Effects of Changing Monetary and Regulatory Policy on Overnight Money Markets

Finance and Economics Discussion Series Divisions of Research & Statistics and Monetary Affairs Federal Reserve Board, Washington, D.C. Effects of Changing Monetary and Regulatory Policy on Overnight Money Markets Elizabeth Klee, Zeynep Senyuz, and Emre Yoldas 2016-084 Please cite this paper as: Klee, Elizabeth, Zeynep Senyuz, and Emre Yoldas (2016). “Effects of Changing Monetary and Regulatory Policy on Overnight Money Markets,” Finance and Economics Discussion Series 2016-084. Washington: Board of Governors of the Federal Reserve System, https://doi.org/10.17016/FEDS.2016.084. 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.

Effects of Changing Monetary and Regulatory Policy on Overnight Money Markets∗ Elizabeth Klee† Zeynep Senyuz‡ Emre Yoldas§ September 2016 Abstract Money markets have been operating under a new monetary policy implementation framework since the Federal Reserve started paying interest on bank reserves in late 2008. The regulatory environment has also evolved substantially over this period. We develop and test hypotheses regarding the effects of changes in the monetary and regulatory policy on dynamicsofkeyovernightfundingmarkets.Wefindthatthefederalfundsratecontinuedtoprovide an anchor, albeit weaker, for unsecured funding rates amid substantial decline in activity and changingcompositionoftrades,whileitstransmissiontotherepomarkethadbeenhampered. The overnight reverse repurchase (ON RRP) operations that started in late 2013 contributed tostrongerco-movementamongovernightfundingratesandmarkedlyreducedtheirvolatility. The change in the FDIC assessment fees and Basel III leverage ratio regulations have exacerbated financial-reporting-day effects in unsecured markets. In contrast, consistent with lower dealer leverage in the post-crisis period, such effects have weakened in the repo market, especiallyaftertheinceptionoftheONRRPfacility. Finally,superabundantbankreservesappear tohavesignificantlydiminishedtheeffectsofreserve-maintenanceonthemoneymarketrates. Keywords: Overnightmoneymarkets,federalfunds,repo,Eurodollar,commercialpaper,VAR models, GARCH models. JEL Classification: C32, E43, E52 ∗We would like to thank James Clouse, Jane Ihrig, participants of the 2016 International Finance and Banking Society Barcelona Conference, and 2016 Western Economic Association International Conference for helpful comments. WethankRichardSambasivamforhisassistancewiththedataset. Theviewsexpressedinthispaper are solely the responsibility of the authors and should not be interpreted as reflecting the views of the Board of Governors of the Federal Reserve System, or other members of its staff. †Federal Reserve Board, E-mail: elizabeth.c.klee@frb.gov ‡Federal Reserve Board, E-mail: zeynep.senyuz@frb.gov §Federal Reserve Board, E-mail: emre.yoldas@frb.gov

1 Introduction The response of the Federal Reserve (Fed) to the global financial crisis significantly altered the backdrop against which monetary policy is implemented in the United States. With successive rate cuts that began in mid-2007, the federal funds rate target was reduced from 5.25 percent in August 2007 to its effective lower bound (ELB) of 0 to 0.25 percent in December 2008. The federal funds rate, as well as other overnight money market rates, remained at the ELB for the nextsevenyears. Throughoutthecrisisanditsaftermath, theFedusedavarietyofnewfacilities to provide liquidity to the financial system as well as unconventional tools, such as large-scale asset purchases (LSAPs), to stimulate the economy. As a result of these efforts, reserves in the banking system have reached unprecedented levels. Marking a significant shift in its monetary policy implementation framework, the Fed started paying interest on bank reserves (IOR) in late 2008 to achieve monetary control in an environment of superabundant reserves. The elevated reserves and the new monetary policy tools changed trading dynamics in the federal funds market. Some institutions, like government-sponsored enterprises (GSEs) that are not eligible to earn IOR, became the primary lenders, while large foreign banks borrowed funds below IOR for arbitrage purposes. Mainly because of this fragmented structure of the federal funds market, the IOR could not set a lower bound on the effective federal funds rate (EFFR). Againstthisbackdrop, theFedintroducedasupplementarytool, theovernightreverserepurchase agreement (ON RRP) facility in September 2013 to set a soft floor on interest rates and enhance monetary control. The changing regulatory environment also created new incentives for money market participants, and a substantial decline in the leverage of securities dealers contributed to the new landscape of these markets. Among the new regulations, the change in the assessment base for the Federal Deposit Insurance Corporation (FDIC) deposit insurance and the Basel III leverage ratio requirement are of particular importance. The former made wholesale funding more costly for U.S. chartered banks relative to that of U.S. branches and agencies of foreign banks, creating an incentive for domestic banks to reduce their borrowing in the money market. The latter resulted in an incentive for foreign banks to dynamically deleverage through money market activity due to differences in implementation of the leverage ratio across major jurisdictions. Meanwhile, 1

both leverage levels and net repo liabilities of the broker-dealer sector decreased notably, creating an important contrast to the pre-crisis period during which major institutions in this sector operated outside the regulated banking system subject to lighter regulations. As dealers dynamically manage their balance sheets mainly through repos, these developments suggest reduced scope for financial-reporting-day dynamics in the repo market on net. Superabundantreserves, the newmonetarypolicyimplementationframework, andthechangingregulatoryenvironment,whichcreatednewincentivesformarketparticipants,haveimportant implicationsfordynamicsoftheEFFRandotherkeyovernightmoneymarketinterestrates. One of the most important questions in this context is how and to what extent the pass-through from the EFFR to other money market interest rates has been affected over time. Moreover, such changes have potentially important effects on the level and volatility of money market rates, particularly on financial-reporting days. In this paper, we shed light on the effects of the new monetary policy tools and financial regulations on the dynamics of overnight money market rates that are relevant for monetary policy implementation. We first provide a detailed summary of these changes in the monetary policyandregulatoryenvironment, andlayoutspecifictestablehypothesesregardingtheireffects on overnight money markets. We then empirically analyze these hypotheses using daily data on key money market rates from 2001 to 2015. In particular, we estimate systems of dynamic modelsforovernightfundingratesforthepre-crisisandtheELBperiods,wheretheformersample serves as a benchmark. Our models incorporate the long-run relationship of the EFFR with the other overnight rates during the pre-crisis period and allow for potentially different dynamics around financial reporting dates. We also explicitly model time-variation in the volatilities and correlations of rates in a multivariate GARCH framework. We also focus on the evolution of money markets through the ELB period and analyze dynamics of the two subsamples split by the beginning of the ON RRP operations in September 2013. This date provides a natural structural breakpoint, as it corresponds to an expansion of the Fed’s monetary policy toolkit. Moreover, the announcement and implementation of the new leverage ratio requirements roughly correspond to the post-ON RRP period. We find that the EFFR continued to provide an anchor for unsecured overnight rates during the ELB period, but co-movement among these rates weakened somewhat. The transmission to 2

the repo rates from the EFFR has been hampered significantly in the ELB period. In addition, newregulationsappeartohavesubstantiallyalteredthedynamicsofunsecuredratesonfinancialreporting dates. Specifically, rates that represent unsecured wholesale funding costs for banks became markedly lower and more volatile on quarter-ends. Contrary to the case of unsecured rates, the quarter-end effects have weakened in the repo market on net, reflecting lower dealer leverage and reduced net repo financing. Another notable change in the ELB period relative to the pre-crisis era has been the disappearance of the day-of-maintenance-period effects on the EFFR, mainly due to the abundance of reserve balances. The money markets went through notable changes within the ELB period as well. We find that the ON RRP operations have contributed positively to the overall co-movement of rates as well as the transmission from the federal funds market to other unsecured funding markets. Moreover, volatility of all rates dampened with an especially notable decline in the repo market. The latter ELB sample also corresponds to the period during which the new leverage ratio regulations were announced and implemented. We find that the tendency of foreign banks to reduce their overnight borrowing on financial-reporting-related dates, combined with the search by cash lenders for alternative investment opportunities, exacerbated month-end and quarter-end effects on the EFFR and Eurodollar rates. The availability of the ON RRP as a viable investment on financial reporting dates, when other investment options may be limited, reduces the potential for sharp drops in the repo rate, as empirically verified in our analysis. In related literature on money market dynamics, Afonso et al. (2011) analyze activity in the federal funds market during the global financial crisis, while Copeland et al. (2014b) and Gorton and Metrick (2012) focus on the repo market in the context of runs. Yoldas and Senyuz (2015) model the behavior of term money market rates and quantify stress thresholds. Although the literature on monetary policy transmission to the economy is vast, there is relatively limited researchonhowthetargetrateistransmittedtootherovernightinterestrates. Bechetal.(2014) examine the link between the federal funds and repo markets and find evidence of deterioration of the pass-through from the EFFR to the repo rate. Kroeger and Sarkar (2016) suggest that this pass-through improved with the introduction of the ON RRP operations. Another strand of literature that is related to our work documents the effects of certain calendar days on money market rates. Spindt and Hoffmeister (1988), Griffiths and Winters 3

