Monetary Policy and Bank Risk-Taking: Evidence from the Corporate Loan Market Teodora Paligorova∗ Bank of Canada E-mail: [email protected]

Jo˜ao A. C. Santos∗ Federal Reserve Bank of New York and Nova School of Business and Economics E-mail: [email protected]

November 24, 2014

∗

The authors thank Jason Allen, Jose Berrospide, Philip Bond, Christa Bouwman, Daniel Carvalho, Todd Gormley, Song Han, Scott Hendry, Kim Huynh, Paul Kupiec, David Martinez-Miera, Pablo Moran, Alisa Roell, Philip Strahan, Jonathan Witmer and seminar participants at Nova School of Business and Economics, SFU Beedie School of Business, Queens Business School, University of Alberta, Instituto Superior de Economia e Gest˜ ao, the 2012 FIRS Meeting in Minneapolis, the 2012 Bank of Spain and Bank of Canada “International Financial Markets” Workshop, the 2012 Northern Finance Association Meeting, the 2013 Day Ahead Conference, the 2013 European Winter Finance Conference, the 2013 Western Finance Association for useful comments. We thank Vitaly Bord for outstanding research assistance. The views stated herein are those of the authors and are not necessarily the views of the Bank of Canada, the Federal Reserve Bank of New York or the Federal Reserve System.

Monetary Policy and Bank Risk-Taking: Evidence from the Corporate Loan Market Abstract Our study of corporate loan pricing policies of U.S. banks over the past two decades shows that the difference in loan spreads between riskier and safer borrowers decreases in periods of monetary policy easing compared to monetary policy tightening. Using individual bank information about lending standards from the Senior Loan Officers Opinion Survey (SLOOS), we unveil evidence that the interest rate discount for riskier borrowers in periods of easing monetary policy is prevalent among banks with greater risk appetite. These findings provide evidence of the bank risk-taking channel of monetary policy.

JEL classification: G21 Key words: Monetary policy, risk-taking channel, loan spreads

1

Introduction

The effect of monetary policy on banks’ risk-taking incentives have received wide attention in the wake of the latest financial crisis, following claims that the Federal Reserve accommodative policies spurred risk-taking among financial intermediaries.1 Banks often enter into long-term contracts that commit them to producing high nominal rates of return. In periods of low interest rates, these contractual rates may exceed the yields available on safe assets, leading them to invest in risky assets to earn excess returns.2 Banks’ search-for-yield is important because it can lead to the buildup of risky credit in the economy (e.g., Stein (2013), Yellen (2011)). In this paper, we contribute to the ongoing debate on the effect of monetary policy on banks’ risk-taking incentives by investigating whether U.S. monetary policy have led banks to ‘underprice’ risk of corporate loans over the past two decades. We study the risk-taking channel of monetary policy by investigating U.S. banks’ loan pricing policies in different monetary policy regimes. As has been recognized by other studies of the bank risk-taking channel of monetary policy, one benefit of looking at the loan pricing (and loan quantities) is that they should summarize (bank) risk-taking incentives reflected in various non-price effects. A novel feature of our paper is that we use a direct measure of bank risk appetite based on survey-based responses from loan officers on whether bank lending standards have eased because of greater tolerance to risk. This information is important because it allows us to directly quantify the risk-taking incentives of each bank. All of the existing studies of the bank-risk taking channel of monetary policy rely on bank balance sheet information to infer whether banks have greater appetite for risk. Accounting balance sheet data, however, makes it hard to separate the effect of bankers’ incentives from more general bank balance sheet movements that may be seemingly consistent with risk-taking behavior but driven by other underlying factors. Our results show that banks charge risky borrowers lower loan spreads (compared to safe borrowers) in periods of easing relative to periods of tightening monetary policy after 1

See Rajan (2006), Borio and Zhu (2008), Brunnermeier (2009), and Diamond and Rajan (2009).

2

Low interest rates may also lead to more bank risk-taking because of their positive impact on collateral values and on banks’ risk-taking capacity (Adrian and Shin (2009)), or by virtue of their negative effect on banks’ screening incentives (Dell′ Ariccia and Marquez (2009)), or through a money illusion effect (Akerlof and Shiller (2009)).

1

controlling for a large set of borrower-, loan- and bank-specific factors as well as a set of macroeconomic factors known to affect loan rates.3 A one standard deviation increase in one of our measures of monetary policy easing is associated with 11% decrease in the loan spread for riskier borrowers. While consistent with the existence of a risk-taking channel of monetary policy, this finding may also derive from other explanations. In the remainder of the paper we try not only to rule out several alternative explanations for that finding, but also to present direct evidence that it is driven precisely by banks’ risk-taking appetite. The first concern is that the monetary policy and economic conditions are not easily separable because they arise endogenously. For that purpose we rely on several measures of monetary policy shocks that are arguably uncorrelated with the macroeconomic environment. Since there is no consensus on the best measure of monetary policy shocks, we show that our findings continue to hold when we consider various measures employed in previous studies such as changes in the federal funds rate, Taylor rule residuals, and Romer and Romer residuals (Jimenez et al. (2014), Coibion (2012), Kashyap and Stein (2000)).4 Romer and Romer (2004) take a narrative approach to constructing the monetary policy shocks (residuals). These residuals have a meaningful interpretation—they reflect the evolution of the Federal Reserve’s operating procedures, policy makers beliefs about the workings of the economy, and the Federal Reserve’s tastes and goals. The next concern we address is to make sure that our findings do not derive from changes in the demand for credit across different monetary policy regimes. We use a sample of individual loan level data matched with time-varying borrower and bank characteristics. To the extent that our firm controls account for changes in firm demand over time, including them in the loan spread regression should reduce concerns related to the effect of credit demand. Additionally, we study loan quantities reasoning that lower loan spreads together with larger loan amounts to riskier borrowers are consistent with loan supply shifts. We find that more risky borrowers receive relatively larger loans than safer borrowers in easing monetary policy 3

We use the terms “low interest rates,” “loose monetary policy,” and “easy monetary policy” interchangeably. The same applies for “high interest rates” and “tight monetary policy.” 4

Jimenez et al. (2014) deal with this problem by relying on the observation that monetary policy in Spain is exogenous during the sample period. In turn, Ioannidou et al. (2009) do so by looking at the effect of U.S. monetary policy on Bolivian banks, pointing that U.S. monetary policy affects the Bolivian economy through the exchange rate peg, but the Bolivian economy has no role in the U.S. monetary policy.

2

regimes after controlling for firm, loan, bank and macroeconomic characteristics. Further, to account for unobservable changes in loan demand we include firm-time and firm-bank-time fixed effects in the loan spread regressions. In this case, the identification comes either from comparing loan spreads for the same firm in the same time period across different banks (firmtime) or within the same bank (firm-bank-time). Controlling for time-varying bank characteristics is important because it allows us to isolate the portion of loan pricing that is due to changes in bank balance sheet constraints possibly induced by changes in monetary policy. Further, our use of bank fixed effects will account for unobservable supply effects in loan interest rates. This approach, however, does not account for unobservable changes in the pool of borrowers across different monetary policy regimes. For this reason, we show that our finding continues to hold when we control for bankfirm fixed effects by comparing loan spreads for the same borrower and bank under different economic and monetary conditions. To the extent that credit supply and demand are driven by endogenous matching between lenders and borrowers, including bank-firm fixed effects should account for it. Finally, the lower loan spread differential between risky and less risky borrowers in different monetary policy regimes may not arises from a change in banks’ risk appetite per se. Existing studies have tried to address this question by making use of bank balance sheet proxies for risk-taking incentives. For example, Jimenez et al. (2014) follow the agency theory and assume that banks with high capital are less prone to take on risk. Well-capitalized banks can indeed be more risk-averse when capital is used as a cushion against adverse contingencies (e.g., Repullo (2000)). However, well-capitalized banks can also be less risk-averse when capital is more costly than other sources of funding, giving rise to a typical moral hazard problem (e.g., Rochet (1992)). Similarly, the theory put forth by Froot and Stein (1993) and Froot and Stein (1998), argues that costs of external finance make banks effectively risk averse, and that a bank’s risk aversion decreases in its capital level. We take a different approach by using direct bank-level information about risk appetite from the Senior Loan Officers Opinion Survey (SLOOS).5 Specifically, we use bank-quarter 5 Peydro and Maddaloni (2011) also consider information from the SLOOS. However, in contrast to us, they rely on the aggregated information from the publicly available SLOOS to examine whether banks adopt softer

3

survey information on whether banks ease their lending standards, whether they have eased their lending standards because of greater risk tolerance, and whether banks are willing to grant the maximum size of the requested loans. Using such cross-sectional bank information is important because it allows us to better identify the loan pricing behavior of risk-taking and risk-averse banks as defined by the survey across different monetary policy regimes. We find that banks that report ‘easier’ lending standards charge lower loan spreads compared to banks that apply stricter lending standards but only so in times of easing monetary policy. Furthermore, we find that those banks that report easier lending standards because they are willing to take on more risk charge riskier borrowers lower loan spreads compared to banks with lower risk appetite, but only in times of easing monetary policy. We also find that banks that are willing to lend the maximum requested amount provide larger loans to riskier borrowers in times of easing versus times of tightening monetary policy. These results are robust to bank, firm, loan controls, bank and bank-firm fixed effects. They also hold when we use banks’ responses to SLOOS directly in our loan spread model and in a two-step procedure we design to isolate the portion of the bank responses that is not driven by macro and bank controls. Our paper makes three important contributions to the recent empirical literature on bank risk-taking and monetary policy. First, it expands this literature to include an investigation of this monetary policy channel in the U.S. manifested through pricing and volumes of credit. Previous literature has examined the impact of monetary policy on the volume of credit (Kashyap and Stein (2000), Bernanke and Blinder (1992)) and the composition of credit (Gertler and Gilchrist (1994)). More recently, the focus has shifted to the potential impact of monetary policy on banks’ risk-taking incentives. It has been documented that a link between protracted periods of low interest rates and bank risk-taking by investigating European and U.S. banks’ expected default frequencies (Altunbas et al. (2010)), banks’ lending surveys (Peydro and Maddaloni (2011)), Spanish banks’ lending decisions to borrowers with bad credit histories (Jimenez et al. (2014)), and Bolivian bank-lending policies (Ioannidou et al. (2009)). Second, it provides evidence that the interest rate discount banks offer riskier borrowers in times of easing monetary policy is driven by banks’ risk appetite. All of the existing studies lending standards when interest rates remain low for a prolonged period.

4

use only indirect bank balance sheet proxies such as capital and liquidity (e.g., Jimenez et al. (2014)) while we rely on measures of banks’ risk appetite based on SLOOS data. Last, we attempt to detect the effect on banks’ risk-taking by looking at banks’ loan pricing policies. Ioannidou et al. (2009) also focuses on banks’ loan pricing policies, but our investigation differs in many important respects. We control for loan-, bank- and borrower-specific characteristics. They too control for loan- and bank-specific characteristics, but for confidentiality reasons are not able to control for time-varying borrower-specific characteristics. Since borrowers may be both balance-sheet constrained and bank-dependent (Gertler and Gilchrist (1994)), any analysis based either on firm-level or on bank-level data suffers from an omitted-variable problem. In addition, our measure of borrower risk is the market-based and thus forwardlooking probability of default, which takes into account the expected performance of the firm. Ioannidou et al. (2009) instead use estimated time to default based on a hazard model or alternatively the bank’s internal rating. The remainder of our paper is organized as follows. Section 2 discusses the data, empirical strategy and sample characteristics. Section 3 examines the impact of the monetary policy regime on loan spreads. Section 4 reports the results of our investigation of the effect of loan demand on loan pricing. Section 5 reports the results of several robustness tests. Section 6 presents the results of tests that build on the SLOOS data. Section 7 concludes the paper with some final remarks.

2 2.1

Data, methodology, and sample characterization Data

The data for this project come from several sources, including the Loan Pricing Corporation’s Dealscan database (LPC), Compustat, the stock price data of the Center for Research on Securities Prices (CRSP), Merrill Lynch’s bond yield indices, the Reports of Condition and Income compiled by the Federal Reserve System, the Federal Deposit Insurance Corporation, and the Comptroller of the Currency, the Federal Reserve System’s Senior Loan Officer Opinion Survey (SLOOS), and finally SDC’s Domestic New Bond Issuances database. We use LPC’s Dealscan database of business loans from 1990 to 2008. Our sample ends 5

in December 2008, to avoid the effects of the Emergency Economic Stabilization Act enacted in October of that year, which authorized the Treasury to spend up to $700 billion to purchase distressed assets and to inject capital into banks, and the effects of the extraordinary measures the Fed undertook since the onset of the crisis, in particular the quantity easing programs, as they both likely had profound effects on banks lending policies. We obtain information on individual loans, including the date of loan origination, the loan spread over LIBOR, maturity, seniority status, purpose and type of contract; information about the borrowing firm, including its sector of activity and legal status (private or public firm); and information on the lending syndicate, including the identity of the lead arranger in the loan syndicate. We use Compustat to get information about firms’ balance sheets. Although LPC contains loans from both privately held firms and publicly listed firms, given that Compustat contains information about publicly held firms, we exclude loans to private firms from our sample. We rely on the CRSP database to link companies and subsidiaries that are part of the same firm and to link companies over time that went through mergers, acquisitions or name changes.6 We then use these links to merge the LPC and Compustat databases to find out the financial condition of the firm at the time it borrowed from banks. We also use CRSP to gather data on firms’ stock prices. We use Merrill Lynch’s yield indexes on new long-term corporate bonds to control for changes in the risk premium in the credit markets. We consider the indexes on yields of triple-A and triple-B rated bonds because these go further back in time than the indexes on the investment-grade and below-grade bonds. We rely on the Reports of Condition and Income to obtain bank data, including capitalto-asset ratio, size, profitability and risk, for the lead bank(s) in each loan syndicate. Wherever possible, we get these data at the bank holding company level using Y9C reports. When these reports are not available, we rely on Call Reports, which have data at the bank level. Finally, we use the Senior Loan Officer Opinion Survey to get information on bank non-price lending terms. Since the late 1960s, the Fed has collected quarterly information on loan officers judgments about changes of non-price lending practices. The survey collects 6

We adopted a conservative criterion and dropped companies that could not be reasonably linked.

