The Transmission of Monetary Policy through Bank Lending: The Floating Rate Channel

Finance and Economics Discussion Series Divisions of Research & Statistics and Monetary Affairs Federal Reserve Board, Washington, D.C. The Transmission of Monetary Policy through Bank Lending: The Floating Rate Channel Ippolito, F., A. K. Ozdagli, and A. Perez-Orive. 2017-026 Please cite this paper as: Ippolito, F., A. K. Ozdagli, and A. Perez-Orive. (2017). “The Transmission of Monetary Policy through Bank Lending: The Floating Rate Channel,” Finance and Economics Discussion Series 2017-026. Washington: Board of Governors of the Federal Reserve System, https://doi.org/10.17016/FEDS.2017.026. 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.

The Transmission of Monetary Policy through Bank Lending: The Floating Rate Channel* Filippo Ippolito Ali K. Ozdagli Ander Perez-Orive y Universitat Pompeu Fabra, Federal Reserve Bank of Boston Federal Reserve Board Barcelona GSE & CEPR September 2016 Abstract We examine both theoretically and empirically a mechanism through which outstanding bank loans a⁄ect the (cid:133)rm balance sheet channel of monetary policy transmission. Unlike other debt, most bank loans have (cid:135)oating rates mechanically tied to monetary policy rates. Hence, monetary policy-induced changes to (cid:135)oating rates a⁄ect the liquidity, balance sheet strength, and investment of (cid:133)nancially constrained (cid:133)rms that use bank debt. We show that (cid:133)rms(cid:151)especially (cid:133)nancially constrained (cid:133)rms(cid:151)with more unhedged bank debt display a stronger sensitivity of their stock price, cash holdings, sales, inventory, and (cid:133)xed capital investment to monetary policy. This e⁄ect disappears when policy rates are at the zero lower bound, which further supports the (cid:135)oating rate mechanism and reveals a new limitation of unconventional monetary policy. We argue that the (cid:135)oating rate channel can have a signi(cid:133)cant macroeconomic e⁄ect due to the large size of the aggregate stock of unhedged (cid:135)oating-ratebusinessdebt, ane⁄ectthatisatleastasimportantasthebanklendingchannel that operates through new loans. Keywords: monetary policy transmission, (cid:133)rm balance sheet channel, bank debt, (cid:135)oating interest rates, (cid:133)nancial constraints, hedging JEL classi(cid:133)cation: G21, G32, E52 * Earlier versions of the paper have been distributed with the title "Is Bank Debt Special for the Transmission of Monetary Policy? Evidence from the Stock Market." We thank Stefan Pitschner, Miguel Karlo De Jesus, and Yifan Yu for excellent research assistance. We are grateful to Adrien Auclert, Juliane Begenau, John Duca, Michael Faulkender, Je⁄ Fuhrer, Simon Gilchrist, Refet Gurkaynak, Satadru Hore, Victoria Ivashina, Sebnem Kalemli-Ozcan, Anil Kashyap, Anna Kovner, Alex Levkov, Juan Pablo Nicolini, Dino Palazzo, Daniel Paravisini, Joe Peek, Marcello Pericoli, Jose Luis Peydr(cid:243), Matt Pritsker, Manju Puri, Christina Romer, David Romer, Kristle Romero Cortes, Steve Sharpe, Nancy Stokey, Geo⁄Tootell, Scott Walker,ChristinaWang,MichaelWeber,PaulWillen,andaudiencesattheBostonFed,BostonCollege,the University of Illinois, Federal Reserve Board, Oxford University, Cass Business School, Queen Mary, UPF, the Bank of Spain, the Atlanta Fed, the 2013 NASM of the Econometric Society, the 2013 Meeting of the Society of Economic Dynamics, the 2013 NBER Summer Institute in Corporate Finance, the 2013 NBER Summer Institute in Monetary Economics, the 2013 Gerzensee ESSFM, the Barcelona GSE "II Asset Prices andtheBusinessCycleWorkshop,"the16thAnnualDNBResearchConference,the24thUNCAnnualCFEA Meeting, the 2015 NY Fed-NYU Stern Conference on Financial Intermediation, the 2015 FIRS conference, the Federal Reserve Board conference on (cid:147)Monetary Policy Implementation and Transmission in the Post- Crisis Period,(cid:148)and the 2016 AEA meetings for helpful comments. All remaining errors are our own. Ander Perez acknowledges (cid:133)nancial support from the Ministry of Economics of Spain grant ECO2012-32434, and from the Bank of Spain Programme of Excellence in Monetary, Financial, and Banking Economics. The views expressed in this paper are the authors(cid:146)and do not necessarily re(cid:135)ect those of the Federal Reserve Bank of Boston, the Federal Reserve System, or the Federal Open Market Committee (FOMC). Corresponding author. ali.ozdagli@bos.frb.org, 600 Atlantic Ave, Boston MA 02210. y

1 Introduction The (cid:133)rm balance sheet channel is one of the main mechanisms through which monetary policy is thought to interact with credit market imperfections to in(cid:135)uence (cid:133)rms(cid:146)investment, hiring, and output, and it operates by a⁄ecting (cid:133)rms(cid:146)balance sheet strength and ability to access new external (cid:133)nance (Bernanke and Gertler (1995), Mishkin (1995)). In this paper we examine, both theoretically and empirically, a mechanism in which outstanding bank loans are an important component of the (cid:133)rm balance sheet channel, motivated by two observations typically overlooked in the monetary economics literature: Monetary policy drives the reference rate underlying (cid:135)oating-rate loan arrangements (Figure 1), and the vast majority of corporate loans from banks feature (cid:135)oating interest rates (Figure 2). Does monetary policy have a strong e⁄ect on (cid:133)rms(cid:146)liquidity positions and their ability to (cid:133)nance future projects by causing changes in the debt service burden of existing (cid:135)oating-rate bank loans? We answer this question through the lens of both stock prices and balance sheet variables by theoretically analyzing a (cid:133)rm that can borrow at (cid:135)oating and (cid:133)xed rates, and by empirically studying (cid:133)rm-level information on the usage of bank debt and (cid:135)oating-rate debt and a new database of (cid:133)rms(cid:146)hedging activity. [FIGURES 1 & 2 ABOUT HERE] We introduce a theoretical framework that considers a (cid:133)rm(cid:146)s choice of debt structure, investment, and dividends. To be able to address our main questions, it is crucial that our analysis features long-term debt, an interest rate exposure decision through a (cid:135)oating vs. (cid:133)xed rate debt choice, and (cid:133)nancing constraints. We start with a stylized two-period model that has the advantage of o⁄ering an analytical solution, while still providing the key insights of our thesis. While the optimal investment of a (cid:133)nancially unconstrained (cid:133)rm is insensitive to internal funds, the amount of internal funds matters for the investment of a constrained (cid:133)rm. In the presence of (cid:135)oating-rate debt, policy rate changes a⁄ect the (cid:133)rm(cid:146)s interest expense on existing debt and therefore internal funds. This di⁄erential e⁄ect on investment between constrained and unconstrained (cid:133)rms translates into a corresponding di⁄erential stock market reaction to an unexpected change in monetary policy. We integrate these ideas into a more general dynamic model that also takes into account important issues such as monetary policy persistence, rationally anticipated monetary policy shocks, e⁄ects of costly distress, and the quantitative strength and duration of the e⁄ects of our mechanism. First, the dynamic model provides a quantitative assessment of the (cid:135)oating rate channel which is broadly consistent with the economic signi(cid:133)cance that we obtain in our empirical regressions. Second, the dynamic model suggests that the results from the stylized 1

simple model are robust to considering persistent and rationally anticipated monetary policy shocks. Third, themodel haspredictionsaboutthee⁄ectsthatchangesininterestrateshave on the expected likelihood and cost of (cid:133)nancial distress and shows that this link ampli(cid:133)es movements in stock prices. Finally, the theoretical framework makes it clear that a very general notion of (cid:133)nancial constraints is su¢ cient to generate our results. In particular, it su¢ ces that (cid:133)nancially constrained (cid:133)rms display some sensitivity of their investment to internal funds and that they are more productive on the margin than unconstrained (cid:133)rms.1 Ourempirical(cid:133)ndingsprovidesupporttothepredictionsofthemodel. We(cid:133)rstdocument that corporations borrow from banks mostly at a (cid:135)oating rate, whereas they mostly issue other forms of debt at a (cid:133)xed rate.2 Using market-based monetary policy surprise measures as in Kuttner (2001) and G(cid:252)rkaynak, Sack, and Swanson (2005), we (cid:133)nd that while a typical (cid:133)rm(cid:146)s stock price decreases about 4 to 5 percent in response to a 100 basis point (bp) surprise increase in the federal funds rate, the stock price of a (cid:133)rm that has one standard deviation more bank debt relative to assets decreases about 1.6 percent more. Crucially, all of the additional stock price decline due to the use of bank debt comes from the sample of unhedged (cid:133)rms, consistent with the (cid:135)oating rate channel, as seen in Figure 3. Our results are robust to controlling for the determinants of bank debt usage and hedging and to using instrumental variables analysis to deal with any possible endogeneity of the bank debt usage and hedging decisions.3 [FIGURE 3 ABOUT HERE] In the absence of (cid:133)nancial frictions, our evidence could be interpreted as a simple cash transfer between a (cid:133)rm(cid:146)s shareholders and its creditors, with no real e⁄ects. In the presence 1Although our focus is on the e⁄ects through existing loans, this channel may be conceptually similar to one operating through new loans. In this alternative case, the movements in internal funds would be caused by the issuance of new debt or re(cid:133)nancing of existing debt at new interest rates. In principle, both mechanisms are not mutually exclusive and might, in fact, reinforce each other. We discuss in detail in the literature review (Section 1) the di⁄erences between both mechanisms. 2As Figure 2 illustrates, 76 percent of the debt of (cid:133)rms that borrow solely from banks has a (cid:135)oating rate, compared with 9 percent of debt for those (cid:133)rms that have only nonbank debt. This result is in line with Faulkender (2005), who (cid:133)nds that about 90 percent of syndicated bank loans to chemical corporations are issued at a (cid:135)oating rate, and with Vickery (2008), who (cid:133)nds that about 70 percent of C&I loans from commercial banks have a (cid:135)oating rate in the Federal Reserve(cid:146)s Survey of Terms of Business Lending. 3To deal with the possibility that omitted variables drive both the choice of bank debt usage or hedging and the responsiveness to monetary policy, we control for all the (cid:133)rm characteristics that have been shown to in(cid:135)uence debt structure: (cid:133)rm size, leverage, pro(cid:133)tability, growth opportunities (market-to-book ratio), risk (CAPM Beta, cash-(cid:135)ow volatility, demand sensitivity to interest rates), cash holdings, and (cid:133)nancial constraint measures. In addition, we instrument for bank debt usage, following Faulkender and Petersen (2006) and Santos and Winton (2008), using proxies for (cid:133)rm visibility and (cid:133)rm uniqueness, both of which drive the ability to issue public debt (and thus the dependence on bank debt) and can be argued to be orthogonal to our dependent variable. 2

of (cid:133)nancing frictions, however, the additional interest expense may a⁄ect the (cid:133)rm(cid:146)s liquidity position, leverage, and overall balance sheet strength, which in turn could a⁄ect the (cid:133)rm(cid:146)s ability to (cid:133)nance pro(cid:133)table investment opportunities, as our theory predicts. We (cid:133)nd that (cid:133)nancialconstraintsincreasethepolicyratesensitivityofstockpricesofunhedgedbankdebt users signi(cid:133)cantly. However, (cid:133)nancial constraints do not change this sensitivity for hedged bank debt users, a (cid:133)nding that suggests an ampli(cid:133)cation of the (cid:135)oating rate channel through the e⁄ect of (cid:133)nancing constraints. Next, we provide further evidence consistent with our theoretical mechanism by using data on real and (cid:133)nancial decisions of (cid:133)rms. First, we show that the interest coverage ratio of a (cid:133)rm responds signi(cid:133)cantly more strongly to monetary policy as the share of bank loans over total assets increases, but only for (cid:133)rms that do not hedge against interest rate risk. The e⁄ect is sizable and persists for up to six quarters. This (cid:133)nding suggests that the exposure to interest rate (cid:135)uctuations through unhedged bank debt exposes (cid:133)rms to signi(cid:133)cant liquidity shocks. We con(cid:133)rm this argument by showing that the cash holdings of (cid:133)nancially constrained (cid:133)rms that use bank debt and do not hedge are very sensitive to monetary policy while those of (cid:133)nancially unconstrained or hedged bank debt users are not. This (cid:133)nding suggests that a monetary policy tightening might hurt (cid:133)rms exposed to the (cid:135)oating rate channel by draining internal liquid resources of (cid:133)rms with limited access to external (cid:133)nance. Consistent with this implication, we show that there is a strong positive relationship between bank debt usage and the sensitivity of inventory investment, (cid:133)xed investment,andsalestomonetarypolicychangesfor(cid:133)nanciallyconstrained(cid:133)rmsthatdonot hedge, butthatthesee⁄ectsaresigni(cid:133)cantlysmallerorabsentwhen(cid:133)rmshedgeinterestrate risk or do not face signi(cid:133)cant (cid:133)nancial constraints. The e⁄ects are quantitatively large: Six quarters after a 100bp monetary policy tightening, (cid:133)nancial constraints are associated with additional decreases ininventories and (cid:133)xed investment of 22.1%and 15.8%, respectively, for ahypothetical(cid:133)rmfully(cid:133)nancedbybankdebtandunhedged, buttheseadditional decreases are reduced to less than half when (cid:133)rms are hedged. Taken together, our evidence suggests that the e⁄ect of the (cid:135)oating rate channel extends beyond a simple reallocation of cash (cid:135)ows between lenders and shareholders and has signi(cid:133)cant real implications for the a⁄ected (cid:133)rms. The potential macroeconomic relevance of our monetary policy transmission mechanism is supported by the large amount of debt that is exposed to interest rate risk. We estimate thatintheUnitedStatesthelowerboundforthedebtexposedtointerestrateriskisbetween $3.2 and $4.1 trillion of the $12.5 trillion of total debt of non(cid:133)nancial businesses as of yearend 2015, and represents roughly 20% of annual GDP ($18.0 tn in 2015).4 We also provide 4Note that this is a lower bound estimate of the amount of debt exposed to interest rate risk, because we are basing our estimates on the fraction of debt that is tied to LIBOR, which is the most common but not the only base rate for (cid:135)oating rate arrangements. An example of a common alternative base rate is the 3

some measure of the macroeconomic importance of our proposed mechanism by comparing it to the traditional bank lending channel and show that the (cid:135)oating rate channel is at least as important as the traditional bank lending channel. Finally, as additional evidence regarding the importance of the (cid:135)oating rate channel, we study the recent zero-lower-bound environment. During this period, the reference rates of (cid:135)oatingrateloans wereboundfrombelowatzeroandthereforeanye⁄ect of bankdebt usage should work through channels other than the (cid:135)oating rate channel. We show that bank debt and hedging have had no e⁄ect on the monetary policy sensitivity of stock prices during the unconventional policy period. Combined with the importance of the (cid:135)oating rate channel during periods of conventional monetary policy, this (cid:133)nding suggests that the absence of the (cid:135)oating rate channel might have limited the e¢ cacy of unconventional monetary policy during the recent period. This result could shed light on the uncertainty regarding the costs and bene(cid:133)ts of unconventional policy, a topic that has gained increased attention recently (e.g., Evans, Fisher, Gourio, and Krane (2015)) as the Federal Reserve contemplates further rate hikes following the recent target rate lifto⁄. Related Literature Theliteratureonthecreditchannelofmonetarypolicyhasputforwardtwomainchannels to explain why (cid:133)nancing constraints of (cid:133)rms might amplify the e⁄ects of monetary policy (Bernanke and Gertler (1995)). The (cid:133)rst channel, the (cid:133)rm balance sheet channel, captures direct and indirect e⁄ects of monetary policy on (cid:133)rms(cid:146)balance sheet strength and ease of access to external (cid:133)nance. Gertler and Gilchrist (1994) (cid:133)nd that inventory investment, sales, and short-term debt of an aggregate of small (cid:133)rms are more responsive to changes in monetary policy than those of an aggregate of large (cid:133)rms. Ashcraft and Campello (2007) and Ciccarelli, Maddaloni, and Peydr(cid:243) (2014) control for the possibility that these results might be driven by a contraction of bank lending supply, and both (cid:133)nd evidence of a strong (cid:133)rm balance sheet channel. None of these papers speci(cid:133)es the precise mechanisms through which the (cid:133)rm balance sheet channel operates, however, which is an important contribution of our paper. We show that a quantitatively signi(cid:133)cant (cid:133)rm balance sheet channel operates through the e⁄ect of monetary policy on (cid:133)rms(cid:146)debt service burden when they use bank debt as a source of (cid:133)nance and retain exposure to interest rate risk by not hedging. Although our focus is on the e⁄ects through existing loans, this channel may be conceptually similar to one operating through new loans. In this alternative case, the movements in internal funds would be caused by the issuance of new debt or re(cid:133)nancing of existing debt at new interest rates. In principle, both mechanisms are not mutually exclusive and might, in fact, reinforce each other. There are however some important di⁄erences between prime rate (displayed in Figure 1), which is also closely tied to policy rates. 4

both mechanisms that we should highlight. First, we (cid:133)nd that short-term debt does not signi(cid:133)cantly increase the sensitivity of (cid:133)rms(cid:146)stock prices to monetary policy, in contrast to bank debt. This suggests that a channel operating through the re(cid:133)nancing of maturing debt might not be as strong as our mechanism. Second, there are important amplifying mechanisms in our channel which would be absent in a channel operating through new loans. For example, the literature has long recognized that long-term debt can create important agency costs, such as underinvestment and risk-shifting (Myers (1977), Bodie and Taggart (1978), and Himmelberg and Morgan (1995)), which means that a monetary policy tightening might worsen these agency costs, particularly through an increased debt service burden under long-term(cid:135)oating rate bank debt. As another example, the e⁄ect of (cid:135)oating rate bank debt on the interest coverage ratio is more likely to lead to covenant violations, which have important implications for (cid:133)rms(cid:146)capital expenditures, as shown in Nini, Su(cid:133), and Smith (2012). Finally, while the pass-through of policy rates to (cid:135)oating interest rates of long-term debt is complete and occurs at frequent resetting dates, the pass-through to short-term bank (cid:133)nancing rates has been shown to be slow (De Bondt, Mojon, and Valla (2005), Illes and Lombardi (2013)). The second channel, the bank lending channel, has focused on why bank lending to (cid:133)rms might be special for the transmission of monetary policy to the real economy (Bernanke and Blinder (1988), Bernanke and Gertler (1995), Stein (1998), Van den Heuvel (2002), and Bolton and Freixas (2006)). All of these theories focus on how the supply of new bank credit might be a⁄ected by monetary policy due to the presence of bank (cid:133)nancing frictions.5 Our proposed mechanism focuses instead on the transmission through loans outstanding at the time of monetary policy actions. Also, our mechanism is unrelated to how much banks su⁄er from(cid:133)nancingconstraints, soitcouldbeactivethroughallbanksatalltimes, unlikeexisting mechanisms, whose potency may be restricted to a subset of banks during periods of credit market distress. Our proposed mechanism is closely related to the burgeoning literature that introduces a similar transmission channel for households. Analogous to our mechanism, this literature suggests that monetary policy has real implications by in(cid:135)uencing households(cid:146)cost of servicing their (cid:135)oating rate debt and, as a result, their disposable income and consumption (Calza, Monacelli, and Stracca (2013), Di Maggio, Kermani, and Ramcharan (2014)). The extensive literature on the relationship between (cid:133)rm fundamentals and debt structure helps us control for determinants of bank debt usage with su¢ cient accuracy, thereby 5Consistent with a role for bank (cid:133)nancial health, the contraction in the supply of lending following a tightening of monetary policy has been found to be stronger in small, less liquid, and more leveraged banks (Kashyap and Stein (2000), Kishan and Opiela (2000), and Jimenez, Ongena, Peydr(cid:243), and Saurina (2012)), and in banks that are not a¢ liated with multibank holding companies (Ashcraft (2006)). 5

alleviating concerns regarding omitted variables. Most of the theoretical literature argues thatbankshaveanadvantageintheresolutionofinformationasymmetriesandrenegotiation of debt contracts compared to holders of public debt because banks have better monitoring ability and do not su⁄er from coordination problem of dispersed bondholders.6 Hence, (cid:133)rms withahighdegreeofinformationasymmetryshouldrelymoreonbankdebt. Incontrast, the models in Diamond (1991) and Rajan (1992) suggest that this prediction holds for high and medium credit quality (cid:133)rms, whereas for low-quality (cid:133)rms the costs of bank monitoring may outweigh the bene(cid:133)ts that would make public debt (cid:150)e.g., junk bonds(cid:150)more preferable. This nonlinear relationship between credit quality and bank debt usage helps alleviate the concern that our results may be driven by (cid:133)nancial constraints. Consistent with this argument, we con(cid:133)rm that our results are not driven by credit quality proxies used in the empirical literature on debt structure (Denis and Mihov (2003) and Lin, Ma, Malatesta, and Xuan (2013)), such as (cid:133)rm size, pro(cid:133)tability, and market-to-book ratio (growth opportunities), as well as by other measures that capture the (cid:133)nancial situation of a (cid:133)rm, such as leverage, cash holdings, and risk (CAPM beta and cash (cid:135)ow volatility), or by debt maturity.7 There is also a good understanding of the determinants of hedging, which allows us to control for them and to use some of the arguably exogenous determinants as instruments. Existing theory predicts that hedging activities are positively related to the severity of (cid:133)nancing constraints (Stulz (1984), Froot, Scharfstein, and Stein (1993)). Although some recent evidence, and our own data, cast doubt on the sign of this relationship (Stulz (1996), Rampini, Su(cid:133), and Viswanathan (2014)), it is clear that (cid:133)nancial constraints can be an important driver of hedging, and we control for them using various measures. Still, several determinants of hedging do not have a direct relationship with the responsiveness of stock returns to monetary policy, which enables us to use them as instruments. In particular, we follow the instrumental variables approach in Campello, Lin, Ma, and Zou (2011), who focus on institutional features of the U.S. tax system. The kinks or discontinuities of the tax schedule create a convexity of tax rates, which enables (cid:133)rms to reduce their expected tax liabilities by hedging in order to minimize income volatility (Smith and Stulz (1985), Graham and Smith (1999), and Petersen and Thiagarajan (2000)). One important question is why most bank lending arrangements involve a (cid:135)oating rate instead of a (cid:133)xed rate despite the fact that many (cid:133)rms hedge the interest rate risk associated with these loans. One answer could arise from the trade-o⁄between (cid:133)rms(cid:146)needs and banks(cid:146) 6SeeDiamond(1984),Fama(1985),HolmstromandTirole(1997),BootandThakor(2009),Rajan(1992), Bolton and Scharfstein (1996), Bolton and Freixas (2000). 7In addition, we instrument for bank debt usage, following Faulkender and Petersen (2006) and Santos and Winton (2008), using proxies for (cid:133)rm visibility and (cid:133)rm uniqueness, both of which drive the ability to issue public debt and can be argued to be orthogonal to our dependent variable. 6

cost of capital. A (cid:133)rm that wants to borrow at a (cid:133)xed rate may have limited access to other (cid:133)xed-rate sources of (cid:133)nancing, such as bonds, whereas the bank might prefer to lend at (cid:135)oating rates, in which case hedging bridges the gap between the desire of the bank and the (cid:133)rm. As discussed by Vickery (2008), there are at least two reasons why banks might prefertolendat(cid:135)oatingrates. First, risinginterestratescancausedepositout(cid:135)owsfromthe banks, and it is costly for banks to replace these out(cid:135)ows with other sources of (cid:133)nancing. Lending at a (cid:135)oating rate would provide a partial hedge against these out(cid:135)ows. Second, (cid:135)oating rate business loans can be used to hedge the maturity mismatch between deposits and long-term mortgage loans. Another piece of evidence that banks are likely willing to lend corporations only at (cid:135)oating rate comes from the fact that even for (cid:133)rms that have access to both bonds and bank debt, most of the bonds are (cid:133)xed rate whereas most of their bank debt is (cid:135)oating rate. If the (cid:135)oating vs. (cid:133)xed rate choice for bank debt were driven by (cid:133)rm characteristics, the (cid:133)rms would likely choose similar rate arrangements for their bonds and bank loans, which does not seem to be the case. Finally, this paper is related to a recent literature that uses the relationship between stock prices and monetary policy to shed light on questions that are otherwise di¢ cult to answer. For example, the relationship between stock prices and monetary policy surprises is used by Gorodnichenko and Weber (2014) to identify the cost of price stickiness; by English, Van den Heuvel, and Zakrajsek (2014) to study the e⁄ect of monetary policy on bankpro(cid:133)tabilitythroughmaturitytransformation; andbyChodorow-Reich(2014) tostudy the e⁄ect of unconventional monetary policy on (cid:133)nancial institutions. The rest of the paper is organized as follows. In Section 2 we introduce and analyze our theoretical results. In Section 3 we describe our data. Our empirical results on stock returns and on balance sheet variables are discussed in Sections 4 and 5, respectively, and in Section 6 we analyze the macroeconomic relevance of our proposed channel. Finally, Section 7 concludes. 2 Theoretical Framework 2.1 Simple Model This section aims to provide a simple setting with a closed form solution that motivates the (cid:135)oating rate channel we study in our empirical analysis. Therefore, we make some simplifyingassumptions that arerelaxedinourdynamicsettinginSection2.2. Inparticular, (cid:133)rms do not issue equity, there is no costly distress, and (cid:133)rms are identical except for their debt structure and (cid:133)nancial constraints. We consider a two-period (three-date) economy with dates t = 0;1;2 . Firms invest a f g 7

(cid:133)xed amount K at time t = 0, which produces a return f (K ) in t = 1, and a variable 0 0 amount K in t = 1, which produces a return f (K ) in t = 2. For simplicity, we assume that 1 1 K is (cid:133)nanced exclusively with long-term debt, which can be (cid:135)oating rate debt (bank loans), 0 L , or (cid:133)xed-rate debt (bonds or hedged bank loans), B . Let l = L =K be the fraction of 0 0 0 0 (cid:135)oating rate debt, so that K = L +B = lK +(1 l)K : (1) 0 0 0 0 0 (cid:0) Floating rate debt requires the payment of interest r L at time t = 1, and of interest 1 0 and principal (1+r )L in t = 2. Fixed rate debt requires the payment of a (cid:133)xed coupon 2 0 r B at time t = 1, and of the (cid:133)xed coupon and principal (1+r )B in t = 2. c 0 c 0 We model monetary policy in the simplest way possible to provide a clear exposition of our mechanism. The rate r su⁄ers an unexpected change after choices are made in t = 0, 1 and we identify this shock as a monetary policy action. We assume that r is una⁄ected by 2 monetary policy.8 A (cid:133)rm(cid:146)s internal funds at the end of the (cid:133)rst period (in t = 1) is N = f (K ) r B r L : (2) 1 0 c 0 1 0 (cid:0) (cid:0) The (cid:133)rm can borrow b in t = 1, subject to an exogenous borrowing constraint 1 (cid:22) b b; (3) 1 (cid:20) whereb is one-perioddebtthat requires arepayment of (1+r )b int = 2. The(cid:133)rminvests 1 2 1 again in t = 1 an amount K = N +b d ; (4) 1 1 1 1 (cid:0) where d are dividends paid in t = 1. A timeline of events is described in Figure 4. 1 [FIGURE 4 ABOUT HERE] The (cid:133)rm maximizes the present value of dividends in t = 0. Dividends, d , are paid in t 8This is essentially a comparative statics exercise, informally referred to as an "MIT shock" by, for example, Guerrieri and Uhlig (Handbook of Macroeconomics, forthcoming) and commonly used in the macroeconomics literature (see (Gertler and Kiyotaki (2010), Eggertson and Krugman (2012), or Guerrieri and Lorenzoni (2016)). Moreover, while monetary policy a⁄ects r directly, it could have persistent e⁄ects and 1 cause changes in r as well. We consider interest rate persistence and rationally anticipated interest rate 2 shocks in the dynamic model of Section 2.2. 8

t = 1 and t = 2, and given by d = N +b K ; and (5) 1 1 1 1 (cid:0) d = f (K ) R B R L R b = f (K ) R (1 l)K R lK R b ; (6) 2 1 c 0 2 0 2 1 1 c 0 2 0 2 1 (cid:0) (cid:0) (cid:0) (cid:0) (cid:0) (cid:0) (cid:0) where R = 1+r , R = 1+r , and R = 1+r represent the gross interest rates. 1 1 2 2 c c (cid:22) A (cid:133)rm that is (cid:133)nancially constrained in t = 1 (b = b) will optimally set d = 0 and invest 1 (cid:22) K = N +b: (7) 1 1 An unconstrained (cid:133)rm instead invests according to the neoclassical investment rule, f (K ) = 1+r : (8) 0 1 2 We are interested in how the (cid:133)rm value and investment react to changes in r once the 1 long-term (cid:133)nancing choices are made. The following proposition is central to our empirical analysis: Proposition 1 Floating rate debt usage increases the monetary policy sensitivity of stock prices and investment of (cid:133)nancially constrained (cid:133)rms. In particular, (i) (cid:135)oating rate debt usage increases the policy rate sensitivity of stock prices for all (cid:133)rms, but the e⁄ect is stronger for (cid:133)nancially constrained (cid:133)rms and (ii) (cid:135)oating rate debt usage increases the policy rate sensitivity of investment (K ) of (cid:133)- 1 nancially constrained (cid:133)rms, while it does not a⁄ect the sensitivity of investment of (cid:133)nancially unconstrained (cid:133)rms. Appendix A provides a formal proof of this proposition, and here we o⁄er an intuitive explanation. The investment of (cid:133)nancially constrained (cid:133)rms in t = 1, given by equation (7), depends on the (cid:133)rm(cid:146)s internal funds N at that point (equation (2)), which in turn are 1 in(cid:135)uenced by the interest expense incurred (r B + r L ). An increase in interest rate r c 0 1 0 1 leads to a reduction in the (cid:133)rm(cid:146)s internal funds at the end of t = 1if the (cid:133)rm uses any (cid:135)oating rate debt, which in turn leads to lower investment, K , in the following period. It 1 is clear from equation (2) that this e⁄ect is stronger for (cid:133)rms with more (cid:135)oating rate debt L . Financially unconstrained (cid:133)rms, however, invest in t = 1 an amount given by (8), which 0 equates the marginal product of capital to the interest rate in the second period, r . For 2 their investment, therefore, their internal funds in t = 1 and the amount of (cid:135)oating rate debt L are irrelevant.9 0 9Ifmonetarypolicyhaspersistente⁄ects,theinvestmentoftheunconstrained(cid:133)rmcanreacttomonetary 9

