Upload
adminali
View
218
Download
0
Embed Size (px)
Citation preview
8/10/2019 2014-033
1/44
Research DivisionFederal Reserve Bank of St. Louis
Working Paper Series
Can risk explain the profitability of technical trading in
currency markets?
Yuliya IvanovaChris Neely
David Rapach
and
Paul Weller
Working Paper 2014-033Ahttp://research.stlouisfed.org/wp/2014/2014-033.pdf
October 2014
FEDERAL RESERVE BANK OF ST. LOUISResearch Division
P.O. Box 442
St. Louis, MO 63166
______________________________________________________________________________________
The views expressed are those of the individual authors and do not necessarily reflect official positions of
the Federal Reserve Bank of St. Louis, the Federal Reserve System, or the Board of Governors.
Federal Reserve Bank of St. Louis Working Papers are preliminary materials circulated to stimulate
discussion and critical comment. References in publications to Federal Reserve Bank of St. Louis Working
Papers (other than an acknowledgment that the writer has had access to unpublished material) should be
cleared with the author or authors.
8/10/2019 2014-033
2/44
Can
risk
explain
the
profitability
of
technical
trading
in
currency
markets?
YuliyaIvanova*
ChrisNeely
DavidRapach
Paul
Weller
October30,2014
JELCodes: F31,G11,G12,G14
Keywords:Exchangerate;Technicalanalysis;Technicaltrading;Efficient
markets
hypothesis;
Risk;
Stochastic
Discount
Factor;
Adaptive
markets
hypothesis;Carrytrade;
*Ph.D. Candidate, Department of Finance, University of Iowa, Iowa city, IA 52245; e-mail: [email protected]; phone: +1-319-335-0926; fax: +1-319-335-3690. Assistant Vice President, Federal Reserve Bank of St. Louis, St. Louis, MO 63166; e-mail:[email protected]; phone: +1-314-444-8568; fax: +1-314-444-8731.
Professor of Economics,
Saint Louis University, St. Louis, MO 63108; e-mail: [email protected]; phone:
+1-314-977-3601; fax: +1-314-997-1478.John F. Murray Professor of Finance Emeritus, University of Iowa, Iowa city, IA 52245; e-mail: [email protected]; phone: +1-319-335-0948; fax: +1-319-335-3690.
Chris Neely is the corresponding author. We thank seminar participants at the Federal Reserve Bank of St.Louis for helpful comments. The usual disclaimer applies. The views expressed in this paper are those of
the authors and do not reflect those of the Federal Reserve Bank of St. Louis or the Federal Reserve
System.
8/10/2019 2014-033
3/44
AbstractItisarobustfindingthattechnicaltradingrulesappliedtoforeignexchangemarkets
haveearnedsubstantialexcessreturnsoverlongperiodsoftime.However,theapproachto
riskadjustmenthas typicallybeen rathercursory,andhas tended to focuson theCAPM.
Weexaminethereturnstoasetofdynamictradingrulesandlookattheexplanatorypower
ofawiderangeofmodels:CAPM,quadraticCAPM,CCAPM,Carharts4factormodel,an
extendedCCAPMwithdurableconsumption,LustigVerdelhan(LV)factors,volatilityand
skewness.Althoughskewnesshassomemodestexplanatorypowerfortheobservedexcess
returns, no model can plausibly account for the very strong evidence in favor of the
profitabilityof technical analysis in the foreign exchangemarket.We conclude that thesefindings strengthen the case for consideringmodels incorporating cognitivebias and the
processesoflearningandadaptation,asexemplifiedintheAdaptiveMarketsHypothesis.
8/10/2019 2014-033
4/44
1
1.
Introduction
It is a stylized fact that excess returns for currencyrelated trading strategies,
especially the carry trade, areweakly correlatedwith traditional risk factors, such as the
CAPMsequitymarketfactor.Tobettermeasureabnormalreturnsincurrencymarketsand
assess market efficiency, recent studies propose a variety of risk factors for carry trade
portfolios. These risk factors include consumption growth (Lustig and Verdelhan, 2007),
crash risk (Brunnermeier, Nagel, and Pedersen, 2008), a forward premium slope factor
(Lustig,Roussanov, andVerdelhan, 2011), economic size (Hassan, 2013),global exchange
rate volatility (Menkhoff, Sarno, Schmeling, and Schrimpf, 2012), and liquidity (Mancini,
Ranaldo,andWrampelmeyer,2013).
Theserecentlyproposedcurrencyriskfactorsusefullyexplainthereturnstoacross
sectionofcarrytradeportfolios.Nevertheless,theeconomiccaseforthesefactorswouldbe
morecompellingiftheycouldalsoexplainexcessreturnsforinvestmentstrategiesbeyond
thecarrytrade(Burnside,2011).Suchexplanatoryabilitywouldallaydataminingconcerns
andbetterestablishtheeconomicrelevanceofthenewlyproposedcurrencyriskfactors.In
this spirit, the present paper investigates the ability of recently proposed currency risk
factorstoaccountfortheexcessreturnsoftechnicaltradingstrategies.Technicaltradinghas
received less attention than the carry tradedespitebeing as successful,welldocumented
and mysterious. Technical analysis (or trend following) is very popular among currency
market participants (Menkhoff and Taylor, 2007), and a sizable literature indicates that
traditional risk factors fail to explain the profitability of a variety of technical strategies.
Fromthisstandpoint,theabilityofnewlyproposedcurrencyriskfactorstoaccountforthe
8/10/2019 2014-033
5/44
2
profitabilityof technicalstrategiesprovidesan informativetestofthefactorsrelevance.If
newriskfactorsfailtoadequatelyaccountforthebehavioroftechnicalstrategies,thenthese
factorsarelessappealingeconomicallyandthesearchformorerobustcurrencyriskfactors
shouldcontinue.
Weinvestigatetheabilityofabroadarrayofcurrencyriskfactorstoexplainexcess
returnsfor the technicalportfoliosdeveloped inNeelyandWeller (2013).Theseportfolios
arebased on a variety of popular technical indicators and provide a realistic picture of
returns for trendfollowing practitioners. We consider the following models: CAPM,
quadraticCAPM,CCAPM,Carharts4factormodel,anextendedCCAPMwithdurable
consumption, LustigVerdelhen (LV) factors, volatility and skewness. In a nutshell, our
resultsshowthatrecentlyproposedcurrencyriskfactorshavelittleexplanatorypowerfor
technicalportfolioreturns.Thenewriskfactorsidentifiedintherecentliteraturethusdonot
appearrelevantforanimportantclassofportfoliosinthecurrencyspace.Wehighlightthe
dimensionsalongwhich thenew risk factors fail toaccount for thebehaviorof technical
portfolios. The inadequacies of extant currency risk factors highlight the challenges in
explainingtechnicalportfolioreturns.
The restof thepaper isorganizedas follows.We firstdescribe theconstructionof
currencyportfolios.We thendescribe thedifferent currency risk factors thatwe consider
andeconometricmethodology.Ourempiricalresultsfollow.
8/10/2019 2014-033
6/44
3
2.
Trading
Rules
and
Data
2.1TradingRules
Thegoalofourpaperistoexaminewhetherrecentadvancesinriskadjustmentcan
explain the seemingly very strong performance of traditional technical trading rules in
foreign exchange markets. We would like our trading rules to represent those that the
academic literature has investigated but also to be chosen dynamically, in an ex ante
fashion, to mimic the backtesting habits of practitioners and also to exploit changing
patternsinadaptivemarkets.Inordertodoso,webasicallyfollowNeelyandWeller(2013)
who dynamically constructed portfolio strategies from an underlying pool of frequently
studied rules 7 filter rules, 3moving average rules, 3momentum rules, and 3 channel
ruleson19dollarand21crossexchangerates.5There isonenotabledifferencebetween
therulesusedinthispaperandthoseinNeelyandWeller(2013):toisolatethedeterminants
oftechnicaltradingrules,thepresentpaperdoesnotincludecarrytradereturnsamongthe
rules.
Menkhoffetal.(2012b)havepreviouslystudiedtheriskadjustmentofcertaintypes
ofcurrencymomentumstrategies. Onemightbeconcerned that the rulesstudiedhere
technicalruleswouldbeverysimilartothemomentumrulesstudiedbyMenkhoffetal.
(2012b). But Menkhoff et al. (2012b) empirically evaluate this concern and argue that
technicalrulesandmomentumrulesarequitedifferent.
