Trade Uncertainty and U.S. Bank Lending

Board of Governors of the Federal Reserve System International Finance Discussion Papers ISSN 1073-2500 (Print) ISSN 2767-4509 (Online) Number 1383 November 2023 Trade Uncertainty and U.S. Bank Lending Ricardo Correa, Julian di Giovanni, Linda S. Goldberg, and Camelia Minoiu Please cite this paper as: Correa, Ricardo, Julian di Giovanni, Linda S. Goldberg, and Camelia Minoiu (2023). “Trade Uncertainty and U.S. Bank Lending,” International Finance Discussion Papers 1383. Washington: Board of Governors of the Federal Reserve System, https://doi.org/10.17016/IFDP.2023.1383. NOTE: International Finance Discussion Papers (IFDPs) are preliminary materials circulated to stimulate discussion and critical comment. The analysis and conclusions set forth are those of the authors and do not indicate concurrence by other members of the research staff or the Board of Governors. References in publications to the International Finance Discussion Papers Series (other than acknowledgement) should be cleared with the author(s) to protect the tentative character of these papers. Recent IFDPs are available on the Web at www.federalreserve.gov/pubs/ifdp/. This paper can be downloaded without charge from the Social Science Research Network electronic library at www.ssrn.com.

Trade Uncertainty and U.S. Bank Lending Ricardo Correa Julian di Giovanni Federal Reserve Board Federal Reserve Bank of New York of Governors and CEPR Linda S. Goldberg Camelia Minoiu Federal Reserve Bank of New York Federal Reserve Bank NBER and CEPR of Atlanta November 1, 2023 Abstract ThispaperusesU.S.loan-levelcreditregisterdataandthe2018–2019TradeWartotestforthe effects of international trade uncertainty on domestic credit supply. We exploit cross-sectional heterogeneityinbanks’ex-anteexposuretotradeuncertaintyandfindthatanincreaseintrade uncertainty is associated with a contraction in bank lending to all firms irrespective of the uncertainty that the firms face. This baseline result holds for lending at the intensive and extensive margins. We document two channels underlying the estimated credit supply effect: a wait-and-see channel by which exposed banks assess their borrowers as riskier and reduce the maturity of their loans and a financial frictions channel by which exposed banks facing relatively higher balance sheet constraints contract lending more. The decline in credit supply has real effects: firms that borrow from more exposed banks experience lower debt growth and investment rates. These effects are stronger for firms that are more reliant on bank finance. JEL Classifications: G21, F34, F42 Keywords: Trade uncertainty, bank loans, trade finance, global value chains, trade war ∗Contact: Ricardo Correa (Ricardo.Correa@frb.gov), Julian di Giovanni (juliandigiovanni@gmail.com), Linda S. Goldberg (Linda.Goldberg@ny.frb.org), Camelia Minoiu (Camelia.Minoiu@atl.frb.org). We are grateful to Michelle Alexopoulos, Chris Boehm, Nick Bloom, Valentina Bruno, Steven Davis, Lorenzo Garlappi, Kristine Hankins, Tarek Hassan,DalidaKadyrzhanova,MatteoIacovellio,AbelIglesias,SeungLee,RalfMeisenzahl(discussant),LubosPastor,DianePierret(discussant),AndreaPolo(discussant),AndreaPresbitero,VeronicaRappoport(discussant),Brad Setser(discussant),BoSun,EugeneTan,LenaTonzer(discussant),LilianaVarela(discussant),FrankWarnock,and participants at IBRN workshops and meetings, the Global Risk, Uncertainty, and Volatility (GRUV) workshop at the Federal Reserve Board, 29th CEPR European Summer Symposium in International Macroeconomics (ESSIM), IFABSannualconference,EuropeanFinanceAssociation(EFA)AnnualMeeting,FRBDallasConferenceonSupply Chains in a Changing Global Landscape, FRB New York Global Research Forum on International Macroeconomics andFinance,5thEBRD-CEPRResearchSymposium,IMFConferenceonGeoeconomicFragmentation,Spring2023 NBER Conference on “International Fragmentation, Supply Chains, and Financial Frictions,” Swedish House of Finance Conference on “The Effects of New Geopolitical Risks on Financial Markets and Firms,” 2023 European Economic Association Annual Meeting, Stanford University 2023 SITE Conference “The Macroeconomics of Uncertainty and Volatility,” and seminars at the Norges Bank, Bank of Canada, Bank of England, and Bank of Italy for useful suggestions. We thank Stephanie Sezen, Diego Silva, and Kelsey Shipman for research assistance. The views expressedinthispaperarethoseoftheauthorsanddonotnecessarilyrepresentthoseoftheFederalReserveBankof New York, Federal Reserve Bank of Atlanta, the Board of Governors of the Federal Reserve, or the Federal Reserve System.

1 Introduction The recent era of trade globalization witnessed firms’ foreign activities proliferate as they entered new markets and sourced more intermediate inputs from abroad. This exponential expansion of international trade ended after the Global Financial Crisis (GFC), with events such as Brexit, trade wars, and the COVID-19 pandemic being major sources of increased trade uncertainty. This uncertainty may also impact financial intermediaries’ given their important role in financing global transactions. In particular, an increase in trade uncertainty can affect firms’ creditworthiness and bank balance sheets, which in turn can induce changes in banks’ lending behavior and their supply of credit. Indeed, according to a Federal Reserve survey, U.S. banks expected to take a range of actions in 2019 to mitigate the impact of international trade developments on their balance sheets, including tightening lending standards and hedging credit risks through derivatives.1 Against this backdrop, we ask how the effects of international trade uncertainty on the domestic economy may be propagated and amplified by banks. This paper assesses the effects of trade uncertainty on U.S. banks’ credit supply by exploiting the spike in trade uncertainty that occurred during the 2018–2019 Trade War. A priori, it is theoretically ambiguous how uncertainty associated with international trade developments will affect banks. On the one hand, banks could serve as shock dampeners if they internalize the disruptions in their borrowers’ activities caused by trade uncertainty. On the other hand, banks may contract lending if they are worried about the prospect of balance sheet losses. We investigate theseissues,startingwiththeconstructionofanovelmeasureofbankexposuretotradeuncertainty by combining firm-level information on trade uncertainty with detailed data on U.S. banks’ loan exposurestodomesticborrowers. Weexploitthecross-sectionalbankheterogeneityinthisexposure to test for the credit supply effect of the increase in uncertainty, while controlling for firm-level credit demand. We next investigate the key mechanisms through which banks’ exposure to trade uncertainty affects their credit supply. Banks’ behavior might be driven by a wait-and-see strategy, whereby the exposed banks are more prone to pull back from risk-taking and to shorten loan maturities. Responses might also be driven by a financial frictions channel by which banks’ credit 1Details on the April 2019 Senior Loan Officer Opinion Survey conducted by the Federal Reserve are available here,includingreferencestothespecialquestionsinvestigatingC&Ilendingtofirmsthatareexposedtodevelopments in Asia or Europe. 1

supply depends on balance sheet constraints. Finally, we ask whether the estimated changes in credit supply have real effects on firms. Our first novel finding is that an increase in trade uncertainty is associated with a larger credit contractionatthebank-firmlevelformoreexposedbanks,thatis,thosebankswithalargerex-ante share of loans to firms in sectors facing a greater increase in ex-post trade uncertainty. This result holds even when we restrict the set of borrowers to firms that are relatively less exposed to an increase in trade uncertainty. Second, the contraction in credit supply is stronger for banks that face larger financial frictions and is also consistent with exposed banks adopting a wait-and-see attitude on lending by evaluating all borrowers—even those in low-uncertainty sectors—as being riskier. Third, firm characteristics affect how banks adjust lending in the face of changes in trade uncertainty. Notably, banks exposed to trade uncertainty contract lending more to firms that are less protected by trade policy. The real outcomes for firms are worse when they borrow from the more exposed banks, with this result stronger for those firms that are more reliant on bank credit. Our analysis uses a comprehensive loan-level data set collected through the Federal Reserve (FR) Y-14Q form (known as the “U.S. credit register”). The data are comprised of quarterly bankfirm loan commitments of minimum size $1 million extended to domestic (public and private) firms bytheU.S.banksthataresubjecttoannualstresstests(thosebankswithassetsabove$50billion). Weusethisdatasettoexamineawiderangeofoutcomesassociatedwiththeintensiveandextensive margins of lending, including lending volumes and spreads, maturities, and the probability of new loan originations. We also analyze the probabilities of default assigned by banks to individual borrowers. Furthermore, we use these data to construct our key measure of bank exposure to trade uncertainty by combining loan exposures with firm-level measures of trade uncertainty. Firm-level tradeuncertaintymeasuresaresourcedfromHassanetal.(2019),Hassanetal.(2020a),andHassan et al. (2020b) and are based on textual analysis of the transcripts of listed firms’ quarterly earnings calls. Given that the firms in the credit register and the uncertainty data do not overlap perfectly, we take a three-step approach in constructing the bank exposure to trade uncertainty variable. First, we aggregate the firm-level uncertainty measures to the sector-level. Second, we assign these sector-level uncertainty measures to borrowers in the credit register based on their sectoral classification. Finally, we aggregate this information at the bank level by taking the average change in uncertainty between 2016–2017 and 2018–2019 across sectors, weighted by initial loan shares in 2

a given sector. The loan shares are taken to be averages over 2014–2015 so they are lagged relative to the start of the sample and hence unlikely affected by the 2018–2019 Trade War. This approach makes the bank exposure measure more likely predetermined with respect to economic conditions during the sample period. Weuseadifference-in-differencesestimationframework. Ourbaselinespecificationregressesthe growth rate in outstanding loans at the bank-firm loan level on the measure of bank exposure to trade uncertainty interacted with a Post dummy taking the value of one for the years of heightened trade uncertainty in 2018 and 2019, and zero for the years 2016 and 2017. To corroborate that the shifts in loan quantities are consistent with a shift in the supply of credit, we estimate complementary specifications using loan spreads as the dependent variable. We make sure that our results are not confounded by standard determinants of banks’ lending decisions by controlling for bank size, capital, core deposits, and sectoral specialization (defined as in Paravisini et al., 2023) in levels and interacted with the Post dummy. We further show that the bank exposure measure is unrelated to these control variables in each yearly cross-section of banks over the sample period, which provides additional support to the validity of the assumption that the bank exposure measure is unrelated to bank attributes that might also affect lending. Akeyempiricalchallengeinisolatingtheeffectsoftradeuncertaintyoncreditsupplyisthefact that credit supply by banks and credit demand by firms may change simultaneously in response to changesinthetradeenvironment. Internationaltradeisimportantforthebankingsectoraschanges in firms’ foreign activities often shift their credit demand (Amiti and Weinstein, 2011). To address this issue, we exploit the granular nature of our data, at the bank-firm loan-level, with controls for firm×quarterfixedeffectstoabsorbtime-varyingcreditdemandshiftsforagivenfirm(Khwajaand Mian, 2008; Jim´enez et al., 2020). We also control for firm×bank fixed effects to account for timeinvariant bank-specific loan demand for individual firms and for potential endogenous matching between banks and firms (Chodorow-Reich, 2014; Farinha et al., 2022; Paravisini et al., 2023). Placebo tests indicate that banks with different levels of exposure to trade uncertainty have similar lending patterns before the sample period, suggesting that unobservable bank characteristics do not explain our results. Throughout the analyses, we reinforce the importance of controlling for credit demand by presenting results on bank lending for two borrower samples: (i) all firms, and (ii) firms that are in low-uncertainty sectors and less likely to have strong endogenous shifts in 3

credit demand.2 We have three sets of main results. Our first result is that an increase in trade uncertainty is associatedwithalargercreditcontractionformoreexposedbanksvis-`a-visall borrowers, including those that are less exposed to an increase in trade uncertainty. This spillover effect through banks is evident on both the intensive and extensive margins of lending: more exposed banks reduce loan growth, charge higher spreads, and are less likely to grant new loans than other banks. The credit supply contraction is economically meaningful. The point estimates from regressions for the full sample imply that a one standard deviation increase in bank exposure to trade uncertainty is associated with a 2.6 percentage point (ppt) decline in loan growth (compared to 0% median loan growth for the sample) and an increase in loan spreads by 6.5 basis points (bps) (compared to 185 bpsmedianloanspreadforthesample). Numbersaresimilarwhenrestrictingtheregressionsample to low-uncertainty firms: a 2.8 ppt contraction in loan growth and a 7.1 bps rise in loan spreads. A one standard deviation increase in bank exposure to trade uncertainty cuts the probability of new loan origination by 0.5%. The second set of results addresses the mechanisms through which trade uncertainty can affect banks’ credit supply. Consistent with real-options theory and adopting a wait-and-see attitude (DixitandPindyck,1994), moreexposedbanksreducethematurityofloansandshifttowardtypes of loans that can be called in early by banks (so-called demandable loans). Moreover, given that exposed banks anticipate a wider dispersion in loan returns and may have difficulties forecasting revenues and capital needs, they downgrade the perceived creditworthiness of firms, as reflected in higher assessed probabilities of default.3 Exposed banks also contract their lending more strongly to firms that are perceived as likely to be adversely affected by the Trade War and hence riskier ex ante, which we measure in two ways: those firms in manufacturing sectors that receive low import protection and those firms in sectors with high import dependence. The financial constraints channel is supported as well, as exposed banks with lower levels of current and stressed capital levels contract their lending by more than other banks. Consistent with both mechanisms, we find that exposed banks rotate their balance sheets away from loans and into safer assets, notably 2Inaddition, weshowthat creditdemand, as reflectedin creditline utilizationrates, actually goes upduring the Trade War for firms in high-uncertainty sectors. 3Infact, aFederalReservesurveyrevealedinApril2019thatU.S.bankswithsizableloancommitmentstofirms exposedtointernationaltradedevelopmentsexpectedtheoutlookforloanlossestodeteriorateoverthecourseofthe year (as discussed further in Section 4.3). 4

