ABehavioralFinanceExplanationfortheSuccessofLowVolatilityPortfolios

MalcolmBaker HarvardBusinessSchoolandNBER BrendanBradley AcadianAssetManagement JeffreyWurgler NYUSternSchoolofBusinessandNBER December18,2009

Abstract
Arguablythemostremarkableanomalyinfinanceistheviolationoftheriskreturntradeoffwithinthe stock market: Over the past 40 years, high volatility and high beta stocks in U.S. markets have substantially underperformed low volatility and low beta stocks. We propose an explanation that combinestheaverageinvestor'spreferenceforrisk andthe typicalinstitutionalinvestor'smandate to maximize the ratio of excess returns to tracking error relative to a fixed benchmark (the information ratio)ratherthantheSharperatio.Modelsofdelegatedassetmanagementshowthatsuchmandates discourage arbitrage activity in both high alpha, low beta stocks and low alpha, high beta stocks. This explanation is consistent with several aspects of the low volatility anomaly including why it has only strengthenedevenasinstitutionalinvestorshavebecomemorenumerous.

Introduction Overthepast40years,lowvolatilityportfolioshaveofferedanenviablecombinationofhighaverage returnsandsmalldrawdowns.Weusetheprinciplesofbehavioralfinancetoexaminethesourceofthis anomalous performance and to consider the chance that it will persist. Behavioral models of security pricescombinetwoingredients:theirrationalityofsomemarketparticipantsandthelimitstoarbitrage whythesmartmoneydoesnotoffsetanyirrationaldemand.Specifically,webelievethatapreference forlotteriesandwellestablishedbiasesofrepresentativenessandoverconfidenceleadtoanirrational demandforhighervolatilitystocks.Moreover,mostinstitutionalinvestorsthatareinapositiontooffset thisirrationaldemandhavefixedbenchmarkmandates,whichbytheirnaturediscourageinvestmentsin lowvolatilitystocks.ApplyingthelogicofBrennans(1993)modelofagencyandassetprices,weshow thatfixedbenchmarkmandatescauseinstitutionalinvestorstopassupthesuperiorriskreturntradeoff oflowvolatilityportfolios.Forthatreason,thestrongperformanceofthesestrategiesislikelytopersist.

Figure1.Returnsbyvolatilityquintile.

100

Bottom Quintile

Value of $1 invested in 1968

10

1 1968 1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008

Top Quintile

0.1

Source:AcadiancalculationusingdatafromtheCenterforResearchonSecurityPrices(CRSP).Ineachmonth,wesortallpubliclytradedstocks trackedbyCRSPwithatleast24monthsofreturnhistoryintofiveequalquintilesaccordingtotrailingvolatility.Volatilityismeasuredasthe standarddeviationofupto60monthsoftrailingreturns.InJanuary1968,$1isinvested,accordingtocapitalizationweights,inthestocksof, forexample,thebottomquintile.Attheendofeachmonth,eachportfolioisrebalanced,withnotransactioncostsincluded.

BehavioralFinanceandLowVolatilityPortfolios In an efficient market, investors only realize above average returns by taking above average risks. In other words, risky stocks have high future returns on average, and safe stocks do not. This simple empiricalpropositionhasbeenhardtofindinthehistoryofUSstockreturns.Mostintuitivemeasuresof risk go in the wrong direction. For example, we take the last 40 years of data from the Center for Research on Security Prices (CRSP) and divide all stocks into five groups in each month according to trailingvolatility.Figure1showsthatadollarinvestedinthehighestriskportfolioinJanuary1968would

