The Cross Section of Money Market Fund Risks and Financial Crises

Finance and Economics Discussion Series Divisions of Research & Statistics and Monetary Affairs Federal Reserve Board, Washington, D.C. The Cross Section of Money Market Fund Risks and Financial Crises Patrick E. McCabe 2010-51 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 Cross Section of Money Market Fund Risks ∗ and Financial Crises PatrickMcCabe BoardofGovernorsoftheFederalReserveSystem September12,2010 Abstract Thispaperexaminestherelationshipbetweenmoneymarketfund(MMF)risksandoutcomes during crises, with a focus on the ABCP crisis in 2007 and the run on money funds in 2008. I analyze three broad types of MMF risks: portfolio risks arising from a fund’s assets, investor risk reflecting the likelihood that a fund’s shareholders will redeem shares disruptively,andsponsorriskduetouncertaintyaboutMMFsponsors’supportfordistressedfunds. IfindthatduringtherunonMMFsinSeptemberandOctober2008,outflowswerelargerfor MMFs that had previously exhibited greater degrees of all three types of risk. In contrast, as theasset-backedcommercialpaper(ABCP)crisisunfoldedin2007,manyMMFssufferedcapitallosses,butinvestorflowswererelativelyunresponsivetorisks,probablybecauseinvestors correctlybelievedthatsponsorswouldabsorbthelosses. However,theconsequencesofMMF riskswerequitecostlyforsomesponsors: Usingauniquedatasetofsponsorinterventions,I showthatsponsorfinancialsupportwasmorelikelyforMMFsthatpreviouslyearnedhigher grossyields(ameasureofportfoliorisk)andfundswithbank-affiliatedsponsors.Funds’gross yieldsandbankaffiliation(butnotfunds’ratings)alsowouldhavehelpedforecastholdingsof distressedABCP.Thispaperprovidessomeusefullessonsforinvestorsandpolicymakers.The significanceofMMFrisksinpredictingpooroutcomesinpastcriseshighlightstheimportance ofmonitoringsuchrisks,andIoffersomeusefulproxiesfordoingso. Thepaperalsoargues for greater attention to the systemic risks posed by the industry’s reliance on discretionary sponsorsupport. ∗TheopinionsexpressedaremineanddonotnecessarilyreflectthoseoftheFederalReserveBoardoritsstaff.Ithank AlyciaChinforexcellentresearchassistance. Thispaperalsohasbenefittedfromhelpfulcommentsandsuggestions fromSteffanieBrady,LaiChingChan,JimClouse,JoshuaGallin,PeterHolirads,MichaelPalumbo,RobertPlaze,Paula Tkac,andseminarparticipantsattheFederalReserveBoardandtheFederalReserveBankofKansasCity. Errorsand shortcomingsare,ofcourse,minealone.

1 Introduction Money market funds (MMFs or “money funds”) have an impressive record of price stability. From the introduction of the rules specifically governing these funds in 1983 until the Lehman bankruptcy in September 2008, only one small MMF lost money for investors, and that loss, in 1994,hadlittlebroaderimpactontheindustry. AlthoughMMFprospectusesandadvertisements must warn that “it is possible to lose money by investing in the Fund” (U.S. Securities and Exchange Commission, 1998a, 2003), investors virtually never lost anything. Indeed, the perceived safetyofMMFstypicallypromptedinflowstothefundsduringperiodsofheighteneduncertainty and motivated some academic researchers to suggest that money funds might function well as “narrow banks” that provide liquidity services (Gorton and Pennacchi, 1993; Scott, 1998; Miles, 2001;Pennacchi,2006). However, two crises in the MMF industry during the financial turmoil that began in 2007 underlined the importance of money fund risks for MMF investors and sponsors, as well as for the broader financial system. The meltdown of the market for asset-backed commercial paper (ABCP) that began in August 2007 caused capital losses for many money funds that held ABCP, butthelosseswereabsorbedbyMMFsponsors(thatis,assetmanagementfirmsandtheirparents and affiliates), so MMF investors lost nothing. In contrast, losses on Lehman Brothers debt followingthatfirm’sbankruptcyinSeptember2008causedtheReservePrimaryFundto“breakthe buck”—itssharepricefellbelow$1—andcostitsshareholdersliquidityaswellasprincipal(asof thiswriting,theassetsofthefundstillhadnotbeencompletelydistributed). Moreover,thedamagequicklyspreadbeyondReserveanditsinvestorsamidabroaderrunonMMFs. Othermoney fund investors were put at risk as concerns about the funds’ vulnerabilities prompted a vicious cycle of redemptions, efforts by MMFs to sell assets, declines in prices for money market instruments,andthepossibilityofcapitallossesthatmotivatedfurtherredemptions. Abroaderliquidity crisis developed as MMF managers, facing enormous redemptions, curtailed their lending to firms and institutions. The run on MMFs appears to have been slowed only by announcements on September 19 of unprecedented government interventions to support MMFs and short-term fundingmarkets. ThispaperexaminestherelationshipbetweenMMFrisksandoutcomesduringcrises,with afocusontheABCPcrisisin2007andtherunonmoneyfundsin2008. Idescribethreebroadtypes ofMMFrisks: (1)portfoliorisksarisingfromthecredit,liquidity,andinterest-raterisksposedbya fund’sassets;(2)investorriskduetothecompositionofanMMF’sinvestorsandthelikelihoodthat theywillsuddenlyanddisruptivelyredeemshares;and(3)sponsorriskthatreflectsthepossibility that an MMF sponsor will not provide financial support for an ailing fund. For each type of risk, I review the relevant academic literature and develop proxies for use in empirical analyses. In addition,IconstructauniquedatasetofsponsorsupportactionsandMMFholdingsofdistressed assets. I employ this data set and the risk proxies in an empirical study of the links between 1

MMFrisksandthecrosssectionofMMFcrisisoutcomes,includingnetredemptionsandsponsor support actions. I find that MMF risks measured well before the unfolding of each crisis had important predictive power for outcomes during the crisis, and I conclude by discussing some policyimplicationsoftheseresults. ThetwocrisesIstudyprovidedifferentperspectivesontheimportanceofMMFrisks. The runonMMFsin2008wasnotindiscriminate;IfindthatredemptionsfromprimeMMFsmarketed toinstitutionalinvestorswerecorrelatedsignificantlywithexanteindicatorsforeachofthethree typesofrisk. Forexample,outflowswerelargerforMMFsthathadpaidhighergrossyieldsinthe previousyearandthuswerelikelycarryinggreaterportfoliorisks,forfundswithlargerpre-crisis flow volatility that signified greater investor risk, and for funds that had sponsors with wider credit default swap (CDS) spreads and hence greater sponsor risk. Meanwhile, net redemptions fromretailprimeMMFsduringtherunvariedwithinvestorriskproxiesbutnotsignificantlywith portfolio or sponsor risk measures, perhaps because retail investors—who generally did not redeemsharesenmasse—werelesscognizantofMMFvulnerabilitiesandposedlowerinvestorrisk for the funds. Indeed, one lesson from the distinction between institutional and retail investors’ behavior during the run is the interactions among fund risks: MMFs with greater investor risks werealsomoresensitivetoportfolioandsponsorrisks. Interactions among fund risks were also consequential during the ABCP crisis, as widespread sponsor support absorbed funds’ losses. With sponsor risks apparently dormant, other MMFrisks—atleastasperceivedbyfundshareholders—alsoremainedlatent,andthefundssaw onlymodestnetoutflowsthatexhibitedlittlecross-sectionalcorrelationwithexanterisks. However, MMF risks were consequential for money fund sponsors; their financial support for their funds reflected concerns about actual or expected losses in funds’ portfolios as well as concerns about investors’ potential responses to those risks. Using a unique data set of sponsor support actions in the wake of the ABCP crisis, I show that portfolio risks, as measured by gross yields in the year prior to the crisis, are useful for predicting whether sponsors intervened to support their funds. A separate analysis of MMFs’ holdings of distressed ABCP corroborates this result. Interestingly,sponsorriskplayedamorecomplexroleinABCPcrisisthanduringtherunin2008. MMFswithbank-affiliatedsponsors,whichpresumablyhadparticularlydeeppocketswithwhich tosupportailingmoneyfunds,weremorelikelybothtoholdtroubledABCPandtoreceivefinancial support to absorb losses. However, controlling for bank affiliation, riskier sponsors (those withhigherpre-crisisCDSspreads)weremorelikelytoexperienceproblems. My findings provide some useful lessons for MMF shareholders and policymakers alike. The significance of MMF risks in predicting poor outcomes in past crises underscores the importanceofmonitoringtheserisks,andthispaperofferssomeusefulproxiesfordoingso. Forexample,shareholdersandregulatorsmighttrackfunds’grossyieldsforearlysignsofundueportfolio risks,particularlyinlightofassetmanagers’incentivestotakeonriskstoboostyields. Thispaper 2

alsoshowsthatMMFrisksarebroaderthantheportfoliorisksthatarethefocusofthecurrentregulatoryframeworkformoneyfunds. Theimportanceofinvestorriskduringtherunin2008lends somesupportfortheSecuritiesandExchangeCommission’s(SEC’s)2009proposalstorequireadditional liquidity for funds that are marketed to riskier investors, such as institutional investors, and the proxies for investor risk that I employ may be useful for identifying funds with riskier clienteles. ThelinkbetweensponsorriskandholdingsofdistressedpaperduringtheABCPcrisis indicates that the sponsor-support option may distort incentives for portfolio managers, and the role of sponsor risk in channeling concerns about financial institutions to their off-balance-sheet MMFs during the 2008 run suggests that expectations for such support may contribute to transmissionoffinancialshocks. Theseconcernsatleastwarrantgreaterattentiontothesystemicrisks posedbytheMMFindustry’srelianceonsponsorsupport. Importantly,theanalysisinthispaperfocusesonrisksthatexplaincross-sectionalvariation inmoneyfunds’experiencesduringepisodesoffinancialturmoil. Assuch,Idonotfocusonsome ofthegenericfeaturesofMMFsthatmakethemstructurallyvulnerabletoruns—suchasMMFs’ practice of rounding their net asset values (NAVs) to the nearest penny—or analyze the types of events that can trigger runs. My results do suggest that fund-specific risks are important during crisesandthatsuchrisksmayinteractwiththestructuralvulnerabilitiesofMMFstocontributeto broaderrisks. Forexample,incentivesforinvestorstoredeemsharesinfundswithroundedNAVs arestrengthenedwhenportfoliorisksarelarger,whenotherMMFinvestorsaremoreriskaverse and more sophisticated (and hence more likely to redeem shares quickly), and when sponsors are less likely to bail out troubled MMFs. The aggregate importance of cross-sectional variation was also evident in the example of the Reserve Primary Fund, which demonstrated that even a single poorly run fund may impose significant costs on the rest of the industry as well as on the broaderfinancialsystem. Hence,byhighlightingsomeexanteriskmeasuresthatpredictedpoor outcomesforindividualMMFsduringpastcrises,thispaperprovidesguidanceonmonitoring— andperhapsmitigating—MMFriskstoreducethelikelihoodofanothercrisis. Section 2 of this paper describes data sources. Section 3 reviews the historical significance ofMMFrisks,academicliteraturerelevanttotheirstudy,andthetwocrisesthatIexamineinthis paper. Section 4 introduces proxies for portfolio, investor, and sponsor risks. Section 5 describes my empirical analysis of the run in 2008, and section 6 turns to an analysis of the ABCP crisis in 2007. Section7concludesthepaperanddiscussespolicyimplications. 2 Datasources The U.S. MMF industry comprises three basic types of funds: (i) prime funds, which chiefly investinshort-termprivatedebtinstrumentssuchascommercialpaper,bankcertificatesofdeposit (CDs), and floating-rate notes issued by private firms; (ii) government-only MMFs, which typicallyholdonlyobligationsoftheTreasury,U.S.governmentagencies,andgovernment-sponsored 3

enterprises, as well as repurchase agreements collateralized with such instruments; and (iii) taxexempt MMFs, which generally hold municipal securities. This paper focuses on prime MMFs, whichborethebruntofthestrainsduringtheMMFcrisesin2007and2008. Primefundsaccount for most MMF assets; Investment Company Institute (ICI) weekly data show that assets under managementinU.S.MMFstotaled$3.58trillionasofSeptember10,2008(justbeforetheLehman bankruptcy and the run on MMFs): $2.18 trillion in prime MMFs, $0.89 trillion in governmentonlyfunds,and$0.52trillionintax-exemptfunds.1 The fund-level data used in this paper mostly come from iMoneyNet, a data vendor that provides weekly and monthly data on each MMF’s assets under management, yields, expense ratios, portfolio composition, and weighted average maturity (WAM). The iMoneyNet data are survivorship-bias free, as they include defunct funds. iMoneyNet also classifies MMFs by fund typeandinvestortype—MMFsareconsideredeither“institutional”or“retail,”dependingonthe types of investors to whom the funds are marketed. I make several adjustments and corrections to the iMoneyNet data. For example, I analyze only MMFs with at least 12 months of pre-crisis dataandfundswithatleast$100millioninassets, andImakeseveralcorrectionstoMMFassets recordedduringtherunin2008. TheseadjustmentsaredescribedinsectionA.1oftheappendix. I obtained some fund and sponsor data from other sources. For example, data on creditdefault swap (CDS) spreads for MMF sponsors came from Markit, and I used Moody’s data to corroborateiMoneyNetMMFratingsinformation. Animportantcontributionofthispaperismycompilationofauniquedatasetofsponsorsupport actions and the securities that prompted such actions. The data come both from public sources and from confidential SEC records. Many of the sponsor-support records are public because sponsors’ actions required some relief from prohibitions under the Investment Company Act of 1940 (ICA). That relief typically came in the form of a “no-action” letter from SEC staff, andtheSEChaspublishedtheseno-actionlettersonitswebsite.2 Anadditionalsourceofpublic information on sponsor support is the financial statements of asset management companies and theirparentfirms.3 TheSECalsoobtainsdataonsponsorsupportarrangementsthatremainconfidential. For example,SECrule2a-7requiresafundtonotifytheCommissionintheeventofadefaultofanysecurityrepresentingmorethan0.5percentofthefund’sassets,andthefundmustalsodescribehow it intends to respond to the default. In other cases, funds have voluntarily apprised SEC staff of supportactionsthatdidnottriggernotificationrequirements. SECstaffprovidedmewithacomplete list of the sponsor support actions for MMFs for which the Commission had been notified 1Seehttp://www.ici.org/pdf/mm_data_2010.pdf. 2TheICArestrictstransactionsbetweenmutualfundsandtheirsponsors.Inparticular,no-actionreliefwasneeded whenadvisersortheiraffiliatespurchasedfrommoneyfundssecuritieswhichwerestilltechnicallyeligibleasMMF portfoliosecurities.Seehttp://www.sec.gov/divisions/investment/im-noaction.shtml#money. 3PeterCranehascompiledausefullistingofsupportinformationcollectedfrompubliclyavailableSECdataand fromfinancialstatements(Crane,2009). 4

betweenAugust2007andMarch2009. TheseincludesupportarrangementstoprotectMMFsfrom lossesorrunsintheaftermathofboththeABCPcrisisandtheLehmanbankruptcy. Togetherwith the public information described above, these data represent the most complete available record of sponsor support for MMFs affected by the financial turmoil that began in 2007—although the fact that some support can occur without notification of the SEC suggests that the data still may notreflecteveryinstanceofsupport. Information about MMFs’ holdings of specific securities (such as individual ABCP issues and Lehman debt securities) comes from two sources. The first is each fund’s closest previous quarterlyportfolio-holdingsfilingwiththeSEC,whichisavailableontheSEC’sEDGARdatabase. Where appropriate, I supplement those data with information garnered from records of sponsor support and other SEC notifications described above, since these records typically name the distressedsecuritiesatissue. TheunitofobservationinmyanalysisistheMMF.Onemoneymarketfundmaycomprise multiple share classes, each of which is a claim on the same portfolio assets. All share classes in anMMFhavethesamegrossyield—thatis,theyieldearnedonportfolioinstrumentsbeforefees. However, different share classes have different fee structures (expense ratios) and pay investors differentnetyields(netyieldisgrossyieldlesstheexpenseratio).4 3 Moneymarketfundrisks: historyandliterature Thefinancialturmoilthatbeganin2007heightenedawarenessofMMFrisksamonginvestors,academics,andpolicymakers. PriortotheABCPcrisisandtherunin2008,MMFrisks—particularly investor and sponsor risks—appear to have been viewed as unimportant or latent. MMF assets grewatanaverageannualrateof13percentoverthequartercenturyfromtheendof1983tothe eve of the Lehman crisis (Investment Company Institute, 2010, p. 160), and the funds’ perceived safetywasundoubtedlypartoftheirappealtoinvestors. MMFs’longrecordofstabilityalsoimpressedsomeacademics,whoadvocatedMMFsasalternativestobanksasprovidersofliquidity intheeconomy. ThissectionprovidesabriefoverviewoftheperceptionsofMMFrisksandhow theseviewshavebeenshapedbythecrisesthatIstudyinthispaper. 3.1 PerspectivesonMMFriskspriortothefinancialturmoilin2007 The importance of governing MMF portfolio risks always has been central to money fund regulation, operations, and monitoring. The provisions of the SEC’s rule 2a-7, which applies only 4Toderiveasingleobservationforafundwithmultipleshareclasses,Icumulatetheassetsforallofitsshareclasses andcomputeasset-weightedmeasuresofotherfundcharacteristics(forexample,expenseratios)thatvaryacrossshare classes. Formoneymarketfundswithbothretailandinstitutionalshareclasses,Icreatedtwofundobservationsfor eachmonthbycumulatingtheretailandinstitutionalshareclassesseparately. 5

to MMFs and hence distinguishes them from other types of mutual funds, are primarily limitationsonMMFs’credit,interest-rate,andliquidityrisksconsistentwiththefunds’maintenanceof a stable NAV. As financial innovation has introduced new risks to MMFs, the SEC has tightened portfolio restrictions. For example, the Commission adopted new diversification requirements, new limits on holdings of lower-rated (“second-tier”) assets, and a more stringent WAM limit in 1991, and it issued guidance to clarify that certain derivative instruments were inappropriate for MMFportfoliosin1994. MMFsthatreceive“AAA”(orequivalent)ratingsfromratingsorganizations must satisfy additional restrictions, beside those in rule 2a-7, that are primarily constraints onportfoliorisks(see,forexample,Moody’sInvestorsService,2005;Standard&Poor’s,2007;and FitchRatings,2009). When, despite the risk-limiting provisions of rule 2a-7, MMF assets have lost value, fund sponsorshistoricallyhaveabsorbedthelossestopreventfunds’NAVsfromdeclining. Forexample,sponsorspickeduplossesonthreecommercialpaperissuesin1989and1990,and,in1994,25 MMFsponsorsintervenedtosupporttheirfundswhenajumpinshort-terminterestratescaused assets held by the funds to lose value (Investment Company Institute, 2009, pp. 175-176). Nothing required these sponsors to provide support, but because allowing a fund to break the buck wouldhavebeendestructivetoasponsor’sreputationandfranchise,sponsorsbackstoppedtheir funds voluntarily. Asset managers’ incentive to preserve reputation and their record of support for MMFs led industry observers to view sponsors as a source of stability—not risk—for MMFs andtoregardsponsorsupportasalmostagiven. Stigum’sMoneyMarket,aclassicreferenceguide tomoneymarkets,notesthat“...amoneyfundrunbyanentitywithdeeppockets,whileitmay nothavefederalinsurance,certainlyhassomethingakintoprivateinsurance...[and]thatinsuranceislikelytoproveadequatetocoveranylossessustainedbythefund”(StigumandCrescenzi, 2007,p. 1117). Because of the relative safety of their portfolios and sponsors’ practice of absorbing losses whentheyhaveoccurred,MMFsareusuallyrecipientsofflight-to-qualityinflowsduringperiods ofhighuncertaintyandmarketturmoil. Forexample, Babaetal.(2009)showedthatMMFstypically attracted inflows when stock market implied volatility, as measured by the VIX, was high.5 The largest weekly inflow to MMFs (as a fraction of assets) recorded by ICI in the past decade came in the week following the September 11, 2001 terrorist attacks, when MMF assets grew 3.8 percent. PerhapsbecauseMMFshistoricallyhavebeensosafe, MMFriskshaveattractedonlylimitedattentionintheacademicliteratureonMMFs,particularlybeforethefinancialturmoilerupted in 2007. Portfolio risks have garnered most of the discussion, but even these have often been viewed as tangential or benign. Domian (1992) and DeGennaro and Domian (1996) found that MMFportfoliomanagerscouldnotearnextra-normalreturnsbyanticipatinginterest-ratechanges, 5More generally, Carpenter and Lange (2003) found that heightened equity market volatility boosted total M2 assets,whichincluderetailMMFs. 6

