Institutional Investors, the Dollar, and U.S. Credit

K.7 Institutional Investors, the Dollar, and U.S. Credit Conditions Niepmann, Friederike and Tim Schmidt-Eisenlohr Please cite paper as: Niepmann, Friederike and Tim Schmidt-Eisenlohr (2019). Institutional Investors, the Dollar, and U.S. Credit Conditions. International Finance Discussion Papers 1246. https://doi.org/10.17016/IFDP.2019.1246 International Finance Discussion Papers Board of Governors of the Federal Reserve System Number 1246 April 2019

Board of Governors of the Federal Reserve System International Finance Discussion Papers Number 1246 April 2019 Institutional Investors, the Dollar, and U.S. Credit Conditions Friederike Niepmann and Tim Schmidt-Eisenlohr Notes: International Finance Discussion Papers (IFDPs) are preliminary materials circulated to stimulate discussion and critical comment. References to IFDPs (other than an acknowledgment that the writer has had access to unpublished material) should be cleared with the author or authors. Recent IFDPs are available on the web at https://www.federalreserve.gov/econres/ifdp/. This paper can be downloaded without charge from the Social Science Research Network electronic library at https://www.ssrn.com.

“Institutional Investors, the Dollar, and U.S. Credit Conditions” Friederike Niepmanna,b,c and Tim Schmidt-Eisenlohra,b Abstract: This paper documents that an appreciation of the U.S. dollar is associated with a reduction in the supply of commercial and industrial loans by U.S. banks. An increase in the broad dollar index by 2.5 points (one standard deviation) reduces U.S. banks' corporate loan originations by 10 percent. This decline is driven by a reduction in the demand for loans on the secondary market where prices fall and liquidity worsens when the dollar appreciates, with stronger effects for riskier loans. Today, the main buyers of U.S. corporate loans---and, hence, suppliers of funding for these loans---are institutional investors, in particular mutual funds, which experience outflows when the dollar appreciates. A shift of traditional financial intermediation to these relatively unregulated entities, which are more sensitive to global developments, has led to the emergence of this new channel through which the dollar affects the U.S. economy, which we term the secondary market channel. Keywords: Leveraged loan market, commercial and industrial loans, U.S. dollar exchange rate, credit standards, institutional investors JEL classifications: E44, F31, G15, G21, G23 a Federal Reserve Board; b CESifo; c C.E.P.R.. The authors are grateful to Mark Carey, Ricardo Correa, Daniel Dias, Wenxin Du, Egemen Eren, Seung Lee, Ryan Monarch, Hyun Song Shin, Viktors Stebunovs, and Egon Zakrajsek, as well as participants in workshops at the Federal Reserve Board for helpful comments. The authors also thank Inaki Aldasoro, Andreas Schrimpf, Nathan Converse, Pablo Cuba Borda, Wenxin Du, Juan-Miguel Londono, and Ralf Meisenzahl for making available some data used in this research, as well as Elizabeth Doppelt and Beau Bressler for excellent research assistance. The views in this paper are solely the responsibility of the author(s) and should not be interpreted as reflecting the views of the Board of Governors of the Federal Reserve System or of any other person associated with the Federal Reserve System.

1 Introduction Commercial and industrial loans (C&I) are an important source of credit for U.S. corporates. C&I loans on the balance sheet of U.S. banks stood at more than $2.1 trillion at the end of 2017. That said, the total volume of these loans is significantly bigger because banks sell a large portion of the loans they originate, especially the riskier ones, on the secondary market to institutional investors within 30 days of origination (Lee et al. (2017)). The size of the syndicated loan market, which captures the majority of loans that are originated and then sold, has increased tremendously since the 1990s, and so has the role of institutional investors as buyers of these loans, as figure 1 documents.1 While roughly 65 percent of the buyers were U.S. banks in 1993, this share fell to 18 percent in 2014.2 Today the largest buyers on the secondary market are mutual funds and CLOs.3 This paper shows that the increased importance of institutional investors as a funding source for C&I loans has led to the emergence of a new macro-financial channel that we term the secondary market channel. Because these investors are sensitive to global developments, and, in particular, to dollar movements, credit conditions for U.S. corporates are affected by the dollar exchange rate: When the dollar appreciates, institutional investors reduce their demand for loans on the secondary market. As a consequence, U.S. banks tighten credit standards and originate fewer C&I loans. A tightening in credit standards and a reduction in the supply of loans have real effects on investment, employment, and GDP (see, for example, Peek and Rosengren (2000), Chodorow-Reich (2013), Bassett et al. (2014), Greenstone et al. (2014)). Therefore, a stronger dollar weighs negatively on U.S. economic activity through the secondary market channel, which is distinct from traditional channels linked to the terms of trade or monetary policy. We start by documenting the surprising correlation between U.S. banks’ corporate loan originations and the dollar, illustrated in figure 2. The solid line shows the log difference of 1Whenaloanissyndicated,theleadbankscreenstheborrowerandlooksforparticipantsinthesyndicate, firstamongotherbanks,thenamonginstitutionalinvestors. Theshareofnon-bankparticipantshasincreased over time and is estimated at around 20 percent (Lee et al. (2017)). 2Data underlying figure 1 come from Irani et al. (2018). 3For more information on the secondary market for U.S. corporate loans, see, for example, Yago and McCarthy (2004) and Ivashina and Sun (2011). 1

thevalueofcorporateloanoriginationsfor16majorU.S.banksataquarterlyfrequency. The dashed line depicts the quarterly change in the broad dollar index, a trade-weighted index that includes the bilateral dollar exchange rates of major U.S. trading partners. A clear relationship is apparent: When the dollar appreciates, U.S. banks reduce lending and vice versa. A one-standard deviation rise in the index (equivalent to an increase of 2.5 points) leads to a 10 percent reduction of new loan originations. Further analysis that employs detailed information on banks’ internal risk ratings of loans reveals that banks not only lend less but also shift to safer borrowers when the dollar appreciates. The same picture emerges from an analysis of credit standards. Based on data from the Senior Loan Officer Opinion Survey (SLOOS), an appreciation of the dollar is associated with a tightening of credit standards for C&I loans by U.S. banks both before and after the 2008/2009 Global Financial Crisis (GFC). We also run a vector autogression (VAR) in which the National Financial Conditions Credit (NFCI Credit) Subindex is explained by standard macroeconomic and financial variables to check for additional explanatory power of the dollar for U.S. financial conditions.4 Figure 3 shows the impulse response function of the monthly change in the NFCI Credit to a one standard deviation change in the monthly dollar change for the post-GFC period (from October 2009 to March 2018). The contemporaneous effect is significant at a 95-percent significance level and quantitatively relevant. A one-standard-deviation increase in the dollar leads to a decrease of the NFCI Credit by 0.012 points in the same month (9 percent of a standard deviation of D NFCI Credit) and 0.025 after 8 months (19 percent of a standard deviation of D NFCI Credit). We obtain very similar results when using the overall National Financial Conditions Index (NFCI) as shown in the right panel. Having established this striking correlation between U.S. credit conditions for corporates and the dollar, we provide evidence for the channel that drives this relationship. The SLOOS asks loan officers for the reasons why they tightened or eased credit standards for C&I loans. 4More specifically, we run a VAR where the following variables explain the monthly change in the NFCI Credit (in the following order): the change in the fed funds rate, the change in the term spread, the change in the excess bond premium, the log difference of the VIX, the log difference of the CPI, the log difference of industrial output, the log difference of nonfarm employment, and the change in the dollar. We allow for two lags of all endogenous variables. 2

The only reason that is correlated with dollar movements is the availability of liquidity in the secondary market. That is, when the dollar appreciates, demand in the secondary market drops, worsening the conditions under which banks can offload their loans. In line with loan officers’ explanations, a bank’s loan originations and credit standards are more sensitive to dollar movements when a larger share of the bank’s loans are held for sale, a proxy for the extent to which the bank sells its loans on the secondary market. Weprovidefurtherevidencefortheproposedchannelwithpricingdataofloanstradedon thesecondarymarketavailablefromtheLoanSyndicationsandTradingAssociation(LSTA). Regressionsofmonthlypricechangesandchangesinbid-askspreads, whichmeasureliquidity conditions, show that prices fall and liquidity on the secondary market deteriorates when the dollar appreciates. Consistent with a shift to safer borrowers, price effects are stronger for the riskier portion of the traded loans—those with a price below 97 percent of the par value. Thus, when the dollar appreciates, prices for traded loans fall at the same time as U.S. banks contract lending, which implies that the dollar shifts the demand for loans on the secondary market. In fact, regressing the log change of U.S. banks’ loan origination on the change in the U.S. Leveraged Loan 100 Index and instrumenting the index by dollar movements delivers quantitatively similar results to those from the reduced-form regression of changes in loan originations on changes in the broad dollar index. That is, quantitatively, the secondary market channel can fully explain the link between the dollar exchange rate and U.S. banks’ corporate lending that we uncover. As mentioned before, U.S. banks only represent a small portion of the buyers on the secondary market for C&I loans and their share of syndicated loan purchases has fallen steadily since the 90s.5 Today, institutional investors dominate this market. This evolution is crucial for the correlation between U.S. credit conditions and the dollar. Precisely, the correlations of both prices and bid-ask spreads with the dollar have increased as the share of U.S. banks as buyers has fallen. To show this, we interact changes in the dollar index with the yearly share of syndicated loans bought by U.S. banks and include this term in the price and liquidity regressions, which show highly significant coefficients. We conclude that the 5See Demsetz (2000), Yago and McCarthy (2004), Irani et al. (2018), Drucker and Puri (2008), Lee et al. (2017), and Lee et al. (2015). 3

growing importance of institutional investors in the secondary loan market has led to the emergence of the secondary market channel. We can push the analysis one step further by looking at investment flows to mutual funds, one of the most important type of buyer of syndicated loans in the secondary market, with a share of around 22 percent in 2014. For this, we use EPFR data on U.S. domiciled, U.S. dollar-denominated mutual funds that invest in bank loans exclusively in the United States. Controlling for fund performance in the preceding month, an increase in the broad dollar index is associated with lower net inflows into these funds. That is, when the dollar strengthens, institutional investors withdraw from mutual funds specialized in buying U.S. bank loans on the secondary market, with the effect of lowering the demand in the secondary market and, consequently, reducing C&I lending by U.S. banks. A key challenge for interpreting our results is that the dollar is endogenous and closely linked to macroeconomic and financial conditions. To shed more light on the mechanism and check the robustness of our results, we follow four strategies. First, we exploit heterogeneity across banks in the share of loans held for sale to estimate a specification, where we look for a differential effect of the dollar on lending across banks. In that specification, we can directly control for all macroeconomic effects by including a time-fixed effect. We further estimate this specification at the bank-borrower-time level, which allows for the inclusion of borrower-time fixed effects that control for any demand effects at the borrower level. Second, we directly control for changes in U.S. economic and financial conditions as well as the U.S. economic and financial outlook. The variables we include are the U.S. unemployment rate and real GDP, the U.S. unemployment outlook and real GDP outlook, the oil price, U.S. short-term interest rates, and the term-spread, as well as future expected interest rates.6 Moreover, we add various measures of risk aversion and uncertainty, namely: the CBOE Volatility Index (VIX), the U.S. excess bond premium (Gilchrist and Zakrajˇsek (2012)), theU.S.varianceriskpremium(LondonoandZhou(2017)), the3-yearU.S.treasury basis (Du et al. (2018b)), and an estimate of the world recession probability (Cuba-Borda et al. (2018)). Adding these controls neither materially affects the significance nor the 6Values of expected short rates and the term spread are from the Survey of Professional Forecasters. 4

magnitude of our results. Third, we instrument changes in the broad dollar index with the response of the dollar exchange rate to monetary policy announcements by five advanced countries’ central banks within a 90-minute window of the announcements, available from Ferrari et al. (2017) and Cieslak and Schrimpf (2019). These surprise dollar movements should be exogeneous to developments in the United States and are unlikely to reveal any news about the U.S. economy.7 Results on credit standards continue to hold when the broad dollar index is instrumented with surprise dollar movements triggered by foreign monetary policy changes. Therefore, dollar movements caused by foreign developments affect the credit standards set by U.S. banks.8 Finally, we directly test for alternative hypotheses. First, when the dollar appreciates, credit risk on banks’ books might rise which may make banks more risk averse. However, risk ratings of existing loans on banks’ books do not move with the dollar. Second, even if loans to U.S. firms might not get riskier, a dollar appreciation could raise credit risk for banks with dollar loans to foreign borrowers (for example, Yesin (2013), Bruno and Shin (2014)). When the dollar appreciates, foreign firms might have a harder time repaying their loans if they have a currency mismatch and are not fully hedged. However, the correlation between loan originations and the dollar even holds for U.S. banks with basically zero foreign business. Moreover, the relationship between lending and the dollar is uncorrelated with the share of foreign assets on a bank’s balance sheet. Third, a stronger dollar might make it harder for U.S. firms to sell their goods abroad. Hence, these firms might invest less and demand less credit. But banks reduce lending to all firms, even to firms in industries that do not import and export, whose business should not be directly affected by dollar movements. In addition, the results go though when we control explicitly for changes in the demand for credit. Combining the different pieces of evidence, we conclude that U.S. banks do not 7Monetary policy suprises abroad might have a direct effect on U.S. credit conditions. While this could potentially invalidate the exlusion restriction, any bias arising from a direct spillover through the interest ratechannelshouldhavetheoppositesignofthesecondarymarketchannelweareinterestedin. Foramore detailed discussion see section 4.1. 8In this regard, the U.S. economy might also be affected by what has been termed the global financial cycle. In this context, see works by Baskaya et al. (2017), Miranda-Agrippino and Rey (2015), and Cerutti et al. (2014), for example. 5

respond directly to dollar movements or changes in the demand for credit that may come with it but instead to conditions on the secondary market for bank loans, which become less favorable when the dollar appreciates. The key open question that then remains is why institutional investors invest less in U.S. corporate loans when the dollar appreciates. To shed further light on this question, we split the broad dollar index into an emerging markets dollar index and an advanced economies dollar index. Results are mainly driven by the emerging markets dollar index, especially post-GFC. This finding is consistent with the idea that institutional investors could be sensitive to the dollar because of exposures to dollar-denominated emerging market corporate debt (Bruno and Shin (2014)). The issuance of this type of debt has surged since 2009, and it becomes riskier when the dollar appreciates because firms are often not fully hedged (Miyajima and Shim (2014), Chui et al. (2014), Niepmann and Schmidt-Eisenlohr (2017)). That said, our findings might also be explained by theories of financial intermediary asset pricing, if institutional investors pull out of U.S. corporate debt at the same time as they make other adjustments to their portfolios that affect the dollar.9 Whatever the underlying mechanisms are, the dollar is undoubtedly closely linked to the risk preferences of institutional investors (Avdjiev et al. (forthcoming)), a fact that is also reflected in U.S. corporate bond yields as we show in section 5.2.10 As institutional investors have become more involved in traditional financial intermediation in the United States, they are changing macro-financial linkages as demonstrated in this paper. Through the secondary market channel, U.S. banks’ credit supply to U.S. firms has become dependent on non-bank investors, who adjust their investments in response to foreign developments. It is crucial that these unregulated entities, whose behaviour clearly differs from traditional banks, are studied further and monitored. More data on their asset allocations are needed to better understand their investment behavior, and more broadly, to ensure the stable supply of credit to U.S. firms. 9See Gabaix and Maggiori (2015), Du et al. (2017), Adrian et al. (2014), Jiang et al. (2018a), Verdelhan (2018), He et al. (2017), and Malamud and Schrimpf (2016). 10Inthatpaper,changesinthedollarandcross-borderbankflowshavealsobeenlinkedtoCIPviolations. In this context, see also Braeuning and Ivashina (2017a) and Du et al. (2018a). 6