beworth$0.61attheendofDecember2008,assuming notransactioncosts.Adollarinvestedinthe lowest risk portfolio would be worth $56.38. The path to that higher dollar value is also considerably smoother.Addingtothepuzzle,muchofthegapcomesinrecentyearsafter1983overaperiodwhen institutional investment managers have become more numerous, better capitalized, and more quantitativelysophisticated. WhatExplainsThisDifferenceinPerformance? Onepossibilityisthatwehavethewrongmeasureofrisk.Maybevolatilityisthewrongidea.Individual securities are not typically held in isolation. So, the right measure of risk is not the level of risk for a particularstockbutitscontributiontotheriskofadiversifiedportfolioofstocks.Thisisthelogicbehind theCapitalAssetPricingModel(CAPM).Theriskofastock,oranyothersecurityforthatmatter,isits beta, which measures the contribution to risk in a broadly diversified market portfolio. This is a less intuitive, but theoretically more appealing notion of risk. Unfortunately, this refined measure of risk goesinthewrongdirection,too.Ifinsteadwedividestocksintofivegroupsineachmonthaccordingto trailing beta, the picture looks better. But, not by much. Figure 2 shows that a dollar invested in the highestrisk portfolioreturns$8.07,whileadollarinvestedinthelowestriskportfolioreturns$54.78. With the exception of the technology bubble in the late 1990s, more of the gap again appears after 1983. Thisisnotentirelynew.Inthe1970s,Black(1972)andBlack,Jensen,andScholes(1972)notedthatthe relationshipbetweenriskandreturnwasflatterthanpredictedbytheCAPM.FamaandFrench(1992) extendedthisanalysisoutto1992andfoundthattherelationshipwasflat,pronouncingthatbetais dead.Morerecently,Ang,Hodrick,Ying,andZhang(2006,2009)setoffanewwaveofresearch,finding thathighvolatilitystocks haveabysmallylowreturnsinboth USandinternationaldata.Ifanything, theevidenceforariskreturntradeoffalongthelinesoftheCAPMhasdeterioratedinthelast30years.

These patterns are hard to explain with traditional, rational theories of asset prices. In principle, beta might simply be the wrong measure of risk, too. The CAPM is just one equilibrium model of risk and return, with unrealistic assumptions. For the past few decades, finance academics have devoted considerable energy to developing rational models, searching for the right measure of risk. Most of these newer models make the mathematics of the CAPM look quaint. But, despite superior computational firepower, this is a steep uphill battle. After all, the task is to prove that high volatility stocks are much less risky. Granted a less risky stock might not be less volatile, but it must at least provideinsuranceagainstbadevents.Alowaveragereturnonhomeownersinsuranceisacceptable,to makeananalogy,becauseitprovidesaveryhighreturnattherightmoment.
Figure1.Returnsbybetaquintile.

100

Bottom Quintile

Value of $1 invested in 1968

10

Top Quintile

1 1968 1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008

0.1

Source:AcadiancalculationusingdatafromtheCenterforResearchonSecurityPrices(CRSP).Ineachmonth,wesortallpubliclytradedstocks tracked by CRSP with at least 24 months of return history into five equal quintiles according to trailing beta. Volatility is measured as the standarddeviationofupto60monthsoftrailingreturns.InJanuary1968,$1isinvested,accordingtocapitalizationweights,inthestocksof, forexample,thebottomquintile.Attheendofeachmonth,eachportfolioisrebalanced,withnotransactioncostsincluded.

A closer look at the pattern of returns reveals why this is such a hard argument to make for high volatilitystocks.Thetopquintileasagroupprovidedaverylowreturninpreciselythoseperiodswhen aninsurancepaymentwouldhavebeenmostwelcome. Underperformanceisvisuallyapparentinthe downturnsof19721974and20002002,thecrashof1987andthefinancialcrisisinthefallof2008.The ingenuity of financial economists applying axioms of rationality should never be underestimated. But, they seem unlikely to be able to explain Figure 1, where investors appear to be paying an insurance premiumeachmonthonlytoloseevenmorewhenthemarketandtheeconomyasawholearemelting down. ABehavioralExplanation If the answer is not risk, then what is going on? We think three things are behind the data here: less thanfullyrationalinvestorbehavior,thelimitstoarbitrage,andthesimplelawofcompounding. Westartwiththefirsttwoandcometothemoremundaneissueofcompoundinglater.Inaninefficient market, mispricing comes from the combination of two things. First, some investors are not fully rational. This assumption is not hard to swallow, with a little inspection or introspection. Call these investors noise traders. Second, there must be limits to arbitrage. That means that the smart money mustbelessthanfullycompetitiveintakingadvantageofthemispricingcreatedbythenoisetraders. Thisbasiclogicisthefoundationofbehavioralfinance,laidoutinsurveyslikeShleifer(2000),Barberis andThaler(2003),andBakerandWurgler(2007). Itiseasytosaythatnoisetradersandthelimitstoarbitragearetoblameforanyanomalyinthedata.It issomewhathardertoexplainthespecificmechanism.Whatistheunderlyingpsychologythatleadstoa preferenceforvolatilestocks?And,whydontsmartinstitutionsfillthegap,overweightinglowvolatility stocksandunderweightinghighvolatilitystocksbyjustenoughtooffsetanyirrationaldemand?

