The Effects of Liquidity Regulation on Bank Demand in Monetary Policy Operations

Finance and Economics Discussion Series Divisions of Research & Statistics and Monetary Affairs Federal Reserve Board, Washington, D.C. The Effects of Liquidity Regulation on Bank Demand in Monetary Policy Operations Marcelo Rezende, Mary-Frances Styczynski, and Cindy M. Vojtech 2016-090 Please cite this paper as: Rezende, Marcelo, Mary-Frances Styczynski, and Cindy M. Vojtech (2016). “The Effects of Liquidity Regulation on Bank Demand in Monetary Policy Operations,” Finance and Economics Discussion Series 2016-090. Washington: Board of Governors of the Federal Reserve System, https://doi.org/10.17016/FEDS.2016.090. 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 Effects of Liquidity Regulation on Bank Demand in Monetary Policy Operations Marcelo Rezende, Mary-Frances Styczynski, and Cindy M. Vojtech∗ October 24, 2016 Abstract We estimate the effects of the liquidity coverage ratio (LCR), a liquidity requirement for banks,onthetendersthatbankssubmitinTermDepositFacilityoperations,aFederalReserve tool created to manage the quantity of bank reserves. We identify these effects using variation in LCR requirements across banks and a change over time that allowed term deposits to count toward the LCR. Banks subject to the LCR submit tenders more often and submit larger tenders than exempt banks when term deposits qualify for the LCR. These results suggest that liquidity regulation affects bank demand in monetary policy operations. JEL Codes: E52, E58, G21, G28 Key words: Liquidity Coverage Ratio, Term Deposit Facility, Monetary Policy, Excess Reserves, Basel III We thank our conference discussants Raluca Roman, Chester Spatt, Adi Sunderam, Guillaume Vuillemey, and JulietaYung,aswellasCarlosCarvalho,JamesClouse,BurcuDuygan-Bump,ClaudioFerraz,JeffreyHuther,Ruth Judson, Beth Klee, Jonathan Parker, Pedro Souza, and seminar participants at the Federal Reserve Board, Office of Financial Research, PUC-Rio, the 2016 Fixed Income and Financial Institutions Conference, the 2016 Summer MeetingoftheEconometricSociety,the2016SummerMeetingoftheInternationalBanking,EconomicsandFinance Association,the2016YaleProgramonFinancialStabilityAnnualConference,the2016EuropeanFinanceAssociation Annual Meeting, and the 2016 MIT Golub Center for Finance and Policy Annual Conference for helpful comments. We also thank Kamran Gupta and Becky Zhang for excellent research assistance. A previous version of this paper circulated under the title “The Effects of Liquidity Regulation on Monetary Policy Implementation.” The views expressed in this paper are those of the authors and not necessarily those of the Federal Reserve Board or research staff. ∗Federal Reserve Board. Address: 20th and Constitution Ave. NW, Washington, DC 20551. E-mail: marcelo.rezende@frb.gov, mary-frances.styczynski@frb.gov, and cindy.m.vojtech@frb.gov.

1 Introduction In response to the recent financial crisis, many countries have adopted liquidity regulations. Academics and policymakers, however, have argued that liquidity regulation may affect bank behavior, thereby interfering with monetary policy (Cœur´e, 2013; Gagnon and Sack, 2014; Bech and Keister, 2015; Committee on the Global Financial System and Markets Committee, 2015; Duffie and Krishnamurthy, 2016; Potter, 2016). In particular, some monetary policy tools remove liquidity from the banking system, but liquidity regulation requires that banks hold liquid assets, which may affect their demand in monetary policy operations. This interaction could lead to less-effective policy tools and weakened financial stability. Although there is some recent theoretical work in this area, to our knowledge, there is no empirical evidence of a causal effect of liquidity regulation on bank demand in monetary policy operations. This paper attempts to fill this gap by estimating the effects of the liquidity coverage ratio (LCR), a liquidity requirement, on the participation of banks in Term Deposit Facility (TDF) operations, a tool created by the Federal Reserve to reduce the quantity of reserves in order to control interest rates.1 Because banks’ reserves in excess of their mandatory reserve requirements help them meet the LCR requirement and because term deposits awarded in TDF operations are deducted from banks’ reserve accounts for the life of the term deposits, the LCR could determine whether banks participate in TDF operations, that is, whether banks submit tenders for term deposits.2 Estimating causal effects of the LCR on TDF participation is challenging because unobservable characteristics of banks may determine both their LCR requirements and whether they decide to offer tenders. Thus, banks’ LCR requirements and their decisions to participate in TDF operations are possibly endogenous, which can bias estimates of the effects of the LCR on TDF participation. Thispotentialbiasimpliesthatweneedanempiricalstrategytobreakthisendogeneityandidentify 1TheFederalReservemayalsousetheovernightreverserepurchaseprogram(ONRRP)andtermRRPtoreduce reserves. Alternatively, reserves can be reduced by selling securities in the System Open Market Account (SOMA) portfolio. More information about the Federal Reserve’s normalization principles and plans can be found on the Board of Governors’ website www.federalreserve.gov/monetarypolicy/policy-normalization.htm. 2Excess reserves account for a large share of the liquid assets held by banks subject to the LCR. By the beginning of 2014, domestic LCR banking organizations had acquired $850 billion of excess reserves, which were about 35 percent of the liquid assets used by those banks to meet the LCR requirement. Foreign depository institutions hold approximately half of the excess reserves in the system. 1

these effects. OuridentificationstrategyreliesoncharacteristicsoftheLCRandoftheTDFoperations. Inthe United States, the LCR takes two forms: standard and modified. The standard LCR applies to all banking organizations with $250 billion or more in total consolidated assets or $10 billion or more in on-balance-sheet foreign exposures and to these banking organizations’ subsidiary depository institutions with consolidated assets of $10 billion or more. The modified LCR, a less-stringent versionoftheLCR,appliestobankholdingcompaniesandsavingsandloanholdingcompaniesthat do not meet the standard thresholds but have $50 billion or more in consolidated assets. Based on themodifiedLCRassetthreshold, verysimilarbanksmayormaynotbesubjecttotheLCR,which generates exogenous variation in LCR requirements that we can use to estimate the effects of the LCR on TDF participation. Moreover, neither the standard nor modified U.S. LCR requirements apply to U.S. branches and agencies of foreign banks, providing us with another comparison group. In addition, the characteristics of the TDF changed over time. Although early operations did not allow banks to withdraw funds prior to maturity, all operations from October 2014 to the present include an early withdrawal feature (EWF). An EWF allows term deposits to count toward the LCR, thereby increasing the incentives for banks subject to the LCR in the United States (LCR banks, for conciseness) to participate in TDF operations. This change over time helps to separate the effects of the LCR from the effects of other policies that are also triggered by the $50 billion asset threshold.3 More specifically, we compare changes in TDF participation rates of standard and modified LCR banks before and after an EWF was introduced with the same changes for banks not subject to the LCR (non-LCR banks).4 We also compare changes between LCR banks and large U.S. branches and agencies of foreign banks (large foreign banks), which are not covered by the U.S. version of the LCR. We distinguish between large and small foreign banks using a $50 billion asset threshold, which is equal to the modified LCR threshold. We estimate that the odds that an LCR bank participates in an operation increases between 3For example, banking organizations above the $50 billion threshold must also participate in the annual stress tests and get approval on their capital plans. 4The fact that the U.S. LCR rule uses a $50 billion threshold to determine which banks are subject to the standardorthemodifiedLCRsuggeststhataregressiondiscontinuityresearchdesignmaybeappropriatetoaddress the question that we study. However, because of the small number of banks with assets close to this threshold, we consider the empirical strategy described in Section 4 more appropriate. Still, a regression discontinuity research design generates results similar to those presented in this paper. 2

2.3 and 7.1 percentage points relative to a non-LCR bank per operation after an EWF is added. These odds imply that by the fifth operation with an EWF, the difference between participation rates of LCR and non-LCR banks increases between 11.5 and 35.5 percentage points. We also find evidence that an EWF increases the tender amounts of LCR banks compared to non-LCR banks. These estimates confirm the material changes in the dollar amounts awarded and the participation rates after an EWF was introduced. For example, although the June 9 and October 20, 2014, TDF operations had similar characteristics, the October operation, which offered an EWF, awarded a much larger amount and had stronger participation.5 In fact, between the two operations, the aggregate size more than doubled, from $78 billion to $172 billion, while the number of participants increased by more than half, from 40 to 66. Our finding that LCR banks participate more often and submit larger tenders with the introduction of an EWF is robust to several changes to the specifications, including narrowing the window around the LCR asset threshold and narrowing or expanding the sample used based on bank characteristics that may affect participation. This paper contributes to the nascent literature on liquidity regulation. Allen (2014) and Diamond and Kashyap (2015) survey this literature and argue that more research is needed to answer fundamental questions on liquidity regulation. In particular, Diamond and Kashyap (2015) note that “in implementing the new liquidity regulations it seems fair to say we are in a situation where practice is far ahead of both theory and measurement.” We contribute to this literature by estimating the effects of a new liquidity regulation on bank demand in monetary policy operations. Asfarasweknow,BonnerandEijffinger(2013)andBanerjeeandMio(2015)haveconductedthe only empirical studies about the impact of the LCR so far. Bonner and Eijffinger (2013) investigate the effects of non-compliance with a liquidity requirement similar to the LCR by Dutch banks on their borrowing and lending terms and volumes. Banerjee and Mio (2015) use the variation over time and across banks in the implementation of a liquidity requirement also similar to the LCR to study the impact of liquidity regulation on bank balance sheets and bank lending in the United Kingdom. The empirical strategies of these papers, however, differ significantly from ours because we exploit characteristics that are specific to the TDF and the U.S. LCR rule to identify the impact of the LCR on bank behavior. In addition, these papers study the effects of the LCR 5The two operations did not overlap with other operations, had the same caps on tender amounts ($10 billion), and offered term deposits with the same interest rate (26 basis points) and the same maturity (7 days). 3

