Firm Financial Conditions and the Transmission of Monetary Policy

Finance and Economics Discussion Series Federal Reserve Board, Washington, D.C. ISSN 1936-2854 (Print) ISSN 2767-3898 (Online) Firm Financial Conditions and the Transmission of Monetary Policy Thiago R.T. Ferreira; Daniel A. Ostry; John Rogers 2023-037 Please cite this paper as: Ferreira, Thiago R.T., Daniel A. Ostry, and John Rogers (2023). “Firm Financial Conditions and the Transmission of Monetary Policy,” Finance and Economics Discussion Series 2023-037. Washington: Board of Governors of the Federal Reserve System, https://doi.org/10.17016/FEDS.2023.037. 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.

Firm Financial Conditions and the ∗ Transmission of Monetary Policy † ‡ § Thiago R.T. Ferreira Daniel A. Ostry John Rogers May 31, 2023 Abstract We study how the transmission of monetary policy to firms’ investment and credit spreads depends on their financial conditions, finding a major role for their excess bond premia (EBPs), the component of credit spreads in excess of default risk. While monetary policy easing shocks compress credit spreads more for firms with higher exanteEBPs,itislower-EBPfirmsthatinvestmore.Werationalizethesefindingsusing a model with financial frictions in which lower-EBP firms have flatter marginal product of capital curves. We also show empirically that the cross-sectional distribution of firm EBPs determines the aggregate effectiveness of monetary policy. Key Words: Monetary Policy, Investment, Credit Spreads, Excess Bond Premium, Firm Heterogeneity. JEL Classification: E22, E44, E50. ∗WearegratefultoCarolynDavinandCaitlinDuttaforoutstandingresearchassistance.Wealsothank Andrea Ajello, Florin Bilbiie, Vasco Carvalho, Ambrogio Cesa-Bianchi, Giancarlo Corsetti, Carlos Vianna de Carvalho, Maarten De Ridder, Luca Fornaro, Maren Froemel, Etienne Gagnon, Nils Gornemann, Matteo Iacoviello, Aeimit Lakdawala, Simon Lloyd, Ali Ozdagli, H´el`ene Rey, Michael Smolyansky, Christian Wolf and presentation attendees at the University of Cambridge, Federal Reserve Board, RCEA 2022, SNDE 2022, CEMLA 2022, Barcelona School of Economics Summer Forum 2022, IAAE 2022, CICF 2022, Banque de France–CEPR–Sciences Po–OFCE EME Workshop 2022, Wake Forest Empirical MacroeconomicsWorkshop2023,andQueenMaryUniversityofLondonWorkshopinQuantitativeMacroeconomics 2023. The views expressed in this paper are solely those of the authors and should not be interpreted as reflecting the views of the the Board of Governors of the Federal Reserve System, or of any other person associated with the Federal Reserve System, nor the Bank of England or its committees. †Federal Reserve Board, International Finance Division; Washington, DC, 20551, USA; Email address: thiago.r.teixeiraferreira@frb.gov. ‡University of Cambridge and Bank of England; Threadneedle Street, London, EC2R8AH, UK; Email address: daniel.ostry@bankofengland.co.uk. §Fudan University; Shanghai, China; Email address: johnrogers@fudan.edu.cn.

1 Introduction How do firms’ investment responses to monetary policy depend on their financial conditions? Most of the large literature addressing this question is informed by theories in which firms’ access to external funds is subject to financial frictions (e.g., Bernanke and Gertler, 1989 and Kiyotaki and Moore, 1997). On the empirical front, the literature has proxied for the severity of firms’ financial frictions using various firm characteristics, such as size (Gertler and Gilchrist, 1994), default risk (Ottonello and Winberry, 2020), age (Cloyne et al., 2023), and liability structure (Gu¨rkaynak et al., 2022). The message of this research is that firms’ financial frictions, reflected in their marginal cost curves (Bernanke et al., 1999), play an important role in shaping their heterogeneous responses to monetary policy. In this paper, we find that differences in firms’ marginal benefit curves for capital are a major driver of their heterogeneous responses to monetary policy. Motivated by the evidence that firms’ future marginal productivity can be inferred from credit spreads (Philippon, 2009), we proxy for differences in firms’ marginal benefit curves using their excess bond premia (EBPs), the component of their credit spreads in excess of default risk (Gilchrist and Zakrajˇsek, 2012). Empirically, we show that while monetary policy easing shocks compress credit spreads more for firms with higher ex-ante EBPs, it is firms with lower EBPs that invest more. We rationalize these findings using a model with leverage constraints on financial intermediaries in which lower EBPs are afforded to firms whose marginal products of capital diminish relatively slowly as they invest, that is, to firms with flatter marginal benefit curves. In this setup, monetary policy easings shift firms’ marginal cost curves outward along their differentially-sloped marginal benefit curves, leading to movements in credit spreads and investment that match our empirical findings. We also show, consistent with our model, that the transmission of monetary policy to aggregate investment depends on the moments, especially the skewness, of the cross-sectional EBP distribution. This result empirically ties the aggregate potency of monetary policy to granular, firm financial conditions. We begin by estimating the heterogeneous responses of firms’ credit spreads and investment to monetary policy shocks. We do so by building a unique data set that combines 1

bond-level corporate yields and firm-level balance sheets for U.S. non-financial firms from 1973 to 2021 with a monetary policy shock series that bridges periods of conventional and unconventional policy. We find, on the one hand, that monetary policy easing shocks compress credit spreads more for firms with higher ex-ante EBPs—that is, for firms with tighter ex-ante financial conditions. On the other hand, we find that monetary policy easing shocks induce larger investment responses from firms with lower ex-ante EBPs. In both cases, the heterogeneity is economically significant: the peak response of investment and credit spreads for a firm with an EBP one standard deviation from the firm-level mean is about twice the size of the mean firm’s response. We also show that, as a state variable for monetary policy, a firm’s EBP plays a larger role than its default risk, measured both by “distance to default” (Merton, 1974) and leverage, and is statistically distinct from other firm characteristics tied to financial frictions such as size, share of liquid assets, and age. We then build a model that rationalizes our empirical results for credit spreads and investment. In the model, firms differ in how rapidly their marginal products of capital diminish as they invest, implying heterogeneity in the slopes of their marginal benefit curves for capital. The model also features an upward-sloping marginal cost curve for financing capital, which arises from leverage constraints on financial intermediaries (Gertler and Kiyotaki, 2010, Gertler and Karadi, 2011 and Anderson and Cesa-Bianchi, 2021). We show that the slope of firms’ marginal benefit curves, in the presence of segmented markets, pins down their EBPs, such that firms with flatter marginal benefit curves have lower EBPs in equilibrium. We find empirical support for this theoretical result by estimating distinct production functions for low- and high-EBP firms. Overall, our model highlights that while the average EBP across firms reflects financial intermediaries’ risk-bearing capacity (Gilchrist and Zakrajˇsek, 2012), differences in firms’ EBPs arise from differences in the resilience of firms’ marginal productivity to further investment, which is consistent with models linking credit spreads to Tobin’s Q (Philippon, 2009) and with the considerable predictive power of the EBP for future economic activity (Favara et al., 2016 and Lo´pez-Salido et al., 2017). Using our model, we show the importance of firms’ EBPs for their responsiveness to monetary policy with a comparative statics exercise. By increasing financial intermediaries’ net worth, a monetary policy easing leads to an outward shift in firms’ marginal cost curve 2

that traces along their respective marginal benefit curves. Thus, a monetary easing engenders a relatively large increase in investment by lower-EBP firms—due to their flatter marginal benefit curves—despite a relatively mild fall in their credit spreads. Conversely, higher-EBP firms increase investment relatively little despite a larger fall in their credit spreads. These results match our empirical findings and establish that the slopes of firms’ marginal benefit curves, as captured by their EBPs, are central to determining the sensitivity of firms’ investment and spreads to monetary policy. We provide support for the model’s economic mechanism by showing that two additional implications of the model hold empirically. First, the slope of firms’ marginal benefit curves should be relevant not just for the transmission of monetary policy, but also for any shift in the marginal cost curve. To test this hypothesis, we build on the inverse relationship between firm-level credit spreads and investment documented by several studies (e.g., Gilchrist and Zakrajˇsek, 2007), which is consistent with credit supply shocks being dominant in capital markets. In this case, lower-EBP firms should invest more following a reduction in their credit spreads, due to their flatter marginal benefit curves. We find robust evidence supporting this hypothesis in the data. From micro to macro, the second implication of our model is that the cross-sectional distribution of firm EBPs should influence the aggregate effectiveness of monetary policy. Specifically, when a larger mass of firms has lower EBPs—i.e., is on a flatter segment of their marginal benefit curves—the transmission of monetary policy to aggregate investment should be more potent. We test this prediction using moments of the cross-sectional EBP distributionasaggregatestatevariablesandinteractthemwithourmonetarypolicyshocks. Consistent with the model, in times when the EBP distribution is more left-skewed, expansionary monetary policy shocks induce larger increases in aggregate investment growth. This implies that variations in the aggregate potency of monetary policy emerge from fluctuations in granular, firm EBPs. Literature Review: Our paper relates to three strands in the literature. The first investigates firms’ heterogeneous responses to monetary policy. Much of this literature is motivated by theories in which firms’ access to external funds is subject to financial frictions, 3

such as agency costs (Bernanke and Gertler, 1989, and Bernanke et al., 1999), collateral constraints tied to firms’ physical capital (Kiyotaki and Moore, 1997) and earnings (Lian and Ma, 2021), as well as frictions in financial intermediation (e.g., Gertler and Kiyotaki, 2010, and Gertler and Karadi, 2011). Importantly—as highlighted by Ottonello and Winberry (2020), for example—financial frictions influence the shape of the marginal cost curve faced by firms. On the empirical front, the literature has used many firm-level characteristics to proxy for the severity of these financial frictions, such as liability structure (Ippolito et al., 2018; Gu¨rkaynak et al., 2022), age (Bahaj et al., 2022; Durante et al., 2022), age & dividends (Cloyne et al., 2023), size (Gertler and Gilchrist, 1994; Crouzet and Mehrotra, 2020), leverage (Anderson and Cesa-Bianchi, 2021; Caglio et al., 2021; Wu, 2018; Lakdawala and Moreland, 2021), credit default swap spreads (Palazzo and Yamarthy, 2022), liquid assets (Jeenas, 2019; Jeenas and Lagos, 2022), liquidity-constraints (Kashyap et al., 1994), marginal productivity (Gonza´lez et al., 2021), and information frictions (Ozdagli, 2018; Chava and Hsu, 2020).1 We contribute to this literature by showing that a firm’s EBP is an important determinant of its responsiveness to monetary policy. Moreover, we provide evidence that firm EBPs convey the slope of their marginal benefit curves for capital, making them distinct from financial frictions summarized in firm marginal cost curves. Second, our paper adds to the longstanding literature on the determinants of investment, especially the user cost of capital theory (Jorgenson, 1963) and the Q theory (Tobin, 1969).2 To address the empirical weakness of Q theory when assessed using equity prices, Philippon(2009)buildsamodelinwhichthe“bondmarket’sQ”iscapturedpredominantly byfirmcreditspreads,whichhefindstobeastrongpredictorofU.S.aggregateinvestment.3 Relatedly, Gilchrist and Zakrajˇsek (2007) and Gilchrist et al. (2014) find similar results for firm-level credit spreads, which are the main source of variation in firms’ user-cost of capital. Gilchrist and Zakrajˇsek (2012) clarify that it is the non-default-risk component of 1Focusing on firm cyclicality, Crouzet and Mehrotra (2020) highlight that as a state variable, firm size may not be capturing the extent of firms’ financial frictions, but rather their industry scope. Jeenas and Lagos, 2022 also focus on a non-financial-frictions channel by studying the effects of an instrumented Tobin’s q on firm equity issuance and investment conditional on firms’ asset liquidity. 2These literatures have their roots in the prima facie incompatibility between the stock and flow theories of capital and investment, respectively (e.g. Clark, 1899, Fisher, 1930, Keynes, 1936, Hayek, 1941). Beginning with Lerner (1953), q-theory has appealed to adjustment costs to resolve this incompatibility (see e.g. Lucas and Prescott, 1971, Abel, 1979 and Hayashi, 1982). 3Lin et al. (2018) extend the model to stochastic interest rates and empirically support their theory. 4

credit spreads, the EBP, that best predicts aggregate economic activity. Our contribution to this literature is twofold: (i) we show that the sensitivity of firms’ investment to changes in credit spreads depends on their ex-ante EBP; and (ii) we provide evidence that firms’ EBPs are linked to the slope of their marginal product of capital curves. Third, our paper contributes to the literature investigating the time-varying aggregate effects of monetary policy, especially its weaker effects during recessions. Vavra (2014) and McKay and Wieland (2021) build models in which monetary policy is less effective in recessions due to cyclicality in the cross-sectional distribution of price adjustments and durable expenditures, respectively. Tenreyro and Thwaites (2016) document that the decreased power of U.S. monetary policy in recessions is particularly evident for durables expenditure and business investment, while Jord`a et al. (2020) show this pattern holds internationally. Our paper contributes to this literature by providing a new firm-level rationale for monetary policy’s time-varying aggregate effects and its weaker transmission in recessions: variation in the slope of firms’ marginal benefit curves for capital, as reflected in the moments of the cross-sectional distribution of firm EBPs. 2 Data and Descriptive Statistics In this section, we describe the baseline monetary policy shock series (Section 2.1); discuss the EBP calculation (Section 2.2); document how the cross-sectional EBP distribution evolves over time and relates to other firm characteristics (Section 2.3); and detail the common features of our regression specifications (Section 2.4). 2.1 Monetary Policy Shocks As a baseline, we use the monetary policy shocks in Bu et al. (2021). These shocks combine three appealing features, which together distinguish them from other monetary policy shocks in the literature. First, by extracting high-frequency interest-rate movements from the entire U.S. Treasury yield curve, these shocks stably bridge periods of conventional 5

and unconventional monetary policy. Second, these shocks are devoid of the central bank information effect, the notion that monetary policy announcements, in addition to providing a pure monetary policy surprise, may also reveal information about the central bank’s views on the macroeconomy. Third, the shocks are not predicted ex-ante by available information, such as Blue Chip forecasts, “big data” measures of economic activity, news releases, and consumer sentiment.4 We calculate these shocks for the period January 1985 to December 2021, and, for regressions at a monthly (quarterly) frequency, aggregate the shocks by summing them within the month (quarter). In our regressions, we normalize the shocks so that positive values refer to monetary policy easings. See Appendix A.1 for more details. Appendix B.5 shows that our results are robust to using alternative monetary policy shocks. 2.2 Data Sources and EBP Calculation To provide a comprehensive picture of the firm, we use four databases: (i) the Center for Research in Security Prices (CRSP) Database and (ii) the CRSP/Compustat Merged Database, Wharton Research Data Services, for firms’ equity prices and balance sheets, respectively; (iii) the Arthur D. Warga, Lehman Brothers Fixed Income Database and (iv) the Interactive Data Corporation, ICE Pricing and Reference Data, for monthly corporate bond yields quoted in secondary markets. Merging these databases enables our unique investigation into monetary policy’s effects on U.S. non-financial firms’ quantities (investment) and prices (credit spreads). The combined sample period of these databases is 1973 to 2021, which is a significantly longer time period than is used in the existing literature on monetary policy and firm heterogeneity. To calculate the excess bond premium, we follow an approach similar to Gilchrist and Zakrajˇsek (2012). We first compute the credit spread S on the bond k issued by firm i ikt at time t as the difference between the bond’s yield and the yield on a U.S. Treasury that shares the same maturity, with the latter calculated by Gu¨rkaynak et al. (2007). Then, we 4For critiques of earlier monetary policy shocks that exhibited predictability, see, for example, Ramey (2016), Miranda-Agrippino (2016), and Bauer and Swanson (2020). 6

decompose each bond’s credit spread S into two components. The first is driven by the ikt firm’s default risk, as well as a vector of bond characteristics, and is termed the predicted ˆ spread S . The second, and residual, component is the excess bond premium, EBP . ikt ikt More precisely, we assume the following decomposition for bond-level credit spreads: logS = βDD +γ(cid:48)Z +υ , (1) ikt it ikt ikt whereDD isfirmi’sdistancetodefault,whichcapturesfirmi’sexpecteddefaultprobabilit ity (Merton, 1974); Z includes a vector of the bond’s characteristics, such as its duration, ikt par value and age, as well as industry and credit rating fixed effects; and υ is the erikt ror term. We estimate regression (1) by ordinary least squares (OLS) and compute the ˆ predicted credit spread S as ikt (cid:104) σˆ2(cid:105) S ˆ = exp β ˆ DD +γˆ(cid:48)Z + , (2) ikt it ikt 2 where β ˆ and γˆ denote the OLS estimates from regression (1) and σˆ2 denotes the estimated variance of the error term, which we assume to be normally distributed. While the model is simple, it explains a significant share of the variation in credit spreads—the R2 is 0.68— driven largely by the firm’s default risk. We define the excess bond premium (EBP) of firm i’s bond k at time t as ˆ EBP = S −S . (3) ikt ikt ikt Thus, the EBP is the component of the bond’s credit spread that is unexplained by the ikt firm’s default risk and the bond’s salient characteristics.5 A higher EBP implies that, ikt controlling for its default risk, the firm faces a higher marginal borrowing rate on its debt, and, thus, faces tighter financial conditions.6 Appendix B.6 shows that our results in the subsequent sections are robust to using a modified EBP that accounts for a potential ikt nonlinear relationship between spreads and distance to default. 5In Appendix A.3, we document that the correlation between our mean credit spreads and that of GilchristandZakrajˇsek(2012)is96%.ThecorrelationbetweenourEBPandthatofthoseauthorsis86%. 6See Appendix A.3 for more details on the EBP and distance to default calculations. 7

Figure 1 Cross-Sectional Distribution of Bond-Level EBPs over Time EBP in Percentage Points 15 10 5 0 -5 5th & 95th Percentiles 10th & 90th Percentiles Mean 1973m1 1980m1 1990m1 2000m1 2010m1 2020m1 Note.Figure1showsthemeanandselectedpercentiles(5th,10th,90th,and95th)ofthecross-sectional distributionofmonthlybond-levelEBPs.Shadedcolumnscorrespondtoperiodsclassifiedasrecessions by the National Bureau of Economic Research. After implementing this procedure for the bonds in the Lehman-Warga (1973–1998) and ICE (1997–2021) databases whose firm’s balance sheet information and equity prices are available in Compustat and CRSP, respectively, our dataset contains 11,913 bonds from 1,872 firms at a monthly frequency from 1973 to 2021.7 While our focus on bondfinanced firms tilts our sample towards large firms, inspecting firms’ marginal borrowing rates is crucial to understand the transmission of monetary policy to firms’ investment. Further, large firms have been shown to play an outsized role in driving U.S. business cycles (Carvalho and Grassi, 2019). For more details about our dataset, including variable definitions, sample selection and summary statistics, see Appendix A. 2.3 The Cross-Sectional EBP Distribution We document that the cross-sectional EBP distribution displays considerable heterogeneity and contains important information beyond what is reflected by the mean EBP (Gilchrist 7We clean the data as in Gilchrist and Zakrajˇsek (2012); see Appendix A.2 for details. 8

Table 1 Transition Matrix for Monthly Bond-Level EBPs EBP Quintiles ik,t+1 1 2 3 4 5 1 0.85 0.11 0.02 0.01 0.01 EBP 2 0.13 0.67 0.16 0.03 0.02 ik,t Quintiles 3 0.02 0.18 0.62 0.16 0.02 4 0.01 0.04 0.18 0.66 0.11 5 0.01 0.01 0.02 0.13 0.83 Note.Table1providestransitionprobabilitiesformonthlybond-levelEBPsbasedon5states.Entry inrowiandcolumnjreferstotheprobabilityoftransitioningfromstate(quintile)itostate(quintile) j in the subsequent month. Probabilities are calculated as an average over the sample. and Zakrajˇsek, 2012). Figure 1 plots the bond-level cross-sectional EBP distribution over the period 1973–2021. For most of this period, the left-tail percentiles are below zero, indicating that an appreciable segment of bonds receive a discount on their credit spreads relative to their default risk. Left-tail percentiles also have more muted cyclical fluctuations than the mean EBP, with a noticeable rise above zero only during the 2008 crisis. In contrast, right-tail percentiles are not only more volatile than the mean, but are also generally greater than zero. Thus, right-tail firms usually pay a premium on their borrowing costs relative to their default risk, especially in recessions. In what follows, we argue that firm EBPs contain firm-specific information related to their economic prospects. Although the percentiles of the EBP distribution vary considerably over time, a bond’s place within the EBP distribution is reasonably persistent. Table 1 displays the Markov transitionmatrixforbond-levelEBPs.Itshowsthattheprobabilityofabond’sEBPstaying in its quintile in the next month (diagonal entries) is much higher than transitioning to any other quintile, with this result being particularly strong in the lowest and highest quintiles of the distribution. We see this result as necessary, but not sufficient, for firm-level EBPs to encode important information about the economic state of firms. We also document the cross-sectional relationship between firm EBPs and other firm characteristics (Figure 2). Specifically, we focus on the average relationship between the 9

Figure 2 Firm EBPs vs. Firm Characteristics in the Cross-Section Mean EBP in Percentage Points .3 Tobin's Average Q .2 Asset Size Liquid Assets Leverage Age .1 0 1 2 3 4 5 Quintile of Firm Characteristics Note. Figure 2 reports firms’ average EBP (y-axis) in each quintile of the following firm characteristics (x-axis): leverage (debt over assets), liquid assets (cash over assets), age (months since IPO), size (assets), and Tobin’s average Q (market over book value of assets). Lines of lighter colors correspond to 90% confidence intervals. For each firm characteristic, (i) we sort firms into quintiles using the historical average of the characteristic, then (ii) we calculate the average EBP (and associated confidence interval) for the firms in each quintile. EBP and the following variables: leverage (debt over assets), liquid assets (cash over assets), age (time since IPO), size (asset value), and average Tobin’s Q (market over book value of assets). First, there is limited cross-sectional association between firms’ EBPs and their leverage or liquid asset share, two prominent measures of firms’ financial constraints. In contrast, older and larger firms tend to have lower EBPs, suggesting that firms’ age and size encode information beyond the extent of their financial frictions (Crouzet and Mehrotra, 2020). Finally, we see that firms with higher average Tobin’s Q—as calculated from equity prices—generally have lower EBPs, which is consistent with our interpretation of firms’ EBPs as capturing their future investment prospects. Despite these cross-sectional correlations, the results that follow highlight that the information contained in firms’ EBPs are statistically and economically distinct from these other variables. 10

