ContentslistsavailableatScienceDirect
North
American
Journal
of
Economics
and
Finance
Aggregate
volatility
expectations
and
threshold
CAPM
Yakup
Eser
Arısoy
a,∗,
Aslıhan
Altay-Salih
b,1,
Levent
Akdeniz
b,2aUniversitéParis-Dauphine,DRMFinance,75775ParisCedex16,France bBilkentUniversityFacultyofBusinessAdministration,06533Ankara,Turkey
a
r
t
i
c
l
e
i
n
f
o
Articlehistory:Received9March2015
Receivedinrevisedform16September 2015
Accepted21September2015 Availableonline9October2015
JELclassification: C13 G12 Keywords: Aggregatevolatility Thresholdregression ConditionalCAPM Range VIX
a
b
s
t
r
a
c
t
Weproposeavolatility-basedcapitalassetpricingmodel(V-CAPM)
inwhichassetbetaschangediscretelywithrespecttochanges
ininvestors’expectationsregardingnear-termaggregate
volatil-ity.Usinganovelmeasuretoproxyuncertaintyaboutexpected
changesinaggregatevolatility,i.e.monthlyrangeoftheVIXindex
(RVIX),we find that portfoliobetas change significantly when
uncertaintyaboutaggregatevolatilityexpectationsisbeyonda
cer-tainthresholdlevel.Duetochangesintheirmarketbetas,smalland
valuestocksareperceivedasriskierthantheirbigandgrowth
coun-terpartsinbadtimes,whenuncertaintyaboutaggregatevolatility
expectationsishigh.Theproposedmodelyieldsapositiveand
sig-nificantmarketriskpremiumduringperiodswheninvestorsdonot
expectsignificantuncertaintyinnear-termaggregatevolatility.Our
findingssupportavolatility-basedtime-varyingriskexplanation.
©2015ElsevierInc.Allrightsreserved.
∗ Correspondingauthor.Tel.:+33144054360.
E-mailaddresses:eser.arisoy@dauphine.fr(Y.E.Arısoy),asalih@bilkent.edu.tr(A.Altay-Salih),akdeniz@bilkent.edu.tr (L.Akdeniz).
1 Tel:+903122902047. 2 Tel:+903122902202.
http://dx.doi.org/10.1016/j.najef.2015.09.013 1062-9408/©2015ElsevierInc.Allrightsreserved.
1. Introduction
The capitalassetpricing model(CAPM)assumesthat afirm’sriskiness, which iscapturedby itsbeta,isconstantthroughtime.However,changesinbusinessconditions,technology,andtaste mightinduceshiftsintheinvestmentopportunitysetandinvestors’associatedrisk-returntradeoffs (Jagannathan&Wang,1996).Manystudiesmodelthevariationinbetasusingcontinuous approx-imationandthetheoreticalframeworkoftheconditionalCAPM1.Yet,despiteastrongtheoryand considerableevidenceontimevariationinbetas,thereisnoconsensusonhowthisvariationshouldbe modelled.
Inthispaper,wemodelassetbetasneitherasstaticnorasacontinuousapproximationimpliedby conditionalmodels,insteadweassumethatassetbetaschangediscretelyintime2.Ourapproach follows thespirit of regime-switching models,which have been extensively used in modelling financialtime-series3. More particularly, we posit that investors re-assessfirms’ systematic risk withrespecttoexpectedchangesinaggregateriskconditionsbasedontheirexpectations regard-inguncertaintyaboutfutureaggregatevolatility.Thereareseveralreasonswhyweassumebetas shouldchangewithrespecttouncertaintyaboutaggregatevolatilityexpectations.First,itiswell documentedthat both equity andaggregate volatility istime-varying4. Therefore,an asset pric-ingmodelthatincorporatestime-variationinaggregatevolatilitywouldnaturallyimplythatasset betasalsochangeaccordingly5,6.Second,time-varyingriskliteraturesuggeststhatstockshave dif-ferentexposurestomarketriskduringrecessionsandexpansions(Lettau&Ludvigson,2001;Petkova &Zhang, 2005). Given the fact that change in aggregate volatility is tightly linkedto business cycles,ourmodel isabletocapturethis linkbyconditioningchanges in assetbetasonchanges in aggregate volatility expectations7. Drivenby the fact that option-impliedvolatility measures aregood forecastsoffuture volatility, weconditiontime variation inbetas based onan option-impliedmeasure,whichsummarizesinvestors’expectationswithrespecttochangesinnear-term aggregatevolatility8.Ratherthanusingmacrovariables asinpreviousstudies,ourapproach con-tributestotheliteraturebyproposinganovelconditioningvariable,whichhasaforward-looking
1SeeHarvey(1989),FersonandHarvey(1991,1993,1999),FersonandKorajczyk(1995),JagannathanandWang(1996),and PetkovaandZhang(2005).
2Theintuitionbehinddiscretechangesinbetaswithrespecttotwodifferentregimessimilartodownside-upsidebeta approachinAng,Cheng,andXing(2006a)whoshowthatassetbetaschangeduringdownsideandupsidemarketsanddownside riskispriced.Methodologically,ourapproachisalsorelatedtoMarkovchainregimeswitchingmodelsasinGuidolinand Timmermann(2008)andChen,Gerlach,andLin(2011),andoptimalchangepointapproachasinBollenandWhaley(2009), andPattenandRamadorai(2013).
3SeeHamilton(1989),HamiltonandLin(1996),andAngandBekaert(2002)fordetailsandapplicationsofregime-switching modelsindifferentsettings.
4Fortheoreticalbackgroundandempiricalevidenceonstochasticvolatilityofequityandstockmarketreturns,seeEngle andBollerslev(1986),French,Schwert,andStambaugh(1987),Schwert(1989),EngleandNg(1993),CaninaandFiglewski (1993),Duffee(1995),Braun,Nelson,andSunier(1995),Andersen(1996),BollerslevandMikkelsen(1999)andBekaertandWu (2000).
5Giventhedefinitionofbeta,i.e.,ˇ i=
Cov (ri,rm)/Var (rm)
,anyuncertaintyininvestors’aggregatevolatilityexpectations (i.e.thedenominator)isexpectedtoaffectreturnsinthecross-sectionthroughbetasonthemarketportfolio.Hence,our approachisdifferentfromAngetal.(2006a),Ang,Hodrick,Xing,andZhang(2006b)whoconditionexpectedreturnsdirectly onaggregatevolatilityandwhomodelaggregatevolatilityasaseparateriskfactor.OurapproachisalsodifferentfromWang andMa(2014)whoexaminetheeffectofexcessvolatilityattheindividualstocklevel.Weinvestigatetheimplicationsof uncertaintyaboutaggregatevolatilityonbetasandonthecross-sectionofexpectedreturns.6PolletandWilson(2010)showthatincreasesinmarketvolatilitycanbeduetoeitherincreasesinaveragevolatilitiesor averagecorrelations,orboth.Furthermore,Buraschi,Porchia,andTrojani(2010)andBuraschi,Trojani,andVedolin(2014)show thatinvestorshavehedgingdemandsagainstbothstochasticaggregatevolatilityriskandstochasticcorrelationrisk.Intheir model,bothriskfactorsstemasaresultofuncertainty(disagreementinbeliefsacrossagents)intheeconomy.Wedonottake adirectstandonthecorrelationstructureinthispaper.However,becauseourmeasureRVIXessentiallycapturesuncertainty inexpectedvolatilityofthemarketportfolio,itiscloselylinkedtobothsourcesofrisk.
7SeeHsuandLi(2009)whodocumentcounter-cyclicalityofvolatilityacrossdifferentassetclasses.
8Forthepredictiveabilityofoption-impliedvolatilitymeasuresrangingfromintra-dayforecaststoone-yearaheadforecasts, andindifferentmarketssuchasforeignexchange,stock,bond,andoptionmarkets,seePoonandGranger(2005),Taylor,Yadav, andZhang(2010),Busch,Christensen,andNielsen(2011),HanandPark(2013)andBianconi,MacLachlan,andSammon(2015).
featurebyconstructionandwhichmodelstimevariationinanasset’sriskinessinaparsimonious way9,10.
Inparticular,weproposeavolatility-basedthresholdCAPM(V-CAPM)whereassetbetaschange withrespecttoinvestors’assessmentofaggregateriskconditions,proxiedbyuncertaintyabout mar-ket’sexpectationsregardingchangesinaggregatevolatility.ThecontributionoftheproposedV-CAPM isfourfold.First,weproposeanovelmeasuretoproxyexpectedchangesinaggregateriskconditions, i.e.rangeoftheVIXindex(RVIX)11.VIXisinherentlyaforward-lookingvolatilitymeasureandit revealsimportantinformationaboutinvestors’expectationsofneartermvolatilityinthemarket12. DefinedasthedifferencebetweenthemaximumandminimumleveloftheVIXindex,RVIXessentially capturesexpectedchangesinnear-termaggregatevolatility,orputdifferentlythedegreeof uncer-taintyinfutureaggregatevolatility.Inarecentpaper,Baltussen,Bekkum,andGrient(2015)show thatthevolatilityofvolatility(vol-of-vol)isanimportantfactorinthecross-sectionofstockreturns. Usingthevolatilityofoption-impliedvolatilityasameasureofuncertaintyaboutvolatility,they doc-umentthatstockswithhighvol-of-volunderperformstockswithlowvol-of-vol.Theauthorsargue thatthevolatilityofoption-impliedvolatilityisanintuitivemeasure,whichistightlyrelatedtothe literaturethatmodelsuncertaintyassecond-orderbeliefs.Inasimilarvein,RVIXcanbeinterpreted asameasurethatcapturesuncertaintyregardingfutureaggregatevolatility.Hence,ifinvestorshold secondorderbeliefsandcareaboutthisuncertainty,thenstocksorportfolioswithdifferent sensi-tivitiestochangesinuncertaintycouldhavedifferentrisk-returndynamics,whichcouldalsoimply differentmarketriskpremiumdynamicsattimesofincreaseduncertaintyaboutexpectedaggregate volatility.
