Financial Constraints and Corporate Environmental Policies ∗ Pengjie Gao†

Taehyun Kim‡

Qiping Xu§

December 8, 2017

Abstract We show that financial constraints affect corporate environmental policies because of significant abatement costs associated with hazardous waste management. Exploiting two experiments in which a firm’s external financing constraints are likely exogenously impacted, we find that relaxation (or tightening) of external financing constraints reduces (or increases) the firm’s toxics release. The impact of financial constraints on toxics releases is amplified by weaker regulatory monitoring and enforcement, by myopic managers with short horizons, and when firms are under pressure to meet earnings targets. Overall, our evidence highlights the real effects of financial frictions in the form of environmental pollution.

∗

We would like to thank Paulo Fulghieri, Nandini Gupta, Paul Schultz, and seminar participants at the University of Notre Dame, Wabash River Finance Conference for valuable comments and suggestions. We thank Andriy Bodnaruk, Gerald Hoberg, Tim Loughran, Max Maksimovic, and Bill McDonald for making their textual-based financial constraint measures available to us. We also thank David Yeh and Pete Pietraszewski for assistance with the data. We are grateful to Tim Antisdel from EPA for answering us questions about TRI Program. We are responsible for remaining errors. † University of Notre Dame, [email protected] ‡ University of Notre Dame, [email protected] § University of Notre Dame, [email protected]

“As man proceeds toward his announced goal of the conquest of nature, he is writing a depressing record of destruction–destruction of the earth he inhabits and destruction of the life that shares it with him.” Rachel Carson in “Silent Spring” (1962)

1

Introduction

In the modern production process, firms often generate byproducts that have adverse impacts on the environment and public health. In 2015, the United States produced 27.24 billion pounds of toxic chemicals in production-related wastes. In a world without environmental regulations, firms that maximize shareholder values have little incentive to protect the environment. Corporations nowadays are required by laws and regulations to internalize a large part of the social costs of pollution. However, environmental protection is rather expensive. In 2005, the last year with official data, U.S. manufacturing firms spent over 26.57 billion dollars in pollution abatement. This is a sizable number: it is about 1% of the US manufacturing shipment value, or above 20% of the manufacturing sector’s total capital expenditure. In this paper, we examine how financial frictions affect corporate environmental policies. Waste treatment requires substantial inputs of energy, labor, contracted services, and raw materials, and the process is deeply integrated into every aspect of corporate decision-making. The tension between environmental protection and the firm’s bottom line lies precisely in the fact that environmentally friendly waste management methods are costly, while the least costly methods such as direct releases and disposal have undesirable environmental consequences. We hypothesize that when a firm is financially constrained and engages in cost-cutting, environmental protection is often sacrificed for the benefits of shareholders and other financial stakeholders. In other words, financial frictions exert and exacerbate externalities in the form of environmental pollution. Exploring the Toxics Release Inventory (TRI) establishment level micro-data from the Environmental Protection Agency (EPA), we first show economically large and statistically signifi1

cant correlations between the amount of toxics released and a number of off-the-shelf financial constraint measures, including two text-based financial constraint measures recently developed by Bodnaruk, Loughran and McDonald (2015) and Hoberg and Maksimovic (2014). These text-based measures extract qualitative information about financial constraints from corporate disclosure documents for essentially all firms and interpret such vast information using welldefined algorithms. It is comforting to see that the positive correlation survives establishment, industry-year and state-year fixed effects, where industry and state are granularly defined at the establishment level. A one-standard-deviation increase in the financial constraint measure is associated with a 2-3% increase in total toxic release. Our results are robust to alternative toxic release under different EPA regulations. To establish causal impact of a firm’s financial constraints on environmental policies, we first explore flow induced price pressure (FIPP) to generate an exogenous shock. Coval and Stafford (2007) and Lou (2012) show that mutual fund managers proportionally scale up or down existing stock positions when fund investors move capital in and out of funds. Flow induced price pressure generates relatively exogenous valuation changes, because it is associated with who is buying or selling due to liquidity reasons as opposed to what is being bought or sold. Because fund flows are rather persistent, the temporary price pressure lasts for several quarters (i.e., “return continuation”). Over time, as the price pressure dissipates, security prices revert back to its fundamental. The price deviation from fundamentals and corrections offers windows of opportunity for firms to issue securities (Khan, Kogan and Serafeim (2012)), which relaxes firms’ financial constraints and improves corporate environmental policies. On the other hand, large outflows pushes stock prices down and make it more difficult for the firm to access external financing, which results in a negative impact on corporate environmental policies. We construct hypothetical flow induced trades based on mutual funds’ existing portfolio, which further ensure the flows are unrelated to firm fundamentals (Edmans, Goldstein and Jiang (2012)). Our estimates show that a one-standard-deviation change in FIPP leads to toxic release changes by 3% – 3.5% relative to the average annual toxic release per

2

establishment in our sample that experience large inflows or outflows. Our second identification strategy explores the collateral channel of firms’ real estate assets. Real estate presents a major part of tangible assets that firms can pledge as collateral, and an increase in collateral pledgeability reduces external financing friction. We examine its effect on corporate environmental policies. Specifically, our identification strategy uses variations in real estate values of land-holding firms driven by local real estate prices (Chaney, Sraer and Thesmar (2012)). To better deal with the concern that local real estate values might be correlated with some demand side factors that also affect corporate environmental policies, we use the MSA-level housing supply elasticity (Saiz (2010)) interacted with U.S. 30-year mortgage rates as an instrument for the MSA-level housing price index (HPI)(Mian and Sufi (2011)). The intuition is that as housing demand increases due to changes in the mortgage rate, home prices are less sensitive to demand in markets with elastic supply because these areas capitalize demand shocks into quantities rather than prices. Our evidence suggests that higher real estate values lead to lower toxic release. In terms of economic magnitude, a onestandard-deviation increase in real estate values leads to the reduction in total toxic release that is between 3%-4% of the sample mean. Furthermore, the impact of the collateral channel is significantly stronger for firms that are initially more financially constrained. After establishing the causality, we step back and explore a number of cross-sectional variations in the impact of financial constraints on toxic release to shed light on the inner workings of corporate environmental policies. Our cross-sectional analysis suggests that government regulation and oversight does play a rather significant role in reducing pollution. In fact, when it comes to making environmental policy choices, firms are keenly aware of and act accordingly to these regulations. In addition, our evidence highlights that the external pressures of the capital markets also shape corporate environmental policies. Overall, our sample of managers are opportunistic: they strategically choose when and where to pollute. Our first set of cross-sectional tests focus on variations of regulatory and monitoring environments. For example, if a certain geographic region is designated as “nonattainment” by the

3

EPA, environmental laws and regulations mandate enhanced monitoring and costly ramifications. Similarly, facilities that release a large quantity of toxics are often under the scrutiny of government agencies, environmental protection organizations, and news media. We show that corporations indeed pollute less when these mitigating measures are in place. Moreover, we show that when facing financial constraints, firms strategically reduce abatement and pollute more where the external monitoring and regulations are weak. Our second set of cross-sectional tests examine how managerial incentives affects firms’ environmental policies. We first infer managerial decision horizons by observing the portfolio composition of institutional investors, motivated by the strong correlation between these two (see, Bushee (2001), Gaspar, Massa and Matos (2005), among others). Our evidence shows that financial constraints increase toxic release significantly more when the managerial decision horizon is short. Then we examine how toxic release responds to managerial incentives to meet earnings targets. The evidence points to a subset of firms that have tried “everything possible,” including squeezing costs from abatement, emit significantly more toxic releases when earnings fall short of annalists’ forecasts. Our paper contributes to the literature on the real effects of financial frictions. Our evidence complements and extends earlier work that focus on financial frictions and investment activities (see, Baker, Stein and Wurgler (2003), Chaney, Sraer and Thesmar (2012), among others) and employment (Chodorow-Reich (2013)). Our paper shows that managers carefully evaluate costs and benefits associated with the implementation of environmental policies. In other words, a firm’s environmental externalities are affected by financial frictions. Our paper also complements recent work on corporate social responsibilities (CSR). First, the environment represents one of the most important corporate non-financial stakeholders. Investigation of corporate environmental policies is important by itself alone, and contributes to the burgeoning literature on the economics of climate change (see, Stern (2008), for a recent review and synthesis.). Second, focusing on corporate environmental policies allows us access to a set of well-defined and high quality performance metrics in a large cross-section over a

4

long period of time. Finally, we also contribute to the debate about corporations that are “doing good by doing well” versus “doing well by doing good” (Hong, Kubik and Scheinkman (2012), Cohn and Wardlaw (2016), Caskey and Ozel (2017)). Traditional literature, especially the management and organizational behavioral literature, juxtaposes corporate environmental policies and corporate profitability. The idea here is that when corporations act as ‘responsible” partners with the environment and other non-financial stakeholders, corporations’ bottom lines benefit. However, as we argued earlier, implementation of corporate environmental policies is costly. Firms trade off “doing well” vs. “doing good.” When the regulation and external non-stakeholder monitoring is weak, actions of “doing well” seems to dominate the motives of “doing good.” The rest of the paper is organized as follows. Section II describes institutional background, introduces data we use in this study, and provides some summary statistics. Our baseline correlation studies are presented in Section III. Section IV introduces two identification strategies to establish causality of financial constraints on corporate environmental policies. A number of cross-sectional tests are presented in Section V. Section VI concludes.

