a sampling-window approach to · 2015-03-03 · 1 1. introduction this note considers an approach...
TRANSCRIPT
![Page 1: A Sampling-Window Approach to · 2015-03-03 · 1 1. Introduction This note considers an approach to constructing Libor fixings using transactions data and multi‐day sampling windows.1](https://reader034.vdocuments.net/reader034/viewer/2022050419/5f8ea6bf84ff263a3951bccb/html5/thumbnails/1.jpg)
This paper presents preliminary fi ndings and is being distributed to economists and other interested readers solely to stimulate discussion and elicit comments. The views expressed in this paper are those of the authors and are not necessarily refl ective of views at the Federal Reserve Bank of New York or the Federal Reserve System. Any errors or omissions are the responsibility of the authors.
Federal Reserve Bank of New YorkStaff Reports
Staff Report No. 596February 2013
Darrell Duffi eDavid Skeie
James Vickery
A Sampling-Window Approach to Transactions-Based Libor Fixing
REPORTS
FRBNY
Staff
![Page 2: A Sampling-Window Approach to · 2015-03-03 · 1 1. Introduction This note considers an approach to constructing Libor fixings using transactions data and multi‐day sampling windows.1](https://reader034.vdocuments.net/reader034/viewer/2022050419/5f8ea6bf84ff263a3951bccb/html5/thumbnails/2.jpg)
Duffie: Stanford University (e-mail: [email protected]). Skeie, Vickery: Federal Reserve Bank of New York (e-mail: [email protected], [email protected]). The authors thank David Hou and Ali Palida for outstanding research assistance, as well as Spence Hilton, Antoine Martin, Jamie McAndrews, and Simon Potter for comments. The views expressed in this paper are those of the authors and do not necessarily reflect the position of the Federal Reserve Bank of New York or the Federal Reserve System.
Abstract
We examine the properties of a method for fixing Libor rates that is based on transactions data and multi-day sampling windows. The use of a sampling window may mitigate problems caused by thin transaction volumes in unsecured wholesale term funding markets. Using two partial data sets of loan transactions, we estimate how the use of different sampling windows could affect the statistical properties of Libor fixings at various maturities. Our methodology, which is based on a multiplicative estimate of sampling noise that avoids the need for interest rate data, uses only the timing and sizes of transactions. Limitations of this sampling-window approach are also discussed.
Key words: shadow banking, financial intermediation
A Sampling-Window Approach to Transactions-Based Libor FixingDarrell Duffie, David Skeie, and James VickeryFederal Reserve Bank of New York Staff Reports, no. 596February 2013JEL classification: G01, G10, G18, G28
![Page 3: A Sampling-Window Approach to · 2015-03-03 · 1 1. Introduction This note considers an approach to constructing Libor fixings using transactions data and multi‐day sampling windows.1](https://reader034.vdocuments.net/reader034/viewer/2022050419/5f8ea6bf84ff263a3951bccb/html5/thumbnails/3.jpg)
1
1.IntroductionThisnoteconsidersanapproachtoconstructingLiborfixingsusingtransactionsdataandmulti‐daysamplingwindows.1Forinstance,onecouldfixthe3‐monthLiborrateonagivendateastheaverageoftheactualinterestratesonall3‐monthloansintherelevanthistoricalsamplewhosetransactionsdatesarewithinthetrailing10businessdays.This“10‐daysamplingwindow”ismerelyforpurposesofillustratingtheconcept.Wewillexaminetheinfluenceofthesamplingwindowonsamplingnoiseandconsideradditionaltechniquesfor“fattening”thesampleandweightingthedatasoastoreducesamplingnoiseandmitigatebiases.Wealsoconsiderthepotentialrangeofapplicationsofthisapproach,andsomeofitsdisadvantages.LiborprovidesanestimateoftheinterestrateatwhichmajorbanksactiveinLondonmayborrowfromotherbanksonanunsecuredbasis.TheBritishBankersAssociation(BBA)currentlyreportsLiboronadailybasisfor10currenciesand15maturitiesbetweenovernightandoneyear.2Thesedailyinterestrate“fixings”areconstructedbasedonbanksubmissions.Eachofapanelofbanksself‐reportsitsownestimatedhypotheticalborrowingratesateachtenor.Notably,Liborisnotcurrentlycomputeddirectlyfromactualloantransactionrates.PublishedLiborratesarereferencedinthesettlementofmanyformsoffinancialcontracts,includingcorporatebondsandloans,mortgages,aswellasinterest‐ratefutures,swapsandoptions.AttentionhasrecentlyfocusedonthepotentialtoaddressshortcomingsofthesurveyapproachtoLiborwithafixingmethodthatissomehowbaseddirectlyonactualloantransactionsdata.Whileadvocatingfortheretentionofasubmission‐basedapproach,theWheatleyReviewofLibor(H.M.Treasury,2012)recommendsthatLiborsubmissionsshouldbe“explicitlyandtransparentlysupportedbytransactiondata.”ItalsooutlinesguidelinesforhowthisprincipleshouldbeimplementedinpracticebyLiborpanelbanks.3Thejudgmentandexpertiseofsubmittingbanksstillplaysaroleunderthisapproach.AnalternativewouldbetocomputeLibordirectlyasanaverageofindividualtransactionrates.Oneconcernoversuchanapproach,however,istherelativesparsenessofdailyinterbankunsecuredloantransactionsatcertainmaturities,
1Liborstandsfor“LondonInterbankOfferedRate”.2ThenumberofcurrenciesandmaturitiesisplannedtobereducedinthefutureinlinewiththerecommendationsoftheWheatleyReviewofLibor(H.M.Treasury,2012).Seesection2.3Theseguidelines(section4.8oftheWheatleyReview)layoutahierarchyoftransactiontypesthatbanksshouldusewhendeterminingtheirsubmissions.Highestpriorityisgiventotransactionsintheunsecuredinterbankdepositmarket,particularlythoseundertakenbythecontributingbank.Intheabsenceofrelevanttransactiondatatheguidelinessuggestthatexpertjudgmentshouldbeusedtodeterminethebank’ssubmission.Theyalsostatethat“submissionsmayalsoincludeadjustmentsinconsiderationofothervariables,toensurethesubmissionisrepresentativeofandconsistentwiththemarketforinter‐bankdeposits”,suchasplacinglessweightonnon‐representativetransactions.