Allofthebilateralrulesborrowinonecurrencyandlendintheother.Thustheyall
5Dooley and Shafer (1984) and Sweeney (1986) look at filter rules; Levich and Thomas (1993) look at filter
and moving average rules; Jegadeesh and Titman (1993) consider momentum rules in equities, citing Bernard
(1984) on the topic; and Taylor (1994) tests channel rules, for example.
8/10/2019 2014-033
7/44
4
produceexcessreturns.Wewillfirstdescribethetypesoftradingrulesbeforedetailingthe
dynamicrebalancingprocedureforcurrencytradingstrategies.
Afilterrulegeneratesabuysignalforaforeigncurrencywhentheexchangerate
(domesticpriceofforeigncurrency)hasrisenbymorethanypercentaboveitsmostrecent
low. It generates a sell signalwhen the exchange ratehas fallenbymore than the same
percentagefromitsmostrecenthigh.Thus,
(1)
where is an indicator variable that takes the value +1 for a long position in foreign
currencyand1 fora shortposition.nt is themost recent localminimumof andxt the
most recent localmaximum.The seven filter ruleshave filter sizes (y)of0.005,0.01,0.02,
0.03,0.04,0.05,and0.1.
Amovingaveragerulegeneratesabuysignalwhenashorthorizonmovingaverage
ofpastexchangeratescrossesalonghorizonmovingaveragefrombelow.Itgeneratesasell
signalwhen the shortmoving average crosses the longmoving average from above.We
denotetheserulesbyMA(S,L),whereSandLarethenumberofdaysintheshortandlong
moving averages, respectively. The moving average rules are MA(1, 5), MA(5, 20), and
MA(1, 200). Thus, MA(1, 5) compares the current exchange rate with its 5day moving
averageandrecordsabuy(sell)signaliftheexchangerateiscurrentlyabove(below)its5
daymovingaverage.
8/10/2019 2014-033
8/44
5
Ourmomentumrules takea long (short)position inanexchangeratewhen then
daycumulativereturnispositive(negative).Weconsiderwindowsof5,20and60daysfor
themomentumrules.
Achannelruletakesalong(short)positioniftheexchangerateexceeds(islessthan)
themaximum (minimum)over thepreviousndaysplus (minus) thebandof inaction (x).
Wesetntobe5,10,and20,andxtobe0.001forallrules.Thus,
(2)
We apply these 16 bilateral rules 7 filter rules, 3 moving average rules, 3
momentumrules,and3channelrulesto19dollarand21crossexchangerates, listed in
Table1.TheseriesfortheDEMwassplicedwiththatfortheEURafterJanuary1,1999.For
simplicitywerefertothisseriesthroughoutastheEUR.Notallexchangeratesaretradable
throughout the sample. Table 1 details the dates on which we permit trading in each
exchangerate.
Inanystudyoftradingperformanceespeciallywhenusingexoticcurrenciesone
mustpay closeattention to transaction costs.Rulesand strategies thatmay appear tobe
profitablewhensuchcostsareignoredturnoutnottobeoncetheappropriateadjustments
have been made. We follow the methods in Neely and Weller (2013) and calculate
transactionscostsusinghistoricalestimatesforsuchcostsinthedistantpastandfractionsof
Bloomberg spreads after those were available. Neely and Weller (2013) detail these
calculations.
otherwise
1,...,minif
1,...,maxif
1
1
21
21
1
xSSSS
xSSSS
z
zntttt
ntttt
t
t
8/10/2019 2014-033
9/44
6
2.2DynamicTradingStrategies
We would like to construct dynamic strategies to mimic the actions of foreign
exchange traders who backtest potential rules on historical data to determine trading
strategies. Accurately modeling potential trading returns provides the most realistic
environmentforassessingwhetherriskadjustmentexplainssuchreturns.
Therefore,weconstructdynamictradingstrategiesasfollows:
1. Ateachpoint in the sample,weapply the16bilateral rules toallavailable
exchangerates,calculatingthehistoricalreturnstatisticsforeachexchangeraterulepairat
eachpoint.There isamaximumof(16*40=)640exchangeraterulesonanygivenday,but
missingdataforsomeexchangeratesoftenleavefewerthanhalfthatnumberofcurrency
rulepairs.
2. Starting 500 days into the sample, we evaluate the Sharpe ratios of all
exchangeraterulepairswithatleast250daysofdatasincethebeginningoftherespective
samples.WethensorttheraterulepairsbytheirexpostSharperatios,rankingtheraterule
pairsbySharperatiofrom1to640.Wethenmeasuretheperformanceofthestrategiesover
thenext20days.
3. Every 20business days, we evaluate, sort and rank all available raterule
pairsusingthecompletesampleofdataavailabletothatpoint.Thus,thereturnsonthetop
ranked strategy pair willbe generatedby a given trading rule applied to a particular
exchangerateforaminimumof20days,atwhichpointitmay(ormaynot)bereplacedby
anotherruleappliedtothesameoradifferentcurrency.
Werefertotheseriesofsortedpairsasdynamicstrategies.Thatis,theseriesofthe
8/10/2019 2014-033
10/44
7
bestexante ruleexchangeratepairs isknownasstrategy1, thesecondbestexante rule
exchangeratepairsisknownasstrategy2,etc.
Although we select the raterule pairs for the dynamic strategies based upon
historicalperformance,asdescribedabove,weevaluatethestrategiesperformanceoverthe
periodsaftertheyareselected. Thatis,allreturnperformancestatisticsinthispaperarefor
strategiesthatwerechosenexanteandarethusimplementable.
2.3Currencyportfolios
As is customary in related literature,we examine the riskadjustment of technical
tradingrulesinthefollowingway:Usingstrategies1to300touseastestassets,weform12
equallyweightedportfoliosof25strategiesperportfolio.Thusportfoliop1attimetconsists
ofthe25currencyrulepairswithSharperatiosranked1to25.Portfoliop2consistsofthe25
currencyrule pairs with Sharpe ratios ranked 26 to 50, and so on. The makeup of the
portfolios of currencyrule pairs may change from period to period with ex post Sharpe
ratiorankings.6
Iftechnicalanalysishaspredictivevalue,onewouldexpectthehigherranked(lower
numbered) ex ante strategies to also do better ex post than the lower ranked (higher
number) strategies. Figure 1 confirms a positive relation between ex ante and ex post
performance over the sample. The figure shows the spread in excess return, standard
deviationandSharperatiosoverthe12portfolios.Thetoppanelshowsthatall12portfolios
6Alternatively, one could examine riskadjusted returns to a portfolio of the top N strategies. Althoughwe omit the full results for brevity, we also riskadjust the return to four such portfolios, using equal andex ante, meanvariance optimal weights and N = 10, 50. We will note results from these exercises whereappropriate.
8/10/2019 2014-033
11/44
8
havepositiveexcessreturns,generallydecliningasonegoesfromp1(4.61%perannum)to
p12(1.08%perannum). Themiddlepanelshowsthatthehighrankedportfoliosalsotend
tohavemorevolatile returns, though the relation isnotassteepas for returns.The third
panelconfirmsthis:expostSharperatiosarehigherfor theportfolioswithhigherexante
rankings,rangingfrom0.92forp1to0.29forp12.
3. Theoreticalframeworkforriskadjustment
3.1GeneralFramework
Toprovideageneralframeworkwithinwhichtomeasureriskexposureweneedto
characterize equilibrium in the foreign exchange market. We assume the existence of a
representative investorbased in theUS,and introducea stochasticdiscount factor (SDF),
that prices payoffs in dollars.7It represents a marginal rate of substitutionbetween
presentandfutureconsumptionindifferentstatesoftheworld.Thefirstorderconditions
forutilitymaximizationsubjecttoan intertemporalbudgetconstraintimplythatanyasset
returnmustsatisfy
|=1 (3)
wheredenotes the informationavailable to the investor at time t.Varyingassumptions
about the content ofproduce thedifferent categories ofmarket efficiency advancedby
Fama. Since we are modeling the risk exposure of technical trading rules we will be
primarilyconcernedwithweakform
efficiency inwhich the information setcontainsonly
pastprices.
7 Although we motivate the SDF framework with a representative investor, much weaker assumptions are
sufficient. In particular, absence of arbitrage implies the existence of a SDF framework, as in equation (3).
8/10/2019 2014-033
12/44
9
Equation(3)impliesthattheriskfreeassetreturnisgivenby
= | (4)
Using(3),(4)andthedefinitionofcovariance,itfollowsthat
|= ,|| . (5)
That is, the excess return to any asset or trading strategy will be proportional to the
covarianceoftheassetreturnwiththeSDF.