(1995), Hamilton (1996), Carpenter and Demiralp (2006), and Judson and Klee (2010) show that the EFFR exhibits calendar-day effects associated with the maintenance period as well as quarter-ends. More recently, Munyan (2015) documents the effects of window dressing activity on financial reporting dates in the repo market. The rest of the paper is organized as follows. We summarize major changes in the federal funds market, monetary policy implementation framework, and the new regulations as well as their likely effects in Section 2. Section 3 describes the data set while Section 4 lays out the methodological framework. Estimation results are presented and discussed in Section 5. Section 6 concludes. 2 Changes in Monetary and Regulatory Policy and Implications 2.1 Bank Reserves and Activity in the Federal Funds Market Banks are required to maintain a minimum level of reserves at the Federal Reserve Banks in their Districts.1 Historically, banks avoided holding excess reserves, as such balances did not earn any interest. Indeed, total reserve balances in the banking system averaged about $10 billion in 2007, whiletotalbankassetswerecloseto$10trillionoverthesameperiod. AscanbeseeninFigure1, reserves in the system increased to more than $800 billion at the end of 2008 as the Fed provided ampleliquidityduringthefinancialcrisisthroughseveralfacilities.2 Followingsubsequentrounds of LSAPs from 2009 to 2014, total reserve balances averaged nearly $2.5 trillion in the first half of 2016.3 Theunprecedentedincreaseinthereservebalanceschangedthelandscapeforthefederalfunds market.4 Historically, transactions in the federal funds market facilitated the redistribution of 1Wewillbereferringtodepositoryinstitutionswithreserveaccountssimplyasbanks. SeeRegulationDReserve Requirementsforafulllistofdepositoryinstitutions,availableathttps://www.federalreserve.gov/boarddocs/ supmanual/cch/int_depos.pdf. 2Seehttps://www.federalreserve.gov/monetarypolicy/bst_crisisresponse.htmfordetailsontheFed’scrisis response, and https://www.federalreserve.gov/monetarypolicy/expiredtools.htm for a list of expired liquidity provision facilities. 3Between November 2008 and October 2014, the Fed purchased nearly $1.7 trillion in Treasury securities and about $2 trillion in agency mortgage-backed securities, as well as $170 billion in agency debt securities in order to put downward pressure on longer-term interest rates. See d’Amico et al. (2012) and Krishnamurthy and Vissing- Jorgensen (2011) for a discussion of the economic rationale and effects of LSAPs. 4Federal funds are unsecured loans of reserve balances between banks and other eligible institutions, mainly GSEs. Federalfundstransactionsaretypicallyconductedforanovernighttermandarecarriedouteitherdirectly between the institutions or through third-party brokers. 4

reserve balances, whereby banks with reserve balances in excess of the required levels lent to banks in need of reserves. The surge in reserve balances led to a substantial decline in banks’ need for short-term borrowing to cover idiosyncratic funding shortfalls. To ensure monetary control and promote efficiency in the banking system, the Fed introduced the IOR as a new monetary policy tool in October 2008. As a result, incentives for banks to lend federal funds at rates below the IOR were largely eliminated. Against this backdrop, the outstanding amount of federal funds borrowed by banks declined to roughly one fourth of the level observed prior to the global financial crisis by 2011, and it has remained low since then (Figure 1). Moreover, volume in the federal funds market declined from $200 billion per day in 2007 to $60 billion per day at the end of 2012 according to Afonso et al. (2013b): They estimate that banks that provided more than half of the federal funds sold before the crisis account for only a fraction of the lending activity after 2008. GSEs that are not eligible toearnIORhavebeenthemainlendersinthepost-crisisperiod.5 Ontheborrowingside, Afonso et al. (2013a) show that mostly banks under the umbrella of bank holding companies (BHCs) and foreign banking organizations (FBOs) have been purchasing federal funds from GSEs for arbitrage purposes.6 These institutions borrow federal funds at rates below the IOR and place the cash in their reserve accounts to earn the spread between the IOR and the EFFR. These transactions have been relatively more profitable for FBOs as they are not subject to assessment by the FDIC.7 These changes in the federal funds market raise the important issue of whether the passthrough from the FFR to other overnight rates has been affected over time. In addition, superabundant reserves may have implications for effects of cash flows on the EFFR associated with days of reserve maintenance period.8 In the pre-crisis era, activity in the federal funds market was,inpart,drivenbymaintenanceperioddynamics,asshownbySpindtandHoffmeister(1988), 5Specifically, it is the Federal Home Loan Bank (FHLB) System that dominated the supply side of the federal funds market. See Ashcraft et al. (2010) for a detailed description of the FHLB system and its role as a liquidity provider to banks. 6In the current context, the FBOs are U.S. branches and agencies of foreign banking institutions. 7In 2011, the FDIC changed the assessment base for its deposit insurance scheme from domestic deposits to totalassetsminusequity,makinglargerbalancesmorecostlyfordomesticbanksregardlessoffundingsource. See Kreicher et al. (2013) for a detailed discussion. 8See the Reserve Maintenance Manual for reporting requirements as well as calculation and maintenance of reserve balances, available at http://www.federalreserve.gov/monetarypolicy/files/ reserve-maintenance-manual.pdf. 5

Griffiths and Winters (1995), and Hamilton (1996). Ennis and Wolman (2015) find that reserves in the system have been fairly widely distributed across banks since mid-2009. This finding along with the extremely elevated level of reserves in the system, may suggest that calendar effects associated with reserve maintenance significantly diminished in the post-crisis era. 2.2 Monetary Policy Implementation Framework Historically, adjustment of the level of reserve balances in the banking system to move the EFFR towardthetargetlevelsetbytheFOMCwasthecentralpillarofmonetarypolicyimplementation. The main policy tool was the open market operations (OMOs) to manage the amount of reserve balancesavailabletothebankingsystem.These operationswouldinfluencetherateinthefederal funds market, where banks experiencing shortfalls could borrow from banks with excess reserves. Given the small volume of reserves at the Fed, around $10 billion, even small OMOs could significantly affect the EFFR. Changes in the federal funds rate would then be transmitted to other short-term interest rates, to longer-term interest rates, and eventually to inflation and economic activity. This framework worked seamlessly while the Fed was operating with a balance sheet of less than $1 trillion before the crisis. The global financial crisis forced changes in the operational framework of the Fed.9 In an environment with superabundant reserves, the conventional approach based on changing the quantity of reserves via OMOs would not work. As a result, the Fed extended its monetary policy toolkit. In the fall of 2008, the Fed started paying interest on banks’ reserve balances, which became the primary tool of its new monetary policy implementation framework in controlling short-term interest rates. While adjusting the IOR is an effective way to move market interest rates in an environment of superabundant reserves, federal funds have generally traded below this rate, mainly due to the fact that only banks can earn the IOR. GSEs still have an incentive to lend at rates below the IOR, as they do not receive interest on their reserve accounts. Moreover, FDIC fees and balance sheet constraints limit arbitrage activity by banks that would push the EFFR toward the IOR. In order to enhance monetary control and put an effective floor under short-term interest rates, the 9SeeIhrigetal.(2015)foranindepthdiscussionoftheevolutionoftheFed’smonetarypolicyimplementation framework through the financial crisis and its aftermath. 6

Fed introduced the ON RRP facility as a supplementary tool for its implementation of monetary policy.10 ON RRPs are offered to a broad set of financial institutions, including money mutual funds (MMFs) that do not have access to the federal funds market. In September 2014, the FOMC issued a statement summarizing the new operating framework, and in December 2015, it successfully lifted the EFFR from its ELB in this framework.11 The primary tool of the new operating framework, IOR, has important implications for the transmission of monetary policy from federal funds to the repo market. In the pre-crisis era, the active presence of large banks in both the federal funds and repo markets was crucial to the co-movement of these two rates. The unsecured nature of the federal funds transactions in which the collateral constraints, which impose limitations on borrowing in the repo market by arbitrageur banks, typically resulted in a small and positive spread between EFFR and rates on repo transactions where the underlying collateral is a U.S. Treasury or agency security. In contrast, EFFR printing below the repo rates became a frequent phenomenon, especially early in theELBperiod. ThisfactreflectsreducedscopeforarbitrageactivityduetoIOR,asidefromthe dramaticreductioninbanks’needsforshort-termborrowing,asdiscussedpreviously. Specifically, when the repo rates were greater than the EFFR in the past, banks could borrow in the federal funds market and place the cash in the repo market, creating downward pressure on the repo rates and pushing the EFFR up. However, in the presence of the IOR, the incentive for banks to engage in arbitrage activity across the federal funds and repo markets exists only when the repo rates are above the IOR. Although GSEs may also engage in this type of arbitrage, frictions— such as internal restrictions or intra-day timing considerations—likely limit such activity. As a result, we expect a weaker link between the EFFR and the repo rates in the ELB sample on net. Thesupplementarymonetarypolicytoolofthenewframework, theONRRPfacility, hasalso been affecting overnight funding dynamics since its inception in September 2013. The Fed has been offering ON RRPs on a daily basis at a pre-announced offering rate. Through this facility, 10Arepoisthesaleofsecuritieswithanagreementtorepurchasethemataspecifiedpriceonalaterdate. Hence, itiseffectivelyacollateralizedloanwherethelenderofthecashreceivesthesecurityascollateralandtheborrower paysthelenderinterestontheloan. Cashborrowersintherepomarketincludebanksandsecuritiesdealers,while money market mutual funds and government-sponsored enterprises are among the cash lenders. Fed transactions in the repo market are defined from the point of view of the market participants—that is a transaction in which securities are lent by the Fed in lieu of cash is called a reverse repo. 11See https://www.federalreserve.gov/monetarypolicy/policy-normalization.htm for further details on policy normalization. Anderson et al. (2016) provide an overview of movements in money markets after the liftoff. 7

