Options, Equity Risks, and the Value of Capital Structure Adjustments

Finance and Economics Discussion Series Divisions of Research & Statistics and Monetary Affairs Federal Reserve Board, Washington, D.C. Options, Equity Risks, and the Value of Capital Structure Adjustments Paul Borochin and Jie Yang 2016-097 Please cite this paper as: Borochin,PaulandJieYang(2016). “Options,EquityRisks,andtheValueofCapitalStructureAdjustments,”FinanceandEconomicsDiscussionSeries2016-097. Washington: Board of Governors of the Federal Reserve System, https://doi.org/10.17016/FEDS.2016.097. NOTE: Staff working papers in the Finance and Economics Discussion Series (FEDS) are preliminary materials circulated to stimulate discussion and critical comment. The analysis and conclusions set forth are those of the authors and do not indicate concurrence by other members of the research staff or the Board of Governors. References in publications to the Finance and Economics Discussion Series (other than acknowledgement) should be cleared with the author(s) to protect the tentative character of these papers.

Options, Equity Risks, and the Value of Capital Structure Adjustments∗ Paul Borochin Jie Yang School of Business Board of Governors of the University of Connecticut† Federal Reserve System‡ This version: October, 2016 Abstract We use exchange-traded options to identify risks relevant to capital structure adjustments in firms. These forward-looking market-based risk measures provide significant explanatory power in predicting net leverage changes in excess of accounting data. They matter most during contractionary periods and for growth firms. We form market-based indices that capture firms’ magnitudes of, and propensityfor, netleverageincreases. Firmswithlargerpredictedleverageincreases outperformfirmswithlowerpredictedincreasesby3.1%to3.9%peryearinbuy-andhold abnormal returns. Finally, consistent with the quality, leverage, and distress risk puzzles, firms with lower predicted leverage increases are riskier but earn lower abnormal returns. Keywords: Capital Structure, Financial Leverage, Options, Implied Volatility JEL classification: G30, G32, G12, G14 ∗We thank Jeff Netter (Editor), an anonymous referee, Turan Bali, Assaf Eisdorfer, John Graham, Philip Strahan, Boris Vallee, Toni Whited, Rohan Williamson, seminar participants at Georgetown University, University ofConnecticut,andparticipantsatthe2014FMAAnnualMeetings,2014OptionMetricsResearchConference,2015 Financial Intermediation Research Society Conference, and 2015 Eastern Finance Association Meetings for helpful comments. The authors acknowledge financial support from the Center for Financial Markets and Policy at the McDonoughSchoolofBusinessatGeorgetownUniversity. Theideasinthispaperaresolelythoseoftheauthorsand do not necessarily reflect the view of the Federal Reserve System. All remaining errors are our own. †Storrs, CT 06269. Phone: (860) 486-2774. Email: paul.borochin@uconn.edu. ‡Washington, DC 20551. Phone: (202) 736-1939. Email: jie.yang@frb.gov.

Options, Equity Risks, and the Value of Capital Structure Adjustments This version: October, 2016 Abstract We use exchange-traded options to identify risks relevant to capital structure adjustments in firms. These forward-looking market-based risk measures provide significant explanatory power in predicting net leverage changes in excess of accounting data. They matter most during contractionary periods and for growth firms. We form market-based indices that capture firms’ magnitudes of, and propensityfor, netleverageincreases. Firmswithlargerpredictedleverageincreases outperformfirmswithlowerpredictedincreasesby3.1%to3.9%peryearinbuy-andhold abnormal returns. Finally, consistent with the quality, leverage, and distress risk puzzles, firms with lower predicted leverage increases are riskier but earn lower abnormal returns. Keywords: Capital Structure, Financial Leverage, Options, Implied Volatility JEL classification: G30, G32, G12, G14

1 Introduction Firms adjust their capital structure by balancing the benefits and costs of using debt. As cash flow risk increases, the likelihood of a firm entering a state of default increases, thereby increasing the expected costs of bankruptcy. Thus, all else equal, increases in cash flow risk should to lead to increases in the cost of debt and consequently to decreases in firm leverage. Empirical research on capital structure has relied primarily on firm characteristics obtained from accounting statements to proxy for firm risks that factor into the cost of capital, leaving substantial variation in capital structure unexplained.1 The equity market, in comparison, is a relatively untapped source of information on firm risks for explaining leverage dynamics.2 TheintuitionforthisapproachcomesfromtheLeland(1994)modelofoptimalcapitalstructure, “Equityreturnvolatilitywillbestochastic,changingwiththeleveloffirmassetvalue,V”(p. 1249). Priorliteraturehasnotedtheusefulnessofequitymarketsincapturingunderlyingfirmriskrelevant to financing decisions (e.g., Myers, 1977; Myers, 1984; Marsh, 1982; Loughran and Ritter, 1995; Leary and Roberts, 2014; Schwert and Strebulaev, 2015; Chen, Wang, and Zhou, 2015). Using the Campbell (1991) decomposition of stock returns into to cash flow news and expected returns news, Vuolteenaho (2002) finds that firm-level stock returns are mainly driven by cash flow news, rather than expected returns news. Cash flow news affect the firm’s bankruptcy cost and while they are not directly observable, they will affect risk measures from the equity market. Thus, these cash flow risk measures should provide useful information for capital structure decisions. More recently, Welch (2004) finds that stock returns explain 40% of capital structure dynamics and Frank and Goyal (2009) show that both stock returns and stock volatility correlate negatively to leverage levels. In this paper, we study the ability of risk measures derived from the equity option market to predict changes to a firm’s capital structure. Studying changes in capital structure, rather than levels, enables us to identify a firm’s reaction to relevant risks and better understand the dynamics of capital structure. Specifically, we model net increases in firm leverage using market 1See Harris and Raviv (1991) and Graham and Leary (2011) for surveys of the capital structure literature. 2The literature linking equity markets to capital structure have been largely focused on using leverage to explain the cross-section of returns. See, e.g., Fama and French (1993), Dichev (1998), Vassalou and Xing (2004), Penman, Richardson,andTuna(2007),Campbell,Hilscher,andSzilagyi(2008),GomesandSchmid(2010),GeorgeandHwang (2010), and Kapadia (2011). We reverse the relationship by using equity risk measures to explain leverage changes. 1

expectations about risks estimated from equity options. Previous studies have documented the power of option prices in identifying investor expectations about the future performance of the underlying asset.3 Investor beliefs about the firm’s equity become impounded in option markets, increasing the price and implied volatility of certain option contracts relative to others (Bollen and Whaley, 2004; Garleanu, Pedersen and Poteshman, 2009). We follow recent findings in the option pricing literature to identify risk measures from these volatility differences across three orthogonal dimensions. Despiteinformativenessaboutfutureperformance, optionsdatahaveseenverylimited applications in corporate events, in which ex-ante market expectations should be a valuable signal. To the best of our knowledge, there have only been three such applications. They have been used to predict the likelihood of takeovers (Subramanian, 2004; Barraclough, Robinson, Smith, and Whaley, 2013; Borochin, 2014), to measure the impact of regulatory legislation (Borochin and Golec,2016),andtomeasureuncertaintyaboutthefirmaroundearningsannouncements(Dubinsky and Johannes, 2006). We add to this literature by using options to derive market-based measures of risk and apply these measures to the challenge of explaining, as well as predicting, within-firm capital structure changes. Using options-based risk measures offers three main advantages. First, they are direct and specific measures of risk based on stock and option price volatilities, rather than proxies of risk using firm characteristics. For a single underlying firm, there exists a variety of contracts with different features, such as type (call versus put) and moneyness (in-the-money versus out-of-themoney). These cross-sections of data allow us to examine orthogonal dimensions of risk in the same firm and to evaluate the importance that different types of risk in the firm’s leverage decisions. Second, they are forward looking and take investor attitudes and beliefs regarding the future of the firm into account, making them more relevant than backward looking book proxies of risk. This is important given that investor attitudes factor significantly in a firm’s ability to access external financing (McLean and Zhao, 2014) and adjust its capital structure. Third, market information is available at higher frequencies than a firm’s accounting filings, which are updated, at best, at a quarterly frequency. Given the time-varying nature of leverage changes (see Figure 1), this makes high-frequency explanatory variables potentially more informative, and therefore better able to 3See, e.g., Bakshi, Cao, and Chen (1997), Ait-Sahalia, Wang, and Yared (2001), Liu, Pan, and Wang (2005), Broadie, Chernov, and Johannes (2007), Cremers and Weinbaum (2010), and Xing, Zhang, and Zhao (2011). 2

address standing challenges in capital structure research.4 For these reasons, it is worthwhile to ask whether and how risk measures from the equity market can explain capital structure decisions. We use options data to derive three market-based measures that capture different dimensions of risk potentially relevant to a firm’s capital structure: 1) the spread between the implied volatility extrapolated from long maturity call options and realized volatility from historical returns to capture changes in perceived riskiness of the firm, 2) the implied volatility spread between short maturity out-of-the-money (OTM) and in-the-money (ITM) puts to capture expectations about a left tail or “crash” event in stock prices, and 3) the spread between the implied volatilities in short maturity calls and short maturity puts to capture expectations about the direction of future stock performance. These three measures are studied extensively in the asset pricing literature.5 Here, we demonstrate their power in explaining capital structure adjustments. To the extent that market expectations about these risks reflect or impact the firm’s cost of capital and access to financing, they will impact the firm’s decision to change its capital structure. Our main measure for capital structure changes is the net levering up ratio, defined as net debt issuance plus net share repurchases over total assets. This variable captures the debt and equity dynamics that lead to net changes in capital structure and summarizes these changes in the direction of increasing leverage in the firm. Importantly, net levering ratio eliminates mechanical changes in leverage driven by changes in equity value, isolating direct managerial decisions about leverage change. This measure (and its variations) has been used in prior capital structure research for this purpose (e.g. Kisgen, 2006; Binsbergen, Graham, and Yang, 2010; Leary and Roberts, 2014). Using the net levering up ratio to measure capital structure adjustments, we examine the power of our options-based measures in predicting capital structure decisions within a firm over the next quarter. To the extent that investors respond to perceived changes in a firm’s cash flow risk, we expect these measures to impact the firm’s capital structure decisions. Specifically, the riskier the firm, the larger the decreases in net leverage. We find a consistently negative relationship between measures of equity risk and future leverage 4Intheirrecentsurveyofthecapitalstructureliterature,GrahamandLeary(2011)drawattentiontotheinability ofstandardfirmcharacteristicstoexplainleveragechangeswithinfirms,aswellastheirdecliningexplanatorypower within industries. 5See, e.g., Goyal and Saretto (2009), Bali and Hovakimian (2009), Cremers and Weinbaum (2010), Xing, Zhang and Zhao (2010). 3

changes. The implied volatility spread between short maturity OTM and ITM puts, measuring equity tail risk, provides the strongest and most robust effect as a significant predictor of future leverage decreases. Additionally, higher spreads between the implied volatility of long maturity calls and realized volatility, as well as realized volatility by itself, also predict leverage decreases. A higher spread between the implied volatilities of short term calls and puts, as a measure of investor optimism, predicts leverage increases. These results are statistically and economically significant. In other words, firms with more (less) risky equity, and therefore more (less) cash flow risk, will decrease (increase) net leverage. It is important to note that this finding is opposite to what a mechanical relationship between implied volatility and leverage would imply, eliminating market anticipation of capital structure changes as a driving explanation. Furthermore, these option-based risk measures provide unique explanatory power in addition to that provided by standard controls using firm characteristics and backward-looking measures of risk based on accounting data. In other words, equity-based risk measures contain relevant information on firm risk for predicting capital structure decisions that is not available in existing accounting-based proxies for firm risk. The results indicate that market-based measures predict changes in capital structure in ways consistent with risk-based interpretations. To the extent that equity markets will influence the information regarding risk or availability of funds and resources, it will impact the firm’s real or perceived cash flow risk and therefore its corporate decisions such as access to external financing (McLean and Zhao, 2014) and, therefore, capital structure. Recent evidence by Foucault and Fresard (2016) also suggests that even firm managers rely on the equity market as a source of information regarding the firm. This strand of literature suggest that both firm insiders and outsiders can learn about firm risks from the equity market and influence decision-making within the firm. Our results are consistent with this interpretation. Theresultingchangesincapitalstructurepolicyreflectthecombinationofbothsupply-sideand demand-side financing concerns. To isolate supply-driven and demand-driven effects, we examine the predictive power of our measures in sub-samples of variable macroeconomic supply and firm demand for capital, respectively. In addition, expansionary and recessionary periods enable us to better identify the importance of our risk measures on levering up by exogenously shocking the supply of capital and injecting volatility into or removing it from the equity markets. We expect 4

our measures to have the most predictive power over sub-samples where cash flow risk matters most. Our results show that the predictive power of market-based measures on capital structure adjustments is strongest during periods of economic recession, when default is more likely and the marginal utility of investors is highest, and among growth firms, which have the most significant needforfinancing. Thisisconsistentwiththecounter-cyclicalityofleverageinKorajczykandLevy (2003), Hackbarth, Miao, and Morellec (2006), and Chen (2010). We obtain the highest adjusted R2 for high growth firms during economic downturns. That is, our options-based measures are most informative when predicting capital structure adjustments for high demand firms during low supply periods. One potential concern is that option risks will change simply due to market anticipation of capital structure change. If that were the case, we should observe a positive relationship between option-based risks and leverage, as a more levered firm is inherently more risky. We observe a consistently negative relationship, indicating that our findings are not merely driven by the anticipationofcapitalstructurepolicy,butaremeasuresofequityrisksthatlimitleverageincreases. This is consistent with Chen, Wang, and Zhou (2015) and Schwert and Strebulaev (2015). Another concern in using options-based measures is that only firms for which options data is available are included. We corroborate our main results by using simplified market-based measures thatdonotrelyonhavingdetailedoptionsdata. Inaddition, asstatedpreviously, oneadvantageof using market-based measures is their availability at higher frequencies. We therefore also consider a monthly, rather than quarterly, aggregation of our measures to predict net levering up decisions. Finally, we use a binary version of the net levering up ratio as the dependent variable to capture the direction, rather than magnitude, of capital structure changes and estimate the propensity to lever up using logistic analysis. This allows us to focus on the ability to lever up and abstract away from the decision of how much to lever up. In all cases, market-based measures retain significant explanatory power for net leverage changes (in excess of accounting controls) with more risky firms decreasing net leverage. Using our results, we propose new indices for predicting net leverage increases by taking linear combinations of our options-based measures, with and without controls. Sorting the indices into terciles, we examine the characteristics of firms that increase leverage. Firms that fall into the top 5

tercile of net levering up are larger, with higher Altman (1968) Z-scores, larger cash flows, and more likely to pay dividends and have credit ratings than firms that fall into the bottom tercile. In addition,firmsthatleveruphavecharacteristicsconsistentwiththoseassociatedwithlowfinancing constraints in the literature, and score lower on the Whited and Wu (2006) and Hadlock and Pierce (2010) indices of financing constraint. Next, we use these indices of net levering up to quantify the value of capital structure adjustments for shareholders. Sorting on the net leverage increase indices, we form equal-weighted portfolios of firms that fall into the top tercile of net levering up (HIGH) and firms that fall into the bottom tercile of net levering up (LOW). We follow their buy-and-hold abnormal returns over the following year as a measure of firm performance. On average, high levering up portfolios earn between 2.8% to 3.1% in abnormal returns over one year, while low levering up portfolios realize slightly negative abnormal returns over the year (averaging from -0.4% to -1.0%). As a result, a buy-and-holdtradingstrategyofbuyingthetoptercileandsellingthebottomtercilenetsabnormal returnsof3.2%to3.9%overthefollowingyear. Inotherwords, HIGHleveringupfirmsoutperform LOW levering up firms on average. The above results suggest that higher net leverage increases are associated with lower firm risk and higher returns. To explore the pricing implications of this observation, we create zerocost portfolios long on firms with LOW levering up and short on firms with HIGH levering up from our predictive indices. We find the zero-cost portfolios produce negative and significant monthly alphas when using equal-weighed portfolios and mostly insignificant monthly alphas when using value-weighted portfolios relative to the P´astor and Stambaugh (2003) 5-factor model. This corroborates the long-term performance results above. Although we use changes in leverage, these results are also consistent with the leverage and distress risk puzzles that find lower returns for firms with higher levels of leverage or higher bankruptcy risk.6 More broadly, they are consistent with superior performance by quality (lower-risk) firms in terms of profitability, share repurchases, low beta, high growth, and low accruals.7 6See, e.g., Fama and French (1993), Dichev (1998), Vassalou and Xing (2004), Penman, Richardson, and Tuna (2007),Campbell,Hilscher,andSzilagyi(2008),GomesandSchmid(2010),GeorgeandHwang(2010),andKapadia (2011). 7See Sloan (1996), Baker and Wurgler (2002), Mohanram (2005), Richarson, Sloan, Soliman and Tuna (2005), Pontiff and Woodgate (2008), McLean, Pontiff and Watanabe (2009), Novy-Marx (2013), Frazzini and Pedersen (2014), and Asness, Frazzini and Pedersen (2015). 6

To the best of our knowledge, this study is the first to use options-based measures of equity risk to predict changes in capital structure and to propose market-based indices for predicting leverage changes. Our results establish the usefulness of options data in providing risk-based interpretations for and predicting capital structure decisions. This allows us to create new market-based measures for capital structure adjustments. The predictive power of our measures establishes a promising connection between market expectations and capital structure decisions. This is particularly relevant in the wake of the financial crisis as investors and regulators re-evaluate the timeliness of accounting-based measures of firm risk (e.g., bank capital ratios) and consider market-based alternatives. Using market-based measures allows us to more directly study specific risk channels that impact financing behavior rather than rely on accounting proxies. Additionally, our approach allows us to bypass the issue of measuring changes in capital structure using limited and lowfrequency data due to real-time availability and updating of market data. Existing theories on the relationship between capital structure and firm risk, following the seminal work of Modigliani and Miller (1958) and the subsequent static tradeoff (Kraus and Litzenberger,1973)anddynamictradeoff(Fischer,Heinkel,andZechner,1989)approaches,suggest a positive relationship between leverage and firm risk. That is, if equity investors are reacting to expected managerial decisions to increase leverage, rather than managers reacting to increased investor perceived risks to the cash flows of the firm, we should observe a positive relationship between equity risk and leverage changes. In contrast, our main empirical result finds a negative relationship between leverage increases and risk. This apparent contradiction is addressed by George and Hwang (2010) who find that firms with low (high) bankruptcy costs and low (high) systematic risk are the ones that choose to hold high (low) leverage, resolving the leverage and distress puzzles. Notably, the theoretically implied positive relationship between leverage and equity risk suggests that our findings of a negative relationship between risk and future leverage changesaredrivenbyinvestorexpectationsoffirmriskandnotbymarketexpectationsofincreased leverage. 7