The implication for technical trading strategies that take positions in foreign
currencies based on past prices is then straightforward. Positive excess returns are
consistentwithmarketefficiencyonlyiftheexcessreturnalsocovariesnegativelywiththe
SDF. This then raises the question of how to model the SDF and how to test whether
equation(5)explainsreturns.
Therearepotentiallyseveralways inwhichonecould test theextent towhich the
SDFframeworkcanexplainexcessreturnstothetradingrules. Themostdirectwouldbeto
model theSDF,, in (3)witha specificutility functionandcalibratedparametersand
test whether the errors from (3) are mean zero. Alternatively, one could estimate the
parameters of with some nonlinear optimization method, such as the generalized
method of moments (GMM), and test the overidentifying restrictions. Finally, one could
linearize theSDF,,with aTaylor series expansion, estimate a linear time seriesor a
returnbetamodelandevaluatewhether therisk factorsexplain theexpected returns.The
nextsubsectionsdetailthosetestingprocedures.
8/10/2019 2014-033
13/44
10
3.2TestingaCalibratedSDF
Initially,wewillfollowLustigandVerdelhan(2007)inusinganextendedversionof
the CCAPM to adjust returns for risk. They use the model of Yogo (2006) in which a
representative agent has EpsteinZin preferences over durable consumption and
nondurableconsumption.Utilityisgivenby
1 ,
(6)
whereisthetimediscountfactor,isameasureofriskaversion,istheelasticity
ofintertemporalsubstitutioninconsumption,and 1 1 1 .Theoneperiodutilityfunctionisgivenby
, 1 (7)
where is the weight on durable consumption and is the elasticity of substitution
between durable and nondurable consumption. Yogo (2006) shows that the stochastic
discountfactortakestheform
, (8)
Where 1
and,isthereturnonthemarketportfolio.
Thismodel, theEpsteinZindurable consumptionCAPM (EZDCAPM)nests two
othermodelsofinterest,thedurableconsumptionCAPM(DCAPM)andtheCCAPM.The
DCAPMholds ifwe impose the restriction 1 .TheCCAPMholds if inadditionwe
impose .Thestochasticdiscountfactorin(8)satisfiesthefamiliarEulerequationin(3).
8/10/2019 2014-033
14/44
11
Toinitiallyassesstheseconsumptionbasedmodelsofrisk,wecarryoutacalibration
exercise similar to that in Lustig and Verdelhan (2007). We choose parameter values
identified in Yogo (2006): 0.023, 0.802, 0.700. Then we use sample data on
durableandnondurableconsumptionand thereturnon themarketportfolio togenerate
pricing errors = 0, whereR is the excess return to portfolio pi, and i
1, ,12.8The coefficient of relative risk aversion is chosen to minimize the sum of
squaredpricingerrorsintheEZDCAPM.Table2presentstheseresults.Allmodelsclearly
perform very poorly; theis negative in every case.9The maximum Sharpe ratios and
priceofriskdiffersubstantiallyfromthoseinTable4ofLustigandVerdelhen(2007).Since
their testassetsare currencyportfolios sortedaccording to interestdifferentialwewould
expect thesenumbers tobesimilar.Theportfolioswith thehighest returnshavenegative
betas(p1hasabetaof 1.97).Thisimpliesthattheportfolioreturncovariespositivelywith
M.SinceMishighinbadtimeswhenmarginalutilityishighandconsumptionislow,these
portfoliosactasconsumptionhedgesandwouldbeexpectedtoearnrelativelylowreturns.
3.3LinearFactorModels
ResearcherstypicallylinearizetheSDFwithafirstorderTaylorapproximationand
thenassessthemodelsfitofthedatawiththatlinearsystem. Althoughitisnotclearhow
well the linearmodel approximates the SDF, thisproceduremakes estimation somewhat
easierandisconsistentwiththeliterature.
Therefore,weconsidertheclassoflinearSDFsthattaketheform
8Recall that portfolios p1 to p12 each consist of 25 currency-rule pairs, ranked every 20 days by ex-ante Sharpe
ratio. P1 contains strategies 1 to 25; p2 contains strategies 26 to 50 and so on.9The R2s are negative because the models predictions covary negatively with the mean excess returns of the
test assets.
8/10/2019 2014-033
15/44
12
(9)
whereis a scalar, is a 1vector of parameters andis a 1vector of demeaned
factors that explain assetprice returns.The constantin (9) isnot identifiedandwe can
normalizeittounity.10WehypothesizethattheSDFpricesportfoliosofexcesstradingrule
returns,,inwhichcaseequation(3)impliesthat
| 0 (10)
Theunconditionalversionof(10)is
1 0 (11)
fromwhichitfollowsthat
(12)
whereis the factor covariancematrix, is a 1vectorof coefficients in a
regressionofonand isa 1vectoroffactorriskpremia.
The model (12) is a returnbeta representation. It implies that an assets expected
return isproportional to itscovariancewith therisk factors. Thebetasaredefinedas the
timeseriesregressioncoefficients
1, ,, 1, , (13)
whereis the nondemeaned factor at time t. In the special case that the factors,, are
excessreturns,thentheintercepts()inthetimeseriesrepresentation(13)arezero.Wecan
seethisbyfirstnotingthattheexpectationofthefactormustsatisfy(12)becausewehave
assumedthatthefactorisalsoareturn.
10For any pair, {a,b}, such that 0, any {ca, cb} where cis a real constant would also satisfythe equation because 0. Therefore, only the ratio b/amatters and one can normalize ato 1or to any other constant.
8/10/2019 2014-033
16/44
13
(14)
wherethesecondequalityfollowsbecausemustequalonebecausethefactorcovaries
perfectlywithitself.Second,wetakeexpectationsin(13)andsolvefortheconstant
0 (15)
Thesecondequalityin(15)uses(12)and(14).
Forthesecases inwhichthefactor isitselfanexcessreturne.g.,theCAPMone
cantestthemodelbyregressingasetofNexcessreturnstotestassetsonthefactor,asin
(13),andthendirectlytestingwhethertheconstants()arezero. Iftheconstantsarenon
zero,thenthemodelfailstoexplaintheexcessreturntothetestassets.
For testsofmoregeneralsetsof factors,FamaMacBeth (1973)suggesta twostage
procedurethatfirstestimatesthesforeachtestassetwiththetimeseriesregression(13).11
Thesecondstagethenestimatesthefactorprices fromacrosssectionalregressionofmean
excessreturnsfromthetestassetsonthebetas.
, (16)
where is the regression coefficient tobe estimated ands are the pricing errors. The
model impliesnoconstant in (16)butone isoften includedwiththereasoning that itwill
pickupestimationerrorintherisklessrate.Alargevalueof,orasignificantchangeinthe
fitofthemodelwithaconstantindicatesapoorfit(Burnside(2011)).Foragivensetoftest
assets,thevariationofthebetasin(14)determinestheprecisionoftheestimatedfactorrisk
11Fama-MacBeth (1973) originally used rolling regressions to estimate the s and cross-sectional regressions at
each point in time to estimate a and for each time period, then using averages of those estimates to getoverall estimates. The time series of and estimates could then be used to estimate standard errors for theoverall estimates that correct for cross-sectional correlation.
8/10/2019 2014-033
17/44
14
premia, . If thebetas do not vary sufficiently, then is not identified and the test is
inconclusive.
ThestandarderrorsinthesecondstageoftheoriginalFamaMacBethproceduredo
not account for the fact that the regressors () are generated regressors. Shanken (1992)
suggestsacorrectiontoaccountforthisissue.OnecanuseGMMtosimultaneouslyestimate
both (13) and (16), obtaining the identical point estimates as the 2stage procedurebut
properlyaccounting for crosssectional correlation,heteroskedasticityand theuncertainty
aboutin the covariancematrix of theparameters (seeCochrane, 2005 chapter 10).The
momentrestrictionsaregivenby
, 0, 0
, 0 for 1,2,and 1,2, (17)
Inourempiricalexercises,weestimatethebetareturnmodelswithGMMandpresentthree
setsofstandarderrorsOLS,ShankenandGMMintheinterestofcomparison.
4. Results
4.1CAPMmodelsofthereturnstop1throughp12
Figure1showedthattheexpostSharperatiosofthetechnicalstrategiesvariedwith
their ex ante rank. That is, past returns tend to predict future returns. Does the risk
adjustmentaffecttheexpectedcrosssectionofreturns?