the Fed borrows cash from eligible counterparties in exchange for Treasury securities from its portfolio. These operations provide an investment vehicle for money market participants who compare the facility’s offering rate with other rates and determine whether to bid in the ON RRP operation. TheONRRPoperations,areinessence,similartothetemporaryOMOsintheformofreverse repos conducted by the Fed prior to the crisis,—however, there are also important differences. Participation in the ON RRP operations are open to a wide range of entities, including MMFs, banks, and GSEs, in addition to the primary securities dealers. Indeed, Frost et al. (2015) show that MMFs have been the dominant lenders in ON RRP operations. Therefore, by expanding the set of alternative investments available to MMFs and GSEs, the ON RRP is expected to contribute to improved alignment of secured and unsecured funding rates. The second important difference of the ON RRP from conventional temporary OMOs is that the latter was conducted to move the EFFR close to the FOMC’s target, while the former is intended to set a floor for the EFFR and other overnight rates. The mechanism is similar to that of IOR for banks in the federal funds market; ON RRP counterparties do not have an incentive to invest in alternative sources unless they offer the ON RRP rate or higher. Indeed, Potter (2015) shows that the ON RRP has established a soft floor, as the FOMC intended—that is, although some trades likely occur below the ON RRP rate, volume-weighted average overnight funding rates have mostly been above the offering rate. A general reduction in the volatility of overnight rates is a direct expected effect of the soft floor set by the ON RRP. Such effects are likely to be especially important on financial-reporting days when borrowers contract the size of their balance sheets, leaving cash lenders looking for alternative safe investment options. Take-upattheONRRPfacilitytrendedupforaboutayearfollowingitsinceptioninSeptember 2013, as can be seen from Figure 2. In September 2014, the FOMC reduced the overall limit on the facility substantially (from $1.4 trillion to $300 billion) and introduced an auction process to allocate reverse repos in the event that the overall limit is binding. This change led to a sharp drop in money market rates on that quarter-end as cash lenders scrambled for alternative investments. InOctober2014, theFOMCauthorizedaseriesoftermRRPsspanningyear-endto help address downward pressure on rates. In contrast to the third quarter, money market rates 8

generally stayed at or above the ON RRP rate at year-end, suggesting that perceived investment capacity is an important factor in determining the effectiveness of RRPs in supporting rates. At the time of the rate hike in December 2015, the aggregate cap on ON RRP operations was temporarily suspended. Currently, the ON RRP operations are limited only by the value of the Treasury securities in the Fed’s open market portfolio that are available for these operations, which stand around $2 trillion. 2.3 New Banking Regulations and Dealer Leverage TheannouncementandimplementationofBaselIIIcapitalandliquidityreformshadasignificant effect on the post-crisis financial landscape. Among the Basel III reforms, effects of the liquidity coverageratio(LCR)andtheleverageratioonmoneymarketsareofparticularinterest. TheLCR rule requires banks to hold high-quality liquid assets (HQLA) to meet cash outflows under a 30day stress scenario. Therefore, it has potential implications for bank activity in overnight money markets as many assets and liabilities closely tied to these markets are under the jurisdiction of the LCR. U.S. banking regulators proposed an LCR rule in October 2013 and finalized it about a year later. Although lending in the federal funds market reduces the LCR numerator because reserves are counted as HQLA, cash inflow assumptions applied to regulated financial institutions imply typically limited or no impact of such activity on the LCR on net.12 Similarly, treatment of collateral in case of repo transactions for LCR purposes implies that lending in the repo market (in which underlying collateral is in the HQLA category) has no effect on a bank’s LCR. On the borrowingside, fundingnon-HQLAassetsthrougheitherunsecuredinterbankborrowingorrepos causesa deteriorationintheLCR,creating anincentivefor bankstoreducetheir reliance onsuch financing. However, by the time the initial LCR announcement was made, banks had already reduced their reliance on wholesale (that is, nonretail) funding substantially—for example, Choi and Choi (2016). The IOR arbitrage trades described previously actually increase a bank’s LCR, as the borrowed cash is parked in the arbitrageur bank’s reserve account, which is treated 12In the LCR calculation, cash inflows and outflows over the 30-day stress period are aggregated and netted. There are specific outflow and inflow rates applicable to different assets and liabilities. Funds receivable from regulated banks have an inflow rate of 100 percent. However, cash inflows are capped at 75 percent of outflows in order to ensure a sufficient amount of HQLA. Hence, in some cases lending in the federal funds market may decrease the LCR of a lending bank. 9

as HQLA with no haircuts, and the cash outflow assumption associated with borrowings from GSEs results in a less-than-proportional increase in the denominator. All told, we do not expect the marginal effect of the LCR through IOR arbitrage trades to be material for overnight money market dynamics in the context of our analysis. AnothernotableaspectofBaselIIIformoneymarketactivityistheintroductionofaleverage ratio requirement. This framework requires banks to hold Tier 1 equity equivalent to at least 3 percent of their leverage exposure calculated using their on- and off-balance-sheet assets, including reserves. The Supplementary Leverage Ratio, the regulation that implements the Basel III leverage ratio provisions in the United States, bases the relevant calculations on averages of daily values for on-balance-sheet items. In contrast, for most foreign banks, disclosures are based on month- or quarter-end levels, increasing the incentive to expand balance sheets on non-reporting dates and contract on reporting dates. Therefore, foreign banks can engage in IOR arbitrage trades in the federal funds market without incurring balance sheet costs on non-reporting days. Such trades are always costly for domestic banks from the leverage ratio perspective. Although the leverage ratio requirement will not become binding until 2018, it was announced in mid-2013, and banks started disclosing their leverage ratios to public in January 2015 including three quarters of historical data. Becoming compliant before public disclosures began was an important motivationforbankstomakeadjustmentstotheirbalancesheets. Asaresult,weexpectstronger financial reporting day effects in the federal funds and Eurodollar markets in the ELB sample after the introduction of the leverage ratio requirements. Declining leverage of securities broker dealers has been an important feature of the post crisis landscape (Adrian et al. (2013)). These institutions were not subject to leverage limits prior to the crisis as they were outside the regulated banking system. However, four out of the five major stand-alone investment banks with dealer arms have been integrated into BHCs either via acquisitionsorconversions. Thischangehasbeenamongthemaindriversoflowerdealerleverage along with generally increased risk aversion in the aftermath of the crisis. Dealers dynamically adjusttheirbalancesheetsmainlythroughshort-termborrowingintheformofreposasdiscussed in Adrian and Shin (2010). Along with overall leverage, repo activity of dealers also declined relative to the pre-crisis norms. As can be seen from Figure 3, although repo-based lending by dealers has been relatively stable since 2001, their borrowings through repos have been notably 10

lowersince2007onnet. Thechangeinnet repofinancingismoredramatic: Theratioofnetrepo liabilities to total liabilities for dealers has been steadily decreasing since its peak in 2007 and reached about 8 percent in 2015, almost one fourth of its level in 2007. Against this backdrop, we expect weaker quarter-end effects on repo rates in the ELB sample, as the aforementioned developments likely reduced the scope for quarter-end window dressing compared with the precrisis era. Moreover, the ON RRP facility further limits the effects of financial-reporting days on repo rates by setting a floor. We summarize all the aforementioned changes in the monetary policy and the regulatory environment, as well as their anticipated effects on overnight money market dynamics, in Table 1. 3 Data We exclude the period from mid-2007 to late 2008 from our analysis as it is associated with unprecedented movements in the rates driven by the financial crisis, and focus on two main samples: the pre-crisis sample that spans from January 2, 2001, to July 31, 2007, and the ELB sample that runs from December 17, 2008, to August 28, 2015. The former is associated with the conventional monetary policy operating framework and serves as a benchmark while the latter is a period during which overnight money markets were subject to the significant changes that we discussed in the preceding section. Our data set consists of four overnight money market interest rates. The first one is EFFR, which is calculated as a volume-weighted average of rates on brokered federal funds trades and published by the Federal Reserve Bank of New York (FRBNY). The second series represents the rate on a major alternative unsecured funding source for large banks: the Eurodollar rate (EDR).EurodollarsareU.S.dollar-denominateddepositsheldinabankorabankbranchlocated outside of the United States. U.S. banks and FBOs cannot directly borrow in the Eurodollar market but can take Eurodollar deposits, mainly through their Caribbean branches, and transfer them onshore to fund U.S. operations. Eurodollar deposits that remain outside the United States are not covered by FDIC deposit insurance, while those that are transferred to an insured U.S. affiliate are included in the deposit insurance assessment base. In terms of reserve requirements, they are treated effectively the same as federal funds; because of their unsecured nature and 11

regulatory treatment, Eurodollar deposits constitute a close substitute to federal funds. However, theEurodollarmarkethasamorediversesetofparticipantscomparedwiththefedfundsmarket, as participants do not have to have an account at the Fed. Cipriani and Gouny (2015) estimate that the average volume in the brokered Eurodollar market is three to four times larger than the brokered federal funds market. We use the EDR data that the FRBNY started collecting in March 2010. Prior to this date, we use the EDR series obtained from Wrightson ICAP in our ELB sample. For our analysis of the pre-crisis period, we substitute the overnight LIBOR, which is obtained from Bloomberg, for EDR because the latter is not available.13 The third key rate in our analysis is a representative rate of secured funding from the repo market. A repurchase agreement (repo) is effectively a collateralized loan in which the lender of the cash receives the security as a collateral and the borrower pays the lender interest on the loan. We use the volume-weighted average rate for Treasury GC repo obtained from the FRBNY, which we will refer to as RPR.14 The final segment of the money market we consider is the commercial paper market, in which large corporations issue unsecured or asset-backed short-term securities for a fixed maturity. Although commercial paper is unsecured, it is considered a very safe investment as typically only creditworthy companies with high ratings issue such securities. Commercial paper is especially attractive for institutional investors like MMFs as they are liquid and essentially riskless. We use the overnight AA nonfinancial commercial paper rate (CPR) from the commercial paper data release of the Federal Reserve Board.15 CPR represents an unsecured funding rate not directly affected by the changing monetary policy framework and new banking regulations outlined previously. Visual investigation suggests very strong co-movement among the rates during normal times (Figure 4). Moreover, the sample means and standard deviations of the rates are remarkably closeinthisperiod,ascanbeseenfromPanelAofTable2. However,asonecaninferfromFigure 13LIBOR is a commonly-used indicator for the average rate at which banks may get short-term loans in the London interbank market. It is the benchmark reference rate for various debt instruments. See Hou and Skeie (2014) for a detailed description of the rate-setting mechanism and efforts to reform the LIBOR. 14The repo market can broadly be divided into two parts: the bilateral market where the two parties interact directly, and the triparty market where clearing/brokerage services of a third-party is involved. Total volume of theTreasuryrepomarketiswellabove$2trillion. SeeCopelandetal.(2014a),Baklanovaetal.(2016)forspecific estimates and breakdowns into different segments. 15Data are available at www.federalreserve.gov/releases/cp/. 12