2 Data and Hypothesis Development 2.1 Options-based Measures of Risk We use previously established connections between the equity risk and capital structure dynamics (e.g., Myers, 1977; Myers, 1984; Marsh, 1982; Leland, 1994; Loughran and Ritter, 1995), mediated through the channel of cash flow risk (Campbell, 1991; Vuolteenaho, 2002), to create option-based measures of risk relevant to capital structure changes. Cash flow risk, whether factual or merely perceived, will increase the cost of debt and reduce the benefits of leverage. As a firm’s cash flow risk is not directly observable, our measures of equity risk provide potentially valuable proxies. Specifically, using options allows us to study the impact of different dimensions of cash flow risk on leverage changes. To form our option-based measures of equity risk, we use daily single-stock option data from OptionMetrics which covers all exchange-traded puts and calls and reports closing bid and ask prices and implied volatilities from 1996 onward. We aggregate daily implied volatility data from 1996to2012intoquarterlyaveragesbyoptiontype(callsversusputs),maturity(longversusshort), and moneyness (in-the-money versus out-of-the-money) to match the frequency of our accounting data. Accordingtothepricepressureargument(BollenandWhaley,2004; Garleanu,Pedersen,and Poteshman, 2009), the shapes of the implied volatility functions reflect excess demand for certain types of options. As a result, the implied volatility reflects the level of risk associated with the underlying asset of an option contract for specific values of option type, maturity, and moneyness. To isolate risks associated with these dimensions, we take differences in average quarterly implied volatilities by firm across each dimension. From these differences we construct three firm-specific implied volatility spread variables that capture expectations about the riskiness of the firm that we hypothesize to be relevant to capital structure adjustments. Each variable is calculated for all firms i in quarters t. 8

2.1.1 (Perceived) Changes in Risk Our first variable is the difference between the average implied volatility of long-term calls over the quarter and the historical, realized volatility over the year: IVspread = IV −RealizedVolatility (1) hist,i,t c,long,i,t i,t whereIVspread measuresthechangesinoverall(perceived)risksofafirm. Long-termoptions hist,i,t are those with more than 200 days to expiration. Excluding a maturity filter produces similar, but slightly weaker, results. RealizedVolatility , the historical volatility for the preceding year, is i,t a measure of the historical risk level of the firm. This measure provides the benchmark level for overall existing firm risk and therefore should be relevant to leverage decisions by itself. Goyal and Saretto (2009) and Bali and Hovakimian (2009) demonstrate that option implied volatilities deviate from historical levels based on investor beliefs about firm risk. Goyal and Saretto (2009) find evidence consistent with the Barberis and Huang (2001) hypothesis of investor overreaction: investors expect firms that have realized losses to be riskier in the future than firms that have realized gains, which causes a divergence and a subsequent reconvergence of implied and historical volatilities. We take an agnostic position on whether the deviation of implied volatility from the historical level is an overreaction or a rational expectation of firm risk on the part of investors and focus on this difference as an indicator of perceptions about changes in the riskiness of firms. A perceived increase in the expected riskiness of a firm, regardless of whether it is accurate, may be sufficient to limit the firm in its ability to increase leverage. Thus, the difference between implied and historical volatility is a relevant measure of firm risk and we hypothesize that a positive difference between implied and realized volatilities negatively affects a firm’s ability to increase leverage, consistent with a perceived increase in firm risk. Hypothesis 1. A positive spread between implied and realized volatilities is positively related with cash flow risk and therefore negatively correlated with an increase in firm leverage. 9

2.1.2 Tail Risk Oursecondvariableisthedifferencebetweenthequarterlyaverageimpliedvolatilitiesofshort-term, out-of-the-money (OTM) puts and short-term, in-the-money (ITM) puts: IVspread = IV −IV (2) mon,i,t p,OTM,i,t p,ITM,i,t where moneyness is defined as the ratio of the spot price to the strike price. Short-term options are those with less than 40 days to expiration. Out-of-the-money puts are those with moneyness less than 0.8 and in-the-money puts are those with moneyness greater than 1.2.8 Conceptually, IVspread captures the risk of left-tail or “crash” events. mon,i,t This measure is motivated by the famous implied volatility “smile” in index options, which is explained either by a price pressure argument on OTM put options as a form of insurance against the risk of a “crash” event (Bollen and Whaley, 2004; Garleanu, Pedersen, and Poteshman, 2009) or from the perspective of price drops due to stochastic volatility and jump processes (Bakshi, Cao, and Chen, 1997; Bates, 2000; Ait-Sahalia, Wang, and Yared, 2001; Liu, Pan, and Wang, 2005; Broadie, Chernov, and Johannes, 2007). In both interpretations a negative slope in the implied volatility function is indicative of the possibility of a crash: the more negative the slope, the bigger the crash. Therefore we consider the presence of an implied volatility smile in single-stock options as a signal of left-tail “crash” risk. If IVspread is positive, the out of the money puts are mon,i,t more valuable than ones in the money, indicating market concern about left-tail “crash” risk and therefore negatively affecting the firm’s ability to increase leverage. Hypothesis 2. A positive spread between short-term OTM and ITM put implied volatilities is positively correlated with cash flow risk and therefore negatively correlated with an increase in firm leverage. 8For robustness, we alternatively define {OTM, ITM} puts as those with moneyness: i) {less than 0.7, greater than 1.3}, and ii) {less than 0.9, greater than 1.1}. Our results are similar using these definitions. Defining OTM lessthan0.8andITMgreaterthan1.2yieldsthehighestexplanatorypowerbymakingtheoptimaltradeoffbetween dispersion in implied volatilities and number of observations. 10

2.1.3 Growth Expectations Finally, our third measure is the difference between quarterly averages of short-term call implied volatility and short-term put implied volatility: IVspread = IV −IV (3) cp,i,t c,short,i,t p,short,i,t where IVspread reflects expectations about the direction of firm performance. Short-term cp,i,t options are those with less than 40 days to expiration. Cremers and Weinbaum (2010) find that differences in call and put implied volatilities are a predictoroffuturefirmperformance. Informedinvestorsbuy(sell)acall(put)optionifperformance is expected to be positive, and buy (sell) a put (call) if it is to be negative. This price pressure causes the implied volatilities of call options to exceed those of puts for firms whose investors have optimistic outlooks, and the opposite for those whose investors are pessimistic. This gap in implied volatilities acts as a barometer of investor sentiment about the firm and provides an indicator of growth expectations. Regardless of whether this expectation is realized, it alone may affect a firm’s ability to obtain funds and change its capital structure.9 For example, McLean and Zhao (2014) find that a firm’s ability to obtain external financing is sensitive to the Baker and Wurgler (2006) investor sentiment index. As such, we hypothesize that a positive difference between the implied volatility of calls and puts positively affects a firm’s ability to lever up, consistent with a positive signal about expected performance. Hypothesis 3. A positive spread between short-term call and put implied volatilities indicates positive growth expectations, and therefore is positively correlated with an increase in firm leverage. 2.2 Modeling Changes in Capital Structure We examine the impact of the options-based measures on capital structure by studying the effect these measures have on the change in net leverage. We follow prior literature (e.g. Kisgen, 2006; Binsbergen, Graham, and Yang, 2010; Leary and Roberts, 2014) in using a deflated measure of changes in capital structure to summarize the decisions made within the firm. Firms may choose 9Cremers and Weinbaum (2010) do find significant abnormal performance in firms classified using their measure of difference in call and put implied volatilities, suggesting these expectations generally materialize. 11

to adjust their capital structure through issuing and paying down debt, and issuing or repurchasing equity. To account for these changes, we use the net levering up ratio of the firm as our main dependentvariable. Netleveringupratioisdefinedasnetdebtissuancesplusnetequityreductions (i.e., share repurchases) as a fraction of total assets: (D −D )+(E −E ) iss,i,t red,i,t red,i,t iss,i,t NLEVR = (4) i,t TA i,t where D is the long-term debt issuance for firm i over quarter t, D is the long-term debt iss,i,t red,i,t reduction,E istheequityreduction,andE istheequityissuance.10 Thisvariableaccounts red,i,t iss,i,t forbothdebtandequitycapitalstructureadjustmentsinthedirectionofincreasingleverageforthe firm. One advantage in using NLEVR is that it allows us to capture capital structure adjustments as explicitly reported by the firm, rather than rely on calculating changes in leverage ratios. That is, using such a measure allows us to exclude changes in leverage ratios that may be induced mechanically and not reflect actual capital structure adjustments.11 Equation(5)presentsthebaselinemodelweusetoexaminehowmarket-basedmeasuresexplain and predict changes in capital structure using the net levering up ratio, NLEVR, as our dependent variable:12 NLEVR = α+β X +fe +ε (5) i,t 1 i,t−1 t i,t where X reflects our three options-based measures, {IVspread , IVspread , and i,t−1 hist mon IVspread }, for firm i in quarter t − 1 (i.e., lagged one quarter). To obtain the previous cp quarter’s market-based variables, we calculate averages of the 3-month, 4-month, and 5-month lags of the options-based and historical volatility variables at monthly frequency.13 We expect 10For robustness, we also deflate using the firm’s total assets from the previous quarter. All results hold. 11Forexample,poorfirmperformancecanreducethevalueofequity,resultinginachangeinleverageratiowithout any capital structure decisions being made by the management. 12Studies of capital structure adjustments commonly employ partial adjustment models to study the speed of adjustmenttotargetleverages. See,e.g.,LearyandRoberts(2005),FlanneryandRangan(2006),HuangandRitter (2009), O¨ztekin and Flannery (2012), Faulkender, Flannery, Hankins, and Smith (2012). Given the use of NLEVR, webypasstheneedtocomputeatargetleverageortorelyoncomputedchangestoleverageratios. Whenfollowinga partialadjustmentmodelusingleverageratiosandtheBlundell-Bond(1998)GMMestimationframeworkwithfirm fixed effects, reassuringly, we find qualitatively similar results. 13We cannot use the 2-month, 1-month, or 0-month lags of option data in estimating the model since we want to establish a predictive relationship between market data and leverage change. This precludes the use of the current quarter’s price data since firm leverage may have changed at any point over the current quarter. However, contemporaneous option data may be used in explanatory (rather than predictive) applications. 12

IVspread and IVspread to be negatively correlated with net levering up behavior, as stated hist mon in Hypotheses 1 and 2, respectively, and IVspread to be positively correlated with net levering cp up behavior, in accord with Hypothesis 3. We examine all measures individually and jointly in our model specifications. We also include year fixed effects to absorb any other time-varying trends in the liquidity of capital markets and investor risk aversion, as well as quarter fixed effects to absorb cyclicality. Allstandarderrorsaretwo-wayclusteredbyfirmandyear-quarterasinPetersen(2009). It is important to note that our options-based measures reflect spreads. As such, when estimating equation (5) using our options-based measures, we also include the corresponding righthand side variable that is being differenced away. For example, when estimating equation (5) using IVspread , we also include RealizedVolatility in the estimation; when using IVspread , we hist mon include IV ; when using IVspread , we also include IV . This allows us to control for p,ITM cp p,short the baseline level of risk. Since factors other than the options-based risk measures may impact the capital structure decision, we include control variables commonly believed to impact capital structure in our full model, specified in equation (6). Prior literature has relied on accounting measures to proxy for cash flow risk.14 We include two such measures: the volatility of the prior five years (i.e., 20 quarters) of earnings normalized by total assets and the volatility of the prior five years of sales normalized by total assets.15 Additional firm-specific controls include the firm’s returns over the prior year, firm size, book-to-market ratio, Altman’s Z-score, the Blouin, Core, Guay (2010) marginal tax rate, long-term debt ratio, and the Whited and Wu (2006) financing constraint index. We also include the 3-digit SIC industry long-term debt ratio to control for industry influences and the credit spread to control for the economy-wide lending environment, in addition to time fixed effects. NLEVR = α+β X +β RealizedReturn +β lnTA +β BTM i,t 1 i,t−1 2 i,t−1 3 i,t−1 4 i,t−1 +β Zscore +β MTR +β σ +β σ +β LTDR (6) 5 i,t−1 6 i,t−1 7 Earnings,t−1 8 Sales,t−1 9 i,t−1 +β IndLTDR +β CredSpread +β WW +fe +ε 10 i,t−1 11 t−1 12 i,t−1 t i,t 14See, e.g. Bradley, Jarrell, and Kim (1984), Titman and Wessels (1988), Leary and Roberts (2005), Lemmon, Roberts, and Zender (2008), and Graham and Leary (2011). 15We also include the volatility of the prior five years of cash flows normalized by total assets. As it is highly correlated with the two mentioned, we remove this variable to reduce multicollinearity. 13

where X includes our three option-based measures and RealizedVolatility , RealizedReturn i,t−1 i is the firm’s cumulative monthly stock return over the prior year, lnTA is firm size measured by the natural log of total assets, BTM is the ratio of book equity to market equity, Zscore is the Altman’s (1968) Z-score that measures the financial health of a firm, σ is the 5-year Earnings volatility ofearningsnormalizedby totalassets, σ isthe 5-yearvolatility ofsalesnormalized by Sales total assets, MTR is the Blouin, Core, and Guay (2010) post-financing marginal tax rate, LTDR isthefirm’slong-termdebtratio, IndLTDR isthefirm’s3-digitSICindustrylong-termdebtratio, CredSpread is the credit spread between Moody’s Baa bonds and Moody’s Aaa bonds, and WW is the Whited and Wu (2006) measure of firm financing constraints. All control variables are lagged one quarter. As before, we include year and quarter fixed effects and the model is estimated with standard errors double clustered by firm and by quarter. Previous literature in capital structure suggests that large, value firms that are not in financial distress have higher leverage ratios. In addition, based on the static tradeoff theory of capital structure, interest deductibility of debt offers a tax benefit for using debt. Therefore, marginal tax rates and changes in tax rates are useful for isolating the demand for debt (Binsbergen, Graham and Yang, 2010; Farre-Mensa and Ljungqvist, 2014).16 The firm’s long-term debt ratio serves to benchmark the firm’s target leverage ratio and to control for any persistence in leverage levels (Lemmon, Roberts, and Zender, 2008). Furthermore, Frank and Goyal (2009) find that a firm’s long-term debt ratio is largely determined by its industry’s long-term debt ratio and Leary and Roberts (2014) find that firms tend to mimic the industry leverage ratio, suggesting that industry leverageplaysalargeroleinafirm’schosencapitalstructure. Thecreditspreadreflectsthecurrent macroeconomic environment and, as a result, the availability of funds in the economy. Finally, the Whited and Wu (2006) index captures a firm’s financial constraints, which may affect the firm’s ability to adjust its capital structure.17 Next, we modify our full model to study both the level and change effects of our main variables and control variables to the net levering up decision. To do this, we decompose the first lags of our option-based risk measures and control variables into the second lags of levels and first 16Graham and Mills (2008) note that while the pre-financing marginal tax rate is useful for explaining leverage levels, the post-financing marginal tax rate is appropriate for predicting changes in leverage. 17Forrobustness,weusetheHadlockandPierce(2010)size-ageindexinplaceoftheWhitedandWu(2006)index. All results hold. Both measures are highly correlated with firm size and with each other. 14

lags of changes following the identity relationship: X ≡ X +∆X . This modified i,t−1 i,t−2 i,t−2,t−1 specification splits each explanatory variable in equation (6) into two terms, a second lag of the level and a first lag of the difference: (cid:88) NLEVR = α+βX +γ∆X + (β c +γ ∆c )+fe +ε (7) i,t i,t−2 i,t−2,t−1 c i,t−2 c i,t−2,t−1 t i,t c∈C where X are our three option-based risk measures and RealizedVolatility , and C are our control i i i variables as previously defined in equation (6). The above analysis predicts the continuous measure of net levering up behavior, NLEVR. For robustness, we also define a dummy variable to capture any net increase in leverage as:    1 if NLEVR i,t > 0, NLEVD = (8) i,t   0 otherwise. While NLEVR allows us to study whether our various measures can explain the magnitude or degree of capital structure changes, NLEVD allows us to test whether our measures have power in explaining the levering up decision. Indeed, while it can be argued that the magnitude of capital structure adjustments is a combination of a firm’s demand and supply for financing, the binary decision of whether a firm levers up or not may be more indicative of its ability and restrictions to doing so. We use a logistic model to examine the impact of our market based measures and control variables on NLEVD: NLEVD = α+β X +β RealizedReturn +β lnTA +β BTM i,t 1 i,t−1 2 i,t−1 3 i,t−1 4 i,t−1 +β Zscore +β MTR +β σ +β σ +β LTDR (9) 5 i,t−1 6 i,t−1 7 Earnings,t−1 8 Sales,t−1 9 i,t−1 +β IndLTDR +β CredSpread +β WW +fe +ε 10 i,t−1 11 t−1 12 i,t−1 t i,t The predicted value from the logistic analysis provides us with the propensity score for whether the firm is likely to lever up. Panels A and B of Figure 1 describe the time-series variations of the net levering up ratio, NLEVR,andthebinarydecisiontonetleverup,NLEVD,respectively. Bothseriesofnetleverage changes exhibit significant time-series variation. While leverage levels are persistent (Lemmon, 15

Roberts and Zender, 2008), Figure 1 demonstrates that leverage changes vary substantially across time. This is consistent with the findings of Graham and Leary (2011) and highlights the potential usefulness of higher-frequency measures in explaining capital structure dynamics. 2.3 Financial Statement and Returns Data We obtain corporate financial statement data from Standard & Poor’s Compustat North American quarterly database from 1996 to 2012 and Moody’s Baa and Aaa rates from the Federal Reserve Board historical interest rate website. These databases are used to construct the net levering up ratio (NLEVR) and control variables discussed above. All dollar amounts are chained to 2000 dollars using CPI to adjust for inflation. We remove any firms with negative book asset value, market equity, book equity, capital stock, sales, dividends, debt, and inventory. Such firms have either unreliable Compustat data or are likely to be distressed or severely unprofitable. Although distressed and unprofitable firms are likely to be restricted from increasing leverage, financially constrained firms need not be distressed or unprofitable in general.18 In addition, we delete observations in which book assets or sales growth over the quarter is greater than 1 or less than -1 and firms worth less than $5 million in 2000 dollars in book value or market value to remove observations that have abnormally large changes due to acquisitions or small asset bases. Next, we remove outliers defined as firm-quarter observations that are in the first and 99th percentile tails for all relevant variables used in our analysis. Following standard practice in the literature, we remove all firms in the financial and insurance, utilities, and public administration industries as they tend to be heavily regulated. Our returns data comes from the daily and monthly CRSP database from 1995 to 2012. We measure realized volatility, RealizedVolatility , on the first of each month using a one-year i,t backward-lookingwindowofdailyreturns. Weannualizetheresultingstandarddeviationtoobtain the realized volatility for the preceding year. This captures the historical level of firm risk and is an input into IVspread , our measure for the perception of change in firm risk.19 Additionally, hist,i,t 18The distinction between financial distress and financing constraint has been drawn by prior work, e.g. Kaplan and Zingales (1997), Kisgen (2006), and Whited and Wu (2006). 19For robustness, we also consider conditional value-at-risk (CVaR) calculated over the previous year at the 1% levelasanalternativemeasureofhistoricalfirmrisk. Asexpected,thetwomeasuresarehighlynegativelycorrelated and yield consistent results. We retain realized volatility in our main specification due to superior significance and explanatory power. 16

we use monthly CRSP returns to compute RealizedReturn , by compounding monthly returns i,t over the prior year. The monthly CRSP database is also used for expected and abnormal returns following the five-factor returns model that includes the Pastor and Stambaugh (2003) liquidity factor. Finally,requiringtheresultingsampletocontainatleastonenon-missingoptions-basedmeasure gives us a sample of 5,087 firms spanning 110,456 firm-quarter observations between 1996 to 2012. To more accurately compare across model specifications, we restrict our sample to those with nonmissing observations for all relevant variables, giving us a sample of 3,700 firms spanning 56,041 firm-quarter observations. Variable definitions are provided in Appendix A. Table I provides the summary statistics for all relevant variables for both samples. Reassuringly, both samples appear to be statistically similar and with no obvious biases when restricting the sample. Table II provides the pairwise correlation between all relevant variables. The pairwise correlations between the three implied volatility spreads (rows (19) through (21)) are under 10%, consistent with a partitioning of risk into unique components and with previous findings from the option pricing literature. Furthermore, while the IV spread measures are largely uncorrelated, the implied volatility levels (rows (14) through (18)) are highly correlated with each other. In addition, the implied volatility levels are highly correlated with RealizedVolatility (row (13)), with correlations ranging from 71.1% to 83.3%. 3 Predicting Changes in Capital Structure In order to examine and compare the impact of the options-based risk measures on changes in net leverage, we test the significance and power of each measure, individually and jointly, on predicting a firm’s net levering up behavior as detailed in Section 2.2 using the restricted sample. 3.1 Baseline Model We start with our baseline model by regressing the net levering up ratio, NLEVR, on our optionsbased measures without any controls, as defined in equation (5). Table III present the results for our baseline model. Columns (1) through (3) report the coefficients for our three implied volatility spreads, IVspread , IVspread , and IVspread , and their corresponding right-hand side hist mon cp 17

volatility levels, respectively. If options-based measures contain unique information on investor beliefs regarding the cash flow risk of the firm, rather than simply reflecting expected managerial leverage decisions, we should expect to see higher risk levels resulting in decreases in net leverage. That is, we expect the coefficients on all volatility level variables to be negative as they reflect risk levels. Indeed, the results confirm that the riskier the firm actually is or perceived to be in terms of realized and implied volatility, the less the firm will lever up. Based on Hypotheses 1 and 2 motivated in Section 2.1, we expect the coefficients on IVspread and IVspread to be hist mon negative in columns (1) and (2), respectively, and find corroborating results. However, contrary to Hypotheses 3, the coefficient on IVspread is negative in column (3) of the baseline model. cp In column (4) of Table III, we include all options-based measures in one specification. Table II documentedthattheimpliedvolatilitylevelvariablesarehighlycorrelated,causingmulticollinearity concerns when combined into one model. To alleviate this issue, we use RealizedVolatility in place of all volatility level variables from columns (1) through (3) to account for the baseline risk level of the firm. This measure is negative and significant at the 1% level, consistent with the idea that firms with higher historical total risk are less likely to increase leverage going forward. Consistent with previous results, IVspread and IVspread remain negative and significant hist mon atthe1%and5%levels, respectively. However, thecoefficientonIVspread becomespositiveand cp significant at the 5% level, consistent with Hypothesis 3 and existing literature. This suggests that the previous negative coefficient in column (3) may contain omitted risks captured in IVspread hist and IVspread that is now controlled for under column (4). mon 3.2 Full Model Next, we present the findings for our options-based risk measures alongside accounting-based proxies of cash flow risk and other common control variables in Table IV. Column (1) shows the significance and explanatory power of the control variables alone in predicting levering up in firms. In general, the coefficients on the control variables are consistent with existing literature on capital structure and financial constraints. Large, financially healthy firms engage in larger net levering up. Moreover, firmswithhigherbook-basedmeasuresofcashflowrisk, earningsvolatilityandsales volatility, decrease net leverage. Consistent with Binsbergen, Graham and Yang (2010) and Farre- 18