6

information through multiple- or dichotomous-choice questions. Over the years the survey sample as well as its format was adjusted several times. Also, the Fed has added questions to capture the various aspects of banks’ lending policies that became relevant, including questions about mortgage lending and consumer loans. However, the part of the survey that is of interest to us as well as the set of banks surveyed remains unchanged during our sample period.7 We use four questions from the survey. The first question asks whether credit standards for approving applications for commercial and industrial (C&I) loans were, on net, tighter, easier, or unchanged from three months earlier. The second question is about the importance of “increased tolerance for risk” when a bank eases its lending terms for C&I loans. The third question asks bank loan officers whether the terms for accepting the maximum size of C&I loan applications have eased over the past three months. The last question that we consider is whether the demand for C&I loans weakened or strengthened (apart from normal seasonal variation) over the past three months.

2.2

Methodology

Our methodology has two parts. The first part investigates whether banks’ risk-taking incentives vary with the stance of monetary policy and in particular whether their risk appetite is stronger in periods of easing monetary policy. To that end, we investigate whether the difference between the loan spreads for risky and safe borrowers varies with the stance of monetary policy, controlling for a set of factors known to explain loan spreads. Although the first part of our methodology controls for a large set of bank factors and relies on within-bank results, the question may still arise as to whether differences in spreads indeed capture differences in banks’ risk appetite. In addition, we design tests using information from SLOOS that aim at determining whether banks that soften their lending standards are willing to offer an interest rate discount to risky borrowers in periods of monetary policy easing compared to period of tightening. 7

For further details on the Senior Loan Officer Opinion Survey, see Schreft and Owens (1991).

7

2.2.1

Banks’ risk-taking and the stance of monetary policy

To ascertain whether the loan spread difference between riskier and safer borrowers decreases in periods of easing monetary policy than in periods of tightening monetary policy, we investigate the following model of loan spreads:

L LOAN SP Df,l,b,t = c + αM P OLICYt + βP DEF AU LTf,t−1 + γM P OLICYt × P DEF AU LTf,t−1 + ψXi,f,t + νYj,f,t−1 + ηZk,b,t−1 + ζMu,t−1 + ϵf,l,b,t

(1)

L LOAN SP Df,l,b,t is the natural log of the all-in-drawn spread over LIBOR of loan l to firm f from bank b at issue date t. Each observation in our sample is a loan-bank pair where a loan is matched only to the lead bank. The all-in-drawn spread is a measure of the overall cost of the loan, expressed as a spread over LIBOR, because it takes into account both fixed and variable fees associated with the loan. M P OLICYt is our measure of the stance of monetary policy at the time of loan issuance. We consider three alternative measures of the stance of monetary policy. Our first measure is the level change of federal funds target rate, F F DIF., between consecutive FOMC meetings. We multiply the change in the target rate by minus one so that we can interpret greater values of F F DIF as a decline in short-term interest rates. Although we control for a set of macroeconomic factors, M, since monetary policy contractions and recessions often coincide, one may wonder whether the effect of F F DIF on loan spreads reflects only the stance of the monetary policy. To overcome this issue, we consider two additional measures of monetary policy shocks: one based on Taylor rule, T AY LOR, and the other based on an approach developed by Romer and Romer, ROM ERS. Taylor rule residuals are calculated from a regression of the federal funds target rate on output gap and the inflation rate. A positive residual indicates tight monetary policy relative to the rule, while a negative residual indicates relatively loose policy. It has been pointed out that this method may not explain well the behavior of the funds rate because of its reduced functional form. Romer and Romer (2004) propose a novel procedure to identify monetary policy shocks. 8

They use the narrative approach to extract measures of the changes in the Fed’s target interest rate at each FOMC meeting. Then this measure is regressed on the Fed’s real time forecasts of past, current and future inflation, output growth and unemployment. The sources of the estimated shocks are policy makers’ beliefs about workings of the economy, tastes and goals, politics, and the pursuit of other objectives. These shock series contain a large amount of variation that is appropriate for identifying the interest rate effect of monetary policy on loan spreads for risky and safe borrowers. One advantage of the Romer and Romer monetary policy shocks is that they are considered to be relatively free from anticipatory movements (e.g., Coibion (2012)). In addition to considering these continuous measures of monetary policy shocks, we also investigate the effect of discrete measures since some views on the risk-taking channel of monetary policy build on the idea that banks are more willing to pursue risk during prolonged episodes of monetary policy easing versus quarter-by-quarter changes. The first discrete measure we use is a categorical variable, LOW RAT E, which is equal to one if the fed funds rate is lower than the median policy rate.8 The other two discrete measures are the dummy variables T AY LOR IN D and ROM ERS IN D, which are equal to one if T AY LOR and ROM ERS indicate loose monetary policy, respectively. Compared to the continuous variables, these indicator variables by construction identify longer periods of easying monetary policy. However, since they have little variation they may not allow to identify the effect of monetary policy per se but rather capture confounding factors. For these reasons, we consider all of the six alternative measures of the monetary policy stance in our model of loan pricing and investigate whether the difference in loan rates between risky and less risky borrowers decreases in periods of easing versus periods of tightening monetary policy. We measure borrower risk by its probability of default computed at the quarter prior to the loan date, P DEF AU LTf,t−1 , following Bharath and Shumway (2008)’s “naive” estimate of the firm’s probability of default, which is a “simple” implementation of Merton (1974)’s model of corporate bankruptcy. It is driven by the firm leverage and by the return and volatility 8

We have also experimented with other cutoff points of the federal funds rate distribution such as the 30th and 20th percentiles. Alternatively, we consider time-variant median policy rates according to which we define low and high interest rate regimes.

9

of stock returns.9 This measure of risk carries two features which are critical for our study. First, it is possible to compute it for all firms that are publicly listed. Second, it is likely a forward-looking measure of risk of default because it is driven by market information. This feature is important because it will likely take into account the expected performance of the firm.10 The key variable in our model of loan spreads is the interaction between our measure of the monetary policy stance and the borrower’s risk of default, M P OLICY ×P DEF AU LT. If banks seek to take on more risk during periods of easing compared to tightening monetary policy, then we should expect γ < 0. In other words, the difference in loan spreads between risky and less risky borrowers shrinks in periods of easing versus periods of tightening monetary policy. In testing this hypothesis we include a number of firm-specific controls Y, loan-specific controls, X, bank-specific controls, Z, and macro factors, M, which may affect loan spreads. We discuss these controls next, starting with our set of firm-specific variables. We use the log of the firm’s sales in hundreds of millions of dollars, L SALES, to control for the overall risk of the firm. Given that larger firms are usually better diversified across customers, suppliers, and regions, we expect this diversification to have a negative effect on loan spreads. The next set of variables proxy for the risk of the firm’s debt. P ROF M ARGIN is the firm’s profit margin (net income divided by sales). L IN T COV is the firm’s interest coverage, which we measure as the log of one plus the interest coverage ratio (i.e., earnings before interest, taxes, depreciation, and amortization divided by interest expense). More profitable firms as well as firms with higher interest coverage have a greater cushion for servicing debt and should therefore pay lower loan spreads. Another aspect of credit risk includes losses to debt holders in the event of default. To capture this risk, we include several variables that measure the size and quality of the asset base that debt holders can draw on in default. T AN GIBLES is the firm’s tangible 9

It is for this reason that we do no use these variables as separate controls in our model of loan spreads. See the Online Appendix to our paper for details on how we compute borrowers’ probability of default. 10

This measure, however, is subject to the criticism that the estimated default probabilities are too low. When we consider credit ratings our results are very similar, which suggests that credit risk measures are not likely to drive our conclusions.

10

assets—inventories plus plant, property, and equipment—as a fraction of total assets. Because tangible assets lose less of their value in default than intangible assets such as brand equity do, we expect this variable to have a negative effect on spreads. ADV ERT ISIN G is the firm’s advertising expense divided by sales; because this variable proxies for the firm’s brand equity, which is intangible, we expect it to have a positive effect on spreads. Similarly, R&D is the firm’s research and development expenses divided by sales; because this variable proxies for intellectual capital, which is intangible, we expect it to have a positive effect on spreads.11 N W C is the firm’s net working capital (current assets less current liabilities) divided by total debt; given that NWC measures the liquid asset base, which is less likely to lose value in the case of default, we expect it to have a negative effect on spreads. M KT BOOK is the firm’s market-to-book ratio, which proxies for the value the firm is expected to gain by future growth. Although growth opportunities are vulnerable to financial distress, we already have controls for the tangibility of the book value of a firm’s assets. Thus, this variable could have a negative effect on spreads if it proxies the additional value (over and above book value) that debt holders can partially access in the event of default. We now discuss our loan-specific variables Y. We include dummy variables equal to one if the loan has restrictions on paying dividends (DIV REST RICT ) and is secured (SECU RED). All else equal, any of these features should make the loan safer, decreasing the spread, but it is well known that lenders are more likely to require these features if they consider the firm to be riskier (see for example Berger and Udell (1990)), so the relationship may be reversed. We also include the loan maturity in years, M AT U RIT Y, and the log of loan amount in millions of dollars, L AM OU N T. Loans with longer maturities may face greater credit risk, but they are more likely to be granted to firms that are thought to be more creditworthy; so, the effect on spreads is ambiguous. Larger loans may impose more credit risk, raising the loan rate, but they may also allow economies of scale in processing and monitoring the loan; again, the effect of this variable on loan spreads is ambiguous. Because the purpose of the loan likely affects credit spreads, we include dummy variables for loans taken out for corporate purposes (CORP P U RP OSES), to repay existing debt 11

Firms are required to report advertising expenses only when they exceed a certain value. For this reason, this variable is sometimes missing in Compustat. The same is true for research and development expenses. In either case, when the variable is missing, we set its value equal to zero.

11

(DEBT REP AY ), and for working capital (W ORKCAP IT AL). Similarly, we include dummy variables to account for the type of the loan—whether it is a line of credit (CREDIT LIN E) or a term loan (T ERM LOAN ). Since most of the loans in the sample are syndicated loans, we attempt to control for the potential effect that the composition of the syndicate may have on loan spread. To that end we include the log of the number of lenders in the syndicate, L LEN DERS, and the portion of the loan that the lead bank in the syndicate retains on its balance sheet, RET AIN ED SHARE. Next, we turn our attention to the set of bank-specific variables Z. These variables are important to control for factors that may affect banks’ willingness and ability to supply funds. LASSET S BK, the log of the bank’s total assets controls for bank size. Arguably, larger banks may be better-diversified or have better access to funding markets, leading to a lower cost of funds and (potentially) lower loan spreads. Similarly, a bank’s return on assets (ROA BK) may proxy for a bank’s improved financial position, again leading to a lower loan spread. In contrast, indicators of bank risk such as the volatility of return on assets (ROA V OL BK) or net loan charge-offs as a fraction of assets (CHARGEOF F S BK) may mean that the bank faces a higher cost of funds, suggesting a positive impact on spreads.12 We control for the bank’s capital-to-assets ratio, CAP IT AL BK, and expect this variable to be negatively related to loan interest rates. According to Boot et al. (1993), banks with low capital are more willing to consume reputational capital to build up financial capital and thus are more likely to renege on implicit guarantees, including the guarantee not to exploit their informational monopoly. This negative relationship may also arise following Froot and Stein (1993) and Froot and Stein (1998), who argue that the costs of external finance make banks effectively risk averse, and that a bank’s risk aversion decreases in its capital level. On the other hand, since capital does not have the tax benefits of debt funding the relationship may be reversed. Additionally, we control for bank holdings of cash and marketable securities as a fraction of total assets, LIQU IDIT Y BK, and for the bank’s access to public debt markets through the fraction of the bank’s subordinated debt to total assets, SU BDEBT BK. Banks with more 12

We use the volatility of return on assets (ROA V OL BK) rather than the stock return because a large number of the banks in the sample are not listed on the stock market.