The amount of (cid:135)oating rate debt can a⁄ect the (cid:133)rm(cid:146)s stock market valuation through two channels. First, an increase in interest rates increases the interest payments on (cid:135)oating rate debt at the expense of dividends paid to shareholders, which a⁄ects all (cid:133)rms irrespective of their (cid:133)nancial constraints. Second, as discussed previously, (cid:133)nancially constrained (cid:133)rms su⁄er, in addition, a reduction in their second-period investment K as a result of the 1 increased interest expense and reduced internal funds, which further hurts future dividends and current stock valuation.10 Summing up, this simple model shows how the (cid:135)oating rate channel a⁄ects both the investment and the stock value of (cid:133)nancially constrained (cid:133)rms more than those of unconstrained (cid:133)rms. 2.2 Dynamic Model The simple model introduced in Section 2.1 formalizes the essential intuition of our mechanism and delivers closed-form expressions. We now integrate those ideas into a more generaldynamicmodelthatalsotakesintoaccountimportantissuessuchasmonetarypolicy persistence, e⁄ectsofcostlydistress, andthequantitativestrengthanddurationofthee⁄ects of our mechanism. To be able to address these issues, we build on Gomes, Jermann, and Schmid (2016) and introduce a partial equilibrium dynamic model of (cid:133)rm investment and (cid:133)nancing decisions that features long-term debt, an interest rate exposure decision through issuance of (cid:135)oating or (cid:133)xed rate debt, and (cid:133)nancing constraints, in an environment with uncertainty arising from interest rate (monetary policy) shocks and operating cost shocks.11 The model considers both (cid:133)nancially constrained and (cid:133)nancially unconstrained (cid:133)rms, and we use the model(cid:146)s stationary distribution of (cid:133)rms to run standard regressions capturing the e⁄ect of monetary policy on stock returns and investment. 2.2.1 Financially Constrained Firms Technology Capital k invested in period t 1 produces output in period t and depreciates fully in t (cid:0) one period. Each (cid:133)rm produces according to the function y = Ak(cid:11); (9) t t policy through changes in R as well. This classic interest rate channel, however, does not depend on how 2 much (cid:135)oating rate debt the (cid:133)rm has and therefore is not relevant to our (cid:135)oating rate channel. 10The dynamic model in Section 2.2 describes how e⁄ects on the expected likelihood and cost of (cid:133)nancial distress constitute another channel through which changes in interest rates interact with the presence of (cid:135)oating rate debt and (cid:133)nancial constraints to explain movements in stock prices. 11Gomes,Jermann,andSchmid(2016)considertheimplicationsofnominaldebtinamodelwithin(cid:135)ation dynamics. Our paper abstracts from these issues, as they are not part of our mechanism, and considers an economy where debt repayments are determined in real terms. 10

where A is a constant productivity parameter. Firm-level pro(cid:133)ts are also subject to additive idiosyncratic shocks, z , so that operating pro(cid:133)ts are equal to t (cid:25) = y z : (10) t t t (cid:0) We assume that z is i.i.d. across (cid:133)rms and time, and takes values over the interval [0;z], t z z with (cid:30)(z)dz = d(cid:8)(z). We think of these as direct negative shocks to (cid:133)rms(cid:146)operating Z0 Z0 income and not necessarily output. They summarize the overall (cid:133)rm-speci(cid:133)c component of their business risk. They are not proportional to the size of the (cid:133)rm, suggesting some element of (cid:133)xed costs and allowing for the possibility of negative pro(cid:133)ts. Financing Firmsfundthemselvesbyissuingbothequityandmulti-perioddebt,whichcanbe(cid:135)oating or (cid:133)xed rate. Let b denote the stock of outstanding debt at the beginning of period t. To t capturethefactthatoutstandingdebtisof(cid:133)nitematurity,weassume,asinGomes,Jermann, and Schmid (2016), that in every period t a fraction (cid:21) of the principal is paid back. The remaining (1 (cid:21)) fraction remains outstanding and cannot be repaid early, so as a result (cid:0) the debt has an expected life of 1=(cid:21). In addition to principal amortization, the (cid:133)rm is also required to pay a periodic coupon c per unit of outstanding debt. The coupon depends on t the amount of (cid:135)oating rate debt: c = (cid:18) r +(1 (cid:18) )(r+ ); (11) t t t t (cid:0) where (cid:18) is the share of debt that is (cid:135)oating rate, which pays an interest rate r set by t t monetary policy. The (cid:133)xed interest rate r + payable to the remaining 1 (cid:18) fraction of t (cid:0) debt is equal to the unconditional average of the (cid:135)oating interest rate plus a (cid:133)xed amount > 0 that captures the cost of hedging. The law of motion for (cid:135)oating rate debt is given by (cid:18) b = (cid:18) (1 (cid:21))b +(cid:18)NEW (b (1 (cid:21))b ); (12) t+1 t+1 t (cid:0) t t+1 t+1 (cid:0) (cid:0) t where for simplicity we assume that all new debt, b (1 (cid:21))b , has to be either (cid:133)xed or t+1 t (cid:0) (cid:0) (cid:135)oating rate, so that (cid:18)NEW 0;1 . Debt is assumed to be safe, so the coupon paid, c , t+1 2 f g t does not feature a risk spread. We assume that (cid:133)rms are subject to a borrowing constraint: b b: (13) t (cid:20) To be able to discuss the (cid:133)rm(cid:146)s equity issuance, we (cid:133)rst de(cid:133)ne n to be the internal funds t 11

of the (cid:133)rm at the beginning of the period, given by n = (cid:25) (c +(cid:21))b : (14) t t t t (cid:0) Wheninternalfunds,n ,arepositive,the(cid:133)rmpaysdividendsequaltod = (cid:26)n ,where(cid:26)is t t t a reduced-form way of capturing (cid:133)rms(cid:146)dividend policy.12 When n is negative, however, the t (cid:133)rm is considered to be in violation of a net worth (cid:133)nancial covenant imposed by creditors and has to issue equity (d < 0). Instances of n < 0 thus capture periods of (cid:133)nancial t t distress. When n < 0, debtholders force the (cid:133)rm to issue equity to (cid:133)nance the negative t internal funds. Following Gomes (2001), equity issues carry a proportional issuance cost (cid:17), so that the dividend net of equity issuance costs is given by d = 1+(cid:17)1 (cid:26)n ; (15) t nt<0 t f g where (cid:17) > 0 and 1 is an indicato(cid:0)r function of(cid:1)strictly negative internal funds.13 We nt<0 f g interpret (cid:17) in our context as a comprehensive parameter that captures di⁄erent costs of (cid:133)nancial distress. They can be considered as reallocation of part of the (cid:133)rm(cid:146)s surplus from shareholderstolendersuponviolationofsuchacovenantduetoatransferofcontrolrightsto creditors, as (cid:135)otation costs of issuing equity, or as other costs of (cid:133)nancial distress.14 Because we only allow for covenant violations as "technical defaults," bondholders are always paid in full, which allows us to avoid the calculation of the price of risky debt. This simplifying assumption allows us to impose a cost on (cid:133)rms with low internal funds, and builds on similar notions of (cid:133)nancial distress introduced in Asquith et al (1994) or Strebulaev (2007).15 Other moreexplicitwaysofmodellingdebtdefaultwouldhavesimilarimplicationsforourpurposes. Our model choice assumes that (cid:133)rms are unable to issue equity in circumstances other than instances of negative internal funds, and, even then, only in an amount needed to cover those negative internal resources. The empirical motivation for this choice is that there are 12Intheabsenceofthispassivedividendpayoutrule, (cid:133)nanciallyconstrained(cid:133)rmsinourmodelwouldnot pay any dividends, but this would be in stark contrast with the data. Our payout rule captures a reduced form version of agency frictions and information asymmetry that encourage dividend payments in the real world (Lintner (1956), Fama and French (2001), Floyd, Li and Skinner (2015)). 13Gomes (2001) also estimates a (cid:133)xed cost component in equity (cid:135)oation costs, which we ignore for clarity of exposition as it is unlikely to have any signi(cid:133)cant qualitative e⁄ects on our results. 14Therecentsurgeinliteratureoncovenantviolationsshowsthattheseeventshaveimportantimplications for(cid:133)rm(cid:146)s(cid:133)nancingandinvestmentbehaviorsandthattheyhappenmuchmorefrequentlythanbankruptcies (ChavaandRoberts(2008),Nini,SmithandSu(cid:133)(2012),andAcharya,Almeida,Ippolito,andPerez(2014)). One of the most important e⁄ects highlighted by this literature is the (cid:133)rms(cid:146)need to deleverage following a covenant violation, through, amongst other means, equity issuances, as in our model. 15In Strebulaev (2007), for example, the managers of (cid:133)rms whose (cid:133)nancial condition deteriorates su¢ ciently must take corrective action in the form of costly (cid:133)re sales of assets. 12

very few secondary issues of equity in the data.16 On a more conceptual basis, adding a realistically calibrated endogenous equity choice would be unlikely to alter our results, and would come at the cost of a loss of tractability. Optimization Firms maximize the present value of dividends d paid to shareholders. The dynamic t programming problem for a (cid:133)rm consists of a choice of dividends d , next period capital t k , debt b , and the share of new debt which is (cid:135)oating rate (cid:18)NEW, taking as given the t+1 t+1 t+1 two exogenous shocks: idiosyncratic productivity shock z and interest rate r , the latter of t t which follows a Markov process.17 We claim, and check in our numerical simulations, that (cid:133)nancially constrained (cid:133)rms are permanently credit constrained so that (13) is binding, b = b, and, as a result, the amount t+1 of capital, k ; follows the law of motion18 t+1 (cid:22) k = (1 (cid:26))n +(cid:21)b: (16) t+1 t (cid:0) Finally, dividends are given by rule (15). The only choice that is not a corner solution is the share of new debt which is (cid:135)oating rate, (cid:18)NEW, which is our main focus. t+1 The value of the (cid:133)rm to its shareholders, denoted V , is the present value of the dividend c distributions: 1 z V (k ;b ;(cid:18) ;z ;r ) = max d +E V (k b ;(cid:18) ;z ;r )d(cid:8)(z) (17) c t t t t t t t c t+1; t+1 t+1 t+1 t+1 (cid:18)N t+ E 1 W (cid:26) 1+r t+1 Z0 (cid:27) where the conditional expectation E is taken over the distribution of shocks to r .19 t t+1 2.2.2 Financially Unconstrained Firms Financially unconstrained (cid:133)rms are identical to constrained (cid:133)rms except in their ability to access external (cid:133)nance. In particular, unconstrained (cid:133)rms can also fund themselves by issuing both equity and multi-period debt, and they face no restriction in the amount of debt 16According to SDC Platinum data (Calomiris and Tsoutsoura, 2010), non(cid:133)nancial non-utility (cid:133)rms have about 100 seasoned equity o⁄erings per year, which implies that only about 2% of (cid:133)rms issue equity in a given year. 17Note that the assumption on the process for r means that agents do not learn about the interest rate t on b until date t+1. t+1 18We ensure that the largest possible operating cost shock z is low enough so that k is always positive. t+1 19We ensure the nonnegativity of V, without loss of generality, by assuming that there is a positive probability of a very valuable growth option that materializes in the future and has a (cid:133)xed value orthogonal to the current state of the (cid:133)rm. A nonnegative V implies that equity holders will never want to default voluntarily on their credit obligations. 13

(cid:133)nance they can obtain and no costs of issuing equity. The associated value of the (cid:133)rm is 1 z V (k b ;(cid:18) ;z ;r ) = max d +E V (k b ;(cid:18) ;z ;r )d(cid:8)(z) ; u t; t t t t t t u t+1; t+1 t+1 t+1 t+1 dt;kt+1;bt+1;(cid:18)N t+ E 1 W (cid:26) 1+r t+1 Z0 (cid:27) (18) where for simplicity we assume that d = n .20 t t The Modigliani-Miller Theorem applies, and (cid:133)rms(cid:146)investment decisions are independent of their (cid:133)nancing arrangements. Their investment choice satis(cid:133)es 1+E (r ) = f (k ): (19) t t+1 0 t+1 2.2.3 Simulation Strategy and Regression Speci(cid:133)cation We are interested in understanding how the exposure to (cid:135)oating interest rate debt a⁄ects the reaction of (cid:133)rms(cid:146)stock price and investment di⁄erently depending on their degree of (cid:133)nancial constraints. To do so, we run regressions on the simulated model economy by generating a random path of the interest rate, r , for 75 quarters, common to all (cid:133)rms, and t one random path of the operating cost shock for each of the 1;000 (cid:133)rms we simulate.21 A period (t to t + 1) is to be identi(cid:133)ed with one quarter, and includes typically two Federal Open Market Committee (FOMC) cycles that determine interest rates in the United States. To test our hypotheses, we run the following speci(cid:133)cation for stock returns, where subscript i refers to each (cid:133)rm: rS = (cid:11) +(cid:11) s +(cid:11) (cid:18) +(cid:11) s (cid:18) +(cid:11) 1( ) +(cid:11) 1( ) s i;t 0 1 t 2 i;t 3 t i;t 4 constrained i;t 5 constrained i;t t +(cid:11) 1( ) (cid:18) +(cid:11) 1( ) s (cid:18) 6 constrained i;t i;t 7 constrained i;t t i;t +(cid:13) controls +(cid:13) 1( ) controls +(cid:13) s controls 1 i;t 2 constrained i;t i;t 3 t i;t +(cid:13) ( ) s controls +" ; (20) 4 constrained i;t t i;t t whererS isthestockreturnof(cid:133)rmioverquarter , ands = r E [r ]istheratesurprise. i;t t t t (cid:0) t (cid:0) 1 t The indicator function 1( ) takes a value of 1 if the (cid:133)rm is (cid:133)nancially constrained, constrained t as described in Section 2.2.1, and a value of 0 if the (cid:133)rm is unconstrained, as described in Section 2.2.2. Our theory predicts that in regression (20) (cid:11) and (cid:11) should be negative, 3 7 i.e., contractionary rate surprises should reduce the market value of the (cid:133)rm and more so 20Notethatthereisnooptimaldistributionofdividendpaymentsof(cid:133)nanciallyunconstrained(cid:133)rmsacross time, as long as they can invest funds at the same interest rate as shareholders. 21Wediscardthe(cid:133)rst30periodsofthesimulationtoavoidthein(cid:135)uenceofinitialconditions. Weintroduce 500 (cid:133)nancially constrained (cid:133)rms and 500 unconstrained (cid:133)rms, consistent with our empirical analysis in the following sections that uses the median of our (cid:133)nancial constraints proxies as the cut-o⁄point to separate (cid:133)nanciallyconstrainedandunconstrained(cid:133)rms. Thisisalsoclosetothefractionof(cid:133)rmsreportedbyHadlock and Pierce (2010) as not (cid:133)nancially constrained (10%) and likely not (cid:133)nancially constrained (50%). 14

for (cid:133)nancially constrained (cid:133)rms.22 The vector controls includes (cid:133)rm leverage bt and (cid:133)rm i;t kt size n . The de(cid:133)nitions of all the variables used are standard in the investment and capital t structure literatures, and are described in detail in Appendix B. The corresponding equation for investment is i i;t = (cid:12) +(cid:12) (cid:1)r +(cid:12) (cid:18) +(cid:12) (cid:1)r (cid:18) +(cid:12) 1( ) +(cid:12) 1( ) (cid:1)r k 0 1 t 2 i;t 3 t i;t 4 constrained t 5 constrained t t i;t 1 (cid:0) +(cid:12) 1( ) (cid:18) +(cid:12) 1( ) (cid:1)r (cid:18) +(cid:13) Q +" ; (21) 6 constrained t t 7 constrained t t i;t 1 i;t t where we add Tobin(cid:146)s Q (Q ) as a determinant of investment and include it as a control i;t variable. We use interest rate changes (cid:1)r in our investment regression but interest rate t surprise changes s in our stock price regression, given that stock prices are forward looking, t an approach we also employ for our empirical analysis. Finally, we also study whether changes in interest rates a⁄ect the interest rate coverage ratio and the likelihood and costs of (cid:133)nancial distress. The de(cid:133)nitions of these variables can also be found in Appendix B. 2.2.4 Calibration In order to assess the quantitative properties of the model we calibrate it at a quarterly frequency. Some model parameters are set to obtain a reasonable match between the model simulated data moments and their real data counterparts, while others are set to common values in the literature. Table I summarizes our baseline parameter choices. Overall, most of our parameter choices are relatively conservative, to avoid concerns that any of the results might be driven by an overstatement of some of the factors that drive the (cid:135)oating rate channel. [TABLE I ABOUT HERE] We (cid:133)rst specify the production technology. We normalize the (cid:133)rm(cid:146)s productivity parameter A to 1, and, in line with most of the literature, set the capital share (cid:11) to 0.4.23 The idiosyncratic operating cost shock is an i.i.d. process that can take values z i;t 2 22In our empirical regressions using real data in Sections 4 and 5 we scale bank debt over total assets, as we think itis a betterproxy for a(cid:133)rm(cid:146)sbank debt usage incomparison other sources of(cid:133)nancing, including debt and equity, but also provide robustness checks at the end of Appendix F using bank debt over total debt. In our simulated model, total assets feature much more variation than bank debt does, which makes most of the variation in bank debt usage come from variation in total assets. For this reason, we scale bankdebtbytotaldebtinthemodelsimulatedbenchmarkregressions. Alternatively, onecouldintroducea capital adjustment cost to subdue the movements in total assets. This would complicate the model without additional insights. Having said this, we provide robustness checks in Table A15 in Appendix B using bank debt over total assets in our simulated model regressions. All our results go through, although statistical and economic signi(cid:133)cances are slightly weaker. 23Corrado, Hulten and Sichel (2009) estimate the income shares of labor and capital in the U.S. over the period 1993-2005 to be 60% and 40%, respectively. We normalize labor to one. 15

0;0:09;0:18;0:27;0:36 with equal probability. This stochastic process delivers a probaf g bility of negative internal funds n of 7.8%, which is in line with the empirical estimate of t the quarterly likelihood of a covenant violation. Nini, Smith and Su(cid:133)(2012) estimate that on average 6.9% of (cid:133)rms are in violation of a (cid:133)nancial covenant in any given quarter in the Compustat sample between 1997 and 2008. Debt maturity, which is driven by the repayment rate (cid:21), has a signi(cid:133)cant in(cid:135)uence on the strength of our channel. Debt in our model comprises both bank loans and bonds, both of which are relevant from a macroeconomic perspective. The Federal Reserve(cid:146)s Flow of Funds states that of the $7.5 trillion of outstanding non(cid:133)nancial corporate business debt in the United States, around $5 trillion are corporate bonds and $2.5 trillion are corporate loans.24 Estimates for bank loan maturities are in the order of 4-5 years.25 Estimates for average corporate bond maturities are signi(cid:133)cantly larger, ranging from 11 to 13 years.26 We choose (cid:21) = 0:035, which delivers an average maturity in our simulations of 7.1 years, below the weighted average empirical estimate using the data above of around 9 years. This is a conservative calibration given that our results would be stronger with larger average maturity due to a longer lasting e⁄ect of monetary policy shocks through existing (cid:135)oating rate debt. We select a cost of hedging of 0.5%, which is in line with the average swap spread for maturities between 5 and 30 years estimated by Jermann (2016) in the period 1998-2008.27 The borrowing constraint parameter b is set to be in line with (cid:133)rms(cid:146)average market leverage, calculated as total debt over total debt plus the market value of equity. Colla, Ippolito, and Li (2013) (cid:133)nd that average market leverage for Compustat leveraged (cid:133)rms is 0.25, the same value we obtain.28 The payout ratio (cid:26) for (cid:133)rms with positive earnings is set to be in line with the average payout ratio of U.S. corporations. Floyd, Li and Skinner (2015) estimate that the payout ratio, calculated as dividends plus share repurchases over income before extraordinary items 24Thesecalculationsignorenoncorporatebusinessdebt,forwhichmaturitydataisveryscarce. Wediscuss the aggregate importance of noncorporate debt in Section 6.1. 25Hollander and Verriest (2016) estimate an average maturity of 5 years for syndicated bank loans in the US between 2005 and 2008, and Gomes, Jermann and Schmid (2016) calibrate debt maturity on the basis of the observed maturity of commercial and industrial loans to be 5 years. Paligorova and Santos (2014) calculate an average maturity at origination of 4 years for corporate bank loans between 1990 and 2010. 26Gilchrist and Zakrajsek (2012) (cid:133)nd that the maturity at origination for public corporate bonds was on average 13 years, between 1971 and 2010. Kwan and Carleton (2011) (cid:133)nd that privately placed bonds have a shorter maturity of around 11 years on average, in a sample between 1985 and 1994. 27The U.S. dollar swap spread is the di⁄erence between the (cid:133)xed rate paid on an interest rate swap and the yield on a U.S. Treasury security of equivalent maturity. Jermann (2016) highlights that this spread has been around 0% following 2008, a feature he associates with limits to arbitrage. 28Amorecomprehensivemeasurethatincludesprivate(cid:133)rmsiscalculatedbyGomes,JermannandSchmid (2016)usingFlowofFundsdata, whoobtainaveragecorporatedebtratios(scaledbyassetsatcurrentcost) slightly above 50% between 2005 and 2009. 16

forallnon(cid:133)nancialUS(cid:133)rmsinCompustat, tobeonaverage73%over2000-2012. Ourpayout is on average equal to 46% of pro(cid:133)ts.29 A large dividend payout worsens (cid:133)rms(cid:146)(cid:133)nancial constraints by keeping their internal funds low. To dispel concerns that our (cid:133)nancially constrained (cid:133)rms are excessively constrained, we keep the payout ratio signi(cid:133)cantly lower than what we observe in the data. [TABLE I ABOUT HERE] Our measure of the cost of issuing equity, (cid:17), is meant to broadly capture costs of (cid:133)nancial distress for shareholders. In our model, it is associated with (cid:133)nancial covenant violations. Nini, Smith and Su(cid:133)(2012) study violations of (cid:133)nancial debt covenants, and show that they are associated with a decrease in market to book ratios of around 15% on average, and a loss of net worth of around 13% of total assets. Covenant violations often lead managers to deleverage by issuing equity (Nini, Smith and Su(cid:133) (2012)), which is re(cid:135)ected in our model, and our parametrization also builds on literature estimates of (cid:135)otation costs. Gomes (2001) estimates that (cid:135)otation costs are equal to around 3% of the issue amount plus a (cid:133)xed component. Other studies show that underwriting fees range between 3% and 8% of the issue amount and that an information asymmetry e⁄ect causes a decrease in stock value of more than 2% on average (Lee and Masulis (2009)). We set (cid:17) conservatively so that the average equity issuance costs incurred in a period in which n < 0 are in the order of 6% of t total assets, in line with the estimates described above. Our results are stronger with larger (cid:17), and hence this is a conservative estimate.30 Finally, we describe the stochastic process for the interest rate r , which is driven by t monetary policy. We estimate that the Federal Funds Target Rate has a quarterly autocorrelation of 0.96, a mean of 0.04, and a normalized standard deviation of 48.9%, using data from 1994 to 2008. In our simulations, r follows a Markov process in which the interest t rate can take values in r 1%;2%;3%;4%;5% , with transition probabilities that deliver t 2 f g a quarterly autocorrelation of 0.95, and a mean and standard deviation of 0.03 and 43.8%, respectively. We choose a smaller mean due to smaller interest rates during our sample period in our empirical analysis. 29Notethatourdividendruleestablishesthatthe(cid:133)rmpaysout(cid:26)=0:9ofinternalfunds,andthatinternal funds are equal to pro(cid:133)ts minus principal debt repayments that period. This is why the payout ratio based on pro(cid:133)ts is signi(cid:133)cantly lower than the value of (cid:26). 30Note that there is no bankruptcy in our model, so our costs of covenant violation events are also meant tocapturebankruptcycosts. Estimatesforthecostsofbankruptcyasashareoftotalassetsareintherange of 20% (Altman (1984), Bris, Welch and Zhu (2006)), to 36% (Alderson and Betker (1995)). 17

2.2.5 Results and Discussion Is the increase in the responsiveness of V and k to monetary policy when a (cid:133)rm t+1 increases its usage of (cid:135)oating rate debt stronger for (cid:133)nancially constrained (cid:133)rms? If so, what is the mechanism behind this e⁄ect? [TABLE II ABOUT HERE] Our results for stock returns are displayed in Panel A of Table II. An increase in the share of (cid:135)oating-rate debt as a share of total debt increases the responsiveness to monetary policy of the stock returns for both constrained and unconstrained (cid:133)rms. In the case of unconstrained(cid:133)rms, thee⁄ectoccursbecauseatighteningofmonetarypolicyinthepresence of(cid:135)oatingratedebtincreasestheshareofcash(cid:135)owsallocatedtobondholders, attheexpense of shareholders. That channel is also present for constrained (cid:133)rms, but, in addition, they su⁄er two other consequences. First, a monetary policy tightening when a (cid:133)rm is exposed to interest rate variations through (cid:135)oating rate debt increases interest expense, decreases cash (cid:135)ows, and increases the likelihood of su⁄ering bankruptcy costs (which in our model correspond to having a negative liquidity position that needs to be covered by issuing costly equity). Second, a (cid:133)nancially constrained (cid:133)rm that su⁄ers an increase in its interest expense throughits(cid:135)oatingratedebtwillsu⁄eralossinitsnetworthanditsinvestmentcapacitywill decrease, su⁄ering from lower future pro(cid:133)ts and dividends. The economic magnitudes are large. While increasing (cid:135)oating-rate debt usage from 0% to 100% of total debt is associated with an additional e⁄ect of a 1 bp surprise tightening of monetary policy on stock returns of unconstrained (cid:133)rms of around 21 to 23 bp, the same increase in (cid:135)oating-rate debt usage increases the responsiveness of (cid:133)nancially constrained (cid:133)rms by around 28 (21.58+7.25) bp, leading to a roughly one-thirdlarger additional e⁄ect that is strongly statistically signi(cid:133)cant. For reference, an unconstrained (cid:133)rm with no (cid:135)oating-rate debt su⁄ers around a 20 bp stock return fall on average following a 1 bp policy rate increase.31 We also repeat the simulated regression exercise using a lower distress cost in column 3, which we will discuss in more detail at the end. Our results for investment are displayed in Panel B of Table II. An increase in the share of (cid:135)oating-rate debt as a share of total debt increases the responsiveness to monetary policy of the investment of constrained (cid:133)rms but has no consistent e⁄ect for unconstrained (cid:133)rms. In the case of unconstrained (cid:133)rms, investment depends only on the interest rate, as the Modigliani-Miller conditions apply to these (cid:133)rms and their capital structure is irrelevant for investment decisions. Constrained (cid:133)rms(cid:146)investment, however, is determined by their liquid 31Notice that the coe¢ cient on Surprise in column 2, which displays the regression with controls, does t not have a meaningful interpretation because we interact all controls with surprise. 18

resources, whichinturnarea⁄ectedbytheirinterestexpenses. Amonetarypolicytightening when a (cid:133)rm is exposed to interest rate variations decreases cash (cid:135)ows and decreases the resources available for investment. For (cid:133)nancially constrained (cid:133)rms, increasing (cid:135)oatingrate debt usage from 0% to 100% of total debt is associated with an additional decrease in investment caused by a 1 bp tightening of monetary policy of around 0.1 to 0.2 bp at a four-quarter horizon, and 0.4 bp at a six-quarter horizon. [TABLE III ABOUT HERE] Table III studies how changes in interest rates interact with the presence of (cid:135)oating rate debt and (cid:133)nancial constraints to a⁄ect the interest rate coverage ratio and the likelihood of (cid:133)nancial distress. Increasing (cid:135)oating-rate debt usage from 0% to 100% of total debt is associated with an increase in the likelihood of an episode of (cid:133)nancial distress caused by a 1 percentage point tightening of monetary policy of around 2.4 percentage points (column (1)), and with an increase in costs of (cid:133)nancial distress caused by a 1 percentage point tightening of monetary policy of around 1.6% of the market value of assets (column (2)). The importance of the relationship between interest rates, (cid:135)oating rate debt, and (cid:133)nancial constraints suggests that this link may amplify the relationship between (cid:135)oating rate debt usage and policy sensitivity of stock prices. We (cid:133)nd evidence consistent with this conjecture. Inparticular,whenwerunasimulationinwhichwedecreaseequityissuacecostsby33%,and rerunourstockreturnregressions(column(3)ofPanelAofTableII), we(cid:133)ndthatincreasing (cid:135)oating-rate debt usage from 0% to 100% of total debt is associated with an additional e⁄ect of a 1 bp surprise tightening of monetary policy on stock returns of constrained (cid:133)rms of around 20 bp (15.99+3.54). This e⁄ect is lower than the 28 bp e⁄ect in column 1 that uses our benchmark distress costs. Further evidence consistent with the speci(cid:133)c mechanism we suggest in our (cid:135)oating rate channel is displayed in column (3) of Table III, which shows that the presence of (cid:135)oating rate debt has a strong in(cid:135)uence on the sensitivity of the interest rate coverage ratio to monetary policy shocks. 3 Data Description and Summary Statistics Our theoretical results imply that the presence of (cid:135)oating rate debt has signi(cid:133)cant in(cid:135)uence on the transmission of monetary policy to (cid:133)rm stock prices and balance sheet variables, especially for (cid:133)nancially constrained (cid:133)rms. Since a majority of bank debt is (cid:135)oating rate and most of the nonbank debt is (cid:133)xed rate, our results suggest that bank debt may play a special role in monetary policy transmission due to its (cid:135)oating rate nature. Testing these implications in the data requires not only information on (cid:133)rms(cid:146)investment, liquidity posi- 19

tion, and stock prices, which are readily available from sources that have been widely used before, but also information on how much bank debt and (cid:135)oating rate debt a (cid:133)rm uses and on (cid:133)rms(cid:146)hedging behavior. We address this challenge by using a new dataset on debt structure (Capital IQ) and by creating a new dataset on the hedging behavior of publicly listed companies. This section describes these e⁄orts in detail. 3.1 Firm-level Data The sample for our main analysis consists of U.S. (cid:133)rms covered by CRSP, Compustat, and Capital IQ (CIQ), excluding utilities (SIC codes 4900(cid:150)4949) and (cid:133)nancials (SIC codes 6000(cid:150)6999). WhileCRSPandCompustatarewellknownandwidelyused, theCIQdatabase is relatively new. CIQ compiles detailed information on capital and debt structure from the footnotes of 10-K Securities and Exchange Commission (SEC) (cid:133)lings. In particular, from CIQ we obtain data on the amount of bank debt (cid:133)rms have in their liabilities. Our main measure of bank debt usage, BankDebt=At, is de(cid:133)ned as total bank debt, which we calculate as drawn credit lines (CL) plus term loans (TL), divided by the total value of book assets (Compustat item AT). For robustness, we also employ two additional measures of bank debt usage: CL plus TL divided by total debt, and TL plus CL plus undrawn credit lines, divided by the total value of book assets. We focus annual CIQ (cid:133)les because they are more densely populated. We focus on the period from 2004 to 2008 because of the lack of wide coverage of bank debt data in CIQ before 2003 and because the federal funds target rate hit the zero lower bound in 2008, after which the quantitative easing program of the Federal Reserve replaced the federal funds target rate as the main monetary policy tool.32 In an extension in Section 6.3, in which we study the quantitative easing period separately, we extend our sample to alsocover2008to2011. Weremoveobservationswithnegativerevenues, missinginformation on total assets, or a value of total assets under $10 million. For our stock price analysis, we also discard penny stocks, de(cid:133)ned as those with a price of less than $5, as in Amihud (2002). Moreover, we follow the convention in the literature (De Bondt and Thaler (1990), Kashyap, Lamont, and Stein (1994) or Polk and Sapienza (2009)) and focus on (cid:133)rms whose (cid:133)scal year ends in December so that balance sheet information about di⁄erent (cid:133)rms is available to investors at the same time. Various degrees of staleness of balancesheet informationacross di⁄erent (cid:133)rms might a⁄ect ourresults, especiallybecause some (cid:133)rm characteristics might be seasonal, not only over the calendar year, but also over the (cid:133)scal year.33 Our main results remain similar when we include all (cid:133)rms. We use two-day 32Because data availability limits our sample, in Appendix C we make sure that the reaction of stock prices to monetary policy shocks in our sample is similar to the e⁄ect of monetary policy on stock prices before 2003 and in the CRSP universe. 33For example, Oyer (1998) (cid:133)nds that in addition to varying with the calendar business cycle, manufac- 20

stock returns for each (cid:133)rm on each of the FOMC meeting dates. For our analysis of balance sheet variables, we use quarterly data.34 After the above (cid:133)lters, the sample for our analysis of stock returns contains 9,746 (cid:133)rm-year observations comprising 2,368 unique (cid:133)rms, and the sample for our analysis of balance sheet variables contains 45,694 (cid:133)rm-quarter observations comprising 3,146 unique (cid:133)rms. Exact variable de- (cid:133)nitions are given in Table A1 in the Appendix. Following common practice in the empirical (cid:133)nance literature, all variables are winsorized at the 1 percent level in both tails of the distribution to prevent extreme values from overin(cid:135)uencing our regressions.35 Throughout the analysis, we use demeaned (cid:133)rm-level variables in regressions with interaction terms to facilitate the interpretation of the coe¢ cient estimates of the policy action as the reaction of the average (cid:133)rm. Table IV provides key statistics for the balance sheet variables we employ in our study. Across the entire sample (column 1), bank debt represents on average 7.22 percent of the book value of assets and 37.51 percent of total debt. For the subset of (cid:133)rms with some bank debt (columns 3 and 4), the above ratios rise to 10.33 percent and 58.89 percent for nonhedgersand15.52percentand50.35percentforhedgers. Inbothsamples, approximately half of bank borrowing is in the form of drawn credit lines and the other half in the form of term loans. [TABLE IV ABOUT HERE] A comparison between leveraged (cid:133)rms without bank debt (column 2) and leveraged (cid:133)rms with bank debt (columns 3 and 4) reveals that (cid:133)rms with bank debt do not seem to display characteristicsthatsuggestthattheyareclearlymoresensitivetomonetarypolicy,compared to leveraged (cid:133)rms without bank debt (i.e., those that use other sources of debt). They have similar size, age, and likelihood of being rated. Nevertheless, within bank debt users, the unhedged ones (column 4) have slightly lower size, age, and probability of being rated compared to leveraged (cid:133)rms without bank debt, although they have similar pro(cid:133)tability and riskiness, as captured by lower average CAPM beta and cash (cid:135)ow volatility. We will discuss the di⁄erences between hedgers and nonhedgers in more detail in the next subsection where we discuss the interest rate hedging data. turing (cid:133)rms(cid:146)sales are higher at the end of the (cid:133)scal year. 34Asaresult,wehavefourobservationsperyear,whichishalfasmuchastheeventstudywithstockprices allowsbecausethereisusuallymorethanoneFOMCannouncementperquarter. ImposingaDecember(cid:133)scal year-end would reduce the number of observations even further. Moreover, the availability of balance sheet informationtoinvestorsisnotasimportantforrealvariablesasitisforstockprices,becausetheinformation is internally available to managers. Therefore, we also include (cid:133)rms whose (cid:133)scal year ends in March, June, September, or December, so that our variables match the measure of monetary policy that we use for that analysis, which is a quarterly aggregate of monetary policy changes. 35See, for example, Fama and French (1992) and Su(cid:133)(2009). 21