Mensa and Ljungqvist (2014), firms with a higher marginal tax rate increase net leverage. Firms with high long-term debt ratios reduce net leverage, consistent with mean reversion of leverage (Lemmon,Roberts,Zender2008). Book-to-marketandfinancingconstraintsdonothavesignificant explanatory power for leverage changes controlling for the other common factors. Neither the industry leverage level nor credit spreads explain changes in leverage as well. Incolumns(2)through(4), weexaminetheeffectofourimpliedvolatilityspreads. Wecontinue to substitute RealizedVolatility, the historical level of firm risk, in all instances of the righthand side volatility level to alleviate multicollinearity concerns and for ease of interpretation. With the inclusion of control variables, the negative coefficient on IVspread in column (2) hist is -0.0103 significant at the 1% level, consistent with Hypothesis 1. This coefficient is highly economically significant: a one standard deviation change in IVspread results in a 150% (= hist 0.0103*0.146/0.001) change in net leverage increase relative to the median level of -0.001. It is important to note that this is a 150% change in the leverage change, not a 150% increase in leverage level. The coefficient on IVspread in column (3) is negative and significant at the 1% mon level. A one standard deviation change in IVspread results in a 66% change in net leverage mon increase relative to the median level. The coefficient on IVspread in column (4), though positive cp per Hypothesis 3, is insignificant both statistically and economically. The increase in adjusted R2 and reduction of significance in the intercept term in columns (2) through (4) relative to column (1) provides further evidence that the options-based measures have explanatory power in excess of existing controls for capital structure. In column (5) of Table IV, we include all three options-based measures along with the controls into one specification. The coefficients on IVspread and IVspread remain negative and hist mon highly significant at the 1% level. Similar to column (4) of Table III, when all three risk measures arecombinedthecoefficientonIVspread becomespositiveandsignificant, atthe10%level. This cp full model produces an adjusted R2 of 4.19%, which is the highest of all specifications considered. Our R2 results do not necessarily indicate that equation (6) is the optimal model specification for studying changes in leverage. However, they confirm that forward-looking option-based measures of cash flow risk provide unique information in predicting leverage increases that is not available in specifications that use only backward-looking book measures, and have high economic significance. 19

Next, wepresenttheresultstothemodifiedversionofthefullmodelinTableVbydecomposing the first lagged levels of the explanatory variables into first lagged differences and second lagged levels as in equation (7). This decomposition allows us to jointly examine the impact that changes of risk as well as levels of risk have on net leverage increases. The results are largely consistent with those previously discussed. Reassuringly, both levels and changes in our options-based measures IVspread and IVspread retain their power in explaining and predicting capital structure hist mon changesoverthenextquarterinexcessofthatprovidedbylevelsandchangesincontrols. Although IVspread becomes insignificant, it remains positive in both the change and level. These results cp confirmthattheoptions-basedriskmeasureshavepredictivepowerforleveragechangesinboththe change and level in addition to any risks proxied by lagged levels or changes in firm characteristics. Overall, our findings suggest that when investors expect firm risk to increase in the future relative to now (IVspread ) or when the firm is expected to experience higher probabilities of a hist crash event (IVspread ), the firm will decrease net leverage. On the other hand, when investors mon are more optimistic than pessimistic regarding the future of the firm (IVspread ), the firm will cp increase net leverage. These effects are significant on top of traditional accounting-based measures in predicting changes in capital structure. In other words, we document a negative relationship between risk and leverage increases. Higher levels, as well as increases, in options-based risk measures reflect investor belief regarding increases in the future cash flow risk of the firm, leading to higher costs of debt and, as a result, to reductions in net levering up behavior. However, if investors are expecting managers to increase leverage and these capital structure expectations, rather than cash flow risk expectations, drive options trading behavior, we should expect to see a positive relationship between the options-based measures of risk and net levering up behavior, based on existing capital structure theories that suggest leverage increases lead to higher equity risk. As such, our results suggest that options-based risk measures have significant informativeness about net increases in leverage, in excess of that provided by accounting-based control variables. Furthermore, this result is economically significant. 20

3.3 Supply and Demand Analysis So far, we show that our risk-based measures are broadly useful in predicting changes in capital structure. Because existing literature demonstrates that firm leverage decisions are made at the intersection of supply and demand (e.g. Faulkender and Petersen, 2005; Binsbergen, Graham and Yang,2010; Farre-MensaandLjungqvist,2014),itisimportanttounderstandthechannelsthrough which these risks affect changes in capital structure and the situations under which these measures may have the most impact on capital structure. To do this, we test the informativeness of our market-based measures for net increase in leverage in sub-samples that isolate variation in supply and demand. We consider variation in the supply of capital by repeating our main analysis during periods of macroeconomic expansion and contraction. Periods of macroeconomic expansion provide a greater availability of financing with more relaxed lending standards, with opposite effects during periods of macroeconomic contractions (e.g. Asea and Blomberg, 1998) and have been shown to matter for capital structure and capital raising (Korajczyk and Levy, 2003; Hackbarth, Miao, and Morellec, 2006; Chen, 2010; Almeida, Campello, Laranjeira, and Weisbenner, 2011; Erel, Julio, Kim, and Weisbach, 2012; Campello and Graham, 2013). In addition, expansionary and recessionary periods enable us to better identify the importance of our risk measures on levering up by exogenously shocking the supply of capital and injecting volatility into or removing it from the equity markets. We examine variation in demand by repeating our analysis on sub-samples of growth and value firms. Growth firms have a higher demand for additional financing compared to value firms with fewer growth opportunities. 3.3.1 Univariate Case First, we consider the effects of univariate variation of supply and demand for financing using the option-based measures. Table VI applies our full model in equation (6) to four independent subsamples. Columns (1) and (2) of Table VI display coefficient estimates for variation in capital supply. To capture this variation, we select sub-sample periods of macroeconomic expansion (1996Q1 through 1999Q4 and 2005Q1 through 2007Q2) and contraction (2001Q1 through 2002Q4 and2007Q3through2009Q2). Weselecttheperiodsrightbeforeandafterthedot-comandfinancial 21

marketcrashestoisolatethefastestratesofexpansionandcontraction. Theseeventshavebeenused for this purpose in existing literature. Campello and Graham (2013) use the technology bubble as a positive supply shock for financing; while Duchin, Ozbas, and Sensoy (2010) and Almeida, Campello, Laranjeira, and Weisbenner (2011) use the financial crisis as a negative supply shock. Our expansion and contraction periods capture the same shocks to tease apart the effects of supply variation.20 Column (1) of Table VI estimates the full model from equation (6) for net levering up over the expansion years and column (2) provides the estimates over the contraction years. All control variables have coefficients largely consistent with those observed in Table IV. Importantly, the coefficientestimatesfortheoptions-basedvariablesarelargelysimilartothoseobservedinTablesIII and IV. Specifically, both IVspread and IVspread are negative and significant in both high hist mon and low capital supply environments. While, IVspread is positive, it is insignificant for both cp expansionary and contractionary periods. The adjusted R2 for the boom years, 3.93%, is lower than that of the bust years, 4.70%. The higher quality of fit in column (2) relative to column (1) suggests that a risk-based model of net change in leverage has more predictive power in supply contractions.21 Consistent with Asea and Blomberg (1998), a relaxation of lending standards in boom periods enables a larger pool of firms to obtain financing, regardless of risk. Conversely, a tightening of lending standards means that firm risk becomes a more significant determinant of access to financing. In other words, cash flow risk matters more when capital supply contracts than when it expands, consistent with Hackbarth, Miao, and Morellec (2006) and Chen (2010). Columns (3) and (4) of Table VI examine the variation in firm demand for capital by sorting firms each quarter into terciles based on their book to market ratio. We expect low (high) BTM firms, or high (low) growth firms, to have stronger (weaker) demand for leverage in order to fund the expected growth. In performing this analysis, we implicitly assume that growth and value firms 20An alternative definition of boom and bust based on terciles of the credit spread as indicator of macroeconomic credit risk yields similar, though slightly weaker, results. 21One potential concern with market-based measures of risk is that they may fail when the market becomes more illiquid, such as during an economic recession. Our market-based measures have higher explanatory power during theeconomicdownturnsub-samplethaninboththefullsampleandeconomicexpansionsub-sample,demonstrating that this is not a problem in our application. 22

have different investment profiles which potentially leads to different capital structure dynamics.22 That is, we expect a growth firm’s capital structure to respond differently to changes in firm risk than that of a value firm at any given point in time. Therefore, we expect to see better (worse) predictive power of our measures among low (high) BTM firms. We find results consistent with this assumption. Column (3) of Table VI presents the results for low BTM (high growth) firms and column (4) presents the results for high BTM (low growth) firms. All three of the options-based measures are significant for the high growth firms in column (3) with IVspread and IVspread negative hist mon and significant at the 5% and 1% levels respectively and IVspread positive and significant cp at the 5% level, consistent with all three hypotheses. In contrast, there is weak explanatory power for the high BTM (low growth) firms with IVspread being negative and marginally hist significant at the 10% level. IVspread is insignificant with a coefficient close to 0 and although mon IVspread is significant at the 10% level, the coefficient is negative and runs counter to a riskcp based interpretation. Furthermore, the adjusted R2 for low BTM (high growth) firms is 6.68%, the highest of all samples previously considered; while the adjusted R2 for high BTM (low growth) firms is 2.29%, the lowest of all samples previously considered. These sub-sample results provide intuition and a sensibility check to our interpretation that implied volatility spreads contain useful information about firm cash flow risk. Realized volatility, representing total historical risk, is a significantly and consistently negative predictor of leverage increases in all supply and demand environments. However, forward-looking estimates of specific risks from our implied volatility spreads are able to explain net levering up behavior substantially better in environments where macroeconomic conditions are poor (low supply) and firms are more likely to seek out financing (high demand). That is, our model fits best during recessions and for high growth firms, during when and for whom cash flow risk matters most. 3.3.2 Bivariate Case Next,weinteractvariationinsupplywithdemandbyfurtherbreakingoursampleintofourbivariate sub-samples combining variation in supply with demand. Specifically, we create four sub-samples 22One motivation for this assumption is the q-theory of investment. See, e.g., Hayashi (1985). 23

permuting high and low growth firms within boom and bust periods, with each sub-sample as previously defined. Table VII present estimates of our full model in equation (6) to the four subsamples of bivariate supply/demand variation. Column (1) of Table VII, consisting of high growth firms in boom years, finds results largely consistent with prior analysis. However, IVspread , measuring tail risk, falls in significance mon to the 10% level, while IVspread , measuring (perceived) changes in risk, rises in significance hist at 1% level, relative to the results in column (3) of Table VI. In other words, when both supply and demand for capital are high, crash risk is somewhat less relevant to levering up, while overall changes in investor beliefs become more important. Column (2) repeats the analysis for low growth firms in boom years, with a substantial reduction in explanatory power, as expected based on the results from the univariate analysis. All options-based risk measures become insignificant and the R2 falls to 2.34%. This suggests that the model has less predictive power for changes in capital structure in firms with a low demand for financing, even during times of high capital supply. Column (3) of Table VII considers high growth firms in bust years. These are firms with the strongestneedforfinancingduringperiodswiththetightestcapitalsupply. Assuch,weexpectthat investor expectations about firm risks to be substantially valuable and informative in predicting leverage changes. Indeed, the explanatory power of our model is the highest among any of the previously considered samples with an R2 of 7.37%. This supports the results from the univariate analysis above: our market-based measures are most informative during situations where cash flow risk matters most. Consistent with solvency concerns during recessions, IVspread , or tail risk, mon is the sole significant determinant of leverage increases among the three options-based measures when demand is high but supply is low. Neither IVspread nor IVspread are significant; hist cp even RealizedVolatility is insignificant. Finally, column (4) presents results for low demand in low supply environments. As expected based on previous results, IVspread is negative and hist significant at the 10% level, IVspread is essentially zero and insignificant, and IVspread is mon cp negative and significant at the 10% level. The R2 in the sample is low at 2.27%. To test whether the coefficients on the options-based measures are statistically different from each other in the above sub-samples, shown in columns (1) through (4) of Table VII, we run seemingly unrelated regressions for each of the supply/demand sub-samples, and perform the χ2 24

testofdifferencesincoefficientsestimatedforoursupplyanddemandvariationsub-samples. These tests of differences between pairs of our four samples are reported in columns (5) through (8) of Table VII. Column (5) compares the boom period, high-growth sample coefficients from column (1) to those from the bust, high-growth sample in column (3). The 10% statistically significant difference on the coefficient on realized volatility between the two sub-samples suggests the importance of the overall level of firm risk in predicting capital structure changes due to variation in capital supply. That is, the predictive power of realized volatility is sensitive to the pro-cyclical supply of capital when the demand for capital is high. Column (6) compares the boom period, low-growth subsample coefficients in column (2) to those from the bust period, low-growth sub-sample in column (4). The absence of significant differences confirms that fluctuations in supply do not affect the importance of our measures for low-growth firms with low demand for financing. Column (7) compares the boom period, high-growth coefficients from column (1) to the boom period,low-growthcoefficientsfromcolumn(2),findingasignificantdifferenceinexpecteddirection of risk, IVspread . Comparing the bust period, high-growth coefficients in column (3) and bust cp period, low-growth coefficients in column (4), reported in column (8), finds this difference also as well as a stronger difference in IVspread . This suggests that variations in capital demand mon drive the predictability of IVspread and IVspread . In particular, directional risk captured cp mon by IVspread matters less for low-growth firms with low demand for financing. As observed cp previously,theabilityofIVspread ,tailrisk,inpredictingchangesinnetleveringupisstrongest mon among high-growth firms during bust periods when the supply of capital is tight. We compute Ftestsforoveralldifferencesincoefficientsonallfourofourmarket-basedriskmeasures. Ofthese,the difference between bust period, high-growth coefficients in column (3) and bust period, low-growth coefficients in column (4) is significant at the 10% level. The bottom line is that differences in the significance of the market-based risk measures exist between firms with high versus low growth in both boom and bust periods. However, differences in significance between boom and bust periods exist only for the high growth firms, but not for low growth firms. This is consistent with demand variation being the stronger determinant of the relevance of market-based risk measures in predicting capital structure adjustments. 25