andDeGennaroandDomianconcludedthat“MMFsingeneralcannotoutperformsimilarinvestments without taking commensurately higher risks.” Although they added that they could not ruleoutthepossibilitythatindividualmanagersmightoutperformthemarket,theydidnotstudy risktaking. DomianandReichenstein(1998),analyzingthevariationinMMFs’netyields,emphasizedtheimportanceoftheexpenseratioschargedbythefunds,ratherthantheriskandreturnof MMFportfolioassets. Theseauthorsconcludedthat, controllingforinvestmentobjective, MMFs withmorethan$300millioninassets“areessentiallycommodities”withlittlemeaningfuldifferentiation in portfolios. Advocates of MMFs as “narrow banks” have pointed to the transparency and liquidity of MMF portfolios as evidence that runs on the funds should be unlikely (Gorton andPennacchi,1993;Scott,1998). However, MMF portfolio risks were central in Collins and Mack (1994), who examined changesinaggregateMMFportfolioriskinaneventstudyoftheSEC’sadoptionin1991ofamendmentstorule2a-7thatfurtherrestrictedfunds’exposurestocreditandinterest-raterisks. Collins and Mack estimated the time-series relationship between funds’ yields and market risk premiumsandconcludedthatrisk-takingdidindeeddecline. Koppenhaver(1999)examinedthecrosssectionofMMFnetandgrossyieldsandfoundthatbothvariedpositivelywithMMFs’holdings ofagencysecuritiesandcommercialpaperandwithfunds’WAMs. WithMMFstypicallyattractinginflowsduringcrises,investorrisk—theriskreflectingthe compositionofafund’sshareholdersandthelikelihoodthattheywillredeemsharessuddenly— doesnotappeartohavebeenaconcernamongacademicians. Indeed,researchfocusingonMMFs’ suitability as “narrow banks” suggested that MMF investors’ behavior might mitigate broader risks. Gorton and Pennacchi (1993) noted that MMF investors did not run from funds followingcommercialpaperdefaults, Scott(1998)emphasizedthatthetransparencyofMMFportfolios would prevent an “ignorance-driven panic,” and Miles (2001) showed that, following monetary policy shocks, MMFs attracted net inflows and increased lending to private borrowers. Miles concluded that MMFs “are, if anything, perceived as safer than commercial banks and certainly less risky than smaller depository institutions.” Similarly, Pennacchi (2006) showed that MMFs attracted net inflows following commercial paper-Treasury bill spread increases (which he interpreted as liquidity shocks) and used this evidence to argue that a system of insurance for MMF shares,ratherthanbankdeposits,mightimprovemoneymarketliquidityfollowingsuchashock. 3.2 TheABCPcrisisin2007 The ABCP crisis, which caused substantial losses for many MMFs, nonetheless appears to have reinforcedpriornotionsaboutMMFsafetyandthestabilizingeffectofsponsorsupport. Thecrisis beganinearnestonAugust6,2007,whentheAmericanHomeMortgageInvestmentCorporation filedforbankruptcyanditsBroadhollowABCPprogramexercisedtheoptiontoextenditsmaturity. AnotherABCPprogram, OttimoFundingLtd., wasunabletorollitspaperonthesameday 7

andbegantoliquidate(Standard&Poor’s,2008b). Amidheightenedconcernsaboutexposuresof someABCPtodistressedsecurities,includingsubprimemortgages,manyABCPprogramsbegan toencounterdifficultiesrollingtheirpaper,andtwoABCPprogramsdefaultedinAugust. Covitz et al. (2009) show evidence that an indiscriminate run on ABCP began that month, as investors becameunwillingtopurchasereissuedpaperregardlessofprogramcharacteristics. As large investors in the ABCP market, MMFs were vulnerable to the credit and liquidity risksposedbydistressedprograms,andmanyMMFportfolioholdingslostvalue. However,MMF sponsorsabsorbedtheselossesbypurchasingsecuritiesoutoftheirfundsatabove-marketprices and by entering capital support agreements to guarantee securities still in the funds. The SEC reportedthatsponsorsintervenedtosupportatleast44MMFsbecauseofexposurestodistressed ABCP,andnoMMFbrokethebuck(U.S.SecuritiesandExchangeCommission,2009a,note38). Sponsors’ actions evidently allayed investors’ concerns; despite the exposures of many MMFs to troubled ABCP, MMF investors responded with only a modest pullback from prime MMFsinAugust2007. Asshownbythesolidblacklineinfigure1,panelA,primeMMFs,which mainlyinvestinprivatedebtinstrumentssuchasABCP,sawonlyverysmallnetoutflows(about $14billion,or0.8percentofassets)inthethreeweeksendingAugust29,2007.6 Tobesure,some individualfundssufferedmoresignificantoutflows;theWallStreetJournalreportedthatonevery largeCreditSuisseMMFsawredemptionsofmorethanhalfitsassetsbeginninginAugust2007, and the authors cited the fund’s exposure to distressed ABCP and its unstable investor base as factorsbehindthelargeoutflows(SmithandLauricella,2007). Butsection6.1ofthispapershows that net flows of individual funds during this episode generally did not vary significantly with proxies for MMF portfolio risk. Surprisingly, given the role of sponsor support in shoring up MMFsduringthiscrisis,netflowsalsodidnotvarysignificantlywithsponsorrisk. Inanycase,primefunds’smallaggregateoutflowswereshort-lived,asassetsbegangrowingrapidlyinSeptember2007. AsillustratedinpanelAoffigure1, primeMMFassetsincreased $425billion(24percent)fromAugust2007toAugust2008,asinvestorsmovedcashintovehicles that were seen as safe.7 The poor performance of alternative cash-management vehicles, such as “enhanced cash” funds and auction-rate securities, likely boosted MMF inflows over this period (seeInvestmentCompanyInstitute,2009,pp. 46-50andBabaetal.,2009). One MMF sponsor that appears to have exploited investors’ confidence in MMFs and the funds’growthfollowingtheABCPcrisiswasReserveManagement,Inc.,whichoperatedtheReserve Primary Fund. Reserve’s cofounders introduced the first MMF in 1970 and for decades reportedly maintained a relatively conservative investment policy, even among MMFs (Nocera, 1994; U.S.SecuritiesandExchangeCommission,2009b). Asshownbythethinsolidblacklinein 6MorenotablethantheoutflowsfromprimeMMFsinAugust2007werethesimultaneousflight-to-qualityinflows togovernment-onlyMMFs,which—asillustratedbythedashedbluelineinpanelA—grew$115billion(23percent)in thethreeweeksfollowingtheAmericanHomeMortgagebankruptcy. 7Government-onlyMMFsgrewevenfasteroverthisperiod;theirassetsexpaned46percent. 8

panelBoffigure1,inthedecadebeforemid-2007,thePrimaryFundtypicallyearnedagrossyield belowthatofitsaveragecompetitor. ButtheICIhasdocumentedachangeinReserve’sportfolio management beginning in the summer of 2007, when the Primary Fund began purchasing commercial paper—including Lehman debt securities—and its gross yield jumped relative to that of its peers (Investment Company Institute, 2009, pp. 53-57; see also U.S. Securities and Exchange Commission,2009b). Netyield,thedashedblackline,shotup,too,andinvestorstooknotice: Between August 2007 and August 2008, the Primary Fund’s market share surged as its assets more thantripled(panelsBandC). 3.3 TherunonMMFsin2008 The Reserve Primary Fund was holding Lehman Brothers debt valued at $785 million (at amortized cost) when Lehman declared bankruptcy early on Monday September 15, 2008. Losses on this exposure caused the Primary Fund to break the buck. The run that followed underlined the importanceofallthreetypesofMMFrisks: portfolio,investor,andsponsorrisks. Clearly, portfolio risk was central to Reserve’s demise. Moreover, aggregate flows to different types of MMFs reflected broader concerns about portfolio risks. As shown in panel A of figure1,nearlyallofthemassivenetredemptionsfromMMFsduringtherunin2008camefrom prime MMFs (the black line), which—unlike government-only and tax-exempt funds—mainly held the debt instruments of private issuers such as Lehman Brothers. Prime funds’ aggregate assetsdropped$450billion(21percent)overthefourweeksbeginningonSeptember10. Amida generalflighttoquality,government-onlyfunds(thebluedashedline)attractedverylargeinflows overthesameperiod. Tax-exemptfunds(thegreendottedline)hadmoderatenetoutflows. Inaddition,thestrikingdistinctionbetweeninstitutionalandretailinvestors’behaviorduringthecrisishighlightedtheimportanceofinvestorrisks. AsofSeptember10,2008,institutional fundsaccountedfor63percent($2.17trillion)oftotalassetsundermanagementinMMFsand63 percentofprimeMMFassets($1.32trillion). Asillustratedinfigure2,panelA,primefundsmarketedtoinstitutionalinvestorssufferedthebruntoftherun,withnetredemptionsof$410billion (30percentofassetsundermanagement)inthefourweeksbeginningSeptember10. PrimeMMFs forretailinvestorssawoutflowsofjust$40billion(5percentofassets)overthesameperiod. Sponsor risk also played a critical role in the run. The Primary Fund was not unique in holding Lehman’s debt at the time of its bankruptcy, but other MMFs with Lehman exposures obtained sufficient sponsor support to avoid breaking the buck (Investment Company Institute, 2009,pp. 60-62). WhatdistinguishedthePrimaryFundwastheinabilityofReservetoprovidethe capitalneeded toabsorbthe MMF’slosses.8 Indeed, sponsor supportburgeonedduring therun: 8ThecrucialimportanceofsponsorsupportapparentlywasnotlostonReserve’sprincipals,who,accordingtoan SECcomplaintonthematter,falselyassuredinvestors,ratingsorganizations,andthepressimmediatelyafterLehman’s bankruptcyonSeptember15thatsponsorsupportwouldprotectthePrimaryFund’s$1NAV.Suchsupportwasnot forthcoming,however,andtheseassurancesultimatelyfailedtopreventmassiveredemptionsrequests. Reservean- 9

TheSECreportedthatalmost20percentofallMMFsreceivedsupportinSeptemberandOctober 2008(U.S.SecuritiesandExchangeCommission,2009a,p. 20). Reserve’s inability to support its Primary Fund after the Lehman Brothers bankruptcy clearly undermined investor confidence in sponsor support. Panel B of figure 2 depicts daily net flowstoprimefundsduringthecrisis.9 Heavyredemptions—atleastfrominstitutionalMMFs— began immediately following Lehman’s bankruptcy on September 15, but outflows surged after Reserve’sannouncementlateonSeptember16thatthePrimaryFundhadbrokenthebuck. The run hit the MMF industry unevenly. As noted above, outflows were concentrated in institutional prime funds, but even among these funds, there was much cross-sectional disparity innetflows. TheredbarsinpanelCoffigure2showthedistributionofoutflowsfrominstitutional prime MMFs from September 9 to October 7. (Retail fund flows, the blue bars, exhibited much lessdispersion.) Ofthe116institutionalprimefundsinmysample,tensawassetsshrinkbymore than 50 percent during this period and another twelve lost between 40 and 50 percent of assets, but 20 had net inflows over the same period. This heterogeneity provides an opportunity to test hypothesesaboutthelinksbetweenindividualMMFs’risksandoutcomesduringthecrisis. 3.4 Post-crisisperspectives AllthreetypesofMMFriskhaveattractednewattentionsincethebeginningofthefinancialturmoil in 2007. Researchers and policymakers have focused on portfolio risk, in particular. Jank andWedow(2008)examinedtheperformanceofGermanMMFsbeforeandduringthesubprime crisisthatbeganin2007.10 TheyfoundthatMMFsthattookonmoreportfolioriskearnedhigher returns than their less risky counterparts during periods of normal money market liquidity but that the riskier MMFs fared relatively worse during periods of low liquidity. As noted in section 3.2,ICIdocumentedasubstantialincreaseintheportfolioriskoftheReservePrimaryFundbeginninginmid-2007,justoverayearbeforethatfundbrokethebuck(InvestmentCompanyInstitute, 2009). In a study of the effectiveness of the Federal Reserve’s Asset-Backed Commercial Paper Money Market Mutual Fund Liquidity Facility (AMLF), Duygan-Bump et al. (2010) found that MMFs with greater ABCP exposures initially suffered larger outflows during the run in 2008 but recovered quickly with the announcement of the AMLF. Furthermore, the SEC in 2010 adopted amendmentstorule2a-7thatimposefurtherconstraintsonportfoliorisk,includingnewliquidity nouncedthatthePrimaryFundhadbrokenthebuckat4pmonTuesday,September16(U.S.SecuritiesandExchange Commission,2009b). 9Net flows depicted in this panel reflect my adjustments to daily assets data for several MMFs. I describe the adjustmentsinsectionA.1oftheappendix. 10GermanMMFs,unlikeUSfunds,reportmark-to-market(accumulating)NAVs,soindividualMMFreturnswerea usefulmeasureofoutcomesduringthecrisis.Asaproxyforportfoliorisk,JankandWedowusedeachfund’sshareof assetsheldin“debtsecurities,”aresidualcategorythatexcludesGermanTreasurybills,bankdeposits,CP,and“other assets.” (TheauthorsarguethattheshareofdebtsecuritiesinanMMF’sportfolioisameasureoftheMMF’sliquidity risk,butthismeasurelikelycomprisesbroaderportfoliorisks,includingcreditrisk.) 10

requirementsforMMFassets,“tomakemoneymarketfundsmoreresilientandlesslikelytobreak abuckasaresultofdisruptionssuchasthosethatoccurredinthefallof2008”(U.S.Securitiesand ExchangeCommission,2010). Investor risk also has received greater attention, especially in the aftermath of the run in 2008. Disparities in outflows from institutional and retail funds prompted the SEC to propose newliquidityrequirementsforMMFsthatwouldhavebeenmorestringentforinstitutionalfunds than for retail funds.11 In addition, ratings organizations, the SEC, and the ICI have highlighted theinvestorrisksarisingfrom“hotmoney”thatflowsintoandoutofMMFsinresponsetoeven small yield changes (Fitch Ratings, 2009; U.S. Securities and Exchange Commission, 2009a, pp. 56, 66-68; Investment Company Institute, 2009, pp. 83-84). For example, investors’ growing use of MMF portals, which facilitate comparisons of MMF yields and the transfer of monies from fund to fund, may have heightened investor risks (Baba et al., 2009). Fitch’s MMF rating criteria, publishedayearaftertherun,notethatafundwith“anover-relianceonhotmoneysourcessuch asportalsmayneedanadditionalliquiditybuffer”(FitchRatings,2009). TheMMFcrisesatissueinthispaperhavealsobroughtanincreasedawarenessoftheimportance of sponsor risk. Baba et al. (2009) reported that during the run in 2008, “[t]he largest redemptionsoccurredatinstitutionalprimefundsmanagedbytheremainingsecuritiesfirmsand small independent managers, which investors doubted could support their funds.” In assigning MMF ratings, ratings agencies began focusing more specifically on sponsors’ resources and abilities to support their MMFs (Moody’s Investors Service, 2008; Fitch Ratings, 2009; Standard & Poor’s, 2007, 2008a). The SEC’s new rules for MMFs, adopted in 2010, facilitate sponsor support forfundsduringemergencies(U.S.SecuritiesandExchangeCommission,2010,p. 94-97).12 4 Moneymarketfundrisksandproxies Prime MMFs are financial intermediaries that provide both maturity and credit transformation. Thefundsholdriskyassetsthatmaymatureinayearormorebutissuesharesthatareredeemable 11TheSECproposednewrequirementsforholdingsof“dailyliquidassets”and“weeklyliquidassets”andsuggested that the liquidity requirements for institutional funds be double those for retail funds. The Commission’s rationaleforthedistinctionbetweeninstitutionalandretailfundswasbasedonthegreaterhistoricalliquidityneedsof institutionalfunds,asreflected,forexample,intheirhigherflowvolatilityandlargeroutflowsduringtherunin2008 (seeU.S.SecuritiesandExchangeCommission,2009a,pp. 55-65). TheSECultimatelyadoptednewliquidityrequirementsforMMFsthatwerethesameforinstitutionalandretailfunds,butincludeda“generalliquidityrequirement” that an MMF “hold securities that are sufficiently liquid to meet reasonably foreseeable shareholder redemptions,” whichpresumablywouldcauseinstitutionalfundstoholdmoreliquidassetsonaveragethanretailfunds(U.S.SecuritiesandExchangeCommission,2010,pp.51-54,56-67). 12Specifically,thenewrulesallowanMMFsponsortopurchasefromafundsecuritiesthathavedefaultedorare “distressed,” but which are still “eligible” as MMF portfolio holdings under rule 2a-7. Purchases of such securities aregenerallyrestrictedundersection17oftheICAbecauseofconcernsthatatransactionbetweenanMMFandan affiliatedpartymightnotbeintheinterestsofthefund. Inthepast,theSEChadroutinelyprovidedno-actionrelief forMMFsponsorsthatpurchasedsuchsecuritiesfromMMFstoprotectthefunds,butthenewruleswereintendedto preventdelaysinsponsorsupportbyallowingasponsortointervenewithoutgettingno-actionrelieffirst. 11

ondemand,andMMFshareshistoricallyhavebeensaferthanmanyoftheassetsthefundshold, even though MMFs (unlike some other financial intermediaries) do not have capital cushions or insurancetoguaranteethestablevalueoftheirliabilities. Instead,MMFs’recordofsafetyreflects regulatory constraints on portfolio composition, conservative asset management, broad investor bases,andfinancialsupportfromsponsors. Yet,noneoftheseisfail-safe,soMMFsaresubjectto theportfolio, investor, andsponsorrisksthatarethefocusofthispaper. Inthissection, Idiscuss theserisksanddescribeproxiesthatIusefortheminmyempiricalanalyses. 4.1 Portfoliorisk Portfolio risks comprise credit, liquidity, and interest-rate risks, all of which have played significantrolesinpastepisodesofMMFstress. Creditrisksthatturnedoutbadlytriggeredsubstantial sponsorsupporttoabsorbMMFlossesin1989and1990aswellasduringtheABCPcrisisnearly two decades later, and Reserve’s poor management of credit risks in the Primary Fund’s portfolio helped to start the run on MMFs in 2008. Liquidity risks came into the spotlight during the run, as MMFs sought to sell securities to meet redemption requests but encountered unreceptive secondary markets for instruments such as term commercial paper—in part, because the dealers whotypicallymakethosemarketswerethemselvesshortofliquidity. AtleastacoupleofMMFs thatreportedlyhadnoexposurestoLehmanBrothersorotherdistressedissuersnonethelesshad totakeextraordinarysteps(oneclosedandtheotherbeganpayingredemptionsin-kind)because sales of securities into illiquid markets would have caused capital losses (Investment Company Institute, 2009, p. 62). Finally, interest-rate risk was responsible for the first instance of an MMF breaking the buck, in 1994, when the Community Bankers Mutual Fund, a small institutional MMF,incurredlossesonasizablepositionininterest-sensitivestructurednotesasratesrosesuddenly (U.S. Securities and Exchange Commission, 1998b).13 Although all three types of portfolio riskareimportantinassessingMMFrisks,theserisksblurduringcrises,whenunusualdiscounts on securities (and hence, higher interest rates) may reflect impaired credit quality, impaired liquidity,orboth. Thus,Igenerallydonotattempttodistinguishamongdifferenttypesofportfolio risksinthispaper. 4.1.1 Grossyieldasaproxyforportfoliorisk MMFs’ history of maintaining stable NAVs precludes the use of price variation as a measure of portfolio risk. However, given the typical relationship between risk and return, one alternative proxy for a fund’s portfolio risk is the gross yield that it earns on its assets. MMF gross yields do not vary much, in part because of rule 2a-7’s constraints on fund portfolios. As shown in the second column of table 1, the cross-sectional standard deviation of annual gross yields for prime 13Poorlymanagedinterest-rateriskalsocausedanMMFtobreakthebuckin1977(GortonandPennacchi,1993, note13),butthiseventoccurredbeforetheSECadoptedrule2a-7. 12

funds (line 1) was only 12 basis points in the five years from 2004 to 2008. The much larger 29 basispointcross-sectionalstandarddeviationinthenetyieldsthatthefundspaidtoshareholders (line2)primarilyreflectedvariationinexpenseratios(line3).14 Even so, gross yields at prime funds varied with the composition and maturity of their assetsinwaysthatsuggestthatgrossyieldsareausefulproxyforportfoliorisk. SectionA.2ofthe appendix shows that, over a sample period from 2004 to 2008, holdings of relatively safe assets, suchasTreasuryandagencysecuritiesandrepurchaseagreements,tendedtoreduceMMFgross yields. At the same time, larger holdings of riskier assets, such as ABCP, floating-rate notes, and otherbankobligations,boostedfunds’grossyields. LongerWAMalsoincreasedgrossyields. Why would MMF sponsors take on greater portfolio risk? Higher portfolio risks increase the likelihood that an MMF will experience capital losses and shortfalls in liquidity, but—for a given level of fees charged—a higher gross yield attracts cash inflows to a fund and boosts its expected future fee revenue (see, for example, Christoffersen, 2001; Christoffersen and Musto, 2002; Jank and Wedow, 2008).15 In section A.3 of the appendix, I analyze the effects of gross yields on subsequent net flows and find that, over a sample period extending from January 1997 to August 2008, an institutional fund that, all else equal, maintained for a year a gross yield that was one standard deviation above average would have expected to attract net flows that were 6 percentage points (of assets) larger than average. For retail MMFs, the analogous effect was a smaller 2 percentage point gain in net flows.16 How MMF sponsors and portfolio managers chose to balance the potential costs and benefits of greater portfolio risk is beyond the scope of this paper, but there is evidence that some sponsors—including, notably, Reserve Management Company—electedtotakeongreaterriskthantheirpeersinordertoattractgreaterinflows. Tables2aand2blistmeans, standarddeviations, andthe10thand90thpercentilesforthe risk proxies used in my empirical analyses. As shown on line 1 of table 2a, the distributions of grossyieldsforinstitutionalandretailfundsintheyearpriortotherunin2008wereverysimilar.17 Gross yields for both types averaged 3.88 percent (columns 1 and 5), and standard deviations (columns 2 and 6) were also similar. (Statistics for expense ratios, which appear on line 4 of the 14The fact that variation in MMFs’ net yields was primarily due to variation in expense ratios, rather than gross yields,promptedDomianandReichenstein’s(1998)conclusionthatMMFs“areessentiallycommodities.” 15Therelationshipbetweenmutualfundperformanceandsubsequentinflowsisthesubjectofanextensiveliterature that mostly analyzes long-term mutual funds (see, for example, Wharton School, University of Pennsylvania, 1962; Friend et al., 1970; Smith, 1978; Ippolito, 1992; Chevalier and Ellison, 1997; Sirri and Tufano, 1998; Del Guercio and Tkac,2002). 16Becausehighergrossyieldsattractnewcashfromperformance-sensitiveinvestors,afund’sgrossyieldmayalso reflect investor risk. However, the empirical evidence suggests that gross yield, especially leading up to the run in 2008,waspredominantlyanindicatorofportfoliorisk. Simplecorrelationsbetweengrossyieldandseveralmeasures ofinvestorriskareallinsignificant.Moreover,myanalysesincludemultiplecontrolsforinvestorrisk.Seesections4.4 and5.2. 17Grossyields,likeseveralotherriskproxiesreportedontables2aand2a,arecomputedovertheyearendingjust priortotheonsetofeachcrisis.Ayear’sdataareusefulincomputingseveraloftheseproxies(suchasthesensitivityof flowtoyield,line7),andmayhelpsmooththroughsomeoftheseasonalvariationinMMFyieldsandflowsdescribed inFarinellaandKoch(2000). 13