2 2.1

Data and Summary Statistics Institutional Background

Born in the wake of elevated concern about environmental pollution, the EPA was established in 1970 to consolidate a variety of federal research, monitoring, standard-setting and enforcement activities into one agency. A number of laws serve as the EPA’s foundation for protecting the environment and public health, with a few Presidential Executive Orders (EOs) also playing a central role. Congress authorizes the EPA to develop and enact regulations, as well as explain the critical details necessary to implement environmental laws. Figure 1 Panel A illustrates how different EPA rules regulate end products and by-products from the production 5

process. The EPA works with state and local authorities to ensure the compliance of environmental regulations, enforce penalties, and launch sanctions. If investigations by the EPA and state agencies uncover willful violations, civil or criminal trials and penalties are sought. Figure 1 Panel B presents the so-called “Waste Management Hierarchy,” which intuitively describes EPA’s waste management guidelines. Source reduction refers to the process of maximizing or reducing the use of natural resources at the beginning of an industrial process, thereby eliminating the amount of waste produced by the process. Source reduction is the EPA’s preferred method of waste management. Recycling refers to the series of activities through which discarded materials are collected, sorted, processed, and converted into raw materials and used in the production of new products.1 Energy recovery is the process of generating energy from the combustion of wastes, including at waste-to-energy combustion facilities and landfill-gas-to-energy facilities.2 Treatment means using various processes, such as incineration or oxidation, to alter the properties or composition of hazardous wastes. Direct disposal or other releases is the least preferred methods of waste management. Unfortunately, direct disposal is often the least costly method of waste management from a firm’s perspective. The tension between environmental protection and the firm’s bottom line lies precisely in the fact that preferred waste management methods are costly, while less preferred methods are harmful to the environment. Consider a typical coal-fired plant. To minimize its environmental impact, it can uses higher-grade (or cleaner) coal that reduces pollutants from the very beginning, which will significantly increase production costs.3 A coal-fired plant can even switch to a natural gas-fired plant but this would require a substantial investment. Indianapolis Power and Light (IPL), a subsidiary of American Energy Service (AES), a publicly traded Fortune 500 company, serves as a good example of the high costs of waste management. IPL is currently 1 The definition excludes use of these materials as a fuel substitute or for energy production. (National Recycling Coalition, 1995) 2 It is often associated with electricity generation, although it can also offset fossil fuels used at industrial sites. 3 Coal prices differ by rank and grade. According the US Energy Information Administration, in 2015, the average annual sale price of coal at mines ranged from $14.63 per short ton (2000 pounds) for subbituminous coal, to $97.91 for anthracite coal. Two other types of coal are lignite coal which costs $22.36 per short ton and bituminous coal which costs $51.57. As one may expect, lignite and bituminous coal generate much more SO2 and smog than anthracite coal.

6

ranked as the nation’s number one “super-emitter” of toxic release and greenhouse gases (Public Integrity, 2016).4 In its 10-K filing with the Security and Exchange Commission (SEC) for the fiscal year ending in December, 2016, IPL states the following:

“On July 29, 2015, the IURC [Indiana Utility Regulatory Commission] issued a CPCN [Certificate of Public Convenience and Necessity], granting IPL authority to convert Unit 7 at the Harding Street Station from coal-fired to natural gas-fired at a cost of up to $70.9 million (the IURC later approved IPL’s updated cost estimate for the Harding Street Station refuels including $64.3 million for Unit 7), and also to install and operate wastewater treatment technologies at Harding Street Station and Petersburg Generation Station in southern Indiana at a cost of up to $325.7 million.” (IPL’s 10-K, contents in brackets are added by authors) It is useful to put these numbers into perspective: by the end of fiscal year 2016, which is the latest information available, IPL’s net income from operating activities was merely $324 million. The above example may leave us with the impression that environmental protection is all about fixed-asset investment. However, this could not be further from the truth. The EPA Pollution Abatement Costs and Expenditures (PACE) survey for the manufacturing sector provide a breakdown breakdown of different components of the abatement operating costs and expenditures, which is presented in Panel C of Figure 1. Perhaps contrary to some conventional wisdom, depreciation of investment is not the most significant source of abatement costs. In fact, it counts for only 14% of operating costs. Costs associated with energy, contract work, labor, and materials and supplies make up the vast majority of the costs.5 The arithmetic behind this simple decomposition points out that the impact of financial constraints on environmental policy is not a sideshow of financial constraints on corporate investment policies, which includes investment into pollution abatement but only accounts for less than 5% of total capital expenditure. To the contrary, environmental policies run much deeper into every stage 4

The source of information is available from https://www.publicintegrity.org/2016/09/29/20248/america-ssuper-polluters. 5 The source of data is Pollution Abatement Costs and Expenditure: 2005, published by the US Census Bureau, available from https://www.census.gov/prod/2008pubs/ma200-05.pdf.

7

of operations for modern corporations. Understanding the impact of financial constraints on environmental policies has its own importance, which deserves careful investigation. In some of our empirical tests, we make use of a county’s attainment or nonattainment status as one of our identification strategies. We provide a brief discussion of the designation of such status and its implications for the firm. Under the 1977 amendment for Clean Air Act (CAAA), each year every county is classified by the EPA as attainment or nonattainment of the national standards for criterion pollutants. The threshold for excessive pollution is applied uniformly across the United States.6 In any given year, some counties generate pollution over these thresholds while others do not. Figure 2 presents the latest version (September 2017) of the nonattainment map. EPA has mandatory and discretionary sanctions for out-of-attainment areas. For example, EPA can impose the mandatory sanction of highway funding moratorium through the Federal Highway Administration. Discretionary sanctions mandate that local plants emitting the pollutant to adopt “lowest achievable emission rates” (LAER) technologies, which requires the installation of the cleanest available technologies regardless of costs. Furthermore, if any new plants plan to locate in the nonattainment county, the EPA forces them to offset their releases from another polluting source within the county. In contrast, for areas designated as “attainment,” large polluting plants are only required to use “best available control technology” (BACT), which is significantly less costly than LAER technology. In summary, a nonattainment status results in more stringent regulations to reduce toxic release without regard to cost (Becker and Henderson (2000), Walker (2013)).

2.2

Data

Our main source of data comes from facility-level (i.e., establishment) Toxics Release Inventory (TRI) program administered by the US Environmental Protection Agency (EPA) from 19902014. For any facility located in the US, if it falls within a TRI-reportable industry sector, has ten 6 Ambient air pollution is measured by EPA pollution monitors that take hourly/daily readings, and the choice and management of monitoring location is not subject to local authorities.

8

or more employees, and the amount of manufactured or processed TRI-listed chemicals is above a certain quantity threshold, then it is required to report information about the toxic release. The EPA conducts an extensive data quality analysis after TRI reporting forms are received and provides analytical support for enforcement efforts led by EPA’s Office of Enforcement and Compliance Assurance (OECA). According to the EPA, it first identifies TRI forms containing potential errors, then EPA staff contact the facilities that submitted these reports for inquiries. If errors are found, the facilities then submit a corrected report to the EPA. In addition, the Office of Inspector General is an independent office within the EPA that performs audits, evaluations, and investigations of the Agency and its contractors to prevent and detect fraud, waste, and abuse.7 In our empirical exercises, we mainly focus on the total quantity of toxic release. As robustness check, we also consider toxic release under the Clean Air Act (CAA), the Clean Water Act (CWA), the Comprehensive Environmental Response, Compensation, and Liability Act (CERCLA, also known as the "Superfund"), as well as the Occupational Safety and Health Act (OSHA). In comparison to firm level or industry level data, working with establishment level data allows us to study corporate environmental policies at the finest granularity with a relatively high precision of measurement. We remove the establishment with total toxic release amount below 100 pounds to make sure small polluters are not driving our results. We further remove establishments with fewer than three consecutive observations. We extract facility information from the National Establishment Time-Series (NETS) database produced by Walls & Associates, which is a continuous annual compilation of different vintages of Dun & Bradstreet (D&B) Million Dollar Directory database. The organizational structure of the NETS database shares many similarities with those of the Longitude Business Database (LDB) maintained by the US Census Bureau. The main difference between these two databases is that NETS database has a wider coverage of small business, while LDB database has more in7

Section 325(c) authorizes civil and administrative penalties for noncompliance with TRI reporting requirements. Section 1101 of Title 18 of the United States Code makes it a criminal offense to falsify information given to the United States Government (including intentionally false records maintained for inspection).

9

depth coverage of economic activities. Due to its production lag time, the most recent version of the NETS database – which is the version we use in this study – provides facility information between 1990 and 2014. We draw firm-level accounting information from the COMPUSTAT database and stock market related information from the Center for Research in Security Prices (CRSP) database. We also draw flow induced price pressure from Lou (2012), MSA-level Home Price Index (HPI) from Federal Housing Finance Agency(FHFA), MSA-level supply local housing supply elasticity (Saiz (2010)), and the 30-year US fixed mortgage rate to identify the causal link between firms’ financial constraints and toxic releases. To measure the incentives of meeting earnings targets, we obtain earnings forecasts of sell-side analysts and realized earnings from the Institutional Brokerage Estimation System (I/B/E/S). Institutional holding data are obtained from Thomson Reuters’ 13F filing database. To classify asset owners into different investment and holding horizons, we adopt the classification scheme in Bushee (2001). Linking these databases poses a challenge because there is no common and consistent linking keys that exist among the EPA TRI report, NETS, and COMPUSTAT /CRSP databases. To establish linkage among these databases, we first link the EPA TRI report with NETS database at the facility level, using a link file provided by the EPA with facility-level D&B numbers (also known as "DUNS number"). Our further check reveals that these links are reliable. In the second step, we link the EPA TRI parent company information with COMPUSTAT /CRSP databases using a historical name matching algorithm. It is crucial to use historical name information which is time-varying to plant opening, close, and ownership changes. We obtain the historical company name from CRSP, supplemented by the historical name and address information obtained from the 10K, 10-Q and 8-K filings using the SEC Analytical Package provided by the Wharton Research Data Service (WRDS).