3.4 Robustness So far, we have shown the usefulness of market-based measures in predicting changes in capital structure in the direction of increasing leverage. Here, we examine the robustness of these results. First, we relax the sample selection criteria to reduce potential sample bias. The restricted sample weuseintheaboveanalysisisdependentonhavingoptionimpliedvolatilitydata. Wetestwhether simpler market-based measures have explanatory power for predicting the net levering up ratio in cases where implied volatility data is unavailable. Second, one advantage that market-based measures have over accounting-based measures is the more frequent availability of data. We test whether the higher updating frequency of the market-based measures does in fact contribute to their superior explanatory power relative to accounting-based measures by examining monthly, rather than quarterly, data. Finally, as mentioned in Section 2.2, we use an indicator variable for levering up, NLEVD, as our dependant variable, rather than the continuous net levering up ratio, NLEVR. This allows us to test whether our measures have power in explaining the levering up decision. 3.4.1 Simplified Market-Based Measures Our previous analyses rely on firms having available options data to compute the three implied volatility spreads (IVspread , IVspread , and IVspread ). One concern is whether this hist mon cp limits the applicability of using market-based measures in predicting capital structure adjustments. A related concern is that by selecting firms with liquid option markets necessary to compute implied volatilities at both long and short horizons we may potentially have a biased sample. Here, we consider more basic characteristics of the option market and examine simpler measures that are more widely available. This serves three purposes: providing a less restrictive and therefore less biased sample, a consistency check for the hypothesis that options convey information about leverage changes, and more general market-based measures applicable to firms with sparse or unavailable option data. We introduce four simple options-based measures. First, we create a dummy variable, HasTradeableOptions, that indicates whether the firm has any positive open interest option contracts within the past three months. Second, we define LogTotalTradeableOptions as the 26

logarithm of the firm’s total option open interest contracts for both puts and calls. We take the daily count for both calls and puts and compute the quarterly average of the three monthly averagesofthiscount. Finally,wecreatetwomeasuresfortheliquidityofeachfirm’soptionmarket, LogTotalOpenInterest and LogTotalVolume. We calculate LogTotalOpenInterest by taking the logarithm of the quarterly average of the three monthly averages of daily open interest amount. LogTotalVolume is the logarithm of the traded volume for both puts and calls. These variables are set to zero for the firms without applicable options data, enabling us to analyze a substantially larger sample size with 183,032 firm-quarters in Table VIII. In other words, we consider whether the existence, size, and liquidity of a firm’s options market can increase the firm’s ability to lever up, hypothesizing a positive relation through superior information transmission. Column (1) of Table VIII reports that the existence of an option market on the firm’s stock, HasTradeableOptions, is insignificant to the degree of net levering up beyond that explained by RealizedVolatility and accounting controls. However, in column (2), the number of tradeable options, LogTotalTradeableOptions, increases net levering up behavior, with a positive coefficient that is significant at the 1% level. Results are similar for LogTotalOpenInterest, and LogTotalVolume, eachhavingapositiveandhighlysignificantcoefficient, aspresentedincolumns (3) and (4), respectively. These results support the idea that even the simpler measures based on the options market have power in explaining and predicting net levering up behavior. They also demonstrate how our methodology can be applied to firms with sparse or nonexistent implied volatility data that precludes the use of our main implied volatility spread measures. 3.4.2 Monthly Market-based Measures The preceding analysis finds that market-based measures contain information relevant to capital structure decisions in excess of that obtainable from accounting-based controls. Tables III through VII use the average implied volatility measures over the past quarter, i.e., the average implied volatility spread from five months ago to three months ago. We take the quarterly average for two reasons: first, to improve the number of observations in our sample to include firms that may not have data for all three months in the past quarter and second, to smooth any kinks in the options data. While it is reassuring to find relevance in market-based measures for predicting net 27

levering up behavior using the past quarterly average, in taking the average we lose one of the key features of using market data - the more frequent availability of information. For robustness, we restrictoursampletofirmswithoptionsdatainallthreemonthsinthepastquarterandrepeatour analysis using monthly option averages instead of quarterly averages in Table IX.23 This reduces our sample to 22,964 firm-quarter observations. If market-based measures indeed contain useful information about changes in net leverage, we should see that options data at monthly frequency, at minimum, retains explanatory power relative to the quarterly frequency and, at best, improves on it. Furthermore, we can observe how the informativeness of these measures in predicting capital structure changes evolves through time. Column (1) of Panel A of Table IX replicates the quarterly results from column (5) of Table IV for comparison with the higher frequency data. Column (2) of Panel A uses the fifth lag of monthly market-based measures, i.e., the end of the first month in the past quarter. This is the month immediately following the release of accounting data from two quarters ago. While IVspread hist and IVspread remain negative and highly significant at the 1% and 5% levels, respectively, mon IVspread loses significance entirely. However, R2 increases to 4.86% from 4.19% in column (1). cp Column (3) of Panel A uses the fourth lag of monthly market-based measures, i.e., the second month in the past quarter. Though the market has not received any new accounting information, the options data advanced by one month and we should expect significant findings if this fresher data is more informative. Indeed, we see that the significance for IVspread remains high at the hist 1% level. Moreover, the significance of IVspread improves to the 1% level. The adjusted R2, mon 4.82%, remains superior to using quarterly data in column (1). Finally, column (4) of Panel A uses the third lag of monthly market-based measures, i.e., the last month in the past quarter. This coincides with a release of new accounting information from onequarterago. IVspread losesitssignificance, suggestingtheimpactofconcurrentaccounting hist data. However, IVspread remains highly significant at the 1% level and adjusted R2 remains mon higherthancolumn(1)at4.74%. Itisreassuringthatusingthefifthandfourthlagsofmarket-based measuresprovidesmoresignificantresultscomparedtothethirdlagintermsofinformationcontent, as that data contains updated market information without corresponding updates to accounting 23We require all three months of data to be present to enable a meaningful comparison of the quality of fit across months. 28

information. This highlights the information advantage of using timely market-based measures. Onepossibleconcernisthatcapitalstructurechangesmaybedeterminedmorethanonequarter in advance, resulting in information leakage that is picked up by the prior quarter’s option prices, driving the results in previous tables. While this would not detract from the usefulness of options relative to accounting data in predicting firm leverage, it would reverse the relationship between risk and leverage increases, as discussed previously. We repeat the analysis in column (1) of Table IX with options data from two quarters ago (suppressed for brevity) and find similarly strong significance for the market-based risk measures on net levering up. This supports our interpretation that equity risks contain information that drives leverage changes (rather than the other way around). 3.4.3 Propensity to Lever Up So far, we use NLEVR as our dependent variable in measuring changes in capital structure. This capturesnotonlythedirectionbutalsothemagnitudethatmarket-basedriskmeasureshaveonthe net levering up ratio of the firm. For robustness, we define a binary variable, NLEVD, in equation (8) that takes the value of 1 for firms that increases net leverage in a particular quarter and 0 otherwise. This allows us to focus on the effectiveness of the options-based measures in predicting the decision to lever up while abstracting away from the decision on the degree of levering up. Columns (5) through (8) of Table IX present the results using NLEVD as the dependent variable in a logistic model, as detailed in equation (9). The control variables generally have the same qualitative effect on NLEVD as in the continuous case. However, the Whited and Wu (2006) index of financing constraints, WW, becomes significant, consistent with the interpretation of NLEVD as being more indicative of a firm’s ability and restrictions to levering up. In addition, the constant term becomes significant suggesting that some variation in NLEVD is not as well explained as in NLEVR. Column (5) of Panel B uses the quarterly averages of options data to predict the likelihood of levering up, analogous to column (1) for the continuous measure of net levering up. Using quarterly data, all three IVspread measures as well as RealizedVolatility are significant and with the expected signs. Columns (6) through (8) predicts the likelihood of levering up using the fifth, fourth, and third 29

lag of monthly market-based measures, respectively. Similar to the results using NLEVR, the monthly variables at higher frequency produce a higher quality of fit compared with the quarterly measures, with improved R2’s. Furthermore, both IVspread and IVspread are significant hist mon for the fifth and fourth lags while IVspread is not. As before, when examining the third lag, cp only IVspread and realized volatility remain significant, consistent with market measure losing mon some, but not all, predictive power with a new, concurrent release of accounting information. 4 Market-Based Indices of Leverage Increases The results from Section 3 provide evidence that market-based risk measures are useful predictors of changes in capital structure. In this section, we use our previous results to create market-based indices that predict net leverage increases in firms. These indices are linear combinations of our options-basedmeasuresandcontrolvariables. Itshouldbenotedthattheseindicesreflectaspecific and unique linear combination of its constituents (based on our previous results) and are therefore indices of predicted net levering up behavior.24 As such, these indices may or may not behave in ways different from its individual constituents. 4.1 Degree of Net Leverage Increase Indices In creating our indices, we use the variable coefficients from column (5) of Table IV. We retain statistically significant coefficients to create two market-based indices for net leverage increases. In the first index, we include only the three options-based measures and RealizedVolatility: LIMkt = −0.0100∗IVspread −0.0047∗IVspread 1,i,t hist,i,t−1 mon,i,t−1 (10) +0.0041∗IVspread −0.0177∗RealizedVolatility cp,i,t−1 i,t−1 where IVspread is the difference between the implied volatility of long-term calls and realized hist volatility, IVspread is the difference between the implied volatility of short-term out-of-themon money puts and short-term in-the-money puts, IVspread is the difference between the implied cp volatility of short-term calls and short-term puts, and RealizedVolatility is the realized volatility 24Forotherexamplesofpopularindicesproposedintheliterature,seeAltman(1968),KaplanandZingales(1997), Lamont,Polk,andSa´a-Requejo(2001),WhitedandWu(2006),HadlockandPierce(2010),andBinsbergen,Graham and Yang (2010). 30

of the underlying asset over the past year. The first measure gives us an index based only on market-based risk measures while controlling for common firm characteristics. In the second index for leverage increase we use the first index and additionally include the significant control variables: LIMkt = LIMkt −0.0022∗RealizedReturn +0.0017∗LnTA 2,i,t 1,i,t i,t−1 i,t−1 +0.0009∗Zscore −0.0114∗σ (11) i,t−1 Sales,i,t−1 −0.0166∗LTDR +0.0067∗IndLTDR i,t−1 i,t−1 where RealizedReturn is the realized stock return of the underlying asset over the prior year, LnTA is the natural log of total assets, Zscore is the Altman (1968) Z-score, σ is the 5- Sales year volatility of sales normalized by total assets, LTDR is the firm’s long-term debt ratio, and IndLTDR is the firms’ 3-digit SIC industry level of leverage. Both measures are increasing in net leverage and capture the direction and magnitude of net leverage increases. 4.2 Propensity to Lever Up Indices We use the predicted values from the logistic analysis using quarterly data from column (5) of Table IX to calculate the probability of an increase in net leverage, regardless of magnitude. This providesuswithtwoadditionalmarket-basedindicesthatcapturewhetherislikelytochangeitsnet leverage at all. As with the first two indices, we create an index using only the three options-based IVspread measures and RealizedVolatility and an index that also includes the control variables. This gives us our third and fourth measures for leverage changes: LIMkt = −0.8579−0.3344∗IVspread −0.2829∗IVspread 3,i,t hist,i,t−1 mon,i,t−1 (12) +0.3527∗IVspread −1.4776∗RealizedVolatility cp,i,t−1 i,t−1 and LIMkt = LIMkt −0.1186∗RealizedReturn +0.1215∗LnTA 4,i,t 3,i,t i,t−1 i,t−1 (13) +0.0425∗Zscore −0.9517∗WW i,t−1 i,t−1 31

where IVspread is the difference between the implied volatility of long-term calls and realized hist volatility, IVspread is the difference between the implied volatility of short-term out-of-themon money puts and short-term in-the-money puts, IVspread is the difference between the implied cp volatility of short-term calls and short-term puts, and RealizedVolatility is the realized volatility of the underlying asset over the past year. RealizedReturn is the realized stock return of the underlying asset over the prior year, LnTA is firm size measured by the natural log of total assets, Zscore is the Altman’s (1976) Z-score that measures the financial health of a firm, and WW is the WhitedandWu(2006)indexoffinancingconstraints. Theseindicesareincreasinginthelikelihood (rather than magnitude) of levering up. It is interesting to note that the Whited and Wu (2006) index of financing constraints affects ability of the firm to obtain leverage, but not the degree to which it increases leverage. Table X presents the descriptive statistics for our four market-based indices of leverage increase in Panel A. Panel B displays the pairwise correlation coefficients. The four market-based indices are strongly correlated with each other, ranging from 74.32% to 98.01%. 4.3 Risk Characteristics of Firms that Increase Leverage Equipped with four potential market-based indices for net leverage increases, LIMkt through 1 LIMkt based on equations (10) through (13) respectively, we examine the characteristics of firms 4 associated with increasing leverage. To do this, in each quarter we sort firms based on LIMkt 1 into three equal-sized bins: LOW, MED, and HIGH. The LOW bin reflects firms with values of LIMkt falling into the bottom tercile in any given quarter, i.e., firms with lower net levering up 1 ratios. The HIGH bin reflects firms with values of LIMkt ranking in the top tercile in any given 1 quarter, i.e., firm with higher net levering up ratios. We repeat this procedure to create LOW, MED, and HIGH bins based on the other three LIMkt measures. Table XI compares common characteristics for firms in the LOW and HIGH bins of LIMkt , 1 which uses only market-based risk measures, and for LIMkt which uses both market-based risk 2 measures and control variables.25 The differences in characteristics between LOW and HIGH firms in Table XI demonstrate significant spreads in firm quality and riskiness, as expected given the 25TheresultsforLIMkt andLIMkt areverysimilar,consistentwiththehighcorrelationsobservedinTableX, 3 4 and are therefore suppressed for brevity. 32

creation of these indices based on our option-based measures of risk. Firms ranking in the bottom tercile of net levering up based on LIMkt are smaller with lower Altman Z-scores and lower cash 1 flows than firms ranking in the upper tercile of LIMkt . These firms are also less likely to pay 1 dividends,tohaveinvestment-gradecreditratings,andtohaveinvestment-graderatingsconditional on having a rating. Furthermore, firms ranking LOW hold more cash and have higher Whited and Wu (2006) and Hadlock and Pierce (2009) size-age indices, consistent with facing more financing constraints. The LOW LIMkt firms also have lower leverage, consistent with George and Hwang 1 (2010) who find that leverage and firm risk are endogenous and firms with higher leverage are less risky resulting in their ability to obtain leverage. All differences in mean firm characteristics between the LOW and HIGH firms are statistically significant at the 1% level. Many of these quality and characteristics have been documented to have performance implications in several distinct areas of the literature. Their consistent observation is that highquality firms exhibit stronger abnormal performance. Firms that rank HIGH in net leverage increases have several key characteristics associated with low risk and abnormally positive performance: lower leverage (Penman, Richardson and Tuna, 2007; George and Hwang, 2010), higher Z-score and credit ratings (Altman, 1968; Campbell, Hilscher, Szilagyi, 2008), higher profitability (Novy-Marx, 2013), and higher dividend payouts (Asness, Frazzini, Pedersen, 2015). All these characteristics are associated with low risk, high quality, and, paradoxically, superior performance (Asness, Frazzini, Pedersen, 2015). Selecting firms by predicted leverage increases based on the LIMkt index captures all of these desirable characteristics simultaneously, and from 1 using market data alone. The comparisons when sorting on LIMkt are largely the same with the exception of leverage. 2 Recall that LIMkt includes the significant control variables, one of which is the long-term debt 2 ratio. That is, LIMkt also controls for existing debt usage. Interestingly, firms ranking LOW 2 based on LIMkt have more leverage than firms ranking HIGH on LIMkt . This may reflect 2 2 increased expected default risks due to having higher leverage, as evidence by the substantially lower Altman’s Z-score. This is consistent with Whited and Wu (2006) who find that existing leverage is a significant determinant of financing constraints. In either case, firms that rank LOW in net levering up have characteristics consistent with having higher cash flow risk and being riskier 33

firms and firms that rank HIGH in net levering up are associated with characteristics consistent with lower cash flow risk and being less risky firms. These results suggest that the market-based LIMkt indices are useful for identifying firms with high-risk characteristics associated with limited access to capital markets. 5 The Value of Capital Structure Adjustments In their survey, Graham and Harvey (2001) find that the ability to adjust capital structure is ranked as a top concern by CFOs. Our results establish a strong connection between equity risks, firm quality, and capital structure adjustments. The tercile sorts in Table XI identify the HIGH net leverage increase tercile as significantly less risky than those in the LOW net leverage increase tercile, and these same risk characteristics have been shown in prior literature to have significant performance implications. Therefore, the firm’s ability to increase leverage should have an economically significant impact for shareholders. In this section we explore the performance and value implications of the market-based leverage increase indices defined previously. That is, we measure firm performance by quantifying shareholder value of the firm’s expected increase in net leverage. For this section of the analysis we create rolling versions of our LIMkt indices based on the past five years (20 quarters). This avoids any look-ahead bias in our performance findings, and provides out-of-sample testing for the indices. The results are qualitatively very similar to those obtained by using the coefficients in equations (10) through (13) estimated on the full sample. 5.1 Predicted Leverage Increases and Firm Performance To measure firm performance and quantify the value of the firm’s expected net increase in leverage, we consider whether a buy-and-hold strategy using our market-based indices generates abnormal returns. First, we create monthly indices for our market-based indices introduced above, LIMkt 1 to LIMkt in equations (10) through (13), using information available at the beginning of the 4 month. As in Table XI, we sort each of our four LIMkt indices into equal-sized LOW, MED, and HIGH bins, with monthly rebalancing. We follow a buy-and-hold strategy by compounding abnormal returns over the 12 months following portfolio creation. Abnormal returns are calculated 34

based on the coefficients obtained from the P´astor and Stambaugh (2003) 5-factor model using a rolling window of the prior 60-months. Panels A through D of Figure 2 plot the buy-and-hold abnormal returns for the HIGH, LOW, and HIGH-LOW leverage increase portfolios based on the four LIMkt indices respectively. For LIMkt in Panel A, buy-and-hold abnormal returns for the HIGH bin reaches 1.5% by month six 1 and 2.8% by month 12, while the LOW bin declines slightly to -0.4% by month six and remains at -0.4% by month 12. Similarly, for LIMkt through LIMkt , the HIGH leverage increase 2 4 portfolio generates buy-and-hold abnormal returns of 3.1%, 2.9%, and 2.9% over the following year, respectively, while the LOW leverage increase portfolio returns -0.4%, -0.9%, and -1.0% over the same period. In other words, the HIGH leverage increase portfolios outperforms the LOW leverage increase portfolios over the following year and a buy-and-hold trading strategy of buying the tercile with the highest net leverage increase and selling the tercile with the lowest net leverage increase nets abnormal returns of 3.18%, 3.48%, 3.84%, and 3.89% over one year based on LIMkt 1 through LIMkt , respectively. These abnormal returns are driven almost entirely by the HIGH 4 leverageincreaseportfolios. Furthermore, LIMkt generatesaHIGHminusLOWabnormalreturn 1 of 3.18% in comparison to the 3.48% generated by LIMkt . This suggests that the abnormal 2 returns are almost entirely driven by the market-based risk measures (IVspread , IVspread , hist mon IVspread , and RealizedVolatility) rather than by the control variables. cp Atfirstglancetheseresultsmayseemtrivial,consideringthatourindicesarefunctionsofimplied volatilityspreadsthathavealreadybeenshowntopredictabnormalreturnsintheoptionsliterature from which they originate.26 However, recall that we use these implied volatility spreads to first predict the degree of leverage increases in LIMkt and LIMkt or propensity to increase leverage 1 2 inLIMkt andLIMkt . Then, weformportfoliosbasedonthepredictedleverageincreaseindices. 3 4 As a result, we isolate net leverage increases as the channel affecting future realized returns, using specific linear combinations of these implied volatility spreads. While each individual IV spread has been shown to explain returns in the literature, this by itself does not imply that our indices of predicted net levering up would do so as well. In other words, it is the predicted net increase in leverage, consistent with lower risk and higher quality as demonstrated in Table XI, that results in 26See Bali and Hovakimian (2009), Cremers and Weinbaum (2010), and Xing, Zhang, and Zhao (2010). 35

between 3.2% and 3.9% abnormal firm performance. 5.2 Net Levering Up Indices and the Quality Anomaly Previous results document a robust, negative relationship between options-based risk measures and net levering up. This relationship is corroborated by Table XI, which shows that firms that fall into the bottom tercile of levering up sorting on the LIMkt indices are smaller, have lower Altman’s Z-scores, lower cashflows, and are less likely to pay dividends and have credit ratings, all consistent with having higher risk and lower quality. In other words, firms that are more (less) risky in terms of the three volatility spreads and realized volatility lever up less (more). Given these findings, it is natural to ask whether the quality anomaly observed in similar characteristics by Asness, Frazzini, and Pedersen (2015) also exists here. As before, we start by creating monthly indices for LIMkt through LIMkt using information 1 4 available at the beginning of each month and sort each of the four LIMkt indices into equal-sized LOW,MED,andHIGHbinseachmonth. Weformzero-costportfoliosbytakingtheaveragereturn of the LOW (higher risk and lower quality) bin minus the average return of the HIGH (lower risk and higher quality) bin for each index, giving us a time series of monthly returns for each portfolio longonfirms thatfallintothetop tercileofpredictednetleverageincreasesandshort onfirmsthat fall into the bottom tercile of predicted net leverage increases.27 If a quality anomaly exists here, we should expect to see significant and negative alpha consistent with a negative risk premium on the riskier firms that have low net leverage increases. We create both equal-weighted and valueweighted portfolios and benchmark performance using the P´astor and Stambaugh (2003) 5-factor expected returns model estimated using a rolling five-year window. Table XII presents the alphas for portfolios formed on sorts of LIMkt through LIMkt . 1 4 Panel A of Table XII presents the monthly alphas for the zero-cost portfolio in our full sample usingequal-weightedportfoliosincolumns(1)through(4)andvalue-weightedportfoliosincolumns (5) through (8). An equal-weighted portfolio based on LIMkt generates a -0.98% monthly alpha 1 in column (1), which is significant at the 1% level. Zero-cost portfolios based on LIMkt through 2 LIMkt generatemonthlyalphasof-1.54%,-1.39%,and-1.48%respectivelyincolumns(2)through 4 27Toavoidpossiblecontaminationduetodelayedreleaseofthepreviousquarter’saccountingdata,wealsomeasure returns three months after portfolio formation. The results are similar to those presented. 36