![Page 4: A Sampling-Window Approach to · 2015-03-03 · 1 1. Introduction This note considers an approach to constructing Libor fixings using transactions data and multi‐day sampling windows.1](https://reader034.vdocuments.net/reader034/viewer/2022050419/5f8ea6bf84ff263a3951bccb/html5/thumbnails/4.jpg)
2
particularlyduringperiodsoffinancialstress.Afixingthatisbasedonrelativelyfewtransactionscouldhaveexcessivesamplingnoiseandcouldalsocreateaheightenedincentiveforsomemarketparticipantstotransactwiththepurposeofinfluencingthedailyfixing.(Inastock‐marketcontext,Carhart,Kaniel,Musto,andReed(2002)discussevidenceoftransactionsdesignedto“paintthetape.”)TheWheatleyReport(H.M.Treasury,2012)indicatesthattherearetoofewtransactionstosupportLiborinmanyofthecurrency‐maturitypairsforwhichLiboriscurrentlyreported.4Weshow,however,thatatleastforsomeofthemoreactiveU.S.dollarmaturities,theuseofasampling‐windowapproachwouldsignificantlyreducethenoisinessoftransactions‐basedaverageinterestrates.Thisapproachwouldalsoimproverobustnesstomisreportingincentives.Theapproachcouldbeexploitedeitherasthebasisforanewfixingrateforasubsetofcurrenciesandmaturities,orasasourceofadditionalinformationinjudgingthevalidityofotherfixingmethods.Weillustratethetransaction‐windowapproachusingtwopartialdatasetsmeasuringunsecuredwholesalelendingactivity.Thefirstisahistoricaldatasetofbrokeredinterbankloans.ThesecondisasetofputativeunsecuredloansinferredfromFedwirepaymentsusingastatisticalalgorithmdevelopedbystaffoftheResearchGroupoftheFederalReserveBankofNewYorkthatextendstheworkofFurfine(1999).(SeeKuo,Skeie,VickeryandYoule,2013foradetaileddescriptionofthisalgorithm.)Wenotethatwhilethesedatasetsareusefulforillustratingourapproach,neithercouldbeusedinpracticeasthebasisforconstructingatransaction‐basedindexofbankfundingcosts.Inparticular,weemphasizethattheKuoetal.statisticalalgorithmidentifiesterminterbankloanswitherror.Historically,algorithmsbasedontheworkofFurfinehavebeenusedasamethodofidentifyingovernightortermfederalfundstransactions.TheResearchGroupoftheFederalReserveBankofNewYorkhasrecentlyconcludedthattheoutputofitsalgorithmbasedontheworkofFurfine5maynotbeareliablemethodofidentifyingfederalfundstransactions.6ThispaperthereforereferstothetransactionsthatareidentifiedusingtheResearchGroup’salgorithmasovernightortermloansmadeorintermediatedbybanks.Useoftheterm“overnightortermloansmadeorintermediatedbybanks”inthispapertodescribetheoutputoftheResearchGroup’salgorithmisnotintendedtobeandshouldnotbeunderstoodtobeasubstituteforortorefertofederalfundstransactions.
4Forthisreason,andbecauseoftheirlowusage,theWheatleyReviewrecommendsdiscontinuingLiborfortenorsof4,5,7,8,10and11months,anddiscontinuingLiborentirelyforfivecurrencies.ReportingofLiboristocontinuefortheUSDollar,Euro,JapaneseYen,UKPoundandSwissFranc.5Itshouldbenotedthatforitscalculationoftheeffectivefederalfundsrate,theFederalReserveBankofNewYorkreliesondifferentsourcesofdata,notonthealgorithmoutput.6Theoutputofthealgorithmmayincludetransactionsthatarenotfedfundstradesandmaydiscardtransactionsthatarefedfundstrades.Someevidencesuggeststhatthesetypesoferrorsinidentifyingfedfundstradesbysomebanksmaybelarge.
![Page 5: A Sampling-Window Approach to · 2015-03-03 · 1 1. Introduction This note considers an approach to constructing Libor fixings using transactions data and multi‐day sampling windows.1](https://reader034.vdocuments.net/reader034/viewer/2022050419/5f8ea6bf84ff263a3951bccb/html5/thumbnails/5.jpg)
3
Giventhelimitationsofexistingdatasets,atransaction‐basedindexwouldrequireconstructingacentralizedandauditablerepositoryofrelevantinterbanktransactions.OnepossiblemethodologyforreportingthenecessarydataistheTradeReportingandComplianceEngine(TRACE),developedbyFINRAforthereportingofindividualtradesincertaintypesoffixed‐incomesecurities.Insection4ofthispaperwealsohighlightanumberofconceptuallimitationsofthesamplingwindowapproach,andconsiderpotentialsolutions.Oneimportantissueisthatafixingbasedonalaggedmovingwindowwillreflectstaleinformationduringperiodswhenmarketconditionschangerapidly,forexampleaftermonetarypolicyannouncements,orattheonsetofafinancialcrisis.Forapplicationsthatallowhindsight,suchasex‐postcorroborationofothermethodsforfixingLibor,atwo‐sidedsamplingwindowcouldbeused,incorporatingtransactiondatafromboththedaysbeforeandafterthefixingdate.Thiscouldmitigatethestaleness.Atwo‐sidedsamplingwindowisofcourseinfeasibleifthefixingneedstobepubliclyreleasedinrealtime.Asecondpotentialconcernisthattheavailablesampleofunderlyingwholesaleloantransactionsmaybesmallevenwithamulti‐daysamplingwindow,particularlyduringperiodsofmarketstress.Onewaytomitigatethisproblemcouldbetoconsiderawiderrangeofunsecuredfundinginstrumentswhenconstructingthetransaction‐basedindex.2.WiderSamplingWindowsSupposethereisasourceofactualtransactionsdataonlargeunsecuredloanstobanksinthedesiredborrowerclass.Incasethevolumeofinterbankloantransactionsisviewedasinsufficient,onemaywishtoconsiderawiderrangeofsourcesofunsecured“wholesale”fundingtomajorbanks,perhapsincludingcertificatesofdeposit,commercialpaperandsoon.EvenforaglobalcurrencysuchastheU.S.dollar,thereareextremelyfewlargeunsecuredloantransactionsatmanyofthematuritiesatwhichLiboriscurrentlyfixed.Evenasampling‐windowapproachwouldnotbereliableinsuchcases.Alternativesforthese“sparselypopulated”maturitiesincludeinterpolation,improvingthecurrentsurvey‐basedapproach,oracessationofLiborfixings,asrecommendedbytheWheatleyReviewofLibor.Fortunately,thematuritiesatwhichtherearefewtransactionssuitablefordeterminingareferenceratearealsolessimportantforapplications.Forexample,therearerelativelyfewderivatives,bonds,andotherinstrumentsthatreference9‐monthLibor.ThemostcommonlyreferencedLiborratesinmajorcurrenciesarethosewithmaturitiesofonemonth,threemonthsand,toalesserextent,sixmonths,asindicatedbyasurveyappearingintheWheatleyReviewofLibor.Wefocusoncurrenciesandmaturitiesforwhichtheaggregate‐sampletransactionsfrequencyispotentiallysufficienttoconsiderforafixing,orforvalidationofafixing.
