight the role of mass media in alleviating informational friction because of the ubiquity of mass media. They study the relationship between media coverage and expected stock returns and find that investors and stock prices are affected even by fake news. Particularly, they observe the "no-media premium," in which stocks with no media coverage earn higher returns than stocks with high media coverage after controlling for various risk factors.
Meanwhile, Engelberg and Parsons [7] propose the problem with a causal assumption on media reports and disentangle the causal impact of media reports from those of the events being reported. To disentangle the causal impact of media reports, they use different media coverage of the same information events and compare investors' behavior. They find that local media coverage is strongly related to local trading volume for earnings announcements of the Standard & Poor's 500 index companies. Griffin, Hirschey, and Kelly [8] expand the literature and test the relationship between financial media and stock prices in 56 countries. The authors document that firms in the most developed markets experience greater fluctuations in their stock prices when news about them is public. By contrast, stock return volatilities of firms in emerging markets are not significantly different on news and non-news days. They also test several hypotheses for their findings and suggest that cross-country differences in stock price reactions are best explained by insider trading.
Studies on the impact of mass media have also been conducted in the research stream on social issues. Hilgartner and Bosk [13] propose a "public arenas model," suggesting that mass media is one of the key public arenas in which social problems are framed and where they grow and sometimes fall. Similarly, Anderson [14] illustrates that the media has played a key role in shaping public perceptions and policy agendas on climate change. Boykoff and Boykoff [15] argue that mass media coverage of climate change is a social interaction between scientists, policy actors, and the public. They argue that the misrepresentation of the scientific perspective on climate change by influential newspaper and television sources in the US perpetrated information bias on anthropogenic climate change.
Outside the US, Mikami et al. [16] investigate how television news and newspapers in Japan influenced the public awareness of the global environmental problems during and before the Earth Summit (UNCED) in 1992. They find gradual-rather than immediate-and cumulative effects of media coverage on public awareness. They also find that the amount of television viewing has a positive association with public awareness [16]. Xu et al. [17] find that corporate environmental violations with high levels of media attention results in greater losses in the Chinese stock market. Interestingly, they only find a weak impact and insignificant results in their previous study [40], in which they do not control for the level of media coverage for each environmental violation event.
Olsen, Carstensen, and Hoyen [18] study the volume of emergency donations attracted by a humanitarian crisis, such as the Indian cyclone of October 1999 and the Mozambique floods of late-January 2000. They argue that the intensity of media coverage, the degree of political interest of donor governments, and the strength of humanitarian non-government and international organizations present in the country ravaged by the crisis are the three main determinants of the amount of assistance rendered. In the same vein, Brown and Minty [19] use Internet donations data after the 2004 tsunami in the Indian Ocean to study the impact of media coverage on charitable donations. They find that an additional minute in the nightly news and an additional story in major newspapers increase the amount of donations substantially even after controlling for various factors and using the instrumental variable approach. They also highlight that tax incentives play a role in raising charitable donations.
Unlike previous studies, Ahern and Sosyura [10] highlight a reverse causality issue of media coverage on important corporate events such as a merger negotiation. They argue that firms do have an incentive to manage media coverage to manipulate stock prices. In their study, bidders in stock mergers are likely to create more news immediately after the start of negotiations, optimizing the stock exchange ratio and the takeover price. Solomon [11] investigates how investor relation (IR) firms affect the relationship between media coverage and stock prices of their client firms. He finds that IR firms can manage the news about their client firms, generating more media coverage of good rather than bad corporate news [11]. In other words, media coverage is manipulated by IR firms to make them more favorable for their client firms. Thus, the previous two papers provide an interesting argument that media coverage can be managed by firms for their private benefits such as desirable stock prices. More recently, Solomon, Soltes, and Sosyura [41] examine the effect of media coverage of mutual fund holdings on investors' asset allocation. They show that winner stocks covered by major national newspapers in the US, including The Wall Street Journal, The New York Times, The Washington Post, and USA Today, attract more capital inflows compared to winner stocks without media coverage. Specifically, media coverage encourages investors to chase past returns rather than facilitate the processing of useful information in fund portfolios. This is consistent with the notion that media coverage strongly influences investor biases and sentiment. Cahan et al. [12] investigate whether firms that act in more socially responsible ways receive more favorable media coverage. As expected, they find that more socially responsible firms receive more favorable news reportage and have a more positive media image. However, at the same time, these firms are likely to exploit the strong relationship between corporate social responsibility (CSR) and media favorability when they have incentives to improve their media image.
Although the causality issue of media coverage remains unsolved, none of the previous studies deny the substantial impact of mass media on people. Besides, the extant research on mass media implies that the influential media or news often represents public sentiment. Therefore, it could be the same for popular movies/films.

The Effects and Drivers of CEP
Given the acceleration in climate change and the increasing attention on environmental issues across the world, the role of not only an individual but also corporations in society and the environment comes to the fore and is currently emphasized. Prior studies on corporate environmental activities can be divided into two types, namely, those on the causes/drivers of CEP and those on their results/outcomes. In this research stream, the signs of the relationship and causal direction between CEP and financial performance remain controversial. For example, some studies on the effects of CEP argue that improvement in environmental performance may require a significant amount of money and time, and thus, is a cost burden on firms, whereas others insist that the investment in environmental performance can have bottom-line benefits exceeding the costs in the long-run. Nevertheless, according to the recent studies, the latter view (i.e., the positive impact view) that the benefits of environmental performance outweigh the costs, thereby resulting in better financial performance, is gaining more support because of the aggravation of the global warming challenge.
At the early stage of the research stream on the effect of CEP on financial performance, Spicer [42], and Mahapatra [43] examined the relationship between corporate performance on pollution control and financial indicators, although they presented conflicting results. Spicer [42] finds that companies in the pulp and paper industry with better pollution control records tend to have higher profitability, lower total risk, lower systematic risk, and higher price/earnings ratios than companies with poorer pollution control records. By contrast, using larger sample size and longer time horizon across six different industries, Mahapatra [43] shows that pollution control expenditures are a drain on resources that could have been invested profitably, and do not reward the companies for their environmentally responsible behavior. More recently, Klassen and McLaughlin [44], Cohen, Fenn, and Naimon [45], and Xu et al. [17] examine the economic consequences of media disclosure related to CEP. All of them support the positive impact view, finding that good environmental performance predicts better financial performance and higher stock returns, whereas negative environmental events such as environmental violations cause significant drops in stock prices and financial performance.
The other side of the research stream on CEP includes the study of Dalhammar, Kogg, and Mont [46], who identify internal and external factors that may act as drivers of and/or barriers to the development of green products. They suggest that customers and government legislation are the main actors in environmentally-friendly products besides the chief executive officers (CEOs) and competitors. Cronqvist and Yu [47] provide an interesting study on the role of CEO characteristics in CSR, a more comprehensive concept than CEP. They find that if a CEO has a daughter, then the company has about 9.1% higher CSR rating than the median firm. Even after controlling for several endogenous sources, the results are robust and the strongest for the responsibility/performance on the categories of diversity, environment, and employee relations. Dummett [48] also discusses drivers and barriers of corporate environmental responsibility (CER) by conducting faceto-face interviews with 25 senior business leaders from major Australian and international companies. He enumerates potential drivers of CEP from prior studies, such as government legislation, pressure from consumers, and societal expectations, and concludes that the threat of legislation is found to be the primary driver of CER. Surprisingly, he finds that business leaders want national governments to intervene more actively to encourage and even force higher CER. In other words, many corporations are still reluctant to voluntarily engage in environmental activities although CER is also an important aspect of corporate policy. Therefore, as shown in Dalhammar, Kogg, and Mont [46] and Dummett [48], firms need "stimuli" to encourage improvement in their environmental performance, rather than just the expectation of better financial outcomes.
In this paper, we argue that movies about anthropogenic (human-made) environmental disasters may also be a "stimulus" for CEP because if these movies are successful at the box office, they would strongly influence the public and investor sentiments, resulting in upward pressure on firms' environmental performance.