10

2.3

Measures of Financial Constraints

It is notoriously difficult to measure financial constraints (Farre-Mensa and Ljungqvist (2016)). In our empirical analysis, we made a conscious choice and avoid using accounting-based measures of financial constraints because they are highly correlated with sales/production, which is a key factor determining the total toxic release amount. Some other accounting-based measures, such as the SA-index developed by Hadlock and Pierce (2010), incorporates information on firm size. Size correlates with production, which in turn positively correlates with toxic release. Instead we mainly rely on two text-based financial constraint measures recently developed by Hoberg and Maksimovic (2014) and Bodnaruk, Loughran and McDonald (2015). To construct the financial constraint measure, Bodnaruk, Loughran and McDonald (2015) first define a copula of words that describe financial constraints. Equipped with such a dictionary, they search the entire 10-K and use a simple “bag-of-words” approach to delineate the “tone” of management discussion and disclosure. In general, they classify a firm-year as more constrained when the list of “financial constraint” words occur more often. Bodnaruk, Loughran and McDonald (2015) validate their measure by its predictive power of future dividend omission, dividend increase, equity recycling (dividend payout and repurchase of common/preferred stocks), and pension underfunding, which are events generally believed to be syndromes of financial constraint. On the other hand, many accounting-based financial constraint measures do not have any predictive power. Hoberg and Maksimovic (2014) take a different approach. They focus on mandated disclosures regarding each firm’s liquidity, as well as the discussion of the sources of capital each firm intends to use in addressing its financing needs. Based on the disclosures in the Management’s Discussion and Analysis (MD&A) section of the 10-K,8 Hoberg and Maksimovic (2014) evaluate financial constraints by accounting instances when a firm was constrained from raising capital through different markets: equity, debt and private placement. The distinction of 8 Not every firm provides liquidity and capitalization resource subsection in MD&A section. Hoberg and Maksimovic (2014) show that these firms are generally healthy firms that have few liquidity issues to disclose.

11

financing channel is important because it underscores how different financing channels serve different aspects of corporate operation, and how financial constraints may arise for different economic reasons. In our empirical analysis, we focus on the debt-market constraint measure which describes a firm’s intension to issue debt to fund their investment and solve their liquidity problems but potentially face excess liabilities.9 In contrast, equity-market constrained firms mainly face the difficulty of raising equity financing to fund the growth of the firm. In summary, we recognize that none of these financial constraint metrics is perfect and when we interpret our empirical evidence, we shall be cautious. It is our hope that the following set of cross-sectional analyses are able to capture some aspects of financial constraints in an intuitive manner.

2.4

Summary Statistics

Our final sample includes 8,294 establishments that belong to 1,544 US public firms over the sample period between 1990 and 2014. 89,464 establishment-year observations are included. Table 1 presents summary statistics for our sample firm-level (Panel A) and compare it to overall Compustat non-financial firms during the sample period (Panel B). Establishment-level summary statistics of key variables are presented in Panel C. Compared to the overall Compustat universe during the same period, our sample firms tend to be larger in size. Compustat median asset size is $191.27 million while our sample median asset is $838.56 million. Our firms also have more tangible assets (Compustat median tangible-to-assets ratio is 23% while our sample has 29%) and less cash (Compustat median cash to asset ratio is 9% while our sample median is only 4%). Overall, Table 1 shows that our sample overweights the manufacturing sector, as those are the firms included in the TRI program. The average amount of toxic release by establishment differ by program definition. Toxics 9 Specifically, when we compare the sample of firms with EPA TRI reporting to the whole universe of COMPUSTAT firms, we indeed find that our sample of firms with EPA TRI reporting have higher leverage, less cash holding, lower Tobin’s Q, lower capital expenditure, and higher fraction of tangible assets (see, Panel A and Panel B from Table 1). These summary statistics hint that our sample of firms with EPA TRI reporting tilt towards mature firms and fit into what Hoberg and Maksimovic (2014) describe as debt-market constrained firms.

12

release by CERCLA, CAA and CWA includes a somewhat wider spectrum of compounds than the release defined by the Occupational Safety and Health Administration (OSHA). The mean total toxic release per establishment per year is approximately 115 tons, and the average total toxic release per establishment per year based on definition of CAA, CWA and CERCLA range from 72 to 109 tons per year, while OSHA release is about 17.82 tons per year. The amount of release is highly skewed because of some of the so-called “super-polluters.” In our analysis, we use the natural logarithm of the release amount defined by different programs to address the skewness in the data. Figure 3 presents the time-series plot of aggregate toxic release of our sample establishments by year.10 The total amount of toxic release exhibit declining trends over time. This is consistent with findings of earlier literature. Some important factors behind the decline in toxic release include more stringent environmental regulations and the higher pollution tax that firms are paying (Henderson (1996), Shapiro and Walker (2015); Levinson (2009)), migration of heavy polluting industries to other countries (Copeland and Taylor (2004)), and introduction of greener technologies (Levinson (2015)). Because of the above time-series attributes, we include year-related fixed effects in all of our specifications. Table A.2 summarizes total toxic release by Fama-French 48-industry classification. As one may expect, chemical; construction materials; steel; machinery; and auto industries have the largest number of facilities with the EPA TRI’s toxic release, followed by consumer products; food products; rubber and plastic petroleum; and public utility industries. Another feature of the data is that, for certain industries such as precious metal; metal mining; and utilities, the average emission per establishment is much higher than for other industries. These summary 10 There was a major expansion of reporting industries in year 1998. Seven new industry sectors were required to report to TRI, including metal mining, coal mining, electric utilities, chemical wholesale distributors, petroleum bulk storage and terminals, hazardous waste management facilities, and solve recovery facilities. In this figure we removed a few new sectors added to the TRI program in 1998 to keep the number of sectors constant through out the time period. However, in the subsequent empirical analysis, we include those sectors. There are a number of smaller expansions of reporting requirements between 2000 and 2014. However, most of the expansion is related to newly added carcinogen toxics based on the National Toxicology Program (NTP) in their Report on Carcinogens (ROC). Using the gradual introduction of newly discovered carcinogen as events that increase corporate liabilities, Gormley and Matsa (2011) explore managerial responses.

13

statistics highlight the necessity to account for establishment or industry characteristics.

3

Baseline

In this section, we describe our baseline ordinary least square (OLS) regression model that relates firms’ toxic release under different categories to various financial constraint measures. The purpose of the correlation analysis is to establish some empirical regularities and benchmark cases. We leave more formal discussion of identification in the next section of the paper. The baseline regression is as follows: Toxic Releasei,t = α + βFinancial Constraintsc,t−1 + γFirm Controlsc,t−1

(3.1)

+ κEstablishment Controli,t−1 + FEs + εi,t where i indexes each EPA establishment, c indexes each firm (headquarter), t indexes year. Firm controls include Log (Asset), Tobin’s Q, Leverage, Tangibility, Capital Investment, and Cash holding. Establishment controls include Log (sales) and Log(employment) to account for production amount at the establishment level. Table (2) presents regression estimates of financial constraints on the total toxic release (measured by metric ton in logarithm) as our main outcome variable. The financial constraint measure (Text FC) in columns (1) to (3) is the textualbased measures developed in Bodnaruk, Loughran and McDonald (2015), and the measure (HM Debt) in columns (4) to (6) is the debt-market constraint measure taken from Hoberg and Maksimovic (2014). We hypothesize the coefficient β to be positive – the more financially constrained firms are, the more toxics firms are likely to release. All specifications include establishment fixed effects to account for time-invariant unobservable establishment attributes. Standard errors are clustered at the firm level. Specification in column (1) in Table (2) also includes the year fixed effects. The key coefficient of 0.221 implies that one standard deviation (0.20) increase in the text measure of financial constraint (Bodnaruk, Loughran and McDonald (2015)) is associated with an ap14

proximately 0.44 increase in the log number of total toxic release, which corresponds to 2.5% (= 0.221 × 0.20/1.75) of our sample mean in logarithm. Similarly, column (4) of Panel A in Table (2) shows that one standard deviation (0.06) increase in the text measure of debt focus (Hoberg and Maksimovic (2014)) is associated with a 4% increase in the total toxic release, which corresponds to a 2.4% of our sample mean. We include industry-year (Fama-French 48-industry classification) fixed effects in specifications (2) and (5), and state-year fixed effects in specifications (3) and (6) where the state is where the establishment is located at that point in time. This set of fixed effects effectively account for time-varying unobservable industry or state characteristics such as state-level environmental regulations and enforcements. Overall, baseline models with establishment and year fixed effects absorb most of the time-series and cross-sectional variations. As shown by R-squared statistics, adding industry-year fixed effects or state-year fixed effect does not significantly alter the R-squared. These alternative specifications barely change our key parameter estimates. All of the point estimates of the impact of financial constraints on toxic release are of similar economic magnitudes, ranging from 1.8% (column(2)) to 2.2% (column(3))of the sample mean. Table A.3 presents regression estimates of toxic release amount under other EPA categories with similar regression specifications as robustness checks. Across different EPA categories of toxic release emission, the coefficients on our two text-based financial constraint measures are similar. In addition, Appendix Table A.4 presents regression estimates of a dummy variable indicating establishment’s EPA compliance status, which takes the value of one being non-compliant and zero otherwise, on two financial constraint measures. Establishment fixed effects are included in all specifications. Coefficients across all specifications are close to zero and none of them is statistically significant at conventional significance level. We find this non-result actually interesting. One possible interpretation of the result is that while financial constraints adversely affect corporate environmental policies, visibly and publicly violating EPA’s compli-

15

ance status is costly (both in economic terms like fines and penalties, as well as reputational concerns).11 As a consequence, a firm will try to avoid being labeled as non-compliant even if it is financially constrained. This observation highlights the potential shortcomings of using the EPA’s compliant status as a sufficient statistic of corporate environmental policies in our context.