![Page 6: A Sampling-Window Approach to · 2015-03-03 · 1 1. Introduction This note considers an approach to constructing Libor fixings using transactions data and multi‐day sampling windows.1](https://reader034.vdocuments.net/reader034/viewer/2022050419/5f8ea6bf84ff263a3951bccb/html5/thumbnails/6.jpg)
4
EvenforrelativelyactiveU.S.dollarloanmaturitiessuchas1monthand3months,wewillshowthatasubstantialproportionalreductioninthesamplingnoiseassociatedwithatransactions‐basedfixingcanbeachievedwiththeuseofarollingsamplingwindow.Thisisnotsurprising,buttheempiricalmagnitudeoftheeffectisnotable.Moreover,ourmethodologydoesnotrelyonaccesstotheinterest‐ratedatathemselves,butratheronthetimesandsizesoftransactions.Ourapproachisinthespiritofstatisticalfiltersthatattempttoextractlonger‐frequencymovementsintime‐seriesdata(suchastheHodrick‐PrescottfilterortheKalmanfilter).Thisapproachcouldbeemployedinatleasttwoways:1)toprovideareplacementtothecurrentquote‐basedapproachfordeterminingtheLiborfixing;2)incorroborationofaquote‐basedorpoll‐basedLiborfixing,forexampleaspartoftheprocessofstrengtheningoversightofLibor.Inthefirstapplication,itwouldbenecessarytouseaone‐sidedlaggingwindow,sincethefixingwouldneedtobeannouncedinrealtime.Forex‐postvalidationpurposes,however,itwouldbepossibletouseatwo‐sidedsamplingwindowtoconstructthefixing,incorporatingbothpastandfuturedata.7Ournumericalexamplesbelowfocusonaone‐sidedwindow.Fromastatisticalfilteringpointofview,atwo‐sidedsamplingwindowwouldloweraveragethedegreeofsamplingerror.Oursimpleillustrativeexampleisafixingofthe3‐monthrateonagivendateastheaverageoftheratesonall3‐monthloansintherelevanthistoricalsamplewhosetransactionsdateiswithinthetrailing10businessdays.Onemayalsowishtouseasamplingwindowbasedonmaturity.Weelaborateandgeneralizeasfollows.SupposeonewantstocreateanestimateR(t,m)ofa“representative”m‐monthmaturityloanrateondayt.LetS(t,m;w,d)bethesubsetofallloansintheentirerelevanthistoricalsampleavailableonthefixingdatetwhosetransactiondateiswithinthetrailingwdaysandwhosematurityiswithinddaysofm.OnecouldfixR(t,m)asthevolume‐weightedaverageoftheloanratesinthisfixingsampleS(t,m;w,d).Forexample,foralag(w)of10daysandamaturitywindow(d)of5days,thefixingsampleforthethree‐monthborrowingrate(thatis,m=3months)onagivenday(t),sayMarch15,2013,wouldincludealltransactionsintherelevantpoolwithloanoriginationdatesbetweenMarch1,2013andMarch15,2013,inclusive(thatis,laggingbynomorethan10businessdays),withloanmaturitiesofthreemonthsplusorminus5businessdays.Inchoosingthelaggingtransaction‐datewindowwandthecenteredmaturitywindowd,onecantradeoffthebenefitofincreasedsamplesizeagainstthecostofbiasesassociatedwithincreasinglystaleoroff‐maturitydata.Inthelastsection,weexplorethebenefitsandcostsofreducingtheweightsappliedtothetransactionsaccordingtothetimelag,inordertomitigatestalenessbias.Inpractice,therelevanttermloanmaturitiesappeartobetightlyconcentratedaroundthestandardmaturitiesof1month,3month,and6months.Itmaybe7WethankSimonPotterforalertingustothispoint.
![Page 7: A Sampling-Window Approach to · 2015-03-03 · 1 1. Introduction This note considers an approach to constructing Libor fixings using transactions data and multi‐day sampling windows.1](https://reader034.vdocuments.net/reader034/viewer/2022050419/5f8ea6bf84ff263a3951bccb/html5/thumbnails/7.jpg)
5
arguedthatitisrelativelypointlesstouseanon‐trivialmaturitywindow.Ontheotherhand,fatteningupthesamplebyincludingsimilar‐maturityloantransactionswouldlowersamplingnoisesomewhatandseemsunlikelytocreateimportantbiases.Theuseofamaturitywindowalsolowersthepotentialincentiveforloanmarketparticipantswhosetransactionsaresampledtocustomizetheirmaturitydatessoastoavoidenteringthefixingsample.3.EmpiricalMethodsandResultsInthissectionwepresentaproportionalsampling‐noisemeasureandourempiricalevidenceregardinghowvariationinthesamplingwindowandotherdatafiltersaffectsthe“thinness”ofthedataunderlyingapotentialtransaction‐basedLiborindex.3.1DatasourcesWedonothaveaccesstoacomprehensivetransaction‐leveldatabaseofunsecuredwholesaleloans.Intheabsenceofsuchdata,weillustrateourapproachusingtwopartialdatasources:
1. Adatasetofbrokeredinterbanktransactionsfromtheperiod2000‐04.2. Statisticallyinferredtransactionsbasedoninterbanktransfersoffederal
reservespassingoverFedwireFundsService(“Fedwire”),alarge‐valuepaymentsystemoperatedbytheFederalReserve,fromtheperiod2007‐12.