(4),allsignificantatthe1%level,consistentwiththequalityanomalydiscussedinAsness,Frazzini, Pedersen (2015). Panels B and C of Table XII present the 5-factor alphas for the LOW and HIGH halves of the zero-cost portfolios from Panel A, respectively. There are highly significant negative alphas for all LOW portfolios in columns (1) through (4) of Panel B and significant positive alphas for the equal-weighted HIGH portfolios in Panel C, consistent with the results in Figure 2. When we turn to the value-weighted portfolios in columns (5) through (8) of panel A, for LIMkt , which uses only the market-based risk measures without controls, there is a positive 1 alpha, significant at the 5% level, inconsistent with a quality anomaly. However, for LIMkt 2 through LIMkt , there are insignificant alphas. The lack of abnormal returns is driven by the 4 positive and significant alphas generated in both the LOW and HIGH groups in panels B and C off-setting each other, in contrast to the equal-weighted portfolios. Value-weighting emphasizes the effect of large firms in both terciles, compressing the cross-section of quality which correlates with size, as demonstrated in Table XI, and therefore compresses the variation of returns. Taken altogether, the findings in Table XII suggest that, unconditional on firm size (as is the case with equal-weighted portfolios), more risky, leverage decreasing (LOW) firms display significant underperformance which drives the negative alphas of the zero-cost portfolios. Theseresultsareconsistentwiththeleverageanddistressriskpuzzlesthatfindlowerreturnsfor firms with higher leverage or higher default risk (Fama and French, 1993; Dichev, 1998; Vassalou and Xing, 2004; Penman, Richardson, and Tuna, 2007; Campbell, Hilscher and Szilagyi, 2008; George and Hwang 2010). As the ability to increase leverage correlates with financial constraints (as seen in Table XI), this also relates to the financial constraints puzzle which finds that firms with higher financing constraints earn lower returns than firms with lower constraints (Lamont and Polk, 2001; Whitedand Wu, 2006). Theseresults arealsoconsistentwithrecentfindingsonpricing anomalies in which quality firms with lower risk, as measured across several firm characteristics, outperform their high-risk counterparts (Novy-Marx, 2013; Frazzini and Pedersen, 2014; Asness, Frazzini, and Pedersen, 2015). According to our findings the leverage decision, as forecasted using options market data, encapsulates these characteristics. 37

6 Conclusion We provide new evidence connecting market-based measures of equity risk to firm capital structure decisions. To the extent that equity market prices are informative about the firm’s cash flow risk, market-based measures of equity risk should be informative regarding the firm’s capital structure decisions. We recover investor expectations about risks relevant to financing constraint from option prices and demonstrate their predictive power for changes in firm leverage. Using options-based measures to explore changes in leverage has three main advantages. First, options-based measures reflectinvestorandmarketattitudesandexpectationsregardingfuturerisksofthefirmandassuch are forward-looking. Second, market information is updated and available more frequently than book-based measures. Third, having various types of options associated with one underlying asset allows us to measure distinct dimensions of risk for the same firm, enabling us to directly measure risks rather than proxy for them using accounting-based firm characteristics. Optionimpliedvolatilityspreadsthatcaptureinvestorperceptionsofthechangeinrisk,left-tail “crash” risk, and direction of risk have significant predictive power for future leverage changes at the firm. This informativeness persists in the presence of historical volatility and accounting-based controls including book-based measures of cash flow risk and financing constraints. These effects become particularly strong when firm demand for financing is high, especially during periods of economic contraction. The results are robust to using simpler option measures and monthly option measures. Furthermore, option based measures are useful for explaining both the degree as well as the decision to lever up. We demonstrate that our market-based measures identify lower-quality firms with higher credit risk, lower cash flows, lower payouts, and smaller size as significantly more unlikely to increase leverage. This is consistent with cash flow risk as the channel linking equity risks identified in our implied volatility measures to capital structure decisions, and supports earlier findings (e.g. George and Hwang, 2010) suggesting that capital structure is endogenous in firm risk. Finally, weexamineabnormalbuy-and-holdfirmperformancetoquantifythevalueofthefirm’s ability to make capital structure adjustments, identified as qualitatively valuable in the Graham and Harvey (2001) CFO survey. The 12-month abnormal return on buying a portfolio of firms in the upper tercile of predicted net leverage increases while shorting the portfolio of firms in the 38

lower tercile of predicted net leverage increases earns from 3.2% to 3.9% over the next year. We find that a zero-cost portfolio long firms that are not predicted to increase leverage and short firms that are generates a negative risk premium using equal-weighted portfolios, suggesting a negative risk premium on the inability to increase leverage, consistent with the leverage and distress puzzles as well as other firm quality-related anomalies. These findings provide promising insight into the link between market-based estimates of investor expectations outside the firm and managerial decision-making within it. Our results show that these market-based measures capture information that is not contained in established accounting-based measures and provide better risk based measures than those proxied from firm characteristics. They are also more broadly suggestive of additional potential applications for the use of market data in estimating corporate decisions previously treated only with book-based measures. One logical extension is to the area of financing constraints. The Whited and Wu (2006) accounting-based index has been shown to have predictive power for the binary decision to increase leverage, alongside our market-based risk measures. Furthermore, the firms we identify as likely to increase leverage have higher Whited and Wu (2006) and Hadlock and Pierce (2010) scores. The present application of estimating the otherwise ex ante unobservable ability of firms to increase leverage is just one case of a potentially large set of connections between investor expectations and firm operations and the risks and frictions that moderate these connections. 39

References Ait-Sahalia,Y.,Y.Wang,andF.Yared,2001,DoOptionsMarketsCorrectlyPricetheProbabilities of Movement of the Underlying Asset?, Journal of Econometrics, 102, 67-110 Almeida, H., M. Campello, B. Laranjeira, and S. Weisbenner, 2011, Corporate Debt Maturity and the Real Effects of the 2007 Credit Crisis, Critical Finance Review, 1, 3-58. Almeida, H., M. Campello, and M. Weisbach, 2004, The Cash Flow Sensitivity of Cash, Journal of Finance, 59, 1777-1804. Altman, E., 1968, Financial Ratios, Discriminant Analysis, and the Prediction of Corporate Bankruptcy, Journal of Finance, 23, 589-609. Asea, P., and B. Blomberg, 1998, Lending cycles, Journal of Econometrics, 83, 89-128. Asness, C., A. Frazzini, and L. Pedersen, 2015, Quality Minus Junk, working paper. Bakshi, G., C. Cao, and Z. Chen, 1997, Empirical Performance of Alternative Option Pricing Models, Journal of Finance, 52, 2003-2049. Bali, T., and A. Hovakimian, 2009, Volatility Spreads and Expected Stock Returns, Management Science, 55, 1797-1812. Baker, M., and J. Wurgler, 2002, Market Timing and Capital Structure, Journal of Finance 57, 1-32. Baker, M., and J. Wurgler, 2006, Investor Sentiment and the Cross-Section of Stock Returns, Journal of Finance 61, 1645-1680. Barberis, N., and M. Huang, 2001, Mental Accounting, Loss Aversion, and Individual Stock Returns, Journal of Finance, 56, 1247-1292. Barraclough, K., D. Robinson, T. Smith, and R. Whaley, 2013, Using Option Prices to Infer Overpayments and Synergies in M&A Transactions, Review of Financial Studies, 26, 695- 722. Bates, D., 2000, Post-’87 Crash Fears in the S&P 500 Futures Options Market, Journal of Econometrics, 94, 181-238. Binsbergen, J.H. van, J.R. Graham, and J. Yang, 2010, The Cost of Debt, Journal of Finance, 65, 2089-2136. Blouin, J., J. Core, and W. Guay, 2010, Have the Tax Benefits of Debt Been Overestimated?, Journal of Financial Economics, 98, 195-213. Blundell, R. and S. Bond, 1998, Initial Conditions and Moment Restrictions in Dynamic Panel Models, Journal of Econometrics, 87, 115-143. Bollen, N., and R. Whaley, 2004, Does Net Buying Pressure Affect the Shape of Implied Volatility Functions?, Journal of Finance, 59, 711-753. Borochin, P., 2014, When Does A Merger Create Value? Using Option Prices to Elicit Market Beliefs, Financial Management, 43, 445-466. 40

Borochin, P., and J. Golec, 2016, Using Options to Measure the Full Value-Effect of an Event: Application to Obamacare, Journal of Financial Economics, 120, 169-193. Bradley, M., G. Jarrell, and E. Kim, 1984, On the Existence of an Optimal Capital Structure: Theory and Evidence. Journal of Finance, 39, 857-878. Broadie, M., M. Chernov, and M. Johannes, 2007, Model Specification and Risk Premia: Evidence from Futures Options, Journal of Finance, 62, 1453-1490. Campbell, J.Y., 1991, A Variance Decomposition for Stock Returns, Economic Journal, 101, 157- 179. Campbell, J., J. Hilscher, and J. Szilagyi, 2008, In Search of Distress Risk, Journal of Finance, 63, 2899-2939. Campello, M., and J.R. Graham, 2013, Do Stock Prices Influence Corporate Decisions? Evidence from the Technology Bubble, Journal of Financial Economics, 107(1), 89-110. Chen, H., H. Wang, and H. Zhou, 2015, Stock Return Volatility and Capital Structure Decisions, working paper. Chen, H., 2010, Macroeconomic Conditions and the Puzzles of Credit Spreads and Capital Structure, Journal of Finance, 65, 2171-2212. Cremers, M., and D. Weinbaum, 2010, Deviations from Put-Call Parity and Stock Return Predictability, Journal of Financial and Quantitative Analysis, 45, 335-367. Dichev, I., 1998, Is the Risk of Bankruptcy a Systematic Risk?, Journal of Finance, 53, 1131-1147. Dubinsky, A., and M. Johannes, 2006, Fundamental Uncertainty, Earnings Announcements, and Equity Prices, Columbia University working paper. Duchin, R., O. Ozbas, and B. Sensoy, 2010, Costly External Finance, Corporate Investment, and Subprime Mortgage Credit Crisis, Journal of Financial Economics, 97, 418-435. Erel, I., B. Julio, W. Kim, and M.S. Weisbach, 2012, Macroeconomic Conditions and Capital Raising, Review of Financial Studies, 25, 341-376. Fama, E., andK.French, 1993, CommonRiskFactorsintheReturnsonStocksandBonds, Journal of Financial Economics, 33, 3-56. Farre-Mensa, J., and A. Ljungqvist, 2014, Do Measures of Financial Constraints Measure Financial Constraints?, working paper. Faulkender, M., M.J. Flannery, K.W. Hankins, and J.M. Smith, 2012, Cash Flows and Leverage Adjustment, Journal of Financial Economics, 103, 632-646. Faulkender, M., and M.A. Petersen, 2006, Does the Source of Capital Affect Capital Structure?, Review of Financial Studies, 19, 45-79. Fischer, E., R. Heinkel, and J. Zechner, 1989, Optimal Dynamic Capital Structure Choice: Theory and Tests, Journal of Finance, 44, 19-40. Flannery, M.J., K.P. Rangan, 2006, Partial adjustment toward target capital structures, Journal of Financial Economics, 79, 469-506. 41

Foucault,T.,andL.Fresard,2016,CorporateStrategy,Conformism,andtheStockMarket,working paper. Frank, M.Z., and V.K. Goyal, 2009, Capital Structure Decisions: Which Factors are Reliably Important?, Financial Management, 38, 1-37. Frazzini, A., and L. Pedersen, 2014, Betting against Beta, Journal of Financial Economics, 111(1), 1-25. Friesen, G., Y. Zhang, and T. Zorn, 2012, Heterogeneous Beliefs and Risk-Neutral Skewness, Journal of Financial And Quantitative Analysis, 47, 851-872. Garleanu, N., L. Pedersen, and A. Poteshman, 2009, Demand-Based Option Pricing, Review of Financial Studies, 22, 4259-4299. George, T.J., and C. Hwang, 2010, A Resolution of the Distress Risk and Leverage Puzzles in the Cross Section of Stock Returns, Journal of Financial Economics, 96, 56-79. Gomes, J.F., L. Schmid, 2010, Levered Returns, Journal of Finance, 65, 467-494. Goyal, A., and A. Saretto, 2009, Cross-Section of Option Returns and Volatility, Journal of Financial Economics, 94, 310-326. Graham, J. R., and C. Harvey, 2001, The theory and practice of corporate finance: evidence from the field, Journal of Financial Economics, 60(2), 187-243. Graham, J.R., and M.T. Leary, 2011, A Review of Capital Structure Research and Directions for the Future, Annual Review of Financial Economics, 3, 309-345. Graham, J.R., and L. Mills, 2008, Using Tax Return Data to Simulate Corporate Marginal Tax Rates, Journal of Accounting and Economics, 46, 366-388. Hackbarth, D., J. Miao, and E. Morellec, 2006, Capital structure, credit risk, and macroeconomic conditions, Journal of Financial Economics, 82, 519-550. Hadlock, C.J., and J.R. Pierce, 2010, New Evidence on Measuring Financial Constraints: Moving Beyond the KZ Index, Review of Financial Studies, 23, 1909-1940. Harris, M., and A. Raviv, 1991, The Theory of Capital Structure, Journal of Finance, 46, 297-355. Hayashi, F., 1985, Corporate Finance Side of the Q Theory of Investment, Journal of Public Economics, 27, 261-280. Huang, R. and J.R. Ritter, 2009, Testing Theories of Capital Structure and Estimating the Speed of Adjustment, Journal of Financial and Quantitative Analysis, 44, 237-271. Kapadia, N., 2011, Tracking Down Distress Risk, Journal of Financial Economics, 102, 167-182. Kaplan, S., andL.Zingales, 1997, DoFinancingConstraintsExplainWhyInvestmentisCorrelated with Cash Flow?, Quarterly Journal of Economics, 112, 169-215. Kisgen, D.J., 2006, Credit Ratings and Capital Structure, Journal of Finance, 61, 1035-1072. Korajczyk, R.A., and A. Levy, 2003, Capital Structure Choice: Macroeconomic Conditions and Financial Constraints, Journal of Financial Economics, 68, 75-109. 42

Kraus, A., and R.H. Litzenberger, 1973, A State-Preference Model of Optimal Financial Leverage, Journal of Finance, 28, 911-922. Lamont, O., C. Polk, and J. Sa´a-Requejo, 2001, Financial Constraints and Stock Returns, Review of Financial Studies, 14, 529-554. Leary,M.,andM.Roberts,2005,DoFirmsRebalancetheirCapitalStructures? JournalofFinance, 60, 2575-2619. Leary, M., and M. Roberts, 2014, Do Peer Firms Affect Corporate Financial Policy?, Journal of Finance, 69, 139-178. Leland, H.E., 1994, Corporate Debt Value, Bond Convenants, and Optimal Capital Structure, Journal of Finance, 49, 1213-1252. Lemmon, M., M. Roberts, and J. Zender, 2008, Back to the Beginning Persistence and the Cross- Section of Corporate Capital Structure, Journal of Finance, 63, 1575-1608. Livdan, D., H. Sapriza, and L. Zhang, 2009, Financially Constrained Stock Returns, Journal of Finance, 64, 1827-1862. Liu,J.,J.Pan,andT.Wang,2005,AnEquilibriumModelofRare-EventPremiaandItsImplication for Option Smirks, Review of Financial Studies, 18, 131-164. Loughran, T., and J. Ritter, 1995, The New Issues Puzzle, Journal of Finance, 50, 23-51 Marsh, P., 1982, The Choice between Equity and Debt: An empirical study, Journal of Finance, 37, 121144. McLean, R.D., J. Pontiff, and A. Watanabe, 2009, Share Issuance and Cross-Sectional Returns: International Evidence, Journal of Financial Economics, 94, 1-17. McLean, R.D. and M. Zhao, 2014, The Business Cycle, Investor Sentiment, and Costly External Finance, Journal of Finance, 69, 1377-1409. Modigliani, F., and M. Miller, 1958, The Cost of Capital, Corporation Finance and the Theory of Investment, American Economic Review, 261-297. Mohanram, P., 2005, Separating Winners from Losers among Low Book-to-Market Stocks using Financial Statement Analysis, Review of Accounting Studies, 10, 133-170. Myers,S.,1977,DeterminantsofCorporateBorrowing,JournalofFinancialEconomics,5,147-175. Myers, S., 1984, The Capital Structure Puzzle, Journal of Finance, 39, 575-592. Novy-Marx, R., 2013, The Other Side of Value: The Gross Profitability Premium, Journal of Financial Economics, 108(1), 1-28. O¨ztekin, O. and M.J. Flannery, 2012, Institutional Determinants of Capital Structure Adjustment Speeds, Journal of Financial Economics, 103, 88-112. P´astor, L. and R.F. Stambaugh, 2003, Liquidity Risk and Expected Stock Returns, Journal of Political Economy, 111, 642-685. 43

Penman, S., S. Richardson, I. Tuna, 2007, The Book-to-Price Effect in Stock Returns: Accounting for Leverage, Journal of Accounting Research, 45, 427-467. Petersen, M.A., 2009, Estimating Standard Errors in Finance Panel Data Sets: Comparing Approaches, Review of Financial Studies, 22, 435-480. Pontiff, J., and W. Woodgate, 2008, Share Issuance and Cross-sectional Returns, Journal of Finance, 63, 921-945 Richardson, S., R. Sloan, M. Soliman, and I. Tuna, 2005, Accrual Reliability, Earnings Persistence and Stock Prices, Journal of Accounting and Economics, 39 (3), 437-485. Schwert, M., and I. Strebulaev, 2015, Capital Structure and Systematic Risk, working paper. Sloan, R., 1996, Do Stock Prices Reflect Information in Accruals and Cash Flows About Future Earnings?, The Accounting Review, 71, 289-315 Subramanian, A., 2004, Option Pricing on Stocks in Mergers and Acquisitions, Journal of Finance, 59, 795-829. Titman, S., and R. Wessels, 1988, The Determinants of Capital Structure Choice. Journal of Finance, 43, 1-19. Vassalou, M., and Y. Xing, 2004, Default Risk in Equity Returns, Review of Financial Studies, 24, 831868. Vuolteenaho, T., 2002, What Drives Firm-Level Stock Returns?, Journal of Finance, 57, 233-264. Welch, I., 2004, Capital Structure and Stock Returns, Journal of Political Economy, 112, 106-131. Whited,T.,andG.Wu,2006,FinancialConstraintsRisk,Review of Financial Studies,19,531-559. Xing, Y., Z. Zhang, and R. Zhao, 2010, What Does the Individual Option Volatility Smirk Tell Us About Future Equity Returns?, Journal of Financial and Quantitative Analysis, 45, 641-662. 44