12

liquid assets should find it easier to fund loans on the margin, leading to lower loan spreads. Similarly, banks with access to the bond market may be able to raise funds at a lower cost, again leading to lower loan spreads. Subordinated debt may also act as a substitute for equity capital, in which case we should also expect the impact on loan spreads to be negative. Our last set of controls, M, account for macroeconomic conditions, equity markets’ volatility and credit market conditions. To that end, we control for the quarterly GDP growth rate and for VIX (Chicago Board Options Exchange Market Volatility Index). We control for the firm’s cost to access the bond market by including the difference between the current yields on BBB- and AAA-rated bonds, L SP READ. We include the slope of the yield curve (SLOP E Y C), computed as the daily yield difference between the five- and one- year zerocoupon bond, to control for expected changes in short-term rates. In addition, we add year and quarter fixed effects to account for time effects at the yearly and quarterly levels. We estimate all our models with robust standard errors clustered at the firm and bank levels. We focus on models estimated with bank fixed effects to reduce concerns about unobserved heterogeneity at the bank level that may affect loan pricing policies. We also report the results of models estimated with bank-firm fixed effects. In this case, the variation in spreads comes from differences in loan pricing policies for the same bank and the same borrower across different monetary policy regimes. Further, we check the robustness of our results by including firm-time and firm-bank-time fixed effects which account for unobservable changes in loan demand (see Section 4). In this case, the identification comes either from comparing loan spreads for the same firm in the same time period across different banks (timefirm) or within the same bank (time-firm-bank). 2.2.2

Banks’ lending standards, risk-taking, and the stance of monetary policy

To assure us that the findings we derive in the first part of our methodology are indeed the result of a change in banks’ risk appetite across different monetary policy regimes, we design two tests using the information banks provide in the SLOOS. This survey is particularly valuable for the purpose of this study because it contains information on banks’ standards for approving loan applications in each quarter. Importantly, banks’ reports in the SLOOS are confidential and do not trigger interventions by supervisors, which may give banks incentives to reveal truthful 13

information. Further, because banks’ responses are anonymous, we are insulated from issues such as banks’ strategic answers aimed at hiding information from and/or signaling to their competitors. We start by using SLOOS information in a one-step test of the relationship between banks’ lending standards, loan spreads and firm risk. Since banks’ lending standards may be affected by macro- and bank-specific factors, we also design a two-step test which attempts first to isolate the part of variation in banks’ lending standards driven by these characteristics and then use the residual variation in lending standards to investigate its effect on the loan spreads. We describe these two tests in more detail below. One-step procedure In order to investigate how banks’ lending standards affect their loan pricing depending on the stance of monetary policy, we start by considering the answers banks provide to the SLOOS’ question of whether bank credit standards for approving applications for C&I loans were, on net, tighter, easier, or unchanged from three months earlier. Using this information we define the dummy variable LS EASIN Gb,t which takes the value one in the quarters in which the bank indicates that its standards for approving loans were on net easier than in the previous three months, and estimate the following model of loan spreads:13 13

The two-category dummy variable LS EASIN Gb,t implies that bank-quarters with tight lending standards are lumped together with bank-quarters that do not change their lending standards. We have also considered a three-category dummy variable in which the bank-quarters without change in lending standards takes a separate category. The results based on the one-step procedure are robust to this definition. Similarly, our conclusions are persevered if we classify in one group easing bank-quarters and the subsequent quarters with no change, and in another group bank-quarters with tightening and subsequent quarters with no change. However, we rely on the dichotomous dummy variable because in the two-step procedure, we use the residuals from the probit model of lending standards which have more tractable properties than the residuals from the multinomial probit.

14

L LOAN SP Df,l,b,t = c + αM P OLICYt + βP DEF AU LTf,t−1 + µLS EASIN Gb,t + γM P OLICYt × P DEF AU LTf,t−1 + ϕb,t−1 LSEASIN Gb,t × M P OLICYt + λLS EASIN Gb,t × P DEF AU LTf,t−1 + θLS EASIN Gb,t × M P OLICYt × P DEF AU LTf,t−1 + ψXi,f,t + νYj,f,t−1 + ηZk,b,t−1 + ζMu,t−1 + ϵf,l,b,t

(2)

Compared to model (1), model (2) adds the new variable LS EASIN Gb,t and its interaction with the borrower risk, P DEF AU LTf,t−1 , and with the stance of the monetary policy at the time of the loan, M P OLICYt . Model (2) allows us to investigate the impact of lending standards on spreads for borrowers with different default risk across different monetary policy regimes. The key effect of interest—the coefficient on the triple interaction, θ—is essentially a difference-in-differencein-differences estimator. This coefficient tells us whether the difference in loan spreads between risky and safe borrowers for banks that soften their lending compared to banks that tighten lending standards in periods of easing monetary policy is larger or smaller than the loan spread difference between risky and safe borrowers when banks soften their lending standards compared to tightening banks in periods of tightening monetary policy. To the extent that banks soften their lending standards somewhat independently of the monetary policy regime, we can estimate θ. In addition to being asked whether the bank’s credit standards for approving applications for C&I loans were, on net, tighter, easier or unchanged from three months earlier, banks are also asked about the importance of “increased tolerance for risk” when they soften the terms for C&I loans. Using this information, we construct the dummy variable RISK T OL, which takes the value one in the quarters a bank indicates that it has eased its lending standards and that “increased tolerance for risk” was very important or somewhat important for easing the terms for C&I loans. We then repeat the investigation described above with the dummy variable RISK T OLb,t . A key advantage of RISK T OL over LS EASIN G is that it isolates those periods when banks not only indicate that their lending standards are looser but 15

they further specify that increased risk tolerance plays a key role in easing those standards. Therefore, if the stance of monetary policy affects banks’ risk-taking incentives, we should find evidence of this link when we use RISK T OL in our loan pricing model. We use yet a third question in the SLOOS survey to advance our a step further in understanding the risk-taking channel of monetary policy. Banks are asked whether the terms to accepting the maximum size of C&I loan application have eased over the past three months. This question is relevant because we can uncover whether banks that are easing lending standards in terms of loan size do so in regards to riskier borrowers relative to less risky ones across different monetary policy regimes. We construct the dummy variable LSize EASIN G, which takes the value one in the quarters a bank indicates it is willing to accept the maximum loan size application. We repeat the analysis described above, however, having the loan amount as a dependent variable. If the stance of monetary policy affects banks’ risk-taking incentives we should find that banks that are easing in terms of accepting the maximum size of the loan applications originate relatively larger loans to riskier borrowers in times of monetary policy easing versus tightening. Finally, we use the question from the SLOOS survey that asks the banks whether C&I loan demand was strong or weak for reasons other than seasonal changes. Using these answers, we construct the dummy variable LS DEM AN D, which takes the value one in the quarters a bank indicates that C&I loan demand was weak for reasons other than seasonal changes. As with the previous questions, we expand our model (1) of loan spreads to include the dummy variable LS DEM AN Db,t , and its interactions with the borrower risk, P DEF AU LTf,t−1 , and with the stance of the monetary policy, M P OLICYt . If our firm and macroeconomic controls explain most of the differences in loan demand, then LS DEM AN D should not affect the loan spreads of risky and safe borrowers in periods of monetary policy easing compared to monetary policy tightening. Two-step procedure A potential concern with the results of our one-step procedure is that banks’ lending standards reflect bank and macro-specific characteristics rather than bank’s appetite for risk. We employ the following two-step procedure to address this concern. In the first step, we use bank 16

and macroeconomic information to isolate the variation in lending standards that is likely attributable to these observable factors. The residual from this first stage regression is then used in the second step to ascertain whether banks’ willingness to take on more risk leads to commensurate changes in their loan pricing policies. We estimate the following probit model of the bank’s lending standards in the first step:

LS EASIN Gb,t = c + ηZk,b,t−1 + ζMu,t−1 + ϵb,t .

(3)

In the second step, we use the residual EASIN GRES retrieved of the first step in the following model of loan spreads:

L LOAN SP Df,l,b,t = c + αM P OLICYt + βP DEF AU LTf,t−1 + µLS EASIN GRES b,t + γM P OLICYt × P DEF AU LTf,t−1 + ϕLS EASIN GRES b,t × M P OLICYt + λLS EASIN GRES b,t × P DEF AU LTf,t−1 + θLS EASIN GRES b,t × M P OLICYt × P DEF AU LTf,t−1 + ψXi,f,t + νYj,f,t−1 + ηZk,b,t−1 + ζMu,t−1 + ϵf,l,b,t

(4)

Model (4) is similar to model (2) except for the fact that rather than using the original variable LS EASIN G to control for banks’ lending standards, we now use the LS EASIN GRES , extracted from the first step (2). We extract the generalized residual EASIN GRES from the first stage, following Gourieroux et al. (1987). These residuals are uncorrelated with the explanatory variables in equation (3) by construction. The inclusion of the generalized residual accounts for the correlation between the error terms in equations (3) and (4), suggesting that if these residuals play a role in the loan spread regressions, it is not through changes in bank and macro factors. Rather, any impact of LS EASIN GRES on loan spreads is due to unobservables associated with a bank’s decision to ease its lending standards. Hence, the residual captures the bank’s choice to ease the standards for approving loan applications for reasons other than its financial and macroeconomic conditions. We view this measure as a 17

proxy for a bank’s decision to rely on lax standards for approving loan applications.14 As in our one-step procedure, the key effect of interest is identified by the coefficient on the triple interaction, θ. This coefficient tells us whether the difference in loan spreads between risky and safe borrowers for banks that soften their lending compared to banks that tighten lending standards in periods of easing monetary policy is larger or smaller than the loan spread difference between risky and safe borrowers when banks soften their lending standards compared to tightening banks in periods of tightening monetary policy. We develop similar two-step procedure tests for RISK T OL and LSize EASIN G. Further, as in the first part of our methodology, in these tests we focus on models estimated with bank-fixed effects or with bank-firm fixed effects.

2.3

Sample characterization

Table 1 presents the characteristics of our sample which consists of 17,974 loans taken out by 3,878 publicly listed nonfinancial corporations between 1990 and 2008 from 235 banks. All variables are winsorized at the 1% level to mitigate the effects of outliers. As is common in corporate samples, many variables in our sample are positively skewed, with mean values greater than median values. That is the case of market-to-book value, the fraction of tangible assets, R&D and advertising expenditures, interest rate coverage, net working capital, and probability of default. Only two of our firm controls, profit margin and sales, are negatively skewed. The median firm has a profit margin of 0.032 and log sales of 1.591, while the mean values are -0.046 and 1.576 respectively. Turning our attention to the loan controls, we find that the mean of the log of loan amount is 4.067 and the median is slightly larger 4.094. Similarly, the loan spread (over LIBOR) is positively skewed with a median of 225 basis points over LIBOR and a mean of 238 basis points over LIBOR. The mean and median maturity is four years. Roughly a third of the loans (33 percent) are for corporate purposes. With regards to the type of contract, 29 14 This part of our analysis has some similarities with Bassett et al. (2014) who use SLOOS to identify bank loan supply shocks. They extract the residuals from a model that estimates the lending standards on macro and bank factors. They then aggregate the residuals to a quarterly index to examine how exogenous bank supply shocks affect real gross domestic product (GDP) and core lending capacity in a VAR framework. We extract the generalized residuals instead at the bank-quarter level and examine their impact on loan pricing in times of easing and tightening monetary policy.

18

percent of loans are term loans, and 55 percent are credit lines. On average, the lead arranger holds 41 percent of the loan, but the median share is only 26 percent. This difference derives from a significant presence of non-syndicated loans in which banks hold the entire loan.15 Next, we turn our attention to the set of bank controls. We measure these controls at the holding company level, and not at the bank level, to capture any potential effects that may arise from ownership transfers between entities of the same holding company. For the ease of exposition, we continue to refer to these as bank controls. Banks are significantly larger than their borrowers: median log of bank assets is 17.048 and mean log bank assets ia 16.547. The average bank has an equity-to-assets ratio of about 8 percent. In contrast, subordinated debt accounts for only about 1 percent of the funding used by the average bank. Both the return on assets, the volatility of the return on assets, and net charge-offs have a mean and a median of about 0.1 percent. Looking at the lending standards variables from the SLOOS, we note that on average about 38% percent of the banks indicate that their standards for approving loans are on net easier than in the previous quarter (LS EASING). Furthermore, based on RISK T OL on average about 27% percent of banks specifically indicate that “increased tolerance for risk” was very important or somewhat important for easing the terms for C&I loans. Based on LSize EASING, 10% of banks indicate easing of the terms for approving the maximum size of the requested loans. In terms of the role of demand for C&I loans (LS DEM AN D,) 28% of the banks indicate that weak demand is an important factor. Finally, looking at our macroeconomic controls we see that the mean value of the L SP READ is 1.882 log points and the mean slope of the yield curve is 0.836. On average the quarterly GDP growth rate is 1.222 percent and the daily VIX index is 19.690. F F DIF that is the difference in the target fed funds rate for two consecutive announcement dates is 0.07% on average. We multiply the fed funds differential by minus one in order to interpret the positive 15 The lead arranger share is missing for a large percentage of loans in Dealscan. To alleviate this problem, in the regression analysis we complemented the lead bank’s share in Dealscan with information from the Shared National Credit program. Since this procedure still leaves some missing data, we apply the so called “dummy variable adjustment” approach in which we plug in an arbitrary value for the cases of missing retained share and then include a dummy variable coded one if data in the original variable was missing and zero otherwise, and its interaction the modified variable. We also apply alternative methods of filling in the missing observations such as using the subgroup mean share by bank-quarter and we obtain similar estimates. The retained share and the method of refilling the data do not seem to affect our main estimates of interest.

19

values as decreasing FF rates for consecutive quarters; 42% of the loans are originated in LOW RATE regime. ROM ERS is the monetary policy shock from Romer and Romer (2004).16 Note that we again multiply the original values by minus one so that we can interpret higher values as easy monetary policy. The mean ROMERS shock is -0.029 and the median is -0.028. ROM ER IN D takes one if the monetary policy is easing according to the ROM ERS residual (i.e. the residual multiplied by minus one is greater than zero). We also document that the average duration of the same monetary policy regime according to the ROM ERS shock (i.e, the shock remains positive/negative) is 6.6 months. The mean Taylor rule residual, T AY LOR takes the value of 0.024. The value is multiplied by minus one to ensure that we interpret higher values as easing monetary policy. We also construct a dummy variable T AY LOR IN D that takes one if the monetary policy is easing and zero other wise: 42% of the loans are originated in easing periods as defined by the Taylor rule residual.

3

Do risky firms receive a loan discount when monetary policy is easing?

In this section, we study the risk-taking channel of monetary policy by investigating how banks’ loan pricing policies vary with the stance of monetary policy. Since there is little consensus on the best way to capture the stance of monetary policy, we consider three continuous and three dummy-based measures of monetary policy. The continuous variables allows us to better identify the effect of monetary policy on bank lending behavior. The indicator variables, in contrast, give us a better way to identify the prolonged periods of monetary policy regimes, which is important for our investigation. Table 2 reports the results of our investigation when we measure the monetary policy stance with a continuous variable. Recall that under our hypothesis if monetary policy easing affects banks’ risk-taking incentives, then in times of easing the spread difference between risky and less risk borrowers will be lower compared to the difference in times of tightening. To test this hypothesis, in addition to controlling for the monetary policy stance at the time of the 16

We thank Chris Crowe for making this variable available to us. For details see Barakchian and Crowe (2010).