ThefirstofthesedatasourceswaspreviouslyusedbyBartolini,HiltonandMcAndrews(2010)andobtainedfromBGCBrokers,oneofthefourlargestU.S.interbankbrokeragefirms.Thesedatarepresentoneoftheonlydirecttransaction‐levelresearchdatasetsforUS‐dollar‐denominatedinterbankloansavailableforresearch.However,thisdatasethasanumberoflimitations.First,thedataareavailableonlyforahistoricaltimeperiodfromJanuary1,2000untilSeptember27,2004.Thissamplepre‐datesthe2007‐08financialcrisisandthepost‐crisisperiod.Second,thedatacoveronlybrokeredloans,whichrepresentonlyasubsetoftheinterbankmarket,andrepresentonlytradesnegotiatedthroughasinglebroker.Theidentitiesoftradecounterpartiesarenotprovided.Finally,thedatacoveronlyinterbankloans,andthusdonotincludeotherunsecuredfundinginstruments(suchaswholesaletimedeposits)thatmaybeusefulforconstructingatransaction‐basedLiborfixing.TheseconddatasourceisasetoftermloansmadeorintermediatedbybanksinferredfrompaymentspassingoverFedwireusingastatisticalalgorithmdevelopedinKuo,Skeie,Vickery,andYoule(2012)(KSVY).TheKSVYalgorithmisageneralizationofFurfine(1999),whoappliedthemethodtoidentifypotentialovernightloans,nottermloans.TheideabehindtheKSVYalgorithmisthatmostwholesaleinterbankloansaresettledoveralarge‐valuepaymentsystem.Inthe
![Page 8: A Sampling-Window Approach to · 2015-03-03 · 1 1. Introduction This note considers an approach to constructing Libor fixings using transactions data and multi‐day sampling windows.1](https://reader034.vdocuments.net/reader034/viewer/2022050419/5f8ea6bf84ff263a3951bccb/html5/thumbnails/8.jpg)
6
caseofUS‐dollarloans,thisislikelytobeeitherFedwireorClearingHouseInterbankPaymentsSystem(“CHIPS”).TheKSVYalgorithmsearchesfortransactionpairsconsistingofa“send”leg(frompartyAtopartyB)foralargeround‐lotamount,anda“return”leg(fromBtoA)onasubsequentdateforaslightlylargeramount,suchthattheimpliedannualizedinterestrateisawholenumberofbasispointsandsuchthatthetransactionpairmeetscertainothercharacteristics.Forthepurposesofthispaper,thealgorithmisusedtoidentifyputativeinterbanktransactionsforwhichboththesendingandreturnlegpassoverFedwirebetweenJanuary1,2007andMay1,2012.ThemostimportantdisadvantageoftheKSVYinferencesisthatthesetofidentifiedtransactionpairsareinferences,notdirectobservationsoftermloans.Itisdifficulttoverifyatthispointhowwellorpoorlythesepairscorrespondtoactualunsecuredtransactions.KSVYdohoweverpresentsometestssuggestingthattheresultsofthealgorithmareinformative.Forexample,KSVYshowthatpriortotheonsetofthefinancialcrisis,thedistributionofimpliedinterestratesoftheseputativeloansisclusteredtightlyaroundtheLiborfixingrate,implyingthattheresultsarenotstatisticalnoise.Aswediscussedintheintroductiontothispaper,itisimportanttoemphasizethatthismethodissubjecttobothType‐IandType‐IIclassificationerrors(failurestodiscoveractualloans,andinferredloansthatarenotactualloans).OneparticularconcernisthattheproximatecounterpartiesidentifiedintheFedwiredatamaybeactingonlyascorrespondents,ratherthanbeingtheultimateborrowerandlenderoffunds.Thisisespeciallyrelevantifauserofthedatawantstorestricttheirsampletoaparticularsubsetofborrowers.Notably,recentresearchbyArmantierandCopeland(2012)concludesthattherelatedovernightFurfinealgorithmperformspoorlyinidentifyingovernightfederalfundsloansconductedbytwolargebanks.8(Note:FederalfundsloansareasubcategoryofinterbankloanswhicharenotsubjecttoU.S.reserverequirements.)Giventheissuesdescribedabove,weemphasizethatneitherofthedatasourcesweconsidercouldreliablybeusedinpracticeasthebasisforcomputingatransaction‐basedreplacementforLibor.Inpractice,suchafixingwouldpresumablyrequirethecreationofarecordlogofactualwholesaleloans(whetherrestrictedtointerbankloans,orencompassingawidersetofunsecuredinstruments),whichcouldbeaggregatedorauditedbyregulatorsorotheroutsideparties.Inthemeantime,however,intheabsenceofasuitabledatabaseofactualterminterbankloans,ananalysisofthesetwodatasetsprovidesatleastaroughideaoftheeffectofthesizeofthesamplewindowandotherfiltersontherobustnessofthesampling‐windowapproach.Giventhelimitationsofthedatasources,wedonot
8Inpartbecauseoftheseconcernswedonotmakeuseofmeasuredinterestratesinthispaper,foreitherdatasource.Instead,werestrictouruseofthesedatasourcestotransactiontimes,maturities,andsizes.
![Page 9: A Sampling-Window Approach to · 2015-03-03 · 1 1. Introduction This note considers an approach to constructing Libor fixings using transactions data and multi‐day sampling windows.1](https://reader034.vdocuments.net/reader034/viewer/2022050419/5f8ea6bf84ff263a3951bccb/html5/thumbnails/9.jpg)
7
presentsampling‐windowestimatesoftheinterbankrateitself,insteadwefocusonhowasamplingwindowapproachwouldaffecttherelativesamplingnoiseassociatedwithatransaction‐basedinterbankindex.3.2ResultsBearinginmindtheimportantcaveatsdescribedabove,weusethesetwodatasourcestocomputeestimatesoftherelativesamplingnoiseassociatedwithanillustrativeUS‐dollarindexrate,forvariousdatafiltersandmaturities.Figure1andTable1illustratetheeffectofchangingthesamplingwindowfortheimpliedsample‐volatilitymultiplierV(t),aproportionalsamplingnoisemeasurethatisbasedonthenumberandrelativesizesofloansinthefixingsampleS(t,m;w,d).Specifically,V(t)isthesquarerootofthesumofthesquareddollar‐sizeweightsoftheloansinS(t,m;w;d).Forexample,ifthefixingsampleS(t,m;w,d)includestwoloans,ofamounts$40millionand$60million,thentherelativesizeweightsare0.4and0.6.Thesumofthesquaredweightsis0.16+0.36=0.52,soV(t)is0.72.Ifoneweretoassumethat,conditionalon“fundamental”loan‐marketinformation,theindividualloanratesinagivenday’sfixingsampleareuncorrelatedandhavethesamestandarddeviationD(t),thenthefixingR(t,m)hasaconditionalstandarddeviationofD(t)V(t).Undertheseconditions,intheaboveexampleofafixingsamplewithtwoloansofamounts$40millionand$60million,thesamplevolatilitymultiplierof0.72meansthattheassociatedsize‐weightedaverageinterestratehasastandarddeviationthatis72%ofthatforafixingratebasedonasingleloantransaction.Thesestatisticalassumptionsdonotapplyinpracticeandwedonotrelyonthem,butthesample‐volatilitymultiplierV(t)neverthelessgivesusagoodideaoftherelativeeffectofthelengthofthesamplingwindowontherobustnessofthesample.ArelativelyhighsamplingvolatilitymultiplierV(t)meansthattherearerelativelyfewloansdominatingthesample,andthereforelittleopportunityfor“diversification”ofthesamplingnoise.Atitsmaximum,forthecaseofasinglesampledloan,V(t)=1.Asthenumberofloansbecomeslargeandthefractionofanyoneloansizerelativetothetotalquantityofloansbecomessmall,V(t)approacheszero,bythelawoflargenumbers.WeemphasizethatV(t)saysnothingaboutthelevelsorvolatilitiesofinterestratesintheinferred‐loansample.Rather,V(t)isdeterminedentirelybythenumberandrelativesizesoftheloansinthefixingsamplefordatet.Withinterest‐ratedatafromactualtransactions,onecouldalsodirectlystudythesamplestandarddeviationsoftheratesinthefixingsamples,andtheeffectofthesamplingwindowonbiasesandrelativenoise.GiventhepotentialformisclassificationusingtheKSVYalgorithm,weavoidusingtheinferredloaninterestrates.