2.4 Common Features of Regression Specifications To estimate the effects of monetary policy conditional on a firm’s characteristic, we follow Jeenas (2019) by averaging the characteristic’s value over the previous year. For example, EBPma denotes the average EBP of firm i’s bond k at time t over the previous year.8 This ikt helps purge uninformative high-frequency variation in our conditioning variables, as well as possible seasonality. Our conclusions, however, are not tied to this particular functional form. In Appendix B.2, we show that our results are robust to conditioning on a dummy variable for whether the value of a firm’s characteristic is above or below the associated median across all firms in a given period (Cloyne et al., 2023, Anderson and Cesa-Bianchi, 2021). For interpretability, we also standardize the conditioning variables to have zero mean and unit variance over the entire sample. We then run local projections (Jorda`, 2005) featuring the interaction between firm characteristics, notably the EBPma, and monetary ikt policy shocks to gauge the heterogenous effects of monetary policy on firm outcomes. Throughout the paper, our specifications include both firm-level and aggregate controls, which we denote by Z . Firm-level controls are leverage, size, sales growth, age, it share of liquid assets, short-term asset share (current over total assets), and Tobin’s average Q. Aggregate controls focus on economic and financial conditions using three lags of the following variables: Chicago Fed’s national activity index for monthly regressions and GDP growth for quarterly regressions, the economic policy uncertainty index of Baker et al. (2016), and the first three principal components of the U.S. Treasury yield curve. Our baseline regressions use macro-financial controls because they allow us to compare the unconditional effect of monetary policy shocks with the effects conditional on firms’ characteristics. That said, our results for the effects of monetary policy conditional on a firm’s EBP are robust to including sector-time fixed effects, as shown in Appendix B.1. Finally, for all panel regressions, the sample is from 1985 to 2021 and inference is conducted using standard errors that are two-way clustered by firm and time period. 8This corresponds to the previous 12 months for monthly data and 4 quarters for quarterly data. 11

3 Monetary Policy and Bond-Level Credit Spreads In this section, we document that expansionary monetary policy shocks decrease credit spreads more for high-EBP bonds than for low-EBP bonds. We also show that the sensitivity of credit spreads to monetary policy shocks is primarily determined by a bond’s EBP, rather its firm’s default risk. Our baseline specification estimates the transmission of monetary policy to bond-level credit spreads both unconditionally and conditional on a bond’s ex-ante EBP. Specifically, we estimate the following regressions at a monthly frequency for a series of horizons h: S −S = βh +βhεm +βhEBPma ×εm +γhZ +e , (4) ikt+h ikt−1 k 1 t 2 ikt−1 t it−1 ikth where S denotes firm i’s bond k credit spread; εm refers to the monetary policy shock ikt t (where positive values reflect easings); EBPma represents firm i’s standardized EBP as ikt−1 conveyed by its bond k; βh is a bond fixed effect; and Z is the vector of control varik it−1 ables described in Section 2.4, plus EBPma . Importantly, EBPma is lagged, as are the ikt−1 ikt−1 controls, to ensure they are not influenced by the contemporaneous monetary policy shock. Figure 3 shows that monetary policy has quantitatively important effects on credit spreads. Panel 3a traces the average response of credit spreads to a surprise monetary easing (βh). We find that a 1 percentage point easing shock induces a decline in the average 1 bond’s credit spreads of nearly 4 percentage points, which occurs eight months after the shock. This result points to a delayed peak effect of monetary policy on firms’ marginal borrowing rates, an issue overlooked by short-horizon studies.9 Panel3bshowsthattheeffectofamonetarypolicyeasingoncreditspreadsislargerfor high-EBP bonds, that is, for firms facing tighter ex-ante financial conditions. In particular, firms whose bonds carry an EBP one standard deviation above the sample mean face an additional decline in their credit spreads of nearly 4 percentage points. Similar to the unconditional effects, this EBP-dependent decline in credit spreads builds up over time, 9This delayed peak effect of monetary policy on bond-level credit spreads is in line with the findings in aggregate studies e.g., Jarocin´ski and Karadi (2020) and Bu et al. (2021). 12

Figure 3 Monetary Policy’s Effect on Bond-Level Credit Spreads (a) Unconditional (b) Conditional on EBP Marginal Effects Marginal Effects 0 0 -2 -2 -4 -4 -6 -6 0 5 10 15 20 0 5 10 15 20 Months after Shock Months after Shock Note. Figure 3 reports the dynamic effects of a monetary policy easing shock εm on the h-month t change in bond credit spreads, S −S , which we estimate using regression (4). Panel 3a ikt+h ikt−1 shows the unconditional effects, βh. Panel 3b shows the effects conditional on EBPma , βh, which 1 ik,t−1 2 measures the additional response of the outcome variable for a firm with a conditioning variable one standard deviation above the sample mean. Inner and outer shaded areas are, respectively, 68% and 90% confidence intervals constructed using two-way clustered standard errors by firm and month. reaching its maximum effect between five and seven months after the shock. We also show that it is mainly the EBP, rather than default risk, that regulates the response of credit spreads to monetary policy. To demonstrate this, we run a “horserace” between our EBP interaction, EBPma ×εm, and a default-risk interaction, xma ×εm: ikt−1 t it−1 t S −S = βh +βhεm +βhEBPma ×εm +βhxma ×εm +γhZ +e , (5) ikt+h ikt−1 k 1 t 2 ikt−1 t 3 it−1 t it−1 ikth where xma is the yearly moving-average of firm i’s default risk, which we measure in two it−1 ways: (i) firm i’s distance to default; and (ii) firm i’s leverage.10 Panels 4a and 4b report the EBP and default-risk interaction coefficients, respectively, when xma is measured by it−1 distance to default, while Panels 4c and 4d do the same for leverage. In both cases, we find that the sensitivity of firms’ credit spreads to monetary policy is primarily a function of their EBPs, rather than their default risk.11 Moreover, the conditional effects by EBP are 10Note that, in this case, both EBPma and xma are also included in Z . ikt−1 it−1 it−1 11To provide comparability with other studies, we also establish that default-risk is a statistically significant state variable for the transmission of monetary policy to credit spread when the EBP is not included as a competing state variable (Appendix B.3). 13

Figure 4 Monetary Policy’s Effect on Bond Credit Spreads: EBP vs. Default Risk (a) Conditional on EBP (b) Conditional on Distance to Default Marginal Effects Marginal Effects 6 0 4 -2 2 -4 0 -6 0 5 10 15 20 0 5 10 15 20 Months after Shock Months after Shock (c) Conditional on EBP (d) Conditional on Leverage Marginal Effects Marginal Effects 0 0 -2 -2 -4 -4 -6 -6 0 5 10 15 20 0 5 10 15 20 Months after Shock Months after Shock Note. Figure 4 reports the dynamic effects of a monetary policy easing shock εm on the h-month t change in bond credit spreads, S −S , which we estimate using two versions of regression ikt+h ikt−1 (5). First, panels 4a and 4b report, respectively, the interaction coefficients on EBPma (βh) and ikt−1 2 our first proxy for default risk xma, the distance to default, (βh). Second, panels 4c and 4d report, it 3 respectively,theinteractioncoefficientsonEBPma (βh)andoursecondproxyfordefaultriskxma, ikt−1 2 it leverage, (βh). In all cases, the interaction terms measure the additional response of the outcome 3 variable for a firm with a conditioning variable one standard deviation above the sample mean. Inner and outer shaded areas are, respectively, 68% and 90% confidence intervals constructed using two-way clustered standard errors by firm and month. largely unchanged relative to our baseline results in Figure 3b. Robustness: We also show that our results are robust to many variants of our empirical approach, including: (i) controlling for time-sector fixed effects (Appendix B.1); (ii) conditioning on the EBP using dummy variables (Appendix B.2); (iii) conditioning on other state variables emphasized in the literature, namely age, liquid asset share, credit rating, 14

Tobin’s average Q, size, and sales growth (Appendix B.4); (iv) using alternative monetary policy shocks (Appendix B.5); and (v) conditioning on an EBP purged of its potential higher-order dependence on distance to default (Appendix B.6). 4 Monetary Policy and Firm-Level Investment Inthissection,wedocumentthatexpansionarymonetarypolicyshocksincreaseinvestment more for low-EBP firms than for high-EBP ones. Thus, conditional on EBP, firms whose investment is more responsive to monetary policy issue bonds whose credit spreads are less responsive. Moreover, we again highlight that the sensitivity of firms’ investment to monetary policy is mainly a function of their EBP, rather than their default risk. To evaluate the dynamic response of firm-level investment to monetary policy, our baseline specification measures both the unconditional effect of a monetary policy shock, aswellastheeffectconditionalonfirms’ex-anteEBP.Specifically,weestimatethefollowing local projections at a quarterly frequency for a series of horizons h: (cid:18) (cid:19) K log it+h = βh +βhεm +βhEBPma ×εm +γhZ +e , (6) K i 1 t 2 it−1 t it−1 ith it−1 where K is the real book value of firm i’s tangible capital stock (as in Ottonello and it Winberry, 2020), EBPma is the average EBPma on firm i’s bonds within a given quarter; it−1 ikt−1 βh are firm fixed effects; and Z is the vector of control variables described in Section i it−1 2.4 plus EBPma . it−1 Figure 5 displays firms’ investment responses to monetary policy shocks. The unconditional response, displayed in Panel 5a, is hump-shaped. Quantitatively, a 1 percentage point monetary easing induces a 10 percentage point increase in investment for the average firm at the peak, which occurs seven quarters after the shock.12 The negative marginal effects in Panel 5b imply that the increase in investment is diminished for firms with higher ex-ante EBPs. This dampened response for higher-EBP firms is economically significant 12The magnitude of this unconditional effect lies between the estimates of Jeenas (2018) and Ottonello and Winberry (2020). 15

Figure 5 Monetary Policy’s Effect on Firm-Level Investment (a) Unconditional (b) Conditional on EBP Marginal Effects Marginal Effects 20 0 10 -10 0 -20 0 4 8 12 0 4 8 12 Quarters after Shock Quarters after Shock Note. Figure 5 reports the dynamic effects of a monetary policy easing shock εm on the h-quarter t cumulative investment of firm i, log(K /K ), which we estimate using regression (6). Panel 5a it+h it−1 shows the unconditional effects, βh. Panel 5b shows the effects conditional on EBPma , βh, which 1 it−1 2 measures the additional response of the outcome variable for a firm with a conditioning variable one standard deviation above the sample mean. Inner and outer shaded areas are, respectively, 68% and 90% confidence intervals constructed using two-way clustered standard errors by firm and quarter. and reaches its largest magnitude ten quarters after the shock.13 We find once more that a firm’s EBP supersedes its default risk as a state-variable for the transmission of monetary policy, this time for investment. As in the previous section, we do so by running a horserace between the interactions of these two firm characteristics with a monetary policy shock: (cid:18) (cid:19) K log it+h = βh +βhεm +βhEBPma ×εm +βhxma ×εm +γhZ +e , (7) K i 1 t 2 it−1 t 3 it−1 t it−1 ith it−1 where default risk xma is again measured in two ways: distance to default and leverage.14 it−1 As shown in Figure 6, the sensitivity of firms’ investment response to monetary policy is primarily a function of their EBPs (Panels 6a and 6c) rather than their default risk (Panels 6b and 6d).15 And, once again, the conditional effects by firm EBP are largely unchanged 13Appendix B.2 presents separate impulse responses for low- and high-EBP firms, and shows they are always either statistically greater than or equal to zero. 14Again, both EBPma and xma are included in Z . it−1 it−1 it−1 15To provide comparability with other studies, we again establish that default-risk is a statistically significant state variable for the transmission of monetary policy to investment when the EBP is not included as a competing state variable (Appendix B.3). 16

Figure 6 Monetary Policy’s Effect on Firm Investment: EBP vs. Default Risk (a) Conditional on EBP (b) Conditional on Distance to Default Marginal Effects Marginal Effects 10 0 -10 0 -20 -10 0 4 8 12 0 4 8 12 Quarters after Shock Quarters after Shock (c) Conditional on EBP (d) Conditional on Leverage Marginal Effects Marginal Effects 10 0 0 -10 -20 -10 0 4 8 12 0 4 8 12 Quarters after Shock Quarters after Shock Note. Figure 6 reports the dynamic effects of a monetary policy easing shock εm on the h-quarter t cumulative investment of firm i, log(K /logK ), which we estimate using 2 versions of regresit+h it−1 sion (7). First, Panels 6a and 6b report, respectively, the interaction coefficients on EBPma (βh) it−1 2 and our first proxy for default risk xma , the distance to default, (βh). Second, panels 6c and 6d it−1 3 report, respectively, the interaction coefficients on EBPma (βh) and our second proxy for default it−1 2 risk xma , leverage, (βh). In all cases, the interaction terms measure the additional response of the it−1 3 outcome variable for a firm with a conditioning variable one standard deviation above the sample mean. Inner and outer shaded areas are, respectively, 68% and 90% confidence intervals constructed using two-way clustered standard errors by firm and quarter. relative to our baseline results. When viewed through the lens of the financial accelerator mechanism presented in other models (e.g., Bernanke et al., 1999 and Ottonello and Winberry, 2020), our results from this section seem at odds with our findings from Section 3. Specifically, we have shown thatwhilefirmsfacingtightfinancialconditions—highEBPs—experiencelargedecreasesin 17

their credit spreads in response to monetary easings, these high-EBP firms increase investment only modestly. Conversely, low-EBP firms experience mild declines in their marginal borrowing costs, and, nevertheless, increase investment considerably.16 The discrepancy between these results and the predictions of financial accelerator models owes to the latter’s emphasis on differences in firms’ default risk and hence the slope of their marginal cost of capital curves. Instead, in the next section, we rationalize our findings with a model in which firms’ heterogeneous responses to monetary policy, conditional on their EBPs, are due to differences in their investment prospects as reflected in their marginal benefit curves. Robustness: We also show that our results are robust to: (i) controlling for time-sector fixed effects (Appendix B.1); (ii) conditioning on EBP using dummy variables (Appendix B.2); (iii) conditioning on other state variables emphasized in the literature: age, liquid asset share, credit rating, Tobin’s average Q, size, and sales growth (Appendix B.4); (iv) using alternative monetary policy shocks (Appendix B.5); and (v) conditioning on an EBP purged of potential higher-order dependence on distance to default (Appendix B.6). 5 Interpretation of Empirical Results In Section 5.1, we build a stylized model in which a firm’s EBP is pinned down by the slope its marginal benefit curve for capital. We provide evidence supporting this result by estimating production functions for low- and high-EBP firms. In Section 5.2, we use our model to demonstrate that the responsiveness of firms’ investment and credit spreads to monetary policy depends on their EBPs in a manner consistent with our empirical results. 5.1 Theoretical Setup: Firm EBPs and Marginal Benefit Curves Our framework focuses on two types of agents: firms who demand capital for production and financial intermediaries who, subject to financial frictions similar to those proposed by Gertler and Kiyotaki (2010) and Gertler and Karadi (2011), supply capital to firms. 16In Appendix B.7, we show that low-EBP firms also borrow more via debt financing than do high-EBP firms in response to a monetary easing, despite the smaller fall in their marginal borrowing costs. 18

Different from previous papers, we highlight the importance of firms’ capital demand for the transmission of the monetary policy. Financial intermediaries are endowed with net worth N and issue deposits D to households (not explicitly modeled here) at an exogenous interest rate R.17 These intermediaries have access to a capital producing technology that transforms N and D on a one-to-one basis into capital K , which they supply to firms for a return R . As long as this re- S K turn on capital exceeds the deposit rate (R > R), intermediaries have an incentive to K leverage-up to increase the return on their equity. Motivated by real-world regulatory capital requirements and risk-management practices, we assume that intermediaries face a constraint that requires them to have sufficient skin in the game when lending to firms. This is modelled as an agency friction in which intermediaries can abscond with a fraction θ of their revenue R K . Similar to Gabaix and Maggiori (2015), we assume that this frac- K S tion is increasing in the size of intermediaries’ balance sheet: θ = θ(K ) and θ(cid:48)(K ) > 0. In S S turn,householdsonlyfundintermediariesthatsatisfyanincentivecompatibilityconstraint: R K −RD ≥ θR K . The optimization problem of the intermediaries is then K S K S max R K −RD s.t. R K −RD ≥ θR K and K = D+N. K S K S K S S KS The solution to the problem above provides a schedule of how much capital intermediaries supply to firms for a given credit spread R /R. We focus on equilibria in which K R ≥ R. When R > R, intermediaries leverage-up until the point in which the skin- K K in-the-game constraint binds. Additionally, when R = R, financial intermediaries are K indifferent between any level of deposits satisfying the skin-in-the-game constraint. Thus, we obtain the following capital supply curve:  R K =   K K S S (1 − − N θ) K S ≥ N θ (8) R  1 K < N, S θ whereK = N/θisthecutoffvalueofcapitalsupplyforwhichtheintermediaries’constraint S 17For simplicity, we set R=1. 19

binds. Importantly, in the region where K ≥ N/θ, the capital supply curve is upward S sloping in credit spreads. Of note, this capital supply curve is also the marginal cost of capital curve (MC) faced by firms. Goods-producing firms use a decreasing returns to scale production technology Kα, D with their profit maximization problem taking the form: max Kα −R K , D K D KD where,asinGertlerandKaradi(2011),firmsborrowatinterestrateR becauseweassume K there are no frictions on their side that limit their access to intermediary funds. The first order condition of this problem yields the marginal benefit curve for capital (MB): R 1 K = αKα−1. (9) R R D As we set α ∈ (0,1), firms’ marginal benefit curves are downward sloping in credit spreads. We focus on the relationship between a firm’s capital intensity α and the level and slope of its marginal benefit curve for capital in equilibrium. While the level of the curve traced in equation (9) reflects firms’ marginal product of capital, the slope is the rate at which this marginal product decreases as firms invest. A flatter-sloped marginal benefit curve implies that, as firms invest, their marginal products of capital remain relatively high compared to their previous equilibrium value. We therefore refer to these firms with flatter marginal benefit curves as having “more-resilient” investment prospects As shown in Figure 7, higher-α firms have flatter marginal benefit curves, and thus more resilient investment prospects. Similar to other papers in the literature (e.g., Anderson and Cesa-Bianchi, 2021), we assume capital markets are segmented by firm type. For simplicity, we consider two islands that are each populated by a continuum of financial intermediaries and firms, which implies capital markets are perfectly competitive on each island. While intermediaries are assumed to be ex-ante identical across islands, we postulate that firms differ across islands in their capital intensities of production, α, leading to differences in the slopes of their 20

Figure 7 Capital Market Equilibrium (a) Low-EBP Firm (b) High-EBP Firm Note. Figure 7 displays the capital market equilibrium on two islands in which firms differ in their capital intensities (α) and hence the slopes of their marginal benefit curves for capital (MB). We set α = 0.90 in Panel 7a and α = 0.71 in Panel 7b, which are calibrated by estimating production functions for firms in the bottom and top quintiles of the EBP distribution using regression (10) (see Table C.5 in Appendix C.3). Panel 7a shows that the high-α firm with the flatter marginal benefit curve has the lower EBP in equilibrium, relative to the low-α in Panel 7b. We describe the calibration of the remainder of the parameters in Appendix C.1. marginal benefit curves for capital. Further, we assume that financial intermediaries supply capital to firms only in their own island. This hypothesis of market segmentation captures the idea that intermediaries such as pension funds and insurance companies may have different lending strategies, due for example to differences in their risk sentiment/appetite for particular assets (e.g., Chernenko and Sunderam, 2012 and Lo´pez-Salido et al., 2017), which lead them to hold portfolios with specific asset shares (e.g., Greenwood and Vissing- Jorgensen, 2018 and Gabaix and Koijen, 2021). Using this model, we demonstrate that variation in capital intensities across firms createsalinkbetweenfirms’EBPsandtheslopesoftheirmarginalbenefitcurvesforcapital. Figure 7 displays the capital market equilibrium on each island. As firms face the same marginal cost curves and markets are segmented across islands, the slope of firms’ marginal benefit curves pins down their equilibrium credit spreads. In particular, firms with higher capital intensities (higher α) have flatter marginal benefit curves and lower credit spreads 21

in equilibrium (Panel 7a), while firms with lower capital intensities have steeper curves and higherspreadsinequilibrium(Panel7b).Sincefirmscarrynodefaultriskinourframework, firms’ credit spreads may be interpreted as EBPs because they arise from intermediaries’ shadow cost of leverage. Thus, the central theoretical result from this section is that, while EBPs arise from financial intermediaries’ leverage constraints (Gilchrist and Zakrajˇsek, 2012), differences in EBPs across firms reflect differences in their investment prospects, with lower EBPs afforded to firms with flatter marginal benefit curves. In Appendix C.2, we show that this result holds for most levels of intermediary net worth.18 We provide empirical evidence to support this theoretical link between firms’ capital intensities and their EBPs by estimating production functions for low- and high-EBP firms. Since firms in our model have access to a Cobb-Douglas production technology with capital as the single input, we estimate the following firm-level panel specification separately for low- and high-EBP firms: logY = β +αlogK +ε , (10) i,t i i,t i,t where output Y is measured as real sales and β is a firm fixed effect. Specifically, we i,t i estimate regression (10) separately for firms whose EBPma are below and above the firmit level median each period. The estimates of α for low- and high-EBP firms from regression (10) are displayed in the first two columns of Table 2. Consistent with our model, we find that the capital intensities of firms with below-median EBPs are significantly larger than those of firms with above-median EBPs, implying that these low-EBP firms have meaningfully flatter marginal benefit curves for capital. Appendix C.3 describes several robustness results to this specification.19 18The only exception is if intermediary net worth is so high that the low-α firm has a credit spread (EBP) of nearly 1, the minimum possible value, which is rare in the data. 19Appendix C.3 provides additional results showing that the α estimates for below- and above-median EBPfirmsarestatisticallydistinctfromeachother.Wealsoshowthattheseresultsarerobusttoincluding both time and sector-time fixed effects, as well as to varying the observation threshold for firms to be includedinthesample.TableC.5providesestimatesofαforfirmsinthebottomandtopterciles,quartiles and quintiles of the EBP distribution. 22