We also obtain information on the percentage of (cid:133)rms(cid:146)debt that is (cid:135)oating or (cid:133)xed rate from CIQ. Unlike the bank debt variable, which is measured with precision (Colla, Ippolito and Li (2013)), the (cid:135)oating rate debt variable is measured with some error and we only use it as an approximation. It is manually collected by CIQ from the footnotes in 10-K (cid:133)lings, which could lead to errors that are hard to detect systematically. Indeed, the sum of (cid:135)oating plus (cid:133)xed rate debt often does not add up to total debt. The 5th percentile of the ratio of (cid:135)oating plus (cid:133)xed rate debt to total debt is 0.83, and the 95th percentile is 1.04. This caveat notwithstanding, our (cid:135)oating rate debt measure is useful to illustrate a key distinctionof bankvs. non-bankdebt. Acomparisonbetweencolumns2withcolumns 3and 4 also reveals that bank debt is more likely to feature (cid:135)oating interest rates than non-bank debt. Floating rate debt represents 12.75 percent of the value of the assets of bank debt users, compared with 1.59 percent for the (cid:133)rms with only non-bank debt. Figure 2 explores in more detail the relation between bank debt and (cid:135)oating rate debt. On the horizontal axis, (cid:133)rm-year observations are grouped into percentile bins of bank debt as a percentage of total debt. On the vertical axis, we report (cid:135)oating rate debt as a percentage of total debt. The (cid:133)gure shows a striking correlation between bank debt and (cid:135)oating rate debt. For those (cid:133)rms forwhichtheentirestockofdebtconsistsofbankdebt, about76percentofitis(cid:135)oatingrate. Forthose(cid:133)rmswhosedebtisentirelyfromnon-banksources, however, onlyaround9percent of debt is (cid:135)oating rate. These (cid:133)gures are consistent with Faulkender (2005), according to which 89.9 percent of bank loans are issued with a (cid:135)oating rate, compared to only 7% of (cid:135)oating rate bonds. Aslan and Kumar (2012) report that all of the syndicated bank loans in their comprehensive sample drawn from the Loan Pricing Corporation(cid:146)s (LPC) Dealscan database from 1996 through 2007 have (cid:135)oating interest rates. Given that the CIQ bank debt variable is measured with precision and that the evidence above suggests that a vast majority of bank debt features (cid:135)oating interest rates, while most non-bank debt is issued with (cid:133)xed rates, we focus in our analysis on bank debt as a proxy for the amount of (cid:135)oating rate debt in (cid:133)rms(cid:146)balance sheet. For robustness, we also make sure that our results are similar when we use CIQ(cid:146)s (cid:135)oating rate debt measure. 3.2 Interest Rate Hedging Data We collect data on interest rate hedging activities of U.S. (cid:133)rms using a text-search algorithm that scans 10-K corporate (cid:133)lings with the SEC. Disclosure of derivative hedging is mandatory under the 1998 Financial Reporting Release (FRR) No. 48 of the SEC and the 2001 SFAS No. 133. We do a detailed search of multiple phrases consistent with the use of interest rate derivatives (such as "hedge against interest rate," "hedge interest rate," "interest rate swap"), and then, for those (cid:133)lings for which we have a preliminary reading consistentwithinterestratehedging, wecheckforfalsepositivesbycontrollingfornegations, 22

such as "not use any interest-rate swaps," "not use interest-rate swaps," "not currently use any interest-rate swaps," "not hedge interest rate," "not use derivative (cid:133)nancial instruments as a hedge against interest rate," "termination of interest rate swap," "(cid:133)xed to (cid:135)oating interest rate swap," or "do not currently use interest rate swap." Appendix D provides examples of the types of discussions on hedging activities that we (cid:133)nd in the 10-K (cid:133)les. Table IV reports the summary statistics for our hedging dummy. Overall, about 35 percent of (cid:133)rm-years in our sample feature the usage of (cid:135)oating to (cid:133)xed rate hedging. For bank debt users (columns 3 and 4) this number is closer to 50 percent whereas for leveraged (cid:133)rms without bank debt (column 2) this number is only 26 percent. In other words, (cid:133)rms that use bank debt are about twice as likely to hedge than those that only use other types of debt. The binary nature of our hedging variable has two implications for our analysis. First, the di⁄erence between bank debt users and other (cid:133)rms is likely an understatement of the relative e⁄ect of hedging. Faulkender(cid:146)s (2005) (cid:133)nding that (cid:135)oating-to-(cid:133)xed rate hedging a⁄ects only 0.5% of the bonds suggests that the net e⁄ect of hedging for (cid:133)rms without bank debt is negligible. This is a reason that makes bank debt a more suitable variable for the study of the (cid:135)oating rate channel compared to CIQ(cid:146)s (cid:135)oating rate debt measure (in addition to the measurement error mentioned before): in our regressions with (cid:135)oating rate debt (instead of bank debt), two hedgers with the same amount of (cid:135)oating rate debt will be treated the same even if one of them has mostly bank debt and the other has mostly non-bank debt although the e⁄ect of hedging on the latter group is negligible. This leads to a noisier estimation of the (cid:135)oating rate channel when using (cid:135)oating rate debt. Second, since not necessarily all bank debt will be hedged at once, our hedging dummy overvalues the protective e⁄ect of hedging and hence the associated coe¢ cient of interest (Hedging*(Bank Debt /At)) is likely underestimated. Therefore, our results should be considered a lower bound for the e⁄ect of hedging and the (cid:135)oating rate channel. We also report statistics about (Hedging*(Bank Debt /At)) and (Hedging*(Floating Rate Debt /At)) in Table IV as the coe¢ cient of this variable is of primary interest to us in the following sections. It is instructive to compare three groups of (cid:133)rms(cid:150)leveraged without bank debt (column 2 of Table IV), bank debt hedgers (column 3), and bank debt non-hedgers (column 4). Bank debt hedgers are signi(cid:133)cantly more leveraged, more pro(cid:133)table, older, less risky, as measured by cash (cid:135)ow volatility and CAPM beta, and less (cid:133)nancially constrained than the other two groups, although their size, on average, is not signi(cid:133)cantly di⁄erent. Their decision to hedge might be precisely because they have large amounts of bank debt, and they might hold less cash because interest rate hedging reduces the precautionary demand for liquidity. They also have signi(cid:133)cantly lower growth opportunities (lower inventory, sales and PPE, and low market-to-book), but are already very pro(cid:133)table with their assets already in place. 23

The leveraged (cid:133)rms without bank debt have on average higher market-to-book asset ratios relative to bank debt users, and also signi(cid:133)cantly more cash holdings. Finally, it is possible that (cid:133)rms(cid:146)hedging choice is associated with bank characteristics. For example, certain banks are more likely to force their borrowers to hedge their interest rate risk. In order to deal with this concern, we calculate the exposure of a large subset of our(cid:133)rms totheirdi⁄erent lenders usingLPCDealscan, whichreports the lendingallocations for banks in a syndicate. We construct several (cid:133)rm-year variables that capture the weighted average of a series of characteristics of the banks a (cid:133)rm is borrowing from. We obtain the bank balance sheet data from the quarterly FFIEC Call Reports, which all regulated U.S. commercial banks are required to (cid:133)le. The results are reported in Table IV. We (cid:133)nd that the banks lending to (cid:133)rms that hedge are very similar to those lending to (cid:133)rms that do not hedge, when studying bank characteristics such as size, capital ratio, deposit ratio, or liquidity ratio. 3.3 Monetary Policy Data Because the equity market will already have responded to anticipated policy actions, we follow the approach of Kuttner (2001) and Bernanke and Kuttner (2005) to dissect the monetary policy actions into the unexpected (surprise) component and the anticipated (expected) component on an FOMC meeting or an announced change in the federal funds target rate. The identi(cid:133)cation of the surprise element in the target rate change relies on the price of the current month 30-day federal funds futures contracts, a price that encompasses marketexpectationsofthee⁄ectivefederalfundsrate. Wefollowthismethodbecausefederal fundsfuturesoutperformtargetrateforecastsbasedonother(cid:133)nancialmarketinstrumentsor based on alternative methods, such as sophisticated time series speci(cid:133)cations and monetary policy rules.36 Another advantage of looking at one-day changes in near-dated federal funds futures is that federal funds futures do not exhibit predictable time-varying risk premia (and forecast errors) over daily frequencies.37 We obtain the data for the decomposition of the federal funds target rate changes from Kenneth Kuttner(cid:146)s webpage, which provides data covering up to June 2008 because "since late 2008 the funds rate has been very close to zero, and the FOMC no longer reports a point target for the rate."38 Appendix E summarizes the process which generates this decomposition. 36See Evans (1998) and G(cid:252)rkaynak, Sack and Swanson (2005) for details. 37See, for example, Piazzesi and Swanson (2008). 38http://econ.williams.edu/faculty-pages/research/ 24

4 The Floating Rate Channel: Evidence from Stock Prices Our evidence in support of the (cid:135)oating rate channel is based on stock prices and balance sheet data. Both types of data have advantages and disadvantages, and studying both provides a more robust test of our proposed channel. Stock prices, to the extent that they e¢ ciently re(cid:135)ect (cid:133)rms(cid:146)underlying fundamentals, can provide a more precise identi(cid:133)cation of a speci(cid:133)c transmission mechanism because they react rapidly to policy changes and can be measured immediately after the policy shock occurs, compared to balance sheet variables, suchas(cid:133)xedinvestmentorinventoryinvestment,whichmightreactslowlyduetoadjustment costs and are measured a long time after the shock has occurred. Their slow reaction might prevent the identi(cid:133)cation of the full policy impact if the e⁄ects occur with a signi(cid:133)cant lag. And because they are not available immediately, other shocks and mechanisms might come into play, making identi(cid:133)cation di¢ cult. Additionally, stock prices provide a welfarerelevant measure of the e⁄ects, by capturing their present value. Moreover, understanding how and why stock prices react to monetary policy has been an important question since, at least, Tobin (1969) and Modigliani (1971) due to its important implications for consumption and investment. For these reasons, we follow Gorodnichenko and Weber (2014), English, Van den Heuvel, and Zakrajsek (2014) and Chodorow-Reich (2014), who also study stock prices to shed light on the transmission of monetary policy. Balance sheet variables, on the other hand, allow us to test more speci(cid:133)c implications of our proposed transmission channel, enabling us to more convincingly distinguish our mechanism from alternative ones. We start by focusing on stock prices, and proceed in three steps. We (cid:133)rst show that bank debt usage makes (cid:133)rms signi(cid:133)cantly more responsive to monetary policy, and then show that this additional responsiveness is concentrated in the (cid:133)rms that do not hedge their interest risk, and especially so in the (cid:133)nancially constrained ones. 4.1 The E⁄ect of Bank Debt Usage As our (cid:133)rst step, we analyze whether a (cid:133)rm i(cid:146)s stock price change Ret over the day i;t t in which a monetary policy shock Surprise occurs and the day after depends on the t importance of bank debt as a source of (cid:133)nancing, (BankDebt=At) .39 For this purpose, i;t 1 (cid:0) we use the following regression, 39We measure returns over a 2-day window for two reasons. First, as Figure 3 shows, it takes more than a day for the full e⁄ect of bank debt usage to be incorporated in stock prices. Moreover, the FOMC blackout period, during which Federal Reserve employees are not allowed to comment on current monetary policy, endsonthedayfollowingtheFOMCannouncement,potentiallymakingtheinferencebasedonawiderthan two-day event window not as reliable. 25

Ret = (cid:12) +(cid:12) Surprise +(cid:12) (BankDebt=At) i;t 0 1 t 2 i;t 1 (cid:0) +(cid:12) Surprise (BankDebt=At) 3 t (cid:3) i;t 1 (cid:0) +(cid:13)Controls +(cid:21)Surprise Controls +" ; (22) i;t 1 t i;t 1 i;t (cid:0) (cid:3) (cid:0) whereControls isavectorof(cid:133)rmcharacteristics. Inthisregression, Ret andSurprise i;t 1 i;t t (cid:0) refertostockreturnsandmonetarypolicysurpriseonthedayoftheFOMCannouncementat date t. Since (BankDebt=At) is available yearly and since we want (BankDebt=At) i;t 1 i;t 1 (cid:0) (cid:0) and Controls to be available to investors simultaneously, we use the last (cid:133)scal yeari;t 1 (cid:0) end data available before the date of the monetary policy event in order to capture the information available to investors at the time of the monetary policy announcement, in line with most of the cross-sectional asset pricing literature dating back at least to Fama and French (1992). With a slight abuse of notation, in the case of (cid:133)rm characteristics, t 1 refers (cid:0) to the most recent (cid:133)scal year-end prior to the federal funds rate target announcement date, meaning that we use December accounting variables for all the FOMC meeting dates and the corresponding stock returns in the following year. In the case of Surprise and Ret , t i;t t refers to the monetary policy announcement date, of which there are eight scheduled ones in any given year.40 Our (cid:133)rm-level controls include book leverage, (cid:133)rm size, the market-to-book ratio, profitability, and interest rate sensitivity, all of which are described in Table A1 in detail.41 We control for book leverage because bank debt users are more likely to be highly leveraged, and as such might be more sensitive to monetary policy. We control for (cid:133)rm size and marketto-book ratios because these variables are well-known risk factors for asset prices since the seminal paper of Fama and French (1992), and they can also a⁄ect the reaction of stock prices to policy surprises because they are related to (cid:133)nancial constraints and investment opportunities.42 Pro(cid:133)tability is included because, as shown in Fama and French (1995), the market-to-book ratio is associated with persistent di⁄erences in pro(cid:133)tability and (cid:133)rms with bank debt tend to be more pro(cid:133)table, as shown in Table IV. In addition, Ehrmann and Fratzscher (2004) report strong evidence that (cid:133)rms with low pro(cid:133)tability are more responsive to monetary policy when pro(cid:133)tability is measured as cash (cid:135)ow divided by income. Finally, we control for the interest rate sensitivity of operating pro(cid:133)ts because it might in(cid:135)u- 40There are also a couple of unscheduled meetings during this time period. Following Bernanke and Kuttner (2005), we use in(cid:135)uence statistics to eliminate meetings that constitute outliers, as described in Appendix C. 41Wedonotcontrolformarketleveragebecause, asshowninOzdagli(2012), thevalueofmarketleverage can be pinned down using book leverage and the market-to-book ratio, leading to collinearity. 42We also add CAPM betas, calculated as in Fama and French (1992), as an additional control in some speci(cid:133)cations. See Appendix F for details. 26

ence the propensity to borrow from banks. This would generate a correlation between bank debt usage and the reaction of stock prices to monetary policy even if there were no causal relationship between these variables.43 The results of this regression under various alternative speci(cid:133)cations are presented in Table A4 and discussed in detail in Appendix F. The main takeaway is that bank debt usage increases the responsiveness of (cid:133)rms(cid:146)stock prices to monetary policy signi(cid:133)cantly. Focusing on column 1, we observe that a one standard deviation (0:114) increase in our bank debt usage measure causes the stock price to increase 1:6 (= 14 0:114) percentage points more (cid:0) (cid:3) in response to a 1 percentage point surprise decrease in the federal funds rate. To put this e⁄ect in perspective, the same surprise decrease in the federal funds rate causes the stock price of the (cid:133)rm with the average amount of bank debt over assets (7:22%) to increase about 4:97 percent on average. This result is very robust, and survives various speci(cid:133)cations that control for potential omitted variables. In particular, we eliminate the possibility that the additional responsiveness occurs because (cid:133)rms that use bank debt are potentially more highlyleveraged,moreseverely(cid:133)nanciallyconstrained,ormorereliantonshort-termdebt,by controlling for the relevant (cid:133)rmcharacteristics. We also use (cid:133)rmand industry characteristics that predict access to public debt markets as instruments and show that our results remain similar. We also replace bank debt usage, (BankDebt=At), with (BankDebt=Debt); to deal with the possibility that using book leverage as a separate control variable is not enough to argue that the e⁄ect of bank debt goes beyond the e⁄ect of other types of debt. Overall, we conclude that bank debt usage is important for the responsiveness of a (cid:133)rm to monetary policy and this importance cannot be attributed to other (cid:133)rm characteristics. 4.2 The Floating Rate Channel of Bank Debt The results obtained in Section 4.1 suggest that bank debt is special for the transmission of monetary policy to stock prices. In this section, we present evidence that suggests that a signi(cid:133)cant part of the e⁄ect is driven by the (cid:135)oating rate nature of bank debt, a transmission mechanismthat we introducedinour theoretical motivationinSection2 andcall the (cid:135)oating rate channel. The (cid:135)oating rate nature of most bank debt suggests that monetary policy actions should be re(cid:135)ected mechanically in the interest expense associated with existing bank loans because 43Our particular concern is that (cid:133)rms that use bank debt are special in that they are on average riskier and more interest rate sensitive, which would suggest that we may overestimate the direct impact of bank debt. While controlling for the interest rate sensitivity of operating pro(cid:133)ts of (cid:133)rms, as we do later, should addresstheseconcerns,welookdeeperintotherelationshipbetweena(cid:133)rm(cid:146)sriskinessanditsbank(cid:133)nancing behavior in Table A2. Contrary to our concerns, columns 1 and 2 show that bank debt usage is weakly negatively associated with cash (cid:135)ow volatility (a statistically insigni(cid:133)cant relationship) and that there is no statistically signi(cid:133)cant relationship with the interest rate sensitivity of operating pro(cid:133)ts. 27

these actions induce changes in the reference rates used in the (cid:135)oating rate agreements.44 Following this logic, our empirical strategy provides evidence for the (cid:135)oating rate channel by exploiting the variation across (cid:133)rms in their (cid:135)oating-to-(cid:133)xed interest rate hedging of their bank debt or (cid:135)oating rate debt. If the (cid:135)oating rate channel is quantitatively relevant, we shouldobservethatthee⁄ectofbankdebtusageonthesensitivityofstockpricestomonetary policy should diminish signi(cid:133)cantly for (cid:133)rms that engage in (cid:135)oating-to-(cid:133)xed interest rate hedging. We restrict our sample to those (cid:133)rms that have variable rate debt outstanding in excess of 1% of total assets, to isolate those (cid:133)rms that might have an incentive to engage in interest rate risk hedging as insurance against (cid:135)uctuations in the interest payments of the existing debt, rather than as a speculative investment opportunity. We divide this sample into those (cid:133)rms that hedge interest rate risk and those that do not, and we test the prediction that (cid:133)rmsthatusebankdebtor(cid:135)oatingratedebtandthathedgeagainstinterestrateriskshould be, all else equal, less responsive to monetary policy shocks than those that do not hedge, by running speci(cid:133)cation (22) separately on each subsample. The results of our main tests, provided in panel A of Table V, are consistent with our predictions.45 In columns 1 and 2, we test whether hedging a⁄ects the impact of bank debt usage on the sensitivity to monetary policy. We (cid:133)nd that while for the subsample of hedgers bank debt usage does not a⁄ect the sensitivity of stock prices to monetary policy, those that do not hedge are signi(cid:133)cantly more responsive to surprise changes in the federal funds rate. In particular, column 1 shows that the e⁄ect of bank debt usage becomes about twice as signi(cid:133)cant for the subsample of non-hedgers in comparison to our results for the full sample in Section 4.1, whereas the e⁄ect for the subsample of hedgers in column 2 becomes both economically and statistically insigni(cid:133)cant.46 Columns 3 and 4 show that this result is robust to the inclusion of a full set of (cid:133)rm level controls, both interacted with surprise and uninteracted, the introduction of (cid:133)rm (cid:133)xed e⁄ects, and clustering errors at the date-industry level. Finally, we interact all regressors in regression (22) with the hedging dummy in order 44The period starting in the fall of 2008 in which the Federal Funds Target Rate reached the zero lower bound was one in which the (cid:135)oating rate channel of monetary policy was unlikely to be operative. Figure 2 clearly shows that both the LIBOR rate and the prime rate have been very stable during this period. We pursuethisquestioninSection6.3andshowthatthe(cid:135)oatingratechannelwasmuteduringthistimeperiod. 45In these regressions, we are using the control variables in our main speci(cid:133)cation from column 3 of Table A4, which are the ones commonly used in the asset pricing literature, and include interest rate sensitivity for completeness. 46This does not necessarily mean that for non-hedgers bank debt does not play any role for the transmission of monetary policy. It could be the case that multiple transmission channels exist that operate in opposite directions and cancel each other out on average for this particular group of (cid:133)rms. An example of a transmission mechanism that would make bank debt using (cid:133)rms less responsive to monetary policy is one in which bank-(cid:133)rm relationships enable (cid:133)rms to bene(cid:133)t from some degree of insurance against changes in credit availability (Puri, Rocholl and Ste⁄en (2013)). 28

to assess statistical signi(cid:133)cance of the di⁄erence between hedgers and non-hedgers using the following regression Ret = (cid:12) +(cid:12) Surprise +(cid:12) Surprise (BankDebt=At) t 0 1 t 2 t (cid:3) t 1 (cid:0) +(cid:12) Surprise (BankDebt=At) Hedge 3 t (cid:3) t 1 (cid:3) t (cid:0) +(cid:21)Surprise Controls Hedge t t 1 t (cid:3) (cid:0) (cid:3) +Uninteracted terms and second order interactions+" ; (23) t in which we drop the reference to the (cid:133)rm-level subindex i for ease of exposition. We (cid:133)nd that the di⁄erence between hedgers (Hedge = 1) and non-hedgers (Hedge = 0), captured t t by (cid:12) ; is statistically signi(cid:133)cant. 3 Since we argue that the (cid:135)oating rate channel works via (cid:135)oating rate nature of bank debt, we test the robustness of our results by directly looking at (cid:135)oating rate debt. Accordingly, we repeat our exercise by replacing bank debt with (cid:135)oating rate debt in columns 5-8, and obtain similar qualitative results: usage of (cid:135)oating rate debt increases the responsiveness of stock prices to monetary policy only for those (cid:133)rms that do not hedge interest rate risk. We also note that the results are slightly weaker than the ones obtained with bank debt usage. As discussed in Section 3.2, this relative weakness is expected because non-bank (cid:135)oating rate debt is less likely to be hedged (Faulkender (2005)) and our hedging variable is a dummy variable that captures whether the (cid:133)rm hedges any of its (cid:135)oating rate debt rather than the total amount of (cid:135)oating rate debt that has been hedged. In our regressions, two hedgers with the same amount of (cid:135)oating rate debt will be treated the same even if one of them has mostly bank debt and the other has mostly non-bank debt. In addition, there is a moderate degree of mismeasurement in our (cid:135)oating rate variable. Both factors lead to a potential underestimation of the (cid:135)oating rate channel when using (cid:135)oating rate debt. [TABLE V ABOUT HERE] Finally,panelBofTableVpresentsthecoe¢ cientsofSurprise Controls fordi⁄erent t t 1 (cid:3) (cid:0) groups. None of these coe¢ cients di⁄er between hedgers and non-hedgers as much as the coe¢ cients of Surprise (BankDebt=At) or Surprise (FloatingRateDebt=At) do. t (cid:3) t (cid:0) 1 t (cid:3) t (cid:0) 1 This is also con(cid:133)rmed by the large p-value of the test that the di⁄erences of coe¢ cients for these variables between hedgers and non-hedgers - i.e., (cid:21) in equation (23) - are jointly zero. This result con(cid:133)rms that the (cid:135)oating rate nature of bank debt is the important channel to focus on. As an additional robustness check, we note that our sample period included a large number of rate increases, in 20 out of 43 meetings. This may create the concern that our 29

results are speci(cid:133)c to rate increases, which would put the external validity of our results into question. Therefore, we repeat the exercise in Table V after discarding those FOMC statements with positive rate changes from the sample. Our results, presented in Table A5, remain very similar quantitatively, alleviating concerns regarding external validity. Asaplaceboexperiment,weusethesamespeci(cid:133)cationbutreplacethedependentvariable with the last two-day returns before the FOMC. Because of the blackout period leading to an FOMC announcement and the resulting few number, if any, of FOMC related news prior to an announcement, this would be a suitable pseudo-control sample where we expect no signi(cid:133)cant di⁄erence in policy sensitivity caused by bank debt usage. This expectation is con(cid:133)rmed in Table A6, according to which the coe¢ cient of Surprise (BankDebt=At) t (cid:3) t 1 (cid:0) is indistinguishable between hedgers and nonhedgers and, if anything, it goes in the opposite direction for Surprise (FloatingRateDebt=At) . t (cid:3) t 1 (cid:0) 4.3 Instrumental Variables Regression In this section, we address the concern that the previous regression estimates might be biased due to the potential endogeneity of (cid:133)rms(cid:146)hedging decisions. Endogeneity concerns have to do mostly with omitted variables bias, rather than reverse causality, as it is unlikely thatstockreturnstodaycana⁄ectpasthedgingdecisions. Inprinciple,omittedvariablesbias in our context could be either positive or negative. First, (cid:133)rms whose operating earnings are more interest rate sensitive (because the demand for their goods or services is more interest rate sensitive, for example) might be more reluctant to expose themselves to interest rate risk through (cid:135)oating rate liabilities, and decide to hedge more. Through this e⁄ect, the coe¢ cient of Surprise (BankDebt=At) should be more negative for hedgers, leading to t (cid:3) t 1 (cid:0) an underestimation of the di⁄erences between hedgers and non-hedgers. Second, it might be that, because hedging is costly, (cid:133)rms facing greater (cid:133)nancial constraints are less likely to hedge, as suggested by Rampini, Su(cid:133), and Viswanathan (2014). If (cid:133)nancial constraints directly increase the policy sensitivity of stock prices and (cid:133)nancially constrained (cid:133)rms are less likely to hedge, we could be overestimating the di⁄erence between hedgers and nonhedgers. To reduce endogeneity bias concerns we introduce in this section an instrumental variables approach.47 Campello, Lin, Ma, and Zou (2011) show that the kinks and discontinuities stemming frominstitutionalfeaturesoftheU.S.taxsystemcanaddresstheendogeneityininterestrate hedging decisions, and use this idea to study the consequences of interest rate hedging for 47We deal with these concerns also by including proxies for the main potential omitted variables, such as (cid:133)nancial constraints and interest rate sensitivity of operating pro(cid:133)ts. For example, Table V includes interest rate sensitivity as a control. In addition, Appendix G discusses the relationship between hedging and (cid:133)nancial constraints in more detail and shows that our results are also obtained in a regression in which hedging and (cid:133)nancial constraints are included together. 30

loan spreads. The kinks or discontinuities of the tax schedule, especially at the zero income level due to loss o⁄set provisions, create a convexity of tax rates as a function of taxable income. As as result, (cid:133)rms can reduce their expected tax liabilities by minimizing income volatility.48 Because interest rate exposure increases income volatility, the convexity of tax rates provides (cid:133)rms with incentives to hedge against interest rate (cid:135)uctuations (relevance condition). At the same time, tax convexity is unlikely to have a direct (cid:133)rst-order e⁄ect on the sensitivity of stock prices to monetary policy shocks (exclusion restriction). The same instrument has been used in other papers, such as Chen and King (2014), to study the e⁄ect of hedgingonthe cost of public debt. We elaborate belowonthe reasons whyourinstrument is likely to satisfy the exclusion restriction. Our hedging variable is instrumented by (tax) Convexity, derived from the following formula as in Graham and Smith (1999), and Campello, Lin, Ma, and Zou (2011), Convexity = 4:88+0:019 Vol 5:50 Corr (cid:2) (cid:0) (cid:2) 1:28 DITC +3:29 DNOL+7:15 DSmallNeg (cid:0) (cid:2) (cid:2) (cid:2) +1:60 DSmallPos 4:77 DNOL DSmallNeg (cid:2) (cid:0) (cid:2) (cid:2) 1:93 DNOL DSmallPos (24) (cid:0) (cid:2) (cid:2) where Vol is the volatility of taxable income, Corr is the (cid:133)rst-order serial correlation of taxable income, DITC is a dummy for the existence of investment tax credits, DNOL is a dummy for net operating losses, and DSmallNeg (DSmallPos) is a dummy for small negative (positive) taxable income less than $500,000. We calculate the volatility of taxable income and the serial correlation of taxable income on a rolling basis, using historical annual data up to the year of interest, starting in 1989. The Convexity measure comes from the regression in Graham and Smith (1999) where the dependent variable is the expected percentage tax savings from a (cid:133)ve percent reduction in the volatility of taxable income and therebyprovidesameasurefortheincentiveofa(cid:133)rmtoreduceincomevolatility,forexample, through hedging. Campello et. al. (2011) argue that this tax convexity measure satis(cid:133)es the relevance and exclusion restrictions for the e⁄ect of interest rate hedging on loan spreads. We believe that their arguments also suggest that the tax convexity is an adequate instrument in our setting. As described in Campello et. al. (2011), the tax convexity formula uses data around a zero-income tax kink and instrument identi(cid:133)cation in that region comes fromthe nonlinear 48See discussions in Smith and Stulz (1985), Graham and Smith (1999), and Petersen and Thiagarajan (2000). 31

form of the income taxation function, rather than the income level itself. For example, a small negative or small positive taxable income indicates the proximity of the (cid:133)rm to the zero income tax kink, which would make it more bene(cid:133)cial for the (cid:133)rm to reduce the income volatility to minimize taxes, whereas the proximity to this kink is unlikely to have a direct e⁄ectonstockpricesensitivitytothemonetarypolicybeyonditse⁄ectonhedgingincentives. Similarly, the existence of net operating loss, DNOL, increases the incentives to hedge (DeAngeloandMasulis, 1990)butthesee⁄ectsaresmallerforthose(cid:133)rmswithsmallpositive and small negative taxable income, DNOL DSmallNeg and DNOL DSmallPos, (cid:2) (cid:2) because a hedging (cid:133)rm tightens its distribution of pro(cid:133)ts, reducing the expected bene(cid:133)t provided by its carryforwards. Moreover, a more negative serial autocorrelation would make the (cid:133)rm more likely to switch between pro(cid:133)ts and losses, giving higher incentive to hedge. It is unlikely that these variables have a direct e⁄ect on stock price sensitivity to monetary policy. Finally, our addition of pro(cid:133)tability as an additional control should address any remaining concerns that tax convexity might be correlated with policy sensitivity of stock prices through pro(cid:133)tability.49 Nevertheless, asinCampelloet. al. (2011), werepeattheinstrumental variableapproach excluding cash (cid:135)ow volatility to be on the conservative side, and we show that our results are not a⁄ected. We also use a variant of this approach with the lagged hedging dummy as an instrument, which would be a suitable instrument to the extent that it is not forward looking.50 WefollowasimplelinearinstrumentalvariableregressionasadvocatedbyAngrist and Pischke (2009, Ch. 4) and use Z, Surprise Z, and Surprise (BankDebt=At) (cid:3) (cid:3) (cid:3) Z as instruments where Z is the lagged hedging dummy, the convexity variable, and the components of the latter. [TABLE VI ABOUT HERE] Panel Aof TableVI presents ourresults forbankdebt usage. Incolumn1, wepresent the standard (cid:133)xed e⁄ects regression where we interact our bank debt usage with hedging for the subsampleofobservationsthathavevaluesforlaggedhedgingdummyandcolumn3doesthe same thing with Convexity. Both of these results are comparable to Table V quantitatively, and also imply that hedgers(cid:146)reaction to monetary policy barely increases with greater bank debtusage. Column2showsthattheresultsremainverysimilarwhenweuselaggedhedging dummy and its interaction with Surprise and Surprise (BankDebt=At) as an instrument (cid:3) 49Theoretically, pro(cid:133)tability could bias our results in either direction. As Graham and Rogers (2002) put it: "Pro(cid:133)tability might be inversely related to hedging if less pro(cid:133)table (cid:133)rms have a higher probability of encountering distress. Conversely, the option value of equity might encourage unpro(cid:133)table (cid:133)rms to hedge less than their nondistressed counterparts." 50We thank an anonymous referee for this suggestion. 32

for hedging, the di⁄erence across columns is very small both in terms of magnitude and statistical signi(cid:133)cance, as indicated by large Hausman p-value from Hausman test (0.934).51 Column 4 of Panel A repeats the same exercise using the variables underlying the convexity measure, except Vol, and column 5 also includes Vol as the instrumental variable. Column 6, on the other hand, uses the actual Convexity measure from formula (24). The resultsincolumns4, 5, and6areverysimilartoeachotheralthoughwelosestatisticalsignificance in column 6, not surprisingly because our sample is di⁄erent from Graham and Smith (1999) where formula (24) comes from. While the instrumental variable results seem quantitatively di⁄erent from standard (cid:133)xed e⁄ect regressions, the Hausman test cannot reject the hypothesis that they are the same, suggesting that the endogeneity of hedging is not a big concern. The (cid:133)nal column repeats the instrumental variable regression using both lagged hedging dummy and Convexity, with results very similar from the standard regressions. Moreover, the qualitative result from all these regressions is the same because the sum of the coe¢ cients of Surprise (BankDebt=At) and Surprise (BankDebt=At) Hedge t (cid:3) t 1 t (cid:3) t 1 (cid:3) t (cid:0) (cid:0) add up to a number statistically insigni(cid:133)cantly di⁄erent from zero, implying that bank debt usage does not signi(cid:133)cantly a⁄ect the sensitivity of stock prices to monetary policy shocks for hedgers. Panel B of Table VI repeats the same exercise for (cid:135)oating rate debt usage and the results are very similar to those from Panel A. Therefore, we conclude that our results are, at least qualitatively, robust to the potential endogeneity of the interest rate risk hedging decision. 4.4 The Floating Rate Channel for Constrained vs. Unconstrained Firms In the absence of (cid:133)nancial frictions, our evidence on the (cid:135)oating rate channel would be interpreted as a simple transfer of cash between a (cid:133)rm(cid:146)s shareholders and its creditors because monetary policy a⁄ects the benchmark rates underlying (cid:135)oating rate liabilities. In this case, the e⁄ect of bank debt usage on stock prices would simply represent the expected present value of this transfer over the lifetime of the loan. In the presence of (cid:133)nancing frictions, however, the impact could be ampli(cid:133)ed through the e⁄ect of variations in the interest expense on the (cid:133)rm(cid:146)s liquidity position and overall balance sheet strength, which in turn could a⁄ect the (cid:133)rm(cid:146)s ability to (cid:133)nance pro(cid:133)table investment opportunities.52 51For the (cid:133)rst stage regression results, we refer the reader to Table A7. The coe¢ cients of instruments go in the expected direction, i.e. past hedging predicts future hedging, and higher convexity and values of variables that predict higher convexity are associated with higher hedging activity. The autocorrelation of cash (cid:135)ows turns out to be more signi(cid:133)cant in the case of (cid:135)oating rate debt. A high R2 and F-statistic also lend support to the relevance condition. 52Other frictions might also result in real implications of the interaction of monetary policy actions and (cid:135)oating rate debt. For example, an existing debt overhang problem (Myers 1977) might be worsened by an increase in the claims of banks following a monetary policy tightening. Or an asset substitution problem 33