![Page 10: A Sampling-Window Approach to · 2015-03-03 · 1 1. Introduction This note considers an approach to constructing Libor fixings using transactions data and multi‐day sampling windows.1](https://reader034.vdocuments.net/reader034/viewer/2022050419/5f8ea6bf84ff263a3951bccb/html5/thumbnails/10.jpg)
8
Forthe3‐monthmaturity,Figure1belowplotsthetimeseriesofV(t)basedona10‐daysamplingwindowfromthetwotransaction‐leveldatasources.9
Figure1:Time‐seriesplotofV(t)
ThedailysamplevolatilitymultiplierV(t)for3‐monthmaturityloans.Thesampleisbasedonaminimumtransactionsizeof$25manda10‐daysamplingwindow.Thesampleperiodis2000‐2004forthebrokereddata,and2007‐2012fortheFedwireinferences.A.Brokeredinterbankdata
B.Fedwireinferences
9ThebrokereddatasampleusedtoconstructFigure1aswellassubsequentfiguresandtablesincludesbothEurodollarandtermFederalfundsinferredtransactions(asdiscussedinBartolinietal.,2010,thedatasetincludesaflagwhichindicatesthetransactiontype;weretainbothcategories).Similarly,fortheFedwireinferences,wepresentresultsbasedontheentiredatasetofinterbankloaninferences,ratherthanattemptingtorestrictthesampletoaparticularloantype.
0
.2
.4
.6
.8
1
SVM
01jan2000 01jul2001 01jan2003 01jul2004Date
0
.2
.4
.6
.8
1
SVM
01jan2007 01jan2008 01jan2009 01jan2010 01jan2011 01jan2012Date
![Page 11: A Sampling-Window Approach to · 2015-03-03 · 1 1. Introduction This note considers an approach to constructing Libor fixings using transactions data and multi‐day sampling windows.1](https://reader034.vdocuments.net/reader034/viewer/2022050419/5f8ea6bf84ff263a3951bccb/html5/thumbnails/11.jpg)
9
Figure1showssubstantialvariationovertimeinthedailysample‐volatilitymultiplierV(t),forbothdatasources.Thesample‐volatilitymultipliermeasuredfromthebrokereddataisconsistentlyhigherthanthatforFedwire‐inferreddata.Thisisnotsurprising,giventhatthebrokereddatacaptureonlyasmallsegmentofthemarket(thosebrokeredinterbankloansintermediatedbyasinglebroker).ThedifferenceinV(t)betweenthetwodatasourcescouldalsopartiallyreflectfalse“matches”intheFedwireinferences,differencesinthesampleperiod,andotherfactors.Table1showsthemedianacrosstheperiodofthesamplevolatilitymultiplierV(t),forvariousmaturitiesandsamplingwindowslags,normalizedbythemedianofV(t)for3‐monthmaturityloansandasamplewindowlagof10days.Wevariedthesamplingwindowfromtwodaysto20days,andconsideredmaturitiesof1,3,and6months.(Thenormalizingcellassociatedwitha10‐daysamplingwindowand3‐monthmaturitythusalwaysshowsavalueof1.)Thetablealsoreportssummarystatisticsfromthetwodatasources.Table1:RelativevaluesofV(t)fordifferentmaturitiesandsamplingwindows
MedianvaluesofthesamplevolatilitymultiplierV(t),forvariouscombinationsoflagwindowandmaturity,normalizedbythemedianvalueofV(t)foralagwindowof10daysandamaturityof3months.Thesampleperiodis2000‐2004forthebrokereddata,and2007‐2012fortheFedwireinferences.Brokeredinterbankloans Maturity
1 month 3 months 6 months
Lag window (days) 2 1.04 1.57 2.22
5 0.81 1.31 1.66
10 0.61 1.00 1.36
15 0.51 0.85 1.17
20 0.46 0.77 1.05
Fedwireinferences Maturity
1 month 3 months 6 months
Lag window (days) 2 1.16 1.68 2.50
5 0.88 1.33 2.13
10 0.67 1.00 1.63
15 0.56 0.84 1.37
20 0.50 0.76 1.23
![Page 12: A Sampling-Window Approach to · 2015-03-03 · 1 1. Introduction This note considers an approach to constructing Libor fixings using transactions data and multi‐day sampling windows.1](https://reader034.vdocuments.net/reader034/viewer/2022050419/5f8ea6bf84ff263a3951bccb/html5/thumbnails/12.jpg)
10
Table1showsthatinbothdatasources,thesamplingnoiseasmeasuredbyV(t)issignificantlygreateratlongermaturitiesandforshortersamplingwindows.Forbothdatasources,V(t)istwotothreetimeslargerforsix‐monthloansthanforone‐monthloans.Thisisnaturalinpartfromthefactthatlonger‐termloansrolloverlessoftenthanshorter‐termloans.(Thatis,theratiooftheflowofloanstothestockofloansislowerinsteadystateforlonger‐maturityloans.)Inanycase,ourpreliminaryresultssuggestcautionoverwhetheritwouldbepossibletoconstructarobustLiborfixingfromunderlyingloantransactionsforlonger‐termloanssuchassixmonths.Table2presentssummarystatisticsofthedatausedtoconstructthesampling‐windowLiborindex.Forbothdatasources,theaverageacrossthesampleperiodofthenumberofinferred3‐monthloantransactionswithina10‐daysamplingwindowislow,8and25transactionsrespectivelyforthebrokereddataandFedwireinferences.Again,careshouldbetakenininterpretingthesestatisticsgiventhatneitherdatasourceiscomprehensive.Table2:Summarystatistics(10daywindow,3monthmaturity)SummarystatisticsfortheestimatedsamplevolatilitymultiplierV(t),aswellasthenumberoftransactionswithinthe10daysamplingwindow,andtheaveragetransactionsize.Sampleperiodis2000‐04forthebrokereddata,and2007‐12fortheFedwireinferences.p10,p25etc.referstopercentilesoftherelevantdistribution.Brokereddata
Mean p10 p25 p50 p75 p90 StDev
SVM 0.48 0.29 0.35 0.45 0.56 0.71 0.17
# of Transactions in Window 8.13 2 4 7 11 16 5.65
Transaction Size ($mm) 78.69 25 40 50 100 150 59.47
Fedwireinferences
Mean p10 p25 p50 p75 p90 StDev
SVM 0.30 0.21 0.24 0.28 0.34 0.41 0.10
# of Transactions in Window 25.45 13 18 24 31 41 10.59
Transaction Size ($mm) 110.81 25 38 54 110 246 213.74
3.3AlternativespecificationsWehaveexperimentedwithvariousotherdatafilters.Intheappendix,wepresenttwovariations.Thefirstconsidersaminimumtransactionsizeof$100million,ratherthan$25million.Applyingthishighersizecutoffinevitablyreducesthenumberofeligibletransactionsatanypointintime,andthusraisesV(t).Onebearsinmind,however,thatthe“root‐mean‐squared”definitionofV(t)impliesthataloan