Table 2 Production Function Estimates for Low- and High-EBP Firms (1) (2) (3) (4) logY Low-EBP High-EBP Low-EBP High-EBP i,t logK 0.88∗∗∗ 0.77∗∗∗ 0.19∗∗∗ 0.14 i,t (.037) (.037) (.043) (.099) logM 0.56∗∗∗ 0.59∗∗∗ i,t (.038) (.037) logO 0.27∗∗∗ 0.29∗∗∗ i,t (.020) (.013) Note: Table 2 presents estimates of the capital intensity (α) of firms with EBPma below and it above the firm-level median each period (labeled as “Low-EBP” and “High-EBP”, respectively). The first two columns report estimates from regression (10). The last two columns report estimates from regression (11), which also include the elasticities with respect to intermediate inputs (γ and δ). Standard errors are two-way clustered by firm and quarter in columns (2) and (3) and bootstrapped in columns (4) and (5). *** denotes statistical significance at the 1% level. Further details and robustness can be found in Appendix C.3. While the estimates of α from specification (10) map cleanly to our model, firms’ production functions also include inputs that can be frictionlessly adjusted (Hall, 1986, 1988) and depend on firms’ unobservable idiosyncratic productivity (Olley and Pakes, 1996). Thus, for further robustness, we also estimate the following production functions for lowand high-EBP firms: logY = β +αlogK +ω +γlogM +δlogO +ε , (11) i,t i i,t i,t i,t i,t i,t where ω is firm i’s unobservable idiosyncratic productivity, and M and O are real varii,t i,t i,t able inputs—intermediate goods (e.g., materials) and other operating expenses (including salaries), respectively—which may be correlated with ω . Following Levinsohn and Petrin i,t (2003) and Ackerberg et al. (2015), consistent estimates of the factor elasticities can be achieved by (i) instrumenting the variable inputs with their lags, and (ii) using the variable inputs as proxy variables for unobserved productivity.20 We provide the estimates of regression (11) for below- and above-median EBP firms in the final two columns of Table 2. The main distinction between the production function 20M and O are measured as real cost of goods sold and selling, general and administrative expenses i,t i,t from Compustat, respectively. We use M as the proxy variable. i,t 23

estimates in this case is the capital intensity α. While low-EBP firms’ capital elasticity is statistically greater than zero, that of high-EBP firms is smaller and indistinguishable from zero.21 These findings are thus consistent with those from estimating regression (10). Overall, these empirical findings match the main result of our theoretical model that differences in EBPs across firms reflect differences in the resilience of their investment prospects. Importantly, our conclusions from this section are also consistent with the interpretation from Gilchrist and Zakrajˇsek (2012) for the EBP, since the average EBP across firms captures fluctuations in the aggregate risk-bearing capacity of financial intermediaries and nets out cross-sectional differences in investment prospects (which we show to be important in Sections 3 and 4). 5.2 Monetary Policy Comparative Statics by Firm EBPs We now use our theoretical framework to study how the transmission of monetary policy to firms’ credit spreads and investment depends on their EBPs. Motivated by the large literature documenting monetary policy’s effects on credit supply (e.g., Bernanke and Gertler, 1995, and Kashyap and Stein, 2000), we model a monetary policy easing as an increase in the net worth of financial intermediaries.22 This increase in intermediary net worth leads to a rightward shift in intermediaries’ supply of capital curve, which is also firms’ marginal cost curve, as seen in both panels of Figure 8. Importantly, the response of firms’ investment and credit spreads to this shift in the marginal cost curve depends on the slope of their marginal benefit curves. Specifically, low- EBP firms with flatter marginal benefit curves invest considerably following a monetary easing, despite a relatively mild fall in their credit spreads (Panel 8a). This is due to the relative resilience of these firms’ marginal product of capital, which decreases at a relatively slow rate as they invest. Conversely, high-EBP firms with steeper marginal benefit curves are afforded a larger fall in their credit spreads, but invest relatively little due to the 21Estimatedcapitalelasticitiesareknowntobesmallerwhenincludingotherinputs(Petrinetal.,2004). 22Foraliteraturereviewonthedifferentchannelsofmonetarypolicy,especiallytheintermediarylending channel, see Adrian and Liang (2018). 24

Figure 8 Monetary Policy’s Effect on Credit Spreads and Investment by Firm EBP (a) Low-EBP Firm (b) High-EBP Firm Note.Figure8presentsthecomparativestaticstoamonetarypolicyeasing,modelledasanincrease inintermediaries’networth,forfirmsontwoislandsthatdifferintheircapitalintensitiesandhence in the slopes of their marginal benefit curves and their EBPs. We set α = 0.90 in Panel 7a and α = 0.71 in Panel 7b, which are calibrated by estimating production functions for firms in the bottom and top quintiles of the EBP distribution using regression (10) (see Table C.5 in Appendix C.3). We describe the calibration of the remainder of the parameters in Appendix C.1. rapid depletion of their sufficiently productive investment opportunities (Panel 8b). This comparative statics exercise rationalizes our empirical results for the sensitivity of firms’ investment and credit spreads to monetary policy, conditional on their EBPs, by appealing to the slope of their marginal benefit curves for capital. While our model is intentionally simple to emphasize the role of heterogeneity in firms’ marginal benefit curves for the transmission of monetary policy, our findings in Sections 5.1 and 5.2 are robust to two important extensions. First, if financial intermediaries across islands differ in either their net worth N or lending constraints (sentiment) θ, our main conclusions are unchanged provided that it is the higher-α firms that face the outward-shifted or flatter marginal cost curves coming from higher net worth/less constrained financial intermediaries. In such a case, the higher-α firms would have an even lower EBP—consistent withbothourmodel’spredictionandtheproductionfunctionestimatesofSection5.1—and would be on an even flatter segment of their marginal benefit curves—thereby amplifying 25

(dampening) the responsiveness of their investment (credit spreads) relative to the results of Section 5.2. Second, our main results are also unchanged if firms differ across islands in their level of default risk, provided that it is higher-α firms that have lower default risk. Specifically, similar to the first extension above, lower default risk would flatten the marginal cost of capital curves faced by the higher-α firms (see Ottonello and Winberry, 2020) and thus place them on a flatter segment of their marginal benefit curves. Such a relationship between default risk and investment prospects (α) would also allow our model to match the positive correlation between firms’ credit spreads and EBPs in the data. 6 Micro- and Macro-economic Implications In this section, we test two implications of our model—one at the firm-level and one in the aggregate—to provide further support our model’s mechanism. Specifically, we show that a firm’s EBP regulates the sensitivity of its investment to movements in its credit spreads (Section 6.1) and that the cross-sectional EBP distribution determines the aggregate effectiveness of monetary policy (Section 6.2). This second result suggests a granular origin for monetary policy’s time-varying aggregate effects. 6.1 Firm-level Credit Spreads and Investment ThetheoreticalframeworkoutlinedinSection5.1illustratesthattheslopeoffirms’marginal benefit curves matters not just for their sensitivity to monetary policy, but more generally fortheirresponsivenesstoanyshiftintheirmarginalcostcurves.Inthissection,webuildon the well-documented negative correlation between firms’ credit spreads and their future investment (e.g. Gilchrist and Zakrajˇsek, 2007), which is consistent with credit supply shocks being dominant in capital markets. Specifically, we investigate how this spreads-investment relationship depends on our proxy for the slope of firms’ marginal benefit curves—their EBPs. We find that increases in credit spreads are associated with smaller declines in investment for high-EBP firms. 26

Figure 9 Firm-Level Credit Spreads and Investment (a) Unconditional (b) Conditional on EBP Marginal Effects Marginal Effects 1.5 0 1 -2 -4 .5 -6 0 0 4 8 12 0 4 8 12 Quarters after Shock Quarters after Shock Note. Figure 9 reports the dynamic response of the h-quarter cumulative investment of firm i, log(K /logK ),toachangeinfirmi’screditspread∆S ,whichweestimateusingregression it+h it−1 i,t (12).Figure9ashowsunconditionaleffects,βh.Figure9bshowseffectsconditionalonEBPma ,βh, 1 it−1 2 which measures the additional response of the outcome variable for a firm with a conditioning variable one standard deviation above the sample mean. Inner and outer shaded areas are, respectively, 68% and 90% confidence intervals constructed using two-way clustered standard errors by firm and quarter. To show this, we estimate quarterly firm-level regressions of investment on changes in credit spreads, using a firm’s ex-ante EBP as a state variable: (cid:18) (cid:19) K log it+h = βh +βh∆S +βh∆S ×EBPma +γhZ +e , (12) K i 1 it 2 it it−1 it−1 ith it−1 where Z includes the controls discussed in Section 2.4, plus EBPma . As before, Our it−1 it−1 results are robust to including time-sector fixed effects (Appendix B.1), to conditioning on EBP using dummy variables (Appendix B.2), and to conditioning on an EBP purged of its potential higher-order dependence on distance to default (Appendix B.6). Consistent with credit supply being the primary source of variation in bond markets, Panel 9a highlights that increases in firms’ credit spreads are associated with significant and persistent declines in their investment. Furthermore, Panel 9b highlights that increases in credit spreads predict less-pronounced declines in investment for firms with higher EBPs, that is, for those with steeper marginal benefit curves. While many papers have explored the firm-level relationship between credit spreads and investment (e.g., Gilchrist et al., 27

2014 and Lin et al., 2018), we document which firms’ investment is most responsive to movements in their marginal borrowing costs by conditioning on firms’ EBPs. These results highlight that the slope of a firm’s marginal benefit curve is a key determinant of its responsiveness to changes in credit supply. To further support this claim, Appendix B.8 uses the intermediary capital risk factor from He et al. (2017) as a source of exogenous variation in credit supply. Consistent with our model, we find that the responses of credit spreads and investment, conditional on firms’ EBPs, from this shock to intermediary net worth are qualitatively similar to those from a monetary policy shock. 6.2 EBP Distribution and Monetary Policy’s Aggregate Effects In this section, we provide evidence that the cross-sectional distribution of firm EBPs, which are tied to the slopes of firms’ marginal benefit curves in our model, determines the effectiveness of the transmission of monetary policy to the macroeconomy. Figure 10 displays one motivation for our investigation, showing a considerable shift in mass from the left tail to the right tail of the EBP distribution during recessions. Further, the distribution of firms’ EBPs should be particularly relevant for monetary policy in the aggregate since our firm-level results are based on a sample titled towards large (bond-financed) firms, who play an outsized role in driving the U.S. business cycle (Carvalho and Grassi, 2019). Our argument extends the framework from Section 5.1 with two types of firms to one with a continuum of firms that differ in their EBPs. In such a heterogeneous firm setup, the response of aggregate investment to monetary policy would depend on the crosssectional distribution of firm EBPs. Specifically, monetary policy should be less effective at stimulating aggregate investment when a larger mass of firms are on a steeper segment of their marginal benefit curves (higher EBPs) and more effective when a larger mass of firms are on a flatter segment (lower EBPs). To evaluate these predictions, we use local projections similar to those from previous sections, but with two important modifications: (i) we use annualized U.S. aggregate investment growth as our dependent variable, and (ii) we use the first three cross-sectional 28

Figure 10 Bond-level EBP Distribution in Recessions and Expansions Density 1 .8 Expansions .6 Recessions .4 .2 0 -4 -2 0 2 4 EBP Note.Figure10presentskernel-densityestimatedbond-levelEBPdistributionsduringNBER-classified recessions and expansions over the period 1973:M1 to 2021:M12. moments of the EBP distribution as state variables. Greater skewness, all else equal, implies a shift in mass from the left tail to the right tail of the EBP distribution, which should render the transmission of monetary policy to aggregate investment less potent. Similarly, allelseequal,ahighermedianEBPimpliesarightwardshiftoftheEBPdistribution,which should also make monetary policy less effective. Conversely, while the effect of a more dispersed EBP distribution is ex-ante ambiguous, it provides an indication of whether firm EBPs in the left or right tail exert a greater influence over the aggregate effectiveness of monetary policy. Specifically, we estimate the following local projection at a quarterly frequency: (cid:18) (cid:19) 400 I log t+h =βh +βhεm +βhMma ×εm +δhY +e , (13) h+1 I 0 1 t 2 t−1 t l t−1 th t−1 whereI isU.S.aggregateinvestment,Mma isavectorthatcontainsthemedian,dispersion t t−1 and Kelly-skewness of the bond-level cross-sectional EBP distribution, and Y includes t−1 the aggregate controls of Section 2.4 plus the vector Mma. 23 In our baseline, we measure t−1 23Forthisregression,wesubstituteGDPgrowthforinvestmentgrowthintheaggregatecontrolsY to t−1 againalignourselveswiththeexistingliterature(e.g.,GilchristandZakrajˇsek,2012).Forthesamereason, 29

Figure 11 Monetary Policy’s Effect on Aggregate Investment Growth (a) Unconditional (b) Conditional on EBP Skewness Marginal Effects Marginal Effects 30 15 15 0 0 -15 -15 -30 0 4 8 12 0 4 8 12 Quarters after Shock Quarters after Shock (c) Conditional on Median EBP (d) Conditional on EBP Dispersion Marginal Effects Marginal Effects 15 45 0 30 -15 15 -30 0 -45 -15 0 4 8 12 0 4 8 12 Quarters after Shock Quarters after Shock Note.Figure11reportsthedynamiceffectsofamonetarypolicyeasingshockεm onh-quarterannut alized aggregate investment growth, 400/(h+1)log(I /I ), which we estimate using regression t+h t−1 (13). Panel 11a shows unconditional effects, βh. Panels 11b, 11c and 11d show the effects condi- 1 tional on the skewness, median and dispersion of the EBP distribution, the three elements in βh, 2 respectively. The conditional effects measure the additional response of the outcome variable when the conditioning variable is one standard deviation above the sample mean. Inner and outer shaded areas are, respectively, 68% and 90% confidence intervals constructed using Newey-West standard errors with 12 lags. dispersion and skewness using 10th and 90th percentiles of the EBP distribution. We use Newey-West standard errors with 12 lags. The results are displayed in Figure 11 and are consistent with the predictions of our model. First, Panel 11a traces the unconditional response of aggregate investment growth we use annualized aggregate investment growth as our dependent variable, with similar results emerging if we use the level of aggregate investment. 30

to a monetary easing shock. As expected, investment growth increases in a hump-shaped fashion, with a peak response 6 quarters after the shock. Panels 11b, 11c and 11d chart the effects of a monetary policy easing shock conditional on the skewness, median, and dispersion of the EBP distribution, respectively. Focusing first on skewness, the negative marginal effects highlight that a more right-skewed EBP distribution dampens the effects of monetary policy on aggregate investment growth. Similarly, a higher median EBP also lessens the potency of monetary policy. Finally, a more dispersed EBP distribution amplifies the transmission of monetary policy, suggesting that the added stimulus from a lower left tail of the EBP distribution overcomes the drag from a higher right tail. Thus, the investment prospects of left-tail EBP firms, those with the flattest marginal benefit curves, are more responsible for the transmission of monetary policy to the macroeconomy.24 Overall, these findings highlight the macroeconomic significance of our firm-level results.25 Finally, Appendix B.9 shows that the results from this section are general and are not tied solely to business cycle variation. Specifically, we find that the aggregate effects of monetary policy conditional on the skewness of the EBP distributions are unchanged when controlling for the interaction between monetary policy shocks and recession indicators similar to those used by Tenreyro and Thwaites (2016). This result is consistent with the amplification mechanism for monetary policy we argue for in this paper—the slope of firms’ marginal benefit curves for capital—being distinct from the decreased power of monetary policy in recessions. 7 Conclusion We examine how and why the responsiveness of firms’ credit spreads and investment to monetary policy depends on their financial conditions, as measured by their EBPs. Our paper has three main parts. First, using a comprehensive bond- and firm-level database, we find that while expansionary monetary policy shocks compress credit spreads more for 24We provide further support for this by interacting our monetary policy shock with the percentiles of the EBP distribution in Appendix B.9. 25Ourresultsarerobusttousingalternativemonetarypolicyshocksandtomeasuringthecross-sectional moments using different percentiles (see Appendix B.9). 31

firms with ex-ante higher EBPs, it is firms with lower EBPs that invest more. Second, we rationalize these results using a model in which firms with flatter curves of marginal benefit for capital—i.e., with marginal products of capital that diminish relatively slowly as they invest—have lower EBPs. While these EBPs arise from financial frictions, differences in EBPs across firms therefore reflect differences in the resilience of their investment prospects, which offers a new economic mechanism for monetary policy’s heterogeneous effects. Third, we provide additional empirical support for the importance of firms’ marginal benefit curves for the transmission of economic shocks. Most importantly, we show that the effect of monetary policy on aggregate investment depends on the moments, in particular the skewness, of the cross-sectional EBP distribution. Policymakers and researchers often discuss three key aspects of the transmission of monetary policy: its distributional effects, its aggregate potency, and the channels through which it operates. Our paper contributes to this debate. On the distributional front, we showthatmonetarypolicyislesseffectiveatstimulatingtheinvestmentoffirmswithhigher EBPs, due to their steeper marginal benefit curves. On the aggregate front, our paper not only provides a theoretical argument for monetary policy’s time-varying effects, but also offers a specific observable—the cross-sectional EBP distribution—to monitor them. On the modelling front, our paper provides new empirical evidence on the salience of firms’ marginal benefit curves to feed the construction of richer models of the macroeconomy. 32

References Abel, Andrew B, Investment and the Value of Capital number 65, Dissertations-G, 1979. Ackerberg, Daniel A, Kevin Caves, and Garth Frazer, “Identification properties of recent production function estimators,” Econometrica, 2015, 83 (6), 2411–2451. Adrian, Tobias and Nellie Liang, “Monetary Policy, Financial Conditions, and Financial Stability,” International Journal of Central Banking, 2018, 14 (1), 73–131. Anderson, Gareth and Ambrogio Cesa-Bianchi, “Crossing the Credit Channel: Credit Spreads and Firm Heterogeneity,” working paper, Bank of England, 2021. Bahaj, S., A. Foulis, G. Pinter, and P. Surico, “Employment and the residential collateral channel of monetary policy,” Journal of Monetary Economics, 2022, 131, 26– 44. Baker, Scott R, Nicholas Bloom, and Steven J Davis, “Measuring economic policy uncertainty,” The Quarterly Journal of Economics, 2016, 131 (4), 1593–1636. Bauer, Michael D and Eric T Swanson,“TheFed’sresponsetoeconomicnewsexplains the” Fed information effect”,” Technical Report, National Bureau of Economic Research 2020. Bernanke, Ben S and Mark Gertler, “Agency Costs, Net Worth, and Business Fluctuations,” The American Economic Review, 1989, 79 (1), 14–31. and , “Inside the black box: the credit channel of monetary policy transmission,” Journal of Economic Perspectives, 1995, 9 (4), 27–48. , , and Simon Gilchrist, “The financial accelerator in a quantitative business cycle framework,” Handbook of Macroeconomics, 1999, 1, 1341–1393. Bharath, Sreedhar T and Tyler Shumway, “Forecasting default with the Merton distance to default model,” The Review of Financial Studies, 2008, 21 (3), 1339–1369. Black, Fischer and Myron Scholes, “The pricing of options and corporate liabilities,” Journal of Political Economy, 1973, 81 (3), 637–654. 33

Bu, Chunya, John Rogers, and Wenbin Wu, “A unified measure of Fed monetary policy shocks,” Journal of Monetary Economics, 2021, 118, 331–349. Caglio, Cecilia R, R Matthew Darst, and Sebnem Kalemli-Ozcan, “Risk-taking and monetary policy transmission: Evidence from loans to smes and large firms,” Technical Report, National Bureau of Economic Research 2021. Carvalho, Vasco M and Basile Grassi, “Large firm dynamics and the business cycle,” American Economic Review, 2019, 109 (4), 1375–1425. Chauvet, Marcelle, “An Econometric Characterization of Business Cycle Dynamics with Factor Structure and Regime Switching,” International Economic Review, 1998, 39 (4), 969–96. Chava, Sudheer and Alex Hsu, “Financial constraints, monetary policy shocks, and the cross-section of equity returns,” The Review of Financial Studies, 2020, 33 (9), 4367– 4402. Chernenko, Sergey and Adi Sunderam, “The real consequences of market segmentation,” The Review of Financial Studies, 2012, 25 (7), 2041–2069. Clark, John Bates, The Distribution of Wealth: A Theory of Wages, Interest and Profits, Macmillan, 1899. Cloyne, James, Clodomiro Ferreira, Maren Froemel, and Paolo Surico, “Monetary policy, corporate finance and investment,” Journal of the European Economic Association, 2023, p. jvad009. Crouzet, Nicolas and Neil R Mehrotra, “Small and large firms over the business cycle,” American Economic Review, 2020, 110 (11), 3549–3601. Durante, A., A. Ferrando, and P. Vermeulen, “Monetary policy, investment and firm heterogeneity,” European Economic Review, 2022, 148. Favara, G, S Gilchrist, KF Lewis, and E Zakrajˇsek, “Updating the recession risk and the excess bond premium (No. 2016-10-06),” Board of Governors of the Federal Reserve System (US), 2016. 34

Fisher, Irving, Theory of interest: as determined by impatience to spend income and opportunity to invest it, Augustusm Kelly Publishers, Clifton, 1930. Gabaix, Xavier and Matteo Maggiori, “International liquidity and exchange rate dynamics,” The Quarterly Journal of Economics, 2015, 130 (3), 1369–1420. and Ralph SJ Koijen, “In search of the origins of financial fluctuations: The inelastic markets hypothesis,” Technical Report, National Bureau of Economic Research 2021. Gertler, Mark and Nobuhiro Kiyotaki, “Financial intermediation and credit policy in business cycle analysis,” in “Handbook of Monetary Economics,” Vol. 3, Elsevier, 2010, pp. 547–599. and Peter Karadi,“Amodelofunconventionalmonetarypolicy,”Journal of Monetary Economics, 2011, 58 (1), 17–34. and Simon Gilchrist, “Monetary Policy, Business Cycles, and the Behavior of Small Manufacturing Firms,” Quarterly Journal of Economics, 1994, 109 (2), 309–340. Gilchrist, Simon and Egon Zakrajˇsek, “Investment and the cost of capital: New evidence from the corporate bond market,” 2007. and , “Credit spreads and business cycle fluctuations,” American Economic Review, 2012, 102 (4), 1692–1720. , Jae W Sim, and Egon Zakrajˇsek, “Uncertainty, financial frictions, and investment dynamics,” Technical Report, National Bureau of Economic Research 2014. Gonz´alez, B., G. Nun˜o, D. Thaler, and S. Albrizio, “Firm Heterogeneity, Capital Misallocation and Optimal Monetary Policy,” Technical Report, CESifo, Munich 2021. Greenwood, Robin and Annette Vissing-Jorgensen, “The Impact of Pensions and Insurance on Global Yield Curves,” 2018. Unpublished. Gu¨rkaynak, Refet, Hatice G¨ok¸ce Karasoy-Can, and Sang Seok Lee, “Stock market’s assessment of monetary policy transmission: The cash flow effect,” The Journal of Finance, 2022, 77 (4), 2375–2421. 35