AsabenchmarkwefirstlookatwhethertheCAPMcanexplaintheexcessreturnsto
the12portfolios,P1P12,whichconsistofstrategies125,2650,276300,respectively.The
model has a single factor, the market excess return, and so we consider the following
regressionequationforeachportfolio:
8/10/2019 2014-033
18/44
15
, , (18)
whereistheexcessreturntothedynamicportfoliostrategyandisthemarketexcess
return.Themodelrequiresthattheintercept,alpha,notbesignificantlydifferentfromzero.
Apositivealphawouldimplythatthemodelfailstoexplainthereturn.
PanelAofTable3 shows the results for regression (13) for the12portfolios. The
riskadjustmentleavesthemeanreturn(alphas)essentiallyunchangedforall12portfolios.
PortfolioP1,forexample,hasahighlysignificantmonthlyalphaof0.39,or4.68percentper
annum.Thehighlysignificantalphas indicate that themarket factorcannotexplain the
excessreturnstothetechnicalportfoliosandthenegativebetasindicatethatthereturnsare
notevenpositivelycorrelatedwiththemarketreturns.WeconcludethattheCAPMfailsto
explaintheexcessreturnsofthedynamicportfoliostrategies.
WeturntothequadraticCAPMinTable3.
, , , (19)
Because the squared excess return is not the return to a tradable portfolio, we cannot
interpret theconstant in the timeseriesregressionasariskadjustedreturn.Here too, the
coefficients on the market (are all negative, which would tend to indicate that the
marketriskdidnotexplaintheforexportfolioreturns,butthequadratictermsusuallyhave
significantlypositivecoefficients( thatsuggestthatequitymarketvolatilityinfluences
dynamic technical portfolio returns. Menkhoff et al. (2012a) document a similar
phenomenon forcarry tradereturns.Further, therightmostcolumnsshow that these
coefficientsjointly differ from zero and from each other, which potentially allows us to
identifythepriceofriskforthequadraticCAPMfactor.
8/10/2019 2014-033
19/44
16
TheQuadCAPM columnsofPanelB show thataddingaconstant to thecross
sectional regression raises the R2 from 0.11 to 0.18 but the constant is not statistically
significantwithGMMstandarderrors.Thepriceofriskforthequadraticmarketreturnis
0.31 and statistically significant. These statistics indicate that the quadratic CAPM might
partiallyexplainsomeoftheexcessreturnsofthedynamicportfoliostrategies.
WenextexamineCarharts(1997)fourfactorextensionofthethreefactormodelof
FamaandFrench(1993)wheretheriskfactorsaretheexcessreturnontheU.S.stockmarket
(),thesizefactor(),thevaluefactor()andthemomentumfactor().12
, , , , (20)
Panel A of Table 3 shows that the alphas for the top portfolios are positive and
highlysignificantandthecoefficientsontheregressorsaregenerallynegative(,and
)orinsignificant().TherightmostcolumnsofPanelAshowthatonecannotreject
the nulls of no variation in any of the fourbeta vectors. This indicates that one cannot
conclusively identify the price of risk for these factors. Perhapsbecause of this lack of
identification,thepricesoffactorriskestimatedbycrosssectionalregressionforbothSMB
andHMLarenegative,whichisinconsistentwithestimatesderivedfromthesamplemean.
Asnotedabove,whenthefactorsaretradableexcessreturnsthenfactormeansareequalto
pricesofrisk.The factormeans inoursampleare0.58% (),0.24% (),0.29% ()
and 0.68% ( ). There is no evidence that the fourfactor model explains the excess
12 Fama and French (1993) showed that 3-factors, market return, firm size and book-to-market ratios very
effectively explained the returns of certain test assets. The factors used in equation (15) are returns to zero-
investment portfolios that are simultaneously long/short in stocks that are in the highest/lowest quantiles of thesorted distribution. For example, the small-minus-big (SMB) portfolio takes a long position in small firms and a
short position in large firms.
8/10/2019 2014-033
20/44
17
returnstothedynamicportfoliostrategies.
4.2Consumptionbasedmodelsofthereturnstop1throughp12
Wenowexaminewhetherconsumptionbasedmodelsofassetpricingexplainthe
returnstothe12portfoliosofdynamicstrategies.TheCCAPMrelatesassetreturnstothe
realconsumptiongrowthofariskaverserepresentativeagent,asinequation(8).
Wefirstconsiderthreevariationsofthelinearapproximationofthefactormodelin
(8),theCCAPM,DCAPMandEZDCAPM(Yogo(2006)).
, , , , (21)
whereandare lognondurableanddurableconsumptiongrowth respectively,and
,isthelogreturnonthemarketportfolio.Thelinearapproximationforthemostgeneral
of thesenestedmodels,theEZDCAPM,usesnondurablesplusservices,durablesandthe
marketexcessreturnasfactors.Onecanestimatethefactorprices,,andportfoliobetas,i
withthebetarepresentationgivenbyequations(13)and(16).
, , , ,, , (22)
, , , , (23)
where , , , , ,is the factor covariance matrix and is
thecovariancematrixbetweenthetestreturnsandthefactors.TheDCAPMandCCAPM
restrict, andthepair,{, ,,}toequalzero,respectively.13
Table4,PanelApresentstheresultsof the timesseriesregressions.Allmodels,C
CAPM,DCAPMandEZDCAPMperformpoorly.FortheCCAPMthebeta forportfolio
p1issignificantatthe10percentlevelbuthasthewrong(negative)sign.Mostofthebetas
13One could infer the utility function parameter values from the estimates of the coefficients on the factors in
the linear model, as in Lustig and Verdelhan (2007), see equation (4) in that paper.
8/10/2019 2014-033
21/44
18
for the DCAPM and EZDCAPM are either negative or insignificant. The rightmost
columnsofPanelA show that thebetavectors allvary significantly from zeroand from
eachother,permittingidentificationofthepricesofrisk.
PanelBshowstheresultsofthecrosssectionalregressionsforeachmodelestimated
withandwithoutaconstant.IncludingconstantsappearstochangethefitoftheCCAPM
andDCAPMmodels:Theconstantsarenotsignificantattheonesided10percentleveland
change the R2s from negative to positive levels. The negative R2s are damning: They
indicatethatasimpleconstantwouldexplaintheexpectedreturnsbetterthanthemodel.
TheEZDCAPMmodelappearstofitbetterthantherestrictedmodels.Theconstant
isnotsignificantandtheR2issizable,at0.61anddoesnotchangemuchwiththeadditionof
the constant.Theprice of risk fornondurable consumption in theEZDCAPMmodel is
statisticallysignificantbutnegative,2.24percent.Thisnegativepriceofriskisimplausible
becausetheorypredictsthatitshouldbepositivewhenthecoefficientofriskaversion 1,
andtheelasticityofsubstitutioninconsumption 1.Yogo(2007)estimates tobe0.210
fortheEZDCAPMwithequityassets.
We find in all cases that the coefficient of risk aversion is estimated to be
significantlynegative.The reason for thiscanbe seenclearly in thecaseof theCCAPM.
Equation (5) implies that , ; the Fundamental Theorem of Asset Pricing
requirestobepositiveandsoitfollowsthatifanassethasapositiveexpectedexcess
return itmust covarynegativelywith theSDF.TheSDF (M), in turn, ismarginalutility,
which will covary negatively with consumption. In the case of the CCAPM where the
singlefactorisconsumptiongrowth,thismeansthatthemodelpredictsthatreturnstothe
8/10/2019 2014-033
22/44
19
higher ranking dynamic portfolio strategies should be positively correlated with
consumptiongrowth. In fact,we find thereverse: thedynamicportfoliostrategies tend to
performwellinstateswhenconsumptiongrowthislow,andthusprovideahedgeagainst
consumption risk. Tobe consistent with the model such strategies should earn negative
excessreturns.
Figure2displaystheactualmeanexcessreturntoeachportfolio()versusthe
predicted return ( for the crosssectional regressions that include a constant.
Intriguingly, there does appear tobe a positive relationshipbetween the predicted and
actualreturnforallthreemodels,CCAPM,DCAPMandEZDCAPM. Recall,however,that
the prices of risk are implausibly signed and that the crosssection regression should
excludetheconstant.
If one excludes the constant from the regression, the actual returns exceed the
predicted return for all but one return for the CCAPM and DCAPM cases (Figure 3).
Although the lower leftpanelofFigure3appears to illustrateanicepositive relationship
betweenEZDCAPMpredictionsandactual returns, this relation ismisleading: It ignores
thenegativeR2sandnegativepricesofnondurableconsumptionrisk.