table,indicatethatretailMMFschargedmuchhigherfeesthaninstitutionalfunds,soretailfunds earnedsmallernetyields.) Asshownincolumns1and5onthefirstlineoftable2b, grossyields for both institutional and retail funds averaged 5.37 percent in the year before the ABCP crisis. ThedistributionofgrossyieldsforMMFswasverytightduringthisperiod: Standarddeviations forinstitutionalandretailMMFswerejust2and3basispoints,respectively(columns2and6). 4.1.2 MMFratingsasaproxyforportfoliorisk Many MMF carry “AAA” or similar top ratings from ratings organizations, particularly S&P, Moody’s,andFitch. Forexample,Fitch’s“AAAmmf”ratingrepresentsitsjudgmentthatanMMF has “[e]xtremely strong capacity to achieve fund’s investment objective of preserving principal and providing shareholder liquidity through limiting credit, market, and liquidity risk” (Fitch Ratings, 2009). Thus, AAA ratings may be viewed as rating agency opinions that MMF portfolio risks are relatively low.18 I include, as a binary proxy for (low) portfolio risk, the existence of a AAAratingfromatleastoneofthethreemajorratingsorganizations. Asshownonline2oftable 2a,53percentofinstitutionalMMFshadtriple-Aratingsontheeveoftherunin2008,butonly18 percentofretailfundshadsuchratings. 4.1.3 Controllingforportfolioholdingsofspecificdistressedsecurities Money funds’ exposures to several specific distressed securities, such as Lehman Brothers’ obligations and troubled ABCP issues, played central roles in the two MMF crises that I study in this paper. Past holdings of problematic securities are not promising predictors of future MMF strains—funds are, for example, unlikely to begin purchasing Lehman paper again. However, controllingforsuchexposuresmaybeimportantinunderstandingthecross-sectionalvariationin outcomes(particularlynetflows)ofMMFsduringpastcrises. ThecriticalroleoftheLehmanBrothersbankruptcyintherunin2008motivatescontrolling for MMFs’ holdings of Lehman debt in analyzing redemptions during that crisis. To do so, I use the sources described in section 2 to identify MMFs that held Lehman’s obligations at the time of its bankruptcy. As shown on line 3 of table 2a, those sources indicate that 12 percent of both institutionalfunds(column1)andretailfunds(column5)heldLehmandebtontheeveoftherun. MultipleABCPissuescausedproblemsforMMFsduringtheABCPcrisis. Asabroadproxy for exposures to distressed securities in this episode, I use an indicator variable equal to one for 18Increasingly, ratings organizations are considering investor and sponsor risks in assigning ratings to MMFs (Moody’sInvestorsService,2008;FitchRatings,2009). Thus,particularlyintheaftermathofthe2008run,topMMF ratings may be viewed as opinions on each MMF’s combination of portfolio, investor, and sponsor risks, although portfoliorisksappeartoremaincentraltotheratingsorganizations’analysisofMMFs. Fortheperiodcoveredbymy analysis,however,ratingsagenciesseemtohavefocusedalmostexclusivelyonportfoliorisks.(Theexactnatureofthe risksanalyzedbytheagenciesmatterslittlefortheinterpretationofmyresults,asIfindthatAAAratingsgenerally hadlittlepredictivepowerfortheoutcomesforMMFsduringthetwocrisesthatIstudy.) 14

MMFs that held ABCP issues that ultimately prompted sponsor-support actions for at least one MMF.Line3oftable2bshowsthat57percentofinstitutionalfunds(column1)and44percentof retailfunds(column5)heldproblematicpaperasthecrisisunfolded. 4.1.4 Alternativeproxiesforportfoliorisk Grossyieldsmaybemorethancompensationforrisk;ahighergrossyieldmayreflectaportfolio manager’s ability to obtain superior risk-adjusted yield. One approach to capturing portfolio risk moredirectlywouldbetouseportfoliocharacteristics,suchasWAMandsharesofassetsheldin TreasurysecuritiesandABCP(andothercharacteristicsdiscussedinsectionA.2oftheappendix) as proxies. Indeed, Jank and Wedow (2008) used the share of individual German MMFs’ assets in oneapparently riskyasset classas a proxy for thesefunds’ liquidityrisk. Instead ofanalyzing everyconceivablepermutationoffundcharacteristicsinmyanalyses,Iemploy—asanalternative to my main specification—the fund portfolio measures in an instrumental variables framework. Thatis,Iuseportfoliocharacteristicstoinstrumentforgrossyieldsandhencecapture,asaproxy for portfolio risk, the component of gross yields that is measured by portfolio attributes. Section 5.2discussesthisapproachinmoredetail. Another approach to measuring portfolio risk in MMFs would be to estimate risk-factor loadingsusinganarbitrage-pricingmodelappliedtoMMFgrossyields. CollinsandMack(1994) used such an approach in an event-study analysis of changes in the aggregate riskiness of MMFs followingtheadoptionofamendmentstorule2a-7in1991. Onehurdletousingthisapproachto pinpoint cross-sectional MMF risks, however, is that amortized-cost accounting prevents funds’ yields from fluctuating directly with the market yields of the assets they hold, so MMFs’ factor loadings would be biased down (and their “alphas” biased up). Even if yields did move with marketrates, theriskinessofindividualfundsthatinvestedinmoreexoticinstrumentsmightbe understated by loadings on conventional market risk factors. Hence, an arbitrage-pricing frameworkisprobablynotwellsuitedtostudyofcross-sectionalvariationinrisksofstable-NAVMMFs. 4.2 Investorrisk19 As figure 2 illustrates, institutional and retail shareholders subjected MMFs to starkly different degrees of investor risk during the run in 2008. But why? Compared with their retail counterparts, many institutional MMF shareholders may be particularly risk-averse and, at the same time, possessgreatersophisticationandresourcestomonitorMMFscarefullyandredeemshares preemptivelyatthefirstsignofanyheightenedrisk. Institutionalinvestorsmayfaceverystrong 19Broadlyspeaking,investorriskisaformofliquidityrisk,whichalsoincludesportfoliorisksarisingfromtheilliquidityof(some)portfolioassets. However, Iusetheterm“investorrisk”tohighlighttheliquidityrisksattributable specificallytoafund’sinvestors,andIemploy“portfoliorisk”toencompasstheliquidityrisksarisingfromtheilliquidityofassets. Thisterminologynotonlyeasesexpositionbutalsoreflectsthedifficultyofdistinguishingportfolio liquidityrisksfromotherportfoliorisks,particularlyduringcrises. 15

incentives to avoid losses on MMF shares; capital losses in client accounts may expose institutionalinvestorswithfiduciaryresponsibilitiestolegalliability,andlossesmayjeopardizecareers of corporate treasurers. Some institutional investors are precluded by law, regulation, or policy from investing in funds without stable NAVs (Investment Company Institute, 2009), so any risk ofNAVfluctuationsmayprompttheseinvestorstoredeemMMFshares. Sophisticated,risk-averseinvestorsalsomaybeparticularlyattunedtodifferencesbetween MMFshareprices,whichareroundedtothenearestpenny,andtheirmark-to-marketvalues(Standard & Poor’s, 2007, pp. 27). Rounded NAVs help MMFs maintain stable NAVs, but they also create destabilizing arbitrage opportunities for shareholders when a fund suffers a small capital loss (less than 0.5 percent of assets) that is rounded away. When redeeming investors receive $1 forsharesthatareworthless,themark-to-marketvalueofremainingsharesdeclinesaslossesare concentrated among fewer shares, and the fund moves closer to breaking the buck. So, any loss (orexpectedloss)inarounded-NAVfundtriggersanincentiveforshareholderstoredeembefore othersdo. NetflowstoMMFsfollowingchangesinshort-terminterestrates—whichaffectthemarkto-market value of MMF shares but not their $1 share prices—indicate that some investors, particularly institutional investors, exploit even small discrepancies between prices and values. For example, Lyon (1984) estimated that arbitrage flows reduced the net yields earned by passive investors in institutional MMFs by about 10 basis points annually. Monetary policy easing cycles present numerous opportunities for investors to take advantage of MMFs’ rounded prices, and theresultingarbitrageflowshaveoccurredmostlyininstitutionalfunds.20 Sophisticatedinvestors whoaremorepronetoexploitshare-priceinefficienciesprobablysubjectfundstoheightenedinvestorriskduringcrises. The relatively higher degree of investor risk in institutional funds may also reflect greater flow volatility in these funds and a greater responsiveness of cash flows to net yield (see section A.3). That is, institutional funds may have relatively larger shares of “hot money” than retail MMFs. Asnotedinsection3.4,intheaftermathoftherunin2008,regulatorsandratingsorganizationshaveidentified“hotmoney”asasourceofinvestorrisksformoneyfunds. Other than the distinction between institutional and retail funds, publicly available data provide little direct information about individual funds’ investors and the risks they represent. Below,IsuggestfourproxiesforthedegreeofinvestorriskinMMFs. 20Reductionsinthefederalfundsrate(FFR)targetthatraisethevalue(butnottheprice)ofMMFsharesmaybe especiallyeasytoexploitbecausetheprofitabletransactionisapurchaseofMMFshares(evenafterthepolicyaction), soinvestorsneednotholdsharesinadvancetoexploittheopportunity. OfthethirteenreductionsintheFFRtarget between2001and2003,thesecond-to-last,a50-basis-pointcutinthetargetonNovember6,2002,promptedthelargest inflows, perhaps because institutional investors had learned from others’ behavior as the easing cycle progressed. Assetsundermanagementinmoneyfunds,whichhadbeentrendingdownforalmostayearbeforethispolicyaction amid very low short-term interest rates, jumped $168 billion (8 percent of assets) in the three weeks following the reductioninthepolicyrate.Institutionalfundsattractedvirtuallyall($167billion)oftheinflow. 16

4.2.1 MMFexpenseratiosasaproxyforinvestorrisk One indicator of the sophistication of the investors in a money fund—and the risks of large redemptions should the fund encounter difficulties—is the fund’s expense ratio. Domian and Reichenstein (1998) found that the net yields of prime funds with more than $300 million in assets “are driven exclusively or almost exclusively by expenses,” and they warned that “expenses are a dead-weight loss to investors.” Still, someone must hold shares in high-expense MMFs, and ChristoffersenandMusto(2002)arguedthatsuchinvestors“havedistinguishedthemselvesfrom the population in general ...as relatively more willing to pay high prices for bad prospects.” Hortac¸suandSyverson(2004)arguedthatlowerexpenseratios(inequityfunds)attractmoresophisticatedinvestorswithlowersearchcostsandlessneedforcostlyservices. Inaddition, lower expense ratios may reflect lower shareholder servicing costs (per dollar of assets under management) that are associated with larger average account sizes. While holders of larger accounts are not necessarily more sophisticated than smaller shareholders, investors with more at stake in a fundmayhavegreaterincentivestomonitoritcarefully. Asshownonline4oftable2a,theaverageexpenseratioforinstitutionalfundsintheyear priortotherun(column1)waslessthanhalfofthatforretailfunds(column5). Atthesametime, there was wide dispersion in expense ratios for both types of funds (columns 2 through 4 and 5 through8). Line4oftable2bshowssummarystatisticsfortheyearbeforetheABCPcrisis. 4.2.2 MMFgrowth,flowvolatility,andyieldsensitivityasproxiesforinvestorrisk I use three proxies for the investor risks posed by MMFs’ exposure to “hot money” and other sources of flow volatility. The first is the growth of the fund over the previous year, which is probably only a crude indicator of the hot money in a fund, since many factors might contribute toitsgrowth. (Imeasuregrowthandflowsthroughoutthispaperintermsofthelogarithmofnet flows, that is, the natural log of the sum of one plus the ratio of net flow to lagged assets.21) The secondmeasureisafund’sweeklyflowvolatility(thestandarddeviationofweeklylognetflows forthepreviousyear),whichmayreflectafund’sexposurestohotmoneyandtoconcentrationsof similarinvestorswithcorrelatedtransactionspatterns. Thethird,thesensitivityofafund’sweekly netflowsinthepreviousyeartoitsownlaggednetyields,isdesignedtocapturethecomponent of a fund’s flow variability that is due to hot-money flows. For each fund, this measure is the estimated coefficient from a regression of the fund’s weekly log net flow on its relative net yield inthepreviousweek. Lines5through7oftable2alistsummarystatisticsforthethree“hotmoney”investorrisk proxiesfortheyearendingAugust2008. Allthreemeasureswereconsiderablylarger,onaverage, 21I multiply by 100 to make this measure similar to net flows as a percentage of lagged assets: Log net flow = 100×ln(1+ netflow ).Netflowisthechangeinassets,netofincreasesduetoaccrualofyield. laggedassets 17

for institutional funds (column 1) than for retail funds (column 5). Table 2b shows analogous summarystatisticsfortheyearendingJuly2007. 4.3 Sponsorrisk When the mark-to-market value of an MMF’s shares falls below 99.5 cents, the fund’s sponsor (that is, its adviser and the adviser’s affiliates or parent firm) has two choices: provide capital supporttobringthevaluebacktoatleast99.5cents,orallowthefundtorepriceitssharesbelow $1—thatis,breakthebuck.22 Sponsorsupportisexpresslyvoluntary,andasponsor’sabilityand commitment to support its funds is not officially monitored or regulated, so MMF investors can never be certain that a sponsor will support an ailing fund. Historically, however, sponsors with theresourcestosupporttheirfundshavedoneso,andevenassetmanagersthatsubsequentlyexitedtheMMFbusinesshavesupportedtheirmoneyfundstopreservetheirbroaderreputations.23 InbothinstancesinwhichMMFsdidbreakthebuck,thesponsorsimplylackedthewherewithal toabsorbthefund’slosses(Eaton,1994;U.S.SecuritiesandExchangeCommission,2009b)andits reputation as an investment adviser was destroyed. Hence, sponsors have gone to such lengths topreventpassingcapitallossesalongtoMMFshareholders,andsponsorsupporthascometobe seen as a form of “private insurance” (Stigum and Crescenzi, 2007). However, as is the case for other forms of private insurance, the surety of sponsor support for MMFs depends on the financialstrengthofthesponsor. Sponsorriskreflectsthepossibilitythatasponsorwillnotbeableto supportitsfunds. 4.3.1 Bankaffiliationasaproxyforsponsorrisk Asoneproxyforsponsorrisk,IuseadummyindicatingwhetheranMMF’ssponsorwasaffiliated with a bank. Bank affiliation presumably would have been associated with lower sponsor risk becausebankshaveaccesstolender-of-last-resortliquiditythroughtheFederalReserve’sdiscount window and may have greater resources for supporting affiliated funds during a crisis. Indeed, many MMFs received support from bank affiliates during the ABCP crisis in 2007 and the run in 2008, and some sizable non-bank firms obtained bank charters during the turmoil that followed 22TheMMF’sboardofdirectors—notitssponsor—ultimatelymustdeterminewhethertorepricesharesbelow$1. Thatdecision,accordingtorule2a-7,shouldbebasedontheboard’s“promptconsideration”ofwhether“theextent ofanydeviationfromthemoneymarketfund’samortizedcostpricepersharemayresultinmaterialdilutionorother unfairresultstoinvestorsorexistingshareholders.” However,thesponsormustdecidewhethertoprovidefinancial supporttopreventsuchadeviation. (SponsorsgenerallydonotcommitinadvancetosupportanMMFthatmight latersufferlosses,becauseanexplicitcommitmentwouldsubjectthesponsortopotentiallylargecontingentliabilities andlikelyforcethesponsortoconsolidatethefundonitsbalancesheet.) 23CreditSuissereportedatotalof$1.8billioninlossesrelatedtosupportofitsMMFsonitsincomestatementsin 2007and2008,eventhoughthefirmexitedtheMMFbusinessinlate2008(CreditSuisse,2008,pp.65-66;CreditSuisse, 2009,pp.27,45,434). 18

the Lehman bankruptcy, presumably to obtain access to the discount window.24 Indeed, Baba et al. (2009) reported that bank-affiliated MMFs experienced smaller-than-average net redemptions during the run. Line 8 of tables 2a and 2b show that about half of both institutional and retail primeMMFswereaffiliatedwithbanksattheonsetofeachcrisisepisode. 4.3.2 SponsorCDSspreadsasaproxyforsponsorrisk As a second proxy for sponsor risk, I use the five-year CDS spread of the sponsor’s senior debt securities. BecausethefinancialconditionofsomelargeinstitutionsthatsponsoredMMFsdeterioratedquicklybeforethebeginningoftherunin2008,Iuseeachsponsor’sCDSspreadsaveraged over the first week of September 2008. (Spreads during the week after the Lehman bankruptcy might be endogenous if MMF outflows worsened strains on sponsors). CDS spreads are only available for sponsors of 43 percent of prime MMFs (line 11 of table 2a). Mean spreads for sponsorsofbothinstitutionalandretailfundsaveragedabout140basispoints(line9). Boththeaverage CDS spread and the dispersion of spreads were much larger immediately before the run in 2008 thantheyhadbeenjustpriortotheonsetoftheABCPcrisis(table2b,line9). 4.4 Interactionsamongriskproxies Tables3aand3bprovidesimplecorrelationsfortheMMFriskproxiesusedinthispaper,asmeasuredbeforetherunin2008andpriortotheonsetoftheABCPcrisisin2007,respectively. PanelsA andBofeachtableshowcorrelationsforinstitutionalfundsandretailfunds,respectively. (Numberingofthevariablesintherowsandcolumnsofthesetablesisthesameasthatfortables2aand 2b). Forclarity,Ireportonlycorrelationsthataresignificantatthe5percentlevel. Thecorrelationspointtosomedistinctionsamongtheproxiesforportfoliorisk. Forexample, among institutional funds in the year before the run in 2008 (table 3a, panel A), none of the pairwisecorrelationsamongtheportfolio-riskproxies—grossyield,triple-Aratings,andLehman exposure (the first three lines and columns)—was significant. Intriguingly, table 3b shows that triple-A ratings for both institutional and retail funds were uncorrelated with distressed ABCP exposure during the ABCP crisis (line 3, column 2). However, MMFs with such exposures did exhibithighergrossyieldsintheyearbeforetheABCPcrisis(line3,column1). Theevidenceforlinksamongtheproxiesforinvestorriskwasalsomixed. Lowerexpense ratios in institutional funds were correlated with other measures of investor risk both before the run in 2008 (table 3a, lines 6 and 7, column 4) and before the ABCP crisis (table 3b, lines 5 and 6, column 4), but other pairwise correlations were generally not significant. The two proxies for 24In analyzing the run in September and October 2008, I treat MMF sponsors that obtained bank charters after Lehman’sbankruptcyasnon-bankentities,exceptintheweeklyregressionsdiscussedinsection5.3. (Inthoseregressions, a sponsor’s bank-affiliation is recorded as of the beginning of each week.) Relabeling sponsors that obtained bankchartersduringtherunasbank-affiliateddoesnotmateriallyaltermyresults. 19