Gu¨rkaynak, Refet S, Brian Sack, and Jonathan H Wright, “The US Treasury yield curve: 1961 to the present,” Journal of Monetary Economics, 2007, 54 (8), 2291–2304. Hall, Robert E., “Market Structure and Macroeconomic Fluctuations,” Brookings Papers on Economic Activity, 1986, 17 (2), 285–338. , “The relation between price and marginal cost in US industry,” Journal of political Economy, 1988, 96 (5), 921–947. Hayashi, Fumio, “Tobin’s marginal q and average q: A neoclassical interpretation,” Econometrica, 1982, pp. 213–224. Hayek, Friedrich August, The pure theory of capital, Routledge, 1941. He, Zhiguo, Bryan Kelly, and Asaf Manela,“Intermediaryassetpricing:Newevidence from many asset classes,” Journal of Financial Economics, 2017, 126 (1), 1–35. Ippolito, Filippo, Ali K Ozdagli, and Ander Perez-Orive, “The transmission of monetary policy through bank lending: The floating rate channel,” Journal of Monetary Economics, 2018, 95, 49–71. Jarocin´ski, Marek and Peter Karadi, “Deconstructing monetary policy surprises—the role of information shocks,” American Economic Journal: Macroeconomics, 2020, 12 (2), 1–43. Jeenas, Priit, “Monetary policy shocks, financial structure, and firm activity: A panel approach,”Financial Structure, and Firm Activity: A Panel Approach (January 5, 2018), 2018. , “Firm Balance Sheet Liquidity, Monetary Policy Shocks, and Investment Dynamics,” working paper, Pompeu Fabra, 2019. and Ricardo Lagos, “Q-Monetary Transmission,” Technical Report, National Bureau of Economic Research 2022. Jord`a, O`scar,“Estimationandinferenceofimpulseresponsesbylocalprojections,”American Economic Review, 2005, 95 (1), 161–182. 36

, Moritz Schularick, and Alan M Taylor, “The effects of quasi-random monetary experiments,” Journal of Monetary Economics, 2020, 112, 22–40. Jorgenson, Dale Weldeau, “Investment: Capital theory and investment behavior,” American Economic Review, 1963, 53. Kashyap, Anil K and Jeremy C Stein, “What do a million observations on banks say about the transmission of monetary policy?,” American Economic Review, 2000, 90 (3), 407–428. , Owen A Lamont, and Jeremy C Stein,“Creditconditionsandthecyclicalbehavior of inventories,” The Quarterly Journal of Economics, 1994, 109 (3), 565–592. Keynes, J. M., The General Theory of Employment, Interest and Money, Macmillan, 1936. 14th edition, 1973. Kiyotaki, Nobuhiro and John Moore, “Credit cycles,” Journal of Political Economy, 1997, 105 (2), 211–248. Lakdawala, Aeimit and Timothy Moreland,“Monetarypolicyandfirmheterogeneity: The role of leverage since the financial crisis,” Available at SSRN 3405420, 2021. Lerner, Abba P, “On the marginal product of capital and the marginal efficiency of investment,” Journal of Political Economy, 1953, 61 (1), 1–14. Levinsohn, James and Amil Petrin, “Estimating production functions using inputs to control for unobservables,” The review of economic studies, 2003, 70 (2), 317–341. Lian, Chen and Yueran Ma, “Anatomy of corporate borrowing constraints,” The Quarterly Journal of Economics, 2021, 136 (1), 229–291. Lin, Xiaoji, Chong Wang, Neng Wang, and Jinqiang Yang, “Investment, Tobin’sq, and interest rates,” Journal of Financial Economics, 2018, 130 (3), 620–640. L´opez-Salido, David, Jeremy C Stein, and Egon Zakrajˇsek, “Credit-market sentiment and the business cycle,” The Quarterly Journal of Economics, 2017, 132 (3), 1373–1426. 37

Lucas, Robert E and Edward C Prescott, “Investment under uncertainty,” Econometrica, 1971, pp. 659–681. McKay, Alisdair and Johannes F Wieland, “Lumpy durable consumption demand and the limited ammunition of monetary policy,” Econometrica, 2021, 89 (6), 2717–2749. Merton, Robert C, “On the pricing of corporate debt: The risk structure of interest rates,” The Journal of finance, 1974, 29 (2), 449–470. Miranda-Agrippino, Silvia, “Unsurprising shocks: information, premia, and the monetary transmission,” 2016. Olley, G Steven and Ariel Pakes, “The Dynamics of Productivity in the Telecommunications Equipment,” Econometrica, 1996, 64 (6), 1263–1297. Ottonello, Pablo and Thomas Winberry,“FinancialHeterogeneityandtheInvestment Channel of Monetary Policy,” Econometrica, 2020, 88 (6), 2473–2502. Ozdagli, Ali K, “Financial frictions and the stock price reaction to monetary policy,” The Review of Financial Studies, 2018, 31 (10), 3895–3936. Palazzo, Berardino and Ram Yamarthy, “Credit risk and the transmission of interest rate shocks,” Journal of Monetary Economics, 2022, 130, 120–136. Petrin, Amil, Brian P Poi, and James Levinsohn,“Productionfunctionestimationin Stata using inputs to control for unobservables,” The Stata Journal, 2004, 4 (2), 113–123. Philippon, Thomas, “The bond market’s q,” The Quarterly Journal of Economics, 2009, 124 (3), 1011–1056. Ramey, Valerie A, “Macroeconomic shocks and their propagation,” Handbook of Macroeconomics, 2016, 2, 71–162. Swanson, Eric T, “Measuring the effects of federal reserve forward guidance and asset purchases on financial markets,” Journal of Monetary Economics, 2021, 118, 32–53. 38

Tenreyro, Silvana and Gregory Thwaites, “Pushing on a string: US monetary policy is less powerful in recessions,” American Economic Journal: Macroeconomics, 2016, 8 (4), 43–74. Tobin, James, “A general equilibrium approach to monetary theory,” Journal of Money, Credit and Banking, 1969, 1 (1), 15–29. Vassalou, Maria and Yuhang Xing, “Default risk in equity returns,” The Journal of Finance, 2004, 59 (2), 831–868. Vavra, Joseph, “Inflation dynamics and time-varying volatility: New evidence and an ss interpretation,” The Quarterly Journal of Economics, 2014, 129 (1), 215–258. Wu, Wenbin, “The credit channel at the zero lower bound through the lens of equity prices,” Journal of Money, Credit and Banking, 2018, 50 (2-3), 435–448. 39

Internet Appendix (Intended for online publication only) Firm Financial Conditions and the Transmission of Monetary Policy by T. Ferreira, D. Ostry, J. Rogers May 31, 2023 A Data Summary A.1 Monetary Policy Shocks A.2 Variable Definitions and Sample Selection A.3 Calculating Distance to Default and the EBP A.4 Summary Statistics B Additional Empirical Results and Robustness B.1 EBP Heterogeneity with Sector-Time Fixed Effects B.2 EBP Heterogeneity using Dummy Variables B.3 Default-Risk as a State Variable B.4 Monetary Policy’s Effect by EBP vs. other Characteristics B.5 Alternative Monetary Policy Shocks B.6 Heterogeneity by EBP purged of Higher-Order Default Risk B.7 Monetary Policy’s Effect on Firm-Level Debt Issuance

B.8 Intermediary Net Worth Shocks and EBP Heterogeneity B.9 Aggregate Implications of EBP Heterogeneity C Model Appendix C.1 Model Parameterization C.2 Firm EBPs and Marginal Benefit Curves in the Model C.3 Firm EBPs and Marginal Benefit Curves in the Data C.4 Firm EBPs and Capital Stock: Model and Data 1

A Data Summary In this section, we present further details on our baseline monetary policy shock series (AppendixA.1),providevariabledefinitionsandoutlineoursample(AppendixA.2),discuss in more detail the EBP and distance to default calculations (Appendix A.3), and provide summary statistics for our main variables of interest (Appendix A.4). A.1 Monetary Policy Shocks This section provides more details about the Bu, Rogers and Wu (2021) monetary policy shocks, which we use in our baseline specifications throughout the paper. The start-date of our sample is January 1985, while the end-date is December 2021. Figure A.1 shows the times series of shocks at a monthly frequency. This “extended” series is longer than the original series of the paper, which runs from January 1994 to September 2019. Figure A.1 Monetary Policy Shocks Monetary Policy Shocks in Percentage Points .15 .1 .05 0 -.05 -.1 -.15 1990m1 2000m1 2010m1 2020m1 Note. Figure A.1 plots the time series of Bu et al. (2021) monetary policy shocks at a monthly frequencyfromJanuary1985toJuly2021.Positivevalueshererepresenttightenings.Shadedcolumns represent periods classified as recessions by the National Bureau of Economic Research. 2

As discussed in the original paper, the Bu et al. (2021) monetary policy shocks are constructed using a two-step Fama-Macbeth procedure with identification achieved via a heteroskedasticity-based instrumental variable approach. The resulting shocks display a moderately-high correlation with other shock series in the literature, but have a number of notable properties: (i) they stably bridges periods of conventional and unconventional policy,providinguswithasignificantlylargersamplethanotherempiricalworkinthisarea; (ii) they are devoid of the central bank information effects; and (iii) they are unpredictable fromtheinformationsetavailableatthetimeoftheshock.Thatsaid,asshowninAppendix B.5, our results are robust to using the Swanson (2021) shocks. For more details on the calculation of the Bu et al. (2021) shock series, see the original paper. Summary statistics for the Bu et al. (2021) monetary policy shock series are presented in Appendix A.4. A.2 Variable Definitions and Sample Selection In this subsection, we first define the variables used in our paper and then discuss our sample. All variable definitions are standard in the literature; we draw particularly on those used in Ottonello and Winberry (2020) and Gilchrist and Zakrajˇsek (2012). The variables are: 1. Real Investment: defined as log(K it+h) for h = 0,1,2..., where K denotes the book Kit−1 it−1 value of the nominal capital stock of firm i at the end of period t−1 deflated by the BLS implicit price deflator (IPDNBS in FRED database). This is the same timing convention as Ottonello and Winberry (2020), although they label the real capital stock of firm i at the end of period t−1 as K . As in Ottonello and Winberry (2020), for each firm, we set it the first value of their nominal capital stock to be the level of gross plant, property, and equipment (ppegtq in Compustat) in the first period in which this variable is reported in Compustat. From this period onwards, we compute the evolution of the capital stock using the changes of net plant, property, and equipment (ppentq in Compustat), which is a measure of net of depreciation investment with significantly more observations than ppegtq. If a firm has a missing observation of ppentq located between two periods with non-missing observations we estimate its value by linear interpolation. We consider only 3

investment spells of 20 quarters or more. 2. Credit spread: defined as S = y −yT, where y is the yield quoted in the secondary ikt ikt t ikt market of corporate bond k issued by firm i in month t from the Lehman-Warga and ICE databases and yT is the yield on a U.S. Treasury with the exact same maturity as t the corporate bond k, using estimates from Gu¨rkaynak et al. (2007). 3. Distance to default: firm’s expected default risk defined by Merton (1974) model. Calculated as in Gilchrist and Zakrajˇsek (2012); see Appendix A.3 for further details. ˆ ˆ 4. EBP: defined as EBP = S −S where S is the predicted value of firm i’s bond ikt ikt ikt ikt k credit spread at time t, which as in Gilchrist and Zakrajˇsek (2012), is calculated from a regression of log(S ) on firm i’s distance to default and bond k’s characteristics. See ikt Appendix A.3 for further details. 5. Leverage: defined as the ratio of total debt (sum of dlcq and dlttq in Compustat) to total assets (atq in Compustat). 6. Share of liquid assets: defined as the ratio of cash and short-term investments (cheq in Compustat) to total assets (atq in Compustat), as in Jeenas (2019). 7. Size: measured as total assets (atq in Compustat) deflated using the BLS implicit price deflator (IPDNBS in FRED database). 8. Sales growth: measured as the log-difference of sales (saleq in Compustat) deflated using the BLS implicit price deflator (IPDNBS in FRED database). 9. Age: defined as age since initial public offering (begdat in Compustat). 10. Tobin’s (average) Q: defined as the ratio market value of assets to book value of assets. Market value of assets is equal to (i) book value of assets (atq in Compustat) plus (ii) market capitalization (share price times oustanding shares) minus common equity plus deferred taxes ((prccq * cshoq) - ceqq + txditcq, in Compustat), as in Cloyne et al. (2023). Since txditcq is sparsely available and is also a relatively small component of Tobin’s q, we impute the value to be zero if an observation is missing. 4

11. Short-Term Assets: defined as the ratio of current assets (actq in Compustat) to total assets (atq in Compustat). 12. Sectors: we use 4-digit SIC codes. 13. GDPandAggregateInvestment:measuredasrealchainedgrossdomesticproduct(GDPC1 in FRED) and real chained investment (RINV in FRED). Growth rates calculated as log-differences. Sample selection: we focus on the non-financial firms whose equity prices are available in the Center for Research in Security Prices (CRSP) database, whose balance sheets are available from the CRSP/Compustat Merged Database, Wharton Research Data Services and whose bond yields data are available in the Arthur D. Warga, Lehman Brothers Fixed Income Database and the Interactive Data Corporation, ICE Pricing and Reference Data. To clean the data, similar to Gilchrist and Zakrajˇsek (2012), we first drop bond-time observations that display any of the following characteristics: they are puttable; they have spreadslargerthan35%orbelow0%;theyhavearesidualmaturityoflessthan6monthsor more than30 years. Afterthis, we drop bonds that have nospellsof at leastone yearof consecutiveobservations. Wethenmergethisbond-leveldatasetwiththefirm-levelCompustat and CRSP databases for non-financial firms. To determine whether a firm is non-financial, we make use of both their NAICS/SIC code as well as the classification scheme internal to the Lehman-Warga/ICE databases. Specifically, if the NAICS/SIC code is available, we exclude those firms classified as financial according to their NAICS/SIC code; otherwise, we exclude firms classified as financial according to the Lehman-Warga/ICE databases. A.3 Calculating Distance to Default and the EBP Our starting point is the credit spread S for bond k issued by firm i at time t, which we ikt calculate in a similar fashion to Gilchrist and Zakrajˇsek (2012). Figure A.2 plots the time series of our mean credit spread and that of Gilchrist and Zakrajˇsek (2012) and highlights that the correlation is 96%. 5

Figure A.2 Credit Spreads: Comparison with Gilchrist and Zakrajˇsek (2012) EBP Percentiles in Percentage Points 10 Corr(SpreadFOR , SpreadGZ) = 0.96 8 6 4 2 0 SpreadFOR SpreadGZ 1973m1 1980m1 1990m1 2000m1 2010m1 2020m1 Note. Figure A.2 compares the mean credit spread calculated in this paper, in red, with the mean creditspreadcalculatedbyGilchristandZakrajˇsek(2012),inblue.Shadedcolumnsrepresentperiods classified as recessions by the National Bureau of Economic Research. To derive each bond’s EBP , as discussed in the main text, following Gilchrist and ikt Zakrajˇsek (2012), we estimate: logS = βDD +γ(cid:48)Z +υ , (A.1) ikt it ikt ikt where DD is firm i’s distance to default (Merton, 1974), and Z includes: (i) the bond’s it ikt duration, age, par value, coupon rate (all in logs); (ii) a dummy for if the bond is callable; (iii) interactions between the characteristics listed in (i) and the call dummy in (ii); (iv) interactions between the call dummy in (ii) and DD , the first three principal components it of the U.S. Treasury yield curve, and the volatility of the 10-year Treasury yield; and (v) industry and credit rating fixed effects. Table A.1 provides the results from estimating regression (A.1). We discuss how we calculate DD later in this section. it ˆ Assuming the error term is normally distributed, the predicted spread S is given by: ikt (cid:104) σˆ2(cid:105) S ˆ = exp β ˆ DD +γˆ(cid:48)Z + , (A.2) ikt it ikt 2 6

Table A.1 Bond-Level Credit Spreads and Firm Default Risk log(S ) Est. S.E. T-stat ikt DD -0.022 0.002 -13.37 it log(Dur ) 0.170 0.018 9.47 ikt log(Age ) 0.094 0.010 9.51 ikt log(Par ) 0.085 0.014 6.25 ikt log(Coupon ) 0.040 0.043 0.94 ikt 1 0.057 0.149 0.39 Call ikt DD ×1 0.010 0.001 7.27 it Call ikt log(Dur )×1 0.030 0.018 1.65 ikt Call ikt log(Age )×1 -0.110 0.011 -9.89 ikt Call ikt log(Par )×1 -0.094 0.015 -6.05 ikt Call ikt log(Coupon )×1 0.503 0.045 11.28 ikt Call ikt LEV ×1 -0.042 0.007 -6.07 t Call ikt SLP ×1 -0.009 0.029 -0.29 t Call ikt CRV ×1 0.191 0.087 2.17 t Call ikt VOL ×1 0.002 0.000 8.37 t Call ikt Adj. R2 0.679 Industry Fixed Effects Yes Credit-Rating Fixed Effects Yes Note. Table A.1 present the estimated coefficients, standard errors and T-statistics from estimating regression(A.1)byOLS.ThesampleperiodisOctober1973toDecember2021andincludes682,316 observations. LEV , SLP , CRV refer to the level, slope and curvature (first three principal comt t t ponents) of the U.S. Treasury Yield Curve (Gu¨rkaynak et al., 2007); VOL refers to the realized t volatilityofdaily10-yearTreasuryyield.Standarderrorsaretwo-wayclusteredbyfirmandmonth. whereβ ˆ andγˆ denotetheOLSestimatedparametersandσˆ2 denotestheestimatedvariance of the error term. Finally, we define the excess bond premium as ˆ EBP = S −S . (A.3) ikt ikt ikt ImplementingthisprocedureforthebondsinICEandLehman-Wargawhosefirm’sbalance sheet data and equity prices are available from Compustat and CRSP, respectively, yields, 7

Figure A.3 Excess Bond Premium: Comparison with Gilchrist and Zakrajˇsek (2012) EBP Percentiles in Percentage Points 6 Corr(EBPFOR , EBPGZ) = 0.86 4 2 0 -2 EBPFOR EBPGZ 1973m1 1980m1 1990m1 2000m1 2010m1 2020m1 Note. Figure A.3 compares the mean EBP calculated in this paper, in red, with the mean EBP calculatedbyGilchristandZakrajˇsek(2012),inblue.Shadedcolumnsrepresentperiodsclassifiedas recessions by the National Bureau of Economic Research. after data cleaning as described in Appendix A.2, a sample of monthly EBPs for 11,913 bonds from 1,872 firms. Figure A.3 plots the time series of our mean EBP and that of Gilchrist and Zakrajˇsek (2012) and highlights that the correlation is 86%. The key predictor in the Gilchrist and Zakrajˇsek (2012) credit spread model is the firm’s Merton (1974) distance to default (DD), an indicator of the firm’s expected default risk. The DD framework assumes that the total value of the firm, denoted by V, is governed by following the stochastic differential equation: dV = µ Vdt+σ VdZ , (A.4) V V t where µ is the expected growth rate of V, σ is the volatility of V, and Z denotes the V V t standard Brownian motion. Assuming that the firm issues a single bond with face-value D that matures in T periods, Merton (1974) shows that the value of the firm’s equity E can be viewed as a call option on the underlying value of the firm V, with a strike price equal to the face-value of the firm’s debt D maturing at T. 8

Using the Black and Scholes (1973) pricing formula for a call option, the value of the firm’s equity is then E = VΦ(δ )−e−rTDΦ(δ ) (A.5) 1 2 where r denotes the risk-free interest rate, Φ(.) denotes the cdf of standard normal distribution, and log(V/D)+(r+0.5σ2)T √ δ = √ V and δ = δ −σ T. 1 2 1 V σ2 T V Using equation (A.5), by Ito’s lemma, one can relate the volatility of the firm’s value to the volatility of the firm’s equity V σ = Φ(δ )σ (A.6) E 1 V E Assuming a time to maturity of one year (T = 1) and daily data on one-year Treasury yields r, the face value of firm debt D, the market value of firm equity E, and its one-year historical volatility σ , equations (A.5) and (A.6) provide a two equation system that can E be used to solve for the two unknowns V and σ .26 Due to the issues raised in Vassalou V and Xing (2004), we follow Gilchrist and Zakrajˇsek (2012) by implementing the two-step iterativeprocedureofBharathandShumway(2008).First,wesetσ = σ foreachdayina V E one-year rolling window and then substitute σ into equation (A.5) to solve for the market V value V for each of these days. Second, from our new estimated V series, we calculate a year-long series of daily log-returns to the firm’s value, ∆logV, which we then use to compute a new estimate for σ as well as for µ .27 We then iterate on σ until convergence. V V V Given solutions (V,σ ,µ ) to the Merton DD model, we are able to calculate the V V 26DailydataforE isfromCRSP(prc∗shrout)andisusedtocalculateadaily252-dayhistoricalrollingwindowequityvolatilityσ .QuarterlydataonfirmdebtD =Current Liabilities+1Long-Term Liabilities E 2 is from Compustat (dlcq+0.5∗dlttq) and is linearly interpolated to form a daily series. √ 27Using the formulas σ = 252∗σ(∆logV) and µ =252∗µ(∆logV). V V 9

firm’s Distance to Default over a one-year horizon as log(V/D)+(µ −0.5σ2) DD = V V (A.7) σ V Since default at T occurs when a firm’s value falls below the value of its debt (log(V/D) < 0), the DD captures the distance a firm is above default, given an expected asset growth rate µ and volatility σ until T, in units of standard deviations. V V A.4 Summary Statistics In this section, we provide summary statistics for our main monthly bond-level and quarterly firm-level variables of interest, as well as for the monetary policy shocks at both a monthly and quarterly frequency. These are displayed in Table A.2. The first columns in Panels A.2a and A.2b report summary statistics for bond-level EBPs at a monthly frequency and firm-level EBPs at a quarterly frequency, respectively. The quarterly firm-level EBP series is constructed by averaging the bond-level EBP series across a firm’s outstanding bonds in a given month and then across the months in a given quarter.28 Thesummarystatisticsforthemonthlybond-levelandquarterlyfirm-levelEBPs are broadly in line with one another. Further, unsurprisingly given the results documented in Appendix A.3, our mean monthly bond-level EBP is very similar to the corresponding mean value from Gilchrist and Zakrajˇsek (2012). The second columns in Panels A.2a and A.2b report summary statistics for our dependent variables of interest, monthly bond-level credit spreads and quarterly firm-level investment,respectively.AswiththeEBP,thevalueofourmeanbond-levelcreditspread—about 2 percentage points—is very similar to the corresponding mean value from Gilchrist and Zakrajˇsek (2012). Similarly, the average level of firms’ investment in our sample—about 0.5 percent—is nearly identical to the corresponding mean value documented by Ottonello and Winberry (2020). The remainder of our summary statistics for firms’ investment are 28The difference in the number of observations between the quarterly firm-level EBP series and the monthly bond-level EBP series reflects these two levels of averaging. 10