![Page 13: A Sampling-Window Approach to · 2015-03-03 · 1 1. Introduction This note considers an approach to constructing Libor fixings using transactions data and multi‐day sampling windows.1](https://reader034.vdocuments.net/reader034/viewer/2022050419/5f8ea6bf84ff263a3951bccb/html5/thumbnails/13.jpg)
11
ofsize$100millionhasarelativeimpactonV(t)thatis16timesthatofa$25millionloan,whenbothsizesarepresentinafixingsample.Secondly,wehaveexperimentedwithanapproachinwhichmoreweightisgiventotransactionsclosertodatet.Seesection4belowforadiscussion.Inunreportedcalculations,wealsoexperimentedwithexpandingthewidthofthematuritywindow(byfivedaysineachdirection).Wefoundthatthishasonlyasmalleffectonthenumberofeligibletransactions.4.SomeDisadvantagesofThisApproach,andTheirMitigationInthissectionwediscusssomeimportantpotentialdisadvantagesofafixingthatisbasedonasampling‐windowapproach:(i)theeffectofusinglaggeddataonthetimelinessoftheresultingLiborfixing,(ii)theriskofalackofunderlyingtransactionsdata,evenwithinasamplingwindow,and(iii)possiblecalendar‐dateeffects.Wealsoconsidersomemitigantsoftheseproblems.Afirstdisadvantageofthesampling‐windowapproachisthatthefixingannouncedonagivendaywouldbebasedinpartonlaggeddatathatmaynolongerberepresentativeofmarketconditions.Thatis,thefixingratecouldbesomewhatstaleduringperiodsofrapidchangesinmarketconditions,forexamplearoundthetimesofsignificantcentral‐bankmonetarypolicyannouncements,orattheonsetofafinancialcrisisorotherperiodinwhichbankfundingcostsareshiftingrapidly,suchasAugust9,2007andtheperiodfollowingit.Theinformationthatmarketparticipantsandregulatorslearnfromtheresulting“Libor”reportcouldthereforebestale.Thereisnosingle“true”interbankborrowingrate,andnosamplingmethodisperfect.Onemaywishtocomparethebiasandsamplingnoiseofthesampling‐windowtransactions‐basedapproachthatwehavedescribedwiththoseofotherfeasiblemethods,includingthecurrentmethodforfixingLibor.Forapplicationsinvolvingbondorswapcontracts,thestalenessintroducedbyasamplingwindowmeasuredindaysisrelativelyunimportant.Afterall,aninvestorholdingapositioninswapsorfloating‐ratenotesisconcernedwiththelevelof3‐monthloanratesthatisgenerallylikelytoprevailseveralyearsintothefuture,andisprobablynotsointerestedinvariationin3‐monthloanrateswithinasmalltimewindowthatbeginsinseveralyears.Apartfromitsroleinfinancialcontracting,Liborisalsousefulforassessingcurrentmarketconditions.However,evenduringtherecentfinancialcrisis,Kuo,SkeieandVickery(2012)showthatmovementsinLiboroverallcommovequitecloselywithanumberofotherpubliclyavailableindices(suchassecondary‐marketCDratesandEurodollaryieldsreportedintheFederalReserveH.15report).Thesealternative
![Page 14: A Sampling-Window Approach to · 2015-03-03 · 1 1. Introduction This note considers an approach to constructing Libor fixings using transactions data and multi‐day sampling windows.1](https://reader034.vdocuments.net/reader034/viewer/2022050419/5f8ea6bf84ff263a3951bccb/html5/thumbnails/14.jpg)
12
indices,whichwouldbemoresensitivetoshort‐termmarketshocks,wouldremainavailabletopolicymakersandmarketparticipants.Wealsonotethatintermsofrevealinginformationtomarketparticipants,asampling‐windowfixingapproachallowstherecoveryofmostofthe“fresh”marketinformationthatispresentintheunderlyingdata.Giventhatthedifferencebetweenthefixingrateondaytandthatonthepreviousdayt‐1iscausedbydroppingobservationsfromdatet‐w(foralagwindowofw)andaddingobservationsfromthelatestdatet,observerscanapproximatelyinvertthemoving‐averageproceduresoastoestimatetheimpliedaveragerateoftransactionsthatoccurredonthelatestavailabledate.Ofcourse,itwouldalsobepossibletosimplyreleasetheaveragetransactionrateforeachday,asdiscussedfurtherbelow.Onecouldreducethebiasassociatedwithstalenessbyweightingthedatawithinthefixingsamplebasedonthetimelag,usingweightsthatdecaywiththelag,sayexponentially.Inordertoillustratetheimpactonsamplingnoiseofde‐weightingstaledata,weexploredtheeffectofanexponentialdecayintransactionweightsthatgivesobservationswitha10‐daylagonly50%oftheweightappliedtoobservationsonthecurrentday.(Thiscorrespondstoaweightfactorof0.933raisedtothepowerofthenumberofdayslagging.)Thisdegreeofde‐weightingofstaletransactionscausesarelativelysmalldegradationinsamplingnoise.10Forexample,for3‐monthinferredtransactionsobtainedfromFedwiredatafor2007‐2012,wesawinTable2thatthemeansamplevolatilitymultiplieris0.30.Withaweightdecayfactorof0.933perdayoflag(50%de‐weightingof10‐dayoldobservations),thesamedataareassociatedwithameansamplevolatilitymultiplierof0.31,about3%higher.Theestimatedeffectsonsamplingnoiseofde‐weightingstaledataaresimilarlymutedinallofthecasesthatwehaveexamined,asdemonstratedinadditionalchartsandtablesfoundintheappendices.Itistobecautionedthattheseresultsarepreliminaryandonlyforillustrativepurposes.Inadditiontopublishingthesampling‐window‐basedfixingrate,onecouldalsopublishsomepropertiesoftheunderlyingdata,suchasthedailyaveragerate,thedailynumberoftransactions,orthesample‐volatilitymeasure.Whilefinancialcontractswouldpresumablybetiedtothefixingrate,otherpublishedinformationbasedonthesamplecouldprovideadditionalusefulinformationandcould
10Inordertogainsomeintuitionforthelimitedimpactofdecayingweightsonthesamplevolatilitymultiplier,considerarelativelyadversecaseinwhichthetransactionsareconcentratedatthefirstandlastdateofa10‐daysamplewindow.Twoequallysizedtransactionsateachendofthe10‐daysamplingwindow,withoutdecay,wouldhaveasamplevolatilitymultiplierofV(t)=(0.52+0.52)0.5=0.707.Withweightsdecayingproportionatelybyafactorof0.933perday,or50%over10days,wewouldhaveV(t)=[(0.5/k)2+(0.5×0.5/k)2]0.5,wherek=0.5+0.25=0.75,implyingV(t)=0.74.So,indeed,eveninthisrelativelyextremesituation,theelevationofthesamplevolatilitymultiplierV(t)duetodecayisonlyabout5%.