Secondcontributionis ourapproach tomodellingtime variation in betas.In standard condi-tionalCAPMsetting,betaspracticallychangeateach pointintime,howeverthisapproach might havea tendency tooverstatethetime variation in betasand resultin estimates thatare highly volatile.OursettingdiffersfromthestandardconditionalCAPMmodelsbyallowingbetastochange onlywhenthedegreeofuncertainty aboutexpectedaggregatevolatilitymovesbeyonda certain thresholdlevel,admitting a discretevariation inbetas intwo distinctregimes.Third,ourmodel implicitlyallowsfortimevariationinaggregatevolatilitywhichisnotpossibleinthestaticCAPM setting.Byendogenouslyincorporatingchangesininvestors’expectationsaboutaggregate volatil-ityinadynamicway,weallowbetastocapturepotentialshiftsintheinvestmentopportunityset linkedtoexpectedchangesinaggregateriskconditions13.Ourfourthcontributioniseconometric. Weformallytestthehypothesisontheexistenceofarelationbetweenbetasanduncertaintyabout aggregatevolatilityusingHansen’s(2000)thresholdregressionmethodology,whichisintuitiveand fullysupportedbytheeconometrictheory14.Theproposedmodelisrichinitspredictionsandoffers
9Amongthemostwidelyusedmacroconditioningvariablesintheliteraturearethedividendyield(Fama&French,1988), defaultspread(Keim&Stambaugh,1986),termspread(Campbell,1987),shorttermtreasurybillrate(Fama&Schwert,1977) andlogconsumption-wealthratio(Lettau&Ludvigson,2001).
10Ourstudyisalsorelatedtotherecentstrandofliteratureshowingthatthedifferenceinoption-impliedvolatilitymeasures hassignificantexplanatorypowerinthecross-sectionofstockreturns.AmongthemarestudiesbyBaliandHovakimian(2009), Bollerslev,Tauchen,andZhou(2009),CremersandWeinbaum(2010)andAtilgan,Bali,andDemirtas(2015).
11Weexaminewhetherusingotherconditioningvariablesdocumentedintheliterature(suchasonemonthT-billrate, aggregatedividendyield,inflationrate,termspreadandcreditspread)resultinsignificantregimechangesinportfoliobetas asimpliedbythethresholdCAPMmodel.Noneoftheexaminedvariablesyieldsignificantregimeshiftsinportfoliobetasas strongasRVIX.
12Oftenreferred toasthe“fear ¨or“marketsentiment¨index,VIXestimatesnear-term(roughly next30-day)expected volatility by weighted-averaging the prices of puts and calls written onthe S&P 500 index over a range of strike prices.
13Wealsoinvestigatewhetherthedocumentedregimechangesinbetaswithrespecttouncertaintyaboutaggregatevolatility expectationsislinkedtodownsideriskasinAngetal.(2006a)andBali,Demirtas,andLevy(2009).Usingupsideanddownside betasofAngetal.(2006a),wedonotfindastrongcorrelationbetweenhighuncertaintyvs.downsidebetasandlowuncertainty vs.upsidebetas.Furthermore,dividingthesampleperiodintomonthsthatcorrespondtohighvs.lowuncertaintyabout aggregatevolatilityandmonthsthatcorrespondtoupsideanddownsidemarkets,wefindmajordifferencesbetweenthe correspondingtimeperiods.
avolatility-basedexplanationtosomeoftheempiricalassetpricinganomaliesdocumentedinthe literature.
UsingRVIXasproxyforuncertaintyaboutaggregatevolatilityexpectations,andportfoliossorted withrespecttomarketcapitalizationsandbook-to-marketratiosastestassets,ourresultscanbe sum-marizedasfollows15.First,usingthemodifiedsupLMtestsuggestedbyHansen(1996),wedocument significanttimevariationinbetas.15outof22testportfolioshavesignificantbootstrapp-valuesat5% level16.Theevidenceconfirmstheexistenceofaggregatevolatilityrelatedchangesinbetasofmost portfolios,mostparticularlyfortheextremesizeandbook-to-marketportfoliosaswellasSMBand HMLportfolios.Theinitialresultssupportthehypothesisthatassetbetaschangediscretelyintime anduncertaintyaboutaggregatevolatilityexpectationsisakeydeterminantofinvestors’assessment aboutanasset’ssystematicrisk.
Next,wetestwhetherdifferentsizeandbook-to-marketportfolioshavedifferentbetasensitivities withrespecttoinvestors’expectationsaboutuncertaintyofaggregatevolatilityandriskconditions. Thethresholdestimatessuggestthatinvestorsoverwhelminglyupdatetheirbetariskassessments whenmonthlyrangeoftheVIXindexisbeyond9.33points17.Whatmakestheresultsfurther remark-ableisthedirectionofthisupdate.Lookingatthechangesinportfoliobetas,onecanseethatstocksin small(andvalue)portfolioshaveconsistentlyhigherbetasattimeswhenuncertaintyaboutexpected aggregatevolatilityishigh(i.e.whenRVIXinagivenmonthisabovethethresholdlevel).Onthe con-trary,theportfoliooflargestmarketcapitalizationstocks(andthegrowthportfolio)exhibitslower betasduringtheseuncertaintimes.Theincreaseinbetasismostpronouncedforthesmallestdecile, highestbook-to-marketdecile,SMBandHMLportfolios.
Frenchetal.(1987),andCampbellandHentschel(1992)documentthatperiodsofhighvolatility usuallycoincidewithdownwardmarketmoves.Furthermore,risk-averseinvestorsarereluctantto losewealthinperiodsofhighvolatilitybecauseitrepresentsadeteriorationininvestment opportu-nities,whichusuallycoincideswithperiodsoflowconsumption(recessions)18.Theincreaseinbetas ofsmallandvalueportfoliosimpliesthatstockswiththesecharacteristicsareperceivedtoberiskier attimesofincreaseduncertaintyaboutexpectedvolatility.ThisalsoholdsforSMBandHML port-folioswhosesensitivitiestomarketreturnsbecomehigherduringthosetimes.Investorsviewsmall andvaluefirmsriskierbecausetheirreturnscorrelatestronglywithmarketreturnsinepisodeswhen uncertaintyaboutexpectedaggregatevolatilityishigh.Ontheotherhand,returnsonbigandgrowth stockscorrelatelesswithmarketreturnsduringthosetimes.Ourresultsareconsistentwiththose ofLettauandLudvigson(2001)andPetkovaandZhang(2005)whofindthatvalueandsmallstocks correlatemorewiththeconsumptiongrowth(marketreturns)duringbadtimesrelativetobigand growthstocks,whiletheoppositeholdsduringgoodtimes.Wearguethatinvestorsviewsmalland valuestocksriskierthantheirbigandgrowthcounterpartsbecausetheirreturnsaremuchmore sen-sitivetomarketriskattimesofincreaseduncertaintyaboutaggregatevolatilityandadversemarket conditions.
Totesttherobustnessoftheaboveresultsandtofurtherexaminetheeffectoftimevariation ininvestors’expectationsofneartermvolatilityonassetrisk-returndynamics,wenextcalculate Jensen’salphas and Sharperatios ofour test assets in thefull sample aswell asin two differ-entregimes(i.e.highandlowuncertaintyaboutvolatility)impliedbythethresholdlevelofRVIX estimated via theV-CAPM. It is well-documented that small and value stocks (and the associ-atedSMBand HMLstrategies) producesignificantly higheraveragereturns thantheirlargeand growthcounterparts. Looking at Jensen’s alphas and betas of differentsize and book-to-market portfolios,we confirmthepreviousfindingsthat thestaticCAPMis unabletooffera risk-based
15WealsouseanorthogonalizedversionofRVIX(RVIXORTH)inordertoensurethattheresultsarenotdrivenbypotential variablesthathavebeendocumentedtobeimportantintheliterature.RVIXORTHisdefinedastheresidualtermobtainedfrom regressingRVIXonaggregatedividendyield,thedefaultspread,thetermspread,theshort-termtreasurybillrate,andtheVIX. TheresultsarerobusttouseofRVIXORTHastheconditioningvariable.
169(17)portfoliosexhibitsignificantchangeinbetasat1%(10%)level.
17Thethresholdestimates(whichcouldpracticallybeanypositiverealnumber)arequitestableforportfoliosthatexhibit significantbetachanges(rangingfrom6.07to11.10),confirmingtherobustnessofthechosenthresholdvariable,RVIX.
explanationtoSMBandHMLreturndifferentials.Ontheotherhand,theanalysisofalphasimplied by the proposed V-CAPM helps us uncover an important aspect of size and value vs. growth puzzles. In particular,we document that size and value strategies yield significantand positive risk-adjustedreturnsduringcalmtimeswhenuncertaintyaboutnear-termaggregatevolatilityis low.Ontheotherhand,thetrade-offforsizeandvaluestrategiesisthattheyhaveextremelybad (significantandnegative)risk-adjustedreturnsattimesofhighuncertaintyaboutexpected volatil-ity.