Table A.2 Monthly Bond-level and Quarterly Firm-level Summary Statistics (a) Monthly Variables (b) Quarterly Variables EBP S εm EBP ∆log(K ) εm ikt ikt t it it t Mean .084 1.98 -.003 Mean .166 .490 -.008 Median -.071 1.28 0 Median -.065 -.006 -.007 S.D. 1.58 2.37 0.028 S.D. 2.01 6.75 .048 5th Perc. -1.32 .380 -.045 5th Perc. -1.76 -3.84 -.091 95th Perc. 1.81 5.66 .042 95th Perc. 2.57 6.27 .072 # Obs. 682,297 750,722 439 # Obs. 67,500 49,281 147 Note.TableA.2presentssummarystatisticsforourmainmonthlybond-levelvariablesandthemonetary policy shock series at a monthly frequency (Panel A.2a) and for our main quarterly firm-level variables and the monetary policy shock series at a quarterly frequency (Panel A.2b) from 1973 to 2021 (1985 to 2021 for the monetary policy shocks). Values are in percentage points, except for investment ∆log(K ) it which is in percent, and are calculated from the fully cleaned and merged dataset (see Appendix A.2). Themonthlymonetarypolicyshockseriesissummedwithineachquartertogeneratethequarterlyseries. Ofnote,themeanabsolute valueofthemonthly(quarterly)monetarypolicyshockseriesis1.7(3.6)basis points, which is an order of magnitude larger than the mean values reported above. For each firm, the monthly bond-level EBP is averaged across the firm’s bonds in a given quarter to generate the quarterly firm-level series. The monthly bond-level EBP (spread) panel includes 11913 (13439) bonds issued by 1872 (2216) firms. The quarterly firm-level EBP (investment) series includes 1866 (998) firms. also consistent with those documented by Ottonello and Winberry (2020), but with a moderately lower standard deviation and tighter tails. As mentioned previously, our analysis focuses on publicly-listed U.S. firms who issue debt in corporate bond markets. While this tilts our sample towards large firms relative to Ottonello and Winberry (2020)’s sample, data on firms’ credit spreads are crucial to inspect the transmission of monetary policy to firms’ investment. Further, large firms have been shown to be the primary driver of U.S. business cycles (e.g., Carvalho and Grassi, 2019). Still, relative to both the literatures on monetary policy’s effects on firm-level investment (e.g., Ottonello and Winberry, 2020) and on bond-level credit spreads (e.g., Anderson and Cesa-Bianchi, 2021), our use of the Lehman-Warga database and a monetary policy shock seriesthatspansperiodsofconventionalandunconventionalpolicyaffordsusasignificantly 11

longer sample.29 This longer sample is made evident by the large number of observations we have for the monetary policy shock series, whose summary statistics at a monthly and quarterly frequency are tabulated in the third columns of Panels A.2a and A.2b, respectively. The quarterly monetary policy shock series is generated by summing the monthly series within each quarter. Of note, the mean absolute value of the monthly (quarterly) monetary policy shock series is 1.7 (3.6) basis points, which is an order of magnitude larger than the mean values reported in the table. 29In addition, relative to Anderson and Cesa-Bianchi (2021) who also focus on publicly-listed U.S. firms who issue debt in corporate bond markets, our bond-level EBPs are calculated for 2500 more bonds and about 900 more firms. 12

B Additional Empirical Results and Robustness In this section, we offer additional empirical results and robustness to complement our findings from the main text. In Section B.1, we show that our results are robust to including time-sector fixed effects. In Section B.2, we show our results are robust to conditioning on bond/firm EBPs using dummy variables. In Section B.3, we show that, when not conditioning on the EBP, default risk indeed regulates firms’ responses to monetary policy. In Section B.4, we highlight that heterogeneous responses by EBP are robust to controlling for monetary policy’s effects conditional on other firm characteristics. In Section B.5, we re-estimate our main specifications with alternative monetary policy shocks. In Section B.6, we re-estimate our results using an EBP purged of its higher-order dependence on default risk. In Section B.7, we document monetary policy’s effects on firm debt issuance by EBP. In Section B.8, we study the conditioning effects of EBP for intermediary net worth shocks. Finally, in Section B.9, we showcase the robustness of our results linking the EBP distribution to the aggregate effectiveness of monetary policy. B.1 EBP Heterogeneity with Sector-Time Fixed Effects We begin by showing that our results for the heterogeneous responses conditional on EBP are robust to controlling for time-sector fixed effects. Indeed, we show that the spreads of high-EBP bonds and investment of low-EBP firms remain more sensitive to monetary policy shocks. In addition, we show the investment of low-EBP firms is more responsive to movements in their credit spreads. To show this, we define W as the vector of firm-level it−1 controls contained in Z (see Section 2.4), but excluding the macro-level controls. it−1 Monetary Policy on Credit Spreads: Beginning with monetary policy’s effect on credit spreads, we include sector-time fixed 13

Figure B.1 Monetary Policy’s Effect on Bond-Level Credit Spreads Depending on EBP (a) Macro-Financial Controls (b) Sector-Time Fixed Effects Marginal Effects Marginal Effects 0 0 -2 -2 -4 -4 -6 -6 0 5 10 15 20 0 5 10 15 20 Months after Shock Months after Shock Note.FigureB.1comparestheeffectsofthedynamicinteraction(βh)betweenEBP andtheBu 2 ikt−1 et al. (2021) monetary policy shock (εm) on the h-period change in credit spreads, S −S , t ikt+h ikt−1 for two different specifications: one that controls for macro-financial controls as in the main text (4) in Panel B.1a and one that includes time-sector fixed effects (B.1) in Panel B.1b. The frequency of the data is monthly. Conditional results describe the additional effect of having the variable one standard deviation above the sample mean. The inner and outer shaded areas correspond to the 68%and90%confidenceintervalsconstructedusingtwo-wayclusteredstandarderrorsbyfirmiand month t, respectively. effects αh in the following specification (B.1):30 s,t S −S = βh +αh +βhεm +βhEBPma ×εm +γhW +e , (B.1) ikt+h ikt−1 k s,t 1 t 2 ikt−1 t it−1 ikth The interaction effects (βhs) are displayed in Panel B.1b of Figure B.1, alongside the results 2 from the original specification in Panel B.1a, which have been recopied from Panel 3b of Figure 3 for comparison. Figure B.1 highlights that our results from section 3 are robust to controlling for sector-time fixed effects: credit spreads of high-EBP bonds are more responsive to monetary policy. Monetary Policy on Firm Investment: Next, turning to monetary policy’s effects on investment, we include sector-time fixed 30Note that macro-financial controls, in addition to the unconditional monetary policy shock εm, would t be absorbed by αh . s,t 14

Figure B.2 Monetary Policy’s Effect on Firm-Level Investment Depending on EBP (a) Macro-Financial Controls (b) Sector-Time Fixed Effects Marginal Effects Marginal Effects 0 0 -10 -10 -20 -20 0 4 8 12 0 4 8 12 Quarters after Shock Quarters after Shock Note. Figure B.2 compares the effects of the dynamic interaction (βh) between EBP and the 2 it−1 Bu et al. (2021) monetary policy shock (εm) on h-period investment of firm i, logK −logK , t it+h it−1 for two different specifications: one that controls for macro-financial controls as in the main text (6) in Panel B.2a and one that includes time-sector fixed effects (B.2) in Panel B.2b. The frequency of the data is quarterly. Conditional results describe the additional effect of having the variable one standard deviation above the sample mean. The inner and outer shaded areas correspond to the 68%and90%confidenceintervalsconstructedusingtwo-wayclusteredstandarderrorsbyfirmiand quarter t, respectively. effects αh in the following specification (B.2): s,t (cid:18) (cid:19) K log it+h = βh +αh +βhεm +βhEBPma ×εm +γhW +e . (B.2) K i s,t 1 t 2 it−1 t it−1 ith it−1 The interaction effect (βhs) are displayed in Panel B.2b of Figure B.2, alongside the 2 results from the original specification in Panel B.2a, which have been recopied from Panel 5b of Figure 5 for comparison. Figure B.2 highlights that our results from Section 4 are robust to controlling for sector-time fixed effects: investment by low-EBP firms is more sensitive to monetary policy shocks. Firm Credit Spreads on Firm Investment: We assess the robustness of our results relating firms’ investment responses to changes in their credit to the inclusion of sector-time fixed effects αh using the following specificas,t 15

Figure B.3 Credit Spreads’ Effects on Firm Investment Depending on EBP (a) Macro-Financial Controls (b) Sector-Time Fixed Effects Marginal Effects Marginal Effects 1.5 1.5 1 1 .5 .5 0 0 0 4 8 12 0 4 8 12 Quarters after Shock Quarters after Shock Note. Figure B.3 compares the effects of the dynamic effect (βh) of a movement in credit spreads 1 ∆S onh-periodinvestmentoffirmi,logK −logK ,fortwodifferentspecifications:onethat it it+h it−1 controlsformacro-financialcontrolsasinthemaintext(12)inPanelB.3aandonethatincludestimesector fixed effects (B.3) in Panel B.3b. The frequency of the data is quarterly. Conditional results describe the additional effect of having the variable one standard deviation above the sample mean. The inner and outer shaded areas correspond to the 68% and 90% confidence intervals constructed using two-way clustered standard errors by firm i and quarter t, respectively. tion (B.3): (cid:18) (cid:19) K log it+h = βh +αh +βh∆S +βh∆S ×EBPma +γhW +e , (B.3) K i s,t 1 i,t 2 i,t it−1 it−1 ith it−1 The interaction effect (βhs) are displayed in Panel B.3b of Figure B.3, alongside the 2 results from the original specification in Panel B.3a, which have been recopied from Panel 9b of Figure 9 for comparison. As before, Figure B.3 highlights that our results from section 6arerobusttocontrollingforsector-timefixedeffects:investmentbylow-EBPfirmsismore sensitive to movements in their credit spreads. B.2 EBP Heterogeneity with Dummy Variables In this subsection, we demonstrate that our findings from the main text are not tied to the functional form of our EBP state variable, the moving yearly mean used by Jeenas (2019). In particular, we perform the same analysis as in the main text using the dummy variable 16

approach used by Cloyne et al. (2023) and Anderson and Cesa-Bianchi (2021), and show that our conclusions are unchanged. Denote by EBP the EBP on firm i’s bond k in period t. Then, define 1EBPlow ikt ikt as a dummy variable taking the value of 1 if EBP lies below the median of the EBP ikt distribution in period t and 0 otherwise. Similarly, define 1EBPhigh as a dummy variable ikt taking the value of 1 if EBP lies above the median of the EBP distribution in period t ikt and 0 otherwise. Now, we reconsider the results from sections 3, 4, and 6. When re-assessing eachsection,weevaluatetwospecifications.Thefirstallowsustotracethedistinctdynamic responses of spreads or investment to either monetary policy shocks or changes in spreads for 1EBPlow and 1EBPhigh firms. The second specification allows us to assess the relative ikt ikt response of these two types of firms. Monetary Policy on Credit Spreads: To assess the distinct responses of credit spreads from monetary policy shocks for lowand high-EBP bonds, we estimate: S −S = βh +βhεm ×1EBPlow +βhεm ×1EBPhigh +γhZ +e , (B.4) ikt+h ikt−1 k 1 t ikt−1 2 t ikt−1 it−1 ikth where Z includes the controls from the main text, plus 1EBPlow and 1EBPhigh . it−1 ikt−1 ikt−1 TheimpulseresponsesaredisplayedinFigureB.4,whereweseethatthecreditspreads of high-EBP bonds are significantly more responsive to monetary policy than are the spreads of low-EBP bonds. This is consistent with our findings from the main text. To see whether these two responses are distinct from one another, we estimate the adapted specification: S −S = βh +βhεm +βhεm ×1EBPhigh +γhZ +e , (B.5) ikt+h ikt−1 k 1 t 2 t ikt−1 it−1 ikth where Z includes the controls from the main text, plus 1EBPhigh . Since we have init−1 ikt−1 cluded the monetary policy shock εm on its own, the interaction coefficient βh’s interpretat 2 tion is now the response of the high-EBP bond’s spread relative to low-EBP bond’s spread 17

Figure B.4 Monetary Policy’s Effect on Credit Spreads for Low- vs High-EBP Bonds (a) Low-EBP (b) High-EBP Marginal Effects Marginal Effects 0 0 -3 -3 -6 -6 -9 -9 -12 -12 0 5 10 15 20 0 5 10 15 20 Months after Shock Months after Shock Note. Figure B.4 traces the response of spreads for low-EBP (1EBPlow) bonds in Panel B.4a and high-EBP (1EBPhigh) bonds in Panel B.4b to a Bu et al. (2021) monetary policy shock (εm), from t regression(B.4),wherethefrequencyismonthly.Conditionalresultsdescribetheadditionaleffectof havingthevariableonestandarddeviationabovethesamplemean.Theinnerandoutershadedareas correspond to the 68% and 90% confidence intervals constructed using two-way clustered standard errors by firm i and month t, respectively. following a shock monetary policy easing. The interaction effect is displayed in Figure B.5 and highlights, as in the main text, that high-EBP bonds’ spreads fall by more following a monetary easing than low-EBP bonds’ spreads. This showcases that, under an alternative functional form for our state variable, the conclusions from Section 3 are unchanged. Monetary Policy on Investment: Proceeding as before, to assess the distinct investment responses to monetary policy shocks for low- and high-EBP firms, we estimate: (cid:18) (cid:19) K log it+h = βh +βhεm ×1EBPlow +βhεm ×1EBPhigh +γhZ +e , (B.6) K i 1 t it−1 2 t it−1 it−1 ith it−1 where Z includes the controls from the main text, plus 1EBPlow and 1EBPhigh. it−1 it−1 it−1 The impulse responses are displayed in Figure B.6, where we see that the investment of low-EBP firms is significantly more responsive to monetary policy than are the investment of high-EBP bonds. This is consistent with our findings from the main text. 18

Figure B.5 Monetary Policy’s Effect on Bond-Level Credit Spreads Depends on EBP Marginal Effects 0 -2 -4 -6 0 5 10 15 20 Months after Shock Note. Figure B.5 traces the relative response (βh) of high-EBP 1EBPhigh bond’s spreads relative 2 ikt−1 to low-EBP 1EBPlow bond’s spreads from a Bu et al. (2021) monetary policy shock (εm), from ikt−1 t regression(B.5),wherethefrequencyismonthly.Conditionalresultsdescribetheadditionaleffectof havingthevariableonestandarddeviationabovethesamplemean.Theinnerandoutershadedareas correspond to the 68% and 90% confidence intervals constructed using two-way clustered standard errors by firm i and month t. Again, to see whether these two responses are distinct from one another, we estimate: (cid:18) (cid:19) K log it+h = βh +βhεm +βhεm ×1EBPlow +γhZ +e , (B.7) K i 1 t 2 t it−1 it−1 ith it−1 whereZ includesthecontrolsfromthemaintext,plus1EBPlow .Sincewehaveincluded it−1 it−1 the monetary policy shock εm on its own, the interaction coefficient’s (βh) interpretation t 2 is now the response of low-EBP firms’ investment relative to high-EBP firms’ investment to a shock monetary policy easing. The interaction effect is displayed in Figure B.7 and highlights that a shock monetary policy easing increases investment more for low-EBP firms than for high-EBP firms. This signifies, as before, that our conclusions from Section 4 are unchanged when using an alternative functional form for our state variable. 19

Figure B.6 Monetary Policy’s Effect on Firm Investment for Low- vs High-EBP Firms (a) Low-EBP Firms (b) High-EBP Firms Marginal Effects Marginal Effects 20 20 10 10 0 0 -10 -10 0 4 8 12 0 4 8 12 Quarters after Shock Quarters after Shock Note. Figure B.6 traces the response of investment for low-EBP (1EBPlow) bonds in Panel B.6a and high-EBP (1EBPhigh) bonds in Panel B.6b to a Bu et al. (2021) monetary policy shock (εm), t from regression (B.6), where the frequency is quarterly. The frequency of the data is quarterly. Conditionalresultsdescribetheadditionaleffectofhavingthevariableonestandarddeviationabove the sample mean. The inner and outer shaded areas correspond to the 68% and 90% confidence intervals constructed using two-way clustered standard errors by firm i and quarter t, respectively. Credit Spreads on Investment: Finally, we assess the distinct investment responses to movements in credit spreads for low- and high-EBP firms by estimating: (cid:18) (cid:19) K log it+h = βh +βh∆S ×1EBPlow +βh∆S ×1EBPhigh +γhZ +e , (B.8) K i 1 it it−1 2 it it−1 it−1 ith it−1 where Z includes the controls from the main text, plus 1EBPlow and 1EBPhigh. it−1 it−1 it−1 The impulse responses are displayed in Figure B.8, where we see that the investment of low-EBP firms is significantly more responsive to movements in their credit spreads compared to the investment of high-EBP firms. This is consistent with our findings from the main text. To see whether these two responses are distinct from one another, we estimate: (cid:18) (cid:19) K log it+h = βh +βh∆S +βh∆S ×1EBPlow +γhZ +e , (B.9) K i 1 it 2 it it−1 it−1 ith it−1 20

Figure B.7 Monetary Policy’s Effect on Firm-Level Investment Depends on EBP Marginal Effects 20 10 0 0 4 8 12 Quarters after Shock Note. Figure B.7 traces the relative response (βh) of low-EBP 1EBPlow firms’ investment relative 2 it−1 to high-EBP 1EBPhigh firms’ investment from a Bu et al. (2021) monetary policy shock (εm), from it−1 t regression(B.7),wherethefrequencyisquarterly.Conditionalresultsdescribetheadditionaleffectof havingthevariableonestandarddeviationabovethesamplemean.Theinnerandoutershadedareas correspond to the 68% and 90% confidence intervals constructed using two-way clustered standard errors by firm i and quarter t. where Z includes the controls from the main text, plus 1EBPlow . Again, because we it−1 it−1 have included the credit spread shock ∆S on its own, the interaction coefficient’s (βh) it 2 interpretation is now the response of low-EBP firms’ investment relative to high-EBP firms’ investment to movements in credit spreads ∆S . it The interaction effect is displayed in Figure B.9 and highlights, as in the main text, that low-EBP firms’ investment falls by more following an increase in their credit spreads relatve to high-EBP firms’ investment, just as in Section 6. 21

Figure B.8 Credit Spread Shocks and Firm Investment for Low- vs High-EBP Firms (a) Low-EBP Firms (b) High-EBP Firms Marginal Effects Marginal Effects 0 0 -2 -2 -4 -4 -6 -6 -8 -8 0 4 8 12 0 4 8 12 Quarters after Shock Quarters after Shock Note. Figure B.8 traces the response of investment for low-EBP (1EBPlow) firms in Panel B.8a and high-EBP (1EBPhigh) firms in Panel B.8b to a change in credit spreads ∆S , from regression it (B.8), where the frequency is quarterly. Conditional results describe the additional effect of having the variable one standard deviation above the sample mean. The inner and outer shaded areas correspond to the 68% and 90% confidence intervals constructed using two-way clustered standard errors by firm i and quarter t, respectively. 22

Figure B.9 Credit Spread’s Effect on Firm-Level Investment Depends on EBP Marginal Effects 0 -2 -4 -6 0 4 8 12 Quarters after Shock Note. Figure B.9 traces the relative response (βh) of low-EBP 1EBPlow firms’ investment relative 2 it−1 to high-EBP 1EBPhigh firms’ investment from a change in credit spreads ∆S , from regression it−1 it (B.9), where the frequency is quarterly. Conditional results describe the additional effect of having the variable one standard deviation above the sample mean. The inner and outer shaded areas correspond to the 68% and 90% confidence intervals constructed using two-way clustered standard errors by firm i and quarter t. 23

B.3 Default Risk as a State Variable In this section, we document that, when not controlling for heterogeneity by EBP, defaultrisk does indeed regulate the response of firms’ credit spreads and investment to monetary policyshocksinamannerconsistentwiththefindingsofAndersonandCesa-Bianchi(2021) and Ottonello and Winberry (2020), respectively. To demonstrate this, we use the dummy variable approach outlined in the previous section, since this is the functional form used by Anderson and Cesa-Bianchi (2021). Ottonello and Winberry (2020) use a linear functional form that purges firms’ default risk of their in-sample firm-specific mean, which is motivated by firms being ex-ante identical in their model. To make our results comparable across as many studies as possible, we use the dummy variable approach. Monetary Policy on Credit Spreads: We begin by assessing the responses of bond-level credit spreads to monetary policy shocks for low- vs. high-default-risk firms. Recall that low distance to default and high leverage firms are viewed as having high default-risk. We begin by estimating the following specification at a monthly frequency: S −S = βh +βhεm ×1xlow +βhεm ×1xhigh +γhZ +e , (B.10) ikt+h ikt−1 k 1 t it−1 2 t it−1 it−1 ikth where x denotes either distance to default or leverage. In the notation of the previous section, 1xlow is a dummy variable taking the value of 1 if x lies below the median of the it it firm-level distance to default or leverage distribution in period t and 0 otherwise. Similarly, define 1xhigh as a dummy variable taking the value of 1 if x lies above the median of the it it firm-level distance to default or leverage distribution in period t and 0 otherwise. Note also that Z includes the controls from the main text, plus 1xlow and 1xhigh. it−1 it−1 it−1 The impulse responses are displayed in Figure B.10, where the Panels B.10a and B.10c trace βh and βh, respectively, when x is distance to default while Panels B.10b and B.10d 1 2 trace βh and βh, respectively, when x is leverage. Clearly, we see that the marginal effects 2 1 24

