value at risk多媒体教学课件.ppt

*****************************************QuadraticModel(continued) OncewehavedonethiswecanusetheCornishFisherexpansiontocalculatefractilesofthedistributionofDPMonteCarloSimulation

(p.355)ThestagesareasfollowsValueportfoliotodaySampleoncefromthemultivariatedistributionsoftheDxiUsetheDxitodeterminemarketvariablesatendofonedayRevaluetheportfolioattheendofdayMonteCarloSimulationCalculateDPRepeatmanytimestobuildupaprobabilitydistributionforDPVaRistheappropriatefractileofthedistributiontimessquarerootofNForexample,with1,000trialthe1percentileisthe10thworstcase.SpeedingUpMonteCarlo UsethequadraticapproximationtocalculateDPHistoricalSimulation

(p.356)Createadatabaseofthedailymovementsinallmarketvariables.ThefirstsimulationtrialassumesthatthepercentagechangesinallmarketvariablesareasonthefirstdayThesecondsimulationtrialassumesthatthepercentagechangesinallmarketvariablesareasontheseconddayandsoonStressTesting

(p.357)Thisinvolvestestinghowwellaportfolioperformsundersomeofthemostextrememarketmovesseeninthelast10to20yearsBack-Testing

(p.357)TestshowwellVaRestimateswouldhaveperformedinthepastWecouldaskthequestion:Howoftenwasthelossgreaterthanthe99%/10dayVaR?PrincipalComponentsAnalysis

(p.357)Supposethataportfoliodependsonanumberofrelatedvariables(eginterestrates)Wedefinefactorsasascenarioswherethereisacertainmovementineachmarketvariable(Themovementsareknownasfactorloadings)TheobservationsonthevariablescanoftenbelargelyexplainedbytwoorthreefactorsResultsforInterestRates

(Table14.4)Thefirstfactorisaroughlyparallelshift(83.1%ofvariationexplained)Thesecondfactorisatwist(10%ofvariationexplained)Thethirdfactorisabowing(2.8%o