sponsor risk, CDS spread and bank affiliation, were significantly negatively correlated only for institutionalMMFsintheyearbeforetheABCPcrisis. Some of the correlations among proxies for different types of risk are relevant to interpretationofmyresults. Forexample,intheyearpriortotherunin2008,MMFgrossyieldswerenot significantly correlated with three of the measures of investor risk and were negatively correlated withthestandarddeviationofweeklyflows(seebothpanelsoftable3a,lines4through7,column 1). Although higher gross yields may attract riskier investors, the correlations support the view thatgrossyieldduringthisepisodewasaproxyforportfoliorisk,ratherthaninvestorrisk. Thecorrelations ontable3a alsopreview thelinksbetween MMFrisksand outflowsfrom institutionalandretailfundsduringtherunin2008. Forexample,asshownonline12ofpanelA, several risk factors were associated with larger outflows from institutional funds during the run in 2008: higher gross yield (column 1), a lower expense ratio (column 4), greater flow sensitivity (column 7), a higher CDS spread (column 9), and larger size (column 10). Correlations between netflowsandrisksforretailfunds(panelB)werequitedifferentfromthoseforinstitutionalfunds. Thispointisexploredinmoredetailinsection5.1. The size of an MMF was an important predictor of outflows during the run in 2008 and, particularlyforinstitutionalfunds,wascorrelatedwithmultipleriskproxiesinbothepisodes(see line10andcolumn10inthetoppanelsoftables3aand3b). Forexample,inbothepisodes,larger institutional funds had lower expense ratios and thus probably faced greater investor risk from relatively sophisticated shareholders. Even controlling for expense ratios, larger, sophisticated investors may have been more likely to invest in bigger funds. Larger institutional funds were morelikelytohavehadtriple-Aratings,anindicatoroflowerportfoliorisk,butlargerfundsalso were more likely to have held Lehman debt in 2008 and distressed ABCP in 2007. In addition, fund size may have intensified concerns about sponsor risk because greater amounts of capital might have been needed to absorb losses in larger funds. Thus, links between size and multiple forms of MMF risk complicate the interpretation of the role of fund size in explaining outcomes duringcrises. 5 MMFrisksandthecrosssectionofoutflowsduringthe2008run ToanalyzeprimeMMFflowsduringtherunthatfollowedtheLehmanbankruptcy,Irunaseries of cross-sectional regressions in which the dependent variable is net flows from September 9 to October 7, 2008 (the Lehman bankruptcy occurred on September 15, and ICI aggregate data indicate that prime funds had large outflows through the first week of October). As discussed in section2,IusedatafromiMoneyNetforprimemoneyfundsthathadbeeninexistenceforatleast ayear,hadassetsexceeding$100million,andhadreliableflowsdataduringtherun. Thesample includes 116 institutional funds and 135 retail funds. Explanatory variables include the proxies 20

forfundrisks(andfundsize),aslistedontables2aand4: flowi = β gyieldi +β AAAi +β Lehmani run 1 t∈P 2 t=P¯ 3 t=P¯ +β exprati +β growthi +β flowSDi +β flowSensi 4 t∈P 5 t∈P 6 t∈P 7 t∈P +β banki +β CDSi +β assetsi +constant+εi. (1) 8 t=P¯ 9 t=P¯ 10 t=P¯ Here, flowi isfundi’slognetflowduringtherun, gyieldi isthegrossyieldoffundioverthe run t∈P year ending August 2008 (period P), and AAAi is a dummy variable equal to one if and only t=P¯ iffund i hada triple-Aratingfrom atleast oneratingsorganization asofthe endof August2008 (at time P¯). Other explanatory variables are listed on tables 2a and 4 in the same order that they appearinequation(1). Table4reportsresultsfromsixregressions. TheseincludeseparateestimationsforinstitutionalandretailMMFs,becausetheiroutflowdynamicsdifferedsostarklyduringtherun,aswell asselectedsignificanceresultsfromapooledregressionthatincludesbothinstitutionalandretail funds. BecauseCDSspreadsareavailableforsponsorsoflessthanhalfofthemoneyfundsinthe sample,Irunanextrasetofregressionstoincludethesespreads. 5.1 Empiricalresults Thefirstthreecolumnsoftable4summarizetheresultsofregressionsthatexcludeCDSspreads. Column1showsresultsforinstitutionalfunds. Line1showsthatgreaterportfoliorisk,asmeasuredbyhighergrossyieldsintheyearprior to September 2008, was associated with significantly larger outflows during the run. Other indicators of portfolio risk—whether the fund had a triple-A rating (line 2) or had Lehman exposure (line 3)—had no significant effect on outflows. The insignificance of the Lehman dummy is surprising,giventheroleofLehman’sbankruptcyintriggeringtherun. However,Ishowinsection 5.3belowthatLehmanexposuresdidhaveasignificantnegativeeffectonfundflowsearlyinthe run,butthattheimpactwasreversedinlaterweeks. InvestorriskalsowasanimportantpredictorofinstitutionalMMFoutflowsduringtherun in 2008: Three of the four investor risk proxies had significant effects (with the expected signs) onnetflows. Line4indicatesthatfundswithhigherexpenseratioshadlargernetflows(smaller outflows)duringtherun—aresultthatstandsincontrastwiththeusualnegativeeffectofexpense ratiosonflows(seesectionA.3oftheappendix). Althoughafund’sgrowthoverthepreviousyear (line 5) had no significant effect, lines 6 and 7 show that institutional funds with greater weekly flowvolatilityandhighersensitivitytoyieldshadsignificantlygreateroutflowsduringtherun. In fact, three indicators of investor risk (the expense ratio, flow volatility, and flow sensitivity) can explain much of the enormous difference in the outflows from institutional and retail funds during the run in 2008. The regression results for institutional funds imply that an institu- 21

tionalfundwithinvestorriskssimilartothoseofaretailfundwouldhavehadnetflowsthatwere 20.2“percentagepoints” largerthananinstitutionalfundwith averageinvestorrisk.25 Thistotal isaverysubstantialportionofthe20.8percentagepointdifferenceinmeanflowforinstitutional and retail funds.26 The calculation is merely illustrative, as it only takes into account differences in investor risk proxies and ignores the striking differences in institutional and retail investors’ responsestorisks. Still,theexplanatorypoweroftheseriskproxiesshowsthatasimpledistinction betweeninstitutionalandretailinvestorsisnottheonlyimportantmeasureofinvestorrisk. Bankaffiliation(line8)didnothaveasignificanteffectonnetflow. However,line10shows that larger institutional MMFs suffered proportionally more severe outflows during the run; indeed,twelvelargeMMFsaccountedformorethanhalfoftheaggregateoutflowfrominstitutional funds. Asdiscussedinsection4.4,fundsizemayhavebeenlinkedtomultipleformsofMMFrisk. Column 3 shows results for retail prime funds. A couple of risk factors with significant explanatorypowerforinstitutionalMMFoutflowsduringtherunwerelesshelpfulinexplaining retail fund flows, and R-squares for retail MMFs are lower. The estimated coefficient on gross yield,line1,isinsignificant. Higherexpenseratioswereassociatedwithsmalleroutflows(line4), but the magnitude of the effect was much smaller than it was for institutional MMFs. Indeed, as indicatedincolumn2,estimatedcoefficientsongrossyieldandtheexpenseratioaresignificantly different for institutional and retail funds. This column reports significance levels for tests that institutional and retail fund coefficients are different, based on a pooled regression of both types offundsinwhichaninstitutionalfunddummyisinteractedwitheachexplanatoryvariable.27 Nonetheless, investor risk does appear to have been important in explaining the variation inretailMMFflowsduringtherun. Inadditiontotheexpenseratio,flowvolatility(line6)enters theregressionsignificantlywiththeexpectedsign. Finally,fundsize(line10)wasalsoasignificant predictorofoutflowsfromretailfunds. Sponsor risk, too, had significant consequences for institutional prime funds during the run. Columns4and6reportresultsfromregressionsthatincludesponsorCDSspreads,atacostof considerablereductionsinsamplesize.28 Asshownincolumn4,highersponsorCDSspreads(line 9)wereassociatedwithsignificantlygreateroutflows. (Inthissample,thecoefficientonthebankaffiliation dummy, an indicator of low sponsor risk, is positive but marginally significant.) For 25Thisresultisbasedonregressioncoefficientsfromthefull-sampleinstitutionalregression(lines4,6,and7inthe firstcolumnoftable4).Imultiplycoefficientsforeachofthesethreeinvestor-riskproxiesbythedifferencebetweenthe proxy’ssamplemeanamongretailfundsanditssamplemeanamonginstitutionalfunds(columns5and1,respectively, of table 2a) and sum these products: 20.2 = 39.30(0.63−0.29)−2.06(2.76−5.13)−0.38(1.51−6.91). For ease of exposition, Irefertotheunitsof logflowsintermsofpercentagepoints. Theactual unitsare100timesthenatural logarithmofthesumofoneplustheratioofnetflowtolaggedassets.Seenote21. 26Comparecolumns1and5online12oftable2a. 27AnF-testrejects(witha p-valueofessentiallyzero)thenullhypothesisthatthecoefficientsforinstitutionaland retailfundsinthispooledregressionareallequal. 28Ialsoranregressionssimilartothosereportedincolumns1and3oftable4,butwithdummyvariablesindicating whetherCDSspreadswereavailableforeachfund’ssponsor.Thedummiesweregenerallynegativebutinsignificant. 22

retail funds, however, sponsor risk was not important in explaining outflows (column 6). Other results for the CDS spreads sample are similar to those for the full sample, although coefficients ontheinvestorriskproxiesaregenerallyestimatedwithlessprecision(forexample,weeklyflow sensitivityisnolongersignificantintheinstitutional-fundregression).29 The significance of proxies for portfolio, investor, and sponsor risk in these regressions showthattherunin2008wasnotsimplyanindiscriminatepanic. MMFswithgreaterrisksbore the brunt of the outflows during the run. R-squares from the regressions indicate that risk proxies explain a sizable portion of the cross-sectional variance in flows, particularly for institutional funds. Moreover, the particularly strong links between risk proxies and outflows from institutionalfundspointtointeractionsinMMFrisks: Fundswithgreaterinvestorrisks(due,inpart,to theirmoresophisticatedinvestors)werealsomoresensitivetoportfolioandsponsorrisks. Table 5 provides some additional insight into the economic significance of the regression resultsbyreportingestimatedimpactsonnetflowof1-standard-deviationincreasesineachrighthand-sidevariable(fordummyvariables,whichareindicatedwithanasterisk(*),theentryrepresentstheestimatedeffectofa0-to-1changeinthatvariableonnetflow,whichtypicallyrepresents about a 2-standard-deviation change).30 For example, line 1 of column 1 shows that, for institutional prime funds, a 1-standard-deviation increase in gross yield paid over the year ending in August 2008 was, all else equal, associated with 9 percentage points greater outflow during the run. Normalizedchangesinotherriskproxiesalsohadsubstantialpredictedeffectsonnetflows forinstitutionalfunds(columns1and3). Forretailfunds(columns2and4),normalizedchanges inmostriskproxieshadmuchsmallerpredictedeffectsonnetflows. 5.2 Doesthelinkbetweenoutflowsandgrossyieldsreflectportfoliorisk? The role of gross yields in explaining outflows from institutional MMFs during the run in 2008 underlinestheimportanceofinterpretingtheinformationcontainedintheseyields. Section4.1.1 and section A.2 of the appendix offer evidence that an MMF’s gross yield is a useful summary statistic for its portfolio risk. As noted in section 4.4, one alternative interpretation—that higher gross yields indicated greater investor risk—is not supported by any significant pairwise correlation between gross yield and the proxies for investor risk. Moreover, the multiple controls for investorriskinequation(1)shouldpickupeffectsofinvestorriskthatarerelatedtogrossyield. Another possibility, discussed in section 4.1.4, is that higher gross yields also reflected greaterportfoliomanagertalent. Onewaytocontrolforanyconfoundingeffectofmanagerability istouseportfoliocharacteristicstoinstrumentforgrossyieldsinregressionsoftheformofequation (1). As noted in section A.2 of the appendix, gross yields of individual prime MMFs varied 29Again, an F-test rejects (with a near-zero p-value) the null hypothesis that the coefficients for institutional and retailfundsareallequalinapooledregressionincludesCDSspreads. 30Seecolumns2and6intable2aforstandarddeviationsoftheexplanatoryvariables. 23

significantly with the shares of their assets held in different types of instruments and with their WAMs. Manager ability should have been orthogonal to these broad portfolio measures. Thus, I run a two-stage least-squares (2SLS) regression in which the portfolio measures from table A1 serveasinstrumentsforgrossyields.31 Results are shown in table 6. For comparison, column 1 repeats the output of the baseline regression for institutional funds from the first column of table 4. The 2SLS regression results, listed in column 2, are quite similar; the estimated coefficient on gross yield is highly significant butalittlesmallerthanthatfromtheregressioninwhichgrossyieldappearsdirectly. TheinstrumentalvariablesresultscorroboratetheviewthatinstitutionalMMFswithhighergrossyieldshad largeroutflowsbecauseofinvestors’concernsaboutgreaterportfoliorisks. 5.3 MMFrisksandoutflowsbyweekduringthe2008run The run on MMFs in September and October 2008 was punctuated by very significant financial and policy events. As indicated by panel B of figure 2, these included the Lehman bankruptcy early on Monday, September 15; Reserve’s announcement at 4 p.m. on Tuesday, September 16 that the Primary Fund had repriced its shares at 97 cents; the announcements of the Treasury’s Temporary Guarantee Program for MMFs and the Federal Reserve’s AMLF on the morning of Friday,September19;andtheannouncementoftheFederalReserve’sCommercialPaperFunding Facility (CPFF) on October 7. Hence, a closer look at the timing of the links between MMF risks andoutflowsmaybeinformativeinunderstandingtheroleofrisksduringtherun. Todoso,Irunweeklyregressionssimilartoequation(1)butinwhichthedependentvariableisnetflowmeasuredoveraparticularweek.32 Themoreprecisetimingoftheflowsinthese regressions allows some modifications to the sample, which are described more fully in section A.1. For example, several MMFs that are excluded from the full-sample regressions because of measurement problems can be included in a subset of the weekly regressions, and each weekly regression includes CDS spreads measured over the previous week, to capture the effects of any deterioration in sponsors’ financial condition. For each specification (including and excluding CDS), I run a separate regression for each of the seven weeks ending on Tuesdays from September 2 to October 14, reflecting the timing of iMoneyNet’s weekly taxable MMF data. Results are 31Analternativeapproachistoincludeportfoliomeasuresinsteadof (ratherthaninstrumentsfor)grossyield. For example,includingonlyfunds’ABCPportfoliosharescorroboratesDuygan-Bumpetal.’s(2010)finding—basedonan analysisofallprimefundsandadifferentsampleperiodthanIuse—thatfundswithhigherABCPexposureshadlarger outflowsduringtherun.(IfgrossyieldandABCPsharearebothincludedinmybaselinespecificationforinstitutional funds,thecoefficientongrossyieldremainssignificantlynegativeandABCPshareisnegativeandsignificantatthe 12 percent level.) A temptation here is to run every permutation of the portfolio characteristics in a “horse race” to pinpointthekeycharacteristicsthatspookedinvestors. Myaimissimpler—toshowthatportfolioriskwasimportant inpromptingoutflowsduringtherun. 32iMoneyNetprovidesdailyassetsdataforasubsetofthefundsittracks, althoughformanyMMFshareclasses, dailyassetobservationsappeartobeflatlinedforeachweek. Ialsoranregressionsatadailyfrequency,butthedaily resultsaddedlittletotheinsightsthattheweeklyregressionsprovide. 24

depictedgraphicallyinfigures3a(forregressionsthatomitCDSspreads)and3b(regressionsthat includeCDSspreads). Forbrevity,IonlyreportresultsforinstitutionalMMFs.33 The figures illustrate the timing of both the economic and statistical significance of each of the risk proxies. The solid line in each panel plots, for each week, the predicted effect of a one-standard-deviationincreaseinanexplanatoryvariableonnetflowasapercentageoflagged assets.34 (The plots also show predicted effects of one-standard-deviation changes in dummy variables,ratherthan0-to-1changes.) Thedashedlinesaretheupperandlowerboundsofaquasi 95 percent confidence interval for the predicted effects derived by using a two-standard-error confidenceintervalforeachestimatedcoefficient.35 Figure3ashowsthat,inthefull-sampleregressionsthatexcludeCDS,theriskproxieshad virtually no significant effect on fund flows in the two weeks prior to the Lehman bankruptcy (the weeks ending September 2 and 9). However, by the third week (September 10-16), which included the Lehman bankruptcy and ended just as the Reserve Primary Fund announced that it had broken the buck, risk proxies began to have significant effects. Two investor risk proxies (theexpenseratio,panel4,andflowsensitivitytoyield,panel7),size(panel10),and—notably— the Lehman-exposure dummy (panel 3) are all significant with the predicted signs. Institutional MMFs suffered their worst outflows of the entire episode during the fourth week (September 17-23), and, on the 19th, the government announced both the Treasury Guarantee Program for MMFsandtheFederalReserve’sAMLF.Inthatweek,highergrossyields(panel1)begantoenter significantly and two investor risk proxies remained significant (panels 4 and 7). However, the coefficient on the Lehman-exposure dummy (panel 3) was positive! By the fifth week, with the governmentprogramsinplace,onlyacoupleofriskproxiesweresignificant,andinthefinaltwo weeks(endinginOctober),almostnonewere. When CDS spreads are added to the regression, the patterns of significance among the proxieschangesomewhat(inpartbecausethesamplesizedeclines),buttheoverallpicture(figure 3b) is similar. Gross yield (panel 1) was a significant predictor of flows one week earlier (during theweekendingSeptember16)thanitwasinthefull-sampleregression. Twoinvestorriskproxies (panels6and7)andtheLehman-exposuredummies(panel3)arenolongersignificant. TheCDS spreaditselfwassignificantatleastatthe10percentlevelforfourweeksspanningSeptember3- 30—startingevenbeforetheLehmanbankruptcy—althoughthemagnitudeofthepredictedeffect increased substantially in the week that included the Lehman collapse and Reserve’s break-thebuckannouncement.36 33Resultsforretailfunds,aswellasresultsforallMMFsintabularform,areavailableuponrequest. 34Predictedeffectsreportedherehavebeenconvertedfromlognetflowstonetflowsasapercentoflaggedassets. 35Forexample,upperboundsforthepredictedeffectsofhighergrossyield(panel1)arecomputedbyaddingtwo standarderrorstoeachweeklyestimatedcoefficientongrossyieldandmultiplyingtheresultbythestandarddeviation ofgrossyieldforinstitutionalprimefunds. 36p-valuesforthenullhypothesisthattheCDSspreadwaszerofortheweeksendingSeptember9,16,23,and30 were0.044,0.093,0.057,and0.002,respectively. 25

In summary, the week-by-week regressions indicate that proxies for risk were generally mostsignificant—andhadthelargestpredictedeffects—justastherunwaspeaking. Threeadditionalpointsareworthnoting. First,theLehman-exposuredummypredictedsignificantlygreater outflows (at leastin the full sample)in the week from September 9-16 butsmalleroutflows in the followingweek. OnepossibleexplanationisthatinstitutionalinvestorsredeemedsharespreemptivelyinthedaysjustbeforetheReservePrimaryFundbrokethebuckbutsubsequentannouncements of support actions by sponsors of MMFs (other than Reserve) with Lehman exposure allayed investor concerns and slowed the outflows. For example, on September 17, BNY Mellon announced support for its Dreyfus MMFs that held Lehman debt (Condon, 2008). This may explainwhytheLehman-exposuredummywasnotsignificantintheregressionsthatcoverthefull periodoftherun. Second, the weekly graphs highlight that fund size had a strikingly large and significant effectonoutflowsatthepeakoftherun. Asnotedinsection4.4,sizeprobablyreflectedmultiple formsofMMFrisk. Third, a weekly analysis of retail funds (not shown) indicates that MMF risk proxies did havesignificanteffectsonfundflows,butonlyratherlateintherun. FortheweekendingSeptember30,coefficientsforLehmanexposure,weeklyflowvolatility,weeklyflowsensitivity,andfund sizearenegativeandsignificant. Thedelayedresponseofretailinvestorsisalsoapparentinpanel Boffigure2;outflowsfromretailfundspeakedlaterthanthosefrominstitutionalfunds. 6 TheABCPcrisisandMMFrisks The crisis in the market for ABCP that erupted in August 2007 caused severe strains for some MMFs and for short-term funding markets. However, the broader consequences were far less momentous than those of the run in 2008, in part because capital losses of MMFs in 2007 were absorbed by their sponsors, no fund broke the buck, and MMFs suffered no large-scale run. As such, MMF risks played a different role in this crisis than they did during the run a year later. Below,IexaminelinksbetweenMMFrisksandthreetypesofABCPcrisisoutcomesforindividual MMFs: primefunds’netflowsinthethreeweeksfromAugust7to28,2007;sponsorinterventions tosupportMMFsthathelddistressedABCP;andfundexposurestodistressedABCP. 6.1 MMFrisksandthecross-sectionofoutflowsduringtheABCPcrisis Given the links between MMF risks and outflows during the run in 2008, a natural question is whetherthesmallaggregateoutflowfromprimefundsduringtheABCPcrisismaskedsignificant cross-sectional correlation between fund risks and flows. Anecdotal evidence suggests such a story; as noted in section 3.2, at least one MMF suffered very large outflows because of holdings of distressed ABCP. Still, the standard deviation of prime institutional MMF flows was much 26

smallerduringtheABCPcrisis(table2b,line12,column2)thanitwasduringthe2008run(table 2a). The dispersion of retail fund flows as the ABCP crisis unfolded (table 2b, line 12, column 6) nearlymatchedthatofinstitutionalfunds. TotesttherelationshipbetweenMMFrisksandnetflowsduringtheABCPcrisis,Iestimate slightlyrevisedversionsofequation(1),withnetflowstoinstitutionalandretailprimeMMFsin thethreeweeksendingAugust28,2007asdependentvariables. Explanatoryvariablesaretherisk proxies described in section 4 and are listed in table 2b. The only important differences between theseriskfactorsandthoseusedtoanalyzetherunin2008arethedatesoverwhichthemeasures are computed and the specific distressed-asset exposures of concern (line 3). For analysis of the ABCPcrisis,therelevantexposuresweretodistressedABCPissuesthatcausedproblemsforother MMFs. Correlationsamongtheriskproxiesappearontable3b. Regression results are shown in table 7. Whereas multiple risk proxies were useful in predicting the net flows of institutional prime MMFs during the 2008 run, net flows to these funds exhibitedlittlesignificantresponsivenesstorisksduringtheABCPcrisis. Acoupleofexplanatory variablesaremarginallysignificantinthefull-sampleregression(column1),butoneofthese(flow sensitivity, line 7) has the “wrong” sign. In the regression that includes CDS spreads (column 4), thesespreadsareonlymarginallysignificant(atthe11percentlevel).37 Oddly,retailfundflowsappeartohavebeensomewhatmoreresponsivetoMMFrisks,but onlyinthesmallersamplethatincludesCDSspreads(column6). Inthatspecification,holdingsof distressedABCP(line3)hadastatisticallysignificantnegativeeffectonnetflow,andtheinvestor riskproxieswereallsignificantatleastatthe10percentlevel—buttwoofthem(theexpenseratio, line4,andgrowthoverthepreviousyear,line5)hadthe“wrong”signs. Noneoftheriskproxies issignificantatthe5percentlevelinthefull-sampleregressionforretailfunds(column3).38 Takentogether,theriskproxieshadfarlessexplanatorypowerforthecross-sectionalvariationinnetflowsduringtheABCPcrisisthantheydidforvariationinflowsduringtherunin2008: Three of the four R-squares on table 7 are less than half those for the corresponding regressions ontable4. ForMMFinvestors,moneyfundrisksapparentlyremainedmostlylatentthroughthe ABCPcrisis,amidthewidespreadinterventionsbysponsorstoabsorblosses. Again,theevidence points to important interactions among MMF risks: When sponsor risk was perceived to be low, portfolioandinvestorriskshadlittleimpactonnetflows. 37Although the estimated coefficient on CDS spreads for institutional prime MMFs in this regression is only marginally significant, it is nearly as large in magnitude as that on CDS spreads for such funds during the run in 2008. However,evenifwetakethesemagnitudesasgiven,highersponsorCDSspreadsduringtheABCPcrisiswere muchlessconsequentialforMMFsthansuchspreadswereduringtherunin2008. Aone-standard-deviationincrease inCDSspreadinthefirstweekofAugust2007wasonly25basispointsandhencewouldpredictonly3.5percentage pointsmoreoutflowfromthesefunds. Asshownonline9,column3oftable5,aone-standard-deviationincreasein sponsorCDSspreadinearlySeptember2008wasassociatedwith12percentagepointsmoreoutflowduringtherun. 38AnF-testbasedonapooledinstitutional-retailregressionthatexcludesCDSspreadsdoesnotrejectthenullhypothesisthatthecoefficientsforinstitutionalandretailfunds(reportedincolumns1and3oftable7)arejointlyequal. However,suchatestdoesrejectthenull(atthe5percentlevel)whenCDSspreadsareincluded. 27