To explore this, we analyze the stock price reaction of hedgers vs. non-hedgers within groups of (cid:133)rms with di⁄erent degrees of (cid:133)nancial constraints. We explore whether hedging a⁄ects the policy sensitivity of stock prices of (cid:133)nancially constrained bank debt users more than it does those of less (cid:133)nancially constrained bank debt users, which would be consistent with the ampli(cid:133)cation of the (cid:135)oating rate channel through the e⁄ect of (cid:133)nancing constraints. Therefore, we run our original regression (22) separately for hedgers and non-hedgers that face di⁄erent degrees of constraints, measured by age and the Hadlock and Pierce (2010) (HP) index. [TABLE VII ABOUT HERE] The (cid:133)rst two rows of Table VII give the coe¢ cients of interest from these regressions. We predict that our (cid:135)oating channel is mute among the hedgers regardless of the degree of (cid:133)nancial constraints which is con(cid:133)rmed by the coe¢ cient of Surprise (BankDebt=At) in (cid:3) columns 3, 4, 7, and 8 in Table VII as these coe¢ cients are statistically and economically insigni(cid:133)cant. Moreover, our ampli(cid:133)cation mechanism through (cid:133)nancial frictions predicts that (cid:133)nancially constrained non-hedgers should react the strongest. This is con(cid:133)rmed by columns1, 2, 5, and6inTableVIIasthecoe¢ cientofSurprise (BankDebt=At) islargerin (cid:3) magnitude among the more constrained (young and high-HP) non-hedgers in comparison to less constrained non-hedgers. In other words, (cid:133)nancing constraints only matter signi(cid:133)cantly for the e⁄ect of bank debt usage on the responsiveness to monetary policy when (cid:133)rms are exposed to interest rate risk because they do not hedge. For completeness, Table A8 repeats this exercise with the Whited-Wu constraint index and presents very similar results. The results do not go through with the Kaplan-Zingales index, which is consistent with the fact that the poor performance of the KZ index is the reason that Hadlock and Pierce developed their HP index.53 Table A9 also compares (cid:133)rms with di⁄erent liquidity positions, using the current ratio and the coverage ratio, and shows that nonhedgers with poor liquidity are the ones with highest policy rate sensitivity, con- (Jensen and Meckling (1976)) might arise as the same increase in interest rates might increase the convexity of shareholders(cid:146)claims and enhance a distortion towards risky investment. In addition, we refer in this papertotherealeconomicoutcomesdirectlycausedby(cid:133)rms(cid:146)decisions, butweshouldnotethateveninthe absenceofchangesin(cid:133)rms(cid:146)decisionsthecash-(cid:135)owreallocationcausedbythe(cid:135)oatingratechannelcanhave macroeconomic e⁄ects to the extent that debtholders and equityholders have di⁄erent consumption-savings behavior because of di⁄erences in demographic characteristics or risk aversion. These additional e⁄ects are outside the scope of our paper. 53WeprefertheHPmeasureamongothercandidates,suchasKaplanandZingales(KZ,1997)andWhited andWu(WW,2006),becauseHadlockandPierce(2010)showthattheKZandWWindiceshaveverylittle power to predict (cid:133)nancial constraints and any power they have comes from (cid:133)rm size and age, the two variables they use to create their composite HP index. 34

sistent with our other results.54 Since the coverage ratio is directly related with a (cid:133)rm(cid:146)s interest expense, we study this variable in more detail in Section 5.1. This evidence suggests that the e⁄ect of our new (cid:135)oating rate channel goes beyond a simple reallocation of cash (cid:135)ows between lenders and shareholders following monetary policy events, possibly re(cid:135)ecting a (cid:133)nancial ampli(cid:133)cation mechanism that works through a (cid:133)rm(cid:146)s interest expense on existing (cid:135)oating rate debt and its liquidity position. If so, our (cid:135)oating rate channel might bear implications for the (cid:133)nancing and production choices of the (cid:133)rms as well. We explore this next in Section 5.2. 5 The Floating Rate Channel: Evidence from Balance Sheet Variables 5.1 Impact on Firms(cid:146)Liquidity Position In this section, we explore the mechanism through which our (cid:135)oating rate mechanism a⁄ects (cid:133)rms(cid:146)balance sheet strength. We conjecture that monetary policy can have a strong impact on the liquidity positions of (cid:133)rms exposed to interest rate risk because their cash (cid:135)ows will be a⁄ected by changes in their interest expense. We focus on the behavior of the interest rate coverage ratio and cash holdings of (cid:133)rms following monetary policy events. We present here a summary of our results, and refer the reader to Appendix H for a detailed description of the tests. The interest rate coverage ratio, de(cid:133)ned as the ratio of a (cid:133)rm(cid:146)s interest expense to the sum of interest expense plus cash (cid:135)ow, is a proxy for (cid:133)rm (cid:133)nancial distress often used in the empirical literature on (cid:133)rm (cid:133)nancial constraints (Whited (1992), Gertler and Gilchrist (1994), and Campello and Chen (2010), for example).55 A high coverage ratio indicates that the (cid:133)rm may face di¢ culties trying to meet interest rate payments with current cash (cid:135)ows and may need to access external (cid:133)nance, make use of retained earnings, or decrease investment and hiring to avoid default. The main channel through which our (cid:135)oating rate mechanism operates is by a⁄ecting this coverage ratio. We test whether a higher bank debt usage as a share of total assets increases the respon- 54We also consider cash-to-assets ratio as an additional measure of liqudity. For the study of (cid:133)nancial constraints, however, this measure is likely to su⁄er from endogeneity because (cid:133)nancially constrained (cid:133)rms tend to hoard cash (Opler et. al., 1999; Bates, Kahle, Stulz, 2009). Consistent with this we see that bank debtusagehasthestrongeste⁄ectfornonhedgerswithhighcashholdings. Notethatthisdoesnotinvalidate the results in the next section which uses the (cid:146)change(cid:146)in cash which is a suitable measure to capture the change in liqudity position of a given (cid:133)rm. 55Part of the literature (see eg. Fazzari, Hubbard and Petersen (2000) or Gertler and Gilchrist (1994)) measures the interest coverage ratio as cash (cid:135)ow over interest expense, instead of as interest expense over cash (cid:135)ow as we (and others, such as Whited (1992)) do. We (cid:133)nd that our choice is more natural in our context, as it allows us to discuss increases in interest expense in terms of increases in our coverage ratio. 35

siveness of (cid:133)rms(cid:146)interest rate coverage ratios following monetary policy actions, due to the higher likelihood of this debt being (cid:135)oating rate.56 Our mechanism predicts that bank debt usage increases the sensitivity of the coverage ratio for non-hedgers, but does not a⁄ect the sensitivity of hedged bank debt users. [TABLE VIII ABOUT HERE] The results are displayed in Table VIII. Consistent with our prediction, the estimate of the coe¢ cient on (Sum)Change BankDebt=At for the hedged sample is statistically t t 1 (cid:3) (cid:0) insigni(cid:133)cant at all horizons, while for the unhedged sample it is always positive after the second quarter following the monetary policy shock, and statistically signi(cid:133)cant at horizons of 3 and 5 quarters. The di⁄erence between the estimates across subsamples is large and statistically signi(cid:133)cant at horizons of 5 and 6 quarters. In terms of economic magnitude, a 100bp tightening of monetary policy is associated with an increase in the coverage ratio of 0.09 (0.12) for an unhedged (cid:133)rm fully (cid:133)nanced with bank debt, relative to a hedged (cid:133)rm fully (cid:133)nanced with bank debt, at a horizon of 5 (6) quarters. This e⁄ect on coverage ratio is important not only because su¢ ciently large increases in the coverage ratio following a monetary policy tightening might force (cid:133)rms to access additional (cid:133)nancing to meet interest rate paymentsandfundtheirinvestmentandhiringplansbutitmightalsoincreasethelikelihood of a covenant violation, which has important implications for (cid:133)rms(cid:146)capital expenditures, as shown in Nini, Su(cid:133), and Smith (2012). Inthepresenceof(cid:133)nancingconstraints, (cid:133)rmsmightneedtotapintoretainedearningsinstead. Totestthisprediction, wecomputethechangeincashholdings, andevaluatewhether theimpactofbankdebtusageonthesensitivityofcashholdingstomonetarypolicyissignificantly stronger for (cid:133)nancially constrained (cid:133)rms than for unconstrained (cid:133)rms in the sample of unhedged (cid:133)rms, and whether this di⁄erence is absent or is at least signi(cid:133)cantly smaller in the sample of hedged (cid:133)rms. We classify (cid:133)rms as (cid:133)nancially constrained (unconstrained) if their value of the Hadlock and Pierce (2010) (HP) index is above (below) the median, and report the di⁄erence between the estimates across constrained and unconstrained (cid:133)rms, within each of the subsamples of hedgers and non-hedgers, and also report the statistical signi(cid:133)cance of the di⁄erence. [TABLE IX ABOUT HERE] 56Monetarypolicyactionsarecalculatedascumulativequarterlychangeintheinterestrate,asinAshcraft and Campello (2007) and JimØnez, Ongena, Peydr(cid:243), and Saurina (2012, 2014), instead of the cumulative surprise component because cash (cid:135)ow and the interest rate expense on existing debt is not forward-looking the way stock prices are. 36

Our results are displayed in Table IX, in which coe¢ cient estimates for (cid:12) for the four 3 subsamples of (cid:133)rms (hedgers/non-hedgers, constrained/unconstrained) and the estimates for the di⁄erence between groups, the coe¢ cient on C\hange (BankDebt=At) Constrained , t t 1 t (cid:0) are shown for horizons of four and six quarters.57 Being (cid:133)nancially constrained only matters, in a statistically signi(cid:133)cant way, for the response of cash holdings of bank debt users to monetary policy when (cid:133)rms do not hedge their interest rate risk. More speci(cid:133)cally, after four quarters following a 1 percentage point increase in the federal funds rate, a constrained (cid:133)rm fully (cid:133)nanced with bank debt experiences on average a 4.2 percent stronger drop in its cash holdings relative total assets than an unconstrained (cid:133)rm if both (cid:133)rms are unhedged. This di⁄erence increases to 9.5 percent at a horizon of 6 quarters. Constraints, however, do not a⁄ect the responsiveness of cash holdings to monetary policy of bank debt users if they are hedged, at any horizon. Taken together, the evidence in this section highlights the nature of our (cid:135)oating rate mechanism as a source of economically signi(cid:133)cant liquidity shocks for (cid:133)rms. 5.2 Real Implications In this section, we test whether the stronger e⁄ect of the (cid:135)oating rate channel for (cid:133)nancially constrained (cid:133)rms identi(cid:133)ed using stock prices, the interest rate coverage ratio, and cash holdings, is associated with signi(cid:133)cant real outcomes in the a⁄ected (cid:133)rms. We focus on the implications for (cid:133)rms(cid:146)inventory investment, (cid:133)xed investment, and sales. The nature of our (cid:135)oating rate mechanism as a liquidity event means it is particularly likely to manifest itself in the behavior of inventory investment, one of the most liquid components of (cid:133)rms(cid:146)balance sheets. We follow Kashyap, Lamont and Stein (1994) and adopt their empirical speci(cid:133)cation for our inventory investment regressions, which we report in Table X. [TABLE X ABOUT HERE] There is a statistically strong negative relationship between bank debt usage and the sensitivity of inventory investment to monetary policy changes for (cid:133)rms in the unhedgedconstrained category. The economic magnitude of the relationship is large for this subgroup: after 6 quarters following a 1 percentage point increase in the Federal funds rate, increasing bank debt usage from 0% to 100% of total assets is associated with a decrease in inventories of on average 21.2%. Bank debt usage instead does not a⁄ect the sensitivity of inventory investment to monetary policy in a statistically signi(cid:133)cant way if (cid:133)rms are (cid:133)nancially un- 57The di⁄erence between groups is obtained by interacting all variables, including (cid:133)xed e⁄ects, with the (cid:133)nancial constraint dummy. 37

constrained or if they are not exposed to interest rate risk in their bank debt because they hedge. Previous empirical studies have shown that the inventory investment of (cid:133)nancially constrained(cid:133)rmsismoresensitivetomonetarypolicythanthatoflargeandrated(cid:133)rms(Gertler and Gilchrist (1994), Kashyap, Lamont and Stein (1994)). Our evidence shows that (cid:133)nancial constraints only increase (cid:133)rms(cid:146)sensitivity to monetary policy if these (cid:133)rms are exposed to interest rate risk through their bank debt, suggesting that our (cid:135)oating rate mechanism is a potentially important driver of this result. This result might be particularly relevant from a macroeconomic perspective given that inventories constitute the most volatile component of GDP (Blinder and Maccini (1991), Davis and Kahn (2008)). We next study the behavior of sales, which we interpret, in line with existing literature, as a proxy for (cid:133)rm-level output (Gertler and Gilchrist (1994), Bond, Elston, Mairesse, and Mulkay (2003)). The results are displayed in Table XI, and are in line with our previous evidence. Being (cid:133)nancially constrained has twice the impact on the sensitivity of sales to monetary policy of unhedged bank debt users than on the sensitivity of hedged bank debt users. Increasing bank debt usage from 0% to 100% of assets is associated with an additional decrease in sales after four (six) quarters of 19.1% (17.8%) following a 100bp monetary policy tightening when the (cid:133)rm is (cid:133)nancially constrained and unhedged, relative to when it is unconstrained and unhedged. For hedged (cid:133)rms, this same di⁄erence is only 7.9% (9.7%). [TABLE XI ABOUT HERE] Finally, we explore the behavior of (cid:133)xed investment. A large body of empirical research documents the di¢ culty of (cid:133)nding a relationship between (cid:133)xed investment and interest rates (Caballero(1999), SharpeandSuarez(2014)), suggestingthattheimpactofmonetarypolicy on (cid:133)xed investment, to the extent that it is signi(cid:133)cant, might occur mostly through indirect channels such as the one discussed in this paper. To test this prediction, we expand our baseline empirical speci(cid:133)cation to include the main factors that have been identi(cid:133)ed in the empiricalliteraturetomatterfor(cid:133)rminvestment(Eberly,RebeloandVincent(2012)),which are Tobin(cid:146)s Q, cash (cid:135)ow, and lagged investment. The results are displayed in Table XII. Consistent with our mechanism, (cid:133)nancial constraints have a signi(cid:133)cant e⁄ect on the impact of bank debt usage on monetary policy sensitivity of (cid:133)xed investment only for the subsample of (cid:133)rms that do not hedge against interest rate risk. The economic magnitude of the relationship is large for this subgroup: after 6 quarters following a 1 percentage point increase in the Federal funds rate, a hypothetical (cid:133)nanciallyconstrained(cid:133)rmthatisfully(cid:133)nancedwithbankdebtsu⁄ersachangeintotalcap- 38

italwhichisonaverage15.8percentagepointslowerthantheonea(cid:133)nanciallyunconstrained bank debt user experiences. Financial constraints however do not signi(cid:133)cantly in(cid:135)uence the responsiveness of (cid:133)xed investment to monetary policy for the subsample of (cid:133)rms that are not exposed to interest rate risk. The e⁄ects after four quarters are not statistically signi(cid:133)cant, which might not be surprising given that investment in tangible capital is more likely to su⁄er from adjustment costs compared to inventory investment, and this might delay any possible e⁄ects.58 [TABLE XII ABOUT HERE] Taken together, the evidence discussed in this section suggests that the e⁄ect of the (cid:135)oating rate channel goes beyond a simple reallocation of cash (cid:135)ows between lenders and shareholders and has real implications for the a⁄ected (cid:133)rms. The impact of our channel on employment, which we have not analyzed due to the absence of reliable quarterly data on number of workers in our databases, might be signi(cid:133)cant as well, depending on the costs of adjusting the workforce along both the intensive and extensive margins. Finally, our strong results on the sensitivity of cash holdings in Section 5.1 suggest other latent and subtle mechanisms that are harder to test for: (cid:133)rms may choose to build large cash bu⁄ers instead of investing in anticipation of future increases in the rates on their (cid:135)oating rate debt and these ex-ante e⁄ects on investment and employment could be large. It is important to point out two potential caveats of the analysis in this section. First, while the identi(cid:133)cation of our proposed e⁄ects can be argued to be strong in our stock return regressions, endogeneity biases are more likely to remain in our regressions dealing with balance sheet variables because of the substantial lag between the monetary policy event date and the date in which the e⁄ects are measured (4 or 6 quarters later). Second, our estimation might also be a⁄ected by the fact that there are several FOMC announcements in the 4 or 6 quarters after our FOMC announcement of interest, which makes it harder to establish causality. As such, our results on real variables are suggestive. 58The coe¢ cient (cid:11) on the interacted term C\hange (BankDebt=At) is positive for most subsamples 3 t t 1 andhorizons,andinTableA10we(cid:133)ndthatitisalsooftenpositivewhenus(cid:0)ingasurprisemeasureofmonetary policy. This means that bank debt usage makes (cid:133)xed investment relatively less sensitive to monetary policy on average, at horizons of between 4 and 6 quarters, for some subsamples. One possible explanation for this (cid:133)nding is that banks protect their borrowers from a tightening in credit conditions in the context of the lending relationships that they form with their clients. See Rajan and Zingales (1998) and Ehrmann, et al. (2001)foradiscussionoftheroleoflendingrelationshipsinalleviatingtheimpactofcontractionarymonetary policy actions on bank borrowers. In the context of the recent crisis, several papers provide evidence that one channel through which banks protect their relationship borrowers in times of credit market distress is through precommitted credit (Ivashina and Scharfstein (2010), and Campello, Giambona, Graham, and Harvey (2011)). This credit insurance role is compatible with our (cid:135)oating rate channel, and both channels might be operating in parallel. 39

6 How Important is the Floating Rate Channel? 6.1 The Aggregate Business Exposure to the Floating Rate Channel The macroeconomic relevance of our monetary policy transmission mechanism depends on the aggregate amount of business debt that is exposed to interest rate risk. The Flow of Funds states that there were roughly $12.5 trillion of outstanding non(cid:133)nancial business debt in the United States at the end of 2015. Of this amount, around $5 trillion are corporate bonds, a majority of which are issued with (cid:133)xed interest rates.59 Of the remaining $7 trillion, 2.5 trillion correspond to corporate loans, of which around 75% are estimated to be referenced to LIBOR.60 Noncorporate business loans account for $4.5 trillion, and Du¢ e and Stein (2015) estimate that 30-50% of them are tied to LIBOR. Note that this is a lower bound estimate of the amount of debt exposed to interest rate risk, because we are basing our estimates on the fraction of debt that is tied to LIBOR, which is the most common but not the only base rate for (cid:135)oating rate arrangements. An example of a common alternative base rate is the prime rate (displayed in Figure 3), which is also closely tied to policy rates. Thee⁄ectiveexposuretointerestratesalsodependson(cid:133)rms(cid:146)usageofinterestrateswaps, but the data suggests that hedging does not substantially alter the aggregate exposures. In our own sample, less than 50% of bank debt users use some amount of hedging. Bretscher, Schmid, and Vedolin (2016) estimate the precise amounts of debt being hedged in a handcollected sample and (cid:133)nd that the percentage of (cid:135)oating rate debt swapped to a (cid:133)xed rate is under 10% for large public (cid:133)rms. Small and medium-sized (cid:133)rms outside these samples, which are not publicly listed and hold a large share of bank debt, make little use of hedging derivatives (Vickery (2008)). Putting this information together, we arrive at an estimate of between $3.2 and $4.1 trillion of outstanding debt exposed to a (cid:135)oating interest rate, or between 26 and 33% of the total.61 This means that aggregate U.S. business non(cid:133)nancial debt exposed to LIBOR 59Inourownsample,we(cid:133)ndthatforthose(cid:133)rmswhosedebtisentirelyfromnon-banksourcesonlyaround 9 percent of debt is (cid:135)oating rate. These (cid:133)gures are consistent with Faulkender (2005), according to whom 7% of bonds are issued with a (cid:135)oating rate. Ogden, Palomino, Sinha, and Yook (2016) put this (cid:133)gure at around 2% in recent years. 60See Du¢ e and Stein (2015). Our own estimates based on our sample of public (cid:133)rms suggest that about 76 percent of bank loans are (cid:135)oating rate. As discussed in Section 3.1, Faulkender (2005) and Aslan and Kumar(2012)estimateevenhigherpercentagesinparticularsubsetsofcorporateloans. Overall,thesevalues underestimate slightly the total share of loans referenced to rates that are linked to the fed funds rate, such as those loans directly referenced to the fed funds rate or prime rate. 61This (cid:133)gure is consistent with other estimates. Chernenko and Faulkender (2011) report that (cid:135)oating ratedebtrepresents32.7%oftotaldebtofthecorporatesectorduringthe1993(cid:150)2003period. Thepercentof total debt of all (cid:133)rms (in other words, the value-weighted average) in our sample, made up of all Compustat 40

(cid:135)uctuations was equivalent to roughly 20% of GDP ($18.0 trn) in 2015, which lends support to the notion that our mechanism has the potential to have signi(cid:133)cant macroeconomic relevance. 6.2 Comparison with the Bank Lending Channel Is the (cid:135)oating rate channel quantitatively relevant from a macroeconomic perspective? One possible answer comes from studying how large we would expect the e⁄ect to be for the overall economy if all (cid:133)rms had borrowed at a (cid:133)xed rate or had access to hedging. However, because our empirical analysis focuses on local e⁄ects it is hard to argue that this analysis provides the true overall e⁄ect, given the potential general equilibrium e⁄ects underlying this counterfactual. Instead, we compare the e⁄ect of the (cid:135)oating rate channel with that of the bank lending channel, as studies on the latter typically focus on local e⁄ects as well. In Appendix I we introduce an analysis based on results from the existing literature and (cid:133)nd that a (cid:133)rm that usuallyborrows$100from(cid:133)nancialintermediarieswillexperiencealong-runcumulative$0.3 external (cid:133)nancing shortfall, as an upper bound, if the federal funds rate increases by one percentage point (Oliner and Rudebusch (1996), Holod and Peek (2007)). We also calculate that the same rate hike would cause, through the (cid:135)oating rate channel, a cash shortfall of between $0.32 (minimumover a one year period) to $0.88 (maximumover a two year period) on a $100 loan. Overall, these calculations suggest that the aggregate (cid:133)nancing shortfall caused by the (cid:135)oating rate channel of monetary policy is likely to be at least as large as the shortfall caused by the bank lending channel. For both channels, the total actual e⁄ect of this shortfall will be determined by similar ampli(cid:133)cation mechanisms, such as the borrowers(cid:146)(cid:133)nancial health. The (cid:135)oating rate channel will also have one additional subtle, but potentially important, ampli(cid:133)cation mechanism: it causes an internal cash shortfall, whereas the bank lending channel causes an external cash shortfall. The external cash shortfall (loss of access) due to the bank lending channel forces the (cid:133)rm to forgo investment projects without a⁄ecting its equity position. However, the internal cash shortfall due to the (cid:135)oating rate channel will always reduce the (cid:133)rm(cid:146)s equity and liquidity position and hence potentially have stronger e⁄ects on the balance sheet health of the (cid:133)rm. 6.3 Evidence from the Unconventional Policy Period As an alternative approach to the importance of the (cid:135)oating rate channel, we apply our benchmark regression to a period during which we do not expect the (cid:135)oating rate channel to be operative, so that any remaining e⁄ect can be attributed to other banking channels. (cid:133)rms excluding utilities and (cid:133)nancials, that is bank debt ((cid:135)oating-rate debt) is 31.6% (25.4%) in 2008. 41

Since late 2008, when the federal funds target rate hit the zero lower bound, the Federal Reserve has focused on alternative policy measures in order to stimulate the U.S. economy. These measures, typically referred to as quantitative easing or unconventional monetary policy tools, have involved large scale purchases of assets with long maturities. As seen in Figure 1, these purchases did not a⁄ect the short-term benchmark interest rates underlying the (cid:135)oating rate bank debt arrangements, as these rates are already at their lowest possible level. If the (cid:135)oating rate channel is important we would expect bank debt usage to have a much less prominent role during the unconventional monetary policy period. Therefore, testing the e⁄ect of bank debt usage in the unconventional monetary policy period is useful to gauge the importance of the (cid:135)oating rate channel. The main challenge for this approach stems from (cid:133)nding a measure of the overall stance of unconventional monetary policy in general, and the surprise component of the Federal Reserve(cid:146)s actions in particular. While one could use the Federal Reserve(cid:146)s balance sheet as a proxy, many of the Fed(cid:146)s actions were announced in advance and hence this would not provide a suitable measure of the monetary policy surprises in the unconventional period. Instead, wefollowWright(2012)andusethehigh-frequencypricechangesinlonger-maturity Treasury futures on a very tight event window around FOMC announcements during the unconventional period to capture the unanticipated changes in the stance of monetary policy, as these tight windows do not include any other macroeconomic news. We prefer this identi(cid:133)cation strategy for unconventional monetary policy surprises over alternative strategies, such as vector-auto-regressions, because the monetary policy surprises identi(cid:133)ed in this fashion are less model-dependent and the regression results are easier to interpret and to compare to our event study results from the previous sections.62 Wright (2012) uses intraday data on two-, (cid:133)ve-, ten-, and thirty-year Treasury bond futures trading in the Chicago Mercantile Exchange and identi(cid:133)es a particular set of FOMC announcement dates. The monetarypolicysurprises onthose dates are computedas the (cid:133)rst principal component of yield changes from 15 minutes before each of these announcements to 1 hour and 45 minutes afterwards. The announcement dates range from November 25, 2008 to September 21, 2011 and the associated monetary policy surprises calculated this way are presented in Table 5 of Wright (2012). These surprises are scaled so that one unit of the shock leads to a 12bp increase in the ten-year Treasury according to Wright (2012), and this is roughly equivalent to the e⁄ect of a 100bp increase in the fed funds target on the ten-year Treasury yield during the conventional period. While the scale is the same as the 62Other event studies that focus on the e⁄ects of unconventional monetary policy are either more descriptive in nature and do not provide a measure of the monetary policy surprise (e.g. Gagnon et al. (2010) and Krishnamurthy and Vissing-Jorgenson (2011)), or base the surprise on a subset of the assets employed by Wright (2012) (e.g. Chodorow-Reich (2014)). 42

shocks presented in Table 5 of Wright (2012), the sign is inverted so that a positive surprise in the following regressions should be interpreted as a contractionary shock, consistent with the other regressions in our paper. Table XIII repeats our benchmark regression given in equation (22) by substituting the conventional monetary policy surprise with the unconventional monetary surprise after applying the same (cid:133)lters to (cid:133)rms as in our previous analysis. The (cid:133)rst column shows that the e⁄ect of an unconventional monetary policy surprise that increases the ten-year Treasury yield by 12bp, decreases the stock price of a (cid:133)rm by about 35bp on average. This number is close to the number (55bp) reported in Wright (2012) for the intraday returns of the S&P 500 futures. Any di⁄erence stems from our use of panel data regressions with a more comprehensive sample and our use of two-day returns, following the strategy we have employed in the previous sections. [TABLE XIII ABOUT HERE] More interestingly, the second column shows that the e⁄ect of bank debt usage on the transmission of monetary policy to stock prices not only diminishes but also goes in the opposite direction of what we observe in the conventional period. In terms of economic magnitude, a one standard deviation (0.13) increase in bank debt usage leads to about a 6bp lower reaction, in comparison to a 35bp reaction of the average (cid:133)rm(cid:146)s stock. One explanation for this pattern might be that bank-(cid:133)rm relationships enable (cid:133)rms to bene(cid:133)t fromsomedegreeofinsuranceprovidedbytheirlendersagainstchangesincreditavailability. The increased importance of this insurance during the recent (cid:133)nancial crisis, combined with the absence of the (cid:135)oating rate channel in the unconventional monetary policy period, can lead to the positive coe¢ cient observed in column 2. Another explanation, which is easier to test, is that we simply need to control for additional variables. Indeed, the third column shows that after including our original control variables, we (cid:133)nd that bank debt usage has an economically and statistically insigni(cid:133)cant e⁄ect on the responsiveness of stock prices to monetary policy shocks. Either interpretation of our results is consistent with the reduced e⁄ectofbankdebtusageintheunconventionalpolicyperiodduetotheabsenceofthe(cid:135)oating rate channel, a channel that previous sections have proven to be particularly important during the conventional monetary policy period. As a (cid:133)nal test, we look at the e⁄ect of hedging on the responsiveness of stock prices to monetary policy shocks. If the di⁄erence between hedgers and non-hedgers we presented earlier in Table V truly re(cid:135)ects the importance of the (cid:135)oating rate channel for the e⁄ect of bank debt usage on the transmission of monetary policy, we should (cid:133)nd that hedging should notin(cid:135)uencethee⁄ectofbankdebtusageduringtheunconventionalmonetarypolicyperiod. 43

Therefore, the last two columns look at the e⁄ect of bank debt usage for hedgers and nonhedgers separately like we did for the conventional period in Table V. We (cid:133)nd that the di⁄erence between hedgers and non-hedgers actually goes in the opposite direction of what we observe in Table V, with bank debt usage increasing the responsiveness of hedgers and decreasing the responsiveness of non-hedgers. Nevertheless, the e⁄ect of bank debt usage is statistically insigni(cid:133)cant for bothhedgers and non-hedgers. This result is further in line with our argument that the (cid:135)oating rate channel is important for the e⁄ect of bank debt usage on the transmission of monetary policy, and that this important channel is mute during the unconventional monetary policy period. Finally, we would like to discuss the generalizability of our analysis of unconventional policy. Although there does not seem to be any evidence of asymmetric e⁄ects for rate increases and decreases in the conventional policy period, as discussed in Section 4.2, our (cid:133)ndings about unconventional monetary policy might be limited to instances when the Federal Reserve is loosening. However, unconventional policy may play a more signi(cid:133)cant role once the Federal Reserve starts tightening using unconventional policy tools. Since this has not happened so far, it would be di¢ cult for us to make a claim about this. 7 Conclusion According to the (cid:133)rmbalance sheet channel of monetary policy, a tightening in monetary policy increases the debt-service burden of borrowers and reduces the value of their collateral andinternalfunds,therebyincreasingtheexternal(cid:133)nancepremiumof(cid:133)nanciallyconstrained (cid:133)rms. Our results con(cid:133)rm that bank lending plays an important role in this transmission mechanism. We use (cid:133)rms(cid:146)hedging activity to provide evidence that an important portion of this transmission is driven by the mechanical relationship between monetary policy and the reference rates for the (cid:135)oating rate arrangements underlying most bank loans to businesses. This channel, which we call the (cid:135)oating rate channel, is distinct fromearlier channels studied in the empirical literature in that it works through existing debt rather than new debt. Our results also contribute to the debate about the e¢ cacy of large scale asset purchases (LSAP)asanalternativetool of monetarypolicy. Thisdebatehasidenti(cid:133)edseveral channels throughwhichLSAPmaya⁄ectpricesofdi⁄erent(cid:133)nancialassets.63 Financialintermediation does not play a signi(cid:133)cant role in any of the most relevant channels, which is perhaps not 63KrishnamurthyandVissing-Jorgensen(2012)showthatquantitativeeasingmightoperatethroughchannels related to signaling, demand for long-term safe assets, in(cid:135)ation, mortgage-backed securities (MBS) prepayment, or corporate bond default risk. See also Gagnon et al. (2011), Joyce et al. (2011), Krishnamurthy and Vissing-Jorgensen (2011), Vayanos and Vila (2009), Hamilton and Wu (2011), Christensen and Rudebusch (2012), Swanson (2011), Li and Wei (2013), and D(cid:146)Amico and King (2013). 44

surprising since aggregate bank loan growth relative to total deposits has been low.64 Our results reveal another reason why LSAP might have a limited impact through bank lending. We (cid:133)ndthat the (cid:135)oating rate channel is not operative inthe unconventional monetarypolicy period, and hence that bank debt usage plays a much less important role in the transmission of monetary policy during this period, consistent with the fact that the zero lower bound signi(cid:133)cantly limited the ability of the Federal Reserve to a⁄ect the short-term benchmark rates underlying (cid:135)oating rate bank debt. It is important to point out some potential caveats of our study. First, as with any other microeconomic analysis of the importance of (cid:133)nancial frictions, general equilibrium forces could a⁄ect the macroeconomic importance of our channel. Large, (cid:133)nancially unconstrained (cid:133)rms could be bene(cid:133)ting from the downsizing of distressed or (cid:133)nancially constrained (cid:133)rms followingamonetarypolicytightening,andtakinguppartoftheirlostmarketshare,through general equilibrium e⁄ects mediating through prices of goods and services or wages, for example. Ourlocalestimateswouldmissthesetotale⁄ectsofcomparingourcurrenteconomy to a hypothetical one in which no (cid:133)rm faced our (cid:135)oating rate channel e⁄ect. Second, there might be other factors at play that interact with our (cid:135)oating rate channel. For example, an increase in in(cid:135)ation and in(cid:135)ation expectations induced by looser monetary policy can reduce the real debt burden, as discussed in Jermann, Gomes, and Schmid (2016). This e⁄ect would be stronger for (cid:133)xed rate liabilities than for (cid:135)oating rate liabilities, which would counteract the direct cash (cid:135)ow e⁄ect of the (cid:135)oating rate channel and suggest that our estimates are a lower bound. We hope that our results stimulate further research in this direction to provide a better understanding of how conventional and unconventional monetary policies di⁄er in terms of their transmission to the real economy. 64http://www.forbes.com/sites/francescoppola/2014/01/21/banks-dont-lend-out-reserves/ 45

References [1] Acharya, V., H. Almeida, F. Ippolito and A. PØrez-Orive (2014). (cid:147)Bank Lines of Credit as Contingent Liquidity: A Study of Covenant Violations and their Implications(cid:148). ECB Working Paper No. 1702. [2] Alderson, Michael J., and Brian L. Betker. "Liquidation costs and capital structure." Journal of Financial Economics 39.1 (1995): 45-69. [3] Altman, Edward I. "A further empirical investigation of the bankruptcy cost question." The Journal of Finance 39.4 (1984): 1067-1089. [4] Altman, E. I., Gande, A., & Saunders, A. (2010). Bank debt versus bond debt: Evidence from secondary market prices. Journal of Money, Credit and Banking, 42(4), 755-767. [5] Amihud, Y. (2002). "Illiquidity and stock returns: cross-section and time-series e⁄ects", Journal of Financial Markets 5, pp. 31-56 [6] Angrist, J. D., & Pischke, J. S. (2008). "Mostly harmless econometrics: An empiricist(cid:146)s companion". Princeton University Press. [7] Ashcraft,A.B.(2006).NewEvidenceontheLendingChannel.JournalofMoney, CreditandBanking, 38(3), 751-775. [8] Ashcraft, A. B., & Campello, M. (2007). Firm balance sheets and monetary policy transmission. Journal of Monetary Economics, 54(6), 1515-1528. [9] Aslan, Hadiye, and Praveen Kumar. "Strategic ownership structure and the cost of debt." Review of Financial Studies 25.7 (2012): 2257-2299. [10] Asquith, Paul, Robert Gertner, and David Scharfstein, 1994, Anatomy of (cid:133)nancial distress: An examination of junk bond issuers, Quarterly Journal of Economics 109, 625(cid:150)658. [11] Bates, T. W., K. M. Kahle, and R. M. Stulz. 2009. Why Do U.S. Firms Hold So Much More Cash Than They Used To? Journal of Finance, 1985-2021. [12] Bernanke, Ben S., and Alan S. Blinder (1988), (cid:147)Credit, Money, and Aggregate Demand," American Economic Review, 78, 435(cid:150)439. [13] Bernanke, Ben S., and Mark Gertler (1995), (cid:147)Inside the Black Box: The Credit Channel of Monetary Policy Transmission," Journal of Economic Perspectives, 9, 27(cid:150)48. [14] Bernanke, Ben S., and Kenneth N. Kuttner (2005), (cid:147)What Explains the Stock Market(cid:146)s Reaction to Federal Reserve Policy?," Journal of Finance, 60, 1221(cid:150)1257. [15] Blinder, A. S., & Maccini, L. J. (1991). Taking stock: a critical assessment of recent research on inventories. The Journal of Economic Perspectives, 73-96. [16] Bodie,Z.,&Taggart,R.A.(1978).Futureinvestmentopportunitiesandthevalueofthecallprovision on a bond. The Journal of Finance, 33(4), 1187-1200. [17] Bolton, P., and D. S. Scharfstein. (1996). "Optimal Debt Structure and The Number of Creditors." Journal of Political Economy, pp. 1(cid:150)25. [18] Bolton, P., and X. Freixas. (2000). (cid:147)Equity, Bonds, and Bank Debt: Capital Structure and Financial Market Equilibrium under Asymmetric Information." Journal of Political Economy, 108(2), PP. 324(cid:150) 351. [19] Bolton,Patrick,andXavierFreixas(2006),(cid:147)CorporateFinanceandtheMonetaryTransmissionMechanism," Review of Financial Studies, 19, 829(cid:150)870. 46

