The 2023 Banking Turmoil and the Bank Term Funding Program

Finance and Economics Discussion Series Federal Reserve Board, Washington, D.C. ISSN 1936-2854 (Print) ISSN 2767-3898 (Online) The 2023 Banking Turmoil and the Bank Term Funding Program David Glancy, Felicia Ionescu, Elizabeth Klee, Antonis Kotidis, Michael Siemer, Andrei Zlate 2024-045 Please cite this paper as: Glancy,David,FeliciaIonescu,ElizabethKlee,AntonisKotidis,MichaelSiemer,andAndrei Zlate (2024). “The 2023 Banking Turmoil and the Bank Term Funding Program,” Finance and Economics Discussion Series 2024-045. Washington: Board of Governors of the Federal Reserve System, https://doi.org/10.17016/FEDS.2024.045. NOTE: Staff working papers in the Finance and Economics Discussion Series (FEDS) are preliminary materials circulated to stimulate discussion and critical comment. The analysis and conclusions set forth are those of the authors and do not indicate concurrence by other members of the research staff or the Board of Governors. References in publications to the Finance and Economics Discussion Series (other than acknowledgement) should be cleared with the author(s) to protect the tentative character of these papers.

∗ The 2023 Banking Turmoil and the Bank Term Funding Program David Glancy Felicia Ionescu Elizabeth Klee Antonis Kotidis Michael Siemer Andrei Zlate June 12, 2024 Abstract Weusehigh-frequencydatatoexaminetheeffectivenessoftheBankTermFundingProgram (BTFP)insupportingtheliquiditypositionsofvulnerablebanksduringtheMarch2023banking turmoil. Weuncoverthreekeyfindings. First,ourhigh-frequencydataconfirmthatbankswith high reliance on uninsured deposits and large unrealized losses on securities holdings suffered larger deposit outflows at the onset of the episode. Second, the BTFP played an outsized role in meeting these outflows at banks with larger securities losses, reflecting the at-par valuation ofsecuritiescollateralattheBTFP(banksatthe90thpercentileinsecuritieslossesreplaced26 cents of every dollar of outflows with BTFP borrowing, compared to only 7 cents on average). Third, in addition to funding loan growth and deposit outflows, banks used the BTFP to build cash holdings, indicating that the program enabled banks to position themselves against potential future funding needs. Overall, we demonstrate that the BTFP enabled banks to meet funding needs and preserve liquidity during the period of stress. Keywords: 2023 banking turmoil, deposit outflows, uninsured deposits, securities losses, Bank Term Funding Program (BTFP), emergency liquidity facilities JEL Classifications: E52, E58, G01, G21 ∗All authors are at the Federal Reserve Board. We thank Mark Carlson, Virginia Lewis, David Lowe, Matt Malloy, Francis Martinez and Alexandros Vardoulakis, as well as seminar participants at the Federal Reserve Board forhelpfulcomments. BohanWangprovidedoutstandingresearchassistance. Theviewsexpressedinthispaperare those of the authors and do not necessarily represent those of the Federal Reserve Board of Governors or anyone in the Federal Reserve System. 1

1 Introduction A key function of banks is maturity transformation. They borrow short-term deposits and provide long-term credit to firms and households. This function also hedges banks against interest rate risk (Drechsler et al. (2021)). Deposits usually bear below-market rates when interest rates rise, thus hedging the interest rate exposure of long-term assets (Hannan and Berger (1991); Neumark and Sharpe (1991); Drechsler et al. (2017)). However, hedging only works so long as deposits stay in the bank. If the deposit franchise value becomes larger than the value of the bank’s longterm assets when interest rates rise, uninsured depositors may run (Jiang et al. (2023), Drechsler et al. (2023), Haddad et al. (2023)). The failure of the Bank of the Commonwealth in 1973 is an example of interest rate risk that resulted from exposure to long-duration, fixed-rate assets and an unstable funding base (Prescott (2024)). The banking turmoil in March 2023 was the most recent manifestation of this phenomenon. What actions can manage the risk of a run in the presence of interest rate risk? Ex-ante, banks could shorten the maturity of their assets, but this would undo their interest rate hedge. The value of short-term assets does not appreciate enough to offset the decline in the value of the deposit franchise when interest rates fall (Drechsler et al. (2023)). Banks could also invest in swaps that pay fixed and receive floating rate. Recent evidence, however, suggests only limited hedging of bank assets using interest rate swap positions (e.g. McPhail et al. (2023); Granja et al. (2024)). This paper establishes that an ex-post lender of last resort (LOLR) intervention calibrated to preserve banks’ asset values by providing liquidity against the par value of collateral can help contain a run on banks’ deposit franchise in the presence of interest rate risk. In particular, we document and discuss the effectiveness of the Bank Term Funding Program (BTFP) introduced in March 2023 as an ex-post intervention to address run risk. As yields started rising in the spring of 2022, the value of securities held by depository institutions declined and banks started to accumulate significant unrealized losses on their balance sheets (Figure 1, left panel). Following the failures of Silicon Valley Bank (SVB) and Signature Bank in March 2023, concerns mounted over the health of other banks with similar characteristics—those with heavy reliance on uninsured deposits and large losses in their securities portfolios. Broader contagion caused some of those vulnerable banks to face rapid deposit outflows (Figure 1, right panel). In response to the turmoil, the Federal Reserve Board established the BTFP, with approval by the Secretary of the Treasury, pursuant to section 13(3) of the Federal Reserve Act. The BTFP extended loans for up to one year at a fixed rate of 10 basis points over the one-year overnightindexedswap(OIS)rate. Importantly,BTFPcollateralwasvaluedatparratherthanmarketvalue. This paper examines banks’ experience with deposit outflows during the 2023 banking turmoil and 2

Figure 1: Unrealized Securities Losses and Deposit Outflows 1 0 −1 −2 −3 0 2 4 6 8 0 2 4 6 8 0 2 4 0 0 0 0 0 1 1 1 1 1 2 2 2 0 0 0 0 0 0 0 0 0 0 0 0 0 2 2 2 2 2 2 2 2 2 2 2 2 2 Date (Quarterly) stessA latoT fo % 16.00 15.75 15.50 15.25 15.00 J a n Fe b M ar A pr M ay J u n J ul Au g S e p Oct N ov D ec J a n Fe b M ar A pr Date (Weekly) )lirT $( stisopeD eroC Available−for−Sale Securities Held−to−Maturity Securities Core Deposits ($ Tril) Note: The left figure shows unrealized securities losses (available-for-sale (AFS) and held-to-maturity (HTM)) for all U.S. commercial banks from 2000Q1 to 2023Q3 as a percent of total assets. The right figure shows the levels of core deposits for U.S. domestically-chartered commercial banks from January 2023 to March 2024. Source: author calculations, Call Reports, H.8. the effectiveness of the BTFP. We bring together several confidential, bank-level, high-frequency datasets, including data on banks’ assets and liabilities from the Federal Reserve’s FR 2644 reporting form and on borrowings from the BTFP and the discount window. The analysis centers around three main questions. First, we examine what risk factors contributed to banks’ core deposit outflows during the stress episode. We find that banks with high reliance on uninsured deposits as a source of funding and large marked-to-market losses on securities holdings faced relatively more deposit outflows. These effects of uninsured deposits and security losses were short-lived, however, and dissipated by early May. Second, we ask how banks managed these core deposit outflows, such as by reducing liquid assets or borrowing wholesale. We find that banks primarily reduced cash holdings or borrowed from non-Fed sources, mainly Federal Home Loan Banks (FHLBs), when core deposit outflows were realized. BTFP borrowings played a relatively modest role in how the average bank met funding needs. However, BTFP was an important source of funding for the vulnerable banks, such as those with large securities losses or high reliance on uninsured deposits. In particular, banks with relatively large security losses used BTFP borrowings to a greater extent to fund core deposit outflows, and turned to other liquidity sources to a lesser extent than the average bank, including to non-Fed borrowings, large time deposits, cash, and securities. On average, banks covered 7 cents for every dollar of deposit outflows with BTFP borrowings and 32 cents with non- Fed borrowings; in contrast, banks at the 90th percentile in securities losses covered 26 cents for every dollar of core deposit outflows with BTFP borrowings and 28 cents with non-Fed borrowings. 3

Vulnerable banks’ relative preference for BTFP over non-Fed borrowings reflects the advantage of at-par collateral valuation at the BTFP. Furthermore, reliance on BTFP borrowings was especially large among regional banks with total assets between $100 and $250 billion. These banks were more impacted by the turmoil, and used the BTFP to cover almost 50 cents for every dollar of their core deposit outflows. Moreover, vulnerable banks pledged substantially more collateral than they subsequently encumbered with BTFP loans. This collateral pledging was concentrated in mortgage-backed securities (MBS) and collateralized mortgage obligations (CMOs), which suffered disproportionately large fair value declines from January 2022 to March 2023. Third, to understand how banks used wholesale borrowings from the BTFP and other sources, we investigate the dynamic relationship between banks’ wholesale borrowings and deposits, cash, and loans over time. Starting with shifts in the composition of bank liabilities, we find that banks increasednon-Fedborrowingsimmediatelyintheweeksthatcoredepositsfell, butonlyusedBTFP borrowing after a period of sustained deposit outflows. This result suggests that the BTFP usage may have entailed a fixed cost so banks only used BTFP if a significant enough funding need arose. On the asset side, we find that banks that increased their wholesale borrowings built up their cash buffers, suggesting a precautionary reason for wholesale borrowing. Additionally, banks that borrowed from either the BTFP or non-Fed sources exhibited stronger loan growth both before and after borrowing, indicating that some of the borrowing was used to fund new lending. Our findings have several important implications. First, BTFP was successful in targeting the more vulnerable banks, whose BTFP borrowing was motivated by both precautionary motives and theneedtofundtherealizeddepositoutflows. TheprecautionarymotivebehindBTFPborrowings is reflected by our finding that, for a given amount of realized deposit outflows, vulnerable banks with large security losses and high reliance on uninsured deposits borrowed more from the BTFP than other banks. Vulnerable banks also pledged substantially more collateral at the BTFP than theyborrowed. Second,theat-parvaluationofcollateralprovidedanadvantagetobankswithlarger security losses, given that such banks used more BTFP relative to non-Fed borrowings compared with other banks. The at-par collateral valuation advantage of BTFP was especially important for vulnerable banks, which pledged larger amounts of MBS and CMO collateral—securities with relativelylargerfairvaluelossesthanTreasuryandagencysecurities, giventheirnegativeconvexity to rising interest rates—in line with the larger shares of these securities in banks’ portfolios. Third, the finding that banks borrowing from either BTFP or non-Fed sources experienced similar trend growthinloanbalancessuggeststhatBTFPsupportedthelendingofmorevulnerablebanks,which were more likely to borrow from the BTFP and to benefit from the at-par collateral advantage. Conceivably in the absence of BTFP borrowing, vulnerable banks may have been forced to execute costlier balance sheet adjustments, which could have impacted lending. 4