5 and Panel B of Table 2, the co-movement of rates appears to have weakened somewhat over the ELB period, on net. For example, RPR remained especially elevated relative to unsecured rates for some time in late 2012. This behavior of the RPR was reportedly due to longer dealer positioning in Treasury securities that coincided with the Fed’s Maturity Extension Program (MEP) as well as higher Treasury debt issuance.16 In addition to weaker co-movement, calendar effects relative to the level of the rates seem stronger, on average, over the ELB period and the samplemomentsalsoshowmorevariationacrosstherates. Inthenextsection,wespecifymodels to quantify such differences and analyze them in detail. Another important difference between the two sample periods is related to the stationarity of the interest rates. As can be seen in Table 3, in the pre-crisis sample, we cannot reject the null of a unit root in the interest rates at any conventional significance level with respect to both the augmented Dickey and Fuller (1979) (ADF) test statistic and the Elliott et al. (1996) (ERS) point-optimal test statistic. In contrast, we reject the null of unit root for all rates in the ELB sample according to the ADF test statistics, with the exception of CPR, and for all rates according to the ERS test statistic. Therefore, the interest rates are well approximated by integrated processes with a likely common stochastic trend in the pre-crisis sample, reflecting the fact that this period contains a full monetary policy cycle with easing early in the period followed by a gradual tightening beginning in 2004. In the ELB period, the rates are persistent but not integrated against the backdrop of no change in the FFR target. Our modeling strategy incorporates this important difference in rate dynamics. 4 Models We specify models that account for persistence and co-movement of the rates as well as timevariationintheirvolatilitiesandcross-correlations.Wealsoallowforvariouscalendarfactorsthat are known to affect dynamics of rates on specific days. We estimate two different models for the pre-crisisandELBperiodsasunitroottestssuggestthattheinterestratesarewell-approximated 16DuringtheMEP,theFedsoldabout$650billionofshort-termsecuritiesandusedtheproceedstobuylongertermsecurities. Byextendingtheaveragematurityofthesecuritiesinitsportfolio,theFedaimedtoputdownward pressure on longer-term interest rates to contribute to a broader easing in financial market conditions. 13

by integrated processes in the pre-crisis sample while they are persistent but stationary during the ELB sample. The pre-crisis model is a vector error correction (VEC) specification that incorporates the long-run equilibrium relationship of overnight money market rates. Let y denote the vector t of the interest rates at time t, that is, y = (EFFR ,RPR ,LIBOR ,CPR )(cid:48) in the pre-crisis t t t t t sample. The interest rate dynamics are characterized by the following VEC model: p (cid:88) ∆y = Ad +β∆TFFR + Φ ∆y +Θz +(cid:15) , (1) t t t j t−j t−1 t j=1 where d is a vector of indicator variables for calendar effects, which we will explain in detail; t TFFR is the target federal funds rate; z is a vector of error correction terms; and (cid:15) is a zerot t mean martingale difference vector process, which is possibly heteroskedastic. Reflecting the pre-crisis monetary policy operating framework, we impose the restriction that there are three distinct co-integrating relationships with each between EFFR and one of the three remaining rates. Formally, we have z = y −(c +γ y ) where i = 1,2,3.17 it 1t i i i+1,t The vector of calendar effects, d , contains 10 indicator variables to account for reserve maint tenance period days, 2 indicators for elevated payment days within a month (15th and 25th), 2 for financial reporting days (month-end and quarter-end), and a dummy variable to control for the brief disturbance in money markets caused by the September 2001 terror attacks. As a result, the model does not contain a constant vector because it cannot be separately identified given the set of maintenance period indicators. We set p = 4 based on Schwarz information criterion. Therefore, the total number of parameters to be estimated is equal to 140, which results in approximately 46 observations per parameter. There exists a mapping from this VEC system to a VAR that can be defined for the level of interest rates. This mapping allows us to directly compare the results from the pre-crisis period with the ELB period as the model for the latter sample is a VAR in levels. Let Ψ for j j = 1,...,p+1 denote the autoregressive coefficient matrices in the implied VAR. Then we have Ψ = Φ +I +ΘΓ where I is an identity matrix, Γ = (i,−diag{−γ}), i is a vector of ones, γ 1 1 17We obtain very similar results when we estimate the number of co-integrating relationships as well as the co-integration parameters in a less restricted fashion as in Johansen (1995). 14

is the vector of co-integration slopes given previously, and diag(.) indicates a diagonal matrix, Ψ = Φ −Φ for j = 2,...,p, and Ψ = −Φ .18 j j j−1 p+1 p FortheELBperiod,wespecifythefollowingVARmodelinlevelsgiventhestationarybehavior of interest rates in this sample: p (cid:88) y = Πd + Ξ y +(cid:15) , (2) t t j t−j t j=1 whered isnowa9×1vectorthatcontainsmonth-end,quarter-end,day-of-the-week,andelevated t payment flow-day indicators.19 Note that the EDR replaces the LIBOR in this sample, so that y = (EFFR ,RPR ,EDR ,CPR )(cid:48). We set p = 3 based on Schwarz model selection criteria. t t t t t This model has 84 parameters to be estimated, resulting in 78 observations per parameter. Bothvisualinvestigationandformaltestingofthemodelresidualssuggestsignificantvolatility clustering in both sample periods. Hence, we estimate multivariate GARCH specifications to model the second moments. Our modeling strategy closely follows that of Bollerslev (1990); however, instead of assuming a constant conditional correlation matrix, we allow for different correlation structures on financial reporting days. Therefore, our specification can be thought of as a hybrid of the constant correlation model and the dynamic correlation model of Engle (2002), who postulates a fully time-varying conditional correlation matrix. Let E((cid:15) (cid:15)(cid:48)|Ω ) = H where t t t−1 t Ω is the information set at time t, then we can write: t H = D R D , (3) t t t t √ (cid:8) (cid:9) where D = diag h , h = Var((cid:15) |Ω ) and R = Corr((cid:15) |Ω ). The individual variances t it it it t−1 t t t−1 are modeled via the following GARCH specification: h = ω +τ (cid:15)2 +δ h +λ I +λ I , i = 1,...,4, (4) it i i i,t−1 i i,t−1 i,1 m,t i,2 q,t 18Acaveatisthatinthepre-crisismodel,shocksarepermanentduetothemodelingofinterestratesasintegrated processes. 19Day-of-the-week indicators replace those for maintenance period days as the latter become insignificant amid abundant reserves in the ELB period. 15

where I and I are month-end and quarter-end indicators, respectively. In this specification, m q the variance at time t is essentially a weighted average of its lagged value, the new information at time t−1 that is captured by the most recent squared residual, the long-run unconditional variance, and the level shifts in volatility on financial reporting dates. We estimate the GARCH equation under variance targeting so that ω is a function of the sample variance of (cid:15) and the i i,t mean vector of the indicator series. Finally, the correlation matrix R is specified as follows: t R = I R +I R +(1−I −I )R , (5) t m,t m q,t q m,t q,t n −1/2 where R , R , and R are correlation matrices of GARCH residuals, that is, h (cid:15) , at monthm q n it t ends, quarter-ends, and all other days, respectively. 5 Empirical Results 5.1 Relationship of FFR with Other Rates and Monetary Policy Transmission Our estimates for the pre-crisis sample are consistent with the conventional monetary policy implementation framework. As shown in Panel A of Table 4 lags of the EFFR are significant in all other rate equations, implying that interest rates were adjusting in response to changes in the EFFR. In addition, the EFFR was not responding to changes in the other rates as implied by the insignificance of other lagged interest rates in the EFFR equation. The magnitude of response to changes in EFFR is estimated to be somewhat small in the case of the LIBOR, likely reflectingacombinationofnon-synchronoustradingaswellasfactorsthatmayonlyaffectoffshore U.S. dollar funding markets. Other than the EFFR, no other interest rate in the system had predictive power for the remaining interest rates. Moreover, as we see in Panel B, changes in the target federal funds rate are highly significant in all equations of the VEC model. Overall, these results show that funding rates were adjusting in response to policy intervention and dynamics in the federal funds market and are consistent with the view that the overnight money markets were tightly connected through the federal funds market in the pre-crisis period. The estimates from the ELB sample shown in Panel A of Table 5 paint a different picture. The federal funds and Eurodollar markets appear to be closely connected as indicated by the 16

statistical and economic significance of the EFFR coefficients in the EDR equation. Similarly, the EFFR is linked to the CPR, which is the other unsecured rate in the system, although to a lesser extent than the EDR. Therefore, the EFFR continued to be an anchor for unsecured rates in the ELB period, although its transmission has been weaker relative to pre-crisis norms, especially in case of the CPR. The most dramatic change across the two periods concerns the transmission from the federal funds to the repo market. The EFFR is neither an economically nor statistically important predictor of the RPR movements in the ELB period. Another difference is that dynamics in the repo and commercial paper markets appear to affect those in the federal funds market, although such effects are not economically large. Therefore, we conclude that co-movement of the EFFR with other rates became noticeably weaker in the ELB sample amid superabundant reserves, subdued trading, and dominance of IOR arbitrage trades in the federal funds market. Moreover, the disconnect between the EFFR and the RPR emphasizes the diminished role of banks as arbitrageurs, as discussed in Section 2.1. To assess the effects of the ON RRP on money market dynamics, we now focus on the ELB period and estimate VAR models for the two subsamples separated by the inception of the ON RRP facility on September 23, 2013. Although the facility has initially been limited in terms of the overall size and the number of participants, this date provides a natural structural break point in the ELB sample. Moreover, our objective is to obtain estimates for the average effects of the ON RRP over a sufficiently long time period, so this split provides a good empirical setup to achieve that goal. The comparison of the results summarized in Tables 6 and 7 suggests that the ON RRP have hadtwoimportanteffects. First,transmissionfromtheEFFRtotheotherunsecuredratesclearly improved: The sum of lagged EFFR terms increased from 0.23 to 0.29 in the case of the EDR and from 0.16 and 0.33 in the case of the CPR. Second, the RPR became a significant predictor of the EFFR movements, in contrast to the pre-crisis relationship where RPR was moving in response to changes in the EFFR, mainly as a result of cross-market arbitrage. Interestingly, the RPR has also become highly significant in the EDR and CPR equations. Hence, it appears that the ON RRP markedly improved the overall co-movement of overnight interest rates and transmission from the federal funds market to other segments of unsecured funding markets. 17