[20] Bond,S.,Elston,J.A.,Mairesse,J.,&Mulkay,B.(2003).FinancialfactorsandinvestmentinBelgium, France, Germany, and the United Kingdom: A comparison using company panel data. Review of Economics and Statistics, 85(1), 153-165. [21] Boot, A. W., and A. V. Thakor. (2010). (cid:147)The Accelerating Integration of Banks and Markets and Its Implications for Regulation." The Oxford Handbook of Banking, pp. 58(cid:150)90. [22] Bretscher, Lorenzo, Lukas Schmid, and Andrea Vedolin. "Interest Rate Uncertainty, Hedging, and Real Activity." Hedging, and Real Activity (January 17, 2016) (2016). [23] Bris,Arturo,IvoWelch,andNingZhu."Thecostsofbankruptcy: Chapter7liquidationversusChapter 11 reorganization." The Journal of Finance 61.3 (2006): 1253-1303. [24] Caballero, R. J. (1999). Aggregate investment. Handbook of macroeconomics, 1, 813-862. [25] Calza, A., Monacelli, T., & Stracca, L. (2013). Housing (cid:133)nance and monetary policy. Journal of the European Economic Association, 11(s1), 101-122. [26] Calomiris, Charles W., and Margarita Tsoutsoura. Underwriting Costs of Seasoned Equity O⁄erings: Cross-Sectional Determinants and Technological Change, 1980(cid:150)2008. Columbia University Working Paper, 2010. [27] Campello, M., & Chen, L. (2010). Are (cid:133)nancial constraints priced? Evidence from (cid:133)rm fundamentals and stock returns. Journal of Money, Credit and Banking, 42(6), 1185-1198. [28] Campello, M., Lin, C., Ma, Y., & Zou, H. (2011). The real and (cid:133)nancial implications of corporate hedging. The Journal of Finance, 66(5), 1615-1647. [29] Campello, M., Giambona, E., Graham, J. R., & Harvey, C. R. (2011). Liquidity management and corporate investment during a (cid:133)nancial crisis. Review of Financial Studies, 24(6), 1944-1979. [30] Chava,Sudheer,andAmiyatoshPurnanandam."Determinantsofthe(cid:135)oating-to-(cid:133)xedratedebtstructure of (cid:133)rms." Journal of Financial Economics 85.3 (2007): 755-786. [31] Chava, Sudheer, and Michael R. Roberts. "How does (cid:133)nancing impact investment? The role of debt covenants." The Journal of Finance 63.5 (2008): 2085-2121. [32] Chen,Jun,andTao-HsienDollyKing."Corporatehedgingandthecostofdebt."JournalofCorporate Finance 29 (2014): 221-245. [33] Chernenko,Sergey,andMichaelFaulkender(2011),(cid:147)TheTwoSidesofDerivativesUsage: Hedgingand SpeculatingwithInterestRateSwaps,"Journal of Financial and Quantitative Analysis,46,1727(cid:150)1754 [34] Chodorow-Reich,Gabriel.2014.(cid:147)E⁄ectsofUnconventionalMonetaryPolicyonFinancialInstitutions.(cid:148) Brookings Papers on Economic Activity (Spring): 155-204 [35] Christensen, J. H., & Rudebusch, G. D. (2012). The Response of Interest Rates to US and UK Quantitative Easing*. The Economic Journal, 122(564), F385-F414. [36] Ciccarelli, Matteo, Angela Maddaloni, and Jose-Luis Peydro. "Trusting the bankers: A new look at the credit channel of monetary policy." Review of Economic Dynamics 18.4 (2015): 979-1002. [37] Colla, Paolo, Filippo Ippolito, and Kai Li. "Debt specialization." The Journal of Finance 68.5 (2013): 2117-2141. [38] Corrado, Carol, Charles Hulten, and Daniel Sichel. "Intangible capital and US economic growth." Review of income and wealth 55.3 (2009): 661-685. 47

[39] D(cid:146)Amico, S., & King, T. B.. Flow and stock e⁄ects of large-scale treasury purchases: Evidence on the importance of local supply. Journal of Financial Economics, 108(2), 425-448. [40] Daniel, Kent D., and Sheridan Titman (1997), (cid:147)Evidence on the Characteristics of Cross Sectional Variation in Stock Returns," The Journal of Finance 52, 1(cid:150)33. [41] Davis, S. J., & Kahn, J. A. (2008). "Interpreting the Great Moderation: Changes in the Volatility of Economic Activity at the Macro and Micro Levels." Journal of Economic Perspectives, 22(4), 155-80. [42] De Bondt, G., Mojon, B., & Valla, N. (2005). "Term structure and the sluggishness of retail bank interest rates in euro area countries". ECB Working Paper 451 [43] De Bondt, W. F., & Thaler, R. H. (1990). Do security analysts overreact?. The American Economic Review, 52-57. [44] Denis, D. J., and V. T. Mihov. 2003. (cid:147)The Choice Among Bank Debt, Non-Bank Private Debt, and Public Debt: Evidence from New Corporate Borrowings." Journal of Financial Economics, 70(1), pp. 3(cid:150)28. [45] Di Maggio, M., Kermani, A., & Ramcharan, R. (2014). Monetary Policy Pass-Through: Household Consumption and Voluntary Deleveraging. Columbia Business School Research Paper, (14-24). [46] Diamond, D. W. (1984). (cid:147)Financial Intermediation and Delegated Monitoring." Review of Economic Studies, 51(3), pp. 393(cid:150)414. [47] Diamond, D. W. (1991). (cid:147)Monitoring and Reputation: The Choice between Bank Loans and Directly Placed Debt." Journal of Political Economy, pp. 689(cid:150)721. [48] Du¢ e, Darrell, and Jeremy C. Stein. "Reforming LIBOR and other (cid:133)nancial market benchmarks." The Journal of Economic Perspectives 29.2 (2015): 191-212. [49] Eberly, J., Rebelo, S., & Vincent, N. (2012). What explains the lagged-investment e⁄ect?. Journal of Monetary Economics, 59(4), 370-380. [50] Eggertsson, Gauti B., and Paul Krugman. "Debt, deleveraging, and the liquidity trap: A Fisher- Minsky-Koo approach." The Quarterly Journal of Economics 127.3 (2012): 1469-1513. [51] Ehrmann, Michael, and Marcel Fratzscher (2004), "Taking Stock: Monetary Policy Transmission to Equity Markets," Journal of Money, Credit, and Banking, 36: 719(cid:150)737. [52] Ehrmann,M.,L.Gambacorta,J.Martinez-PagØs,P.Sevestre,andA.Worms,2001,"FinancialSystems and the Role of Banks in Monetary Policy Transmission in the Euro Area", ECB Working paper 105, Dec 2001 [53] English, William B., Skander J. Van den Heuvel, and Egon Zakrajsek (2014). "Interest Rate Risk and Bank Equity Valuations," Working Paper Series 14-05. Wharton Financial Institutions Center. [54] Evans,CharlesL.(1998),(cid:147)Real-TimeTaylorRulesandtheFederalFundsFuturesMarket,"Economic Perspectives, 22, 44(cid:150)55. [55] Evans, Charles, et al. "Risk management for monetary policy near the zero lower bound." Brookings Papers on Economic Activity 2015.1 (2016): 141-219. [56] Fama, E. F. (1985). (cid:147)What(cid:146)s Di⁄erent about Banks?" Journal of Monetary Economics, 15(1), 29(cid:150)39. [57] Fama, Eugene F., and Kenneth R. French (1992), (cid:147)The Cross-Section of Expected Stock Returns." Journal of Finance, 47, 427(cid:150)465. 48

[58] (cid:151)(cid:151)(cid:151) (1995), (cid:147)Size and Book-to-Market Factors in Earnings and Returns," Journal of Finance, 50, 131(cid:150)155. [59] (cid:151)(cid:151)(cid:151) (2001). "Disappearing dividends: changing (cid:133)rm characteristics or lower propensity to pay?." Journal of Financial economics 60.1 (2001): 3-43. [60] Faulkender, Michael (2005), (cid:147)Hedging or Market Timing? Selecting the Interest Rate Exposure of Corporate Debt," Journal of Finance, 60, 931(cid:150)962. [61] Faulkender, M., & Petersen, M. A. (2006). Does the source of capital a⁄ect capital structure?. Review of Financial Studies, 19(1), 45-79. [62] Fazzari, Steven M., R. Glenn Hubbard, and Bruce C. Petersen. "Investment-cash (cid:135)ow sensitivities are useful: A comment on Kaplan and Zingales." Quarterly journal of Economics (2000): 695-705. [63] Ferson, Wayne E., and Campbell R. Harvey (1999), (cid:147)Conditioning Variables and the Cross Section of Stock Returns," Journal of Finance, 54, 1325(cid:150)1360. [64] Floyd, Eric, Nan Li, and Douglas J. Skinner. "Payout policy through the (cid:133)nancial crisis: The growth of repurchases and the resilience of dividends." Journal of Financial Economics 118.2 (2015): 299-316. [65] Froot, KennethA., David S.Scharfstein, and JeremyC. Stein(1993), (cid:147)Risk Management: Coordinating Corporate Investments and Financing Policies." Journal of Finance, 5, 1629(cid:150)1658. [66] Gagnon, J., Raskin, M., Remache, J. and Sack, B. (2010).(cid:145)Large-scale asset purchases by the Federal Reserve: did they work?(cid:146), Federal Reserve Bank of New York Sta⁄Report No. 441. [67] Gagnon, Joseph, Matthew Raskin, Julie Remache, and Brian Sack. (2011). (cid:147)The Financial Market E⁄ectsoftheFederalReserve(cid:146)sLarge-ScaleAssetPurchases.(cid:148)InternationalJournalofCentralBanking 7, pp. 3(cid:150)43. [68] Gertler, Mark, and Simon Gilchrist (1994),(cid:147)Monetary Policy, Business Cycles, and the Behavior of Small Manufacturing Firms," Quarterly Journal of Economics, 109: 309(cid:150)340. [69] Gertler, Mark, and Nobuhiro Kiyotaki. "Financial intermediation and credit policy in business cycle analysis." Handbook of monetary economics 3.3 (2010): 547-599. [70] Gilchrist,Simon,andEgonZakraj(cid:154)ek."Creditspreadsandbusinesscycle(cid:135)uctuations."TheAmerican Economic Review 102.4 (2012): 1692-1720. [71] Gomes, Joao F. "Financing investment." American Economic Review (2001): 1263-1285. [72] Gomes, Joao, Urban Jermann, and Lukas Schmid. "Sticky Leverage." (2016) Working Paper. [73] Gorodnichenko,Y.,&Weber,M(2014)."AreStickyPricesCostly? EvidenceFromTheStockMarket." working paper. [74] Graham, John R., and Daniel A. Rogers (2002), (cid:147)Do Firms Hedge in Response to Tax Incentives?," Journal of Finance, 57, 815(cid:150)839. [75] Graham, John R., and Cli⁄ord W. Smith, (1999), Tax incentives to hedge, Journal of Finance 54, 2241(cid:150)2262. [76] Guerrieri, Veronica, and Guido Lorenzoni, (2016). Credit crises, precautionary savings, and the liquidity trap. Working Paper. [77] Gurkaynak,RefetS.,BrianSack,andEricT.Swanson(2005),(cid:147)DoActionsSpeakLouderthanWords? The Response of Asset Prices to Monetary Policy Actions and Statements," International Journal of Central Banking, 1, 55(cid:150)93. 49

[78] Hadlock,CharlesJ.,andJoshuaR.Pierce(2010),(cid:147)NewEvidenceonMeasuringFinancialConstraints: Moving Beyond the KZ Index," Review of Financial Studies, 23, 1909 (cid:150)1940. [79] Hamilton, James, and Jing Cynthia Wu. (2011). (cid:147)The E⁄ectiveness of Alternative Monetary Policy Tools in a Zero Lower Bound Environment.(cid:148)Journal of Money, Credit, and Banking 44, pp. 3(cid:150)46. [80] Himmelberg, C. P., & Morgan, D. P. (1995, June). Is bank lending special?. In CONFERENCE SERIES-FEDERAL RESERVE BANK OF BOSTON (Vol. 39, pp. 15-36). [81] Hollander, Stephan, and Arnt Verriest. "Bridging the gap: the design of bank loan contracts and distance." Journal of Financial Economics 119.2 (2016): 399-419. [82] Holmstrom, B., and J. Tirole. (1997). (cid:147)Financial Intermediation, Loanable Funds, and The Real Sector." Quarterly Journal of Economics, 112(3), pp. 663(cid:150)691. [83] Holod, D., & Peek, J. (2007). Asymmetric information and liquidity constraints: a new test. Journal of Banking & Finance, 31(8), 2425-2451. [84] Illes, A., &Lombardi, M.J.(2013).Interestratepass-throughsincethe(cid:133)nancialcrisis.BISQuarterly Review, September. [85] Inklaar, R., & Wang, J. C. (2013). Real output of bank services: what counts is what banks do, not what they own. Economica, 80(317), 96-117. [86] Ivashina, V.(2009).Asymmetricinformatione⁄ectsonloanspreads.Journal of Financial Economics, 92(2), 300-319. [87] Ivashina, V., & Scharfstein, D. (2010). Bank lending during the (cid:133)nancial crisis of 2008. Journal of Financial Economics, 97(3), 319-338. [88] Jensen,M.C. andW.H.Meckling(1976)."Theoryofthe(cid:133)rm: Managerialbehavior, agencycostsand ownership structure." Journal of Financial Economics, 3, pp. 305(cid:150)360 [89] Jermann, Urban J. "Negative Swap Spreads and Limited Arbitrage." Available at SSRN (2016). [90] JimØnez, Gabriel, Steven Ongena, Jose-Luis Peydr(cid:243), and Jesus Saurina (2012), (cid:147)Credit Supply and Monetary Policy: Identifying the Bank Balance-Sheet Channel with Loan Applications," American Economic Review, 102, 2301(cid:150)2326. [91] JimØnez, G., Ongena, S., Peydr(cid:243), J. L., & Saurina, J. (2014). Hazardous Times for Monetary Policy: What Do Twenty-Three Million Bank Loans Say About the E⁄ects of Monetary Policy on Credit Risk-Taking?. Econometrica, 82(2), 463-505. [92] Joyce, Michael, Ana Lasaosa, Ibrahim Stevens, and Matthew Tong. 2011. (cid:147)The Financial Market Impact of Quantitative Easing.(cid:148)International Journal of Central Banking 7(3), pp. 113(cid:150)161. [93] Kaplan, S. N., & Zingales, L. (1997). Do investment-cash (cid:135)ow sensitivities provide useful measures of (cid:133)nancing constraints?. The Quarterly Journal of Economics, 169-215. [94] Kashyap,AnilK.,OwenA.Lamont,andJeremyC.SteinCreditConditionsandtheCyclicalBehavior of Inventories The Quarterly Journal of Economics (1994) 109 (3): 565-592 [95] Kashyap, Anil K. , and Jeremy C. Stein, (cid:147)What Do a Million Observations on Banks Say About the Transmission of Monetary Policy?," American Economic Review, 90 (2000), 407(cid:150)428. [96] Kishan,RubyP.,andTimothyP.Opiela,(cid:147)BankSize,BankCapital,andtheBankLendingChannel," Journal of Money, Credit and Banking, 32 (2000), 121-141. 50

[97] Krishnamurthy,Arvind,andAnnetteVissing-Jorgensen.2011.(cid:147)TheE⁄ectsofQuantitativeEasingon Interest Rates.(cid:148)Brookings Papers on Economic Activity, Fall, pp. 215(cid:150)265. [98] Krishnamurthy, Arvind, and Annette Vissing-Jorgensen. 2012. (cid:147)The Aggregate Demand for Treasury Debt.(cid:148)Journal of Political Economy 120(2), pp. 233(cid:150)267. [99] Kuttner, Kenneth N., (cid:147)Monetary Policy Surprises and Interest Rates: Evidence from the Fed Funds Futures Market," Journal of Monetary Economics, 47 (2001), 523(cid:150)544. [100] Kwan, Simon H., and Willard T. Carleton. "Financial contracting and the choice between private placement and publicly o⁄ered bonds." Journal of Money, Credit and Banking 42.5 (2010): 907-929. [101] Lee, Gemma, and Ronald W. Masulis. "Seasoned equity o⁄erings: Quality of accounting information and expected (cid:135)otation costs." Journal of Financial Economics 92.3 (2009): 443-469. [102] Li, Canlin, and Min Wei. 2013. (cid:147)Term Structure Modeling with Supply Factors and the Federal Reserve(cid:146)s Large-Scale Asset Purchase Programs.(cid:148)International Journal of Central Banking 9(1), pp. 3(cid:150)39. [103] Lin, Chen, Yue Ma, Paul Malatesta, and Yuhai Xuan. 2012. "Corporate Ownership Structure and Bank Loan Syndicate Structure." Journal of Financial Economics, 104(1), pp. 1(cid:150)22. [104] Lintner, J. (1956). (cid:147)Distribution of Incomes of Corporations Among Dividends, Retained Earnings, and Taxes." American Economic Review, 46(2), pp. 97(cid:150)113. [105] Mian, Shehzad L., (cid:147)Evidence on Corporate Hedging Policy," Journal of Financial and Quantitative Analysis, 31 (1996), 419-439. [106] Mishkin, F. S. (1995). " Symposium on the Monetary Transmission Mechanism. The Journal of Economic Perspectives, 3-10. [107] Modigliani, F. (1971). "Monetary policy and consumption". Consumer spending and monetary policy: the linkages, 9-84. [108] Myers, S. C. (1977). Determinants of corporate borrowing. Journal of Financial Economics, 5(2), 147-175. [109] Nance, Deana R., Cli⁄ord W. Smith, and Charles W. Smithson, (cid:147)On the Determinants of Corporate Hedging," Journal of Finance 48 (1993), 267(cid:150)284. [110] Nichols, D. C., Wahlen, J. M., & Wieland, M. M. (2009). "Publicly traded versus privately held: implications for conditional conservatism in bank accounting. Review of accounting studies, 14(1), 88-122. [111] Nini, G., Smith, D. C., & Su(cid:133), A. (2012). Creditor control rights, corporate governance, and (cid:133)rm value. Review of Financial Studies, 25(6), 1713-1761. [112] Ogden,Rick,etal.CorporateBondIssuers(cid:146)SwapExposuretoRisingInterestRates.No.2016-05-26-1. Board of Governors of the Federal Reserve System (US), 2016. [113] Oliner, S. D., & Rudebusch, G. D. (1996). Monetary policy and credit conditions: evidence from the composition of external (cid:133)nance: comment. The American Economic Review, 300-309. [114] Opler, T., L. Pinkowtiz, R. Stulz, and R. Williamson. 1999. The Determinants and Implications of Corporate Cash Holdings. Journal of Financial Economics 52 (1): 3(cid:150)46. [115] Oyer, P.(1998).Fiscalyearendsandnonlinearincentivecontracts: Thee⁄ectonbusinessseasonality. Quarterly Journal of Economics, 149-185. 51

[116] Ozdagli,AliK.,2012,Financialleverage,corporateinvestment,andstockreturns.ReviewofFinancial Studies 25:1033(cid:150)69 [117] Paligorova, Teodora, and Joao AC Santos. "Rollover risk and the maturity transformation function of banks." Available at SSRN 2403418 (2014). [118] Peek, Joe & Eric S. Rosengren, 2013. "The role of banks in the transmission of monetary policy," Public Policy Discussion Paper 13-5, Federal Reserve Bank of Boston. [119] Petersen, Mitchell A., and S. Ramu Thiagarajan, 2000, Risk measurement and hedging: With and without derivatives, Financial Management 29, 5(cid:150)30. [120] Piazzesi, Monika, and Eric T. Swanson, (cid:147)Futures Prices as Risk-Adjusted Forecasts of Monetary Policy," Journal of Monetary Economics, 55 (2008), 677(cid:150)691. [121] Polk,C.,&Sapienza,P.(2009).Thestockmarketandcorporateinvestment: Atestofcateringtheory. Review of Financial Studies, 22(1), 187-217. [122] Puri,M.,Rocholl,J.,&Ste⁄en,S.(2013)."Whatkindsofbank-clientrelationshipsmatterinreducing loan defaults and why?." working paper [123] Rajan,R.G.(1992).(cid:147)InsidersandOutsiders: TheChoicebetweenInformedandArm(cid:146)s-LengthDebt." Journal of Finance, 47(4), pp. 1367(cid:150)1400. [124] Rajan, R. G., & Zingales, L. (1998). Which capitalism? Lessons form the east Asian crisis. Journal of Applied Corporate Finance, 11(3), 40-48. [125] Rampini, A. A., Su(cid:133), A., & Viswanathan, S. (2014). Dynamic risk management. Journal of Financial Economics, 111(2), 271-296. [126] Rigobon, Roberto, and Brian Sack, (cid:147)The E⁄ects of War Risk on US Financial Markets," Journal of Banking and Finance, 29 (2005), 1769(cid:150)1789. [127] Santos,J.A.,&Winton,A.(2008).Bankloans,bonds,andinformationmonopoliesacrossthebusiness cycle. The Journal of Finance, 63(3), 1315-1359. [128] Sharpe, Steve A. and Gustavo A. Suarez. "The insensitivity of investment to interest rates: Evidence from a survey of CFOs". Finance and Economics Discussion Series 2014-02 [129] Smith, Cli⁄ord W., and RenØ M. Stulz, (cid:147)The Determinants of Firms(cid:146)Hedging Policies," Journal of Financial and Quantitative Analysis, 28 (1985), 391(cid:150)405. [130] Stein, Jeremy, (cid:147)An Adverse-Selection Model of Bank Asset and Liability Management with Implications for the Transmission of Monetary Policy," RAND Journal of Economics, 29 (1998), 466-486. [131] Strebulaev,IlyaA."Dotestsofcapitalstructuretheorymeanwhattheysay?."TheJournalofFinance 62.4 (2007): 1747-1787. [132] Stulz,RenØM.,(cid:147)OptimalHedgingPolicies,"JournalofFinancialandQuantitativeAnalysis,19(1984), 127(cid:150)140. [133] (cid:151)(cid:151)(cid:151), (cid:147)Rethinking Risk Management," Journal of Applied Corporate Finance, 9 (1996), 8(cid:150)25. [134] Su(cid:133), Amir,(cid:147)Bank Lines of Credit in Corporate Finance: An Empirical Analysis," Review of Financial Studies, 22 (2009), 1057(cid:150)1088. [135] Swanson,Eric.2011.(cid:147)Let(cid:146)sTwistAgain: AHigh-FrequencyEvent-StudyAnalysisofOperationTwist and Its Implications for QE2.(cid:148)Brookings Papers on Economic Activity, Spring, pp. 151(cid:150)188. 52

[136] Tobin, J. (1969). A general equilibrium approach to monetary theory. Journal of Money, Credit and Banking, 1(1), 15-29. [137] Van den Heuvel, Skander J., (cid:147)Does Bank Capital Matter for Monetary Transmission?," Economic Policy Review, Federal Reserve Bank of New York, Issue May (2002), 259(cid:150)265. [138] Vayanos, Dimitri, and Jean-Luc Vila. 2009. (cid:147)A Preferred-Habitat Model of the Term Structure of Interest Rates.(cid:148)NBER Working Paper 15487. [139] Vickery, J. (2008). How and why do small (cid:133)rms manage interest rate risk?. Journal of Financial Economics, 87(2), 446-470. [140] Whited, T. M. (1992). Debt, liquidity constraints, and corporate investment: Evidence from panel data. The Journal of Finance, 47(4), 1425-1460. [141] Whited, T. M., & Wu, G. (2006). Financial constraints risk. Review of Financial Studies, 19(2), 531- 559. [142] Wright, J. H. (2012), What does Monetary Policy do to Long-term Interest Rates at the Zero Lower Bound?. The Economic Journal, 122: F447(cid:150)F466 53

APPENDIX A - Sensitivity of Stock Returns and Investment to Monetary Policy in the Simple Model We are interested in how the reaction of the market value of equity and of investment @lnV @lnK 0 1 to monetary policy, and respectively, change with the share of (cid:135)oating rate @lnR @lnR 1 1 debt, l, and the role (cid:133)nancial constraints play.65 A (cid:133)rm(cid:146)s t = 0 value is given by the present discounted value of dividends, d d 1 2 V = + ; (25) 0 R R R 1 1 2 and this expression can be further simpli(cid:133)ed because the (cid:133)rm will optimally set d = 0, 1 because investors in (cid:133)nancially constrained (cid:133)rms will value funds inside the (cid:133)rm more than they do outside the (cid:133)rm. Investors of unconstrained (cid:133)rms are indi⁄erent between receiving a positive dividend d > 0 in t = 1, or having the (cid:133)rm save that amount at rate r and paying 1 2 it out in t = 2, so we can also assume for them that d = 0 too. To isolate our mechanism 1 of interest, we assume we are in the steady state, in which r = r = r . (26) 2 c 1 The value of the (cid:133)rm is thus f (K ) R (1 l)K R lK R b 1 c 0 2 0 2 1 V = (cid:0) (cid:0) (cid:0) (cid:0) ; (27) 0 R R 1 2 and after applying (26) and simplifying we get f (K ) R (K +b ) 1 2 0 1 V = (cid:0) ; (28) 0 R R 1 2 subject to equations (2)(cid:150)(4). The reaction to a monetary policy shock of an unconstrained (cid:133)rm in t = 1 (i.e. for whom constraint (3) is not binding) is given by @lnV lK 0 0 = 1 ; (29) @lnR (cid:0) (cid:0) V 1 0 wherewehaveappliedtheenvelopetheorem. The(cid:133)rsttermcapturesthestrongerdiscounting offuturepro(cid:133)ts, andthesecondtermcapturesthee⁄ectofanincreaseinthecash(cid:135)owclaims 65Note that dlnR =dr . 1 (cid:24) 1 1

of debtholders. This sensitivity changes with the usage of (cid:135)oating rate debt according to: @2lnV K 0 0 = ; (30) @lnR @l (cid:0)V 1 0 which simply captures the increase in the e⁄ect of a raise in interest rates on the transfer of cash-(cid:135)ow rights to debtholders when (cid:135)oating rate debt is larger. We now turn to the case of a (cid:133)nancially constrained (cid:133)rm. For simplicity, we assume, (cid:22) without loss of generality, they are not able to borrow at all, so that b = b = 0. The value 1 of the (cid:133)rm is then f (K ) R K 1 2 0 V = (cid:0) : (31) 0 R R 1 2 In this case: @lnV lK f (K ) 0 0 0 1 = 1 (32) @lnR (cid:0) (cid:0) V R 1 0 2 where, as with the unconstrained, the (cid:133)rst term captures the stronger discounting of future pro(cid:133)ts. The additional factor in the second term, f (K )=R , captures the e⁄ect of an 0 1 2 increase in the interest expense in t = 1 on investment in t = 1. Therefore, the second term is di⁄erent for constrained (cid:133)rms because they invest at a marginal rate of investment which is potentially di⁄erent from R . As long as f (K ) > R , which is the case because 2 0 1 2 the (cid:133)rm is now assumed to be (cid:133)nancially constrained, then this e⁄ect is stronger than for unconstrained (cid:133)rms. As a result, when there is some (cid:135)oating rate debt, a change in interest rates has a stronger impact on a (cid:133)rm(cid:146)s equity value if this (cid:133)rm is constrained. This sensitivity changes according to the usage of (cid:135)oating rate debt according to: @2lnV K f (K ) 0 0 1 = ; (33) @lnR @l (cid:0)V R 1 0 2 where we can observe how an increase in (cid:135)oating rate debt exposes the (cid:133)rm to a loss in value which is stronger the higher the marginal product of capital in the second period. This marginal product of capital is higher the more (cid:133)nancially constrained the (cid:133)rm is. This ensures, as a comparison of (30) and (33) makes clear, that the e⁄ect of (cid:135)oating rate debt usage on the sensitivity of stock prices to monetary policy is always greater for (cid:133)nancially constrained (cid:133)rms. Intuitively, each additional dollar that a constrained (cid:133)rm needs to pay to (cid:135)oating rate debtholders has a value which is a multiple of f 0 (K1) times more for them, R2 compared to unconstrained (cid:133)rms. Note that the constrained (cid:133)rm(cid:146)s problem becomes the same as unconstrained (cid:133)rm when the constraint is not binding (i.e., f (K ) = R ). 0 1 2 The reaction of investment is more straightforward. In particular, we have @lnK K 1 0 = l ; (34) @lnR (cid:0) K 1 1 2

if the (cid:133)rm is (cid:133)nancially constrained, and @lnK 1 = 0; (35) @lnR 1 if the (cid:133)rm is (cid:133)nancially unconstrained. Equation (34) follows immediately from the fact that for the constrained (cid:133)rm (cid:22) (cid:22) K = N +b = f (K ) r (1 l)K r lK +b; (36) 1 1 0 c 0 1 0 (cid:0) (cid:0) (cid:0) and equation (35) follows from the fact that the investment of the unconstrained (cid:133)rm is determined by the neoclassical investment rule f (K ) = R . 0 2 2 For r = r = r , we have that for a (cid:133)nancially constrained (cid:133)rm 2 c 1 @lnK K 1 0 = ; (37) @lnR @l (cid:0)K 1 1 while for an unconstrained (cid:133)rm this derivative is 0. Therefore, the e⁄ect of monetary policy on investment is stronger for (cid:133)rms with high (cid:135)oating rate debt, l, especially if they are (cid:133)nancially constrained, i.e. have low K . 1 B - Simulated Regressions of the Dynamic Model B.1. De(cid:133)nition of Variables The de(cid:133)nitions of the variables used in the dynamic model of Section 2.2 are standard in the investment and capital structure literatures, and are given in Table A14. [TABLE A14 ABOUT HERE] B.2. Robustness of Results Table A15 displays regression results using simulated data from our dynamic model in Section 2.2 in which we use (cid:135)oating rate debt over total assets as our proxy for exposure to (cid:135)oating rate debt, as opposed to (cid:135)oating rate debt over total debt as in our main analysis. All our results go through, although statistical and economic signi(cid:133)cance is slightly weaker. [TABLE A15 ABOUT HERE] C - Subsample Robustness of the E⁄ect of Monetary Policy on Stock Prices WestartwiththereactionoftheaggregateCRSPvalue-weightedindexbetweenFebruary 3

1994 and June 2008. Following Rigobon and Sack (2005), we focus on this period for two reasons. First, starting in February 1994, the FOMC(cid:146)s policy of announcing target rate changes at pre-scheduled dates virtually eliminated the timing ambiguity associated with rate changes prior to this date. Second, after June 2008, the Federal Reserve switched from announcing a speci(cid:133)c target rate to announcing a range for the target rate. Table A3 o⁄ers a comparison between the responses of equity prices to federal funds rate changes in di⁄erent samples, after discarding outliers as in Bernanke and Kuttner (2005). Columns 1 and 2 show that on the day of an FOMC announcement, a 100bp surprise increase in the federal funds rate decreases stock prices by around 300bp when we look either at the valueweighted returns or the individual returns of the entire CRSP universe between 1994 and 2008. This result is comparable to the numbers reported in Bernanke and Kuttner (2005). Columns 3 and 4 show that the reaction of equity prices to surprise changes in monetary policy is stronger in the sample of 2003(cid:150)2008 than in previous years. However, the sign and signi(cid:133)cance of the coe¢ cient of surprise is the same for both samples. A comparison of columns 4 and 5 reveals that (cid:133)rms in the sample for which we have bank debt usage data have a reaction to monetary policy shocks very similar to that of the overall CRSP universe during the 2003(cid:150)2008 period. [TABLE A3 ABOUT HERE] D- Examples from 10-K (cid:133)les on hedging activities The following two paragraphs are examples of the type of discussion on hedging activities that we (cid:133)nd in the 10-K (cid:133)les. In (cid:133)scal year 2008, BioFuel Energy Corp reports as follows: We are subject to interest rate risk in connection with our bank facility. Under the facility, our bank borrowings bear interest at a (cid:135)oating rate based, at our option, on LIBOR or an alternate base rate. (...). In September 2007, the Operating Company, through its subsidiaries, entered into an interest rate swap for a two-year period. The contract is for $60.0 million principal with a (cid:133)xed interest rate of 4.65%, payable by the Operating Company and the variable interest rate, the one-month LIBOR, payable by the third party. Similarly, in (cid:133)scal year 2006 Netsmart Technologies reports: In October 2005, we entered into a revolving credit and term loan agreement with the Bank of America (...). This (cid:133)nancing provides us with a (cid:133)ve-year term loan of $2.5 million. The term loan bears interest at LIBOR plus 2.25%. We 4

have entered into an interest rate swap agreement with the Bank for the amount outstanding under the term loan whereby we converted our variable rate on the termloantoa(cid:133)xedrateof7.1%inordertoreducetheinterestrateriskassociated with these borrowings. E- Monetary Policy Surprise Calculation Procedure Following Bernanke and Kuttner(cid:146)s analysis, we de(cid:133)ne an event as either an FOMC meeting or an announced change in the funds target rate. Kuttner (2001) and Bernanke and Kuttner (2005) obtain the corresponding surprise change in the target rate by (cid:133)rst calculating the change in the rate implied by the corresponding futures contract, given by 100 minus the futures contract price, and then scaling this result by a factor associated with the number of days of the month in which the event occurred because the payo⁄ of the contract is determined by the average realized federal funds e⁄ective rate during the month. Accordingly, the unexpected target rate change, for an event taking place on day d of month m, is given by D (cid:1)iu = (f0 f0 ); D d m;d (cid:0) m;d 1 (cid:0) (cid:0) where f0 f0 is the change in the current-month implied futures rate, and D is the m;d (cid:0) m;d 1 (cid:0) number of days in the month. To suppress the end-of-month noise in the federal funds rate, the unscaled change in the implied futures rate is used as the measure of target rate surprise when the event occurs on the last three days of a month. If the event happens on the (cid:133)rst day of a month, f1 is used instead of f0 . The expected federal funds rate change is m 1;D m;d 1 (cid:0) (cid:0) de(cid:133)ned as the di⁄erence between the actual change minus the surprise: (cid:1)ie = (cid:1)i (cid:1)iu; (cid:0) where (cid:1)i is the actual federal funds rate change. The data for the decomposition of the federal funds target rate changes can be obtained from Kenneth Kuttner(cid:146)s webpage.66 F - The E⁄ect of Bank Debt Usage on Monetary Policy Sensitivity of Stock Prices Table A4 presents the results of regression (22) using alternative speci(cid:133)cations that become more restrictive across columns. The (cid:133)rst column of Table A4 contains the result from a basic random-e⁄ects panel regression with no controls and suggests that a one standard deviation (0:114) increase in our bank debt usage measure causes the stock price to increase 66http://econ.williams.edu/people/knk1/research 5