6.2 MMFrisksandsponsorfinancialsupportfollowingtheABCPcrisis For some MMFsponsors, however, the consequences ofmoney fund risks were quitecostly duringtheABCPcrisis(Crane,2009;CreditSuisse,2008,2009). Indeed, sponsors’recordoffinancial supportfortheirfundsprovidesanotherperspectiveonMMFoutcomesduringthisepisode. Becausesponsors’actionsrevealtheirconcernsaboutactualorexpectedlossesinfunds’portfolios, as well as concerns about investors’ reactions to fund risks, the support actions indicate where NAVfluctuationsorlargeoutflowslikelywouldhaveoccurredintheabsenceofinterventions.39 SomeofthepredictedrelationshipsbetweenMMFrisksandsponsorsupportactionsshould be similar to the links between risks and outflows during the run in 2008. Greater portfolio risk, allelseequal,shouldbeassociatedwithahigherprobabilityofasponsorbailout. Supportactions mightalsobelinkedtoinvestorrisk,sinceheavyredemptionscanexacerbatelossesforMMFs. However, a key distinction between MMF outflows and sponsor support as outcomes of interestisintheirlinkstosponsorrisk. Section5showsthatMMFinvestorsrespondedtosponsor risk by disproportionately redeeming shares from MMFs with riskier sponsors during the run in 2008. As such, sponsor risk is an important dimension of MMF risk. But when the outcome of concern is a sponsor’s intervention to support an ailing fund, sponsor risk may have multiple roles. AsponsorwithdeepfinancialresourcesmaybemorelikelytobailoutatroubledMMF,so lowersponsorriskmaypredictthe“bad”outcome(support). Butalinkbetweensponsorfinancial strength and MMF support also might reflect moral hazard, if portfolio managers made riskier investments believing that deep-pocketed affiliates could absorb losses. In addition, anticipation ofasponsor’sneedtoprovidesupportforits MMFsmightdriveupitsCDSspreads, sosponsor riskitselfmaybeendogenous. Asdescribedinsection2,Iobtainedinformationaboutsponsor-supportinterventionsfrom publicsources,includingtheSEC’swebsiteandthefinancialstatementsofassetmanagers,aswell as from confidential SEC records. These data are thus a more complete record of support than is available from public records, although, as noted earlier, some sponsor interventions still may be unrecorded. Table 8 provides some summary statistics, measured just before the ABCP crisis, for all prime funds in my sample (column 1) and for the funds with records of sponsor support (column2).40 ThetablecombinesdataforinstitutionalandretailfundsbecauseIfindnostatistical evidence that the relationship between MMF risks and support actions was different for the two typesoffunds. 39Ideally,sponsorsupportrecordswouldquantifythemarketvalueofeachsponsor’scontributiontoafund’sassets toallowestimationofthemagnitudeofNAVdeclinethatwouldhaveoccurredwithoutintervention. However,most oftherecordsprovide,atbest,informationaboutthetotalamountsofdistressedassetspurchasedfromfundsorthe notionalsizeofwrapsprovidedfordistressedassets.Thus,Iwasnotabletoconstructacontinuousmeasureofsupport applicabletoallthefundsthatreceivedit. 40Asnotedinsection2,thesampleincludesfundswithatleast$100millioninassetsand12monthsofhistoryin theiMoneyNetdatabaseasofJuly31,2007.Theunitofanalysisisthemoneymarketfund;multipleshareclassesfora singlefundarecombinedintooneobservation(seenote4). 28

Ofthe249primemoneyfundsinthesampleasofJuly31,2007(line1),39funds(16percent) hadrecordsofsponsorsupportduetoholdingsofdistressedABCP.41 Line2showsthat20ofthe 116institutionalprimeMMFsreceivedsupport,asdid19ofthe133retailprimefunds. Supported funds were somewhat larger than average (line 3) and had earned gross yields that were just a touchhigherthanaverageintheyearbeforetheABCPcrisisbegan(line4). Supportedfundsalso weremorelikelythanotherprimefundstohavehadatriple-Arating: One-thirdofallMMFshad such a rating, but triple-A funds accounted for almost half the funds that received support (line 5). Thenextfourlinesindicatethatsponsor-supportedfundsexhibitedgreaterinvestorriskthan the rest of the prime fund industry by several measures: They had lower expense ratios (line 6), morerapidgrowthinthepreviousyear(line7),andgreaterflowvolatilityandsensitivitytoyield (lines8and9). Supportedfundsweremorelikelythanaveragetobebank-affiliated(line10)and tohavesponsorswithCDSspreadsintheMarkitdatabase(line11). TheCDSspreadsofsponsors ofMMFsthatreceivedsupportweresomewhathigherthanaverage, bothintheweekbeforethe onset of the crisis (line 12) and in the year beforehand (line 13). In addition (not shown), CDS spreads for bank sponsors averaged 16 basis points less than those of other sponsors in the first weekofAugust2007,and10basispointslessintheyearbeforetheonsetoftheABCPcrisis. To test the relationship between sponsor support and MMF characteristics, including risk proxies,Idefine: (cid:40) 1 iffundireceivedsponsorsupportduetoexposuretodistressedABCP Si ≡ 0 otherwise. Iuseaprobitregressiontoestimate: (cid:18) Pr(Si = 1) = Φ β gyieldi +β AAAi +β exprati +β growthi 1 t∈P 2 t=P¯ 3 t∈P 4 t∈P +β flowSDi +β flowSensi +β banki +β CDSi +β CDSi 5 t∈P 6 t∈P 7 t=P¯ 8 t=P¯ 9 t∈P (cid:19) +β insti +β assetsi +constant . (2) 10 t=P¯ 11 t=P¯ Explanatory variables are listed, in the order in which they appear in equation (2), in table 9. I estimateonepooledregressionforinstitutionalandretailfundsbecausetherelationshipbetween supportandMMFcharacteristicswasnotstatisticallydifferentforinstitutionalandretailfunds.42 Theexplanatoryvariablesarethesameasthoseusedtoestimateequation(1),withthreechanges. 41The SEC indicated that 44 funds had received such support (U.S. Securities and Exchange Commission, 2009a, note38).Ifoundrecordsofsupportfor43uniqueMMFs,buttwowereforgovernment-onlyfunds,onewasforafund withnoiMoneyNetdata,andoneforafundthatwaslessthanoneyearold. 42Inapooledregressionofinstitutionalandretailfunds(notshown),inwhichaninstitutional-funddummywasinteractedwitheachexplanatoryvariable,noneoftheindividualestimatedcoefficientswassignificantlydifferent(atthe 5percentlevel)forthetwotypesoffunds,andtheinteractiveinstitutionaldummyvariableswerejointlyinsignificant. 29

First, I do not include the indicator variable for exposures to distressed securities, which I constructed using securities identified in sponsor-support records. Second, equation (2) includes a dummyforinstitutionalfunds. AthirdchangeismotivatedbyconcernsabouttheendogeneityofCDSspreads. Although sponsors’early-AugustCDSspreads(CDSi )areusefulforexaminingthesponsor-riskeffecton t=P¯ subsequentnetflows,spreadsmeasuredontheeveoftheABCPcrisismayhavealreadyreflected concerns about the costs of MMF bailouts for sponsors. If so, early-August spreads should enter theregressionpositively,whileCDSspreadsmeasuredearliershouldbelesspredictive—orenter negatively, if sponsor financial strength caused moral hazard for managers. Hence, I include in theregressionaverageCDSspreadsmeasuredovertheprecedingyear(CDSi ). t∈P Thefirsttwocolumnsoftable9listtheresultsofestimatingequation(2). Column1shows results for the full sample of MMFs when CDS spreads are excluded, and column 2 provides resultsforthesubsampleforwhichsponsorCDSspreadsareavailable. As shown on line 1, portfolio risk—as indicated by funds’ gross yields earned over the previousyear—wasastatisticallysignificantpredictorofsponsorsupportduringtheABCPcrisis. It was also economically important: Based on the full-sample result (column 1), a one-standarddeviation increase in a fund’s gross yield (only 3 basis points in the pooled sample of all prime funds) is associated with a 13 percentage point increase in the probability that the fund was the recipient of sponsor support. (Based on the regression that includes CDS spreads, the predicted effectis39percentagepoints.) Theimportanceofgrossyieldinpredictingsponsorsupportmakes sense: Riskierportfoliosweremorelikelytoexperiencelossesthatsponsorsultimatelyabsorbed. Incontrast,atriple-Arating(line2)hadnosignificantpredictivepowerinthefullsampleandhad the “wrong” sign in the CDS sample: Controlling for CDS spreads, funds with triple-A ratings weremorelikelytohavebeentherecipientsofsponsorsupport.43 The evidence for a link between investor risk and sponsor bailouts is less compelling. Among the indicators of investor risk, only the volatility of net flows over the previous year (line5)wassignificant(marginallysointheCDSregression). Moreover,asnotedabove,testsfor institutional-retaildifferencesintherelationshipbetweensponsorsupportandMMFrisksfound no evidence for any distinction. Given the overwhelming evidence during the run in 2008 that institutional funds represented greater investor risk, this suggests that such risk was not a factor insponsorinterventions. Oneinterpretationoftheseresultsisthatsponsorsupportwasprovided whenportfolioassetssoured,regardlessoftheadditionalrisksthatinvestorsmayhaveposed. Sponsor risk had a significant but somewhat ambiguous role in predicting sponsor support. Bank-affiliatedMMFsweremorelikelytoreceivesupport(line7). Indeed,Babaetal.(2009) 43Onepossibleexplanationisthatatriple-Aratingreflectedasponsor’smotivation(conditionalonfinancialstrength) tosupportanailingfund,althoughtheratingsorganizations’narrowfocusonportfoliorisks,particularlypriorto2008, makessuchanexplanationseemdubious.Asshowninsection6.3below,triple-Aratingswerenothelpfulinpredicting whetherMMFshelddistressedABCP. 30

notedthatbank-affiliatedfundmanagers“wereover-representedamongsupportproviders.” As notedabove,thelinkbetweenbanksandMMFsupportmayreflectagreaterpropensityofdeeppocketedsponsorstobailouttroubledfunds,conditionalonsimilarexposurestodistressedsecurities, or it may reveal a moral hazard problem for bank-affiliated portfolio managers. An additionalconsiderationisthatthebanksmayhavebeenmorelikelytodisclosefinancialsupportfor affiliated MMFs because banks face more rigorous regulatory oversight and disclosure requirementsthansomeotherfinancialservicesfirms. ThelinkbetweensponsorCDSspreadsandsupportismorepuzzling: Theestimatedcoefficient onearly-August spreads (line 8) isnegative and insignificant, while that onspreads registered over the preceding year is positive and highly significant. This pattern is exactly opposite whatwewouldexpectifspreadswereendogenouslyreflectingthelikelihoodofsponsorsupport. But the role of sponsor risk is odd: Apparently, aside from a sponsor’s bank affiliation, riskier sponsorsweremorelikelytointervenelatertosupporttheirfunds.44 6.3 MMFrisksandholdingsofdistressedABCP Analysis of MMFs’ holdings of distressed ABCP can help interpret links between risks and poor outcomes for the funds. For example, a finding that bank-affiliated advisers were more likely to haveinvestedinproblematicABCPwouldbeevidenceinfavorofamoral-hazardexplanationfor the link between bank affiliation and sponsor support. The potential incompleteness of sponsorsupportrecordsalsomakesportfolioexposuresinformationvaluableforthestudyofMMFsduringthisepisode. Usingfunds’ mostrecentSEC filings(seesection 2)prior totheonset oftheABCP crisis, I identified124MMFsthatheldABCPthatultimatelytriggeredsponsorsupportactionsforatleast one MMF. The third column of table 8 provides some summary statistics on these funds, which accounted for nearly half of the money funds in the sample—more than triple the number that reportedlyreceivedsponsorsupport. Comparedwiththefullsample(column1),fundsthatheld distressedpaperwereabitmorelikelytobeinstitutionalfunds,werelarger,hadflowsthatwere more sensitive to yield, and were more likely to be affiliated with banks. Otherwise, funds that helddistressedpaperweresimilartootherMMFs. Finally,asshownincolumn4ofthetable,Icombinetheindicatorsofsponsorsupportand exposure to distressed paper to capture MMFs that apparently held problematic securities that were not identified using the portfolio records. This combination adds just four funds to the list ofthosethathelddistressedsecurities. To examine the relationship between exposures to distressed ABCP and the riskiness and 44Ifbankaffiliationisdroppedfromtheregression,coefficientsonsponsorCDSaresimilartobutsmallerinmagnitudethanthosereportedincolumn2(thecoefficientonCDSspreadsoverthepreviousyearremainssignificant). 31

othercharacteristicsofMMFs,Iredefine: (cid:40) 1 iffundihelddistressedABCP Si ≡ 0 otherwise. Using this definition of Si, I reestimate equation (2) separately for the two criteria for identifying funds that held distressed ABCP: portfolio disclosures alone and portfolio disclosures plus sponsorsupportrecords. Resultsappearincolumns3through6oftable9. Again, as shown on line 1, portfolio risk—as measured by gross yield—was an important predictor of MMF problems. Gross yield is statistically and economically significant in every regressionreportedhere;aone-standard-deviationincreaseingrossyieldispredictedtoraisethe likelihood that a fund held distressed paper by 15 to 18 basis points, depending on specification. Incontrast,fundratings(line2)werenothelpfulinpredictingexposurestoproblematicABCP. Thelinkbetweeninvestorriskandholdingsofdistressedpaperislessclear. Oneindicator of lower investor risk (the expense ratio, line 3) and one of greater investor risk (flow sensitivity toyield,line6)arepositivelyrelatedtoholdingsofdistressedABCPinmostofthespecifications listedincolumns3through6. Takenliterally,thismightimplythatfundswithlesssophisticated, hot-moneyinvestorsweremorelikelytoholddistressedsecurities. Aninstitutional-funddummy (line9)ispositiveandsignificantinacoupleofspecifications(onlymarginallysoinone),buttests of the null hypothesis that all estimated coefficients for institutional and retail funds are jointly equaldonotrejectthenullforanyoftheregressionsshownontable9. As line 7 shows, MMFs with bank-affiliated sponsors were significantly more likely to hold distressed ABCP than other funds. Depending on specification and sample, bank affiliationincreasedtheprobabilitythatafundhelddistressedpaperbybetween26and41percentage points. Thestrengthofthisresultaidsininterpretingthelinkbetweenbankaffiliationandsponsor support—bank-affiliated funds evidently were more likely to receive support because they were morelikelytoholdproblematicABCP—andpointstoapotentialmoralhazardproblemforbankaffiliatedMMFmanagers. Moralhazardisnottheonlypossibleexplanation,butsomeothersare no more charitable. For example, it is possible that bank-affiliated managers were more likely to purchase risky ABCP for their funds because they had more institutional familiarity than other managerswithcomplexinstrumentslikepaperissuedbystructuredinvestmentvehicles(SIVs). The role of sponsor CDS spreads in predicting holdings of distressed paper again seems puzzling. In the regression defining Si based solely on portfolio disclosures (column 4), CDS spreadsareinsignificant. WhenSi ≡ 1alsoforfundsthatreceivedsupport(column6),coefficients on sponsors’ CDS spreads in the year before the crisis are positive and significant (line 9) but spreads immediately beforehand (line 8) are insignificant. Controlling for bank affiliation, funds withriskiersponsorsweremorelikelytopurchaseproblematicassets.45 45Whenbankaffiliationisdroppedfromtheregression,sponsorCDSisinsignificant.Asnotedabove,bank-affiliated 32