InvestorBiasesThatLeadtoaPreferenceforVolatileStocks We think that three biases are at work here: a preference for lotteries, representativeness, and overconfidence.EachhasitsrootsintheearlyworkoftheNobelprizewinningpsychologistsKahneman andTversky(KT).Forourpurposeshere,eachleadstoalessthanfullyrationalincreaseinthedemand forhighvolatilitystocks. Preferenceforlotteries.Thepreferenceforlotteriesisthesimplest.Wouldyoutakeagamblewhereyou lose $100 with 50% probability and win $110 with 50% probability? Most people say no. Despite the positiveexpectedpayoff,thepossibilityoflosing$100isenoughtodeterparticipation,evenwhen$100 is trivial in the context of wealth or income. KT called this loss aversion. Taken on its own, loss aversion suggests investors would shy away from volatility for fear of realizing a loss. But, something strangehappensastheprobabilitiesshift.Supposeinsteadthechoiceisbetweenacertaingainof$5or alargepayoffof$5,000withonlya0.1%probability.Mostpeopletakethegamble.Theamountspent onlotteriesandroulettewheelsisaclearmanifestationofthistendency. Tobestatisticallyprecise,thisismoreaboutpositiveskewness,wherelargepositivepayoffsaremore likelythanlargenegativeones,thanitisaboutvolatility.But,MittonandVorkink(2007)havepointed out that volatile individual stocks, with limited liability, also happen to be positively skewed. Buying a low priced, volatile stock is like a lottery ticket: There is some small chance of doubling or tripling in value,ormuchmore,withinamonth,andthereisamuchlargerprobabilitythatitwilldeclineinvalue. Representativeness. One way to explain representativeness is with a 1983 experiment. KT described a woman named Linda as single, outspoken, and very bright. She majored in philosophy. As a student, shewasdeeplyconcernedwithissuesofdiscriminationandsocialjustice,andalsoparticipatedinanti nucleardemonstrations.Thentheyasked:Whichismoreprobable?A:Lindaisabankteller.OrB:Linda isabanktellerwhoisactiveinthewomensmovement.ThefactthatmanysubjectschooseBsuggests

thatprobabilitytheoryandtheapplicationofBayesruleisnotaningrainedskill.IfLindaisabankteller whoisactiveinthewomensmovement(Bistrue),shemustalsobeabankteller(Aistrue).Suppose Lindais70%likelytobeactiveinthewomensmovement.SupposealsothatLindahasa1%chanceof being a bank teller. B, at 0.7% = 1% x 70%, is less probable than A. The mistake arises, KT theorized, becausethesecondjobfitsthedescriptionorismorerepresentativeofLinda.Ifwethinkofabank tellerwhoisactiveinthewomensmovement,apersonlikeLindacomestomindmorereadily. What does this have to do with stocks and volatility? A similar problem arises when defining the characteristics of great investments. The layman and the quant take two different approaches. The laymantriestothinkofgreatinvestmentsmaybebuyingMicrosoftorGenzymeattheirIPOsin1986 anddrawstheconclusionthatsmallandspeculativeinvestmentsinnewtechnologiesvolatilestocks are the path to investment success. Microsoft returned 70% per year in its first five years as a public company.TheproblemwiththislogicissimilartotheLindaquestion.Subjectswhoanswerincorrectly ignorethebaserateprobabilityofbeingabanktelleranditseffectontheprobabilityofB.Thelayman herelargelyignoresthelargenumberofsmallandspeculativeinvestmentsthatfailedandasaresult isinclinedtooverpayofvolatilestocks.ThequantexaminesthefullsampleofstockslikeMicrosoftand Genzyme in an analysis like Figure 1 and concludes that, without the knowledge to separate the Microsoftsfromthelosers,thisgroupofinvestmentshasfaredpoorly. Overconfidence. Another bias underlying the preference for high volatility stocks is overconfidence. Experimentersasksubjectstoestimate,forexample,thepopulationofMassachusetts,andtoprovidea 90%confidenceintervalaroundthisanswer.Theconfidenceintervalshouldbewideenoughsothat,for every10questionsofthistype,nineshouldcontaintherightanswer.Theexperimentalevidenceshows that most people form confidence intervals that are far too narrow. In other words, they are