The Impact of Disasters
Furthermore, the effect of disasters, including economic and political crises and natural catastrophes, on people and economy is examined to some extent. For example, Berkman, Jacobsen, and Lee [49] investigate the relationship between political crises and stock returns internationally using the International Crisis Behavior (ICB) project database. They show that an increase in the average number of international political crises per month leads to a significant impact on world market volatility, and also an economically large and negative impact on stock returns.
Meanwhile, Gourio [50] presents a theoretical model with business cycle and economic disaster risk. He also tests his model with actual data and finds that an increase in a risk of economic disaster such as the Great Depression leads to a decline of economic outputs and an increase in risky asset prices. Chiu et al. [51] examine the impact of investor sentiment on equity liquidity and trading behavior during the subprime financial crisis in 2008. They show that pessimistic sentiment caused by the financial crisis increased the quoted bid-ask spread, lead to the evaporation of equity liquidity, and eventually intensified the net-selling pressure during the period. Similarly, Abdelhédi-Zouch, Abbes, and Boujelbéne [52] and Ryu, Ryu, and Yang [53] highlight the dominating effect of the subprime financial crisis on investor sentiment.
Gao, Liu, and Shi [54] examine the degree of people's risk awareness after experiencing catastrophic disasters in Japan. They find that people become relatively insensitive (sensitive) to risk when they experience disasters that have lower (higher) fatalities than expected. Moreover, Ding et al. [55] investigate the impact of coronavirus disease 2019 (COVID-19) on the reaction of stock returns across 61 economies. They document that the impact may vary based on corporate characteristics such as cash holdings, corporate leverage, existence of the global supply chain, profitability, CSR activities, and corporate governance, but the impact is still substantial to every economy.
Likewise, we believe that environmental disasters depicted in the sample movies can significantly influence people and environmental sentiment, especially when they are successful at the box office and grasp people's attention, even though most of environmental disasters in the movies are unreal or not occurred yet.

Hypotheses Development
Previous studies on mass media consistently show that mass media has a significant impact on the public, and even shapes public perception [13][14][15][16]18,19]. However, they all focus on narrow forms of mass media such as news broadcasts and newspapers. Thus, little is known about the movie industry, although it is also an important type of mass media. In this study, we provide new insights into the role of movies beyond just offering leisure. Among various types and genres of movies, we focus on anthropogenic environmental disaster movies because they are socially reflective and realistic. Scientists concur that global warming is a grave concern and its serious consequences are imminent. Building upon the previous studies on the role of mass media and public sentiment, we formulate the first hypothesis: Hypothesis 1 (H1). Environmental sentiment, measured by the box office performance of anthropogenic environmental disaster movie(s), is likely to increase CEP.
As stated in this hypothesis, the box office performance of the movie reflects the level of contemporary environmental sentiment. That is, an environmental movie's success at the box office implies a high level of environmental sentiment. Therefore, we include several characteristics of movies, including box office performance, production budget, and the number of theaters that screen the movies, as proxies of environmental sentiment. For robustness tests of the relationship, we use alternative measures for CEP such as carbon dioxide (CO 2 ) emissions, greenhouse gas (GHG) emissions, and environmental costs of a firm provided by the Trucost database.
Moreover, we test the relationship between CEP and financial performance as in previous literature but mitigate the previous reverse causality issue by introducing the release of environmental movies as an exogenous shock to all the US firms. Therefore, building upon the results of recent studies on CEP and financial performance, we formulate the second and the most important hypothesis of this study: Hypothesis 2 (H2). With a high level of environmental sentiment in the public, measured by the box office performance of anthropogenic environmental disaster movies, firms with better environmental performance have better financial performance in the subsequent year.
Besides the amplifying effect of the environmental sentiment on corporate financial performance, a high level of environmental sentiment might play an important role in reducing the risk associated with firms regarded as environmentally responsible. Testing this notion, we formulate the last hypothesis:

Hypothesis 3 (H3).
With a high level of environmental sentiment in the public, firms with better environmental performance experience lower risk in the stock market in the subsequent year.
Finally, we perform several robustness tests to further support these hypotheses. For H 1 , we employ the two-stage least squares (2SLS) regression method using an instrumental variable to mitigate possible endogeneity concerns and use alternative movie variables related to the movie performance. For H 2 , we select the industries that are related and not related to the environment and conduct a subsample analysis to verify the effect of the environmental sentiment on the CEP-financial performance relationship. For H 3 also, we conduct a subsample analysis after dividing environmentally related and unrelated industries and investigate the changes in institutional stock ownership per the level of the environmental sentiment and CEP.