1:6 (= 14 0:114) percentage points more in response to a 1 percentage point surprise de- (cid:0) (cid:3) crease in the federal funds rate. To put this e⁄ect in perspective, the same surprise decrease in the federal funds rate causes the stock price of the (cid:133)rm with the average amount of bank debt over assets (7:22%) to increase about 4:97 percent on average. [TABLE A4 ABOUT HERE] Columns 2a-2c provide the same regression with di⁄erent adjustments for well-known risk factors in asset pricing. Column 2a repeats the same regression as column 1 with the excess market return as an additional control variable and shows that the coe¢ cient of Surprise (BankDebt=At) remains very similar. Column 2b repeats the regression in (cid:3) column 1 after replacing the dependent variable with the stock return in excess of that predicted by the Capital Asset Pricing Model (CAPM), which corrects for the correlation of individual stock returns with the aggregate market return. The coe¢ cient of interest ( 10:5) is within one standard deviation of the coe¢ cient in column 1 ( 14:1). However, (cid:0) (cid:0) the coe¢ cient of Surprise changes signi(cid:133)cantly and actually turns positive. This does not mean that a positive (contractionary) rate surprise causes an increase in the stock price of the average (cid:133)rm. We observe this pattern because part of the market return on the FOMC date can be attributed to monetary policy so the CAPM correction of returns prevents us from measuring the full e⁄ect of monetary policy. The same pattern can also be observed when we do the risk adjustment with the Fama-French three-factor model instead of CAPM, asshownincolumn2c. Tocircumventthisproblem, wecontinuewiththeunadjustedreturns as in column 1 and use the CAPM beta as a control variable for the correlation of individual returnswiththemarketreturninotherregressionsofTableA4. Forthesamereason, instead of correcting returns using the Fama-French 3-factor model we use the (cid:133)rm characteristics underlying the 3-factor model (size, market-to-book, CAPM beta) as control variables.67 Another advantage of this approach stems from the observation that (cid:133)rm characteristics subsume the e⁄ect of the Fama-French risk factors in explaining stock returns, as discussed in Daniel and Titman (1997) and Ferson and Harvey (1999). In order to address potential identi(cid:133)cation issues, such as non-spherical disturbances and omitted variables, we progressively add controls, industry (cid:133)xed e⁄ects, both interacted anduninteracted, standarderrorsclusteredattheevent-industrylevel, and(cid:133)rm(cid:133)xede⁄ects. Non-sphericalitywouldprimarilya⁄ectthestandarderrorsofourestimatesratherthantheir 67For size, we use book value of assets, rather than market value of equity for two reasons. First, the market equity size premium has declined signi(cid:133)cantly in the last three decades. Second, since Gertler and Gilchrist (1994), balance sheet (rather than market) size has been widely used as a proxy of (cid:133)rms(cid:146)(cid:133)nancial constraints,whichisconsideredanimportantfactorbehindthetransmissionofmonetarypolicyandtherefore more suitable for our purpose. 6

consistency,whichisthemainreasonwhyweuseclusterederrors. However,omittedvariables can in(cid:135)uence our inference by a⁄ecting both the standard errors and the consistency of our estimates. Therefore, controls and (cid:133)rm-level (cid:133)xed e⁄ects speci(cid:133)cations aim at distinguishing between bank debt being special, or bank debt users being special, for reasons that are not captured in our basic regression in column 1 of Table A4. We also include an instrumental variable analysis. More speci(cid:133)cally, column 3 introduces (cid:133)rm controls and year (cid:133)xed e⁄ects and shows that the coe¢ cient in column 1 remains e⁄ectively unchanged. This is also robust to alternative speci(cid:133)cations, as shown in columns 4 to 8. In column 4, industry (cid:133)xed-e⁄ects enter the regressionbothinteractedwithsurpriseanduninteracted,withindustriesclassi(cid:133)edaccording to Fama-French 48 sectors available from Kenneth French(cid:146)s website, and errors are clustered at the event-industry level to address possible time-and-cross-section heteroskedasticity in the errors. Column 5 extends the de(cid:133)nition of bank debt to include undrawn credit lines. Another concern stems from the possibility that bank debt usage is caused by cash (cid:135)ow risk, (cid:133)nancial constraints or the interest rate sensitivity of (cid:133)rms(cid:146)demand and our results re(cid:135)ect the importance of these factors rather than bank debt usage, so in column 6 we introduce measures that control for these factors. We follow Faulkender (2005) and measure the interest rate sensitivity of (cid:133)rms(cid:146)operating pro(cid:133)ts as the correlation between quarterly (cid:133)rmearningsbeforeinterest, tax, depreciationandamortization(EBITDA)andthree-month average LIBOR rates. We introduce cash (cid:135)ow volatility and CAPM beta as controls for (cid:133)rm risk, and cash holdings as a measure of the ability of the (cid:133)rm to withstand liquidity shocks associated with monetary policy. Finally, to capture (cid:133)nancing constraints, we follow Hadlock and Pierce (2010), who show that (cid:133)rm size and age are very useful predictors of the severity of (cid:133)nancial constraints, and introduce a measure based solely on these two (cid:133)rm characteristics.68 We call this measure of (cid:133)nancial constraints the HP index. To further deal with this concern, Column 7 replaces industry (cid:133)xed e⁄ects with (cid:133)rm (cid:133)xed e⁄ects to capture unobserved time invariant omitted (cid:133)rm characteristics. Overall the coe¢ cient of bank debt usage barely changes, which adds robustness to the evidence that bank debt usage makes (cid:133)rms more responsive to monetary policy shocks. Column 8 includes an instrumental variable regression with (cid:133)xed e⁄ects to deal further with the potential endogeneity problem of bank debt usage. Following Faulkender and 68We use Hadlock and Pierce(cid:146)s estimates for 2000-2004 from Table 6 in their paper because this period is closer to our sample. The HP index is calculated as (-0.548*Size) + (0.025*Size^2) (cid:150)(0.031*Age), where size is the log of in(cid:135)ation adjusted (to 2004) book assets, and age is the number of years for which the (cid:133)rm has stock returns in CRSP to ensure that we have few missing observations. An alternative measure can be derived using the IPO date, but this leads to multiple missing observations, though the results are qualitativelysimilar. Incalculatingthisindex,sizeisreplacedwithlog($4.5billion)andagewiththirty-seven years if the actual values exceed these thresholds. 7

Petersen (2006) and Santos and Winton (2008) we instrument for bank debt usage using proxies for (cid:133)rmvisibility and (cid:133)rmuniqueness. These authors use these proxies to instrument for(cid:133)rmaccesstocorporatebondmarketsandarguethat(cid:133)rmsthatarehighlyvisibleandless uniquearemorelikelytoberated. Asproxiesforvisibility, theyintroduce(cid:133)rms(cid:146)membership oftheS&P500indexoroftheNewYorkStockExchange(NYSE).Asaproxyforuniqueness, they use a measure of the number of (cid:133)rms in their same industry other than itself that have credit ratings. We instead use them as instruments for bank debt usage following the argument that (cid:133)rms that do not have access to bond markets are likely to be stronger users of bank debt. Our last instrument relies on the observation that banking regulation limits the amount of unsecured loans a bank can issue (Ivashina (2009)). Consistent with this observation, Altman, Gande, and Saunders (2010) (cid:133)nd that about 70 percent of the bank loans to corporations are secured, in contrast to 3 percent of bonds. Therefore, we would expect that the collateral of a (cid:133)rm is an important determinant of how much the (cid:133)rm can borrow from a bank, and hence use tangibility as an additional instrument.69 We (cid:133)nd that while we lose statistical signi(cid:133)cance, the coe¢ cients of the regression are e⁄ectively unchanged and the Hausman test cannot reject the hypothesis that they are the same, suggesting that endogeneity is not a big concern. Finally, column 9 provides the same speci(cid:133)cation as in column 7 but replaces bank debt with (cid:135)oating rate debt. While the statistical signi(cid:133)cance of (cid:135)oating rate debt drops somewhat, as expected due to the points mentioned in the data section, it is similar in size to the coe¢ cient of bank debt in column 7 and still statistically signi(cid:133)cant. A possible concern in Table A4 is that bank debt may be proxying for the use of shorttermdebt. Thisconcern(cid:133)ndssupportinthedescriptivestatisticsreportedinTableIVwhich shows that bank debt users have a higher percentage of short-term debt than nonbank-debt users (3:71 percent versus 1:09 percent, calculated as a share of total assets) and that the Pearson pair-wise correlation between these two variables is 0:27. For example, to the extent that changes in monetary policy a⁄ect primarily the short end of the yield curve, one can expect (cid:133)rms with a shorter average maturity of debt to be more sensitive to increases in interest rates. To test this hypothesis, we rewrite the speci(cid:133)cation provided in equation (22) in terms of short-term debt divided by the book value of assets, STDebt=At. Formally, the complete 69Because we are interested in how bank debt usage a⁄ects (cid:133)rms(cid:146)sensitivity to monetary policy shocks and because this term in our regression is non-linear, given by Surprise (BankDebt=At) , we cannot t (cid:3) i;t 1 usethetraditionalinstrumentalvariableapproachwherethe(cid:133)rststageestimatesof(BankDeb(cid:0)t=At) are i;t 1 used to replace our endogenous variable in the second stage, which is dubbed "the forbidden regressio(cid:0)n" in the literature. Therefore, we use an alternative approach suggested in Angrist and Pischke (2009, Ch. 4) where we use Surprise (Instrument) as an instrument for Surprise (BankDebt=At) : t (cid:3) i;t 1 t (cid:3) i;t 1 (cid:0) (cid:0) 8

regression speci(cid:133)cation is: Ret = (cid:12) +(cid:12) Surprise +(cid:12) (BankDebt=At) +(cid:12) (STDebt=At) t 0 1 t 2 t 1 3 t 1 (cid:0) (cid:0) +(cid:12) Surprise (BankDebt=At) +(cid:12) Surprise (STDebt=At) 4 t (cid:3) t 1 5 t (cid:3) t 1 (cid:0) (cid:0) +(cid:13)Controls +(cid:21)Surprise Controls +" : (38) t 1 t t 1 t (cid:0) (cid:3) (cid:0) Table A11 provides the empirical results of this test. Columns 1 and 2 show the results of a version of regression (38) in which the terms containing (BankDebt=At) are removed. t 1 (cid:0) We observe that the amount of short-term debt in a (cid:133)rm(cid:146)s balance sheet is not signi(cid:133)cantly associated with the strength of the sensitivity to surprises in the federal funds rate. The coe¢ cient in column 1 ( 10:30) is not only statistically insigni(cid:133)cant but also economically (cid:0) very small: one standard deviation in short-term debt usage leads to only a (10:30 0:05 =) (cid:3) 0:5 percentage point increase in the sensitivity to monetary policy of stock prices, a far lower (cid:133)gure than the 1:6 percentage points for bank debt usage. Introducing additional controls in column2 makes this e⁄ect evensmaller. Columns 3and4provide acomplete speci(cid:133)cationof (38), including bank debt, both interacted and not interacted with Surprise. The coe¢ cient (cid:12) remains insigni(cid:133)cant, while the coe¢ cient (cid:12) retains the sign, size, and signi(cid:133)cance of the 5 4 speci(cid:133)cations reported in Table A4. We conclude that the higher sensitivity of bank debt users to federal funds rate surprises is not due to their higher exposure to short-term debt. Finally, one potential concern is that our bank debt usage variable, BankDebt=At, might be correlated with other leverage variables and our results re(cid:135)ect the importance of debt in general rather than bank debt in particular. One way we address this concern is by adding book leverage directly in Table A4. As another way to address the same issue, we normalize total bank debt with total debt to create an alternative measure of bank debt usage, BankDebt=Debt. We use this new measure to replace our original measure and repeat the regressions in Table A4. The new results, presented in Table A12 in the appendix con(cid:133)rm our results from Table A4. In particular, we (cid:133)nd that a one standard deviation in BankDebt=Debt (0:40) leads to approximately a (0:40 3 =) 1:2 percentage point increase (cid:3) in the responsiveness of stock prices to monetary policy surprises, which is in the ballpark of our previous result (1:6 percentage points) using BankDebt=At.70 Hence, we continue to use BankDebt=At in the following analysis. G - E⁄ect of Hedging: Controlling for Financial Frictions 70Asbefore,the(cid:135)oatingratedebtcoe¢ cientissomewhatsmallerduetomismeasurement. Asimilarissue also appears when we include undrawn credit lines in column 4 because there are some (cid:133)rms in our sample that have very little debt but very large undrawn credit lines. 9

One concern in our previous regressions is that the estimates might be biased due to an omitted variable bias associated with the relationship between (cid:133)nancial constraints and (cid:133)rms(cid:146)hedging behavior. Some theories predict that hedging activities are positively related to the severity of (cid:133)nancing constraints. If external (cid:133)nance is costly, (cid:133)rms may (cid:133)nd it optimal tohedgeagainstlowcash(cid:135)owrealizationstoavoidhavingtoforgopositivenet-present-value (NPV) projects (Froot, Scharfstein, and Stein (1993)) or to avoid nonlinear costs of (cid:133)nancial distress (Stulz (1984)).71 The empirical evidence, however, does not provide support for this prediction, and has documented that (cid:133)rms that are more likely to face (cid:133)nancial constraints, such as small (cid:133)rms, are less likely to manage risk (Stulz (1996)).72 Motivated by these (cid:133)ndings, Rampini, Su(cid:133), and Viswanathan (2014) introduce and test a theory that suggests there is a trade-o⁄between hedging and (cid:133)nancing, because both activities compete for the same collateral. In equilibrium, (cid:133)rms that are more (cid:133)nancially constrained hedge less. The importantroleof(cid:133)nancingconstraintsin(cid:133)rms(cid:146)willingnessandabilitytohedgesuggeststhat one should control for (cid:133)nancial constraints and how these constraints interact with both the ability to raise debt and to hedge.73 To deal with this concern, we study how much of the e⁄ect of hedging survives after controlling for (cid:133)nancial frictions, which we measure using (cid:133)rm age and the HP index. In particular, we estimate an expanded version of regression (23) 71Othermotivationsfortheuseofhedginghavetodowithcorporategovernanceandmanagerialincentives (Chava and Purnanandam (2007)), and with market timing (Faulkender (2005)). More generally, the value creation of hedging has been examined by Nance, Smith, and Smithson (1993), Mian (1996) and Graham and Rogers (2002). 72Columns5and6ofTableA2showthatthisisthecasealsoinoursample. Larger,rated,morepro(cid:133)table and less (cid:133)nancially constrained (low HP index) (cid:133)rms are more likely to hedge than their smaller, unrated, less pro(cid:133)table or more (cid:133)nancially constrained counterparts. 73Oneimportantquestioniswhymostbanklendingarrangementsinvolvea(cid:135)oatingrateinsteadofa(cid:133)xed rate despite the fact that many (cid:133)rms hedge the interest rate risk associated with these loans. One answer could arise from the trade-o⁄between (cid:133)rms(cid:146)needs and banks(cid:146)cost of capital. A (cid:133)rm that wants to borrow at a (cid:133)xed rate may have limited access to other (cid:133)xed-rate sources of (cid:133)nancing, such as bonds, whereas the bank might prefer to lend at (cid:135)oating rates, in which case hedging bridges the gap between the desire of the bank and the (cid:133)rm. As discussed by Vickery (2008), there are at least two reasons why banks might prefer to lend at (cid:135)oating rates. First, rising interest rates can cause deposit out(cid:135)ows from the banks and it is costly for banks to replace these out(cid:135)ows with other sources of (cid:133)nancing. Lending at a (cid:135)oating rate would provide apartial hedge againsttheseout(cid:135)ows. Second, (cid:135)oating ratebusinessloans can beused to hedgethe maturity mismatch between deposits and long-term mortgage loans. 10

Ret = (cid:12) +(cid:12) Surprise +(cid:12) Surprise (BankDebt=At ) t 0 1 t 2 t t 1 (cid:3) (cid:0) +(cid:12) Surprise (BankDebt=At) Hedge 3 t (cid:3) t 1 (cid:3) t (cid:0) +(cid:12) Surprise (BankDebt=At) FinFrictions 4 t (cid:3) t (cid:0) 1 (cid:3) t (cid:0) 1 +(cid:21) Surprise Controls Hedge 1 t t 1 t (cid:3) (cid:0) (cid:3) +(cid:21) Surprise Controls FinFrictions 2 t t 1 t 1 (cid:3) (cid:0) (cid:3) (cid:0) +Uninteracted terms and Second Order Interactions+" (39) t and check if (cid:12) , the coe¢ cient of Surprise (BankDebt=At) Hedge , remains similar to 3 t (cid:3) t 1(cid:3) t (cid:0) what we have calculated before in Table Vand whether it still eliminates the full e⁄ect of the Surprise (BankDebt=At) , i.e., (cid:12) +(cid:12) = 0. Columns 1 and 2 of Table A13 show that t (cid:3) t 1 3 2 (cid:0) the estimates of (cid:12) have a magnitude very similar to the one we have calculated in Table V 3 for the coe¢ cient on the triple interaction term Surprise (BankDebt=At) Hedge and t (cid:3) t 1(cid:3) t (cid:0) that it nulli(cid:133)es the e⁄ect of Surprise (BankDebt=At ): Similar evidence is obtained in t t 1 (cid:3) (cid:0) columns 3 and 4 when we use (cid:135)oating rate debt instead of bank debt. These results suggest that the (cid:135)oating rate channel is a unique and distinct channel that works separately from possible e⁄ects of (cid:133)nancial frictions. H - Description of Tests based on Balance Sheet Variables In this Appendix, we describe in detail the empirical tests discussed in Section 5. Interest Coverage Ratio Tests We compute the interest coverage ratio at the quarterly level as the sum of interest expenses (Compustat item, XINTQ) and cash (cid:135)ow divided by interest expenses, where cash (cid:135)ow is equal to earnings before extraordinary items (IBQ) plus depreciation (DPQ). We test whether a higher bank debt usage as a share of total assets increases the responsiveness of (cid:133)rms(cid:146)interest rate coverage ratios following monetary policy actions, due to the higher likelihood of this debt being (cid:135)oating rate. We use the following empirical speci(cid:133)cation: (cid:1)CoverageRatio = (cid:12) +(cid:12) C\hange t 1;t+x 0 1 t (cid:0) +(cid:12) (BankDebt=At) +(cid:12) C\hange (BankDebt=At) 2 t 1 3 t t 1 (cid:0) (cid:0) +(cid:13)Controls +(cid:21)C\hange (Controls )+" ; (40) t 1 t t 1 t (cid:0) (cid:0) where C\hange is the cumulative quarterly change in the interest rate, as in Ashcraft and t Campello (2007) and JimØnez, Ongena, Peydr(cid:243), and Saurina (2012, 2014), instead of the 11

cumulative surprise component because cash (cid:135)ow and the interest rate expense on existing debt is not forward-looking the way stock prices are.74 Our (cid:133)rm-level controls include book leverage, (cid:133)rm size, market-to-book ratio, pro(cid:133)tability, interest rate sensitivity, and shortterm debt. We include (cid:133)rm and year-quarter (cid:133)xed e⁄ects, and we cluster errors at the industry-quarter level.75 The dependent variable ((cid:1)CoverageRatio ) is calculated as the change between t 1;t+x (cid:0) the coverage ratio in the quarter before the monetary policy shock and x 1;2;3;4;5;6 2 f g quarters ahead. The timing of e⁄ects of the (cid:135)oating rate channel is in(cid:135)uenced by the frequency with which interest rates of (cid:135)oating rate bank loans are reset to adjust to movements in the reference rate. Because this frequency can range from 1 day to 1 year (Inklaar and Wang (2013)), the e⁄ects might occur with a lag of several quarters, although a majority of commercial and industrial (C&I) loans have a resetting frequency of one month or less, according to the Federal Reserve(cid:146)s Survey of Terms of Business Lending. As in Sections 4.2 and 4.4, we restrict our sample to include (cid:133)rms that have outstanding variableratedebtequivalenttoatleast1%oftotalassetstoeliminate(cid:133)rmsthatmaybeusing interest-rate derivatives for speculative purposes, and we run speci(cid:133)cation (40) separately for subsamples of hedgers and non-hedgers. Cash Holdings Tests We compute the change in cash holdings as the di⁄erence between total cash and shortterm investments at the end of quarter t+x (where x 1;2;3;4;5;6 ) and at the end of 2 f g quartert 1,scaledbytotalassetsattheendofthequartert 1. Changesincashholdingsare (cid:0) (cid:0) expressed in basis points, and we use the same regression speci(cid:133)cation as (40). Firm controls are taken from the empirical literature that focuses on corporate cash accumulation (Bates, Kahle and Stulz (2009)), and include (cid:133)rm size, leverage, market-to-book ratio, and cash (cid:135)ow risk. We test whether the impact of bank debt usage on the sensitivity of cash holdings to monetary policy (coe¢ cient (cid:12) ) is signi(cid:133)cantly stronger for (cid:133)nancially constrained (cid:133)rms 3 than for unconstrained (cid:133)rms in the sample of unhedged (cid:133)rms, and whether this di⁄erence is absent or is at least signi(cid:133)cantly smaller in the sample of hedged (cid:133)rms. We classify (cid:133)rms as (cid:133)nancially constrained (unconstrained) if their value of the Hadlock and Pierce (2010) (HP) index is above (below) the median, and report the di⁄erence between the estimates of (cid:12) across constrained and unconstrained (cid:133)rms, within each of the subsamples of hedgers 3 74This argument also holds for other balance sheet variables in this section, as they are more likely to respond to the anticipated component of monetary policy changes as well because adjustment costs prevent themfromreactingrapidlytochangesinexpectationsaboutfuturepolicyrates,particularlywhenthechange intheexpectationhappensshortlybeforetheFOMCannouncement. Still, ourresultsinthissectionremain qualitatively similar when we use the sum of the monetary policy surprises on the FOMC announcements dates in a given year, as in Gorodnichenko and Weber (2014). 75Year-quarter (cid:133)xed e⁄ects also control for possible seasonality occurring at the quarterly frequency. 12

and non-hedgers, and also report the statistical signi(cid:133)cance of the di⁄erence.76 Results are analyzed for horizons of four and six quarters. Inventory Investment Tests We follow Kashyap, Lamont and Stein (1994) and adopt their empirical speci(cid:133)cation for our inventory investment regressions, which we augment to introduce monetary policy changes, bank debt usage, and our (cid:133)rm level controls: Inventories ln t+x = (cid:12) +(cid:12) C\hange Inventories 0 1 t (cid:18) t 1(cid:19) (cid:0) +(cid:12) (BankDebt=At) +(cid:12) C\hange (BankDebt=At) 2 t 1 3 t t 1 (cid:0) (cid:0) +(cid:13)Controls +(cid:21)C\hange Controls t 1 t t 1 (cid:0) (cid:0) Sales Inventories t;t+x t 1 +ln +ln( (cid:0) )+" t : (41) Sales Sales (cid:18) t x 1;t 1(cid:19) t 1 (cid:0) (cid:0) (cid:0) (cid:0) Our (cid:133)rm-level controls include, as before, book leverage, (cid:133)rm size, market-to-book ratio, pro(cid:133)tability, interest rate sensitivity, and short-term debt. We add, following Kashyap, Lamont and Stein (1994), the cash to total assets ratio at the end of quarter t 1, separately (cid:0) and interacted with change, and also the di⁄erence between the log of total sales during quarters t to t+x and the log of total sales during quarters t x 1 to t 1, and the log of (cid:0) (cid:0) (cid:0) the inventory to sales ratio at the end of quarter t 1. We express our dependent variable (cid:0) in basis points. We include (cid:133)rm and year-quarter (cid:133)xed e⁄ects, and we cluster errors at the industry-quarter level. Results are analyzed for horizons of four and six quarters. Sales Tests Weinterpretsales,inlinewithexistingliterature,asaproxyfor(cid:133)rm-leveloutput(Gertler and Gilchrist (1994), Bond, Elston, Mairesse, and Mulkay (2003)). We employ empirical speci(cid:133)cation (40) and use the di⁄erence between the log of total sales during quarters t to t+x and the log of total sales during quarters t x 1 to t 1 as our dependent variable. (cid:0) (cid:0) (cid:0) As before, we introduce year-quarter (cid:133)xed e⁄ects, which also control for possible seasonality occurring at the quarterly frequency. Fixed Investment Tests To test ourpredictionon(cid:133)xedinvestment, we expandourbaseline empirical speci(cid:133)cation (40)toincludethemainfactorsthathavebeenidenti(cid:133)edintheempiricalliteraturetomatter for (cid:133)rm investment (Eberly, Rebelo and Vincent (2012)). These are Tobin(cid:146)s Q, proxied by 76As explained in Section 4.4, we choose the HP measure because Hadlock and Pierce (2010) show that other indices have very little power to predict (cid:133)nancial constraints, and any power they have comes from (cid:133)rm size and age, the two variables they use to create their composite HP index. 13

the market-to-book ratio, cash (cid:135)ow, and lagged investment. We run the following regression: K ln t+x = (cid:11) +(cid:11) C\hange +(cid:11) (BankDebt=At) +(cid:11) C\hange (BankDebt=At) K 0 1 t 2 t 1 3 t t 1 (cid:18) t 1(cid:19) (cid:0) (cid:0) (cid:0) +(cid:21)(FirmControls) +(cid:13)C\hange (FirmControls) t 1 t t 1 (cid:0) (cid:0) CF I t t 1 +(cid:11) 4 Q t +(cid:11) 5 +(cid:11) 6 (cid:0) +" t ; (42) K K (cid:18) t (cid:19) (cid:18) t 1(cid:19) (cid:0) where our dependent variable is computed as the di⁄erence between the log of total (cid:133)xed capital (measured as property, plant and equipment) x quarters ahead, and the log of capital one quarter before the monetary policy event. We express our dependent variable in basis points. Our(cid:133)rm-levelcontrolsalsoincludebookleverage, (cid:133)rmsize, pro(cid:133)tability, interestrate sensitivity, and short-term debt, all interacted with C\hange and also introduced separately. t We include (cid:133)rm and year-quarter (cid:133)xed e⁄ects, and we cluster errors at the industry-quarter level. I - Bank Lending Channel vs Floating Rate Channel AccordingtotheestimationsinHolodandPeek(2007),aonepercentagepointpermanent rate hike would be associated with a 0.01 (for publicly-held banks) to 0.1 percentage point (for privately-held banks) decrease in C&I loans as a fraction of total bank assets over a four quarter period. Moreover, C&I loans are about 10 percent of total assets for both public and private banks, and public banks hold about 80 percent of total assets (Nichols, Wahlen, and Wieland (2009)). Therefore, the weighted average e⁄ect is a 0.3 percent decrease in C&I loans over a four quarter period ((0.01/0.1)*0.8+(0.1/0.1)*0.2=0.3). This result is also in line with Oliner and Rudebusch (1996): over eight quarters they (cid:133)nd a 0.6 percent decrease in C&I loans to small (cid:133)rms who hold half the bank loans whereas no loan decline is observed for large (cid:133)rms, implying a 0.3 percent decrease for total loans. Moreover, since this e⁄ect is the same as the e⁄ect from Holod and Peek (2007) over the (cid:133)rst year, we can conclude that this is the long-run e⁄ect of a permanent increase in the federal funds rate. Overall, this suggests that a (cid:133)rm that usually borrows $100 from the bank will experience a cumulative $0.3 shortfall if the federal funds rate increases by one percentage point. This e⁄ect includes any ampli(cid:133)cation mechanism that works through bank balance sheets. Of course, this e⁄ect might be an upper bound for the true supply e⁄ect because part of the e⁄ect can be attributed to the potential decline in loan demand due to reduced demand for goods and services associated with the tightening of monetary policy. Moreover, Oliner and Rudebusch (1996) (cid:133)nd that loans to large (cid:133)rms actually increase after a tightening, suggesting a reallocation of credit from small (cid:133)rms to large (cid:133)rms that further alleviates the 14

e⁄ects of the bank lending channel. How does this compare to the (cid:135)oating rate channel? A one percentage point rate hike would increase the interest expense by $1 on a $100 loan. However, some loans are (cid:133)xed rate and some (cid:135)oating rate loans are hedged, which we will take into account. Since the papers about the bank lending channel in the previous paragraph focus on C&I loans, we will do the same here. As shown in Vickery (2008), about 70 percent of total bank loans in the Federal Reserve(cid:146)s survey of business lending is made at a (cid:135)oating rate which is close to what Figure2suggests(about75percent)inoursample. Ifwecontinuewiththesameassumption that half of C&I loans are made to small businesses this would imply that about 65 percent of C&I loans to small businesses are at a (cid:135)oating rate (0.65*0.5+0.75*0.5=0.7). Moreover, we (cid:133)nd that (cid:133)rms that do not use hedging derivatives hold about 30 percent of the loans in our sample, where almost all (cid:133)rms are large according to the de(cid:133)nition of the Quarterly Financial Report for Manufacturing, Mining, and Trade Corporations (QFR) used by Oliner andRudebusch(1996),whichhasacuto⁄of$25millionintotalassetstoqualify(cid:133)rmsaslarge in2014. Smallandmediumsized(cid:133)rmsusuallymakelittleuseofhedgingderivatives(Vickery (2008)). Therefore, as an initial approximation, we assume that no small (cid:133)rm hedges. As a second approximation, we will assume that small (cid:133)rms have the same hedging behavior as the (cid:133)rms with less than 25 million dollars in assets in our sample (37 percent in terms of total loan size) which is an upper bound for hedging derivative usage because these (cid:133)rms are on the upper tail of the size distribution. Using these numbers, we calculate the total percentage of C&I loans that are (cid:135)oating rate and unhedged as a number between 0.32 and 0.44 (0.65*(1-0.37)*0.5+0.75*0.3*0.5=0.32 and 0.65*1*0.5+0.75*0.3*0.5=0.44). Therefore, the average cash shortfall after the one percentage point rate hike would be between $0.32 and $0.44 on a $100 loan over four quarters, or $0.64 to 0.88 over two years, assuming the e⁄ect on the interest expense persists over this time horizon, which seems to be the case according to unreported regressions in which we study the e⁄ect of bank debt usage and hedging on the response of the interest expense of (cid:133)rms to monetary policy.77 77One implicit assumption is that the rate on the C&I loans is reset frequently. According to the Survey of Terms of Business Lending of November 2006, 27 percent of the C&I loans are subject to repricing at any time, 28percentoftheloanshavedailyrepricing, and23percentoftheloanshavearepricingperiodof2-30 days. 15

TABLES AND FIGURES

Figure 1 The relation between the Federal Funds Target Rate and Floating-Rate Debt Reference Rates This figure displays the relation between two of the most common reference rates used in floating rate loans, the 3- Month London Interbank Offered Rate (LIBOR) and the Bank Prime Loan Rate, and the Federal Funds Target Rate, from January 1986 until January 2015. The data is from the Federal Reserve Bank of St. Louis FRED Economic Data. Figure 2 The relation between bank debt and floating-rate debt This figure displays the relation between bank debt and floating-rate debt as a percentage of a firm’s total debt. Firms are grouped in the horizontal axis according to bank debt as a percentage of total debt. The vertical axis shows the corresponding percentages of floating-rate debt as a percentage of total debt.

Figure 3 Cumulative Reaction to monetary policy tightening associated with Bank Debt Usage: Hedgers vs Non-hedgers This figure displays the average additional effect of a 1 percentage point surprise increase in the Federal Funds target rate on the cumulative stock price return of a hypothetical firm that is financed exclusively with bank debt, relative to a firm with no bank debt. In the bottom (top) panel the sample consists of firms that hedge (do not hedge) interest rate risk. The estimates are a result of running regression (1) with the cumulative stock return over multiple trading days as the dependent variable. Dotted lines capture the 95% confidence interval around our estimates.