2 Data The data used in this paper come from a variety of sources. We discuss three of the main data sources in this section. Other data that complement the analysis are detailed in the data appendix. 2.1 Introducing key data sources SLOOS The Senior Loan Officer Survey (SLOOS) is conducted by the Federal Reserve on a quarterly basis. The survey asks loan officers about their lending practices and reasons for any changes over a three-month period. The aggregated survey responses are published on the Federal Reserve’s website.11 Confidential bank-level responses are kept at the Federal Reserve and are available to researchers there. The survey was established in 1990 and has beenwidelyusedinresearch(forexample, Bassettetal.(2014), Hirtle(2009), andPaligorova and Santos (2017)). The reporting panel consists of up to 80 large U.S. chartered commercial banks and up to 24 U.S. branches and agencies of foreign banks. This paper will largely rely on information about credit standards for commercial and industrial loans, including the reported reasons for changes in standards. U.S. loan-level data Loan-level data for large U.S. banks come from the so called FR Y-14 reports available at the Federal Reserve.12 All Bank Holding Companies (BHCs) that participate in the U.S. stress tests must report detailed, confidential information on their corporateloansandleaseswithacommittedexposureabove$1milliononaquarterlybasis.13 The data start in 2011:Q3, and the sample used in this paper runs through 2017:Q4. Among others, the data provide information on committed and utilized exposures, the date when the loan was originated, the probability of default and/or the rating that the bank assigns to the loan, and the industry in which the borrower is located. From these data, we obtain bank-level loan originations with information on total volumes, the number of loans, and 11https://www.federalreserve.gov/data/sloos/about.htm. 12https://www.federalreserve.gov/apps/reportforms/reportdetail.aspx. 13These data have been used in Bidder et al. (2017) and Brown et al. (2017), for example. 7

weighted-average risk ratings.14 Data on the secondary loan market For information on pricing and the liquidity of loans in the secondary loan market, we draw on data from the Loan Syndications and Trading Association (LSTA). These data are daily and contain information on the name of the borrower as well as the mean bid and mean ask price of each loan, among other things. As a measure of liquidity, we compute the spread between the mean ask and mean bid price of a loan (for example, Santos and Shao (2018)). Prices are defined as the average of the mean bid and mean ask price by loan. In addition to the loan-level data, we employ the U.S. Leverage Loan 100 Index, also provided by the LSTA. Loan pricing data are available from 1998 onward. Other data sources For balance sheet data of U.S. banks, we draw on the FR Y-9C reports. Information on dollar exchange rates is from the Federal Reserve, which calculates several trade-weighted dollar indexes, most importantly the broad dollar index.15 Information on the economic and the interest rate outlook are from the Survey of Professional Forecasters. For more information on additional data sources, see the data appendix. 2.2 Lending standards and loan originations In this paper, we will use credit standards as reported by banks and their quarterly loan originations to characterize U.S. credit conditions. Table 1 demonstrates that these two measures are highly correlated: Easier lending standards predict an increase in loan originations at the bank level in the same quarter and in the following quarter. Column (1) of the table shows results of a bank-level regression, where the log of new loans issued by bank 𝑏 to U.S. firms is regressed on a dummy variable and its lag, which takes the value of 1 if the bank reported that it eased credit standards for medium-sized and large firms in a 14A loan is labeled a new loan if its reported origination date lies in the quarter in which the loan was first reported by the bank. 15The broad dollar idex is a weighted average of the foreign exchange value of the U.S. dollar against the currencies of the following major U.S. trading partners: Euro Area, Canada, Japan, Mexico, China, United Kingdom, Taiwan, Korea, Singapore, Hong Kong, Malaysia, Brazil, Switzerland, Thailand, Philippines, Australia,Indonesia,India,Israel,SaudiArabia,Russia,Sweden,Argentina,Venezuela,Chile,andColombia. 8

given quarter or zero otherwise. The regression also includes a dummy variable and its lag that takes the value of 1 if the bank reported that it tightened credit standards for these firms.16 The coefficients associated with easing credit standards are positive and significant at the 5-percent level, suggesting that easier credit standards in quarter 𝑡 and quarter 𝑡−1 are associated with larger loan originations in quarter 𝑡. Columns (2) and (3) confirm this result in the aggregate. The dependent variable in each of these columns is the log volume of all loan originations in the Y-14 data. The right-hand side variables were computed by first multiplying the bank-level dummy variables with each bank’s average total C&I loans from 2011 to 2017, then summing them over all banks and taking the log values.17 Column (3) controls for the total volume of C&I loans of the banks that answered the question that quarter. These regressions confirm that an easing of credit standards is associated with higher loan orginations today and in the next three months. We can therefore interpret both SLOOS credit standards and loan originations from the Y-14 data as measures of current U.S. credit conditions. 3 U.S. Credit Conditions and the Dollar ThissectionexplorestherelationshipbetweencreditconditionsforU.S.firms, liquidityinthe secondary market for U.S. bank loans, and the dollar. It starts by documenting the striking correlation between U.S. banks’ loan originations and credit standards and the dollar, and then presents the reason for the correlation: a drop in the demand for U.S. bank loans by institutional investors in the secondary market. The regressions presented in this section are largely run on aggregate data to emphasize the aggregate relevance of the correlations shown and include only a few key macro and financial control variables. Robustness of the results and the discussion of possible confounding factors can be found in section 4. 16SLOOS responses, which cover credit standards in the first quarter and are published in the second quarter, are related to new loans in the first quarter. 17The Y-14 data are reported by BHCs, while the SLOOS data are reported by commercial banks. We map commercial banks to BHCs and sum responses over all commercial banks by BHC, assigning an equal weight to each commercial bank owned by the same BHC. Note that 1 was added to the quarterly weighted average dummy variables before taking the log to avoid dropping zero values. 9

3.1 A stronger dollar implies tighter U.S. credit conditions Corporate loan originations decline As shown in figure 2 in the introduction, U.S. banks’ loan originations are highly correlated with the dollar. Table 2 presents this result in the form of regression analysis. In column (1), the quarterly log change of domestic loan originations of the 16 U.S. banks that are continuously in the sample is regressed on the change in the average quarterly broad dollar index. The estimated coefficient of -0.04 indicates that a one standard deviation rise in the broad dollar index (equivalent to an increase of 2.5 points and reflecting an appreciation of the dollar) leads to a 10 percent reduction in new loans issued by U.S. banks. The 𝑅2 of this simple time series regression is remarkably high, at 18 percent. Of course, the dollar is related to other factors that also affect banks’ loan supply. First, when U.S. monetary policy tightens and interest rates rise, the dollar appreciates at the same time as bank lending declines. We therefore include in the regressions the spread between the 10-year and 3-month treasury rate as well as the change in the effective federal funds rate, which we set equal to the Wu-Xia shadow rate (Wu and Xia (2016)) at the zero lower bound. Second, the dollar tends to appreciate in times of uncertainty, when banks may de-lever and issue fewer loans. Therefore, we control for the change in the log of the CBOE Volatility Index (VIX). The dollar might also reflect expectations about the future performance of the U.S. economy and, thereby, affect loan supply. To control for this channel, we exploit the excess bond premium introduced by Gilchrist and Zakrajˇsek (2012), which is a leading indicator of U.S. economic activity. We also obtain the U.S. unemployment outlook through the Survey of Professional Forecasters.18 Both variables are included in the regressions as simple quarterly changes. We work with this set of control variables to produce the baseline results and later expand the set in the robustness section. The effect of the dollar on loan originations is little changed when the control variables are included (column (2) of table 2). As columns (3) through (6) highlight, the correlation between the dollar and loan originations stems mainly from the extensive margin. The 18Unemployment outlook is defined as the mean 4-quarter-forward forecasted unemployment rate, minus the mean “nowcasted” unemployment rate. 10

coefficient associated with the dollar is significant and equal to -0.3 when the log change in the number of loans is the dependent variable (columns (3) and (4)). In contrast, it is not significant in the regression of the change in the average loan size and smaller, estimated at -0.1 (columns (5) and (6)). Accordingly, the extensive margin accounts for 75 to 80 percent of the effect. Credit standards tighten While the U.S. loan-level data from the FR Y-14 reports are onlyavailablefrom2011:Q3,theSLOOSdataareavailableforalongertimeseries,goingback to 1990. Table 3 shows regressions that employ these data. In each column, the dependent variable, the net percentage of banks that report tightening in credit standards for C&I loans to large and medium-sized corporations, is regressed on changes in the broad dollar index (and control variables). Because survey responses are persistent, the lagged value is also included as a regressor. In columns (1) and (2), the coefficients associated with the broad dollar index are insignificant. However, once we exclude U.S. recessions from the sample, the dollar coefficient becomes highly significant (columns (3) and (4)).19 The coefficient of 0.92 in column (4) implies that a one-standard deviation rise in the broad dollar index increases the net percentage of survey respondents that report tightening credit standards by 2.3 percentage points, corresponding to a beta coefficient of 15 percent. This compares to a beta coefficient of the log VIX of 19 percent. Thus the dollar effect is almost as strong as the effect of the VIX. Table 26 in the appendix splits the SLOOS sample into a pre- and post-GFC period, showing that the correlation between U.S. banks’ credit standards and the dollar holds during both time periods with relatively stable dollar coefficients. Because we are interested in the relationship between the dollar and credit supply in normal times, we exclude recession quarters from now on in all regressions.20 19Specifically, we drop all quarters during NBER recessions and the Global Financial Crisis from the sample, that is, 1990, 1991, 2001, and 2007:Q3 to 2009:Q2. Results are very similar when we only drop the Global Financial Crisis. 20TheSLOOSnotonlyprovidesinformationoncreditstandardsforC&Iloans, butalsoforothertypesof lending. Appendix table 25 highlights that credit standards for C&I loans to small firms as well as those for commercial real estate are also correlated with the dollar. We also ran regressions on credit standards for consumerloansandresidentialrealestatelending, forwhichthedollarcoefficientswerealwaysinsignificant. 11

Banks lend to safer borrowers Not only the quantity of new loans and credit standards are affected when the dollar appreciates, but also the riskiness of the new loans. Table 4 documents that banks lend to safer borrowers when the dollar appreciates. In the table, the change in the weighted average probability of default of new loans is regressed on the change in the dollar index.21 Changes in the dollar index alone explain 33 percent of the variation in the change in the weighted average probability of default. According to the estimate in column (2), a one-standard deviation increase in the broad dollar index reduces the average probability of default that banks assign to their new loans by 9 basis points. This compares to an average probability of default of 100 basis points. 3.2 The secondary market as the driver Loan officer survey points to secondary market as the cause Having documented that U.S. credit conditions for corporations tighten when the dollar appreciates, we search for an explanation. Fortunately, the SLOOS asks banks for their reasons for tightening or easing credit standards. Table 5 presents regressions of the share of banks that mentioned a particular reason for tightening (top panel) or easing (bottom panel) credit standards out of all banks that reported tightening or easing credit standards on the change in the broad dollar index. Banks are free to mention any of the following reasons: (i) changes in the bank’s capital position, (ii) changes in competition from other lenders, (iii) legislative changes, supervisory actions, or changes in accounting standards, (iv) changes in the bank’s current or expected liquidity position, (v) changes in the economic outlook, (vi) changes in the bank’s risk tolerance, (vii) changes in liquidity in the secondary market for these loans, (viii) industry-specific problems. As the table shows, the single reason that is significantly associated with dollar movements is “changes in the liquidity in the secondary market for these loans”. When the dollar appreciates, fewer banks mention increased liquidity in the secondary market as a reason for easing. Table 8 explores this result further, displaying the 21The probability of default is from the banks’ internal risk models and captures the probability that a borrower defaults on its obligations within the next 12 months. The weighted average probability of default was constructed as follows: First, loans for which the probability of default was missing were assigned the average probability default of loans with the same rating. Second, probabilities of default above 10 percent were dropped. Third, only loans of the 16 banks that are always in the sample were included in the calculation. 12

full regression in column (1) and including the set of control variables in column (2). As in earlier regressions, the dollar coefficient continues to be significant and its magnitude is little changed when these variables are added to the regression. According to the coefficient of -0.012 in column (2), the share of banks that report secondary market liquidity as a reason for easing decreases by 3 percentage points when the dollar index rises by 2.5 points.22 While we explore secondary market prices and liquidity of U.S. corporate loans below, figure 4 visually confirms loan officers’ responses. The figure plots the three-quarter moving averageshareofbanksthatreportedsecondarymarketliquidityasareasonforeasingagainst the three-quarter moving average S&P/LSTA Leverage Loan 100 Index, which reflects the performance of the largest facilities in the U.S. leveraged loan market. Clearly more loan officers report favorable conditions on the secondary market for corporate loans as a reason for easing when prices on the secondary market are higher and vice versa. Share of loans held for sale explains bank’s sensitivity to the dollar To confirm thecentralroleofthesecondarymarket, westudytheeffectsofthedollaronU.S.banks’loan originations at the bank level. From FR Y-9C data, we obtain the average share of loans held for sale over a bank’s total loans and lease financing receivables and interact this share with changes in the dollar index. Results of these bank-level regressions are presented in table 7. Column (1) includes the interaction term that tests whether banks’ sensitivities to the dollar depend on the extent to which they sell off loans. The interaction term is significant at the 1 percent level, indicating that loan originations of banks that pass on more loans to outside investors are more responsive to dollar movements, entirely consistent with the role of the secondary market for the presence of the documented correlation. This result is robust to controlling for a bank’s Tier1-capital ratio and its ratio of wholesale funding to total loans (column 2) as well as time-fixed effects (column 3).23 Of note, the regressions shown in these columns indicate that conditional on a bank’s share of loans held for sale, the bank’s loan originations are less sensitive to the dollar when its Tier1-capital ratio is higher. In this case, 22It is somewhat surprising that changes in secondary market liquidity as a reason for tightening are not associated with dollar movements, which may be related to banks’ reporting incentives and practices. 23A bank’s wholesale funding share was computed following Choi and Choi (2016). 13

the bank might be less constrained to keep risky loans on its balance sheet.24 Wealsofinddifferencesinthesensitivitiesofbanks’creditstandardstodollarmovements. In column (4), individual bank responses are regressed on changes in the dollar and their interaction with a bank’s share of loans that are held for sale. The dependent variable now takes three values: -1 corresponds to easing standards, 0 corresponds to no change, and +1 represents tightening standards. The regressions based on SLOOS responses draw the same picture as those based on Y-14 data. The positive and statistically significant coefficient in column (4) indicates that the credit standards of banks with a higher share of loans held for sale are more sensitive to the dollar, a finding that is robust to the inclusion of macro and financial control variables and time-fixed effects (columns (5) and (6)).25 Prices and liquidity fall in the secondary market Next, we take a closer look at the secondary market for U.S. corporate loans, employing data from the Loan Syndication and Trading Association (LSTA). We start by regressing log changes in the S&P/LSTA Leverage Loan 100 Index, which reflects the performance of the largest facilities in the U.S. leveraged loan market, on changes in the dollar index at a monthly frequency.26 According to column (1) of table 8, 10.4 percent of the variation in monthly changes in the index are explained by changes in the dollar index. Column (2) adds macro and financial control variables, still delivering a significant dollar coefficient. Columns (3) and (4) replicate the analysis with individual prices of loans issued to U.S. borrowers and deliver very similar results.27 The coefficient of -0.0017 in column (2) implies that a one-standard-deviation increase in the dollar decreases the leveraged loan index by 0.4 percentage points with a beta coefficient 24To strengthen identification and control for changes in the demand for loans, we present additional regressions using Y-14 data in the robustness section. There, regressions are run at the loan level, which allowsustotestfordifferentialsensitivitiestodollarmovementsacrossbankswhilecontrollingforborrowertime fixed effects. 25For Columns (1) through (3), the average share is computed over the period 2011 to 2017. In columns (4)to(6),theaverageshareisayearlyaverageovertheperiod1992to2017. Forthisreason,thesevariables are not picked up by the bank-fixed effects in columns (4) through (6) and we need to include them directly in the regressions. 26We prefer to run regressions at this higher frequency for identification, but results on quarterly data are very similar and are reported in the appendix. All control variables included vary at a monthly frequency except the unemployment outlook, which varies at a quarterly frequency. 27Observations with only one quote are dropped from the sample. Price changes and change in bid-ask spreads in the first and 99th percentile are also excluded. 14

of 21 percent. Because we know that banks are originating fewer loans and tighten loan standards when the dollar appreciates, we can interpret the negative effect of the dollar on prices as driven by demand: Non-bank investors, the main buyers on the secondary market, buy fewer loans when the dollar appreciates. In line with the response of prices, we see liquidity in the loan market deteriorate when the dollar appreciates. Columns (5) and (6) show regressions of the average monthly spread between the daily mean bid and mean ask price at the loan level. Again, the dollar coefficient is highly statistically significant across all specifications. We conclude that demand for U.S. corporate loans on the secondary market, which is dominated by non-bank investors, falls when the dollar appreciates. This has an effect on the supply of loans by U.S. banks, which subsequently originate fewer loans and tighten credit standards.28 Effects are stronger for riskier loans As shown previously, U.S. banks not only originate fewer loans when the dollar appreciates, they also shift to safer borrowers. Consistent with this finding, the effects of the dollar on loan prices in the secondary market are stronger for riskier loans. Table 9 provides evidence for this pattern based on monthly data. In the table, the dollar is interacted with a dummy variable that takes the value of 1 if the average price of the loan is 97 percent of the par value or below, which proxies loan risk. For risky loans, the price effect of a one-standard deviation increase in the dollar is -0.7 percentage points compared with an effect of -0.04 percentage points for the less risky category. Differences in the effect of the dollar on liquidity are less pronounced and not significant when control variables or time-fixed effects are included in the regression, suggesting that the effects on liquidity are similar across loans. 3.3 The role of institutional investors Effects strengthen with participation of institutional investors The role of nonbank investors in the primary and secondary market for syndicated loans has steadily in- 28ThismechanismisconsistentwiththatdescribedinBrucheetal.(2017). Bankshaveloansinthepipeline of which they retain larger shares when investors are willing to pay less than expected. They subsequently arrange fewer loans and lend less. 15

creased over time. Table 10 shows that the effect of the dollar on secondary market prices and liquidity is linked to the share of investors that are U.S. banks (shown in figure 1 as the dashed line). In the table, changes in the dollar index are interacted with the yearly share of buyers in the secondary market that are U.S. banks. The sample period runs from March 2007 to December 2017, excluding periods during the GFC. The interaction terms are highly significant and economically relevant. According to the estimates, the correlation between the dollar and secondary market prices (bid-ask spreads) becomes negative (positive) when the share of U.S. banks drops below 26 (23) percent, which happened for the first time around 2001. We acknowledge that there is a time trend in the participation of institutional investors in the data, but the presented evidence strongly suggests that the participation of non-banks and foreign investors in the secondary market for corporate loans is the driver of the negative correlation between the dollar and demand in this market. Investors reduce positions in funds that buy U.S. bank loans Data from Irani et al. (2018) suggest that mutual funds are the second largest buyer group on the secondary market with an estimated share of about 22 percent in 2014. Using weekly EPFR data on mutual funds flows, we compile a sample of U.S. domiciled, U.S. dollar-denominated mutual funds that invest in bank loans exclusively in the United States. In December 2017, these funds had total assets under management of $146 billion. Based on this sample, we analyze the correlation between net flows into these funds and the dollar. Figure 5 presents graphical evidence, documenting a clear negative correlation between fund inflows and the broad dollar index.29 Table 11 presents regression results where the dependent variable is the ratio of monthly aggregate net inflows over assets under management for all funds in the sample. Column (1) shows the simple correlation, column (2) controls for a fund’s performance in the previous quarter, and column (3) adds the baseline control variables. When the dollar appreciates, funds specialized in U.S. bank loans experience outflows, as investors pull money out. Faced with these outflows, funds then need to reduce their holdings of U.S. corporate bank loans, which leads to a fall in the demand for these 29The graph shows total flows into all funds reported by EPFR, including funds that enter during the sample period. If we only look at funds that operate continuously throughout the sample period, the qualitiative picture is unchanged, but quantities are smaller. 16

loans on the secondary market. Effects are material. A one-standard-deviation (monthly) appreciation of the dollar implies an outflow of 0.77 percent of assets under management. To compare the implied quantities to the estimated reduction in loan issuance by U.S. banks, we estimate effects again at a quarterly frequency. A one-standard-deviation rise in the dollar then leads to outflows of around $3.2 billion.30 Given that mutual funds make up around 22 percent of the secondary market, this effect would scale to about $11.9 billion if all other non-bank buyers in the secondary market reacted similarly to the dollar.31 3.4 Quantitative aggregate implications for the U.S. economy As discussed before, a 2.5 point increase in the dollar index, reduces U.S. banks’ loan originations by 10 percent. At the end of 2017, the utilized exposures of U.S. banks’ total loan originations totaled $125 billion. A reduction in the growth of loan originations of 10 percentage points implies $12.5 billion less lending. The broad dollar index increased by more than 20 points from 2014:Q1 to 2017:Q1. We calculate that U.S. banks would have lent an additional $100 billion to U.S. firms had the dollar remained flat over this period. These quantifications likely underestimate the effect on U.S. corporate loans, primarily, because loan originations on the balance sheets of U.S. banks do not correspond to the total C&I loans originated in the quarter. As mentioned earlier, a large portion of the loans is quickly sold and may not appear on bank balance sheets. Second, Y-14 data only include banks participating in Fed-run stress tests, but there are many more banks involved in C&I loanoriginationsanddistribution. Third,tighteningofcreditstandardsinonequarterweighs on loan originations in the next quarter. Therefore, we conjecture that dollar movements haveeconomicallynotableeffectsonU.S.banks’creditsupplythroughthedescribedchannel. And, as an extensive academic literature has shown, credit supply shocks have economically significant effects on investment, employment and output.32 30At a quarterly frequency, the estimated dollar coefficient is 0.0012. 31The effect increases further when we regress flows on the emerging markets dollar index, which delivers acoefficientof-0.0211. Combinedwiththeslightlylargerstandarddeviationoftheemergingmarketsdollar indexof2.9,thiscoefficientimpliesoutflowsofabout$7billionfromthesecondarymarketfrommutualfund flows alone. 32PeekandRosengren(2000),Chodorow-Reich(2013),Bassettetal.(2014),Greenstoneetal.(2014),and Gilchrist et al. (2017). 17