PortfolioSharperatiosalsoofferasimilarvolatility-basedtime-varyingriskexplanationtosizeand valuevs.growthanomalies.Inperiodswhenuncertaintyaboutaggregatevolatilityislow,thestrategy inthesmallest(value)decileportfolioscommandhigherreward-to-variabilityratiosasopposedtothe biggest(growth)decileportfolios.However,inperiodsofhighuncertaintyaboutaggregatevolatility, investorsexperiencemuchworsereward-to-variabilityratiosforthesmallest(value)decileportfolio relativetobiggest(growth)decileportfolio.Theresultsconfirmourhypothesisthatmarket’s expec-tationofuncertaintyaboutaggregatevolatilityisanimportantdeterminantofinvestors’assessment ofriskandexpectedreturns.Changesinbetasandrisk-adjustedreturnsduringperiodswith differ-entlevelsofuncertaintyaboutaggregatevolatilityexpectationscontributetoourunderstandingof whysmallandvaluestocksonaverageearnhigherreturnsthantheirbigandgrowthcounterparts. Weshowthatinvestinginsmallandvaluestocksareriskystrategiesinperiodswhenthereishigh uncertaintyaboutexpectedaggregatevolatility,andthusinvestorsgetcompensatedfortheriskthat theyaretakingagainstthisuncertainty.
We finallytest thepricingimplicationsof theproposedV-CAPMbydividingthesample into periods of highand low uncertainty aboutexpected aggregate volatility and by estimating the betas and the corresponding risk premium in the cross-section. To avoid the problem of fac-tor structure related biases in the estimation procedure, we estimate the betas and the risk premia at the individual stock level rather than at the portfolio level. We start with examin-ingtherelationshipbetweenstocks’betasand futurereturns.To thatend, wefirstestimatethe beta loadings via monthly regressions using daily returns as in Ang et al. (2006a,b). Classify-ingbetasashighuncertainty(RVIX>9.33)vs.lowuncertainty(RVIX<9.33)betas,weformdecile portfolios each month by sorting individual stocks according to theirbetas. We then examine out-of-sampleaveragedecilereturnsforthefollowingmonthtoinvestigatewhetherstocks’beta exposures determined by uncertainty about aggregate volatility explain thecross-sectional dis-persionintheirnext-monthreturns.Univariateportfoliosortsindicatethatstocksinthehighest uncertainty(calm)betadecileunderperform(outperform)stocksinthelowestuncertainty(calm) beta decile by 0.72% (0.60%) per month. Furthermore, the differences in risk-adjusted returns (CAPM and FF 3-factor alphas) of portfolios with highest and lowest exposure to uncertainty betasarealsonegativeandstatisticallysignificant.Theresultsarerobusttousingvalue-weighted returns.
Wefinallyestimatethecorrespondingriskpremiaforuncertaintyandcalmbetasusingthe stan-dardFamaandMacBeth(1973)cross-sectionalregressionmethodology.Consistentwiththeprevious studies,wedocumentaninsignificantmarketriskpremiumthroughoutthefullsampleperiod, con-firmingtheinabilityofastaticversionofCAPMtoexplainthecross-sectionofstockreturns.Onthe otherhand,wedocumentapositiveandsignificantmarketriskpremiumduringcalmtimeswhen uncertaintyaboutaggregatevolatilityislow.Thisresultisrobusttotheinclusionofdifferentfactor exposuressuchasSMBbeta,HMLbetaandMOMbetaaswellasvariousfirmcharacteristicssuchas idiosyncraticvolatility,size,book-to-marketratio,andfirm-levelmomentum.Onthecontrary, dur-ingperiodsofhighuncertaintyaboutaggregatevolatility,themarketriskcommandsasignificantand negativepremium,howeveritssignificancedisappearswhendifferentportfolioexposuresandfirm characteristicsareincluded.Theresultsimplyamajorimprovementoverpricingrelativetostatic CAPMbyre-establishingapositivemarketriskpremiumduringcalmperiodswhenuncertaintyabout aggregatevolatilityisfairlylow.
Theremainderofthepaperisorganizedasfollows.Section2introducesthethresholdV-CAPM andtherelatedeconometricframework.Section3presentsdataandsomestylizedfacts.Section4 documentsempiricalfindingsfortime-seriesandcross-sectionaltestsoftheproposedV-CAPM.The finalsectionoffersconcludingremarks.
2. ThethresholdCAPM
Tocapturetheeffectofuncertaintyaboutaggregatevolatilityexpectationsonmarketbeta,we startwiththefollowingconditionalCAPM:
E
ri,t+1Zt=˛i+ˇtErm,t+1|Zt+εit+1, (1)whereri,t+1istheexcessreturnonasseti,rm,t+1istheexcessreturnonthemarketportfolioandE
istheexpectationoperator.ˇtcapturestime-variationinmarketbetas,andZt istheconditioning
informationoninvestors’assessmentofnear-termaggregatevolatilityrisk.Usingmonthlyrangeof theVIXindexasaproxyforinvestors’informationsetforexpectedchangesinaggregatevolatility,we modeltime-variationinbetasasinFersonandHarvey(1999)19:
ˇt=ˇ11{Zt≤}+ˇ21{Zt>}, (2)
where1{}istheindicatorfunctionandisthethresholdparameterforaggregatevolatility.Combining Eqs.(1)and(2),wehavethefollowingthresholdvolatilityCAPM:
ri,t+1=
˛11{Zt≤} +˛21{Zt>} +ˇ11{Zt≤}+ˇ21{Zt>} rm,t+1+εi,t+1, (3)whereZtisthemonthlyrangeoftheVIXindex(RVIX)thatsummarizesinvestors’informationset
regardingtheevolutionofuncertaintyaboutnear-termaggregatevolatility. 2.1. Econometricmodel
Theobservedsampleis{rt+1,rm,t+1,Zt},t=1,...,T−1.Therandomvariablesrt,rm,t,andZtare
real-valued.ThethresholdvariableZtisassumedtohaveacontinuousdistribution20.Thethreshold
regressionhasthesameformatasinEq.(3),whichcanberewrittenas
rt+1=xt+1+ıxt+1() +et+1 (4)
wherext+1=rm,t+1,xt+1() =xt+11{zt≤},=ˇ2andı=ˇ1−ˇ2.
Theabovemodelcanfurtherbegeneralizedtothecasewhereonlyasubsetofparametersswitches betweentheregimesandtothecasewheresomeregressorsonlyenterinoneofthetworegimes. Also,takesvaluesinaboundedsubsetoftherealline,.Thisappliestothecaseofourconditioning variableRVIX,whichisboundedbelowbyzerobydefinition.Weassumermt,Zt,andetarestrictly
stationaryergodicand-mixing21. 2.2. Testingforathreshold
Weusetheheteroskedasticity-consistentLagrangeMultiplier(LM)testforathreshold,asinHansen (1996).WetestforthenullofH0:ı=0againstH1:␦ /= 0.Ifthenullisrejected,thisimpliesasignificant
changeinbetaswithrespecttolevelsaboveorbelowthresholdRVIX.
Forall∈ wehavethefollowingLMstatisticsforthenullofnothreshold: LMT() =T
R ()ˆ R ˆV∗T() R
−1R ()ˆ
,19SeeSection2fordetailsontheconstructionoftheconditioningvariableRVIX. 20SeeHansen(2000)fordetailedexplanationsrelatedtotheassumptions. 21The-mixingcoefficientssatisfy
m1/2<∞.The-mixingassumptioncontrolsthedegreeoftimeseriesdependence andallowstheprocessestobeautocorrelatedandheteroskedastic,andissufficientlyflexibletoembracemanynon-lineartime seriesprocesses,includingthresholdautoregressions.
where, R= [0,I] , ˆ ()=
ˆ(), ˆı()= T t=1 x∗ t+1()x∗t+1() −1
T t=1 x∗ t+1()rt+1 , x∗ t+1()= [xt+1,xt+1()] , ˆ V∗ T()=MT()−1V˜T()MT()−1, MT()=T1 T t=1 x∗ t+1()x∗t+1(), ˜ VT()= 1T T t=1 x∗ t+1()x∗t+1()e˜2t+1,
andwhere ˜etisobtainedfromtherestrictedleastsquares.OnelimitationoftheLMtestisthelarge
samplelimitforthesup-LM,whichisnotnuisancefreebecausethethresholdisnotidentifiedunder thenullofno-thresholdeffect.Becauseofthisissue,Hansen(1996)suggestsabootstrapanalogofthe sup-LMtestandshowsthatthisbootstrapmethodyieldsasymptoticallycorrectp-values.Weusethe bootstrapanalogfollowingthestepsoutlinedinHansen(1996)andestimatetheunknownthreshold parameter,,asinHansen(2000).
3. Data
ThemarketandstockreturndataisfromCenterforResearchin SecurityPrices(CRSP) value-weightedmarketindexforallNYSE,AMEX,andNASDAQstocks.Therisk-freerateistheone-month T-BillrateobtainedfromIbbotsonAssociates.DataonVIXandVXOisobtainedfromChicagoBoard OptionsExchange’s(CBOE).ThesamplecoverstheperiodfromJanuary1986toDecember2012,witha totalof324months22.Thetestportfoliosconsistofstockssortedaccordingtotheirmarket capitaliza-tions,andbook-to-marketratios.Moreprecisely,weuse10portfoliossortedaccordingtotheirmarket capitalizations,10portfoliossortedaccordingtotheirbook-to-marketratios,and2factorportfolios SMBandHML23.Forcross-sectionaltestsperformedinSection4.3,weusetheCRSPuniversecovering allNYSE/AMEX/NASDAQcommonstockswithsharecodes10and11.