securities. The third set of results focuses on the consequences of exposed banks’ credit contraction for the real sector. Our analysis of real effects uses a loan-weighted average of each firm’s exposure to their banks’ exposure to trade uncertainty. We test whether firms that are more exposed to trade uncertainty through their banks are affected in terms of their investment and total debt growth. We find that the more exposed firms are unable to substitute for reduced bank lending through alternative sources of finance and these firms exhibit lower total debt growth and investment rates. A one standard deviation increase in firms’ exposure to trade uncertainty via their relationship with exposed banks is associated with an economically meaningful decrease of the growth rate of the firms’ total debt and of their investment ratio in 2018–2019 by 2.4 and 2.7 ppts, respectively. These results are consistent with a credit supply contraction having a material adverse effect on exposed firms’ real outcomes. We also find that private firms—more likely to depend on bank financing— and firms with a higher share of bank debt experience relatively worse real outcomes, which confirms banks as a conduit for amplifying the effects of trade uncertainty. We conduct additional tests to increase confidence in the interpretation of our results. First, we present evidence to allay the potential concern that our results are driven by the effects of the Trade War on realized and expected returns on loans (a first-moment effect) instead of the uncertainty regarding loan returns (a second-moment effect). Specifically, we show that the results are invariant to controlling for two measures of returns on loans—bank exposure to changes in actual trade policy (that is, the loan share to tariffs-hit sectors) and bank exposure to changes in overall sentiment (constructed in the same way as the baseline exposure measure). Results do not change when we additionally control for bank exposure to changes in non-trade uncertainty (that is, political uncertainty in sectors other than trade). Second, we show that our results are robust to other potential explanations for our baseline findings, including the possibility that changes in macroeconomic conditions—such as fluctuations in the value of the U.S. dollar and in commodity prices—may correlate with the trade environment and affect banks’ lending decisions during the sample period. Our main results hold up when controlling for bank cyclicality, for bank exposures to tradable-goods producing sectors and to firms integrated in global value chains (arguably more exposed to exchange rate fluctuations), or when dropping oil companies from the sample (as the oil sector experienced a protracted credit contraction starting in 2015). 5

Additional results and alternative methodological choices further support our baseline findings. We show our results are not limited to the standard terms of loan contracts—volumes, spreads, andmaturities—butalsoextendtoothermargins, withmoreexposedbanksconsistentlytightening collateral requirements on loans to all borrowers compared to other banks. Finally, the baseline findings are invariant to specification changes such as (a) including no fixed effects; (b) including loan-type×quarter and firm×loan-type×quarter fixed effects for trade finance and other loans; (c) using a weighted-least-squares estimation that accounts for variations in the precision of sectoral estimates of trade uncertainty; and (d) varying the period of analysis to allow for potential anticipation effects of the Trade War. Related literature Our paper contributes to several strands of literature. Prior studies provide evidence that banks facilitating international trade amplify the effects of trade shocks on firms and households (Amiti and Weinstein, 2011; Niepmann and Schmidt-Eisenlohr, 2017a,b; Niepmann, 2015; Michalski and Ors, 2012; Paravisini et al., 2023). Our focus is instead on the direction of linkage from trade to banks, which has received little attention. Federico et al. (2020) document that policy actions associated with China’s accession to the World Trade Organization in 2001 had sizeable effects on bank loan supply to Italian firms. The authors find that endogenous financial frictions arise as a result of the trade shock’s negative effects on bank loan portfolios. Hankins et al. (2022) examine the effects of metal and steel tariffs enacted in 2018 on the supply of auto loans by U.S. finance companies and document negative spillover effects of these policies on consumer credit. Our contribution emphasizes the effects of trade uncertainty on bank commercial lending, and establishes a rich set of mechanisms underlying the real consequences of the credit supply response. Ourworkalsorelatestotheliteratureontherealandfinancialeffectsofuncertainty(Kavianiet al.,2020;Bergeretal.,2020;Hustedetal.,2020;Bakeretal.,2016;Bloom,2014;Buchetal.,2015). Global banks play an important role in the international transmission of financial stresses through lendingandliquidityflows(AmitiandWeinstein,2018;DeHaasandVanHoren,2013;Cetorelliand Goldberg, 2012; Schnabl, 2012; Peek and Rosengren, 2000). Some papers document consequences of uncertainty for bank lending (Crozet et al., 2022; Jasova et al., 2021; Wu and Suardi, 2021; Soto, 2021; Alessandri and Bottero, 2020; Bordo et al., 2016; Valencia, 2017), while others relate 6

uncertainty to global liquidity or capital flows (Rey, 2015; Avdjiev et al., 2020; Kalemli-O¨zcan and Kwak, 2020). The latter literature emphasizes different reasons why aggregate risk conditions may affectbankcredit, includingthroughbanks’value-at-riskconstraintsandleverage(BrunoandShin, 2015). Relative to this strand of literature, we focus on a specific type of uncertainty—around the trade environment—with potentially crucial implications for the global activities of banks and the integration of trade and finance. Trade uncertainty differs from aggregate uncertainty because of itssectoralandgeographicspecificity, whichallowsustodelvedeeperintothemechanismsatwork. Beyond international trade and uncertainty, our paper also speaks to the literature on bankintermediated spillovers of sectoral shocks to broader groups of borrowers (see Gilje et al., 2016; Cort´es and Strahan, 2017; Huber, 2018; Dell’Ariccia et al., 2021; Mayordomo and Rachedi, 2022, among others). For instance, Galaasen et al. (2021) use administrative data from Norway to show that granular credit shocks to firm balance sheets can have large and significant effects on portfolio level return on bank loans, and pass through to non-granular firms. Mart´ın et al. (2021) document a crowding-out effect of the pre-GFC housing boom in Spain on bank commercial credit to other sectors. Many studies in this literature trace the effects of shocks to bank assets and lending opportunities to the real economy. Our contribution is to examine the effects of a sudden and unanticipated increase in sectoral uncertainty to firms in sectors experiencing both high and low changes in uncertainty, and to document the key mechanisms explaining bank lending behaviors. Finally, our work builds on the insights of a growing literature on the economic effects of trade wars, which has a particular emphasis on U.S.-China trade relations. Evidence has been building on the real effects of the 2018–2019 tariffs (Handley and Limao, 2017; Caldara et al., 2020; Novy and Taylor, 2020; Fajgelbaum et al., 2023) and supply chain disruptions (Schiller, 2017; Huang et al., 2019; Amiti et al., 2019; Grossman et al., 2023, see Antr`as and Chor (2022) for a survey). Research documents almost complete pass-through of the tariff burden to U.S. prices (Amiti et al., 2019; Cavallo et al., 2021) and adverse effects on consumption (Waugh, 2019), investment (Amiti et al., 2020), and employment (Flaaen and Pierce, 2019). Our focus on a banking channel of trade uncertainty transmission emphasizes that, importantly, the effects of international trade uncertainty come on top of the documented effects of tariffs. The paper proceeds as follows. Section 2 describes data sources and our approach to constructing measures of bank exposure to trade uncertainty. Section 3 develops the conceptual framework 7

and offers our three conjectures on the mechanisms for and the consequences of bank credit supply adjustment following a rise in trade uncertainty. Section 4 discusses the headline results, placebo and additional identification tests, the evidence on the mechanisms, and real effects for borrowing firms. Section 5 presents additional tests that entertain and rule out alternative explanations and additional measures of exposure to uncertainty. Section 6 concludes. 2 Data and Bank Exposure to Trade Uncertainty 2.1 The U.S. “Credit Register” Our empirical tests require representative and detailed information on the terms of commercial loans for lenders and borrowers. To this end, we rely on micro-level bank data akin to a credit register. Our main data source contains information at the loan level and comes from the FR Y- 14Q H1 “Wholesale credit schedule” (see here for more details). These data are collected quarterly from U.S. and foreign Bank Holding Companies (henceforth BHCs, which we refer to as banks throughout the paper for simplicity) as part of the annual Dodd-Frank Stress Test (DFAST) and Comprehensive Capital Analysis and Review (CCAR). As banking organizations with assets above $50 billion were required to report these schedules during our sample period, these data cover the near-universe of commercial loans from large U.S. banks, which account for three-quarters of outstanding loan balances (Favara et al., 2021) and close to 90% of total banking sector assets (Frame et al., 2023). The reporting panel of banks fluctuates between 30 and 35 banks between 2016:Q1 and 2019:Q4. The FR Y-14Q data set contains loan-level information on commercial and industrial loans to domestic borrowers held by reporting banks. We use information on the value of loans outstanding to non-financial firms (firms in the utilities and financial sectors are excluded from the sample). We observe other characteristics of the loans, such as the type of loan (e.g., line of credit or term loan) and loan purpose (e.g., trade finance loan, etc.),4 interest rates, maturity, collateral requirements (whether the loan is secured), and collateral type (fixed assets and real estate, cash, accounts receivables and inventory, blanket liens). For each loan, banks report their own estimates of the probability of default over a one-year horizon, computed in line with the Basel II guidelines. 4We classify credit facilities with purpose “Trade financing” as trade finance loans. 8

Borrower-specific probability of default is derived from internal risk ratings-based models approved by supervisors. In addition, banks report a wide range of annual borrower characteristics such as total assets, profitability, cash holdings, tangibility, sales revenue, and total debt. The vast majority of the bank borrowers in the data set, which account for 64% of non-financial business debt liabilities and 80% of U.S. output (Caglio et al., 2021), are privately-held firms. We merge the loan-level data with quarterly bank balance sheet and income statement items for each bank from form FR Y-9C. Descriptive statistics for the loans, banks, and firms in our main regression sample are shown in Table 1. The median loan in our sample has a size of $10 million and a spread of 185 bps (over the prime bank rate or LIBOR). Median loan growth across bank-firm pairs in the regression sample, computed relative to the start of the sample period (2016:Q1), is 0% (average growth is -23% for multi-lender firms and 1.1% for single-lender firms). In aggregate bank balance sheet data, average C&Iloangrowthatthe39largestBHCswas3.1%during2016:Q1-2019:Q4. Medianremainingtime to maturity is 2.5 years, 13.4% of loans are demandable (with no specified maturity), and 7.2% of loans are new originations. Almost 60% of observations are credit lines and 2.4% are trade finance loans. There is significant variation in bank capital as measured by the ratio of common equity to total assets, which has an average of 11.5%. Close to 70% of firms belong to low-uncertainty sectors and 10% are publicly-traded. 2.2 Bank Exposure to Trade Uncertainty Akeyelementofouranalysisisthemeasureofbankexposuretotradeuncertainty. Constructionof this variable proceeds in three steps. First, we use estimates of firm-level trade risk and uncertainty forU.S.firmsfromHassanetal.(2019)toobtaintradeuncertaintymeasuresthatvaryatthesector level. Second, we assign these sector-level uncertainty measures to borrowers in the credit register based on their sectoral classification. Third, we aggregate this information at the bank level using banks’ initial loan shares to firms across sectors. Hassanetal.(2019)relyontextualanalysisthatextractsinformationonthefrequencyofterms concerning trade and uncertainty for publicly-listed firms. This approach leverages computational linguistics tools applied to the transcripts of quarterly earnings conference calls to construct measures of risks facing listed firms. Textual analysis allows the authors to calculate the share of 9

earnings calls language that identifies risks associated with specific topics. Key for our analysis is one such topic—trade risk and uncertainty—that captures discussions related to international trade and potential risk and uncertainty jointly (e.g., the words “tariffs” and “uncertain” occurring in a call).5 Figure 1 shows the evolution of this measure between 2014 and 2019. As seen in panel A, trade uncertainty spikes in 2018 and remains high through 2019. Moreover, as shown in panel B, trade uncertainty rises considerably more than other sectoral risks such as those classified as political, environmental, or economic.6 Caldara et al. (2020) examine the evolution of trade policy uncertainty using newspaper coverage and earnings-calls-based measures (see Figure OA-2) and confirm a sharp increase in uncertainty after 2017, which they link to concerns about “supply chain disruptions” and “higher costs of raw materials” amid hikes in tariff rates. They also argue that the main source of risks in 2017, when trade uncertainty indexes increase notably for the first time, was related to changes in corporate tax policy, notably the 2017 border tax adjustment proposal. Combined with the fact that increases in tariffs by the United States on its major trading partners started in February 2018 and paused in December 2019 with the U.S.-China agreement on the Phase One deal, we settle on the period between 2018:Q1 and 2019:Q4 as the period of “heightened trade uncertainty” or Trade War for purposes of the analysis. Benguria et al. (2022) and Grossman etal.(2023),amongothers,arguethatthe2018–2019cycleofretaliatorytradeactionsdramatically increased uncertainty in trade-oriented sectors by reversing decades of trade liberalization.7 Firm-level indicators of trade uncertainty are available only for listed firms in the Hassan et al. (2019) data set, while the credit register covers a large set of both public and private firms. Therefore, in the first step we merge the uncertainty measures to the credit register by sector. We obtainaverageuncertaintyatthe3-digitNAICSsectorlevelastheaverageoffirm-leveluncertainty 5The top biagrams for trade in the training library used by the authors include trade agreement, barriers, free trade, markets, trade relations, duties, globalization, labor standards, and policy objectives. Bigrams for risk and uncertaintyincluderisk/risks,uncertainty,variable,change,possibility,uncertain/uncertainty,doubt,prospect,variability, exposed, probability, unknown, unpredictable, and speculative, among others. 6Figure OA-1 depicts trade uncertainty relative to overall, political, and nonpolitical sentiment and shows that while trade uncertainty rose materially during 2018–2019, increases in measures of sentiment were more muted. 7Our choice of Trade War period is also corroborated by the findings of Hassan et al. (2021), who use textual analysis of earnings calls for firms worldwide to identify marked increases in perceived country risk. Their analysis identifies a spike in country risk for China during the U.S.-China trade tensions between 2018:Q4 and 2019:Q4. Furthermore, given that trade uncertainty starts rising in 2017, we check that our headline results are robust when we drop data for the year 2017 from the analysis and compare lending outcomes in 2015–2016 versus 2018–2019. 10