To strengthen our quantitative analysis, we run two-stage least squares regressions. Specifically, we regress changes in loan originations and credit standards on changes in secondary market prices instrumented by changes in the broad dollar index. Results are shown in table 12. Using the first stage combined with the second stage coefficients (columns (2) and (3)), we obtain exactly the same effect from a one-standard deviation increase in the broad dollar index as compared to the reduced form regression (column (1) of table 2). The same holds for regressions with changes in credit standards as the dependent variable. This exercise implies that the movements in quantities and credit standards we observe can be fully explained by changes in prices in response to dollar movements. 4 Addressing Endogeneity Concerns The analysis so far has treated the dollar exchange rate as an exogenous variable but, of course, it is an equilibrium object. A myriad of factors, even though not well understood, drive the dollar, and whatever moves the dollar might also affect banks’ credit supply and non-bank financial intermediaries’ investment decisions.33 We have already addressed multiple possible third factors that could drive the presented correlations by controlling for the federal funds rate, the term spread, the VIX, the excess bond premium, and the unemploymentoutlook. Notealsothatseveralkeyresultsrelyonvariationacrossbanksandarerobust to the inclusion of time-fixed effects. Nevertheless, to address concerns further, this section presents an instrumental variable strategy where the broad dollar index is instrumented with movements of the dollar in response to foreign monetary policy announcements. This exercise excludes that U.S.-specific developments correlated with the dollar are responsible for our results. Finally, the dollar could affect credit supply through additional channels. This section also tests for these alternative channels but does not find support for their relevance. 33The difficulty to find fundamental drivers of the exchange rate has been termed the “exchange rate disconnect puzzle“. See Frankel and Rose (1995) and Froot and Rogoff (1995). 18

4.1 Instrumenting the dollar To explore whether an unobservable third factor associated with U.S. economic or financial conditions drives the correlation between U.S. credit conditions and the dollar, we take an instrumental variable approach. In our case, the instrument needs to broadly move the dollar but should not affect U.S. economic and financial conditions. Such an instrument is inherently hard to find. Foreign monetary policy changes, which affect the dollar, qualify in principle as instruments. Cieslak and Schrimpf (2019) provides the response of the dollar exchange rate from 15 minutes before to 90 minutes after monetary policy announcements in Australia, Canada, the euro area, Japan, Switzerland, and the United Kingdom. Because foreign central banks do not have private information about the U.S. economy, monetary policy decisions in these countries should not move expectations about the U.S. economy. Also, importantly, any direct effect of foreign monetary policy on U.S. interest rates should have the opposite effect on lending from our channel. When the foreign authorities loosen monetary policy more than expected, this should lower interest rates in the United States, which would ease U.S. credit conditions. However, the same policy should lead to an appreciation of the dollar, and should, through the secondary market channel, tighten U.S. credit conditions. We instrument quarterly changes in the broad dollar index with the cumulative response of the respective bilateral dollar exchange rate to announcements by five major central banks within the same time period. Data on the response of the dollar exchange rates are available from 2000 to 2017. Table 13 presents the results of the IV regression for credit standards. Column (1) of table 13 shows the OLS regression for the reduced sample. The first stage and second stage regressions are shown in columns (2) and (3), respectively. The IV coefficient associatedwiththedollarissignificantata5percentlevelandaround50percenthigherthan theOLScoefficient. WetakethisasevidencethatcreditstandardsofU.S.banksaresensitive to the component of the dollar exchange rate that is driven by non-U.S. developments. IVresultsforloanoriginationsandsecondarymarketpricesarenotsignificantatstandard levels, likely because of the limited power of the instrument for the post-GFC period. As we will show later, most of the results for the time period after the GFC are driven by 19

the emerging markets dollar index. Therefore, it is not surprising that advanced economy monetary policy shocks do not have enough power to serve as instruments for the regressions thatemployY-14andLSTAdata. Unfortunately, wecurrentlydonothavedataonexchange rate responses to monetary policy announcements in emerging markets, but we will keep working on constructing instruments for the emerging markets dollar index. 4.2 Excluding alternative explanations Monetary policy, additional risk and uncertainty proxies, and the U.S. economic outlook As argued before, the dollar is affected by economic risk and uncertainty, U.S. monetary policy and interest rates, and expectations about the performance of the U.S. economy. Weincludedseveralcontrolsinpriorregressions, butwecanaddadditionalproxies that capture these factors. As additional risk and uncertainty measures, we incorporate the 3-year treasury basis from Du et al. (2018b), the world recession probability from Cuba- Borda et al. (2018), the U.S. variance risk premium from Londono and Zhou (2017), as well as the oil price. To capture changes in interest rate expectations, we add changes in the one-year ahead forecast of the 3-month treasury yield and the forecast spread between the 10-year and 3-month treasury yield from the Survey of Professional Forecasters to the regressions. From the same data source, we obtain changes in the 4-quarter ahead forecast of real GDP, and we add the contemporaneous real annual GDP growth and the annual change in the unemployment rate. Table 27 through 32 in the appendix show that the dollar coefficient is largely robust to the inclusion of these financial and macroeconomic controls.34 Export channel and broader demand channels The fact that U.S. banks’ loan originations decline when the price for corporate loans on the secondary market falls tells us that the dollar acts as a driver of change in demand for loans on the secondary market and, through this channel, as a shifter of banks’ credit supply. That said, in principle, U.S. firms’ credit demand could also react to the dollar. In particular, firms that export might have 34The oil price is highly correlated with the dollar for some quarters after the GFC. Therefore, the dollar coefficient becomes generally less significant when the oil price is included. However, in most regressions, the point estimates are little changed. 20

a harder time selling their goods abroad when facing a stronger dollar and might therefore invest less and demand less credit. At the same time, firms that import might find it more attractive to source goods from abroad.35 The demand of firms that neither import nor export should not be affected by dollar movements in a major way. To test for demand effects associated with the dollar, we obtain data from the Bureau of Economic Analysis and the U.S. Census Bureau to calculate import and export intensities of U.S. industries at the 3-digit NAICS level for the pre-sample year 2010.36 We merge these export and import intensities with data on U.S. banks’ loan originations from Y-14 reports aggregated to the bank-industry level and interact the change in the dollar index with an industry’s export and import intensity, respectively. Table 14 presents the results. Column (1) shows the baseline regression of the change in loan originations on the dollar for the industry-bank-level data. Columns (2) and (3) include the two interactions terms, which are insignificant at standard significance levels. The dollar coefficient in turn is unaffected and becomes slightly larger as macro and financial control variables are added. The remaining columns provide sample splits. Columns (3) and (4) show results for loan originations to firms in industries that neither import nor export. Columns (5) and (6) show results for industries with positive imports or exports. Given that the dollar coefficients are the same across these two groups, we find no evidence that the dollar affects the demand for credit differentially across firms in different industries. To address concerns that all firms could change their demand for credit in response to dollar movements, we include proxies for changes in credit demand in the regressions. Employing the Y-14 data, we compute the bank-level change in the utilized exposures of pre-existing loans.37 The idea is that changes in the demand of credit from firms should be reflected in a higher share of committed exposures drawn. Results of a regression that include this variable are shown in column (2) of table 15. The coefficient on the change in 35Goods are often priced in USD. It is therefore unclear whether the import channel is really operative at a quarterly frequency (Casas et al. (2016)). 36An industry’s import and export intensity is computed as exports or imports over domestic absorption (production + imports - exports). 37Specifically, we calculate the change in utilized exposures of pre-existing credit lines at the bank level, normalizing the change by the lagged sum of committed exposures. We drop the top and bottom percentile of the obtained variable to reduce the impact of outliers. 21

exposures drawn is positive and significant at the 10 percent level, suggesting that when firms draw down more credit, banks’ loan originations also increase. The SLOOS offers another demand control. The survey asks banks about any changes in the demand for credit by large and medium enterprises. Column (4) of table 15 includes the net percentage of firms that said that demand decreased as well as its lagged value in a regression of the net percentage of firms that said that they tightened credit standards on the change in the dollar. Compared to the baseline regression presented in column (3), the dollar coefficient increases slightly. Of note, the demand controls are not significant at standard levels. Finally, we check the robustness of the results that explain banks’ sensitivities to dollar movements with their shares of loans held for sale in total loans. To this end, we exploit the Y-14 data at the loan-level to control for borrower-time fixed effects, which should absorb any factors that affect the firm-level demand for loans. Results are presented in table 16. In the regressions, we look at the extensive margin of lending at the bank-borrower-time level.38 In line with the analysis performed on bank-level data, the negative correlation between the dollar and loan originations is stronger for banks with a larger share of loans held for sale. Based on this analysis, we conclude that demand effects associated with dollar movements do not play a large role in explaining changes in credit standards and do not confound our results. Credit risk of U.S. firms The instrumental variable exercise presented in the previous subsection highlights that credit conditions respond to components of the dollar that are driven by foreign developments. To address further concerns that changes in U.S. conditions, specifically changes in the risk environment for U.S. firms, could be responsible for tightening credit standards, we analyze the credit risk in U.S. banks’ portfolios. If banks contract lending because U.S. firms have become riskier, this should be reflected in the risk ratings banks assign to the firms they lend to. The Y-14 data allow us to track loans on banks’ 38More specifically, the dependent variable takes the value of 1 if a bank originates a loan to a specific borrower in a quarter but did not originate a loan to that borrower in the previous quarter. The variable is 0iftherewasnochangeinoriginationbehavior,and-1ifabankoriginatedaloantoaborrowerlastquarter but does not originate a loan this quarter. 22

books, their risk rating, and the probability of defaults banks assign over time. Columns (1) and (2) of table 17 present regressions where the dependent variable takes the value of 1 if a firm was upgraded, zero if the firm’s rating remained unchanged, and -1 if the firm was downgraded. Asindicatedbythestatisticallyinsignificantdollarcoefficients, anappreciation of the dollar is not associated with an increase in credit risk of existing loans. Results are simliar for changes to the probability of default assigned by banks (columns (3) and (4)). Therefore, we reject the hypothesis that banks contract lending because credit risk in their U.S. loan book has gone up. It also likely means that the riskiness of the pool of domestic loans the banks could originate is unchanged.39 U.S. banks’ foreign operations While credit risk associated with U.S. borrowers is unaffected when the dollar moves, this might not be the case for foreign borrowers. One prominent story why bank lending responds to dollar movements is formulated in Bruno and Shin (2014). There, banks provide lending in dollars to foreign firms that earn revenues in theirlocalcurrency. Asaresult, thefirms’debtburdenincreaseswhenthedollarappreciates. This, in turn, raises the credit risk in banks’ books. Under a value-at-risk constraint, the balance sheet capacity of banks goes down, meaning that credit supply contracts. While this mechanism could, in principle, directly affect U.S. banks, it does not appear to be quantitatively relevant for them (Niepmann and Schmidt-Eisenlohr (2017)).40 Table 18 presents evidence for this. The dependent variable in this table is the change in U.S. banks’ domestic loan originations. It is regressed on changes in the dollar index interacted with different measures of banks’ foreign activities derived from U.S. banks’ FFIEC009 Foreign Exposure Reports. Column (1) includes the average share of a bank’s foreign claims over total assets. Foreign claims are defined as claims on foreign residents held both by the parent company as well as on the balance sheet of foreign branches and subsidiaries. Column 39This conclusion is based on the premise that banks’ reported risk ratings and probabilities of default reflectbanks’currentriskassessments. BecausetheFederalReservecloselymonitorsbanks’Y-14submissions andanalyzes theappropriatenessof theirriskmanagementandrisk models, we believethat riskratingsand probabilities of default of existing loans would reflect broad shifts in banks’ credit risk assessments. That said, Plosser and Santos (2018) raise doubts about the accuracy of banks’ reported risk ratings. 40As we discuss later, this mechanism could explain why global investors invest less in U.S. bank loans when the dollar appreciates. However, it does not appear relevant for U.S. banks themselves. 23

(2) has a bank’s share of international claims, which exclude claims held in foreign offices in local currency from total foreign claims. Column (3) uses banks’ local claims over total assets, those held by foreign offices. Column (4) employs the average share of local liabilities over total liabilities. The interaction terms between the dollar and these different measures are always insignificant. Column (5) shows the baseline regression for a sample of banks with foreign claims below 1 percent of total assets, while column (6) only includes banks without foreign offices. Even for these banks, which clearly are focused on the United States, loan originations are negatively correlated with the dollar. Under the story proposed in this paper, this finding is easy to explain: Because most U.S. banks offload a portion of their loans to other financial intermediaries on the secondary market, they face worse conditions when the dollar appreciates and reduce lending, independently of their foreign claims positions. Another way to investigate whether the foreign operations of U.S. banks might play a role in explaining their sensitivities to the dollar is to analyze the sensitivities of their cross-border loan issuance to dollar movements. U.S. banks not only lend to U.S. firms, but also to corporations in foreign countries, either from their U.S. offices or through branches and subsidiaries located abroad. Some of these loans are also syndicated and sold on the secondary market.41 Table 19 studies changes in aggregate foreign loan originations. The response of cross-border lending is essentially the same as that of U.S. lending, and there is no difference between lending to emerging and advanced economies.42 U.S. banks reduce loan originations to all borrowers symmetrically when the dollar appreciates. That said, cross-border loan originations decline more when the excess bond premium rises compared with U.S. loan originations (coefficient of -0.02 versus -0.4 and -0.5). An increase in the excess bond premium reflects a reduction in the risk-bearing capacity of the U.S. financial sector (Gilchrist and Zakrajˇsek (2012)). Therefore, in line with much of the literature on 41Cerutti et al. (2015) estimate that about one third of all cross-border lending corresponds to syndicated lending. De Haas and Van Horen (2013) estimate that two-thirds of cross-border lending to emerging markets is in the form of syndicated lending. In this context, see also Lee et al. (2017) and Niepmann and Schmidt-Eisenlohr (2017). 42We are aware that the dollar has a mechanical effect on the non-U.S. dollar denominated portion of cross-borderloanoriginations. Unfortunately,informationonthecurrencydenominationoftheloansisonly available starting from 2014:Q3. Analysis of dollar-denominated loans where available indicates the same responsetothedollarastotalcross-borderloanoriginations. Notealsothatonlyaround25percentofloans in the Y-14 data are denominated in local currency. 24

cross-border lending, we find that this lending is more volatile and contracts more when banks become more risk averse.43 U.S. banks’ funding costs As a final step, we check whether funding conditions for U.S. banks change when the dollar appreciates. Monthly data from CRANE delivers information on U.S. banks’ funding obtained from money market mutual funds (MMMF) in the form of repos, commercial papers, or certificates of deposit.44 The rates on MMMF funding reflect banks’ marginal funding costs. Analysis of the correlation between rate changes and changes in the dollar can therefore shed light on funding pressures banks might face. Table 20 shows the results based on a sample of 13 U.S. banks covering the period from February 2011 to December 2017. All regressions shown in the table include instrument-fixed effects a well as the weighted average maturity of a bank’s instruments outstanding. The significant negative dollar coefficient implies that the interest rates that U.S. banks pay for funding go down when the dollar appreciates. According to the findings in this paper, the reason might be that banks’ willingness to pay goes down when they cut lending and shift to safer borrowers as a result of a dollar appreciation.45 In any case, banks do not face higher marginal funding costs when the dollar appreciates. 5 Why Are Institutional Investors Sensitive to the Dollar? The key open question that remains is why global investors change their demand for U.S. corporate loans when the dollar moves. Because of a lack of data on these investors’ balance sheets, we are not able to delve deeper and provide a conclusive answer. However, the results presentedbelowsuggestthatinvestors’holdingsofemergingmarketdollardenominateddebt could play a role. 43See, for example, Allen (2011) and Avdjiev et al. (2012). 44We thank Inaki Aldasoro, Torsten Ehlers, and Egemen Eren for valuable support with the funding cost analysis. For their work with CRANE data on the price of dollar funding for global banks, see Aldasoro et al. (2018). 45Regressions of the volume of funds that U.S. banks borrow from MMMF on the dollar result in insignificant coefficients. It appears that the amount of funding banks obtain from MMMF does not changes with the dollar. 25