Appendix A Wedetailtheconstructionofourvariablesbelow. Summarystatisticsofthesevariablesarereported in Table I. Realized Volatility Annualized standard deviation from the first of each month using a one-year backward-looking window of daily returns Long Call Implied Vol-Realized Vol Diff (IVspread ) hist Spread between quarterly average implied volatility from long-maturity call options and Realized Volatility. Long maturity options are those with >200 days to expiration. See Section 2.1 for details. Short OTM Put-ITM Implied Vol Diff (IVspread ) mon Spreadbetweenquarterlyaverageimpliedvolatilityfromshort-termout-of-the-moneyputoptions and in-the-money put options. Short maturity options are those with <40 days to expiration. Out-of-the-money options are those with spot/strike < 0.8 and in-the-money options are those with spot/strike > 1.2. See Section 2.1 for details. Short Call-Short Put Implied Vol Diff (IVspread ) cp Spread between quarterly average implied volatility from short-maturity call options and shortmaturityputoptions. Shortmaturityoptionsarethosewith<40daystoexpiration. SeeSection2.1 for details. Net Levering Up Ratio (NLEVR) (Diss,i,t−Dred,i,t)+(Ered,i,t−Eiss,i,t) TAi,t where D is long-term debt issuance (DLTIS), D is long-term debt reduction (DLTR), E is iss red red equity repurchases (PRSTKC), and E is equity issuance (SSTK). See Section 2.2 for details. iss Net Levering Up Dummy (NLEVD) is 1 if NLEVR > 0, and 0 otherwise. See Section 2.2 for details. Total Assets Assets - Total (ATQ) * Adjustment to 2000 Dollars Total Market Capitalization Price-Close-Quarter (PRCCQ) * Common Shares Outstanding (CSHOQ) * Adjustment to 2000 Dollars Realized Return Annualized returns using a one-year backward-looking window of monthly returns Ln Total Assets (LnTA) ln{Assets - Total (ATQ) * Adjustment to 2000 Dollars} Book-to-Market Ratio (BTM) TotalCommonEquity(CEQQ) Price-Close-Quarter(PRCCQ)*CommonSharesOutstanding(CSHOQ) 45

Altman’s Zscore (Zscore) 3.3*PretaxIncome(PIQ)+1.0*NetSales(SALEQ)+1.4*RetainedEarnings(REQ)+1.2*WorkingCapital TotalAssets(ATQ) where Working Capital = Current Assets-Total (ACTQ) - Current Liabilities-Total (LCTQ) BCG Marginal Tax Rate (MTR) Blouin, Core, Guay (2010)’s post-financing marginal tax rate Earnings / TA 5-Yr Volatility (σ ) earnings Standard deviation of past 20 quarters of IncomeBeforeExtraordinaryItems(IBQ) TotalAssets(ATQ) Sales / TA 5-Yr Volatility (σ ) sales Standard deviation of past 20 quarters of Sales(SALEQ) TotalAssets(ATQ) Long-term Debt / TA (LTDR) Long-TermDebt-Total(DLTTQ) TotalAssets(ATQ) SIC3 Long-term Debt / TA (IndLTDR) (cid:80)N i=1 Long-TermDebt-Total(DLTTQ) for each firm i in its SIC3 industry (cid:80)N TotalAssets(ATQ) i=1 Credit Rating Spread (CS) Moody’s Baa Rate - Moody’s Aaa Rate Whited and Wu (2006) Financing Constraint Index (WW) −0.091×CF −0.062×DDIV +0.021×LTD −0.044×SIZE +0.102×ISG i,t i,t i,t i,t i,t −0.035×SG i,t where CF is cashflows over total assets, DDIV is an indicator for a dividend-paying firm, LTD is long-term debt over total assets, SIZE is the natural log of book assets, ISG is the sales growth in the firm’s 3-digit SIC industry, and SG is the firm’s one quarter sales growth Hadlock and Pierce (2010) Size-Age Financing Constraint Index −0.737×FirmSize +0.043×FirmSize2 −0.040×FirmAge i,t i,t i,t whereFirmSizeisthelogofbookassetsadjustedforinflationusing2004dollarsandreplacedwith log($4.5billion)iftheactualvalueisgreater,andFirmAgeisthenumberofyearsthefirmhasbeen on Compustat with a non-missing stock price and replaced with 37 if the actual age is greater 46

Table I: Sample statistics of options-based measures and common firm characteristics. Options-based measures are definedinSection2.1. CommonfirmcharacteristicsandcontrolvariablesaredefinedinAppendixA.PanelApresents the summary statistics for the full sample with at least one non-missing options variable and panel B presents the summary statistics for the sample restricted to those with non-missing observations for all relevant variables. PanelA:UnrestrictedSamplew/AtLeastOneNon-MissingOptions-BasedMeasure No. Obs Mean StdDev 1% 25% 50% 75% 99% TotalAssets($millions) 110456 5587.9 19479.8 35.7 337.0 1012.3 3358.4 81981.0 TotalMarketCapitalization($millions) 110456 6355.2 21159.2 62.7 436.8 1154.3 3595.7 102294.5 LogTotalAssets 107787 6.933 1.672 3.563 5.730 6.815 7.997 11.125 Book-to-MarketRatio 107787 0.500 0.377 0.070 0.251 0.408 0.638 1.933 Altman’sZscore 100080 1.558 4.696 -19.379 0.691 2.456 4.036 7.491 BCGMarginalTaxRate 94841 0.284 0.095 0.008 0.259 0.333 0.346 0.352 Earnings/TA5-YrVolatility 107686 0.022 0.020 0.002 0.008 0.016 0.029 0.090 Sales/TA5-YrVolatility 107541 0.055 0.047 0.003 0.023 0.040 0.070 0.231 Long-termDebt/TA 106344 0.168 0.167 0.000 0.002 0.136 0.280 0.637 SIC3Long-termDebt/TA 110456 0.196 0.096 0.044 0.126 0.172 0.247 0.519 CreditRatingSpread 107787 1.054 0.487 0.550 0.780 0.920 1.210 3.090 Whited-WuIndex 97962 -0.322 0.094 -0.526 -0.388 -0.316 -0.255 -0.107 NetLeveringUp/TA 110456 0.000 0.047 -0.195 -0.008 -0.001 0.009 0.138 NetLeveringUp>0 110456 0.372 0.483 0.000 0.000 0.000 1.000 1.000 RealizedReturn 110455 0.192 0.815 -0.772 -0.199 0.072 0.378 3.026 RealizedVolatility 110452 0.530 0.265 0.174 0.347 0.471 0.650 1.406 ImpliedVol: LongCalls 103913 0.487 0.208 0.187 0.335 0.442 0.595 1.147 ImpliedVol: ShortCalls 106946 0.566 0.225 0.233 0.409 0.522 0.674 1.299 ImpliedVol: ShortPuts 105364 0.593 0.232 0.246 0.434 0.546 0.701 1.374 ImpliedVol: OTMPuts 101920 0.618 0.219 0.275 0.466 0.580 0.727 1.319 ImpliedVol: ITMPuts 87662 0.598 0.244 0.233 0.423 0.550 0.723 1.369 IVspread 103906 -0.042 0.151 -0.517 -0.092 -0.021 0.030 0.282 hist IVspreadmon 81303 0.050 0.141 -0.408 -0.004 0.061 0.119 0.368 IVspreadcp 104929 -0.028 0.086 -0.303 -0.053 -0.019 0.004 0.165 PanelB:RestrictedSamplew/AllNon-missingRelevantVariables No. Obs Mean StdDev 1% 25% 50% 75% 99% TotalAssets($millions) 56041 4457.5 10890.3 41.0 348.0 1039.4 3510.7 47482.0 TotalMarketCapitalization($millions) 56041 5815.3 18045.9 74.0 466.1 1195.9 3685.0 80309.0 LogTotalAssets 56041 6.939 1.617 3.724 5.748 6.817 8.027 10.624 Book-to-MarketRatio 56041 0.491 0.363 0.069 0.246 0.402 0.630 1.842 Altman’sZscore 56041 1.654 4.523 -18.290 0.710 2.477 4.108 7.554 BCGMarginalTaxRate 56041 0.284 0.093 0.012 0.255 0.331 0.346 0.352 Earnings/TA5-YrVolatility 56041 0.023 0.020 0.002 0.009 0.016 0.030 0.091 Sales/TA5-YrVolatility 56041 0.056 0.047 0.006 0.024 0.042 0.072 0.226 Long-termDebt/TA 56041 0.163 0.167 0.000 0.001 0.127 0.274 0.636 SIC3Long-termDebt/TA 56041 0.189 0.093 0.038 0.125 0.161 0.236 0.519 CreditRatingSpread 56041 1.088 0.508 0.550 0.800 0.930 1.250 3.090 Whited-WuIndex 56041 -0.320 0.094 -0.529 -0.385 -0.312 -0.253 -0.107 NetLeveringUp/TA 56041 0.001 0.044 -0.174 -0.008 -0.001 0.008 0.137 NetLeveringUp>0 56041 0.362 0.481 0.000 0.000 0.000 1.000 1.000 RealizedReturn 56041 0.156 0.778 -0.771 -0.239 0.024 0.342 2.945 RealizedVolatility 56041 0.558 0.261 0.184 0.377 0.502 0.678 1.406 ImpliedVol: LongCalls 56041 0.507 0.194 0.210 0.367 0.468 0.610 1.116 ImpliedVol: ShortCalls 56041 0.594 0.219 0.247 0.441 0.554 0.702 1.305 ImpliedVol: ShortPuts 56041 0.617 0.229 0.255 0.458 0.573 0.729 1.377 ImpliedVol: OTMPuts 56041 0.632 0.210 0.290 0.486 0.599 0.739 1.296 ImpliedVol: ITMPuts 56041 0.581 0.231 0.231 0.416 0.537 0.700 1.321 IVspread 56041 -0.051 0.146 -0.516 -0.105 -0.026 0.029 0.237 hist IVspreadmon 56041 0.053 0.138 -0.401 0.000 0.065 0.122 0.356 IVspreadcp 56041 -0.023 0.070 -0.245 -0.044 -0.016 0.004 0.130 47

Table II: Pairwise correlation matrix of options-based measures and common firm characteristics. Options-based measures are defined in Section 2.1. Common firm characteristics and control variables are defined in Appendix A. (1) (2) (3) (4) (5) (6) (7) (1)NetLeveringUp/TA (2)LogTotalAssets 0.1113 (3)Book-to-MarketRatio 0.0054 0.1051 (4)Altman’sZscore 0.1466 0.3067 0.0366 (5)BCGMarginalTaxRate 0.1328 0.3333 -0.0674 0.6840 (6)Earnings/TA5-YrVolatility -0.1090 -0.4113 -0.0893 -0.4727 -0.4622 (7)Sales/TA5-YrVolatility -0.0310 -0.2094 0.0022 0.0461 0.0002 0.2540 (8)Long-termDebt/TA -0.0199 0.3776 0.1272 -0.0262 -0.0250 -0.1770 -0.1457 (9)SIC3Long-termDebt/TA 0.0168 0.2120 0.1478 0.0384 0.0333 -0.1989 -0.1146 (10)CreditRatingSpread -0.0096 0.0858 0.1905 0.0335 -0.0183 -0.0930 -0.1075 (11)Whited-WuIndex -0.1075 -0.8793 -0.0820 -0.3435 -0.3566 0.4099 0.1941 (12)RealizedReturn -0.0451 -0.0652 -0.2824 -0.0040 -0.0263 0.0792 0.0353 (13)RealizedVolatility -0.1442 -0.4450 0.0717 -0.3046 -0.4022 0.4001 0.1956 (14)ImpliedVol: LongCalls -0.1443 -0.5169 0.1322 -0.3723 -0.4260 0.4377 0.2124 (15)ImpliedVol: ShortCalls -0.1277 -0.4701 0.1328 -0.2988 -0.3438 0.3727 0.1929 (16)ImpliedVol: ShortPuts -0.1246 -0.4553 0.1294 -0.2826 -0.3271 0.3554 0.1904 (17)ImpliedVol: OTMPuts -0.1426 -0.4744 0.1323 -0.3177 -0.3816 0.3806 0.1850 (18)ImpliedVol: ITMPuts -0.1160 -0.4808 0.1337 -0.2913 -0.3268 0.3509 0.1900 (19)IVspread 0.0653 0.1078 0.0468 0.0488 0.1514 -0.1326 -0.0666 hist (20)IVspreadmon -0.0221 0.0808 -0.0209 0.0050 -0.0322 -0.0079 -0.0355 (21)IVspreadcp 0.0095 0.0191 -0.0054 -0.0099 -0.0071 0.0016 -0.0195 (8) (9) (10) (11) (12) (13) (14) (9)SIC3Long-termDebt/TA 0.4791 (10)CreditRatingSpread 0.0420 0.0699 (11)Whited-WuIndex -0.2881 -0.1915 -0.1114 (12)RealizedReturn -0.0518 -0.0453 -0.1789 0.0572 (13)RealizedVolatility -0.1360 -0.1205 0.0022 0.4545 0.0930 (14)ImpliedVol: LongCalls -0.1306 -0.1228 0.0566 0.5282 -0.0032 0.8330 (15)ImpliedVol: ShortCalls -0.1205 -0.1029 0.1201 0.4766 -0.0237 0.7783 0.9135 (16)ImpliedVol: ShortPuts -0.1120 -0.0891 0.1396 0.4583 -0.0269 0.7595 0.8846 (17)ImpliedVol: OTMPuts -0.1058 -0.0869 0.1602 0.4781 0.0061 0.7805 0.8929 (18)ImpliedVol: ITMPuts -0.1126 -0.0797 0.0371 0.4769 -0.0341 0.7109 0.8291 (19)IVspread 0.0692 0.0516 0.0671 -0.1096 -0.1670 -0.6778 -0.1606 hist (20)IVspreadmon 0.0289 0.0030 0.1815 -0.0695 0.0709 -0.0027 -0.0285 (21)IVspreadcp -0.0101 -0.0306 -0.0808 -0.0091 0.0138 -0.0549 -0.0395 (15) (16) (17) (18) (19) (20) (16)ImpliedVol: ShortPuts 0.9516 (17)ImpliedVol: OTMPuts 0.9055 0.9196 (18)ImpliedVol: ITMPuts 0.8392 0.8673 0.8168 (19)IVspread -0.1770 -0.1820 -0.2082 -0.1684 hist (20)IVspreadmon -0.0264 -0.0529 0.1522 -0.4256 -0.0328 (21)IVspreadcp 0.0101 -0.2863 -0.1770 -0.2142 0.0460 0.0921 48

Table III: Results from the estimation of the baseline model without controls, as in equation (5). The dependent variable is the net levering up ratio (NLEVR), defined as long-term debt issuance net of long-term debt reductions minus equity issuancenet of equity reductions, as a ratio tototal book assets. IVspread measures the difference hist between the implied volatility on long-term call options and the realized, historical volatility for the firm. Realized volatility is the average historical volatility of the firm’s returns over the past year. IVspread measures the mon difference between out-of-the-money and in-the-money put options. IV is the implied volatility of the in-thep,ITM money put options. IVspread is the difference between the implied volatility of short-term call options and the cp short-term put options for a firm. IV is the implied volatility of the long-term put options. All market-based p,short measuresreflecttheaverageofthethreemonthsinthelaggedquarter. Standarderrorsarereportedintheparentheses and clustered by both firm and year-quarter as in Petersen (2009). Significance at the 10% level is indicated by *, 5% level by **, and 1% level by ***. NetLeveringUp/TA(NLEVR) (1) (2) (3) (4) IVspread -0.0261*** -0.0261*** hist,t−1 (0.0024) (0.0024) IVspreadmon,t−1 -0.0271*** -0.0032** (0.0020) (0.0015) IVspreadcp,t−1 -0.0203*** 0.0056 ** (0.0027) (0.0024) RealizedVolatility -0.0387*** -0.0386*** t−1 (0.0022) (0.0022) IVp,ITM,t−1 -0.0350*** (0.0020) IV -0.0319*** p,short,t−1 (0.0022) Constant 0.0192 *** 0.0200 *** 0.0187 *** 0.0191 *** (0.0016) (0.0016) (0.0016) (0.0016) QuarterFixedEffects? Y Y Y Y YearFixedEffects? Y Y Y Y No. Obs. 77389 77389 77389 77389 AdjustedR2 0.0302 0.0272 0.0240 0.0303 49

Table IV: Results from the estimation of the full model with controls, as in equation (6). The dependent variable is thenetleveringupratio(NLEVR),definedaslong-termdebtissuancenetoflong-termdebtreductionsminusequity issuance net of equity reductions, as a ratio to total book assets. IVspread measures the difference between the hist impliedvolatilityonlong-termcalloptionsandtherealized,historicalvolatilityforthefirm. IVspread measures mon the difference between out-of-the-money and in-the-money put options. IVspread is the difference between the cp implied volatility of short-term call options and the short-term put options for a firm. Realized volatility is the average historical volatility of the firm’s returns over the past year. All market-based measures reflect the average of the three months in the lagged quarter. All controls are defined in Appendix A. Standard errors are reported in the parentheses and clustered by both firm and year-quarter as in Petersen (2009). Significance at the 10% level is indicated by *, 5% level by **, and 1% level by ***. NetLeveringUp/TA(NLEVR) (1) (2) (3) (4) (5) IVspread -0.0103*** -0.0100*** hist,t−1 (0.0027) (0.0027) IVspreadmon,t−1 -0.0048*** -0.0047*** (0.0014) (0.0014) IVspreadcp,t−1 0.0028 0.0041 * (0.0023) (0.0023) RealizedVolatility -0.0180*** -0.0124*** -0.0125*** -0.0177*** t−1 (0.0025) (0.0018) (0.0019) (0.0025) RealizedReturnt−1 -0.0026*** -0.0022*** -0.0021*** -0.0022*** -0.0022*** (0.0004) (0.0004) (0.0004) (0.0004) (0.0004) LogTotalAssets 0.0021 *** 0.0017 *** 0.0019 *** 0.0018 *** 0.0017 *** t−1 (0.0004) (0.0004) (0.0004) (0.0004) (0.0004) Book-to-MarketRatiot−1 -0.0008 0.0005 0.0000 0.0001 0.0004 (0.0008) (0.0008) (0.0008) (0.0008) (0.0008) Altman’sZscoret−1 0.0010 *** 0.0009 *** 0.0009 *** 0.0009 *** 0.0009 *** (0.0001) (0.0001) (0.0001) (0.0001) (0.0001) BCGMarginalTaxRate 0.0082 ** 0.0008 0.0018 0.0018 0.0008 t−1 (0.0040) (0.0040) (0.0040) (0.0040) (0.0040) Earnings/TA5-YrVolatility -0.0401** -0.0122 -0.0175 -0.0181 -0.0121 t−1 (0.0186) (0.0177) (0.0180) (0.0181) (0.0176) Sales/TA5-YrVolatility -0.0162*** -0.0116* -0.0125** -0.0125** -0.0114* t−1 (0.0060) (0.0062) (0.0061) (0.0061) (0.0061) Long-termDebt/TA -0.0171*** -0.0167*** -0.0169*** -0.0170*** -0.0166*** t−1 (0.0022) (0.0022) (0.0022) (0.0022) (0.0022) SIC3Long-termDebt/TA 0.0092 *** 0.0067 ** 0.0072 ** 0.0073 ** 0.0067 ** t−1 (0.0032) (0.0031) (0.0031) (0.0031) (0.0031) CreditRatingSpread -0.0015 -0.0011 -0.0018* -0.0018* -0.0010 t−1 (0.0012) (0.0011) (0.0010) (0.0010) (0.0011) Whited-WuIndext−1 -0.0032 0.0038 0.0027 0.0022 0.0039 (0.0062) (0.0062) (0.0061) (0.0061) (0.0062) Constant -0.0121*** 0.0033 -0.0010 -0.0005 0.0027 (0.0025) (0.0034) (0.0031) (0.0031) (0.0034) QuarterFixedEffects? Y Y Y Y Y YearFixedEffects? Y Y Y Y Y No. Obs. 56041 56041 56041 56041 56041 AdjustedR2 0.0389 0.0417 0.0415 0.0413 0.0419 50