acrossfirmsineachsector.8 Fortheimputationofaverageuncertaintyfromlistedfirmstoallfirms, we rely on recent evidence that listed firms’ equity valuations strongly predict economic activity at the industry level, especially for manufacturing sectors (Flynn and Ghent, 2022), which are overrepresented in banks’ loan portfolios. We then calculate the change in average trade uncertainty for each sector between 2016–2017 (before the Trade War) and 2018–2019 (during the Trade War). Firms in the manufacturing and transportation sectors account for a larger fraction of those that are most affected.9 Critical for our identification strategy is assessing whether the firms in sectors that were more affected by rising trade uncertainty had similar performance relative to firms that were in less affected sectors before the Trade War. To test this identifying assumption, we rank sectors by this measure and construct an indicator variable for those sectors above the 75th percentile of the distribution of change in trade uncertainty. We then classify firms in the top quartile sectors as “high-uncertainty” firms and test whether the sales growth of these firms differed systematically fromthatofotherfirmsbefore2018. Theresultsofthis“paralleltrends”testareshowninFigure2, where we find no statistically significant difference in the sales growth of firms in high- and lowuncertainty sectors in 2016 and 2017, but a significant difference in the years thereafter. This figure suggests that the performance of firms exposed to large increases in uncertainty was not different before the Trade War and therefore that their performance during the Trade War cannot be attributed to differential pre-existing trends. The second step to construct a measure of bank exposure to trade uncertainty involves merging the sectoral measures of trade uncertainty with banks’ initial loan exposures to individual sectors. Theinitialbankshareofloanstofirmsinindividualsectorsiscomputedrelativetototalbankloans andistheaverageover2014–2015. Thisaveragehelps(a)toconstructapre-determinedmeasureof bankexposure(beforethestartofthesampleperiod)thatislikelyunrelatedtoeconomicconditions during the Trade War and (b) to avoid relying on a single year of data which may result in a noisy measure. Combining these two inputs yields a continuous measure of bank-level exposure to trade 8For this aggregation we use sectoral classifications from S&P Compustat for the firms. We aggregate the firmleveluncertaintyinformationatthe3-digitNAICSlevelandnotamoregranularleveltohavesufficientfirmsineach sectorfortheaveragetobereliable. Wecheckthatourresultsarerobusttoaccountingforthesparsefirm-leveldata in some sectors with a weighted least squares estimation in the Online Appendix. 9Figure OA-3 reports the change in trade uncertainty across all sectors in our sample and Table OA-1 lists the most and least affected sectors. 11

uncertainty for bank-sector pair {b,s} defined as: (cid:88) Bank ExposureU = ω ×∆Uncertainty , b,s bs′,2014-15 s′,2018-19/2016-17 s′̸=s where s′ represents any given sector except sector s. The exposure measure thus leaves out direct information on uncertainty for sector s and instead creates a loan share-weighted sum of changes in uncertainty of all other sectors that bank b lends to, where the term ω captures the share bs′,2014–15 of the sum of loans to firms in sector s′ in bank b’s loan portfolio and ∆Uncertainty s′,2018–19/2016–17 measures the change in trade uncertainty for sector s′.10 In the cross-section of banks, the average and median bank loan exposures to trade uncertainty are positive, which means that the average bank has an initial loan portfolio that is tilted towards sectors facing higher trade uncertainty during the sample period (see Table 1). It is important for our identification strategy to check if the bank exposure to trade uncertainty is correlated with bank characteristics that may influence lending decisions. The identifying assumption for unbiased estimation of the effect of bank exposure to trade uncertainty on credit is that this exposure is not systematically correlated with other bank-level shocks. That is, banks should not sort into certain sectors such that unobserved bank-level shocks are correlated to both a decline in credit supply and increases in uncertainty in those same sectors (Borusyak et al., 2022). To check this assumption, in Table OA-2 we regress bank exposure to trade uncertainty on bank size, leverage, the share of core deposits in liabilities, and sectoral specialization. The regression are run in the yearly cross-sections of banks as well as in a panel that stacks the data across all the years in the sample period. We find that the bank exposure measure is unrelated to these characteristics, nevertheless, we include them as controls in the baseline specification. 3 Conceptual Framework In a setting when trade uncertainty potentially affects a large portion of bank borrowers across sectors, banks may have to decide between providing additional lending to these affected borrowers or scaling back their exposures. This choice depends on two opposing forces. On the one hand, 10This approach for generating the bank-sector exposure measure closely follows the “leave-one-sector-out” approach suggested in Borusyak et al. (2022) and implemented, for instance, in Federico et al. (2020). 12

bank specialization in lending generates incentives for banks to lend more to their main borrowers in periods of stress to limitpotential losses from defaults (Favara andGianetti, 2017; Giannetti and Saidi, 2019; Agarwal et al., 2020; Blickle et al., 2023). Therefore, firms outside distressed sectors may see a reduction in credit from specialized banks if those banks allocate more lending capacity to core sectors. On the other hand, banks may react to higher uncertainty by adopting the behaviors of nonfinancial firms predicted by corporate finance theory. Studies of investment under uncertainty highlighthowtheirreversiblefeaturesoffixedassetpurchasesaffectthetimingofthoseinvestments in periods of uncertainty (Bernanke, 1983; Pindyck, 1991; Caballero and Pindyck, 1992; Dixit and Pindyck,1994). Thesestudiesestablishanegativelinkbetweenuncertaintyandfirminvestment,as firms tend to postpone investment until uncertainty about future conditions declines (Bloom et al., 2007; Bloom, 2009; Handley and Limao, 2015). In a similar vein, banks may react to heightened uncertainty by deciding to pull back from lending or by adopting risk management tools that increase the flexibility of loan agreements. For instance, they may reduce loan maturities so as to evaluate borrower creditworthiness more frequently (typically borrowers have annual reviews) and modify loan contracting features while uncertainty persists. An increase in trade uncertainty can generate different bank behaviors compared to a sectoral shock such an economic, financial, or policy event that affects realized and expected returns to lending to that particular sector (a first-moment effect). This is because uncertainty increases the dispersion of returns to lending and raises the prospect of future balance sheet gains or losses without them necessarily materializing. As a result, it is possible that banks’ reactions to a rise in uncertainty(asecond-momenteffect)differfromreactionstochangesinactualorexpectedreturns. The literature shows that banks often respond to adverse sectoral shocks by shrinking exposures to affected sectors and reallocating lending capacity to less-affected sectors.11 By contrast, increased uncertainty may induce banks to curtail loan exposures to all borrowers because it may be difficult to assess the range and magnitude of potential gains or losses from lending and their effects on capital ratios. Indeed, standard portfolio allocation models predict that an increase in volatility of asset payoffs leads to a reduction in the risky portfolio share (Markowitz, 1952). 11Forevidenceonbanklendingresponsestoadverseshocksinrealestatemarkets,theoilindustry,ortrade-intensive sectors,see,amongothers,PeekandRosengren(2000);Federicoetal.(2020);Bidderetal.(2021);Caoetal.(2022); Federico et al. (2023). 13

Our research design is structured around three conjectures that we discuss below. The first conjecture focuses on the direction of the effect of trade uncertainty on bank lending. Conjecture 1: Banks respond to an increase in trade uncertainty by reducing credit supply across all borrowers. This conjecture posits that, once we control for firm credit demand, banks that are more exposed to trade uncertainty will have behaviors similar to those observed in the investment-underuncertaintyliterature(DixitandPindyck,1994). Thealternativetothisconjecturewouldbebanks lending less only to one group of borrowers and reallocating lending capacity to other groups of borrowers. Banks may also try to safeguard profitability by re-balancing portfolios toward asset classes that are less affected by the rise in uncertainty, such as securities investments. Delving deeper into mechanisms, banks’ adoption of a wait-and-see attitude in the face of increased uncertainty may also manifest along other dimensions. For instance, as forecasting the distribution of returns from lending across sectors becomes more difficult, banks may assess all of their borrowers as riskier and assign them higher probability of default. They may also adopt other risk mitigation strategies such as shortening loan maturities and extending more demandable loans that can be called back on a short notice. While it might be difficult to pick “winners and losers” from the Trade War, banks may nevertheless curtail exposures to borrowers whom they perceive as likely to be negatively affected by tariffs. Such borrowers may include, for instance, firms in sectors less protected by trade policy or firms more dependent on imported intermediate goods and potentially more likely to experience higher cost of imported inputs. Additionally, credit supply adjustments could be associated with financial constraints at banks. Banks’expectationsoffuturebalancesheetgainsorlossesmayplayaroleintheirlendingdecisions particularly when the bank anticipate losses that erode capital and constrain new lending. The transmission of uncertainty shocks to lending decisions may be driven by the external finance premium for banks, which relates banks’ financial health to their ability to raise external financing (Bernanke, 2007). Banks with smaller capital buffers and greater capital erosion in bad states of the world (as reflected, for instance, in severely adverse scenarios of stress tests) could face tighter financing constraints and would thus be less able or willing to bear risks as uncertainty rises. This mechanism suggests stronger credit supply contractions for banks exposed to uncertainty when 14

they have lower levels of current or stressed capital. Concretely, we examine evidence for these channels within the following conjecture: Conjecture 2: Two channels drive bank credit supply responses to trade uncertainty exposures. a) Consistent with real-options theory, exposed banks adopt a wait-and-see attitude, reducing credit supply, downgrading the expected creditworthiness of firms, shortening the maturity of loans, and paring back exposure to riskier firms. b) Consistent with financial frictions, lower capitalized banks exposed to trade uncertainty contract lending by more. A final conjecture pertains to the real implications for the firms that borrow from banks exposed to trade uncertainty. This issue is especially relevant in a world with frictions that limit firms’ ability to substitute their credit financing across banks or to other sources of funds. An overall decline in credit supply to borrowing firms may occur, for instance, when the borrowers experience hold-up problems because of the information monopolies of their banks (Rajan, 1992). An extensive literature documents the close link between banks’ financial health and the performance of their bank-dependent borrowers (see, e.g., Slovin et al. (1993); Kang and Stulz (2000); Chava and Purnanandam (2011); Chodorow-Reich (2014); Schwert (2018)). Accordingly, we conjecture the following: Conjecture 3: Real outcomes are worse for firms that borrow from banks with higher exposures to trade uncertainty. 4 Main Results This section presents the empirical specifications and results of the estimations testing the conjectures. The results first assess whether trade uncertainty affects the supply of bank credit to U.S. firms (Sections 4.1 and 4.2). Then we test the potential mechanisms underlying the link between tradeuncertaintyandthecontractionofbanklending(Section4.3)andweexaminehowbanksmay reallocate their assets when faced with increased trade uncertainty. Lastly, we present evidence of real effects for borrowing firms (Section 4.4). 15

4.1 Trade uncertainty and bank credit supply Specification According to Conjecture 1, an increase in bank exposure to trade uncertainty reduces the supply of bank credit broadly across firms. We test this conjecture by estimating a difference-in-differences specification linking trade uncertainty to lending outcomes on the intensive margin: y = β Bank Exposure ×Post +β X +β X ×Post +γ +δ +e , (1) b,i,s,t 1 b,s t 2 b,t−1 3 b,t−1 t i,t b,i b,i,s,t where the dependent variable y in the baseline regressions is defined as either the loan growth b,i,s,t (the growth of loan commitments from bank b to firm i in sector s relative to the beginning of the sample period) or the corresponding loan spread. The sample period includes all loans between 2016:Q1and2019:Q4. WedefinePost asanindicatorvariableequaltooneduring2018:Q1through t 2019:Q4, and zero during 2016:Q1 through 2017:Q4. Bank Exposure is our measure of bank b,s exposure to trade uncertainty as defined in Section 2.2. The coefficient of interest is β . A negative 1 value for β in the loan growth specification (and a positive one in the loan spread specification) 1 would provide evidence supporting the conjecture. We examine this specification in the full sample of firms and separately for low-uncertainty firms. Coefficients are estimated with Ordinary Least Squares (OLS) and standard errors are double clustered by bank-firm and quarter. Specification (1) includes firm×quarter fixed effects (γ ) that i,t allowustokeeploandemandconstantatthefirmlevelovertime,andhenceexaminethedifferential lending behavior of banks with varying degrees of exposure to uncertainty vis-`a-vis a given firm in a given year. We also report specifications with firm×bank fixed effects (δ ), which allow for the b,i possibility that loan demand is specific to the bank-firm pair. This may be the case when banks specialize in certain types of credit (such as trade credit) or certain types of borrowers (such as large exporters)—see, e.g., Ivashina et al. (2021) and Paravisini et al. (2023). These fixed effects aim to allay concerns that the coefficient on bank exposure, β , captures the effects of firm-specific 1 factors such as credit demand, as opposed to banks’ supply-side lending decisions. Specificationsincludestandarddeterminantsofbanklendingdecisions(X ),suchas(lagged) b,t−1 size(log-totalassets),capital(commonequitydividedbytotalassets),andcoredeposits(inpercent of total liabilities). Given that bank exposure to trade uncertainty may capture some form of 16