5.1 The role of the emerging markets dollar index Until now, we have employed the broad dollar index in the regressions. To explore whether a particular component of the dollar exchange rate drives the demand for U.S. corporate loans by non-bank financial investors, we use two different trade-weighted dollar indices as independent variables in table 21, one that includes dollar exchange rates vis-a-vis emerging economies (EMEs) and one vis-a-vis advanced economies (AFEs).46 Altogether the results in the table highlight the role of the emerging market dollar index for our proposed mechanism. Specifically, only the coefficient of the emerging economies dollar index is statistically significant when changes in the riskiness of new loans, secondary market liquidity of loans as a reason for tightening, and changes in the U.S. Leverage Loan 100 Index are regressed on the two dollar indices (see columns (5) through (8)). Both indices have predictive power for loan price changes in column 8. However, the point estimate is considerably larger for the emerging economies dollar index than for the advanced economies dollar index. The regressions based on SLOOS credit standards show that the emerging market dollar index explains tightening of credit standards after the GFC, while the advanced economies dollar index is statistically significant prior to 2008 (columns (3) and (4)). The result for loan originations, presented in column (1), is less clear cut. The loadings on the two dollar indices are very similar both in magnitude and in statistical significance. Table 22 presents regressions of net fund inflows from the EFPR data on the two dollar indices. Again, only the emerging market dollar index coefficient is statistically significant. In sum, the emerging market dollar appears to be the driver of the secondary market channel post-GFC. When emerging market currencies depreciate vis-a-vis the dollar, bank loan funds experienceoutflows, pricesandliquidityonthesecondarymarketfall, andU.S.bankstighten credit standards. 46TheAFElistis: Canada,Euro,Japan,Sweden,Switzerland,UnitedKingdomandAustralia. TheEME list is: Argentina, Brazil, China, Chile, Colombia, Hong Kong, India, Indonesia, Israel Korea, Malaysia, Mexico, Philippines, Saudi Arabia, Singapore, Russia, Taiwan, Thailand, and Venezuela. 26

5.2 Possible explanations and current literature As this paper highlights, the dollar is closely linked to the investment behavior of non-bank intermediaries. As has been argued before, the dollar can be understood as a “barometer of the risk-taking capacity in global capital markets” (Avdjiev et al. (forthcoming)), where investors move to safer assets when the dollar appreciates. This risk factor is not reflected in any other proxy of risk sentiment that has been used in the literature.47 In light of the instrumental variable results and the relevance of the emerging market dollar index documented in this paper, the dollar appears to be a global risk factor that is not primarily linked to developments in the United States and the risk-bearing capacity of U.S. banks, but associated with institutional investors’ risk attitudes, which respond also to foreign developments. Several explanations have been put forward to rationalize why the dollar may be a risk factor. In particular, researchers at the BIS have argued that the dollar has emerged as a risk factor because of the significant amount of emerging market corporatedebtwhichhasbeenissuedsince2009(BrunoandShin(2014), MiyajimaandShim (2014), Bruno and Shin (2017)). When the dollar appreciates, this debt becomes riskier and investors might want to reduce their exposures. Other work has highlighted how the dollar might endogenously move with changes in the risk aversion of intermediaries (Gabaix and Maggiori (2015), Jiang et al. (2018b), He et al. (2017)). Camanho et al. (2018) find evidence that portfolio rebalancing flows affect exchange rates. If institutional investors retrench from risky U.S. bank loans when the dollar appreciates, other debt instruments might also be sensitive to the dollar. Table 23 shows that this is indeed the case. Columns (1) through (3) present regressions of monthly changes in bond yields on changes in the broad dollar index. The sample consists of indices of bond yields of 5- and 10-year maturities and different risk ratings, where a higher number indicates higher risk. The sample runs from January 1992 to December 2017. The significant coefficients on the dollar and its interaction with the ratings variable in column (1) indicate that the yields of bonds with ratings below AA are positively correlated with the dollar. When the dollar 47The dollar’s role as a risk factor almost certainly stems from its dominant role in global banking and the world economy more broadly. See Gopinath and Stein (2018) for a recent contribution explaining the existence of a dominant currency. 27

appreciates, yields rise (prices fall)—the more so, the riskier the bond. For a BB bond, a 2.5 point increase in the broad dollar index increases the yield by 10 basis points. Column (2) indicates that yields are particularly correlated with the emerging markets dollar index in line with earlier results on bank loans. Column (3) confirms that the interaction term is robust to the inclusion of time-fixed effects. These results for U.S. corporate bonds mirror the results for bank loans and reaffirm that credit conditions for U.S. corporations worsen when the dollar appreciates, with stronger effects for riskier firms. 6 Conclusions Recent academic literature links the dollar to the risk aversion of financial intermediaries and global credit conditions. This relationship might be conventional wisdom for market observers who often describe the connection between the dollar and institutional investors’ asset allocations as risk-off/risk-on episodes. These episodes are typically thought to mainly concern emerging markets, whose bonds are particularly risky.48 This paper shows, however, that credit to U.S. firms also depends on the behavior of global investors and, hence, the dollar because of these investors’ increasing role for traditional financial intermediation in the United States. Because U.S. banks offload a large portion of their loans to non-bank investors, conditions on the secondary market affect U.S. banks’ credit supply, which worsen when the dollar appreciates. In line with Bruche et al. (2017), not only credit to those firms that borrow through the syndicated lending market may be affected. Because it takes time to originate loans, banks are forced to hold larger shares of risky loans on their books when secondary market conditions worsen. In response, they might reduce lending to all firms. Shocks to the credit supply can have real consequence as the GFC and numerous academic papers in its aftermath have shown. As traditional financial intermediation moves to nonbank financial institutions such as mutual funds and CLOs, new macro-financial channels are emerging. As this paper shows, foreign developments that affect the dollar have effects on credit supply to U.S. firms because of the response of institutional investors. To insure that credit supply remains stable, especially in times of turmoil, attention needs to be paid 48For example, Braeuning and Ivashina (2017b). 28

to these relatively unregulated entities. References Adrian, Tobias, Erkko Etula, and Tyler Muir, “Financial Intermediaries and the Cross-Section of Asset Returns,” The Journal of Finance, 2014, 69 (6), 2557–2596. Aldasoro, Inaki, Torsten Ehlers, and Egemen Eren, “Business models and dollar funding of global banks,” BIS Working Papers 708, Bank for International Settlements, March 2018. Allen, Franklin, Cross-border banking in Europe: implications for financial stability and macroeconomic policies, CEPR, 2011. Avdjiev, Stefan, Wenxin Du, Catherine Koch, and Hyun Song Shin, “The dollar, bankleverageandthedeviationfromcoveredinterestparity,”American Economic Review: Insights, forthcoming. , Zsolt Kuti, and El˝od Tak´ats, “The euro area crisis and cross-border bank lending to emerging markets1,” BIS Quarterly Review, 2012, p. 37. Baskaya, Yusuf Soner, Julian di Giovanni, Sebnem Kalemli-Ozcan, and Mehmet Fatih Ulu, “International spillovers and local credit cycles,” Working Paper 23149, National Bureau of Economic Research February 2017. Bassett, William F, Mary Beth Chosak, John C Driscoll, and Egon Zakrajˇsek, “Changes in bank lending standards and the macroeconomy,” Journal of Monetary Economics, 2014, 62 (C), 23–40. Bidder, Rhys, John Krainer, and Adam Shapiro, “De-leveraging or de-risking? How bankscopewithloss,”WorkingPaperSeries2017-3,FederalReserveBankofSanFrancisco 2017. Braeuning, Falk and Victoria Ivashina, “Monetary Policy and Global Banking,” Working Paper 23316, National Bureau of Economic Research April 2017. 29

and , “U. S. monetary policy and emerging market credit cycles,” Working Papers 17-9, Federal Reserve Bank of Boston August 2017. Brown, James R, Matthew Gustafson, and Ivan Ivanov, “Weathering Cash Flow Shocks,” 2017. Bruche, Max, Frederic Malherbe, and Ralf Meisenzahl, “Pipeline Risk in Leveraged Loan Syndication,” CEPR Discussion Papers 11956, C.E.P.R. Discussion Papers 2017. Bruno, Valentina and Hyun Song Shin, “Cross-border banking and global liquidity,” The Review of Economic Studies, 2014, p. rdu042. and , “Global dollar credit and carry trades: a firm-level analysis,” The Review of Financial Studies, 2017, 30 (3), 703–749. Camanho, Nelson, Harald Hau, and Helene Rey, “Global Portfolio Rebalancing and Exchange Rates,” Mimeo 2018. Casas, Camila, Federico Diez, Gita Gopinath, and Pierre-Olivier Gourinchas, “Dominant Currency Paradigm,” NBER Working Papers 22943, National Bureau of Economic Research, Inc 2016. Cerutti, Eugenio, Galina Hale, and Camelia Minoiu, “Financial crises and the composition of cross-border lending,” Journal of International Money and Finance, 2015, 52, 60 – 81. Adjustment in the Aftermath of the Global Crisis 2008-09: New Global Order? , Stijn Claessens, and Lev Ratnovski, “Global Liquidity and Drivers of Cross-Border Bank Flows,” IMF Working Papers 14/69, International Monetary Fund 2014. Chodorow-Reich, Gabriel, “The employment effects of credit market disruptions: Firmlevelevidencefromthe2008–9financialcrisis,” The Quarterly Journal of Economics, 2013, 129 (1), 1–59. Choi, Dong Boem and Hyun-Soo Choi, “The effect of monetary policy on bank wholesale funding,” Staff Reports 759, Federal Reserve Bank of New York January 2016. 30

Chui, Michael, Ingo Fender, and Vladyslav Sushko, “Risks related to EME corporate balance sheets: the role of leverage and currency mismatch,” BIS Quarterly Review, 2014. Cieslak, Anna and Andreas Schrimpf, “Non-monetary news in central Bank communication,” Journal of International Economics, 2019. Cuba-Borda, Pablo, Alexander Mechanick, and Andrea Raffo, “Monitoring the WorldEconomy: AGlobalConditionsIndex,”IFDPNotes2018-06-15,BoardofGovernors of the Federal Reserve System (U.S.) 2018. Demsetz, Rebecca S., “Bank Loan Sales: A New Look At The Motivations For Secondary Market Activity,” Journal of Financial Research, June 2000, 23 (2), 197–222. Drucker, Steven and Manju Puri, “On loan sales, loan contracting, and lending relationships,” The Review of Financial Studies, 2008, 22 (7), 2835–2872. Du, Wenxin, Alexander Tepper, and Adrien Verdelhan, “Deviations from Covered Interest Rate Parity,” NBER Working Papers 23170, National Bureau of Economic Research, Inc 2017. , , and , “Deviations from covered interest rate parity,” The Journal of Finance, 2018, 73 (3), 915–957. , Joanne Im, and Jesse Schreger, “The US Treasury Premium,” Journal of International Economics, 2018, 112, 167–181. Ferrari, Massimo, Jonathan Kearns, and Andreas Schrimpf, “Monetary policy’s rising FX impact in the era of ultra-low rates,” BIS Working Papers 626, Bank for International Settlements 2017. Frankel, Jeffrey A and Andrew K Rose, “Empirical research on nominal exchange rates,” Handbook of international economics, 1995, 3, 1689–1729. Froot, Kenneth A and Kenneth Rogoff, “Perspectives on PPP and long-run real exchange rates,” Handbook of international economics, 1995, 3, 1647–1688. 31

Gabaix, Xavier and Matteo Maggiori, “International liquidity and exchange rate dynamics,” The Quarterly Journal of Economics, 2015, 130 (3), 1369–1420. Gilchrist, Simon and Egon Zakrajˇsek, “Credit spreads and business cycle fluctuations,” American Economic Review, 2012, 102 (4), 1692–1720. , Michael Siemer, and Egon Zakrajˇsek, “The Real Effects of Credit Booms and Busts,” 2017. Gopinath, Gita and Jeremy C. Stein, “Banking, Trade, and the making of a Dominant Currency,” NBER Working Papers 24485, National Bureau of Economic Research, Inc April 2018. Greenstone, Michael, Alexandre Mas, and Hoai-Luu Nguyen, “Do Credit Market Shocks affect the Real Economy? Quasi-Experimental Evidence from the Great Recession and Normal Economic Times,” NBER Working Papers 20704, National Bureau of Economic Research, Inc 2014. Haas, Ralph De and Neeltje Van Horen, “Running for the exit? International bank lending during a financial crisis,” The Review of Financial Studies, 2013, 26 (1), 244–285. He, Zhiguo, Bryan Kelly, and Asaf Manela, “Intermediary asset pricing: New evidence from many asset classes,” Journal of Financial Economics, 2017, 126 (1), 1–35. Hirtle, Beverly, “Credit derivatives and bank credit supply,” Journal of Financial Intermediation, 2009, 18 (2), 125 – 150. Irani, Rustom M, Rajkamal Iyer, Ralf Meisenzahl, and Jose-Luis Peydro, “The Rise of Shadow Banking: Evidence from Capital Regulation,” CEPR Discussion Papers 12913, C.E.P.R. Discussion Papers 2018. Ivashina, Victoria and Zheng Sun, “Institutional demand pressure and the cost of corporate loans,” Journal of Financial Economics, 2011, 99 (3), 500 – 522. 32

Jiang, Zhengyang, Arvind Krishnamurthy, and Hanno Lustig, “Foreign Safe Asset Demand and the Dollar Exchange Rate,” NBER Working Papers 24439, National Bureau of Economic Research, Inc 2018. , , and , “Foreign Safe Asset Demand for US Treasurys and the Dollar,” in “AEA Papers and Proceedings,” Vol. 108 2018, pp. 537–41. Lee, Seung Jung, Lucy Q Liu, and Viktors Stebunovs, “Risk Taking and Interest Rates: Evidence from Decades in the Global Syndicated Loan Market,” Technical Report 2015. , Lucy Qian Liu, and Viktors Stebunovs, “Risk Taking and Interest Rates; Evidence from Decades in the Global Syndicated Loan Market,” IMF Working Papers 17/16, International Monetary Fund January 2017. Londono, Juan M. and Hao Zhou, “Variance risk premiums and the forward premium puzzle,” Journal of Financial Economics, 2017, 124 (2), 415 – 440. Malamud, Semyon and Andreas Schrimpf, “Intermediation Markups and Monetary Policy Passthrough,” Swiss Finance Institute Research Paper Series 16-75, Swiss Finance Institute 2016. Miranda-Agrippino, Silvia and Helene Rey, “World Asset Markets and the Global Financial Cycle,” CEPR Discussion Papers 10936, C.E.P.R. Discussion Papers November 2015. Miyajima, Ken and Ilhyock Shim, “Asset managers in emerging market economies,” BIS Quarterly Review, September 2014. Niepmann, Friederike and Tim Schmidt-Eisenlohr, “Foreign Currency Loans and Credit Risk: Evidence from U.S. Banks,” Technical Report 2017. Paligorova, Teodora and Joo A.C. Santos, “Monetary policy and bank risk-taking: Evidence from the corporate loan market,” Journal of Financial Intermediation, 2017, 30, 35 – 49. 33

Peek, Joe and Eric S Rosengren, “Collateral damage: Effects of the Japanese bank crisis on real activity in the United States,” American Economic Review, 2000, 90 (1), 30–45. Plosser, Matthew C and Joao A C Santos, “Banks’ Incentives and Inconsistent Risk Models,” The Review of Financial Studies, 2018, 31 (6), 2080–2112. Santos, Jo˜ao AC and Pei Shao, “Loan Ownership and Liquidity in the Secondary Loan Market,” 2018. Verdelhan, Adrien, “The share of systematic variation in bilateral exchange rates,” The Journal of Finance, 2018, 73 (1), 375–418. Wu, Jing Cynthia and Fan Dora Xia, “Measuring the macroeconomic impact of monetary policy at the zero lower bound,” Journal of Money, Credit and Banking, 2016, 48 (2-3), 253–291. Yago, Glenn and Donald McCarthy, The US leveraged loan market: A primer, Milken Institute Santa Monica, CA, 2004. Yesin, Pinar, “Foreign Currency Loans and Systemic Risk in Europe,” Working Paper, Study Center Gerzensee 13.06, Gerzensee 2013. 34