on bank balance sheets and bank behavior in the interbank money market and in private sector lending, whereas we investigate effects on bank demand in monetary policy operations. Our paper is related to theoretical work on the impact of the LCR through banks on the effectiveness of monetary policy, which includes Bech and Keister (2015) and Duffie and Krishnamurthy (2016). Bech and Keister (2015) present a model of monetary policy implementation with term funding and a liquidity requirement similar to the LCR. Their study shows that the LCR may change bank demand for central bank reserves, thereby also changing the effects of monetary policy operations on equilibrium interest rates. We contribute to their work showing empirical evidence that confirms that the LCR affects bank demand for central bank reserves. Indeed, because term deposits awarded in TDF operations are deducted from banks’ reserve accounts at the Federal Reserve, our finding that the LCR affects bank demand for term deposits implies that the LCR affects bank demand for reserves. Duffie and Krishnamurthy (2016) build a model to examine how the LCR may attenuate the pass-through effectiveness of monetary policy by the Federal Reserve. When the LCR is binding, banks reduce their demand for assets that count the least towards the LCR compared with their demand for assets that count the most. This shift in demand distorts the spread between the rate of return on those assets, thereby reducing the pass-through effectiveness of monetary policy. We contributetotheirworkbyprovidingempiricalevidencethattheLCRaffectsthedemandforassets that do not count towards the LCR (term deposits without an EWF) relative to assets that count (term deposits with an EWF). In addition, our results show that an EWF attenuates the impact of the LCR on bank demand for term deposits, confirming the findings from Bech and Keister (2015) and Duffie and Krishnamurthy (2016) that central banks can adjust their monetary policy tools to accommodate the LCR. Our paper is also related to research that analyzes other effects of liquidity regulation. Adrian andBoyarchenko(2013)andCovasandDriscoll(2014)developdynamicgeneralequilibriummodels to study the interactions between a liquidity requirement and a capital requirement for the banking sector and the effects of those requirements on consumption and bank risk. Diamond and Kashyap (2015) investigate the effects of liquidity regulation on bank runs. Walther (2016) examines the effects of a liquidity regulation similar to the net stable funding ratio and capital requirements on systemic risk. We contribute to these papers with empirical evidence that liquidity regulation 4

affects bank behavior materially, which is an important feature of their models.6 Moreover, our paper contributes to the recent literature on monetary policy implementation in an environment where banks hold large amounts of excess reserves (Bech and Klee, 2011; Kashyap and Stein, 2012; Martin, McAndrews, Palida, and Skeie, 2013; Chen, Clouse, Ihrig, and Klee, 2014; Ennis, 2014; Armenter and Lester, 2015; Ihrig, Meade, and Weinbach, 2015; Williamson, 2015; Kandrac and Schlusche, 2016). As far as we know, the only papers in this literature with empirical analysis are Bech and Klee (2011) and Kandrac and Schlusche (2016). However, their topics are different from ours: Bech and Klee (2011) study the determinants of bargaining power in the federal funds market, and Kandrac and Schlusche (2016) analyze the effects of excess reserves on bank lending and risk taking. In contrast, we investigate the relation between liquidity regulation and participation in TDF operations. To our best knowledge, the only papers in this literature that study the TDF are Martin, McAndrews, Palida, and Skeie (2013); Chen, Clouse, Ihrig, and Klee (2014); and Ihrig, Meade, and Weinbach (2015). Martin, McAndrews, Palida, and Skeie (2013) use a theoretical model to investigate in what situations it would be optimal to use new monetary policy tools, including the TDF, to control interest rates. Chen, Clouse, Ihrig, and Klee (2014) present a theoretical model to evaluate whether these tools would allow the Federal Reserve to raise short-term interest rates in an environment with abundant excess reserves. Ihrig, Meade, and Weinbach (2015) describe how these tools may be used for monetary policy implementation. Our paper differs from these three papers because we study the TDF empirically. We contribute to this literature by showing empirical evidence that liquidity requirements affect bank demand in monetary policy operations. The rest of this paper is organized as follows: Section 2 presents some background on the TDF and the LCR, Section 3 summarizes our data, Section 4 describes our empirical strategy, Section 5 presents our results, and Section 6 concludes. 6More generally, our paper contributes to recent work on bank regulation and monetary policy implementation throughbanks. BianchiandBigio(2014)studyhowcapitalrequirements,shockstothebankingsystem,andmonetary policyalterthetradeoffbetweenprofitingfromlendingandincreasingliquidityrisk. However,theyanalyzetheeffects of capital regulation, whereas we study the effects of liquidity regulation. Still, our empirical results confirm that bank regulation affects bank demand in monetary policy operations, which is an important result of their paper. 5

2 Background on the TDF and the LCR 2.1 TDF The TDF is a tool created by the Federal Reserve to manage the aggregate quantity of reserves held by depository institutions.7 Only depository institutions eligible to receive interest from the Federal Reserve are allowed access to term deposits through the TDF.8 The funds placed in a term deposit are deducted from the institution’s reserve account for the life of the deposit, effectively draining those reserves from the banking system and putting upward pressure on interest rates. From the institutions’ perspective, term deposits are an alternative to excess reserves as an asset that can be held with the central bank, with the advantage that term deposits pay a slightly higher interest rate. Althoughoperationshavesofarbeenmostlyintendedtoensuretheoperationalreadinessandto giveinstitutionsfamiliaritywiththefunctionality,theTDFcouldbeanimportanttoolinthefuture. For instance, the Federal Open Market Committee (FOMC) indicated plans to use supplementary tools, such as the TDF, as needed to control the federal funds rate during the process of monetary policy normalization (Federal Open Market Committee, 2014).9 The Federal Reserve determines the characteristics of the operations confidentially and announces them prior to the operation dates. As part of TDF testing, the Federal Reserve has changed many characteristics of the term deposits over successive operations. Characteristics that have varied include operation format (offering a fixed or floating interest rate), maturity, interest rate, maximum deposit amount, and the time between the operation and its settlement. Table 1 shows the main changes in the characteristics of the 16 operations held between May and December of 2014, which we study in this paper.10 The offered rates and the maximum tender amounts 7The Federal Reserve started offering TDF operations in 2010. Operation details are available on the Board of Governors’ website under Monetary Policy and Policy Tools (www.federalreserve.gov/monetarypolicy/tdf.htm). 8AninstitutioncanparticipateinTDFoperationsifitiseligibletoreceiveinterestfromtheFederalReserveand ifithasalocationtosettlefundsfromTDFtransactions. InteresteligibilityisdefinedbyRegulationD.Institutions eligible to receive interest from the Federal Reserve include commercial banks, thrifts, and credit unions. To submit a tender, the institution must also have a means of accessing the TDF application. 9Federal Open Market Committee (2014) defines monetary policy normalization as the steps to raise the federal funds rate and other short-term interest rates to more normal levels and to reduce the Federal Reserve’s securities holdings. 10We focus on these 16 operations because, with the exception of their EWF status, they had very similar characteristics. Operations conducted before and after this time period differ substantially from these 16 operations. For example, operations conducted before May 2014 had much lower maximum tender amounts, and the first three 6

Table 1: TDF Operation Details EWF Operation Rate Maximum tender date (basis points) amount ($ billions) No May 19 26 3 No May 27 26 5 No June 2 26 7 No June 9 26 10 No June 16 27 10 No June 23 28 10 No June 30 29 10 No July 7 30 10 Yes October 14 26 5 Yes October 20 26 10 Yes October 27 26 15 Yes November 3 26 20 Yes November 10 27 20 Yes November 17 28 20 Yes November 24 29 20 Yes December 1 30 20 Note: Thistableshowssummarystatisticsofall16TDFoperationsconductedfromMaytoDecember 2014. All operations offered seven-day term deposits with fixed interest rates, full-allotment tenders, and a minimum tender amount of $10,000. increase between May and July, drop between July and October, and increase again between October and December. In addition, operations before July do not allow institutions to withdraw funds prior to maturity, and all operations from October to present include an EWF subject to a pecuniary penalty, which consists of the forfeiture of all interest and an annual penalty rate of 0.75 percent applied to the principal amount. However, someimportantcharacteristicsremainedunchangedthroughoutthisperiod: Allthese operations offered fixed interest rates and a minimum tender amount of $10,000. Also, all these operations followed a full-allotment policy, that is, the amount of term deposits for each bank was limitedonlybythemaximumtenderamount. Moreover,alltheseoperationsofferedseven-dayterm deposits, except for two operations, which offered six-day (November 17) and eight-day (November 24) term deposits to account for a federal holiday. operations that followed the December 2014 operation had overlapping maturities. 7

2.2 LCR The LCR is the ratio between a bank’s high-quality liquid assets (HQLA) and its projected net cash outflow over a 30-day stress scenario. As implied by the name, assets that qualify as HQLA must be easily and immediately convertible to cash with little to no loss of value. Banks subject to the LCR requirement must meet a minimum LCR ratio. Thus, the LCR promotes short-term resilience in the financial system by requiring banks to hold HQLA sufficient to cover a short-term liquidity stress scenario.11 TheLCRappliestolargeandinternationallyactivebankingorganizationsandwillbegradually implemented between 2015 and 2017. The standard LCR applies to all banking organizations with $250 billion or more in total consolidated assets or $10 billion or more in on-balance-sheet foreign exposures and to these banking organizations’ subsidiary depository institutions with consolidated assets of $10 billion or more. The modified LCR, a less-stringent version of the LCR, applies to bank holding companies and savings and loan holding companies that do not meet these thresholds but have $50 billion or more in consolidated assets. Banks subject to the standard version must have an LCR of at least 80, 90, and 100 percent by January 2015, 2016, and 2017, respectively, while banks subject to the modified version must have an LCR of at least 90 and 100 percent by January 2016 and 2017, respectively. Thus, in 2014, when the TDF operations we study were conducted, the LCR did not apply. However,asfigure1shows,LCRbanksweresteadilyincreasingtheirholdingsofHQLAthroughout 2014. This figure shows quarterly data on the HQLA-to-total assets ratio for each of the LCR groups from the first quarter of 2010 through the fourth quarter of 2015. The HQLA ratio began to increase in late 2013 at banks subject to the LCR (either standard or modified), but remained about flat over time at banks not subject to the LCR. These changes in the HQLA ratio indicate that LCR banks were already accumulating HQLA in response to the LCR requirement before those banks were actually required to meet the LCR. This evidence has an important implication for the empirical strategy of this paper because, if in 2014 banks were already accumulating HQLA to meet the LCR, then we can interpret differences in how LCR and non-LCR banks respond to the EWF as consequences of the LCR. 11Basel Committee on Banking Supervision (2008, 2013) contain the Basel III background and guidelines for the LCR. The final U.S. rules are available in Federal Register (2014). 8