Finally, larger MMFs (line 10) were significantly more likely to hold distressed paper than their smaller counterparts. A one standard deviation increase in fund assets boosted the predicted likelihood of such holdings by about 15 percentage points in the regressions summarized incolumns3through6. 7 Conclusionsandpolicyimplications This paper argues that MMFs are subject to three types of risk: (1) portfolio risks arising from the credit, liquidity, and interest-rate risks posed by a fund’s assets; (2) investor risk due to the composition of an MMF’s investors and the likelihood that they will suddenly and disruptively redeem shares; and (3) sponsor risk that reflects the possibility that an MMF sponsor will not provide financial support for an ailing fund. I describe proxies for each type of risk and provide evidence that these measures were useful in explaining the cross-sectional variation in outcomes forfundsduringcrises,particularlythecatastrophicrunonMMFsinSeptemberandOctober2008. Indeed, one important finding of this paper is that the run on MMFs was not indiscriminate; ex anteriskproxiesexplainalargeportionofthesubstantialvarianceinoutflowsduringthecrisis. Portfolio risk is the focus of much of the risk-limiting regulation that governs MMFs as well as the ratings criteria applied to MMFs by ratings organizations. I show that portfolio risk, as measured by gross yield, was a significant and economically important predictor of outflows during the run in 2008. Moreover, the portfolio risks that motivated redemptions during the run were broader than direct exposure to Lehman debt, which prompted larger outflows only early in the run. Gross yield was also useful in predicting sponsor support for ailing funds during the ABCP crisis in 2007 and for predicting holdings of distressed ABCP during that episode. In retrospect,theseresultsareintuitive: RiskysecuritiesshouldpayhigherratesthatboostMMFportfolios’ gross yields and should—if yields reflect systematic risk—perform poorly during crises. The ABCP crisis and the run on MMFs a year later provided a laboratory to test whether higher yields in MMFs were, in fact, conveying information about risks that was observable well before thecrises. Ifindthathigheryieldsdidoffersuchasignal. However, I find that another possible indicator of portfolio risk—whether a fund had a triple-Arating—wasoflittleuseinpredictingcrisisoutcomes,includingoutflowsduringtherun in 2008 or exposure to distressed paper during the ABCP crisis. This is perhaps surprising, as ratingsorganizations’publicationssuggestthatatopratingshouldbeusefulasanindicatorofan MMF’s(low)risk,particularlyasreflectedinitsportfolioquality. ThelinkbetweenotherformsofMMFriskandoutcomesismorecomplex. Investorsgenerallyhavefledto,notfrom,MMFsduringepisodesoffinancialturmoil,andsponsorshavebeen sponsorshadlowerCDSspreadsthanthoseofothersponsorsjustpriortotheABCPcrisis.Becausethebank-affiliated funds were more likely to hold distressed ABCP, failing to control for bank affiliation misses the broader positive correlationbetweensponsorCDSspreadsintheyearbeforethecrisisandholdingsofdistressedassets. 33

seenasasourceofstabilityratherthanrisk. Hence,investorandsponsorrisksofMMFshaveremained largely latent, even during crises. Not surprisingly, these risks historically have received lessattentionfromregulators,ratingsorganizations,andacademicsthanhaveportfoliorisks. Investors’ muted response to the ABCP crisis illustrates the usual latency of investor and sponsorrisks. SouringpaperputconsiderablestrainsonMMFs,andmanywouldhavebrokenthe buck without sponsor support. Nonetheless, investors’ net flows to MMFs as the crisis erupted in August 2007 were not significantly related to sponsor and portfolio risks, and there was no consistentpatternofassociationbetweenoutflowsandinvestorrisk. In contrast, the severity of the financial crisis in September 2008 and Reserve’s inability to absorb losses in its Primary Fund undermined confidence in sponsors’ ability to support MMFs andbroughtsponsorandinvestorriskstothefore. Institutionalinvestorsreactedtosponsorrisk by redeeming shares more aggressively from MMFs with sponsors that had higher CDS spreads ontheeveoftherun. Meanwhile,theconsequencesofinvestorriskwereapparentintheaggregate pattern of outflows during the run, as institutional investors redeemed shares at ten times the pace of retail redemptions. Even among institutional funds, I find that proxies for investor risk, including funds’ expense ratios, flow volatility, and flow sensitivity to yield, helped predict the crosssectionofoutflows. The two crises also underline important interactions among the three types of MMF risks. For example, the particularly strong links between risk proxies and outflows from institutional fundsduringtherunin2008showthatfundswithgreaterinvestorriskswerealsomoresensitive to portfolio and sponsor risks. In addition, the contrast between the responsiveness of institutionalfundflowstoallthreetypesofMMFrisksduringtherunin2008andtherelativeinertness of funds’ flows during the ABCP crisis in 2007 indicates that heightened sponsor risk in 2008 intensifiedtheconsequencesofotherrisks. Although sponsor risk was not a significant factor in the cross-section of net flows during the ABCP crisis, one proxy for sponsor risk—whether an MMF was affiliated with a bank—was a significant predictor of poor outcomes during this episode. Bank-affiliated money funds were morelikelytoreceivesponsorsupportandtoholddistressedABCPintheirportfolios. Somewhat morepuzzlingisthat,controllingforbankaffiliation,riskiersponsors—thosewithhigherpre-crisis CDSspreads—weremorelikelytohaveexperiencedproblemsintheirMMFs. Thispaperprovidessomeusefullessonsbothforpolicymakersandinvestors. ThesignificanceofMMFrisksinexplainingpooroutcomesinpastcriseshighlightstheimportanceofmonitoring these risks, and I offer some useful proxies for doing so. For example, shareholders and regulators might track funds’ gross yields for early signs of problematic portfolio risks, particularlygivenassetmanagers’incentivestoboostyields. Myindicatorsofinvestorriskmaybeuseful intheSEC’songoingeffortstoreflectsuchriskinsettingliquidityrequirementsforMMFs.46 46InadoptingnewliquidityrequirementsforMMFs,theSECnotedthatinstitutionalandretailMMFshavesubstan- 34

This paper’s findings also raise concerns about the systemic risks associated with sponsor support actions for MMFs and the expectations of safety that these actions have fostered among investors. Clearly,sponsorsupportofMMFswascriticalinhelpingfundsweathertheABCPcrisis in 2007 and the run in 2008. But the extensive record of sponsor support has probably attracted many highly risk-averse investors who would not hold MMFs without the conviction that the funds are effectively (privately) insured. Hence, sponsor support has likely increased investor risk for MMFs. The fact that funds with bank sponsors were more likely to have held distressed ABCPandtohavereceivedsponsorbailoutsinthewakeoftheABCPcrisisalsosuggeststhatthe possibilityofsponsorsupportmayundermineincentivesforprudentassetmanagement. Furthermore, during the run in 2008, concerns about the ability of sponsors to support their MMFs evidently prompted heavier redemptions from money funds with weaker sponsors, andthustransmittedthesponsors’strainstooff-balance-sheetMMFsandintoshort-termfunding markets. Thus, by fostering expectations of implicit recourse to sponsors, past support actions hadcreatedachannelforthetransmissionduringcrisesofstrainsbetweenentitiesthatshouldnot havebeenrelated. Whetherornotsuchsupportwasactuallydelivered,itmayhavecontributedto financialstrains. BailoutsofMMFsduringtherunrequiredscarcecapitalfromsponsorsatatime when liquidity was in short supply and worsened some sponsors’ financial condition (Standard & Poor’s, 2008a). But Reserve’s failure to provide support that investors had come to expect was catastrophicfortheReservefranchiseanddestabilizingforthefinancialsystem. Moreover,despite theapparentimportanceofsponsorsupportforMMFs,thepracticeisdiscretionary,unregulated, and opaque, and it is probably most unreliable when systemic risks are most salient. Indeed, otherformsofdiscretionaryfinancialsupportdidnotfarewellduringthefinancialturmoil,such asdealers’supportforauction-ratesecurities(HanandLi,2009). Thus, my findings argue for additional attention to the systemic risks posed by the MMF industry’s reliance on sponsor support. However, a full discussion of the strengths and weaknesses of the current “system” of discretionary sponsor support for MMFs is well beyond the scopeofthispaper. Simplisticresponsestotherisksposedbysponsorsupport—suchasabanon support actions—maydo more harm thangood. The SEC’s action in2010 to requiremore extensive disclosure of sponsor interventions should at least improve researchers’ and policymakers’ understandingofthepractice. tiallydifferentinvestorrisksandliquidityneeds,butsetasinglerequirementforbothtypesoffunds,inpartbecause ofthe”difficultyindrawingbrightlines”betweenthem(seenote11). Still,theSECindicatedthatitwouldcontinue tolookfora“workableobjectivedefinitionthatwouldaccuratelyidentifyfundswithlowerliquidityneeds”(U.S.SecuritiesandExchangeCommission,2010, pp. 58-61). Ofcourse, proxiesthatareusefulformeasuringriskmayhave drawbacksifusedtodetermineregulatorydistinctions: AlthoughMMFswithlowerexpenseratiosexhibitedgreater investorrisksduringtherun, thisauthorwouldnotsuggestimposingmorestringentliquiditystandardsforMMFs becausetheychargelowerfees! 35

References Baba, Naohiko, Robert N. McCauley, and Srichander Ramaswamy, “US dollar money market fundsandnon-USbanks,”BISQuarterlyReview,March2009,pp.65–81. Carpenter,SethB.andJoeLange,“MoneyDemandandEquityMarkets,”WorkingPaper2003-3, FederalReserveBoardFinanceandEconomicsDiscussionSeries,February2003. Chevalier,JudithandGlennEllison,“RiskTakingbyMutualFundsasaResponsetoIncentives,” JournalofPoliticalEconomy,December1997,105(6),1167–2000. Christoffersen, Susan E. K., “Why Do Money Fund Managers Voluntarily Waive Their Fees?,” JournalofFinance,June2001,56(3),1117–1140. andDavidK.Musto,“DemandCurvesandthePricingofMoneyManagement,”TheReviewof FinancialStudies,Winter2002,15(5),1499–1524. Collins, Sean S. and Phillip R. Mack, “Avoiding Runs in Money Market Mutual Funds: Have RegulatoryReformsReducedthePotentialforaCrash,”WorkingPaper94-14,FederalReserve BoardFinanceandEconomicsDiscussionSeries,June1994. Condon, Christopher, “BNY Mellon Cash Fund Hit by Losses from Lehman Debt (Update 1),” Bloomberg.com, 18 September 2008. http://www.bloomberg.com/apps/news?pid= 20601087&refer=home&sid=aNriOV.A5w1E. Covitz,DanielM.,NellieLiang,andGustavoSuarez,“TheEvolutionofaFinancialCrisis: Panic intheAsset-BackedCommercialPaperMarket,”WorkingPaper2009-36,FederalReserveBoard FinanceandEconomicsDiscussionSeries,18August2009. Crane, Peter G., “Money Market Mutual Fund Update: In the Aftermath of Breaking the Buck,” 13January2009. PresentationattheFederalReserveBoard,Washington,DC. CreditSuisse,“AnnualReport2007,”2008. ,“AnnualReport2008,”2009. DeGennaro,RamonP.andDaleL.Domian,“MarketEfficiencyandMoneyMarketFundPortfolioManagers: BeliefsVersusReality,”TheFinancialReview,May1996,31(2),453–474. Del Guercio, Diane and Paula A. Tkac, “The Determinants of the Flow of Funds of Managed Portfolios: Mutual Funds vs. Pension Funds,” Journal of Financial and Quantitative Analysis, December2002,37(4),523–557. 36

Domian, Dale L., “Money Market Mutual Fund Maturity and Interest Rates,” Journal of Money, CreditandBanking,November1992,24(4),519–527. and William Reichenstein, “Performance and Persistence in Money Market Fund Returns,” FinancialServicesReview,November1998,6(5),169–183. Duygan-Bump,Burcu,PatrickM.Parkinson,EricS.Rosengren,GustavoA.Suarez,andPaulS. Willen, “How Effective Were the Federal Reserve Emergency Liquidity Facilities? Evidence from the Asset-Backed Commercial Paper Money Market Mutual Fund Liquidity Facility,” WorkingPaperQAU10-3,FederalReserveBankofBoston,29April2010. Eaton, Leslie, “New Caution about Money Market Funds,” The New York Times, 29 September 1994. Farinella,JosephA.andTimothyW.Koch,“SeasonalPatternsinMoneyMarketMutualFunds,” ReviewofQuantitativeFinanceandAccounting,May2000,14(3),261–276. Federated Investors and Putnam Investments, “Federated Investors, Inc. and Putnam InvestmentsAnnounceTransactiontoBenefitMoneyMarketFundShareholders,”25September2008. http://www.sec.gov/Archives/edgar/data/1056288/000119312508200683/dex991.htm. FitchRatings,“GlobalMoneyMarketFundRatingCriteria,”5October2009. Friend,Irwin,MarshallBlume,andJeanCrockett,MutualFundsandOtherInstitutionalInvestors: ANewPerspective,McGraw-Hill,1970. Gorton,GaryandGeorgePennacchi,“MoneyMarketFundsandFinanceCompanies: AreThey the Banks of the Future?,” in Michael Klausner and Lawrence J. White, eds., Structural Change inBanking,NewYorkUniversity,1993,pp.173–214. Han,SongandDanLi,“LiquidityCrisis,Runs,andSecurityDesign: LessonsfromtheCollapseof theAuctionRateSecuritiesMarket,”WorkingPaper,FederalReserveBoard,15February2009. Hortac¸su,AliandChadSyverson,“ProductDifferentiation,SearchCosts,andCompetitioninthe Mutual Fund Industry: A Case Study of S&P 500 Index Funds,” Quarterly Journal of Economics, May2004,119(2),403–456. Investment Company Institute, “Report of the Money Market Working Group,” available at http://www.ici.org/pdf/ppr_09_mmwg.pdf, 17March2009. , “Investment Company Fact Book,” available at http://www.ici.org/pdf/2010_factbook. pdf, 2010. 37

Ippolito, Richard A., “Consumer Reaction to Measures of Poor Quality: Evidence from the MutualFundIndustry,”JournalofLawandEconomics,April1992,35(1),45–70. Jank, Stephan and Michael Wedow, “Sturm und Drang in money market funds: when money market funds cease to be narrow,” Discussion Paper Series 2: Banking and Financial Studies, No.2008/20,DeutscheBundesbank,2008. Koppenhaver, Gary D., “Circle Unbroken: Bank-Affiliated Money Market Mutual Funds,” in “Global Financial Crises: Implications for Banking and Regulation, Proceedings of the 35th AnnualConferenceonBankStructureandCompetition”FederalReserveBankofChicago,May 1999,pp.430–447. Lyon,AndrewB.,“MoneyMarketFundsandShareholderDilution,”JournalofFinance,September 1984,39(5),1011–1020. Miles, William, “Can Money Market Mutual Funds Provide Sufficient Liquidity to Replace DepositInsurance?,”JournalofEconomicsandFinance,Fall2001,25(3),328–342. Moody’s Investors Service, “Frequently Asked Questions about Moody’s Ratings of Managed Funds,”20July2005. ,“ParentalSupportinMoneyMarketFunds: Moody’sPerspective,”November2008. Nocera,Joseph,APieceoftheAction: HowtheMiddleClassJoinedtheMoneyClass,Simon&Schuster, 1994. Pennacchi,George,“DepositInsurance,BankRegulation,andFinancialSystemRisks,”Journalof MonetaryEconomics,January2006,53(1),1–30. Scott, Kenneth E., “Mutual Funds as an Alternative Banking System,” Journal of Institutional and TheoreticalEconomics,March1998,154(1),86–96. Sirri,ErikR.andPeterTufano,“CostlySearchandMutualFundFlows,”JournalofFinance,October1998,53(5),1589–1622. Smith, Keith V., “Is Fund Growth Related to Performance?,” Journal of Portfolio Management, Spring1978,5,49–54. Smith, RandallandTomLauricella, “HowTurmoilMeltedaMoneyFund,” TheWallStreetJournal,29December2007. Standard&Poor’s,“FundRatingsCriteria,”2007. , “Money-Market Fund Support Weighs on Fund Managers and Bank Sponsors,” 4 December 2008. 38

, “The U.S. Asset-Backed Commercial Paper Market May Be Down, But It’s Not Out,” 7 July 2008. Stigum,MarciaandAnthonyCrescenzi,Stigum’sMoneyMarket,4thed.,McGraw-Hill,2007. The Reserve, “Reserve Primary Fund Makes Initial Distribution of $26 billion to Primary Fund Shareholders,” 30 October 2008. http://www.reservefunds.com/pdfs/Press%20Release% 20PrimDist%202008_1030.pdf. U.S. Securities and Exchange Commission, “Final Rule: Registration Form Used by Open-End ManagementInvestmentCompanies,”availableatwww.sec.gov/rules/final/33-7512r.htm, 18March1998. FileNo.S7-10-97. ,“IntheMatterofCraigS.VanucciandBrianK.Andrew,Respondents: OrderInstitutingPublic AdministrativeandCease-and-DesistProceedings,” 11January1998. AdministrativeProceedingFileNo.3-9804. ,“FinalRule: AmendmentstoInvestmentCompanyAdvertisingRules,”availableatwww.sec. gov/rules/final/33-8294.htm, 1October2003. FileNo.S7-17-02. , “Money Market Fund Reform: Proposed Rule,” available at www.sec.gov/rules/proposed/ 2009/ic-28807.pdf, 30June2009. Releaseno.IC-28807. , “Securities and Exchange Commission v. Reserve Management Company, Inc. et al,” 5 May 2009. CivilActionNo.09-CV-4346. , “Money Market Fund Reform: Final Rule,” available at www.sec.gov/rules/final/2010/ ic-29132.pdf, 23February2010. Releaseno.IC-29132. Wharton School, University of Pennsylvania, “A Study of Mutual Funds,” 1962. Report of the CommitteeonInterstateandForeignCommerce(87thCongress,2ndSession,HouseReportNo. 2274). 39

Table 1. Gross yields, net yields, and expense ratios of prime MMFs, 2004‐2008 (percent) (1) (2) Cross‐sectional MMF characteristic Mean1 standard deviation2 1. Gross yield 3.57 0.12 2. Net yield 3.08 0.29 3. Expense ratio 0.49 0.27 Notes. 1. Sample averages of unadjusted fund‐level annual data for all prime MMFs (1348 observations over 5 years). 2. For each variable, the standard deviation is calculated after subtracting from each observation the annual asset‐weighted average for that variable for all prime MMFs in the same year.

Table 2a. MMF characteristics and risk proxies before and during the run in 2008 Institutional funds (N=116) Retail funds (N=135) (1) (2) (3) (4) (5) (6) (7) (8) Mean S.D. 10th pctl. 90th pctl. Mean S.D. 10th pctl. 90th pctl. Portfolio risk proxies 1. Gross yield1 3.88 0.14 3.69 4.03 3.88 0.16 3.68 4.06 2. Triple‐A rating?2,3 0.53 0.50 0.00 1.00 0.20 0.40 0.00 1.00 3. Lehman exposure?3 0.12 0.33 0.00 1.00 0.12 0.32 0.00 1.00 Investor risk proxies 4. Expense ratio1 0.29 0.18 0.13 0.51 0.63 0.20 0.39 0.91 5. Growth (log net flow), previous year1,4 13.31 37.18 ‐34.36 56.69 4.11 29.26 ‐18.52 26.12 6. Standard deviation of weekly flow1 5.13 3.18 1.78 8.98 2.76 3.20 0.65 5.52 7. Weekly flow sensitivity to relative yield1 6.91 14.24 ‐5.54 25.68 1.51 5.38 ‐4.37 8.87 Sponsor risk proxies 8. Bank‐affiliated fund?3,5 0.53 0.50 0.00 1.00 0.48 0.50 0.00 1.00 9. Average CDS spread, Sept. 2‐9, 20086 1.40 0.81 0.84 2.18 1.39 0.95 0.81 2.18 Other 10. Log of assets (in $ billions) as of Sept. 9, 2008 1.25 1.68 ‐1.07 3.38 0.41 1.50 ‐1.58 2.40 11. Has CDS spread data, Sept. 2‐9, 2008?3 0.43 0.50 0.00 1.00 0.43 0.50 0.00 1.00 Memo: Log net flow, Sept. 9 ‐ Oct. 7, 20084 ‐25.81 31.13 ‐62.31 5.30 ‐4.98 14.06 ‐15.63 8.75 Notes. 1. Year from September 1, 2007 to August 31, 2008. 2. As of August 31, 2008. 3. Indicator variable: 1=yes and 0=no. 4. Log net flow is 100 times the natural logarithm of the sum of 1 and the ratio of net flow over this episode to assets at beginning of episode, that is, 100*ln(1+flow/lagged assets). 5. Bank‐affiliation data are as of September 14, 2008 (when Bank of America purchased Merrill Lynch). 6. CDS spreads are available only for 50 observations for institutional funds and for 58 observations for retail funds.

Table 2b. MMF characteristics and risk proxies before and during the ABCP crisis in 2007 Institutional funds (N=116) Retail funds (N=133) (1) (2) (3) (4) (5) (6) (7) (8) Mean S.D. 10th pctl. 90th pctl. Mean S.D. 10th pctl. 90th pctl. Portfolio risk proxies 1. Gross yield1 5.37 0.02 5.34 5.39 5.37 0.03 5.32 5.40 2. Triple‐A rating?2,3 0.53 0.50 0.00 1.00 0.18 0.39 0.00 1.00 3. Distressed ABCP exposure?3 0.57 0.50 0.00 1.00 0.44 0.50 0.00 1.00 Investor risk proxies 4. Expense ratio1 0.29 0.17 0.13 0.51 0.65 0.20 0.42 0.92 5. Growth (log net flow), previous year1,4 18.80 36.98 ‐14.39 62.73 2.41 32.91 ‐21.13 26.34 6. Standard deviation of weekly flow1 5.16 3.18 2.04 9.37 2.54 2.71 0.71 4.74 7. Weekly flow sensitivity to relative yield1 2.85 65.48 ‐56.38 72.36 ‐1.59 18.07 ‐18.55 14.09 Sponsor risk proxies 8. Bank‐affiliated fund?2,3 0.52 0.50 0.00 1.00 0.47 0.50 0.00 1.00 9. Average CDS spread, August 1‐7, 20075 0.41 0.25 0.20 0.83 0.35 0.19 0.18 0.58 Other 10. Log of assets (in $ billions) as of August 7, 2007 1.22 1.57 ‐0.97 3.15 0.35 1.53 ‐1.57 2.39 11. Has CDS spread data, August 1‐7, 2007?3 0.53 0.50 0.00 1.00 0.47 0.50 0.00 1.00 Memo: Log net flow, August 7‐28, 20074 ‐0.80 11.50 ‐14.41 10.32 0.46 10.33 ‐6.03 7.64 Notes. 1. Year from July 1, 2006 to June 30, 2007. 2. As of June 30, 2007. 3. Indicator variable: 1=yes and 0=no. 4. Log net flow is 100 times the natural logarithm of the sum of 1 and the ratio of net flow over this episode to assets at beginning of episode, that is, 100*ln(1+flow/lagged assets). 5. CDS spreads are available only for 61 observations for institutional funds and for 63 observations for retail funds.