Inordertoproxyinvestors’expectationsabouttheevolutionofnear-termaggregatevolatility,we usethemonthlyrangeoftheVIXindex(RVIX).SimilartoChou(2005),wedefineRVIXinagiven monthas:
RVIXt= Max
VIX− Min
VIX, = 1,2,...,T (5) wheredenotestradingdaysinagivenmonth,andtdenotesmonths.Takingthedifferencebetween themaximumandminimumlevelofVIXindexinagivenmonth,RVIXsummarizesinvestors’ expec-tationsregardingchangesinnear-termaggregatevolatility.
22VIXdataisavailablefromJanuary1990onwards.Inordertohaveasmuchdataaspossible,weusetheVXOindex(whichis basedonS&P100indexoptions)fromJanuary1986toDecember1989,andtheVIXindexfromitsintroductioninJanuary1990 onwards.Theresultsremainunaffectedwhenwelimitthesampleperiodto1990–2012usingtheVIXindexonly,oromitting VIXandusingVXOthroughoutthe1986–2012period.
23SMB(SmallMinusBig)istheaveragereturnonthethreesmallportfoliosminustheaveragereturnonthethreebig portfolios,andHML(HighMinusLow)istheaveragereturnonthetwovalueportfoliosminustheaveragereturnonthetwo growthportfolios.
Rangebasedvolatilitymeasureshavegainedrecentinterest,andtheyfarequitewellinpredicting futurevolatility24.Tothebestofourknowledge,this isthefirststudytoproposea rangebased measureoftheVIXindex25.Wefurthertestedabatteryofvolatilitymeasuresrangingfromstatistical andhistoricalmeasuresofvolatilitysuchasstandarddeviationofreturns,squaredreturns,GARCH basedvolatilityestimatestoforward-lookingmeasuresofvolatilitysuchaschangeintheVIXindex andS&P500straddlereturns26.TheproposedRVIXtogetherwithS&P500straddlereturnsarethe mostsuccessfulincapturingtime-variationinbetas.
WearguethatthesuccessofRVIXindetectingchangesinbetasisduetoitsabilitytocharacterize uncertaintyaboutfutureaggregatevolatilitymuchbetterthanalternativemeasures.Byusinga mea-surewhichessentiallycapturesvolatilityofoption-impliedaggregatevolatility,wehavetheadvantage offirstidentifyinginvestors’expectationsabouttheevolutionofnear-termaggregatevolatility(VIX), andsecondmeasuringthedegreeofuncertaintyinexpectedaggregatevolatilitycapturedbytherange ofthethisforward-lookingoption-impliedaggregatevolatilitymeasure(RVIX).Aggregatevolatility riskhasbeendocumentedtobeanimportantfactorthatdeterminesinvestors’risk-returntradeoff andtime-variationininvestmentopportunityset27.BecauseRVIXisessentiallyaproxyforthedegree ofuncertaintyregardingaggregatevolatilityexpectations,weexpectRVIXtobeastrongconditioning variablethatcapturesuncertaintyininvestors’informationsetregardingaggregatevolatilityrisk,and hencetohaveimplicationsregardingassetpricing,portfolioallocation,andstockreturnpredictability asanimportantconditioningvariable.
WefurthertestwhetherthechosenconditioningvariableRVIXiscorrelatedwithotherbusiness cyclemeasuresdocumentedintheliterature28.Towardsthatend,weareparticularlyinterestedin thedividendyield(DIV)oftheS&P500index,thedefaultspread(DEF)whichthespreadbetweenBAA andAAAratedcorporatebondyields,thetermspread(TERM)whichisthespreadbetween10-year, 1-yearU.S.governmentbondyieldsandtheshort-termtreasurybillrate(TB)andtheVIXindex,allof whichhavebeendocumentedasstrongpredictorsofbusinesscyclesandhencetheconditionalCAPM informationset29.Tocheckwhetherourresultsarenotaffectedfromapotentialcorrelationwiththe businesscyclevariablesdocumentedintheliterature,wecreateanorthogonalizedmeasureofRVIX (RVIXORTH),whichisdefinedastheresidualtermfromthefollowingregression:
RVIXt=˛+ˇMKTMKTt+ˇSMBDIVt+ˇDEFDEFt+ˇTERMTERMt+ˇTBTBt+ˇVIXVIXt+εt (6)
Table1reportsthesummarystatisticsofRVIX,themarketportfolio,aswellastheorthogonalized versionofRVIX(RVIXORTH),andmacrovariablesusedintheanalysis.Lookingatthemean(6.66), median(4.98),theminimum(0.92)andthemaximum(129.04)oftheRVIX,onecansaythatVIX index(expectationsofnear-termmarketvolatility)and itsrange(expectedchangesinnear-term aggregatevolatility)arequitestableanddonotmovesignificantlyinmostofthemonthsduringour sampleperiod.Withoutmuchsurprise,themaximumlevelofRVIXwasrecordedinOctober1987, wheretheVXOindexskyrocketedfromitsminimumvalueon3rdOctober1987of21.15pointsto itshistoricalmaximumof150.19pointsonblackMonday.Finally,similartothenegativecorrelation
24SeeAlizadeh,Brandt,andDiebold(2002),Chou(2005),BrandtandJones(2006),ChouandLiu(2010),Harris,Stoja,and Yilmaz(2011)andBannouh,Martens,andvanDijk(2013)forarticlesthatmotivatetheuseofrangebasedvolatilitymeasures indifferentsettings.
25PreviousstudiessuchasGarmanandKlass(1980)andParkinson(1980)aswellasmanyothersusethelogarithmic trans-formationofthestockpriceasameasureofstockvolatility.Althoughtheasymptoticpropertiesandforecastingpowerof price-basedrangemeasureshasbeenextensivelydocumented,asfarastheauthorsareawareof,thisisthefirstpaperthat appliestheconceptofrangetoimpliedvolatility.TohaveabetterunderstandingofthepropertiesofRVIX,wecheckthe predic-tiveabilityofRVIXtoforecastrealizedvariance(VAR)asdefinedinWelchandGoyal(2008).Thepairwisecorrelationbetween RVIXinmontht-andVARinmontht+1is0.27.
26ThereaderisreferredtoSection4.4foradetaileddiscussionofresults.
27SeeCampbell(1993),Chen(2002),Angetal.(2006a,b)andBarinov(2012)forpapersthatdocumentpricingofaggregate volatilityrisk.
28SeeAngandChen(2002)andAngandBekaert(2002)forstudiesthatdocumentincreaseincorrelationsduringrecessions. 29DividendyielddataisfromRobertShiller’swebsite(http://www.econ.yale.edu/∼shiller/data.htm),governmentand cor-poratebondyieldsarefromSt.LouisFedwebsite(https://research.stlouisfed.org/fred2/),andshort-termTreasurybillratesare fromKenFrench’sdatalibrary,(http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/datalibrary.html).
Table1
Descriptivestatistics.
PanelA:summarystatistics
MKT RVIX RVIXORTH TERM DEF DIV TB Mean 0.89 6.66 0 1.43 1.00 0.02 0.31 Median 1.49 4.98 0.04 1.47 0.91 0.02 0.37 Maximum 12.88 129.04 84.04 3.40 3.38 0.39 0.79 Minimum −22.64 0.92 −16.67 −0.41 0.55 0.01 0 Std.Dev 4.61 8.36 6.11 1.07 0.40 0.01 0.20 Skewness −0.89 10.25 8.60 0.10 2.87 0.34 −0.11 Kurtosis 5.48 144.02 121.09 1.82 14.94 1.85 2.08 PanelB:correlations MKT 1 RVIX −0.41 1 RVIXORTH 0 0.73 1 TERM −0.04 0.09 0 1 DEF −0.08 0.28 0 0.29 1 DIV 0.02 0.08 0 0.13 0.32 1 TB 0.04 −0.05 0 −0.71 −0.34 0.36 1
Thistablereportsthedescriptivestatisticsformonthlyreturnsonthemarketportfolio(MKT),andmonthlyrangeofthe VIXindex(RVIX),aswellasorthogonalizedmeasureofRVIX(RVIXORTH),and4businesscyclerelatedmeasures.The mar-ketportfolioistheCRSPvalue-weightedindexforallNYSE,AMEX,andNASDAQstocks.RVIXisthedifferencebetween maximumandminimumlevelofVIXinagivenmonth,i.e.RVIXt=Max{VIX}−Min{VIX},=1,2,...,T,wheredenotes tradingdaysinagivenmonth,andtdenotesmonths.DIVisthedividendyieldoftheS&P500index.DEFisdefinedasthe spreadbetweenBAAandAAAratedcorporatebondyieldandTERMisthespreadbetween10-yearand1-yearU.S. gov-ernmentbondyields.TBistheone-monthTreasuryBillrate.RVIXORTHistheresidualtermfromthefollowingregression: RVIXt=˛+ˇMKTMKTt+ˇSMBDIVt+ˇDEFDEFt+ˇTERMTERMt+ˇTBTBt+ˇVIXVIXt+εtThesamplecoverstheperiodfromJanuary 1986toDecember2012(324months).FortheperiodcoveringJanuary1986toDecember1989,VIXisreplacedbyVXOwhich isbasedonS&P100indexoptions.Allreturnfiguresareinpercentages.
documentedinpreviousstudiesbetweentheVIXindexandmarketreturns,thecorrelationbetween
RVIXandthemarketis−0.41.
3.1. Stylizedfacts
Thissectiondocumentssomestylizedfactsaboutthechosenthresholdparameter,marketreturns,
andtheempiricallydocumentedsizeandvaluevs.growthanomalies.