lending specialization, we also include a bank specialization measure that identifies banks with outsized exposures to individual sectors.12 All the control variables enter both in levels (X ) b,t−1 as well as interacted with the Post dummy variable (X × Post ) to make sure that the β t b,t−1 t 1 coefficient is not contaminated by bank size, capital, deposit funding, or specialization. Baseline: Intensive margin Table 2 reports estimates based on specification (1) estimated for the full sample of borrowers (panel A) and for low-uncertainty firms (panel B). In columns 1 and 2, the coefficient of interest on the difference-in-differences term Bank Exposure ×Post is b,s t negative and statistically significant at conventional levels, and shows that rising trade uncertainty isassociatedwithlowerloangrowthformoreexposedbanks,bothforthefullsampleoffirmsandfor low-uncertainty firms. The coefficient magnitudes are economically sizeable. Using the coefficients in column 2 where we use the most stringent set of fixed effects, an increase in bank exposure to trade uncertainty by one standard deviation (0.25) is associated with an average decline in loan growth by between 2.6 and 2.8 ppts (relative to the median growth rate of loan commitments of 0% over the sample period). The estimates in columns 3 and 4 of Table 2 show that banks with higher exposure to trade uncertainty charge higher loan spreads than other banks. The coefficient estimates are statically significant at conventional levels and economically meaningful. Using the coefficients in column 4 where we again use the most stringent set of fixed effects, an increase in bank exposure to trade uncertainty by one standard deviation leads to an average increase in lending spreads of 6.5 and 7.1 bps for all and low-uncertainty firms, respectively. Although these changes are relatively small compared to the median spread in the sample (185 bps), the directional movement supports the conjecture that the supply of credit from banks exposed to trade uncertainty shifted inward.13 Overall, our baseline results suggest that trade uncertainty is associated with a contraction in 12We obtain this measure as follows. We start by calculating the shares of loans for each bank in our sample at end-2017 to individual sectors using the 3-digit NAICS classification. Then, we calculate the 75th percentile of thatdistributionplus 1.5 interquantileranges. Thebankspecialization variableisdefinedasa dummyvariablethat takes a value of one for bank-sector observations for which the share exceeds that threshold—these are the banks “specialized” in that particular sector—and zero otherwise. As shown in Table OA-2, this measure is uncorrelated with bank exposure to trade uncertainty. Furthermore, our baseline results are virtually identical if we use the specialization measure computed as the average of 2014 and 2015 instead of the one for 2017. 13The credit register data further allows us to explore whether exposed banks are more likely to tighten collateral requirementstohedgeagainstpotentialloanlosses. Repeatingthebaselineregressionswithadummyvariabletaking value one for secured loans (corresponding to about three-quarters of all loans in the sample) as the dependent variable,wefindthat,indeed,moreexposedbanksaremorelikelytorequireloanriskmitigantsduringtheperiodof heightened trade uncertainty (see Table OA-3). 17

credit supply. This result holds for both loan volumes and loan spreads, and is present in the intensive margin of lending (we explore the extensive margin and additional features of bank-level results in the next sections). The Online Appendix presents several robustness checks on our baseline results, including alternative samples of firms, fixed effects, and estimation methods. 4.2 Parallel trends and threats to identification Parallel trends Akeyidentifyingassumptionbehindtheunbiasedestimationofβ isthatbanks 1 made similar lending decisions before the 2018–2019 period regardless of their exposure to sectors lateraffectedbyrisingtradeuncertainty. Totestthevalidityofthisassumption,wefirstexplorethe dynamic difference-in-differences effects in our baseline regressions. Figure 3 plots the individual coefficientsfortheinteractiontermbetweenthebankexposuremeasureandquarterlydummiesover the sample period and their confidence intervals. The coefficients on the difference-in-differences termbeforethetradeuncertaintyshockarestatisticallyindistinguishablefromzeroinmostperiods before the Trade War, suggesting a lack of anticipation effects and pre-shock lending adjustment by banks (either in volume or spreads of loans). By contrast, during 2018–2019 we observe a statistically significant contraction in loan volumes (panel A) and a rise in spreads (panel B), both of which become stronger over time.14 Wealsotestthevalidityoftheparalleltrendsassumptionwithformalplacebotests. Thesetests are meant to ensure that bank exposure to trade uncertainty does not capture the effects of bank unobservables—if it did, then we would find patterns similar to our baseline results in previous periods. As shown in Table OA-5, when we shift the sample period back by one or two years, we find no systematic association between bank exposure to trade uncertainty and lending outcomes. These findings allay potential concerns that our baseline results capture the effects of unobserved bank characteristics rather than those of trade uncertainty itself. First versus second moment effects One concern might be that results are driven by the effect of the Trade War on realized or expected returns on loans (first-moment effect) instead of 14We run an additional test to check for evidence that banks may have anticipated the Trade War and started adjustingtheirlendingexposuresaheadoftime. Tothisend,wedroploanobservationsfrom2017fromourregression sample and run the regressions by comparing lending outcomes during 2015–2016 versus 2018–2019. The estimates arereportedinTableOA-4andshowthatthebaselineresultsremainunchanged,suggestingthatbanksdidnotreact in anticipation of the heightened uncertainty associated with the Trade War. 18

the uncertainty regarding expected returns (second-moment effect). We approach this issue with two new specifications that control for (a) bank exposure to changes in actual trade policy (that is, the loan share to tariffs-hit sectors) or (b) bank exposure to changes in overall sentiment. First, we construct a measure of bank exposure to sectoral tariff changes as the end-2017 as the share of loan commitments to firms in sectors that received tariffs during 2018–2019, sourced from Flaaen and Pierce (2019). The binned scatterplot in Figure OA-4 depicts a positive correlation between the exposure of banks to trade uncertainty and their exposure to sectors that experienced tariff changes during the Trade War. Similarly, Benguria et al. (2022) show that textual measures of exposure to trade policy uncertainty are highly correlated with actual trade war exposures in the cross-section of firms. We then include this measure in a horse-race regression with bank exposure to trade uncertainty, as shown in panel A of Table 3. The point estimates on trade uncertainty exposure barely change relative to the baseline results in Table 2. Second,weproxyforbanks’perceptionsoffutureloanreturnswithsentimentmeasurescaptured in firms’ earnings call transcripts. In Figure OA-1 we plot three measures of sentiment—overall, politicalandnonpoliticalsentiment—overthesampleperiodalongwiththetheaggregatetraderisk measure. TheplotshowsthatchangesinsentimentduringtheTradeWarweresmallrelativetothe rise in trade uncertainty, making it unlikely that it would confound our main results. To formally test whether sentiment matters, we construct a measure of bank exposure to overall sentiment similar to that calculated for trade uncertainty. As shown in panel B of Table 3, the coefficients on this measure are statistically insignificant, while the coefficients on bank exposure to trade uncertainty are statistically significant at the conventional levels and robust across specifications. 4.3 Trade uncertainty and evidence on channels of transmission to lending Beyond the baseline effects of trade uncertainty on loan quantities and prices, we investigate the potential channels that may drive banks to change lending behaviors as trade uncertainty rises. In particular, asnotedinConjecture2, weexplorewhetherthisreactionisexplainedbyacombination of “wait-and-see” and financial constraints channels. 19

4.3.1 Wait-and-see behaviors Westartbyassessingwhetherbanksadjusttheirlendingactivitiesconsistentwitha“wait-and-see” approach. It is difficult to directly test for this type of behavior, therefore we compile a portfolio of evidence suggestive of this channel using information on the extensive margin of lending, additional lending terms, banks’ own internal risk assessments, and heterogeneity in banks’ perceptions of which sets of firms are likely to lose or gain from the Trade War. Extensive margin The first tests focus on the extensive margin of adjustment. A wait-and-see approach would predict that banks more exposed to trade uncertainty postpone or stop lending to some of their borrowers. In Table 4 we report regression results for new loan originations, where all specifications include the most restrictive set of fixed effects and we again display results for all and for low-uncertainty firms. Specifications in columns 1–2 vs. 3–4 differ on the construction of the dependent variable and the aggregation level of the data; specifically, in columns 1–2 we run regressions using loan-level data with a loan-origination dummy as the dependent variable, and thus capture a “pure” extensive margin effect. By contrast, in columns 3–4 we run regressions aggregating the data up to the bank-firm level and specifying the dependent variable as the share of new loan volume in total loans outstanding, thus capturing a mixture of extensive and intensive margin effects. Across specifications, the estimated coefficient on the difference-in-differences term is negative and statistically significant, implying that bank exposure to trade uncertainty affects the extensive margin of lending as well. In terms of economic relevance, a one standard deviation increase in banks’ exposure to trade uncertainty is associated with a probability of a new loan origination that is lower by approximately 0.5%. This contrasts with an unconditional probability of a new loan origination of about 5% over our sample period. Banks’ private risk assessments and loan maturities Totestwhetherexposedbanksadjust their own internal risk assessments and other lending terms as a result of rising trade uncertainty, we estimate specification (1) but with distinct dependent variables. First, we examine the link between banks’ exposure to trade uncertainty and their forward-looking assessments of borrower creditworthiness based on borrower-level probabilities of default over a one-year horizon. Second, we assess whether exposed bank reduce the maturities of their loans, which could be 20

a sign they are decreasing the “irreversibility” of loan commitments (alternatively, increasing the frequency with which they conduct borrower reviews, and making loan modifications). In this case, the dependent variables are the remaining time to maturity (the median maturity and time to maturity of loans in the credit register are 5 and 2.5 years, respectively). We also use an indicator variable that categorizes loans as demandable as an additional dependent variable. A demandable loan allows the lender to react swiftly to any concerns about the firm and recall the loan. Once notified, the borrower must repay the principal and any associated interest. In specifications that examine loan maturities, we follow Li et al. (2023) and include the following loan controls: the log of loan size (total loan commitment) and dummy variables for floating rate loans, secured loans, and loans with prepayment penalty. Conjecture 2 would be supported by a negative coefficient on the difference-in-differences term for loan maturity and a positive one for demandable loans. Results are reported in Table 5. Estimates in columns 1–2 suggest that banks with greater exposure to trade uncertainty assess their borrowers as having increased default risk, suggesting potential concerns of higher credit risk and potential balance sheet losses. This finding is consistent with anecdotal evidence drawn from the Federal Reserve’s Senior Loan Officer Opinion Survey in the first quarter of 2019, according to which large and regional U.S. banks with significant loan commitments to firms exposed to international developments (of at least 40% of the loan book) expected the outlook for loan losses to deteriorate over the course of 2019. These concerns do not appear to have materialized. In unreported regressions using bank balance sheet data over 2016:Q1–2019:Q4, we examine the dynamics of loan loss reserves, nonperforming loans, and net charge-offs, and do not find any evidence of a deterioration in loan performance nor of higher provisioning at exposed banks. Incolumns3–6ofTable5weexaminetheeffectsonloanmaturities. Theestimatessuggestthat more exposed banks shorten the maturity of loans more than other banks, including for the lowuncertainty firms (columns 3–4). Moreover, estimates in columns 5–6 indicate that exposed banks are more likely to grant demandable loans, which increase lenders’ flexibility when borrowers show signs of distress. Overall, these results corroborate Conjecture 2 and suggest that, as uncertainty rises, more exposed banks try to increase the flexibility of their lending by shortening the maturity of loan contracts and more frequently re-assessing the creditworthiness of their borrowers. 21

Heterogeneous effects across borrowers We check if exposed banks curtail exposures relatively more to those borrowers whom they perceive as likely to be adversely affected by trade tariffs and hence riskier ex ante. We focus on firms in manufacturing sectors that receive low import protection via tariffs, as they are more likely to experience a worsening of growth prospects and higher credit risk during the Trade War. Furthermore, we identify firms for which tariffs could generate a substantial increase in production costs, that is, firms that rely heavily on imported intermediate inputs and are thus more integrated in global value chains. We use the following specification: (cid:88) y = β Bank Exposure ×Post ×Firm Type b,i,s,t τ b,s t i,τ τ=1,2 (2) +β X +β X ×Post +γ +δ +e , 3 b,t−1 4 b,t−1 t i,t b,i b,i,s,t where τ = 1 indicates a Low-Protection Firm, τ = 2 indicates a High-Protection Firm, and the coefficients of interest are β and β . Low-protection firms as those firms in manufacturing sectors 1 2 where the new import tariff rate as a share of consumption falls below the 75th percentile of the distribution (at the 4-digit NAICS classification level for 2018), and, respectively, firms in sectors with above-median total imports as a share of industry output (at the 3-digit NAICS classification level for 2014–2015). The results for these tests are reported in Table 6. Panel A divides firms into low- and highimport protection bins and panel B splits the borrowers into high- and low-import dependence.15 Theestimatessuggestthatbanksexposedtotradeuncertaintyactivelymanagetheirloanportfolios by reducing risks. As seen in panel A, firms in sectors with low tariff protection experience a larger decline in loan growth from exposed banks relative to those banks’ high-protection borrowers (the p-values of one-sided tests in columns 1–2 indicate that the differential contraction in loan growth at exposed banks is statistically significant only for low-protection firms). Moreover, the estimates in panel B show that firms that are more dependent on imports for production experience lower loan growth compared to other firms borrowing from the same banks. However, the effects on loan spreads are not significantly different across low- and high-protection sectors. Overall, the estimation results suggest that uncertainty-exposed banks de-risk their loan portfolios on the loan 15The tariff data are only available for the manufacturing sectors and thus decreases the regression sample size relative to our baseline specifications. 22