Table 3a. Correlations of MMF characteristics and risk proxies before and during the run in 2008 (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) A. Institutional Gross Aaa Held Expense Growth SD of Flow Bank‐ CDS Log Has MMFs (N=116) yield rated? Lehman? ratio prev. yr. flow sens. affil.? spread1 assets CDS? 1. Gross yield 1 2. Triple‐A rating? 1 3. Lehman exposure? 1 4. Expense ratio ‐0.345 * 1 5. Growth, prev. year 1 6. SD weekly flow ‐0.338 * 0.223 ‐0.319 * 1 7. Flow sensitivity 0.199 ‐0.254 * 1 8. Bank‐affiliated? 1 9. CDS spread1 1 10. Log of assets 0.285 * 0.261 * ‐0.338 * 0.257 * 1 11. Has CDS spreads? 0.186 1 12. Log flow (event) ‐0.311 * 0.424 * ‐0.368 * ‐0.310 ‐0.537 * (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) B. Retail Gross Aaa Held Expense Growth SD of Flow Bank‐ CDS Log Has MMFs (N=135) yield rated? Lehman? ratio prev. yr. flow sens. affil.? spread2 assets CDS? 1. Gross yield 1 2. Triple‐A rating? ‐0.237 * 1 3. Lehman exposure? 0.188 1 4. Expense ratio 1 5. Growth, prev. year 1 6. SD weekly flow ‐0.247 * 0.377 * 1 7. Flow sensitivity 0.188 1 8. Bank‐affiliated? 0.185 0.204 1 9. CDS spread2 ‐0.272 1 10. Log of assets 0.196 0.188 1 11. Has CDS spreads? 0.272 * 1 12. Log flow (event) ‐0.297 * 0.234 * ‐0.490 * ‐0.183 ‐0.237 * ‐0.172 Notes. See table 2a for variable descriptions. Correlations shown are significant at the 5 percent level (* indicates signficance at the 1 percent level). 1. Based on 50 observations with CDS spreads. 2. Based on 58 observations with CDS spreads.

Table 3b. Correlations of MMF characteristics and risk proxies before and during the ABCP crisis in 2007 (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) A. Institutional Gross Aaa ABCP Expense Growth SD of Flow Bank‐ CDS Log Has MMFs (N=116) yield rated? problem? ratio prev. yr. flow sens. affil.? spread1 assets CDS? 1. Gross yield 1 2. Triple‐A rating? 1 3. ABCP exposure? 0.295 * 1 4. Expense ratio ‐0.323 * 1 5. Growth, prev. year ‐0.319 * 1 6. SD weekly flow ‐0.307 * 1 7. Flow sensitivity 0.255 * 0.290 * ‐0.291 * 1 8. Bank‐affiliated? ‐0.272 * 1 9. CDS spread1 ‐0.554 * 1 10. Log of assets 0.285 * 0.260 * 0.259 * ‐0.326 * 0.275 * ‐0.199 1 11. Has CDS spreads? 0.326 * 1 12. Log flow (event) 0.213 ‐0.252 * ‐0.263 ‐0.238 * (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) B. Retail Gross Aaa ABCP Expense Growth SD of Flow Bank‐ CDS Log Has MMFs (N=133) yield rated? problem? ratio prev. yr. flow sens. affil.? spread2 assets CDS? 1. Gross yield 1 2. Triple‐A rating? 1 3. ABCP exposure? 0.266 * 1 4. Expense ratio ‐0.188 1 5. Growth, prev. year ‐0.170 1 6. SD weekly flow ‐0.179 0.426 * ‐0.172 1 7. Flow sensitivity ‐0.227 * 1 8. Bank‐affiliated? 0.189 0.272 * 0.178 ‐0.369 * 1 9. CDS spread2 ‐0.302 1 10. Log of assets 0.292 * ‐0.179 0.257 * ‐0.236 * 1 11. Has CDS spreads? 0.289 * 0.184 0.381 * 1 12. Log flow (event) ‐0.172 ‐0.174 Notes. See table 2b for variable descriptions. Correlations shown are significant at the 5 percent level (* indicates signficance at the 1 percent level). 1. Based on 61 observations with CDS spreads. 2. Based on 63 observations with CDS spreads.

Table 4. Regression results: MMF risk proxies and net flows in September and October 2008 Model: No CDS Spreads Includes CDS spread (1) (2) (3) (4) (5) (6) Investor type: Institutional * Retail Institutional * Retail Portfolio risk proxies 1. Gross yield (percentage points) ‐67.60 *** ‐4.05 ‐161.23 *** ‐5.25 September 2007 ‐ August 2008 (‐4.40) (‐0.76) (‐4.69) (‐0.49) 2. Triple‐A rating? 3.43 ‐3.26 6.73 ‐0.46 (1=yes, 0=no) (0.77) (‐0.76) (0.93) (‐0.07) 3. Lehman exposure? 3.50 ‐0.30 1.85 ‐0.92 (1=yes, 0=no) (0.48) (‐0.17) (0.20) (‐0.20) Investor risk proxies 4. Expense ratio (percentage points) 39.30 ** 11.11 57.46 ** 15.08 September 2007 ‐ August 2008 (3.48) (2.44) (3.58) (1.55) 5. Growth (log net flow) 0.06 0.04 0.04 0.01 September 2007 ‐ August 2008 (0.83) (0.86) (0.38) (0.08) 6. Standard deviation of weekly flow ‐2.06 ‐1.95 ‐3.00 ‐2.14 (percentage points) (‐2.84) (‐2.77) (‐1.67) (‐2.78) 7. Weekly flow sensitivity to ‐0.38 ‐0.28 ‐0.30 0.06 relative net yield (‐2.34) (‐1.10) (‐1.00) (0.23) Sponsor‐risk proxies 8. Bank affiliated? 4.62 ‐3.58 12.91 ** ‐6.61 (1=yes, 0=no) (1.00) (‐1.73) (1.92) (‐1.58) 9. Average CDS spread (pctg. points) ‐15.85 ** ‐1.05 September 2‐9, 2008 (‐2.84) (‐0.69) Other 10. Log of assets ‐8.12 *** ‐1.95 ‐8.29 ** ‐3.78 September 9, 2008 (‐5.74) (‐2.99) (‐4.51) (‐5.13) 11. Constant 242.97 *** 12.56 619.83 *** 19.52 (4.06) (0.62) (4.44) (0.45) Number of observations 116 135 50 58 R‐squared 0.501 0.369 0.649 0.475 Notes. Dependent variable is log flow from September 9 to October 7, 2008. Log flow is defined as 100*ln(1+flow/lagged assets), where flow is equal to the change in assets less accrued yield. t‐statistics in parentheses are based on robust standard errors. * t‐tests of the hypothesis that coefficients for institutional and retail funds are the same, based on a pooled institutional‐ retail fund regression with interactive coefficients (an indicator variable for institutional funds multiplied by each coefficient). */**/*** denotes significance at the 10/5/1 percent level.

Table 5. Regression results: Estimated effects of 1‐standard‐deviation changes in explanatory variables on net flows during the run in 2008 (percentage points of flow) Model: No CDS Spreads Includes CDS spread (1) (2) (3) (4) Investor type: Institutional Retail Institutional Retail Portfolio risk proxies 1. Gross yield (percentage points) ‐8.9 ‐0.6 ‐16.4 ‐0.8 September 2007 ‐ August 2008 2. Triple‐A rating?* 3.5 ‐3.2 7.0 ‐0.5 3. Lehman exposure?* 3.6 ‐0.3 1.9 ‐0.9 Investor risk proxies 4. Expense ratio (percentage points) 7.2 2.2 10.3 2.8 September 2007 ‐ August 2008 5. Growth (log net flow) 2.3 1.2 1.9 0.2 September 2007 ‐ August 2008 6. Standard deviation of weekly flow ‐6.3 ‐6.0 ‐8.1 ‐8.9 7. Weekly flow sensitivity to ‐5.2 ‐1.5 ‐4.2 0.3 relative net yield Sponsor‐risk proxies 8. Bank affiliated?* 4.7 ‐3.5 13.8 ‐6.4 9. Average CDS spread (pctg. points) ‐12.0 ‐1.0 September 2‐9, 2008 Other 10. Log of assets ‐12.8 ‐2.9 ‐12.9 ‐5.4 September 9, 2008 Notes. Estimates based on regressions summarized in table 4. For each continuous variable, estimated effect is computed by multiplying the estimated coefficient on the variable by the standard deviation of the variable. *Estimated effect for each dummy variable is the estimated coefficient on the variable converted to percentage points of assets.

Table 6. MMF risk proxies, instruments for gross yield, and net flows to institutional prime MMFs in September and October 2008 (1) (2) Specification OLS 2SLS1 Portfolio risk proxies 1. Gross yield (percentage points) ‐67.60 ‐54.74 September 2007 ‐ August 2008 (‐4.40) (‐3.10) 2. Triple‐A rating? 3.43 2.92 (1=yes, 0=no) (0.77) (0.63) 3. Lehman exposure? 3.50 0.52 (1=yes, 0=no) (0.48) (0.07) Investor risk proxies 4. Expense ratio (percentage points) 39.30 38.80 September 2007 ‐ August 2008 (3.48) (3.31) 5. Growth (log net flow) 0.06 0.06 September 2007 ‐ August 2008 (0.83) (0.84) 6. Standard deviation of weekly flow ‐2.06 ‐1.63 (percentage points) (‐2.84) (‐1.91) 7. Weekly flow sensitivity to ‐0.38 ‐0.39 relative net yield (‐2.34) (‐2.11) Sponsor‐risk proxies 8. Bank affiliated? 4.62 4.20 (1=yes, 0=no) (1.00) (0.91) Other 9. Log of assets ‐8.12 ‐8.51 September 9, 2008 (‐5.74) (‐5.82) 10. Constant 242.97 191.92 (4.06) (2.74) Number of observations 116 112 R‐squared 0.501 0.500 Notes. Dependent variable is log flow from September 9 to October 7, 2008. Log flow is defined as 100*ln(1+flow/lagged assets), where flow is equal to the change in assets less accrued yield. t‐statistics in parentheses are based on robust standard errors. 1. Instruments for gross yield are portfolio shares of various instruments (Treasury and agency securities, repurchase agreements, time deposits, ABCP, floating rate notes, other domestic‐bank obligations, other foreign‐ bank obligations), WAM, and the share of assets maturing in seven days or less. The instruments are listed on lines 1‐7 and lines 9‐10 of table A1 in the appendix.

Table 7. Regression results: MMF risk proxies and net flows in August 2007 Model: No CDS Spreads Includes CDS spread (1) (2) (3) (4) (5) (6) Investor type: Institutional * Retail Institutional * Retail Portfolio risk proxies 1. Gross yield (percentage points) 7.42 44.19 69.76 ‐1.57 August 2006 ‐ July 2007 (0.17) (1.13) (1.26) (‐0.07) 2. Triple‐A rating? 3.79 1.44 2.21 0.80 (1=yes, 0=no) (1.56) (0.78) (0.68) (0.31) 3. Distressed ABCP exposure? ‐1.91 ‐3.62 ‐2.59 ‐3.59 (1=yes, 0=no) (‐0.66) (‐1.11) (‐0.53) (‐2.23) Investor risk proxies 4. Expense ratio (percentage points) 10.46 ** ‐7.42 18.11 ** ‐8.00 August 2006 ‐ July 2007 (1.82) (‐1.36) (1.50) (‐1.83) 5. Growth (log net flow) ‐0.06 0.04 ‐0.06 0.05 August 2006 ‐ July 2007 (‐1.39) (0.47) (‐0.90) (2.04) 6. Standard deviation of weekly flow ‐0.03 ‐0.72 0.27 ‐0.86 (percentage points) (‐0.09) (‐1.70) (0.42) (‐2.43) 7. Weekly flow sensitivity to 0.03 ** ‐0.07 0.02 ** ‐0.10 relative net yield (1.64) (‐1.66) (0.68) (‐2.47) Sponsor‐risk proxies 8. Bank affiliated? 0.88 ‐0.37 ‐1.77 ‐2.21 (1=yes, 0=no) (0.44) (‐0.22) (‐0.53) (‐1.51) 9. Average CDS spread (pctg. points) ‐14.46 0.87 August 1‐7, 2007 (‐1.64) (0.22) Other 10. Log of assets ‐1.46 ‐0.35 0.08 ‐0.08 August 7, 2007 (‐1.78) (‐0.67) (0.07) (‐0.17) 11. Constant ‐41.95 ‐228.67 ‐374.47 18.84 (‐0.18) (‐1.11) (‐1.27) (0.16) Number of observations 116 133 61 63 R‐squared 0.159 0.135 0.221 0.414 Notes. Dependent variable is log flow from August 7 to August 28, 2007. Log flow is defined as 100*ln(1+flow/lagged assets), where flow is equal to the change in assets less accrued yield. t‐statistics in parentheses are based on robust standard errors. * t‐tests of the hypothesis that coefficients for institutional and retail funds are the same, based on a pooled institutional‐ retail fund regression with interactive coefficients (an indicator variable for institutional funds multiplied by each coefficient). */**/*** denotes significance at the 10/5/1 percent level.

Table 8. Summary statistics for sponsor support and exposures to distressed ABCP (as of July 31, 2007, unless noted) (1) (2) (3) (4) All prime Funds that Funds that Funds that either money received sponsor held received sponsor market support for distressed support for or held funds distressed ABCP ABCP distressed ABCP 1. Number of funds 249 39 124 128 (share of all MMFs, percent) (100) (16) (50) (51) 2. Number of institutional funds 116 20 66 68 (share of line 1, percent) (47) (51) (53) (53) 3. Average assets under management 6.66 7.77 8.81 8.81 (billions of dollars) 4. Average gross yield, year ending 5.37 5.38 5.38 5.38 July 31, 2007 (percent) 5. Number of triple‐A rated funds 85 19 45 46 (share of line 1, percent) (34) (49) (36) (36) 6. Average expense ratio, year ending 0.48 0.43 0.47 0.47 July 31, 2007 (percent) 7. Average growth, year ending 10.56 24.09 13.67 14.92 July 31, 2007 (percent) 8. Average standard deviation of log flow, 3.76 4.78 3.77 3.76 year ending July 31, 2007 (percent) 9. Average flow sensitivity to net yield, 0.48 8.56 9.30 10.24 year ending July 31, 2007 10. Number of bank‐affiliated funds 122 26 72 74 (share of line 1, percent) (49) (67) (58) (58) 11. Number of funds with CDS spread data 124 33 75 79 (share of line 1, percent) (50) (85) (60) (62) 12. Average CDS spread 0.38 0.40 0.37 0.38 week of August 1‐7, 2007 (pctg. pts.) 13. Average CDS spread 0.15 0.18 0.15 0.16 year ending July 31, 2007 (pctg. pts.)

Table 9. Probit analysis of sponsor support and holdings of distressed securities Funds that Funds that either received Funds that held received sponsor Criterion: sponsor support for distressed ABCP support for or held distressed ABCP distressed ABCP (1) (2) (3) (4) (5) (6) No CDS Includes No CDS Includes No CDS Includes Explanatory variables Spreads CDS Spreads CDS Spreads CDS Portfolio risk proxies 1. Gross yield (percentage points) 17.44 40.65 14.68 17.04 16.24 24.47 August 2006 ‐ July 2007 (4.34) (4.00) (4.46) (3.33) (4.70) (3.76) 2. Triple‐A rating? 0.25 0.80 ‐0.02 0.19 ‐0.04 0.24 (1=yes, 0=no) (1.05) (2.11) (‐0.08) (0.65) (‐0.20) (0.78) Investor risk proxies 3. Expense ratio (percentage points) ‐0.09 0.33 1.27 1.41 1.21 1.34 August 2006 ‐ July 2007 (‐0.17) (0.39) (2.35) (1.70) (2.18) (1.54) 4. Growth (log net flow) 0.00 0.00 0.00 0.00 0.00 0.01 August 2006 ‐ July 2007 (1.29) (0.50) (‐0.65) (0.41) (‐0.06) (1.41) 5. Standard deviation of weekly flow 0.08 0.07 0.04 0.05 0.05 0.04 (percentage points) (2.29) (1.84) (1.26) (1.24) (1.35) (0.96) 6. Weekly flow sensitivity to 0.00 0.00 0.01 0.00 0.01 0.01 relative net yield (0.76) (‐0.21) (2.53) (1.26) (3.18) (2.16) Sponsor‐risk proxies 7. Bank affiliated? 0.57 1.33 0.67 0.85 0.71 1.14 (1=yes, 0=no) (2.71) (3.61) (3.61) (2.67) (3.80) (3.15) 8. Average CDS spread (pctg. points) ‐0.80 ‐0.34 ‐0.24 August 1‐7, 2007 (‐0.90) (‐0.46) (‐0.31) 9. Average CDS spread (pctg. points) 10.36 3.18 5.21 August 2006 ‐ July 2007 (3.22) (1.48) (1.99) Other 10. Institutional fund? ‐0.38 ‐0.17 0.52 0.15 0.49 0.12 (1=yes, 0=no) (‐1.23) (‐0.43) (2.01) (0.42) (1.86) (0.33) 11. Log of assets 0.10 ‐0.02 0.25 0.28 0.25 0.28 August 7, 2007 (1.24) (‐0.15) (3.96) (3.12) (3.91) (2.81) 12. Constant ‐95.35 ‐221.94 ‐80.29 ‐93.31 ‐88.60 ‐133.64 (‐4.40) (‐4.04) (‐4.53) (‐3.37) (‐4.77) (‐3.80) Number of observations 249 124 249 124 249 124 Pseudo R‐squared 0.161 0.325 0.192 0.216 0.215 0.301 Notes. Dependent variable for each regression is equal to 1 for money market funds that met the criterion indicated in the column header and 0 for all others. z‐statistics based on robust standard errors are in parentheses.

Figure 1 A. Assets under Management in Money Market Funds by Investment Objective Billions of dollars 2200 2200 Weekly 2000 2000 Start of Prime 1800 Government-only ABCP Crisis 1800 1600 Tax-exempt 8/25 1600 1400 1400 1200 1200 1000 1000 800 800 Lehman 600 bankruptcy 600 400 400 200 200 0 0 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 Source: Investment Company Institute. B. Reserve Primary Fund: Assets and Relative Yields* Billions of dollars Percent, annualized 60 0.6 Monthly 50 0.5 40 Assets (left axis) 0.4 Relative net yield (right axis) 30 Relative gross yield (right axis) 0.3 20 0.2 8/31 10 0.1 0 0.0 -0.1 -0.2 -0.3 -0.4 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 Source: iMoneyNet and author’s calculations. *Institutional share classes only. Note: Relative net (gross) yield is net (gross) yield less asset-weighted average net (gross) yield for all institutional prime money market funds. C. Reserve Primary Fund: Market Share and Relative Yields* Percent Percent, annualized 5 0.5 Monthly 4 Market share (left axis) 0.4 Relative net yield (right axis) 3 Relative gross yield (right axis) 0.3 2 0.2 8/31 1 0.1 0 0.0 -0.1 -0.2 -0.3 -0.4 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 Source: iMoneyNet and author’s calculations. *Institutional share classes only. Note: Relative net (gross) yield is net (gross) yield less asset-weighted average net (gross) yield for all institutional prime money market funds.

Figure 2 A. Assets under Management in Prime Money Market Funds by Investor Type Billions of dollars 1600 1600 Weekly Start of Lehman 1400 ABCP Crisis bankruptcy 1400 1200 1200 8/4 1000 1000 Institutional 800 Retail 800 600 600 400 400 200 200 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 Source: Investment Company Institute. B. Aggregate Daily Net Flows to Prime Money Market Funds Billions of dollars Reserve breaks 40 Daily the buck Lehman Treasury Guarantee, CPFF 20 bankruptcy AMLF announced 0 -20 -40 -60 -80 Institutional Retail -100 Total -120 8/26 9/2 9/8 9/12 9/18 9/24 9/30 10/6 10/10 Source: iMoneyNet and author’s adjustments (see section A.1 of the appendix). C. Distribution of Net Flows to Prime Money Market Funds Number of funds 70 Net flow from September 9 to October 7, 2008 60 50 Institutional Retail 40 30 20 10 0 -50 -40 -30 -20 -10 0 10 20 Net flow (percent of assets as of September 9) Source: iMoneyNet. Excludes funds with less than $100 million in assets, funds less than one year old as of August 2008, and funds with asset-reporting problems (see section A.1 of the appendix).

Figure 3a Predicted Effects of Risk Proxies by Week Institutional funds (no CDS spreads) 1. Gross yield Percent 2. Triple-A rating? Percent Upper bound 6 6 Predicted effect Weekly Lower bound Weekly 3 3 0 0 -3 -3 Upper bound -6 Predicted effect -6 Lower bound -9 -9 -12 -12 9/2 9/9 9/16 9/23 9/30 10/7 10/14 9/2 9/9 9/16 9/23 9/30 10/7 10/14 Week ending Week ending 3. Lehman exposure Percent 4. Expense ratio Percent 6 6 3 3 0 0 -3 -3 -6 -6 -9 -9 -12 -12 9/2 9/9 9/16 9/23 9/30 10/7 10/14 9/2 9/9 9/16 9/23 9/30 10/7 10/14 5. Growth, previous year Percent 6. Standard deviation of weekly flow Percent 6 6 3 3 0 0 -3 -3 -6 -6 -9 -9 -12 -12 9/2 9/9 9/16 9/23 9/30 10/7 10/14 9/2 9/9 9/16 9/23 9/30 10/7 10/14 7. Weekly flow sensitivity 8. Bank affiliated? Percent to relative net yield Percent 6 6 3 3 0 0 -3 -3 -6 -6 -9 -9 -12 -12 9/2 9/9 9/16 9/23 9/30 10/7 10/14 9/2 9/9 9/16 9/23 9/30 10/7 10/14 10. Log of assets Percent 6 3 0 -3 -6 -9 -12 9/2 9/9 9/16 9/23 9/30 10/7 10/14 Note.This figure plots predicted effects based on seven weekly regressions described in section 5.3 of the text. The solid line in each panel plots---for each of the seven weeks ending on the specified dates---the predicted effects on net flows of one-standard-deviation increases in the explanatory variable in the panel title. Predicted effects on net flows are expressed as percentages of lagged assets. The dashed lines in each panel are the upper and lower bounds of quasi 95 percent confidence intervals for the predicted effects. Confidence intervals are computed by using a two-standard-error confidence interval for each estimated coefficient.