Related literature: Our findings contribute to the literature on the interplay between banks’ liquidity and interest rate risk. In theory, the risk of runs on a bank’s deposit franchise increases when interest rates are high and banks rely on uninsured deposits. This is because the value of the bank’s assets declines relative to the value of the uninsured deposit franchise (e.g., Drechsler et al. (2023); Jiang et al. (2023); Haddad et al. (2023)). Our results highlight the role of banks’ markedto-market losses on securities holdings, which reduced the value of their assets when interest rates increased, and also the role of uninsured deposits in prompting deposit runs. Our results also show that banks with larger losses tapped more into BTFP funding—obtaining funding at the par value of collateral—to restore their asset values and contain deposit runs. Relatedly, our findings contribute to the literature on LOLR by analyzing the BTFP in two important ways. First, a large theoretical literature discusses the benefits of LOLR interventions duringfinancialcrises(DiamondandDybvig(1983);Freixasetal.(2004);RochetandVives(2004)). By supplying term funding (Jasova et al. (2021)), expanding collateral eligibility requirements (Van Bekkum et al. (2018)), or directly purchasing commercial loans from banks (Minoiu et al. (2024)), LOLR can help banks support the supply of credit and prevent a crunch when private liquidity freezes. Our finding that BTFP supported banks’ loan balances is consistent with this literature. This is largely due to BTFP’s one-year fixed-rate funding at 10 basis points spread over the market rate, which resembles the value of the deposit franchise when deposits stay with the bank (e.g., Drechsler et al. (2021)). Second, our finding that banks with large securities losses disproportionately used the BTFP to meet deposit outflows suggests that the par valuation of collateral was a key determinant of the program’s ability to encourage utilization and contain deposit runs. By helping such banks avoid realizing losses from potential sales of securities and maximize the amount of liquidity obtained against collateral, the BTFP helped prevent them from becoming undercapitalized, which would have further exacerbated the bank run (Egan et al. (2017); Jiang et al. (2023)). In other words, the par valuation of collateral helped contain the run, even in the presence of significant interest rate risk. This finding marks an important contribution to the literature on the design and transmission of central bank liquidity provision (Drechsler et al. (2016); Carlson and Macchiavelli (2020); Carpinelli and Crosignani (2021); Kotidis et al. (2022)). More generally, our paper relates to the literature on bank runs and panics. Our finding that depositors withdrew because of a bank’s higher securities losses is consistent with the standard models of bank runs (e.g., Diamond and Dybvig (1983)) and theories where solvency concerns increase the likelihood of a run (e.g., Diamond and Rajan (2011); Brunnermeier and Pedersen (2009)). However, our paper is more directly related to newer theories in which bank runs could occur even when assets are fully liquid and high-quality (e.g., Drechsler et al. (2023), Jiang et al. 5

(2023)). The March 2023 banking crisis represents a case in which the interaction between higher franchise values, uninsured deposits, and market-to-market losses on otherwise liquid security holdings prompted runs, which the BTFP helped contain. Our evidence that the run was concentrated on banks with higher shares of uninsured deposits also complements studies of past episodes of depositor behavior during bank runs (e.g., Iyer and Puri (2012); Iyer et al. (2016)). Finally, our analysis complements findings in an emerging literature that studies the 2023 banking turmoil events. For example, Choi et al. (2023) study the contagion effects following the run on SVB, while Caglio et al. (2023) present evidence on flight to safety from regional banks towards larger banks. Cookson et al. (2023) highlight social media as a key spillover mechanism. While Cipriani et al. (2024) study the role of fundamentals and public signals on prompting runs in a study contemporaneous with ours, we also focus on banks’ balance sheet adjustment strategies in the aftermath of deposit runs, distinguishing between BTFP and other sources of wholesale funding, as well as highlighting the role of par collateral valuation at the BTFP in containing runs. 2 Institutional Background Silicon Valley Bank (SVB) was a $200 billion bank with a concentrated customer base in the technology industry. On March 8, 2023, SVB announced significant losses in its securities portfolio. Coupledwiththebank’sconcentratedcustomerbase,theannouncementledtorapidandsubstantial deposit outflows. Meanwhile, Signature Bank and Silvergate Bank faced similar liquidity pressures, amid investors’ concerns about their cryptocurrency industry exposures. While Silvergate Bank decided to wind down its operations voluntarily, SVB and Signature Bank collapsed on March 10 andMarch12,respectively,whenitbecameclearthattheyhadinsufficientliquiditytomeetdeposit outflows. The speed and magnitude of the deposit runs were unprecedented (Rose, 2023). For example, SVB informed regulators on the morning of March 10 that $100 billion in deposit withdrawals were expected for that day. In contrast, depositors withdrew about $17 billion over the course of eight business days in September 2008 during the run on Washington Mutual, a much larger bank. While SVB and Signature Bank were placed in receivership, concerns over broader contagion led to sizable declines in bank stock prices. The KBW Bank Index and KBW Regional Bank Index fell by as much as 20% and 25% on subsequent days. The declines in bank stock prices led to yet more deposit outflows and eroded confidence in the banking system. To avert potentially far-reaching consequences of these developments, the Federal Reserve and other agencies took decisive actions to protect depositors and maintain confidence in the banking sector. On March 12, the Federal Reserve announced the establishment of the BTFP. The BTFP alloweddepositoryinstitutionsthatwereeligibleforprimarycreditattheFederalReserve’sdiscount 6

window to borrow for up to one year at a fixed rate of 10 basis points over the one-year overnightindexed swap (OIS) rate. The longer loan term helped provide a stable source of funding for depository institutions. In addition, BTFP collateral was valued at par, rather than market value. Eligible collateral was limited to open-market operation eligible collateral, and included Treasury securities,agencydebt,andagencyMBSandCMOs. Nohaircutswereappliedtoeligiblecollateral. TheleftpanelofFigure2showsthattheinterestrateonBTFPloansincreasedsteadilybetween March and October before it gradually declined in late 2023, reflecting market expectations for interest rates over the next year. The interest rate applicable to new BTFP loans was adjusted on January 24, 2024, to be no lower than the interest rate on reserve balances (IORB) on the same day the loan was made, partially reflecting the stabilization of market funding conditions. The right panel of Figure 2 plots the outstanding amount of BTFP borrowing (in orange) and of collateral pledged (in blue) over the program. The outstanding amount of BTFP loans and the collateral pledged at the BTFP increased rapidly in the first few months of the program, before stabilizing during the second half of 2023. Interestingly, depository institutions pledged substantially more collateral than what they used to borrow, which may reflect a precautionary motive to access the BTFP if and when they need to. After January 24, 2024, when the Federal Reserve announced that the BTFP would cease making new loans as scheduled on March 11, the amount of pledged collateral fell while borrowing gradually declined. Figure 2: BTFP rate and uptake 5.6 5.4 5.2 5.0 4.8 4.6 4.4 M ar A pr M ay J u n J ul Au g S e p Oct N ov D ec J a n Fe b M ar Date (Daily) )%( etaR PFTB 300 600 200 400 100 200 0 0 M ar A pr M ay J u n J ul Au g S e p Oct N ov D ec J a n Fe b M ar Date (Weekly) )liB $( gniworroB PFTB BTFP Collateral Pledged ($ Bil) BTFP Rate (Daily) BTFP Borrowing ($ Bil) BTFP Collateral Pledged ($ Bil) Note: The left panel of the figure shows BTFP rate (in %). The right panel of the figure shows BTFP borrowing and collateral pledging ($B). The time period is between March 12, 2023 and March 11, 2024. 7

3 Methodology and Data 3.1 Methodology Our analysis centers around three questions. First, what risk factors contributed to core deposit outflows during the banking turmoil? Second, how did banks manage these deposit outflows? Third, how did banks use BTFP and other funds to fund loans and accumulate reserve balances? To address the first question, we run cross-sectional regressions predicting changes in banks’ core deposits over four eight-week intervals. We principally focus on changes between March 8 and May 3, because this period covers the onset of the banking turmoil. We also analyze changes between May 3 and June 28, June 28 to August 23, and August 23 to October 18 to study deposit outflows during the period in which banking conditions stabilized. We regress the change in core deposits on bank-specific risk factors measured as of 2022q4. Formally, we estimate the equation: ∆Deposits Uninsured deposits HTM Securities Losses i,t,t(cid:48) = β i,22q4 +β i,22q4 +γ(cid:48)X +ε (1) 1 2 i,t i,t Assets Assets Assets i,t i,22q4 i,22q4 where i indexes banks, t and t(cid:48) denote the start and end of the eight-week window under consideration, and uninsured deposits and HTM securities losses (both normalized by a bank’s total assets as of 2022q4) are the primary risk factors of interest. We additionally consider specifications interacting the two risk factors to capture the fact that HTM securities losses may predominantly affect the incentives of uninsured depositors to withdraw funds (Jiang et al., 2023). We focus on the HTM security portfolio since banks are restricted in hedging interest rate risk for HTM securities and face greater impediments to selling these assets, given the marked-to-market valuations involved that would result in accounting losses (Granja et al., 2024). However, the findings are broadly similar when considering security losses combined between AFS and HTM portfolios.1 X i,t is a vector of controls that includes loan growth, ∆Loans /Assets , to account for the fact that i,t,t(cid:48) i,t loan growth can cause deposit growth if funds are initially held in the borrower’s deposit account (Glancy et al., 2020; Cooperman et al., 2023). We also include balance sheet items from 2022q4 Call Report filings: liquid asset holdings (the sum of cash, federal funds, reverse repo and securities holdings) as a share of total assets, deposit growth leading into the period of the banking turmoil (i.e. 2022q1 to 2022q4), average funding cost (annualized interest expense as a percent of total liabilities), and FHLB advances as a share of total assets.2 Finally, we include bank size indicators 1AppendixTableB1presentsresultsfromtheinteractionofuninsureddepositswithbothHTMandAFSsecurities losses. It shows that AFS security losses also interact with uninsured deposits, but the effect is smaller. 2Controlling for loan growth in this regression could conceptually be problematic if banks reduced loans to meet deposit outflows. However, banks are unlikely to adjust lending at the high frequencies studied in this paper. When banks change lending policies, it tends to affect loan balances over the course of the following years rather than 8