20

loan and for the borrower credit risk (as measured by its probability of default), we include the interaction of these two variables as specified in our loan pricing model in Section 2.2.1. Models (1) to (3) include bank-fixed effects, and models (4) to (6) include bank-firm fixed effects. All models include firm-, bank-, and macroeconomic controls with robust standard errors clustered by bank and by firm. We start in column (1) of Table 2 with a measure of monetary policy that captures the difference between the short-term policy rate for two consecutive FOMC meetings. The changes in short-term policy rates are often used to gauge the stance of monetary policy (e.g., Jimenez et al. (2014)). Higher values of F F DIF indicate that policy rates are lower in the current period compared to the previous period (F F DIF is multiplied by (-1)). Looking at column (1), we see that the coefficient of interest on the interaction term P DEF AU LT × F F DIF is -0.288. It indicates that for one standard deviation decrease in F F DIF , the sensitivity of loan spreads to firm risk decreases by approximately 9.7% (0.288×0.336). Since monetary policy easing and recessions often coincide, one may wonder whether monetary policy alone drives the effect of the short-term policy rate on loan spreads we identify in column (1). We include a set of macroeconomic controls—the triple-B spread in the bond market, the slope of the Treasury yield curve, the growth rate of GDP and VIX—as well as time dummies. To further address this concern, in columns (2) and (3) we use instead the Taylor rule (T AY LOR) and Romer and Romer (ROM ERS) residuals to capture the stance of monetary policy.17 Looking at columns (2) and (3), we see that the estimates of interest P DEF AU LT ×T AY LOR and P DEF AU LT ×ROM ERS are both negative and statistically significant, confirming that loan spreads for risky borrowers decrease when monetary policy is easing. One standard deviation increase in ROM ERS leads to 11% (0.228×0.473) decrease in the sensitivity between loan spreads and firm risk. In columns (4) to (6) we estimate the same specifications as in columns (1) to (3) using bank-firm fixed effects, which allows us to control for unobserved heterogeneity at the bank-firm level. We are able to use bank-firm fixed effects because 70 percent of all bankfirm pairs occur more than once in the sample period. Looking at the results, all estimates of the interaction terms of the monetary policy shocks with the borrower’s probability of 17

See Section 2.2 for details on the choice of these measures.

21

default continue to be negative and statistically significant. Further, they all have larger magnitudes. In column (6), for example, one standard deviation increase in ROM ERS is associated with a 19% (0.835×0.473) reduction in loan spreads as borrowers become riskier. For all specifications, monetary policy easing is negatively related to loan spreads for borrowers with average probability of default as suggested by the negative sign of the total effect of P DEF AU LT × M P OLICY and M P OLICY. A quick look at coefficients of the remaining controls indicates that they have reasonable effects on loan spreads and are generally consistent with other studies of loan spreads.18 In the interest of space we leave out a detailed discussion of these estimates and turn our attention to the estimates of borrower risk and monetary policy. Column (1) of Table 3 reports the results when we use the dummy variable LOW RAT E, which takes the value one if the federal funds rate at the time of the loan is below its sample median for the 1990-2008 period. Once again, the interaction term LOW RAT E × P DEF AU LT is negative and significant, implying that the compensation for risk that banks demand is lower when interest rates are low versus when they are high. In model (1) the estimate on P DEF AU LT is 0.974 and the estimate on P DEF AU LT × LOW RAT E is -0.238, indicating that the sensitivity of loan spread to firm risk is 24% smaller in the low interest rate regime relative to the high interest rate regime. In column (2), we rely on the dummy variable of monetary policy shocks T AY LOR IN D, which takes the value one if the residuals are positive (easing) and zero otherwise. The coefficient on the interaction term P DEF AU LT × T AY LOR IN D is -0.11, indicating that the loan spread for risky relative to safe borrowers in periods of easing is around 11% smaller than in the periods of tightening. In column (3), we rely on an indicator variable of monetary policy shocks ROM ERS IN D, which takes the value one if the residuals are positive (easing) and zero otherwise. The coefficient on the interaction term P DEF AU LT × ROM ERS IN D is -0.20, implying that the loan spread for risky relative to safe borrowers in periods of easing is around 20% smaller than in periods of tightening. In columns (4) to (6) of Table 3 we include bank-firm fixed 18 For other studies of loan spreads see, for example, Santos and Winton (2008), Hale and Santos (2009), and Santos (2011).

22

effects. Overall, all estimates of interest preserve the expected signs and magnitudes with the exception of P DEF AU LT × T AY LOR IN D, which loses its significance. The results we have presented thus far show that when relying on either continuous or indicator variables to capture the stance of monetary policy during periods of easing monetary policy, banks charge riskier borrowers lower loan spreads compared to safer ones. As we noted above, relying on shocks that remove anticipatory developments about the economy such as the Romer and Romer shocks ensures that our estimates capture the effect of the stance of monetary policy regime on loan spreads per se. The dummy variable definitions are vulnerable to confounding effects of other macroeconomic conditions. For this reason, in the rest of the paper we use the continuous version of the Romer and Romer shocks to control for the stance of the monetary policy. Looking ahead, we emphasize that according to our findings banks charge riskier borrowers relatively lower spreads when monetary policy regime is easing compared to tightening regime. Next, we investigate whether our results are driven by changes in the demand for loans. In the interest of space, in what follows, we do not report the results for the various firm-, loan-, and bank-specific controls or the results for the macroeconomic controls.

4

Can credit demand explain loan discount to risky borrowers?

The effect of monetary policy shock on loan spreads for different levels of borrowers’ risk we identified in the previous section may also be the result of credit demand. For example, periods of monetary policy easing may be accompanied with poor economic conditions, weaker investment opportunities and hence weaker demand for credit. Moreover, because poor economic conditions often have a bigger effect on the demand for credit by riskier borrowers, the relatively lower spreads for riskier borrowers could therefore be the result of lower demand for credit (e.g., Erel et al. (2012)). We provide three qualitatively different tests below that show that credit demand is likely not the driver of our findings on loan spreads. First, to investigate the hypothesis that risky borrowers have possibly lower demand for credit in times of monetary policy easing is to examine loan quantities. If the credit supply curve shifts outward, holding everything else equal, we should observe lower prices of credit

23

to be accompanied with higher credit amount at any given price. Alternatively, if lower prices are observed together with lower loan amounts, then supply effects are difficult to reconcile. To investigate this hypothesis, we estimate the same specification as in equation (1) except for having loan amounts as a dependent variable. The results of this investigation are reported in columns (1) and (2) of Table 4. Column (1) reports the results of the model estimated with bank-fixed effects, and column (2) reports the results estimated with bank-firm fixed effects. Both specifications show that riskier borrowers obtain larger loans than less risky borrowers when the monetary policy regime is easing. In terms of economic impact of the estimate in column (1), one standard deviation increase in ROM ERS is associated with 27% increase in loan amount on average. This increase in loan amounts goes together with the 11% decrease in loan spreads reported in the previous section (column (4), Table 2). The increase in loan amounts and the decline in loan spreads for risky relative to less risky borrowers does not support the credit demand hypothesis. Second, we investigate whether our results hold across borrowers with different size. This test is important because large and small firms may be subject to different unobserved firm-specific demand shocks that happen to be correlated with monetary policy regimes. Therefore, if the estimate of ROM ERS×P DEF AU LT is correlated with uncontrolled demand shocks that are different for large and small firms, we run the risk of observing that the estimate is negative and significant for either large or small firms. This outcome suggests that demand rather than supply factors are at play. To test this hypothesis, we estimate loan spread regressions separately for large and small firms. The results of this investigation are reported in columns (3) to (6) of Table 4. We observe in columns (3) and (4) (bank fixed effects) and in columns (5) and (6) (bank-firm fixed effects) that the estimates on ROM ERS×P DEF AU LT are negative and significant for both large and small firms, which is contrary to the credit demand hypothesis. Looking at the magnitude of the estimates, they suggest that large firms relative to small firms experience on average a greater loan spread undercut when monetary policy is easing. One possible explanation is that large firms may have alternative sources of financing and banks decrease their loan spreads more aggressively to provide them with incentives to borrow. Alternatively, according to the broad credit channel, loose monetary policy causes borrowers’ 24

balance sheets to improve their collateral values.19 To the extent that small firms improve their net worth relatively more when interest rates are low, we would expect their loan spreads to decrease by more than those for large firms. This explanation, however, is not supported by the data because it is the large risky firms that experience a greater cut than the small risky ones.20 Our final test accounts for time-variant unobservable changes in loan demand. To that end, we add to our model time-firm fixed effects. This allows us to control for loan demand shocks within a specific time period. The identification comes from comparing loan spreads for the same firm in the same time period across different banks. If demand shocks are correlated with the monetary policy regime and/or a firm’s riskiness, time-firm fixed effects would account for the effect of credit demand. We report the results of this test in column (7) of Table 4. In our sample each firm takes five loans on average and the time period between two loans is two years. To ensure that the same firm takes more than one loan over a specific time period, we define time as a two-year period. The results of this test also show that the loan spread between risky and safe borrowers is lower during easy monetary policy regimes. In column (8), we go a step further and include firm-bank-time fixed effect. The identification comes from comparing spreads within the same firm, bank and time period. In such a way we control for unobserved time-varying heterogeneity at the bank-firm-time level. These results further reinforce the conclusion that our findings go through even when we account for demand effects by using different types of fixed effects.

5

Robustness tests

In this section, we report several robustness tests of our key finding. These tests aim to address issues related to the regression specification, the definition of the monetary policy stance, and concerns with improvements in collateral values or in borrowers’ future financial condition. 19

In the balance sheet channel of monetary policy the external finance premium, defined as the costs of funds raised externally and funds raised internally, decreases when interest rates are low precisely because of higher collateral values (see Bernanke and Gertler (1995)). 20

See Section 5.3 for further details on the role of collateral.

25

5.1

Full interaction

Our first robustness test attempts to address concerns with the main regression specification we use to investigate the effect of the stance of monetary policy. In this specification, we interact monetary policy only with the firm’s probability of default. However, it may be that other firm characteristics affect loan spreads together with the monetary policy regime. To address this concern, we re-estimate our loan spread model with interactions of monetary policy with all of the variables in the regression. This specification also allows to verify whether a firm’s probability of default affects spreads through monetary policy regimes even after conditioning on all interactive terms. The results of this investigation are presented in Table 5. Column (1) reports the results estimated with bank fixed effects, while column (2) reports the results estimated with bankfirm fixed effects. The estimate on ROM ERS×P DEF AU LT in column (1) is -0.344 implying that even after controlling for the interactive effects of monetary policy with all other variables, the probability of default has significantly lower impact on loan spreads as monetary policy is easing. The same conclusion holds when we consider bank-firm fixed effects. These findings show that our main result is not vulnerable to omitted variables such as the interaction terms of monetary policy regimes and all other variables. They also highlight that our measure of firm probability of default retains significant predictive power after we control for other proxies of firm risk. We continue the analysis using the more parsimonious specification that accounts only for the interaction of monetary policy and the probability of default.

5.2

Low-for-long monetary policy

One important point in the debate on the risk-taking channel of monetary policy is the understanding that banks have incentives to take on more risk when interest rates remain low for prolonged periods. Although our continuous measures of monetary policy ensure that the regimes of easing/tighenig last for at least several months, in this section we explicitly examine the effect of the duration of the monetary policy regime.21 We start by using ROM ERS IN D, which takes the value one if monetary policy is easing (ROM ERS is positive) and zero other21

One advantageous feature of the ROMERS residuals is that they tend to be relatively more persistent from quarter to quarter—there are no frequent changes from easing to tightening and vice versa.

26

wise. Next, to investigate the importance of the duration of the regime, we count the number of consecutive months for which the regime does not change (M ON T HS ROM ERS IN D). The longest easing period in our sample lasts seventeen months and the longest tightening period lasts twenty seven months.22 We then add the triple interaction between P DEF AU LT , ROM ERS IN D and M ON T HS ROM ERS IN D to the basic specification. The results of this investigation are reported in Table 6. Column (1) reports the results estimated with bank fixed effects and column (2) reports the results estimated with bank-firm fixed effects. In both cases we see that the triple interaction is negative and significant, indicating that the prolonged easing regimes are associated with higher the interest rate discounts for risky borrowers. This result further confirms that the discount of loan spreads for risky borrowers is more pronounced the longer the monetary policy regime remains low.

5.3 5.3.1

Other robustness tests Post-loan firm probability of default

Finding that loan spreads are relatively lower for riskier borrowers in times of easing monetary policy compared to times of tightening monetary policy is consistent with a bank risktaking channel of monetary policy, but it could also be the result of the following situation. Loan spreads of risky borrowers are relatively lower when monetary policy is easing because banks expect borrowers to improve their future risk profiles. It is unlikely that this explains our finding because our measure of borrower risk—the borrower’s probability of default—is a forward-looking measure that captures the expected probability of default. Nevertheless, to further reduce concerns with this alternative explanation, we examine whether the probability of default declines during the three years after the loan origination compared to three years before the loan origination. The results of this investigation, which are reported in the accompanying Online Appendix to our paper, suggest that banks do not discount the spreads of risky borrowers in times of easing relative to times of tightening anticipating risky borrowers to have lower probability of default in the near future after the origination of the loan. A caveat to this test is that the 22 As a robustness check, in unreported regressions we exclude periods for which the monetary policy regime is shorter than three months and the results remain intact.