Figure 3b Predicted Effects of Risk Proxies by Week Institutional funds (including CDS spreads) 1. Gross yield Percent 2. Triple-A rating? Percent Upper bound 10 10 Predicted effect Weekly Lower bound 5 Weekly 5 0 0 -5 -5 Lower bound Predicted effect -10 Upper bound -10 -15 -15 9/2 9/9 9/16 9/23 9/30 10/7 10/14 9/2 9/9 9/16 9/23 9/30 10/7 10/14 Week ending Week ending 3. Lehman exposure Percent 4. Expense ratio Percent 10 10 5 5 0 0 -5 -5 -10 -10 -15 -15 9/2 9/9 9/16 9/23 9/30 10/7 10/14 9/2 9/9 9/16 9/23 9/30 10/7 10/14 5. Growth, previous year Percent 6. Standard deviation of weekly flow Percent 10 10 5 5 0 0 -5 -5 -10 -10 -15 -15 9/2 9/9 9/16 9/23 9/30 10/7 10/14 9/2 9/9 9/16 9/23 9/30 10/7 10/14 7. Weekly flow sensitivity 8. Bank affiliated? Percent to relative net yield Percent 10 10 5 5 0 0 -5 -5 -10 -10 -15 -15 9/2 9/9 9/16 9/23 9/30 10/7 10/14 9/2 9/9 9/16 9/23 9/30 10/7 10/14 9. Average CDS spread, previous week Percent 10. Log of assets Percent 10 10 5 5 0 0 -5 -5 -10 -10 -15 -15 9/2 9/9 9/16 9/23 9/30 10/7 10/14 9/2 9/9 9/16 9/23 9/30 10/7 10/14 Note.This figure plots predicted effects based on seven weekly regressions described in section 5.3 of the text. The solid line in each panel plots---for each of the seven weeks ending on the specified dates---the predicted effects on net flows of one-standard-deviation increases in the explanatory variable in the panel title. Predicted effects on net flows are expressed as percentages of lagged assets. The dashed lines in each panel are the upper and lower bounds of quasi 95 percent confidence intervals for the predicted effects. Confidence intervals are computed by using a two-standard-error confidence interval for each estimated coefficient.

Appendix A.1 AdjustmentstoiMoneyNetdata ImakeseveraltypesofadjustmentstotheiMoneyNetdata. First,somevariablesintheiMoneyNet dataset—notablyMMFratingsandthebank-affiliationofMMFsponsors—representonlycurrent (nothistorical)information. iMoneyNetprovidedhistoricalsnapshotsofthesedatatoassistinmy compilationofreal-timeinformation. Second, I drop several funds from the sample. For the analysis of each crisis episode, I includeonlyMMFswithatleast12monthsofiMoneyNetdatapriortotheonsetofthecrisisand at least $100 million in assets (the average prime MMF had $7 billion in assets on the eve of the Lehmanbankruptcy). Theverysmallfundstypicallyhadflowsthatwerehighlyvolatilerelative totheirassetsand,giventheeconomiesofscaleintheMMFindustry(DomianandReichenstein, 1998),probablyserveddifferentrolesforassetmanagersthanfundsofmoretypicalsize. Foranalysis of the run in 2008, I drop two extreme outliers with inflows exceeding 100 percent of assets (nootherfundinthesamplegrewmorethan28percentduringtherun),althoughdroppingthese funds has no material effect on my results. I also drop five MMFs with flow measurement problemsduringtherun(includingfundsthatdisappeared),butIincludethemwhereappropriate—in some cases, after correcting assets data (see below)—in the analysis of weekly flows discussed in section5.3. Alltold,the54fundsIexcludefromtheanalysisoftheABCPcrisismanaged1percent ofprimeMMFassetsatthetime, andthe50fundsdroppedfromthebaselineanalysisoftherun in2008accountedfor7percentofassetsontheeveofthatcrisis. Third, I make corrections to assets (and net flows) reported for three MMFs in September andOctober2008: theReservePrimaryFund,theFederatedPrimeObligationsFund,andthePutnam Prime Money Market Fund. I use these corrected assets data onlyin the weekly regressions reported in section 5.3 (the funds are excluded from the full-sample regressions). Weekly results aresimilarwhetherdataforthesefundsareincludedornot. According to iMoneyNet data, the Reserve Primary Fund’s assets dropped from $62.6 billion on Friday, September 12, 2008 to $39.8 billion on Monday the 15th, $35.3 billion on the 16th, and$7.1billiononthe17th. However,ReserveannouncedinOctober2008that“[t]heFund’stotal assetshavebeenapproximately$51billionsincethecloseofbusinessonSeptember15”(TheReserve,2008),apparentlybecausethefundfailedtohonorredemptionrequests(U.S.Securitiesand Exchange Commission, 2009b). To estimate the assets of individual Primary Fund share classes (which included both institutional and retail classes) on September 15 and the following days, I distribute the $11.6 billion decline in assets that Reserve reported for September 15 to different shareclassesinproportiontotheirdeclinesinassetsthatdayasreportedbyiMoneyNet. The Putnam Prime Money Market Fund was closed on September 17, and its $12.3 billion in assets were absorbed by the Federated Prime Obligations Fund on September 24 (Federated A-1

Investors and Putnam Investments, 2008). iMoneyNet reports only $8.4 billion in assets for the Putnam fund as of September 17 (down from $12.3 billion the previous day) and a jump in the Federated MMF’s assets from $21.8 billion on September 23 to $34.1 billion on September 24. Hence,totrackaggregatedailyandweeklyassetsandnetflows,IholdthePutnamfund’sassetsat $12.3billionfromSeptember16to23. ThedisappearanceofthePutnamfund’sassetsonSeptember 24 offsets that day’s jump in assets for the Federated fund. However, I do not use net flows datafortheFederatedPrimeObligationsFundafterSeptember23inmycross-sectionalanalyses because the merger appears to have affected net flows to that fund even after the merger. Although the Federated MMF’s assets had declined only 3.1 percent from September 19 (when the Treasury Guarantee and the AMLF were announced) to September 23, assets fell 23.5 percent in the two days following the merger—perhaps as former Putnam investors redeemed shares after losingliquidityforaweek. Fourth,thenarrowsampleperiodsthatIuseintheweeklyregressionsdiscussedinsection 5.3 allow some modifications to the sample employed for the full-period regressions. Several MMFs, including the three discussed above, that are dropped from the full-period regressions becauseofmeasurementproblemscanbeincludedforsomeoftheweeklyregressions. Thisadds four funds to the institutional sample for the first three weeks that I study (they begin to drop out after that) and one additional fund to the retail sample. The institutional and retail samples thatincludeCDSspreadsareunchanged. Inclusionorexclusionoftheadditionalfundsmakesno materialdifferenceintheweeklyregressionresults. In addition, each weekly regression includes CDS spreads measured over the previous week, to capture the effects of any deterioration in sponsors’ financial condition. Finally, two large MMF sponsors (Morgan Stanley and Goldman Sachs) became bank holding companies on September21; intheweeklyregressions, thebank-affiliationdummyissetasofthebeginningof eachweek. A.2 Grossyieldandportfoliorisk Toexplorethelinkbetweengrossyieldandportfolioriskingreaterdetail,Irunapanelregression ofeachprimeMMF’sannualgrossyieldineachyearfrom2004to2008onanumberofitsportfolio characteristics measured in the same year. The first two columns of table A1 provide summary statisticsfortheexplanatoryvariablesintheregression. Lines1through8listportfoliosharesheld indifferenttypesofassets,includingrelativelysafeassets,suchasTreasuryandagencysecurities (line 1) and repurchase agreements (line 2), and some relatively riskier ones, such as ABCP (line 4), floating-rate notes (line 5), and other domestic and foreign bank obligations (lines 6 and 7).47 Portfolioweightedaveragematurity(line9),whichaveraged41daysinthisperiod,andtheshare 47OtherdomesticandforeignbankobligationsarebankliabilitiesotherthantimedepositsandCP.Theseobligations aremostlyjumboCDs. A-2

ofassetsmaturinginsevendaysorless(line10)areindicatorsofportfoliomaturityandliquidity. Ialsoincludeintheregressionthenaturallogofassets(line11),inpartbecauseotherresearchers have found that MMF yields vary with size (see, for example, Domian and Reichenstein, 1998, and Jank and Wedow, 2008). Finally, the regression includes an indicator of whether a fund was institutional(line12). Iestimateasimplepanelregressionoftheform: ri = ΓXi +constant+εi. (A-1) t t t Each variable in the regression is expressed net of its asset-weighted average among all primeMMFsinthesameyear. Forexample, ri isfund i’s“relative”grossyield—thatis, itsgross t yield in year t less the asset-weighted average gross yield of all other prime funds in year t.48 Xi t is a vector of the fund characteristics listed on lines 1 through 12 of table A1 (other CP, line 8, is omittedfromtheregression). Thethirdcolumnofthetableshowsestimationresults. Asindicatedonlines1and2,larger shares of Treasury and agency securities and repurchase agreements depressed gross yields, on average.49 In contrast, larger shares of fund assets held in riskier asset classes, including ABCP, floating-rate notes, and other bank obligations (lines 4 through 7), were associated with greater relativegrossyields. Forexample,a1-standard-deviation(16.86percentagepoint)increaseinthe shareofanMMF’sportfolioheldinfloating-ratenoteswouldbeassociatedwitha3.3basispoint (16.86percentagepoints× 0.0020percent ) increase in gross yield—a substantial predicted effect, percentagepoint given the small variation in gross yields among MMFs during this period (see table 1). Longer WAMalsoincreasedgrossyields(line9),butotherfundcharacteristicshadnosignificanteffecton grossyieldsoverthisperiod. (Thesignificanceoftheportfoliocharacteristicsinexplaininggross yieldsmotivatesmyuseoftheattributeslistedonlines1-7and9-10oftableA1asinstrumentsfor grossyield. Seesection5.2ofthetext.) One potential shortcoming of gross yield as a measure of portfolio risk is that it may not capture concentration risks among a fund’s assets, although the nature of such risks in MMFs suggeststhatfundswithgreaterconcentrationriskstypicallyalsowouldhavehighergrossyields. SEC rules generally prohibit an MMF from holding more than five percent of its assets in the securities of any single non-government issuer, but the private debt instruments held by prime MMFsgenerallyhavebeenheavilyconcentratedinfinancial-sectorissues. Thus,afund’sholdings of private debt may be a crude measure of its concentration risk.50 Since private debt holdings 48Apanelregressionwithtime(year)fixedeffectsyieldssimilarresultsbutamountstocomputingrelativemeasures usingunweightedmeans. 49Theregressioncoefficientontheshareofaparticularassetcategoryrepresentstheestimatedeffectongrossyield ofraisingtheshareofassetsinthatcategoryby1percentagepointwhilereducingbythesameamounttheshareheld inthecategoryexcludedfromtheregression(otherCP). 50Consider,forexample,theassetcategorieslistedintableA1. Privatedebtinstruments(lines3through8)made upanaverageof81percentofMMFportfoliosoverthefive-yearspancoveredbythistable. Ofthat,atleastone-third (timedeposits, ABCP,andotherdomesticandforeignbankobligations)wasfinancial-sectorexposures. Inaddition, A-3

boostMMFs’grossyields,theseyieldslikelywouldvarypositivelywithconcentrationrisks. A.3 Grossyieldandnetflowtomoneymarketfunds To examine the relationship between gross yield and net flows to MMFs, I use monthly data fromiMoneyNetforasampleperiodextendingfromJanuary1997toAugust2008(justbeforethe Lehmanbankruptcy,Reserve’scapitalloss,andtherunonMMFs). ForeachMMFineachmonth, Idecomposegrossyieldintothreecomponents: (1)Theaverageeffectivefederalfundsrate(FFR) prevailing in that month; (2) the MMF’s category-average gross yield, that is, the asset-weighted meangrossyieldforallMMFsofthesamecategoryinthatmonth,lesstheaverageeffectiveFFR; and (3) the MMF’s relative gross yield, which is defined as its gross yield less its category-average grossyield. (Theanalysiscoversjusttwocategoriesofmoneyfunds: institutionalandretailprime MMFs.) I estimate an empirical model of net new cash flow to each fund as a function of these threecomponentsofgrossyieldandseveralothercontrols: 5 f t i = πr t i −1 +γci t−1 +φFFR t−1 +βExpRatio t−1 + ∑ θ s f t i −s +ΓX t i +εi t . (A-2) s=1 The unit of observation for this regression is the MMF-month, and each fund-level observation combinesdataforalloftheshareclassesofthefund.51 Thedependentvariableintheregression, fi,isfundi’slognetflowinmontht.52 t Themodelincludesthethreecomponentsofgrossyield,laggedonemonth: relativegross yield, r t i −1 , category-average gross yield, ci t−1 , and the effective federal funds rate, FFR t−1 . Also included are the expense ratio and four lags of net flow. The vector Xi is a set of additional t controls, including the logarithm of the MMF’s assets (lagged 5 months); the fund’s 12b-1 fee, if any; the fee waiver, if any; a dummy for whether the fund was waiving fees; fund fixed effects; calendar-yeartimefixedeffects;andcalendar-monthseasonalfixedeffects.53 I estimated equation (A-2) separately for institutional and retail MMFs. Results for institutional funds are shown in the first column of table A2. Lines 1 through 3 show estimated coefficientsforthethreecomponentsofgrossyields: relativegrossyields,category-averagegross yields,andtheeffectiveFFR.Mostrelevanttoacross-sectionalanalysisofMMFrisksisthehighly significant estimated effect of relative gross yield, line 1, on net flow. This effect is also economicallyimportant. Basedontheestimatedcoefficientonrelativegrossyieldof7.29,aone-standarddeviationincreaseingrossyield(8.3basispointsoverthissampleperiod)wasassociatedwithan most “other” (unsecured) CP, which accounted for one-third of MMF assets on average, was likely financial-sector issuance,sincetheshareoffinancialCPintotalunsecuredCPaveraged82percentoverthisperiod. 51Seeinnote4inthetext. 52Thatis, fi =100×ln(1+ netflow ).Seenote21inthetext. t laggedassets 53Christoffersen(2001)foundthatfeewaiversplayedanimportantroleinattractingassetstoretailmoneyfunds,in particular.FarinellaandKoch(2000)documentedseasonalpatternsinMMFnetflows. A-4

0.6 percent increase in net flow per month.54 An institutional fund that maintained a gross yield thatwasonestandarddeviationaboveitscategoryaverageforayearwouldhaveexpectedtoattractnetflowsthatwere6percentagepoints(ofassets)largerthantheflowsreceivedbyatypical competitor.55 TheregressionresultsshownintableA2underscoretheincentivesthatMMFadvisershad to boost gross yields, whether they passed the increase on to shareholders (by keeping expense ratios fixed and allowing net yield to rise) or not. Indeed, the results provide some marginally significantevidencethatafundthatraisedgrossyieldandexpensesbyequalamounts—andthus boostedadviserrevenuebutleftnetyieldunchanged—attractedlargersubsequentinflows. Point estimates of coefficients on relative gross yield (line 1) are larger in magnitude than those for expenseratios(line4),andline13ofthetablereportssignificancelevelsforFtestsofthehypothesis thatπ+β = 0,whichisrejectedatthe10percentlevel. Results for retail funds are shown in column 3. For such funds, the advantages of a relativelyhighgrossyieldarestatisticallysignificant, butlesscompellingthanthoseforinstitutional funds. The difference between retail and institutional funds’ sensitivity to relative gross yields is not only statistically significant, as indicated by column 2,56 but also economically important. A retailfundthatmaintainedaone-standard-deviationgross-yieldspreadoveritscategoryaverage forayearwould,allelseequal,haveexpectedtoattractonly2percentagepointsmoreflowthan itsaveragecompetitor. Thisresultsuggeststhatmanagersofretailfundsfacedweakerincentives totakeportfoliorisksthanmanagersofinstitutionalfunds. Retail and institutional investors responded differently to category-average gross yields, too. Larger category-average gross yields tended to draw assets to all retail funds; indeed, retail investors appear to have been about as sensitive to average MMF yield as they were to the relative yields earned by individual funds. In contrast, institutional investors responded to relative grossyieldbutnottocategory-averagegrossyield,suggestingthattheseinvestorstendedtopurchase shares in the highest yielding funds, regardless of the average yield spread (relative to the FFR) achieved by the MMF industry. One interpretation of these results is that for institutional investors, MMFs served as substitutes for direct holdings of money market instruments, which paidyieldsthatmovedcloselywithaverageMMFyields. Thus,onlyrelativeyieldsdifferentiated MMFs from these substitutes. For retail investors, MMFs competed with bank deposits (InvestmentCompanyInstitute,2010,pp. 34-35),whichpaidratesthatdidnotmovecloselywithfunds’ averageyields,socategory-averageyieldsattractedinflowstoretailfunds. 54Thatis,0.6percent=100(e(0.00083×7.29/100)−1). 55Theboosttoannualflowfromahighergrossyieldissomewhatlessthantheannualizedone-monthimpactbecauseofthenegativeestimatedcoefficientsonlaggednetflow. 56Significance levels reported in this column are those for interactive coefficients from a combined regression of institutionalandretailMMFs(notshown)inwhichaninstitutionalfunddummyisinteractedwitheachexplanatory variablelistedontableA2. A-5

Table A1. Gross yields of prime MMFs and selected fund characteristics, 2004‐2008 (annual data) (1) (2) (3) MMF characteristic Cross‐sectional Gross‐yield regression: (units are percent, unless indicated) Mean1 standard deviation2 Est. coeff. and t‐statistic3 Portfolio share held in: 1. Treasury and agency securities 8.06 12.49 ‐0.0019 (‐7.37) 2. Repurchase agreements 10.64 11.73 ‐0.0008 (‐2.66) 3. Time deposits 2.16 4.03 0.0000 (‐0.01) 4. ABCP 12.23 13.51 0.0011 (4.76) 5. Floating‐rate notes 20.05 16.86 0.0020 (10.03) 6. Other domestic‐bank obligations 7.10 8.79 0.0027 (7.85) 7. Other foreign‐bank obligations 6.17 9.82 0.0012 (3.76) 8. Other CP (omitted from regression) 33.60 21.83 9. Portfolio weighted average maturity (days) 41.29 9.83 0.0015 (4.52) 10. Share of assets maturing in seven days 29.97 14.53 ‐0.0002 (‐0.72) 11. Log of assets (in billions of dollars) 0.40 1.59 ‐0.0035 (‐1.75) 12. Instititional‐fund dummy 0.47 0.50 0.0032 (0.54) 13. Constant ‐0.0005 (‐0.10) Number of observations 1348 Adjusted R‐squared 0.252 Notes. 1. Sample averages of unadjusted fund‐level annual data for all prime MMFs. 2. For each variable, the standard deviation is calculated after subtracting from each observation the annual asset‐weighted average for that variable for all prime MMFs in the same year. 3. Dependent variable is each MMFʹs annual gross yield (in percent) in a given year, less the asset‐weighted mean gross yield for all prime MMFs in the same year. Unit of observation is the fund‐year. All explanatory variables are expressed in terms of deviation from annual asset‐weighted averages for all prime MMFs in the same year. t‐statistics in parentheses are based on robust standard errors.

Table A2. Net Flows to Prime Money Market Funds and Gross Yields in Previous Month (1) (2) (3) Institutional Difference significant? Retail funds (p‐value)* funds 1. Relative gross yield (π) 7.29 0.000 1.85 in previous month (5.73) (3.23) 2. Category‐average gross yield ‐0.93 0.015 1.72 in previous month (‐0.95) (3.68) 3. Effective FFR 0.59 0.884 0.62 in previous month (3.34) (6.37) 4. Expense ratio (β) ‐3.27 0.970 ‐3.34 in previous month (‐1.68) (‐4.58) 5. 1st lag of log flow ‐0.09 0.546 ‐0.08 (‐8.23) (‐8.58) 6. 2nd lag of log flow ‐0.04 0.044 ‐0.01 (‐5.15) (‐1.38) 7. 3rd lag of log flow 0.01 0.497 0.00 (1.09) (0.00) 8. 4th lag of log flow ‐0.02 0.133 ‐0.01 (‐4.08) (‐1.48) 9. 5th lag of log assets ‐2.33 0.002 ‐1.56 (‐11.35) (‐10.98) 10. 12b‐1 fee in previous month ‐1.32 0.134 2.63 (‐0.54) (2.67) 11. Fee waiver 1.29 0.628 0.38 in previous month (0.73) (0.59) 12. Any fee waiver in previous month? ‐0.38 0.791 ‐0.28 (1=yes, 0=no) (‐1.11) (‐1.53) 13. p‐value for test that π +β=0 0.071 0.100 Number of observations 15704 22840 Adj. R‐squared 0.066 0.073 Notes. Dependent variable is monthly log flow to prime MMFs from January 1997 to August 2008. Monthly log flow is defined as 100*ln(1+flow/lagged assets). Flow is equal to the change in assets less accrued yield. t‐statistics in parentheses are based on robust standard errors. Model also includes fund fixed effects, calendar‐ year time fixed effects, and calendar‐month seasonal fixed effects. *p‐values based on a pooled institutional‐retail fund regression with interactive coefficients (an indicator variable for institutional funds multiplied by each coefficient).