Data Description
The data related to anthropogenic environmental disaster movies are collected from several sources. The main data for this study are sourced from a subscription-based database called Internet Movie Database Pro (IMDbPro), which provides detailed information on movies such as individual movie financials, and daily, weekly, monthly, and annual box office statistics. We collected weekly, monthly, and annual gross profits, the number of tickets sold, production budget, running time, number of released weeks, number of theaters that screened the movies, and other relevant information for each environmental movie. We also collected annual gross revenue and the total number of tickets sold in the US box office for the sample period. We then used other websites that provide information on the movie industry and box offices, such as The Numbers and Box Office Mojo, to reconfirm the data collected from IMDbPro and reconstruct the missing information, wherever applicable. Moreover, we utilized The Numbers website to count the number of environmental or disaster movies each year, using "environment", "pollution", "global warming", "climate change", or "disaster" as a keyword, and checked if there is any particular pattern or season in the number of those movies.
In addition to the US annual box office data, we used Google Trends to collect the state-level data on the environmental movies for the robustness tests. Google Trends provides the data on the Google search trends of a certain keyword, and a user can choose a specific time period such as a certain day or year, and even see the relative search frequency between 0 and 100 by country or sub-region (i.e., state). Therefore, we gathered the relative search frequency of 50 states in the US on each of the sample environmental movies in the release year, and scaled the frequency from 0 to 1 to use as the state-level measure of the environmental sentiment. Unfortunately, Google Trends provides the search data after 2004, so the sample period for the additional analysis with the state-level data is from 2004 to 2016.
The CEP data used here is collected from the MSCI ESG KLD STATS (a. k. a. KLD) database. The KLD database contains annual ratings for seven major categories: community, corporate governance, diversity, employee relations, environment, human rights, and product quality and safety. Each category provides positive (i.e., strength) and negative (i.e., concern) indicators. The number of indicators change almost annually and are different across the categories. If a firm does a good deed (harm), then it is listed as a strength (concern) indicator and is assigned a value of 1, and 0 otherwise. We only use the environment category for this study, and the raw CER score is calculated by subtracting the total number of concerns from the total number of strengths. The total number of strengths (concerns) is calculated by summing up all the strength (concern) indicator variables in a given year. A higher CER score indicates better environmental performance.
However, according to Manescu [56], this simple summation approach is not appropriate to compare scores across years because, as mentioned above, the number of strength and concern indicators varies considerably each year. To overcome this issue, we follow Deng, Kang, and Low [57] and calculate the adjusted CER measure by dividing the strength and concern scores by the respective number of strength and concern indicators to derive adjusted strength and concern scores, and then take the difference between the adjusted total strength score and the adjusted total concern score. Both the raw and adjusted CER scores are used in this study. Although the results using the raw CER score are qualitatively similar to those using the adjusted CER score, some argue that both raw and adjusted CER scores vary annually.
To alleviate this issue, we also implement alternative measures of CEP from the Trucost database. Trucost is a part of S&P Global and provides carbon and environmental data on 15,000 companies globally. More specifically, Trucost collects environmental performance data and disclosure metrics such as CO 2 and GHG emissions, water use, and waste disposal from publicly available sources, including each company's financial statements, 10-K reports, SEC filings, and sustainability reports. Using these metrics and estimated economic damage, Trucost calculates direct and indirect environmental costs associated with each firm. Direct environmental costs measure estimated damage caused by a firm's direct operations on six parameters: GHG, water, waste, land and water pollutants, air pollutants, and natural resource use. Indirect environmental costs are also a consequence of a firm's activities and are calculated based on the same set of six parameters, but occur at sources owned by other firms. To calculate the indirect costs, Trucost uses its own methodology to estimate the effects, ranging from the first-tier upstream supply chain (direct suppliers) to the last one. We add direct and indirect CO 2 emissions, GHG emissions, and environmental costs to create total emissions and costs, which provide alternative measures of CEP in this study. Similarly to using the data from Google Trends, the timeframe changes to the period between 2002 and 2015 when using environmental costs data from the Trucost database because it is only available from 2002 to 2015.
Following the previous literature, Compustat, ExecuComp, Thomson Reuters, and CRSP databases are also used for variables related to each firm's financials and stock market information, and other variables that might affect CEP. We collect the financial and accounting data, including firm size (total assets), leverage ratio (debt to equity), investment opportunity (Tobin's Q), cash flow, capital expenditure, and return on assets (ROA) from the Compustat database. CEO stock ownership and the CEO's position on the board of directors are collected from the ExecuComp database. Institutional ownership data of the sample firms are collected from Thomson Reuters. The data related to the stock market such as a firm's market value is collected from the CRSP database. Lastly, the data on annual climate conditions are obtained from the National Centers for Environmental Information (NCEI) of the National Oceanic and Atmospheric Administration and the World Bank. To minimize the effects of outliers from the data, following Harford, Mansi, and Maxwell [58], we winsorize all the variables at the 1% level on either tail. The major sample covers all the listed firms in the US from 1992 to 2016. The sample period is determined by the data availability of CER scores from the KLD database.