5.2 Reserve Maintenance Period Effects In Figure 6, we report point and interval estimates for the coefficients of the effects of reserve maintenance days on the EFFR in the pre-crisis period. Clearly, maintenance period days have had small but economically meaningful and statistically significant effects on the EFFR. Due to elevated payment flows following weekends, the EFFR used to be firmer by 1 to 2 basis points on Mondays. By contrast, funds used to trade softer by a slightly greater magnitude on Fridays, as banks generally tried to avoid an excess position over the weekend during which reserves count for three days toward the reserve requirement. Tuesdays were also associated with softness due to reduced demand towards the middle of the week when payment flows are relatively lighter. These estimates are consistent with those of Hamilton (1996), Carpenter and Demiralp (2006), and Judson and Klee (2010)) that were obtained in different empirical frameworks. IntheELBperiod,althoughwecannotstatisticallyrejectdayoftheweekeffectsinthefederal funds market, our estimates (not reported) indicate economically miniscule effects. When we combine our coefficient estimates with trading volumes reported by Afonso et al. (2013b), we find that the average day-of the-week effect is about only 3 percent of its pre-crisis level in dollar terms. Moreover, when we normalize the estimated effects by the standard deviation of the EFFR residuals to control for the dramatically different level of the average EFFR across the two periods, we find that the day-of-the-week effect is about 70 percent weaker in the ELB period. Therefore, we conclude that given the abundance of reserves and their fairly widespread distribution as reported by Ennis and Wolman (2015), reserve-maintenance effects in the federal funds market diminished substantially. 5.3 Market Dynamics on Financial-Reporting Days The estimated magnitudes of calendar effects are quite different across the pre-crisis and ELB periods as evident from Panel B in Tables 4 and 5. However, the average levels of overnight interest rates are dramatically different across the two samples. To control for the general level of interest rates and allow for a direct comparison between the two periods, we normalize the estimates relative to standard deviations of model residuals associated with the respective equation in the VAR system. 18

Figure 7 shows the normalized estimates for the two main samples. In the pre-crisis sample, all rates were subject to modest upward pressure at month-ends, possibly due to heavier payment flows as well as adjustments related to financial reporting. Most comprehensive financial reports are produced on a quarterly basis, so deleveraging by financial intermediaries on quarter-end is common practice. Indeed, quarter-end effects were more prominent than month-end effects, with the exception of the EFFR. Rates were markedly softer in the repo market, likely because securities-financing demand by dealers grew weaker on quarter-ends as these institutions actively managed their leverage. In contrast, it appears that reduced willingness to lend in unsecured markets on quarter-ends was the dominant factor leading to higher rates on financial reporting days. ThispatternisobservedespeciallyforLIBOR,likelyreflectingbanks’desiretoshowstrong liquidity positions on their financial statements and regulatory filings. Money market dynamics on financial-reporting days changed materially in the ELB sample. First of all, both the EFFR and the EDR have been exhibiting economically and statistically significant softening on quarter-ends. This behavior is mainly due to the IOR-arbitrage trades by large BHCs and FBOs dominating the demand side of the federal funds market and the change in the FDIC assessment scheme. Domestic banks, with their total assets being subject to the FDIC assessment, have a strong incentive to reduce IOR-arbitrage trades on quarterends. Significant balance sheet constraints associated with the Basel III leverage ratio that became prevalent in the later part of the ELB sample also likely contributed to these dynamics. Against this backdrop, cash lenders’ search for alternative investments on quarter-ends amid weaker demand by bank borrowers appears to have led to a material softening in the CPR; the estimate for quarter-end effects on the CPR across the two samples became negative. Contrary to the case of the unsecured rates, the quarter-end effect has become insignificant for the RPR in the ELB period, on net. This phenomenon likely reflects a combination of factors. First, earlier in the ELB period, collateral demand was relatively strong due to flight-to-quality flows, leading to increased willingness to lend cash at lower rates in lieu of Treasury collateral. Second, later in the period, as new regulations were announced and implemented, lower dealer leverage and reduced net repo financing probably reduced the scope of quarter-end deleveraging effects. Finally, the availability of the ON RRP as a viable investment, especially on financial reporting 19

dates when other investment options may be limited, reduced the potential for sharp falls in the repo rates. Giventhetimelineofannouncementandimplementationofthenewleverageratioregulations associated with Basel III, we can gain further insight into their effects by examining the results fromthepre-andpost-ONRRPsamples. Figure8showsthenormalizedmonth-endandquarterend effects on rates for the two periods. Consistent with leverage ratio calculations for some foreignbanksbeingbasedonmonth-endaverages,boththeEFFRandtheEDRstartedtodecline notably at month-ends later in the ELB sample. Moreover, downward pressure on these rates at quarter-ends also became more pronounced, especially for the EDR. This likely reflects the fact that Eurodollars are a relatively more important source of dollar funding for foreign banks, which are subject to a less stringent implementation of the Basel III leverage ratio. These banks are subject to reporting their leverage ratio based on only month-end and quarter-end observations as opposed to U.S. banks that are required to calculate their balance sheet ratios based on daily averages over a quarter. In contrast, quarter-end effects on CPR have been relatively stable across the two ELB subsamples, suggesting limited spillover effects from the federal funds and Eurodollar markets. Given that the aforementioned regulations do not have direct implications for the nonfinancial commercial paper market, the incremental change in month- and quarter-end effects on the EFFR and EDR suggests that the leverage ratio regulation has been the primary driver of these dynamics. 5.4 Volatility and Correlation of Overnight Interest Rates In this subsection, we focus on both general and financial-reporting-driven volatility dynamics across the two main sample periods as well as before and after the introduction of the ON RRP facility. The parameter estimates of the volatility models for the pre-crisis and ELB samples are shown in Table 8. As expected, volatility of all rates declined substantially at the ELB in absolute terms. For example, the volatility of innovations in the EFFR equation declined from 5.6 basis points to only about 1 basis point. Meanwhile, the volatility process for the EFFR has become notably less persistent as captured by the decline in the sum of GARCH parameters (τ+δ)andmoreresponsivetoshocksasmeasuredbytheincreaseinthecoefficientofthesquared innovation (τ). Therefore, aside from calendar effects that we will discuss, volatility clustering 20

has become more prevalent in the EFFR amid subdued trading activity in the federal funds market dominated by IOR arbitrage. In the case of the RPR, the volatility process has become somewhat more persistent, and sensitivity to shocks has slightly increased. Figure 9 shows the estimated month-end and quarter-end effects on volatilities in both of the main sample periods.20 As before, estimates are normalized by dividing by the standard deviations of residuals to allow for direct comparison across the two periods. Prior to the crisis, similar to the calendar effects in the conditional mean models, quarter-ends had a larger effect on the volatility of overnight rates. This pattern was especially the case for the RPR with around 2 to 5 times higher volatility on quarter-ends. This substantial quarter-end volatility clustering in the RPR moderated notably in the ELB period but remained significant. This result, combined with insignificance of the quarter-end effect on the level of RPR, suggests that quarter-end dynamics became more complex and worked in both directions in the post-crisis era. In contrast to the RPR, the estimated quarter-end effect on the EFFR volatility increased substantiallyintheELBperiod. However,theestimateisrelativelyimprecise,asitisstatistically significant only at the 10 percent level. In addition, the change in the month-end volatility drift of the EFFR is quite substantial. Consistent with the soft floor set by the ON RRP, volatility of the overnight interest rates declined 35 to 50 percent in the second ELB subsample, as seen in Table 9. Moreover, the estimated volatility parameters indicate a substantial reduction in the overall volatility clustering of the RPR, mainly led by a dramatic decline in the calendar effects (Table 9 and Figure 10). Indeed, the quarter-end spikes in volatility of the RPR due to collateral squeezes and reduced demand for funds by banks became statistically insignificant. Figure 11 illustrates the striking change in the RPR volatility in full detail. An important caveat is that the unconditional variances in our GARCH specifications are anchored to the corresponding sample variances, so the dramatic level shift right after the ON RRP inception reflects the average effect across the two ELB samples. Elevated-volatility episodes in the ON RRP period are related to the debt limit issue and the government shutdown in the fall of 2013. Similar to the case of the RPR, the quarter-end effect on the CPR also became insignificant in the latter ELB sample. In contrast, 20Althoughbasedonasymptoticnormaldistributions,confidencebandsareasymmetric,asweestimatethemin the variance space and then convert to standard deviations. 21