overconfidentintheaccuracyoftheirknowledge.Moreover,themoreobscurethequestionifitisthe populationofBhutaninsteadofMassachusettsthemorethiscalibrationdeteriorates. Valuing stocks involves this same sort of forecasting. What will revenues be in 2014? Overconfident investors are likely to disagree. They will agree to disagree, sticking with the false precision of their estimates.Theextentofdisagreementishigherformoreuncertainoutcomes.Forexample,stocksthat areeithergrowingquicklyordistressedvolatilestockswillelicitawiderrangeofopinions. Oneextraassumptionisrequiredtoconnectoverconfidence,ormoregenerallyirreducibledifferences ofopinion,tothedemandforvolatilestocks.Weneedpessimisticinvestorstoactlessaggressivelythan optimistic ones. In other words, we must have a general reluctance to short sell stocks. This extra assumption means that prices are set by optimists. So, stocks with a wider range of opinions sell for higherprices,andhavelowerfuturereturns. TheLimitstoInstitutionalArbitrageinLowVolatilityStocks Acombinationofpreferencesforlotteriesandrepresentativenessandoverconfidencebiasesincreases thedemandforhighvolatilitystocksamonglessthanfullyrationalinvestors.Theremaininganddeeper puzzleiswhysophisticatedandwellcapitalizedinstitutionsdonotoffsetthisdemandandcapitalizeon thereturndifferencesinFigure1.Whatisespeciallypuzzlingisthatthispatternhas,ifanything,gained forceoveraperiodwhereinstitutionalmanagementintheU.S.hasdoubledfrom30%toover60%,as thedatafromtheFlowofFundsshowsinFigure3. Onepartofthepuzzleiswhyinstitutionalinvestorsdonotshorttheverypoorperformingtopvolatility quintile.Thishasasimpleanswer.Thesearegenerallysmallstocks(inthefullCRSPuniverse)andthey are costly to trade in large quantity and difficult to short, because the volume of shares available to

borrowislimited.Thesecondandmoreinterestingpartofthepuzzleiswhyinstitutionalinvestorsdo notoverweightthelowerriskandhigherperforminglowvolatilityquintile. We think the limits to arbitrage here come from the typical contract in delegated investment management. The implicit or explicit mandate for the vast majority of institutional managers is to maximize the information ratio relative to a specific and fixed benchmark. For example, if the benchmarkistheS&P500,theinformationratioistheaveragedifferencebetweenthereturnearnedby the investment manager and the return on the S&P 500,scaled by the volatility of the tracking error. Trackingerroristhestandarddeviationofthereturndifferences.
Figure3.InstitutionalOwnership,19682008.

70%

60%

50%

40%

30%

20%

10%

0% 5 4 9 1 7 4 9 1 9 4 9 1 1 5 9 1 3 5 9 1 5 5 9 1 7 5 9 1 9 5 9 1 1 6 9 1 3 6 9 1 5 6 9 1 7 6 9 1 9 6 9 1 1 7 9 1 3 7 9 1 5 7 9 1 7 7 9 1 9 7 9 1 1 8 9 1 3 8 9 1 5 8 9 1 7 8 9 1 9 8 9 1 1 9 9 1 3 9 9 1 5 9 9 1 7 9 9 1 9 9 9 1 1 0 0 2 3 0 0 2 5 0 0 2 7 0 0 2

Source: Institutional ownership from the Flow of Funds Table L.213. Assets managed by insurance companies (lines 12 and 13), public and privatepensionfunds(lines14,15,and16),openandclosedendmutualfunds(lines17and18),andbrokerdealers(line20).Assetsunder managementarescaledbythemarketvalueofdomesticcorporations(line21).

Thiscontracthasanumberofadvantages.Itisarguablyeasiertounderstandtheskillofaninvestment manager and the risks taken by examining the return relative to a benchmark. The ultimate investor cares more simply about risk, not tracking error. But, it comes at a cost. Roll (1992) analyzes the distortions that arise from a fixed benchmark mandate, and Brennan (1993) considers the effect on stock prices. In particular, a benchmark makes institutional investment managers less likely to exploit thepatternsinFigure1.Welaythisoutformallyinanappendixforthemathematicallyinclined.But,the logicissimple. In the SharpeLintner CAPM, investors with common beliefs aim to maximize the expected return on their portfolios and minimize volatility. This leads to a simple relationship between risk and return. A stocks expected return is equal to the riskfree rate of return plus its beta times the market risk premium: E(R)=Rf+xE(RmRf) (1)

Now,imaginethatthereissomeextra,andlessthanfullyrationaldemandforhighvolatilitystocksthat comes from a preference for lotteries and representativeness and overconfidence biases. This extra demand will push up the price of higher risk stocks and drive down their expected returns, and correspondinglypushdownthe priceoflowerriskstocksanddriveuptheirexpectedreturns.Willan institution with a fixed benchmark exploit these mispricings? The short answer, illustrated with an example,islikelyno. Example.Forsimplicity,assumeaninstitutionalmanagerwherethebenchmarkisthemarketportfolio, and suppose the expected return on the market is 5% over the riskfree rate and the volatility of the marketis20%.Takeastockwithaof0.8,andimaginethatitisundervalued,withanexpectedreturn greaterthantheCAPMbenchmarkinEquation(1)byanamount.Overweightingthestockbyasmall amount,say0.1%,willincreasetheexpectedactivereturnby0.1%x{0.25xE(RmRf)}=0.1%x{