Tables Table 1: Correlation between loan originations and credit standards LogUtil. Exp. (1) (2) (3) FEasing 0.230*** (0.0728) FTight. -0.0309 (0.0970) Easing 0.156** (0.0761) Tight. -0.0998 (0.101) FW.Easing 0.219** 0.218** (0.0831) (0.0865) FW.Tight. -0.00234 -0.00146 (0.00490) (0.00510) W.Easing 0.0333*** 0.0150 (0.00424) (0.0111) W.Tight 0.00824 0.00531 (0.00613) (0.00577) FW.AllResp. -1.331 (1.519) W.AllResp. 1.611 (1.204) Constant 21.14*** 20.01*** 13.54 (0.0989) (1.885) (10.76) Observations 655 25 25 𝑅2 0.764 0.528 0.586 Note: This table explores the extent to which loan originations as reported in the Y-14 data are correlated with credit standards for C&I loans as reported in the SLOOS. In column (1), the dependent variable is the quarterly log loan originations (utilized exposures) of bank 𝑏. These are regressed on the response of bank 𝑏’s loan officer regarding credit standards of C&I loans. The variable Easing (Tight.) takes the value of 1 if the loan officer said that the bank eased (tightened) credit standards and zero otherwise. The regression shown includes both contemporaneous values as well as lagged values. Columns (2) and (3) are run at the aggregate level, hence the small number of observations. Here, the aggregate quarterly loan originations of U.S. banks in the Y-14 data are regressed on these banks’ weighted average change in credit standards in logs and their lagged values. The last column controls for the log of the total volume of C&I loans of the surveyed banks. Robust standard errors are in parentheses. *, **, and *** denote significance at the 10%, 5%, and 1% level. 35

Table 2: U.S. banks’ loan originations and the dollar DUtil. Exp. DNumberofLoans DAv. SizeofLoans (1) (2) (3) (4) (5) (6) DDollar -0.0402** -0.0454* -0.0305** -0.0401* -0.00976 -0.00526 (0.0145) (0.0223) (0.0132) (0.0212) (0.00883) (0.00641) DLogVix -0.0463 0.281 -0.328 (0.434) (0.312) (0.196) DExcessBP 0.0304 -0.206 0.236** (0.175) (0.123) (0.0877) DFedFundsR. 0.0700 0.0815 -0.0115 (0.143) (0.132) (0.0692) DTermSpread 0.130 0.126 0.00386 (0.146) (0.124) (0.0732) DUnempl. Outl. 1.396 0.279 1.117*** (0.987) (0.935) (0.356) Constant 0.0530 0.0376 0.0275 0.0365 0.0255 0.00108 (0.0469) (0.0526) (0.0391) (0.0510) (0.0260) (0.0198) Observations 25 25 25 25 25 25 𝑅2 0.181 0.294 0.163 0.250 0.039 0.462 Adjusted𝑅2 0.145 0.059 0.127 -0.000 -0.003 0.282 Note: This table documents the strong correlation between changes in 16 U.S. banks’ aggregate quarterly loan originations and the dollar. In columns (1) and (2), the log change in loan originations is regressed on the change in the broad dollar index. Columns (3) through (6) separately show the correlations of the extensive and intensive margins of lending with the dollar. In columns (3) and (4), the dependent variable is the log change in the number of new loans. In columns (5) and (6), the log change in the average size of new loans is regressed on changes in the broad dollar index. Robust standard errors are in parentheses. *, **, and *** denote significance at the 10%, 5%, and 1% level. Table 3: U.S. banks’ credit standards for C&I loans and the dollar FullSample ExcludeRecessions (1) (2) (3) (4) Sh. Tight. Sh. Tight. Sh. Tight. Sh. Tight. DDollar 0.548 -0.0763 1.379*** 0.918** (0.419) (0.454) (0.420) (0.398) Sh. Tight(prev.) 0.875*** 0.883*** 0.842*** 0.914*** (0.0563) (0.0628) (0.0833) (0.0814) DLogVix 12.09* 15.70*** (6.496) (5.451) DExcessBP 5.992 2.966 (4.341) (3.512) DFedFundsR. 2.366 3.051 (2.914) (2.627) DTermSpread 0.533 -1.559 (2.199) (2.041) DUnempl. Outl. -3.704 9.052 (7.863) (9.504) Constant -0.140 0.106 -1.132 -1.046 (0.797) (0.785) (0.864) (0.960) Observations 111 111 92 92 𝑅2 0.813 0.841 0.716 0.787 Adjusted𝑅2 0.810 0.830 0.709 0.769 Note: Thistabledocumentsthestrongcorrelationbetweencreditstandardsandthedollar. Ineachcolumn, the net percentage of banks that reported tightening credit standards is regressed on its lagged value and thechangeinthebroaddollarindex. Columns(1)and(2)presentresultsforthefullsample. Incolumns(3) and (4), the underlying sample excludes quarters during NBER recessions and the GFC. Robust standard errors are in parentheses. *, **, and *** denote significance at the 10%, 5%, and 1% level. 36

Table 4: The riskiness of new loans and the dollar (1) (2) DPD DPD DDollar -0.000290*** -0.000357*** (0.0000947) (0.000111) DLogVix 0.00113 (0.00156) DExcessBP -0.00107 (0.00123) DFedFundsR. 0.00121** (0.000532) DTermSpread 0.00125* (0.000661) DUnempl. Outl. -0.00108 (0.00404) Constant 0.000201 0.000239 (0.000293) (0.000281) Observations 25 25 𝑅2 0.228 0.308 Adjusted𝑅2 0.195 0.077 Note: ThistableshowsthatU.S.bankslendtosaferborrowerswhenthedollarappreciates. Inbothcolumns, thechangeintheweightedaverageprobabilityofdefaultofnewloansisregressedonthechangeinthebroad dollar index. Column (1) shows the baseline results. Column (2) includes the standard set of macro and financial control variables. Robust standard errors are in parentheses. *, **, and *** denote significance at the 10%, 5%, and 1% level. Table 5: The reasons for easing and tightening credit standards and the dollar Tightening (1) (2) (3) (4) (5) (6) (7) (8) Capital Competit. Legal Liqui. Econ. Outl. Risk Sec. Liqu. Spec. DDollar 0.00117 -0.00269 -0.00649 -0.00189 0.00270 0.00298 -0.00295 0.0159* (0.00526) (0.00551) (0.0112) (0.00720) (0.00846) (0.00866) (0.00970) (0.00829) Observations 81 80 31 35 88 80 66 84 𝑅2 0.001 0.003 0.009 0.002 0.001 0.001 0.001 0.036 Easing DDollar 0.00957 0.00460 0.00315 -0.00823 -0.0106 -0.00423 -0.0151*** 0.00705 (0.00878) (0.00497) (0.00449) (0.00565) (0.0114) (0.00550) (0.00463) (0.0104) Observations 90 82 31 35 90 80 66 91 𝑅2 0.010 0.009 0.017 0.070 0.009 0.005 0.105 0.004 Note: This table provides evidence that the reason why credit standards for C&I loans set by U.S. banks tighten when the dollar appreciates is related to the liquidity of these loans on the secondary market. In each column, the share of banks that mentioned the reason displayed in the title of the column as a reason for easing or tightening in the SLOOS is regressed on the change in the broad dollar index. For example, column (1) shows the coefficient obtained from a regression of the share of banks that said in quarter 𝑡 that theytightenedcreditstandardsbecauseofchangesinthebank’scapitalpositiononthechangeinthebroad dollarindexinquarter𝑡. Thevariousreasonsthatbanksarefreetomentionareasfollows: (i)changesinthe bank’s capital position, (ii) changes in competition from other lenders, (iii) legislative changes, supervisory actions,orchangesinaccountingstandards,(iv)changesinthebank’scurrentorexpectedliquidityposition, (v)changesintheeconomicoutlook,(vi)changesinthebank’srisktolerance,and(vii)changesinliquidityin thesecondarymarketfortheseloans,(viii)industry-specificproblems. Someofthesereasonswereintroduced over time. This explains the varying number of observations across regressions. Robust standard errors are in parentheses. *, **, and *** denote significance at the 10%, 5%, and 1% level. 37

Table 6: The reason for easing in response to dollar movements (1) (2) ShareEasing ShareEasing DDollar -0.0151*** -0.0118** (0.00463) (0.00500) DLogVix 0.0550 (0.0635) DExcessBP -0.155*** (0.0476) DFedFundsR. 0.0484 (0.0415) DTermSpread 0.0452 (0.0339) DUnempl. Outl. 0.168 (0.106) Constant 0.196*** 0.197*** (0.0127) (0.0119) Observations 66 66 𝑅2 0.105 0.250 Adjusted𝑅2 0.091 0.174 Note: This table provides evidence that the reason why credit standards for C&I loans set by U.S. banks changewiththedollarisrelatedtotheliquidityoftheseloansonthesecondarymarket. Heretheregression from column (7) of the lower panel of table 5 is repeated in column (1). Column (2) includes the baseline macro and financial control variables. Robust standard errors are in parentheses. *, **, and *** denote significance at the 10%, 5%, and 1% level. 38

Table 7: Differences in banks’ sensitivities to the dollar and loan securitization DUtil. Exp. NetTight. (1) (2) (3) (4) (5) (6) DDollar -0.0200** -0.157* 0.00340 0.0106 (0.00936) (0.0825) (0.00448) (0.0225) DDollarXAv. HFSShare -0.852** -1.237** -1.294*** 0.327*** 0.374*** 0.399*** (0.367) (0.518) (0.483) (0.126) (0.140) (0.138) DDollarXAv. T1Ratio 0.0112 0.0113* -0.000599 -0.00306 (0.00704) (0.00661) (0.00184) (0.00236) DDollarXAv. Wholes. Fund. Share -0.00315 -0.00360 -0.00302 -0.00140 (0.0119) (0.0106) (0.00282) (0.00277) NetTightening(prev) 0.342*** 0.340*** 0.244*** (0.0345) (0.0346) (0.0357) Av. HFSShare 0.964** 1.097** 0.680 (0.482) (0.513) (0.504) Av. Wholes. Fund. Share -0.0148 -0.00889 (0.0117) (0.0106) Av. T1Ratio 0.00274 0.00320 (0.00559) (0.00872) TimeFE No No Yes No No Yes BankFE Yes Yes Yes Yes Yes Yes Observations 679 679 679 1352 1352 1352 𝑅2 0.048 0.054 0.241 0.200 0.202 0.288 Adjusted𝑅2 -0.000 0.003 0.170 0.181 0.181 0.237 Note: Thistabletestsfordifferencesinthesensitivityofloanoriginationstodollarmovementsacrossbanks. Incolumns(1)through(3),thedependentvariableisthelogchangeinloanoriginations(utilizedexposures) of bank 𝑏 in quarter 𝑡. Explanatory variables are the change in the broad dollar index and its interaction with three bank characteristics: a bank’s average share of loans held for sale in its total loans, a bank’s average Tier1 capital ratio, and its average wholesale funding share. For columns (1) through (3), averages were computed over quarterly values from 2011 to 2017. In columns (4) through (6), the dependent variable is the response of bank 𝑏’s loan officer in the SLOOS to the question whether C&I lending standards of the bank eased or tightened. +1 corresponds to tightening credit standards, 0 means no change, and -1 reflects easing standards. Regressions in these columns also include last quarter’s response, the change in the broad dollar index, and its interactions with bank characteristics. For the regressions with SLOOS data, bank characteristics werecomputed asyearly averages from1992to 2017. Columns(3)and (6)includetime-fixed effects. Robust standard errors are in parentheses. *, **, and *** denote significance at the 10%, 5%, and 1% level. 39

Table 8: The secondary loan market and the dollar Agg. Price Price Bid-Ask (1) (2) (3) (4) (5) (6) DDollar -0.00252*** -0.00165*** -0.00140*** -0.00124*** 0.0126*** 0.00700*** (0.000533) (0.000441) (0.000280) (0.000261) (0.00306) (0.00260) DLogVix -0.0248*** -0.00550* 0.139*** (0.00686) (0.00291) (0.0286) DExcessBP -0.00186 -0.00102 0.0129 (0.00333) (0.00156) (0.0117) DFedFundsR. -0.00471 0.00105 -0.00383 (0.00620) (0.00241) (0.0248) DTermSpread 0.00798** 0.00383** -0.0169 (0.00339) (0.00163) (0.0154) DUnempl. Outl. -0.0203*** -0.0286*** 0.127*** (0.00650) (0.00395) (0.0315) Constant 0.00456*** 0.00492*** 0.000481 0.000435 -0.00284 -0.00269 (0.000714) (0.000667) (0.000369) (0.000316) (0.00341) (0.00283) Observations 168 168 224256 224256 181774 181774 𝑅2 0.104 0.387 0.009 0.032 0.008 0.033 Adjusted𝑅2 0.098 0.365 0.009 0.032 0.008 0.033 Note: This table shows that an appreciation of the dollar is associated with lower prices and worse liquidity conditions for loans sold on the secondary market. Data on secondary market prices and bid-ask spreads come from the Loan Syndication and Trading Association (LSTA). Columns (1) and (2) use as dependent variable the monthly change in the U.S. Leveraged Loan 100 Index. Columns (3) through (6) are based on a loan-level sample. In columns (3) and (4), the change in the monthly average price of a loan traded on the secondary market is regressed on the change in the monthly broad dollar index. Columns (5) and (6) display regressions of the change in the monthly average bid-ask spread of a loan. Columns (1) and (2) have robust standard errors in parentheses. In columns (3) through (6), standard errors are clustered by month. *, **, and *** denote significance at the 10%, 5%, and 1% level. 40

Table 9: The greater sensitivity of riskier loans to the dollar DPrice DSpread (1) (2) (3) (4) (5) (6) DDollar -0.000521*** -0.000458*** 0.00922*** 0.00575** (0.000154) (0.000153) (0.00305) (0.00254) RiskyXDDollar -0.00242*** -0.00224*** -0.00204*** 0.00303 0.000477 0.000533 (0.000529) (0.000544) (0.000441) (0.00245) (0.00269) (0.00166) Risky 0.000391 -0.0000447 -0.000920** -0.00429 -0.00319 -0.000726 (0.000687) (0.000580) (0.000463) (0.00276) (0.00234) (0.00181) c1 0 0 0 0 (.) (.) (.) (.) DLogVix -0.00496 0.118*** (0.00311) (0.0239) DExcessBP -0.00103 0.0101 (0.00155) (0.00967) DFedFundsR. 0.000991 -0.00514 (0.00238) (0.0200) DTermSpread 0.00375** -0.0147 (0.00162) (0.0125) DUnempl. Outl. -0.0285*** 0.112*** (0.00389) (0.0254) Constant 0.000257 0.000379** -0.000495 -0.000954 (0.000172) (0.000176) (0.00305) (0.00257) TimeFE No No Yes No No Yes Observations 224256 224256 224256 223620 223620 223620 𝑅2 0.016 0.037 0.098 0.007 0.028 0.078 Adjusted𝑅2 0.016 0.037 0.097 0.007 0.028 0.078 Note: This table shows that the correlation between secondary market loan prices (liquidity) and the dollar are stronger for riskier loans. The regressions are parallel to those of changes in monthly loan prices and bid-askspreadsincolumns(3)through(6)oftable8butnowincludeaninteractiontermbetweenthechange in the dollar index and a dummy variable that takes the value of 1 if the average price of a loan is at 97 percentoftheparvalueorbelowandzerootherwise. Standarderrorsclusteredbymonthareinparentheses. *, **, and *** denote significance at the 10%, 5%, and 1% level. 41

Table 10: The role of non-bank investors DPrice DSpread (1) (2) (3) (4) DDollar -0.00437*** -0.00362*** 0.0392*** 0.0265*** (0.00114) (0.00111) (0.00791) (0.00788) Sh. USbanksXDDollar 0.0173*** 0.0126** -0.176*** -0.118*** (0.00593) (0.00584) (0.0413) (0.0427) Sh. USbanks -0.0296*** -0.0268*** 0.126*** 0.106*** (0.00703) (0.00562) (0.0457) (0.0399) DLogVix -0.00437 0.115*** (0.00367) (0.0296) DExcessBP -0.000284 0.00538 (0.00169) (0.0112) DFedFundsR. 0.00422* -0.0134 (0.00224) (0.0215) DTermSpread 0.00427** -0.0132 (0.00166) (0.0131) DUnempl. Outl. -0.0263*** 0.0982*** (0.00344) (0.0222) Constant 0.00505*** 0.00460*** -0.0194** -0.0168** (0.00135) (0.00104) (0.00839) (0.00709) Observations 177000 177000 176447 176447 𝑅2 0.025 0.047 0.017 0.036 Adjusted𝑅2 0.025 0.047 0.017 0.036 Note: This tables shows that the correlations between secondary market loan loan prices, liquidity and the dollarhavebecomestrongerastheshareofU.S.banksasbuyersofloansonthesecondarymarkethasfallen over time. The regressions are parallel to those of changes in monthly loan prices and bid-ask spreads in columns (3) through (6) of table 8, but now include an interaction term between the change in the dollar index and the yearly share of syndicated loans that are bought by U.S. banks from Irani et al. (2018). Standard errors clustered by month and are in parentheses. *, **, and *** denote significance at the 10%, 5%, and 1% level. Table 11: Net flows into U.S. mutual funds specialized in U.S. bank loans (1) (2) (3) NetFlows NetFlows NetFlows DDollar -0.00709*** -0.00546** -0.00529** (0.00232) (0.00224) (0.00240) LagAv. Perf. 0.00645** 0.00191 (0.00285) (0.00292) DLogVix 0.00564 (0.0226) DExcessBP 0.0268* (0.0147) DFedFundsR. -0.0479*** (0.0156) DTermSpread 0.101*** (0.0173) DUnempl. Outl. -0.0459 (0.0296) Constant 0.0151*** 0.0118*** 0.0160*** (0.00332) (0.00341) (0.00306) Observations 106 106 106 𝑅2 0.064 0.099 0.398 Note: This table shows that U.S. domiciled, U.S. dollar denominated mutual funds that are specialized in investing in bank loans in the United States experience net outflows when the dollar appreciates. In column (1), net inflows are regressed on the change in the broad dollar index. Column (2) controls for funds’ lagged performance. Column(3)includesthebaselinemacroandfinancialcontrolvariables. Robuststandarderrors are in parentheses. *, **, and *** denote significance at the 10%, 5%, and 1% level. 42