volume margin. Altogether, these findings support the first part of Conjecture 2, namely, banks exposed to trade uncertainty assess all of their borrowers as being potentially riskier, and attempt to reduce their risky by shortening the maturity of the loans they originate and by curtailing credit to those borrowers that may be more adversely affected by trade developments. 4.3.2 Financial constraints The second part of Conjecture 2 focuses on the role that financial constraints at banks, measured by bank capitalization, play in determining the relation between trade uncertainty and loan supply. Specification To examine the differential effects of trade uncertainty on lending behaviors that relates to bank financial frictions, we focus on two measures of bank capital: (a) common equity to assets(i.e.,thesimpleleverageratio)and(b)theStressedCommonEquityTier1(CET1)ratio(the minimum CET1 ratio estimated under the “Supervisory Severely Adverse” scenario of the Dodd- Frank Act stress test (DFAST)). We test the conjecture with a modified version of specification (1): (cid:88) y = β Bank Exposure ×Post ×Bank Type b,i,s,t τ b,s t b,τ τ=1,2 (3) +β X +β X ×Post +γ +δ +e , 3 b,t−1 4 b,t−1 t i,t b,i b,i,s,t where where τ = 1 indicates a Low-Capital Bank, τ = 2 indicates a High-Capital Bank, and high-capital banks are those with a capital ratio above the 75th percentile of the cross-sectional distribution. Evidence of financial frictions would arise if the coefficient of interest β were greater 1 than β . Additional evidence for this channel could come from shifts in banks’ asset allocations 2 conditional on their exposure to trade uncertainty. Heightened uncertainty could induce banks to reallocate capital to non-lending activities, to shrink their balance sheets, or a combination of strategies. If exposed banks anticipate capital constraints to become more binding, they may exhibit lower risk-appetite and change allocations in favor of safer securities rather than making risky commercial loans. To explore this possibility, we also examine changes in broad balance sheet components by degree of bank exposure to trade uncertainty in a bank-level panel for our sample 23

period (2016:Q1–2019:Q4). Results Table 7 provides the main tests for the financial frictions channel. In panel A we define thecapitalratioasequityoverassets,whilepanelB’smeasureisbasedonthebanks’post-stresstest CET1 capital ratio. Consistent with Conjecture 2, the estimates across all specifications indicate that lower-capital, more constrained banks reduce loan growth and increase loan spreads more than other banks. The results in columns 1–2 show that higher-capital banks do not reduce loan growth while lower-capital banks do. By contrast, exposed banks increase loan spreads regardless of capital level (columns 3–4). P-values of one-sided t-tests suggest that the credit contraction effects are relatively stronger for more constrained banks (and statistically significant at least at the 5% level of significance). These estimates are economically meaningful and shed light on the role of capital in dampening the transmission of real shocks through the banking system. We estimate a version of the model in column 1 in panel A using the capital ratio in levels to assess loan growth at exposed banks at different capitalization levels. In particular, we compare the loan growth for a bank with median capital levels before the Trade War (11.6% at end-2017) with a bank with the median capital level before the GFC (8.5% at end-2007) at median exposure to trade uncertainty (1.77). After the increase in trade uncertainty, the average loan growth of the bank with post-GFC capital levels is almost 7 ppts higher than the bank capitalized at pre-GFC levels. This is a material difference compared to the median growth rate of lending over the sample period and highlights the role of higher capital ratios in enhancing the resilience of banks to uncertainty shocks like the Trade War. Overall, these results suggest lower capacity and willingness to bear risk at lower-capital banks that are exposed to trade uncertainty, which is consistent with a financial frictions channel underpinning our baseline effects. In Table OA-6 we examine asset portfolio re-balancing in the bank-quarter panel, for all banks (panel A) and separately for high vs. low capital banks (panels B and C). Regression results in panel A indicate no effect of bank exposure to trade uncertainty on total bank asset growth (column 1). However, loans as a percentage of total assets fall, which is consistent with the results for commercial loans in the credit register data (column 2). In addition, the share of securities in total assets increase at more exposed banks (with a statistically significant coefficient at 10%) 24

while cash holdings remain unchanged (columns 3–4). These results suggest that banks respond to increases in trade uncertainty by shifting their asset-mix away from risky loans towards safer securities. Furthermore, the estimates in panels B and C indicate that these asset shifting patterns are stronger for lower-capital banks, across both definitions of capital ratio considered. 4.4 Real effects for firms Specification Conjecture 3 posits that the credit supply impact of trade uncertainty will affect firms’ real outcomes. To test for this conjecture, we start by gathering firm financial data in a firm-year panel over 2016–2019 and construct a measure of firm exposure to trade uncertainty via the firm’s relationships with uncertainty-exposed banks. This is a continuous variable representing the average uncertainty exposure of a firm’s lenders, weighted by the share of each lender in total borrowing by that firm (at end-2014), defined as: (cid:88) Firm ExposureU i = ω ib,2014 ×Bank ExposureU b , (4) b where ω is firm i’s beginning-of-sample loan share from each bank b, and ExposureU is bank ib,2014 b b’s total exposure to trade uncertainty (defined as the simple average across sectors of the banksector exposure from the baseline specifications). Then, we use a range of firm-level financial data and the following specification to test for real effects: y = β Firm ExposureU ×Post +β X +β X ×Post +γ +δ +e , (5) i,s,c,t 1 i t 2 i,t−1 3 i,t−1 t i s,c,t i,s,c,t and y refers to a range of firm-level outcomes including total debt growth and the investi,s,c,t ment ratio (capital expenditure divided by lagged fixed assets) for firm i in industry s, located in county c and in year t. We control for a wide range of (lagged) firm characteristics and risk attributes (X ). Following the literature (see, e.g., Fazzari, Hubbard and Petersen, 1987; Leary i,t−1 and Roberts, 2014; Dinlersoz, Kalemli-O¨zcan, Hyatt and Penciakova, 2018), we include firm size (log-assets), liquidity (cash and marketable securities as a share of assets), tangibility (tangible assets as a share of assets), interest coverage ratio (EBITDA/total interest expense), return on assets, and a dummy taking value one for firms with a speculative-grade internal risk rating. We 25

alsocontrolforrealsalesgrowth, aproxyforthedemandandgrowthopportunitiesfacingeachfirm (Whited and Wu, 2006). Specifications include firm fixed effects (γ ) and industry×county×year i fixed effects (δ ) to absorb time-varying shifts in macroeconomic conditions affecting all firms s,c,t in a given industry and county. Once again, specifications consider the sample of all firms, and also the sample of low-uncertainty firms. Values for β coefficient estimates that are negative 1 and statistically significant would provide support for Conjecture 3. In addition to testing for the effect of trade uncertainty on firm outcomes, we interact the difference-in-differences term (Firm ExposureU ×Post ) with two measures of bank dependence: (a) a dummy variable that i t captures whether a firm is private or public, anticipating stronger real effects for private firms to the extent that such firms are more bank-dependent and less able to secure financing in public debt markets, and (b) a dummy variable that takes value one for firms with above-median share of CCAR bank debt (approximated with the sum of utilized loan amounts from the banks in the FR Y-14Q sample). Results Real effects results are presented in Table 8. We run the regressions for all firms in columns 1–2 and low-uncertainty firm in columns 3–4. The estimates suggest that higher firm exposure to trade uncertainty via banks is associated with a contraction in firms’ total debt growth and investment rates. The estimated coefficients on the difference-in-differences term are statisticallysignificantinallspecificationsexceptfortotaldebtgrowthforlow-uncertaintyfirmsincolumn 3. The estimates suggest that firms in borrowing relationships with banks more exposed to trade uncertainty are unable to substitute reduced credit from those banks with other sources of financing, as their total debt growth declines. This credit contraction, in turn, has a material effect on their investment rates. In terms of economic magnitudes, the coefficient estimates in columns 1–2 indicate that an increase in firm exposure to trade uncertainty is associated with a reduction in the growth rate of debt and in the investment rate by 2.4 and 2.7 ppts, respectively. These are sizeable effects given that average debt growth and investment rate over the period are 5.5% and 17.3%. Next, we examine whether bank-dependent firms are more adversely affected. We use two measures of bank dependence. First, we divide firms into those that are publicly-traded and those that are privately-held, with the latter group being significantly more bank-dependent than the former (see, e.g., Caglio et al. (2021)). Our assumption is that listed firms are more likely to tap 26

alternative sources of finance, such as public debt markets, when their banks are unable to lend to them. The results in panel A of Table 9 show that higher trade uncertainty has a significant dampeningeffectonprivatefirms’performanceandnosucheffectforlistedfirms(thedifference-indifferences coefficients are statistically significant for private firms in three of four specifications). Second, we define bank dependence as a high (above-median) share of bank debt in the firm’s total debt. The results for this measure are reported in panel B of Table 9 and show a larger and statistically significant credit contraction at firms with higher bank dependence (in three of four specifications), corroborating the finding that bank-dependent firms are relatively more affected by trade uncertainty exposure through their banks. Taken together, the results in this section are consistent with Conjecture 3 and highlight that firms borrowing from exposed banks experience worse economic outcomes as trade uncertainty and tensions rise, which suggests that they cannot costlessly switch to alternative sources of finance. This effect is more pronounced for firms that are more reliant on banks. 5 Ruling Out Alternative Explanations It is important to establish that our results are not driven by changes in macroeconomic conditions that may have occurred simultaneously with the rise in trade uncertainty during 2018–2019. Here we entertain several alternative explanations for our results and supply evidence suggesting that these explanations are not the main driver of our findings. Trade uncertainty versus non-trade uncertainty A possible concern is that the trade uncertainty measure captures risk factors that are unrelated to international trade developments but co-move to generate spurious results. Panel B of Figure 1 suggests such a confounding effect is unlikely given the notable jump in trade uncertainty and not in other sectoral risks. Nevertheless, we run a horse-race regression where we add a measure capturing bank exposure to non-trade uncertainty (in interaction with the Post dummy) as an additional explanatory variable. This measure is computed in the same way as the baseline exposure to trade uncertainty, with the only difference that we obtain non-trade uncertainty measures at the sector level from firm-level risk indicators vis-a-vis all sectors other than trade. Other sectors include economic policy & budget, environ- 27

ment, institutions & political processes, health care, security & defense, tax policy, and technology & infrastructure. The results are reported in Table OA-7, where the estimated coefficients on the difference-in-differencestermsforthenon-tradeexposuremeasureisstatisticallyinsignificant,while our baseline coefficients remain statistically significant and with the expected sign. Exchange rate movements Next, we explore whether our results are driven by exchange rate movements, which may co-move with trade uncertainty, given that the strength of the U.S. dollar affects both banks’ asset quality and trade activities. The Bank of International Settlements (BIS) broad U.S. dollar index appreciated by 4.7% during the high-trade uncertainty period between January 1, 2018 and December 31, 2019. Exchange rate fluctuations affect banks and firms through several traditional mechanisms. When the dollar appreciates, banks may pull back from lending if they expect repayment capacity to deteriorate among their borrowers, especially among those unhedged foreign borrowers with dollar-denominated debts. A stronger dollar also reduces the purchasing power of foreign firms, which can make it harder for some U.S. firms to sell their goods abroad, impairing their growth prospects and profitability. In addition, several financial mechanisms can drive the link between the U.S. dollar and the provision of dollar credit. A stronger dollar is associated with tighter dollar credit conditions (Bruno and Shin, 2023; Niepmann and Schmidt-Eisenlohr, 2019), which implies that foreign exporters more reliant on dollar-funded bank credit, may experience a decline in credit access, higher loan spreads (Meisenzahl et al., 2021), and a slowdown in real activity. This, in turn, may dampen the growth of U.S. firms that rely on imported intermediate inputs for their production, which, in turn, can affect their credit risk as perceived by lenders. To address the possibility that fluctuations in the value of the U.S. dollar explain our results, we conduct two tests. First, we examine whether our main results survive after we control for bank exposure to these alternative mechanisms. To this end, we construct an additional exposure measure representing, for each bank, the end-2017 share of outstanding loans to firms in tradablegoods producing sectors, which arguably are more exposed to U.S. dollar fluctuations than firms in non-tradable goods sectors. We follow Desai et al. (2008) and classify construction, retailers, transportation, and recreation as non-tradable goods producing sectors. We then interact this exposure uncertainty measure with the U.S. dollar broad exchange rate index and include it in the 28

regression with our baseline trade exposure interaction. As shown in Table OA-8, estimates for this specification reveal that including this additional control variable does not affect the statistical and economic significance of the estimated coefficient on our key difference-in-differences term. Second, wetestwhetherbanksdifferentiallycurtailtheircreditsupplyacrosscreditlines(which are mainly used by firms as a source of liquidity insurance) versus term loans (typically used for financing investment). This test allows us to rule out a “credit channel” of dollar movements by which a stronger dollar tightens liquidity conditions in the secondary market for syndicated credits (NiepmannandSchmidt-Eisenlohr,2019). Thischannelpredictsthatourresultsshouldbestronger for term loans, which are more likely to be sold in the secondary market than credit lines (Gatev and Strahan, 2009). When we unpack the baseline difference-in-differences term by credit lines versus term loans, we find that credit lines are relatively more affected by an increase in trade uncertainty (see Table OA-9). For term loans, spreads increase at more exposed banks (columns 3– 4), but loan growth does not change significantly neither in the full sample nor for low-uncertainty firms (columns 1–2). These results are therefore inconsistent with our baseline findings operating through a credit channel of dollar movements. Bank cyclicality An alternative explanation for our findings could be that bank exposure to trade uncertainty captures the degree of bank cyclicality, that is, the sensitivity of a bank’s lending book to monetary and financial conditions. If this were the case, then the results would reflect a standard bank lending channel of monetary policy rather than the effects of trade uncertainty. To address this possibility, we measure the extent of loan book cyclicality, for each bank in our sample, as the long-run correlation of the growth rate of a banks’ total loan commitments and that of the overall banking sector. Our main estimates are robust to controlling for bank cyclicality in interaction with the Post dummy: if anything, bank cyclicality operates in the opposite direction of the uncertainty exposure, with more cyclical banks increasing loan volumes (and leaving spreads unchanged) during the Trade War (Table OA-10 panel A). Commodity prices Following the sharp and sustained oil price decline that started in mid- 2014, U.S. banks with more concentrated exposures in the oil sector experienced losses and cut down lending, especially to firms in the oil sector (Bidder et al., 2021). One might worry that our 29

