FRM二级基础段培训课件：案例（打印版）.docx

7CurrentIssues':2021.,8 3lt 31 FrameworkCurrent Issues1.BeyondLIBOR2.ReplacingLlBOR3.MachineLearning4.AIandMLinfinancialservices5.ClimateChange-PhysicalRiskandEquity6.TheGreenSwan7.WhenSellingBecomesViral8.MarketsintheTimeofCOVID-199.FinancialCrimeinTimesofCOVID-1910.CyberRiskandtheUSFinancialSystem专业事新tt1.BeyondLIBOR4-84Muy色嘛！A.AnIdealReferenceRateAnIdealReferenceRateNotsusceptibletomanipulation.Derivedfromactualtransactionsinliquidmarkets.Serveasabenchmarkforbothtermlendingandfunding.5-84MwrrmaB.ProblemswithLIBoRIssuesthatledtothereplacementofLIBOR3Constructedfromasurveyofbanksreporting.ThiscreatedamplescopeforpanelbankstomanipulateLIBORsubmissions.Sparseactivityininterbankdepositmarkets.1.Thedispersionofindividualbankcreditrisk.1.IBORaimstocapturecommonbankrisk.iRegulatoryandthemarketwanttoreducecounterpartycreditriskininterbankexposures,bankshavealsotiltedtheirfundingmixtowardslessriskysourcesofwholesalefunding.uy*at*maD.RisksofRFRsintheRepoMarketRisksofRFRSintheRepoMarket0/Nreporatecannotreflectbanksmarginalfundingcosts.Banks'asset-liabilitymanagementischallenging.Whenunderstress,reporatescanmoveintheoppositewayofunsecuredrates.Theforcesdrivingunsecured0/Nrates(includingcreditrisk)pulledtheserateshigherastheunsecuredinterbankmarketsfroze.Atthesametime,theforcesdrivingsecured0/NrateswerepullingthemlowerowingtoacollateralshortageandflighttosafetyForlongertenors,termratesbasedonnewRFRsarelikelytodeviatepersistentlyfromtheirLIBORcounterpartseveninnormaltimes.TransitionIssues:themigrationoflegacyLIBOR-Iinkedexposurestothenewbenchmarksafter2021.M亚倒舞m2.ReplacingLIBOR10-84Muy 色嘛 ！A.TheFactsPublicationofUBOR-theLondonInterbankOfferedRate-willlikelyceaseattheendof2021.after202LtheFCAwouldnolongercompelreluctantbankstorespondtotheLIBORsurvey.11 84then,theFCAcoulddeclareLIBORratesUnrepresentative*offinancialrealityanditwillvanish.MwrrmaB.RisksWhenLIBOREndsThesystemicriskposedbythecessationofLIBOR.3ThefirstarisesfromthelegacycontractsreferencingLIBOR.WhenpublicationofLIBORstops(orisexpectedtostop),contractsthatlackadequatefallbackprovisionsmayplungeinvalue.Totheextentthatlarge,leveragedintermediariesareexposed,theresultinglossescouldimpairtheircapital,leavingusinthedarkaboutwhichinstitutionsarehealthyandwhicharenot.3Thesecondissueiswhether,whenLIBORceases,therewillbeanadequatesubstitutethatallowsintermediariesbothtofundthemselvesinaliquidmarketandtoprovidecredit.C.CurrentProblemsWheredothingsstandnow?First,thereremainplentyofdollarLIBORlegacycontractsoutstanding,thelatestavailabledataarenearlythreeyearsoldrandtheindustrycontinuestocreateLIBOR-Iinkedcontracts,westronglysuspectthatthesenumbersunderstatethechallenge.Second,thereisnocentralrepositoryprovidinginformationaboutwhat,ifany,fallbacklanguageexistsinthesecontracts.WithoutLIBOR,whathappens?Inadequatefallbacklanguagefostersuncertaintyaboutthevalueoftheassets,andcouldtriggerawaveoflawsuits.Third,whiletheprocessofcreatingasatisfactoryreplacementfordollarLIBORiswelladvanced,itisfarfromcomplete.Thereislittletimefortesting.D.TheGovernment'sRoleintheTransitionFourveryimportantrolesforgovernmentofficials.3Thegreaterthecertaintyabouttheenddate,thefastertheLIBORtransitionWillbe.AuthoritiescanfurtherintensifywarningsabouttheimprudenceofrelyingonLIBOR.3Supervisorsmustensurethatallsystemicallyimportantbanksandfinancialmarketutilitiesarefullyprepared.?Thereremainsaremarkableabsenceofup-to-datedataonLIBOR-linkedinstruments.regulatorsshouldgatherandpublishdatashowingtheevolutionofLIBORe×posure(includinginformationonfallbacklanguage)atleastquarterly.GBecausetheLlBORtransitionwilldirectlyaffectmanyhouseholdsandsmallbusinesseswithUBOR-Iinkeddebtitisimportantforauthorities(includingtheConsumerFinancialProtectionBureau)topromotepublicawarenessofthechangesunderway.14-84M皿施舞na3.MachineLearningA.IntroductionTheDrivingForcesMoredetailsofreporting.High-frequency,unstructuredlowqualityconsumerdata.BigdataPredictionversusExplanationStatisticalmethodsaregoodforexplanation.MLisgoodforprediction.16-84uy * a ! B. Background to MLSupervised LearningDependent variable y is known.Unsupervised LearningDependent variable y is lacking.17-84su n * maB.BackgroundtoMLMachineLearningMethodsRegressionAsupervisedMLproblem.Topredictacontinuousdependentvariabley.Afactorisaddedtopenalisecomplexityinthemodel.ClassificationAdiscreteproblem.ClusteringAnunsupervisedMLproblem.B.BackgroundtoMLOverfittingFitthedatasampleverywell.Performpoorlywhentestedout-of-sample.Havingtoomanyparameters.WaystoDealwithOverfittingBoosting:overweightscarcerobservationsinatrainingdataset.Bagging:amodelisrunthousandsoftimes,eac