Table 12: Instrumenting secondary market prices with the dollar DUtil. Exp. Sh. Tight. (1) (2) (3) (4) (5) (6) OLS First IV OLS First IV DPrice 2.001 14.13* -147.1*** -332.2** (3.498) (7.243) (38.18) (134.3) DDollar -0.00285** -0.00346*** (0.00128) (0.00128) Sh. Tight(prev.) 0.758*** 0.825*** (0.0587) (0.111) Constant -0.00842 0.0139*** -0.144 -0.0761 0.0143*** 2.918 (0.0547) (0.00315) (0.0935) (0.986) (0.00257) (2.150) Observations 25 25 25 56 56 56 𝑅2 0.017 0.214 . 0.788 0.146 0.685 Adjusted𝑅2 -0.026 0.180 . 0.780 0.130 0.673 Note: This table presents instrumental variable regressions where change in the U.S. Leverage Loan 100 Index are instrumented with changes in the broad dollar index. Columns (1) to (3) show regressions based onloanoriginationsdata. Columns(4)to(6)presentthecorrespondingregressionsfordataonbanks’credit standards. Columns (1) and (4) display simple OLS regressions for the IV samples. Columns (2) and (5) showregressionsofthechangeintheloanpriceindexonthechangeinthebroaddollarindex,thefirststage regressions. Columns (3) and (6) give the second-stage results, displaying the dollar coefficients estimated via IV. Robust standard errors are in parentheses. *, **, and *** denote significance at the 10%, 5%, and 1% level. Table 13: IV regressions for SLOOS credit standards Sh. Tight. (1) (2) (3) OLS First IV DDollar 1.540*** 2.265** (0.414) (1.130) U.S.Mon. Surpr. -0.000787 -0.00307 (0.00383) (0.00857) EMUMon. Surpr. -0.00577** (0.00225) JapanMon. Surpr. -0.00285 (0.00388) UKMon. Surpr. -0.00180 (0.00413) CanadaMon. Surpr. -0.00173 (0.00389) Austr. Mon. Surpr. 0.00592 (0.00421) Sh. Tight(prev.) 0.889*** 0.00821 0.886*** (0.0843) (0.0164) (0.0799) Constant -0.495 -0.179 -0.502 (0.909) (0.273) (0.907) Observations 64 64 64 𝑅2 0.821 0.188 0.811 Note: This table presents the IV results for the relationship between credit standards for C&I loans and the dollar where the broad dollar index is instrumented with the response of dollar exchange rates to monetary policy announcements in five advanced economies. Columns (1) shows the baseline OLS regression of the quarterly net percentage of banks that said that they tightened credit standards on the change in the broad dollarindex. Column(2)presentsthefirst-stageregression, wherethebroaddollarindexisregressedonthe responseofthedollarexchangeratevis-a-vistheCanadiandollar,theEuro,theBritishPound,theYen,and the Australian dollar to monetary policy announcements in the respective advanced economy taken from Ferrari et al. (2017) and Cieslak and Schrimpf (2019). These responses, computed from 15 minutes before to 90 minutes after monetary policy announcements, were summed within a quarter and cover the period from 2000 to 2017. Column (3) displays the second-stage results. As an additional control variable, the IV regressions includes the response of the USD-EUR exchange rate to U.S. monetary policy announcements from the same sources. Robust standard errors are in parentheses. *, **, and *** denote significance at the 10%, 5%, and 1% level. 43

Table 14: Regressions for sectors with different trade intensities Baseline Int. terms NoTrade ExporImp (1) (2) (3) (4) (5) (6) (7) DDollar -0.0387** -0.0375*** -0.0532*** -0.0392** -0.0579** -0.0377*** -0.0472** (0.0151) (0.00621) (0.00750) (0.0176) (0.0211) (0.0134) (0.0212) DUSDXExp. Int. 0.0178 0.0174 (0.0769) (0.0784) DUSDXImp. Int. -0.0249 -0.0258 (0.0576) (0.0584) Exp. Int. -0.0909 -0.0872 (0.166) (0.165) Imp. Int. 0.0305 0.0328 (0.111) (0.111) DLogVix 0.235** 0.308 0.0825 (0.102) (0.297) (0.329) DExcessBP -0.0384 -0.0109 -0.0934 (0.0631) (0.181) (0.185) DFedFundsR. 0.105** 0.103 0.109 (0.0453) (0.161) (0.141) DTermSpread 0.223*** 0.252* 0.166 (0.0523) (0.146) (0.128) DUnempl. Outl. 1.722*** 1.719** 1.725*** (0.237) (0.818) (0.584) Constant 0.0392 0.0419*** 0.0368** 0.0440 0.0445 0.0294 0.0145 (0.0450) (0.0153) (0.0164) (0.0458) (0.0505) (0.0454) (0.0449) Observations 27460 27460 27460 18433 18433 9027 9027 𝑅2 0.004 0.004 0.009 0.004 0.010 0.003 0.007 Note: Thistabletestsfordifferencesinthesensitivityofbanks’loanoriginationstothedollaracrosssectors. Column (1) shows the baseline regression of the log change in loan originations on the change in the broad dollar index for Y-14 data aggregated at the bank-industry level. Loans are grouped by industry according tothe3-digitNAICSindustrytowhichtheborrowingfirmbelongs. Columns(2)and(3)includeinteraction terms between the change in the broad dollar index and the export and import intensity of an industry. Columns (4) and (5) are based on a sample of industries that neither export nor import. The sample underlying results in columns (6) and (7) only include industries that export or import. Standard errors clustered by industry × quarter are in parentheses. *, **, and *** denote significance at the 10%, 5%, and 1% level. 44

Table 15: Controlling for the demand for loans DUtil. Exp. Sh. Tight. (1) (2) (3) (4) DDollar -0.0487** -0.0477** 0.918** 1.130*** (0.0222) (0.0221) (0.398) (0.382) DShareofComm. Expos. Drawn 6.292* (3.518) Sh. Tight(prev.) 0.914*** 0.798*** (0.0814) (0.111) Sh. LessDemand -0.0375 (0.0968) Sh. LessDemand(prev.) -0.101 (0.0806) DLogVix 0.0340 0.0205 15.70*** 12.65* (0.339) (0.337) (5.451) (6.659) DExcessBP -0.00621 -0.00135 2.966 3.513 (0.168) (0.170) (3.512) (3.645) DFedFundsR. 0.113 0.118 3.051 4.123 (0.157) (0.159) (2.627) (2.500) DTermSpread 0.307** 0.305** -1.559 -1.892 (0.145) (0.146) (2.041) (2.085) DUnempl. Outl. 1.334 1.325 9.052 12.70* (0.834) (0.813) (9.504) (7.246) Constant 0.0302 0.0145 -1.046 -0.825 (0.0455) (0.0423) (0.960) (0.804) Observations 605 605 92 92 𝑅2 0.080 0.085 0.787 0.802 Note: This table shows that results are robust to controlling for the demand for loans. Column (1) presents the baseline Y-14 regression run on a sample that varies by bank and quarter. Column (2) includes the change in the exposures drawn by bank and quarter as a demand control. Columns (3) and (4) are based on the SLOOS data. Column (3) displays the baseline regression identical to column (4) of table 3, while column (4) includes the net percentage of banks that said that credit demand increased as well as its lagged value in the regression. Robust standard errors are in parentheses. *, **, and *** denote significance at the 10%, 5%, and 1% level. 45

Table 16: Differences in banks’ sensitivities to the dollar, obligor-level regressions DLoanDummy (1) (2) (3) DDollar -0.0000788* -0.000329** (0.0000415) (0.000155) DDollarXAv. HFSShare -0.00174* -0.00174* (0.000904) (0.000945) DDollarXAv. T1Ratio 0.0000200 0.0000200 (0.0000121) (0.0000130) DDollarXAv. Wholes. Fund. Share 0.0000251 0.0000251 (0.0000220) (0.0000248) Bank-BorrowerFE Yes Yes Yes Borrower-TimeFE No No Yes Observations 57285325 57285325 57285325 𝑅2 0.002 0.002 0.079 Note: This table tests for differences in the sensitivity ofloan originations to dollarmovements across banks at the obligor-bank time level. The dependent variable takes the value of 1 if a bank originates a loan to a borrower in a quarter but did not originate a loan to the same borrower in the previous quarter. It takes the value -1 if the bank issued a loan last quarter to a borrower but does not originate a new loan this quarter. It takes the value 0 if a bank continues to originate loans (to not originate a loan) to a borrower in a quarter. Explanatory variables are the change in the broad dollar index and its interaction with three bank characteristics: a bank’s average share of loans held for sale in its total loans, a bank’s average Tier1 capital ratio, and its average wholesale funding share. The averages were computed over quarterly values from 2011 to 2017. Standard errors are clustered by time in column (1) and by bank-time in columns (2) and (3) and are shown in parentheses. *, **, and *** denote significance at the 10%, 5%, and 1% level. Table 17: Risk ratings of existing loans RatingImpr. PDLower. (1) (2) (3) (4) DDollar -0.000415 0.000616 0.0252 0.0467 (0.000596) (0.000876) (0.0216) (0.0292) DLogVix -0.00131 -0.172 (0.0289) (0.427) DExcessBP -0.0128 0.443 (0.00910) (0.299) DFedFundsR. -0.0148 -0.706 (0.00884) (0.431) DTermSpread -0.0130 -0.354 (0.0112) (0.244) DUnempl. Outl. 0.0437 1.252 (0.0716) (1.358) Constant -0.00558 -0.00601** 0.157*** 0.293** (0.00346) (0.00242) (0.0400) (0.107) Observations 2130057 2130057 1668636 1668636 𝑅2 0.000 0.001 0.019 0.081 Note: This table shows that the risk ratings of existing loans on banks’ books are not correlated with the dollar. Incolumns (1)and (2), thedependent variable isa categorialvariablethat takes thevalueof 1if the borrower associated with a loan was upgraded, 0 if the rating remained unchanged, and -1 if the borrower was downgraded compared to the previous quarter based on the ratings that the banks report in the Y-14 data. For columns (3) and (4), the dependent variable is very similar to the one in the other columns but loans are now classified according to whether the probability of default of the borrower as reported by the bankchanged. Byconstruction,thesampleexcludesanyloansthatwerenewlyissuedinaquarter. Standard errors clustered by quarter are in parentheses. *, **, and *** denote significance at the 10%, 5%, and 1% level. 46

Table 18: Banks’ foreign activities do not matter (1) (2) (3) (4) (5) (6) Forcl Intcl Locl Localliab <1pcforcl Nolocalcl DDollar -0.0400*** -0.0324*** -0.0455*** -0.0467*** -0.0357* -0.0493** (0.00805) (0.00853) (0.00752) (0.00743) (0.0180) (0.0198) Av. For. Cl. Share 0.0381 (0.150) DDollarXAv. For. Cl. Share 0.0347 (0.0634) Av. Int. Cl. Share 0.100 (0.205) DDollarXAv. Int. Cl. Share -0.0412 (0.0922) AvLocalShare 0.0237 (0.313) DDollarXAv. LocalShare 0.199 (0.124) AvLoc. Curr. Share 0.00799 (0.555) DDollarXAv. Loc. Curr. Share 0.326* (0.172) Constant 0.0419** 0.0382* 0.0457** 0.0462** 0.0563 0.0668 (0.0185) (0.0199) (0.0180) (0.0187) (0.0468) (0.0520) Observations 628 628 628 628 101 97 𝑅2 0.034 0.033 0.039 0.043 0.087 0.140 Adjusted𝑅2 0.029 0.028 0.035 0.038 0.077 0.131 Note: This table investigates whether differences in banks’ foreign operations predict the sensitivity of a bank’s loan originations to the dollar. The regression in column (1), run at the bank level, includes an interaction term between the change in the broad dollar index and a bank’s average share of foreign claims intotalassetsovertheyears2011to2017. Incolumn(2),theinteractiontermwiththedollarisconstructed using a bank’s average international claims (foreign claims minus claims in foreign offices in local currency). In column (3), the interaction is computed using a bank’s average local claims, those held in foreign offices. In column (4), the interaction term is based on a bank’s average local liabilities. The regression displayed in column (5) is based on a sample of banks with a share of foreign claims of less than 1 percent of total assets. The sample that underlies column (6), only includes banks with zero claims held in foreign offices. Robust standard errors are in parentheses. *, **, and *** denote significance at the 10%, 5%, and 1% level. 47

Table 19: Cross-border loan originations and the dollar Dom. AFE EME (1) (2) (3) (4) (5) (6) Dom. Dom. AFE AFE EME EME DDollar -0.0402** -0.0454* -0.0403* -0.0358 -0.0337 -0.0423 (0.0145) (0.0223) (0.0201) (0.0223) (0.0328) (0.0315) DLogVix -0.0463 -0.0206 0.280 (0.434) (0.468) (0.277) DExcessBP 0.0304 -0.395* -0.429* (0.175) (0.207) (0.230) DFedFundsR. 0.0700 0.0553 0.224 (0.143) (0.180) (0.145) DTermSpread 0.130 0.205 0.183 (0.146) (0.204) (0.161) DUnempl. Outl. 1.396 1.268 -0.678 (0.987) (1.079) (0.557) Constant 0.0530 0.0376 0.0524 0.0299 0.0176 0.0239 (0.0469) (0.0526) (0.0538) (0.0586) (0.0250) (0.0405) Observations 25 25 25 25 25 25 𝑅2 0.181 0.294 0.126 0.275 0.110 0.286 Adjusted𝑅2 0.145 0.059 0.088 0.034 0.071 0.048 Note: ThistablecomparesthesensitivitiesofnewloansissuedtoU.S.borrowersversusborrowersinadvanced foreigneconomies(AFE)andemergingmarkets(EME).Columns(1)and(2)repeatthebaselineresultsfrom columns(1)and(2)oftable2,whichdocumentthecorrelationbetweenU.S.banks’domesticloanoriginations and the dollar, to facilitate the comparison. Columns (3) and (4) include new loans to borrowers residing in advanced foreign economies. Columns (5) and (6) are for loans to borrowers in emerging markets. Robust standard errors are in parentheses. *, **, and *** denote significance at the 10%, 5%, and 1% level. 48

Table 20: U.S. banks’ rates paid for money market mutual funds funding (1) (2) (3) DDollar -6.038* -6.528** -5.006** (2.827) (2.888) (1.809) WAM -0.0428 -0.0424 0.00113 (0.0310) (0.0321) (0.0285) DlogVIX -0.202 -0.755 0.640 (0.642) (0.680) (0.799) DExcessBP 2.507 -3.892 7.826 (16.42) (15.76) (7.451) DFedFundsR. 36.03** 35.21** -3.466 (15.72) (15.03) (12.76) DTermSpread -14.95 -14.03 -10.22 (18.67) (17.90) (14.92) DWorldRec. Prob 36.31* (19.61) DVRPUS 0.0928 (0.0696) D4qUnempl. Rate 15.47** (6.688) DUnempl. Outlook 172.2*** (22.91) Observations 2,716 2,716 2,716 R-squared 0.035 0.040 0.266 InstrumenttypeFE YES YES YES NoBanks 13 13 13 Robuststandarderrorsinparentheses ***p<0.01,**p<0.05,*p<0.1 Note: This table shows that marginal funding costs of U.S. banks decline when the dollar appreciates. It presents regressions of the change in the interest rate in basis points paid by 13 U.S. banks on Treasury Repos, Government Agency Repos, Other Repos, Certificates of Deposits, Commercial Paper, and Asset- Backed Commercial Paper on the change in the broad dollar index and control variables. All regressions include instrument-type fixed effects. Standard errors clustered by time are in parentheses. *, **, and *** denote significance at the 10%, 5%, and 1% level. 49

Table 21: The emerging and advanced economies dollar indices (1) (2) (3) (4) (5) (6) (7) (8) (9) Y-14 SLOOS SL.Pre-2008 SL.Post-2008 Risk SL.Reason PriceIndex Price Bid-AskSpr. DDollarAFE -0.0296 0.853** 1.330** -0.707 -0.00000102 -0.00317 -0.000425 -0.000379*** 0.000805 (0.0268) (0.407) (0.528) (0.440) (0.000188) (0.00576) (0.000394) (0.0000330) (0.00185) DDollarEME -0.0162 0.0786 -0.0871 1.452** -0.000364** -0.0112 -0.00153*** -0.00108*** 0.00658** (0.0208) (0.281) (0.327) (0.541) (0.000141) (0.00672) (0.000533) (0.0000361) (0.00260) DLogVix -0.0720 16.49*** 23.08* 9.148* 0.00219 0.0851 -0.0222*** -0.00381*** 0.106*** (0.424) (5.389) (11.98) (4.639) (0.00181) (0.0709) (0.00748) (0.000290) (0.0247) DExcessBP 0.0220 2.734 -1.066 -0.0285 -0.000967 -0.160*** -0.00157 -0.000876*** 0.00911 (0.178) (3.676) (6.722) (3.786) (0.00172) (0.0465) (0.00331) (0.000191) (0.00955) DFedFundsR. 0.0677 3.023 6.989 -1.015 0.00131** 0.0436 -0.00450 0.00138*** -0.00632 (0.148) (2.650) (4.389) (3.068) (0.000578) (0.0418) (0.00618) (0.000260) (0.0196) DTermSpread 0.127 -1.887 -3.776 -0.261 0.00144** 0.0477 0.00833** 0.00407*** -0.0166 (0.158) (2.078) (2.963) (2.795) (0.000620) (0.0362) (0.00341) (0.000204) (0.0121) DUnempl. Outl. 1.364 9.520 -0.386 14.34* 0.000758 0.177* -0.0204*** -0.0287*** 0.115*** (1.063) (9.647) (16.19) (7.558) (0.00387) (0.101) (0.00634) (0.000737) (0.0239) Sh. Tight(prev.) 0.920*** 0.952*** 0.770*** (0.0804) (0.0860) (0.157) Constant 0.0349 -0.738 -0.460 -2.380 0.000328 0.201*** 0.00507*** 0.000495*** -0.00260 (0.0532) (0.895) (1.224) (1.500) (0.000272) (0.0128) (0.000676) (0.0000367) (0.00224) Observations 25 92 58 34 25 66 168 224270 223627 𝑅2 0.293 0.789 0.810 0.831 0.359 0.263 0.396 0.033 0.029 Note: This table shows that the emerging markets dollar index is the main driver of our results, especially post-GFC.Thetablerepeatsallbaselineregressions,butregressionsnowincludethechangesintwodifferent dollar indices, one index summarizing bilateral dollar exchange rates vis-a-vis emerging markets (EME), the other one vis-a-vis advanced foreign economies (AFE). The following dependent variables are in each of the respective columns: (1) log change in U.S. banks’ aggregate loan originations; (2) net percentage of U.S. banks that reported tightening credit standards for C&I loans; (3) same as in (2) but sample includes quarters only pre-GFC; (4) same as in (2) but sample includes only quarters post GFC; (5) change in the weightedaverageprobabilityofdefaultofnewloans;(6)theshareofbanksthatmentionedsecondarymarket liquidity of loans as a reason for easing credit standards; (7) monthly change in the U.S. Leverage Loan 100 Index; (8) monthly change in the price of a loan sold on the secondary market; (9) monthly change in the bid-ask spread of a loan sold on the secondary market. Standard errors are in parentheses. Clustering is as in the baseline specifications. *, **, and *** denote significance at the 10%, 5%, and 1% level. 50