Exhibit 1 September 16, 2016 Figure 1: Ratio of HQLA to Total Assets over Time HHiigghh--qquuaalliittyy lliiqquuiidd aasssseettss ((HHQQLLAA)) ttoo ttoottaall aasssseettss rraattiioo PPeerrcceennttaaggee ppooiinnttss 2200 Start of EWF 1188 SSttaannddaarrdd LLCCRR bbaannkkss MMooddiiffiieedd LLCCRR bbaannkkss NNoonn--LLCCRR bbaannkkss 1166 1144 1122 1100 88 QQ44 66 44 22 00 22001100 22001111 22001122 22001133 22001144 22001155 Note: This figure shows quarterly data on the ratio between the estimate of HQLA described in Appendix A and total assets for each of the LCR groups from the first quarter of 2010 through the fourth quarter of 2015. TheshadedareasindicatetheperiodinwhichtheTDFoperationsstudiedinthispaperoccurred,betweenthe second and the fourth quarter of 2014. The HQLA ratio began to increase in late 2013 at banks subject to the LCR (either standard or modified), but remained about flat over time at banks not subject to the LCR. ThesechangesovertimeintheHQLAratiosupportourassumptionthatLCRbankswerealreadyaccumulating HQLAinresponsetotheLCRrequirementbeforethosebankswereactuallyrequiredtomeettheLCR.Source: CallReport(FFIEC031andFFIEC041),theReportofTransactionAccounts,VaultCashandOtherDeposits (FR 2900), and balance data from internal Federal Reserve accounting records. In addition, figure 1 shows that, during the period that we analyze in this paper, HQLA ratios at LCR and non-LCR banks followed different trends, but those trends were apparently stable during this period. In particular, those trends did not change when an EWF was introduced in the TDF. These facts suggest that bank characteristics that might be related to TDF participation did not change simultaneously with the introduction of an EWF. In this case, changes in bank participation before and after the introduction of an EWF could be attributed to an effect of an EWF. 9

2.3 Relationship between the LCR and an EWF The LCR should affect how banks respond to an EWF. In TDF operations, banks exchange excess reserves, which belong to the highest HQLA category, for term deposits, which do not qualify as HQLA if early withdrawal is not possible.12 Thus, when a bank participates in a TDF operation without an EWF, the bank lowers its LCR. For this reason, LCR banks should submit tenders less often, and submit lower amounts, in TDF operations without an EWF than non-LCR banks with similar characteristics. For the same reason, an EWF should increase demand for term deposits among LCR banks in particular because term deposits with an EWF help these banks meet the LCR requirement. An EWF has two characteristics that determine the empirical strategy we follow. First, even though an EWF should be particularly valuable for LCR banks, it is possible that all banks participatemoreoftenintheTDFbecauseofanEWF.Indeed, anEWFmakestermdepositsmoreliquid, which can be valuable for any bank. Thus, our empirical strategy must account for this positive effect of an EWF on participation across all banks. In Section 4, we discuss how we address this issue. Second, banks did not need to meet LCR requirements during the sample period. Thus, one could argue that LCR banks should not have responded to an EWF differently than non-LCR banks because none of them had to meet these liquidity requirements. Still, LCR banks should already be relatively more interested in participating in the TDF than non-LCR banks in order to gain familiarity with the TDF before term deposits became LCR-eligible. Indeed, the operations we study in this paper were intended “to ensure the operational readiness of the TDF and to provide eligible institutions with an opportunity to gain familiarity with term deposit procedures.” (Board of Governors of the Federal Reserve System, 2014). Moreover, as shown in subsection 2.2, LCR banks accumulated a large amount of HQLA in 2013 and 2014, well in advance of the requirement, while non-LCR banks lowered their HQLA holdings as a share of total assets during the same period. Therefore, we assume that the LCR caused all the observed differences in how LCR and non-LCR banks responded to the EWF. Given that this assumption is fundamental to our empirical strategy, in Section 5 we investigate whether our results are robust to various changes 12Excess reserves and term deposits with an EWF are considered level 1 HQLA. Assets in this category are not limited as a share to a bank’s total HQLA and are not discounted in LCR calculations. 10

in the definition of LCR banks and non-LCR banks or, equivalently, to the set of banks assigned to control and treatment groups in our empirical exercises. 3 Data We use a panel in which each observation is a commercial bank-TDF operation pair. We include data on domestic commercial banks and U.S. branches and agencies of foreign commercial banks (foreignbanks)operatingintheUnitedStates. Werestrictthesampletocommercialbanksbecause some of the data used in this paper are not available for other depository institutions. For domestic commercial banks, we also limit the main sample to banks that do not belong to bank holding companies and to lead banks within bank holding companies, and we define a lead bank as the largest bank by total assets within a bank holding company. By limiting the sample to one bank per domestic bank holding company, we ensure that participation decisions are independent across banks. In fact, independence across banks is particularly important in this setting because the lead banks often hold the majority of the excess reserves of their holding companies. The resulting panel is composed of the 3,687 domestic and 189 foreign commercial banks that were eligible to participate in the 16 TDF operations held between May and December 2014.13 Table 2 shows summary statistics of participation rates and tender amounts in TDF operations. In this table, panels 1 to 4 separate observations depending on whether they are from domestic (panels 1 and 2) or foreign (panels 3 and 4) banks. We separate observations into foreign and domesticbanksbecausetheU.S.LCRruleappliestodomesticbanksonly. Withinthedomesticset, we divide data between LCR and non-LCR banks. A bank is considered an LCR bank if the bank or its bank holding company is subject to either the standard or modified LCR requirement. For foreigndata,wedividethesamplebetweenlargeandsmallforeignbanksusinganassetthresholdof $50billion—thesamethresholdthattheU.S.LCRruleusestodefinemodifiedLCRbanks—making the panel of large foreign banks more similar to the panel of domestic LCR banks. As shown in the four panels of table 2, participation rates and tender amounts of domestic LCR 13To be eligible to participate in the TDF a bank must also submit a formal application to the Federal Reserve. Thus, our sample includes banks with and without formal access to the TDF. We use the sample of banks eligible to participate in the TDF instead of the narrower sample of banks with formal access to the TDF because applying foraccesstotheTDFismostlikelyadecisionendogenoustounobservablebankcharacteristics. Still,ourresultsare about the same if we restrict our sample to banks with formal access to the TDF. 11

Table 2: TDF Summary Statistics by Bank Type and Operation Type Variable Obs. Banks Mean Std. Dev. Min. Max. 1. Domestic banks, non-LCR a. NoEWF Submittedatender(percentagepoints) 29,232 3,654 0.36 Tenderamount($millions) 29,232 3,654 0 24 0 1,400 b. WithEWF Submittedatender(percentagepoints) 29,232 3,654 0.55 Tenderamount($millions) 29,232 3,654 1 39 0 2,000 2. Domestic banks, LCR a. NoEWF Submittedatender(percentagepoints) 264 33 22.73 Tenderamount($millions) 264 33 841 2,205 0 10,000 b. WithEWF Submittedatender(percentagepoints) 264 33 39.02 Tenderamount($millions) 264 33 2,971 5,811 0 20,000 3. Foreign banks, small (<$50 billion in assets) a. NoEWF Submittedatender(percentagepoints) 1,361 171 4.04 Tenderamount($millions) 1,361 171 100 723 0 10,000 b. WithEWF Submittedatender(percentagepoints) 1,360 170 7.94 Tenderamount($millions) 1,360 170 270 1,622 0 20,000 4. Foreign banks, large (≥$50 billion in assets) a. NoEWF Submittedatender(percentagepoints) 151 18 32.45 Tenderamount($millions) 151 18 1,779 3,205 0 10,000 b. WithEWF Submittedatender(percentagepoints) 152 19 55.26 Tenderamount($millions) 152 19 4,825 6,862 0 20,000 Note: This table shows summary statistics of participation and tender amounts from the 16 TDF operations conducted betweenMayandDecember2014. Eachobservationisabank-operationpair. Thedataarecomposedofobservationsfrom the3,687domesticand189foreigncommercialbankseligibletoparticipateintheseoperations. banks and large foreign banks are higher than participation rates and tender amounts of domestic non-LCR banks and small foreign banks, respectively. Also, participation and tender amounts increase across all four groups with an EWF. For instance, as shown in panel 2, participation rates are 1.7 times larger (39.02/22.73 = 1.7) and tender amounts are 3.5 times larger (2,971/841 = 3.5) in operations with an EWF compared to operations without an EWF for LCR banks. For the most part, these are the largest proportional differences between operations with and without an EWF across all combinations of banks by ownership and size shown in table 2. Thus, these summary statistics offer some support to the hypothesis that an EWF has a stronger impact on LCR banks. However, other differences between domestic and foreign banks, LCR status, and EWF availability most likely also determine TDF participation. For this reason, in Section 4 we present an empirical 12

strategy intended to account for these differences and adequately identify the effects of the LCR on TDF participation. We add data on bank characteristics to the bank-operation panel. Bank-specific data include the dollar amounts of the banks’ total excess reserves in the most recent week before the respective operation.14 For domestic banks, the data also include the total assets, return on assets, return on equity, total capital ratio, leverage ratio, net interest margin, total loan delinquency ratio, and total net charge-offs ratio from the most recent quarter before each respective operation, which are obtained from quarterly reports of condition and income (Call Reports).15 For foreign banks, the only characteristics included in the data are total assets and excess reserves. We also build an estimate of HQLA for domestic banks using Call Report data. Ideally, we wouldliketoincludeinourdatatheLCRofeachbank. Unfortunately, CallReportsdonotinclude the data necessary to calculate a bank’s projected net cash outflow over a 30-day stress scenario, which is the denominator of the LCR. In addition, banks do not directly report in these forms their HQLA, which is the numerator of the LCR. Thus, we build an estimate of HQLA for each bank, which we describe in Appendix A, and we use the ratio of this estimate to total assets, which we show in figure 1, in our regressions.16 Table 3 summarizes these data. Of note, LCR and non-LCR banks differ substantially in characteristics that should determine how banks participate in TDF operations. For example, domestic LCR banks have ratios of excess reserves to total assets and of HQLA to total assets that are more than twice as large as domestic non-LCR banks, and these differences are statistically significant. However, figures 2 and 3 also show that bank characteristics of LCR and non-LCR banks did not change materially during the sample period. These two figures show the mean of the variables includedintable3(exceptfortheratioofHQLAtototalassets,whichweshowinfigure1)fromthe first quarter of 2010 to the fourth quarter of 2015. The shaded areas indicate the period in which 14A bank’s excess reserves is for the most part equal to its average end-of-day account balances due from Federal ReserveBankslessitsreservebalancerequirement(RBR).BalancedataarefrominternalFederalReserveaccounting records whereas bank-level RBR is calculated based on confidential filings of the FR 2900 Report of Transaction Accounts, Vault Cash and Other Deposits. 15Call Reports are mandatory forms filed quarterly by commercial banks (Consolidated Report of Condition and Income, FFIEC 031 and FFIEC 041). 16HQLAisestimatedprimarilyusingexcessreservesandsecurityassetswitha0percentora20percentriskweight (Call Report Schedule RC-R). 13