month-end and quarter-end effects, especially the former, became more pronounced for the other unsecuredrates,mainlyduetothepullbackfromtheunsecuredmarketsbybankborrowersdriven by the Basel III leverage regulation. CorrelationstructureofVARinnovationscanprovidefurtherinsightsintotheco-movementof overnight interest rates. Table 10 reports estimates obtained for the pre-crisis and ELB samples from the multivariate GARCH framework defined previously. Interestingly, the correlations of the EFFR residuals with those of the three other rates during normal times are fairly close across the two main samples. Hence, it appears that factors exogenous to the dynamic system of four interestrates, suchasTreasurydebtissuanceandrelatedliquidityeffects, continuedtooperatein a similar fashion on net. The differences on month-ends and quarter-ends are more pronounced, but they are subject to substantial uncertainty. For example, the quarter-end EFFR-RPR correlation is notably different with respect to point estimates, but it is actually statistically indistinguishable from zero in both periods. The EFFR innovations are most strongly correlated with those of the EDR in the ELB sample, especially on month-ends. Estimates reported in Table11suggestthatthisislargelyduetheaforementionedeffectsoftheleverageratio. Another notable change across the two ELB subsamples is the substantial decline in the EFFR-RPR correlation. The changing regulatory environment led to some movements in opposite directions in these two rates at month-ends and quarter-ends, while the ON RRP constrained downward movements in the RPR by setting an effective soft floor in the repo market. In the case of other rate pairs, it is difficult to make a reliable comparison, as most estimates are not statistically different from zero. 6 Conclusion We analyze evolving dynamics of key overnight interest rates that play a crucial role for monetary policy implementation in the United States. We estimate systems of dynamic models for a set of money market rates that incorporate co-movement through their long-run relationship, their potentially different dynamics around financial reporting dates, and time-variation in their volatilities and correlations. 22

We show that at the ELB, although the EFFR continued to provide an anchor for unsecured overnight rates, the transmission to the repo rate is hampered. We find that co-movement across the rates has weakened overall compared with the pre-crisis period, especially on financial reportingdates. Moreover,theday-of-maintenance-periodeffectsontheEFFRhavesubstantially diminished likely reflecting the abundance of bank reserves. When we focus on the ELB period, we find evidence of notably different rate dynamics after the inception of the ON RRP facility. Rate movements, especially on financial reporting days, have changed, reflecting the effects of the announcement and implementation of new regulations that also took place during this period. Consistent with the intended effect of ON RRP to set a soft floor for repo rates, volatility in the repo market has substantially declined after the introduction of the facility. Moreover, calendar effects on RPR volatility largely disappeared, likely reflecting the diminished potential for sharp falls in rates, as well as the availability of the ON RRP as a viable investment, especially on financial reporting dates when other investment options may be limited. 23

References Adrian, T., M. Fleming, J. Goldberg, M. Lewis, F. M. Natalucci, and J. J. Wu(2013): “Dealer Balance Sheet Capacity and Market Liquidity during the 2013 Selloff in Fixed Income Markets,” FEDS Notes. Washington: Board of Governors of the Federal Reserve System. Adrian, T. and H. S. Shin (2010): “Liquidity and leverage,” Journal of Financial Intermediation, 19, 418–437. Afonso, G., A. Entz, and E. LeSueur (2013a): “Who’s Borrowing in the Fed Funds Market,” http://libertystreeteconomics.newyorkfed.org/2013/12/ whos-borrowing-in-the-fed-funds-market.html, Liberty Street Economics Blog. ———(2013b): “Who’sLendingintheFedFundsMarket,” http://libertystreeteconomics. newyorkfed.org/2013/12/whos-lending-in-the-fed-funds-market.html, Liberty Street Economics Blog. Afonso, G., A. Kovner, and A. Schoar (2011): “Stressed, not Frozen: The Federal Funds Market in the Financial Crisis,” Journal of Finance, 66, 1109–1139. Anderson, A. G., J. E. Ihrig, G. C. Weinbach, and E. E. Meade (2016): “What Happened in Money Markets After the Fed’s December Rate Increase?” http://dx.doi.org/10. 17016/2380-7172.1713,FEDSNotes.Washington: BoardofGovernorsoftheFederalReserve System. Ashcraft, A. B., M. L. Bech, , and W. S. Frame (2010): “The Federal Home Loan Bank System: The Lender of Next-to-Last Resort?” Journal of Money, Credit and Banking, 42. Baklanova, V., C. Caglio, M. Cipriani, and A. Copeland (2016): “A New Survey of the U.S. Bilateral Repo Market: A Snapshot of Broker-Dealer Activity,” Federal Reserve Bank of New York Staff Report No. 758. Bech, M., E. Klee, and V. Stebunovs (2014): “Arbitrage, Liquidity and Exit: The Repo and Federal Funds Markets before, during, and Emerging from the Financial Crisis,” in Developments in Macro-Finance Yield Curve Modelling, ed.byJ.S.Chadha, A.C.J.Durr, M.A.S. Joyce, and L. Sarno, Cambridge: Cambridge University Press, 293–325. 24

Bollerslev, T. (1990): “Modelling the Coherence in Short-run Nominal Exchange Rates: A Multivariate Generalized ARCH Model,” The Review of Economics and Statistics, 498–505. Carpenter, S. and S. Demiralp (2006): “The Liquidity Effect in the Federal Funds Market,” Journal of Money, Credit and Banking, 38, 901–920. Choi, D. B. and H.-S. Choi(2016): “Theeffectofmonetarypolicyonbankwholesalefunding,” Federal Reserve Bank of New York Staff Report No.. 759. Cipriani, M. and J. Gouny (2015): “The Eurodollar Market in the United States,” http://libertystreeteconomics.newyorkfed.org/2015/05/ the-eurodollar-market-in-the-united-states.html#.V6OE7KIsC1w, Liberty Street Economics Blog. Copeland, A., I. Davis, E. LeSueur, and A. Martin (2014a): “Lifting the Veil on the U.S. Bilateral Repo Market,” http://libertystreeteconomics.newyorkfed.org/2014/ 07/lifting-the-veil-on-the-us-bilateral-repo-market.html#.VzowDuTYgTV, Liberty Street Economics Blog. Copeland, A., A. Martin, and M. Walker (2014b): “Repo Runs: Evidence from the Tri- Party Repo Market,” Journal of Finance, 69, 2343–2380. d’Amico, S., W. English, D. Lo´pez-Salido, and E. Nelson(2012): “TheFederalReserve’s Large-scale Asset Purchase Programmes: Rationale and Effects,” The Economic Journal, 122, F415–F446. Dickey, D. A. and W. A. Fuller (1979): “Distribution of the Estimators for Autoregressive Time Series with A Unit Root,” Journal of the American statistical association, 74, 427–431. Elliott, G., T. J. Rothenberg, and J. H. Stock (1996): “Efficient Tests for an Autoregressive Unit Root,” Econometrica, 64, 813–836. Engle, R. (2002): “Dynamic Conditional Correlation: A Simple Class of Multivariate Generalized Autoregressive Conditional Heteroskedasticity Models,” Journal of Business & Economic Statistics, 20, 339–350. 25

Ennis, H. M. and A. L. Wolman (2015): “Large Excess Reserves in the United States: A ViewfromtheCross-SectionofBanks,” International Journal of Central Banking, 11, 251–287. Frost, J., L. Logan, A. Martin, P. McCabe, F. Natalucci, and J. Remache (2015): “Overnight R P Operations as a Monetary Policy Tool: Some Design Considerations,” Finance and Economics Discussion Series 2015-010. Washington: Board of Governors of the Federal Reserve System. Gorton, G. and A. Metrick (2012): “Securitized Banking and the Run on Repo,” Journal of Financial Economics, 104, 425–451. Griffiths, M. D. and D. B. Winters (1995): “Day-of-the-week effects in federal funds rates: Further empirical findings,” Journal of Banking & Finance, 19, 1265–1284. Hamilton, J. D. (1996): “The Daily Market for Federal Funds,” Journal of Political Economy, 104, 26–56. Hou, D. and D. Skeie (2014): “LIBOR: Origins, Economics, Crisis, Scandal, and Reform,” Federal Reserve Bank of New York Staff Report No. 667. Ihrig, J. E., G. C. Weinbach, and E. E. Meade (2015): “Rewriting Monetary Policy 101: Whats the Feds Preferred Post-Crisis Approach to Raising Interest Rates?” Journal of Economic Perspectives, 29, 177–198. Johansen, S. (1995): Likelihood-based inference in cointegrated vector autoregressive models, Oxford university press. Judson, R. and E. Klee(2010): “WhithertheLiquidityEffect: TheImpactofFederalReserve Open Market Operations in Recent Years,” Journal of Macroeconomics, 32, 713–731. Kreicher, L. L., R. N. McCauley, and P. McGuire (2013): “The 2011 FDIC assessment on banks managed liabilities: interest rate and balance sheet responses,” BIS Working Papers No. 413. 26

Krishnamurthy, A. and A. Vissing-Jorgensen (2011): “The Effects of Quantitative Easing on Interest Rates: Channels and Implications for Policy,” Brookings Papers on Economic Activity, 2011, 215–287. Kroeger, A. and A. Sarkar (2016): “Monetary Policy Transmission before and after the Crisis,” http://libertystreeteconomics.newyorkfed.org/2016/06/ monetary-policy-transmission-before-and-after-the-crisis.html#.V6OF3KIsC1w, Liberty Street Economics Blog. Munyan, B. (2015): “Regulatory Arbitrage in Repo Markets,” Office of Financial Research Working Paper. Potter, S. (2015): “Money Markets and Monetary Policy Normalization,” https://www. newyorkfed.org/newsevents/speeches/2015/pot150415.html,RemarksattheMoneyMarketeers of New York University, New York City. Spindt, P. A. and J. R. Hoffmeister (1988): “The micromechanics of the federal funds market: Implications for day-of-the-week effects in funds rate variability,” Journal of Financial and Quantitative Analysis, 23, 401–416. Yoldas, E. and Z. Senyuz(2015): “FinancialStressandEquilibriumDynamicsinMoneyMarkets,” Finance and Economics Discussion Series 2015-091. Washington: Board of Governors of the Federal Reserve System. 27