4.3Foreignexchangebasedmodelsofthetechnicalreturns
Consumptionbasedmodelshavegenerallyfailedtoexplainriskadjustedreturnsin
severalcontextssotheresultsinTable4comeasnosurprise.Thisfailureofconsumption
basedmodelshas ledresearchersto lookforotherriskfactorsthatmightproxyforfuture
investment opportunities. In the context of stock returns it hasbecome commonplace to
8/10/2019 2014-033
23/44
20
work with factors that are the returns to various stock portfolios (see Fama and French
(1993)andCarhart(1997)).
Lustig,RoussanovandVerdelhan (2011)have recentlyapplied thisgeneral idea to
the foreignexchangemarket.They formcurrencyportfolioson thebasisof interest rates.
Portfolio 1 contains those currencies with the lowest interest rates, portfolio 6 those
currencieswith thehighest interest rates.Theportfoliosare rebalancedat theendofeach
month.FromtheseportfoliosLustigetal.(2011)constructtworiskfactors.Thefirstfactor,
whichtheydenoteRX,istheaveragecurrencyexcessreturntogoingshortinthedollarand
longinthebasketofsixforeigncurrencyportfolios.Thesecondfactor,HMLFX,isthereturn
to a strategy thatborrows low interest rate currencies (portfolio 1) and invests in high
interestratecurrencies(portfolio6),inotherwordsacarrytrade.14
Canthesefactorsexplainthecrosssectionalvariationinexpectedreturnsacrossthe
12technicalportfoliosthatweresortedonpastSharperatio?Weexaminethisquestionwith
theGMMestimationofthebetareturnmodelsusingtheRXfactorandtheHMLFX factoras
riskfactors.Thisprovidesuswithestimatesofthefactorriskpremia(and)and
the parameters of the model. Because the factors are tradable we can test the modelby
comparing theestimatesof the riskpremiawith the factormeans.We reject themodel if
theydiffersignificantly.
Panel A of Table 5 shows that thebetas on the RX factor are small and always
insignificant.However,foreightofthedynamicportfoliosthebetasontheHMLFXfactorare
significantlynegativeatthetenpercentlevelandallofthepointestimatesarenegative.The
14RX and HMLFXare very similar to the first two principal components of the returns to the 6 portfolios.
8/10/2019 2014-033
24/44
21
signsofthebetassuggestthatthereturnstothetechnicaltradingrulestendtobehighwhen
thecarry tradereturnsare low.TheHMLFXbetas forportfoliosp1,p3,p4,andp5arenot
significant, however, and the alphas for the top five portfolios are often significant and
always higher than the unadjusted mean returns. The fact that the constants (i.e., risk
adjusted returns) are always higher than the unadjusted mean returns indicates that
accounting for HMLFX and RX risk actually deepens the puzzle of the profitability of
technical trading rules. These results make it seem unlikely that any risk factor that
explainedthecarrytradecouldalsoexplainthetechnicalreturns.
Tests from the time series regression in Panel A, far right columns, indicate that
neithersetofbetasvariessufficientlytoidentifythepricesofrisk. Nevertheless,wepresent
resultsforthecrosssectionalregressioninPanelBofTable5. Thoseresultsshowthatthe
priceofriskfortheHMLFX factoris2.08andishighlysignificant,andtheR2is0.35(Table5,
PanelB).Buttheconstantintheregressionis0.22andisalsohighlysignificant.Themeanof
theHMLFXis0.69%.Whentheconstantisexcluded,theR2becomesnegative,however.We
concludethattheRXandHMLFXfactorscannotexplainthereturnstothedynamictechnical
portfolios.
Menkhoff,Sarno,SchmelingandSchrimpf(2012a)findthatglobalforeignexchange
volatilityisanimportantfactorinexplainingcarrytradereturns.Toinvestigatethisfactors
explanatory power for our technical returns, we estimate a global volatility factor in a
mannerverysimilarto thatofMenkhoff,Sarno,SchmelingandSchrimpf(2012a).Wefirst
calculatethemonthlyreturnvarianceforeachoftheavailableexchangeratesateachmonth
inour sampleand then calculateaglobal foreign exchangevolatility factor from the first
8/10/2019 2014-033
25/44
22
principalcomponentofthemonthlyvariances.Vol1 istheseriesof innovationsofanAR1
process fit to thisprincipal componentwhileVol2 is the firstdifference in thisprincipal
component. We then estimate abeta representation using these volatility factors and a
dollarexposurefactorthattheynoteisverysimilartotheLustig,RoussanovandVerdelhan
(2011)RX factor.Menkhoff, Sarno, Schmeling and Schrimpf (2012a)denote this theDOL
factorintheirwork.
Table6displaystheresultsfromtheGMMestimationofthebetasystem. PanelAof
Table6shows that thebetasonRXarenotjointlystatisticallysignificantnorsignificantly
differentfromeachother. ThusthepriceofRXriskisnotidentified. Incontrast,betason
thevolatilityfactors(Vol1andVol2)arepositiveandhighlysignificantbutdonotappearto
captureanyofthecrosssectionalspread inreturns.That is,there isnoobviouspatternto
thebetasfromthelow tohighrankingportfolios.
Turning to Panel B of Table 6, when one excludes the constant from the cross
sectionalequation,thepricesofriskonVOL1andVOL2aresmallandinsignificantandthe
R2s of the crosssectional regressions are negative. In addition, the constant terms are
significant. In other words, the volatility factor picks up time series variation in returns
whichiscommontoallportfolios,butthemodeldoesnotexplainthecrosssectionalspread
in returns, or the level of returns for portfolios p1 and p2. There is no evidence for a
negativepriceofvolatilityrisk.
Researchers have also explored skewness as a risk factor for carry trade returns
(Rafferty(2012))andthecrosssectionofequityreturns(Amaya,Christoffersen,Jacobs,and
Vasquez (2011). To investigatewhetherexposure toskewnesscangenerate thereturns to
8/10/2019 2014-033
26/44
23
the technical trading rules, we form a skewness factora tradable portfolio that is long
(short) currencies in thehighest (lowest) skewnessquintile in agivenmonthand then
estimatethemodelsbetarepresentation.Table7presentsthoseresults.
Panel A of Table 7 shows that the skewness betas are all positive and highly
significant.Onecannot,however,rejectthehypothesisthattheyareallequaltoeachother,
precludingstrong identificationofthepricesofrisk. InPanelB,theconstant inthecross
sectionalregressionismarginallysignificantandtheexclusionofthisconstantreducesthe
R2fromahealthy0.25toaverymodest0.09.Withouttheconstant,thepriceofriskis0.82
andstatisticallysignificant.These results suggest thatexposure to skewnessdoesaccount
forsometechnicaltradingrulereturns,althoughitsexplanatorypowerisfairlylimited.
5. DiscussionandConclusion
Researchershave longdocumented theprofitabilityof technicalanalysis in foreign
exchangerates.StudiesthatfoundpositiveresultsincludePoole(1967),DooleyandShafer
(1984), Sweeney (1986), Levich and Thomas (1993), Neely, Weller and Dittmar (1997),
Genay(1999),Lee,GleasonandMathur(2001)andMartin(2001)).
Despitesuchasubstantialrecordofdocumentedgains,thereasonsforthissuccess
remain mysterious. Neely (2002) appears to rule out the central bank intervention
explanation suggestedbyLeBaron (1999).To investigate thepossibilityofdata snooping,
dataminingandpublicationbias,Neely,WellerandUlrich(2009)analyzetheperformance
ofrulesintrueoutofsampleteststhatoccurlongafteranimportantstudy.Theyconclude
thatdatasnooping,dataminingandpublicationbiasareunlikelyexplanationsbutthatthe
adaptivemarketshypothesisisplausible.
8/10/2019 2014-033
27/44
24
It remainspossible, however, that technical trading rule profitability results from
exposuretosomesortofrisk.Recently,severalauthorshaveconsideredmoderntechniques
for riskadjustmentof thecarry tradeormomentum rules in foreignexchangemarkets. If
these factors explain the returns to other trading strategies, they become much more
interesting economically.The goal of thispaperhasbeen to apply abroad range of risk
adjustment techniques to determine whether there is any evidence that exposure to risk
plausiblyexplainstheprofitabilityoftechnicalanalysisintheforeignexchangemarket.
We examine many types of risk adjustment models, including the CAPM, a four
factormodel,severalconsumptionbasedmodels,andfactorsmotivatedbythecarrytrade
puzzle, volatility measures and skewness. Although skewness might have some limited
explanatory power, no model of risk adjustment can plausibly explain the very robust
findingsoftheprofitabilityoftechnicalanalysisintheforeignexchangemarket.