Table 3: Summary Statistics of Bank Characteristics Domestic banks Foreign banks Assets Assets Non-LCR LCR less than at least $50 billion $50 billion Total assets ($ millions) 783 310,561∗∗ 5,147 89,032∗∗ (2,591) (484,280) (8,690) (27,420) Return on assets 0.93 1.10∗ (1.68) (0.97) Return on equity 8.00 9.00 (24.63) (5.98) Net interest margin 3.73 2.84∗∗ (1.07) (1.43) Total capital ratio 18.38 15.59∗∗ (13.60) (5.58) Leverage ratio 10.68 10.54 (4.29) (2.92) Loan delinquency ratio 2.59 2.10∗∗ (2.87) (1.32) Net charge-off ratio 0.20 0.42∗∗ (0.93) (0.59) Excess reserves/total assets 3.53 9.28∗∗ 43.15 47.37 (6.13) (12.27) (83.10) (30.61) HQLA/total assets 5.07 10.42∗∗ (7.32) (8.06) Observations 58,464 528 2,721 303 Banks 3,654 33 170 19 Note: Theunitofobservationisabank-operationpair. Thesampleincludesdomesticcommercialbanksthatdonotbelong to a bank holding company, domestic commercial banks that are the lead bank within a bank holding company, and all foreign commercial banks operating in the United States in 2014. Excess reserves are based on two-week averages of data prior to each TDF operation, covering a period from early May 2014 to late November 2014. All other data are based on quarter-enddatafromCallReportsforMarch31,June30,andSeptember30of2014. HQLAdataareestimatedfromthese forms because they are not directly reported. See Appendix A for more details. All variables are measured in percentage points, except when stated otherwise. * and ** indicate that a two-sided t-test rejects the hypothesis that the mean in the column is the same as the mean in the column on the left at the 5 and 1 percent levels, respectively. T-tests are done separatelyfordomesticandforeignbanks. Standarddeviationsareinparentheses. the TDF operations studied in this paper occurred, between the second and the fourth quarter of 2014. Because this period is shorter than one year, bank characteristics changed little during this time interval. Moreover, trends in characteristics of LCR and non-LCR apparently did not change when an EWF was introduced. The measures of availability of reserves and of profitability in figure 2 remained mostly stable over the sample period. The ratio of excess reserves to total assets (top left panel) remained about 14

Exhibit 1 September 16, 2016 Figure 2: Bank Characteristics over Time EExxcceessss rreesseerrvveess ttoo ttoottaall aasssseettss rraattiioo RReettuurrnn oonn aasssseettss PPeerrcceennttaaggee ppooiinnttss PPeerrcceennttaaggee ppooiinnttss 22 Start of EWF 1144 Start of EWF LLCCRR bbaannkkss LLCCRR bbaannkkss 1122 NNoonn--LLCCRR bbaannkkss NNoonn--LLCCRR bbaannkkss 1100 88 11 QQ44 66 44 QQ44 22 00 00 22001100 22001111 22001122 22001133 22001144 22001155 22001100 22001111 22001122 22001133 22001144 22001155 RReettuurrnn oonn eeqquuiittyy NNeett iinntteerreesstt mmaarrggiinn PPeerrcceennttaaggee ppooiinnttss PPeerrcceennttaaggee ppooiinnttss 1144 66 Start of EWF Start of EWF LLCCRR bbaannkkss 1122 55 NNoonn--LLCCRR bbaannkkss 1100 44 QQ44 QQ44 88 33 66 22 44 LLCCRR bbaannkkss NNoonn--LLCCRR bbaannkkss 22 11 00 00 22001100 22001111 22001122 22001133 22001144 22001155 22001100 22001111 22001122 22001133 22001144 22001155 Note: Thisfigureshowsquarterlydataontheratiobetweenexcessreservesandtotalassets,returnonassets, returnon equity, andnetinterestmargin forLCRand non-LCRdomestic banksfromthe firstquarter of2010 throughthefourthquarterof2015. TheshadedareasindicatetheperiodinwhichtheTDFoperationsstudied in this paper occurred, between the second and the fourth quarter of 2014. Source: Call Report (FFIEC 031 andFFIEC041),theReportofTransactionAccounts,VaultCashandOtherDeposits(FR2900),andbalance data from internal Federal Reserve accounting records. flat for non-LCR banks, but increased for LCR banks. However, the increase in this ratio for LCR banks is about the same in the third and in the fourth quarter of 2014, that is, immediately before and after the introduction of an EWF. Return on assets (top right panel) and return on equity (bottom left panel) were similar and stable for LCR and non-LCR banks from the second to the fourth quarter of 2014. Net interest margins (bottom right panel) of LCR banks changed little throughout the period covered in this figure. The measures of capitalization and delinquency in figure 3 also remained stable between the 15

Exhibit 2 September 16, 2016 Figure 3: Bank Characteristics over Time (continued) TToottaall ccaappiittaall rraattiioo LLeevveerraaggee rraattiioo PPeerrcceennttaaggee ppooiinnttss PPeerrcceennttaaggee ppooiinnttss 2200 1122 QQ44 1188 QQ44 1111 1100 1166 99 1144 Start of EWF 88 1122 77 Start of EWF 1100 66 L N L N CC oo R n R n-- LL bb CC aann RR kk bb ss aannkkss 88 LLCCRR bbaannkkss 4 5 4 5 66 NNoonn--LLCCRR bbaannkkss 33 44 22 22 11 00 00 22001100 22001111 22001122 22001133 22001144 22001155 22001100 22001111 22001122 22001133 22001144 22001155 DDeelliinnqquueennccyy rraattee NNeett cchhaarrggee--ooffff rraattee PPeerrcceennttaaggee ppooiinnttss PPeerrcceennttaaggee ppooiinnttss 88 33 Start of EWF Start of EWF 77 LLCCRR bbaannkkss 66 NNoonn--LLCCRR bbaannkkss 55 L N L N CC oo R n R n-- LL bb CC aann RR kk bb ss aannkkss 22 44 33 11 QQ44 22 11 QQ44 00 00 22001100 22001111 22001122 22001133 22001144 22001155 22001100 22001111 22001122 22001133 22001144 22001155 Note: This figure shows quarterly data on total capital ratio ratio, leverage ratio, delinquency rate, and net charge-offrateforLCRandnon-LCRdomesticbanksfromthefirstquarterof2010throughthefourthquarter of 2015. The shaded areas indicate the period in which the TDF operations studied in this paper occurred, between the second and the fourth quarter of 2014. Source: Call Report (FFIEC 031 and FFIEC 041). second and the fourth quarter of 2014. Total capital ratios (top left panel) and leverage ratios (top right panel) were about unchanged for both LCR and non-LCR banks during the period covered in this figure. Delinquency rates (bottom left panel) and net charge-off rates were similar and stable at low levels for LCR and non-LCR banks during 2014. In summary, bank characteristics of LCR and non-LCR banks are quite similar before and after the introduction of an EWF during the sample and thus most likely did not cause any changes in participation of LCR and non-LCR banks during that period. 16

4 Empirical Strategy In this section, we describe the empirical strategy used to investigate whether TDF participation depends on whether a bank is subject to the LCR and whether the operation has an EWF. We mainlyuseanindicatorofwhetherabanksubmittedatenderinaTDFoperationasthedependent variable to avoid biases in our estimates caused by the censoring of tender amounts. Indeed, tender amounts are censored from above and below because all operations have maximum tender amounts and a minimum tender amount of $10,000. In addition, caps on tender amounts vary across operations, with operations with an EWF having higher caps, on average. In fact, as shown in table 2, some banks submitted tender amounts equal to the maximum allowed. Still, we show that regressions using tender amounts and a participation indicator as the dependent variable yield similar results: tender amounts and participation rates increase with the introduction of an EWF. Figure 4 motivates the empirical strategy. This figure shows bank participation rates over the 16 TDF operations held between May and December 2014. The panels on the left show domestic banks, and the panels on the right show foreign banks. The top panels include data from all banks, while the middle and bottom panels restrict the sample based on bank asset size, narrowing the sample to banks within an interval of assets around the $50 billion threshold. In each panel, a vertical line indicates when an EWF was introduced. Note that as this interval of assets narrows for domestic banks (moving from the top left to the bottomleftpanel),participationratesofLCRbanksandnon-LCRbanksgetcloserinthepre-EWF period, the operations to the left of the vertical line. In fact, participation rates of domestic LCR and non-LCR banks between $25 billion and $100 billion are very similar and are about flat in operations without an EWF. Thus, these panels suggest that non-LCR domestic banks become a better comparison group to estimate the effects of an EWF on LCR domestic banks as we narrow the interval of bank assets. Also of note, participation rates of LCR banks trend upward, while rates of non-LCR banks remain flat after an EWF is introduced. Thus, LCR banks apparently respond differently to an EWFinadirectionconsistentwiththehypothesisthatanEWFincreasesthevalueoftermdeposits forLCRbanksmorethanitdoesfornon-LCRbanks. TheEWFmayhavecausedagradualincrease in participation of LCR banks compared to non-LCR banks instead of an immediate jump, for 17