27

future change in borrowers’ probability of default is driven in part by the interest rates banks charge, thus making it hard to distinguish if firms become more riskier because of the lower cost of loans or because of banks’ lending policies. 5.3.2

Collateral Effects

Bank loans differ along several dimensions other than price. In this section, we focus on the possibility that collateral of loans changes over different monetary policy regimes, which can affect loan spreads. It is possible that when monetary policy is easing the value of pledged collateral increases, leading to a reduction in the cost of funding (i.e., the credit channel of monetary policy described in Bernanke and Gertler (1995)). To the extant that the value of pledged collateral increases relatively more for riskier borrowers and hence the loan discount for riskier borrowers increases, our results could be explained with the effect of the credit channel. We do not observe collateral values in the data, however, we have information of whether a loan is collateralized, and we have information on asset tangibility which is often used as a proxy for collateral (e.g., Almeida and Campello (2007)). If the collateral story were at play, we would expect the interest rate discount to be stronger for collateralized loans relative to uncollateralized loans. Based on the results of our tests in the Online Appendix, we do not find supporting evidence for this hypothesis. In addition, the results we report in Table 4 where we split the sample into small and large firms also cast doubt on the collateral story. Assuming that small firms are less diversified and hence more risky, we would expect a greater decrease in loan spreads for these firms than for large firms if improved collateral values drive the results; the reason is that small firms would experience relatively greater improvement in collateral values than large firms in times of monetary policy easing. Based on the results in Table 4, this conjecture does not seem to be plausible. 5.3.3

Omitted Variables

Next, we are aware that our results may be subject to latent credit risk and investment opportunities that are correlated with monetary policy regimes. While accounting for these factors is virtually impossible, we reason that if these factors were at play their effect would be stronger 28

for unrated firms for which less public information is available than for firms that have a credit rating. For that reason, we estimate our loan spread regressions for rated and unrated borrowers and we find out that the sensitivity of loan spreads to firm risk is similar for rated and unrated borrowers, which ameliorates the concern about the role of omitted factors.23 5.3.4

Bond spreads and risk-taking

Finally, one may still wonder if the loan spread discount for risky borrowers is bank driven or alternatively it is the result of an omitted unobservable factor that can affect the relationship between monetary policy, firm risk and loan spreads. One way to address this concern is to see if riskier borrowers also receive a discount when they raise funding in the bond market in periods of easing monetary policy. Some of the reasons put forth in the literature for banks to search-for-yield in periods of easing monetary policy arguably can also apply to bond investors. However, unlike banks bond investors are not protected by the safety net and hence are less likely to become as risk-taking as banks when monetary policy is loose. To that end, we investigate the effect of monetary policy on bond spreads (over Treasury with the same maturity of the bond). The results reported in the Online Appendix show that bond issuers do not charge risky borrowers lower premium in times of monetary policy easing versus tightening. It appears that in contrast to banks, the risk appetite of bond investors—usually insurance companies, pension funds and other investment managers—is not affected by the stance of the monetary policy in the same way. The evidence we unveiled showing that risk-taking is not present in the bond market in times of easing monetary policy confirms that risk-taking behavior is pertinent to banks, but it does not allow us to pin down its source. We address this issue in the next section by providing evidence that the interest rate discount to riskier borrowers in periods of monetary policy easing is driven by banks’ risk-taking incentives. 23

The results are reported in the Online Appendix.

29

6

Using SLOOS to identify bank risk-taking

Existing studies of the risk-taking channel of monetary policy have proxied for high bank risk-taking incentives with lower capital ratios. This approach, however, depends critically on the validity of the proxy for risk. Indeed according to Repullo (2000) lower capital induce more risk taking; however, there are other theories, including Rochet (1992) and Dell′ Ariccia et al. (2013) that derive the opposite relationship between bank capitalization and risk-taking incentives. We take a different approach and use information from the Senior Loan Officer Opinion Survey. This survey is valuable for an investigation of the bank risk-taking channel of monetary policy because it allows us to extract bank-specific measures of lending standards and risk appetite that by construction are not driven by changes in banks’ balance sheets and macroeconomic conditions.

6.1

Bank lending standards and monetary policy easing: One-stage procedure

We begin this part of our investigation by considering the information banks provide in response to the SLOOS question of whether they have eased their standards for approving loans.24 This measure of lending standards allows us to pin down directly the propensity to ease at the bank level, which gives us the opportunity to investigate whether the suggestive evidence of the bank risk-taking channel we unveiled in our previous tests is indeed driven by banks with soft lending standards. To that end, we expand our model of loan spreads and include the dummy variable LS EASIN G, which takes one if banks are easing their lending standards and zero otherwise. The results of this investigation are reported in Table 7. Looking at column (1), which reports results estimated with bank fixed effects, we see that the coefficient of the triple interaction term ROM ERS × LS EASIN G × P DEF AU LT is equal to -1.704. This result suggests that the average loan spread for risky borrowers is lower when monetary policy is 24 The exact survey question is: “Over the past three months, how have your bank’s credit standards for approving applications for C&I loans or credit lines—other than those to be used to finance mergers and acquisitions—to large and middle-market firms changed (annual sales of $50 million or more)?”

30

easing, but only for banks with softer lending standards. To better understand the drivers of the triple interaction term, in Table 8 we report the marginal effects on loan spreads.25 The upper panel of Table 8 shows that when monetary policy is easing, the loan spread between riskier and safer firms is lower for softening banks than for tightening banks. Based on the third row of column (3), the difference in loan spread for risky and less risky borrowers between easing (SOFT BK) and tightening (TIGHT BK) banks is 0.004. In column (6), this difference is 0.070 when monetary policy is tightening. In column (7), the difference-in-difference-indifference estimate takes the value of -0.066 and corresponds to the triple interaction term LS EASIN G×ROM ERS×P DEF AU LT in column (1) of Table 7. Looking at column (2) of Table 7, which reports results estimated with bank-firm fixed effects, we see that the coefficient of the triple interaction term continues to be negative suggesting that the difference in loan spreads for risky and safe borrowers is lower when monetary policy is easing, but only for banks with soft lending standards. As we discussed in the methodology section, banks with soft lending standards are asked a followup question about the exact reason as to why they have softened their lending standards. We focus on the risk tolerance as a potential reason to ease their lending standards.26 The answers to this question provides useful information because measure banks’ appetite for risk more precisely. Using this information, we define the dummy variable RISK T OL, which takes the value one if the bank specifies that it has softened its lending standards because of greater risk tolerance and zero otherwise. Next, similar to our previous test, we expand our model of loan spreads to distinguish banks’ loan pricing policies depending on their risk tolerance and the stance of monetary policy. The results of this investigation are reported in columns (3) and (4) of Table 7. As in the previous test, these columns report the results estimated with bank and bank-firm fixed effects, respectively. The negative and significant estimate on the triple interaction terms ROM ERS ×RISK T OL×P DEF AU LT in columns (3) and (4) confirm that the loan spread 25

We set LS EASIN G equal to one or zero, and choose the following (arbitrary) values of P DEF AU LT and ROM ERS: A risky firm has a P DEF AU LT of 0.2 and less risky firm has a P DEF AU LT of 0.05; Easy monetary policy is measured at the 75th percentile of ROM ERS and tight monetary policy is measured at the 25th percentile of ROM ERS. 26 The exact survey question is: “If your bank has eased its credit standards or its terms for C&I loans or credit lines over the past three months, how important have been increased tolerance for risk?”

31

between risky and less risky borrowers is lower in times of monetary policy easing compared to tightening, however, more so for banks that indicate greater risk tolerance to be the reason for softer lending standards. In the second panel of Table 10, we report the marginal effects in a similar way as for LS EASIN G discussed above. In column (7) of Table 10 the sensitivity of loan spreads to firm risk for banks with soft lending standards because of greater risk tolerance is lower than for banks with tight lending standards. Our next test focuses on loan amounts. In Section 4 we reported results showing that loan amounts increase while loan spreads decrease to riskier borrowers in easing versus tightening periods which is consistent with the presence of supply side effects. SLOOS gives us an opportunity to test that assertion because one of its questions is whether the terms for loan approval with regards to the maximum size of the loans have eased.27 In this case, the dependent variable in our regression is the loan amount, and we include the dummy variable LSize EASIN G, which takes the value one if the bank indicates it has eased the terms to grant the maximum size of the loans and zero otherwise. The results of this investigation are reported in columns (5) and (6) of Table 7. The positive estimate on the triple interaction term ROM ERS ×LSize EASIN G×P DEF AU LT when we include bank and bank-firm effects, respectively, in columns (5) and (6) imply that banks that ease standards in terms of the maximum size of the loans originate larger loans to riskier borrowers in times of monetary policy easing versus tightening.28 This result adds support to our previous finding that the change in loan amounts in periods of easing monetary policy is likely supply driven. The results reported thus far confirm that the decline in loan spreads between risky and safe borrowers in periods of easing monetary policy is driven by banks’ lending standards in particular by their willingness to take on more risk. In the next section we refine our tests 27

The exact survey question is: “For applications for C&I loans or credit lines—other than those to be used to finance mergers and acquisitions—from large and middle-market firms and from small firms that your bank currently is willing to approve, how have the terms of those loans changed over the past three months in terms of the maximum size of credit lines?” Although this question is related to credit lines’ size, it is often the case that credit lines are originated together with term loans. We employ the full sample for this part of the exercise for the sake of comparability. When we restrict the sample only to credit lines, the conclusions remain unchanged. 28 As we do with the previous two SLOOS questions, we report in the third panel of Table 8 the marginal effects associated with the latest question. The difference-in-difference-in-differences estimate of 0.098 (column (7) of Table 8 indicates that banks extend smaller loans to riskier borrowers in the easing regime compared to the tightening regime.

32

to remove the effects of macro and bank balance sheet factors that may explain banks’ answers to SLOOS questions. We view this procedure as helpful in using cleaner measures of banks’ risk appetite which is related to their lending standards and not driven by observable controls.

6.2

Bank lending standards and monetary policy easing: Two-step procedure

As we discuss in the methodology section, in the first stage we estimate a probit model in which the dependent variable is a categorical variable that takes the value of one if the bank indicates it is easing its lending standards and zero otherwise (LS EASIN G). The independent variables in the first-stage model are the set of bank and macro factors described in Section 2.2.2.29 We construct the generalized residuals from this first stage regression following Gourieroux et al. (1987), and use them as a measure of the bank’s easing of lending standards in the second stage regression. As we cannot point out to the direct reason for easing due to the nature of the question, we view this measure (LS EASIN GRES ) as a general proxy for lending standards. These residuals capture banks’ appetite for risk, tastes and sentiment beyond the effect of bank balance sheet and macro controls on bank-risk taking. In columns (1) and (2) of Table 9 we report the results of the second-stage regressions which now include the residual LS EASIN GRES , its interaction with P DEF AU LT and

ROM ERS. According to the results in column (1), the estimate on LS EASIN GRES ×ROM ERS×P DEF AU is -0.629 and it is statistically significant. This indicates that when monetary policy is easing (high values of ROM ERS) compared to when it is tightening, banks with softer lending standards charge riskier borrowers relatively less than banks with tight lending standards. Following the approach we adopted in our one-step procedure, we report in the upper panel of Table 10 the marginal effects on loan spreads computed in column (1). It is apparent that when monetary policy is easing the relationship between loan default risk and loan spreads in banks with soft lending standards are lower than in bank with tight standards. Based on column (3) in Table 10 the difference in loan spreads for risky and less risky borrowers between easing and tightening banks is -0.047. However, the opposite holds when monetary policy is 29

We do not report the results of the first stage in the interest of space, but they are available from the authors upon request.

33

tightening (column 6). The difference-in-difference-in-difference estimate -0.006 in column (7) corresponds to the triple interaction term LS EASIN GRES ×ROM ERS×P DEF AU LT in column (1) of Table 9. We reach a similar conclusion when including bank-firm fixed effects reported in column (2) of Table 9. One potential concern with our findings from the two-step approach is that banks ease their lending standards together with monetary policy easing.

The distributions of

LS EASIN GRES across easing and tight monetary policy regimes are comparable, which reassures us that banks’ decision to soften lending standards do not depend entirely on the stance of monetary policy. This finding is important because it allows us to isolate the effect of ROM ERS×P DEF AU LT from LS EASIN GRES ×ROM ERS×P DEF AU LT. While LS EASIN GRES captures risk appetite due to easing of the lending standards in general, easing due to greater risk tolerance goes a step further to clarify the exact reason for such easing. We use the residual RISK T OLRES from a first step regression similar to the one for LS EASIN GRES . RISK T OLRES is based on information about whether “increased tolerance for risk” played an important role in the decision to ease lending standards for C&I loans (RISK T OL). The second stage results with RISK T OLRES as a measure of risk-tolerance are reported in columns (3) and (4) of Table 9. The negative sign on the estimate on RISK T OLRES ×ROM ERS×P DEF AU LT in column (3) indicates that risk tolerant banks lower spreads relatively more for risky borrowers in easy compared to tight monetary policy regime. In the second panel of Table 10 we calculate the marginal effects in a similar way to LS EASIN GRES as discussed above. In column (7) of Table 10 the loan spread for risky and less risky borrowers for banks with soft lending standards because of greater risk tolerance is lower than for banks that tighten lending standards. Finally, we report in columns (5) and (6) of Table 9 the second stage results which use the residual LSize EASIN GRES . This residual comes from a first stage regression in which the dependent variable takes the value one if the bank indicates it is easing its terms of granting the maximum size of requested loans and zero otherwise (LSize EASIN G). The dependent variable in the second stage regressions in this exercise is the loan amount. The results show that those banks easing the terms of granting the maximum size of the requested loan originate 34

larger loans for risky borrowers relative to less risky borrowers in times of monetary policy easing. The opposite result holds when monetary policy is tightening. In the third panel of Table 10 based on different values of LSize EASIN GRES , P DEF AU LT , ROM ERS we see that the difference-in-difference-in-differences estimate is 0.204 (column (7) of Table 10), indicating that banks extend larger loans to riskier borrowers in the easing compared to the tightening regime. In sum, our investigation based on banks’ answers to the SLOOS provide three critical pieces of evidence in support of the bank risk-taking channel. First, banks with softer lending standards charge risky borrowers lower loan spreads when monetary policy is easy compared to when it is tight. Second, banks with softer lending standards due to risk tolerance charge risky borrowers relatively lower spreads during monetary policy easing. Third, banks with softer lending standards with regards to the maximum size of requested loan amounts originate larger loans to risky borrowers relative to less risky borrowers in times of easing. Importantly, these findings hold both when we use either one-step or two-step approaches, which speaks to their robustness. Further, since in the two-step approach we control for bank-specific factors and macroeconomic conditions in the first step our findings derived in this way are more likely driven by banks’ intrinsic risk appetite.