First,lookingatFig.1,onecanseethattheproposedconditioningvariableRVIXindeedtracks
significantnegativemarketmoves.Giventheempiricalevidencethatnegativemarketmoves are mostassociatedwithincreasesinaggregatevolatility,ournovelmeasureRVIXisessentiallycapableof providingarelationshipbetweentheevolutionofnear-termmarketvolatilityanddownwardmarket moves.
Next,weconductasimpleexercisetoexaminereturnsondifferentsizeandbook-to-market portfo-liosindifferentvolatilityregimesinmoredetail.UsingthresholdestimatesoftheRVIXindex,wedivide thesampleintotworegimes,whereregime1(2)representscalm(uncertain)monthsinwhichRVIX isbelow(above)theestimatedthresholdleveloftheassociatedportfolio.Thiswayofdecomposing returnsintocalmanduncertainmonthsgivesusinterestinginsightsregardinginvestor’ expecta-tionsaboutuncertaintyinnear-termaggregatevolatilityandportfolioreturndynamics.Forexample, lookingatcolumns5and10ofTable2,onecanseethatassetclasses,regardlessoftheirportfolio characteristics,losemuchmorewhenmarketvolatilityisexpectedtobehighlyvolatile.Thisisinline withHsuandLi(2009)whodocumentthatvolatileperiodscoincidewithbearmarkets.Ontheother hand,columns2,3,7and8documentthetypicalsizeandvaluevs.growthanomalies.More particu-larly,staticCAPMfailstoofferaclearandlinearrelationshipbetweenbetasandportfolioreturns,i.e. high(low)returnsarenotalwaysjustifiedbyhigh(low)CAPMbetas.However,lookingatcolumns 4,5,9,and10ofTable2,onecangaininterestinginsights.Forexample,incalmmonths(whenRVIX
0 20 40 60 80 100 120 -.25 -.20 -.15 -.10 -.05 .00 .05 .10 .15 86 88 90 92 94 96 98 00 02 04 06 08 10 12 MARKET RVIX
Fig.1.Time-seriesofRVIXandmarketreturns.
Table2
Stylizedfactsaboutportfolioreturns.
Size Beta Full sample Regime1 (Calm) Regime2 (Volatile) B/M Beta Full sample Regime1 (Calm) Regime2 (Volatile) Small 1.0163 0.0097 0.0176 −0.0365 High 1.0557 0.0115 0.0187 −0.0298 Decile2 1.1604 0.0096 0.0170 −0.0330 Decile2 0.9864 0.0107 0.0160 −0.0203 Decile3 1.1434 0.0104 0.0173 −0.0297 Decile3 0.9716 0.0091 0.0149 −0.0239 Decile4 1.1262 0.0094 0.0158 −0.0271 Decile4 0.9512 0.0102 0.0147 −0.0156 Decile5 1.1380 0.0104 0.0164 −0.0240 Decile5 0.9467 0.0090 0.0135 −0.0167 Decile6 1.0713 0.0104 0.0157 −0.0213 Decile6 0.9029 0.0096 0.0143 −0.0179 Decile7 1.0673 0.0108 0.0164 −0.0205 Decile7 0.8487 0.0098 0.0153 −0.0221 Decile8 1.0795 0.0103 0.0156 −0.0203 Decile8 0.8445 0.0102 0.0141 −0.0124 Decile9 1.0101 0.0101 0.0148 −0.0170 Decile9 0.9228 0.0093 0.0129 −0.0109 Big 0.9449 0.0087 0.0118 −0.0096 Low 1.0501 0.0090 0.0116 −0.0066 SMB 0.0716 0.0010 0.0058 −0.0269 HML 0.0059 0.0026 0.0071 −0.0231 Market 0.0089 0.0129 −0.0143 Market 0.0089 0.0128 −0.0143 Thistablepresentsthereturnsonseveralportfoliosthathavebeenusedastestassetsinthisstudyandthemarketportfolio duringthefullsampleperiodfromJanuary1986throughDecember2012(324months)andtwodifferentvolatilityregimes. Sizerepresentsportfolioswhichcontainstockssortedwithrespecttotheirmarketcapitalizations.B/Mrepresentsportfolios whichcontainstockssortedwithrespecttotheirbook-to-marketratios.SMBisaportfoliothatislonginstocksinthesmallest decileandshortinstocksinthebiggestdecile.HMLisaportfoliothatislonginstockswhichareinthehighestB/Mdecileand shortinstockswhichareinthelowestB/Mdecile.
isbelowtheestimatedthreshold),onecanseealmostamonotonousdecreaseinreturnsgoingfrom
smallandvalueportfoliosthroughbigandgrowthportfolios.Theoppositeistrueforepisodeswhen
uncertaintyaboutaggregatevolatilityishigh,whensmallandvalueportfoliosbecometheworst
per-formers.Despitetheirhigheraveragereturnsrelativetobigandgrowthportfolios,smallandvalue
stockportfoliosbecometheworstperformersattimesofhighuncertaintyaboutexpectedaggregate
volatilitywhenthemarketisexpectedtodobadly30.Ontheotherhand,bylosinglessthanthe
mar-ketportfolio,bigandgrowthportfolioscanbeseenasrelativelysaferassetclassesduringuncertain periodsaboutaggregatevolatilityandmarketconditions.
Thepreliminaryfindingsinformallyconfirmourhypothesesthatsizeandbook-to-market portfo-lioshavedifferentsensitivitiestomarketriskduringperiodsofdifferentexpectationsregardingthe
30Thisalsoholdsforthezero-costSMBandHMLportfolios,whichearnonaverage58and71basispointspermonthduring calmmarketconditions,butwhichbecomeextremelyriskystrategiesandlose269and231basispoints,respectively,during highexpectedvolatilityperiods.
evolutionofaggregatevolatility.Thus,wepositthatanassetpricingmodelthatcorrectlytakesinto accountthisvolatility-basedtimevariationinriskandreturnsisexpectedtodobetterinpricingand inexplainingsizeandvaluevs.growthanomalies.
4. TestsofV-CAPM
Webeginbyexaminingwhethertherearestatisticallysignificantregimeshiftsinbetasdueto changesininvestors’expectationsregardinguncertaintyaboutaggregatevolatilityandmarket condi-tions.OurconditioningvariableisrangeoftheVIXindexandTable3reportstheassociatedbootstrap p-valuesforthesup-LMtest.Thenullhypothesisisthatthereisnosignificantregimeshiftin portfo-liobetas.Accordingtobootstrapp-valuespresentedinTable3,therearesignificantregimechanges inbetasofmostportfolios.Forportfoliossortedwithrespecttomarketcapitalizations,eightoutof tenexperiencesignificantchangesintheirbetasbetweenuncertainandcalmperiodsaboutexpected aggregatevolatility.Forportfoliossortedwithrespecttobook-to-marketratios,theevidenceindicates aregimeshiftinbetasofsevenoutoftenportfolios.SMBandHMLportfoliosalsoexhibit signifi-cantregimeshiftsinbetas.TheresultsarealsorobusttousingtheorthogonalizedversionofRVIX (RVIXORTH).TakingintoaccountthatthethresholdparameterRVIXcapturesinvestors’expectations aboutuncertaintyinnear-termvolatility,uncertaintyabouttheevolutionofaggregatevolatilityseems tobeanimportantdeterminantoftheirassessmentofaggregateriskconditionsandanasset’s sensitiv-itytooverallmarketrisk.Theresults,ifpersistent,offernewevidenceandanalternativeexplanation totheempiricallyobservedsizeandvaluevs.growthanomalies.
TheconditionalCAPM modelsmayhavea tendencytooverstatethetime variation,and asa result,continuousapproximationsofCAPMproducehighlyvolatilebetaestimates.Thisisfurther confirmedwiththeevidencereportedinBraunetal.(1995),whouseabivariateEGARCHmodelto estimateconditionalbetasanddocumentweakevidenceoftimevariation.Ontheotherhand,our thresholdmethodologyusingRVIXasaconditioningvariablesuggeststhatportfoliobetasare sta-bleduringdifferentexpectedvolatilityregimes,howeverinvestorsupdatetheirbetaestimateswhen theirexpectationsregardingtheuncertaintyaboutnear-termvolatilitychangeconsiderably.
Table3
Bootstrapp-valuesfor10sizeandB/Mportfolios.
Size RVIX RVIXORTH B/M RVIX RVIXORTH Small 0.000*** 0.000*** High 0.012** 0.000*** Decile2 0.000*** 0.024** Decile2 0.009*** 0.001*** Decile3 0.000*** 0.035** Decile3 0.013** 0.000*** Decile4 0.015** 0.042** Decile4 0.144 0.148 Decile5 0.025*** 0.051* Decile5 0.029** 0.009** Decile6 0.140 0.044** Decile6 0.067* 0.022** Decile7 0.016** 0.003*** Decile7 0.008*** 0.030** Decile8 0.144 0.009*** Decile8 0.435 0.642 Decile9 0.056* 0.018** Decile9 0.219 0.167 Big 0.008*** 0.007*** Low 0.000*** 0.000*** SMB 0.009*** 0.058** HML 0.001*** 0.000*** Thistablereportsthebootstrapp-valuesofthemodifiedsup-LMtestsuggestedbyHansen(1996).Wetestthenullhypothesis ofnosignificantregimeshiftsinportfoliobetasduetochangesinthelevelofuncertaintyregardingaggregatevolatility expec-tations,capturedbyRVIX,aswellasanorthogonalizedmeasureofRVIX(RVIXORTH).RVIXORTHistheresidualtermfromthe followingregression:
RVIXt=˛+ˇMKTMKTt+ˇSMBDIVt+ˇDEFDEFt+ˇTERMTERMt+ˇTBTBt+ˇVIXVIXt+εt
Sizerepresentsportfolioswhichcontainstockssortedwithrespecttotheirmarketcapitalizations,andB/Mrepresentsportfolios whichcontainstockssortedwithrespecttotheirbook-to-marketratios,respectively.ThesampleperiodcoversJanuary1986 toDecember2012(324months).