Figure 4: Participation of Banks in TDF Operations All Domestic Banks All Foreign Banks Percentage points Percentage points 80 80 LCR banks At least $50 billion 60 60 Non-LCR banks Less than $50 billion 40 40 Start of EWF 20 Start of EWF 20 0 0 05/19 06/02 06/16 06/30 10/14 10/27 11/10 11/24 05/19 06/02 06/16 06/30 10/14 10/27 11/10 11/24 05/27 06/09 06/23 07/07 10/20 11/03 11/17 12/01 05/27 06/09 06/23 07/07 10/20 11/03 11/17 12/01 Operation Date (MM/DD) Operation Date (MM/DD) Domestic Banks with Assets between Foreign Banks with Assets between $5 billion and $500 billion Percentage points $5 billion and $500 billion Percentage points 80 80 LCR banks At least $50 billion 60 60 Non-LCR banks Less than $50 billion Start of EWF 40 40 20 20 Start of EWF 0 0 05/19 06/02 06/16 06/30 10/14 10/27 11/10 11/24 05/19 06/02 06/16 06/30 10/14 10/27 11/10 11/24 05/27 06/09 06/23 07/07 10/20 11/03 11/17 12/01 05/27 06/09 06/23 07/07 10/20 11/03 11/17 12/01 Operation Date (MM/DD) Operation Date (MM/DD) Domestic Banks with Assets between Foreign Banks with Assets between $25 billion and $100 billion Percentage points $25 billion and $100 billion Percentage points 80 80 LCR banks At least $50 billion 60 60 Non-LCR banks Less than $50 billion Start of EWF 40 40 20 Start of EWF 20 0 0 05/19 06/02 06/16 06/30 10/14 10/27 11/10 11/24 05/19 06/02 06/16 06/30 10/14 10/27 11/10 11/24 05/27 06/09 06/23 07/07 10/20 11/03 11/17 12/01 05/27 06/09 06/23 07/07 10/20 11/03 11/17 12/01 Operation Date (MM/DD) Operation Date (MM/DD) Note: This figure shows bank participation rates over the 16 TDF operations held between May and December2014. Thepanelsontheleftshowdomesticbanks,andthepanelsontherightshowforeignbanks. Thetoppanelsincludedatafromallbanks,whilethemiddleandbottompanelsrestrictthedatalbasedon bankassetsize. Ineachpanel,averticallineindicateswhenanEWFwasintroduced. Also,domesticbanks are divided between those affiliated with banking organizations subject to the LCR (standard or modified) andthoseaffiliatedwithorganizationsthatareexemptfromtheLCR.Participationratesforforeignbanks are divided between those with less than $50 billion and those with at least $50 billion in total assets. 18

example, because banks must apply for access to the TDF and the Federal Reserve must approve their access before banks can participate in the TDF. This application process may have caused a delay for some banks between the date in which they decided to participate in the TDF and the operation in which they actually participated for the first time. In addition, as shown in table 1, the later operations had higher spreads and larger maximum tender amounts, making them more attractive. These differences in participation trends motivate our main empirical strategy in which we estimate whether an EWF caused an increase in the participation trend of domestic LCR banks using domestic non-LCR banks as a control group. Based on this strategy, we interpret estimates of a positive change in that trend as evidence of the LCR’s effect on TDF participation. The panels on the right of figure 4 show participation rates of foreign banks. Participation rates of all foreign banks and foreign banks with assets between $5 billion and $500 billion, shown in the top right and the middle right panels, are very similar to rates of domestic banks, shown in the top left and the middle left panels. Participation rates of foreign banks increase over time and this trend is stronger for larger banks. However, as we narrow the sample to banks with assets between $25 billion and $100 billion, participation rates of foreign banks, shown in the bottom right panel, differ substantially from the rates of domestic banks, shown in the bottom left panel. Rates of foreign banks above and below the $50 billion threshold are, on average, very similar with and without an EWF, while rates of domestic banks subject to and exempt from LCR are also similar without an EWF but have very different trends once an EWF is introduced. These facts are consistent with the hypothesis that an EWF does not increase the value of term deposits for large foreign banks more than it does for small foreign banks because none of them are subject to the U.S. LCR rule. However, several of these foreign banks are subject to an LCR requirement of their home country. In fact, several of the foreign banks that participated in TDF operations publicly disclose a ratio. Given that these disclosed ratios are generally well above 100 percent, the requirement was likely less binding, which explains the growing participation of foreign banks across all operations. Of note, foreign banks held about half of the excess reserves in the system at this time, and if LCR rules are not binding, the slightly higher rate of return on TDF operations over excess reserves was likely enough to motivate participation. Overall, these facts motivate an alternative empirical strategy in which we estimate whether an EWF caused an increase in the participation trend of domestic LCR banks using foreign banks with at least $50 19

billion as a control group. 5 Results 5.1 Participation of Domestic LCR Banks and Domestic Non-LCR Banks We begin exploring the effects of an EWF by testing participation rates of domestic banks. We use an empirical strategy adapted from a differences-in-differences strategy. A differences-in-differences strategy would typically investigate whether an EWF causes an immediate change in TDF participation, but we investigate whether an EWF causes a gradual change in participation as motivated by figure 4. Because of this evidence, we believe that measuring a change in the trend of participation over time is more suitable than measuring an immediate change in participation. We estimate the following equation: Y = αLCRBank +βLCRBank ×t+γLCRBank ×t×EWF +ν +ϕ +ε , (1) ij i i i j i j ij where Y is a dummy variable equal to 1 if bank i submitted a tender in operation j and equal ij to 0 otherwise. LCRBank is a dummy variable equal to one if bank i’s holding company has the i characteristics that will make it subject to either the standard or modified LCR requirement and zero otherwise.17 EWF is the indicator of whether operation j has an EWF, and t is an operation j trend normalized to one in the first operation with an EWF.18 ν is a bank random effect, ϕ is an i j operation fixed effect, and ε is a bank- and operation-specific unobservable error. ν and ε have ij i ij independent normal distributions with a mean of zero and variance of σ2 and σ2. Note that this ν ε specificationisanalogoustoadifferences-in-differencesmodelusing∆Y asthedependentvariable. ij The interaction term γLCRBank ×t×EWF measures the change in the trend of participation i j ratesofLCRbanksrelativetonon-LCRbanksaftertheintroductionofanEWF.Thus, γ estimates the difference between the effects of an EWF on the participation of LCR and non-LCR banks. In Appendix B, we use a simple model to show that, under additional assumptions, the parameter estimated by the differences-in-differences between participation of LCR and non-LCR banks 17The list of banks that met the criteria to be subject to the LCR did not change during the sample period. As a result,theLCRstatusisfixedatthebanklevelthroughoutthesampleperiodandthevariableLCRBank doesnot i need a subscript j. 18The EWF was introduced in the ninth operation in our sample and thus t∈{−7,−6,...,7,8}. 20

Table 4: TDF Participation of Domestic Banks (1) (2) (3) All Banks with Banks with banks assets between assets between $5 billion & $25 billion & $500 billion $100 billion LCRBank 0.186** 0.174* 0.123 (0.071) (0.076) (0.082) LCRBank × t 0.008 -0.001 -0.021 (0.006) (0.007) (0.019) LCRBank × t × EWF 0.023* 0.024* 0.071** (0.010) (0.012) (0.027) Observations 58,992 1,994 344 Banks 3,687 126 24 R-squared 0.13 0.12 0.15 Note: Thistableshowsestimatesofequation(1). Thedependentvariableisadummyvariableequalto1ifbank isubmitsatenderinoperationj andequalto0otherwise. LCRBanki isadummyvariableequaltooneifthe bank’s holding company i is subject to either the standard or modified LCR requirement and zero otherwise. EWFj istheindicatorofwhetheroperationjhasanEWF,andtisanoperationtrendnormalizedtooneinthe firstoperationwithanEWF.Allequationsincludeoperationfixedeffects. Standarderrorsareclusteredatthe banklevelandareshowninparentheses. *and**denotesignificantatthe5and1percentlevels,respectively. in operations with and without an EWF can be interpreted as the shadow cost of the LCR. The shadow cost of the LCR could be an important measure for monetary policy implementation. For example,DuffieandKrishnamurthy(2016)showthatthiscostcaninhibitthecreationofshort-term bank liabilities, thereby reducing the pass-through effectiveness of monetary policy. Table 4 shows the regression results using equation (1). Columns 1 to 3 use the same subsamples from the top left, middle left, and bottom left panels in figure 4, respectively. Column 1 shows the results using data from all banks. The estimate of γ in this column is positive and statistically significant, with a value of 0.023. This result suggests that the probability that an LCR bank participated in a TDF operation relative to a non-LCR bank increased from 2.3 percent to 18.4 percent over the course of the eight EWF operations (2.3×8 = 18.4). Also, the coefficient on the LCR dummy is positive, which is expected based on the summary statistics that show that LCR banks generally have higher participation rates (table 2 and figure 4). The estimate of γ in column 1 of table 4 suggests that an EWF causes an increase in the participation of LCR banks relative to non-LCR banks, but it may also be driven by differences 21

in characteristics of LCR and non-LCR banks that are unrelated to the LCR. Indeed, LCR banks are much larger and complex than non-LCR banks on average, which may explain why LCR banks respond differently to an EWF. Thus, to compare observations from banks with more similar characteristics, columns 2 and 3 narrow the sample to observations from banks with assets close to $50 billion. As we move from column 1, all banks, to column 3, banks with assets between $25 billion and $100 billion, the estimate of γ increases and remains statistically significant, despite the sharp drop in the number of observations. The fact that γ increases as we narrow our sample supports the hypothesis that the stronger effect of an EWF on LCR banks is caused by the LCR, not by differences between unobservable characteristics of LCR and non-LCR banks. 5.2 Tender Amounts of Domestic LCR Banks and Domestic Non-LCR Banks We next test whether bank tender amounts were affected by an EWF. To do this, we estimate a bidding model as follows: B = C , if C < Y ij j j ij = Y , if R ≤ Y ≤ C (2) ij j ij j = 0, if Y < R , ij j whereB isthetenderamountsubmittedbybankiinoperationj,andR andC aretheminimum ij j j and the maximum tender amounts of this operation. Y is still determined by equation (1), but Y ij ij is now the latent value of bank i’s tender amount in operation j. All these variables enter equation (2) as the natural logarithm of the respective variable measured in thousands of U.S. dollars. Table5presentstheseresultsusingthesubsamplesofbankswithassetsabove$1billion,between $5billionand$500billion,andbetween$25billionand$100billionincolumns1to3,respectively.19 The estimates of γ increase as we narrow the interval of assets, moving from column 1 to 3, and becomestatisticallysignificantwhenwerestrictthesampletobankswithassetsbetween$25billion and $100 billion in column 3. This result supports our hypothesis that LCR banks submit larger 19Column1oftable5restrictsthesampletoobservationsfrombankswithassetsabove$1billionbecauseofdifficultiesinthecomputationofthetobitmodeldescribedbyequation(2). Ideally,thiscolumnshoulduseobservations from all banks, as column 1 of table 4 does. However, possibly because of the large fraction of banks with assets below $1 billion that do not participate in TDF operations, estimates of the tobit model with this broader sample could not be computed. 22