Figure B.10 Monetary Policy’s Effect on Spreads for Low vs. High Default-Risk Firms (a) Low Distance to Default (b) High Leverage Marginal Effects Marginal Effects 0 0 -3 -3 -6 -6 -9 -9 -12 -12 0 5 10 15 20 0 5 10 15 20 Months after Shock Months after Shock (c) High Distance to Default (d) Low Leverage Marginal Effects Marginal Effects 0 0 -3 -3 -6 -6 -9 -9 -12 -12 0 5 10 15 20 0 5 10 15 20 Months after Shock Months after Shock Note. Figure B.10 traces the distinct responses of low- and high- distance to default firms’ spreads to a monetary policy shock in Panels B.10a and B.10c from estimating regression (B.10) with x as distance to default, while Panels B.10b and B.10d trace the distinct responses of high- and lowleverage firms’ spreads to a monetary policy shock from estimating regression (B.10) with x as leverage. The frequency of the data is monthly. Conditional results describe the additional effect of havingthevariableonestandarddeviationabovethesamplemean.Theinnerandoutershadedareas correspond to the 68% and 90% confidence intervals constructed using two-way clustered standard errors by firm i and month t, respectively. in the top row (Panels B.10a and B.10b), for low distance to default and high leverage firms, are larger than those in the bottom row. That is, consistent with the findings of Anderson and Cesa-Bianchi (2021), the credit spreads of firms with high default risk are more responsive to monetary policy shocks than are the firms with low default risk. Following a similar path to the one from the previous section, we next assess whether the response of spreads for high- vs. low-default risk firms are statistically different from 25

Figure B.11 Monetary Policy’s Relative Effect on Bond-Level Credit Spreads by Default Risk (a) Low Distance to Default (b) High Leverage Marginal Effects Marginal Effects 0 0 -2 -2 -4 -4 -6 -6 0 5 10 15 20 0 5 10 15 20 Months after Shock Months after Shock Note.FigureB.11tracestheresponseofcreditspreadsforhigh-defaultrisk,lowdistancetodefaultin PanelB.11aandhigh-leverageinPanelB.11b,toamonetarypolicyshockfromestimatingregression (B.11), where the frequency is monthly. Conditional results describe the additional effect of having the variable one standard deviation above the sample mean. The inner and outer shaded areas correspond to the 68% and 90% confidence intervals constructed using two-way clustered standard errors by firm i and month t, respectively. one another using: S −S = βh +βhεm +βhεm ×1xhigh(low) +γhZ +e , (B.11) ikt+h ikt−1 k 1 t 2 t it−1 it−1 ikth where, we include 1xlow when x is distance to default and 1xhigh when x is leverage, to it−1 it−1 keep the responses comparable. As before, because we have included the monetary policy shock εm on its own, the interaction coefficient’s (βh) interpretation is the response of hight 2 default risk firms’ (low distance to default or high leverage) spreads relative to low-default risk firms’ spreads to a monetary policy shock. Note also that Z includes the controls it−1 from the main text, plus 1xhigh(low). it−1 The interaction effect is displayed in Figure B.11 and highlights that high defaultrisk firms’ spreads fall by more following a shock monetary policy easing compared to low default-risk firms’, as is found by Anderson and Cesa-Bianchi (2021). Monetary Policy on Investment: To assess the distinct investment responses to monetary policy shocks for firms with 26

Figure B.12 Monetary Policy’s Effect on Investment for Low vs. High Default-Risk Firms (a) Low Distance to Default (b) High Leverage Marginal Effects Marginal Effects 20 20 10 10 0 0 -10 -10 0 4 8 12 0 4 8 12 Quarters after Shock Quarters after Shock (c) High Distance to Default (d) Low Leverage Marginal Effects Marginal Effects 20 20 10 10 0 0 -10 -10 0 4 8 12 0 4 8 12 Quarters after Shock Quarters after Shock Note.FigureB.12tracesthedistinctresponsesoflow-andhigh-distancetodefaultfirms’investment to a monetary policy shock in Panels B.12a and B.12c from estimating regression (B.12) with x as distance to default, while Panels B.12b and B.12d trace the distinct responses of high- and lowleverage firms’ investment to a monetary policy shock from estimating regression (B.12) with x as leverage. The frequency of the data is quarterly. Conditional results describe the additional effect of havingthevariableonestandarddeviationabovethesamplemean.Theinnerandoutershadedareas correspond to the 68% and 90% confidence intervals constructed using two-way clustered standard errors by firm i and quarter t, respectively. low- vs. high-default risk, we estimate: (cid:18) (cid:19) K log it+h = βh +βhεm ×1xlow +βhεm ×1xhigh +γhZ +e , (B.12) K i 1 t it−1 2 t it−1 it−1 ith it−1 where again x refers either to firms’ distance to default or leverage and Z includes the it−1 controls from the main text, plus 1xlow and 1xhigh. it−1 it−1 27

Figure B.13 Monetary Policy’s Relative Effect on Firm-Level Investment by Default Risk (a) High Distance to Default (b) Low Leverage Marginal Effects Marginal Effects 20 20 10 10 0 0 -10 -10 0 4 8 12 0 4 8 12 Quarters after Shock Quarters after Shock Note. Figure B.13 traces the response of investment for low-default risk, high distance to default in PanelB.13aandlow-leverageinPanelB.13b,toamonetarypolicyshockfromestimatingregression (B.13), where the frequency is quarterly. Conditional results describe the additional effect of having the variable one standard deviation above the sample mean. The inner and outer shaded areas correspond to the 68% and 90% confidence intervals constructed using two-way clustered standard errors by firm i and quarter t, respectively. The impulse responses are displayed in Figure B.12, where we see that only the investment responses of low default-risk—high distance to default (Panel B.12c) and low leverage (Panel B.12d)—firms are statistically different from zero. This is consistent with the findings of Ottonello and Winberry (2020). Finally, to see whether the responses of low vs. high default-risk firms are distinct from one another, we estimate: (cid:18) (cid:19) K log it+h = βh +βhεm +βhεm ×1xlow(high) +γhZ +e , (B.13) K i 1 t 2 t it−1 it−1 ith it−1 where, we include1xlow whenxis leverage and1xhigh whenxis distanceto default, to keep it−1 it−1 the responses comparable. As before, because we have included the monetary policy shock εm on its own, the interaction coefficient’s (βh) interpretation is the response of low-default t 2 risk firms’ (high distance to default or low leverage) investment relative to low-default risk firms’ investment to a monetary policy shock. Note also that Z includes the controls it−1 from the main text, plus 1xlow(high). it−1 28

The impulse responses are traced in Figure B.13 and highlight that point estimates for both leverage and distance to default imply that low-default risk firms’ investment increases by more than high-default risk firms’. However, only when using distance to default (Panel B.13a) is the effect statistically different from zero, albeit at longer horizons. This is consistent with Ottonello and Winberry (2020) who show that distance to default outperforms leverage in regulating firms’ investment response to monetary policy. It is worth pointing out that Jeenas (2019) and Anderson and Cesa-Bianchi (2021) find that it is high-default-risk firms whose investment is more sensitive to monetary policy, while Lakdawala and Moreland (2021) highlight that the sign of heterogeneity by default risk may have changed following the global financial crisis. The differences in results across studies are part of an ongoing debate in the literature, which our results in this section for heterogeneity by default risk contribute to. B.4 Monetary Policy’s Effect by EBP vs. other Characteristics In this section, we show that the importance of firms’ EBPs for determining their responsiveness to monetary policy is robust to conditioning on other competing firm characteristics. We first document that, as for the baseline linear interactions used in the main text, EBP heterogeneity tends to supersede heterogeneity by distance to default and leverage when using the dummy variable approach. Next, we consider heterogeneity by credit rating, age, size (assets) , sales growth, share of liquid assets, and Tobin’s average Q and show that the EBP remains a significant state variable for the transmission of monetary policy when conditioning on these firm characteristics as well. To provide comparability with the existing studies, we use the dummy variable approach when assessing the conditioning effects of firms’ EBPs relative to their credit rating (Ottonello and Winberry, 2020), age (Cloyne et al., 2023), and size and sales growth (Gertler and Gilchrist, 1994), but use our baseline linear interaction for share of liquid assets (Jeenas, 2019) and Tobin’s average q. 29

B.4.1 Distance to Default and Leverage with dummy variables: In the main text, we ran horseraces between linear EBP interactions and linear default risk interactions to highlight that firms’ responsiveness to monetary policy was largely a function of their EBPs.31 In this section, we show that running similar horseraces using the dummy variable approach does not alter our conclusion that a firm’s EBP supersedes its default risk as state variable for the transmission of monetary policy to both credit spreads and investment. Monetary Policy on Credit Spreads: We begin by running a horserace between the EBP and a measure of default risk x (distance to default or leverage) as a conditioning variable for the impact of monetary policy on credit spreads: S −S = βh +βhεm +βhεm ×1EBPhigh +βhεm ×1xlow(high) +γhZ +e , ikt+h ikt−1 k 1 t 2 t ikt−1 3 t it−1 it−1 ikth (B.14) where, as before, because we have included the monetary policy shock εm on its own, the t interaction coefficient associated with 1EBPhigh (βh) is interpreted as the credit spread ikt−1 2 response of high-EBP bonds relative to low-EBP bonds due to a monetary policy shock, controlling for heterogeneity by default risk. An analogous interpretation is associated with βh. As before, we use 1xlow when x is distance to default and 1xhigh when x is leverage, 3 it−1 it−1 so as to capture the relative effect of high default risk firms relative to low default risk firms. Note also that Z includes the controls from the main text, plus 1EBPhigh and it−1 ikt−1 1xlow(high). it−1 The results are displayed in Figure B.14 and highlight, as in the main text, that firms’ EBPs tend to supersede their default risk in regulating the sensitivity of firms’ spreads to monetary policy shocks, and that it is the spreads of firms whose bonds carry high-EBPs that are most responsive. 31Specifically, we used linear interactions between the one-year moving average of a firm’s characteristic (EBP or default risk) and the monetary policy shock, as in Jeenas (2019). 30

Figure B.14 Monetary Policy’s Relative Effect on Spreads by EBP vs. Default Risk (a) High EBP (b) Low Distance to Default Marginal Effects Marginal Effects 0 0 -2 -2 -4 -4 -6 -6 0 5 10 15 20 0 5 10 15 20 Months after Shock Months after Shock (c) High EBP (d) High Leverage Marginal Effects Marginal Effects 0 0 -2 -2 -4 -4 -6 -6 0 5 10 15 20 0 5 10 15 20 Months after Shock Months after Shock Note.FigureB.14displaysdynamicinteractioncoefficientsfromahorseracebetween(A)therelative response of high-EBP bonds’ spreads compared to low-EBP bonds’ (Panels B.14a and B.14c) and (B) the relative response of high-default-risk firms’ spreads compared to low-default-risk firms’ (low distancetodefaultinPanelB.14bandhighleverageinPanelB.14d)fromamonetarypolicyshockεm t from estimating regression (B.14). Frequency is monthly. Conditional results describe the additional effect of having the variable one standard deviation above the sample mean. Inner and outer shaded areascorrespondto68%and90%confidenceintervalsconstructedusingtwo-wayclusteredstandard errors by firm i and month t. Monetary Policy on Investment: Next, we show the same for monetary policy’s effect on investment, using the following 31

local projection: (cid:18) (cid:19) K log it+h = βh +βhεm +βhεm ×1EBPlow +βhεm ×1xhigh(low) +γhZ +e , K i 1 t 2 t it−1 3 t it−1 it−1 ith it−1 (B.15) where, we include 1xhigh when x is distance to default and 1xlow when x is leverage, so it−1 it−1 as to capture the relative effect of low default risk firms’ investment response vs. to high default risk firms, and compare them to the relative response of low-EBP firms’ investment, as compared to high-EBP firms’. Again, Z includes the controls from the main text, it−1 plus 1EBPlow and 1xhigh(low). it−1 it−1 The results are displayed in Figure B.15. As in the main text, we see that firms’ EBPs tend to supersede their default risk in regulating the sensitivity of firms’ investment to monetary policy shocks, and that it is firms with low-EBPs whose investment is most responsive. B.4.2 Credit Rating: In their appendix, Ottonello and Winberry (2020) assess the conditioning power of firms’ default risk as measured by their credit ratings, using the dummy variable approach. Here, we use the dummy variable approach to highlight that heterogeneity by EBP is robust to controlling for heterogeneity by credit rating. Monetary Policy on Credit Spreads: We begin by running the following local projection: S −S = βh +βhεm +βhεm ×1EBPhigh +βhεm ×1Ratelow +γhZ +e , ikt+h ikt−1 k 1 t 2 t ikt−1 3 t it−1 it−1 ikth (B.16) where 1Ratelow denotes a dummy variable taking the value of one if the firms’ credit rating lies below the median of the cross-sectional credit rating distribution in the period prior to the monetary surprise, that is, the firm is viewed as relatively risky. Note again that Z it−1 32

Figure B.15 Monetary Policy’s Relative Effect on Investment by EBP vs. Default Risk (a) Low EBP (b) High Distance to Default Marginal Effects Marginal Effects 20 20 10 10 0 0 -10 -10 0 4 8 12 0 4 8 12 Quarters after Shock Quarters after Shock (c) Low EBP (d) Low Leverage Marginal Effects Marginal Effects 20 20 10 10 0 0 -10 -10 0 4 8 12 0 4 8 12 Quarters after Shock Quarters after Shock Note.FigureB.15displaysdynamicinteractioncoefficientsfromahorseracebetween(A)therelative response of low-EBP firms’ investment compared to high-EBP firms’ (Panels B.15a and B.15c) and (B) the relative response of low-default-risk firms’ investment compared to high-default-risk firms’ (high distance to default in Panel B.15b and low leverage in Panel B.15d) from a monetary policy shock εm from estimating regression (B.15). Frequency is quarterly. Conditional results describe t the additional effect of having the variable one standard deviation above the sample mean. Inner and outer shaded areas correspond to 68% and 90% confidence intervals constructed using two-way clustered standard errors by firm i and quarter t. includes the controls from the main text, plus 1EBPhigh and 1Ratelow . ikt−1 it−1 In Figure B.16, we see that while high-risk firms’ spreads are more responsive to monetary shocks (Panel B.16b), the EBP continues to be an important determinant of the sensitivity of firms’ spreads to monetary policy.32 32Interestingly, since rating agencies rely on the Merton (1974) model as a primary determinant of the credit rating, the impulse responses for credit rating look similar to those for distance to default in this 33

Figure B.16 Monetary Policy’s Relative Effect on Spreads by EBP vs. Credit Rating (a) High EBP (b) Low Credit Rating Marginal Effects Marginal Effects 0 0 -2 -2 -4 -4 -6 -6 0 5 10 15 20 0 5 10 15 20 Months after Shock Months after Shock Note. Figure B.16 displays dynamic interaction coefficients from a horserace between (A) the relative response of high-EBP bonds’ spreads compared to low-EBP bonds’ (Panel B.16a) and (B) the relative response of low-credit-rating (risky) firms’ spreads compared to high-rating (safe) firms’ (Panel B.16b) from a monetary policy shock εm from estimating regression (B.16). Frequency is t monthly. Conditional results describe the additional effect of having the variable one standard deviation above the sample mean. Inner and outer shaded areas correspond to 68% and 90% confidence intervals constructed using two-way clustered standard errors by firm i and month t. Monetary Policy on Investment: Next, we estimate: (cid:18) (cid:19) K log it+h = βh +βhεm +βhεm ×1EBPlow +βhεm ×1Ratehigh +γhZ +e , K i 1 t 2 t it−1 3 t it−1 it−1 ith it−1 (B.17) where Z includes the controls from the main text, plus 1EBPlow and 1Ratehigh. it−1 it−1 it−1 The impulse responses are presented in Figure B.17. We see again that the EBP regulates firms’ investment response to monetary policy (Panel B.17a), as in the main text, superseding heterogeneity by credit rating (Panel B.17b). case. 34

Figure B.17 Monetary Policy’s Relative Effect on Investment by EBP vs. Credit Rating (a) Low EBP (b) High Credit Rating Marginal Effects Marginal Effects 20 20 10 10 0 0 -10 -10 0 4 8 12 0 4 8 12 Quarters after Shock Quarters after Shock Note. Figure B.17 displays dynamic interaction coefficients from a horserace between (A) the relative response of low-EBP firms’ investment compared to high-EBP firms’ (Panel B.17a) and (B) the relativeresponseofhigh-credit-ratingfirms’investmentcomparedtolow-ratingfirms’(PanelB.17b) from a monetary policy shock εm from estimating regression (B.17). Frequency is quarterly. Cont ditional results describe the additional effect of having the variable one standard deviation above the sample mean. Inner and outer shaded areas correspond to 68% and 90% confidence intervals constructed using two-way clustered standard errors by firm i and quarter t. B.4.3 Age: Next, we turn to demonstrate the robustness of our EBP state to firms’ age, which Cloyne et al. (2023) show regulates the sensitivity of firms’ investment to monetary policy shocks. LikeAndersonandCesa-Bianchi(2021),weuseagesinceIPO,sincethisvariableisavailable in the Compustat database. Admittedly, this is different from the age since incorporation variable used by Cloyne et al. (2023). Monetary Policy on Credit Spreads: Cloyne et al. (2023) use the dummy variable approach in establishing their empirical findings, and we follow them in our robustness check and run the following horserace regression: S −S = βh +βhεm +βhεm ×1EBPhigh +βhεm ×1Agelow +γhZ +e , ikt+h ikt−1 k 1 t 2 t ikt−1 3 t it−1 it−1 ikth (B.18) 35

Figure B.18 Monetary Policy’s Relative Effect on Spreads by EBP vs. Age (a) High EBP (b) Low Age (Young) Marginal Effects Marginal Effects 0 0 -2 -2 -4 -4 -6 -6 0 5 10 15 20 0 5 10 15 20 Months after Shock Months after Shock Note.FigureB.18displaysdynamicinteractioncoefficientsfromahorseracebetween(A)therelative responseofhigh-EBPbonds’spreadscomparedtolow-EBPbonds’(PanelB.18a)and(B)therelative response of low-age (young) firms’ spreads compared to high-age (old) firms’ (Panel B.18b) from a monetarypolicyshockεm fromestimating(B.18).Frequencyismonthly.Conditionalresultsdescribe t the additional effect of having the variable one standard deviation above the sample mean. Inner and outer shaded areas correspond to 68% and 90% confidence intervals constructed using two-way clustered standard errors by firm i and month t. where 1Agelow is a dummy variable taking the value of 1 if a firms’ age is below the it−1 median of firms’ age distribution in the period before the monetary surprise, and zero otherwise. Note again that Z includes the controls from the main text, plus 1EBPhigh it−1 ikt−1 and 1Agelow . it−1 Consistent with the direction of the heterogeneity in Cloyne et al. (2023), Panel B.18b of Figure B.18 highlights that the spreads of young firms are relatively more responsive to monetary policy shocks. Still, we see that our findings for heterogeneity by EBP from the main text are robust to conditioning on age (Panel B.18a). Monetary Policy on Investment: Next, we turn to confirm that the heterogeneous effects of monetary policy on invest- 36

Figure B.19 Monetary Policy’s Relative Effect on Investment by EBP vs. Age (a) Low EBP (b) High Age (Old) Marginal Effects Marginal Effects 20 20 10 10 0 0 -10 -10 0 4 8 12 0 4 8 12 Quarters after Shock Quarters after Shock Note.FigureB.19displaysdynamicinteractioncoefficientsfromahorseracebetween(A)therelative response of low-EBP firms’ investment compared to high-EBP firms’ (Panel B.19a) and (B) the relative response of high-age (old) firms’ investment compared to low-age (young) firms’ (Panel B.19b)fromamonetarypolicyshockεm fromestimating(B.19).Frequencyisquarterly.Conditional t results describe the additional effect of having the variable one standard deviation above the sample mean. Inner and outer shaded areas correspond to 68% and 90% confidence intervals constructed using two-way clustered standard errors by firm i and quarter t. ment by firms’ EBPs are robust to controlling for heterogeneity by age. We do so using: (cid:18) (cid:19) K log it+h = βh +βhεm +βhεm ×1EBPlow +βhεm ×1Agehigh +γhZ +e , K i 1 t 2 t it−1 3 t it−1 it−1 ith it−1 (B.19) where Z includes the controls from the main text, plus 1EBPlow and 1Agehigh. The it−1 it−1 it−1 results displayed in Figure B.19 highlight that the EBP indeed continues to regulate the responsiveness of firms’ investment to monetary policy. Surprisingly, we see in Panel B.19b that it is old firms whose investment response is larger compared to young firms following a monetary shock, in contrast to Cloyne et al. (2023), albeit only marginally. There are a few potential explanations. First, Cloyne et al. (2023) use a different measure of investment to what is used by Ottonello and Winberry (2020) and in our paper and, in addition, study investment growth rather than the level of investment. Since our model speaks to investment, we prefer our measure. Second, we focus on firms who use bond finance, which tend to be larger and older firms, such that our samples are not identical. Third, Cloyne 37

et al. (2023)’s monetary policy shocks are constructed from a proxy-VAR. Nonetheless, we show that heterogeneity by EBP is robust to controlling for heterogeneity by age. B.4.4 Size: As in Cloyne et al. (2023), Gertler and Gilchrist (1994) employ a dummy variable approach to assess how a firm’s size determines its sensitivity to monetary policy shocks. In this section, we measure size in assets and, as a measure of growth in size, we use sales growth, and compare each of their abilities to regulate firms’ responses to monetary policy to firms’ EBPs. Monetary Policy on Credit Spreads: We begin with monetary policy’s effect on credit spreads: S −S = βh +βhεm +βhεm ×1EBPhigh +βhεm ×1Sizelow +γhZ +e , ikt+h ikt−1 k 1 t 2 t ikt−1 3 t it−1 it−1 ikth (B.20) where 1Sizelow is a dummy taking the value of 1 if a firms’ assets (sales growth) are below the median in the period before the monetary shock, and 0 otherwise. Note again that Z it−1 includes the controls from the main text, plus 1EBPhigh and 1Sizelow . ikt−1 it−1 The results are displayed in Figure B.20. We see that while firms’ with low assets, that is small firms, have spreads who are more responsive to monetary policy, consistent with the findings in Gertler and Gilchrist (1994), sales growth does not seem to be a key determinant of the sensitivity of spreads. In both cases, heterogeneity by EBP is robust to controlling for the conditioning effects of these measures of (growth in) size. Monetary Policy on Investment: 38