4

Identification Strategies

Section 3 presents a positive correlation between firms’ toxic releases and various financial constraint measures. However, it remains challenging to claim the causal impact of firms’ financial constraint on toxic release. The key concern lies in the omitted variable issues. For example, there might be some firm or industry level unobservables that affect both firms’ financial health and their environmental decisions, which could bias the OLS coefficients in either direction. Alternatively, there could be reverse causality concerns. It is possible that firms with worse environmental consequences can have a higher cost of capital (Chava (2014)). In order to establish the causal link from financial constraint to toxic release decisions, we need to generate an exogenous shock to firms’ financial health, while the shock should be unrelated to firms’ toxic release decisions. In this section, we exploit two experiments to generate this exogenous shock in order to establish the causal link.

4.1

Mutual Fund Flow Induced Price Pressure (FIPP)

A key observation from the mutual fund asset fire sales literature (Coval and Stafford (2007)) is that when investors move capital to or away from mutual funds, their inflows and outflows force mutual fund managers to proportionally scale up or down existing stock positions. Consequently, trades by the mutual fund managers induce price pressure that pushes stock prices up 11

In the full EPA TRI record, only 2.88% of observations have a noncompliance status. The average penalty dollar amount at both the state and federal level is $245 thousand, with the highest record being $40 million. The compliance action costs average as high as $36 million, with the highest being around $5 billion.

16

when there is capital inflow; and pushes stock prices down when there are capital outflows. As the price pressure slowly dissipates afterwards, there is subsequent return reversal. Therefore, the direction of initial capital flow is positively correlated with stock returns in the short-term, but negatively correlated with stock returns in the intermediate-term. A common interpretation of the temporary price pressure is that it represents a source of “non-fundamental” shocks to the firm. A number of recent papers (Khan, Kogan and Serafeim (2012), Edmans, Goldstein and Jiang (2012),Gao and Lou (2013),Hoberg and Maksimovic (2014)) explore the effects of temporary price pressure resulted from mutual fund flow induced trading. One common finding is that positive price pressure induces equity and debt issuance, while negative shocks make security issuance difficult. Thus, flow induced price pressure generates relatively exogenous variations of a firm’s access to external financing due to equity market valuation. Holding demand for external financing constant, for a time-t positive shock of FIPP, this would imply that average firms during the return continuation phase time(t,t+1) face relatively relaxed financial constraints. Exploring the similar setting, we use the extreme flows to generate relatively exogenous variation of prices, and then test the impact of financial constraints on corporate environmental policies. Specifically, we construct a quarterly measure of flow induced price pressure as follows: P F I P P j,t =

i

Shar esi, j,t−1 × Per cF low i,t P i Shar esi, j,t−1

(4.1)

where Shar esi, j,t−1 is the number of shares of stock j held by mutual fund i at the end of previous quarter t − 1, and Per cF low i,t denotes the capital flow to mutual fund i in quarter t as a fraction of its total net assets at the beginning of the quarter. For the normalization purpose, we use total shares outstanding in the denominator. The key feature of the FIPP here is that instead of actual transactions, we use the hypothetical trades based on mutual funds’ previous quarter-end portfolio. The uninformed nature of the trading arises because it captures the change of a mutual fund’s positions that are mechanically induced by fund investor flows, 17

which are very different from the fund manager’s discretionary trading that is potentially driven by information about the firm fundamentals. One may further argue that even the previous quarter-end positions of the fund may contain some fundamental information that is yet to be incorporated into prices. There are two pieces of evidence that are inconsistent with this interpretation. First, if the existing position contains information that is yet to be incorporated into prices, then we expect to find mutual funds on average have strong tendency of short-term performance persistence. Empirical evidence provides little support of pervasive performance persistence. On the contrary, there is evidence of negative performance persistence (Carhart (1997)). Second, the price continuation and price reversal patterns are more consistent with the price pressure rather than slow information diffusion story – in particular, the latter implies no return reversal pattern. We then estimate the following panel regression model, Toxic Releasei, j,t = α + βFIPP j,τ + γFirm Controls j,t−1

(4.2)

+ κEstablishment Controli, j,t−1 + F E + εi, j,t where i indexes each EPA establishment, j indexes each firm, t indexes year, and firm Controls include Log(Assets), Tobin’s Q, Lever age, Tangi bili t y, C api t al I nvest ment and C ashholding. Establishment control includes the natural logarithm of establishment-level sales and employment. The main variable of interest is FIPPi,t , which denotes the average FIPP within a given firm-year. Recall that FIPP occuring contemporaneously or in the immediate past is in the return continuation phase. Therefore, an increase in FIPP in the short-term implies the relative relaxation of financial constraint for firms that experiece large inflows, and the opposite for firms experience large outflows. As we hypothesize the enforcement of corporate environmental policies is negatively correlated with financial constraints, we predict that β >0 for the contemporaneous FIPP. Table 3 presents regression estimates of firms’ toxic release on the FIPP, which is our shock

18

to firms financial constraints. In the first two columns, we estimate the model when FIPP belongs to the extreme positive decile (i.e., decile 10, or the extreme “inflow” decile). Column (1) includes both firm and year fixed effects, and the unit of observation is establishment-firmyear; and column (2) includes year fixed effects to account for potential issuance waves that are clustered in times, and the unit of observation is in firm-year. First, column (2) confirms the evidence in Khan, Kogan and Serafeim (2012), which shows that indeed a large positive FIPP shocks induces equity issuance and therefore relaxes financial constraints. Second, column (1) presents a sizeable and statistically significant impact of FIPP on toxics release. A one-standard-deviation increase in FIPP (0.22) in the inflow sample increase toxics release (in natural logarithm) by 0.6, which represents 3.6% of our inflow sample mean. When the FIPP is extremely negative (i.e., decile 1, or the extreme “outflow” decile), we find the increase of outflow leads to significantly higher toxics release . A one-standard-deviation increase in FIPP (0.095) in the outflow sample toxics release (in natural logarithm) by 0.44, which represents 3.25% of our outflow sample mean (1.42). Overall, results are consistent with the idea that short-term price deviation from fundamental driven by FIPP is associated with changes in financial constraints, which in turn has a significant impact on the corporate environmental policies.

4.2

The Collateral Channel: Real Estate Value Shock

Our second identification strategy exploits the collateral channel of firms’ real estate assets. Real estate presents a major part of the tangible assets that firms can pledge as collaterals, therefore, an increase in real estate assets’ value reduces external financing friction and facilitates lending activities. As shown in Chaney, Sraer and Thesmar (2012), an increase in collateral value leads to more debt issuance and investment activities. This experiment fits our setting well because our sample firms tend to have more tangible and higher leverage compared to the average firms. Given the importance of collateral pledgeability, we examine its effect on corporate environmental policies. 19

This experiment uses variations in real estate collateral value of land-holdings firms driven by local real estate prices between 1993-2007. Before 1993, Compustat provides detailed decomposition of property, plant and equipment (PPE), which includes 3 major categories of real estate assets: Building, Land and Improvement, and Construction in Progress. We retrieve the market value of these three categories for firms in 1993 and then inflate the real estate value using local MSA level Housing Price Index (HPI) provided by the Federal Housing Finance Agency (FHFA) from 1993-2007. A few complications are involved in this process. First, the asset values provided by Compustat in 1993 are book values instead of market values. To generate market value in 1993, we first refer the average age of these assets by calculating the ratio of accumulated depreciation of buildings to historic cost of buildings (assuming a 40 year depreciation schedule), and then approximate the historical cost using CPI before 1975 and state-level HPI after 1975. Second, after getting the market value in 1993, we use headquarter MSA-level HPI as the price inflater to generate real estate value between 19932007.12 Real estate value calculated this way has the advantage of not being affected by firms’ decision of acquiring additional real estate assets after 1993, which helps us to obtain a cleaner identification. There might exist some concerns with respect to omitted variables associated both HPI and firms’ environmental decisions, which will contaminate the collateral channel that we try to establish. To deal with these endogeneity concerns, we use the MSA-level HPI instrumented by the MSA-level supply elasticity (Saiz (2010)) interacted with U.S. 30 year mortgage rates (Mian and Sufi (2011), Chaney, Sraer and Thesmar (2012)). The intuition is that as housing demand increases due to changes in the mortgage rate, home prices are less sensitive to demand in elastically supplied markets because these areas capitalize demand shocks into quantities rather than prices. Whereas in areas with very inelastic supply, demand shocks translate into prices instead of quantity (for example, consider areas like San Francisco or Boston). The first stage 12

Although this choice is partially driven by the lack of precise geographic location for each real estate asset, Chaney, Sraer and Thesmar (2012) collect information regarding the real estate assets using firms’ 10K filing and confirm that facilities do clusters with headquarters in the same MSA area, and a major portion of corporate real estate assets do locate in the headquarters.