Table 5: TDF Tender Amounts of Domestic Banks (1) (2) (3) Banks with Banks with Banks with assets above assets between assets between $1 billion $5 billion & $25 billion & $500 billion $100 billion LCRBank 27.445** 28.490** 50.033* (3.796) (5.422) (20.512) LCRBank × t 0.012 -0.870 -5.782* (0.356) (0.444) (2.596) LCRBank × t × EWF 0.804 1.126 8.205** (0.549) (0.675) (3.138) Observations 7,690 1,994 344 Banks 492 126 24 Log-likelihood -1,163 -764 -164 Note: Thistableshowsestimatesofthemodelcomposedofequations(1)and(2). Thedependentvariableisthe naturallogarithmofbanki’stenderamountinoperationj measuredinthousandsofU.S.dollars. LCRBanki isadummyvariableequaltooneifthebank’sholdingcompanyiissubjecttoeitherthestandardormodified LCRrequirementandzerootherwise. EWFj istheindicatorofwhetheroperationj hasanEWF,andtisan operation trend normalized to one in the first operation with an EWF. All equations include operation fixed effects. Standarderrorsareclusteredatthebanklevelandareshowninparentheses. *and**denotesignificance atthe5and1percentlevels,respectively. bids than non-LCR banks in operations with an EWF. To gauge the impact on tender amounts implied by the estimate of γ in column 3, we perform a simple exercise as follows. Assume that, in the absence of the EWF, an LCR bank would submit a tenderequaltotheminimumamount, equalto$10,000. IfanEWFwereincludedinthatoperation forthefirsttime, thesamebankwouldsubmitatenderequalto$37million. Althoughthisincrease of more than 1,000 times in the tender amount may appear to be too large, we note that, as shown in table 2, the average positive tender amount by LCR banks in operations with an EWF is about $3 billion. Theresultsusingdollaramountsfromtendersasthedependentvariablecomplementtheresults usingaparticipationindicator,givingfurthersupporttothehypothesisthatanEWFhasastronger impact on the demand of TDF deposits by LCR banks relative to non-LCR banks. However, in our view the econometric evidence using a participation indicator is preferable to the evidence using tender amounts in the setting that we study for two main reasons. First, the linear probability model that we use with the participation indicator allows us to interpret directly the effect implied 23

by the coefficient estimates, which is not possible with the nonlinear model that we use with tender amounts. Second, estimates of the nonlinear model that we use, a tobit, are typically less robust in settings such as the one that we study, in which most observations are left-censored (banks do not submit tenders) and in which we must estimate the coefficients of independent binary variables (suchasγ). Forthesereasons, weadoptthespecificationsusingtheparticipationindicator, instead of the tender amounts, as our benchmark. Although the results using dollar amounts from tenders as the dependent variable corroborate the results using a participation indicator, we recognize that all these results may still be driven by unobservable characteristics of LCR and non-LCR banks that determine how they respond to an EWF. In fact, even when we restrict our sample to banks within the interval from $25 billion to $100 billion, LCR banks remain much larger than non-LCR banks, on average. Narrowing this interval much further is not possible because the number of banks in it, 24, is already quite small. For this reason, we next compare the participation of banks of more similar sizes—and, possibly, similar unobservable characteristics—that are subject to and not subject to the U.S. LCR. In the next subsection, we compare participation rates of large domestic banks and large foreign banks. 5.3 Participation of Large Domestic and Large Foreign Banks We next show evidence of the effects of an EWF on participation rates using the domestic LCR banksasatreatmentgroupandforeignbankswithatleast$50billioninassets, whicharenon-LCR banks, as a control group. For clarity, the LCRBank notation is replaced with DOM, making the specification Y = αDOM +βDOM ×t+γDOM ×t×EWF +ν +ϕ +ε , (3) ij i i i j i j ij where Y , EWF , ν , ϕ , and ε are defined as before, and DOM is a dummy variable equal to ij j i j ij i one for domestic banks and equal to zero for foreign banks. Once again, γ measures the difference in the impact of an EWF on the trend of TDF participation between LCR and non-LCR banks. Table 6 shows the results. Columns 1 to 3 use observations from banks with at least $50 billion. The coefficient of interest is the interaction term DOM ×t×EWF . The estimated coefficient i j increases and becomes statistically significant as we narrow the sample from banks with at least 24

Table 6: TDF Participation of Large Domestic and Large Foreign Banks LargeBanks SmallBanks (1) (2) (3) (4) (5) (6) Bankswith Bankswith Bankswith Bankswith Bankswith Bankswith assetsabove assetsbetween assetsbetween assetsbelow assetsbetween assetsbetween $50billion $50billion& $50billion& $50billion $5billion& $25billion& $500billion $100billion $50billion $50billion DOM -0.045 -0.054 -0.138 -0.015 -0.044 -0.192 (0.129) (0.130) (0.146) (0.013) (0.043) (0.166) DOM×t -0.023 -0.028* -0.038* -0.006* -0.008 -0.032 (0.014) (0.014) (0.019) (0.002) (0.006) (0.032) DOM×t×EWF 0.031 0.034 0.082** 0.002 0.005 0.041 (0.019) (0.020) (0.030) (0.003) (0.009) (0.039) Observations 799 735 373 61,217 2,391 314 Banks 51 47 26 3,827 155 23 R-squared 0.08 0.09 0.17 0.02 0.04 0.24 Note: This table shows estimates of equation (3). The dependent variable is a dummy variable equal to 1 if bank i submittedatenderinoperationjandequalto0otherwise. DOMiisadummyvariableequaltoonefordomesticbanks andequaltozeroforforeignbanks. EWFj istheindicatorofwhetheroperationj hasanEWF,andtisanoperation trend normalized to one in the first operation with an EWF. All equations include operation fixed effects. Standard errorsareclusteredatthebanklevelandareshowninparentheses. *and**denotesignificantatthe5and1percent levels,respectively. $50 billion in assets in column 1 to banks between $50 billion and $100 billion in column 3. This increase in γ is consistent with the graphical evidence from figure 4 and the hypothesis that an EWF increases participation of LCR banks more than it does for non-LCR banks. Columns 4 to 6 provide additional support to this hypothesis based on a falsification test. In these columns, we restrict the sample to domestic and foreign banks with less than $50 billion in assets, which are all non-LCR banks. Thus, if the hypothesis that an EWF affects participation of LCR and non-LCR banks differently because of the LCR is correct, γ should not be significant in anyofthesecolumns. Indeed,γ isneversignificant,evenwhenwenarrowthesampletoobservations from banks between $25 billion and $50 billion in total assets in column 6. Restricting the panel to domestic and foreign banks above or below the $50 billion threshold cannot fully eliminate the possibility that our results are driven by unobservable characteristics of domestic and foreign banks that determine how they respond to an EWF. For example, foreign banks typically hold more assets abroad than domestic banks and these assets, which we cannot observe, may cause differences in the willingness of foreign and domestic banks to participate in 25

the TDF. This limitation is similar to the one we discussed in the comparison between larger and smaller domestic banks in the previous subsection. Thus, in the next subsections we provide additional evidence that the stronger impact of an EWF on LCR banks is caused by the LCR and not by differences in unobservable characteristics. 5.4 Robustness Results Inthissubsection,wetesttherobustnessofourbaselineresultsshowninsubsection5.1. Werestrict our attention to domestic banks because some of the data necessary to perform the robustness tests that we conduct in this subsection are not available for foreign banks.20 We make changes to the panel of banks and add control variables. We also perform falsification tests to search for any changes in participation trends of LCR banks relative to non-LCR banks before an EWF was introduced. Overall, the baseline result holds: LCR banks increase their participation in the TDF after an EWF is added. Table 7 shows robustness results for our main specification, presented in column 3 of table 4. By using column 3 of table 4 as the baseline for all robustness tests, we make all results comparable and keep this subsection concise.21 Thus, all columns of table 7 show estimates of equation (1) using data from banks with assets between $25 billion and $100 billion. Column 1 of table 7 restricts the data from column 3 of table 4 to observations from banks with minimum excess reserves of at least $10,000 in at least one of the operations. This restriction limits the sample to observations from banks with sufficient excess reserves to participate in at least one operation because, for all operations, $10,000 was the minimum bid.22 Only one bank is eliminated from the sample and γ remains statistically significant and equal to 0.073, almost the same as the baseline result from table 4. In addition to having sufficient excess funds to meet the minimum bid, a bank needs a location to settle funds from TDF transactions in order to participate. Any bank that is eligible to receive interest from the Federal Reserve can set up such a location. Thus, column 2 of table 7 restricts the 20Still,therobustnesstestsincolumns1and2intable7canbereplicatedwiththesamplesofdomesticandforeign banks used in subsection 5.3. These tests show that the results from that subsection are robust too. These tests are available upon request from the authors. 21Still, robustness tests of other columns in table 4 show that those results are robust too. 22Given the amount of reserves in the banking system, it is relatively easy for banks to acquire excess reserves as desired. Therefore, including all banks in initial testing is not unreasonable. 26