Figure B.20 Monetary Policy’s Relative Effect on Spreads by EBP vs. Size (a) High EBP (b) Low Assets (Small) Marginal Effects Marginal Effects 0 0 -2 -2 -4 -4 -6 -6 0 5 10 15 20 0 5 10 15 20 Months after Shock Months after Shock (c) High EBP (d) Low Sales Growth Marginal Effects Marginal Effects 0 0 -2 -2 -4 -4 -6 -6 0 5 10 15 20 0 5 10 15 20 Months after Shock Months after Shock Note.FigureB.20displaysdynamicinteractioncoefficientsfromahorseracebetween(A)therelative responseofhigh-EBPbonds’spreadscomparedtolow-EBPbonds’(PanelB.20a)and(B)therelative response of low-asset-size (small) firms’ spreads compared to large firms’ (Panel B.20b) from a monetary policy shock εm from estimating (B.20). Panels B.20c and B.20d do the same but replace t small(inassets)firmswithlowsalesgrowthfirms.Frequencyismonthly.Conditionalresultsdescribe the additional effect of having the variable one standard deviation above the sample mean. Inner and outer shaded areas correspond to 68% and 90% confidence intervals constructed using two-way clustered standard errors by firm i and month t. Next, turning to investment, we estimate: (cid:18) (cid:19) K log it+h = βh +βhεm +βhεm ×1EBPlow +βhεm ×1Sizehigh +γhZ +e , K i 1 t 2 t it−1 3 t it−1 it−1 ith it−1 (B.21) where Z includes the controls from the main text, plus 1EBPlow and 1Sizehigh. it−1 it−1 it−1 39

Figure B.21 Monetary Policy’s Relative Effect on Investment by EBP vs. Size (a) Low EBP (b) Low Assets (Small) Marginal Effects Marginal Effects 20 20 10 10 0 0 -10 -10 0 4 8 12 0 4 8 12 Quarters after Shock Quarters after Shock (c) Low EBP (d) Low Sales Growth Marginal Effects Marginal Effects 20 10 10 0 0 -10 -10 -20 0 4 8 12 0 4 8 12 Quarters after Shock Quarters after Shock Note.FigureB.21displaysdynamicinteractioncoefficientsfromahorseracebetween(A)therelative response of low-EBP firms’ investment compared to high-EBP firms’ (Panel B.21a) and (B) the relative response of low-assets (small) firms’ investment compared to large firms’ (Panel B.21b) from a monetary policy shock εm from estimating (B.21). Panels B.21c and B.21d do the same but t replacesmall(inassets)firmswithlowsalesgrowthfirms.Frequencyisquarterly.Conditionalresults describe the additional effect of having the variable one standard deviation above the sample mean. Inner and outer shaded areas correspond to 68% and 90% confidence intervals constructed using two-way clustered standard errors by firm i and quarter t. We display the results in Figure B.21. The point-estimates in Panel B.21b indicate that, consistent with Gertler and Gilchrist (1994), small firms adjust investment more than large firms in response to monetary policy shocks. In addition, firms with high sales growth also adjust investment more following monetary shocks, as seen in Panel B.21d. In both cases, however, the EBP remains significant as a determinant of firms’ investment response to monetary policy. 40

Figure B.22 Monetary Policy’s Effect on Spreads by EBP vs. Liquidity of Assets (a) EBP (b) Liquidity of Assets Marginal Effects Marginal Effects 6 0 4 -2 2 -4 -6 0 0 5 10 15 20 0 5 10 15 20 Months after Shock Months after Shock Note.FigureB.22displaysdynamicinteractioncoefficientsfromahorseracebetweentheinteraction between a monetary policy shock and (A) the EBP (Panel B.22a) and (B) firms’ liquidity of assets (Panel B.22b) on the h-period change in credit spreads, S −S from estimating (B.22). ikt+h ikt−1 Frequency is monthly. Conditional results describe the additional effect of having the variable one standard deviation above the sample mean. Inner and outer shaded areas correspond to 68% and 90% confidence intervals constructed using two-way clustered standard errors by firm i and month t. B.4.5 Liquidity of Assets: Jeenas (2019) documents that the investment response to monetary policy of firms with lower share of liquid assets is relatively large, where liquidity is measured as the ratio of cash and short-term investments to total assets. He does so using our functional form from the main text, so we revert back to the conditioning on the average value of a firms’ characteristic over the previous year. Monetary Policy on Credit Spreads: We start by estimating: S −S = βh +βhεm +βhEBPma ×εm +βhεm ×Liqma +γhZ +e , ikt+h ikt−1 k 1 t 2 ikt−1 t 3 t it−1 it−1 ikth (B.22) where Liqma refers to the average share of liquid assets of firm i over the previous year. it−1 Note again that Z includes the controls from the main text, plus EBPma and Liqma . it−1 ikt−1 it−1 41

Figure B.23 Monetary Policy’s Effect on Investment by EBP vs. Liquidity of Assets (a) EBP (b) Liquidity of Assets Marginal Effects Marginal Effects 20 0 10 -10 0 -20 -10 0 4 8 12 0 4 8 12 Quarters after Shock Quarters after Shock Note.FigureB.23displaysdynamicinteractioncoefficientsfromahorseracebetweentheinteraction between a monetary policy shock and (A) the EBP (Panel B.23a) and (B) firms’ liquidity of assets (Panel B.23b) on h-period cumulative investment logK −logK from estimating (B.23). it+h it−1 Frequency is quarterly. Conditional results describe the additional effect of having the variable one standard deviation above the sample mean. Inner and outer shaded areas correspond to 68% and 90% confidence intervals constructed using two-way clustered standard errors by firm i and quarter t. The results are displayed in Figure B.22. We see that, consistent with the results in Jeenas (2019), firms with lower share of liquid assets experience a larger reduction in their credit spreads following a monetary easing (Panel B.23b), although the effects are relatively small.Bycontrast,theheterogeneouseffectsconditionalonfirms’EBPsarelargerandmore significant. Monetary Policy on Investment: Turning now to investment, we estimate: (cid:18) (cid:19) K log it+h = βh +βhεm +βhEBPma ×εm +βhεm ×Liqma +γhZ +e . (B.23) K i 1 t 2 it−1 t 3 t it−1 it−1 ith it−1 The results are displayed in Figure B.23. We see that controlling for liquidity of assets has little impact on on the EBP’s ability to regulate firms’ investment response to monetary policy. Heterogeneiety by firms’ liquid asset share is not statistically significant. 42

Figure B.24 Monetary Policy’s Effect on Spreads by EBP vs. Tobin’s Average Q (a) EBP (b) Tobin’s Average Q Marginal Effects Marginal Effects 6 0 4 -2 2 -4 0 -6 0 5 10 15 20 0 5 10 15 20 Months after Shock Months after Shock Note.FigureB.24displaysdynamicinteractioncoefficientsfromahorseracebetweentheinteraction between a monetary policy shock and (A) the EBP (Panel B.24a) and (B) firms’ average Tobin’s Q (Panel B.24b) on the h-period change in credit spreads, S −S from estimating (B.24). ikt+h ikt−1 Frequency is monthly. Conditional results describe the additional effect of having the variable one standard deviation above the sample mean. Inner and outer shaded areas correspond to 68% and 90% confidence intervals constructed using two-way clustered standard errors by firm i and month t. B.4.6 Tobin’s average Q: Tobin’s average Q has received comparatively less attention in the recent literature relative to other state variables we have examined in this section. Still, we show that heterogeneity by EBP is robust to controlling for the conditioning effects by Tobin’s average Q. Monetary Policy on Credit Spreads: Webeginbyaugmentingourmainspecificationfromthemaintextwiththeinteraction between the monetary policy shock and Tobin’s average Q: S −S = βh +βhεm +βhEBPma ×εm +βhεm ×Qma +γhZ +e , ikt+h ikt−1 k 1 t 2 ikt−1 t 3 t it−1 it−1 ikth (B.24) where Qma refers to the average Q of the firm over the preceding year, as in Jeenas (2019). it−1 The results are displayed in Figure B.24, and highlight that Tobin’s Q’s impact on the sensitivity of firms’ spreads to monetary policy shocks is not statistically significant. 43

Figure B.25 Monetary Policy’s Effect on Investment by EBP vs. Tobin’s Average Q (a) EBP (b) Tobin’s Average Q Marginal Effects Marginal Effects 20 0 10 -10 0 -20 -10 0 4 8 12 0 4 8 12 Quarters after Shock Quarters after Shock Note.FigureB.25displaysdynamicinteractioncoefficientsfromahorseracebetweentheinteraction between a monetary policy shock and (A) the EBP (Panel B.25a) and (B) firms’ average Tobin’s Q (Panel B.25b) on h-period cumulative investment logK −logK from estimating (B.25). it+h it−1 Frequency is quarterly. Conditional results describe the additional effect of having the variable one standard deviation above the sample mean. Inner and outer shaded areas correspond to 68% and 90% confidence intervals constructed using two-way clustered standard errors by firm i and quarter t. Moreover, this variable does not affect the role of the EBP as a state variable for the transmission of monetary policy to firm credit spreads. Monetary Policy on Investment: Turning now to investment, we estimate: (cid:18) (cid:19) K log it+h = βh +βhεm +βhEBPma ×εm +βhεm ×Qma +γhZ +e , (B.25) K i 1 t 2 it−1 t 3 t it−1 it−1 ith it−1 The results are displayed in Figure B.25. In Panel B.25b, the positive point-estimates, which are more statistically significant than for the credit spread regression, indicate that theinvestmentoffirmswithhigherTobin’sQsaremoresensitivetomonetarypolicyshocks. Still, heterogeneity by EBP is larger and more significant (Panel B.25a). 44

B.5 Alternative Monetary Policy Shocks In this subsection, we demonstrate the robustness of our results to the use of alternative monetary policy shocks, namely those of Swanson (2021). Swanson (2021) constructs a seriesofthreedistincttypesofmonetarypolicyshocks:(i)conventionalinterestrateshocks; (ii) forward guidance shocks; and (iii) asset purchase shocks. To provide comparability with our baseline Bu et al. (2021) monetary policy shock, which provides a unified measure of both conventional and unconventional U.S. monetary shocks, we sum across the three types of Swanson (2021) shocks. In what follows, we show that, as in the main text, the spreads of high-EBP bonds and investment of low-EBP firms are more responsive to this aggregated Swanson (2021) monetary policy shock series. Furthermore, the shapes of the impulse responses are very similar to those in our baseline specification. Monetary Policy on Credit Spreads: We begin by assessing the effects of a monetary policy easing on bond-level credit spreads, both unconditionally and conditional on a bond’s EBP, by estimating the local projections in equation (4) from the main text using the aggregated Swanson (2021) monetary policy shock series. The results are displayed in Figure B.26. They highlight that, as in the main text, a monetary policy easing induces a significant decline in credit spreads for the average firm (Panel B.26a). Moreover, consistent with our baseline results, the decline in credit spreads is larger for firms whose bonds carry a higher ex-ante EBP (Panel B.26b). Monetary Policy on Firm Investment: Next, we turn to evaluate the effects of a monetary policy easing on firm-level investment, both unconditionally and conditional on a bond’s EBP, by estimating the local projections in equation (6) from the main text using the aggregated Swanson (2021) monetary policy shock series. The results are displayed in Figure B.27. As in the main text, we see that a monetary easing induces an increase in investment for the average firm (Panel B.27a ). Furthermore, Panel B.27b highlights that this increase is larger for firms with ex-ante lower EBPs, which is consistent with our findings from the main text. In sum, the results showcase that our findings are not specific to the Bu et al. (2021) shock series. 45

Figure B.26 Monetary Policy’s Effect on Bond-Level Credit Spreads Depends on EBP (a) Unconditional (b) Conditional on EBP Marginal Effects Marginal Effects 0 0 -.5 -.5 -1 -1 -1.5 -1.5 0 5 10 15 20 0 5 10 15 20 Months after Shock Months after Shock Note. Figure B.26 presents the dynamic interaction effects (βh) between EBP and a Swanson 2 ikt−1 (2021) monetary policy shock on the h-period change in credit spreads, S −S from estiikt+h ikt−1 mating regression (4) from the main text. The frequency of the data is monthly. Conditional results describe the additional effect of having the variable one standard deviation above the sample mean. Inner and outer shaded areas correspond to the 68% and 90% confidence intervals constructed using two-way clustered standard errors by firm i and month t, respectively. Figure B.27 Monetary Policy’s Effect on Firm-Level Investment Depends on EBP (a) Unconditional (b) Conditional on EBP Marginal Effects Marginal Effects 5 0 2.5 -2.5 0 -5 0 4 8 12 0 4 8 12 Quarters after Shock Quarters after Shock Note. Figure B.27 presents the dynamic interaction effects (βh) between EBP and a Swanson 2 it−1 (2021) monetary policy shock (εm) series on h-period cumulative investment , logK −logK t it+h it−1 fromestimatingregression(6)fromthemaintext.Thefrequencyofthedataisquarterly.Conditional results describe the additional effect of having the variable one standard deviation above the sample mean.Innerandoutershadedareascorrespondtothe68%and90%confidenceintervalsconstructed using two-way clustered standard errors by firm i and quarter t, respectively. 46

B.6 EBP purged of Higher-Order Default-Risk In this section, we demonstrate that our results from the main text are robust to conditioning on firm EBPs that have been purged of potential higher-order dependence on default risk. Specifically, we re-estimate our credit spread regression (1) with the square of firms’ distance to default (DD2) as an additional regressor. Then, following the same steps as it in the baseline, we output a new EBP that is purged of its dependence on the square of its distance to default. We then re-assess our conclusion from sections 3, 4, and 6 that the EBPs regulate firms’ responsiveness to monetary policy using this new EBP measure. The results are displayed in Figures B.28, B.29 and B.30 for, respectively, the effects of monetary policy on credit spreads, monetary policy on investment, and credit spreads on investment. In all cases, our results are robust to using this new measure of firms’ EBP that is purged of the square of firms’ distance to default. Figure B.28 Monetary Policy’s Effect on Bond-Level Credit Spreads by EBP ex. DD2 (a) Baseline Conditional on EBP (b) Conditional on EBP ex. DD2 Marginal Effects Marginal Effects 0 0 -2 -2 -4 -4 -6 -6 0 5 10 15 20 0 5 10 15 20 Months after Shock Months after Shock Note.FigureB.28comparestheeffectsofthedynamicinteraction(βh)betweenEBP andtheBu 2 ikt−1 et al. (2021) monetary policy shock (εm) on the h-period change in credit spreads, S −S , t ikt+h ikt−1 from estimating regression (4) for 2 different EBPs. The first is our baseline EBP (Panel B.28a) and the second is the EBP purged of DD2 (Panel B.28b). The frequency of the data is monthly. Conditionalresultsdescribetheadditionaleffectofhavingthevariableonestandarddeviationabove the sample mean. Inner and outer shaded areas correspond to the 68% and 90% confidence intervals constructed using two-way clustered standard errors by firm i and month t, respectively. 47

Figure B.29 Monetary Policy’s Effect on Firm-Level Investment by EBP ex. DD2 (a) Baseline Conditional on EBP (b) Conditional on EBP ex. DD2 Marginal Effects Marginal Effects 0 0 -10 -10 -20 -20 0 4 8 12 0 4 8 12 Quarters after Shock Quarters after Shock Note. Figure B.29 compares the effects of the dynamic interaction (βh) between EBP and 2 ikt−1 the Bu et al. (2021) monetary policy shock (εm) on the h-quarter cumulative investment of firm i, t logK −logK , from estimating regression (6) for 2 different EBPs. The first is our baseline it+h it−1 EBP (Panel B.29a) and the second is the EBP purged of DD2 (Panel B.29b). The frequency of the data is quarterly. Conditional results describe the additional effect of having the variable one standard deviation above the sample mean. Inner and outer shaded areas correspond, respectively, tothe68%and90%confidenceintervalsconstructedusingtwo-wayclusteredstandarderrorsbyfirm and quarter. Figure B.30 Credit Spread’s Effects on Firm Investment by EBP ex. DD2 (a) Baseline Conditional on EBP (b) Conditional on EBP ex. DD2 Marginal Effects Marginal Effects 1.5 1.5 1 1 .5 .5 0 0 0 4 8 12 0 4 8 12 Quarters after Shock Quarters after Shock Note.FigureB.30comparestheeffectsofthedynamiceffect(βh)betweenEBP andachangein 2 ikt−1 creditspreads∆S onh-periodInvestmentoffirmi,logK −logK ,fromestimatingregression it it+h it−1 (12) for 2 different EBPs. The first is our baseline EBP (Panel B.30a) and the second is the EBP purged of DD2 (Panel B.30b). The frequency of the data is quarterly. Conditional results describe theadditionaleffectofhavingthevariableonestandarddeviationabovethesamplemean.Innerand outer shaded areas correspond to the 68% and 90% confidence intervals constructed using two-way clustered standard errors by firm i and quarter t, respectively. 48

B.7 Monetary Policy’s Effect on Firm-Level Debt Issuance In this section, we show that—just as how investment increases by more for low-EBP firms following a shock monetary policy easing than for high-EBP firms—low-EBP firms increase debt-issuance compared to high-EBP ones following a monetary easing. We demonstrate this using the dummy-variable conditioning method outlined in Section B.2. Results are similar with our baseline linear functional form for the EBP interaction, but are more noisy. As in our investment specification, to assess the distinct responses of low- and high- EBP firms’ growth in debt issuance following a monetary shock, we estimate: (cid:18) (cid:19) D log it+h = βh +βhεm ×1EBPlow +βhεm ×1EBPhigh +γhZ +e , (B.26) D i 1 t it−1 2 t it−1 it−1 ith it−1 where D is firm i’s real outstanding debt (short- plus long-term) in period t and where i,t Z includes the controls from the main text, plus EBPlow and EBPhigh. The results are it−1 it−1 it−1 displayed in Figure B.31 and highlight that only low-EBP firms increase debt following a monetary easing, which is consistent with our investment results. 49

Figure B.31 Monetary Policy’s Effect on Firm Debt Issuance for Low- vs High-EBP Firms (a) Low-EBP Firms (b) High-EBP Firms Marginal Effects Marginal Effects 30 30 20 20 10 10 0 -10 0 -10 0 4 8 12 0 4 8 12 Quarters after Shock Quarters after Shock Note.FigureB.31tracestheresponseofdebtissuancegrowthforlow-EBP(1EBPlow)firmsinPanel B.31a and high-EBP (1EBPhigh) firms in Panel B.31b to a Bu et al. (2021) monetary policy shock (εm), from estimating regression (B.26), where the frequency is quarterly. The frequency of the data t is quarterly. Conditional results describe the additional effect of having the variable one standard deviation above the sample mean. Inner and outer shaded areas correspond to the 68% and 90% confidence intervals constructed using two-way clustered standard errors by firm i and quarter t, respectively. 50

B.8 Intermediary Net Worth Shocks and EBP Heterogeneity In this section, we study how shocks to the net worth of financial intermediaries influence firms’ credit spreads and investment conditional on their EBPs. We measure these shocks using the orthogonalized intermediary capital risk factor of He et al. (2017). We first assess the effect on credit spreads by replacing the monetary policy shock εm t in our baseline monetary policy specification (4) with the net-worth shock εNW: t S −S = βh +βhεNW +βhEBPma ×εNW +γhZ +e , (B.27) ikt+h ikt−1 k 1 t 2 ikt−1 t it−1 ikth The unconditional (Panel B.32a) and conditional (Panel B.33b) results are displayed in Figure B.32. They show that a shock increase in intermediary net worth lowers firms’ credit spreads, and that this decrease is larger for firms with higher EBPs. Thus, the effects of intermediary net-worth shocks are qualitatively similar to those of monetary policy shocks. Figure B.32 Intermediary Net Worth Shocks on Credit Spreads by EBP (a) Unconditional (b) Conditional on EBP Marginal Effects Marginal Effects .1 0 0 -.1 -.1 -.2 -.2 0 6 12 18 0 6 12 18 Months after Shock Months after Shock Note. Figure B.32 reports the dynamic effects of an intermediary net worth shock εNW on the ht month change in bond credit spreads, S −S , which we estimate using regression (B.27). ikt+h ikt−1 Panel B.32a shows the unconditional effects, βh. Panel B.33b shows the effects conditional on 1 EBPma , βh. Conditional results describe the additional effect of having the variable one stanik,t−1 2 darddeviationabovethesamplemean.Innerandoutershadedareasare,respectively,68%and90% confidence intervals constructed using two-way clustered standard errors by firm and month. 51

Figure B.33 Intermediary Net Worth Shocks on Investment by EBP (a) Unconditional (b) Conditional on EBP Marginal Effects Marginal Effects 2 0 1 -1 0 -2 0 6 12 18 0 6 12 18 Quarters after Shock Quarters after Shock Note. Figure B.33 reports the dynamic effects of an intermediary net worth shock εNW on the t h-quarter cumulative investment of firm i, log(K /K ), which we estimate using regression it+h it−1 (B.28).PanelB.33ashowstheunconditionaleffects,βh.PanelB.33bshowstheeffectsconditionalon 1 EBPma ,βh.Conditionalresultsdescribetheadditionaleffectofhavingthevariableonestandard ik,t−1 2 deviation above the sample mean. Inner and outer shaded areas are, respectively, 68% and 90% confidence intervals constructed using two-way clustered standard errors by firm and quarter. Next, we perform a similar exercise by using the intermediary net worth shock in our baseline investment specification (6): (cid:18) (cid:19) K log it+h = βh +βhεNW +βhEBPma ×εNW +γhZ +e , (B.28) K i 1 t 2 it−1 t it−1 ith it−1 The results are displayed in Figure B.33 and highlight that a shock increase in intermediary net worth leads to an increase in firms’ investment (Panel B.33a) which is larger for firms with lower EBPs (Panel B.33b). Again, this consistent with our baseline monetary policy results. Overall, this exercise reinforces the notion that firm EBPs reflect the slope of firms’ marginal benefit curves for capital and are an important state variable for understanding firms’ responsiveness to shifts in their marginal cost curves. 52