20

specification is as follows,

H P I i,t = α + β Elast ici t yi × M or t gage Rat e t + M SA F E + Year F E + εi,t

(4.3)

In the second stage, we run regression specifications as:

To x icsRel easek, j,i,t = α+β×RE Value j,i,t +γH P I i,t +

X

κn I ni t.C ond j ×H P I i,t +cont r ols+εk, j,i,t

n

(4.4) where i indicates MSA, t indicates year, j indicates firm, and k indicates establishment. The regressions include MSA-level HPI to control for the direct impact of real estate price on toxic releases. It also includes initial characteristics, which are the five quintiles of age, assets, and return on assets interacted with MSA-level HPI to control for the heterogeneous ownership decisions and its potential impact on the sensitivity that we measure. Establishment fixed effects are included in all specifications. Table A.5 presents first stage regression estimates where the interaction terms are significantly positively related to MSA HPI, both in the specification with MSA-level supply elasticity in column (1) and with the quartile indicators in column (2). Table 4 Panel A presents the regression estimates for collateral channels, with column (1) using real estate value inflated by HPI and columns (2)-(4) using real estate value inflated by the instrumented HPI. Across all four columns, higher real estate values leads to lower toxic release amount. The coefficient estimates show that a one-standard-deviation increase in real estate value leads to a 0.056 (1.47*0.037) drop in total release when we include industry-year fixed effects, which is between 3%-4% of the sample mean. To learn more about firms’ resource allocation decision across environmental protection and other important aspects such as investment, in Table A.5 columns (5)-(6) we examine the effect of real estate value changes on investment at the firm level. We include cashflow/Assets and lagged Tobin’s Q as control variables and firm fixed effects, year fixed effects, and initial firm characteristics interacted with MSA-level HPI as in Chaney, Sraer and Thesmar (2012) 21

to make the results comparable to the literature. Within our sample, one standard deviation increase in the real estate value leads to 14% increase in CAPEX/PPE relative to our sample mean. The sensitivity on investment is about 2-3 times larger than the sensitivity on toxic release under the same real estate value shock. In Panel B of Table 4, we take this identification strategy one step further by examining the heterogeneous impact with respect to firms’ ex ante financial constraint level. The variation we use here is that the increase in real estate value should be more beneficial for firms that are ex ante constrained. This extra set of tests make our results robust to concerns that our instrument might be related to some higher order of demand factors that might affect firms’ environmental policies, as it would be hard to argue why these factors would result in different impacts for constrained versus unconstrained firms. Panel B presents the results for the low group, which include the bottom 30% of textual financial constraint measures, versus the rest of our sample. Real estate value does not show significant impact on the total release amount for the unconstrained group. The coefficient for the unconstrained group in column (3) is only 0.001 with t-stat being almost zero, whereas for high group in column (4) the coefficient is 0.087, which is statistically significant at 1% level. This cross-sectional analysis further confirms the impact of financial constraints on firms’ environmental policy through the collateral channel.

5

Cross-sectional Analysis

In addition to the exogenous shocks, we explore a few cross-sectional settings leading to heterogeneous impact of financial constraints on firms’ corporate environmental policies that are in line with our mechanisms. These setting include: the regulatory environment and monitoring, the investment horizon of institutional investors, and earnings management.

22

5.1

Regulatory Environment

As we discussed early, under the Clean Air Act (CAA), if an area is designated as "nonattainment," states must take corrective actions. Through the Federal Highway Administration, the EPA can impose the mandatory sanction of highway funding moratorium. Discretionary sanctions further require facilities located in nonattainment counties to adopt the lowest achievable emission rate (LAER) without cost consideration. Any new emissions are required to be offset from an existing emission source within the same county. This set of regulations generate cross-county variations of monitoring and enforcement in environmental protection. Exploiting such an institutional setting, Table 5 investigates how financial constraints impact toxic release by individual establishments that reside at geographic locations with different external monitoring and enforcement by government agencies. We first examine how nonattainment status is related to the level of toxic releases in Columns (1) and (2). The regression specification include year fixed effects and firm fixed effects, as well as firm-level and establishment-level controls. These two specifications differ by the definition of establishmentlevel industry classifications. Results show that establishments in nonattainment counties on average releases less than their intra-industry peers from attainment counties. The economic magnitude is major: the nonattainment dummy is associated with 20% less toxic releases relative to our sample mean. Columns (3) to (6) of Table 5 answer a different question: when firms are more financially constrained, how do establishments from counties with different attainment status manage the toxic release in response? We separately estimate the impact of financial constraints on toxic release amounts, grouped by the contemporaneous attainment status of the county where the establishment resides. We find that when firms are more financially constrained, establishments in attainment counties (weak regulation) generate more in toxic release amounts. In other words, the sensitivity of toxic release to financial constraint is higher in attainment counties. The sensitivity on average is between 2 to 3 times larger in attainment counties(weak regulation) and statistically significant, while the sensitivity measure is much smaller and not 23

statistically significant in nonattainment countries (strong regulation). This set of tests highlights firms’ active management of toxic release according to local regulatory environment. From a long-term perspective, strict regulatory environmental protections is associated with lower toxic releases amount, which could be driven by more abatement expenditures or even location choice of the establishments in response to regulatory changes(Henderson (1996),Becker and Henderson (2000)). Whereas from a short-term perspective, when facing financial constraints firms shift resources away from locations where regulatory environmental protection and enforcement is relatively weak (i.e., attainment counties). It is worthwhile to point out that such short-term intra-firm resource reallocation is possible precisely due to the abatement costs composition: lumpy fixed-asset investment is not the only option to process the industrial wastes; instead, the majority of the abatement costs are variable costs and can be transfered fairly easily. In addition to the non-attainment status, we explore the polluters’ size as another index for regulatory strictness. Due to historical reasons, large polluters have been the focus for EPA monitoring and enforcement activities as they impose the largest threat to harm people’s health and the environment. In addition, large polluters also face much higher level of public scrutiny of their environmental performance. For example, a number of environmental protection advocacy groups, public concerns groups, and news media routinely scrutinize so-called “super-polluters.”13 Table (6) tests the impact of financial constraints on toxic release in two subsamples: (1) “large” super-polluters, or establishments belonging to the top 30% of total toxic release amount in a year; and (2) “small” polluters, or establishments of the other 70% of total toxic release amount in a year. The most robust finding from Table (6) is that, financial constraints seem to have a much larger and statistically significant impact on smaller polluters, whereas the effects on large polluters are tiny and statistically insignificant. 13

For more information, interested readers are referred http://www.sierraclub.org/compass/2016/11/standing-super-polluters.

24

to

the

following

article:

5.2

Investment Horizons of Institutional Investors

Bushee (2001) shows that high levels of ownership by institutions with short-term investment horizons (i.e., “transient” institutions) are associated with overweighting of the near-term earnings component of value and underweighting of the long-term earnings component. In the context of market for corporate control, evidence in Gaspar, Massa and Matos (2005) show that target firms with short-term shareholders are more likely to receive an acquisition bid, but get lower premiums. Chen, Harford and Li (2007) find that only concentrated holdings by independent long-term institutions (i.e., “dedicated” institutions in the language of Bushee (2001)) are related to post-merger performance and their presence makes withdrawal of bad bids more likely. Taking the evidence together, there exists a strong correlation between the investment horizons of institutional investors and managerial decision horizon, although at this point it is not entirely clear whether such a correlation is driven by managerial catering to outside investors, or outside investor’s preference for managerial styles, or both. Using holdings by institutional investors and their portfolio composition, we calculate each firm’s average investment horizon. Based on the positive correlation established in the literature, we interpret investor’s investment horizon as a proxy for the firm’s managerial horizons. Table (7) sorts firm-year observations into two groups by investment horizons and examines the impact of financial constraints on establishments’ toxic releases. The idea behind this exercise is to explore whether the managerial decision horizon matters for the corporate environmental policies. For example, managers with myopic preference is more likely to “cut corners” in the presence of financial constraints. Empirical evidence supports our conjecture. Columns (1) and (2) sort firm into two groups based on the number of “transient” institutions investors, columns (3) and (4) use the equallyweighted institutional investor’s holding horizon measured by portfolio turnover rate, and columns (5) and (6) use holding value weighted-average institutional investor’s turnover rate. The sensitivity of toxic releases to financial constraint measures here is approximately 2 to 3 times higher for establishments with shorter investment horizon, indexed by more transient 25

holders (Column(1)) and higher turnover ratios (Columns (3) and (5)). For example, in Column (1) the coefficient on Text FC measure is 0.482 with statistical significance at 1% level and a one-standard-deviation increase in Text FC corresponds to 5.5% of our sample mean, whereas the coefficient in Column (2) for low group is only 0.103 and statistically insignificant from zero. The overarching conclusion from this contrast is that financial constraints have an economically sizable and statistically significant impact on toxic release when the investment horizon or managerial decision horizon is short.

5.3

Earnings Management

To provide collaborative evidence on the impact of managerial incentives on toxic release, we examine how toxic release responds to the incentives of meeting earnings targets. Firms exhibit strong tendency to meet earnings targets, as shown in literatures (Healy and Wahlen (1999), Dechow and Skinner (2000), and Fields, Lys and Vincent (2001)). Survey evidence (Graham, Harvey and Rajgopal (2005)) and academic research also finds that managers will manipulate real activities in order to meet earnings benchmarks (Roychowdhury (2006), Cohen, Mashruwala and Zach (2010), Caskey and Ozel (2017), Xu and Zwick (2017)), even though such activities have a negative impact on firms’ long-run performances (Bhojraj et al. (2009)). Given the high costs associated with waste management, we hypothesize environmental policy to be one channel through which managers take actions to reach performance benchmarks. Table 8 puts this idea to the test. We calculate an earnings surprise as the difference between reported earning and the analysis consensus, and define a dummy variable equal to 1 for firm-years that fell short of analysis consensus. The intuition is that firms missing the target are precisely those that have tried “everything possible” including squeezing costs from environmental protection. Columns (1) to (3) in Table 8 present the baseline results. Establishments emit significantly more toxic releases when EPS is lower than analysis forecasts, and the effect size is approximately 2.3% of the sample mean. 26

We explore one more institutional setting to generate cross-sectional variation. Environmental regulations enable the EPA to ban federal agencies from awarding contracts, grants, or loans to plants and facilities that violate the Clean Air Act (CAA) or Clean Water Act (CWA). These particular regulations impose higher costs for government contractors and prevent them from cutting corners in environmental protections for cost savings. Columns (4) and (5) split the sample of firms into two subsamples based on whether the establishment is a US government contractor and test whether the establishments respond differently, given the same kind of earnings target pressure. Our evidence suggests that the increasing toxic release when under earnings pressure primarily shows up for non government contractors. The coefficient in column (5) is only 0.008 for government contractors and not statistically significant, while the coefficient is three time as large for non government contractor in column (6) and statistically significant. Overall, our result is consistent with earnings management and environmental protection being connected decisions, with an active trade-off margin operating between them. In addition, firms exposed to less strict regulatory enforcement strongly favor meeting earnings targets over environmental protection when facing this particular trade off.