Table V: Results from the estimation of the modified full model with first lagged differences and second lagged levels, as in equation (7). The dependent variable is the net levering up ratio (NLEVR), defined as long-term debt issuance net of long-term debt reductions minus equity issuance net of equity reductions, as a ratio to total book assets. IVspread measuresthedifferencebetweentheimpliedvolatilityonlong-termcalloptionsandtherealized, hist historical volatility for the firm. IVspread measures the difference between out-of-the-money and in-the-money mon putoptions. IVspread isthedifferencebetweentheimpliedvolatilityofshort-termcalloptionsandtheshort-term cp put options for a firm. Realized volatility is the average historical volatility of the firm’s returns over the past year. All market-based measures reflect the average of the three months in the lagged quarter. All controls are defined in Appendix A. Standard errors are reported in the parentheses and clustered by both firm and year-quarter as in Petersen (2009). Significance at the 10% level is indicated by *, 5% level by **, and 1% level by ***. NetLeveringUp/TA(NLEVR) (1) (2) (3) (4) (5) ∆IVspread -0.0099*** -0.0097** hist,t−2,t−1 (0.0038) (0.0038) ∆IVspreadmon,t−2,t−1 -0.0043** -0.0038** (0.0019) (0.0019) ∆IVspreadcp,t−2,t−1 0.0015 0.0029 (0.0027) (0.0026) ∆RealizedVolatility -0.0112* -0.0124** -0.0124** -0.0111* t−2,t−1 (0.0062) (0.0053) (0.0053) (0.0062) ∆RealizedReturnt−2,t−1 -0.0037*** -0.0037*** -0.0032*** -0.0033*** -0.0036*** (0.0007) (0.0008) (0.0008) (0.0008) (0.0008) ∆LogTotalAssets -0.0110*** -0.0109*** -0.0107*** -0.0109*** -0.0107*** t−2,t−1 (0.0034) (0.0034) (0.0034) (0.0034) (0.0034) ∆Book-to-MarketRatiot−2,t−1 0.0038 ** 0.0046 *** 0.0040 *** 0.0041 *** 0.0045 *** (0.0016) (0.0016) (0.0015) (0.0015) (0.0016) ∆Altman’sZscoret−2,t−1 0.0011 *** 0.0010 *** 0.0011 *** 0.0011 *** 0.0010 *** (0.0002) (0.0002) (0.0002) (0.0002) (0.0002) ∆BCGMarginalTaxRate -0.0054 -0.0107 -0.0091 -0.0090 -0.0108 t−2,t−1 (0.0090) (0.0095) (0.0094) (0.0094) (0.0095) ∆Earnings/TA5-YrVolatility -0.0710 -0.0412 -0.0492 -0.0480 -0.0422 t−2,t−1 (0.0570) (0.0559) (0.0567) (0.0567) (0.0560) ∆Sales/TA5-YrVolatility -0.0628** -0.0556* -0.0562* -0.0560* -0.0559* t−2,t−1 (0.0310) (0.0312) (0.0311) (0.0311) (0.0311) ∆Long-termDebt/TA 0.0200 *** 0.0200 *** 0.0194 *** 0.0196 *** 0.0199 *** t−2,t−1 (0.0074) (0.0073) (0.0073) (0.0073) (0.0073) ∆SIC3Long-termDebt/TA 0.0183 0.0163 0.0166 0.0169 0.0161 t−2,t−1 (0.0113) (0.0116) (0.0116) (0.0116) (0.0116) ∆CreditRatingSpread -0.0019** -0.0018** -0.0024*** -0.0023*** -0.0018** t−2,t−1 (0.0010) (0.0009) (0.0008) (0.0008) (0.0009) ∆Whited-WuIndext−2,t−1 0.0057 0.0119 * 0.0102 * 0.0098 0.0121 * (0.0062) (0.0063) (0.0062) (0.0062) (0.0063) Continuedbelow. 51

Continuedfromabove. (1) (2) (3) (4) (5) IVspread -0.0172*** -0.0168*** hist,t−2 (0.0035) (0.0035) IVspreadmon,t−2 -0.0060*** -0.0054** (0.0023) (0.0023) IVspreadcp,t−2 0.0008 0.0037 (0.0033) (0.0033) RealizedVolatility -0.0194*** -0.0104*** -0.0105*** -0.0190*** t−2 (0.0027) (0.0018) (0.0018) (0.0028) RealizedReturnt−2 -0.0026*** -0.0022*** -0.0020*** -0.0021*** -0.0021*** (0.0004) (0.0004) (0.0004) (0.0004) (0.0004) LogTotalAssets 0.0024 *** 0.0022 *** 0.0024 *** 0.0023 *** 0.0022 *** t−2 (0.0005) (0.0005) (0.0005) (0.0005) (0.0005) Book-to-MarketRatiot−2 -0.0030*** -0.0015* -0.0022*** -0.0022** -0.0017* (0.0009) (0.0009) (0.0009) (0.0009) (0.0009) Altman’sZscoret−2 0.0009 *** 0.0009 *** 0.0009 *** 0.0009 *** 0.0009 *** (0.0001) (0.0001) (0.0001) (0.0001) (0.0001) BCGMarginalTaxRate 0.0085 * 0.0021 0.0039 0.0040 0.0021 t−2 (0.0045) (0.0045) (0.0046) (0.0046) (0.0045) Earnings/TA5-YrVolatility -0.0396* -0.0128 -0.0211 -0.0216 -0.0126 t−2 (0.0205) (0.0196) (0.0200) (0.0201) (0.0196) Sales/TA5-YrVolatility -0.0116* -0.0069 -0.0084 -0.0085 -0.0067 t−2 (0.0067) (0.0067) (0.0067) (0.0067) (0.0067) Long-termDebt/TA -0.0209*** -0.0204*** -0.0206*** -0.0207*** -0.0203*** t−2 (0.0025) (0.0025) (0.0025) (0.0025) (0.0025) SIC3Long-termDebt/TA 0.0099 *** 0.0077 ** 0.0084 ** 0.0084 ** 0.0077 ** t−2 (0.0037) (0.0036) (0.0036) (0.0036) (0.0036) CreditRatingSpread -0.0045*** -0.0035*** -0.0041*** -0.0043*** -0.0033*** t−2 (0.0012) (0.0013) (0.0012) (0.0012) (0.0013) Whited-WuIndext−2 -0.0028 0.0100 0.0063 0.0056 0.0102 (0.0081) (0.0080) (0.0079) (0.0080) (0.0080) Constant -0.0100*** 0.0054 -0.0013 -0.0006 0.0046 (0.0032) (0.0040) (0.0036) (0.0036) (0.0040) QuarterFixedEffects? Y Y Y Y Y YearFixedEffects? Y Y Y Y Y No. Obs. 44105 44105 44105 44105 44105 AdjustedR2 0.0408 0.0432 0.0426 0.0424 0.0433 52

TableVI:Resultsfromtheestimationofthefullmodel,asinequation(6),usingunconditionalsub-samplesofvarying supply and demand for financing. Columns (1) and (2) sub-sample all observations into boom and bust years. We usetwoeventstodefineboomandbustyears: thedot-combubbleandthefinancialcrisis. Boomperiodsaredefined tobe1996Q1through1999Q4and2005Q1through2007Q2. Bustperiodsaredefinedtobe2001Q1through2002Q4 and2007Q3through2009Q2. Colunns(3)and(4)sub-sampleallfirmsintothreeequal-sizedbinsbasedonthebookto-marketratioeachquarterwithLowBTMdefinedasthebottomtercileandHighBTMdefinedasthetoptercile. The dependent variable is the net levering up ratio (NLEVR), defined as long-term debt issuance net of long-term debtreductionsminusequityissuancenetofequityreductions,asaratiotototalbookassets. IVspread measures hist the difference between the implied volatility on long-term call options and the realized, historical volatility for the firm. IVspread measures the difference between out of the money and at the money put options. IVspread mon cp is the difference between the implied volatility of long-term call options and the long-term put options for a firm. Realizedvolatilityistheaveragehistoricalvolatilityofthefirm’sreturnsoverthepastyear. Allexplanatoryvariables arelaggedatonequarter. Allmarket-basedmeasuresreflecttheaverageofthethreemonthsinthepastquarter. All controls are defined in Appendix A. Standard errors are reported in the parentheses and clustered by both firm and year-quarter as in Petersen (2009). Significance at the 10% level is indicated by *, 5% level by **, and 1% level by ***. NetLeveringUp/TA(NLEVR) BoomYears BustYears LowBTM HighBTM (1) (2) (3) (4) IVspread -0.0149*** -0.0097** -0.0122** -0.0069* hist,t−1 (0.0042) (0.0044) (0.0048) (0.0037) IVspreadmon,t−1 -0.0057** -0.0064** -0.0112*** -0.0003 (0.0025) (0.0028) (0.0030) (0.0017) IVspreadcp,t−1 0.0056 0.0019 0.0159 ** -0.0048* (0.0042) (0.0038) (0.0073) (0.0026) RealizedVolatility -0.0226*** -0.0146*** -0.0194*** -0.0141*** t−1 (0.0039) (0.0036) (0.0035) (0.0028) RealizedReturnt−1 -0.0040*** -0.0020*** -0.0013*** -0.0015** (0.0007) (0.0005) (0.0005) (0.0007) LogTotalAssets 0.0007 0.0025 *** 0.0020 *** 0.0009 * t−1 (0.0007) (0.0006) (0.0007) (0.0005) Book-to-MarketRatiot−1 0.0008 -0.0008 0.0184 *** -0.0012 (0.0016) (0.0011) (0.0049) (0.0009) Altman’sZscoret−1 0.0010 *** 0.0010 *** 0.0011 *** 0.0002 (0.0001) (0.0002) (0.0002) (0.0001) BCGMarginalTaxRate 0.0107 -0.0069 0.0011 0.0045 t−1 (0.0071) (0.0075) (0.0076) (0.0053) Earnings/TA5-YrVolatility -0.0156 -0.0026 -0.0052 0.0093 t−1 (0.0274) (0.0280) (0.0288) (0.0208) Sales/TA5-YrVolatility -0.0148* -0.0092 -0.0023 -0.0176** t−1 (0.0079) (0.0090) (0.0117) (0.0079) Long-termDebt/TA -0.0113*** -0.0246*** -0.0070** -0.0190*** t−1 (0.0041) (0.0035) (0.0035) (0.0029) SIC3Long-termDebt/TA 0.0175 *** 0.0080 * -0.0032 0.0114 *** t−1 (0.0053) (0.0042) (0.0052) (0.0042) CreditRatingSpread 0.0016 -0.0040*** 0.0000 -0.0006 t−1 (0.0042) (0.0010) (0.0015) (0.0009) Whited-WuIndext−1 -0.0017 0.0059 -0.0071 0.0066 (0.0100) (0.0089) (0.0108) (0.0087) Constant 0.0025 0.0046 -0.0066 0.0080 * (0.0068) (0.0056) (0.0049) (0.0044) QuarterFixedEffects? Y Y Y Y YearFixedEffects? Y Y Y Y No. Obs. 17771 18444 18871 18186 AdjustedR2 0.0393 0.0470 0.0668 0.0229 53

snmuloC .gnicnanfirofdnameddnaylppusgniyravfoselpmas-buslanoitidnocgnisu,)6(noitauqenisa,ledomllufehtfonoitamitseehtmorfstluseR:IIVelbaT tekram-ot-koob hgih dna wol yb ,2Q7002 hguorht 1Q5002 dna 4Q9991 hguorht 1Q6991 sa denfied ,sraey moob eht ni snoitavresbo lla elpmas-bus )2( dna )1( MTBhgihdnawolyb,2Q9002hguorht3Q7002dna4Q2002hguorht1Q1002sadenfied,sraeytsubehtnisnoitavresbollaelpmas-bus)4(dna)3(snmuloC .smrfi ehttroper)8(hguorht)5(snmuloC .retrauqhcaeoitartekram-ot-koobnognitrosnehwelicret)pot(mottobehtehtnismrfisadenfiedsiMTB)hgiH(woL .smrfi sadenfied,)RVELN(oitarpugnireveltenehtsielbairavtnednepedehT .ecnacfiingisdna,seulav-p,sisylanadelpmas-busehtneewtebstneicffieocnisecnereffid serusaem daerpsVI .stessa koob latot ot oitar a sa ,snoitcuder ytiuqe fo ten ecnaussi ytiuqe sunim snoitcuder tbed mret-gnol fo ten ecnaussi tbed mret-gnol tsih ecnereffid eht serusaem daerpsVI .mrfi eht rof ytilitalov lacirotsih ,dezilaer eht dna snoitpo llac mret-gnol no ytilitalov deilpmi eht neewteb ecnereffid eht nom mret-gnol eht dna snoitpo llac mret-gnol fo ytilitalov deilpmi eht neewteb ecnereffid eht si daerpsVI .snoitpo tup yenom eht ta dna yenom eht fo tuo neewteb pc eno ta deggal era selbairav yrotanalpxe llA .raey tsap eht revo snruter s’mrfi eht fo ytilitalov lacirotsih egareva eht si ytilitalov dezilaeR .mrfi a rof snoitpo tup era srorre dradnatS .A xidneppA ni denfied era slortnoc llA .retrauq tsap eht ni shtnom eerht eht fo egareva eht tcefler serusaem desab-tekram llA .retrauq tset-F dna tset 2χ no desab stekcarb eht ni detroper era seulav-p .)9002( nesreteP ni sa retrauq-raey dna mrfi htob yb deretsulc dna sesehtnerap eht ni detroper .*** yb level %1 dna ,** yb level %5 ,* yb detacidni si level %01 eht ta ecnacfiingiS .stneicffieoc neewteb ytilauqe fo )RVELN(AT/pUgnireveLteN secnereffiD sraeYtsuB sraeYmooB )4(-)3( )2(-)1( )4(-)2( )3(-)1( MTBhgiH MTBwoL MTBhgiH MTBwoL )8( )7( )6( )5( )4( )3( )2( )1( 6600.0 3110.0- 8300.0 0410.0- *0010.0- 5300.0- 2600.0- ***5710.0- 1−t,tsihdaerpsVI ]1364.0[ ]6003.0[ ]0346.0[ ]7912.0[ )3500.0( )1800.0( )7700.0( )0600.0( **2410.0- 7700.0- 0300.0- 6300.0 1000.0- **3410.0- 0300.0- *7010.0- 1−t,nomdaerpsVI ]1830.0[ ]8342.0[ ]6195.0[ ]9836.0[ )2400.0( )6500.0( )5300.0( )1600.0( *5420.0 *6620.0 2400.0 3600.0 *5010.0- 0410.0 4600.0- *2020.0 1−t,pcdaerpsVI ]0760.0[ ]2160.0[ ]0636.0[ ]7817.0[ )5500.0( )3410.0( )0500.0( )1110.0( 1900.0 1800.0- 7100.0 *5510.0- ***3710.0- 3800.0- **7510.0- ***7320.0ytilitaloVdezilaeR 1−t ]1601.0[ ]8853.0[ ]9887.0[ ]1460.0[ )8300.0( )4500.0( )3600.0( )3500.0( 8000.0- 8000.0- 2300.0- ***7200.0- 1−tnruteRdezilaeR )4100.0( )6000.0( )0200.0( )9000.0( 6000.0 ***2300.0 4000.0 7000.0 stessAlatoTgoL 1−t )8000.0( )1100.0( )0100.0( )8000.0( *1200.0- 3600.0 5100.0- ***3230.0 1−toitaRtekraM-ot-kooB )1100.0( )4500.0( )7100.0( )9900.0( 1000.0 ***5100.0 2000.0 ***0100.0 1−terocsZs’namtlA )2000.0( )3000.0( )4000.0( )3000.0( 7400.0 4510.0- 0810.0 3000.0 etaRxaTlanigraMGCB 1−t )9800.0( )6310.0( )0210.0( )8510.0( 3440.0 1310.0- 1040.0- 4500.0 ytilitaloVrY-5AT/sgninraE 1−t )4720.0( )9640.0( )5050.0( )0930.0( 8910.0- 3400.0- 9120.0- 8800.0ytilitaloVrY-5AT/selaS 1−t )3410.0( )5510.0( )1410.0( )5310.0( ***7120.0- *2210.0- ***5910.0- 3500.0- AT/tbeDmret-gnoL 1−t )1500.0( )3600.0( )0500.0( )0700.0( *3010.0 0500.0- ***1620.0 1300.0 AT/tbeDmret-gnoL3CIS 1−t )6500.0( )0800.0( )3800.0( )9800.0( 1200.0- **6400.0- 1500.0- 0800.0 daerpSgnitaRtiderC 1−t )8100.0( )3200.0( )2500.0( )4600.0( 5300.0 5500.0- 7700.0 **9520.0- 1−txednIuW-detihW )8410.0( )7710.0( )8510.0( )3210.0( ***9510.0 ***5130.0- 3900.0 9700.0tnatsnoC )8500.0( )3800.0( )9800.0( )7900.0( *]8750.0[ ]5922.0[ ]2759.0[ ]8534.0[ tseT-F Y Y Y Y ?stceffEdexiFretrauQ Y Y Y Y ?stceffEdexiFraeY 8795 6116 8295 7806 .sbO .oN 7220.0 7370.0 4320.0 1060.0 2RdetsujdA 54