Figure 4 Timeline of Events in the Simple Model

Table I Dynamic Model Calibration – Parameter Value Choices

Table II The Effect of Monetary Policy – Simulated Data Panel A of this table examines how monetary policy affects firms’ stock returns and how this effect varies with floating rate debt usage and financing constraints. Not displayed in this table are some of the lower order combinations of the triple interaction variable (Constrained*Surprise *Floating Rate/ Total Debt), the constant term, and controls. Column 3 displays a regression in which the costs of distress are 33% lower than in the benchmark calibration. Panel B of this table examines how monetary policy affects firms’ investment and how this effect varies with floating rate debt usage and financing constraints. Not displayed in this table are some of the lower order combinations of the triple interaction variable (Constrained*Change *Floating Rate/ Total Debt), the constant term, and Tobin’s Q. Surprise, Change, Stock Return and Net Investment/K are expressed in basis points. The definitions of all the variables used are described in detail in Appendix B. Parentheses contain t-statistics. The asterisks denote *** for p<0.01, ** for p<0.05, * for p<0.1. Panel A. Effect on Stock Returns Dep variable: Stock Return t (1) (2) (3) Surprise -19.55*** -15.08*** -15.04*** (-54.41) (-46.17) (-76.62) Surprise *Floating Rate/ Total Debt -21.58*** -23.12*** -15.99*** (-37.14) (-44.21) (-43.17) Constrained*Surprise *Floating Rate/ Total Debt -7.25*** -4.23*** -3.54*** (-6.83) (-4.43) (-5.22) Controls NO YES NO Distress Costs Benchmark Benchmark Low R-squared 0.19 0.24 0.24 Observations 38,000 38,000 38,000 Panel B. Effect on Investment Dep variable: Net Investment /K t, t+x t-1 4 Quarters Ahead 6 Quarters Ahead (1) (2) (3) (4) Change -0.19*** -0.47*** -0.46*** -0.64*** (-6.30) (-14.72) (-15.02) (-20.07) Change *Floating Rate/ Total Debt 0.02 -0.24*** 0.19*** 0.02 (0.47) (-4.71) (3.87) (0.34) Constrained*Change *Floating Rate/ -0.12 -0.22** -0.42*** -0.49*** Total Debt (-1.40) (-2.46) (-4.71) (-5.42) Controls NO YES NO YES R-squared 0.01 0.04 0.01 0.03 Observations 38,000 38,000 38,000 38,000

Table III The Effect of Monetary Policy on Financial Distress – Simulated Data This table examines how monetary policy affects firms’ financial distress likelihood (column (1)), costs of financial distress (column (2)), and the interest rate coverage ratio (column (3)), and how these effects vary with floating rate debt usage and financing constraints. Not displayed in this table are some of the lower order combinations of the triple interaction variable (Constrained*Surprise *Floating Rate/ Total Debt), the constant term, and controls. All dependent variables and Change are expressed in basis points. The definitions of all the variables used are described in detail in Appendix B. Parentheses contain tstatistics. The asterisks denote *** for p<0.01, ** for p<0.05, * for p<0.1. Dep variable: Covenant Costs of Interest Rate Violation Financial Coverage Ratio Likelihood Distress (1) (2) (3) Change 1.48*** 0.08 0.18 (5.18) (1.55) (0.72) Change *Floating Rate/ Total Debt 0.21 0.64*** 13.36*** (0.674) (6.39) (29.79) Constrained*Change *Floating Rate/ Total Debt 2.38*** 1.56*** 2.62*** (2.58) (8.48) (3.19) R-squared 0.19 0.24 0.16 Observations 38,000 38,000 38,000

Table IV Descriptive Statistics This table provides summary statistics for the entire sample and for different subsamples. The entire sample consists of U.S. firms covered by Capital IQ, CRSP, and Compustat from 2003 to 2008 with December fiscal year-end, excluding utilities (SIC codes 4900- 4949) and financials (SIC codes 6000-6999). We remove firm-year observations with negative revenues, missing information on total assets, or a value of total assets under 10 million. We also discard penny stocks, defined as those with a price of less than $5. After the above filters, the sample contains 9,746 firm-year observations comprising 2,368 unique firms. Complete variable definitions are given in the appendix. All variables are winsorized at the 1% level in both tails of the distribution, and total assets are expressed in terms of year-2000 dollars. See Table A1 for variable definitions. (1) (2) (3) (4) Leveraged Firms with Bank Debt Entire Sample Leveraged Firms w/out Hedgers Nonhedgers Bank Debt Mean SD Mean SD Mean SD Mean SD Term Loans/At 3.95% 9.02% 0.00% 0.00% 8.93% 12.70% 5.09% 8.88% Drawn Credit Lines/At 3.09% 6.55% 0.00% 0.00% 6.12% 8.51% 5.06% 7.36% Bank Debt /At 7.22% 11.66% 0.00% 0.00% 15.52% 14.33% 10.33% 11.01% Bank Debt / Total Debt 37.51% 40.07% 0.00% 0.00% 50.35% 36.26% 58.89% 37.97% Floating-Rate Debt/At 9.77% 13.44% 1.59% 5.86% 15.37% 15.77% 10.04% 11.73% Float-Rate Debt / Tot. Debt 38.31% 40.47% 8.95% 24.65% 47.04% 38.14% 50.62% 41.26% Undrawn Credit Line/At 9.85% 10.56% 8.19% 9.89% 12.99% 9.61% 11.11% 10.95% (Bank Debt + Und CL)/At 17.14% 16.64% 8.19% 9.89% 28.51% 16.89% 21.43% 15.36% Short-Term Debt /At 2.55% 5.15% 1.85% 4.25% 3.65% 6.13% 3.71% 5.76% Profitability 4.94% 15.73% 4.35% 16.46% 8.91% 7.24% 4.31% 16.40% Size (Total Assets, Million $) 4,274.32 23990 5,404.67 20,292.93 5,071.905 23,784.68 4,677.73 32,800.7 Book Leverage 28.15% 29.58% 26.87% 29.38% 45.19% 27.21% 31.07% 27.71% Earnings-Interest Rate Sensitivity -13.23% 35.46% -11.82% 36.96% -15.63% 33.43% -12.98% 34.44% Rated Dummy 32.98% 47.02% 36.23% 48.08% 57% 49.52% 28.76% 45.27% Market-to-Book Assets 1.98 1.57 2.13 1.61 1.42 0.92 1.79 1.38 Cash/At 22.35% 24.26% 27.19% 23.31% 7.44% 9.83% 17.44% 22.07% CAPM Beta (Monthly) 1.32 1.20 1.37 1.23 1.11 0.93 1.35 1.24 Cash Flow Volatility 1.11% 0.49% 1.14% 0.44% 0.95% 0.42% 1.10% 0.48% Hadlock-Pierce Fin. Con. Measure -2.85 0.59 -2.89 0.61 -3.12 0.49 -2.82 0.57 Age 16.78 17.01 18.08 18.95 20.20 19.31 16.63 15.72 Hedging Dummy 34.80% 47.63% 26.46% 44.12% 100.00% 0.00% 0.00% 0.00% Hedging*(Bank Debt /At) 4.22% 10.18% 0.00% 0.00% 15.52% 14.33% 0.00% 0.00% Hedging*( Floating-Rate Debt/At) 5.75% 12.11% 0.71% 3.94% 15.37% 15.77% 0.00% 0.00% Firm’s lenders’ characteristics: Bank Size (ln(assets)) 19.80 1.92 19.65 2.05 Tier 1 Capital Ratio 8.39% 0.65% 8.46% 0.82% Deposit Ratio 51.03% 7.84% 53.90% 9.55% Liquidity Ratio 22.47% 4.17% 21.32% 5.50% Observations (annual) 9,746 9,746 2,509 2,509 2,463 2,463 2,647 2.647 Quarterly data Interest Rate Coverage Ratio 0.14 0.19 0.12 0.18 0.22 0.20 0.14 0.18 Inventory (quarterly growth %) 2.02% 21.70% 2.37% 23.10% 1.71% 17.96% 1.73% 19.03% Sales (quarterly growth %) 2.16% 21.1% 2.40% 21.19% 1.51% 17.48% 2.04% 21.67% Prop. Pla. & Equip. (q. growth %) 2.13% 5.42% 2.15% 5.50% 1.80% 5.12% 2.05% 5.09% Observations (quarterly) 45,694 45,694 11,932 11,932 10,117 10,117 11,645 11,645

Table V The Role of Bank Debt Usage and Interest Rate Risk Exposure in the Transmission of Monetary Policy This table examines how firms’ bank and floating rate debt usage impacts the effect of monetary policy on stock prices, and how this impact varies with firms’ hedging activity. Hedgers are defined on a yearly basis as those firms that report having swapped their interest rate from floating to fixed in their 10-K annual reports. Only firms with floating rate debt constituting more than 1% of total assets are included. Bank Debt/At is defined as bank debt (term loans plus drawn revolving credit) over the book value of assets (At). FloatingRateDebt /At is defined as floating rate debt over the book value of assets (At). All regressions also include an unreported constant term, as well as ln(assets), book leverage, profitability, market-to-book, interest rate sensitivity, and their interaction with surprise. All firm characteristics are lagged by one year and winsorized at 1%. Parentheses contain t-statistics. The asterisks denote *** for p<0.01, ** for p<0.05, * for p<0.1. (1) (2) (3) (4) (5) (6) (7) (8) ALL Non- Hedgers Non- Hedgers ALL Non- Hedgers Non- Hedgers Panel A - Main Variables Hedgers Hedgers Hedgers Hedgers Surprise -6.04** -4.10*** -8.62*** -5.08* -6.83** -6.20*** -4.59*** -8.22*** -5.76** -6.34** (-2.54) (-4.07) (-9.05) (-1.91) (-2.35) (-2.63) (-4.76) (-8.96) (-2.20) (-2.16) BankDebt/At 1.28** 0.35 0.26 0.13 1.94*** (2.38) (0.86) (1.14) (0.13) (3.12) FloatingRateDebt /At 1.03** 0.25 0.21 0.77 1.19** (2.06) (0.65) (0.96) (0.84) (2.14) Surprise *(BankDebt/At) -13.50* -25.18*** 1.41 -38.02*** 3.45 (-1.76) (-3.04) (0.26) (-3.09) (0.38) Surprise *(FloatingRateDebt /At) -14.70* -20.81*** -2.72 -30.79** -3.71 (-1.92) (-2.60) (-0.53) (-2.36) (-0.40) Surprise*(BankDebt/At)*Hedging 26.71*** 41.46*** (2.71) (2.85) Surprise*(FloatingRateDebt /At)*Hedging 17.78* 27.07* (1.90) (1.74) (continues on the next page)

(ctd from previous page) (1) (2) (3) (4) (5) (6) (7) (8) ALL Non- Hedgers Non- Hedgers ALL Non- Hedgers Non- Hedgers Panel B - Surprise*Other variables Hedgers Hedgers Hedgers Hedgers Surprise *log(Assets) -1.32** -1.95*** -0.55 -1.30** -1.62** -0.77 (-2.34) (-2.69) (-0.77) (-2.35) (-2.21) (-1.11) Surprise *Book Leverage -0.32 5.85 -6.20 0.64 5.73 -4.14 (-0.09) (1.18) (-1.36) (0.18) (1.13) (-0.91) Surprise *Market-to-Book 0.81 -0.22 2.18 0.85 -0.12 2.19 (0.80) (-0.17) (1.60) (0.83) (-0.10) (1.61) Surprise *Profitability -18.28 -16.22 -22.98 -18.04 -16.65 -23.07 (-1.63) (-1.34) (-1.06) (-1.62) (-1.39) (-1.06) Surprise *Int. Rate Sensitivity -7.63 -9.21 -5.61 -7.76 -9.77* -5.84 (-1.50) (-1.58) (-0.89) (-1.54) (-1.69) (-0.92) Test for the coefficients of Surprise*(Other variables)*Hedging are jointly zero (p-value) 0.19 0.39 Firm Controls YES NO NO YES YES YES NO NO YES YES Firm FE YES NO NO YES YES YES NO NO YES YES Surprise*Firm Controls YES NO NO YES YES YES NO NO YES YES Cluster (Fed event*IndustryFF48) YES NO NO YES YES YES NO NO YES YES Observations 24,123 11,796 12,335 11,788 12,335 24,123 11,796 12,335 11,788 12,335

Table VI The Role of Interest Rate Risk Exposure in the Transmission of Monetary Policy: Instrumental Variables Analysis All variables are as defined in Table IV. Column (1) is the fixed effects regression for the sample of firms that have data on lagged hedging dummy and column (3) is the fixed effects regression for the sample of firms that have data on tax convexity, our instrument for hedging from Graham and Smith (1999) and Campello, Lin, Ma, Zou (2011). Column (2) uses lagged hedging dummy as instrument for hedging. Column (4) uses the variables underlying the convexity measure, excluding Vol, whereas column (5) uses all variables as instruments. Column (6) uses the tax convexity measure directly, as given in the text. Column (7) uses both the lagged hedging dummy and the tax convexity measure. Only firms with floating rate debt constituting more than 1% of total assets are included. A constant, non-interacted terms, and the policy surprise interacted with firm size, book leverage, profitability and the market-to-book ratio are included but not reported. All firm characteristics are lagged by one year and winsorized at 1%. Parentheses contain t-statistics. The asterisks denote *** for p<0.01, ** for p<0.05, * for p<0.1. The coefficient of Surprise in columns (4) to (6) are not negative because the linear first stage regression does not restrict hedging between zero and one. Panel A: Bank Debt (1) (2) (3) (4) (5) (6) (7) OLS IV OLS IV1 IV2 IV3 IV L.hedging L.hedging Convexity Convexity Convexity Convexity Both Surprise -5.13*** -3.99*** -5.78*** 1.83 1.21 64.25 -5.40*** (-4.62) (-3.19) (-3.33) (0.25) (0.21) (1.07) (-2.60) Surprise*(BankDebt/At) -38.17*** -30.19** -49.43*** -102.30** -121.91*** -153.31 -46.47*** (-3.90) (-2.47) (-3.73) (-2.33) (-2.95) (-0.33) (-2.66) Surprise*(BankDebt/At)*Hedging 43.50*** 34.95** 59.00*** 173.15*** 183.41*** 501.27 59.60** (3.54) (2.13) (3.53) (2.65) (3.02) (0.82) (2.56) Hausman test (p-value) 0.934 0.786 0.925 0.998 1.000 Firm FE YES YES YES YES YES YES YES Firm Controls YES YES YES YES YES YES YES Surprise*Firm Controls YES YES YES YES YES YES YES Observations 23,413 23,413 12,009 12,009 12,009 12,009 11,665 Panel B: Floating Rate Debt (1) (2) (3) (4) (5) (6) (7) OLS IV OLS IV1 IV2 IV3 IV L.hedging L.hedging Convexity Convexity Convexity Convexity Both Surprise -5.82*** -4.47*** -6.74*** 2.13 0.88 41.97 -6.17*** (-5.38) (-3.66) (-3.93) (0.31) (0.15) (0.87) (-3.02) Surprise*(FloatingRateDebt/At) -28.79*** -25.17** -25.45* -74.93* -88.14** -243.60 -30.58* (-2.96) (-2.06) (-1.95) (-1.74) (-2.22) (-1.55) (-1.83) Surprise*(FloatingRateDebt/At)*Hedging 26.36** 24.32 22.45 136.09** 128.79** 422.53 33.77 (2.18) (1.50) (1.37) (2.03) (2.16) (0.78) (1.52) Hausman test (p-value) 0.882 0.812 0.984 0.956 0.999 Firm FE YES YES YES YES YES YES YES Firm Controls YES YES YES YES YES YES YES Surprise*Firm Controls YES YES YES YES YES YES YES Observations 23,413 23,413 12,009 12,009 12,009 12,009 11,665

Table VII Interest Rate Risk Exposure and the Transmission of Monetary Policy: The Role of Financing Constraints Hedgers are defined on a yearly basis as those firms that report having hedged their interest rate risk from floating to fixed, in their 10-K annual reports. Financial constraints are proxied by the firm’s age and the Hadlock and Pierce (2010) measure given by HP = -0.548*Size+0.025*Size2-0.031*Age. Firm size is defined to be the log of assets (inflation-adjusted to 2004). Age is defined as the current year minus the first year that the firm has a non-missing stock price in CRSP. Firm size and age are at the 1% tails on the low end, and at the $4.5 billion and 37- year points on the high end. The financial constraint measure takes the value of 1 if the firm’s age is below the median or if the firm’s HP statistic is above the median in a given year. Only firms with floating rate debt constituting more than 1% of total assets are included. Bank Debt/At is defined as bank debt (term loans plus drawn revolving credit) over the book value of assets (At). All regressions also include an unreported constant term, as well as ln(assets), book leverage, profitability, market-to-book value, both interacted with surprise and un-interacted. All firm and lender characteristics are lagged by one year and winsorized at 1%. Parentheses contain t-statistics. The asterisks denote p values: *** for p<0.01, ** for p<0.05, and * for p<0.1. (1) (2) (3) (4) (5) (6) (7) (8) Non- Non- Hedgers Hedgers Non- Non- Hedgers Hedgers Hedgers Hedgers Hedgers Hedgers VARIABLES OLD YOUNG OLD YOUNG LOW HP HIGH HP LOW HP HIGH HP Surprise -6.23*** -3.05 -6.33** -7.03*** -1.18 -6.31** -5.90** -9.46*** (-3.73) (-1.48) (-2.52) (-2.74) (-0.52) (-2.40) (-2.29) (-3.05) Surprise*(BankDebt/At) -20.30 -56.73*** 3.81 3.20 -29.19** -46.11*** 4.05 7.01 (-1.49) (-3.49) (0.37) (0.29) (-1.96) (-3.06) (0.46) (0.51) Surprise*(BankDebt/At)*Constrained -36.43* -0.61 -16.92 2.96 (-1.74) (-0.04) (-0.79) (0.19) Firm FE YES YES YES YES YES YES YES YES Firm Controls YES YES YES YES YES YES YES YES Surprise*Firm Controls YES YES YES YES YES YES YES YES Observations 6,713 5,075 7,303 5,032 5,785 6,003 8,561 3,774 R-squared 0.01 0.01 0.01 0.02 0.01 0.01 0.02 0.02 Number of gvkey 432 409 407 337 354 486 469 288

Table VIII The Effect of Monetary Policy on the Interest Coverage Ratio This table examines how monetary policy affects firms’ interest coverage ratio and how this effect varies with bank debt usage and interest rate risk hedging. The quarterly coverage ratio is equal to interest expenses (XINTQ) divided by the sum of interest expenses and cash flow. Cash flow is equal to earnings before extraordinary items (IBQ) plus depreciation (DPQ). The dependent variable is computed as the difference between the coverage ratio x quarters after the monetary policy shock and the coverage ratio during the quarter before the monetary policy shock, where x={4,6}. Change is the sum of all changes in the federal funds rate that occur during a quarter. Hedgers are defined as those firms that report having hedged their interest rate risk from floating to fixed in their 10K annual reports. Only firms with floating rate debt constituting more than 1% of total assets are included. Bank Debt/At is defined as bank debt (term loans plus drawn revolving credit) over the book value of assets (At). All regressions also include an unreported constant term. Unreported controls include ln(assets), book leverage, market-to-book, profitability and interest rate sensitivity of operating income. All firm controls are lagged by one quarter and winsorized at 1%. Parentheses contain t-statistics. The asterisks denote *** for p<0.01, ** for p<0.05, * for p<0.1. Dep variable: CoverageRatio - CoverageRatio t+x t-1 (1) (2) (3) (4) (5) (6) x=1 quarter x=2 quarters x=3 quarters x=4 quarters x=5 quarters x=6 quarters ahead ahead ahead ahead ahead ahead Non-hedgers (Sum) Change* BankDebt/At -0.11 3.56 6.04* 4.69 8.72** 7.88 (-0.04) (1.00) (1.71) (1.46) (2.28) (1.14) Hedgers -3.05 -0.18 1.82 -1.06 -0.33 -3.89 (Sum) Change* BankDebt/At (-0.71) (-0.08) (0.54) (-0.29) (-0.15) (-1.15) Hedger*(Sum) Change* BankDebt/At -2.93 -3.74 -4.21 -5.74 -9.05** -11.77** (-0.72) (-0.87) (-0.76) (-1.01) (-1.98) (-2.08) Firm Controls YES YES YES YES YES YES Firm FE YES YES YES YES YES YES Change*Firm Controls YES YES YES YES YES YES Year-quarter dummies YES YES YES YES YES YES Industry-Quarter Clustering YES YES YES YES YES YES Observations (non-hedgers regressions) 7,669 7,511 7,332 7,193 7,076 6,963 Observations (hedgers regressions) 7,445 7,351 7,238 7,134 7,036 6,941

Table IX The Effect of Monetary Policy on Cash Holdings This table examines how monetary policy affects firm cash holdings and how this effect varies with bank debt usage and interest rate risk hedging. Cash holdings are calculated as cash and short-term investments divided by total assets. Change in Cash Holdings is computed as the difference (in basis points) between t-1,t+x the cash holdings x quarters ahead, and cash holdings at the end of the quarter before the monetary policy change occurs, scaled by total assets at the end of the quarter before the monetary policy change occurs. Change is the sum of all changes in the federal funds rate that occur during a quarter. Hedgers are defined as those firms that report having hedged their interest rate risk from floating to fixed in their 10K annual reports. Only firms with floating rate debt constituting more than 1% of total assets are included. Bank Debt/At is defined as bank debt (term loans plus drawn revolving credit) over the book value of assets (At). All regressions also include an unreported constant term. Unreported controls include ln(assets), book leverage, market-to-book, profitability and interest rate sensitivity of operating income. All firm controls are lagged by one quarter and winsorized at 1%. Parentheses contain t-statistics. The asterisks denote *** for p<0.01, ** for p<0.05, * for p<0.1. Dependent variable (in basis points): 10,000*((Casht+x- Casht-1)/Assetst-1) x=4 quarters ahead x=6 quarters ahead Non-hedgers Hedgers Non-hedgers Hedgers Constrained Unconstrained Constrained Unconstrained Constrained Unconstrained Constrained Unconstrained (high HP) (low HP) (high HP) (low HP) (high HP) (low HP) (high HP) (low HP) BankDebt/At 700.68 140.01 152.40 221.87 420.47 496.87 358.07* 241.59 (1.56) (0.38) (0.82) (1.39) (0.81) (1.36) (1.74) (1.48) (Sum) Change *BankDebt/At -3.22 0.99 -0.53 -0.20 -7.06** 2.39 -1.00 1.00 (-1.43) (0.86) (-0.67) (-0.22) (-2.37) (1.54) (-0.57) (1.06) (Sum) Change* -4.21* -0.33 -9.45*** -1.99 BankDebt/At*Constrained (-1.74) (-0.33) (-3.15) (-1.08) Firm Controls YES YES YES YES YES YES YES YES Firm FE YES YES YES YES YES YES YES YES Change*Firm Controls YES YES YES YES YES YES YES YES Year-quarter dummies YES YES YES YES YES YES YES YES Industry-Quarter Clustering YES YES YES YES YES YES YES YES Observations 3,812 3,764 2,032 5,204 3,663 3,667 1,934 5,075

Table X The Effect of Monetary Policy on Inventory Investment This table examines how monetary policy affects firm inventory investment and how this effect varies with bank debt usage and interest rate risk hedging. Inventories are calculated as Total Inventories (INVTQ), and Change in Inventories is computed as the difference (in basis points) between the log of t-1,t+x inventories x quarters ahead and the log of inventories at the end of the quarter before the monetary policy change occurs. Change is the sum of all changes in the federal funds rate that occur during a quarter. Hedgers are defined as those firms that report having hedged their interest rate risk from floating to fixed in their 10K annual reports. Only firms with floating rate debt constituting more than 1% of total assets are included. Bank Debt/At is defined as bank debt (term loans plus drawn revolving credit) over the book value of assets (At). All regressions also include an unreported constant term. Controls include the inventory to sales ratio, the change in cumulative sales over the x quarters following the monetary policy change and the x quarters before, and cash holdings over assets, and also (unreported): ln(assets), book leverage, market-to-book, profitability and interest rate sensitivity of operating income. All firm controls are lagged by one quarter and winsorized at 1%. Parentheses contain t-statistics. The asterisks denote *** for p<0.01, ** for p<0.05, * for p<0.1. Dependent variable (in basis points): 10,000*(ln(Inventoryt+x) - ln(Inventoryt-1)) x=4 quarters ahead x=6 quarters ahead Non-hedgers Hedgers Non-hedgers Hedgers Constrained Unconstrained Constrained Unconstrained Constrained Unconstrained Constrained Unconstrained (high HP) (low HP) (high HP) (low HP) (high HP) (low HP) (high HP) (low HP) BankDebt/At -55.43 -1,082.20 672.17 1,547.13 1,204.72 -2,817.44 -473.70 1,068.42 (-0.03) (-0.69) (0.61) (1.40) (0.60) (-1.60) (-0.38) (0.84) (Sum) Change -16.81** 7.59 7.78 -6.06 -21.20*** 0.99 5.31 -2.44 *BankDebt/At (-2.05) (0.57) (1.64) (-1.62) (-2.83) (0.09) (1.07) (-0.49) (Sum) Change* -24.39** 13.83** -22.18* 7.74 BankDebt/At*Constrained (-2.21) (2.35) (-1.72) (1.39) ln(Inventoryt-1/ Salest-1) -3,144.56*** -3,824.72*** -4,857.47*** -602.85 -3,626.21*** -4,301.13*** -6,462.38*** -1,388.76 (-9.75) (-7.95) (-7.24) (-0.80) (-10.78) (-7.70) (-9.68) (-1.53) ln(Salest-1,t+x) 0.51*** 0.52*** 0.65*** 0.65*** 0.54*** 0.60*** 0.87*** 0.82*** (10.04) (9.41) (5.45) (7.34) (11.63) (12.14) (10.30) (15.00) Casht-1/Att-1 6,150.26*** 6,864.56*** 18,180.29*** 7,091.18*** 6,494.47*** 7,590.74*** 15,304.41*** 7,736.34*** (4.48) (5.51) (6.06) (3.92) (4.78) (5.36) (4.85) (3.79) Firm Controls YES YES YES YES YES YES YES YES Firm FE YES YES YES YES YES YES YES YES Change*Firm Controls YES YES YES YES YES YES YES YES Year-quarter dummies YES YES YES YES YES YES YES YES Industry-Quarter Clustering YES YES YES YES YES YES YES YES Observations 2,964 3,171 1,448 4,243 2,863 3,082 1,371 4,130

Table XI The Effect of Monetary Policy on Sales This table examines how monetary policy affects firm sales and how this effect varies with bank debt usage and interest rate risk hedging. Change in Sales t-x-1,t+x is calculated as the difference between the ln of the accumulated quarterly sales over x quarters starting in the quarter (t) in which the monetary policy action occurs, and the ln of the accumulated quarterly sales in the x quarters preceding the monetary policy action. Change is the sum of all changes in the federal funds rate that occur during a quarter. Hedgers are defined as those firms that report having hedged their interest rate risk from floating to fixed in their 10K annual reports. Only firms with floating rate debt constituting more than 1% of total assets are included. Bank Debt/At is defined as bank debt (term loans plus drawn revolving credit) over the book value of assets (At). All regressions also include an unreported constant term. Unreported controls include ln(assets), book leverage, market-to-book, profitability and interest rate sensitivity of operating income. All firm controls are lagged by one quarter and winsorized at 1%. Parentheses contain t-statistics. The asterisks denote *** for p<0.01, ** for p<0.05, * for p<0.1. Dependent variable (in basis points): 10,000*(ln(Sales ) - ln(Sales )) t, t+x t-x-1, t-1 x=4 quarters ahead x=6 quarters ahead Non-hedgers Hedgers Non-hedgers Hedgers Constrained Unconstrained Constrained Unconstrained Constrained Unconstrained Constrained Unconstrained (high HP) (low HP) (high HP) (low HP) (high HP) (low HP) (high HP) (low HP) BankDebt/At -2,353.22*** -1,066.41 319.36 -395.46 -2,671.31*** -392.00 938.17** -516.16 (-2.86) (-1.32) (0.77) (-0.77) (-2.95) (-0.43) (2.03) (-0.95) (Sum) Change -6.43 12.71* -5.12** 4.83** -6.29 16.89** -5.51*** 6.31** *BankDebt/At (-1.59) (1.79) (-2.47) (2.14) (-1.60) (2.23) (-2.87) (2.39) (Sum) Change* -19.13*** -9.95*** -23.18*** -11.82*** BankDebt/At*Constrained (-3.71) (-3.27) (-3.59) (-3.84) Firm Controls YES YES YES YES YES YES YES YES Firm FE YES YES YES YES YES YES YES YES Change*Firm Controls YES YES YES YES YES YES YES YES Year-quarter dummies YES YES YES YES YES YES YES YES Industry-Quarter YES YES YES YES YES YES YES YES Clustering Observations 3,813 3,770 2,037 5,207 3,664 3,671 1,940 5,078

Table XII The Effect of Monetary Policy on Fixed Investment This table examines how monetary policy affects firm fixed investment and how this effect varies with bank debt usage and interest rate risk hedging. Inventories are calculated as Total Inventories (INVTQ), and Fixed Investment is computed as the difference (in basis points) between the log of property, plant and t-1,t+x equipment (PPEGTQ) x quarters ahead and the log of PPEGTQ at the end of the quarter before the monetary policy change occurs. Change is the sum of all changes in the federal funds rate that occur during a quarter. Hedgers are defined as those firms that report having hedged their interest rate risk from floating to fixed in their 10K annual reports. Only firms with floating rate debt constituting more than 1% of total assets are included. Bank Debt/At is defined as bank debt (term loans plus drawn revolving credit) over the book value of assets (At). All regressions also include an unreported constant term. Controls include the lagged investment to capital ratio, the lagged cash holdings to capital ratio, and the market to book ratio, and also (unreported): ln(assets), book leverage, market-tobook, profitability and interest rate sensitivity of operating income. All firm controls are lagged by one quarter and winsorized at 1%. Parentheses contain tstatistics. The asterisks denote *** for p<0.01, ** for p<0.05, * for p<0.1. Dependent variable (in basis points): 10,000*(ln(PPE ) - ln(PPE )) t+x t-1 x=4 quarters ahead x=6 quarters ahead Non-hedgers Hedgers Non-hedgers Hedgers Constrained Unconstrained Constrained Unconstrained Constrained Unconstrained Constrained Unconstrained (high HP) (low HP) (high HP) (low HP) (high HP) (low HP) (high HP) (low HP) BankDebt/At -2,007.41 -1,002.06 1,031.09 -119.24 -717.06 -483.32 831.46 -245.13 (-1.20) (-1.02) (1.27) (-0.29) (-0.34) (-0.49) (0.97) (-0.56) (Sum) Change 4.14 9.65* 3.27* -1.15 -1.39 14.44*** 1.52 1.30 *BankDebt/At (0.51) (1.67) (1.74) (-0.58) (-0.20) (2.84) (0.74) (0.62) (Sum) Change* -5.50 4.41 -15.82** 0.21 BankDebt/At*Constrained (-0.79) (1.59) (-2.03) (0.07) Market-to-Book 281.38** 476.11*** -88.34 773.48*** 330.31* 481.26*** -26.30 866.56*** (2.21) (5.23) (-0.54) (7.14) (1.93) (5.27) (-0.13) (7.30) CashFlow/Capital 5,652.95** 3,212.44 6,407.42** 2,888.91** 11,226.57*** 5,846.00*** 3,430.26 5,405.10*** (2.09) (1.19) (2.21) (2.34) (2.93) (2.91) (1.33) (3.26) 16,219.03** 13,070.66*** 9,663.72*** 12,616.35*** 17,210.22*** 13,567.38*** 9,220.25*** 12,262.66*** Lagged * Investment/Capital (5.45) (8.12) (4.60) (9.04) (5.06) (8.17) (4.53) (6.72) Firm Controls YES YES YES YES YES YES YES YES Firm FE YES YES YES YES YES YES YES YES Change*Firm Controls YES YES YES YES YES YES YES YES Year-quarter dummies YES YES YES YES YES YES YES YES Industry-Quarter YES YES YES YES YES YES YES YES Clustering Observations 3,813 3,770 2,037 5,207 3,664 3,671 1,940 5,078

Table XIII Bank Debt Specialness in the Unconventional Period All regressions include firm fixed effects. Hedgers are defined on a yearly basis as those firms that report having hedged their interest rate risk from floating to fixed in their 10-K annual reports. Calculation of other variables is presented in Tables III. Parentheses contain t-statistics. The asterisks denote *** for p<0.01, ** for p<0.05, * for p<0.1. (1) (2) (3) (4) (5) VARIABLES ALL ALL ALL Hedgers Non-Hedgers Surprise -0.33*** -0.35*** -0.31*** -0.24*** -0.24*** (-11.67) (-12.19) (-10.63) (-3.36) (-5.42) Surprise*(BankDebt/At) 0.43** 0.00 -0.23 0.15 (1.98) (0.00) (-0.61) (0.28) Surprise*LnAssets -0.11*** -0.12*** -0.08*** (-5.27) (-3.46) (-2.89) Surprise*Book Leverage 0.24* 0.65*** 0.14 (1.92) (3.03) (0.74) Surprise*Profitability -0.15 -0.87 -0.05 (-0.69) (-1.40) (-0.18) Surprise*M/B -0.12*** -0.19*** -0.09*** (-5.05) (-2.89) (-2.97) Observations 38,097 36,736 36,568 10,918 15,256 R-squared 0.00 0.00 0.01 0.02 0.01 Number of gvkey 1,903 1,792 1,779 679 1,030

APPENDIX

Table A1 Description of Firm Level Variables Item codes are from Compustat. CIQ items come from Capital IQ. All data used in regressions is deflated to year 2000 dollars. Variable Construction Bank Debt/At 1 [Drawn Credit Lines (CIQ) + Term Loans (CIQ)] / Assets (AT) Bank Debt/At 2 [Drawn Credit Lines (CIQ) + Term Loans (CIQ) + Undrawn Credit Lines (CIQ)]/ Assets (AT) Book Leverage (Total Debt (DLC+DLTT)) / (Total Debt + Book Value of Equity) Book Value of Equity Common/Ordinary Equity – Total (CEQ) Cash/At Cash and Short-Term Investments (CHE)/Total Assets (AT) Cash Flow Quarterly level measure: earnings before extraordinary items (IBQ) + depreciation (DPQ). Cash Flow Volatility Standard Deviation of Operating Income Before Depreciation (OIBDP) over Previous 12 Quarters Scaled by Total Assets (AT) CAPM Beta Monthly CAPM Beta using last 60 months. Floating Interest Rate Debt Debt with floating interest rate (CIQ) Hadlock-Pierce (HP) measure HP = -0.548*Size+0.025*Size2-0.031*Age. Size is log of assets (inflation adjusted to 2004). Age is the current year minus the first year that the firm has a non-missing stock price in CRSP. Size (Age) is winsorized at 1% on the low end, and at the $4.5 billion (37 years) on the high end. Hedging Dummy A dummy variable that takes the value 1 if a firms reports floating-to-fixed interest-rate hedging activities in its 10-K Interest Rate Coverage Ratio Quarterly level measure: interest expenses (XINTQ) / (interest expenses (XINTQ) + cash flow) Int. Rate Sensitivity of Earnings Correlation between quarterly firm EBITDA and three-month average LIBOR rates Inventory Investment Quarterly level measure: logarithm of Inventories (INVTQ) in quarter ‘t’ - logarithm of Inventories in quarter ‘t-1’ (Inventory is deflated to base year 2000) Investment in Capital Quarterly level measure: logarithm of Property, Plant and Equipment (PPEGTQ) in quarter ‘t’ - logarithm of Property, Plant and Equipment in quarter ‘t-1’ (Property, Plant and Equipment is deflated to base year 2000) Market-to-Book Assets [Market Value of Equity + Total Debt + Preferred Stock Liquidating Value (PSTKL)] / Total Assets (AT) Market Value of Equity Stock Price (PRCC_F) × Common Shares Used to Calculate EPS (CSHO) Profitability Operating Income before Depreciation (OIBDP) / Total Assets (AT) Rated A dummy variable that takes the value of one if the firm has a long term credit rating from S&P, and zero otherwise Sales Growth Logarithm of total sales (SALEQ) during quarters ‘t’ to ‘t+s’ – logarithm of total sales during quarters “t-s-1” to “t-1” (sales are deflated to base year 2000) Short-Term Debt Debt in current liabilities (DLC) and is equal to the total amount of short-term notes and the current portion of long-term debt that is due in one year. Size (At) Logarithm of Book Value of Total Assets (AT) , deflated to base year 2000 Tangibility (Inventories (INVT) + Net Plant, Property, Equipment (PPENT))/Total Assets (AT) Total Debt Long-Term Debt (DLTT) + Debt in Current Liabilities (DLC)