results pick up the effects of bank exposure to the oil sector, in particular those of the protracted creditcrunchthatfollowedthedeclineinoilprices. Toalleviatethisconcern, wedropoilcompanies from the sample (broadly identified as those in the 2-digit NAICS “Mining, quarrying, and oil and gas extraction” sector). Removing oil companies from the sample leaves the results unchanged (Table OA-10 panel B). Credit demand While our analysis focuses on understanding shifts in bank loan supply, it is equally important to determine how firms adjust credit demand in the face of uncertainty shocks. To this end, we examine the credit utilization rate, defined as the ratio of credit utilized relative to credit committed. We run regressions in data aggregated at the firm-quarter level (where the dependent variable is the average utilization rate on credit lines of firms with multiple revolvers outstandingacrossbanks). RegressionestimatesinTableOA-11indicatethatcreditlineutilization rates are higher for high-uncertainty firms during the Trade War (that is, those firms in sectors with a change in average uncertainty between 2016–2017 and 2018–2019 above-75th percentile). The evidence thus suggests that firms most affected by the rise in trade uncertainty attempted to boost their liquidity positions by defensively drawing down bank credit lines. The rise in loan demand is thus inconsistent with the baseline evidence of declining loan growth and rising loan spreads at more exposed banks, increasing our confidence in a supply-side interpretation of the identified effects. 6 Conclusion This paper shows that trade uncertainty affects U.S. banks’ credit supply along several dimensions. Exploiting the large and unanticipated spike in trade uncertainty during the 2018–2019 Trade War, coupled with supervisory loan-level data for U.S. banks and firms, we document that banks with higher ex-ante exposure to sectors facing a greater increase in trade uncertainty pull back from lending, with negative real effects for bank-dependent firms. Our results highlight an important banking channel for the transmission of uncertainty shocks to the real economy and caution against protectionist trade policies that are a major source of ongoing economic uncertainty. Our analysis also suggests that a full accounting of the macroeconomic effects of trade disputes and other de- 30

globalizingeventsshouldtakeintoaccounttheendogenouscontractionaryresponsesofthefinancial sector. Feedback effects between the financial sector and the real economy that originate with real sector shocks are a promising avenue for future research. 31

References Agarwal,Sumit,RicardoCorrea,BernardoMorais,JessicaRold´an,andClaudiaRuiz-Ortega,“Owe a Bank Millions, the Bank Has a Problem: Credit Concentration in Bad Times,” International Finance Discussion Papers 1288, Board of Governors of the Federal Reserve System (U.S.) July 2020. Alessandri, Piergiorgio and Margherita Bottero, “Bank lending in uncertain times,” European Economic Review, 2020, 128, 103503. Amiti, Mary and David E. Weinstein, “Exports and Financial Shocks,” Quarterly Journal of Economics, 10 2011, 126 (4), 1841–1877. andDavidEWeinstein, “HowMuchDoIdiosyncraticBankShocksAffectInvestment? Evidence from Matched Bank-Firm Loan Data,” Journal of Political Economy, 2018, 126 (2), 525–587. , Sang Hoon Kong, and David Weinstein, “The Effect of the U.S.-China Trade War on U.S. Investment,” May 2020. NBER Working Paper No. 27114. , Stephen J Redding, and David Weinstein, “The Impact of the 2018 Trade War on U.S. Prices and Welfare,” March 2019. NBER Working Paper No. 25672. Antr`as, Pol and Davin Chor, “Chapter 5 - Global value chains,” in Gita Gopinath, Elhanan Helpman, and Kenneth Rogoff, eds., Handbook of International Economics: International Trade, Volume 5, Vol. 5 of Handbook of International Economics, Elsevier, 2022, pp. 297–376. Avdjiev, Stefan, Leonardo Gambacorta, Linda S Goldberg, and Stefano Schiaffi, “The shifting drivers of global liquidity,” Journal of International Economics, 2020, 125, 103324. Baker, Scott R., Nicholas Bloom, and Steven J. Davis, “Measuring Economic Policy Uncertainty,” Quarterly Journal of Economics, 07 2016, 131 (4), 1593–1636. Benguria, Felipe, Jaerim Choi, Deborah L Swenson, and Mingzhi Jimmy Xu, “Anxiety or pain? TheimpactoftariffsanduncertaintyonChinesefirmsinthetradewar,”Journal of International Economics, 2022, 137, 103608. Berger, Allen N, Omrane Guedhami, Hugh H Kim, and Xinming Li, “Economic policy uncertainty and bank liquidity hoarding,” Journal of Financial Intermediation, 2020, p. 100893. Bernanke, Ben S., “Irreversibility, Uncertainty, and Cyclical Investment,” Quarterly Journal of Economics, 1983, 98 (1), 85–106. , “The financial accelerator and the credit channel,” 2007. Speech delivered at The Credit Channel of Monetary Policy in the Twenty-first Century Conference, Federal Reserve Bank of Atlanta, Atlanta, GA, June 15. Bidder, Rhys M, John R Krainer, and Adam Hale Shapiro, “De-leveraging or de-risking? How banks cope with loss,” Review of Economic Dynamics, 2021, 39, 100–127. Blickle, Kristian, Cecilia Parlatore, and Anthony Saunders, “Specialization in banking,” 2023. NBER Working Paper No. 31077. Bloom, Nicholas, “The impact of uncertainty shocks,” Econometrica, 2009, 77 (3), 623–685. , “Fluctuations in uncertainty,” Journal of Economic Perspectives, 2014, 28 (2), 153–176. Bloom, Nick, Stephen Bond, and John Van Reenen, “Uncertainty and investment dynamics,” Review of Economic Studies, 2007, 74 (2), 391–415. Bordo,MichaelD,JohnVDuca,andChristofferKoch,“Economicpolicyuncertaintyandthecredit channel: Aggregate and bank-level U.S. evidence over several decades,” Journal of Financial Stability, 2016, 26, 90–106. 32

Borusyak, Kirill, PeterHull, andXavierJaravel, “Quasi-experimentalshift-shareresearchdesigns,” The Review of Economic Studies, 2022, 89 (1), 181–213. Bruno, Valentina and Hyun Song Shin, “Cross-border banking and global liquidity,” Review of Economic Studies, 2015, 82 (2), 535–564. and , “Dollar and exports,” Review of Financial Studies, 2023, 36 (8), 2963–2996. Buch, Claudia M., Manuel Buchholz, and Lena Tonzer, “Uncertainty, Bank Lending, and Bank- Level Heterogeneity,” IMF Economic Review, 2015, 63 (4), 919–954. Caballero, Ricardo J. and Robert S. Pindyck, “Uncertainty, investment, and industry evolution,” International Economic Review, 1992, 37 (3), 641–662. Caglio, Cecilia, Matthew Darst, and S¸ebnem Kalemli-O¨zcan, “Collateral Heterogeneity and Monetary Policy Transmission: Evidence from Loans to SMEs and Large Firms,” 2021. NBER Working Paper No. 28685. Caldara, Dario, Matteo Iacoviello, Patrick Molligo, Andrea Prestipino, and Andrea Raffo, “The economic effects of trade policy uncertainty,” Journal of Monetary Economics, 2020, 109, 38–59. Cao, Jin, Valeriya Dinger, Ragnar Juelsrud, and Karolis Liaudinskas, “Trade conflicts and credit supply spillovers: Evidence from the Nobel Peace Prize trade shock,” 2022. BOFIT Discussion Paper No. 2208. Cavallo, Alberto, GitaGopinath, BrentNeiman, andJennyTang, “TariffPass-ThroughattheBorder and at the Store: Evidence from U.S. Trade Policy,” American Economic Review: Insights, March 2021, 3 (1), 19–34. Cetorelli, Nicola and Linda S. Goldberg, “Liquidity management of U.S. global banks: Internal capital markets in the great recession,” Journal of International Economics, 2012, 88 (2), 299– 311. Chava, Sudheer and Amiyatosh Purnanandam, “The effect of banking crisis on bank-dependent borrowers,” Journal of Financial Economics, 2011, 99 (1), 116–135. Chodorow-Reich, Gabriel, “The employment effects of credit market disruptions: Firm-level evidence from the 2008-9 financial crisis,” Quarterly Journal of Economics, 2014, 129 (1), 1–59. Cort´es,KristleRomeroandPhilipEStrahan,“Tracingoutcapitalflows: Howfinanciallyintegrated banks respond to natural disasters,” Journal of Financial Economics, 2017, 125 (1), 182–199. Crozet, Matthieu, Banu Demir, and Beata Javorcik, “International trade and letters of credit: A double-edged sword in times of crises,” IMF Economic Review, 2022, pp. 1–27. Dell’Ariccia, Giovanni, Dalida Kadyrzhanova, Camelia Minoiu, and Lev Ratnovski, “Bank lending in the knowledge economy,” Review of Financial Studies, 2021, 34 (10), 5036–5076. Desai, Mihir A, C Fritz Foley, and Kristin J Forbes, “Financial constraints and growth: Multinational and local firm responses to currency depreciations,” Review of Financial Studies, 2008, 21 (6), 2857–2888. Dinlersoz, Emin, S¸ebnem Kalemli-O¨zcan, Henry Hyatt, and Veronika Penciakova, “Leverage over the life cycle and implications for firm growth and shock responsiveness,” 2018. NBER Working Paper No. 25226. Dixit, Avinash K. and Robert S. Pindyck, Investment under Uncertainty, Princeton University Press, 1994. Fajgelbaum,Pablo,PinelopiKGoldberg,PatrickJKennedy,AmitKhandelwal,andDariaTaglioni, “TheU.S.-Chinatradewarandglobalreallocations,”American Economic Review: Insights (fortcoming), 2023. 33

Farinha, Luisa, Sotirios Kokas, Serafeim Tsoukas, and Enrico Sette, “Real effects of imperfect bank-firm matching,” 2022. Bank of Portugal Working Paper w202210. Favara, Giovanni and Mariassunta Gianetti, “Forced Asset Sales and the Concentration of Outstanding Debt: Evidence from the Mortgage Market,” Journal of Finance, 2017, 72 (3), 1081– 1118. , Ivan Ivanov, and Marcelo Rezende, “GSIB surcharges and bank lending: Evidence from U.S. corporate loan data,” Journal of Financial Economics, 2021, 142 (3), 1426–1443. Fazzari, Steven, R Glenn Hubbard, and Bruce C Petersen, “Financing constraints and corporate investment,” Brookings Papers on Economic Activity, 1987, (1), 141–206. Federico, Stefano, Fadi Hassan, and Veronica Rappoport, “Trade Shocks and Credit Reallocation,” July 2020. Bank of Italy Working Paper No. 1289. , Giuseppe Marinelli, and Francesco Palazzo, “The 2014 Russia shock and its effects on Italian firms and banks,” 2023. NBER Working Paper No. w31171. Flaaen, Aaron and Justin Pierce, “Disentangling the Effects of the 2018-2019 Tariffs on a Globally Connected U.S. Manufacturing Sector,” May 2019. FEDS 2019-086. Flynn, Sean J and Andra Ghent, “Does Main Street Benefit from What Benefits Wall Street?,” Journal of Financial and Quantitative Analysis, 2022, pp. 1–37. Frame, W Scott, Ping McLemore, and Atanas Mihov, “Haste makes waste: Banking organization growth and operational risk,” 2023. FRB Dallas Research Department Working Paper Series No. 2023. Galaasen, Sigurd, Rumstam Jamilov, Ragnar Juelsrud, and Helene Rey, “Granular Credit Risk,” December 2021. NBER Working Paper No. 27994. Gatev, Evan and Philip E Strahan, “Liquidity risk and syndicate structure,” Journal of Financial Economics, 2009, 93 (3), 490–504. Giannetti, Mariassunta and Farzad Saidi, “Shock propagation and banking structure,” Review of Financial Studies, 2019, 32 (7), 2499–2540. Gilje, Erik P, Elena Loutskina, and Philip E Strahan, “Exporting liquidity: Branch banking and financial integration,” Journal of Finance, 2016, 71 (3), 1159–1184. Grossman, Gene, Elhanan Helpman, and Stephen Redding, “When tariffs disrupt global supply chains,” 2023. Princeton University, Dept. of Economics Working Papers 2021-73. Haas,RalphDeandNeeltjeVanHoren,“RunningfortheExit? InternationalBankLendingDuring a Financial Crisis,” Review of Financial Studies, 01 2013, 26 (1), 244–285. Handley, Kyle and Nuno Limao, “Trade and investment under policy uncertainty: Theory and firm evidence,” American Economic Journal: Economic Policy, 2015, 7 (4), 189–222. and ,“PolicyUncertainty,Trade,andWelfare: TheoryandEvidenceforChinaandtheUnited States,” American Economic Review, September 2017, 107 (9), 2731–83. Hankins,KristineWatson,MortezaMomeniShahraki,andDavidSovich,“DoesTradePolicyAffect Consumer Credit? The Role of Captive Finance Lenders,” University of Kentucky (Department of Finance), 2022. Hassan,TarekA,StephanHollander,LaurencevanLent,andAhmedTahoun,“Firm-LevelPolitical Risk: Measurement and Effects,” Quarterly Journal of Economics, 08 2019, 134 (4), 2135–2202. Hassan, Tarek Alexander, Jesse Schreger, Markus Schwedeler, and Ahmed Tahoun, “Sources and transmission of country risk,” Review of Economic Studies (forthcoming), 2021. 34