that equal one if a bank is a community bank (under $10 billion in total assets) or regional bank (between $10 and $250 billion in assets).3 Second, we investigate how banks managed core deposit outflows, either by borrowing funds from the BTFP or other sources or by reducing liquid asset holdings. We replace the left-hand side variable of specification (1) with changes in either BTFP borrowing to total assets or changes in other balance sheet items that could be used to meet funding needs (e.g., FHLB advances, securities sales), and test whether banks that had characteristics that made them vulnerable to deposit outflows adjusted more to compensate. Formally, we estimate the regression: ∆Y ∆Deposits HTM Securities Losses i,t,t(cid:48) i,t,t(cid:48) i,22q4 = β +β 1 2 Assets Assets Assets i,t i,t i,22q4 (2) Uninsured deposits +β i,22q4 +γ(cid:48)X +ε , 3 i,t i,t Assets i,22q4 where ∆Y /Assets is the change in some balance sheet item over the eight-week interval of i,t,t(cid:48) i,t interest and ∆Deposits /Assets is the change in core deposits during this time. β estimates i,t,t(cid:48) i,t 1 how much banks adjust balance sheet item Y in response to a dollar of deposit outflows. In some specifications,weinteractdepositoutflowswithHTMsecuritylossestoinvestigatehowtheresponse to deposit outflows differs across banks. Finally, X is a vector of the same variables included as i,t controls in equation (1).4 Third,tobetterunderstandthedynamicsinhowbanksusedwholesaleborrowing,weinvestigate the relationship between banks’ borrowing decisions and changes in deposits, cash holdings, and loans over time. Specifically, instead of using a fixed window to estimate the relationship between borrowing and other balance sheet developments, we use local projections to trace out changes in bank balance sheet items at various horizons surrounding borrowing. Specifically, we estimate: y −y ∆BTFP i,t+h i,t−1 = βh i,t +τh+(cid:15)h (3) Assets Assets t i,t i,t−1 i,t−1 whereτh isaweekfixedeffect. {βh}tracesthepathofbalancesheetvariablessurroundingincrease t in BTFP balances relative to other banks that did not change BTFP usage in a particular week. We also estimate a similar specification for non-Fed borrowings. The sample for this analysis includes t ranging from March 15th to July 26th. We investigate leads/lags up to 10 weeks (i.e., h ∈ [−10,−9,...,10]), so the estimation data includes balance sheet changes from January 4th to following weeks (Bassett et al., 2014; Castro et al., 2022). 3The omitted category is banks with total assets that exceed $250 billion. 4For robustness, we also estimate equation (2) with Excess Loan Growth—defined as (∆Loans i,t,t(cid:48) − ∆Deposits )/Assets —as the main independent variable. This measure reflects the additional funding need i,t,t(cid:48) i,t the bank needs to raise (besides from core deposits) to fund the growth in their loan portfolio. 9

October 4th. 3.2 Data Our analysis primarily draws on form FR 2644, which provides detailed data on banks’ assets and liabilities at a weekly frequency. To focus attention on institutions participating in conventional banking activities, we restrict the sample to domestically chartered banks, and omit banks in companies principally involved in investment banking or custodian activities.5 On the asset side, the data provide a detailed classification of different loan types, as well as securities and cash holdings. On the liabilities side, the data include core and large time deposits balances, as well as aggregate wholesale borrowings.6 Wholesale borrowings include FHLB advances, repos, federal funds, discount window and BTFP borrowing. We isolate Federal Reserve borrowing from other wholesale borrowings by separately obtaining discount window and BTFP borrowings data from internal Federal Reserve records.7 Outside of the largest banks, which are more active in repo markets,theremainingpartofwholesaleborrowingsismostlycomposedofFHLBadvances. Finally, we rely on bank Call Reports data as of December 2022 to measure ex-ante vulnerabilities (e.g., uninsured deposit shares and security losses). 3.3 Summary statistics To gauge the scope of coverage of our sample, Figures A1-A8 of Appendix A compare certain balance sheet characteristics of FR 2644 filers to the universe of banks (Call Reports data). In general, FR 2644 banks have very similar characteristics to the corresponding sample of all commercial banks (core deposits, FHLB advances, unrealized securities loss relative to total assets). The most significant differences are total assets and the share of uninsured deposits (Figures A1 and A8), where our FR 2644 sample banks are, on average, larger and rely more on uninsured deposits than the average commercial bank in the Call Reports universe. Therefore, our sample is relevant to study the regional banks and banks with high reliance on uninsured deposits, which were disproportionately affected by the banking episode in March 2023. Table 1 provides additional background information for the banks in our sample. It distinguishes between borrowers and non-borrowers from the BTFP to explore differences in balance sheet characteristics for the banks that accessed the program. Three observations stand out. First, 5Though credit unions also participated meaningfully in the BTFP, we restrict our attention to banks due to the lack of high frequency credit union data. 6“Core”depositsaredefinedasliquid(non-CD)depositsplussmall(under$100,000)timedeposits. Thisdefinition includesbrokereddepositsthathavebeenparticipatedoutinsharesoflessthan$100,000,sowemayunderestimate depositoutflowstotheextentthatbanksoffsetsomeofthedepositoutflowsbyraisingbrokereddepositsofthistype. 7These are daily data, which we aggregate at the weekly level following the timing convention in FR 2644 forms (i.e, Wednesday close of business levels). 10

BTFP borrowers are substantially smaller than non-borrowers. In general, smaller banks are less able to access market-based funding than larger banks, both in normal times and under stress. Second, BTFP borrowers appear to have greater reliance on uninsured deposits as a source of funding and larger unrealized HTM securities losses relative to assets. And third, possibly related to these risk factors, BTFP borrowers appear to have weaker core deposit growth and a greater reliance on FHLB advances in the run-up to banking turmoil. Table 1: Summary Statistics Uninsured CoreDeposit UnrealizedHTM FHLBadv FHLB Assets Deposit # Institutions Growth SecuritiesLosses toAssets membershr ($B) toAssets (%) (%) (%) (%) (%) A. 2022 Q4 FR2644 Banks 668 20.74 −4.52 28.57 .40 3.01 94 Borrowers 82 6.26 −7.57 32.22 .65 3.62 98 Non-borrowers 586 22.77 −4.09 28.05 .36 2.93 93 B. 2023 Q1 FR2644 Banks 668 20.98 −9.55 26.56 .34 3.45 94 Borrowers 82 6.48 −16.13 29.05 .57 3.76 98 Non-borrowers 586 23.00 −8.63 26.21 .31 3.41 93 C. 2023 Q2 FR2644 Banks 668 20.87 −9.25 25.19 .38 3.46 94 Borrowers 82 6.41 −8.25 27.54 .63 2.94 98 Non-borrowers 586 22.9 −9.39 24.85 .35 3.53 93 D. 2023 Q3 FR2644 Banks 668 21.00 −0.78 25.3 .50 3.24 94 Borrowers 82 6.45 0.90 27.45 .76 2.93 98 Non-borrowers 586 23.07 −1.02 24.99 .46 3.29 93 Notes: Thetableshowssummarystatisticsfor2022Q4CallReportdataforbanksandthesubsetofbanksincludedintheFR2644, broken down by their respective BTFP borrower status (as of 5/3/2023). The uninsured deposits share is defined as uninsured depositsrelativetototaldeposits;unrealizedsecuritieslosses,FHLBadvancesarerelativetototalassets. 4 Results 4.1 Which banks experienced core deposit outflows? In this section, we investigate what risk factors were associated with deposit outflows and, consequently, financing needs that could be addressed by the BTFP. Table 2 regresses changes in core deposits on banks’ 2022q4 uninsured deposit volumes and HTM security losses (all measured as a share of total assets). Columns 1-2 investigate changes in the first eight weeks of stress (from March 8th to May 3rd), while columns 3-8 investigate changes over the following 24 weeks (divided in three intervals of less acute stress). Column 1 shows that uninsured deposits played a significant role in prompting deposit outflows 11

during the period of most acute stress. The estimates imply that core deposits fell by about 4 cents for every dollar of uninsured deposits at a bank. This effect is sizable: a one standard deviation increaseinuninsureddepositsasashareofassets(12.8%)resultsinabouta50basispointdeclinein core deposits as a fraction of assets (relative to a sample mean of -1.7%). Securities losses, however, appear to have little effect on deposit outflows: a one standard deviation increase in HTM losses to assets (0.9%) only raises deposit outflows by 2 basis points as a fraction of assets, and the result is not statistically significant. Uninsured depositors would have a stronger incentive to withdraw if there were concerns about the risk of failure at their bank. To reflect differences in risk, column 2 interacts the uninsured deposit share with securities losses. The findings show that the effects of having high uninsured deposits are predominantly driven by banks with higher securities losses. The estimates imply that deposit outflows for banks with no security losses exhibit a slightly smaller sensitivity to uninsured deposit shares (core deposits at such banks fall by 3 cents per dollar of uninsured deposits.) However, deposit outflows for a bank where securities losses were 2.4 percent of assets (roughly the 90th percentile of the distribution for the sample) are estimated to be nearly twice as sensitive to uninsured deposits as overall sample, with core deposits falling by about 10 cents for every additional dollar of uninsured deposits. These effects of uninsured deposits and securities losses appear to be predominantly confined to the two months following the bank failures. Columns 3-8 present estimates from the same specification, but predicting changes in deposits over the course of May and June, July and August or September and October. Uninsured deposits and securities losses areno longer predictive of core deposit outflows during these intervals. In unreported regressions, we also conduct similar analysis of the period leading into the runs (January 11 to March 8). Uninsured deposits were weakly associated with deposit outflows during this period, likely reflecting those borrowers being more rate sensitive. In contrast, held-to-maturity securities losses and its interaction with uninsured deposits were not predictive of outflows. 12