Table A2 Which Firms Use Bank, Floating-Rate and Short-Term Debt, and Interest Rate Hedging This table examines the use of bank debt, floating-rate debt, short-term debt and interest rate hedging, using firm-year data. A constant is included but not reported. All firm characteristics are winsorized at the 1% level. Standard errors are clustered at the firm level. Columns 1-4 use an OLS specification, as in Lemmon, Roberts and Zender (2008), while columns 5 and 6 use a Probit specification. Parentheses contain t-statistics. The asterisks denote *** for p<0.01, ** for p<0.05, * for p<0.1. (1) (2) (3) (4) (5) (6) Bank/At Bank/At Floating-Rate Debt/At Short-Term Debt/At Hedging Hedging LnAssets 0.01*** 0.00** -0.00*** 0.21*** (16.24) (2.55) (-10.13) (40.54) Profitability -0.03*** -0.02*** 0.01*** -0.01*** 1.42*** 1.43*** (-10.08) (-7.45) (4.69) (-4.67) (19.78) (20.10) Market to Book -0.00*** -0.00*** -0.00*** -0.00 -0.14*** -0.14*** (-3.86) (-7.22) (-3.11) (-1.27) (-22.25) (-22.98) Book Leverage 0.18*** 0.18*** 0.04*** 0.05*** 0.38*** 0.40*** (113.33) (114.16) (34.31) (44.82) (14.27) (14.99) Unrated 0.03*** 0.03*** 0.00 0.00*** -0.37*** -0.58*** (20.13) (17.13) (1.31) (3.35) (-21.33) (-37.48) Interest Rate Sensitivity 0.01 0.00 0.00 -0.00 -0.01 -0.05*** (1.56) (0.83) (0.05) (-0.12) (-0.73) (-2.68) Cash Flow Volatility -0.50 -0.70 0.33 -0.52** -12.79*** -13.74*** (-1.04) (-1.46) (1.10) (-2.33) (-6.31) (-6.82) Tangibility 0.03*** 0.03*** 0.01*** 0.01*** 0.25*** 0.28*** (6.66) (7.16) (2.94) (3.71) (6.73) (7.58) Age -0.00*** -0.00*** 0.00*** 0.00*** (-5.35) (-6.44) (6.14) (3.18) HP Index -0.01*** -0.48*** (-4.78) (-34.32) Bank/At 0.76*** 0.09*** 3.04*** 2.92*** (268.07) (41.96) (48.57) (47.29) Year FE YES YES YES YES YES YES Industry FF48 FE YES YES YES YES YES YES Observations 69,179 69,179 67,127 69,179 64,654 64,654 Number of gvkey 2,564 2,564 2,503 2,564

Table A3 Response of Equity Prices to Federal Funds Rate Changes: Comparison across Samples The table reports the results from regressions of equity returns on the surprise and expected components of the change in the federal funds rate, all expressed in percentage terms. Outliers are excluded following the analysis of Bernanke and Kuttner (2005) based on a Cook’s D statistic greater than 0.1. As in Bernanke and Kuttner (2005), for the period 1994-2002 outliers include October 15, 1998, January 3, 2001, March 20, 2001, April 18, 2001, and September 17, 2001 which are discussed in their paper. For the period 2003-2008, outlier dates are January 22, 2008, and March 18, 2008. Both of which are characterized by very large rate cuts. On January 21, 2008, in response to deteriorating market conditions, the Federal Open Market Committee (FOMC) held an unscheduled meeting (conference call) despite the national holiday (Martin Luther King day). They decided on a rate cut of 75 basis points (bp), which they announced shortly before the opening of U.S. markets. Although the rate cut was almost entirely unexpected, with an unprecedented surprise of -74bp, stock prices declined by almost 100bp compared to their closing price before the holidays. Shortly after, on March 18, 2008, the FOMC announced another unusually large cut in the federal funds rate (-75bp) in response to turmoil in the markets and the collapse of Bear Stearns. Stocks rallied in response, although the federal funds futures data suggested that some market participants had expected an even larger rate cut (about 100bp). Column 1 contains returns for a value-weighted equity index. Columns 2-5 report returns for individual firm-date observations over different sample periods. Column 5 includes only observations for which data on bank debt is available. The firm level regressions contain random effects. Parentheses contain t-statistics. The asterisks denote *** for p<0.01, ** for p<0.05, * for p<0.1. (1) (2) (3) (4) (5) Daily Value- Daily Returns Daily Returns Daily Returns Two-day Returns weighted Index All Firms All Firms All Firms Our Sample 1994-2008 1994-2008 1994-2002 2003-2008 2003-2008 Expected 0.421 0.209*** 0.193*** 0.133*** -0.641*** (1.00) (8.40) (5.73) (3.90) (8.13) Surprise -3.359** -2.704*** -2.424*** -4.665*** -4.451*** (-2.05) (-32.46) (-25.67) (-25.64) (11.60) # Observations 115 536,357 363,290 173,067 66,200

Table A4 The Role of Bank Debt Usage and Interest Rate Risk Exposure in the Transmission of Monetary Policy This table examines how the reaction of firm equity prices to surprise changes in the federal funds rate varies with their level of bank dependence. The sample consists of U.S. firms covered by Capital IQ, CRSP and Compustat from 2003 to 2008, excluding utilities (SIC 4900-4949) and financials (SIC 6000-6999). We focus on firms with December fiscal year end to avoid asynchronous balance sheet items and use 2-day returns in order to allow the effect of bank-debt to be fully incorporated in stock prices. We remove firm-year observations with negative revenues, missing information on total assets, or a value of total assets under 10 million. We also discard penny stocks, defined as those with a price of less than $5. The sample comprises 43 monetary policy events from 2003 to 2008. Firm characteristics are demeaned and are lagged by one year and winsorized at the 1% level. The regression specification is as in equation (1). Unreported terms include a constant and non-interacted coefficients. In specification (6) we add undrawn credit lines to bank debt and normalize the resulting ratio to have the same standard deviation as the original BankDebt/At. Standard errors are clustered at the date level in specifications (1)-(2) and two-way clustered at the date and industry levels in specifications (4)-(7). Industries are Fama-French 48 industries. Square brackets around the estimates of the coefficient of surprise in columns (4)-(7) are introduced to indicate that, due to the interaction of surprise with industry fixed effects, these estimates cannot be interpreted as the estimate applicable to the average firm. Parentheses contain t-statistics. The asterisks denote *** for p<0.01, ** for p<0.05, * for p<0.1. Hadlock and Pierce (2010) measure given by HP = -0.548*Size+0.025*Size2-0.031*Age. (1) (2a) (2b) (2c) (3) (4) (5) (6) (7) (8) (9) No Mkt. Return CAPM Fama-French Other Industry FE & Inc. Undrawn Other Firm Fixed Instrumental Floating Rate Controls Control Control Control Controls Event-Industry Credit Lines Controls Effects Variable Debt Clustering Surprise -4.97*** 2.93*** 1.28*** -0.09 -8.02*** [-7.44] [-8.07] [-9.83] -8.04*** -8.07*** -8.81*** (-13.03) (8.22) (3.66) (-0.25) (-17.72) (-0.83) (-0.90) (-1.10) (-3.33) (-17.12) (-3.63) Surprise*(BankDebt/At) -14.10*** -12.31*** -10.52*** -8.09*** -16.34*** -16.77*** -14.62*** -15.22*** -16.37*** -14.50 -13.79** (-4.35) (-4.15) (-3.54) (-2.63) (-4.17) (-3.82) (-3.10) (-3.30) (-2.69) (-0.58) (-2.48) Surprise*LnAssets -0.95*** -1.12*** -1.06*** -0.07 -0.94*** -1.00** -1.01** (-3.67) (-4.19) (-3.99) (-0.15) (-2.64) (-2.06) (-2.53) Surprise*Book Leverage 3.28** 3.83* 2.59 4.24** 3.15 2.44 4.49* (1.96) (1.85) (1.32) (1.98) (1.28) (0.40) (1.71) Surprise*Profitability -16.10*** -11.49** -11.08** -8.16 -15.36** -15.68*** -16.54 (-6.10) (-2.19) (-2.13) (-1.33) (-2.08) (-4.06) (-1.45) Surprise*M/B -0.02 -0.41 -0.41 -0.71 0.01 0.10 0.43 (-0.08) (-0.77) (-0.78) (-1.31) (0.01) (0.25) (0.42) Surprise*Int Rate Sensitivity -7.05** (-2.24) Surprise*Cash-Flow Volatility -77.57 (-0.55) Surprise*Beta 1.47** (2.16) Surprise*Cash Holdings 3.37 (0.96) Surprise*HP 4.22*** (3.42) Firm FE NO NO NO NO NO NO NO NO YES YES YES FF48 Industry FE NO NO NO NO NO YES YES YES NO NO NO Year FE NO NO NO NO YES YES YES YES YES YES YES Surprise*FF48 Industry FE NO NO NO NO NO YES YES YES NO NO NO Cluster (Fed event*IndustryFF48) NO NO NO NO NO YES YES YES YES NO YES Observations 64,682 64,682 64,557 64,549 64,428 62,871 62,746 55,506 64,428 63,626 41,710

Table A5 The Role of Bank Debt Usage and Interest Rate Risk Exposure in the Transmission of Monetary Policy: Excluding Positive Rate Changes This table repeats Table II after discarding those FOMC announcements with positive rate changes. Hedgers are defined on a yearly basis as those firms that report having hedged their interest rate risk from floating to fixed in their 10-K annual reports. Only firms with floating rate debt constituting more than 1% of total assets are included. Bank Debt/At is defined as bank debt (term loans plus drawn revolving credit) over the book value of assets (At). FloatingRateDebt /At is defined as floating rate debt over the book value of assets (At). All regressions also include an unreported constant term, as well as ln(assets), book leverage, profitability, market-to-book, interest rate sensitivity, and their interaction with surprise. All firm characteristics are lagged by one year and winsorized at 1%. Parentheses contain t-statistics. The asterisks denote *** for p<0.01, ** for p<0.05, * for p<0.1. (1) (2) (3) (4) (5) (6) (7) (8) Non- Hedgers Non- Hedgers Non- Hedgers Non- Hedgers Main Variables Hedgers Hedgers Hedgers Hedgers Surprise -3.44*** -7.84*** -4.42 -5.96* -3.94*** -7.41*** -5.14* -5.41* (-3.28) (-7.71) (-1.62) (-1.94) (-3.92) (-7.56) (-1.91) (-1.74) Surprise *(BankDebt/At) -25.20*** 0.59 -38.68*** 1.96 (-2.91) (0.10) (-3.13) (0.20) Surprise *(FloatingRateDebt /At) -20.64** -3.98 -29.97** -6.35 (-2.47) (-0.73) (-2.26) (-0.64) 25.61** 40.64*** 16.32* 23.62 (2.48) (2.75) (1.66) (1.43) Surprise *log(Assets) -2.02*** -0.70 -1.65** -0.98 (-2.82) (-0.97) (-2.26) (-1.37) Surprise *Book Leverage 6.43 -6.13 6.02 -3.62 (1.29) (-1.32) (1.16) (-0.78) Surprise *Market-to-Book -0.25 2.34* -0.16 2.35* (-0.19) (1.66) (-0.12) (1.67) Surprise *Profitability -13.14 -24.14 -13.79 -24.12 (-1.07) (-1.07) (-1.13) (-1.06) Surprise *Int. Rate Sensitivity -9.80* -5.33 -10.38* -5.57 (-1.67) (-0.80) (-1.79) (-0.84) Firm Controls NO NO YES YES NO NO YES YES Firm FE NO NO YES YES NO NO YES YES Surprise*Firm Controls NO NO YES YES NO NO YES YES Cluster (Fed event*IndustryFF48) NO NO YES YES NO NO YES YES Observations 7,067 7,585 7,067 7,585 7,067 7,585 7,067 7,585

Table A6 The Role of Bank Debt Usage and Interest Rate Risk Exposure in the Transmission of Monetary Policy Cumulative Returns in two Days before the FOMC announcement This table provides a placebo experiment by repeating Table II after replacing the dependent variable with the cumulative returns over the two days before the FOMC announcement. Hedgers are defined on a yearly basis as those firms that report having hedged their interest rate risk from floating to fixed in their 10-K annual reports. Only firms with floating rate debt constituting more than 1% of total assets are included. Bank Debt/At is defined as bank debt (term loans plus drawn revolving credit) over the book value of assets (At). FloatingRateDebt /At is defined as floating rate debt over the book value of assets (At). All regressions also include an unreported constant term, as well as ln(assets), book leverage, profitability, market-to-book, interest rate sensitivity, and their interaction with surprise. All firm characteristics are lagged by one year and winsorized at 1%. Parentheses contain t-statistics. The asterisks denote *** for p<0.01, ** for p<0.05, * for p<0.1. (1) (2) (3) (4) Main Variables Non-Hedgers Hedgers Non-Hedgers Hedgers Surprise -1.98 -2.82 -2.30 -2.53 (-1.00) (-1.08) (-1.19) (-0.97) Surprise *(BankDebt/At) 2.45 4.10 (0.26) (0.59) Surprise *(FloatingRateDebt /At) 23.77** 0.61 (2.36) (0.09) Surprise *log(Assets) -0.89 -0.81 -0.50 -0.93 (-1.29) (-1.13) (-0.74) (-1.32) Surprise *Book Leverage -0.26 -3.30 -5.33 -2.40 (-0.07) (-0.73) (-1.29) (-0.53) Surprise *Market-to-Book 2.28** -0.34 2.35** -0.34 (2.36) (-0.22) (2.41) (-0.22) Surprise *Profitability 2.71 23.97 0.09 23.88 (0.38) (1.46) (0.01) (1.45) Surprise *Int. Rate Sensitivity -0.65 -0.48 -0.56 -0.59 (-0.14) (-0.10) (-0.12) (-0.12) Firm Controls YES YES YES YES Firm FE YES YES YES YES Surprise*Firm Controls YES YES YES YES Cluster (Fed event*IndustryFF48) YES YES YES YES Observations 11,788 12,334 11,788 12,334

Table A7 The Role of Interest Rate Risk Exposure in the Transmission of Monetary Policy: First Stage Regressions of Instrumental Variables Analysis Following Graham and Smith (1999) and Campello, Lin, Ma, Zou (2011), Vol is the volatility of taxable income, Corr is the serial correlation of taxable income, DITC is a dummy for investment tax credits, DNOL is a dummy for net operating losses, and DSmallNeg (DSmallPos) is a dummy for small negative (positive) taxable income. We calculate the volatility of taxable income and the serial correlation of taxable income on a rolling basis, using historical annual data up to the year of interest, starting in 1989. Column (1) uses lagged hedging dummy as instrument for hedging. Column (2) uses the variables underlying the convexity measure, excluding Vol, whereas column (3) uses all variables. Column (4) uses the tax convexity measure, Convexity, directly, as given in the text. Column (5) uses both the lagged hedging dummy and the tax convexity measure. Only firms with floating rate debt constituting more than 1% of total assets are included. A constant, non-interacted terms, and the policy surprise interacted with firm size, book leverage, profitability and the market-to-book ratio are included but not reported for brevity. All firm characteristics are lagged by one year and winsorized at 1%. Parentheses contain tstatistics. The asterisks denote *** for p<0.01, ** for p<0.05, * for p<0.1. Panel A: Bank Debt (1) (2) (3) (4) (5) IV IV1 IV2 IV3 IV VARIABLES L.hedging Convexity Convexity Convexity Both Surprise*(BankDebt/At)* L.hedgingdummy 0.69*** 0.67*** (116.06) (73.74) Corr -0.01 -0.00 (-0.71) (-0.31) DITC -0.23*** -0.23*** (-12.08) (-11.74) DSmallNeg -0.87 -0.74 (-0.42) (-0.36) DNOL 0.16*** 0.16*** (11.48) (11.58) DNOL*DSmallNeg 2.17 1.77 (0.67) (0.55) DSmallPos 1.41 1.59 (0.03) (0.04) DNOL*DSmallPos -2.00 -2.28 (-0.05) (-0.05) Vol -0.00 (-0.40) Convexity 0.02*** -0.00 (9.75) (-1.62) Observations 23,413 12,009 12,009 12,009 11,665 R-squared 0.86 0.75 0.75 0.74 0.87 Firm FE YES YES YES YES YES Firm Controls YES YES YES YES YES Surprise*Firm Controls YES YES YES YES YES F-stat 6012 441.5 385.0 1390 2024

Panel B: Floating Rate Debt (1) (2) (3) (4) (5) IV IV1 IV2 IV3 IV VARIABLES L.hedging Convexity Convexity Convexity Both Surprise*(FloatingRateDebt/At)* L.hedgingdummy 0.70*** 0.68*** (120.40) (77.82) Corr -0.18*** -0.17*** (-12.35) (-11.99) DITC -0.16*** -0.15*** (-8.30) (-7.88) DSmallNeg 3.25 2.00 (0.49) (0.30) DNOL 0.15*** 0.15*** (10.89) (10.86) DNOL*DSmallNeg -2.78 -1.62 (-0.44) (-0.25) DSmallPos 8.68 8.31 (0.04) (0.04) DNOL*DSmallPos -9.37 -9.20 (-0.05) (-0.05) Vol 0.00 (0.57) Convexity 0.04*** 0.01*** (20.86) (8.36) Observations 23,413 12,009 12,009 12,009 11,665 R-squared 0.87 0.77 0.77 0.76 0.88 Firm FE YES YES YES YES YES Firm Controls YES YES YES YES YES Surprise*Firm Controls YES YES YES YES YES F-stat 6564 477.1 415.8 1553 2346

Table A8 Interest Rate Risk Exposure and the Transmission of Monetary Policy: The Role of Financing Constraints Whited-Wu and Kaplan_Zingales Measures Hedgers are defined on a yearly basis as those firms that report having hedged their interest rate risk from floating to fixed in their 10-K annual reports. Financial constraints are proxied with Whited-Wu (WW) quarterly measure as reported in Whited and Wu (2006) and Kaplan-Zingales (KZ) measure, as reported in Lamont, Polk, Saa-Requejo (2001). Accordingly, KZ = –1.001909[(IB+DP)/lagged PPENT] + 0.2826389[ (AT + PRCC_F×CSHO - CEQ - TXDB)/AT] + 3.139193[(DLTT + DLC)/(DLTT + DLC + SEQ)] – 39.3678[(DVC + DVP)/lagged PPENT] – 1.314759[CHE/lagged PPENT] and WW = –0.091 [(IB + DP)/AT] – 0.062[indicator set to one if DVC + DVP is positive, and zero otherwise] + 0.021[DLTT/AT] – 0.044[log(AT)] + 0.102[average SIC 3-digit industry sales growth each year] – 0.035[sales growth]. All capitalized mnemonics refer to Compustat data items. The financial constraint measure takes value 1 if the corresponding measure is above the median in a given year. Only firms with floating rate debt constituting more than 1% of total assets are included. Bank Debt/At is defined as bank debt (term loans plus drawn revolving credit) over the book value of assets (At). All regressions also include an unreported constant term, as well as ln(assets), book leverage, profitability, market-to-book, interacted with surprise and uninteracted. All firm and lender characteristics are lagged by one year and winsorized at 1%. Parentheses contain t-statistics. The asterisks denote *** for p<0.01, ** for p<0.05, * for p<0.1. (1) (2) (3) (4) (5) (6) (7) (8) LOW WW HIGH WW LOW WW HIGH WW LOW KZ HIGH KZ LOW KZ HIGH KZ VARIABLES NONHEDGER NONHEDGER HEDGER HEDGER NONHEDGER NONHEDGER HEDGER HEDGER Surprise -2.59 -6.66** -6.11** -11.56*** -6.86*** -3.51** -3.24 -7.86*** (-1.08) (-2.54) (-2.31) (-3.40) (-3.24) (-2.04) (-0.89) (-3.68) Surprise*BankDebt/At -18.82 -51.93*** 0.59 6.95 -45.70* -45.79*** -26.92* 7.91 (-1.24) (-3.36) (0.07) (0.47) (-1.84) (-3.76) (-1.72) (0.91) Surpr.*(BankDebt/At)* -33.11 6.36 -0.0834 34.83* Constrained (1.50) (0.39) (0.00) (1.84) Firm FE YES YES YES YES YES YES YES YES Firm Controls YES YES YES YES YES YES YES YES Surprise*Firm Controls YES YES YES YES YES YES YES YES Observations 5,291 5,996 7,893 3,795 3,879 6,909 2,703 8,436 R-squared 0.01 0.01 0.02 0.01 0.01 0.01 0.02 0.02 Number of gvkey 363 489 450 285 344 510 189 516

Table A9 Interest Rate Risk Exposure and the Transmission of Monetary Policy: The Role of Liquidity Constraints Current Ratio and Interest Coverage Ratio Hedgers are defined on a yearly basis as those firms that report having hedged their interest rate risk from floating to fixed in their 10-K annual reports. Liquidity constrained firms are those with current ratio (Current Assets/Current Liabilities) below the median or interest coverage ratio above the median in a given year. The interest coverage ratio is equal to interest expenses (XINT) divided by the sum of cash flow and interest expenses. Cash flow is equal to earnings before extraordinary items (IB) plus depreciation (DP). All capitalized mnemonics refer to Compustat data items. Only firms with floating rate debt constituting more than 1% of total assets are included. Bank Debt/At is defined as bank debt (term loans plus drawn revolving credit) over the book value of assets (At). All regressions also include an unreported constant term, as well as ln(assets), book leverage, profitability, market-to-book, interacted with surprise and uninteracted. All firm and lender characteristics are lagged by one year and winsorized at 1%. Parentheses contain t-statistics. The asterisks denote *** for p<0.01, ** for p<0.05, * for p<0.1. (1) (2) (3) (4) (5) (6) (7) (8) CURRENT CURRENT CURRENT CURRENT COVERAGE COVERAGE COVERAGE COVERAGE RATIO RATIO RATIO RATIO RATIO RATIO RATIO RATIO UNCONSTR. CONSTR. UNCONSTR. CONSTR. UNCONSTR. CONSTR. UNCONSTR. CONSTR. VARIABLES NONHEDGER NONHEDGER HEDGER HEDGER NONHEDGER NONHEDGER HEDGER HEDGER Surprise -7.30*** -3.18* -7.92*** -8.33*** -4.51** -6.94*** -2.83 -11.63*** (-3.73) (-1.80) (-2.84) (-3.59) (-2.35) (-2.70) (-0.93) (-4.94) Surprise*BankDebt/At -17.48 -49.19*** 8.79 -1.06 -30.07* -41.00*** 3.20 7.46 (-0.93) (-3.84) (0.52) (-0.13) (-1.65) (-3.02) (0.18) (0.93) Surpr.*(BankDebt/At)* -31.71 -9.857 -10.93 4.258 Constrained (1.41) (0.55) (0.47) (0.21) Firm FE YES YES YES YES YES YES YES YES Firm Controls YES YES YES YES YES YES YES YES Surprise*Firm Controls YES YES YES YES YES YES YES YES Observations 4,978 6,450 3,320 8,351 5,963 5,457 4,262 7,975 R-squared 0.01 0.01 0.03 0.01 0.01 0.01 0.02 0.02 Number of gvkey 417 496 257 532 459 470 331 517

Table A10 The Effect of Monetary Policy on Fixed Investment – Analysis using Monetary Policy Surprises This table examines how monetary policy affects firm fixed investment and how this effect varies with bank debt usage and interest rate risk hedging. Inventories are calculated as Total Inventories (INVTQ), and Fixed Investment is computed as the difference (in basis points) between the log of property, plant and t-1,t+x equipment (PPEGTQ) x quarters ahead and the log of PPEGTQ at the end of the quarter before the monetary policy surprise occurs. Surprise is the sum of all surprises in the federal funds rate that occur during a quarter. Hedgers are defined as those firms that report having hedged their interest rate risk from floating to fixed in their 10K annual reports. Only firms with floating rate debt constituting more than 1% of total assets are included. Bank Debt/At is defined as bank debt (term loans plus drawn revolving credit) over the book value of assets (At). All regressions also include an unreported constant term. Controls include the lagged investment to capital ratio, the lagged cash holdings to capital ratio, and the market to book ratio, and also (unreported): ln(assets), book leverage, market-tobook, profitability and interest rate sensitivity of operating income. All firm controls are lagged by one quarter and winsorized at 1%. Parentheses contain tstatistics. The asterisks denote *** for p<0.01, ** for p<0.05, * for p<0.1. Dependent variable: ln(PPE t+x) - ln(PPE t-1) x=4 quarters ahead x=6 quarters ahead Non-hedgers Hedgers Non-hedgers Hedgers Constrained Unconstrained Constrained Unconstrained Constrained Unconstrained Constrained Unconstrained (high HP) (low HP) (high HP) (low HP) (high HP) (low HP) (high HP) (low HP) (Sum) Surprise (omitted) BankDebt/At -1,764.02 -828.41 989.51 -93.86 -618.15 -312.49 841.93 -258.83 (-1.07) (-0.84) (1.21) (-0.23) (-0.30) (-0.31) (0.97) (-0.59) (Sum) Surprise 15.27 42.02*** 2.28 -3.08 -18.73 48.24*** 1.99 -1.03 *BankDebt/At (0.64) (2.75) (0.51) (-0.54) (-1.05) (3.27) (0.39) (-0.17) (Sum) Surprise* -26.74 5.36 -66.97** 3.01 BankDebt/At*Constrained (-1.65) (0.74) (-4.19) (0.40) Market-to-Book 283.36** 467.50*** -106.77 769.00*** 337.19** 475.44*** -40.56 863.15*** (2.20) (5.13) (-0.67) (7.15) (1.98) (5.18) (-0.21) (7.30) CashFlow/Capital 5,746.04** 3,162.56 6,435.81** 2,858.27** 11,257.96*** 5,782.55*** 3,500.91 5,374.71*** (2.11) (1.16) (2.23) (2.31) (2.94) (2.87) (1.36) (3.24) 16,215.94*** 13,209.57*** 9,745.25*** 12,662.08*** 17,325.72*** 13,773.10*** 9,317.81*** 12,314.08*** Lagged Investment/Capital (5.44) (8.18) (4.67) (9.06) (5.08) (8.26) (4.59) (6.73) Firm Controls YES YES YES YES YES YES YES YES Firm FE YES YES YES YES YES YES YES YES Surprise*Firm Controls YES YES YES YES YES YES YES YES Year-quarter dummies YES YES YES YES YES YES YES YES Industry-Quarter Clustering YES YES YES YES YES YES YES YES Observations 3,813 3,770 2,037 5,207 3,664 3,671 1,940 5,078

Table A11 Short-Term Debt and the Response of Equity Prices to Federal Funds Rate Changes This table examines how the reaction of firm equity prices to surprise changes in the target federal funds rate varies with their usage of short-term debt. Short-Term Debt/At is defined as debt in current liabilities (item 34) over the book value of assets. Columns 2 and 4 include (unreported) log(assets), profitability, book leverage, the market-tobook ratio, and their interaction with policy surprise. All firm characteristics are lagged by one year, demeaned, and winsorized at 1%. Parentheses contain t-statistics. The asterisks denote *** for p<0.01, ** for p<0.05, * for p<0.1. (1) (2) (3) (4) Surprise -5.04*** -8.11*** -4.97*** -8.02*** (-13.32) (-18.07) (-13.03) (-17.73) Surprise*(ShortTermDebt/At) -10.30 -8.30 -4.36 -5.26 (-1.38) (-1.06) (-0.56) (-0.66) Surprise*(BankDebt/At) -13.64*** -15.99*** (-4.09) (-4.05) Firm Controls NO YES NO YES Surprise*Firm Controls NO YES NO YES Year FE NO YES NO YES Observations 65,893 65,649 64,658 64,428

Table A12 Is Bank Debt Special for the Transmission of Monetary Policy? Normalizing Bank Debt with Total Debt This table examines how the reaction of firm equity prices to changes in the federal funds rate varies with their level of bank dependence. The sample consists of U.S. firms covered by Capital IQ, CRSP and Compustat from 2003 to 2008, excluding utilities (SIC 4900-4949) and financials (SIC 6000-6999). We focus on firms with December fiscal year end to avoid asynchronous balance sheet items and use 2-day returns in order to allow the effect of bank-debt to be fully incorporated in stock prices. We remove firm-year observations with negative revenues, missing information on total assets, or a value of total assets under 10 million. We also discard penny stocks, defined as those with a price of less than $5. The sample comprises 43 monetary policy events from 2003 to 2008. Firm characteristics are demeaned and are lagged by one year and winsorized at the 1% level. The regression specification is as in equation (1). Unreported terms include a constant and non-interacted coefficients. In specification (5) we add undrawn credit lines to bank debt and normalize the resulting ratio to have the same standard deviation as the original BankDebt/At. Standard errors are clustered at the date level in specifications (1)-(2) and two-way clustered at the date and industry levels in specifications (4)-(7). Industries are Fama-French 48 industries. Parentheses contain t-statistics. The asterisks denote *** for p<0.01, ** for p<0.05, * for p<0.1. (1) (2) (3) (4) (5) (6) (7) (8) No With Controls Event-indust. Including Other Firm Fixed Instrumental Floating Rate Controls Clustering Credit Lines Controls Effects Variable Debt Surprise -5.60*** -8.20*** [-8.27] [-8.58] [-10.51] -8.19*** -8.22*** -8.57*** (-13.65) (-16.33) (-0.92) (-0.96) (-1.17) (-3.51) (-14.32) (-3.51) Surprise*(BankDebt/Debt) -2.18** -3.04*** -3.06*** -1.61 -3.14*** -2.85** -3.20 -1.53 (-2.12) (-2.73) (-2.87) (-1.58) (-2.75) (-2.18) (-0.35) (-0.95) Surprise*LnAssets -0.77*** -1.05*** -0.79*** -0.15 -0.73* -0.79 -0.85** (-2.76) (-3.74) (-2.83) (-0.32) (-1.90) (-0.86) (-2.16) Surprise*Book Leverage 0.38 0.52 -0.05 0.80 0.30 0.52 1.32 (0.25) (0.28) (-0.02) (0.40) (0.13) (0.32) (0.50) Surprise*Profitability -21.36*** -14.62** -15.21** -12.29 -21.73** -22.41*** -17.99 (-6.93) (-2.26) (-2.35) (-1.53) (-2.40) (-4.53) (-1.58) Surprise*M/B 0.40 -0.16 -0.02 -0.23 0.45 0.40 0.55 (1.20) (-0.25) (-0.04) (-0.33) (0.57) (0.75) (0.55) Surprise*Int Rate Sensitivity -7.42** (-2.44) Surprise*Cash-Flow Volatility -91.93 (-0.62) Surprise*Beta 1.83** (2.44) Surprise*Cash Holdings 1.19 (0.32) Surprise*HP 3.94*** (3.13) Firm FE NO NO NO NO NO YES YES YES FF48 Industry FE NO NO YES YES YES NO NO NO Year FE NO YES YES YES YES YES YES YES Interacted FF48 Industry FE NO NO YES YES YES NO NO NO Cluster (Fed event*IndustryFF48) NO NO YES YES YES YES NO YES Observations 53,054 53,028 51,963 51,963 45,972 53,028 52,398 41,665

Table A13 Bank Debt Specialness and Firm Financing Constraints This table examines how the effect of monetary policy on firm stock prices varies with their exposure to bank debt and their level of financial constraints. Financial constraints are proxied with the firm’s age and the Hadlock and Pierce (2010) measure given by HP = -0.548*Size+0.025*Size2-0.031*Age. Firm size is defined to be the log of assets (inflation adjusted to 2004). Age is defined as the current year minus the first year that the firm has a nonmissing stock price in CRSP. Firm size and age are at the 1% tails on the low end, and at the $4.5 billion and thirtyseven year points on the high end. The financial constraint measure takes value 1 if the firm’s age is below the median or firm’s HP statistic is above the median in a given year. Only firms with floating rate debt constituting more than 1% of total assets are included. A constant, non-interacted terms, and the policy surprise interacted with firm size book leverage, profitability and the market-to-book ratio are included but not reported. All firm characteristics are lagged by one year and winsorized at the 1% level. Industries are defined according to the Fama French 48 sector grouping. Parentheses contain t-statistics. The asterisks denote *** for p<0.01, ** for p<0.05, * for p<0.1. (1) (2) (3) (4) VARIABLES AGE HP AGE HP Surprise -4.92*** -2.31 -5.47*** -2.80 (-3.25) (-1.18) (-3.69) (-1.47) Surprise*Financial Constraint Measure 0.67 -3.85 0.56 -4.28* (0.36) (-1.57) (0.31) (-1.79) Surprise*Hedging -1.95 -0.83 -1.29 (-0.95) (-0.42) (-0.63) Surprise*(BankDebt/At) -28.00** -29.20** (-2.50) (-2.46) Surprise*(BankDebt/At)*Financial Constraint Measure -16.79 -12.50 (-1.43) (-1.04) Surprise*(BankDebt/At)*Hedging 41.25*** 40.41*** (3.36) (3.27) Surprise*(FloatingRateDebt /At) -19.31* -25.18** (-1.73) (-2.21) Surprise*(FloatingRateDebt /At)*Financial Constraint Measure -17.02 -5.33 (-1.48) (-0.45) Surprise*(FloatingRateDebt /At)*Hedging 24.55** 26.06** (2.04) (2.16) Firm FE YES YES YES YES Firm Controls YES YES YES YES Surprise*Firm Controls YES YES YES YES Observations 24,123 24,123 24,123 24,123 R-squared 0.01 0.01 0.01 0.01 Number of gvkey 1,283 1,283 1,283 1,283

Table A14 – Dynamic Model Simulated Regressions: Variable Definitions

Table A15 The Effect of Monetary Policy using Simulated Data – Robustness using Bank Debt over Assets This table displays regression results using simulated data from our dynamic model in Section 2.2 in which we use floating rate debt over total assets as our proxy for exposure to floating rate debt. It provides robustness for all the results on Tables II and II. Δrate refers to Surprise in stock return regressions, and Change in all other regressions. The definitions of all the variables used are described in detail in Appendix B. Parentheses contain tstatistics. The asterisks denote *** for p<0.01, ** for p<0.05, * for p<0.1. Dep variable: Stock Stock Investment/K Investment/K Covenant Costs of Interest Rate Returns Returns (4Q Ahead) (6Q Ahead) Violation Financial Coverage Likelihood Distress Ratio (1) (2) (3) (4) (5) (6) (7) Δrate -14.52*** -11.60*** -1.81*** -0.99*** -0.21 -0.53*** -1.24*** (-49.05) (-54.70) (-48.09) (-25.52) (-0.88) (-9.38) (-4.53) Δrate *(BankDebt/At) -1.46*** -1.38*** -0.09*** -0.06*** 0.02 0.09*** 0.79*** (-60.03) (-79.83) (-31.87) (-21.18) (1.01) (19.58) (35.15) Constrained* Δrate *(BankDebt/At) -0.25*** -0.22*** -0.02** -0.01 -0.02 0.21*** 0.52*** (-3.95) (-4.93) (-2.32) (-0.35) (-0.40) (17.76) (8.99) Controls YES YES YES YES YES YES YES R-squared 0.72 0.24 0.19 0.07 0.56 0.55 0.48 Observations 38,000 38,000 38,000 38,000 38,000 38,000 38,000