, Stephan Hollander, Laurence van Lent, and Ahmed Tahoun, “The Global Impact of Brexit Uncertainty,” January 2020. NBER Working Paper No. 26609. , , , Markus Schwedeler, and Ahmed Tahoun, “Firm-Level Exposure to Epidemic Diseases: COVID-19, SARS, and H1N1,” April 2020. NBER Working Paper No. 26971. Huang, Yi, Chen Lin, Sibo Liu, and Heiwai Tang, “Trade networks and firm value: Evidence from the U.S.-China trade war,” 2019. Available at SSRN 3504602. Huber, Kilian, “Disentangling the effects of a banking crisis: Evidence from German firms and counties,” American Economic Review, 2018, 108 (3), 868–898. Husted, Lucas, John Rogers, and Bo Sun, “Monetary policy uncertainty,” Journal of Monetary Economics, 2020, 115, 20–36. Ivashina, Victoria, Luc Laeven, and Enrique Moral-Benito, “Loan types and the bank lending channel,” Journal of Monetary Economics, 2021. Jasova, Martina, Caterina Mendicino, and Dominik Supera, “Policy uncertainty, lender of last resort and the real economy,” Journal of Monetary Economics, 2021, 118, 381–398. Jim´enez, Gabriel, Atif Mian, Jos´e-Luis Peydr´o, and Jesu´s Saurina, “The real effects of the bank lending channel,” Journal of Monetary Economics, 2020, 115, 162–179. Kalemli-O¨zcan, S¸ebnem and Jun Hee Kwak, “Capital flows and leverage,” Annual Review of Economics, 2020, 12, 833–846. Kang, Jun-Koo and Rene M Stulz, “Do banking shocks affect borrowing firm performance? An analysis of the Japanese experience,” The Journal of Business, 2000, 73 (1), 1–23. Kaviani, Mahsa S, Lawrence Kryzanowski, Hosein Maleki, and Pavel Savor, “Policy uncertainty and corporate credit spreads,” Journal of Financial Economics, 2020, 138 (3), 838–865. Khwaja, Asim Ijaz and Atif Mian, “Tracing the impact of bank liquidity shocks: Evidence from an emerging market,” American Economic Review, 2008, 98 (4), 1413–1442. Leary, Mark T and Michael R Roberts, “Do peer firms affect corporate financial policy?,” Journal of Finance, 2014, 69 (1), 139–178. Li, Lei, Elena Loutskina, and Philip E Strahan, “Deposit market power, funding stability and long-term credit,” Journal of Monetary Economics, 2023. Loughran, Tim and Bill McDonald, “When is a liability not a liability? Textual analysis, dictionaries, and 10-Ks,” Journal of Finance, 2011, 66 (1), 35–65. Markowitz, Harry, “The utility of wealth,” Journal of Political Economy, 1952, 60 (2), 151–158. Mart´ın, Alberto, Enrique Moral-Benito, and Tom Schmitz, “The financial transmission of housing booms: Evidence from Spain,” American Economic Review, 2021, 111 (3), 1013–1053. Mayordomo, Sergio and Omar Rachedi, “The China Syndrome Affects Banks: The Credit Supply Channel of Foreign Import Competition,” Journal of Financial and Quantitative Analysis, 2022, 57 (8), 3114–3144. Meisenzahl, Ralf, Friederike Niepmann, and Tim Schmidt-Eisenlohr, “The dollar and corporate borrowing costs,” International Finance Discussion Paper, 2021, (1312). Michalski, Tomasz and Evren Ors, “(Interstate) Banking and (interstate) trade: Does real integration follow financial integration?,” Journal of Financial Economics, 2012, 104 (1), 89–117. Niepmann, Friederike, “Bankingacrossborders,” Journal of International Economics, 2015, 96(2), 244–265. 35

and Tim Schmidt-Eisenlohr, “International trade, risk and the role of banks,” Journal of International Economics, 2017, 107, 111–126. and , “No guarantees, no trade: How banks affect export patterns,” Journal of International Economics, 2017, 108, 338–350. and , “Institutional investors, the dollar, and U.S. credit conditions,” Journal of Financial Economics (forthcoming), 2019. Novy, Dennis and Alan M. Taylor, “Trade and Uncertainty,” Review of Economics and Statistics, 10 2020, 102 (4), 749–765. Paravisini, Daniel, Veronica Rappoport, and Philipp Schnabl, “Specialization in bank lending: Evidence from exporting firms,” Journal of Finance, 2023, (4), 2049–2085. 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. Pindyck, RobertS., “Irreversibility, Uncertainty, andInvestment,” Journal of Economic Literature, September 1991, 29 (3), 1110–1148. Rajan,RaghuramG,“Insidersandoutsiders: Thechoicebetweeninformedandarm’s-lengthdebt,” Journal of Finance, 1992, 47 (4), 1367–1400. Rey, H´el`ene, “Dilemma not trilemma: The global financial cycle and monetary policy independence,” 2015. NBER Working Paper No. w21162. Schiller,Christoph,“FinancialContagioninInternationalSupply-ChainNetworks,”February2017. Mimeo, Arizona State University. Schnabl, Philipp, “The international transmission of bank liquidity shocks: Evidence from an emerging market,” Journal of Finance, 2012, 67 (3), 897–932. Schwert, Michael, “Bank capital and lending relationships,” Journal of Finance, 2018, 73 (2), 787–830. Slovin,MyronB,MarieESushka,andJohnAPolonchek,“Thevalueofbankdurability: Borrowers as bank stakeholders,” Journal of Finance, 1993, 48 (1), 247–266. Soto, Paul E., “Breaking the Word Bank: Measurement and Effects of Bank Level Uncertainty,” Journal of Financial Services Research, 2021, 59 (1), 1–45. Valencia, Fabi´an, “Aggregateuncertaintyandthesupplyofcredit,” Journal of Banking & Finance, 2017, 81, 150–165. Waugh, Michael E., “The Consumption Response to Trade Shocks: Evidence from the U.S.-China Trade War,” 2019. NBER Working Paper No. 26353. Whited, Toni M and Guojun Wu, “Financial constraints risk,” Review of Financial Studies, 2006, 19 (2), 531–559. Wu, Wei-Shao and Sandy Suardi, “Economic Uncertainty and Bank Lending,” Journal of Money, Credit and Banking, 2021. 36

Figure 1. Trade and other uncertainty indexes This figure depicts the evolution of the trade uncertainty index compared to aggregate indexes of overall, political, andnonpoliticalrisk(panelA)andsectoralrisk(panelB).Thesemeasuresareconstructedusingtextualanalysisof earningscalltranscriptsbylistedfirmsandcountthefrequencyofmentionsofsynonymsfor“risk”or“uncertainty.” Individual risk indexes shown below are computed from firm-level data as quarterly averages across reporting U.S. firms and are standardized. Sources: Hassan et al. (2019, 2020a,b), and https://sites.google.com/view/firmrisk. A. Trade uncertainty index vs. aggregate uncertainty indexes B. Trade uncertainty index vs. sectoral uncertainty indexes 37

Figure 2. Dynamic sales growth differential at high vs. low-uncertainty firms Thisfigureshowstheeffectsoffirmexposuretotradeuncertaintythroughitslendersonrealsalesgrowthduring2016- 2019. The chartplots theestimated coefficientsand theassociated99% confidencelevelsof adynamic difference-indifferencesmodelthatregressesfirm-levelrealsalesgrowthonadummyvariableforhigh-uncertaintyfirmsinteracted with yearly dummies and firm characteristics (size, leverage, and cash holdings). Sources: FR Y-14Q and Hassan et al. (2019, 2020a,b). 38

Figure 3. Dynamic difference-in-differences coefficient chart for lending outcomes at lowuncertainty firms This figure shows the effects of bank exposure to trade uncertainty on loan growth (panel A) and loan spreads (panelB)forlow-uncertaintyfirmsduringthesampleperiodextendedbackbyanadditionalyear(thatis,2015:Q1– 2019:Q4). Thechartsplottheestimateddifference-in-differencescoefficientsandtheassociated99%confidencelevels of the dynamic variant of the specifications in columns 1 (loan growth) and column 3 (for spreads) in Table 2 with interaction effects between bank exposure and quarterly dummies (with base period 2017:Q4). A. Loan growth B. Loan spreads 39

Table 1. Selected descriptive statistics This table reports selected summary statistics for the loan-level regression sample and variables. Measures of bank exposuretotradeuncertainty,tariffsandtradable-goodsproducingsectorsaredescribedinSection2.2. Loangrowth is computed as log(committed amount /committed amount ). The regression sample at the loan level refers t 2016:Q4 to U.S. BHCs with at least $50 billion in assets that participate in CCAR stress tests and report to the FR Y-14Q before 2019; and domestic non-financial firms. Sources: FR Y-14Q, U.S. Bureau of Economic Analysis (BEA), S&P Compustat, Flaaen and Pierce (2019), and Hassan et al. (2019, 2020a,b). N Mean St. Dev. P25 P50 P75 A. Bank characteristics Exposuretotradeuncertainty 318 1.782 0.248 1.639 1.772 1.915 Exposuretooverallsentiment 318 1.421 0.219 1.294 1.442 1.580 Exposuretotariffs-hitsectors 312 0.338 0.104 0.280 0.333 0.429 Exposuretotradable-goodssectors 275 0.416 0.100 0.353 0.387 0.431 Exposuretonon-tradeuncertainty 318 -0.001 0.002 -0.003 -0.001 0.000 Size(log-assets) 318 19.435 1.035 18.681 19.148 19.924 Capital(commonequity/assets) 318 11.549 2.054 10.102 11.433 13.111 Coredeposits(%liabilities) 318 63.103 17.618 54.561 69.530 75.913 Specialization 318 0.381 0.486 0.000 0.000 1.000 1: Highcapital(commonequity/assets) 318 0.390 0.489 0.000 0.000 1.000 1: Highstress-testCET1ratio 300 0.510 0.501 0.000 1.000 1.000 Cyclicality 318 1.179 1.158 0.661 1.100 1.431 B. Firm characteristics 1: Firminlow-uncertaintysector 212973 0.698 0.459 0.000 1.000 1.000 1: Firmintariffs-hitsector 216311 0.217 0.412 0.000 0.000 0.000 1: Firmisinhigh-importdependencesector 139276 0.197 0.397 0.000 0.000 1.000 1: Firmisinhigh-tariffsprotectionsector 46986 0.241 0.428 0.000 0.000 0.000 1: Firminoilsector 216311 0.023 0.149 0.000 0.000 0.000 Totaldebtgrowth 18917 5.501 49.529 -13.706 0.000 18.643 Investmentrate 18140 17.288 29.020 0.000 2.811 22.175 Firmexposuretouncertainty 18917 1.377 0.619 0.883 1.657 1.830 Size(log-assets) 18917 18.366 2.441 16.523 17.799 19.884 Liquidity(cashandmktbsecurities/assets) 18917 9.357 12.478 1.217 4.647 12.757 Tangibility(tangibleassets/totalassets) 18917 85.753 21.806 79.611 97.618 100.000 Interestcoverageratio(ICR) 18917 0.327 0.699 0.038 0.092 0.231 Returnonassets(ROA) 18917 15.561 21.434 6.012 11.376 18.932 Salesgrowth 18917 11.530 36.631 -1.268 5.626 14.994 1: Firmisspeculative-grade 18917 0.619 0.486 0.000 1.000 1.000 1: Firmispublic 18917 0.108 0.311 0.000 0.000 0.000 1: FirmhashighshareofCCARbankdebt 17825 0.395 0.489 0.000 0.000 1.000 C. Loan characteristics Loanamount(USDmillion) 928768 28.291 71.978 2.770 10.000 31.250 Loangrowth 928768 -0.230 0.937 -0.623 0.000 0.299 Loanspread(ppts) 540067 2.015 1.180 1.250 1.850 2.600 Timetomaturity(years) 1095308 2.563 1.982 0.750 2.500 4.000 1: Loanisdemandable 1095308 0.134 0.341 0.000 0.000 0.000 1: Loanisneworigination 925630 0.072 0.258 0.000 0.000 0.000 1: Loanissecured 927367 0.774 0.418 1.000 1.000 1.000 Probabilityofdefault 868739 0.026 0.092 0.003 0.007 0.017 1: Loanisfortradefinancing 928768 0.024 0.152 0.000 0.000 0.000 1: Loanisacreditline 838702 0.574 0.494 0.000 1.000 1.000 40

Table 2. Baseline Results: The effect of trade uncertainty bank lending to all firms and spillovers to low-uncertainty firms This table shows OLS estimates for a regression of loan growth and spreads on bank exposure to trade uncertainty. Thedataareatthebank-firm-quarterloan-levelandrefertooutstandingloanstodomesticborrowers(non-financial firms) during 2016:Q1–2019:Q4. Bank exposure to trade uncertainty is measured as the average of the difference in trade uncertainty across sectors (between 2016:Q1–2017:Q4 and 2018:Q1–2019:Q4), weighted by initial bank loans shares to those sectors (See Section 2.2 for the construction of the variable). The dummy variable Post takes value of one for the period 2018:Q1-2019:Q4 and zero for the period 2016:Q1-2017:Q4. Bank controls include size (log-total assets), capital (common equity/total assets), deposits (core deposits/liabilities), and specialization, and enter in levels and interacted with Post. Standard errors are double clustered at the quarter and bank-firm level. Significance: *** 1%, **5%, and *10%. (1) (2) (3) (4) Dependent variable Loan growth Loan spread A. All firms Bank exposure × Post -0.133*** -0.102*** 0.321*** 0.260*** (0.038) (0.030) (0.096) (0.085) Observations 928,768 925,465 483,660 481,152 R2 0.240 0.342 0.798 0.856 B. Low-uncertainty firms Bank exposure × Post -0.164*** -0.111*** 0.347*** 0.283** (0.047) (0.036) (0.103) (0.096) Observations 660,608 658,123 339,851 337,955 R2 0.248 0.350 0.804 0.856 Bank controls Y Y Y Y Bank controls × Post Y Y Y Y Bank FE Y Y Y Y Firm × Quarter FE Y Y Y Y Firm × Bank FE Y Y 41