2.5%}.Theextravarianceoftheportfolioisatleast{0.1%}2x{m2x(1 )2}={0.1%}2x0.0016,or the component of tracking error that comes from having a portfolio that is not equal to 1.0. This investmentmanagerwouldnotstartoverweightingthestockuntilitsexceeds2.5%peryear. Another way to see this is to consider the information ratio of the low volatility strategy in Figure 1. Table1showsthattheSharperatioishighat0.38.But,theinformationratio,ortheratiooftheexcess returnoverthefixedbenchmarktothetrackingerror,ismuchlessimpressiveat0.15.
Table1.ReturnsbyVolatilityQuintile,1968.012008.12.

Low PanelA:Sorts GeometricAverageRpRf AverageRpRf SD SharpeRatio AverageRpRm TrackingError InformationRatio t() PanelB:Sorts GeometricAverageRpRf AverageRpRf SD SharpeRatio AverageRpRm TrackingError InformationRatio t() 4.16% 4.85% 12.28% 0.39 0.84% 9.20% 0.09 0.63 2.32% 2.10 4.24% 5.03% 13.15% 0.38 1.02% 6.75% 0.15 0.75 2.02% 2.37

2 3.24% 4.16% 13.85% 0.30 0.15% 6.49% 0.02 0.79 0.97% 1.11 3.27% 4.66% 16.77% 0.28 0.65% 4.59% 0.14 1.01 0.60% 0.84

3 3.37% 4.64% 16.10% 0.29 0.63% 4.91% 0.13 0.96 0.78% 1.02 2.67% 4.99% 21.40% 0.23 0.98% 7.87% 0.12 1.28 0.14% 0.14

4 1.28% 3.35% 20.17% 0.17 0.66% 6.30% 0.11 1.22 1.55% 1.89 0.60% 4.32% 26.95% 0.16 0.31% 14.24% 0.02 1.53 1.81% 1.01

High 0.59% 3.18% 27.12% 0.12 0.83% 13.70% 0.06 1.58 3.14% 1.97 6.69% 1.64% 31.99% 0.05 5.65% 20.36% 0.28 1.70 8.44% 3.16

Source:WeformportfoliosbydividingtheCRSPuniverseintofiveequalsizedquintilesaccordingtotrailingbetainPanelAandtrailingvolatility inPanelB.Betasandstandarddeviationsaremeasuredusinguptofiveyearsofmonthlyreturns.ThereturnonthemarketRmandtheriskfree rateRf aretakenfromKenFrenchswebsite.Averagereturnsaremonthlyaveragesmultipliedby12.Standarddeviationandtrackingerrorare monthlystandarddeviationsmultipliedbythesquarerootof12.

To recap, an investment manager with a fixed benchmark is well suited to exploit mispricings that involve stocks with approximately market risk. A stock with a around 1 and a positive will be overweighted,andastockwithaaround1andanegativewillbeunderweighted.However,theywill dolittletocorrecttheundervaluationoflowstocksandtheovervaluationofhighstocks.Infact,not onlywilltraditionalassetmanagerswithafixedbenchmarkfailtocompensatefornoisetraderdemand for high stocks, they are part of the problem. By the same logic in the example, an investment manager will overweight a stock with a of 1.2, if its is zero or modestly negative. In a simple equilibriumdescribedintheappendixalongthelinesofBrennan(1993),withnoirrationalinvestorsin thesystem,thepresenceofdelegatedinvestmentmanagementwithafixedbenchmarkwillcausethe CAPMrelationshiptofail.Inparticular,itwillbetooflatasshowninFigure4: E(R)={Rf+c}+x{E(RmRf)c} (2)

The constant c depends on aggregate risk aversion, the tracking error mandate of the investment manager, and the fraction of asset management that is delegated. Requiring investment managers to targetaof1willhelp,butitwillnotsolvetheproblemoflimitedarbitrage.Suchamanagerwillstillbe reluctanttooverweightanunderpricedstockwithalessthan1.Asimplecheckofthislogicistolook attheformutualfundmanagers.Overthepasttenyears,mutualfundshaveaveraged1.10.

Figure4.DelegatedinvestmentmanagementwithafixedbenchmarkflattenstheCAPMrelationship.