Empirical Methodology
We implement several empirical methods to test the arguments and hypotheses in this study. First, we conduct an event study on the release of the sample environmental movie, Erin Brockovich, to examine the investor sentiment and corporate outcomes in the stock market. Daily abnormal stock returns (AR) and accumulative abnormal stock returns (CAR) are calculated in an event window around the box office release of the movie, Erin Brockovich.
The AR is calculated as in Equation (1).
where R i,t andR i,t are the actual daily return and expected daily return of stock i on day t. The expected daily returnR i,t is calculated in three ways as Equations (2)-(4).
where R m,t is the market return, SMB t is a small minus big factor, and HML t is a high minus low factor from Fama-French 3-factor model on day t. The parameters in Equations (2)-(4) are estimated over a period of 221 days starting from one year (251 days) prior to the release of corporate prosecution up to 30 days prior to the press release.
In addition, CAR is calculated as in Equation (5).
where t 1 and t 2 indicate the beginning and the end of the event window. Then, we perform univariate tests to alleviate the possible endogeneity concerns, and test the major hypotheses in this study using various regression models. The basic models are presented below: H3 : CEP is measured by the raw and adjusted CER scores from MSCI KLD Stats for firm i for year t. For the robustness check, we also implement environmental variables from the Trucost database. To measure the level of environmental sentiment, we try various measures related to environmental disaster movies, ranging from the box office sales to the movie rating, but we use the ratio of box office sales to the total population in the US as the main variable of the environmental sentiment. For example, Annual Top 20 is a binary variable that is assigned a value of 1 if at least one environmental disaster movie is ranked as one of the top 20 movies of the year, ENV. Movie Performance represents the ratio of the population that watched the movie and is calculated by dividing the total number of the environmental movie tickets sold at the box office by the total US population in the release year, and ENV. Movie Number is the total number of environmental disaster movie(s) in a given year. We also include control variables that may affect the main independent variables following the previous studies. When testing the first hypothesis, control variables include ln (total assets), leverage ratio, Tobin's Q, cash flow normalized by total assets, capital expenditures normalized by total assets, institutional ownership, CEO equity ownership, and CEO duality (if CEO is also the chairperson of the board), which were factors of CEP in the previous literature [47,59,60]. When testing the second hypothesis, we used three different variables to measure corporate financial performance; ROA, operating cash flow, and Tobin's Q. For ROA and operating cash flow, control variables are ln (total assets), leverage ratio, capital expenditures normalized by total assets, institutional ownership and CEO equity ownership, following Jo, Kim, and Park [39] and Lin and Fu [61]. For Tobin's Q, control variables are ln (sales), ROA, leverage ratio, R&D expenditures normalized by total assets, ln (firm age), and CEO equity ownership following Yermack [62] and Daines [63]. Lastly, when testing the third hypothesis, we follow Pan, Wang, and Weisbach [64] and Dhaliwal et al. [65] to include ln (market equity value), ln (firm age), cash flow ratio, market beta, market to book ratio, market leverage, ROA, R&D expenditures, property, equipment, and plant, and dividend payout ratio as control variables for firm risk. More details on variable definitions are described in Appendix A.
In order to conduct the analysis at the state-level, we use Google Trends to record the relative search frequency, 0 to 100, of the environmental disaster movie by states in a release year because the state-level box office data are not available. Then, we divided the search frequency by 100, scaling down to the value to between 0 and 1.
In the regression analyses, we use the lagged terms of movie variables for several reasons. Firstly, it may take several months for enough people to watch the movie and influence the public sentiment on environmental problems. Secondly, it would also take a significant amount of time for firms to respond to environmental sentiment and improve their environmental performance. Lastly, most of the environmental movies in this study are released after June, as shown in the Appendix A. If the quarterly data on CEP were available, it would be possible to analyze the effect of the release of environmental movies more accurately. However, as the environmental data from either KLD Stats or Trucost is only provided annually and environmental sentiment may not act on the firm immediately, it is more reasonable to use the subsequent year's environmental performance rather than the given year's performance. Table A1 in the Appendix A shows the summary statistics for all the variables used in the regression analysis. As shown, the 25th percentile, median, mean, and 75th percentile values of both raw and adjusted CER scores are all zero, implying that environmental performance of most of the sample firms are only mediocre. According to the list of environmental movies presented in the Appendix A, environmental movies are screened in 9 of the 25 years of the sample period from 1992 to 2016, but the mean value of environmental movie dummy is 0.464, indicating that about half the sample observations are from years with the environmental movie(s). However, the mean value of the annual top 20 dummy is much smaller, 0.214, because it only takes a value of 1 if an environmental movie was successful at the box office and ranked among the top 20 movies in the release year. In other words, the annual top 20 dummy is much more conservative than the regular environmental movie dummy. In addition, the environmental movie performance variable is calculated as the ratio of the total number of tickets sold for the environmental movie(s) to the total US population in the release year. We can guess that about 10% of the total population watched the sample environmental movie, on average, because environmental movies were released during half the sample period (9 out of 25 years) and the mean environmental movie performance is around 3% of the total US population. Summary statistics for other variables in Table 1 are not extraordinary and similar to the values in prior studies. On average, institutions hold about 65% of a firm's shares, which is comparable to the mean value, 50-60%, from previous studies [66,67]. Further, a CEO owns about 2.5% of a firm's equity on average. CEO duality is a dummy variable that takes a value of 1 if the CEO is also the chairperson of the firm's board of directors and 0 otherwise. In more than 75% of the sample firms, the CEO also holds the position of the chairperson of the board of directors, implying poor corporate governance and powerful CEOs. Next, we check the correlations among the variables of interest. Table 2 shows the correlation matrix of the variables in the regression analysis. As evident, the dependent variables do not have a strong correlation with the main independent variables or control variables (at most −0.2181 between ROA and leverage ratio). Table 1 shows summary statistics of the main dependent and independent variables used in the study with the mean, median, standard deviation, and 25% and 75% percentile values for the entire firm-year observations over the 1992-2016 period in the US. The main dependent variables, CEP (ENV. score and adjusted ENV. score), are created using the environmental ratings from the MSCI ESG KLD database. Alternative measures of environmental performance, environmental costs, CO 2 emissions, and GHG emissions are from Trucost, and the sample period for these variables are from 2002 to 2015. The units for environmental costs are in dollar million, while those for CO 2 and GHG emissions are in million tons. Other variables related to CEO characteristics and firm characteristics are from Compustat, CRSP, Execucomp, and Thomson Reuters databases. All the variables are winsorized at the 1% level on either tail.  Table 2 shows the correlation matrix for the main dependent and independent variables in this study. Raw Environmental Score, Adj. Environmental Score, ROA, Operating Cash Flow/Total Assets, and Idiosyncratic Stock Return Volatility adjusted by the Fama-French 3-factor model are the main dependent variables used in this study. Other variables are the main independent or control variables used in the study. Institutional Ownership is also used as a dependent variable in the regression analysis to check the channels of the environmental movie(s) that affect the relationship between corporate environmental and financial performance.

Event Study on Pacific Gas and Electric Company (PG&E)
Before analyzing the effect of environmental movies using multivariable regression models, we first test if an environmental movie impacts the stock market through an event study on the sample movie, Erin Brockovich. Whereas most of the sample movies in the study are not confined to a particular firm, Erin Brockovich is an environmental movie about the contamination of drinking water with hexavalent chromium by a single firm, PG&E, in the southern California town of Hinkley in 1993. The movie was released on Friday, 17 March 2000, seven years after the actual incident. Therefore, the effect of media coverage on the incident by other mass media types such as news broadcasts and newspapers may not coincide with the impact of the movie. Table 3 shows the results of the event study of PG&E Corporation around the release of the movie, Erin Brockovich. The stock return of PG&E Corporation is adjusted with the risk-free rate, CAPM, or Fama-French 3 (FF3)factor model. Because the movie was premiered on Friday, 17 March 2000, the event day is set to Monday, 20 March 2000, as it would take some time for the movie to influence investor sentiment and for the stock market to reflect information content of the movie. The estimation window for the analysis is from a year to a month before the event day. Table 3 also shows the long-term (1, 2, and 5 years) buy-and-hold abnormal returns of PG&E stock after the release of the movie, Erin Brockovich. The stock return of PG&E Corporation is adjusted with the risk-free rate or FF3-factor model.
As illustrated in Table 3, PG&E experienced significantly negative abnormal stock returns (between 5% and 10%) around the movie release date even after adjusting the returns with Capital Asset Pricing Model (CAPM) and the Fama-French 3-factor model (FF3 Factors). The event study results indicate that the movie does have an impact on investors and the stock market. To check for the long-term effect of the movie, we also examine the stock performance of PG&E in the long-run. The 1-year, 2-year, and 5-year buy-and-hold abnormal stock returns are also presented in Table 3. The buy-and-hold abnormal returns of PG&E stock for one year and two years are negative, whereas the abnormal return becomes positive after five years, suggesting that holding the stock for one or two years after the movie release incurs negative abnormal returns, but not after five years. There might be other events and factors that contribute to the negative abnormal returns during the study period, but it is the first step toward understanding the impact.  Note: The numbers in parentheses are t-values, and *, **, and *** indicate statistical significance at the 10%, 5%, and 1% levels, respectively.