Table 3. Horse-race with proxies of the first moment This table shows OLS estimates for a regression of loan growth and spreads on bank exposure to trade uncertainty in a horse-race with two variables capturing first-moment effects. The first variable is bank exposure to actual changes in trade policy, that is, to sectors that received tariffs (panel A). The second variable is bank exposure to changesinoverallsentiment(panelB).Bankexposuretotariffs-hitsectorsistheaverageshareofloancommitments to tariffs-hit sectors during 2014–2015. Bank exposure to changes in overall sentiment is computed in the same way as bank exposure to trade uncertainty, but we use the overall sentiment index instead of the trade uncertainty index. All specification details, sample period, and controls as in Table 2. Standard errors are double clustered at the quarter and bank-firm level. Significance: *** 1%, **5%, and *10%. (1) (2) (3) (4) Dependent variable Loan growth Loan spread All Low-uncertainty All Low-uncertainty firms firms firms firms A. Control for bank exposure to tariffs-hit sectors Bank exposure to (trade) uncertainty × Post -0.140*** -0.153*** 0.233** 0.262** (0.029) (0.033) (0.082) (0.092) Bank exposure to tariffs-hit sectors × Post 0.258*** 0.271*** 0.318** 0.252** (0.074) (0.088) (0.110) (0.111) Observations 918,982 653,795 477,573 335,091 R2 0.343 0.350 0.855 0.855 Bank controls Y Y Y Y Bank controls × Post Y Y Y Y Firm × Quarter FE Y Y Y Y Firm × Bank FE Y Y Y Y B. Control for bank exposure to overall sentiment Bank exposure to (trade) uncertainty × Post -0.094** -0.085* 0.284*** 0.317*** (0.036) (0.041) (0.073) (0.078) Bank exposure to overall sentiment × Post -0.013 -0.050 -0.047 -0.066 (0.031) (0.037) (0.063) (0.060) Observations 925,465 658,123 481,152 337,955 R2 0.342 0.350 0.856 0.856 Bank controls Y Y Y Y Bank controls × Post Y Y Y Y Firm × Quarter FE Y Y Y Y Firm × Bank FE Y Y Y Y 42

Table 4. Wait-and-see behaviors: The extensive margin of lending This table shows OLS estimates for a regression of extensive margin of lending outcomes on bank exposure to trade uncertainty. Thedependentvariableisadummyvariablethattakesvalueonefornewloanoriginationsinloan-level data and zero otherwise (panel A) or the share of new loans (volume weighted) in bank-firm-quarter level data (panel B). All specification details, sample period, and controls as in Table 2. Standard errors are double clustered at the quarter and bank-firm level. Significance: *** 1%, **5%, and *10%. (1) (2) (3) (4) Dependent variable Loan is new origination Share of new loan originations (volume-weighted) A. Loan-level data B. Bank-firm level data All Low-uncertainty All Low-uncertainty firms firms firms firms Bank exposure × Post -0.018*** -0.017** -0.017*** -0.019** (0.005) (0.008) (0.005) (0.007) Observations 925,630 658,255 346,388 246,891 R2 0.581 0.588 0.668 0.678 Bank controls Y Y Y Y Bank controls × Post Y Y Y Y Firm × Quarter FE Y Y Y Y Firm × Bank FE Y Y Y Y 43

Table 5. Wait-and-see behaviors: Banks’ assessment of firm default risk and loan maturities This table shows OLS estimates for a regression of banks’ assessment of firm default risk and loan maturities on bank exposure to trade uncertainty. The dependent variable is the probability of default (PD) (columns 1–2); remaining time to maturity in quarters (columns 3–4) and a dummy variable for demandable loans in the extended datasetthatincludessuchloans(columns5–6). Incolumns1–2weincludeallthefirmsinthedataset(thatis,both single- and multi-lender firms) so as to capture banks’ assessment of borrower risk across the entire loan portfolio. Demandable loans are only included in the analysis of loan maturities in columns 3–6 of this table. All specification details, sample period, and controls as in Table 2 and discussed in Section 4.3. Other loan controls in columns 3–6 followLietal.(2023)toinclude: thelogofloansize(totalloancommitment)anddummyvariablesforfloatingrate loans, secured loans, and loans with prepayment penalty. Standard errors are double clustered at the quarter and bank-firm level. Significance: *** 1%, **5%, and *10%. (1) (2) (3) (4) (5) (6) Probability of default Time to maturity (years) Demandable loan All Low-uncertainty All Low-uncertainty All Low-uncertainty firms firms firms firms firms firms Bankexposure×Post 0.010*** 0.013*** -0.143*** -0.095** 0.021** 0.016* (0.003) (0.004) (0.040) (0.039) (0.009) (0.008) Observations 1,432,240 998,525 1,091,466 705,790 1,095,308 708,517 R2 0.012 0.013 0.714 0.678 0.768 0.512 Bankcontrols Y Y Y Y Y Y Bankcontrols×Post Y Y Y Y Y Y Otherloancontrols Y Y Y Y BankFE Y Y Y Y Y Y QuarterFE Y Y Y Y Y Y Firm×QuarterFE Y Y Y Y Firm×BankFE Y Y 44

Table 6. Heterogeneous effects across borrowers ThistableshowsOLSestimatesforaregressionofloangrowthandloanspreadonbankexposuretotradeuncertainty allowingforheterogeneityacrossfirms. PanelAallowsfordifferenteffectsbydegreeofimportprotection,wherelow importprotectionisanindicatorforfirmsinsectorsbelowthe75th percentileofthenewimporttariffrateasashare of consumption distribution (data available at the 4-digit NAICS classification level for manufacturing industries). Panel B allows for different effects depending on the sector-specific degree of import dependence, where high import dependenceisanindicatorforfirmsinsectorswithabove-mediantotalimportsasashareofindustryoutput(atthe 3-digitNAICSclassificationlevel)andzerootherwise. Allspecificationdetails,sampleperiod,andcontrolsasinTable2. Standarderrorsaredoubleclusteredatthequarterandbank-firmlevel. Significance: ***1%,**5%,and*10%. (1) (2) (3) (4) Dependent variable Loan growth Loan spread All Low-uncertainty All Low-uncertainty firms firms firms firms A. Tariff import protection Bank exposure × Post × Low protection (1) -0.187*** -0.214*** 0.268** 0.299** (0.052) (0.059) (0.103) (0.118) Bank exposure × Post × High protection (2) -0.074 -0.153 0.378*** 0.290* (0.069) (0.122) (0.114) (0.140) pvalue test: H :|1|>|2| - - 0.138 0.239 a Observations 288,762 185,478 148,118 95,331 R2 0.338 0.344 0.854 0.855 B. Import dependence Bank exposure × Post × High dependence (1) -0.121*** -0.127** 0.343*** 0.333*** (0.038) (0.047) (0.098) (0.103) Bank exposure × Post × Low dependence (2) -0.090** -0.079 0.192** 0.194** (0.042) (0.045) (0.077) (0.084) pvalue test: H :|1|>|2| 0.151 - 0.442 0.479 a Observations 665,692 470,644 348,858 246,151 R2 0.348 0.361 0.861 0.859 Bank controls Y Y Y Y Bank controls × Post Y Y Y Y Firm × Quarter FE Y Y Y Y Firm × Bank FE Y Y Y Y 45

Table 7. Financial constraints: Role of bank capital This table shows OLS estimates for a regression of loan growth and spreads on bank exposure to trade uncertainty allowing for heterogeneous effects by bank capital. The measure of capital is common equity divided by total assets (at end-2017) in panel A and post stress-test CET1 capital ratio (defined as the minimum CET1 capital ratio estimated under the “Supervisory Severely Adverse” scenario of the Dodd-Frank Act stress test (DFAST)) in panel B. High-capital banks have capital ratios above the 75th percentile. All specification details, sample period, and controls as in Table 2. Standard errors are double clustered at the quarter and bank-firm level. Significance: *** 1%, **5%, and *10%. (1) (2) (3) (4) Dependent variable Loan growth Loan spread All Low-uncertainty All Low-uncertainty firms firms firms firms A. Bank capital: Equity/Assets Bank exposure × Post × Low-capital -0.173*** -0.158*** 0.337** 0.367** (0.036) (0.039) (0.144) (0.167) Bank exposure × Post × High-capital capital -0.011 -0.075 0.164*** 0.172*** (0.037) (0.046) (0.045) (0.041) p-value t-test H :|1|>|2| - - 0.043 0.049 a Observations 925,467 658,123 481,152 337,955 R2 0.740 0.744 0.856 0.856 B. Bank capital: Post-stress test CET1 ratio Bank exposure × Post × Low-capital -0.242*** -0.220*** 0.332* 0.367* (0.041) (0.048) (0.159) (0.177) Bank exposure × Post × High-capital capital 0.017 -0.033 0.188*** 0.197*** (0.033) (0.040) (0.049) (0.046) p-value t-test H :|1|>|2| - - 0.034 0.064 a Observations 886,460 629,384 458,023 320,494 R2 0.742 0.746 0.856 0.857 Bank controls Y Y Y Y Bank controls × Post Y Y Y Y Bank FE Y Y Y Y Firm × Quarter FE Y Y Y Y Firm × Bank FE Y Y Y Y 46

Table 8. Real effects of trade uncertainty through bank lending: Full sample ThistableshowsOLSestimatesforaregressionoffirm-leveltotaldebtgrowthandinvestmentratioonfirmexposure to trade uncertainty through its lenders. Firm exposure to trade uncertainty through its lenders is computed as the average exposure to trade uncertainty of the banks from which a given firm borrows, weighted by relative importance of each bank in the firms’ total bank debt at end-2014. The data are at the firm-year level over the period between 2016 and 2019. The dummy variable Post takes value one for the period 2019–2019 and zero for the period 2016–2017. Firm controls include size (log-assets), liquidity (cash and marketable securities/assets), tangibility (tangible assets as a share of total assets), interest coverage ratio (EBITDA/total interest expense), ROA (return on assets), real sales growth—all lagged one year—and a dummy variable taking value one for firms rated speculative-grade by their lender banks, and enter in levels and interacted with Post. Firm industry is3-digitNAICSclassification. Standarderrorsareclusteredatthefirmlevel. Significance: ***1%,**5%,and*10%. (1) (2) (3) (4) Total debt growth Investment rate All Low-uncertainty All Low-uncertainty firms firms firms firms Firm exposure to trade uncertainty × Post -0.038* -0.022 -0.044*** -0.053*** (0.019) (0.023) (0.010) (0.011) Observations 18,917 13,251 19,978 14,180 R2 0.515 0.502 0.703 0.705 Firm controls Y Y Y Y Firm controls × Post Y Y Y Y Firm FE Y Y Y Y Industry × County × Year FE Y Y Y Y 47

Table 9. Real effects of trade uncertainty through bank lending: Heterogeneity by dependence on bank debt This table shows OLS estimates for a regression of firm-level total debt growth and investment rate on firm exposure to trade uncertainty through its lenders, allowing for heterogeneity by degree of dependence on bank debt. Dependent on bank debt is proxied by firm ownership (private/public) in panel A and by the share of CCAR bank debt (above/below median share of utilized loans from FR Y-14Q reporting banks in the firm’s total debt) in panel B. Firm exposure to trade uncertainty through its lenders is computed as the average exposure to trade uncertainty of the banks from which a given firm borrows, weighted by relative importance of each bank in the firms’ total bank debt at end-2014. The data are at the firm-year level over the period between 2016 and 2019. The dummy variablePosttakesvalueonefortheperiod2019–2019andzerofortheperiod2016–2017. Firmcontrolsincludesize (log-assets), liquidity (cash and marketable securities/assets), tangibility (tangible assets as a share of total assets), interest coverage ratio (EBITDA/total interest expense), ROA (return on assets), real sales growth—all lagged one year—and a dummy variable taking value one for firms rated below investment-grade by their lender banks, and enterinlevelsandinteractedwithPost. Firmindustryis3-digitNAICSclassification. Standarderrorsareclustered at the firm level. Significance: *** 1%, **5%, and *10%. (1) (2) (3) (4) Total debt growth Investment rate All Low-uncertainty All Low-uncertainty firms firms firms firms A. Bank dependence: Private vs. public firms Firm exposure × Private firm (1) -0.038* -0.021 -0.047*** -0.054*** (0.020) (0.023) (0.010) (0.012) Firm exposure × Public firm (2) -0.034 -0.007 -0.023 -0.051 (0.057) (0.068) (0.026) (0.032) Observations 18,917 21,469 19,978 13,251 R2 0.515 0.626 0.703 0.502 B. Bank dependence: Share of CCAR bank debt Firm exposure × Higher bank debt share (1) -0.045** -0.031 -0.058*** -0.070*** (0.020) (0.024) (0.011) (0.013) Firm exposure × Lower bank debt share (2) -0.032 -0.015 -0.047*** -0.055*** (0.021) (0.025) (0.011) (0.013) p-value t-test H :|1|>|2| - - 0.071 0.032 a Observations 18,917 13,251 17,865 12,609 R2 0.515 0.502 0.707 0.709 Firm controls Y Y Y Y Firm controls × Post Y Y Y Y Firm FE Y Y Y Y Industry × County × Year FE Y Y Y Y 48