Delegated Management
n r u t e R

Rm

CAPM Rf

0.0

0.5

1.0 Beta

1.5

2.0

2.5

PuttingthePiecesTogether The combination of irrational investor demand, stemming from a preference for lotteries and representativeness and overconfidence biases, and delegated investment management with fixed benchmarks causes the relationship between risk and return to breakdown. High risk stocks, whether measuredbyor,donotearnacommensuratereturn.And,lowriskstocksoutperform.Despitethis pattern, sophisticated investors are largely on the sideline because the mandate of maximizing return subjecttotrackingerrormakesarbitragingthesemispricingsunattractive.

Theimplicationisclear.Thereisasolidinvestmentthesisforlowvolatilitystrategies.Aslongasfixed benchmark contracts are the dominant form of implicit or explicit contracts between investors and investmentmanagementfirms,theanomalyinFigure1isunlikelytogoaway.
Table2.Riskandreturnacrossassetclasses.

SharpeRatio Average Return ShortTermGovernment IntermediateTerm LongTermGovernment CorporateBonds LargeCompanyStocks SmallStocks 5.70% 7.88% 8.90% 8.48% 9.86% 13.04% Excess Return 0.00% 2.18% 3.20% 2.77% 4.16% 7.34% SD 0.02% 5.58% 10.54% 9.58% 15.33% 21.79% Sharpe

CAPMPerformance t() t()

0.39 0.30 0.29 0.27 0.34

0.04 0.13 0.18 0.95 1.14

2.72 4.44 6.77 131.77 33.50

2.00% 2.67% 2.05% 0.26% 2.65%

2.31 1.65 1.43 0.65 1.41

Source:Wecomputetheaveragereturnandbetabyassetclass,usingdatafromIbbotsonAssociates.ThereturnonthemarketRmandtherisk freerateRfaretakenfromKenFrenchswebsite.Averagereturnsaremonthlyaveragesmultipliedby12.

Therearesomeadditional,moresubtlepredictions.Investmentmanagerswithfixedbenchmarkswillbe unlikely to exploit mispricings where stocks of different risks have similar returns within a particular mandate. But, risk and return are more likely to line up across mandates: for example, between intermediate and longterm bonds; between government and corporate bonds; between stocks and bonds; and, between large and small company stocks. The CAPM works in the right direction at the macrolevelinthedatafromIbbotsonAssociatespresentedbelowinFigure5,thoughthereturnslower riskinvestmentsingovernmentandinvestmentgradebondsarehigherthantheCAPMwouldpredict. CompoundingandTransactionCosts ThelastingredientinFigure1iscompounding.Theaveragemonthlyreturnsarecertainlydifferent.The top quintile has an average return of 41.9 basis points per month, while the bottom quintile has an average return of 13.7 basis points. But, the compound monthly differences are even larger at 35.3 basispointsand55.7basispoints.Theextra35basispointdifferenceincumulativereturnscomesfrom compoundingalowervolatilitymonthlyseries.

Whichistherightmeasureofperformance,thearithmeticmonthlyaverageorthecompoundreturnsin Figure1?Thereissomeambiguity.Byrebalancingeverymonthsaybackto50%incashand50%ina stockportfolioaninvestorcancompoundgrossreturnsatthearithmeticaveragerate,multipliedby 50%ofcourse.But,thisstrategyinvolvesconsiderabletransactioncosts.And,delayingrebalancingeven forafewmonthserodesthereturns. SearchingforLowerVolatility Giventhepowerofcompoundinglowvolatilityreturnsandtheoutperformanceoflowvolatilitystocks, anaturalquestioniswhetherwecandoevenbetterthanthelowvolatilityquintileinFigure1bytaking moreefficientadvantageofthebenefitsofdiversification.Leavingreturnsaside,wecandobetter,ifwe have a good estimate of not only individual firm volatility, but also the correlations among stocks. A portfoliooftwocorrelatedstockseachwithlowvolatilitycanbemorevolatilethanaportfoliooftwo uncorrelatedbutslightlymoreindividuallyvolatilestocks. With this in mind, we construct a minimum variance portfolio that takes advantage of forecast correlationaswellasvolatility.FollowingthemethodofClarke,deSilvaandThorley(2006),weuseonly thetop1000stocksintheCRSPuniversearealisticandimplementablestrategyandwecomparethe returnsonthelowvolatilityportfoliototheperformanceofthelowestquintile(200stocksnow,notthe bottomquintileofthewholeCRSPuniverseasinFigure1)sortedbyvolatility.Theminimumvariance portfoliohaslowervolatilityat11.5%,versus12.8%forthebottom200stocks.Thecompoundannual return is correspondingly higher at 10.7%, versus 10.1%. These patterns are visually apparent in the cumulativereturnsplottedinFigure5below.