6.3

Loan demand: Evidence from SLOOS

To conclude our investigation in this section we take another look at the role of loan demand. As we noted above, this is important because a portion of the risk-taking effect may be attributed to demand factors as opposed to bank risk-taking per se. In Section 4, we reported several tests that aim to ameliorate concerns that demand for loans drives the results. In this section, we investigate the role of loan demand by using banks’ responses to the SLOOS question whether demand for C&I loans has changed over the past three months apart from normal seasonal variation.30 Our test builds on the following idea. If demand factors are already controlled for by including firm characteristics, then we would expect the interaction between LS DEM AN D, 30

The exact survey question is: “Apart from normal seasonal variation, how has demand for C&I loans changed over the past three months?”

35

ROM ERS and P DEF AU LT to be insignificant. Finding such evidence would reassure us that firm controls capture loan demand well and it would show that SLOOS survey provide a reliable information on bank lending policies since the use of qualitatively different information about loan demand yields consistent results.31 In Table 11, we estimate models similar to those described in the previous section. Columns (1) and (2) show the results of the one-step loan regression in which we interact ROM ERS, P DEF AU LT, and LS DEM AN D, the dummy variable that takes the value one if a bank has weak loan demand and zero otherwise. The triple interaction term tells us whether the firms’ probability of default in times of easing versus tightening affects loan spreads together with changed demand conditions as measured by the survey question. When using bank and bank-firm fixed effects in columns (1) and (2) respectively, the triple interaction term is insignificant. The same conclusion holds in columns (3) and (4), which report the results for the two-step procedure using LS DEM AN DRES . These findings confirm that demand factors are not the key driver of the interest rate discount that riskier borrowers receive when they take loans in periods of easing compared to tight monetary policy.

7

Final remarks

Our finding that banks with softer lending standards offer riskier borrowers interest rate discount in periods of easing monetary policy, and perhaps even more importantly, our finding that banks that are more risk tolerant offer riskier borrowers discount in periods of easing monetary policy constitute solid evidence in support of the existence of a bank risk-taking channel of monetary policy. Although the quantitative impact of this channel on the stability of the financial system is unclear, our evidence suggests an additional aspect for potential consideration in the design of monetary policy. Our findings open up several avenues for future research. For instance, our tests focus on banks’ loan pricing policies to existing borrowers. It would be useful to investigate whether the risk-taking incentives brought about by monetary policy also lead banks to change their 31

For further information on the effects of lending standards and their quality on bank lending activity see Lown and Morgan (2006).

36

lending policies to new borrowers. Another avenue for future research is understanding the factors that drive the differences of banks and bond issuers risk-taking incentives. Similarly, our tests do not distinguish new loans from renegotiations of existing loans. Mian and Santos (2011), however, show that credit market conditions are an important factor of firms’ incentives to refinance their existing credits. Therefore, it may be useful to investigate whether banks’ risk-taking incentives play a role in firms’ decisions to refinance and the terms of refinancing. Lastly, our findings suggest that an investigation of the real effects of banks’ risk-taking policies induced by monetary policy is also a fruitful area for future research.

37

References Adrian, T. and Shin, H. (2009). Money, liquidity and monetary policy. American Economic Review, 99:600–605. Akerlof, G. and Shiller, R. (2009). Animal spirits: How human psychology drives the economy and why it matters for global capitalism. Princeton University Press. Almeida, H. and Campello, M. (2007). Financial constraints, asset tangibility, and corporate investment. Review of Financial Studies, 20:1429–1460. Altunbas, Y., Gambacorta, L., and Marques-Ibanez, D. (2010). Bank risk and monetary policy. Journal of Financial Stability, 6:121–129. Barakchian and Crowe, C. (2010). Monetary policy matters: New evidence based on a new shocl measure. IMF Working Paper, 230. Bassett, W., Chosak, M., Driscoll, J., and Zakrajˇ sek, E. (2014). Changes in bank lending standards and the macroeconomy. Journal of Monetary Economics, 62:23–40. Berger, A. and Udell, G. (1990). Collateral, loan quality, and bank risk. Journal of Monetary Economics, 25 (1):21–42. Bernanke, B. and Blinder, A. (1992). The federal funds rate and the channels of monetary transmission. American Economic Review, 82(4):901–921. Bernanke, B. and Gertler, M. (1995). Inside the black box: The credit channel of monetary policy transmission. Journal of Economic Perspectives, 9:27–48. Bharath, S. and Shumway, T. (2008). Forecasting default with the merton distance to default model. Review of Financial Studies, 21:1339–1369. Boot, A., Greenbaum, S., and Thakor, A. (1993). Reputation and discretion in financial contracting. American Economic Review, 83:1165–1183. Borio, C. and Zhu, H. (2008). Capital regulation, risk-taking and monetary policy: a missing link in the transmission mechanism? BIS Working Papers 268. Brunnermeier, M. (2009). Deciphering the liquidity and credit crunch 2007-2008. Journal of Economic Perspectives, 23:77–100. Coibion, O. (2012). Are the effects of monetary policy shocks big or small? American Economic Journal: Macroeconomics, 4(2):1–32. Dell′ Ariccia, G., Laeven, L., and Marquez, R. (2013). Real interest rates, levearge, and bank risk-taking. Journal of Economic Theory, forthcoming. Dell′ Ariccia, G. and Marquez, R. (2009). Lending booms and lending standards. Journal of Finance, 61:2511–2546. Diamond, D. and Rajan, R. (2009). Fear of fire sales and the credit freeze. NBER, 14925. Erel, I., Julio, B., Kim, W., and Weisbach, M. (2012). Macroeconomic conditions and capital raising. Review of Financial Studies, 25:341–376. 38

Froot, K.A., D. S. and Stein, J. (1993). Risk management: Coordinating corporate investment and financing policies. The Journal of Finance, 68(5):1629–1658. Froot, K. and Stein, J. (1998). Risk management, capital budgeting, and capital structure policy for financial institutions: An integrated approach. The Journal of Financial Economics, 47:55–82. Gertler, M. and Gilchrist, S. (1994). Monetary policy, business cycle, and the behavior of small manufacturing firms. Quarterly Journal of Economics, 109:309–340. Gourieroux, C., Monfort, A., Renault, E., and Trognon, A. (1987). Generalized residuals. Journal of Econometrics, 34:5–32. Hale, G. and Santos, J. (2009). Do banks price their informational monopoly? Journal of Financial Economics, 93:185–206. Ioannidou, V., Ongena, S., and Peydro, J. (2009). Monetary policy, risk-takig and pricing: Evidence from a quasi-natural experiment. CentER Discussion Paper Series 31S. Jimenez, G., Saurina, J., Ongena, S., and Peydro, J. (2014). Hazardous times for monetary policy: What do twenty-three million bank loans say about the effects of monetary policy on credit risk-taking? Econometrica, 82 (2):463–505. Kashyap, A. K. and Stein, J. (2000). What do a million observations on banks say about the transmission of monetary policy. American Economic Review, 90:407–428. Lown, C. and Morgan, D. (2006). The credit cycle and the business: New findings using the loan offer opinion survey. Journal of Money, Credit, and Banking, 38:1576–1597. Merton, R. (1974). On the pricing of corporate debt: the risk structure of interest rates. Journal of Finance, 2:449–470. Mian, A. and Santos, J. (2011). Liquidity risk and maturity management over the business cycle. Federal Reserve Bank of New York, mimeo. Peydro, J. and Maddaloni, A. (2011). Bank risk-taking, securitization, supervision, and low interest rates: Evidence from the euro area and U.S. lending standards. Review of Financial Studies, 24:2121–2165. Rajan, R. (2006). Has finance made the world riskier? 12:499–533.

European Financial Management,

Repullo, R. (2000). Who should act as lender of last resort? An incomplete contracts model. Journal of Money,Credit and Banking, 32:580–605. Rochet, J. (1992). Capial requirements and the behaviour of commerical banks. European Economic Review, 36:733–762. Romer, C. and Romer, D. (2004). A new measure of monetary policy shocks: derivatons and implications. American Economic Review, 94:1055–1084. Santos, J. (2011). Bank loan pricing following the subprime crisis. Review of Financial Studies, 24:1916–1943. 39

Santos, J. and Winton, A. (2008). Bank loans, bonds, and informational monopolies across the business cycle. Journal of Finance, 63:1315–1359. Schreft, S. and Owens, R. (1991). Survey evidence of tigher credit conditions: What does it mean. Economic Review, March/April:29–34. Stein, J. (2013). Overheating in credit markets: Origins, measurement, and policy responses. Speech at the Restoring Household Financial Stability after the Great Recession: Why Household Balance Sheets Matter. Yellen, J. (2011). Assessing potential financial imbalances in an era of accommodative monetary policy. Speech at International Conference: Real and Financial Linkage and Monetary Policy, Bank of Japan.

40

Appendix 1: Definition of Variables ADV ERT ISIN G is advertising expenses scaled by a firm’s sales. CAP IT AL BK is the ratio of equity over risk-weighted assets. CHARGEOF F S BK is net charge off over risk weighted assets. CORP P U RP OSES is one if the loan is used for corporate purpose. CREDIT LIN E equals one if the loan is a credit line. DEBT REP AY is one if the loan is for repayment of previous debt. DIV REST RICT is equal to two if the borrower has to meet a dividend restriction, one is there are not such restrictions and zero if the information is missing. EX RET is the one year stock return above the market return. It is used to calculate P DEF AU LT . F F DIF is the difference between two consecutive policy rates. Higher differential indicates a decrease in the policy rate (multiplied by -1). GDP GROW T H is quarterly GDP growth. L ASSET S BK is the natural log of bank assets in hundreds of millions of dollars at the quarter before loan origination. L AM OU N T is the natural log of loan amount in millions of dollars. LEV ERAGE is debt over total assets. It is used to calculate P DEF AU LT . L IN T COV is the natural log of one plus EBITDA over interest expense. LIQU IDIT Y BK is the ratio of liquidity to risk-weighted assets. L LOAN SP D is the natural log of the all-in-drawn loan spread over LIBOR (in basis points) at origination. L LEN DERS N U M is the natural log of the number of lenders in the syndicate LOW RAT E is an indicator variable that takes one if the federal funds rate is lower than the sample median. L SALES is the natural log of the firm’s annual sales in hundred millions of US dollars. LS DEM AN D is equal to one if a bank indicates that demand for loans is strong in the past three months. LS DEM AN DRES is the generalized residual from a probit regression with dependent variable LS DEM AN D. LS EASIN G is equal to one if a bank indicates that its credit standards for approving applications for C&I loans or credit lines–other than those to be used to finance mergers and acquisitions–to large and middle-market firms are easier than in the previous three months. LS EASIN GRES is the generalized residual from a probit regression with dependent variable LS EASIN G. LSize EASIN G is equal to one if a bank indicates that the terms for the maximum size of credit lines for large and middle-market firms has eased compared to the previous quarter. 41

LSize EASIN GRES is the generalized residual from a probit regression with dependent variable LSize EASIN G. L SP READ is the natural log of the difference between the Moody’s indexes on the yields of AAA- and BBB-rated bonds. M AT U RIT Y is the maturity of the loan in years. M KT BOOK is the ratio of market to book value of the firm. M ON T HS ROM ERS IN D is the number of consecutive months for which ROM ERS IN D is either zero or one. N W C is net working capital over debt. P DEF AU LT is the probability of default defined as the cumulative normal distribution of distance-to-default measure proposed by Bharath and Shumway (2008). See online appendix for details. P RIV AT E indicated whether the bond is placed privately. P ROF M ARGIN is the ratio of net income over sales. R&D is research and development expenses scaled by a firm’s sales. RET AIN ED SHARE is the share of the loan retained by the lead arranger. RISK T OL is equal to one in the quarters a bank indicates increased tolerance for risk was very important or somewhat important for easing the terms for C&I loans. RISK T OLRES is the generalized residual from a first-stage probit regression with dependent variable risk tolerance that takes one if risk tolerance is an important reason for easing lending standards for approving applications for C&I loans or credit lines, and zero otherwise. ROA BK is the bank’s net income before taxes over risk weighted assets. ROA V OL BK is the volatility of the bank’s return on assets. ROM ERS is the residuals retrieved from a regression of intended federal funds rate on the Federal Reserve’s internal forecast of inflation and real activity (Romer and Romer (2004)). Higher values indicate stronger monetary policy easing (the residual is multiplied by -1). ROM ERS IN D is an indicator variable that takes one if ROM ERS is higher than zero (i.e, easing) and zero otherwise. SECU RED is equal to two if the loan is secured; one it is not secured, and zero if the information is missing. SLOP E Y C is the difference between the yields of the five and one year zero coupon bond. ST OCK V OL is the one year stock return volatility using daily returns. It is used to calculate P DEF AU LT . SU BDEBT BK is the fraction of the bank’s subordinated debt to total assets. T AN GIBLES is inventories plus plant, property, and equipment over total assets. T AY LOR is the Taylor rule residual based on a regression of the target for the federal funds rate on inflation and unemployment net of the Congressional Budget Office natural unemployment rate. Higher values indicate stronger monetary policy easing (the residual is multiplied by -1). 42

T AY LOR IN D is an indicator variable that takes one if T AY LOR is higher than zero (i.e., easing) and zero otherwise. T ERM LOAN is equal to one if a loan is a term loan. V IX is the quarterly Chicago Board Options Exchange Market Volatility Index based on the implied volatility of S&P 500 index options. W ORK CAP IT AL is one if the loan is used for working capital.