6

Conclusion

Exploiting several novel establishment level datasets, we provide evidence that a firm’s environmental policies are directly affected by its financial constraints. Treating toxic release generated during the production process is costly. When a firm is resource constrained, it faces the tradeoff between environmental protection and shareholder value maximization. Often, corporations have the tendency of “cutting corners” and sacrificing environmental protection. Next, to establish causality, we explore two scenarios where a firm’s external financing constraints are likely to be exogenously impacted — one in equity financing and the other in collateralized borrowing. Evidence shows that relaxation (or tightening) of external financing constraints reduces (or increases) the firm’s toxic release. In addition, cross-sectional tests

27

illustrate that the impact of financial constraints on toxic release is amplified by weaker regulatory monitoring and enforcement, myopic managers with short horizons, and when firms are under pressure to meet earnings targets. Overall, our evidence highlights the externalities of financial frictions in the form of environmental pollution.

28

References Baker, Malcolm, Jeremy C Stein, and Jeffrey Wurgler. 2003. “When does the market matter? Stock prices and the investment of equity-dependent firms.” The Quarterly Journal of Economics, 118(3): 969–1005. Becker, Randy, and Vernon Henderson. 2000. “Effects of air quality regulations on polluting industries.” Journal of political Economy, 108(2): 379–421. Bhojraj, Sanjeev, Paul Hribar, Marc Picconi, and John McInnis. 2009. “Making sense of cents: An examination of firms that marginally miss or beat analyst forecasts.” The Journal of Finance, 64(5): 2361–2388. Bodnaruk, Andriy, Tim Loughran, and Bill McDonald. 2015. “Using 10-k text to gauge financial constraints.” Journal of Financial and Quantitative Analysis, 50(4): 623–646. Bushee, Brian J. 2001. “Do institutional investors prefer near-term earnings over long-run value?” Contemporary Accounting Research, 18(2): 207–246. Carhart, Mark M. 1997. “On persistence in mutual fund performance.” The Journal of finance, 52(1): 57–82. Caskey, Judson, and N Bugra Ozel. 2017. “Earnings expectations and employee safety.” Journal of Accounting and Economics, 63(1): 121–141. Chaney, Thomas, David Sraer, and David Thesmar. 2012. “The collateral channel: How real estate shocks affect corporate investment.” The American Economic Review, 102(6): 2381– 2409. Chava, Sudheer. 2014. “Environmental externalities and cost of capital.” Management Science, 60(9): 2223–2247. Chen, Xia, Jarrad Harford, and Kai Li. 2007. “Monitoring: Which institutions matter?” Journal of financial Economics, 86(2): 279–305. Chodorow-Reich, Gabriel. 2013. “The employment effects of credit market disruptions: Firm-level evidence from the 2008–9 financial crisis.” The Quarterly Journal of Economics, 129(1): 1–59. Cohen, Daniel, Raj Mashruwala, and Tzachi Zach. 2010. “The use of advertising activities to meet earnings benchmarks: Evidence from monthly data.” Review of Accounting Studies, 15(4): 808–832. Cohn, Jonathan B, and Malcolm I Wardlaw. 2016. “Financing constraints and workplace safety.” The Journal of Finance, 71(5): 2017–2058. Copeland, Brian R, and M Scott Taylor. 2004. “Trade, growth, and the environment.” Journal of Economic literature, 42(1): 7–71.

29

Coval, Joshua, and Erik Stafford. 2007. “Asset fire sales (and purchases) in equity markets.” Journal of Financial Economics, 86(2): 479–512. Dechow, Patricia M, and Douglas J Skinner. 2000. “Earnings management: Reconciling the views of accounting academics, practitioners, and regulators.” Accounting horizons, 14(2): 235–250. Edmans, Alex, Itay Goldstein, and Wei Jiang. 2012. “The real effects of financial markets: The impact of prices on takeovers.” The Journal of Finance, 67(3): 933–971. Farre-Mensa, Joan, and Alexander Ljungqvist. 2016. “Do measures of financial constraints measure financial constraints?” The Review of Financial Studies, 29(2): 271–308. Fields, Thomas D, Thomas Z Lys, and Linda Vincent. 2001. “Empirical research on accounting choice.” Journal of accounting and economics, 31(1): 255–307. Gao, Pengjie, and Dong Lou. 2013. “Cross-market timing in security issuance.” Gaspar, José-Miguel, Massimo Massa, and Pedro Matos. 2005. “Shareholder investment horizons and the market for corporate control.” Journal of financial economics, 76(1): 135– 165. Gormley, Todd A, and David A Matsa. 2011. “Growing out of trouble? Corporate responses to liability risk.” The Review of Financial Studies, 24(8): 2781–2821. Graham, John R, Campbell R Harvey, and Shiva Rajgopal. 2005. “The economic implications of corporate financial reporting.” Journal of accounting and economics, 40(1): 3–73. Hadlock, Charles J, and Joshua R Pierce. 2010. “New evidence on measuring financial constraints: Moving beyond the KZ index.” The Review of Financial Studies, 23(5): 1909–1940. Healy, Paul M, and James M Wahlen. 1999. “A review of the earnings management literature and its implications for standard setting.” Accounting horizons, 13(4): 365–383. Henderson, J Vernon. 1996. “Effects of air quality regulation.” The American Economic Review, 86(4): 789. Hoberg, Gerard, and Vojislav Maksimovic. 2014. “Redefining financial constraints: A textbased analysis.” The Review of Financial Studies, 28(5): 1312–1352. Hong, Harrison, Jeffrey D Kubik, and Jose A Scheinkman. 2012. “Financial constraints on corporate goodness.” National Bureau of Economic Research. Khan, Mozaffar, Leonid Kogan, and George Serafeim. 2012. “Mutual fund trading pressure: Firm-level stock price impact and timing of SEOs.” The Journal of Finance, 67(4): 1371–1395. Levinson, Arik. 2009. “Technology, international trade, and pollution from US manufacturing.” The American economic review, 99(5): 2177–2192.

30

Levinson, Arik. 2015. “A direct estimate of the technique effect: changes in the pollution intensity of US manufacturing, 1990–2008.” Journal of the Association of Environmental and Resource Economists, 2(1): 43–56. Lou, Dong. 2012. “A flow-based explanation for return predictability.” The Review of Financial Studies, 25(12): 3457–3489. Mian, Atif, and Amir Sufi. 2011. “House prices, home equity–based borrowing, and the us household leverage crisis.” The American Economic Review, 101(5): 2132–2156. Roychowdhury, Sugata. 2006. “Earnings management through real activities manipulation.” Journal of accounting and economics, 42(3): 335–370. Saiz, Albert. 2010. “The geographic determinants of housing supply.” The Quarterly Journal of Economics, 125(3): 1253–1296. Shapiro, Joseph S, and Reed Walker. 2015. “Why is pollution from US manufacturing declining? The roles of trade, regulation, productivity, and preferences.” National Bureau of Economic Research. Stern, Nicholas. 2008. “The Economics of Climate Change.” American Economic Review, 98(2): 1–37. Walker, W Reed. 2013. “The transitional costs of sectoral reallocation: Evidence from the Clean Air Act and the workforce.” The Quarterly journal of economics, 128(4): 1787–1835. Xu, Qiping, and Eric Zwick. 2017. “Kinky Tax Policy and Abnormal Investment Behavior.”

31

Figure 1: Environmental Protection Agency Panel A presents different Environmental Protection Agency (EPA) regulations that govern toxics release in the United States. Panel B presents the EPA guideline for waste management, where disposal or other releases are the least favorable method. Panel C presents summary of the Pollution Abatement Costs and Expenditures (PACE) 2005 survey (last available) conducted by the US Census Bureau for the manufacturing sector.

A. EPA’s Role in Protecting Public Health

B. Waste Management Hierarchy

C. Abatement Costs and Expenditures

Source: United States Environmental Protection Agency

32

Figure 2: EPA Nonattainment Status This figure displays a map of counties designated “Nonattainment” for the National Ambient Air Quality Standards (NAAQS) pollutants as of September 2017. EPA publishes nonattainment status for each county in its Green Book publication each year.

Source: https://www.epa.gov/green-book

33

Figure 3: Aggregate Toxics Release

0

Emission Amount (000 Tons) 200 400

600

This figure presents the aggregate time series for establishments in our public US firm sample between 19902014. We include the total toxic release amount (in 000 tons) and toxic release under Clean Air Act (CAA) (in 000 tons) in the figure.

1990

1995

2000

Year

Total Release

34

2005

2010 CAA Release

2015

Table 1: Summary Statistics Panel A presents the firm level summary statistics for our sample US public firms from 1990-2014. Panel B summarizes Compustat non-fianncial firms from 1990-2014 as comparison. Panel C summarizes establishment level data, including sales($Mil), employment, toxics release amount (tons) under different EPA categories, and noncomplicance status.