Table4
Thresholdestimatesfor10sizeand10book-to-marketportfolios.
PanelA:10sizeportfolios
CAPMbeta Betaforregime1 Betaforregime2 Thresholdestimate
Small 1.0163 0.9093 1.2350 8.80 Decile2 1.1604 1.0772 1.1662 8.42 Decile3 1.1434 1.0453 1.2060 9.33 Decile4 1.1262 1.0401 1.1874 9.33 Decile5 1.1380 1.0684 1.1884 9.33 Decile6 1.0713 1.0330 1.2516 17.69 Decile7 1.0673 1.0109 1.1067 9.33 Decile8 1.0795 1.0478 1.0919 6.07 Decile9 1.0101 0.9723 1.0190 9.33 Big 0.9449 0.9762 0.9223 9.33 SMB 0.0716 −0.0616 0.3026 9.33 PanelB:10B/Mportfolios
CAPMbeta Betaforregime1 Betaforregime2 Thresholdestimate
High 1.0557 0.9776 1.0831 9.58 Decile2 0.9864 0.9357 1.0513 9.58 Decile3 0.9716 0.9247 1.0606 9.58 Decile4 0.9512 0.9038 1.0399 9.58 Decile5 0.9467 0.8687 1.0648 11.10 Decile6 0.9029 0.8308 1.0127 11.10 Decile7 0.8487 0.7983 0.8906 10.92 Decile8 0.8445 0.7425 0.9266 15.00 Decile9 0.9228 0.8373 0.9944 10.78 Low 1.0501 1.1249 0.9892 9.33 HML 0.0059 −0.1514 0.0930 9.33
ThistablereportstheunconditionalCAPMbetas,thethresholdbetaestimateswithrespecttolowandhighvolatilityregimes, andtheirassociatedthresholdvolatilityestimates,proxiedbyS&P500at-the-moneystraddlereturns.PanelsAandBpresent resultsforportfoliossortedwithrespecttomarketcapitalizations,andbook-to-marketratios,respectively.SMBisaportfolio thatislonginstocksinthesmallestdecileandshortinstocksinthebiggestdecile.HMLisaportfoliothatislonginstocks whichareinthehighestB/MdecileandshortinstockswhichareinthelowestB/Mdecile.Thesamplecoverstheperiodfrom January1986toDecember2012(324months).Regime1(2)correspondstolow(high)uncertaintyaboutvolatilityregimes wheremonthlyRVIXislower(higher)thantheestimatedthresholdlevel.
4.1. Therelationbetweenaggregatevolatilityexpectationsandbeta
Havingdetectedsignificantregimeshiftsinbetasformostoftheportfolios,weproceedtotestthe
magnitudeofthischange,andestimateassetbetasandtheirassociatedthresholdparametersduring
uncertainandcalmperiodsaboutexpectedaggregatevolatility.Table4reportsthestaticCAPMbetas,
betasestimatedviatheV-CAPMincalm(regime1)andhighuncertainty(regime2)regimes,together withthethresholdestimateofRVIX,whichdeterminesthechangeinuncertaintyaboutaggregate volatilityexpectations,above(orbelow)whichinvestorsre-assessastock’sriskiness.
BeforegoingintodetailedanalysisofportfoliobetasTable4,lookingatthelastcolumnofPanelA, onecanseethattheestimatedthresholdlevelofRVIXisverystableacrosssizeportfolios,whichis estimatedat9.33in7ofthe11cases31.GiventhatRVIXcantakeonanypositiverealnumber,this con-sistentlevelofthethresholdestimateaffirmstherobustnessofthethresholdestimationprocedure, theproposedmodel,andthechosenthresholdparameter,RVIX.WearguethatthestabilityofRVIX acrossportfoliossignalstothedegreeofuncertaintythatthemarketviewsascriticalregarding aggre-gatevolatilityexpectations.WhenrangeofVIXindexinagivenmonthisbelow9.33,theuncertainty
31Thethresholdlevelsof17.69and6.07fordeciles6and8mightseemasbigdeviationsfrom9.33atfirstsight,however, notethatthesearethetwoportfolioswherethesup-LMtestwasunabletodetectsignificantregimechanges.Thisisalsothe casefordecile8ofbook-to-marketsortedportfolios,whichisdetectedasaportfoliowithinsignificantregimeshiftandhasa relativelyhighthresholdestimateof15.00.
inthemarketistolerableandbetasremainunaffected.However,whenRVIXismorethan9.33,this indicatesthatuncertaintyinthemarketregardingtheevolutionofexpectedaggregatevolatilityhas increased,andhenceinvestorsupdatetheirinformationsetandrisk-returndynamicswithrespectto thisinformation,whichisalsoreflectedinbetasaccordingly.
Next,adetailedanalysisofcolumns3and4ofPanelArevealsimportantinsightsabouthowthe riskinessofdifferentsizesortedportfolioschangesfromonevolatilityregimetotheother.Wenote significantchangesinbetariskofsizesortedportfolios.Inparticular,betasofsmallstockportfolios increaseconsiderablyattimesofhighuncertaintyaboutexpectedaggregatevolatility.Furthermore, itisonlythebiggestdecileportfolio,whichexhibitsadecreaseinitsbetaduringhighuncertainty episodes.
Theabovefindingsimplythatinvestorsre-assesstheriskinessofsizesortedportfolioswhenrange oftheVIXindexisabove(orbelow)thethresholdlevelof9.33.Forexample,whenaggregatevolatility isexpectedtobevolatilesignificantly(i.e.whenRVIXisabovethethreshold),investorsre-estimate thebetaforthesmallestdecileportfolio,andupdateitfrom0.91incalmperiodsto1.24inuncertain periods.Similarly,theriskinessofthebiggestdecileportfoliochangeswhentheRVIXisabove(or below)thethresholdlevelof9.33.Morespecifically,thebetaforthebiggestportfoliodropsfrom0.98 incalmvolatilityperiodsto0.92inuncertainperiods.Furthermore,thebetadifferentialbetweenthe smallestandbiggestportfolios(SMB)increasesfrom−0.06inthelowexpectedvolatilityregimeto 0.30inthehighuncertaintyregimeaboutexpectedaggregatevolatility.
Ourfindingsimplythatthesensitivityofanasset’sreturnwithrespecttothelevelofuncertainty regardingexpectedaggregatevolatilityisanimportantdeterminantofanasset’sriskiness.Thishas clearimplicationsonpricingandportfolioallocation.Forexample,byhavingalowercovariancewith themarketattimesofhighuncertaintyaboutmarketvolatility,biggestdecileportfoliotendstolose lessthananyothersize-basedstrategyduringvolatileperiods.Alsogiventhatvolatileepisodes usu-allycoincidewithdownwardmarketmovesandrecessions,astrategyinvestedinthebiggestdecile portfolioappearstoberelativelylessriskyforrisk-averseinvestors,whoarereluctanttolosewealth duringthosetimes.Thisimpliesademandforbigstocks,thuspushingtheirpricesupandresulting inloweraveragereturns.Similarly,theriskofsmallstockportfoliosgoesupwhennear-term aggre-gatevolatilityisexpectedtobevolatile.Becauseuncertaintyandincreasesinaggregatevolatilityare mostlyassociatedwithbadmarketconditionsanddeteriorationsininvestorwealth,bycorrelating highlywiththemarketattimesofhighuncertaintyaboutaggregatevolatility,smallstocksareviewed asriskierattimeswhenextradollaroflossismuchmoreimportant.
PanelBofTable4offerssimilarresultsforportfoliossortedwithrespecttobook-to-marketratios. Valueportfolioshaveconsistentlyhigherbetasattimesofhighuncertaintyaboutexpectedvolatility, whereasitisonlythegrowthportfoliowhosebetadecreasesduringthosetimes.Theresultsindicate significanttimevariationintheriskassessmentsofvalueandgrowthportfolioswithrespectinvestors’ expectationsaboutuncertaintyinnear-termaggregatevolatility.Investorsviewvaluestocksmuch riskierbecausetheyhaveahighercorrelationwiththemarketattimesofhighuncertaintyabout aggregatevolatility.Similarly,aportfoliostrategyingrowthstockstendstobelessriskyatthose times.TheresultsareinlinewithLettauandLudvigson(2001)andPetkovaandZhang(2005),whoalso documenttimevariationinriskinessandexpectedreturnsofvalueandgrowthstocks,inconditional CCAPMandconditionalCAPMsettings,respectively.
4.2. Therelationbetweenaggregatevolatilityexpectationsandrisk-adjustedreturns
Thedocumentedevidencesofarindicatesthatassetbetaschangesignificantlybetweendifferent volatilityregimes,dependingonwhetherinvestorsexpectsignificantuncertaintyaboutaggregate volatilityornot.Furthermore,theproposedV-CAPMrevealsadistinctivepatternregardingchangeof betariskamongdifferentassetclasses.Moreparticularly,smallmarketcapitalizationandhigh book-to-market(value)portfoliosbecomeriskierattimesofhighuncertaintyaboutaggregatevolatility expectations.Ontheotherhand,bigmarketcapitalizationandlowbook-to-market(growth)portfolios becomelessriskyatthosetimes.Thefindingsofourmodelofferapotentialremedytothestatic CAPManditsfailureinexplainingthewelldocumentedsizeandvaluevs.growthanomalies.Inorder toexaminetherobustnessoftheproposedvolatility-basedtime-varyingbetariskexplanation,and
Table5
ComparisonofJensen’salphas.