![Page 15: A Sampling-Window Approach to · 2015-03-03 · 1 1. Introduction This note considers an approach to constructing Libor fixings using transactions data and multi‐day sampling windows.1](https://reader034.vdocuments.net/reader034/viewer/2022050419/5f8ea6bf84ff263a3951bccb/html5/thumbnails/15.jpg)
13
potentiallybeusedincontracting,forexampleinordertoallowfinancialcontractstobetiedtomarketliquidityortothequalityofthefixingsample.Aseconddisadvantageofasampling‐windowapproachisthatitisnotguaranteedtoproducereliableresultsunderallmarketconditions.Iftherearetoofewtransactionsatagivenmaturitytoprovideevenareasonableestimateofmajor‐bankborrowingrates,marketparticipantswillneverthelessrequireareferencerateonwhichtobasethesettlementofderivativesandfloating‐rateloancontracts.FortheU.S.dollarmarket,ourresultsbasedonalimiteddatasetsuggestsomehopeforthefeasibilityoftransaction‐basedfixing,usingsamplingwindows,for1‐monthand3‐monthmaturities.Inanycase,onemaywishtointroducerobustnesssafeguardsinthedefinitionofthefixingsampleS(t,m;w,d),suchasexpandingthefixingsamplewheneverthereisinsufficientdataforareliablefixing.Forinstance,onecouldtakethesamplewindowtobeafixednumberofdaysortheminimumnumberofdaysnecessarytoincludeagivenvolumeoftransactions,whicheverisgreater.11AsanalternativetofixingLiborbasedonunsecuredborrowingrates,ithasbeensuggestedthatLibormightbereplacedwithabenchmarkratebasedonsecuredlendingtransactions.Prominentamongthesuggestedsecuredinterestratesis“GCFrepo,”whosemarketisdescribedbyFlemingandGarbade(2003).12Thisapproachwouldintroduceseveralpotentialcomplications,however.First,forGCFrepo,thereremainrobustnessconcernsoverwhetherthereisasufficientvolumeofGCFrepotransactionsattherelevantmaturities.Second,GCFreporatesareonlyindirectlyconnectedtobanks’unsecuredcostoffunds,whichreducestheusefulnessofGCFrepoasthebasisforanindexrateforfinancialcontracting.Forcommercialbanksandbankholdingcompanies,unsecuredborrowingisgenerallyamuchlargersourceofoverallfundingthansecuredborrowing.Unsecuredborrowingisalsotraditionallytheprimarysourceoffundingonthemargin.(Forsecuritiesdealers,securedborrowingisalargersourceoffundingandamoretypicalmarginalsourceoffunding,relativetobanks.)Further,Libor‐basedswapsareheavilyusedforrisk‐managementandpricediscoveryfortheunsecureddebtofnon‐financialcorporations.BasingLiboronasecuredborrowingratewouldreduceitsusefulnesshereaswell.Third,usingasecuredfinancingratesuchasGCFreporaisesthe
11ArelatedconcernisthataLiborfixingbasedonasamplingwindowapproachcouldbecomedistortedaroundkeycalendardates,suchastheendofaquarterorcalendaryear.Counterpartiesmayforexamplelengthenorshortenthematurityofotherwisestandardcontractstoinfluencewhethertheycoverparticularfinancialstatementdates,forwindow‐dressingpurposesorforotherreasons.Thiscouldaffectthesetofcontractswhosematuritiesliewithinagivenrange(d)aroundastandardmaturitysuchasonemonthorthreemonths.Inourexamples,wesetthisdaterangetobeconstant,butitmaybenecessarytoadjustdinsuchsituations.12TheDTCCpublishesanaverageovernightGCFreporateforthreetypesofcollateral:Treasuries,agencyMBS,andagencydebt.TradinginfutureslinkedtotheseindicesbeganinJuly2012.Seehttps://globalderivatives.nyx.com/nyse‐liffe‐us/dtcc‐gcf‐repo‐index‐futures/settlement‐procedures
![Page 16: A Sampling-Window Approach to · 2015-03-03 · 1 1. Introduction This note considers an approach to constructing Libor fixings using transactions data and multi‐day sampling windows.1](https://reader034.vdocuments.net/reader034/viewer/2022050419/5f8ea6bf84ff263a3951bccb/html5/thumbnails/16.jpg)
14
questionofhowtotreatlegacyLibor‐basedfinancialcontracts,ofwhichthereareenormousquantities.AcounterpartyreceivingLiboronalegacycontractwouldnotwillinglyreceiveinsteadtheGCFreporate,whichistypicallymuchlower.Replacing“legacyLibor”withanapproximationofunsecuredratesthatareestimatedfromsecuredfinancingrateswouldlikelyleadtoasubstantialamountofcontractualdispute.Thisalsoraisesthepossibilityoftwoparallelmarkets,atleastduringatransitionperiod,with“legacy”and“new”benchmarksbasedonunsecuredandsecured(repo)rates,respectively.Theassociatedtransitionwouldbeawkwardandlengthy,andinvolvesplittingliquidityacrossthetwomarketswithanattendantlossinmarketefficiency.Inanycase,asampling‐windowapproachcouldalsobeusedfortermreporates,providedtherearesufficientdata.TheWheatleyReport(H.M.Treasury,2012)reviewsotheralternativeapproachesandbenchmarks,suchastheovernightindexswaprate(OIS),andprovidesadescriptionoftheiradvantagesanddisadvantages.