Endogeneity Concerns and Univariate Tests
We also address possible endogeneity concerns with the release of the environmental movie(s). One may argue that environmental movies are produced because of the preexisting environmental sentiment in a particular year, and at the same time, companies also improve their environmental performance as a response to the pre-existing environmental sentiment. However, this concern is not that serious because it takes a significant amount of time to produce a movie and it is very hard to predict the production period.
Yet, one may still claim that production companies of environmental movies wait for the best time for the release after they finish production, deciding to release the environmental movies in the year with high environmental sentiment. In this case, it is not the environmental movie that raises the environmental sentiment, affecting CEP, but is the pre-existing environmental sentiment that affects both the release of the environmental movie and CEP. To check and mitigate this concern, we first count the total number of environmental or disaster movies each year and see if there is any interesting pattern and correlation to the release of the sample environmental movie. As shown in Appendix A, the total number of environmental or disaster movies does not show any specific pattern, and the sample movies do not seem to be released in accordance with the release of other environmental or disaster movies.
Then, we conduct univariate tests to examine if abnormal environmental or climate conditions are associated with the release of environmental movies because abnormal environmental conditions such as natural disaster and abnormal temperature could be associated with pre-existing environmental sentiment. More specifically, we use CO 2 emissions per capita, average annual or monthly abnormal temperatures, and the total number and cost of natural disasters for the abnormal climate conditions. Table 4 presents the results for the univariate tests, indicating that abnormal climate conditions are not significantly different in the years (months) with environmental movies and the years (months) without environmental movies. Therefore, the univariate tests somewhat relieve the possible endogeneity concern that production companies release the environmental movie(s) when the pre-existing environmental sentiment is high. However, endogeneity concerns are not completely resolved, so we include variables related to abnormal climate conditions as the additional control variables, and also employ the 2SLS method. Table 4 shows the univariate test results for the relationship between the release of environmental disaster movies and environmental problems in the US. It also reports t-test statistics for differences in means and Wilcoxon rank-sum (Mann-Whitney) test statistics for differences in medians between sample years (months) with environmental disaster movies and sample years (months) without environmental disaster movies. The values in the parentheses are t-statistics for means and Wilcoxon z-values for medians.