B.9 Aggregate Implications of EBP Heterogeneity In this section, we highlight the robustness of our conclusions from Section 6.2, where we showed that variation in the cross-sectional distribution of firm EBPs has important implications for the aggregate effectiveness of monetary policy. Specifically, we document thatourresultsarerobustto:(i)measuringmomentsoftheEBPdistributionusingdifferent percentiles; (ii) conditioning directly on the percentiles of the EBP distribution; (iii) using the aggregated Swanson (2021) monetary policy shocks; (iv) a horserace between monetary policy’s interaction with the moments of the EBP distribution and its interaction with various recession indicators. First,weshowthatourresultsfromSection6.2arenottiedtotheparticularpercentiles we use to construct the moments of the EBP distribution, the 10th and 90th percentiles. To demonstrate this, we re-estimate regression (13) by constructing our moments using the 5th and 95th percentiles, the 15th and 85th percentiles, the 20th and 80th percentiles, and the 25th and 75th percentiles of the EBP distribution. Figure B.34 presents the results, focusing on the skewness of the EBP distribution. In all cases, we see that an increase in skewness dampens the impact of a monetary easing on aggregate investment, consistent with our conclusions from the main text. Second,ratherthanconditioningonthemomentsoftheEBPdistribution,wecondition on the percentiles used to construct these moments, in particular, the 10th, 50th (median), and 90th percentiles. The results are displayed in Figure B.35 and highlight that on-impact a rise in median EBP and a fall in the 90th percentile of the EBP distribution dampens the effect of monetary policy on aggregate investment. Further, only the left-tail of the EBP distribution matters at medium horizons, where an increase meaningfully dampens the effects of expansionary monetary policy shocks on aggregate investment. This suggests that the 10th percentile of the EBP distribution is responsible for the conditioning effects of the EBP distribution’s skewness and dispersion from our baseline specification. Third, we re-estimate our baseline specification using the aggregated Swanson (2021) monetary policy shocks discussed in Appendix B.5. The impulse responses displayed in 53

Figure B.36 are qualitatively similar to those from the main text. Finally, we examine the extent to which the EBP distributions’s impact on the aggregate effectiveness of monetary policy is related to the well-documented weaker effects of monetary policy in recessions. We do so by running horseraces between our moment interactions and interactions between the monetary policy shocks and two types of (lagged) recession indicators: (i) the smoothed U.S. recession probability measure from Chauvet (1998); (ii) a dummy variable for NBER-classified U.S. recessions. In particular, the Chauvet(1998)measureverycloselytrackstherecessionmeasureusedinTenreyroandThwaites (2016). The results are displayed in Figures B.37 and B.38. There are three key takeaways. First, an increase in the probability of a U.S. recession or the incidence of a recession severely dampens the expansionary power of an easing U.S. monetary policy shock, consistent with the existing evidence. Second, the inclusion of these interactions does not distort the conditioning power of the skewness of the EBP distribution, nor the dispersion, highlighting the generality of the relationship between the slope of firms’ marginal benefit curves and the aggregate effectiveness of monetary policy. Third, the conditioning effects of the median of the EBP distribution are crowded out by the recession indicators. This is consistent with Gilchrist and Zakrajˇsek (2012)’s result that aggregate EBP rises in recessions and suggests a potential new transmission channel for monetary policy’s weaker effects in recessions: the steeper slopes of firms’ marginal benefit curves around equilibrium. 54

Figure B.34 EBP Skewness and Monetary Policy’s Effect on Aggregate Investment (a) Conditional on 95-05 EBP Skewness (b) Conditional on 85-15 EBP Skewness Marginal Effects Marginal Effects 15 15 0 0 -15 -15 -30 -30 0 4 8 12 0 4 8 12 Quarters after Shock Quarters after Shock (c) Conditional on 80-20 EBP Skewness (d) Conditional on 75-25 EBP Skewness Marginal Effects Marginal Effects 15 15 0 0 -15 -15 -30 -30 0 4 8 12 0 4 8 12 Quarters after Shock Quarters after Shock Note. Figure B.34 reports the dynamic effects from monetary policy shocks, conditional on the skewness of the EBP distribution (βh),on the h-quarter cumulative aggregate investment, 400/(h+ 2 1)log(I /I ), estimated using regressions (13). Panel B.34a, B.34b, B.34c, and B.34d measure t+h t−1 skewness using the 95-05, 85-15, 80-20 and 75-25 percentiles of the EBP distribution, respectively. Conditionalresultsdescribetheadditionaleffectofhavingthevariableonestandarddeviationabove the sample mean. Inner and outer shaded areas correspond, respectively, to the 68% and 90% confidence intervals constructed using Newey-West standard errors with 12 lags. 55

Figure B.35 EBP Percentiles and Monetary Policy’s Effect on Aggregate Investment (a) Unconditional (b) Conditional on 10th Percentile EBP Marginal Effects Marginal Effects 50 25 25 0 0 -25 -25 -50 0 4 8 12 0 4 8 12 Quarters after Shock Quarters after Shock (c) Conditional on Median EBP (d) Conditional on 90th Percentile EBP Marginal Effects Marginal Effects 25 75 0 50 -25 25 -50 0 -75 -25 0 4 8 12 0 4 8 12 Quarters after Shock Quarters after Shock Note. Figure B.35 reports the dynamic effects from monetary policy easing shocks on h-quarter cumulative aggregate investment, 400/(h+1)log(I /I ), estimated using a variant of regression t+h t−1 (13). Panel B.35a shows unconditional effects (βh). Panels B.35b, B.35c and B.35d shows effects 1 conditional on the 10, 50 and 90th percentiles of the EBP distribution, respectively. Conditional results describe the additional effect of having the variable one standard deviation above the sample mean.Innerandoutershadedareascorrespond,respectively,tothe68%and90%confidenceintervals constructed using Newey-West standard errors. 56

Figure B.36 Monetary Policy’s Effect on Aggregate Investment Growth (a) Unconditional (b) Conditional on EBP Skewness Marginal Effects Marginal Effects 10 5 5 0 0 -5 -5 -10 0 4 8 12 0 4 8 12 Quarters after Shock Quarters after Shock (c) Conditional on Median EBP (d) Conditional on EBP Dispersion Marginal Effects Marginal Effects 5 10 0 5 -5 0 -10 -5 0 4 8 12 0 4 8 12 Quarters after Shock Quarters after Shock Note. Figure B.36 reports the dynamic effects of a Swanson (2021) monetary policy easing shock εm t onh-quarterannualizedaggregateinvestmentgrowth,400/(h+1)log(I /I ),whichweestimate t+h t−1 using regression (13). Panel B.36a shows unconditional effects, βh. Panels B.36b, B.36c and B.36d 1 showtheeffectsconditionalontheskewness,mediananddispersionoftheEBPdistribution,thethree elements in βh, respectively. Conditional results describe the additional effect of having the variable 2 one standard deviation above the sample mean. Inner and outer shaded areas are, respectively, 68% and 90% confidence intervals constructed using Newey-West standard errors with 12 lags. 57

Figure B.37 Monetary Policy’s Effect on Aggregate Investment Growth (a) Unconditional (b) Conditional on EBP Skewness Marginal Effects Marginal Effects 30 15 0 15 -15 0 -30 -15 0 4 8 12 0 4 8 12 Quarters after Shock Quarters after Shock (c) Conditional on Median EBP (d) Conditional on EBP Dispersion Marginal Effects Marginal Effects 30 30 15 15 0 0 -15 -15 0 4 8 12 0 4 8 12 Quarters after Shock Quarters after Shock (e) Conditional on Recession Probability Marginal Effects 15 0 -15 -30 0 4 8 12 Quarters after Shock Note.FigureB.37reportsthedynamiceffectsofamonetarypolicyeasingshockεm onh-quarterant nualizedaggregateinvestmentgrowth,400/(h+1)log(I /I ),whichweestimateusingregression t+h t−1 (13). Panel B.37a shows unconditional effects, βh. Panels B.37b, B.37c and B.37d show the effects 1 conditional on the skewness, median and dispersion of the EBP distribution, the three elements in βh, respectively. Panel B.37e shows the effects conditional on the probability of a recession. Con- 2 ditional results describe the additional effect of having the variable one standard deviation above the sample mean. Inner and outer shaded areas are, respectively, 68% and 90% confidence intervals 58 constructed using Newey-West standard errors with 12 lags.

Figure B.38 Monetary Policy’s Effect on Aggregate Investment Growth (a) Unconditional (b) Conditional on EBP Skewness Marginal Effects Marginal Effects 15 30 0 15 -15 0 -30 -15 0 4 8 12 0 4 8 12 Quarters after Shock Quarters after Shock (c) Conditional on Median EBP (d) Conditional on EBP Dispersion Marginal Effects Marginal Effects 30 15 30 0 15 0 -15 -15 0 4 8 12 0 4 8 12 Quarters after Shock Quarters after Shock (e) Conditional on NBER Recession Marginal Effects 0 -50 -100 -150 0 4 8 12 Quarters after Shock Note. Figure B.38 reports the dynamic effects of a monetary policy easing shock εm on h-quarter t annualized aggregate investment growth, 400/(h+1)log(I /I ), which we estimate using ret+h t−1 gression (13). Panel B.38a shows unconditional effects, βh. Panels B.38b, B.38c and B.38d show the 1 effects conditional on the skewness, median and dispersion of the EBP distribution, the three elementsinβh,respectively.PanelB.38eshowstheeffectsconditionalonanNBER-classifiedrecession. 2 Conditionalresultsdescribetheadditionaleffectofhavingthevariableonestandarddeviationabove the sample mean. Inner and outer shaded areas are, respectively, 68% and 90% confidence intervals 59 constructed using Newey-West standard errors with 12 lags.

C Model Appendix In this section, we provide further information about our model. In particular, we present the model’s parameterization (Section C.1); provide further details on the relationship between a firm’s EBP and the slope of its marginal benefit curve in the model (Section C.2) and in the data (Section C.3); and discuss the empirical and model-implied link between firm EBPs and their capital stock (Section C.4). C.1 Model Parameterization Table C.1 Benchmark Model Parameterization Parameter Value Description N 0.02 Intermediary Net-Worth Pre-Shock 1 N 0.055 Intermediary Net-Worth Post-Shock 2 R 1 Safe Interest Rate α 0.71 Cobb-Douglas capital elasticity L α 0.9 Cobb-Douglas capital elasticity H θ(K ) 0.9K1.25 Agency Friction t t Table C.1 presents our model’s parameterization. Among the parameters are the net-worth of intermediaries before and after the shock, which we select such that intermediaries’ constraints bind for both firms. The safe interest rate, R, is set to 1 in the model for simplicity. As mentioned in the main text, we vary the slope of firms’ marginal benefit curves for capital by adjusting α, the intensity of capital in firms’ Cobb-Douglas production functions. We calibrate α = 0.71 and α = 0.90 by estimating production functions for L H firms in the bottom (L) and top (H) quintiles of the EBPma distribution using regression it (10). These results are presented in Table C.5 in Appendix C.3. In addition, we follow Gabaix and Maggiori (2015) by assuming that the fraction of theirrevenuesintermediariescandivertisincreasinginthesizeoftheirbalance-sheet:θ(K ). t Thefunctionalform0.9K1.25 isselectedtogeneratean(approximately)linearmarginalcost t 60

of capital curve, which allows us to focus on heterogeneity in the slope of firms’ marginal benefit curves. C.2 Firm EBPs and Marginal Benefit Curves in the Model Figure 7 in the main text documents the relationship between a firm’s EBP and the slope of its marginal benefit curve for capital in our model. Specifically, using our parameterization in Table C.1, firms with flatter marginal benefit curves near the equilibrium have lower equilibrium EBPs. In what follows, we show that this result holds for most levels of intermediary net worth (N). In the equilibria shown in Figure 7, the high-α (α ) firm has both a lower EBP and H a flatter marginal benefit curve (Panel 7a). From inspection, there are two potential ways this result could be violated: (i) intermediaries have sufficiently high net worth; and (ii) intermediaries have sufficiently low net worth. We discuss these two cases in turn. Case (i): intermediaries have sufficiently high net worth. As the marginal benefit curve of the firm with low α (α ) (Panel 7b) intersects the horizontal axis (RK = R) before L the firm with α (Panel 7a), we know that for sufficiently high intermediary net-worth, H the α -firm will have a lower equilibrium EBP. Thus, there exists an equilibrium in which L (a) intermediaries’ net worth is ε > 0 below this level, and (b) the α -firm has both the L lower-EBP and the steeper marginal benefit curve. We now bound this level of intermediary net worth and show that it is almost identical to the intermediary net worth for which the α -firm has a credit spread of 1 under our parameterization. L When intermediary net worth N, and hence equilibrium capital, is sufficiently high, the α -firm always has a flatter marginal benefit curve but only has a lower EBP if H α KαH−1 < α KαL−1, where K and K denote the α - and α -firms’ equilibrium capital H H L L H L H L stock, respectively. The cutoff level of capital stock K∗ for which this ceases to hold occurs at the intersection of the two firms’ marginal benefit curves: (cid:104)α (cid:105) 1 K∗ = L αH−αL. (C.1) α H 61

The N for which the α -firm has K < K∗ can be found from α KαH−1 = KH−N , or: H H H H KH(1−θ) N =K −α KαH(1−θ) H H H (cid:104)α (cid:105) 1 (cid:104)α (cid:105) αH (cid:104)α (cid:105) 1 N < L αH−αL −α L αH−αL(1−θ( L αH−αL)) (C.2) H α α α H H H If N is below the value in (C.2), then the α -firm has both a flatter marginal benefit curve H and a lower EBP in equilibrium. In our baseline parameterization, this is N (cid:47) 0.07, which is nearly identical to the N that makes the α -firm have a credit spread of 1, which is very H rare in practice.33 Case (ii): intermediaries have sufficiently low net worth. This condition, as it turns out, does not have any “bite” under our parameterization. When N (cid:47) 0.07, and especially for small N, the α -firm has the lower EBP but may not have the flatter marginal benefit H curve. We show, in fact, that the α -firm always has the flatter marginal benefit curve for H N ≥ 0 by setting N = 0 and showing: |α (α −1)KαH−2| < |α (α −1)KαL−2|, (C.3) H H H L L L under our paramaterization. Solving for the equilibrium capital stock when N = 0 one finds that inequality (C.3) holds. C.3 Firm EBPs and Marginal Benefit Curves in the Data In our model in the main text, we show that higher-α firms with flatter marginal benefit curves for capital have lower EBPs, which we find support for by estimating production functions for low- and high-EBP firms. In this section, we highlight the robustness of these empiricalresultsandhighlighttheempiricalestimatesweusetocalibratecapitalintensities in our model. To begin, we consider robustness to the threshold minimum number of consecutive 33For this calculation, we set θ = 0.9 rather than θ = 0.9K1.25. The latter would restrict this bound t further. 62

Table C.2 α Estimates for Low- and High-EBP Firms by Minimum Firm Observations (a) Model Analogue (b) Full Specification logY logK logY logK i,t i,t i,t i,t Min. Obs. Low-EBP High-EBP Min. Obs. Low-EBP High-EBP 20 quarters 0.83∗∗∗ 0.73∗∗∗ 20 quarters 0.20∗∗∗ 0.13 (.034) (.034) (.033) (.094) 25 quarters 0.87∗∗∗ 0.77∗∗∗ 25 quarters 0.18∗∗∗ 0.13 (.034) (.035) (.062) (.118) 30 quarters 0.88∗∗∗ 0.77∗∗∗ 30 quarters 0.19∗∗∗ 0.14 (.037) (.037) (.043) (.099) 35 quarters 0.87∗∗∗ 0.76∗∗∗ 35 quarters 0.18∗∗∗ 0.13 (.042) (.039) (.040) (.085) Note. Table C.2 presents estimates of the capital intensity (α) of firms with EBPma below and above it the firm-level median each period (labeled as “Low-EBP” and “High-EBP”, respectively) depending on the threshold minimum number of firms’ consecutive observations. Table C.2a presents the results from estimatingregression(10).TableC.2bpresentstheresultsfromestimatingregression(11).Thethresholds consideredare20,25,30and35quarters.Theresultsfor30quartersarethosepresentedinTable2inthe main text. Standard errors are two-way clustered by firm and quarter in Table C.2a and bootstrapped in Table C.2b. *** denotes statistical significance at the 1% level. observations firms must have to be included in the sample. Table C.2 presents estimates of the capital elasticity α of below- and above-median EBP firms from estimating regression (10) in Table C.2a and regression (11) in Table C.2b for 4 different threshold values: 20, 25, 30 and 35 quarters. The table highlights that, for both the model analogue and full specification case, the empirical result that low-EBP firms have higher capital intensities is very robust to the threshold observation level. One small difference is that, in the model analogue case, the estimated α for both low- and high-EBP firms is slightly lower for the 20 quarter cutoff as compared to the other three, where the α estimates are all very similar. Due to this similarity, and because observing firms at many different levels of capital helps improve the estimates of α, we view these higher cutoffs as more representative of the true α and we select the 30 quarter threshold as our baseline in the main text (Table 2). The remainder of the results from this section are estimated using this threshold as well.34 That being said, all our results carry through if we were to use any of these threshold values. Next, in the model analogue case, we document that our empirical result that low- 34Of note, 20 quarters is the threshold we use for the analysis in Section 4. 63

Table C.3 α Estimates for Low-, High-EBP Firms with Time/Sector-Time Fixed Effects logY Low-EBP High-EBP Low-EBP High-EBP i,t logK 0.81∗∗∗ 0.76∗∗∗ 0.95∗∗∗ 0.90∗∗∗ i,t (.041) (.037) (.057) (.073) Time FE Yes Yes No No Time-Sector FE No No Yes Yes Note: Table C.3 presents estimates of the capital intensity (α) of firms with EBPma below it and above the firm-level median each period (labeled as “Low-EBP” and “High-EBP”, respectively) from estimating regression (10) in the main text but augmented with either Time or Time-Sector Fixed Effects. Standard errors are two-way clustered by firm and quarter and *** denotes statistical significance at the 1% level. Table C.4 Differences in α Estimates Between Low- and High-EBP Firms logY logY logY i,t i,t i,t logK 0.82∗∗∗ 0.78∗∗∗ 0.92∗∗∗ i,t (.031) (.034) (.062) logK ×1EBPma,Low .013∗ .018∗∗ .011∗ i,t i,t (.008) (.008) (.007) Time FE No Yes No Time-Sector FE No No Yes Note: Table C.4 presents estimates of differences in the capital intensity (α) between firms withEBPmabelowandabovethefirm-levelmedianusingamodifiedversionofregression(10) it inthemaintext,logY =β +αlogK +γ logK ×1EBPma,Low+γ 1EBPma,Low+ε , i,t i i,t 1 i,t i,t 2 i,t i,t that we also augment with either Time or Time-Sector Fixed Effects. Standard errors are two-wayclusteredbyfirmandquarterand***,**,and*denotestatisticalsignificanceatthe 1%, 5% and 10% levels, respectively. EBP firms have higher capital intensities is robust to estimating regression (10) including both time as well as sector-time fixed effects. These results are displayed in Table C.3. In addition, we show that the α estimates for below- and above-median EBP firms are statistically distinct from each other. This is true also when including time and sector-time fixed effects. These results are displayed in Table C.4. Finally, we provide estimates in Table C.5 of α for firms in the bottom and top terciles, quartiles and quintiles of the EBP distribution, in addition to above and below the median. We see that the gap between the α estimates widens as we go deeper in the tails, peaking with estimates of α = 0.90 and α = 0.71 for firms with EBPmas in the bottom and top H L i,t 64

Table C.5 Estimates of α for Low- and High-EBP Firms by Percentiles logY logK i,t i,t Low & High Percentile Cutoffs Less than EBPLow Greater than EBPHigh 50 & 50 0.88∗∗∗ 0.77∗∗∗ (.037) (.037) 33 & 67 0.89∗∗∗ 0.75∗∗∗ (.050) (.043) 25 & 75 0.89∗∗∗ 0.71∗∗∗ (.059) (.050) 20 & 80 0.90∗∗∗ 0.71∗∗∗ (.060) (.055) Note: Table C.5 presents estimates of the capital intensity (α) of firms with EBPma below it and above certain percentiles of the firm-level EBPma distribution in a given period from it estimating regression (10). The percentiles considered are (1) below and above the median, (2) below 33rd and above 67th, (3) below 25th and above 75th, (4) below 20th and above 80th. The resultsfor (1),below andabove the median,are thosepresented in(10) inthe main text. The results for (4), below 20th and above 80th, are used to calibrate the α parameters in the model. Standard errors are two-way clustered by firm and quarter and *** denotes statistical significance at the 1% level. quintiles of the EBP distribution. These are the estimates we use to calibrate our model in Section 5. C.4 Firm EBPs and Capital Stock: Model and Data Finally, the model under our baseline parameterization suggests that when the low-EBP firm has a flatter marginal benefit curve, it also has a lower capital stock. Table C.6 highlights that, without controls, this positive relationship between firm EBPs and their capital stock is present in the data. However, when adding controls, we see that the EBP and firms’ capital stock appear unrelated. As alluded to Section 5, one can make firm EBPs unrelated to their capital stock if higher-α firms with flatter marginal benefit curves receive “preferential sentiment” from intermediaries (Lo´pez-Salido et al., 2017), modelled as a looser compatibility constraint, i.e., a lower θ. By pushing the α -firm’s marginal cost curve out- H ward, preferential sentiment would further decrease the α -firm’s EBP and place it on an H even flatter segment of its marginal benefit curve in equilibrium, implying our comparative statics results would continue to hold in a model with both heterogeneity in firms’ capi- 65

Table C.6 Firm EBPs and Capital Stock Vars logK logK logK logK i,t i,t i,t i,t EBPma 0.028*** 0.01 -.010 -.007 i,t (.008) (.006) (.010) (.009) Firm FE Yes Yes Yes Yes Time-Sector FE No No Yes Yes Firm Controls No Yes No Yes Note: Table C.6 presents the marginal effects β from the following regression: logK = 1 it β +α +β EBPma +γhW +ε where W is the vector of firm-level control variables i s,t 1 it it it it described in Section 2.4. Standard errors at two-way clustered by firm i and quarter t. *** denotes statistical significance at the 1% level. tal intensities and preferential sentiment by intermediaries for high-α firms. Furthermore, in addition to making firms’ capital stocks unrelated to their EBPs, this preference by intermediaries for high-α firms could also be the source segmentation across islands. 66