![Page 17: A Sampling-Window Approach to · 2015-03-03 · 1 1. Introduction This note considers an approach to constructing Libor fixings using transactions data and multi‐day sampling windows.1](https://reader034.vdocuments.net/reader034/viewer/2022050419/5f8ea6bf84ff263a3951bccb/html5/thumbnails/17.jpg)
15
ReferencesArmantier,OlivierandAdamCopeland(2012).“AssessingtheQualityof“Furfine‐based”Algorithms,”FederalReserveBankofNewYorkStaffReports,No.575October.Bartolini,Leonardo,SpenceHiltonandJamesJ.McAndrews(2010).“SettlementDelaysintheMoneyMarket”,JournalofBankingandFinance34,934‐945.Carhart,Mark,RonKaniel,DavidMusto,andAdamReed(2002)“LeaningfortheTape:EvidenceofGamingBehaviorinEquityMutualFunds,”JournalofFinance57,661‐693.Fleming,MichaelandKennethGarbade(2003).“TheRepurchaseAgreementRefined:GCFRepo,”FederalReserveBankofNewYorkCurrentIssuesinEconomicsandFinance,9(June).Furfine,Craig(1999)“TheMicrostructureoftheFederalFundsMarket,”FinancialMarkets,Institutions&Instruments8,24‐44.Kuo,Dennis,DavidSkeie,JamesVickery(2012),“AComparisonofLibortoOtherMeasuresofBankBorrowingCosts,”WorkingPaper,FederalReserveBankofNewYork.Kuo,Dennis,DavidSkeie,JamesVickery,andThomasYoule(2012),“IdentifyingTermInterbankLoansfromFedwirePaymentsData,”WorkingPaper,FederalReserveBankofNewYork.H.M.Treasury(2012).“TheWheatleyReviewofLibor:FinalReport,”H.M.Treasury,London,September,2012.
![Page 18: A Sampling-Window Approach to · 2015-03-03 · 1 1. Introduction This note considers an approach to constructing Libor fixings using transactions data and multi‐day sampling windows.1](https://reader034.vdocuments.net/reader034/viewer/2022050419/5f8ea6bf84ff263a3951bccb/html5/thumbnails/18.jpg)
16
Appendix:OtherDataFiltersAppendixA1.Minimumtransactionsizeof$100m(ratherthan$25m)ThestatisticsshownherearecomputedforthesamedataasthoseunderlyingFigure1andTable1,withtheexceptionthatthetransactionssizeshaveaminimumof$100m,ratherthanaminimumof$25m.FigureA1.Time‐seriesplotofV(t)i.Brokereddata
ii.Fedwireinferences
0
.2
.4
.6
.8
1
SVM
01jan2000 01jul2001 01jan2003 01jul2004Date
0
.2
.4
.6
.8
1
SVM
01jan2007 01jan2008 01jan2009 01jan2010 01jan2011 01jan2012Date
![Page 19: A Sampling-Window Approach to · 2015-03-03 · 1 1. Introduction This note considers an approach to constructing Libor fixings using transactions data and multi‐day sampling windows.1](https://reader034.vdocuments.net/reader034/viewer/2022050419/5f8ea6bf84ff263a3951bccb/html5/thumbnails/19.jpg)
17
TableA1.StatisticsforV(t)MedianacrossthesampleperiodofthesamplevolatilitymultiplierV(t)forthematurityandsamplingwindowlengthshown,normalizedbythemedianofV(t)forasamplingwindowof10daysandmaturityof3months.i.Brokereddata Maturity
1 month 3 months 6 months
Lag window (days) 2 1.00 1.41 1.41
5 0.79 1.41 1.41
10 0.54 1.00 1.41
15 0.44 0.78 1.41
20 0.38 0.70 1.41
ii.Fedwireinferences Maturity
1 month 3 months 6 months
Lag window (days) 2 1.22 1.73 2.42
5 0.92 1.42 2.42
10 0.66 1.00 1.71
15 0.54 0.83 1.54
20 0.48 0.74 1.43
![Page 20: A Sampling-Window Approach to · 2015-03-03 · 1 1. Introduction This note considers an approach to constructing Libor fixings using transactions data and multi‐day sampling windows.1](https://reader034.vdocuments.net/reader034/viewer/2022050419/5f8ea6bf84ff263a3951bccb/html5/thumbnails/20.jpg)
18
AppendixA2.UsingexponentialdecayThestatisticsshowninFigureA2andTableA2arecalculatedusingthesamesamplesasthoseofFigure1andTable1,exceptthatweincorporateexponentialdecayoverthesamplingwindow.FigureA2.Time‐seriesplotofV(t)(i)Brokereddata
(ii)Fedwireinferences
0
.2
.4
.6
.8
1
SVM
01jan2000 01jul2001 01jan2003 01jul2004Date
0
.2
.4
.6
.8
1
SVM
01jan2007 01jan2008 01jan2009 01jan2010 01jan2011 01jan2012Date
![Page 21: A Sampling-Window Approach to · 2015-03-03 · 1 1. Introduction This note considers an approach to constructing Libor fixings using transactions data and multi‐day sampling windows.1](https://reader034.vdocuments.net/reader034/viewer/2022050419/5f8ea6bf84ff263a3951bccb/html5/thumbnails/21.jpg)
19
TableA2.StatisticsforV(t)MedianfortheperiodofthesamplevolatilitymultiplierV(t)fortheindicatedmaturityandsamplingwindowlengthshown,normalizedbythemedianforasamplingwindowof10daysandmaturityof3months.(i)Brokereddata Maturity
1 month 3 months 6 months
Lag window (days) 2 1.03 1.55 2.19
5 0.80 1.28 1.63
10 0.61 1.00 1.35
15 0.52 0.86 1.18
20 0.47 0.78 1.09
(ii)Fedwireinferences Maturity
1 month 3 months 6 months
Lag window (days) 2 1.15 1.67 2.48
5 0.88 1.31 2.10
10 0.67 1.00 1.63
15 0.57 0.86 1.38
20 0.53 0.79 1.27