Environmental Sentiment and CEP
After the event study and univariate tests, we conduct multivariable regression analysis to examine the major hypotheses. Table 5 shows the basic ordinary least squares regression results, testing the first hypothesis. Three main independent variables are related to an environmental movie: Annual Top 20, ENV. Movie Number, and ENV. Movie Performance. The annual top 20 variable, Annual Top 20, is a dummy variable that takes a value of 1 if at least one environmental movie in a given year is released and is also ranked among the top 20 movies of the year at the US box office. As mentioned in the earlier section, this variable is much more conservative than the regular environmental dummy variable because it also reflects the performance of the movie. The environmental movie number variable, ENV. Movie Number, is the total number of environmental movies at the box office in a given year. Lastly, the environmental movie performance variable, ENV. Movie Performance, is measured by the ratio of the total number of tickets sold (i.e., the total number of people that watched) for the environmental movie to the total US population in the release year of the movie. Among these three movie variables, Annual Top 20 and ENV. Movie Performance are the main movie variables. We do not use a regular environmental movie dummy because good movie performance at the box office is essential to represent the public sentiment but the regular dummy variable does not reflect the performance at all. The rest of the variables are control variables. We also add firm fixed effects to control for firm-specific variations. When industry fixed effects are included instead of firm fixed effects, the results are not only the same but also stronger. The major limitation of this analysis is that movie variables are annual and have the same values throughout the year. Therefore, we cannot control for year fixed effects because the movie variables possess perfect-collinearity problems with the year fixed effects. Instead, we try to control for time trends using variables related to annual climate conditions, as used in the previous univariate tests. We also cluster standard errors at a firm level.  Note: *, **, and *** indicate statistical significance at the 10%, 5%, and 1% levels, respectively. The numbers in parentheses are t-values.
The coefficients of all the environmental movie variables in Table 5 are consistent with the first hypothesis and significantly positive at the 1% level in most cases (15% in the fifth column), indicating that the level of environmental sentiment, measured by the performance of the sample environmental movie(s), is positively associated with CEP. According to the results, if an environmental movie is ranked as one of the top 20 movies of the year, the raw (adjusted) CER score improves by 0.069 (0.012), which is about two-thirds (one and a half times) of the mean (adjusted) CER score in the sample. In addition, if one more percent of the US population watches the environmental movie, the raw (adjusted) CER score improves by 0.011 (0.003), which is about one-ninth (one-third) of the mean (adjusted) CER score. The impacts are not only statistically significant but also meaningful in magnitude. Interpreting the coefficients of other control variables, the Tobin's Q value is negatively associated with CEP in the subsequent year, implying that firms with high growth opportunities are likely to invest in areas other than the environment. Highly significant and negative coefficients of the capital expenditure ratio indicate that firms that employ more resources on capital expenditures have fewer resources to invest in environmental performance. By contrast, firms that hold more cash can afford to invest in environmental performance in the subsequent year. Similar to Graves and Waddock [68], firms with high institutional stock holdings are more likely to invest in environmental performance as institutional shareholders put pressure on the management to engage in social responsibility. In addition, a CEO needs higher equity ownership, implying stronger power within the corporation, to pursue improvement in CEP. Lastly, it is expected that abnormal environmental conditions would encourage environmental awareness, and in turn, CEP. However, all the annual climate variables-average natural disaster costs, CO 2 emissions per capita, and abnormal temperature-show negative associations with CEP. These coefficients refute the notion that firms invest in and improve their environmental performance in response to abnormal environmental conditions. Table 6 shows multivariable regression results of adjusted CER strengths and concerns on the independent variables. Adj. CER strength (concern) is the number of good (bad) environmental indicators divided by the total number of indicators. As described in the data description part, the raw or adjusted CER scores can be subdivided into strengths and concerns. The raw CER strengths (concerns) indicate the number of good deeds (wrongdoings) of a firm among the list of CER indicators. Alleviating the issue of annual variations in indicators, the adjusted CER strengths (concerns) are calculated by dividing the raw CER strengths (concerns) by the total number of strength (concern) indicators in a given year. Therefore, we subdivide the adjusted CER score into strength and concern parts and conduct the same regression analysis to examine the effect of environmental sentiment on CEP in more detail. The first three columns in Table 6 show the results on the adjusted CER strengths and the last three columns present the adjusted CER concerns. Surprisingly, environmental sentiment positively influences both adjusted CER strengths and concerns, implying that firms increase both good and bad environmental activities with a higher level of environmental sentiment. However, the magnitude and significance of the effect are much stronger (at least three times) for the CER strengths than for the CER concerns. Therefore, environmental sentiment increases the overall CER score.  Note: *, **, and *** indicate statistical significance at the 10%, 5%, and 1% levels, respectively. The numbers in parentheses are t-values.
In addition to the univariate tests, we implement the 2SLS regression method to further address the possible endogeneity issue. As an instrumental variable for the 2SLS regression analysis, we use a logarithmic value of the annual box office total profits, which would be related to the two main movie variables, Annual Top 20 and ENV. Movie Performance, but is unlikely to be associated with CEP. The performance of the environmental movie depends on the current status of the movie industry or box office, and the movie industry or box office has no relation with CEP. In the first stage, we regress the main environmental movie variable, either Annual Top 20 or ENV. Movie Performance, on the instrument variable, the logarithm of the annual total profits of the box office, and all the exogenous variables in the second-stage regression, in which, we regress CEP on the predicted values of environmental movie variables from the first stage regression and other control variables in the regression model. In both the first and second-stage regressions, we control for the firm fixed effects and cluster the standard errors at the firm level. Columns (1) and (3) in Table 7 show the first-stage regression results and high F-statistic values, 233.17 and 655.79. They test the statistical significance of the instrument and indicate that it rejects the null hypothesis that the logarithm of the annual total profits of the box office is a weak instrument. Columns (2) and (4) document the main and second-stage regression results, which still support the first hypothesis. Even after controlling for the possible endogeneity concern using the 2SLS method, the level of the environmental sentiment is still positively associated with CEP.  Note: *, **, and *** indicate statistical significance at the 10%, 5%, and 1% levels, respectively. The numbers in parentheses are t-values.
Additionally, we use alternative measures of CEP because the raw and adjusted CER scores do not vary much during the sample period as shown in Table 1. Corporate environmental costs and emission variables from Trucost are implemented in Table 8, and the results indicate that a high level of environmental sentiment significantly reduces the total CO 2 and GHG emissions and the total environmental costs, which is consistent with the first hypothesis and the results in Table 5. This time, we only include two main movie variables, Annual Top 20 and ENV. Movie Performance, which reflect both existence and performance of the sample environmental movie because the results are the same with other movie variables, including ENV. Movie Number and the regular environmental movie dummy. Table 8 presents multivariable regression results of environmental costs, CO 2 emissions, and GHG emissions on the independent variables. Annual Top 20 is a dummy variable that takes a value of 1 if the movie is ranked among the top 20 in a given year. ENV. Movie Performance measures the environmental movie's influence on the public, proxied by the ratio of the number of tickets sold for the environmental movie to the total population in the movie release year. Control variables include ln (Total Assets), leverage ratio, Tobin's Q, Cash Flow/Total Assets, Capital Expenditure/Total Assets, ROA, Cash Holding Ratio, Institutional Ownership, CEO Equity Ownership, CEO Duality, and variables related to annual environmental conditions such as annual natural disaster costs, CO 2 emissions per capita, and average abnormal temperature. For the regression analysis, we control for the firm fixed effect and cluster standard errors at the firm level. Note: *** indicates statistical significance at the 1% level. The numbers in parentheses are t-values. All the variables are winsorized at the 1% level on either tail.