Table 7: TDF Participation with Additional Bank Characteristics and Observations (1) (2) (3) (4) Banks with Banks with Banks with Banks with assets between assets between assets between assets between $25 billion & $25 billion & $25 billion & $25 billion & $100 billion, $100 billion, $100 billion $100 billion minium excess min. exc. res. reserves & TDF access LCRBank 0.136 0.247 0.143 0.058 (0.094) (0.184) (0.084) (0.098) LCRBank × t -0.024 -0.052 -0.019 -0.020 (0.021) (0.043) (0.018) (0.021) LCRBank × t × EWF 0.073∗∗ 0.108∗ 0.060∗ 0.070∗∗ (0.028) (0.043) (0.024) (0.026) Bank characteristics? No No No Yes All banks in organization? No No Yes No Observations 328 215 392 344 Banks 23 15 28 24 R-squared 0.14 0.18 0.14 0.27 Note: Thistableshowsrobustnessresultsfortheestimatesincolumn3oftable4. Allcolumnsoftable7showestimates ofequation(1)usingapanelofdomesticbankswithassetsbetween$25billionand$100billion. Thedependentvariable is a dummy variable equal to 1 if bank i submitted a tender in operation j and equal to 0 otherwise. LCRBanki is a dummy variable equal to one if the bank’s holding company i is subject to either the standard or modified LCR requirement and zero otherwise. EWFj is the indicator of whether operation j has an EWF, and t is an operation trendnormalizedtooneinthefirstoperationwithanEWF.Column1restrictsthepaneltoobservationsfrombanks with minimum excess reserves of at least $10,000 in at least one of the operations. Column 2 restricts the sample in column 1 to banks with an active location to settle TDF funds at the time of the TDF operation. In column 3, we presentresultsafterrelaxingtheleadbankpanelrestriction,asdefinedearlierasthelargestbankaffiliatedwithabank holdingcompanybytotalassets. Column4addsbankcharacteristicsascontrolvariables: returnonassets, returnon equity, total capital ratio, leverage ratio, net interest margin, total loan delinquency ratio, total net charge-offs ratio, excess reserves-to-assets, and HQLA-to-assets. Excess reserves are measured as averages over the two weeks prior to eachTDFoperation. Allequationsincludeoperationfixedeffects. Standarderrorsareclusteredatthebankleveland areshowninparentheses. *and**denotesignificanceatthe5and1percentlevels,respectively. sample in column 1 to banks with an active location to settle TDF funds at the time of the TDF operation. The number of banks in the sample drops by one-third, and γ rises to 0.108. According to this estimate, LCR banks increased their participation rate by 10.8 percentage points relative to non-LCR banks in the first TDF operation with an EWF. In column 3, we present results relaxing the lead bank restriction. In Section 3, we defined the lead bank as the largest bank affiliated with a bank holding company and we restricted the sample to banks that do not belong to a bank holding company and to the lead banks within bank holding 27

Table 8: Falsification Tests of TDF Participation Changing the EWF Introduction Date (1) (2) (3) Banks with Banks with Banks with assets between assets between assets between $25 billion & $25 billion & $25 billion & $100 billion, $100 billion, $100 billion, true date of one operation two operations EWF introduction before before LCRBank 0.163 0.095 0.110 (0.134) (0.084) (0.083) LCRBank × t -0.026 -0.016 -0.021 (0.025) (0.020) (0.023) LCRBank × t × EWF 0.071* 0.043 0.038 (0.035) (0.032) (0.036) Observations 258 257 256 Banks 24 24 24 R-squared 0.08 0.06 0.05 Note: Thistableshowsrobustnessresultsfortheestimatesincolumn3oftable4. Allcolumnsoftable8show estimatesofequation(1)usingapanelofdomesticbankswithassetsbetween$25billionand$100billion. The dependent variable is a dummy variable equal to 1 if bank i submitted a tender in operation j and equal to 0 otherwise. LCRBankiisadummyvariableequaltooneifthebank’sholdingcompanyiissubjecttoeitherthe standardormodifiedLCRrequirementandzerootherwise. EWFj istheindicatorofwhetheroperationj has anEWF,andtisanoperationtrendnormalizedtooneinthefirstoperationwithanEWF.Incolumn1,EWFj isconstructedusingthetruedatewhenanEWFwasintroducedinTDFoperations,namelyOctober14,2014, the ninth operation in the sample. Columns 2 and 3 present falsification tests in which EWFj is constructed assuminganEWFwasintroducedintheeighthandintheseventhoperationinthesample,respectively. Columns 1to3useobservationsfromthethirdtothefourteenth,thesecondtothethirteenth,andthefirsttothetwelfth operation,respectively,toensurethattheregressionsinallcolumnsuseobservationsfromsixoperationsbefore and six operations after the (true or false) date of introduction of an EWF. All equations include operation fixed effects. Standard errors are clustered at the bank level and are shown in parentheses. * and ** denote significanceatthe5and1percentlevels,respectively. companies. By relaxing this restriction, the number of banks included in the regression increases from 24 to 28. The coefficient estimate of 0.06 is significant and close to the original estimate in table 4. Column 4 of table 7 also uses the same sample of column 3 of table 4, but adds bank characteristics as control variables. Once again, the results remain about unchanged. On balance, these results show that the results from subsection 5.1 are robust. We now perform falsification tests to search for changes in participation trends of LCR banks relative to non-LCR banks before an EWF was introduced. To perform these tests, we restrict the 28

data to the six operations immediately before the introduction of an EWF and the six operations immediately after, including the operation in which an EWF was introduced. This restriction ensures that the number of operations before and after the (true or false) date of introduction of an EWF is the same and that this number remains the same as we move that date from the true date to an earlier date. We then replicate the results from column 3 of table 4 using the true date of introduction and false earlier dates. Table 8 presents the results. In column 1, EWF is constructed using the true date when j an EWF was introduced in TDF operations, namely October 14, 2014, the ninth operation in the sample. Columns 2 and 3 present falsification tests in which EWF is constructed assuming j an EWF was introduced in the eighth and in the seventh operation in the sample, respectively. Columns 1 to 3 use observations from the third to the fourteenth, the second to the thirteenth, and the first to the twelfth operation, respectively, following the data restriction described above. The estimate of γ in column 1 is equal to the estimate in column 3 of table 4, 0.071, and also statistically significant, despite the data restriction imposed in table 8. When we compare the estimate of γ in column 1 of table 8 with the estimates in columns 2 and 3, we observe that the estimates decrease and are not significant as we move the date of introduction of an EWF from the true date in column 1 to one (0.043 in column 2) and two (0.038 in column 3) operations before. These falsification test support the assumption that changes in unobservable characteristics of banks before the introduction of an EWF do not drive our estimates of γ. 6 Conclusion This paper shows that the LCR, a liquidity requirement for banks, affects their participation in the TDF, a Federal Reserve facility created to manage the aggregate quantity of reserves held by depository institutions. Our analysis suggests that banks subject to the LCR participate relatively more often than banks that are exempt from the LCR in TDF operations that allow banks to withdraw funds prior to maturity. LCR banks also increase the size of their bids in operations with an EWF compared to non-LCR banks. Thus, our results imply that liquidity regulation may affect bank demand in monetary policy operations. More broadly, these results suggest that liquidity regulation may affect the transmission of 29

monetary policy through banks. Indeed, given that funds placed in term deposits are deducted from banks’ excess reserves, the evidence that the LCR affects bank demand for term deposits indicates that the LCR also affects the demand for excess reserves. Consequently, this liquidity requirement should create a difference in the demand for excess reserves between banks subject to theLCRandotherbanks,whichmaydeterminehowmonetarypolicyistransmittedtotheeconomy. In fact, Kandrac and Schlusche (2016) provide evidence that the transmission of monetary policy through bank lending varies across banks depending on the amount of excess reserves that each bank holds.23 Together with their evidence, our work suggests that the LCR may affect the impact of monetary policy through banks by changing their demand for excess reserves. Appendix A High-Quality Liquid Assets (HQLA) Inthisappendix,wedescribehowweestimateHQLA.HQLAareestimatedprimarilyusingquarterend balance sheet data reported in two regulatory filings, the Call Report (FFIEC 031 and FFIEC 041) and the Report of Transaction Accounts, Vault Cash and Other Deposits (FR 2900). Given thelackofgranularityintheCallReport,theestimatestendtobeconservative(low). Inparticular, the estimate does not include reverse repurchase agreements (also known as resale agreements). The LCR requirement classifies HQLA into three types: level 1, level 2A, and level 2B. Level 1 assets consist of excess reserve balances, Treasury securities, and Ginnie Mae securities. Excess reserve balances are equal to reserve balances minus required reserves. Reserve balance data are from internal Federal Reserve accounting records, and the bank-level reserve balance requirement is calculated based on confidential filings of the FR 2900.24 Treasury securities and Ginnie Mae securities are estimated using the risk-weighted assets schedule in the Call Report (RC-R Part II). Held-to-maturity securities, available-for-sale securities, and trading assets with a risk weight of zero percent are assumed to qualify as level 1 assets (RCFDB604 + RCFDB609 + RCFDB628). Level 2A assets primarily consist of debt and mortgage-backed securities backed by Fannie Mae and Freddie Mac. These assets are estimated using securities with a risk weight of 20 percent (RC- 23Inaddition,theliteraturehasshownsubstantialevidencethatdifferencesinotherbankcharacteristicsdetermine how monetary policy affects bank credit supply (Kashyap and Stein, 2000; Jim´enez, Ongena, Peydro´, and Saurina, 2014;ScharfsteinandSunderam,2015;Dell’Ariccia,Laeven,andSuarez,2016;Drechsler,Savov,andSchnabl,2016). 24Excess reserves can be proxied using public data in the Call Report RC-A item 4 “Balances due from Federal Reserve Banks” (RCFD0090). 30