43

Table 1: SAMPLE CHARACTERISTICS MEAN L SALES TANGIBILITY R&D ADVERTISING LINTCOV MKTBOOK PROF MARGIN NWC P DEFAULT

1.576 0.726 0.021 0.010 1.908 1.762 -0.046 5.923 0.037

L AMOUNT LOAN SPD (bsp) MATURITY (years) CREDIT LINE TERM LOAN CORP PURPOSE WORK CAPITAL DEBT REPAY SECURED DIV RESTRICT RETAINED SHARE L LENDERS NUM

4.067 238.141 4.030 0.547 0.285 0.329 0.113 0.128 1.21 0.571 40.607 1.230

L ASSETS BK CAPITAL BK SUBDEBT BK LIQUIDITY BK ROA BK ROA VOL BK CHARGEOFFS BK AAA BK AA BK A BK BBB BK BELOW BBB BK UNRATED BK LS EASING RISK TOL LSize EASING LS DEMAND

16.547 0.081 0.014 0.248 0.001 0.001 0.001 0.018 0.242 0.432 0.029 0.001 0.279 0.383 0.267 0.109 0.282

L SPREAD SLOPE YC GDP GROWTH VIX FF DIF LOW RATE ROMERS ROMER IND MONTHS ROMERS IND TAYLOR TAYLOR IND

1.882 0.836 1.222 19.69 0.001 0.422 -0.029 0.447 6.621 0.024 0.428

ST.DEV

25th MEDIAN FIRM CONTROLS 1.983 0.243 1.591 0.375 0.446 0.713 0.055 0.000 0.000 0.024 0.000 0.000 1.240 1.188 1.767 1.039 1.120 1.422 0.653 -0.005 0.032 22.952 0.031 0.433 0.107 0.000 0.000 LOAN CONTROLS 1.645 2.995 4.094 148.265 125.000 225.000 2.406 2.083 4.333 0.498 0.000 1.000 0.452 0.000 0.000 0.470 0.000 0.000 0.316 0.000 0.000 0.334 0.000 0.000 0.857 0.000 1.000 0.879 0.000 0.000 33.828 13.333 26.087 1.078 0.000 1.099 BANK CONTROLS 2.288 15.412 17.048 0.022 0.068 0.078 0.013 0.000 0.012 0.092 0.182 0.241 0.001 0.001 0.001 0.002 0.000 0.001 0.001 0.000 0.001 0.132 0.000 0.000 0.428 0.000 0.000 0.496 0.000 0.000 0.168 0.000 0.000 0.035 0.000 0.000 0.448 0.000 0.000 0.486 0.000 0.000 0.442 0.000 0.000 0.311 0.000 0.000 0.450 0.000 0.000 ECONOMY CONTROLS 0.196 1.737 1.907 0.832 0.103 0.681 0.656 0.986 1.255 7.883 13.951 18.261 0.003 -0.003 0.003 0.494 0.000 0.000 0.228 -0.165 -0.028 0.498 0.000 0.000 6.432 2.000 4.000 0.937 -0.57 0.155 0.494 0.000 0.000

75th 2.937 0.980 0.012 0.006 2.466 2.001 0.071 1.478 0.004 5.192 305.000 5.000 1.000 1.000 1.000 0.000 0.000 2.000 2.000 60.8 2.079 18.063 0.091 0.024 0.306 0.002 0.002 0.001 0.000 0.000 1.000 0.000 0.000 1.000 1.000 1.000 1.000 1.000 2.014 1.581 1.609 23.222 0.003 1.000 0.094 1.000 9.000 0.733 1.000

Table 2: LOAN SPREAD REGRESSIONS: BASE RESULTS I The dependent variable is L LOAN SP D, the log of the all-in-drawn spread over LIBOR at origination. The omitted category for SECURED is unsecured and for DIV RESTRICT, no dividend restrictions. The missing categories for both variables are not reported. All variables are defined in Appendix 1. All models include year, quarter, and bank/bank-firm fixed effects. *** denotes 1% significant level, ** denotes 5% significant level, and * denotes 10% significant level. P DEFAULT FF DIFF P DEFAULT×F F DIF

(1) 0.781*** (0.101) 0.150*** (0.032) -0.288** (0.144)

TAYLOR

(2) 0.878*** (0.045)

ROMERS

R&D ADVERTISING LINTCOV MKTBOOK PROF MARGIN NWC Loan Controls L AMOUNT MATURITY SECURED CREDIT LINE TERM LOAN DIV RESTRICT

(5) 0.865*** (0.055)

-0.136*** (0.006) -0.096*** (0.025) -0.877*** (0.286) -0.652** (0.300) -0.092*** (0.013) -0.100*** (0.013) 0.044*** (0.015) 0.001*** (0.000)

-0.140*** (0.004) -0.102*** (0.012) -0.899*** (0.098) -0.624*** (0.164) -0.096*** (0.004) -0.100*** (0.005) 0.038*** (0.012) 0.001*** (0.000)

-0.057*** -0.057*** (0.006) (0.005) 0.008 0.003 (0.006) (0.002) 0.559*** 0.556*** (0.026) (0.012) -0.384*** -0.455*** (0.067) (0.025) -0.115** -0.169*** (0.050) (0.026) 0.238*** 0.234*** (0.026) (0.014) Continued on Next Page...

45

(6) 0.670*** (0.102)

0.032*** (0.006) -0.139*** (0.053) -0.020 (0.034) -0.473** (0.240)

P DEFAULT×ROMERS

TANGIBLES

(4) 0.730*** (0.132) 0.141*** (0.037) -0.502* (0.263)

0.066*** (0.014) -0.076* (0.043)

P DEFAULT× TAYLOR

Firm Controls LSALES

(3) 0.886*** (0.107)

-0.015 (0.032) -0.835*** (0.264)

-0.139*** (0.007) -0.104*** (0.024) -0.860*** (0.290) -0.660* (0.342) -0.093*** (0.012) -0.100*** (0.013) 0.039** (0.018) 0.001*** (0.000)

-0.155*** (0.011) -0.020 (0.041) -0.836** (0.379) -0.842* (0.505) -0.079*** (0.011) -0.087*** (0.009) 0.075*** (0.027) 0.000 (0.000)

-0.133*** (0.009) -0.030 (0.028) -0.968*** (0.274) -0.708* (0.412) -0.079*** (0.005) -0.094*** (0.007) 0.042** (0.017) -0.000 (0.000)

-0.155*** (0.010) -0.021 (0.036) -0.989** (0.435) -0.745 (0.489) -0.076*** (0.012) -0.086*** (0.010) 0.073*** (0.026) 0.000 (0.000)

-0.057*** (0.006) 0.002 (0.007) 0.558*** (0.031) -0.432*** (0.061) -0.146*** (0.044) 0.236*** (0.029)

-0.038*** (0.005) -0.003 (0.003) 0.345*** (0.015) -0.251*** (0.043) -0.093** (0.045) 0.115*** (0.021)

-0.043*** (0.005) -0.007*** (0.002) 0.345*** (0.012) -0.292*** (0.021) -0.116*** (0.022) 0.111*** (0.013)

-0.040*** (0.005) -0.003 (0.004) 0.339*** (0.015) -0.253*** (0.038) -0.096** (0.041) 0.111*** (0.022)

CORPORATE PURPOSE DEBT REPAY WORK CAPITAL RETAINED SHARE L LENDERS NUM Bank Controls LASSETS BK ROA BK SUBDEBT BK ROAVOL BK CHARGEOFFS BK LIQUIDITY BK CAPITAL BK AAA BK AA BK A BK BBB BK BELOW BBB BK Economy Controls L SPREAD SLOPE YC GDP GROWTH VIX Constant Observations R2

Table 2—Continued 0.001 0.234*** (0.021) (0.014) -0.030* -0.030** (0.016) (0.013) -0.036* -0.050*** (0.020) (0.012) 0.023* 0.034*** (0.013) (0.007) -0.011 -0.013** (0.014) (0.006)

-0.013 (0.023) -0.032* (0.017) -0.050** (0.021) 0.036*** (0.009) -0.017 (0.015)

-0.054*** (0.015) -0.079*** (0.018) -0.104*** (0.015) 0.013 (0.012) -0.011 (0.009)

-0.048*** (0.010) -0.093*** (0.012) -0.093*** (0.012) 0.018*** (0.006) -0.007 (0.006)

-0.062*** (0.015) -0.076*** (0.015) -0.110*** (0.014) 0.013 (0.012) -0.011 (0.011)

0.032 (0.043) 11.280 (8.850) 0.981 (1.388) 4.323 (6.596) 9.758 (7.723) -0.111 (0.191) -0.022*** (0.006) 0.246*** (0.091) -0.125** (0.055) -0.135*** (0.049) -0.067* (0.040) 0.051 (0.121)

0.019 (0.019) 7.183 (6.016) 1.126 (0.872) 0.034 (5.606) 5.375 (6.349) -0.122 (0.110) -0.024*** (0.005) 0.098 (0.212) -0.123** (0.049) -0.130*** (0.048) -0.053 (0.049) 0.065 (0.159)

0.016 (0.044) 7.296 (10.092) 0.811 (1.482) 1.034 (6.610) 0.187 (5.876) -0.139 (0.208) -0.024*** (0.007) 0.128 (0.100) -0.109* (0.058) -0.115** (0.051) -0.044 (0.038) 0.064 (0.152)

-0.012 (0.045) 2.695 (9.053) 0.899 (1.262) 11.307* (6.039) 7.317 (10.378) -0.058 (0.152) -0.009 (0.007) 0.409*** (0.090) 0.008 (0.049) -0.004 (0.050) 0.020 (0.047) -0.210 (0.174)

0.198*** (0.012) 5.676 (5.616) 4.948*** (0.782) 25.923*** (5.356) 1.735 (5.811) 0.183* (0.108) -0.009** (0.004) 0.069 (0.182) 0.033 (0.054) -0.005 (0.053) 0.066 (0.055) -0.180 (0.149)

-0.019 (0.042) -3.092 (8.681) 1.093 (1.111) 6.662 (7.029) -0.130 (9.090) -0.122 (0.141) -0.012 (0.008) 0.194* (0.109) 0.004 (0.052) -0.005 (0.053) 0.035 (0.048) -0.216 (0.171)

0.308*** (0.114) 0.047*** (0.015) -0.008*** (0.003) 0.001 (0.003) 0.489 (0.933) 17,974 0.561

0.096 (0.075) 0.094*** (0.015) -0.007*** (0.002) 0.006*** (0.002) 1.105 (0.920) 17,974 0.565

0.042 (0.092) 0.066*** (0.018) -0.005* (0.003) 0.005* (0.003) 1.576* (0.826) 17,974 0.565

0.327*** (0.118) 0.062*** (0.014) -0.005** (0.002) 0.001 (0.002) 0.668 (0.723) 17,974 0.375

0.028 (0.031) 0.104*** (0.006) -0.013*** (0.002) 0.007*** (0.001) 1.188 (0.856) 17,974 0.351

0.162 (0.126) 0.078*** (0.014) -0.005** (0.002) 0.004*** (0.001) 1.737 (0.982) 17,974 0.3660

46

Table 3: LOAN SPREAD REGRESSIONS: BASE RESULTS II The dependent variable is L LOAN SP D, the log of the all-in-drawn spread over LIBOR at origination. All variables are defined in Appendix 1. All models include year, quarter, and bank/bank-firm fixed effects. *** denotes 1% significant level, ** denotes 5% significant level, and * denotes 10% significant level. P DEFAULT LOW RATE P DEFAULT× LOW RATE

(1) 0.947*** (0.105) 0.070*** (0.022) -0.238*** (0.082)

TAYLOR IND

(2) 0.946*** (0.059)

(4) 0.887*** (0.143) 0.120*** (0.026) -0.421*** (0.125)

0.004 (0.018) -0.110* (0.062)

P DEFAULT× TAYLOR IND ROMERS IND

Yes Yes Yes Yes Yes Yes Yes

Yes Yes Yes Yes Yes Yes Yes

17,974 0.579

17,974 0.565

17,974 0.557

47

(5) 0.816*** (0.068)

(6) 0.772*** (0.102)

0.051*** (0.011) -0.035 (0.090) -0.031*** (0.011) -0.200* (0.105) Yes Yes Yes Yes Yes Yes Yes

P DEFAULT× ROMERS IND Loan Controls Firms Controls Bank Controls Economy Controls Year Quarter Bank FE Bank-Firm FE Observations R2

(3) 0.845*** (0.114)

Yes Yes Yes Yes Yes Yes

Yes Yes Yes Yes Yes Yes

-0.006 (0.014) -0.239** (0.116) Yes Yes Yes Yes Yes Yes

Yes 17,974 0.3698

Yes 17,974 0.351

Yes 17,974 0.365

48

Loan Controls Firms Controls Bank Controls Economy Controls Year Quarter Banks FE Bank-firm FE Firm-time FE Bank-firm-time FE Observations R2

ROMERS×P DEFAULT

ROMERS

P DEFAULT

17,974 0.623

(1) -0.650*** (0.103) -0.023 (0.050) 1.102** (0.558) Yes Yes Yes Yes Yes Yes Yes

L AMOUNT

17,974 0.359

Yes

(2) -0.189 (0.136) -0.129*** (0.046) 0.980** (0.385) Yes Yes Yes Yes Yes Yes

9,239 0.413

8,735 0.634

L LOAN SPD Small Large (3) (4) 0.774*** 1.240*** (0.072) (0.092) 0.009 0.088** (0.054) (0.043) -0.481* -0.770* (0.280) (0.414) Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes

8,735 0.461

Yes

Yes

9,239 0.332

Large (6) 0.920*** (0.068) 0.015 (0.048) -1.349** (0.624) Yes Yes Yes Yes Yes Yes

Small (5) 0.672*** (0.151) 0.123 (0.078) -0.817** (0.394) Yes Yes Yes Yes Yes Yes

17,974 0.352

Yes

(7) 0.805*** (0.096) -0.005 (0.034) -0.769** (0.245) Yes Yes Yes Yes Yes Yes

Yes 17,974 0.309

(8) 0.804*** (0.097) 0.002 (0.035) -0.995*** (0.251) Yes Yes Yes Yes Yes Yes

Table 4: TESTS OF DEMAND FOR LOANS The dependent variable in columns (1) and (2) is loan amount (L AMOUNT) and in columns (3) to (8) is loan spread (LLOAN SP D). All variables are defined in Appendix 1. Bank-firm-time fixed effects denotes fixed effects for the same firm, the same bank in a two-year time period. In columns (4) and (6) the sample is split into large firms (sales higher than the sample median) and in columns (3) and (5) into small firms (sales lower than the sample median). *** denotes 1% significant level, ** denotes 5% significant level, and * denotes 10% significant level.