Environmental Sentiment on Environmental and Financial Performance
Furthermore, we examine the relationship between CEP and financial performance as in the previous literature by introducing the level of environmental sentiment. To investigate the associations between three variables, we first divide the sample firms into two types, namely, environmentally responsible and irresponsible firms. Environmentally responsible firms are those with either the raw environmental scores or adjusted environmental scores above the industry median scores in a given year, and irresponsible firms are those below the industry median values. Next, we create a binary variable, High CER Firm Dummy, which takes a value of one for environmentally responsible firms and zero for irresponsible firms, and multiply it with ENV. Movie Performance, the proxy for the level of environmental sentiment, to generate the interaction term between two. In other words, we examine the effect of environmental sentiment on environmentally responsible firms' financial performance.
As High CER Firm Dummy has perfect-collinearity problems with firm fixed effects and ENV. Movie Performance has the same issues with year fixed effects, we switch those variables with firm or year fixed effects back and forth when performing the analysis in Table 9, and include industry fixed effects when we cannot control for firm fixed effects. The results of testing the second hypothesis are reported in Table 9. Columns (1), (4), and (7) in Table 9 document the regression results only with industry fixed effects, while columns (2), (5), and (8) report the results with firm fixed effects, instead of High CER Firm Dummy. Highly significant and positive coefficients of High CER Firm Dummy in columns (1), (4), and (7) indicate that environmentally responsible firms exhibit better financial performance, measured by ROA, operating cash flow ratio, and Tobin's Q, in the subsequent year compared to irresponsible firms. This is consistent with the prior studies that argued a positive association between CEP and financial performance [17,39,42,44,45]. In columns (3), (6), and (9), both firm fixed and year fixed effects are included, instead of High CER Firm Dummy and ENV. Movie Performance.
Thus, the results in these columns are the most conservative ones. The interaction term between ENV. Movie Performance and High CER Firm Dummy is the main variable of interest in this analysis. The coefficient estimates of the interaction term between ENV. Movie Performance and High CER Firm Dummy in most of the columns are significantly positive (weakly significant only in the first column), at least at the 10% level. These positive coefficients indicate that environmentally responsible firms with higher level of environmental sentiment have better financial performance, supporting the second hypothesis. Examining the economic significance of the coefficient estimates in columns (3) and (6), a 1% increase in the US population that watched the environmental movie is associated with 0.030% (0.040%) growth in ROA (operating cash flow ratio), which is about 0.7% (0.5%) of the mean values. Note: *, **, and *** indicate statistical significance at the 10%, 5%, and 1% levels, respectively.
Moreover, we subdivide the sample and analyze the same regression model to validate the environmental movie performance as a measure of the environmental sentiment and verify the impact of the environmental sentiment. We argue that if the environmental movie performance indeed measures the environmental sentiment and influences the corporate environmental-financial performance link, this environmental sentiment should have a greater impact on firms related to the environment than those that are unrelated. Since firms in polluting industries are the ones that contribute to the environmental destruction and global warming the most, people and the public with environmental awareness would pay more attention to those firms than firms in other industries. Similarly, Jo and Park [69] show that CSR initiatives of firms in controversial industries such as alcohol, gambling, tobacco, and firearms have greater risk-decreasing effect, thereby extending their "social license to operate." Therefore, we select industries closely related to the environment (e.g., polluting industries) and those that seem unrelated (e.g., non-polluting industries) following Becker and Henderson [70] and employing the Trucost data. Becker and Henderson [70] divide polluting industries-which emit volatile organic compounds and nitrogen oxides, resulting in Ozone depletion-and non-polluting industries based on the publications and documents of the US Environmental Protection Agency. In addition, we rank Fama and French's 48 industries in order of direct environmental costs using the Trucost database and select the top (bottom) five industries that have the highest (lowest) direct first-tier environmental costs. The results were the same when using top (bottom) ten industries with the highest (lowest) direct environmental costs. We check if any of the top (bottom) five industries are listed among the non-polluting (polluting) industries from Becker and Henderson [70], and as expected, find no such case.
Finally, we define environment-related (non-related) industries as polluting (nonpolluting) industries from Becker and Henderson [70] or industries with high (low) direct environmental costs in this study. Environment-related industries (i.e., polluting industries) include the chemical industry (14th in Fama-French industry classification), rubber, and plastic product industry (15th in the classification), construction material industry (17th in the classification), other industries related to metal mining and works (19th, 27th, and 28th in the classification, respectively), coal and petroleum and natural gas (29th and 30th in the classification, respectively), and utilities (31st in the classification). More specifically, industrial organic chemical companies (SIC from 2860 to 2869), miscellaneous plastic companies (SIC from 3080 to 3089), and forestry companies (SIC from 0800 to 0899) from Becker and Henderson [70] are included in the chemical, rubber, and plastic product and construction material industries, respectively. We also include polluting industries from Becker and Henderson [70] that have not been included using the environmental costs data. By contrast, non-polluting industries are recreation, entertainment, printing and publishing, medical equipment, personal services, measuring and control equipment, computers, and insurance (6th, 7th, 8th, 12th, 33rd, 35th, 37th, and 45th in the classification, respectively) plus the apparel industry, mattresses, and certain leather products (specifically SIC from 231 to 236, 2515, and 315 to 317, respectively). Table 10 reports the results of the subsample analysis based on the two industry types. The first four columns are about the sample firms in the industries that are non-polluting and unrelated to the environment, while the last four are about the firms in polluting industries. I use two measures of corporate financial performance, ROA, and Tobin's Q. As expected, the coefficients of the interaction terms of firms in industries unrelated to the environment are not significant at all, whereas those of firms in polluting industries are significantly positive at the 10% level for ROA and 5% level for Tobin's Q. These results imply that the environmental sentiment works together with good environmental performance of firms in polluting industries on financial performance whereas the environmental sentiment does not have a significant impact on financial performance of firms in nonpolluting industries with good environmental performance. In other words, people and the public value good environmental performance of firms in polluting industries more than that of firms in non-polluting industries, resulting in better financial performance. Therefore, the results based on the industry types in Table 10 confirm that the performance of environmental disaster movie(s) acts as a good proxy for the environmental sentiment and that the environmental sentiment works as an amplifier in the CEP and financial performance relationship. Note: *, **, and *** indicate statistical significance at the 10%, 5%, and 1% levels, respectively. The numbers in parentheses are t-values.

Environmental Sentiment and Firm Risk
In addition to the financial performance of environmentally responsible firms, we also investigate whether environmental sentiment exerts an influence on firm risk in the stock market. More specifically, we examine the effect of environmental sentiment on the volatility of daily stock returns of environmentally responsible firms. With regard to the regression model, firm risk, measured by the realized (idiosyncratic) stock return volatility in the stock market, becomes the main dependent variable instead of corporate financial performance. The results are reported in Table 11. Again, the main variable of interest is the interaction term between ENV. Movie Performance and High CER Firm Dummy. The dependent variable, SVOL, in columns (1) and (2) is the standard deviation of daily stock returns in the excess of the risk-free rate. In columns (3), (4), (5), and (6), we use standard deviations of "adjusted" daily stock returns, IDVOL CAPM or IDVOL FF3 , which are the residuals from regressing daily stock returns with CAPM or the Fama-French 3-factor model, respectively. The results in columns (2), (4), and (6) are the most conservative ones, controlling for both year and firm fixed effects. The coefficients of the interaction terms are all statistically significant and negative at the 1% level, meaning that environmentally responsible firms experience lower risk in the stock market when the environmental sentiment is higher. Note: *, **, and *** indicate statistical significance at the 10%, 5%, and 1% levels, respectively.