R Part II, RCFDB605 + RCFDB610).25 Level 2B assets consist of corporate debt and equity. Given the ownership restrictions of banks, these holdings are small and are estimated using the securities schedule of the Call Report (RC-B). Specifically, we use item 6.a. “Other domestic debt” and include both held-to-maturity and available-for-sale (RCFD1738 + RCFD1741) securities. According to the final rule, haircuts are applied to level 2 assets: the level 2A haircut is 15 percent, and the level 2B haircut is 50 percent. Thus, we estimate level 1, 2A, and 2B liquid asset amounts (LAA) as follows: Level 1 LAA = Excess reserves+RCFDB604+RCFDB609+RCFDB628, (A.1) Level 2A LAA = 0.85×(RCFDB605+RCFDB610), (A.2) Level 2B LAA = 0.50×(RCFD1738+RCFD1741). (A.3) The LCR rule also imposes caps on level 2 LAA as a share of HQLA: Level 2B LAA assets cannot be more than 15 percent of a bank’s total HQLA, and total Level 2 (2A + 2B) LAA cannot be more than 40 percent of total HQLA. We compute each bank’s HQLA calculating the sum of their respective level 1, 2A, and 2B LAA based on equations (A.1) to (A.3) and subject to these caps.26 Appendix B A Model of TDF Participation In this appendix, we present a simple model of participation in the TDF to illustrate how, under certain assumptions, variation between LCR and non-LCR banks and between operations with and without an EWF identify the shadow cost of the LCR.27 We assume for simplicity that there are only two TDF operations, 0 and 1, such that only operation 1 has an EWF, that is, EWF = 0 and 0 EWF = 1. Because there are only two periods, TDF participation by LCR and non-LCR banks 1 do not have different time trends. Instead, an EWF causes a change in the level of participation of LCR versus non-LCR banks, which is measured by γ. This assumption does not impose any loss 25Note that risk-weighted asset definitions changed with Basel III. Beginning with the March 31, 2015, form, the items are level 1 = RCFDD962 + RCFDD967 + RCFDD977 and level 2A = RCFDD963 + RCFDD968. 26See, for example, http://www.occ.treas.gov/topics/capital-markets/balance-sheet-management/liquidity/Basel- III-LCR-Formulas.pdf for formulas that can be used to calculate this sum. 27This model is based on a conference discussion of our paper by Adi Sunderam. 31

of generality and is intended only to keep the model simple. An extension of this model to include multiple periods, different trends in participation by LCR and non-LCR banks, and to allow the EWFtoaffectthetrendsinparticipationofLCRandnon-LCRbanksdifferentlyisstraightforward, but would make the model more complicated and would not change the results materially given the purpose of this appendix. We also impose assumptions that are not imposed in the empirical model used in the paper. These assumptions allow us to give an economic interpretation to the parameter identified by the differences-in-differences model. We assume that a bank decides whether or not to participate in the TDF comparing the benefits and costs of investing in the TDF as opposed to leaving the respective funds as excess reserves with the Federal Reserve. In particular, we assume that the probability that bank i participates in the TDF in operation j is given by: Y = C +(RateTDF −IOER )−LiquidityCost ij i j j j −ShadowCostLCR×LCRBank ×(1−EWF )+ε , (B.1) i j ij where C is a bank fixed effect, RateTDF is the TDF interest rate, IOER is the interest rate on i j j excessreserves,andRateTDF > IOER > 0. LiquidityCost istheliquiditycostoftermdeposits j j j such that LiquidityCost > LiquidityCost > 0, where the first inequality holds because an EWF 0 1 increases the liquidity of term deposits and thus lowers this cost, and the second inequality holds because even with an EWF, term deposits still include some liquidity cost. Liquidity costs in an operation with an EWF arise because funds can only be withdrawn prior to maturity subject to a penalty. Of note, we assume that LiquidityCost does not depend on whether or not the bank is j subject to the LCR. ShadowCostLCR is the shadow cost for LCR banks of using excess reserves to invest in term deposits given that excess reserves count towards the LCR, while term deposits without an EWF do not. LCRBank is, once again, an indicator of whether the bank is subject i to the LCR. EWF and ε are the same as in subsection 5.1. j ij Based on equation (B.1), the population differences-in-differences is the shadow cost of the 32

LCR: (cid:32) (cid:33) (cid:104) (cid:105) (cid:104) (cid:105) E Y |LCRBank = 1,EWF = 1 −E Y |LCRBank = 0,EWF = 1 − ij i j ij i j (cid:32) (cid:33) (cid:104) (cid:105) (cid:104) (cid:105) E Y |LCRBank = 1,EWF = 0 −E Y |LCRBank = 0,EWF = 0 ij i j ij i j = ShadowCostLCR. (B.2) This model can be extended to include multiple periods and a gradual effect of the EWF on participation in the TDF. In this case, γ could also be interpreted as the shadow cost of the LCR. References Adrian, T., and N. Boyarchenko (2013): “Liquidity Policies and Systemic Risk,” Federal Reserve Bank of New York Staff Report No. 661. Allen, F. (2014): “How Should Bank Liquidity be Regulated?,” Working Paper. Armenter, R., and B. Lester (2015): “Excess Reserves and Monetary Policy Normalization,” Federal Reserve Bank of Philadelphia Working Paper No. 15-35. Banerjee, R. N., and H. Mio (2015): “The Impact of Liquidity Regulation on Banks,” Bank of England Staff Working Paper No. 536. Basel Committee on Banking Supervision (2008): Principles for Sound Liquidity Risk Management and Supervision. Bank for International Settlements Committee Publications. (2013): Basel III: The Liquidity Coverage Ratio and Liquidity Risk Monitoring Tools. Bank for International Settlements Committee Publications. Bech, M. L., and T. Keister(2015): “LiquidityRegulationandtheImplementationofMonetary Policy,” Working Paper. Bech, M. L., and E. Klee (2011): “The Mechanics of a Graceful Exit: Interest on Reserves and Segmentation in the Federal Funds Market,” Journal of Monetary Economics, 58(5), 415–431. 33

Bianchi, J., and S. Bigio(2014): “Banks,LiquidityManagementandMonetaryPolicy,”Working Paper, https://www0.gsb.columbia.edu/faculty/sbigio/papers/Paper_06-30-14.pdf. Board of Governors of the Federal Reserve System (2014): “Press Release, February 21, 2014,” http://www.federalreserve.gov/newsevents/press/monetary/20140221a.htm. Bonner, C., and S. Eijffinger (2013): “The Impact of Liquidity Regulation on Bank Intermediation,” CEPR Discussion Paper No. 9124. Chen, H., J. Clouse, J. Ihrig, and E. Klee (2014): “The Federal Reserve’s Tools for Policy Normalization in a Preferred Habitat Model of Financial Markets,” forthcoming, Journal of Money, Credit and Banking. Cœure´, B. (2013): “Liquidity Regulation and Monetary Policy Implementation: from Theory to Practice,” Speech at the Toulouse School of Economics, Toulouse, October 3. Committee on the Global Financial System and Markets Committee (2015): “Regulatory Change and Monetary Policy,” CGFS Papers No. 54, Bank for International Settlements. Covas, F., and J. C. Driscoll (2014): “Bank Liquidity and Capital Regulation in General Equilibrium,” Finance and Economics Discussion Series Paper No. 2014-85, Federal Reserve Board. Dell’Ariccia, G., L. Laeven, and G. Suarez (2016): “Bank Leverage and Monetary Policy’s Risk-Taking Channel: Evidence from the United States,” forthcoming, Journal of Finance. Diamond, D. W., and A. K. Kashyap (2015): “Liquidity Requirements, Liquidity Choice and Financial Stability,” Working Paper. Drechsler, I., A. Savov, and P. Schnabl (2016): “The Deposit Channel of Monetary Policy,” Working Paper. Duffie, D., and A. Krishnamurthy (2016): “Passthrough Efficiency in the Feds New Monetary Policy Setting,” Working Paper Presented at the 2016 Jackson Hole Symposium of the Federal Reserve Bank of Kansas City. 34

Ennis, H. M. (2014): “A Simple General Equilibrium Model of Large Excess Reserves,” Federal Reserve Bank of Richmond Working Paper No. 14-14. Federal Open Market Committee (2014): “Policy Normalization Principles and Plans,” September 17. Federal Register (2014): “Liquidity Coverage Ratio: Liquidity Risk Measurement Standards, Final Rule,” 79(197), 61439–61541. Gagnon, J. E., and B. Sack (2014): “Monetary Policy with Abundant Liquidity: A New Operating Framework for the Federal Reserve,” Peterson Institute for International Economics Policy Briefs PB14-4. Ihrig, J. E., E. E. Meade, and G. C. Weinbach (2015): “Rewriting Monetary Policy 101: What’s the Fed’s Post-Crisis Approach to Raising Interest Rates,” Journal of Economic Perspectives, 29(4), 177–198. Jime´nez, G., S. Ongena, J.-L. Peydro´, and J. Saurina (2014): “Hazardous Times for Monetary Policy: What do 23 Million Loans Say About the Impact of Monetary Policy on Credit Risk-Taking?,” Econometrica, 82(2), 463–505. Kandrac, J., and B. Schlusche (2016): “Quantitative Easing and Bank Risk Taking: Evidence from Lending,” Working Paper. Kashyap, A. K., and J. C. Stein (2000): “What Do a Million Observations on Banks Say about the Transmission of Monetary Policy?,” American Economic Review, 90(3), 407–428. (2012): “The Optimal Conduct of Monetary Policy with Interest on Reserves,” American Economic Journal: Macroeconomics, 4(1), 266–282. Martin, A., J. McAndrews, A. Palida, and D. Skeie (2013): “Federal Reserve Tools for Managing Rates and Reserves,” Federal Reserve Bank of New York Staff Reports, No. 642. Potter, S. (2016): “Implementing Monetary Policy Post-Crisis: What Have We Learned? What Do We Need to Know?,” Remarks at Columbia University SIPA and Federal Reserve Bank of New York Workshop, New York City, New York, May 4. 35

Scharfstein, D., and A. Sunderam (2015): “Market Power in Mortgage Lending and the Transmission of Monetary Policy,” Working Paper. Walther, A. (2016): “Jointly Optimal Regulation of Bank Capital and Liquidity,” Journal of Money, Credit, and Banking, 48(2-3), 415–448. Williamson, S. D. (2015): “Interest on Reserves, Interbank Lending, and Monetary Policy,” Federal Reserve Bank of St. Louis Working Paper No. 2015-024A. 36