Table 2: Determinants of Core Deposit Outflows ∆Deposits Assets March 8 to May 3 May 3 to June 28 June 28 to August 23 August 23 to October 18 (1) (2) (3) (4) (5) (6) (7) (8) UninsuredDeposits -0.04** -0.03** 0.01 0.01 0.00 0.00 0.01+ 0.02* Assets (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) × HTMSecurityLosses -0.03** -0.00 0.01 -0.01 Assets (0.01) (0.01) (0.01) (0.01) HTMSecurityLosses -0.02 0.89* 0.13 0.23 0.23* -0.07 -0.04 0.26 Assets (0.13) (0.37) (0.11) (0.32) (0.11) (0.33) (0.11) (0.31) ∆Loans 0.27** 0.26** 0.35** 0.35** 0.26** 0.26** 0.22** 0.22** Assets (0.07) (0.07) (0.07) (0.07) (0.08) (0.08) (0.08) (0.08) LiquidAssets -0.02** -0.02** -0.05** -0.05** -0.03** -0.04** -0.02** -0.02** Assets (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) Avg. Funding Cost 0.60** 0.60** -0.21 -0.21 0.55** 0.55** 0.26 0.26 (0.23) (0.23) (0.21) (0.21) (0.20) (0.20) (0.19) (0.19) 2022 Dep. Growth 0.04** 0.04** 0.00 0.00 -0.01 -0.01 0.01 0.01 (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) FHLBAdvances 0.04 0.04 0.01 0.01 -0.03 -0.03 0.00 0.00 Assets (0.03) (0.03) (0.03) (0.03) (0.03) (0.03) (0.02) (0.02) Community Bank 0.88 0.80 -0.24 -0.25 0.95 0.98 1.34* 1.30+ (0.79) (0.78) (0.76) (0.76) (0.72) (0.72) (0.67) (0.67) Regional Bank 1.59+ 1.55+ 0.53 0.52 1.00 1.02 1.26+ 1.24+ (0.82) (0.81) (0.78) (0.78) (0.74) (0.74) (0.69) (0.69) Constant -1.90* -2.14* 0.41 0.39 -0.58 -0.51 -1.47+ -1.53* (0.91) (0.91) (0.85) (0.86) (0.82) (0.83) (0.77) (0.77) R2 0.119 0.127 0.113 0.112 0.068 0.068 0.053 0.053 a Observations 668 668 674 674 678 678 674 674 Notes: Thetableshowsresultsfromcross-sectionalregressionsofchangesinbanks’coredepositsrelativetobeginning onperiodtotalassets(∆Deposits /Assets )overthreeeight-weekintervalsonuninsureddeposits,HTMsecurity i,t,t(cid:48) i,t losses, contemporaneous loan growth (∆Loans i,t,t(cid:48)/Assets i,t ), liquid asset holdings (the sum of cash, federal funds, reverserepoandsecuritiesholdings),averagefundingcost(annualizedinterestexpenseasapercentoftotalliabilities), depositgrowthin2022,FHLBadvances,andbanksizeindicators(i.e.,communitybanksareoneswithassetsunder $10 billion in assets and regional banks ones with assets between $10 and $ 250 billion in assets) measured as of 2022:Q4. Coredepositandloangrowtharewinsorizedatthe1and99percentlevels. Allvariablesexceptindicators, depositgrowthin2022,andaveragefundingcostarerelativetototalassets. +,*,**indicatesignificanceat10%,5%, and 1%, respectively. 13

4.2 How did banks manage core deposit outflows? Wholesale Borrowing from the BTFP and Private Sources: Next, we focus on how banks managed these deposit outflows. Table 3 investigates how banks adjusted non-deposit borrowing from March 8 to May 3 in response to deposit outflows during that interval. Columns 1 and 3 predict changes in BTFP borrowing as a share of total assets, while columns 4 and 6 predict changes in non-Fed borrowing (borrowings besides BTFP and discount window) as a percent of total assets.8 The results in column 1 suggest that a one dollar decline in core deposits, on average, resulted in about an 8 cent increase in BTFP borrowing. Risk factors that were associated with deposit outflows also affected BTFP borrowing beyond their effect on actual deposit outflows (column 2). A one standard deviation increase in HTM security losses and uninsured deposit share were associated with increases of BTFP borrowing of 14 basis points and 26 basis points, respectively, as a share of assets. These results are suggestive of BTFP being used both to meet realized outflows, but also to raise precautionary funding when a bank has features that could make it vulnerable to potential future outflows. Another reason that banks with greater HTM securities losses might access BTFP funding is that such losses make BTFP funding preferable to other alternatives, such as FHLB borrowing. First, high HTM losses may indicate that a bank’s asset holdings are less liquid, causing it to increase borrowing rather shedding liquid assets. That is, banks holding cash or AFS securities wouldbemoreabletoliquidateassetsthanbankspredominantlyholdingHTMsecurities.9 Second, by valuing securities at par, the BTFP increased borrowing capacity more for banks with greater unrealized losses. For both of these reasons, banks with high HTM securities losses might be expected to meet more of their marginal funding needs with BTFP. To examine the effect of unrealized security losses on banks’ marginal borrowing decisions, column 3 interacts deposit outflows with HTM securities losses. This specification allows us to assess how much banks differ in their propensity to fund deposit outflows with BTFP borrowing. The estimates imply that banks with negligible losses on securities only fund about 4 cents of every dollar of deposit outflows with BTFP funding. However, a bank at the 90th percentile of losses (2.4%) would replace about 26 cents of every dollar of deposit outflows with BTFP borrowing. An important question remains: Is this higher demand for BTFP borrowing reflective of the program’s design (including the at-par valuation of collateral) or does it reflect broader demand for wholesale borrowing? To assess this, columns 4, 5 and 6 repeat the same analysis, but for non-Fed 8Variation in discount window borrowing was minimal. While discount window borrowing spiked at the onset of thestress,byMay3rdthemostseriouslystrainedbankswereclosed,andborrowingreturnedtonearpre-SVBlevels. 9Indeed, banks with greater HTM securities losses disproportionately used private borrowing even before the banking stress in March. 14

borrowing, consisting predominantly of FHLB advances. This borrowing was a more prominent source of funding overall, with banks increasing borrowing by about 26 cents for every dollar of outflows. However, the interaction effect between deposit outflows and HTM losses seen for BTFP fundingismissingfornon-Fedborrowing. BankswithgreaterHTMsecuritylosseswereslightlyless likely to meet deposit outflows with non-Fed borrowing, though the difference is not statistically significant. This result indicates that the high utilization of BTFP for banks with greater security losses did not reflect a general increase in the demand for wholesale funding (e.g., to build liquidity buffers in anticipation of outflows), but instead the choice of banks with greater unrealized losses opting to borrow from the BTFP. 15

Table 3: Wholesale Borrowing ∆BTFPBorrowings ∆Non-FedBorrowings Assets Assets (1) (2) (3) (4) (5) (6) ∆Deposits -0.08** -0.07** -0.04 -0.26** -0.32** -0.33** Assets (0.02) (0.03) (0.03) (0.04) (0.04) (0.04) × HTMSecurityLosses -0.09** 0.02 Assets (0.03) (0.04) HTMSecurityLosses 0.15+ -0.04 0.05 0.09 Assets (0.09) (0.11) (0.12) (0.15) UninsuredDeposits 0.02** 0.02* 0.00 0.00 Assets (0.01) (0.01) (0.01) (0.01) ∆Loans 0.09+ 0.10* 0.25** 0.25** Assets (0.05) (0.05) (0.07) (0.07) LiquidAssets 0.01 0.01 -0.04** -0.04** Assets (0.01) (0.01) (0.01) (0.01) Avg. Funding Cost 0.23 0.21 -0.33 -0.32 (0.15) (0.15) (0.22) (0.22) 2022 Dep. Growth -0.02* -0.02* 0.01 0.01 (0.01) (0.01) (0.01) (0.01) FHLBAdvances 0.05* 0.05* -0.01 -0.01 Assets (0.02) (0.02) (0.03) (0.03) Community Bank 0.81 0.70 -2.00** -1.98** (0.52) (0.52) (0.73) (0.74) Regional Bank 0.82 0.76 -0.93 -0.92 (0.54) (0.54) (0.76) (0.76) Constant 0.43** -1.51* -1.27* 0.29* 3.13** 3.08** (0.09) (0.60) (0.61) (0.12) (0.85) (0.86) R2 0.014 0.055 0.066 0.075 0.146 0.145 a Observations 668 668 668 668 668 668 Notes: The table shows results from cross-sectional regressions of changes in banks’ BTFP borrowings and non-Fed borrowings (wholesale borrowings that exclude BTFP and discount window) over the interval March 8—May 3 on contemporaneous deposit growth (∆Deposits /Assets ) uninsured deposits, HTM security losses, contemporai,t,t(cid:48) i,t neous loan growth (∆Loans i,t,t(cid:48)/Assets i,t ), liquid asset holdings (the sum of cash, federal funds, reverse repo and securitiesholdings),averagefundingcost(annualizedinterestexpenseasapercentoftotalliabilities),depositgrowth in 2022, FHLB advances, and bank size indicators (i.e., community banks are ones with assets under $10 billion in assets and regional banks ones with assets between $10 and $ 250 billion in assets) measured as of 2022:Q4. Core deposit and contemporaneous loan growth are winsorized at the 1 and 99 percent levels. All variables except indicators, deposit growth in 2022, and average funding cost are relative to total assets. +,*,** indicate significance at 10%, 5%, and 1%, respectively. 16