The State-Level Environmental Sentiment
For the robustness tests and another way to address the possible endogeneity concerns, we conduct the state-level analysis using the data collected from Google Trends, which provides the state-level search frequency of a certain keyword. We use the state-level search frequency of the title of the sample environmental disaster movie(s) in the release year as the state-level environmental sentiment because the stronger the movie's influence is, the higher the Google search frequency would be. Table 12 demonstrates the effect of the state-level environmental sentiment on corporate environmental performance, and the results are as strong and significant as those in Table 5. Lagged terms of State-level Google Search is the main independent variable, getting a value between 0 and 1. The value of zero (one) in a given state in the release year indicates that the Google search for the sample environmental movie is the least (most) frequent relative to other states. Firm characteristics, including the firm size, leverage ratio, investment opportunity, cash flow, capital expenditures, cash holding ratio, and institutional ownership, are controlled as in Table 5. The annual abnormal climate conditions, including natural disaster costs, CO 2 emissions per capita and abnormal temperature, are also controlled. In addition, we control for the firm fixed effects and cluster standard errors at a firm level. The raw environmental score, strengths, and concerns are used for the dependent variable in the first, third, and fifth columns, respectively, and adjusted score, strengths, and concerns are used in the second, fourth, and sixth columns.
The significantly positive coefficients of State-level Google Search indicate that firms headquartered in the states with higher environmental sentiment have better environmental performance than firms headquartered in the states with lower environmental sentiment do in the subsequent year of the movie release. We also verify the impact of the state-level environmental sentiment on the relations between CEP and financial performance and between CEP and firm risk. Tables 13 and 14 show the results on those relations, and the results are as significant as the results in Tables 9 and 11. The main variable of interest is the interaction term between State-level Google Search and High CER Firm Dummy. The coefficients of the interaction term indicate that environmentally responsible firms headquartered in the states with higher environmental sentiment exhibit higher ROA, operating cash flow, and Tobin's Q, and lower firm risk than firms headquartered in the states with lower environmental sentiment in the subsequent year of the movie release.  Lastly, we also perform a subsample analysis to further alleviate the endogeneity concerns. Huge production budget, large production company, and aggressive advertisement might be other factors, affecting the performance of the movie, and so the public sentiment. In order to differentiate the pure impact of a movie on the public sentiment from other factors, we focus on the environmental movie with a low production budget and unknown production company that recorded incredible success, which is An Inconvenient Truth among the sample environmental movies. Therefore, we narrow down the sample period between 2005 and 2007 because the movie is released in 2006. Now, Annual Top 20 takes a value of 1 if the movie, An Inconvenient Truth, is ranked in the top 20 in the previous year. ENV. Movie Performance is the ratio of the number of tickets sold for An Inconvenient Truth to the total population in the movie release year.
The subsample analysis results in Tables 15 and 16 are still consistent with the previous results, and major hypotheses. The amazing success of the movie with a low production budget and unknown production company, An Inconvenient Truth, creates strong environmental sentiment, improving corporate environmental performance and amplifying the positive association between CEP and financial performance. Besides, we conduct several additional robustness tests using more movie variables, but the results were the same and even more significant.  Note: *, **, and *** indicate statistical significance at the 10%, 5%, and 1% levels, respectively. The numbers in parentheses are t-values.

Discussion
In this study, we examined the role of environmental movies in CEP and financial performance after controlling for other firm-specific factors. The effects of mass media, including television news, newspapers, and magazines, have been studied extensively across various research subjects. However, the role of a movie/film has been rarely examined in-depth and its economic and social impacts are unknown although it is also an important type of mass media. Therefore, we chose a specific genre of movies, anthropogenic environmental disasters, meaning environmental disasters triggered by humans, and investigated its impacts on the environmental sentiment and corporate behaviors because it is widely accepted that mass media shapes the public sentiment and influences corporate outcomes. In addition, we argued that the level of environmental sentiment would depend on the performance of the environmental movies because more people would be affected by the message inside a movie as more people watch the movie.
Using the event study methodology on the sample environmental disaster movie, Erin Brockovich, we first showed that the investor sentiment was strongly influenced by the release of the movie, and the company associated with the movie, PG&E, experienced negative abnormal stock returns around the release of the movie. Then, we used univariate tests to mitigate the possible endogeneity concerns in this study, and implemented multivariable regression analysis, finding that the environmental sentiment, measured by the box office sales of environmental disaster movies in the US, significantly influences corporate environmental performance. The positive association between the environmental sentiment and CEP is still robust after controlling for the possible endogeneity issue with the 2SLS regression method. For other robustness checks, we used alternative measures of CEP, such as the firm's total environmental costs, CO 2 emissions, and GHG emissions, for the dependent variable, and the results were robust and consistent with the main hypothesis. Therefore, this study contributes to the existing literature on CEP by presenting a new driver of CEP.
More importantly, we documented that environmental sentiment intensifies the positive effect of CEP on financial performance, supporting the prior research on the positive association between corporate environmental and financial performance. We found that as the released environmental movie performs better at the box office (i.e., more people watch the movie), the environmental sentiment becomes stronger and the financial performance of environmentally responsible firms gets better. In addition, the environmental sentiment significantly reduces the risks associated with environmentally responsible firms in the stock market as well. Utilizing the state-level data from Google Trends and using subsample analyses based on the industries and a particular movie, An Inconvenient Truth, we further back-up the arguments in this study. To the best of our knowledge, this study is the first to examine the role of a special type of mass media, movies/films, and like other types of mass media, movies/films also have strong influence on public sentiment and corporate behaviors. Therefore, this study also adds to the existing literature on mass media.
For future research, if data on movie profits from other sources such as video home system (VHS) and digital versatile disc (DVD) can be obtained, a more elaborate and comprehensive analysis on the role of movies can be conducted. Lastly, more and more people are watching movies at home through other media services such as Netflix, an American technology and media services provider, rather than at the theater. If data on environmental movie or documentary film through Netflix can be gathered, it would be interesting to compare the impacts of the movies at the box office with those of the movies through the online platform.

Firm level characteristics (CRSP, Compustat, or ExecuComp)
Cash/Total Assets Cash holding ratio, which is calculated as the sum of cash and short-term equivalents (che) divided by total assets (at) Capital Expenditure/Total Assets Capital expenditure ratio, the ratio of capital expenditures (capx) to total assets (at) Cash Flow/Total Assets Cash flow ratio, calculated as the sum of income before extraordinary items (ibc) and depreciation and amortization (dp) divided by total assets (at) of the previous period

CEO Equity Ownership
The stock equity ownership of the CEO CEO Duality Dummy variable, taking a value of 1 if the CEO is also a chairperson of the board of directors, otherwise 0 Total Assets (at) The total value of a firm's assets (in $ million)

Market Value of Firm
The market value of a firm is calculated by adding the total market value (mkvalt) and total assets (at) and subtracting common/ordinary equity (ceq) and deferred taxes (txdb) Tobin's Q Tobin's Q, calculated as the ratio of the market value of assets to the replacement value of assets (book value of total assets), following Fama and French (1992)

Leverage Ratio
The book leverage ratio, calculated as the sum of long-term debt (dltt) and current liabilities (dlc) divided by total assets (at) ROA Return on assets, calculated as the ratio of operating income before depreciation (ni) to total assets (at)

Operating Cash Flow
Operating cash flow, calculated by subtracting total accruals from net income (total accruals = change in current assets − change in cash equivalent − change in current liabilities + change in debt in current liabilities − depreciation and amortization)

Institutional Ownership
The stock equity ownership of financial institutions  This is the list of environmental disaster movies for the sample period from 1992 to 2016. Environmental disaster movies are the movies or films that show man-made environmental disasters or have messages about environmental problems caused by humans. The premiere date is a release date of a movie at the US box office. The highest rank is the best rank of the movie during the release period at the US box office, and annual rank is the rank based on the number of tickets sold at the US box office in the release year. Gross profit is the movie profit (in dollar million) generated from the box office tickets sold.