Figure5.Alowvolatilityportfolioversusaportfoliooflowvolatilitystocks

100

80 Minimum Variance Portfolio 60

Value of $1 invested in 1968

40

Bottom Quintile

20

0 1968 1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008

Source:Thesolidlineisaminimumvarianceportfolioofthetop1000stocksbymarketcapitalizationintheCRSPuniverseusingsixmethods. The covariance matrix is estimated using the method of Clarke, de Silva and Thorley (2006). We limit the individual stock weights to fall between zero and three percent. The dashed line is the portfolio of the lowest quintile by trailing volatility. Volatility is measured as the standarddeviationofupto60monthsoftrailingreturns.

TheBestofBothWorlds Moststockmarketanomaliescanbethoughtofasdifferentreturns,similarrisksanomalies.Valueand momentum strategies, for example, are of this sort; the return differences are emphasized, not risk differences.Institutionalinvestmentmanagersarewellpositionedtotakeadvantageofsuchanomalies because they can generate high excess return while maintaining average risks, matching their benchmarksriskandcontrollingtrackingerror. Butthelowvolatilityanomalyisofaquitedifferentcharacter.Exploitingitinvolvesholdingstockswith moreorlesssimilarreturns,whichdoesnothelptheinvestmentmanagersexcessreturns,anddifferent

risks,whichonlyincreasestheirtrackingerror.So,eventhoughirrationalinvestorsarehappytooverpay forhighriskwhileshunninglowrisk,theinvestmentmanagerisgenerallynotincentivizedtoexploitit. And thus, the anomaly persists. Indeed, it has grown stronger over time, in step with the increase in institutionalinvestmentmanagement. In summary, the behavioral finance explanation for The Greatest Anomaly is that irrational investors prefer total risk while at the same time institutional investors are incentivized to manage risk against theirbenchmarks.Thetheoreticaldiagnosisalsoimpliesthepracticalprescription.Investorswhowant tomaximizereturnsubjecttototalriskmustincentivizetheirmanagerstodojustthat,byfocusingon thebenchmarkfreeSharperatio,nottheinformationratio. In recent years, more and more managers and their sophisticated clients have noticed the appealing riskreturn profiles of Low Volatility and Maximum Sharpe Ratio strategies. For them, our behavioral financeinsightsaregoodnews,becausetheysuggeststhataslongasmostoftheinvestingworldsticks withbenchmarks,theadvantageistheirs.

References Ang,Andrew,R.Hodrick,Y.Xing,X.Zhang."TheCrossSectionofVolatilityandExpectedReturns." JournalofFinance,61(2006),pp.259299. ."HighIdiosyncraticVolatilityandLowReturns:InternationalandFurtherU.S.Evidence."Journalof FinancialEconomics,91(2009),pp.123. Baker,Malcolm,J.Wurgler."InvestorSentimentandtheCrossSectionofStockReturns."Journalof Finance,61(2006),pp.16451680. ."InvestorSentimentintheStockMarket."JournalofEconomicPerspectives,Vol.21,No.2(2007),pp. 129152. Barberis,Nicholas,M.Huang."StocksasLotteries:TheImplicationsofProbabilityWeightingforSecurity Prices."AmericanEconomicReview98(2008),pp.20662100. Barberis,Nicholas,R.Thaler."ASurveyofBehavioralFinance."InG.Constantinides,M.Harris,R.Stulz, eds.,HandbookoftheEconomicsofFinance,NorthHolland. Black,Fischer."CapitalMarketEquilibriumwithRestrictedBorrowing."JournalofBusiness,Vol.45,No. 3(1972),pp.444455. Black,Fischer,M.C.Jensen,M.Scholes."TheCapitalAssetPricingModel:SomeEmpiricalTests."InM. C.Jensen,ed.,StudiesintheTheoryofCapitalMarkets.NewYork:Praeger,1972,pp.79121.