8/10/2019 2014-033
28/44
25
References
Amaya,D.,Christoffersen,P.,Jacobs,K.,&Vasquez,A.,2011,Do realized skewnessand
kurtosispredict thecrosssectionofequity returns (No.201144).SchoolofEconomics
andManagement,UniversityofAarhus.
Brunnermeier, Markus K, Stefan Nagel, and Lasse H Pedersen, 2008, Carry Trades And
Currency Crashes. Kenneth Rogoff, Woodford, Michael, & Acemoglu, Daron. NBER
MacroeconomicsAnnual200823:313347.
Burnside,Craig,2011,Carrytradesandrisk,NBERWorkingPaper17278.
Burnside,Craig,2011,TheCrossSectionofForeignCurrencyRiskPremiaandConsumption
GrowthRisk:Comment.AmericanEconomicReview.
Carhart,MarkM.,1997,Onpersistenceinmutualfundperformance.TheJournalofFinance
52.1:5782.
Dooley, Michael P., and Jeffrey R. Shafer, 1984, Analysis of shortrun exchange rate
behavior:March1973toNovember1981,inDavidBigmanandTeizoTaya,eds.:Floating
Exchange Rates and the State ofWorld Trade Payments (Ballinger Publishing Company,
Cambridge,MA):4370.
Fama,EugeneF.,1970,Efficientcapitalmarkets:A reviewof theoryand empiricalwork.
TheJournalofFinance25.2:383417.
Genay,Ramazan,1999,Linear,nonlinearand essential foreign exchange rateprediction
withsimpletechnicaltradingrules.JournalofInternationalEconomics47.1:91107.
8/10/2019 2014-033
29/44
26
Hassan,Tarek,2013,CountrySize,CurrencyUnions,andInternationalAssetReturns.The
JournalofFinance68.6:22692308.
LeBaron,Blake,1999,Technicaltradingruleprofitabilityandforeignexchangeintervention.
JournalofInternationalEconomics49.1:125143.
Lee, Chun I., Kimberly C. Gleason, and Ike Mathur, 2001, Trading rule profits in Latin
Americancurrencyspotrates.InternationalReviewofFinancialAnalysis10.2:135156.
Levich,RichardM.,andLeeR.ThomasIII,1993,Thesignificanceoftechnicaltradingrule
profits in the foreignexchangemarket:Abootstrapapproach.Journalof International
MoneyandFinance12.5:451474.
Lustig, Hanno, and Adrien Verdelhan, 2011, The cross section of foreign currency risk
premia and consumption growth risk: Reply. The American Economic Review: 3477
3500.
Lustig,Hanno, andAdrienVerdelhan.2007.TheCrossSectionofForeignCurrencyRisk
PremiaandConsumptionGrowthRisk.AmericanEconomicReview,97.1:89117.
Lustig,Hanno,NikolaiRoussanov, andAdrienVerdelhan, 2011,Common risk factors in
currencymarkets. ReviewofFinancialStudies:hhr068.
Lustig,Hanno,NikolaiRoussanov,andAdrienVerdelhan,2014,Countercyclicalcurrency
riskpremia.JournalofFinancialEconomics111.3:527553.
Mancini,L.,Ranaldo,A.andWrampelmeyer,J.,2013,Liquidity in theForeignExchange
Market: Measurement, Commonality, and Risk Premiums. The Journal of Finance
68.5:18051841.
8/10/2019 2014-033
30/44
27
Martin, Anna D., 2001, Technical trading rules in the spot foreign exchange markets of
developingcountries.JournalofMultinationalFinancialManagement11.1:5968.
Lukas Menkhoff & Mark P. Taylor, 2007, The Obstinate Passion of Foreign Exchange
Professionals:TechnicalAnalysis,JournalofEconomicLiterature45.4: 936972.
Menkhoff,Lukas,LucioSarno,MaikSchmelingandAndreasSchrimpf,2012,Carrytrades
andglobalforeignexchangevolatility.JournalofFinance67.2:681718.
Menkhoff, Lukas, Lucio Sarno, Maik Schmeling, and Andreas Schrimpf, 2012, Currency
momentumstrategies.JournalofFinancialEconomics106.3:660684.
Neely,ChristopherJ.,2002,Thetemporalpatternoftradingrulereturnsandexchangerate
intervention: Intervention does not generate technical trading rule profits.Journal of
InternationalEconomics58.1:211232.
Neely, ChristopherJ., and Paul A. Weller, 2013, Lessons from the evolution of foreign
exchangetradingstrategies.JournalofBanking&Finance37.10:37833798.
Neely,ChristopherJ.,PaulA.Weller, andJoshuaM.Ulrich, 2009,The adaptivemarkets
hypothesis: Evidence from the foreign exchange market. Journal of Financial and
QuantitativeAnalysis44.02:467488.
Poole,William,1967,Speculativepricesasrandomwalks:Ananalysisoftentimeseriesof
flexibleexchangerates.SouthernEconomicJournal33:468478.
Rafferty, Barry, 2012, Currency Returns, Skewness and Crash Risk. Skewness and Crash
Risk.Unpublishedmanuscript.
Shanken,Jay.,1992,OntheEstimationofBetaPricingModels.ReviewofFinancialStudies
5.1:133.
8/10/2019 2014-033
31/44
28
Sweeney, RichardJ., 1986, Beating the foreign exchange market.Journal of Finance 41.1:
163182.
Yogo, Motohiro, 2006, A ConsumptionBased Explanation of Expected Stock Returns.
JournalofFinance61.2:53980.
8/10/2019 2014-033
32/44
29
Table1:Datadescription
Notes:The tabledepicts the21exchangeratesversus theUSDand19nonUSDcrossratesused in
oursamplealongwiththestartingandendingdatesofthesamples,numberoftradingdates,average
transactioncost,andstandarddeviationofannualizedlogreturns.
Currency Group Country
Currency abbreviation
versus the USD
# of trading
obs
Trading start
date
Trading end
date Mean TC
STD of Annualized
FX Return
Advanced Australia AUD 9008 4/7/1976 12/31/2012 3.1 11.6
Advanced Canada CAD 9344 1/2/1975 12/31/2012 2.9 6.7
Advanced Euro Area EUR 9717 4/3/1973 12/31/2012 3.0 10.6Advanced Japan JPY 9599 4/3/1973 12/28/2012 3.0 10.5
Advanced New Zealand NZD 6027 8/3/1987 12/31/2012 3.9 12.4
Advanced Norway NOK 6515 1/2/1986 12/31/2012 3.4 11.6
Advanced Sweden SEK 7278 1/3/1983 12/28/2012 3.3 11.4
Advanced Switzerland CHF 9697 4/3/1973 12/31/2012 3.1 12.0
Advanced UK GBP 9338 1/2/1975 12/31/2012 2.9 9.9
Dev. Europe Czech Republic CZK 5049 1/5/1993 12/31/2012 5.2 12.4
Dev. Europe Hungary HUF 4466 1/2/1995 12/28/2012 10.3 14.3
Dev. Europe Hungary/Switzerland HUF_CHF 4165 1/3/1996 12/28/2012 10.5 12.0
Dev. Europe Poland PLN 3918 2/24/1997 12/31/2012 7.1 14.6
Dev. Europe Russia RUB 3055 8/1/2000 12/28/2012 3.6 7.4
Dev. Europe Turkey TRY 2769 1/2/2002 12/31/2012 12.9 15.4
Latin America Brazil BRL 3330 5/3/1999 12/31/2012 6.0 16.8
Latin America Chile CLP 4359 6/1/1995 12/28/2012 5.9 9.5
Latin America Japan/Mexico JPY_MXN 3887 1/4/1996 12/28/2012 4.6 16.9Latin America Mexico MXN 4220 1/4/1996 12/31/2012 4.6 10.5
Latin America Peru PEN 4252 4/1/1996 12/31/2012 5.3 5.0
Other Israel ILS 3750 7/20/1998 12/31/2012 8.1 7.8
Other Israel/Euro Area ILS_EUR 2552 1/2/2003 12/31/2012 8.5 10.2