Overall, the findings in Table 3 suggest that BTFP borrowing was motivated by both precautionary motives and the need to fund deposit outflows. That banks with HTM losses were more reliant on BTFP to meet deposit outflows suggests that the at-par valuation of collateral supported demand for BTFP loans. We also see supporting evidence for these interpretations in banks’ decisions regarding the pledging of collateral to the BTFP. Table B3 presents similar estimates, but with the dependent variable being the amount collateral pledged to BTFP (as a share of assets). Two findings are notable. First, deposit outflows, HTM losses, and uninsured deposits prompt banks to raise collateral pledged to the BTFP by almost twice as much as they actually borrow.10 This result provides further evidence of precautionary motives for BTFP usage, as it shows that banks at more risk of outflows posted collateral to ensure the ability to borrow should the need arise. Second, this additional pledging was predominantly accounted for by MBS and CMO securities.11 Large Time Deposits and Liquid Assets: How did banks adjust balance sheet items besides wholesale borrowing in the face of core deposit outflows? We now document the other margins of adjustment, namely the raising of large time deposits and the liquidation of other assets. Figure 3 presents estimates of how much banks turn to various sources based on regressions alongthelinesofthespecificationinequation(2). Whenthedependentvariableisanasset(namely, cash or security holdings), we flip the sign, so that we are predicting the extent to which banks reduce holdings of that asset to meet funding needs. Estimates in blue reflect how much a bank with no HTM losses (empirically, corresponding to the median amount of HTM losses since over half of banks only hold AFS securities) use a particular source to meet deposit outflows. Estimates in red reflect the predicted elasticity between a particular funding source and deposit outflows for a bank at the 90th percentile in HTM security losses to assets. Estimates predicting changes in borrowing come from columns 3 and 6 of Table 3, while the estimates underlying the other funding sources are displayed in Table B2. As previously discussed, BTFP loans played a small role for banks without HTM losses. Banks without HTM losses only increased BTFP borrowing by about 4 cents for each dollar of core deposit outflows. Instead, these banks predominantly reduced cash holdings or borrowed from non-Fed sources; banks without HTM losses reduced cash by about 47 cents, increased non-Fed borrowings by about 33 cents, and increased large-time deposits by about 7 cents for each dollar of core deposit outflows. 10The estimates from Table B3, column 1 indicate that one dollar in deposit outflows, HTM losses, or uninsured deposits raise total collateral pledged to BTFP by 12 cents, 26 cents and 5 cents, respectively, whereas in Table 3, column 2 BTFP borrowing was only found to increase by 7 cents, 15 cents, and 2 cents, respectively. 11Of the additional collateral pledged in response to deposit outflows, HTM losses, or uninsured deposits, roughly two-thirds are found to be driven by MBS/CMOs. 17

BTFP loans played a larger role for banks with greater HTM losses. Banks with greater HTM losses used BTFP funding more on the margin to fund deposit outflows (26 cents per dollar of core deposit outflows) and turned slightly less to the other four funding sources (reduced cash by 40 cents, increased non-Fed borrowing by 28 cents, raised large-time deposits by 5 cents, and actually increasedsecurityholdingsby4cents). Inshort, bankswithhighHTMlossesreliedmoreonBTFP to meet deposit outflows, but funding decisions were otherwise similar to that of the rest of the banking sector. Figure 4 presents similar estimates, but disaggregates the sample across banks of different sizes. Broad patterns of how banks fund deposit outflows are similar; in the absence of notable HTM security losses, both regional and community banks mostly respond to deposit outflows by raising private borrowing or reducing reserves, with BTFP borrowings having a relatively limited role into the funding mix. However, the shift towards BTFP funding for high-HTM loss banks is mostly driven by regional banks. The estimates imply that a regional bank in the 90th percentile of HTM losses would fund almost half of their deposit outflows by borrowing from the BTFP. Though there is some uncertainty about this exact point estimate due to the smaller subsample and the extrapolation to a relatively high level of HTM-losses, this result provides some useful clarity on which banks utilized BTFP most heavily, namely regional banks with high HTM losses and notable deposit outflows. Tosummarizetheresultsofthissection, mostbanksdidnotuseBTFPfunding. Thosethatdid use BTFP borrowing appear to do so for a combination of precautionary reasons (raising funding orpostingcollateraliftheyhadriskfactorslinkedtodepositoutflows)ortoreplacerealizeddeposit outflows. Though banks typically met such funding needs by raising private borrowing or reducing cashholdings,bankswithhighHTMsecuritylossesreliedmoreheavilyonBTFPfunding,especially regionalbanks. WhilethishigherrelianceonBTFPforhigh-HTM-lossbanksmayreflectnumerous factors—for example, precautionary demand for liquidity, less liquid initial asset holdings, or a greater benefit from at-par collateral valuation—the main takeaway is the same: BTFP provided a backstop source of funding that enabled banks to manage funding pressures during the period of most acute stress. 18

Figure 3: How Banks Fund Core Deposit Outflows: Increase Liabilities and Reduce Assets 60 50 40 30 20 10 0 -10 -20 )stnec( metI teehS ecnalaB ni egnahC swolftuO tisopeD eroC fo ralloD rep Bank HTM Security Loss Median 90th Percentile BTFP Private Large Time Reserve Security Borrowing Wholesale Deposits Reduction Sales Borrowing Notes: Chart presents estimates of how much various funding sources are used (in cents) to meet a $1 increase in funding needs. Blue dots (lines) present point estimates (95% confidence intervals) for how much aparticularsourceisusedforabankwithnoHTMlosses(themedian). Reddotsandlinespresentequivalent estimates for a bank at the 90th percentile in HTM securities losses as a share of assets. The regressions underlying these estimates are in Table B2. Figure4: HowBanksFundDepositOutflows, ByBankSize: IncreaseLiabilitiesandReduceAssets 80 60 40 20 0 -20 )stnec( metI teehS ecnalaB ni egnahC swolftuO tisopeD eroC fo ralloD rep Bank HTM 80 Security Loss Median 90th Percentile 60 40 20 0 -20 BTFP Private Large Time Reserve Security Borrowing Wholesale Deposits Reduction Sales Borrowing (a) Regional Banks )stnec( metI teehS ecnalaB ni egnahC swolftuO tisopeD eroC fo ralloD rep Bank HTM Security Loss Median 90th Percentile BTFP Private Large Time Reserve Security Borrowing Wholesale Deposits Reduction Sales Borrowing (b) Community Banks Notes: Charts present estimates of how much various funding sources are used (in cents) to meet a $1 increase in funding needs. Left figure shows estimates for regional banks (those with assets between $10 and $100 billion) while the right shows estimates for community banks (those with assets under $10 billion). Blue dots (lines) present point estimates (95% confidence intervals) for how much a particular source is used for a bank with no HTM losses (the median). Red dots and lines present equivalent estimates for a bank at the 90th percentile in HTM securities losses as a share of estimates. 19

Robustness Deposit flows are not the only development affecting banks’ needs for funding. Banks similarly need to manage their balance sheets to accommodate lending opportunities. Interesting, the coefficients on Loan Growth in Tables 3 and B2 demonstrate that loan growth has a roughly symmetric effect on banks’ funding decisions: banks increased BTFP borrowing by 10 cents per dollar of loan growth, while reducing cash by 45 cents, increasing private borrowing by 24 cents and increasing large-time deposits by 7 cents. Given this symmetry, we show that the broad patterns shown in Figure 3 also hold when we measure a banks’ funding needs by Excess Loan Growth (the difference between loan growth and deposit growth as a share of initial assets). This measure reflects how banks respond to funding needs generated by either loan growth or deposit outflows. FigureB1andTableB4showthatthemainresultsarerobusttothisalternativemeasure. Deposit Pricing Most of the analysis of this section treated deposit outflows as being outside of banks’ immediate control. However, if banks are facing deposit pressures, one potential response is to increase the rates paid on deposits to stem the outflows rather than trying to adjust other parts of the portfolio. Table B5 estimates from the specification in equation (2), but with changes time deposit pricing at various horizons as the dependent variable. There is some evidence of adjustment along the price dimension; one percentage point higher deposit outflows or HTM losses raise 3-month CD rates by about 2 and 7 basis points, respectively. However, there is little effect on CDs at other time horizons. Additionally, these effects are second-order relative to differences in deposit pricing across banks of different sizes, so the implications of these pricing changes for deposit flows are likely modest over such short time intervals. 4.3 Uses of Wholesale Borrowings: Dynamics of Deposits, Cash, and Loans Finally, we shift our attention from investigating what prompts banks to borrow from BTFP to asking how BTFP funds were used. Instead of investigating changes in bank balance sheets over one specific window, we use local projections to trace out how various balance sheet items change in the period surrounding the increase in borrowing. The next three figures plot coefficient estimates from regressions similar to that in specification (3), showing how core deposits (Figure 5), cash holdings (Figure 6) and loan balances (Figure 7) evolve in the period surrounding an increase in BTFP usage (left panels) or an increase in non-Fed borrowing (right panels). Both the independent and dependent variables are scaled by assets, so the impulses can be interpreted as the predicted change in the balance sheet item in dollars, per dollar increase in borrowing. To preview results, we find that banks use BTFP to build up cash holdings (consistent with BTFP helping banks to satisfy precautionary demand for liquidity) and to meet fund needs arising 20

from loan growth and deposit outflows. These results thus provide some additional support for the narrative proposed in Section 4.2, but with more detail on the timing involved. Dynamics of Core Deposits: Starting with deposit dynamics, Figure 5 shows that banks tend to use BTFP after a sustained period of weak core deposit growth. Banks with one dollar higher borrowingfromBTFPonaveragehadabout40centsmoreincoredepositoutflowsovertheprevious 10 weeks. The week 0 effect gives the change in deposits in the week of actual BTFP usage, which is small and statistically insignificant. The right panel shows that dynamics for non-Fed borrowings are quite different. Banks borrow from these sources in the actual week that they experience deposit outflows. On average, deposits fallabout30centsforeverydollarofnon-Fedborrowinginagivenweek. Thereisnotaparticularly strong trend in core deposits leading into an increase in non-Fed borrowings. Inshort,non-Fedborrowingsappeartobeusedtofunddepositoutflowsintheweekstheyoccur, whereas BTFP appears to be used after a period of sustained outflows. This outcome is consistent with BTFP having some fixed cost that inhibits use despite features that make it desirable relative to other funding sources—for example, pre-payability or having collateral valued at par. This fixed cost causes banks to wait until funding needs to pass some threshold level of need for it to be worth paying the cost to achieve the favorable source of funding. Non-Fed borrowing, on the other hand, is not subject to these frictions, and thus can be used immediately when deposits decline. Figure 5: Relationship Between Core Deposits Growth and Borrowing .6 .4 .2 0 -.2 -.4 stisopeD eroC ni egnahC 0 -.1 -.2 -.3 -.4 -10 -5 0 5 10 Weeks Since BTFP Borrowing (a) Change around BTFP Borrowing stisopeD eroC ni egnahC -10 -5 0 5 10 Weeks since non-Fed Borrowing (b) Change around non-Fed Borrowing Notes: The charts show the dynamic relationship between banks’ weekly core deposits and BTFP borrowings (panel a) or non-Fed borrowings (panel b). The coefficients show the relationship between changes in core deposits at various horizons (in dollars) surrounding a$1 increase in borrowings at each bank in the presence of week fixed effects. The week 0 effect gives the change in deposits in the week of actual borrowings. Dynamics of Cash Holdings: Figure 6 provides similar analysis for cash holdings. One potential use of borrowing is to increase reserves in anticipation of deposit outflows (or replenish reserves 21