Table 5: LOAN SPREAD REGRESSIONS: FULL INTERACTION SPECIFICATIONS The dependent variable is L LOAN SP D, the log of the all-in-drawn spread over LIBOR at origination. In these specification all variables are fully interacted with ROM ERS. The estimates on the variables and their interaction terms are not reported for brevity. Variables are defined in Appendix 1. *** denotes 1% significant level, ** denotes 5% significant level, and * denotes 10% significant level. (1) 0.882*** (0.046) -0.344* (0.201) Yes Yes Yes Yes Yes Yes Yes

P DEFAULT P DEFAULT× ROMERS Loan Controls Firms Controls Bank Controls Economy Controls Year Quarter Bank FE Bank-Firm FE R2 Observations

0.570 17,974

49

(2) 0.645*** (0.097) -1.070*** (0.299) Yes Yes Yes Yes Yes Yes Yes 0.373 16,601

Table 6: MONETARY POLICY EASING FOR LONG PERIODS The dependent variable is L LOAN SP D. MONTHS ROMERS IND is the number of consecutive quarters for which ROMERS IND is smaller or greater than zero. All variables are defined in Appendix 1. *** denotes 1% significant level, ** denotes 5% significant level, and * denotes 10% significant level. P DEFAULT ROMERS IND MONTHS ROMERS IND MONTHS ROMERS IND×P DEF AU LT ROMER IND×M ON T HS ROM ERS IN D P DEFAULT×ROM ERS IN D ROMER IND×M ON T HS ROM ERS IN D × P DEF AU LT Loan Controls Firms Controls Bank Controls Economy Controls Year Quarter Bank FE Bank-Firm FE R2 Observations

(1) 0.729*** (0.128) -0.055*** (0.012) -0.002** (0.001) 0.058*** (0.019) 0.011*** (0.003) 0.138 (0.115) -0.071*** (0.025) Yes Yes Yes Yes Yes Yes Yes 0.566 17,974

50

(2) 0.871*** (0.135) -0.027 (0.018) -0.002 (0.001) 0.016** (0.007) 0.007* (0.004) -0.039 (0.146) -0.065*** (0.023) Yes Yes Yes Yes Yes Yes Yes 0.377 17,974

51

Loan Controls Firms Controls Bank Controls Economy Controls Year Quarter Bank FE Bank-Firm FE R2 Observations

ROMER RES×LSize EASIN G × P DEF AU LT

P DEFAULT×LSize EASIN G

LSize EASING×ROM ERS

LSize EASING

ROMERS×RISK T OL × P DEF AU LT

P DEFAULT×RISK T OL

RISK TOL×ROM ERS

RISK TOL

ROMERS×LS EASIN G × P DEF AU LT

P DEFAULT×LS EASIN G

LS EASING×ROM ERS

LS EASING

P DEFAULT×ROM ERS

P DEFAULT

ROMERS

0.535 15,882

Yes Yes Yes Yes Yes Yes Yes

Yes 0.532 15,882

Yes Yes Yes Yes Yes Yes

Lending Standards (1) (2) -0.069** -0.025 (0.03) (0.042) 1.020*** 0.800*** (0.104) (0.094) 0.289 -0.846 (0.191) (0.313) 0.032** 0.026** (0.009) (0.012) 0.087 0.077 (0.074) (0.056) 0.185 0.130 (0.12) (0.152) -1.704*** -0.910* (0.546) (0.574)

0.456 15,882

Yes Yes Yes Yes Yes Yes Yes

0.020 (0.020) -0.007 (0.052) 0.126 (0.319) -1.809* (0.937)

Yes 0.413 15,882

Yes Yes Yes Yes Yes Yes

0.059*** (0.019) 0.133** (0.065) -0.183 (0.127) -1.641** (0.837)

Risk Tolerance (3) (4) -0.071** -0.068 (0.029) (0.059) 1.277*** 1.211*** (0.149) (0.120) -0.078 -0.819*** (0.357) (0.274)

0.552 15,882

0.063** (0.027) 0.125 (0.231) -0.333 (0.352) 2.521** (1.174) Yes Yes Yes Yes Yes Yes Yes

Amounts (5) -0.023 (0.073) -1.064*** (0.229) -0.149 (0.725)

Yes 0.492 15,882

0.002 (0.025) 0.018 (0.091) 0.446 (0.473) 3.513** (1.593) Yes Yes Yes Yes Yes Yes

(6) -0.165** (0.068) -1.109* (0.639) 0.446 (0.473)

Table 7: LENDING STANDARDS: ONE-STAGE ESTIMATION The dependent variable in columns (1) to (4) is L LOAN SP D; in columns (5) and (6) is L AM OU N T . All other variables are defined in Appendix 1. Standard errors are bootstrapped. *** denotes 1% significant level, ** denotes 5% significant level, and * denotes 10% significant level.

52

RISKY BORROWERS SAFE BORROWERS ∆

RISKY BORROWERS SAFE BORROWERS ∆

RISKY BORROWERS SAFE BORROWERS ∆

EASY MONETARY POLICY SOFT BK TIGHT BK ∆ (1) (2) (3) (1)-(2) Table 0.248 0.203 0.045 0.087 0.046 0.041 0.161 0.157 0.004 Table 0.258 0.247 0.011 0.074 0.057 0.017 0.184 0.190 -0.007 Table -0.162 -0.218 0.056 0.014 -0.056 0.070 -0.176 -0.162 -0.014

TIGHT MONETARY POLICY SOFT BK TIGHT BK ∆ (4) (5) (6) (4)-(5) 7, column (1) 0.317 0.206 0.111 0.101 0.060 0.041 0.216 0.146 0.070 7, column (3) 0.376 0.270 0.106 0.119 0.076 0.042 0.257 0.193 0.064 7, column (5) -0.312 -0.204 -0.107 -0.043 -0.048 0.005 -0.268 -0.156 -0.112

0.163 0.065 0.098

-0.096 -0.025 -0.070

-0.066 0.000 -0.066

(7) (3)-(6)

Table 8: LENDING STANDARDS, ONE-STAGE ESTIMATION: MARGINAL EFFECTS This table reports estimates based on Table 7, columns (1), (3) and (5). In this particular example, SOFT/TIGHT BK indicate that LS EASIN G, RISK T OL, or LSizeEASIN G take one/zero. EASY/TIGHT MONETARY POLICY take the values of the 75th/25th percentile of the distribution of ROMERS. RISKY and SAFE BORROWERS have probability of default (P DEF AU LT ) 0.2 and 0.05, respectively.

53

Loan Controls Firms Controls Bank Controls Economy Controls Year Quarter Bank FE Bank-Firm FE R2 Observations

ROMER RES×LSizeEASIN GRES × P DEF AU LT

P DEFAULT×LSizeEASIN GRES

LSize EASINGRES × ROM ERS

LSize EASINGRES

ROMERS×RISK T OLRES × P DEF AU LT

P DEFAULT×RISK T OLRES

RISK TOLRES × ROM ERS

RISK TOLRES

ROMER SHOCK×LS EASIN GRES × P DEF AU LT

P DEFAULT×LS EASIN GRES

LS EASINGRES × ROM ERS

LS EASINGRES

P DEFAULT×ROM ERS

P DEFAULT

ROMERS

0.473 15,882

Yes Yes Yes Yes Yes Yes Yes

Yes 0.412 15,882

Yes Yes Yes Yes Yes Yes

Lending Standards (1) (2) -0.013 0.017 (0.036) (0.024) 1.184*** 0.921*** (0.113) (0.057) -0.299 -1.154*** (0.243) (0.196) 0.023*** 0.017*** (0.007) (0.006) 0.048 0.025 (0.051) (0.024) -0.023 0.018 (0.069) (0.058) -0.629** -0.460* (0.296) (0.264)

0.473 15,882

Yes Yes Yes Yes Yes Yes Yes

0.031*** (0.009) 0.021 (0.036) -0.013 (0.009) -0.808** (0.402)

Yes 0.456 15,882

Yes Yes Yes Yes Yes Yes

0.021*** (0.007) 0.034 (0.031) -0.073 (0.007) -0.451 (0.386)

Risk Tolerance (3) (4) -0.017 0.015 (0.026) (0.023) 1.193*** 0.921*** (0.047) (0.057) -0.366* -1.135*** (0.200) (0.202)

0.570 15,882

0.007 (0.021) 0.110 (0.107) -0.244 (0.147) 1.594*** (0.370) Yes Yes Yes Yes Yes Yes Yes

Amounts (5) -0.011 (0.065) -0.984*** (0.156) 1.096** (0.534)

Yes 0.456 15,882

-0.029** (0.014) -0.026 (0.070) 0.340* (0.185) 1.390* (0.847) Yes Yes Yes Yes Yes Yes

(6) -0.066 (0.043) -0.488*** (0.112) 0.813** (0.382)

Table 9: LENDING STANDARDS: SECOND STAGE The dependent variable in columns (1) to (4) is L LOAN SP D; in columns (5) and (6) is L AM OU N T . All other variables are defined in Appendix 1. Standard errors are bootstrapped. *** denotes 1% significant level, ** denotes 5% significant level, and * denotes 10% significant level.

54

RISKY BORROWERS SAFE BORROWERS ∆

RISKY BORROWERS SAFE BORROWERS ∆

RISKY BORROWERS SAFE BORROWERS ∆

EASY MONETARY POLICY TIGHT MONETARY POLICY SOFT BK TIGHT BK ∆ SOFT BK TIGHT BK ∆ (1) (2) (3) (4) (5) (6) (1)-(2) (4)-(5) Table 9, column (1) 0.210 0.202 0.008 0.273 0.226 0.047 0.074 0.019 0.055 0.079 0.051 0.028 0.136 0.183 -0.047 0.194 0.176 0.019 Table 9, column (3) 0.252 0.224 0.027 0.317 0.233 0.084 0.092 0.045 0.047 0.107 0.052 0.055 0.160 0.183 -0.023 0.210 0.181 0.029 Table 9, column (5) -0.187 -0.178 -0.009 -0.370 -0.204 -0.166 -0.033 -0.048 0.016 -0.101 -0.048 -0.054 -0.155 -0.130 -0.025 -0.268 -0.156 -0.112

0.157 -0.048 0.204

-0.057 -0.005 -0.052

-0.039 0.027 -0.066

(7) (3)-(6)

Table 10: LENDING STANDARDS, TWO-STEP ESTIMATION: MARGINAL EFFECTS This table reports estimates based on Table 9, columns (1), (3) and (5). In this particular example, SOFT/TIGHT BK take the values of the 75th/25th percentile of the distribution of LS EASIN GRES , RISK T OLRES , and LSizeEASIN GRES . EASY/TIGHT MONETARY POLICY take the values of the 75th/25th percentile of the distribution of ROMERS. RISKY and SAFE BORROWERS have probability of default (P DEF AU LT ) 0.2 and 0.05, respectively.

Table 11: LENDING STANDARDS: LOAN DEMAND CHANGE The dependent variable is L LOAN SP D. LSDEM AN DRES is the residual from a probit model with dependent variable LS DEM AN D. All other variables are defined in Appendix 1. Standard errors are bootstrapped. *** denotes 1% significant level, ** denotes 5% significant level, and * denotes 10% significant level. ROMERS P DEFAULT P DEFAULT×ROM ERS LS DEMAND LS DEMAND×ROM ERS P DEFAULT×LS DEM AN D ROMERS×LS DEM AN D × P DEF AU LT

First Stage -0.030 -0.006 (0.044) (0.041) 1.085*** 0.848*** (0.164) (0.146) -0.204 -1.126*** (0.192) (0.305) 0.014 0.004 (0.017) (0.016) -0.040 0.008 (0.057) (0.047) 0.036 0.097 (0.251) (0.240) -0.294 -0.230 (0.577) (0.679)

LS DEMANDRES LS DEMANDRES × ROM ERS P DEFAULT×LS DEM AN DRES ROMERS×LS DEM AN DRES × P DEF AU LT Loan Controls Firms Controls Bank Controls Economy Controls Year Quarter Bank FE Bank-Firm FE R2 Observations

Yes Yes Yes Yes Yes Yes Yes 0.544 14,442

55

Yes Yes Yes Yes Yes Yes Yes 0.421 14,442

Second -0.050 (0.035) 1.082*** (0.135) -0.302 (0.216)

0.007 (0.011) -0.030 (0.033) 0.085 (0.154) -0.360 (0.343) Yes Yes Yes Yes Yes Yes Yes 0.476 14,442

Stage -0.006 (0.024) 0.893*** (0.065) -1.097*** (0.204)

-0.003 (0.007) -0.016 (0.024) 0.091 (0.067) -0.034 (0.276) Yes Yes Yes Yes Yes Yes Yes 0.426 14,442