Panel A: Firm Characteristics

Assets(Mil) Leverage Cash/Assets Tobin Q CAPEX/PPE Tangible Text FC HM Debt

N

Mean

Median

SD

P25

P75

18,155 18,100 18,149 17,218 17,699 18,146 11,292 8,713

6,148.28 0.41 0.09 1.58 0.21 0.33 0.70 0.02

871.79 0.39 0.04 1.33 0.17 0.29 0.69 0.01

26,249.97 0.27 0.11 0.84 0.17 0.18 0.20 0.06

262.35 0.21 0.01 1.06 0.11 0.19 0.55 -0.02

3,226.08 0.57 0.12 1.81 0.25 0.43 0.82 0.05

Panel B: Compustat Firm Characteristics

Assets - Total Leverage Cash/Assets Tobin Q CAPEX/PPE Tangible

N

Mean

Median

SD

P25

P75

154,083 152,256 153,890 138,277 141,544 153,801

3,097.42 0.35 0.18 1.94 0.43 0.32

191.27 0.30 0.09 1.37 0.22 0.23

16,185.57 0.31 0.23 1.98 0.72 0.26

48.69 0.04 0.02 1.03 0.12 0.09

977.30 0.56 0.26 2.11 0.44 0.49

Panel C: Establishment Characteristics

Sales(Mil) Employment Total Release CAA Release CWA Release CERCLA Release OSHA Release Log Total Release Log CAA Release Log CWA Release Log CERCLE Release Log Osha Release Non Compliance

N

Mean

Median

SD

P25

P75

89,454 89,464 87,292 74,984 80,517 84,165 51,503 87,292 74,984 80,517 84,165 51,503 89,464

115.14 676.30 119.32 71.85 87.60 109.62 17.82 1.75 1.25 1.44 1.61 -0.22 0.04

40.90 275.00 8.42 5.93 6.36 7.56 1.24 2.13 1.78 1.85 2.02 0.21 0.00

256.60 1,403.84 351.14 193.21 254.93 313.38 43.01 3.20 3.37 3.22 3.28 3.55 0.20

15.29 100.00 0.85 0.49 0.59 0.72 0.13 -0.16 -0.71 -0.52 -0.33 -2.08 0.00

105.28 665.00 52.19 37.30 39.22 47.43 11.20 3.95 3.62 3.67 3.86 2.42 0.00

35

Table 2: Total Toxics Release This table present OLS estimates of the log number of total toxics release on two financial constraint measures: text-based financial constraint proxies by Bodnaruk, Loughran and McDonald (2015) and text-based financial constraint proxies by Hoberg and Maksimovic (2014). Firm level controls include log(assets), Tobin’s Q, Tangible, Leverage, Cash/Assets, and CAPEX/PPE. Establishment level controls include log(sales) and log(employment). Firm level controls are lagged by one year and establishment level controls are contemporaneous. Establishment fixed effects are included in all specifications and standard errors are clustered at the firm level.

A Text FC

(1) 0.221∗∗∗ (3.07)

(2) 0.157∗∗ (2.40)

(3) 0.192∗∗∗ (2.78)

HM Debt Observations Adj R-Squared Establishment FE Controls Year FE Industry-Year FE State-Year FE

51,364 0.83 Yes Yes Yes

51,312 0.84 Yes Yes

51,106 0.84 Yes Yes

(4)

(5)

(6)

0.686∗ (1.91) 36,514 0.86 Yes Yes Yes

0.643∗∗ (2.12) 36,463 0.86 Yes Yes

0.648∗∗ (2.09) 36,374 0.86 Yes Yes

Yes

Yes Yes

Yes

Table 3: Flow Induced Price Pressure This table reports regression of firms’ emissions on the flow induced price pressure (FIPP). Columns (1) and (3) use the log of total toxic release as outcome variable with observations at the establishment-year level. Column (2) use net stock issuance scaled by lagged assets as outcome variable with observations at the firm-year level. Firm level controls include log(assets), Tobin’s Q, Tangible, Leverage, Cash/Assets, and CAPEX/PPE. Establishment level controls include log(sales) and log(employment). Firm level controls are lagged by one year and establishment level controls are contemporaneous. Inflow (outflow) sample includes firm-years that experience buying (selling) pressure in the highest decile among firms in the given year, and we exclude observations if they were also in the highest decile in the previous year.

FIPP Observations Adj R-Squared Year FE Firm FE Controls

Inflow (1) (2) Log(total release) Net Stock Issuance -2.712∗ 0.108∗∗ (-1.70) (2.15) 1,787 231 0.35 0.18 Yes Yes Yes Yes Yes

36

Outflow (3) Log(total release) -4.746∗ (-1.87) 2,588 0.29 Yes Yes Yes

Table 4: The Collateral Channel – Real Estate Shock This table reports regression estimates of firms’ total toxics releases on firms’ real estate value for our sample firms between 1993 to 2007. In Panel A, Column (1) presents OLS estimates where firms’ real estate value is calculated as the 1993 market value of real estate assets, inflated by the MSA-level House Price Index (HPI). Columns(2)-(4) present regression estimates where MSA-level HPI is instrumented by the interaction term between MSA-level local housing supply elasticity and 30-year US fixed mortgage rate. MSA-level HPI is included in the regressions to control for the direct impact of HPI. Firm level controls include lagged log(assets), Tobin’s Q, Tangible, Leverage, Cash/Assets, CAPEX/PPE and initial firm characteristics interacted with MSA-level HPI. Establishment level controls include contemporaneous log(sales) and log(employment). Columns(5)-(6) present regression estimates using CAPEX/PPE as the LHS variable, with Cashflow/Assets and lagged Tobin’s Q as controls variables. Standard errors are clustered at the firm level. Panel B presents a subsample test where the low group includes the bottom 30% of textual financial constraint measures each year and the high group includes the rest observations.

A. Real Estate Value Shock (1) -0.041∗∗ (-2.32)

RE Value

(4)

25,152 0.82 Yes

-0.050∗∗∗ (-2.82) 20,558 0.83 Yes

-0.048∗∗∗ (-2.60) 20,394 0.83 Yes

-0.037∗∗ (-2.14) 20,510 0.84 Yes

Yes Yes

Yes Yes

Yes

Yes

RE Value IV Observations Adj R-Squared Establishment FE Firm FE Controls Year FE State-Year FE Industry-Year FE

Log(total release) (2) (3)

CAPEX/PPE (5) (6) 0.019∗∗∗ (3.39)

3,511 0.45

0.021∗∗∗ (3.12) 3,357 0.46

Yes Yes Yes

Yes Yes Yes

Yes Yes

B. Constrained vs Unconstrained Firms

RE Value

(1) Low -0.050 (-0.71)

Text FC (2) (3) High Low -0.083∗∗∗ (-2.67)

RE Value IV Observations Adj R-Squared Establishment FE Controls Year FE

6,051 0.84 Yes Yes Yes

12,381 0.83 Yes Yes Yes 37

-0.001 (-0.01) 4,776 0.83 Yes Yes Yes

(4) High

-0.087∗∗∗ (-2.74) 10,341 0.83 Yes Yes Yes

Table 5: Nonattainment Status This table examines the relation between a firm’s total toxics release choice and the EPA nonattainment status of the county where an establishment resides. Firm level controls include log(assets), Tobin’s Q, Tangible, Leverage, Cash/Assets, and CAPEX/PPE. Establishment level control includes log(sales) and log(employment). Firm level controls are lagged by one year and establishment level controls are contemporaneous. Columns(1)-(2) include firm, year, and establishment-level industry fixed effects. Columns(3)-(4) split the sample according to the nonattainment status, and include establishment fixed effects and year fixed effects.

All

Nonattainment

Nonattainment County (3) (4) (5) (6) Yes No Yes No

(1)

(2)

-0.367∗∗∗ (-4.12)

-0.337∗∗∗ (-4.00)

Text FC

0.131 (1.47)

0.272∗∗∗ (3.03)

HM Debt Observations Adj R-Squared Year FE Controls Firm FE Establishment FE FF48 FE SIC4 FE

52,510 0.41 Yes Yes Yes

52,484 0.46 Yes Yes Yes

Yes Yes

38

20,300 0.87 Yes Yes

30,633 0.84 Yes Yes

0.422 (1.04) 11,424 0.89 Yes Yes

Yes

Yes

Yes

1.110∗∗ (2.28) 17,653 0.87 Yes Yes Yes

Table 6: Large Polluters This table examines the heterogeneous impact of financial constraint on total toxics release for large vs small polluters. Firm level controls include log(assets), Tobin’s Q, Tangible, Leverage, Cash/Assets, and CAPEX/PPE. Establishment level control includes log(sales) and log(employment). Firm level controls are lagged by one year and establishment level controls are contemporaneous. We define an establishment as a large polluter if it belongs to the top 30% of total release amount. The regressions also include establishment fixed effects and year fixed effects.

Text FC

(1) Large 0.053 (1.13)

(2) Small 0.197∗∗ (2.28)

HM Debt Observations Adj R-Squared Year FE Controls Establishment FE

15,846 0.86 Yes Yes Yes

39

34,834 0.72 Yes Yes Yes

(3) Large

(4) Small

-0.036 (-0.19) 10,732 0.88 Yes Yes Yes

0.781∗ (1.85) 25,214 0.77 Yes Yes Yes

Table 7: Institutional Holding Composition This table reports regression estimates of firms’ total toxic chemical releases on firms’ institutional holding composition. Columns(1)-(2) present estimates with subsample split by the median number of transient holders. Columns(3)-(4) present estimates with subsample split by the equal weighted turnover ratio of institutional holders. Columns(3)-(4) present estimates with subsample split by the value weighted turnover ratio of institutional holders. Firm level controls include log(assets), Tobin’s Q, Tangible, Book leverage, Cash/Assets, and CAPEX/PPE. Establishment level control includes log(sales) and log(employment). Firm level controls are lagged by one year and establishment level controls are contemporaneous. Establishment and year fixed effects are included in all regressions and standard errors are clustered at firm level.

Text FC Observations Adj R-Squared Establishment FE Year FE

N TRA Holders (1) (2) High Low 0.482∗∗∗ 0.103 (3.78) (1.07) 25,268 22,661 0.83 0.82 Yes Yes Yes Yes

EW Turnover (3) (4) High Low 0.403∗∗∗ 0.092 (2.99) (1.02) 23,875 23,338 0.83 0.84 Yes Yes Yes Yes

40

VW Turnover (5) (6) High Low 0.356∗∗∗ 0.124 (3.11) (1.14) 23,455 23,667 0.82 0.84 Yes Yes Yes Yes

Table 8: Toxics Release and Earnings Management This table reports OLS estimates of firms’ total toxics release on a dummy variable indicating the reported earnings per share missing the average analysis forecast in that firm-year. Columns(1)-(3) present results for the full sample, Columns(4)-(5) present subsample results by splitting the sample according to government contractor status. Firm level controls include log(assets), Tobin’s Q, Tangible, Book leverage, Cash/Assets, and CAPEX/PPE. Establishment level control includes log(sales) and log(employment). Firm level controls are lagged by one year and establishment level controls are contemporaneous. Establishment fixed effects are included in all regressions and standard errors are clustered at firm level.