that were used to fund earlier outflows). The figure suggests that some of each type of borrowing is used for this function. In the week of BTFP usage, cash holdings increase by about a third of the amountofBTFPusageandthenremainselevatedoverthenext10weeks. Fornon-Fedborrowings, banks on average see a decline in cash the week before borrowing, followed by a larger rebound in the week they borrowed, indicating that borrowing reflects a mixture of replenishing reserves as well as building up reserves to meet future needs. Figure 6: Relationship Between Cash Holdings and Borrowing .6 .4 .2 0 -.2 sgnidloH hsaC ni egnahC .6 .4 .2 0 -.2 -10 -5 0 5 10 Weeks Since BTFP Borrowing (a) Change around BTFP Borrowing sgnidloH hsaC ni egnahC -10 -5 0 5 10 Weeks since non-Fed Borrowing (b) Change around non-Fed Borrowing Notes: The charts show the dynamic relationship between banks’ weekly cash holdings and BTFP borrowings (panel a) or non-Fed borrowings (panel b). The coefficients show the relationship between changes in cash holdingsatvarioushorizons(indollars)surroundinga$1increaseinborrowingsateachbankinthepresence of week fixed effects. The week 0 effect gives the change in cash holdings in the week of actual borrowings. Dynamics of Loan Balances: Figure 7 plots changes in loan balances surrounding a change in borrowings. Overall, the patterns look similar for the two types of borrowings. Banks experience a modest increase in loan balances around the time of borrowing—on the order of 5 cents of loan growth per dollar of borrowing in the couple of months both before and after borrowing—but the estimates are for the most part statistically insignificant. Though banks are likely funding some loans on the margin using BTFP, an increased usage of borrowing may be indicative of funding pressures that would cause banks to restrict credit supply. These offsetting mechanisms would dampen the relationship between BTFP borrowing and loan growth. 22

Figure 7: Relationship Between Loan Growth and Borrowing .2 .1 0 -.1 -.2 secnalaB naoL ni egnahC .2 .1 0 -.1 -.2 -10 -5 0 5 10 Weeks Since BTFP Borrowing (a) Change around BTFP Borrowing secnalaB naoL ni egnahC -10 -5 0 5 10 Weeks since non-Fed Borrowing (b) Change around non-Fed Borrowing Notes: Thechartsshowthedynamicrelationshipbetweenbanks’weeklyloansandBTFPborrowings(panela) or non-Fed borrowings (panel b). The coefficients show the relationship between changes in loans at various horizons (in dollars) surrounding a $1 increase in borrowings at each bank in the presence of week fixed effects. The week 0 effect gives the change in loans in the week of actual borrowings. 23

5 Conclusion This paper examines banks’ experience with deposit outflows during the 2023 banking turmoil and the effectiveness of the BTFP. First, we find that banks with high reliance on uninsured deposit funding and large marked-to-market losses on securities holdings faced relatively more deposit outflows. Second, we find that in response to deposit outflows, banks primarily reduced cash holdings and borrowed from non-Fed sources (e.g., FHLB advances). BTFP borrowing played a relativelymodestroleinhowtheaveragebankmetfundingneeds, butwasamoreimportantsource of funding for banks with relatively larger securities losses or larger reliance on uninsured deposits. These factors were especially prominent for regional banks. Moreover, these risk factors were associatedwithBTFPborrowingevenbeyondtheireffectonrealizeddepositoutflows, suggestinga precautionarymotiveforBTFPborrowing. Inaddition,vulnerablebanksappearedtopreferBTFP to non-Fed borrowings, and shed less cash to cover funding needs, reflecting the advantage of the at-par collateral valuation at the BTFP. Third, banks increased non-Fed borrowings immediately in the weeks that deposits fell, but only used BTFP borrowing after a period of sustained deposit outflows,suggestingthatprogramusageentailedafixedcost. BanksborrowingfromtheBTFPand non-Fed sources displayed similar trend growth in loan balances, possibly suggesting that BTFP supported the loan balances of more vulnerable banks. 24

References Bassett, W. F., Chosak, M. B., Driscoll, J. C. and Zakrajˇsek, E. (2014). Changes in bank lending standards and the macroeconomy. Journal of Monetary Economics, 62, 23–40. Brunnermeier, M. and Pedersen, L. (2009). Market liquidity and funding liquidity. The Review of Financial Studies, 22 (6), 2201–2238. Caglio, C., Dlugosz, J. and Rezende, M. (2023). Flight to safety in the regional bank crisis of 2023. Carlson, M. and Macchiavelli, M. (2020). Emergency loans and collateral upgrades: How broker-dealers used federal reserve credit during the 2008 financial crisis. Journal of Financial Economics, 137 (3), 701–722. Carpinelli, L. and Crosignani, M. (2021). The design and transmission of central bank liquidity provisions. Journal of Financial Economics, 141 (1), 27–47. Castro, A., Glancy, D., Ionescu, F. and Marchal, G. (2022). What happens when banks tighten C&I loan supply? Choi, D. B., Goldsmith-Pinkham, P. and Yorulmazer, T. (2023). Contagion effects of the Silicon Valley bank run. Cipriani, M., Eisenbach, T. M. and Kovner, A. (2024). Tracing Bank Runs in Real Time. Federal Reserve Bank of New York Staff Reports, 1104. Cookson, J. A., Fox, C., Gil-Bazo, J., Imbet, J. F. and Schiller, C. (2023). Social media as a bank run catalyst. Cooperman, H. R., Duffie, D., Luck, S., Wang, Z. Z. and Yang, Y. (2023). Bank funding risk, reference rates, and credit supply. Journal of Finance, forthcoming. Diamond, D. and Dybvig, P. (1983). Bank runs, deposit insurance, and liquidity. Journal of Political Economy, 91 (3), 401–419. — and Rajan, R. (2011). Fear of fire sales, illiquidity seeking, and credit freezes. Quarterly Journal of Economics, 126 (2), 557–591. Drechsler, I., Drechsel, T., Marques-Ibanez, D. and Schnabl, P. (2016). Who borrows from the lender of last resort? Journal of Finance, 71(5), 1933–1974. —,Savov, A.andSchnabl, P.(2017).TheDepositsChannelofMonetaryPolicy.TheQuarterly Journal of Economics, 132(4), 1819–1876. —, — and — (2021). Banking on Deposits: Maturity Transformation without Interest Rate Risk. Journal of Finance, 76(3), 1091–1143. —, —, — and Wang, O. (2023). Banking on Uninsured Deposits. Egan, M., Hortacsu, A. and Matvos, G. (2017). Deposit competition and financial fragility: evidence from the us banking sector. American Economic Review, 107 (1), 169–216. Freixas, X., Parigi, B. and Rochet, J.-C. (2004). The lender of last resort: a twenty-first century approach. Journal of the European Economic Association, 2 (6), 1085–1115. Glancy, D., Gross, M. and Ionescu, F. (2020). How did banks fund c&i drawdowns at the onset of the covid-19 crisis? Granja, J., Jiang, E. X., Matvos, G., Piskorski, T. and Seru, A. (2024). Book value risk management of banks: Limited hedging, HTM accounting, and rising interest rates. (NBER Working Paper No. 32293). 25

Haddad, V., Glaser-Hartman, B. and Muir, T. (2023). Bank Fragility When Depositors Are the Asset. Hannan, T. and Berger, A. (1991). The Rigidity of Prices: Evidence from the Banking Industry. The American Economic Review, 81(4), 938–945. Iyer, R. and Puri, M. (2012). Understanding Bank Runs: The Importance of Depositor-Bank Relationships and Networks. American Economic Review, 102(4), 1414–1445. —, — and Ryan, N. (2016). A Tale of Two Runs: Depositor Responses to Bank Solvency Risk. Journal of Finance, 71(6), 2687–2726. Jasova, M., Mendicino, C. and Supera, D. (2021). Policy uncertainty, lender of last resort and the real economy. Journal of Monetary Economics, 118, 381–398. Jiang, E. X., Matvos, G., Piskorski, T. and Seru, A. (2023). Monetary tightening and u.s. bank fragility in 2023: mark-to-market losses and uninsured depositor runs? Journal of Financial Economics, forthcoming. Kotidis, A., Malliaropulos, D. and Papaioannou, E. (2022). Public and Private Liquidity During Crises Times: Evidence From Emergency Liquidity Assistance (ELA) to Greek Banks. McPhail, L.,Schnabl, P.andTuckman, B.(2023).Dobankshedgeusinginterestrateswaps? (NBER Working Paper No. 31166). Minoiu, C., Zarutskie, R. and Zlate, A. (2024). Motivating banks to lend? credit spillover effects of the main street lending program. FEDS Working Paper 2021-078, Federal Reserve Board. Neumark, D. and Sharpe, S. (1991). Market Structure and the Nature of Price Rigidity: Evidence from the Market for Consumer Deposits. The Quarterly Journal of Economics, 107(2), 657–680. Prescott, E.(2024). The failure of the bank of the commonwealth: An early example of interest rate risk. Economic Commentary. Rochet, J.-C. and Vives, X. (2004). Coordination failures and the lender of last resort: was bagehot right after all? Journal of the European Economic Association, 2 (6), 1116–1147. Rose, J. (2023). Understanding the speed and size of bank runs in historical comparison. Economic Synopses, (12), 1–5. Van Bekkum, S.,Gabarro, M.andIrani, R. M.(2018).Doesalargermenuincreaseappetite? collateral eligibility and credit supply. Review of Financial Studies, 31 (3), 943–979. 26