Table 22: Net flows into U.S. bank loan mutual funds, EME vs. AFE dollar index (1) (2) (3) NetFlows NetFlows NetFlows DDollarEME -0.00759*** -0.00667** -0.00656*** (0.00276) (0.00265) (0.00221) DDollarAFE 0.00146 0.00207 0.00139 (0.00342) (0.00330) (0.00260) LagAv. Perf. 0.00616** 0.00142 (0.00274) (0.00289) DLogVix 0.0175 (0.0232) DExcessBP 0.0274* (0.0149) DFedFundsR. -0.0463*** (0.0161) DTermSpread 0.103*** (0.0176) DUnempl. Outl. -0.0487* (0.0266) Constant 0.0153*** 0.0122*** 0.0164*** (0.00328) (0.00337) (0.00309) Observations 106 106 106 𝑅2 0.095 0.127 0.418 Note: This table shows regressions of net inflows into U.S. domiciled, U.S. dollar denominated funds that invest in U.S. bank loans on the emerging markets dollar index and the advanced foreign economies dollar index. Regressions are parallel to those shown in table 11. Robust standard errors are in parentheses. *, **, and *** denote significance at the 10%, 5%, and 1% level. 51

Table 23: U.S. corporate bond yields and the dollar (1) (2) (3) Baseline EME/AFE EME/AFE DDollar -0.0229 (0.0175) DDollarEME -0.0404*** (0.0140) DDollarAFE 0.00951 (0.0176) Rat. XDDollar 0.0126** (0.00616) Rat. XDDollarEME 0.0100** 0.0101** (0.00490) (0.00417) Rat. XDDollarAFE 0.00423 0.00408 (0.00624) (0.00541) Rat. XDLogVix 0.295*** 0.283*** 0.282*** (0.0640) (0.0616) (0.0541) Rat. XDExcessBP 0.158*** 0.156*** 0.156*** (0.0428) (0.0429) (0.0357) Rat. XDFedFundsR. -0.0370 -0.0384 -0.0393 (0.0572) (0.0576) (0.0499) Rat. XDTermSpread -0.0969** -0.0944** -0.0942*** (0.0413) (0.0415) (0.0353) Rat. XDUnempl. Outl. 0.373** 0.373** 0.370*** (0.164) (0.163) (0.142) DLogVix -0.470*** -0.393** (0.180) (0.173) DExcessBP -0.437*** -0.428*** (0.121) (0.121) DFedFundsR. 0.576*** 0.585*** (0.158) (0.159) DTermSpread 0.892*** 0.878*** (0.116) (0.117) DUnempl. Outl. -0.783* -0.791* (0.465) (0.464) Rating-MaturityFE Yes Yes Yes TimeFE No No Yes Observations 3683 3683 3683 𝑅2 0.185 0.187 0.374 Note: This table shows that yields on the risky category of U.S. corporate bonds also responds to dollar movements. Yields rise when the dollar appreciates. The sample consists of monthly 5- and 10-year U.S. corporate bond yield indices for different rating buckets. Column (1) includes an interaction term between the change in the broad dollar index and a categorial variable that reflects the risk rating of the different bond indices. Higher values of the ratings variable correspond to a worse rating. Column (2) shows results when the dollar index is broken into an EME and and AFE dollar index. Column (3) is the same as column (2)butincludestime-fixedeffects. Standarderrorsclusteredbyrating×timeareinparentheses. *, **, and *** denote significance at the 10%, 5%, and 1% level. 52

Figures Figure 1: Total value of syndicated loans bought and buyer composition in SNC 160 0.6 140 0.5 120 0.4 100 80 0.3 60 0.2 40 0.1 20 0 0 noillib $ ni syub fo eulav latoT syub ni erahs 'sknab .S.U Total Value of Syndicated Loans Bought in the Secondary Market Year U.S. Banks Mutual Funds CLOs Other U.S. Entities Foreign Entities Share of U.S. banks Note: The figure shows the total yearly value of syndicated loans bought by U.S. banks, CLOs, mutual funds, other U.S. entities, and foreign entities in the secondary market (y-axis to the left) derived from SharedNationalCreditdatainIranietal.(2018). Thedashedlineinthefigurerepresentstheshareofloans bought by U.S. banks (y-axis on the right). 53

Figure 2: U.S. banks’ loan originations and the dollar snoitanigiro naol ni egnahc goL 4. 2. 0 2.- 4.- 6 4 2 0 2- 4xedni rallod daorb ni egnahC 2011q3 2013q1 2014q3 2016q1 2017q3 quarter Log change in origin. Change in broad dollar Note: The figures shows the quarterly log change in the total value of loan originations of 16 major U.S. banksaswellasthequarterlychangeinthebroaddollarindex,atrade-weightedindexofthedollarexchange rate vis-a-vis major U.S. trading parters, from 2011 to 2017. Information on U.S. banks’ loan originations is from the Y-14 reports U.S. banks file with the Federal Reserve. Figure 3: The U.S. National Financial Conditions Index (NFCI) and the Dollar Impulse Response Function of D NFCI Credit to D Dollar, post GFC Impulse Response Function of D NFCI to D Dollar, post GFC .04 .04 .02 .02 0 0 -.02 -.02 0 2 4 6 8 0 2 4 6 8 Months Months 95% CI Cumulative Orthogonalized IRF 95% CI Cumulative Orthogonalized IRF Graphs by irfname, impulse variable, and response variable Graphs by irfname, impulse variable, and response variable Note: The left (right) panel of the figure shows the impulse response function of the National Financial Conditions Subindex (National Financial Conditions Index) to a one-standard-deviation increase in the monthly change of the broad dollar index obtained from a vector autoregression (VAR). In the VAR, the following variables explain the monthly change in the credit index in the following order: the change in the fed funds rate, the change in the term spread, the change in the excess bond premium, the log difference of the VIX, the log difference of the CPI, the log difference of industrial output, the log difference of nonfarm employment, and the change in the dollar. The VAR includes two lags of all endogenous variables. 54

Figure 4: Secondary market liquidity as a SLOOS reason and secondary market prices )AM q3( erahS gnisaE soolS 4. 3. 2. 1. 0 1. 50. 0 50.- 1.- )AM q3( xednI ecirP ni egnahC 2002q1 2006q1 2010q1 2014q1 2018q1 Date Sloos Easing Share (3q MA) D Price Index (3q MA) Note: This figure shows the three-quarter moving-average share of banks that said in the SLOOS that they easedcreditstandardsbecauseof“improvedliquidityinthesecondarymarketfortheseloans”togetherwith thethree-quartermoving-averageS&P/LSTAU.S.LeverageLoan100Index,whichreflectstheperformance of the largest facilities in the U.S. leveraged loan market from 2002 to 2017. 55

Figure 5: U.S. mutual funds investing in corporate loans and the dollar )noillib $( swolfnI evitalummuC 001 08 06 04 02 0 031 021 011 001 09 xednI ralloD daorB 2006m1 2008m1 2010m1 2012m1 2014m1 2016m1 2018m1 Date Cummulative Inflows ($ billion) Broad Dollar Index Note: This figure shows the monthly net inflows into U.S. domiciled, U.S. dollar denominated mutual funds thatinvestinbankloansexclusivelyintheUnitedStatestogetherwiththemonthlybroaddollarindexfrom mid-2006 to 2017. Data on these mutual funds is from EFPR. Weekly data was summed within a month and aggregated over all funds in the data. 56

Appendix A: Robustness Tables Table 24: C&I credit standards and the dollar, pre and post crisis Nocrisis,pre-2008 Nocrisis,post-2008 (1) (2) (3) (4) DDollar 1.324** 1.229** 1.258*** 0.785* (0.655) (0.603) (0.403) (0.395) Sh. Tight(prev.) 0.878*** 0.954*** 0.641*** 0.950*** (0.0989) (0.0887) (0.0903) (0.138) DLogVix 21.30* 14.66*** (12.11) (4.580) DExcessBP 0.0791 0.744 (6.625) (4.092) DFedFundsR. 7.901* -2.680 (4.562) (3.427) DTermSpread -2.164 1.429 (3.051) (3.232) DUnempl. Outl. 0.934 22.00*** (15.67) (7.880) Constant -0.324 -1.285 -3.324*** -0.893 (1.139) (1.270) (0.980) (1.497) Observations 58 58 34 34 𝑅2 0.730 0.799 0.638 0.789 Adjusted𝑅2 0.720 0.771 0.615 0.732 Note: This table shows results of sample splits for SLOOS regressions. Columns (1) and (2) are based on a sample that only includes the pre-GFC period. Columns (3) and (4) only include quarters after the GFC. Robust standard errors are in parentheses. *, **, and *** denote significance at the 10%, 5%, and 1% level. Table 25: Credit standards of other types of loans C&ISmall Comm. RealEst. (1) (2) (3) (4) DDollar 0.869** 0.606** 1.089* 1.009** (0.366) (0.266) (0.612) (0.481) DLogVix 22.15*** 25.49** (5.397) (12.03) DExcessBP 1.396 -2.063 (3.307) (5.453) DFedFundsR. 1.898 7.533* (2.016) (4.081) DTermSpread -1.664 -3.861 (2.294) (2.970) DUnempl. Outl. 15.63 -9.468 (10.26) (11.40) Sh. Tight. CIS(prev) 0.755*** 0.863*** (0.134) (0.0969) Sh. Tight. Comm. RE(prev) 0.841*** 0.908*** (0.0800) (0.0638) Constant -0.798 -0.948 0.415 -0.774 (0.957) (0.796) (1.045) (1.097) Observations 74 74 62 62 𝑅2 0.568 0.707 0.663 0.749 Adjusted𝑅2 0.556 0.676 0.651 0.717 Note: Thistableshowsthatalsocreditstandardsforothertypesofloansarealsocorrelatedwiththedollar. Incolumns(1)and(2),thenetpercentageoffirmsthatreportedtighteningcreditstandardsforC&Iloansto small firms is the dependent variable. In columns (3) and (4), it is the net percentage of firms that reported tightening credit standards for commercial real estate loans. Robust standard errors are in parentheses. *, **, and *** denote significance at the 10%, 5%, and 1% level. 57

Table 26: Secondary market regressions, quarterly frequency Agg. Price Price Bid-Ask (1) (2) (3) (4) (5) (6) DDollar -0.00346*** -0.00151** -0.00259*** -0.00143*** 0.0156*** 0.00737** (0.00128) (0.000710) (0.000846) (0.000427) (0.00465) (0.00309) DLogVix -0.0259** -0.00767 0.181*** (0.0102) (0.00516) (0.0450) DExcessBP -0.0285*** -0.0145*** 0.0325 (0.00910) (0.00462) (0.0293) DFedFundsR. 0.000394 -0.0000937 0.00182 (0.00643) (0.00381) (0.0330) DTermSpread 0.00646 0.00579** -0.0355 (0.00405) (0.00242) (0.0213) DUnempl. Outl. -0.0453* -0.0535*** 0.233*** (0.0239) (0.0121) (0.0710) Constant 0.0143*** 0.0128*** 0.00219 0.00164 -0.0155 -0.0124* (0.00257) (0.00171) (0.00176) (0.00102) (0.0104) (0.00684) Observations 56 56 72588 72588 72947 72947 𝑅2 0.146 0.715 0.020 0.087 0.023 0.091 Adjusted𝑅2 0.130 0.680 0.020 0.086 0.023 0.090 Note: This table presents regressions of changes in secondary market loan prices and liquidity on changes in the broad dollar at a quarterly frequency. The table is parallel to table 8, which presents regressions performed on monthly data. Robust standard errors are in parentheses. *, **, and *** denote significance at the 10%, 5%, and 1% level. 58

Table 27: Robustness with additional controls: loan originations (1) (2) (3) (4) (5) DUtil. Exp. DUtil. Exp. DUtil. Exp. DUtil. Exp. DUtil. Exp. DDollar -0.0454* -0.0497 -0.0445* -0.0363 -0.0494* (0.0223) (0.0327) (0.0238) (0.0290) (0.0261) DLogVix -0.0463 -0.998 -0.0437 0.107 0.0551 (0.434) (0.614) (0.476) (0.529) (0.421) DExcessBP 0.0304 -0.121 0.0587 0.0152 0.0144 (0.175) (0.235) (0.178) (0.199) (0.188) DTermSpread 0.130 0.119 0.171 0.0515 0.185 (0.146) (0.195) (0.244) (0.212) (0.161) DUnempl. Outl. 1.396 1.113 1.479 1.238 0.917 (0.987) (1.129) (1.236) (1.118) (0.922) DFedFundsR. 0.0700 0.116 0.0763 0.111 0.0146 (0.143) (0.163) (0.160) (0.174) (0.149) DVRPUS 0.00824 (0.00475) DTreas. Basis3y -0.00315 (0.0114) DWorldRec. Prob. 0.630 (1.145) D3mRateOutlook 0.205 (0.658) DTermSpreadOutlook 0.240 (0.776) D4qRealGDP -0.490 (8.106) D4qUnempl. Rate -0.0816 (0.202) DOilPrice 0.00717 (0.0107) RealGDPOutlook -48.81* (24.35) Constant 0.0376 0.0577 0.0395 -0.0103 0.0573 (0.0526) (0.0602) (0.0585) (0.160) (0.0564) Observations 25 25 25 25 25 𝑅2 0.294 0.380 0.298 0.310 0.391 Adjusted𝑅2 0.059 0.008 -0.053 -0.104 0.141 Note: This table shows the robustness of key results to the inclusion of additional macro and financial control variables. The dependent variable is the log change in U.S. banks’ quarterly loan originations. Robust standard errors are in parentheses. *, **, and *** denote significance at the 10%, 5%, and 1% level. 59

Table 28: Robustness with additional controls: credit standards (1) (2) (3) (4) (5) (6) Sh. Tight. Sh. Tight. Sh. Tight. Sh. Tight. Sh. Tight. Sh. Tight. DDollar 0.918** 0.890** 0.890** 1.299*** 0.918** 1.225** (0.398) (0.436) (0.416) (0.489) (0.399) (0.546) Sh. Tight(prev.) 0.914*** 0.996*** 0.904*** 0.940*** 0.911*** 1.008*** (0.0814) (0.0610) (0.0854) (0.0800) (0.0828) (0.0620) DLogVix 15.70*** 13.14* 16.29*** 17.69*** 15.81*** 16.32 (5.451) (7.466) (5.432) (6.122) (5.584) (9.850) DExcessBP 2.966 -0.781 3.248 1.946 2.958 -1.036 (3.512) (3.632) (3.519) (3.625) (3.578) (3.572) DTermSpread -1.559 -0.385 -1.428 -2.588 -1.847 -0.161 (2.041) (2.613) (3.090) (2.148) (2.315) (3.204) DUnempl. Outl. 9.052 23.13*** 5.696 10.14 11.64 11.50 (9.504) (8.419) (10.31) (9.767) (8.499) (12.84) DFedFundsR. 3.051 -0.184 3.014 3.035 2.911 0.195 (2.627) (1.940) (3.125) (2.786) (2.484) (2.278) DVRPUS -0.0199 -0.0431 (0.0526) (0.0605) DTreas. Basis3y 0.306* 0.317* (0.167) (0.183) DWorldRec. Prob. 6.548 10.32 (9.279) (9.998) D3mRateOutlook -7.753 -4.455 (6.309) (7.679) DTermSpreadOutlook -3.882 -0.137 (8.323) (10.15) D4qRealGDP -41.44 -54.69 (75.34) (90.61) D4qUnempl. Rate -0.848 -0.803 (1.575) (1.631) DOilPrice 0.294** 0.213 (0.135) (0.134) RealGDPOutlook 235.0 -751.5* (561.5) (392.1) Constant -1.046 -0.167 -1.128 -0.521 -1.096 0.571 (0.960) (0.933) (0.993) (2.020) (1.005) (2.509) Observations 92 59 91 92 92 59 𝑅2 0.787 0.908 0.790 0.798 0.787 0.922 Adjusted𝑅2 0.769 0.889 0.766 0.773 0.767 0.892 Note: Thistableshowstherobustnessofkeyresultstotheinclusionofadditionalmacroandfinancialcontrol variables. The dependent variable is the net percentage of U.S. banks that said that they tightened credit standards for C&I loans to large and medium enterprises. Robust standard errors are in parentheses. *, **, and *** denote significance at the 10%, 5%, and 1% level. 60