Other South Africa ZAR 4394 4/3/1995 12/31/2012 8.7 16.4
Other Taiwan TWD 3605 1/5/1998 12/28/2012 5.0 5.3
Adv. Cross Rates Switzerland/UK CHF_GBP 9169 1/3/1975 12/31/2012 3.0 9.8
Adv. Cross Rates Australia/UK AUD_GBP 8920 4/7/1976 12/31/2012 3.2 12.4
Adv. Cross Rates Canada/UK CAD_GBP 9217 1/2/1975 12/31/2012 3.0 10.3
Adv. Cross Rates Japan/UK JPY_GBP 8982 1/2/1975 12/28/2012 3.0 12.2
Adv. Cross Rates Euro Area/UK EUR_GBP 9187 1/2/1975 12/31/2012 3.0 8.1
Adv. Cross Rates Australia/Switzerland AUD_CHF 8848 4/7/1976 12/31/2012 3.2 14.4
Adv. Cross Rates Canada/Switzerland CAD_CHF 9150 1/3/1975 12/31/2012 3.1 12.4
Adv. Cross Rates Japan/Switzerland JPY_CHF 9338 4/3/1973 12/28/2012 3.2 11.7
Adv. Cross Rates Euro Area/Switzerland EUR_CHF 9602 4/3/1973 12/31/2012 3.2 5.9Adv. Cross Rates Canada/Australia CAD_AUD 8894 4/7/1976 12/31/2012 3.2 10.3
Adv. Cross Rates Japan/Australia JPY_AUD 8633 4/7/1976 12/28/2012 3.2 15.3
Adv. Cross Rates Euro Area/Australia EUR_AUD 8861 4/7/1976 12/31/2012 3.2 12.8
Adv. Cross Rates Japan/Canada JPY_CAD 8968 1/2/1975 12/28/2012 3.1 12.7
Adv. Cross Rates Euro Area/Canada EUR_CAD 9158 1/2/1975 12/31/2012 3.1 10.7
Adv. Cross Rates Japan/Euro Area JPY_EUR 9347 4/3/1973 12/28/2012 3.1 11.3
Adv. Cross Rates New Zealand/Australia NZD_AUD 5943 8/3/1987 12/31/2012 3.9 8.7
8/10/2019 2014-033
33/44
30
Table2: Calibration
CCAPM DCAPM EZCCAPM EZDCAPM
stdT[M]/ET[M] 0.93 1.22 0.92 1.22
varT[M]/ET[M] 0.60 0.61 0.59 0.61
MAE(in%) 1.32 0.96 1.33 0.96
R2 1.50 0.13 1.50 0.13
Notes:Thesampleis19782010(annualdata).ThetableshowstheresultsofestimatinganEuler
equation, as in equation (3), with 12 dynamic technical trading strategies as test assets. The
returnsare those toportfoliosp1 top12,asdescribed in the text.The first tworowsreport the
maximumSharperatio(row1,stdT[M]/ET[M])andthepriceofrisk(row2,varT[M]/ET[M]).The
lasttworowsreportthemeanabsolutepricingerror(inpercentagepoints)andtheR2.FollowingYogo (2006), we fixed sigma () at 0.023 (EZCCAPM and EZDCAPM), alpha () at 0.802
(DCAPMandEZDCAPM),delta()at0.98,andrho()at0.700(DCAPM,EZDCAPM).Gamma
()isfixedat41.16tominimizethemeansquaredpricingerrorintheEZDCAPM.
8/10/2019 2014-033
34/44
31
Table3:ResultsforCAPM,QuadraticCAPMandCarhartmodelforportfoliosp1p12
p1 p2 p3 p4 p5 p6 p7 p8 p9 p10 p11CAPM
Constant 0.39 *** 0.25 *** 0.21 *** 0.13 * 0.20 *** 0.07 0.14 ** 0.14 ** 0.12 * 0.13 ** 0.11
Rm 0.04 * 0.05 ** 0.04 * 0.04 * 0.03 0.05 ** 0.05 *** 0.04 * 0.05 *** 0.04 ** 0.04
R2
0.01 0.02 0.01 0.02 0.01 0.02 0.03 0.02 0.03 0.03 0.02
Quad.CAPM
Constant 0.26 *** 0.17 ** 0.12 * 0.05 0.14 ** 0.00 0.09 0.07 0.05 0.09 0.02
Rm 0.03 0.04 * 0.03 0.04 * 0.02 0.04 ** 0.05 *** 0.03 0.05 *** 0.04 *** 0.03
Rm2 0.65 *** 0.39 0.46 0.38 0.25 0.35 0.23 0.32 0.33 0.20 0.60
R2
0.04 0.03 0.03 0.03 0.02 0.04 0.03 0.03 0.04 0.03 0.05
CarhartConstant 0.41 *** 0.27 *** 0.21 *** 0.13 * 0.17 *** 0.07 0.13 * 0.13 ** 0.11 0.12 ** 0.08
Rm 0.04 * 0.05 ** 0.03 0.04 ** 0.02 0.04 * 0.03 0.03 0.04 ** 0.03 0.02
SMB 0.05 * 0.04 * 0.04 0.02 0.02 0.03 0.03 * 0.03 0.04 * 0.04 * 0.03
HML 0.06 * 0.03 0.02 0.02 0.01 0.01 0.01 0.01 0.01 0.01 0.02
UMD 0.03 * 0.02 0.02 0.02 0.03 ** 0.02 0.02 0.02 0.02 * 0.03 ** 0.03
R2
0.04 0.03 0.02 0.02 0.02 0.04 0.03 0.03 0.05 0.04 0.03
Mean R 0.37 0.23 0.19 0.11 0.18 0.05 0.11 0.12 0.09 0.11 0.09
Panel A: Time Series Regressions
8/10/2019 2014-033
35/44
32
Rm 3.34 3.29 2.99 0.61 3.16 2.27
(2.13) * (2.59) ** (1.92) * (0.47) (2.23) * (1.57)
[1.70] [2.08] * [1.44] [0.37] [1.09] [0.70]{1.44} {2.02} * {1.26} {0.33} {1.06} {0.62}
Rm2 0.17 0.31
(1.70) (3.88) ***
[1.31] [3.10] **
{1.31} {3.10} **
SMB 1.64 2.41
(1.01) (2.04) *
[0.49] [0.91]
{0.44} {0.90}
HML 3.76 3.50
(3.76) *** (3.72) ***
[1.83] [1.67]
{1.92} * {1.70}
UMD 4.01 5.37
(2.08) * (3.58) ***
[1.01] [1.60]
{1.08} {1.60}
Constant 0.28 0.18 0.09(3.11) ** (2.00) * (0.82)
[2.55] ** [1.50] [0.38]
{2.33} ** {1.29} {0.35}
R2
0.05 0.16 0.18 0.11 0.73 0.72
Panel B: Cross Sectional Regressions
CarhartQuad. CAPMCAPM
Notes: Monthly data 06/1977 12/2012. The table shows the results of estimating a beta
representationmodelofdynamic technical trading strategiesonequity factors.Factorsare the
excessreturnontheU.S.stockmarket(),thesizefactor(SMB),thevaluefactor()andthemomentum factor (). tstatisticsare indicated inparentheses. ( ) tstatisticsbasedonOLSstandarderrors. []tstatisticsbasedonShankenstandarderrors. {}tstatisticsbasedonGMM
standarderrors. Significancelevels: ***1%,**5%,*10%.
8/10/2019 2014-033
36/44
33
Table4:ResultsforCCAPM,DCAPMandEZDCAPMmodelforportfoliosp1p12
p1 p2 p3 p4 p5 p6 p7 p8 p9 p10 p11
CCAPM
Constant 5.95 *** 3.36 ** 2.56 1.14 2.14 0.89 1.18 2.39 ** 1.12 0.91 0.29
Nondurables 0.99 ** 0.44 0.29 0.09 0.09 0.05 0.19 0.67 0.11 0.20 0.46
R2
0.11 0.02 0.01 0.00 0.00 0.00 0.00 0.07 0.00 0.01 0.03
DCAPM
Constant 4.75 *** 4.66 *** 2.93 3.22 ** 3.90 1.37 2.78 3.76 * 1.83 1.30 0.33
Nondurables 1.55 *** 0.16 0.12 1.05 * 0.90 0.17 0.92 0.04 0.22 0.37 0.48
Durables 0.56 0.61 * 0.17 0.97 ** 0.82 0.23 0.74 * 0.64 0.33 0.18 0.02
R2 0.14 0.07 0.01 0.09 0.09 0.01 0.08 0.14 0.02 0.01 0.03
EZDCAPM `
Constant 5.73 *** 5.46 ** 3.56 4.82 *** 5.24 ** 2.32 3.52 4.35 ** 2.04 1.81 1.49
Nondurables 1.33 ** 0.34 0.02 1.41 * 1.20 0.38 1.09 0.10 0.27 0.49 0.74
Durables 0.37 0.76 0.29 1.29 ** 1.08 0.41 0.89 0.75 0.37 0.28 0.25
Market 0.07 0.05 0.04 0.11 ** 0.09 ** 0.06 0.05 0.04 0.01 0.04 0.08
R2
0.20 0.12 0.04 0.23 0.21 0.07 0.12 0.17 0.02 0.03 0.15
Mean R 4.57 2.76 2.16 1.26 2.27 0.81 1.44 1.46 0.97 1.19 0.94
Panel A: Time Series Regressions
8/10/2019 2014-033
37/44
34
Notes:Annualdata19782010.The tableshowstheresultsofestimatingabetarepresentation
modelofdynamic technical trading strategieson consumption factors.Nondurables () andDurables()arelognondurable(plusservices)anddurableconsumptiongrowthrespectively,and Market ( ,) is the log return on the market portfolio. tstatistics are indicated inparentheses. ( ) t statisticsbased on OLS standard errors. [ ] t statisticsbased on Shanken
standarderrors. {}t statisticsbasedonGMMstandarderrors. Significancelevels: ***1%,**5%,
*10%.