A Appendix Figure A1: Total Assets 8% 6% 4% 2% 0% 7 9 11 13 15 17 19 21 Log of Total Assets snoitavresbO fo % Sample All Banks FR 2644 Banks Note: The Figure shows the distribution of total assets for all commercial banks (4746 banks, shown in blue) and the subset of banks in our sample (668 banks, shown in orange). 27

Figure A2: Core Deposits to Total Assets 6% 4% 2% 0% 0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 100 Core Deposits over Total Assets (%) snoitavresbO fo % Sample All Banks FR 2644 Banks Note: The Figure shows the distribution of core deposits to total assets for all commercial banks (4746 banks, shown in blue) and the subset of banks in our sample (668 banks, shown in orange). Figure A3: FHLB Advances to Total Assets 6% 4% 2% 0% 0 5 10 15 20 FHLB Share of Total Assets snoitavresbo fo % Sample All Banks FR 2644 Banks Note: The Figure shows the distribution of FHLB advances to total assets for all commercial banks (4746 banks, shown in blue) and the subset of banks in our sample (668 banks, shown in orange). 28

Figure A4: Core Deposits Growth 6% 4% 2% 0% −60−55−50−45−40−35−30−25−20−15−10−5 0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95100 2022 Core Deposits Growth (%) snoitavresbO fo % Sample All Banks FR 2644 Banks Note: The Figure shows the distribution of core deposits growth in 2022 for all commercial banks (4746 banks, shown in blue) and the subset of banks in our sample (668 banks, shown in orange). Figure A5: Unrealized Securities Losses to Total Assets 15% 10% 5% 0% −2 −1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Unrealized Securities Loss Share of Total Assets (%) snoitavresbo fo % Sample All Banks FR 2644 Banks Note: TheFigureshowsthedistributionofunrealizedsecuritieslossestototalassetsforallcommercialbanks (4746 banks, shown in blue) and the subset of banks in our sample (668 banks, shown in orange). 29

Figure A6: HTM Unrealized Securities Losses to Total Assets 40% 30% 20% 10% 0% −2 −1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 HTM Loss Share of Total Assets (%) snoitavresbo fo % Sample All Banks FR 2644 Banks Note: TheFigureshowsthedistributionofHTMunrealizedsecuritieslossestototalassetsforallcommercial banks (4746 banks, shown in blue) and the subset of banks in our sample (668 banks, shown in orange). Figure A7: AFS Unrealized Securities Losses to Total Assets 10% 5% 0% −2 −1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 AFS Loss Share of Total Assets (%) snoitavresbo fo % Sample All Banks FR 2644 Banks Note: The Figure shows the distribution of AFS unrealized securities losses to total assets for all commercial banks (4746 banks, shown in blue) and the subset of banks in our sample (668 banks, shown in orange). 30

Figure A8: Uninsured Deposits to Total Assets 5% 4% 3% 2% 1% 0% 0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 100 Uninsured Deposits over Total Assets (%) snoitavresbO fo % Sample All Banks FR 2644 Banks Note: The Figure shows the distribution of uninsured deposits to total assets for all commercial banks (4746 banks, shown in blue) and the subset of banks in our sample (668 banks, shown in orange). 31

Table B1: Determinants of Deposit Outflows - Controlling for AFS Securities Losses ∆Deposits Assets March 8 to May 3 May 3 to June 28 June 28 to August 23 August 23 to October 18 (1) (2) (3) (4) (5) (6) (7) (8) UninsuredDeposits -0.04** -0.01 0.01 0.01 0.00 0.00 0.01+ 0.01 Assets (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) × HTMSecurityLosses -0.03** -0.00 0.01 -0.01 Assets (0.01) (0.01) (0.01) (0.01) × AFSSecurityLosses -0.01* 0.00 -0.00 0.00 Assets (0.00) (0.00) (0.00) (0.00) HTMSecurityLosses -0.02 0.98** 0.16 0.26 0.24* -0.04 -0.01 0.28 Assets (0.14) (0.37) (0.12) (0.33) (0.12) (0.33) (0.11) (0.32) AFSSecurityLosses 0.01 0.29+ 0.05 0.05 0.03 0.05 0.05 -0.00 Assets (0.07) (0.15) (0.06) (0.13) (0.06) (0.13) (0.06) (0.13) ∆Loans 0.27** 0.26** 0.35** 0.35** 0.26** 0.26** 0.22** 0.22** Assets (0.07) (0.07) (0.07) (0.07) (0.08) (0.08) (0.08) (0.08) LiquidAssets -0.02* -0.02* -0.05** -0.05** -0.04** -0.04** -0.03** -0.02** Assets (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) Avg. Funding Cost 0.61* 0.64** -0.18 -0.18 0.57** 0.57** 0.30 0.29 (0.24) (0.23) (0.21) (0.21) (0.21) (0.21) (0.20) (0.20) 2022 Dep. Growth 0.04** 0.04** 0.00 0.00 -0.01 -0.01 0.01 0.01 (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) FHLBAdvances 0.04 0.04 0.01 0.01 -0.03 -0.03 0.00 0.00 Assets (0.03) (0.03) (0.03) (0.03) (0.03) (0.03) (0.02) (0.02) Community Bank 0.87 0.74 -0.31 -0.32 0.90 0.93 1.27+ 1.24+ (0.79) (0.79) (0.76) (0.76) (0.73) (0.73) (0.67) (0.67) Regional Bank 1.58+ 1.50+ 0.48 0.47 0.97 0.98 1.20+ 1.19+ (0.82) (0.82) (0.78) (0.78) (0.75) (0.75) (0.69) (0.69) Constant -1.90* -2.66** 0.42 0.40 -0.58 -0.54 -1.46+ -1.42+ (0.92) (0.95) (0.85) (0.88) (0.82) (0.85) (0.77) (0.80) R2 0.117 0.130 0.113 0.110 0.067 0.065 0.053 0.052 a Observations 668 668 674 674 678 678 674 674 Notes: Thetableshowsresultsfromcross-sectionalregressionsofchangesinbanks’coredepositsrelativetobeginning onperiodtotalassets(∆Deposits /Assets )overthreeeight-weekintervalsonuninsureddeposits,HTMsecurity i,t,t(cid:48) i,t losses, AFS security losses, contemporaneous loan growth (∆Loans i,t,t(cid:48)/Assets i,t ), liquid asset holdings (the sum of cash, federal funds, reverse repo and securities holdings), average funding cost (annualized interest expense as a percentoftotalliabilities),depositgrowthin2022,FHLBadvances,andbanksizeindicators(i.e.,communitybanks are ones with assets under $10 billion in assets and regional banks ones with assets between $10 and $ 250 billion in assets) measured as of 2022:Q4. Core deposit and contemporaneous loan growth are winsorized at the 1 and 99 percent levels. All variables except indicators, deposit growth in 2022, and average funding cost are relative to total assets. +,*,** indicate significance at 10%, 5%, and 1%, respectively. 32

Table B2: Pecking Order of Funding Sources ∆BTFP Borrowings -∆Cash ∆NonFed Borrowings ∆LT. Dep. -∆Securities Assets Assets Assets Assets Assets (1) (2) (3) (4) (5) ∆Deposits -0.04 -0.47** -0.33** -0.07** -0.01 Assets (0.03) (0.05) (0.04) (0.02) (0.02) ×HTMSecurityLosses -0.09** 0.03 0.02 0.01 0.02 Assets (0.03) (0.05) (0.04) (0.02) (0.02) HTMSecurityLosses -0.04 -0.02 0.09 -0.00 0.09 Assets (0.11) (0.19) (0.15) (0.08) (0.06) UninsuredDeposits 0.02* -0.01 0.00 0.00 0.00 Assets (0.01) (0.01) (0.01) (0.00) (0.00) ∆Loans 0.10* 0.45** 0.25** 0.07* -0.01 Assets (0.05) (0.09) (0.07) (0.03) (0.03) LiquidAssets 0.01 0.03** -0.04** -0.02** 0.00 Assets (0.01) (0.01) (0.01) (0.00) (0.00) Avg. Funding Cost 0.21 0.37 -0.32 -0.10 -0.13 (0.15) (0.28) (0.22) (0.11) (0.09) 2022 Dep. Growth -0.02* 0.02+ 0.01 0.00 -0.01* (0.01) (0.01) (0.01) (0.00) (0.00) FHLBAdvances 0.05* -0.01 -0.01 -0.00 0.02+ Assets (0.02) (0.03) (0.03) (0.01) (0.01) Community Bank 0.70 0.67 -1.98** 0.45 -0.18 (0.52) (0.94) (0.74) (0.37) (0.30) Regional Bank 0.76 -0.35 -0.92 0.30 0.02 (0.54) (0.97) (0.76) (0.38) (0.31) Constant -1.27* -2.83** 3.08** 0.81+ 0.16 (0.61) (1.10) (0.86) (0.43) (0.36) R2 0.066 0.169 0.145 0.042 0.020 a Observations 668 668 668 668 668 Notes: Thetableshowsresultsfromregressingchangesinvariousbalancesheetitems—increasesinBTFP(Columns 1–2), reductions in cash holdings (Columns 3–4), increases in non-Fed Borrowings (Columns 5–6), increases in large timedeposits(Columns7–8), andreductionsinsecurityholdings—onbanks’excessloangrowthlikeinspecification (2). Among the explanatory variables, we also include losses on securities holdings (the difference between the value of security holdings when measured at market value and at amortized cost) as a share of assets, both in levels and interacted with excess loan growth. The remaining explanatory variables are like in Table 3. +,*,** indicate significance at 10%, 5%, and 1%, respectively. 33