PanelA:10Sizeportfolios
˛CAPM ˛V-CAPM,Regime1 ˛V-CAPM,Regime2 Small 0.0636(0.25) 0.5601(1.98**) −2.1113(−4.47***) Decile2 −0.0221(−0.11) 0.3322(1.37) −1.5210(−3.50***) Decile3 0.0680(0.41) 0.3908(1.96*) −1.1887(−2.85***) Decile4 −0.0177(−0.12) 0.2389(1.32) −0.9651(−2.48**) Decile5 0.1753(0.58) 0.2750(1.80*) −0.6550(−2.22**) Decile6 0.1089(0.94) 0.2543(2.00) −0.4421(−1.50) Decile7 0.1521(1.43) 0.3319(3.02***) −0.5203(−1.77*) Decile8 0.0938(0.92) 0.2158(1.94*) −0.4439(−1.86*) Decile9 0.1147(1.49) 0.1997(2.35**) −0.2266(−1.25) Big 0.0111(0.20) −0.0792(−1.27) 0.3373(2.37**) SMB −0.3088(−0.85) 0.2543(1.33) −2.6881(−4.48***) PanelB:10B/Mportfolios
˛CAPM ˛V-CAPM,Regime1 ˛V-CAPM,Regime2 High 0.2306(0.94) 0.5930(2.26**) −1.4147(−3.40***) Decile2 0.2241(1.40) 0.4556(2.34**) −0.6294(−2.27**) Decile3 0.1143(0.61) 0.4352(2.29**) −1.1103(−2.54**) Decile4 0.2190(1.33) 0.3595(1.75*) −0.3378(−1.14) Decile5 0.0443(0.35) 0.1756(1.16) −0.1842(−0.63) Decile6 0.1256(0.89) 0.3015(1.77*) −0.4080(−1.21) Decile7 0.1060(0.73) 0.3309(1.95*) −0.7462(−2.50**) Decile8 0.1566(1.58) 0.1791(1.75*) 0.1433(0.49) Decile9 0.0547(0.62) 0.0167(0.15) 0.3516(1.61) Low −0.0222(−0.18) −0.2426(−1.88*) 0.7474(2.66**) HML 0.0600(0.17) 0.4331(2.53**) −2.4016(−4.10***) ThistablereportsJensen’salphasfortheunconditionalCAPMandforthethresholdvolatilitymodel(V-CAPM)withrespectto lowandhighvolatilityregimes.ThesamplecoverstheperiodfromJanuary1986toDecember2012(324months).Regime1 (2)correspondstolow(high)uncertaintyaboutvolatilityregimeswheremonthlyRVIXislower(higher)thantheestimated thresholdlevel.Thenumbersinparenthesesdenotetheassociatedt-statisticswithNewey-Westcorrectedstandarderrors.***, **,and*denotesignificanceat1%,5%,and10%level,respectively.
toseewhetherinvestors’expectationsaboutuncertaintyinaggregatevolatilityhasasimilar
time-varyingeffectonrisk-adjustedreturns,wenextcompareJensen’salphasandSharperatioswithinthe
fullsample,andincalmandhighuncertaintyregimesdeterminedbythethresholdlevelofRVIX.
4.2.1. ComparisonofJensen’salphas
Itiswell-documentedthatsmallandvaluestocksonaverageproducesignificantlyhigherreturns
thantheirlargeandgrowthcounterparts32.However,lookingatJensen’salphasandbetasofdifferent
sizeandbook-to-marketportfoliosinTable5,weconfirmpreviousstudiesthatstaticCAPMisunable toofferarisk-basedexplanationtotheseabnormalreturns.Ontheotherhand,alphasimpliedby differentvolatilityregimeshelpusuncoveranimportantaspectofsizeandvaluevs.growthpuzzles, offeringavolatility-basedtime-varyingriskexplanation.
Weinitiallydocumentthatsizeandvaluestrategiespayoffattimesoflowexpectedvolatility yieldingsignificantandpositiverisk-adjustedreturns.However,thetrade-offforthesestrategiesis thattheyhaveextremelybad(significantandnegative)risk-adjustedreturnsattimesofhigh uncer-taintyaboutexpectedvolatility.Forexample,astrategyinvestedinthesmallestdecileportfolioearns anaveragerisk-adjustedreturnof56basispointsduringcalmmonths,whereasthesamestrategy yieldsarisk-adjustedreturnof-211basispointsinmonthswhenuncertaintyaboutvolatilityishigh. Similarly,astrategyinvestedinthehighestbook-to-market(value)portfolioearnsanaverage risk-adjustedreturnof59basispointsduringcalmmonths,butyieldsanaveragerisk-adjustedreturn 32Althoughexcessreturnsonsmallstocksoverbigstockshavebeendisappearingduringthelasttwodecades,excessreturns onvaluestocksovergrowthstockshavebeensignificantlypersistentoveryears.
of−142basispointsinmonthswhenuncertaintyaboutvolatilityishigh.SMBandHMLstrategies alsoyieldsimilarandsignificantrisk-adjustedreturnsovercalmanduncertaintyperiods.Onthe con-trary,althoughstrategiesinbiggestandlowestbook-to-market(growth)portfoliosdisappointtheir investorsincalmmonthswithaveragerisk-adjustedreturnsof8and24basispoints,respectively, theyyieldpositiveandsignificantrisk-adjustedreturnsinmonthswhenuncertaintyaboutvolatility ishigh(24and75basispoints,respectively).
4.2.2. ComparisonofSharperatios
Next,welookatanotherpopularmeasureofrisk-adjustedreturnproposedbySharpe(1966,1975). Sharperatioisacommonlyusedmeasuretotracktheperformanceofmutualfundsanditcanbeeasily appliedtomeasurethereward-to-variabilityofanyinvestmentassetorportfolio.Byscalinganasset’s excessreturntothestandarddeviationofexcessreturnsontheasset,itisanidealwayofmeasuring ofreward-to-variabilityofamanagedfundandthesensitivityofreturnsonaninvestmentclassora tradingstrategyperunitofrisktaken33.Themeasureismodelfree,henceitprovidesanindirecttest fortherobustnessofourchosenvolatilityparameterRVIX,aswewillcomparetheSharperatiosoftest assetswithinthewholesamplewiththoseobtainedintwodifferentvolatilityregimesdeterminedby RVIX.Analyzingreward-to-variabilityratiosindifferentvolatilityregimeswillgiveusfurtherinsight abouttherisk-returndynamicsofthetestassetswithrespectinvestors’expectationsregardingthe uncertaintyabouttheevolutionofnear-termvolatility.
LookingatPanelAofTable6,onecandetectnoclearpatterninSharperatiosofportfoliossorted withrespecttomarketcapitalizationswithinthefullsample.Onecanevensaythatduringthesample period,aninvestmentstrategybasedonstocksinthesmallestsizedecilecommandsalowerreward perunitofrisktakenasopposedtoastrategybasedonstocksinthehighestdecile,whichisnot consistentwitharationalrisk-basedexplanation.Ontheotherhand,whenwedecomposethesample intotwovolatilityregimesdeterminedbytheRVIX,weseedifferentrisk-returndynamicsacrosssize sortedportfoliosindifferentvolatilityregimes.Incalmperiods,thestrategyinthesmallestdecile port-foliocommandsahigherreward-to-variabilityratiocomparedtothebiggestdecileportfolio(0.2727 vs.0.2250),howeverinperiodswhenuncertaintyaboutvolatilityishigh,thesituationisreversed, investorsexperienceamuchworsereward-to-variabilityratioforthesmallestdecileportfolioagainst thebiggestdecile(−0.4513vs.−0.1727).ThisdifferentpatterninSharperatiosisalsoconsistentwith ourpreviousresultsdocumentingsignificantdifferencesinbetasandJensen’salphasofthose strate-giesandexplainswhyinvestorswouldwanttobecompensatedfortheextrariskthattheyaretaking byinvestinginsmallstocks.
WeobserveasimilarpatternfortheSharperatiosofportfoliossortedwithrespecttobook-to marketratios.AlthoughthereisnotasignificantinSharperatiosinthefullsample,wedocument thatvalueportfolioscommandhigher(lower)reward-to-variabilityratioscomparedtogrowth port-foliosincalm(uncertain)periods,offeringacoherentvolatility-basedrisk-returnexplanationtothe empiricallydocumentedvaluevs.growthanomaly.Theresultsconfirmourhypothesisthatmarket’s expectationofuncertaintyinaggregatevolatilityisanimportantdeterminantofinvestors’assessment ofrisk-returndynamics.Changesinbetasandrisk-adjustedreturnshelpusuncoverwhysmalland valuestocksonaverageearnhigherreturnsthantheirbigandgrowthcounterparts.Byconditioning assetreturnsusinganovelforward-lookingvolatilitymeasure(RVIX),whichsummarizesinvestors’ expectationsabouttheuncertaintyofnear-termaggregatevolatility,theproposedV-CAPMoffersa volatility-basedtime-varyingriskexplanationtothesizeandvaluevs.growthanomalies.
4.3. Cross-sectionalanalysis
Thetime-seriesanalysesattheportfoliolevelinthefirstpartindicatedistinctexposureofsize andbook-to-marketratiosortedportfoliostouncertaintyaboutaggregatevolatilityexpectations, whichmanifestsitselfwithsignificantchangesinportfoliobetasduringlowandhighuncertainty
33TheexcessreturnontheassetcanbeonanybenchmarksuchastheS&P500returnsortherisk-freerate.Asinmoststudies, wechoosereturnsinexcessoftherisk-freeratetomeasureanasset’sexcessreturn
Table6
ComparisonofSharperatios.