 Brennan,MichaelJ."AgencyandAssetPricing."Workingpaper,UniversityofCalifornia,LosAngeles, 1993. "MinimumVariancePortfoliosintheU.S.EquityClarke,RogerG.,H.deSilva,S.Thorley.Market."The JournalofPortfolioManagement,Vol.33,No.1(2006),pp.1024. Fama,EugeneF.,K.R.French."TheCrossSectionofExpectedStockReturns."JournalofFinance,47 (1992),pp.427465. Miller,EdwardM."Risk,Uncertainty,andDivergenceofOpinion."JournalofFinance,32(1977),pp. 11511168. Mitton,Todd,K.Vorkink."EquilibriumUnderdiversificationandthePreferenceforSkewness."Reviewof FinancialStudies,Vol.20,No.4(2007),pp.12551288. Roll,Richard."AMean/VarianceAnalysisofTrackingError."JournalofPortfolioManagement,Vol.18, No.4(1992),pp.1322. Shleifer,A.InefficientMarkets:AnIntroductiontoBehavioralFinance.Oxford,UK:OxfordUP,2000.

Appendix.DelegatedportfoliomanagementandtheCAPM This is a short derivation is similar in setup to Brennan (1993) and shows that delegated portfolio managementwithafixedbenchmarkwilltendtoflattentheCAPMrelationshipandmakelowvolatility and low beta stocks and low volatility portfolios an attractive investment strategy. To show this somewhatmoreformally,weneedtomakeafewassumptions. 1. Stocksandbonds.Therearestocksi=1toNthatpayapersharedividendofdipersharenext period.dhasmean1andcovariance.Eachhas1/Nsharesoutstanding.Forsimplicity,assume that=m2with1=N.Tomakeinvertible,weneede.g.m2 +2I.Thesecondterm dropsoutforlargeN.Thereisariskfreebondthatreturnsrf.Definedollarexpectedreturnwith someabuseofnotationasRiRf1pi(1+rf)anddollarmarketreturnRmRf{1/N}{1(1p(1+rf)}. 2. Investors. There are two representative investors j = 1, 2, who are meanvariance utility maximizers,eachwithcapitalkj. 3. Investmentstrategies.Eachinvestormakesascalarassetallocationdecisionajbetweenstocks and bonds, and a vector portfolio choice decision nj. Both investors are CARA, meanvariance utilitymaximizers(indollars)withariskaversionparameterof. a. Investor1delegateshisportfoliochoice.Investor1allocatesafractiona1ofhiscapital toanintermediarywhochoosesaportfolion1oninvestor1sbehalf. b. Investor2chooseshisownportfolio.Investor1allocatesafractiona2ofhiscapitalto stocksandhechoosesaportfolion2.Thiscanbecollapsedtoasinglechoicevariablen2, witha2=1.

4. Intermediation.Thereisasingleintermediarychoosesaportfolion1tomaximizeE{(n1nb)d} (n1nb)(n1nb),wherenb={1/N}1isthemarketportfolioand(n1nb)1=0. Ifthereareonlyinvestorsoftype2,thentheCAPMholds.Investor2choosesntomaximizeE{nd} nn } + (knp)(1+rf). The first order condition is n = {1/(2)} 1{1p(1+rf)} and the market clearing conditionisn={1/N}1.Puttingthesetwotogether,andrearrangingtermsleadstothefollowingCAPM relationshipbetweenriskandreturn:(3) R1Rf1p(1+rf)={1/N}{1(1p(1+rf)}(RmRf),or RiRf+(RmRf)i Ifthereareinvestorsoftype1andtype2,thingsaresomewhatmorecomplicated.Investor2chooses n2asbefore.Theintermediarychoosesn1tomaxE{(n1nb)d}(n1nb)(n1nb)s.t.(n1nb)1=0.The firstorderconditionisn1=nb+{1/(2)}1{1p(1+rf)1}=11{1p(1+rf)}/111.Investor1chooses a1tomaxa1{E{n1d}a12n1()n1}+(k1a1n1p)(1+rf).Thefirstorderconditionhereisa1={1/(2 n1n1)}{n1(1p(1+rf)}.Combiningthesethreechoicesn2,n1,anda1withthemarketclearingcondition ofa1n1+n2={1/N}1andrearrangingleadstothefollowingrelationshipbetweenriskandreturn: R1Rf{1p(1+rf)}={1/N}{1(1p(1+rf)}+(C/A)(1)(RmRf) whereA={1/(2)+a1/(2)};B=(1a1);C={a1/(2)},or Ri(Rf+c)+(RmRfc)I wherec=(C/A) The important thing to notice is that Equation (4) is a flattened version of Equation (3). Returns are increasingin,butnotquiteatarateequaltoRmRf.Thesecondtermispositiveforstockswithab lessthan1andnegativeforstocksthathaveagreaterthan1.Thatistosay,relativetotheCAPM,low stocksareundervaluedandhighstocksareovervalued. (4)

(3)