Table B3: BTFP Collateral Pledging Total MBS/CMO Agency Securities Treasury Securities (1) (2) (3) (4) (5) (6) (7) (8) ∆Deposits -0.12** -0.06 -0.08* -0.05 -0.03* -0.02 -0.01 0.00 Assets (0.04) (0.05) (0.04) (0.04) (0.01) (0.02) (0.01) (0.01) × HTMSecurityLosses -0.16** -0.08* -0.04* -0.05** Assets (0.05) (0.04) (0.02) (0.01) HTMSecurityLosses 0.26+ -0.11 0.18 0.00 0.06 -0.02 0.01 -0.09* Assets (0.15) (0.19) (0.12) (0.15) (0.05) (0.06) (0.03) (0.04) UninsuredDeposits 0.05** 0.05** 0.04** 0.04** 0.00 0.00 0.01* 0.01* Assets (0.01) (0.01) (0.01) (0.01) (0.00) (0.00) (0.00) (0.00) ∆Loans 0.17* 0.19* 0.09 0.10 0.07* 0.07* 0.02 0.02 Assets (0.08) (0.08) (0.07) (0.07) (0.03) (0.03) (0.02) (0.02) LiquidAssets 0.01 0.01 0.00 0.00 0.01** 0.01** 0.00 0.00 Assets (0.01) (0.01) (0.01) (0.01) (0.00) (0.00) (0.00) (0.00) Avg. Funding Cost -0.25 -0.29 -0.11 -0.12 -0.07 -0.08 -0.08 -0.09 (0.27) (0.27) (0.21) (0.21) (0.09) (0.09) (0.06) (0.06) 2022 Dep. Growth -0.03* -0.03* -0.02* -0.02* -0.00 -0.00 -0.00 -0.00 (0.01) (0.01) (0.01) (0.01) (0.00) (0.00) (0.00) (0.00) FHLBAdvances 0.06+ 0.06+ 0.04 0.04 0.01 0.01 0.01+ 0.01+ Assets (0.03) (0.03) (0.03) (0.03) (0.01) (0.01) (0.01) (0.01) Community Bank -1.22 -1.43 -1.26+ -1.36+ 0.26 0.22 -0.23 -0.29 (0.91) (0.90) (0.72) (0.72) (0.30) (0.30) (0.21) (0.21) Regional Bank -0.31 -0.43 -0.20 -0.26 0.19 0.16 -0.29 -0.33 (0.94) (0.94) (0.75) (0.74) (0.31) (0.31) (0.22) (0.22) Constant 0.11 0.55 0.36 0.58 -0.44 -0.34 0.19 0.32 (1.06) (1.06) (0.83) (0.84) (0.34) (0.35) (0.24) (0.24) R2 0.088 0.101 0.091 0.095 0.029 0.035 0.011 0.032 a Observations 668 668 668 668 668 668 668 668 Notes: Thetableshowsresultsfromcross-sectionalregressionsofchangesinbanks’BTFPcollateralpledgingoverthe interval March 8—May 3 on contemporaneous deposit growth (∆Deposits /Assets ) uninsured deposits, HTM i,t,t(cid:48) i,t securitylosses, contemporaneousloangrowth(∆Loans i,t,t(cid:48)/Assets i,t ), liquidassetholdings(thesumofcash, federal funds, reverse repo and securities holdings), average funding cost (annualized interest expense as a percent of total liabilities), deposit growth in 2022, FHLB advances, and bank size indicators (i.e., community banks are ones with assetsunder$10billioninassetsandregionalbanksoneswithassetsbetween$10and$250billioninassets)measured as of 2022:Q4. Core deposit and contemporaneous loan growth are winsorized at the 1 and 99 percent levels. All variables except indicators, deposit growth in 2022, and average funding cost are relative to total assets. +,*,** indicate significance at 10%, 5%, and 1%, respectively. 34

Figure B1: How Banks Fund Excess Loan Growth: Increase Liabilities or Reduce Assets 60 50 40 30 20 10 0 -10 -20 )stnec( metI teehS ecnalaB ni egnahC htworG naoL ssecxE fo ralloD rep Bank HTM Security Loss Median 90th Percentile BTFP Private Large Time Reserve Security Borrowing Wholesale Deposits Reduction Sales Borrowing Notes: The chart presents estimates of how much various funding sources are used (in cents) to meet a $1 increaseinfundingneeds(asmeasuredbyexcessloangrowth). Bluedots(lines)presentpointestimates(95% confidence intervals) for how much a particular source is used overall. Red dots and lines present equivalent estimatesforabankatthe90thpercentileinsecuritieslossesasashareofassets. Theregressionsunderlying these estimates are in Table B4. 35

Table B4: Funding of Excess Loan Growth ∆BTFP Borrowings -∆Cash ∆NonFed Borrowings ∆LT. Dep. -∆Securities Assets Assets Assets Assets Assets (1) (2) (3) (4) (5) ∆(Loans-Deposits) 0.05* 0.45** 0.31** 0.06** 0.01 Assets (0.02) (0.04) (0.04) (0.02) (0.01) ×HTMSecurityLosses 0.08** -0.05 0.00 -0.00 0.00 Assets (0.03) (0.05) (0.04) (0.02) (0.02) HTMSecurityLosses -0.08 0.06 0.04 -0.00 0.05 Assets (0.12) (0.22) (0.18) (0.09) (0.07) UninsuredDeposits 0.02** -0.00 0.00 0.00 0.00 Assets (0.01) (0.01) (0.01) (0.00) (0.00) LiquidAssets 0.01 0.03** -0.04** -0.02** 0.00 Assets (0.01) (0.01) (0.01) (0.00) (0.00) Avg. Funding Cost 0.22 0.40 -0.31 -0.10 -0.13 (0.15) (0.28) (0.22) (0.11) (0.09) 2022 Dep. Growth -0.02* 0.02 0.00 0.00 -0.01* (0.01) (0.01) (0.01) (0.00) (0.00) FHLBAdvances 0.04* -0.02 -0.02 -0.01 0.02+ Assets (0.02) (0.03) (0.03) (0.01) (0.01) Community Bank 0.74 0.72 -2.04** 0.46 -0.22 (0.52) (0.93) (0.73) (0.37) (0.30) Regional Bank 0.76 -0.35 -1.00 0.29 -0.01 (0.54) (0.97) (0.76) (0.38) (0.31) Constant -1.27* -2.88** 3.12** 0.83+ 0.19 (0.61) (1.10) (0.86) (0.43) (0.36) R2 0.063 0.169 0.150 0.038 0.020 a Observations 668 668 668 668 668 Notes: Thetableshowsresultsfromregressingchangesinvariousbalancesheetitems—increasesinBTFP(Column1), reductionsincashholdings(Column2),increasesinnon-FedBorrowings(Column3),increasesinlargetimedeposits (Column 4), and reductions in security holdings (Column 5)—on banks’ excess loan growth like in specification (2). Among the explanatory variables, we also include losses on securities holdings (the difference between the value of security holdings when measured at market value and at amortized cost) as a share of assets, both in levels and interacted with excess loan growth. The remaining explanatory variables are like in Table 3. +,*,** indicate significance at 10%, 5%, and 1%, respectively. 36

Table B5: Deposit Pricing 3 Month 6 Month 12 Month 24 Month (1) (2) (3) (4) (5) (6) (7) (8) ∆Deposits -2.07* -0.55 -0.41 0.19 0.78 1.49 0.03 0.20 Assets (0.89) (0.97) (0.95) (1.04) (0.86) (0.94) (0.56) (0.61) × HTMSecurityLosses -3.09** -1.36 -1.65+ -0.39 Assets (0.85) (0.97) (0.89) (0.58) HTMSecurityLosses 7.37** 0.74 1.06 -1.90 1.69 -1.89 -2.23 -3.05 Assets (2.64) (3.19) (2.88) (3.56) (2.65) (3.28) (1.73) (2.12) UninsuredDeposits 0.02 -0.12 -0.12 -0.17 0.07 0.02 -0.04 -0.05 Assets (0.23) (0.23) (0.24) (0.24) (0.21) (0.21) (0.14) (0.14) ∆Loans -0.15 0.23 -0.77 -0.62 -0.14 0.05 0.52 0.57 Assets (1.65) (1.63) (1.74) (1.74) (1.60) (1.60) (1.03) (1.03) LiquidAssets 0.34+ 0.33+ 0.39* 0.38+ 0.43* 0.41* 0.28* 0.28* Assets (0.18) (0.18) (0.20) (0.20) (0.18) (0.18) (0.12) (0.12) Avg. Funding Cost 1.28 1.13 -1.47 -1.55 -6.13 -6.48 -4.40 -4.47 (6.25) (6.18) (6.64) (6.63) (5.73) (5.72) (3.71) (3.71) 2022 Dep. Growth -0.15 -0.12 -0.07 -0.06 -0.13 -0.12 0.04 0.05 (0.24) (0.24) (0.26) (0.26) (0.23) (0.23) (0.15) (0.15) FHLBAdvances 0.77 0.68 0.50 0.46 0.27 0.24 0.10 0.09 Assets (0.65) (0.65) (0.72) (0.72) (0.65) (0.65) (0.42) (0.42) Community Bank 71.99** 67.22** -12.06 -14.40 17.80 14.98 13.68 13.05 (20.54) (20.34) (20.85) (20.90) (19.26) (19.28) (12.29) (12.34) Regional Bank 70.42** 67.91** -17.67 -19.08 13.10 11.30 14.37 13.96 (20.93) (20.69) (21.47) (21.47) (19.80) (19.78) (12.62) (12.65) Constant -79.73** -68.20** 19.87 24.98 -15.61 -9.40 -10.05 -8.65 (23.28) (23.22) (23.86) (24.11) (21.98) (22.19) (14.12) (14.28) R2 0.045 0.068 -0.000 0.001 0.007 0.011 0.006 0.005 a Observations 507 507 581 581 593 593 576 576 Notes: Thetableshowsresultsfromcross-sectionalregressionsofbasispointchangesin3,6,12,24monthCDsoffered bybanksovertheintervalMarch8—May3oncontemporaneousdepositgrowth(∆Deposits /Assets )uninsured i,t,t(cid:48) i,t deposits, HTM security losses, contemporaneous loan growth (∆Loans i,t,t(cid:48)/Assets i,t ), liquid asset holdings (the sum of cash, federal funds, reverse repo and securities holdings), average funding cost (annualized interest expense as a percentoftotalliabilities),depositgrowthin2022,FHLBadvances,andbanksizeindicators(i.e.,communitybanks are ones with assets under $10 billion in assets and regional banks ones with assets between $10 and $ 250 billion in assets) measured as of 2022:Q4. Core deposit and contemporaneous loan growth are winsorized at the 1 and 99 percent levels. All variables except indicators, deposit growth in 2022, and average funding cost are relative to total assets. +,*,** indicate significance at 10%, 5%, and 1%, respectively. 37