Figures and Tables Figure 1: Reserves and Federal Funds Panel b: Eurodollar Volume billion USD billion USD Billion USD 2500 F R e e d s e fu rv n e d s B a o l u a t n s c ta e n s d ( i l n e g ft ) (right) 350 2000 1800 300 2000 1600 250 1400 1500 200 1200 1000 150 1000 800 100 600 500 400 50 200 0 0 0 2002 2004 2006 2008 2010 2012 2014 2016 2010 2011 2011 2012 2013 2014 2015 2016 Note: Brokered data from FRBNY. Note: DataarequarterlyandobtainedfromCallreports. Panel c: Repo Volume Panel d: Commercial Paper Issuance Billion USD Billion USD 700 8 650 7 600 550 6 500 450 5 400 4 350 300 3 250 2 200 150 1 100 50 0 2002 2004 2006 2008 2010 2012 2014 2016 2002 2004 2006 2008 2010 2012 2014 2016 Source: FRBNY. Source: DTCC. 28

Figure 2: ON RRP Operations Sep-2013 Dec-2013 Mar-2014 Jun-2014 Sep-2014 Dec-2014 Mar-2015 Jun-2015 DSU noillib 350 300 250 200 150 100 50 0 DSU noillib 2000 Takeup (left) Facility cap (right) 1500 1000 500 0 Note: Data are daily and obtained from FRBNY, available at https://apps. newyorkfed.org/markets/autorates/temp. Figure 3: Repo Financing Activity by Securities Brokers and Dealers 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 tnecrep 70 60 50 40 30 tnecrep 35 RA / TA (left) RL / TL (left) NRL / TL (right) 25 15 5 Note: Data are annual and obtained from the Financial Accounts of the U.S. statistical release (Z.1) of the Federal Reserve Board, available at http://www. federalreserve.gov/releases/z1/. TA (TL) denotes total assets (liabilities), RA (RL) denotes repo assets (liabilities), and NR is net repo financing, that is, RL–RA). 29

7002-1002 :setaR tseretnI tekraM yenoM thginrevO :4 erugiF 7002 6002 5002 4002 3002 2002 1002 tnecrep 7 RFFE RPR 6 ROBIL RPC 5 4 3 2 1 0 laicremmocehtmorfsiRPC .grebmoolBmorfsirobiL.YNBRFmorferaRPRdnaRFFE .yliaderaataD :etoN .draoB evreseR laredeF eht fo esaeler atad repap 30

5102-8002 :setaR tseretnI tekraM yenoM thginrevO :5 erugiF 4102 3102 2102 1102 0102 9002 8002 tnecrep 53.0 RFFE 3.0 RPR RDE RPC 52.0 2.0 51.0 1.0 50.0 0 50.0morf dna ,0102 hcraM retfa YNBRF morf si RDE .YNBRF morf era RPR dna RFFE .yliad era ataD :etoN .draoBevreseRlaredeFehtfoesaeleratadrepaplaicremmocehtmorfsiRPC .etadsihtotroirpPACInosthgirW 31

Figure 6: Day of Maintenance Period Effects on EFFR during the Pre-Crisis Period Basis points ZLB Basis points 4 4 4 4 2 2 2 2 0 0 0 0 −2 −2 −2 −2 −4 −4 −4 −4 R1 F1 M1 T1 W1 R2 F2 M2 T2 W2 R1 F1 M1 T1 W1 R2 F2 M2 T2 W2 Note: Dots indicate point estimates and horizontal lines mark the boundaries of the 95 Calendar Effects: Pre−crisis Calendar Effects: ZLB percentconfidencebands. M,T,W,R,andFdenotedaysoftheweekfromMondayto Pre−crisis Basis points ZLB Basis points Friday. Thesubscriptsindicatewhetherthecorrespondingdateisthefirstorthesecond oneinthem20aintenanceperiod. 20 20 20 10 10 10 10 0 0 0 0 −10 −10 −10 −10 −20 −20 −20 −20 M Q M Q M Q M Q M Q M Q M Q M Q FFR RP LIBOR CP FFR RP ED CP Calendar Effects on Volatility: Pre−crisis Calendar Effects on Volatility: ZLB Pre−crisis Basis points ZLB Basis points 20 20 20 20 10 10 10 10 0 0 0 0 −10 −10 −10 −10 32 −20 −20 −20 −20 M Q M Q M Q M Q M Q M Q M Q M Q FFR RP LIBOR CP FFR RP ED CP

FFR Maintenance Period Effects: Pre−crisis FFR Maintenance Period Effects: ZLB Pre−crisis Standard deviations ZLB Standard deviations 1.0 1.0 1.0 1.0 0.5 0.5 0.5 0.5 0.0 0.0 0.0 0.0 −0.5 −0.5 −0.5 −0.5 −1.0 −1.0 −1.0 −1.0 R1 F1 M1 T1 W1 R2 F2 M2 T2 W2 R1 F1 M1 T1 W1 R2 F2 M2 T2 W2 Figure 7: Month- and Quarter-end Effects Pre−crisis Standard deviations ELB Standard deviations 4 4 4 4 2 2 2 2 0 0 0 0 −2 −2 −2 −2 −4 −4 −4 −4 M Q M Q M Q M Q M Q M Q M Q M Q EFFR RPR LIBOR CPR EFFR RPR EDR CPR Pre−crisis Standard deviations ELB Standard deviations Figure 8: Month- and Quarter-end Effects within the ELB Period 4 4 4 4 2 2 2 2 Before ON RRP Standard deviations After ON RRP Standard deviations 5 5 5 5 0 0 0 0 −2 −2 −2 −2 0 0 0 0 −4 −4 −4 −4 −5 M Q M Q M Q M Q −5 −5 M Q M Q M Q M Q −5 EFFR RPR LIBOR CPR EFFR RPR EDR CPR −10 −10 −10 −10 M Q M Q M Q M Q M Q M Q M Q M Q FFR RPR EDR CPR FFR RPR EDR CPR Before ON RRP Standard deviations After ON RRP Standard deviations 4 Note: Dots indicate point estimates and h4orizonta4l lines mark the boundaries of the 95 4 percent confidence bands. M and Q denote month-end and quarter-end respectively. Effectsarenormalizedwithrespecttothestandarddeviationsofmodelresiduals. 2 2 2 2 0 0 0 0 −2 −2 −2 −2 33 −4 −4 −4 −4 M Q M Q M Q M Q M Q M Q M Q M Q FFR RPR EDR CPR FFR RPR EDR CPR

FFR Maintenance Period Effects: Pre−crisis FFR Maintenance Period Effects: ZLB Pre−crisis Standard deviations ZLB Standard deviations 1.0 1.0 1.0 1.0 0.5 0.5 0.5 0.5 0.0 0.0 0.0 0.0 −0.5 −0.5 −0.5 −0.5 −1.0 −1.0 −1.0 −1.0 R1 F1 M1 T1 W1 R2 F2 M2 T2 W2 R1 F1 M1 T1 W1 R2 F2 M2 T2 W2 Pre−crisis Standard deviations ELB Standard deviations 4 4 4 4 2 2 2 2 0 0 0 0 −2 −2 −2 −2 −4 −4 −4 −4 M Q M Q M Q M Q M Q M Q M Q M Q EFFR RPRFigure 9L:IBOMRonth- anCdPRQuarter-end EffEeFFcRts on VolRaPtRility EDR CPR Pre−crisis Standard deviations ELB Standard deviations 4 4 4 4 Before ON RRP Standard deviations After ON RRP Standard deviations 2 2 2 2 5 5 5 5 0 0 0 0 0 0 0 0 −2 −2 −2 −2 −4 −4 −4 −4 −5 −5 −5 −5 M Q M Q M Q M Q M Q M Q M Q M Q EFFR RPR LIBOR CPR EFFR RPR EDR CPR −10 −10 −10 −10 M Q M Q M Q M Q M Q M Q M Q M Q FFR RPR EDR CPR FFR RPR EDR CPR Figure 10: Month- and Quarter-end Effects on Volatility within the ELB Period Before ON RRP Standard deviations After ON RRP Standard deviations 4 4 4 4 2 2 2 2 0 0 0 0 −2 −2 −2 −2 −4 −4 −4 −4 M Q M Q M Q M Q M Q M Q M Q M Q FFR RPR EDR CPR FFR RPR EDR CPR Note: Dots indicate point estimates and horizontal lines mark the boundaries of the 95 percent confidence bands. M and Q denote month-end and quarter-end respectively. Effectsarenormalizedwithrespecttothestandarddeviationsofmodelresiduals. 34

Figure 11: Repo Rate Volatility and ON RRP Basis points 8 8 ON RRP Started 7 7 6 6 5 5 4 4 3 3 2 2 1 1 2009 2010 2011 2012 2013 2014 2015 35

