Whichofthesemodelsworksthebest?Isbetaagoodproxy
for risk and is it correlated with expected returns? The
answerstothesequestionshavebeendebatedwidelyinthe
pasttwodecades.ThefirsttestsoftheCAPMsuggestedthat
betas and returns were positively related, though other
measures of risk (such as variance) continued to explain
differencesinactualreturns.Thisdiscrepancywasattributed
tolimitationsinthetestingtechniques.Whiletheinitialtests
oftheAPMsuggestedthattheymightprovidemorepromise
intermsofexplainingdifferencesinreturns,adistinctionhas
to be drawn between the use of these models to explain
differencesinpast returnsandtheirusetopredictexpected
returnsinthefuture.ThecompetitorstotheCAPMclearlydo
amuchbetterjobatexplainingpastreturnssincetheydonot
constrainthemselvestoonefactor,astheCAPMdoes.This
extensiontomultiplefactorsdoesbecomemoreofaproblem
whenwetrytoprojectexpectedreturnsintothefuture,since
thebetasandpremiumsofeachofthesefactorsnowhaveto
be estimated. Because the factor premiums and betas are
themselves volatile, theestimationerror may eliminatethe
benefitsthatcouldbegainedbymovingfromtheCAPMto
more complex models.
Ultimately,thesurvivalofthecapitalassetpricingmodelas
the default model for risk in real-world applications is a
testamenttobothitsintuitiveappealandthefailureofmore
complexmodelstodeliversignificantimprovementinterms
of estimating expected returns. We would argue that a
judicious use of the CAPM, without an overreliance on
historicaldata,isstillthemosteffectivewayofdealingwith
risk in valuation.
Estimating Parameters for Risk and Return Models