(1) Miss Earnings Forecast Observations Adj R-Squared Establishment FE Controls Year FE Industry-Year FE State-Year FE

0.046∗∗ (2.14) 73,822 0.77 Yes Yes Yes

All (2)

(3)

0.038∗∗ (2.04) 73,777 0.78 Yes Yes

0.043∗∗ (2.04) 73,276 0.78 Yes Yes

Yes Yes

41

Government Contractor (4) (5) Yes No 0.008 0.046∗∗ (0.19) (2.20) 17,511 57,292 0.76 0.79 Yes Yes Yes Yes Yes Yes

Table A.1: Variable Definition CAPEX/ PPE Cash/Assets Cashflow/Assets Leverage Tobin’s Q Tangible log(employment ) log(sales) Nonattainment Non Compliance Text FC HM Debt N TRA Holders EW Turnover VW Turnover Net Stock Issuance RE Value

Capital Expenditures/ L.PPENT Cash and Short-term Investment/L.Assets Operating income before depreciation and amortization + depreciation and amortization/L.Assets (Debt in Current Liabilities + Long-Term Debt)/ Assets (Total Asset + Common Shares Outstanding × Closing Price (Fiscal Year) − Common Equity − Deferred Taxes)/Asset PPENT/L.Assets Log number of employee at the establishment level Log number of sales dollar amount (inflation adjusted) at the establishment level A dummy equals to one if an establishment resides in a county with nonattainment status A dummy equals to one if an establishment is in assigned as non compliant by the EPA in a year Textual financial constraint measure by Bodnaruk, Loughran and McDonald (2015) Debt focus financial constraint measured by Hoberg and Maksimovic (2014) Number of transient holders according to Bushee (2001) reported on 13-F filings Equal-weighted turnover ratio of institutional investors reported on 13-F filings Value-weighted turnover ratio of institutional investors reported on 13-F filings (Sell of common and preferred stocks-purchases of common and preferred stocks)/L.Assets 1993 market value of real estate assets inflated by the MSA-level House Price Index (HPI)

42

Table A.2: Summary Statistics by Industry The table presents summary statistics of total toxics release (tons) by Industry (Fama-French 48). Industries with fewer than 50 observations are dropped in this table. Total Toxics Release

Agriculture Food Products Candy & Soda Beer & Liquor Tobacco Products Recreation Printing and Publishing Consumer Goods Apparel Medical Equipment Pharmaceutical Products Chemicals Rubber and Plastic Products Textiles Construction Materials Construction Steel Works Etc Fabricated Products Machinery Electrical Equipment Automobiles and Trucks Aircraft Shipbuilding, Railroad Equipment Defense Precious Metals Non-Metallic and Industrial Metal Mining Coal Petroleum and Natural Gas Utilities Personal Services Business Services Computers Electronic Equipment Measuring and Control Equipment Business Supplies Shipping Containers Transportation Wholesale Retail Almost Nothing Total

N

Mean

Median

SD

P25

P75

179 3,953 204 255 162 595 160 4,314 374 1,077 1,590 10,806 3,135 1,121 7,271 364 7,119 2,689 6,766 3,335 4,927 2,224 650 466 152 363 62 2,623 2,773 314 1,910 576 3,342 1,116 4,104 2,211 247 2,512 745 395 87,181

161.18 82.33 15.56 35.10 153.70 59.30 8.52 52.41 23.00 33.73 101.88 152.23 78.70 51.73 46.77 95.63 208.82 25.57 21.92 35.31 73.19 37.46 43.47 37.80 1,079.73 586.05 242.26 180.09 746.93 55.28 87.47 43.53 12.81 14.22 263.02 113.45 88.08 67.93 54.31 362.62 119.37

6.67 13.66 9.26 11.50 42.38 37.71 4.48 9.95 10.21 6.01 9.12 10.25 9.55 8.60 7.04 1.45 8.66 3.87 2.98 3.60 10.88 6.91 20.20 6.30 654.31 39.67 68.17 33.79 430.55 1.01 5.90 5.00 1.55 1.37 29.58 39.50 3.89 2.23 4.86 16.22 8.42

426.08 209.21 24.13 52.61 211.58 76.29 17.49 116.32 32.54 118.25 294.00 415.81 217.41 150.91 149.83 344.72 540.37 80.53 98.88 139.81 166.31 113.10 71.85 126.26 1,015.07 848.03 363.19 339.30 771.09 257.38 254.39 197.31 44.21 48.12 445.00 253.47 249.68 244.89 155.53 691.44 351.30

0.38 4.14 0.44 3.92 9.67 5.60 0.73 0.65 1.81 0.55 1.00 1.70 1.29 1.61 0.43 0.20 0.73 0.35 0.25 0.23 1.01 0.59 4.67 0.59 10.60 2.55 24.10 2.25 108.06 0.13 0.73 0.14 0.13 0.13 4.38 7.13 1.16 0.55 0.72 1.08 0.85

78.00 65.00 19.10 41.67 215.64 87.00 9.76 52.22 28.50 21.28 52.96 66.44 44.35 37.72 29.89 34.98 71.69 21.62 14.39 21.12 53.00 27.78 51.47 31.81 2,248.90 1,014.77 284.00 198.52 1,219.48 12.98 39.07 21.17 7.72 7.50 341.45 116.61 24.97 17.20 23.24 259.53 52.20

43

Table A.3: Toxics Release Under Various EPA Categories This table present OLS estimates of the log number of firms’ toxics release under the Clean Air Act(CAA), the Clean Water Act (CWA), the Comprehensive Environmental Response, Compensation and Liability Act (CERCLA), and the Occupational Safety and Health Act (OSHA) on two financial constraints measures: text-based financial constraint proxies by Bodnaruk, Loughran and McDonald (2015), text-based financial constraint proxies by Hoberg and Maksimovic (2014). Firm level controls include log(assets), Tobin’s Q, Tangible, Book leverage, Cash/Assets, and CAPEX/PPE. Establishment level controls include log(sales) and log(employment). Firm level controls are lagged by one year and establishment level controls are contemporaneous. Establishment fixed effects are included in all specifications and standard errors are clustered at the firm level. (1) CAA 0.253∗∗∗ (3.14)

Text FC

(2) CWA 0.200∗∗∗ (2.66)

(3) CERCLA 0.195∗∗ (2.57)

(4) OSHA 0.345∗∗∗ (3.75)

HM Debt Observations Adj R-Squared Establishment FE Controls Year FE

45,098 0.83 Yes Yes Yes

47,312 0.84 Yes Yes Yes

49,361 0.84 Yes Yes Yes

33,109 0.81 Yes Yes Yes

(5) CAA

(6) CWA

(7) CERCLA

(8) OSHA

0.953∗∗ (2.12) 31,583 0.86 Yes Yes Yes

0.594∗ (1.89) 33,312 0.87 Yes Yes Yes

0.891∗∗ (2.38) 35,227 0.87 Yes Yes Yes

0.738 (1.20) 23,374 0.83 Yes Yes Yes

Table A.4: Compliance Status of Toxics Release and Financial Constraint This table presents OLS regression estimates of a dummy variable indicating establishment’s EPA compliance status, which equals one for non-compliant and zero otherwise, on two financial constraint measures: text-based financial constraint proxies by Bodnaruk, Loughran and McDonald (2015) and text-based financial constraint proxies by Hoberg and Maksimovic (2014). Firm level controls include log(assets), Tobin’s Q, Tangible, Leverage, Cash/Assets, and CAPEX/PPE. Establishment level controls include log(sales) and log(employment). Firm level controls are lagged by one year and establishment level controls are contemporaneous. Establishment fixed effects are included in all specifications and standard errors are clustered at the firm level.

Text FC

(1) -0.003 (-0.46)

(2) 0.002 (0.34)

(3) -0.002 (-0.27)

HM Debt Observations Adj R-Squared Establishment FE Controls Year FE Industry-Year FE State-Year FE

51,364 0.16 Yes Yes Yes

51,312 0.17 Yes Yes

51,106 0.17 Yes Yes

Yes

(4)

(5)

(6)

0.002 (0.07) 36,514 0.16 Yes Yes Yes

0.012 (0.38) 36,463 0.16 Yes Yes

0.013 (0.41) 36,374 0.16 Yes Yes

Yes Yes

44

Yes

Table A.5: The Collateral Channel – Real Estate Shock First Stage This table reports first stage regression estimates of MSA level HPI on the interaction between local housing supply elasticity and 30-year US fixed mortgage rate between 1993 to 2007 (Chaney, Sraer and Thesmar (2012)). Column (1) uses MSA-level local housing supply elasticity (Saiz (2010)) and Column (2) uses quartiles of the elasticity. MSA fixed effects and year fixed effects are included in all specifications. Standard errors are clustered at the MSA level.

MSA Supply Elasticity * Mortgage rate

MSA HPI (1) (2) ∗∗∗ 0.027 (6.11)

1 Quartile Elasticity * Mortgage Rate

-0.062∗∗∗ (-7.75)

2 Quartile Elasticity * Mortgage Rate

-0.046∗∗∗ (-5.87)

3 Quartile Elasticity * Mortgage Rate

-0.014∗∗ (-2.19) 1,246 0.94 Yes Yes

Observations Adj R-Squared Year FE MSA FE

1,246 0.94 Yes Yes

45