Table 29: Robustness with additional controls: riskiness of new loans (1) (2) (3) (4) (5) DPD DPD DPD DPD DPD DDollar -0.000357*** -0.000297** -0.000365*** -0.000544** -0.000338*** (0.000111) (0.000125) (0.000103) (0.000187) (0.000101) DLogVix 0.00113 0.00806 0.000674 -0.00213 0.000646 (0.00156) (0.00608) (0.00168) (0.00278) (0.00139) DExcessBP -0.00107 0.000565 -0.00138 -0.00142 -0.000989 (0.00123) (0.00116) (0.00143) (0.00166) (0.00140) DTermSpread 0.00125* 0.00114 0.000697 0.00316* 0.000982 (0.000661) (0.00104) (0.00135) (0.00152) (0.000585) DUnempl. Outl. -0.00108 -0.000173 -0.000380 0.000470 0.00123 (0.00404) (0.00445) (0.00412) (0.00416) (0.00385) DFedFundsR. 0.00121** 0.000957 0.00103 0.00164 0.00148** (0.000532) (0.000707) (0.000663) (0.00105) (0.000566) DVRPUS -0.0000573 (0.0000487) DTreas. Basis3y 0.0000408 (0.000105) DWorldRec. Prob. -0.0128* (0.00715) D3mRateOutlook 0.000942 (0.00404) DTermSpreadOutlook -0.00234 (0.00627) D4qRealGDP -0.0236 (0.0586) D4qUnempl. Rate -0.00131 (0.00136) DOilPrice -0.000107 (0.0000933) RealGDPOutlook 0.235** (0.0886) Constant 0.000239 0.0000726 0.000169 -0.000372 0.000144 (0.000281) (0.000306) (0.000238) (0.000866) (0.000287) Observations 25 25 25 25 25 𝑅2 0.308 0.466 0.347 0.492 0.363 Adjusted𝑅2 0.077 0.146 0.020 0.187 0.100 Note: Thistableshowstherobustnessofkeyresultstotheinclusionofadditionalmacroandfinancialcontrol variables. The dependent variable is the change in the weighted average probability of default of new loans issued by U.S. banks. Robust standard errors are in parentheses. *, **, and *** denote significance at the 10%, 5%, and 1% level. 61

Table 30: Robustness with additional controls: reasons for easing, secondary market prices and liquidity (1) (2) (3) (4) (5) (6) (7) ShareSec. Liqu. Agg. Price Agg. Price Price Price Bid-Ask Bid-Ask DDollar -0.0123** -0.00138 -0.00294*** -0.00125*** -0.00118*** 0.00688** 0.00687*** (0.00591) (0.000897) (0.000868) (0.000456) (0.000255) (0.00267) (0.00226) DLogVix -0.215* -0.0612*** -0.0288*** -0.0166 -0.00631 0.255*** 0.132*** (0.121) (0.0174) (0.00927) (0.0117) (0.00394) (0.0607) (0.0292) DExcessBP -0.126** -0.0307*** -0.00727 -0.0167*** -0.000253 0.0527* 0.00694 (0.0527) (0.0108) (0.00655) (0.00620) (0.00159) (0.0300) (0.00968) DFedFundsR. 0.0834* -0.00197 0.0325*** -0.00302 0.00252 0.0185 0.00537 (0.0474) (0.00917) (0.0109) (0.00462) (0.00219) (0.0386) (0.0176) DTermSpread 0.0387 -0.00146 0.0188 0.000239 0.00329* -0.0318 -0.0161 (0.0518) (0.00741) (0.0121) (0.00334) (0.00176) (0.0315) (0.0125) DUnempl. Outl. 0.127 -0.0449* -0.0278 -0.0210 -0.0113 0.183* 0.0825 (0.197) (0.0257) (0.0302) (0.0146) (0.00807) (0.104) (0.0510) D3mRateOutlook -0.0319 -0.0227 0.0206 -0.00525 -0.00758 -0.0263 0.0596 (0.115) (0.0218) (0.0186) (0.00956) (0.00589) (0.0646) (0.0419) DTermSpreadOutlook -0.120 -0.0308 0.0198 -0.0231** -0.00773 -0.0168 0.00947 (0.139) (0.0232) (0.0210) (0.0104) (0.00492) (0.0872) (0.0345) D4qRealGDP 0.619 -0.106 -0.0465 0.112 -0.0310 -1.792** -0.267 (1.187) (0.205) (0.156) (0.115) (0.0451) (0.821) (0.243) D4qUnempl. Rate -0.00633 0.00125 0.000733 0.00429** 0.00116** -0.00469 -0.00441 (0.0238) (0.00332) (0.00135) (0.00185) (0.000562) (0.00889) (0.00308) RealGDPOutlook -3.955 0.907 1.659 0.918 0.719* -4.769 -3.123 (6.698) (1.293) (1.152) (0.594) (0.369) (3.160) (2.275) DVRPUS 0.00222*** 0.000258* 0.0000118 0.0000831 0.00000382 -0.000903** -0.000111 (0.000775) (0.000138) (0.0000524) (0.000100) (0.0000205) (0.000377) (0.000139) DTreas. Basis3y -0.00522* -0.000205 -0.000574 -0.000259 -0.0000111 0.00277* -0.000818 (0.00274) (0.000296) (0.000406) (0.000190) (0.0000969) (0.00139) (0.000697) DWorldRec. Prob. -0.291 0.0190 -0.0249 0.00503 -0.00499 0.0421 0.0613** (0.180) (0.0387) (0.0181) (0.0194) (0.00405) (0.0894) (0.0244) Constant 0.195*** 0.0160*** 0.00492 0.00117 0.00152 0.0256 0.00176 (0.0321) (0.00532) (0.00356) (0.00292) (0.00115) (0.0217) (0.00604) Observations 59 56 192 69449 215141 69556 214788 𝑅2 0.460 0.753 0.569 0.098 0.040 0.106 0.033 Adjusted𝑅2 0.289 0.668 0.535 0.098 0.040 0.106 0.033 Note: This table shows the robustness of key results to the inclusion of additional macro and financial control variables. In column (1), the dependent variable is the share of firms that mentioned secondary marketliquidityasareasonforeasingcreditstandardsintheSLOOS.Incolumns(2)and(3),thedependent variable is the log change in the U.S. Leverage Loan 100 Index. Columns (4) and (5) present regressions of the log change in the price of a loan sold on the secondary market on changes in the broad dollar index. Columns (6) and (7) have the monthly log change in the bid-ask spread of a loan sold on the secondary market as the left-hand-side variable. Even columns use quarterly data, while odd columns (except column (1))usemonthlydata. Standarderrorsareclusteredinaccordancewithclusteringinthebaselineregressions. *, **, and *** denote significance at the 10%, 5%, and 1% level. 62

Table 31: Robustness with additional controls, including oil price: reasons for easing, secondary market prices and liquidity (1) (2) (3) (4) (5) (6) (7) ShareSec. Liqu. Agg. Price Agg. Price Price Price Bid-Ask Bid-Ask DDollar -0.0153** -0.000674 -0.000930 -0.000756 -0.000788*** 0.00257 0.00274 (0.00722) (0.000958) (0.000930) (0.000469) (0.000269) (0.00266) (0.00219) DLogVix -0.256* -0.0515** -0.0258*** -0.0102 -0.00497 0.199*** 0.118*** (0.148) (0.0199) (0.00865) (0.0119) (0.00420) (0.0631) (0.0309) DExcessBP -0.118** -0.0322*** -0.00842 -0.0177*** -0.000651 0.0615** 0.0112 (0.0551) (0.0109) (0.00608) (0.00612) (0.00154) (0.0297) (0.00980) DFedFundsR. 0.0788 -0.00100 0.0297*** -0.00223 0.00237 0.0115 0.00709 (0.0482) (0.00895) (0.00955) (0.00442) (0.00209) (0.0369) (0.0161) DTermSpread 0.0468 -0.00350 0.0113 -0.00134 0.00173 -0.0182 0.000510 (0.0539) (0.00763) (0.00986) (0.00343) (0.00199) (0.0285) (0.0137) DUnempl. Outl. 0.102 -0.0393 -0.0204 -0.0170 -0.0100 0.149 0.0692 (0.199) (0.0258) (0.0271) (0.0145) (0.00757) (0.106) (0.0470) D3mRateOutlook -0.0547 -0.0169 0.0224 -0.00120 -0.00524 -0.0616 0.0353 (0.123) (0.0208) (0.0185) (0.00915) (0.00554) (0.0637) (0.0382) DTermSpreadOutlook -0.140 -0.0254 0.0225 -0.0197* -0.00615 -0.0459 -0.00703 (0.143) (0.0225) (0.0203) (0.0102) (0.00498) (0.0799) (0.0344) D4qRealGDP 0.728 -0.115 -0.101 0.0920 -0.0434 -1.622** -0.137 (1.245) (0.199) (0.138) (0.111) (0.0451) (0.781) (0.243) D4qUnempl. Rate -0.00492 0.00115 0.000316 0.00407** 0.00109** -0.00278 -0.00367 (0.0238) (0.00324) (0.00128) (0.00182) (0.000551) (0.00907) (0.00308) DOilPrice -0.00163 0.000345 0.00103*** 0.000250** 0.000225*** -0.00216*** -0.00236*** (0.00203) (0.000225) (0.000291) (0.000103) (0.0000713) (0.000805) (0.000678) RealGDPOutlook -4.732 1.205 2.342** 1.059* 0.805** -5.987* -4.013* (6.967) (1.370) (1.070) (0.592) (0.345) (3.443) (2.194) DVRPUS 0.00246** 0.000201 0.00000852 0.0000448 -0.00000450 -0.000571 -0.0000230 (0.000923) (0.000149) (0.0000404) (0.000101) (0.0000215) (0.000365) (0.000145) DTreas. Basis3y -0.00527* -0.000159 -0.000413 -0.000246 0.00000304 0.00266** -0.000967 (0.00265) (0.000281) (0.000399) (0.000182) (0.0000962) (0.00128) (0.000677) DWorldRec. Prob. -0.295 0.0220 -0.0235* 0.00575 -0.00339 0.0359 0.0444** (0.180) (0.0388) (0.0134) (0.0193) (0.00385) (0.0912) (0.0220) Constant 0.196*** 0.0156*** 0.00568* 0.00103 0.00169 0.0267 -0.00000282 (0.0334) (0.00535) (0.00319) (0.00287) (0.00113) (0.0209) (0.00604) Observations 59 56 192 69449 215141 69556 214788 𝑅2 0.467 0.760 0.619 0.099 0.043 0.109 0.037 Adjusted𝑅2 0.282 0.670 0.586 0.099 0.043 0.109 0.037 Note: This table shows the robustness of key results to the inclusion of additional macro and financial control variables. In column (1), the dependent variable is the share of firms that mentioned secondary marketliquidityasareasonforeasingcreditstandardsintheSLOOS.Incolumns(2)and(3),thedependent variable is the log change in the U.S. Leverage Loan 100 Index. Columns (4) and (5) present regressions of the log change in the price of a loan sold on the secondary market on changes in the broad dollar index. Columns (6) and (7) have the monthly log change in the bid-ask spread of a loan sold on the secondary market as the left-hand-side variable. Even columns use quarterly data, while odd columns (except column (1))usemonthlydata. Standarderrorsareclusteredinaccordancewithclusteringinthebaselineregressions. *, **, and *** denote significance at the 10%, 5%, and 1% level. 63

Table 32: Robustness with additional controls: secondary market prices and liquidity, interactions (1) (2) (3) (4) (5) (6) DPrice DPrice DPrice DSpread DSpread DSpread DDollar -0.0000450 -0.00296** 0.00238 0.0161** (0.000221) (0.00128) (0.00243) (0.00785) RiskyXDDollar -0.00232*** -0.00204*** 0.00113 0.000533 (0.000503) (0.000441) (0.00235) (0.00166) Risky -0.000343 -0.000920** -0.00251 -0.000726 (0.000480) (0.000463) (0.00205) (0.00181) Sh. USbanksXDDollar 0.00916 -0.0531 (0.00706) (0.0427) Sh. USbanks -0.0308*** 0.160*** (0.00750) (0.0475) DLogVix -0.00473 -0.00579 0.118*** 0.131*** (0.00416) (0.00439) (0.0310) (0.0310) DExcessBP -0.000712 0.000245 0.0112 0.00432 (0.00153) (0.00168) (0.00975) (0.0102) DFedFundsR. 0.00193 0.00383* 0.00670 0.0154 (0.00207) (0.00219) (0.0160) (0.0178) DTermSpread 0.00169 0.00262 0.000431 -0.00287 (0.00196) (0.00192) (0.0138) (0.0134) DUnempl. Outl. -0.0106 -0.00926 0.0688 0.0666 (0.00752) (0.00699) (0.0468) (0.0470) D3mRateOutlook -0.00562 -0.00628 0.0353 0.0421 (0.00557) (0.00541) (0.0381) (0.0343) DTermSpreadOutlook -0.00665 -0.00876 -0.00640 -0.00180 (0.00504) (0.00545) (0.0343) (0.0362) D4qRealGDP -0.0334 0.00365 -0.139 -0.401 (0.0430) (0.0509) (0.241) (0.260) D4qUnempl. Rate 0.00119** 0.00142** -0.00337 -0.00709** (0.000539) (0.000604) (0.00301) (0.00335) DOilPrice 0.000222*** 0.000122 -0.00235*** -0.00170** (0.0000684) (0.0000804) (0.000680) (0.000701) RealGDPOutlook 0.797** 0.727** -4.023* -3.547* (0.350) (0.316) (2.185) (2.110) DVRPUS -0.00000296 -0.00000169 -0.0000247 0.00000908 (0.0000205) (0.0000198) (0.000145) (0.000124) DTreas. Basis3y 0.0000676 0.000138 -0.000997 -0.00230*** (0.000100) (0.000114) (0.000666) (0.000844) DWorldRec. Prob. -0.00327 -0.00285 0.0448** 0.0286 (0.00385) (0.00342) (0.0220) (0.0190) c1 0 0 0 0 (.) (.) (.) (.) Constant 0.00157 0.00533*** 0.00109 -0.0162** (0.00103) (0.00134) (0.00625) (0.00689) TimeFE No Yes No No Yes No Observations 215141 224256 167885 214788 223620 167615 𝑅2 0.048 0.098 0.056 0.037 0.078 0.045 Adjusted𝑅2 0.048 0.097 0.056 0.037 0.078 0.045 Note: Thistableshowstherobustnessofkeyresultstotheinclusionofadditionalmacroandfinancialcontrol variables. In columns (1) and (2), the dependent variable is the monthly log change in the price of a loan sold on the secondary market. Columns (3) and (4) have the monthly log change in the bid-ask spread of a loan sold on the secondary market as the left-hand-side variable. Standard errors clustered by time are in parentheses. *, **, and *** denote significance at the 10%, 5%, and 1% level. 64

Appendix B: Data Appendix ∙ FR Y-14: Confidential quarterly data available at the Federal Reserve starting in 2011:Q3. Reported by all banks that participate in CCAR/DFAST. ∙ SLOOS data: Public aggregated responses published quarterly on the Federal Reserve Board’s website; confidential bank-level version available at the Federal Reserve; since 1990. ∙ Loan Syndication and Trading Association Data: Daily information on loans sold on the secondary market, including mean bid and mean ask prices. ∙ FR Y-9C: Quarterly information on the income and balance sheets of U.S. Bank Holding Companies. ∙ Survey of Professional Forecasters: Information on expected unemployment rates, real GDP growth, short-term and long-term interest rates as the means of the 4-quarter ahead forecast values. ∙ Broad dollar index, EME and AFE dollar index: Trade-weighted dollar indices computed and published by the Federal Reserve. ∙ Trade intensities: Calculated as exports/(production + imports - exports) and imports/(production + imports - exports). Production data are from the Bureau of Economic Analysis. Trade data are from the U.S. Census Bureau. ∙ Standardmacroandfinancialvariables: QuarterlyandmonthlyvaluesfromBloomberg and Haver. ∙ Excess bond premium: Downloaded from the Federal Reserve’s website: https://www. federalreserve.gov/econresdata/notes/feds-notes/2016/updating-the-recession-risk-andthe-excess-bond-premium-20161006.html. ∙ ConvenienceyieldsprovidedbyWenxinDu: https://sites.google.com/site/wenxindu/data/ govt-cip and Du et al. (2018b). 65

∙ Variance risk premium provided by Juan-Miguel Londono used in Londono and Zhou (2017). ∙ WorldrecessionprobabilityprovidedbyPablo-CubaBorda: https://www.federalreserve. gov/econres/notes/ifdp-notes/monitoring-the-world-economy-a-global-conditions-index- 20180615.htm. ∙ EFPR data: Weekly information on performance, assets under management, and net flows into mutual funds. ∙ The dollar response to monetary policy shocks: Measured from 15 minutes before to 90 minutes after monetary policy announcements used in Ferrari et al. (2017) and Cieslak and Schrimpf (2019). ∙ Syndicated loan data computed from Shared National Credit Data: Provided by Ralf Meisenzahl as shown in Irani et al. (2018). ∙ AdditionalinputvariablesfortheVARfromFRED:Industrialproduction(https://fred. stlouisfed.org/series/INDPRO),totalnon-farmemployment(https://fred.stlouisfed.org/series/ PAYEMS),NationalFinancialConditionsIndex(https://fred.stlouisfed.org/series/NFCI) and National Financial Conditions Subindex (https://fred.stlouisfed.org/series/NFCI CREDIT), Consumer Price Index (https://fred.stlouisfed.org/series/CPIAUCSL). ∙ FFIEC009 exposure data: Confidential quarterly information on the foreign exposures of U.S. banks available at the Federal Reserve. ∙ CRANE data: Monthly information on money market mutual funds’ investments, including interest rates paid. 66