TableVIII:Resultsfromtheestimationofequation(6)usingoptioncountsinplaceoftheoptions-basedriskmeasures. The dependent variable is the net levering up ratio (NLEVR), defined as long-term debt issuance net of long-term debtreductionsminusequityissuancenetofequityreductions,asaratiotototalbookassets. HasTradeableOptions is a dummy variable that equals 1 if the firm has tradeable options in the market over the three months in the past quarter and 0 otherwise. Log Total Tradeable Options is the natural log of the total number of tradeable options a firm has in the market over a month averaged over the three months in the past quarter. Log Total Open Interest is the natural log of the open interest a firm has in the market over a month averaged over the three months in the past quarter. Log Total Volume is the natural log of the total number of options traded in the market for the firm over a month averaged over the three months in the past quarter. All explanatory variables are lagged one quarter. All controls are defined in Appendix A. Standard errors are reported in the parentheses and clustered by both firm and year-quarter as in Petersen (2009). Significance at the 10% level is indicated by *, 5% level by **, and 1% level by ***. NetLeveringUp/TA(NLEVR) (1) (2) (3) (4) HasTradeableOptions 0.0004 t−1 (0.0004) LogTotalTradeableOptions 0.0004 *** t−1 (0.0002) LogTotalOpenInterest 0.0001 ** t−1 (0.0001) LogTotalVolume 0.0002 *** t−1 (0.0001) RealizedVolatility -0.0038*** -0.0041*** -0.0040*** -0.0042*** t−1 (0.0008) (0.0008) (0.0008) (0.0008) RealizedReturnt−1 -0.0032*** -0.0032*** -0.0032*** -0.0032*** (0.0003) (0.0003) (0.0003) (0.0003) LogTotalAssets 0.0014 *** 0.0011 *** 0.0012 *** 0.0011 *** t−1 (0.0002) (0.0002) (0.0002) (0.0002) Book-to-MarketRatiot−1 0.0014 *** 0.0017 *** 0.0016 *** 0.0016 *** (0.0004) (0.0004) (0.0004) (0.0004) Altman’sZscoret−1 0.0007 *** 0.0007 *** 0.0007 *** 0.0007 *** (0.0001) (0.0001) (0.0001) (0.0001) BCGMarginalTaxRate 0.0053 ** 0.0050 ** 0.0052 ** 0.0052 ** t−1 (0.0024) (0.0024) (0.0024) (0.0024) Earnings/TA5-YrVolatility -0.0489*** -0.0512*** -0.0511*** -0.0514*** t−1 (0.0109) (0.0109) (0.0109) (0.0109) Sales/TA5-YrVolatility -0.0055 -0.0053 -0.0054 -0.0055 t−1 (0.0034) (0.0034) (0.0034) (0.0034) Long-termDebt/TA -0.0153*** -0.0149*** -0.0150*** -0.0149*** t−1 (0.0018) (0.0018) (0.0018) (0.0018) SIC3Long-termDebt/TA 0.0114 *** 0.0117 *** 0.0116 *** 0.0117 *** t−1 (0.0021) (0.0021) (0.0021) (0.0021) CreditRatingSpread -0.0008 -0.0008 -0.0008 -0.0008 t−1 (0.0006) (0.0006) (0.0006) (0.0006) Whited-WuIndext−1 -0.0055* -0.0062* -0.0061* -0.0062* (0.0033) (0.0033) (0.0033) (0.0033) Constant -0.0050*** -0.0037** -0.0040** -1.2994*** (0.0015) (0.0016) (0.0016) (0.0952) QuarterFixedEffects? Y Y Y Y YearFixedEffects? Y Y Y Y No. Obs. 183032 183032 183032 183032 AdjustedR2 0.0332 0.0333 0.0333 0.0333 55

esu)5(dna)1(snmuloC .retrauqtsapehtniserusaemdesab-tekramylhtnomgninimaxe)6(noitauqenisaledomllufehtfonoitamitseehtmorfstluseR:XIelbaT shtnom evfi deggal serusaem desab-tekram esu ew ,nosirapmoc roF .serusaem desab-tekram ruo etaluclac ot retrauq tsap eht ni shtnom eerht eht fo egareva eht deggal dna ,)7( dna )3( snmuloc ni )retrauq tsap eht ni htnom dnoces ,.e.i( shtnom ruof deggal ,)6( dna )2( snmuloc ni )retrauq tsap eht ni htnom tsrfi ,.e.i( ten eht ,)RVELN( AT / pU gnireveL teN si A lenaP ni elbairav tnedneped ehT .)8( hguorht )4( snmuloc ni ,)retrauq tsap eht ni htnom tsal ,.e.i( shtnom eerht otoitarasa,snoitcuderytiuqefotenecnaussiytiuqesunimsnoitcudertbedmret-gnolfotenecnaussitbedmret-gnolsadenfied,mrfiehtforoivahebpugnirevel daerpsVI .DVELN,)esiwrehto=0,0>pUgnireveLteN=1(pUgnireveLteNrofrotacidninagnisunoissergertigolasetamitseBlenaP .stessakooblatot tsih eht serusaem daerpsVI .mrfi eht rof ytilitalov lacirotsih ,dezilaer eht dna snoitpo llac mret-gnol no ytilitalov deilpmi eht neewteb ecnereffid eht serusaem nom dna snoitpo llac mret-gnol fo ytilitalov deilpmi eht neewteb ecnereffid eht si daerpsVI .snoitpo tup yenom eht ta dna yenom eht fo tuo neewteb ecnereffid pc era selbairav yrotanalpxe llA .raey tsap eht revo snruter s’mrfi eht fo ytilitalov lacirotsih egareva eht si ytilitalov dezilaeR .mrfi a rof snoitpo tup mret-gnol eht nisaretrauq-raeydnamrfihtobybderetsulcdnasesehtnerapehtnidetropererasrorredradnatS.AxidneppAnidenfiederaslortnocllA .retrauqenotadeggal .*** yb level %1 dna ,** yb level %5 ,* yb detacidni si level %01 eht ta ecnacfiingiS .)9002( nesreteP )DVELN(esiwrehto=0,0>pUgnireveLteN=1 )RVELN(AT/pUgnireveLteN )8( )7( )6( )5( )4( )3( )2( )1( 4403.0- *2865.0- ***3958.0- *4433.0- 2500.0- ***8110.0- ***9110.0- ***0010.0daerpsVI 1−t,tsih )7492.0( )5103.0( )2382.0( )4571.0( )6300.0( )8300.0( )6300.0( )7200.0( ***8895.0- ***0007.0- ***0359.0- ***9282.0- ***8800.0- ***5700.0- **6500.0- ***7400.0- 1−t,nomdaerpsVI )3191.0( )2841.0( )9671.0( )3090.0( )8200.0( )5200.0( )8200.0( )4100.0( 5870.0 8520.0 2130.0 ** 7253.0 0700.0 8500.0 9300.0 * 1400.0 1−t,pcdaerpsVI )8022.0( )1462.0( )4602.0( )3151.0( )8400.0( )9400.0( )2400.0( )3200.0( ***2124.1- ***9945.1- ***8776.1- ***6774.1- ***0610.0- ***0910.0- ***7020.0- ***7710.0ytilitaloVdezilaeR 1−t )7632.0( )5342.0( )0232.0( )0631.0( )5300.0( )1300.0( )0300.0( )5200.0( **7041.0- 1860.0- 7940.0- ***6811.0- ***6811.0- **0100.0- ***1200.0- ***2200.0- 1−tnruteRdezilaeR )7450.0( )3940.0( )4730.0( )7620.0( )7620.0( )4000.0( )7000.0( )4000.0( *** 1931.0 *** 1631.0 *** 1431.0 *** 5121.0 *** 4100.0 *** 4100.0 *** 4100.0 *** 7100.0 stessAlatoTgoL 1−t )5530.0( )4530.0( )9430.0( )8320.0( )5000.0( )5000.0( )5000.0( )4000.0( *7741.0- 1711.0- 0511.0- 1710.0- 4000.0- 8000.0- 8000.0- 4000.0 1−toitaRtekraM-ot-kooB )1970.0( )5570.0( )3370.0( )7840.0( )1100.0( )1100.0( )1100.0( )8000.0( *** 5940.0 *** 0940.0 *** 8840.0 *** 5240.0 *** 9000.0 *** 8000.0 *** 8000.0 *** 9000.0 1−terocsZs’namtlA )0210.0( )8110.0( )8110.0( )8700.0( )2000.0( )2000.0( )2000.0( )1000.0( 3884.0 7554.0 5983.0 8451.0 3400.0 1400.0 9200.0 8000.0 etaRxaTlanigraMGCB 1−t )2263.0( )4953.0( )7453.0( )0642.0( )0500.0( )0500.0( )9400.0( )0400.0( 9808.2- 2217.2- 5416.2- 1427.1- 1427.1- *9440.0- 4830.0- 1210.0ytilitaloVrY-5AT/sgninraE 1−t )2008.1( )1097.1( )1877.1( )6561.1( )6561.1( )3420.0( )2420.0( )6710.0( 4363.0- 7973.0- 3923.0- 7951.0- 7951.0- 5900.0- 9900.0- *4110.0ytilitaloVrY-5AT/selaS 1−t )6126.0( )4326.0( )8226.0( )9924.0( )9924.0( )2800.0( )1800.0( )1600.0( **6664.0- **2064.0- **5454.0- 0560.0- ***9020.0- ***0120.0- ***1120.0- ***6610.0- AT/tbeDmret-gnoL 1−t )0491.0( )4391.0( )2291.0( )9141.0( )9200.0( )9200.0( )9200.0( )2200.0( 4432.0 9212.0 5591.0 7592.0 4400.0 6400.0 2400.0 ** 7600.0 AT/tbeDmret-gnoL3CIS 1−t )7203.0( )6303.0( )8203.0( )6002.0( )6400.0( )6400.0( )6400.0( )1300.0( 2390.0- 4380.0- 3390.0- 9840.0- 5100.0- 7100.0- 1200.0- 0100.0daerpSgnitaRtiderC 1−t )6370.0( )7270.0( )2660.0( )5150.0( )4100.0( )5100.0( )4100.0( )1100.0( **1631.1- **1101.1- *0960.1- **7159.0- 5100.0 2300.0 7300.0 9300.0 1−txednIuW-detihW )9165.0( )0855.0( )6055.0( )5973.0( )7700.0( )6700.0( )5700.0( )2600.0( ***8202.1- ***7601.1- ***5620.1- ***9758.0- 5600.0 2600.0 * 2800.0 7200.0 tnatsnoC )4103.0( )6203.0( )4992.0( )7171.0( )2400.0( )3400.0( )5400.0( )4300.0( Y Y Y Y Y Y Y Y ?stceffEdexiFretrauQ Y Y Y Y Y Y Y Y ?stceffEdexiFraeY 46922 46922 46922 14065 46922 46922 46922 14065 .sbO .oN 6380.0 2380.0 1480.0 9070.0 4740.0 2840.0 6840.0 9140.0 2RdetsujdA 56

Table X: Options based indices of net leverage increase. Using the coefficients estimated from columns (1) and (5) in Table IX, we create market-based indices to measure net levering up behavior. LIMkt is the options based net 1 leverage increase index calculated according to equation (10), using only the coefficient estimates for IVspread , hist IVspread , IVspread , and Realized Volatility obtained from column (1) of Table IX. LIMkt is calculated mon cp 2 according to equation (11), using LIMkt and the coefficient estimates for the control variables obtained from 1 column (1) of Table IX. LIMkt and LIMkt use the coefficient results from column (5) of Table IX and are 3 4 calculated according to equations (12) and (13), respectively. Panel A provides the summary statistics for our four options based indices. Panel B reports the pairwise correlation matrix of our four options based indices. PanelA:SampleStatistics No. Obs Mean StdDev 1% 25% 50% 75% 99% LIMkt1 77389 -0.010 0.004 -0.022 -0.012 -0.009 -0.007 -0.003 LIMkt2 69311 0.001 0.009 -0.027 -0.003 0.003 0.008 0.017 LIMkt3 77389 -1.689 0.366 -2.859 -1.870 -1.616 -1.434 -1.132 LIMkt4 64093 -0.479 0.654 -2.381 -0.843 -0.404 -0.017 0.680 PanelB:PairwiseCorrelation (1) (2) (3) (1)LIMkt1 (2)LIMkt2 0.7697 (3)LIMkt3 0.9801 0.7432 (4)LIMkt4 0.8509 0.9198 0.8469 57

Table XI: Options based indices of net leverage increase. Using the coefficients estimated from columns (1) and (5) in Table IX, we create market-based indices to measure net levering up behavior. LIMkt is the options 1 based net leverage increase index calculated according to equation (10), using only the coefficient estimates for IVspread , IVspread , IVspread , and Realized Volatility obtained from column (1) of Table IX. LIMkt is hist mon cp 2 calculated according to equation (11), using LIMkt and the coefficient estimates for the control variables obtained 1 from column (1) of Table IX. LIMkt and LIMkt use the coefficient results from column (5) of Table IX and 3 4 are calculated according to equations (12) and (13), respectively. This table compares the means of common firm characteristics associated with firms with low and high leverage increases. We sort each measure into three equal binseachyear-quarter. LOWtercilereflectsthefirmswithlowerleverageincreasesandHIGHreflectsthefirmswith higherleverageincreases. Meansineachtercilearereportedbelowandtestedtoseeiftheyarestatisticallydifferent from each other. Significance at the 10% level is indicated by *, 5% level by **, and 1% level by ***. LIMkt1: OptionsOnly LIMkt2: WithControls LOW HIGH DIFF LOW HIGH DIFF TotalAssets($millions) 1580.1 14653.0 *** 1043.4 15550.9 *** TotalMarketCapitalization($millions) 1411.8 17268.1 *** 1047.0 19650.5 *** LogTotalAssets 5.9347 8.2745 *** 5.8648 8.3922 *** Book-to-MarketRatio 0.5416 0.4662 *** 0.5065 0.4657 *** Altman’sZscore -1.1041 3.1932 *** -2.0232 3.8563 *** TotalDebt/TA 0.1538 0.2334 *** 0.2059 0.1694 *** Long-termDebt/TA 0.1326 0.2001 *** 0.1838 0.1372 *** CashFlow/TA 0.0004 0.0275 *** -0.0014 0.0295 *** Cash/TA 0.3242 0.1134 *** 0.3142 0.1367 *** PaysDividend=1 0.1069 0.6495 *** 0.0832 0.6588 *** HasLong-termDebtCreditRating 0.2205 0.6890 *** 0.2608 0.6524 *** HasInvestmentGradeLong-termDebt 0.0174 0.5387 *** 0.0055 0.5711 *** Whited-WuIndex -0.2640 -0.3908 *** -0.2590 -0.3979 *** Hadlock-PierceSize-AgeIndex -3.2596 -3.9699 *** -3.2383 -4.0097 *** 58

0.8% 0.6% 0.4% 0.2% 0.0% (cid:882)0.2% (cid:882)0.4% (cid:882)0.6% (cid:882)0.8% 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014 stessA(cid:3)latoT(cid:3)/(cid:3)pU(cid:3)gnireveL(cid:3)teN Panel(cid:3)A 50.0% 45.0% 40.0% 35.0% 30.0% 25.0% 20.0% 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014 Year )%((cid:3)0(cid:3)>(cid:3)pU(cid:3)gnireveL(cid:3)teN Panel(cid:3)B Year Figure 1: Time series of net levering up behavior. Net levering up ratio is defined as long-term debt issuance net of long-term debt reductions minus equity issuance net of equity reductions, as a ratio to total book assets. Panel A displays the average net levering up ratio by fiscal year. Panel B displays the proportion of firms that increase net leverage by fiscal year. 59

serusaem desab-tekram ruof ruo no desab smrfi tros eW .esaercni egarevel fo soiloftrop dethgiew-eulav dna dethgiew-lauqe no desab sisylana oiloftroP :IIX elbaT egarevel rehgih gnivah sa defiissalc era elicret )mottob( pot eht otni llaf taht smriF .htnom hcae selicret otni , tkMIL hguorht tkMIL ,esaercni egarevel fo 4 1 eht fo seires-emit eht sserger dna oiloftrop hcae ni nruter egareva dethgiew-lauqe eht dnfi ew ,elicret hcae morf soiloftrop gnimroF .)segarevel rewol( sesaercni eht ni oiloftrop hcae ni nruter dethgiew-eulav eht esu eW .)4( hguorht )1( snmuloc ni ledom rotcaf-5 )3002( hguabmatS dna rotsaP eht no snruter oiloftrop eht morf sahplA eht troper C dna B slenaP .oiloftrop hgiH-woL tsoc-orez eht rof stluser lluf eht stneserp A lenaP .)8( hguorht )5( snmuloc ni ledom rotcaf-5 ,* yb detacidni si level %01 eht ta ecnacfiingiS .sesehtnerap eht ni detroper era srorre dradnatS .ylevitcepser ,soiloftrop hgiH dna woL eht rof noisserger rotcaf .*** yb level %1 dna ,** yb level %5 4tkMIL 3tkMIL 2tkMIL 1tkMIL 4tkMIL 3tkMIL 2tkMIL 1tkMIL WV WV WV WV WE WE WE WE )8( )7( )6( )5( )4( )3( )2( )1( oiloftroPhgiH-woL:AlenaP 9000.0- 0000.0 1100.0- ** 5800.0 ***8410.0- ***9310.0- ***4510.0- ***8900.0ahplA )9400.0( )3400.0( )9300.0( )3400.0( )2300.0( )1300.0( )0300.0( )0300.0( 351 351 351 351 351 351 351 351 .sbO .oN 6283.0 7864.0 4044.0 9673.0 3685.0 7676.0 3665.0 4556.0 2R oiloftroPwoL:BlenaP * 7600.0 3500.0 ** 1600.0 *** 6210.0 ***1110.0- ***5010.0- ***5110.0- ***5800.0ahplA )5300.0( )5300.0( )7200.0( )6300.0( )3200.0( )3200.0( )1200.0( )1200.0( 351 351 351 351 351 351 351 351 .sbO .oN 3538.0 4938.0 2668.0 7608.0 0609.0 4719.0 2019.0 0529.0 2R oiloftroPhgiH:ClenaP *** 7700.0 *** 3500.0 *** 2700.0 ** 2400.0 ** 7300.0 ** 4300.0 ** 9300.0 3100.0 ahplA )4200.0( )9100.0( )3200.0( )9100.0( )6100.0( )5100.0( )6100.0( )6100.0( 351 351 351 351 351 351 351 351 .sbO .oN 8386.0 9247.0 1896.0 4457.0 4478.0 8478.0 4478.0 1188.0 2R 60

5.0% 4.0% 3.0% 2.0% 1.0% 0.0% (cid:882)1.0% (cid:882)2.0% 0 1 2 3 4 5 6 7 8 9 10 11 12 snruteR(cid:3)dloh(cid:882)dna(cid:882)yuB Panel(cid:3)A 5.0% Least(cid:3)Net(cid:3)Levering(cid:3)Up Most(cid:3)Net(cid:3)Levering(cid:3)Up Most(cid:3)Net(cid:3)Levering(cid:3)Up(cid:3)(cid:882)(cid:3)Least(cid:3)Net(cid:3)Levering(cid:3)Up 4.0% 3.0% 2.0% 1.0% 0.0% (cid:882)1.0% (cid:882)2.0% 0 1 2 3 4 5 6 7 8 9 10 11 12 Months snruteR(cid:3)dloh(cid:882)dna(cid:882)yuB Panel(cid:3)B Least(cid:3)Net(cid:3)Levering(cid:3)Up Most(cid:3)Net(cid:3)Levering(cid:3)Up Most(cid:3)Net(cid:3)Levering(cid:3)Up(cid:3)(cid:882)(cid:3)Least(cid:3)Net(cid:3)Levering(cid:3)Up Months 5.0% 4.0% 3.0% 2.0% 1.0% 0.0% (cid:882)1.0% (cid:882)2.0% 0 1 2 3 4 5 6 7 8 9 10 11 12 snruteR(cid:3)dloh(cid:882)dna(cid:882)yuB Panel(cid:3)C 5.0% Least(cid:3)Likely(cid:3)to(cid:3)Net(cid:3)Lever(cid:3)Up Most(cid:3)Likely(cid:3)to(cid:3)Net(cid:3)Lever(cid:3)Up Most(cid:3)Likely(cid:3)(cid:882)(cid:3)Least(cid:3)Likely 4.0% 3.0% 2.0% 1.0% 0.0% (cid:882)1.0% (cid:882)2.0% 0 1 2 3 4 5 6 7 8 9 10 11 12 Months snruteR(cid:3)dloh(cid:882)dna(cid:882)yuB Panel(cid:3)D Least(cid:3)Likely(cid:3)to(cid:3)Net(cid:3)Lever(cid:3)Up Most(cid:3)Likely(cid:3)to(cid:3)Net(cid:3)Lever(cid:3)Up Most(cid:3)Likely(cid:3)(cid:882)(cid:3)Least(cid:3)Likely Months Figure 2: Buy-and-hold portfolios based on the four leverage increase measures. Panel A displays the buy-and-hold returnsoverthefollowingyearfromastrategyofbuyingfirmspredictedtoincreaseleverageandsellingfirmspredicted toreduceitusingthemarket-basedmeasure,LIMkt ,asdefinedinequation(10)usingthesignificantmarket-based 1 variablecoefficientsfromcolumn(9)inTableIII.AbnormalreturnsarecalculatedbasedonthePastorandStambaugh (2003) 5-factor model. Panel B presents the buy-and-hold returns based on the market-based measure, LIMkt , as 2 definedinequation(11)usingsignificantmarket-basedvariableandcontrolcoefficientsfromcolumn(9)inTableIII. Panel C displays the buy-and-hold returns over the following year from a strategy of buying HIGH net leverage increase firms and selling LOW net leverage increase firms using the market-based measure, LIMkt , as defined in 3 equation(12)usingthesignificantmarket-basedvariablecoefficientsfromcolumn(5)inTableIX.Abnormalreturns arecalculatedbasedonthePastorandStambaugh(2003)5-factormodel. PanelDpresentsthebuy-and-holdreturns basedonthemarket-basedmeasure,LIMkt ,asdefinedinequation(13)usingsignificantmarket-basedvariableand 4 control coefficients from column (5) in Table IX. 61