PanelA:10Sizeportfolios
Sfull full SRegime1 Regime1 SRegime2 Regime2 Small 0.1047 6.1800 0.2727 5.2697 −0.4513 8.6296 Decile2 0.0995 6.4755 0.2486 5.5158 −0.3761 9.4052 Decile3 0.1197 6.0513 0.2833 4.9713 −0.3423 9.3709 Decile4 0.1073 5.8601 0.2605 4.8034 −0.3200 9.2201 Decile5 0.1264 5.7654 0.2774 4.7388 −0.2899 9.1022 Decile6 0.1369 5.2880 0.2894 4.3103 −0.2692 8.4899 Decile7 0.1475 5.1879 0.3159 4.1494 −0.2799 8.4778 Decile8 0.1374 5.1926 0.2912 4.2296 −0.2718 8.3341 Decile9 0.1451 4.7889 0.2965 3.8871 −0.2493 7.7732 Big 0.1238 4.4706 0.2250 3.8149 −0.1727 6.9428 SMB −0.0664 3.2503 0.0570 3.2547 −0.5456 2.8644 PanelB:10B/Mportfolios
Sfull full SRegime1 Regime1 SRegime2 Regime2 High 0.1376 6.0808 0.3009 5.1259 −0.3601 8.9403 Decile2 0.1548 4.8706 0.3275 3.9062 −0.2871 7.8923 Decile3 0.1305 4.5919 0.3300 3.5221 −0.3430 7.6930 Decile4 0.1564 4.5171 0.3034 3.7634 −0.2560 7.0277 Decile5 0.1240 4.7423 0.2702 3.7813 −0.2403 7.9366 Decile6 0.1389 4.6370 0.2981 3.7196 −0.2668 7.6018 Decile7 0.1360 4.8830 0.3072 3.9252 −0.3128 7.8432 Decile8 0.1503 4.6758 0.2802 3.8612 −0.1958 7.5451 Decile9 0.1306 4.7558 0.2433 3.9492 −0.1732 7.6979 Low 0.1125 5.1623 0.1829 4.5820 −0.1185 7.5777 HML 0.0180 3.0837 0.0943 2.9435 −0.3368 3.5854 Thistablereportsportfolioex-postSharperatiosandstandarddeviationsforthefullsampleandtwosubsamplesrepresenting twodifferentvolatilityregimes.Regime1(2)correspondstolow(high)uncertaintyaboutvolatilityregimeswheremonthly RVIXislower(higher)thantheestimatedthresholdlevel.PanelsAandBpresentsresultsforportfoliossortedwithrespectto marketcapitalizations,andbook-to-marketratios,respectively.SMBisaportfoliothatislonginstocksinthesmallestdecile andshortinstocksinthebiggestdecile.HMLisaportfoliothatislonginstockswhichareinthehighestB/Mdecileandshort instockswhichareinthelowestB/Mdecile.ThesamplecoverstheperiodfromJanuary1986toDecember2012(324months).
periodsaboutaggregatevolatility.However,itisimportanttonotethatstockscanexhibitsignificant
cross-sectionalvariationwithineachportfolio.Thereforeeventhoughtime-seriesanalysisatthe
port-foliolevelpointtowardsRVIXbeingapotentiallyimportantconditioningvariableinexplainingstock
returns,thisexplanatorypowermightresultfromstocks’othercharacteristics.Inthenextsection,
weexaminewhethercross-sectionaldifferencesinbeta-returnrelationshipareattributabletoRVIX
attheindividualstocklevel,andwhetherbetasimpliedbyhighvs.lowuncertaintyaboutaggregate
volatilityisapricedriskfactorinthecross-section.
4.3.1. UnivariateportfoliosortsbasedonthresholdRVIXbetas
WestartwithexaminingwhetherstockbetasestimatedviathresholdlevelofRVIXcanpredict
thecross-sectionaldifferencesintheirreturns.FollowingAngetal.(2006a,b),weestimateEq.(3)
usingmonthlyregressionwindowswithdailydata.Inlinewithourpreviousresults,weusethemost commonlyobservedRVIXthresholdlevel toidentifylow(high)expectedvolatilityperiods.More particularly,weidentifymonthsinwhichthethresholdparameterRVIXisgreater(less)than9.33 asperiodsofhigh(low)uncertaintyaboutexpectedaggregatevolatility34.Wedenote correspond-ingbetasasˇUNC andˇCALM,respectively.ThesampleistheuniverseofCRSPstockscoveringall
34WefurthertriedsixotherthresholdlevelsofRVIXrangingfrom8.42to11.10,whichcoincidewiththenextcommonly observedRVIXthresholdlevelsinourtestsafter9.33.Theresultsarerobusttodifferentthresholdlevelsfoundinportfoliolevel analyses.
NYSE/AMEX/NASDAQcommonstockswithsharecodes10and11.Theregressionsareestimatedeach monthfromJanuary1986toDecember2010.
Wenextconductportfolio-levelanalysistoinvestigatecross-sectionalpredictivepowerofˇUNC
andˇCALMForeachmonth,fromFebruary1986toDecember2010,stocksaresortedintodecile
port-foliosbasedontheirˇUNCandˇCALM.Ourportfolioformationexerciseusesinformationavailableonly
asoftheformationdate.Henceitavoidspotentiallook-aheadbiasintheestimationofbetas.Decile 1(10)containsstockswiththelowest(highest)betas.Next-monthpost-rankingportfolioreturnsare calculatedusingbothequally-andvalue-weightedweightingschemes,andtheprocedureisrepeated eachmonth.Table7reportsnext-monthreturns,CAPMandFamaandFrench(1993)3-factoralphas ofˇUNCandˇCALM.sorteddeciles.
Univariateportfoliosortsindicateanalmostmonotoneandnegativerelationshipbetweenthe betasandnext-monthaveragereturnswhenuncertaintyaboutexpectedaggregatevolatilityishigh. PortfolioofstockswithlowestˇUNC(portfolio1)earns1.72%permonth,whereasreturnonthe
port-folioofstockswithhighestˇUNC(portfolio10)is1.16%permonth.Thespreadportfoliowhichislong
inthehighestˇUNCstocksandshortinthelowestˇUNCstocks(10−1)losesonaverage0.72%per
monthwithat-statisticof−2.92.Nextmonth’srisk-adjustedreturns(CAPMandFF3-factoralphas) aswellasvalue-weightedreturns(bothrawandrisk-adjusted)alsoimplyanegativebetanext-month stockreturnrelationshipduringperiodsofhighuncertaintyaboutexpectedaggregatevolatility,with alphasandrawreturnsofthespreadportfoliorangingfrom−0.49%to−0.73%withalmostallbutone beingsignificantat5%level.
Ontheotherhand,ourproposedmodelisabletoestablishapositivebetanext-monthstockreturn relationshipduringperiodsoflowuncertaintyaboutexpectedaggregatevolatility.Inparticular, port-folioofstockswithlowestˇCALM(portfolio1)earns1.84%permonth,whereasreturnontheportfolio
ofstockswithhighestˇCALM(portfolio10)is2.44%permonth.Thespreadportfoliowhichislong
inthehighestˇCALMstocksandshortinthelowestˇCALMstocksearnsonaverage0.60%permonth
whichissignificantat5%level.Althoughvalue-weightedreturnsfurtherconfirmapositivebetastock returnrelationshipatperiodsoflowuncertaintyaboutaggregatevolatility,nextmonth’srisk-adjusted returns(CAPMandFF3-factoralphas)ofthespreadportfolioarerelativelylowerandinsignificant. Overall,betasimpliedbythethresholdlevelofRVIXcaptureuncertaintyaboutexpectedaggregate volatilityindicatingthatcross-sectionofexpectedreturnsaresystematicallyrelatedtothelevelof uncertaintyaboutaggregatevolatility.
4.3.2. Fama–MacBeth(1973)testsforriskpremia
Thereisnowaconsensusontimevariationinmarketrisk.TheconditionalCAPMisanattemptto capturethisvariation.However,Ghysels(1998)showsthattheconditionalCAPMisunabletospecify timevariationaccurately,leadingtohigherpricingerrorscomparedtotheunconditionalCAPM.In viewofthesefindings,webelievethatitiscrucialtounderstandthetruedynamicsoftimevariationin betariskandincorporatethisdynamicsinthepricingmodel.Ourpreviousfindingsestablishthatbeta riskexhibitssignificantchangestriggeredbyshiftsininvestors’expectationsregardingtheevolution ofnear-termaggregatevolatility.Hence,thenextnaturalstepwouldbetoanalyzewhetherthisrisk ispricedinthecross-section.
WeemploystandardFamaandMacBeth(1973)two-passregressionmethodology.Thefullmodel tobetestedis rji,t=˛jt+ˇjMKT,i,tjMKT,t+ K
k=1 ˇjk,i,tk k,t+εji,t, j=0,1,2 (7)where’srepresentunconditionalpricesofriskforvariousloadingsandcharacteristics,andj=0,1, and2representfullsample,lowandhighuncertaintyaboutexpectedvolatility,respectively.Inline withourpreviousresults,weusethemostcommonlyobservedRVIXthresholdleveltoidentifylow (high)uncertaintyperiods.Moreparticularly,weidentifymonthsinwhichthethresholdparameter RVIXisgreater(less)than9.33asuncertain(calm)periodsaboutexpectedaggregatevolatility.
Inthefirstpass,betaloadingsareestimatedateachmonthusingdailyobservations.Inthesecond pass,across-sectionalregressionisruneachmonth,withbetaloadingsobtainedfromthefirstpass