Table 1: Changes in Monetary and Regulatory Policy and Implications Superabundant reserves and IOR Lower trading volumes in (i) Weaker co-movement of EFFR with other rates the federal funds market (ii) Increased EFFR volatility Reduced scope for repo-arbitrage Weaker federal funds - repo co-movement trades by banks Abundant reserves and MP effects diminish in the aggregate widespread distribution ON RRP Inclusion of MMFs and GSEs (i) Stronger co-movement of overnight interest rates among counterparties (ii) Lower interest rate volatility (iii) Weaker financial reporting effects on repo rates New Regulations and Lower Dealer Leverage LCR IOR arbitrage trades more attractive, but limited effect due to other regulatory constraints FDIC assessment change Stronger financial-reporting-day effects on unsecured Leverage ratio rates and their volatility Diminishing leverage and repo Weaker financial-reporting-day effects on repo rates financing by dealers 36

Table 2: Descriptive Statistics of Money Market Rates EFFR RPR LIBOR/EDR* CPR Panel A: Jan. 2, 2001-July 31, 2007 Mean 2.937 2.881 2.999 2.927 Stdev 1.660 1.639 1.661 1.662 10th 1.010 0.980 1.058 0.990 50th 2.480 2.440 2.541 2.450 90th 5.250 5.220 5.301 5.250 AC(1) 0.999 0.999 0.999 0.999 Panel B: Dec. 17, 2008-Aug. 28, 2015 Mean 0.129 0.118 0.137 0.107 Stdev 0.042 0.068 0.051 0.058 10th 0.080 0.030 0.080 0.040 50th 0.130 0.110 0.130 0.090 90th 0.190 0.210 0.210 0.190 AC(1) 0.954 0.920 0.950 0.958 Note: Data are daily. Mean, standard deviation and quantiles are reported in percent. AC(1) denotes first order autocorrelation. * LIBOR is used for Panel A calculations and EDR is used in Panel B. Table 3: Unit Root Tests EFFR RPR LIBOR/EDR* CPR Panel A: ADF Test Pre-crisis -1.24 -1.31 -1.00 -1.02 ELB -3.37 -3.17 -2.82 -2.52 Panel B: ERS Test Pre-crisis 251.3 275.6 158.2 195.7 ELB 1.2 2.8 2.5 3.4 Note: ADFistheaugmentedDickeyandFuller(1979)testwiththe1,5,and10percent critical values of -3.44, -2.87, and -2.57, respectively. ERS is the point optimal test of Elliottetal.(1996)withthe1,5,and10percentcriticalvaluesof1.99,3.26,and4.48, respectively. * LIBOR is used for Panel A calculations and EDR is used in Panel B. 37

Table 4: Overnight Money Market Rates before the Financial Crisis EFFR RPR LIBOR CP Panel A Autoregressive terms (sum) EFFR 0.947 0.449 0.345 0.521 (0.00) (0.00) (0.00) (0.00) RPR 0.033 0.694 0.010 0.001 (0.39) (0.00) (0.72) (0.98) LIBOR 0.016 -0.125 0.546 -0.091 (0.91) (0.41) (0.00) (0.43) CP 0.005 -0.021 0.099 0.570 (0.97) (0.92) (0.21) (0.00) Panel B Change in Target FFR and Calendar Effects ∆TFFR 0.454 0.406 0.337 0.416 (0.00) (0.00) (0.00) (0.00) 15th 5.50 6.04 6.10 7.00 (0.00) (0.00) (0.00) (0.00) 25th 4.33 0.69 0.09 1.14 (0.00) (0.24) (0.75) (0.00) Month-end 5.33 4.15 7.43 6.20 (0.00) (0.00) (0.00) (0.00) Quarter-end 5.66 -12.52 17.33 11.17 (0.03) (0.01) (0.00) (0.00) Note: Columns represent equations of the model. The sum of autoregressive terms (cid:80) correspond to Ψ in the notation of section 4. p-values based on robust (HAC) j standard errors are reported in parentheses. Calendar effects are reported in basis points. Daily sample runs from January 2, 2001, to July 31, 2007. 38

Table 5: Overnight Money Market Rates at the ELB EFFR RPR EDR CPR Panel A Autoregressive terms (sum) EFFR 0.911 0.107 0.223 0.153 (0.00) (0.21) (0.00) (0.02) RPR 0.032 0.809 0.014 -0.011 (0.00) (0.00) (0.29) (0.47) EDR -0.024 0.048 0.705 0.000 (0.53) (0.50) (0.00) (1.00) CP 0.036 0.054 0.069 0.881 (0.01) (0.14) (0.00) (0.00) Panel B Calendar Effects 15th 0.80 3.29 0.85 0.96 (0.00) (0.00) (0.00) (0.00) 25th -0.26 0.65 -0.08 0.37 (0.01) (0.08) (0.36) (0.05) Month-end -0.14 3.47 -0.13 0.37 (0.63) (0.00) (0.72) (0.20) Quarter-end -3.21 -0.41 -5.07 -1.58 (0.00) (0.70) (0.00) (0.03) Note: Columns represent equations of the model. The sum of autoregressive terms (cid:80) correspond to Ξ in the notation of section 4. p-values based on robust (HAC) j standard errors are reported in parentheses. Calendar effects are reported in basis points. Daily sample runs from December 17, 2008, to August 28, 2015. 39

Table 6: Overnight Money Market Rates before the ON RRP EFFR RPR EDR CPR Panel A Autoregressive terms (sum) EFFR 0.911 0.112 0.226 0.164 (0.00) (0.26) (0.00) (0.03) RPR 0.018 0.803 0.002 -0.022 (0.13) (0.00) (0.88) (0.23) EDR -0.002 0.045 0.739 0.005 (0.97) (0.59) (0.00) (0.94) CPR 0.028 0.052 0.047 0.880 (0.04) (0.20) (0.01) (0.00) Panel B Calendar Effects 15th 1.11 3.93 1.26 1.42 (0.00) (0.00) (0.00) (0.00) 25th -0.29 0.70 0.01 0.54 (0.03) (0.17) (0.94) (0.02) Month-end 0.79 3.94 1.19 0.71 (0.00) (0.00) (0.00) (0.06) Quarter-end -3.14 -1.06 -4.29 -2.02 (0.00) (0.45) (0.00) (0.04) Note: Columns represent equations of the model. The sum of autoregressive terms (cid:80) correspond to Ξ in the notation of section 4. p-values based on robust (HAC) j standard errors are reported in parentheses. Calendar effects are reported in basis points. Daily sample runs from December 17, 2008, to September 20, 2013. 40

Table 7: Overnight Money Market Rates after the ON RRP EFFR RPR EDR CPR Panel A Autoregressive terms (sum) EFFR 0.823 0.033 0.290 0.333 (0.00) (0.83) (0.06) (0.00) RPR 0.102 0.813 0.096 0.084 (0.00) (0.00) (0.00) (0.00) EDR -0.127 0.101 0.416 -0.145 (0.01) (0.42) (0.00) (0.01) CPR 0.129 0.126 0.113 0.565 (0.04) (0.21) (0.12) (0.00) Panel B Calendar Effects 15th -0.02 1.61 -0.18 -0.24 (0.87) (0.00) (0.28) (0.27) 25th -0.08 0.46 -0.22 -0.12 (0.50) (0.17) (0.09) (0.54) Month-end -2.35 2.42 -3.25 -0.37 (0.00) (0.00) (0.00) (0.15) Quarter-end -3.41 1.10 -6.80 -1.05 (0.00) (0.37) (0.00) (0.02) Note: Columns represent equations of the model. The sum of autoregressive terms (cid:80) correspond to Ξ in the notation of section 4. p-values based on robust (HAC) j standard errors are reported in parentheses. Calendar effects are reported in basis points. Daily sample runs from September 23, 2013, to August 28, 2015. 41

Table 8: Volatility of Rates Pre-crisis ELB EFFR RPR LIBOR CPR EFFR RPR EDR CPR σ 5.64 5.91 3.94 4.30 1.05 2.38 1.22 1.51 (cid:15) τ 0.116 0.306 0.450 0.316 0.253 0.231 0.440 0.414 (0.01) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) δ 0.839 0.173 0.180 0.171 0.396 0.385 0.240 0.282 (0.00) (0.11) (0.03) (0.01) (0.05) (0.00) (0.00) (0.00) Table 9: Volatility of Rates within the ELB Period Before ON RRP After ON RRP EFFR RPR EDR CPR EFFR RPR EDR CPR σ 1.11 2.65 1.27 1.67 0.73 1.40 0.79 0.83 (cid:15) τ 0.212 0.159 0.365 0.383 0.189 0.315 0.458 0.158 (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.02) δ 0.561 0.327 0.368 0.281 0.191 0.465 0.146 0.681 (0.00) (0.00) (0.00) (0.00) (0.02) (0.01) (0.00) (0.00) Note: p-values based on robust standard errors are reported in parentheses. σ are (cid:15) reported in basis points. 42

Table 10: Correlations of VAR Residuals Pre-crisis ELB RPR LIBOR CPR RPR EDR CPR Normal times 0.490 0.586 0.614 0.457 0.545 0.373 (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) Month-end 0.421 0.246 0.341 0.301 0.879 0.395 (0.04) (0.47) (0.22) (0.19) (0.00) (0.37) Quarter-end 0.348 0.334 0.362 -0.056 0.564 0.360 (0.30) (0.23) (0.32) (1.00) (0.03) (0.29) Table 11: Correlations of VAR Residuals within the ELB Period Before ON RRP After ON RRP RPR EDR CPR RPR EDR CPR Normal times 0.502 0.546 0.413 0.128 0.612 0.173 (0.00) (0.00) (0.00) (0.16) (0.00) (0.07) Month-end 0.395 0.596 0.104 -0.291 0.854 0.039 (0.17) (0.05) (0.63) (1.00) (0.00) (0.90) Quarter-end 0.032 0.595 0.358 -0.489 0.356 0.334 (0.95) (0.05) (0.34) (1.00) (0.59) (0.51) Note: Correlations with EFFR. p-values based on robust standard errors are reported in parentheses. 43