Nondurables 1.95 3.09 1.90 3.20 2.03 2.24
(4.08) *** (5.28) *** (4.19) *** (5.19) *** (4.26) *** (5.14) ***
[2.79] ** [2.55] ** [2.94] ** [2.20] * [2.43] ** [2.60] * *
{3.08} ** {2.76} ** {3.29} *** {2.02} * {2.44} ** {2.65} **
Durables 1.76 4.77 1.61 1.89
(2.66) ** (3.72) *** (2.48) ** (2.83) **
[1.85] * [1.51] [1.43] [1.39]
{2.06} * {1.31} {1.43} {1.46}
Market 19.32 26.95
(2.63) ** (3.28) ***
[1.40] [1.47]
{1.56} {1.21}
Constant 1.43 1.48 0.58
(2.42) ** (2.68) ** (0.97)
[1.49] [1.67] [0.49]
{1.54} {1.68} {0.41}
Parameters
84.32 133.94 82.33 139.40 91.89 102.72
(4.08) *** (5.28) *** (4.19) *** (5.17) *** (4.27) *** (5.23) ***
[2.79] ** [2.55] ** [2.93] ** [2.19] * [2.41] ** [2.61] * *
{3.08} ** {2.76} ** {3.28} *** {2.01} * {2.45} ** {2.66} **
0.09 0.11
(3.28) ** (3.32) ***
[1.99] * [1.61]
{2.14} * {1.32}
0.06 0.50 0.01 0.12
(0.26) (2.35) ** (0.06) (0.54)
[0.18] [1.00] [0.06] [0.26]
{0.19} {0.87} {0.04} {0.26}
R2
0.50 1.10 0.50 0.45 0.65 0.61
Panel B: Cross Sectional Regressions
EZDCAPMDCAPMCCAPM
8/10/2019 2014-033
38/44
35
Table5:ResultsforLVmodelforportfoliosp1p12
p1 p2 p3 p4 p5 p6 p7 p8 p9 p10 p11
Constant 0.35 *** 0.20 *** 0.17 ** 0.06 0.12 0.04 0.10 0.09 0.12 0.15 * 0.11
Rx 0.02 0.03 0.04 0.02 0.02 0.02 0.01 0.02 0.06 0.04 0.02
HMLfx 0.03 0.07 * 0.05 0.05 0.03 0.08 ** 0.09 ** 0.07 * 0.09 ** 0.10 ** 0.09
R2
0.00 0.02 0.01 0.01 0.01 0.03 0.04 0.02 0.05 0.06 0.04
Mean R 0.32 0.15 0.12 0.02 0.10 0.02 0.04 0.04 0.04 0.07 0.04
Panel A: Time Series Regressions
Constant 0.22
(2.66) **
[2.04] *
{1.84} *
Rx 0.29 1.73
(0.53) (2.32) **
[0.41] [1.78]
{0.39} {1.68}
HMLfx 2.08 0.32
(2.96) ** (0.49)
[2.29] ** [0.38]
{1.78} {0.31}
R2
0.35 0.24
Panel B: Cross Sectional Regression
Note: Monthly data 11/ 1983 12/2012. The table shows the results of estimating abeta representation mod
strategiesoncurrencymarket factors.Rx theaveragecurrencyexcessreturn togoingshort in thedollarand
currencyportfolios.HMLFX thereturntoastrategythatborrowslowinterestratecurrenciesandinvestsinhigh
words a carry trade. tstatistics are indicated inparentheses. ( ) t statisticsbasedonOLS standard errors. [
standarderrors. {}t statisticsbasedonGMMstandarderrors. Significancelevels: ***1%,**5%,*10%.
Table6:Resultsforvolatilityfactorsforportfoliosp1p12
8/10/2019 2014-033
39/44
36
p1 p2 p3 p4 p5 p6 p7 p8 p9 p10 p11
Constant 0.29 *** 0.12 ** 0.11 * 0.01 0.08 0.05 0.01 0.03 0.02 0.04 0.02
Rx 0.01 0.03 0.03 0.01 0.01 0.01 0.01 0.02 0.05 0.04 0.02
VOL1 2.33 ** 2.15 ** 1.71 2.51 ** 1.70 ** 2.93 *** 1.94 ** 1.75 * 2.60 ** 2.63 ** 2.01
R2
0.03 0.03 0.02 0.05 0.03 0.07 0.03 0.03 0.06 0.07 0.04
Constant 0.32 *** 0.15 ** 0.13 ** 0.02 0.10 0.02 0.04 0.05 0.05 0.08 0.05
Rx 0.01 0.03 0.03 0.01 0.01 0.01 0.02 0.02 0.05 0.03 0.02
VOL2 1.86 ** 1.92 ** 1.67 * 2.39 *** 1.73 *** 2.83 *** 2.22 *** 1.62 ** 2.72 *** 2.64 *** 1.77 **
R2
0.02 0.03 0.03 0.05 0.03 0.08 0.05 0.03 0.08 0.09 0.03
Mean R 0.32 0.15 0.12 0.02 0.10 0.02 0.04 0.04 0.04 0.07 0.04
Panel A: Time Series Regressions
8/10/2019 2014-033
40/44
37
Constant 0.19 0.28
(2.21) * (3.09) **
[2.03] * [2.37] **
{1.91} * {2.42} **
Rx 0.02 0.22 0.55 0.64
(0.04) (0.39) (0.96) (1.10)
[0.04] [0.38] [0.74] [1.04]
{0.03} {0.41} {0.68} {1.06}
VOL1 0.05 0.03
(1.90) * (1.24)
[1.75] [1.20]
{1.62} {1.35}
VOL2 0.10 0.03
(3.25) *** (1.04)
[2.52] ** [0.99]
{2.22} * {1.01}
R2
0.06 0.11 0.22 0.17
Panel B: Cross Sectional Regression
Note: Monthly data 11/ 1983 12/2012. The table shows the results of estimating abeta representation mod
strategiesoncurrencyvolatilityfactors.Rxtheaveragecurrencyexcessreturntogoingshort inthedollarand
currency portfolios. VOL1 volatility innovations measuredby the residuals from AR(1). VOL2 volatility
difference. tstatisticsareindicatedinparentheses.()t statisticsbasedonOLSstandarderrors. []t statisticsba
{}t statisticsbasedonGMMstandarderrors. Significancelevels: ***1%,**5%,*10%.
8/10/2019 2014-033
41/44
38
Table7:Resultsforskewnessfactorforportfoliosp1p12
p1 p2 p3 p4 p5 p6 p7 p8 p9 p10 p11
Constant 0.73 *** 0.78 *** 0.86 *** 0.91 *** 0.71 *** 0.93 *** 0.87 *** 0.91 *** 0.96 *** 0.91 *** 0.85 *
SKEW 0.20 *** 0.20 *** 0.19 *** 0.19 *** 0.16 *** 0.18 *** 0.18 *** 0.19 *** 0.19 *** 0.19 *** 0.17 *
R2
0.18 0.19 0.22 0.22 0.19 0.22 0.21 0.26 0.25 0.27 0.21
Mean R 0.37 0.23 0.19 0.11 0.18 0.05 0.11 0.12 0.09 0.11 0.09
Panel A: Time Series Regressions
Co nstant 0.6 4
(2.78) **
[1.78]
{1.83} *
SKEW 4.31 0.82
(3.45) *** (2.73) **
[2.16] * [2.73] **
{2.33} ** {3.15} ***
R2
0.25 0.09
Panel B: Cross Sectional Regression
Note: Monthly data 06/1977 12/2012. The table shows the results of estimating abeta representation mod
strategiesoncurrencyvolatilityfactors.SKEWreturnofaportfoliothatislongcurrenciesinthehighestskew
currencies in the lowest (negative) skewnessquintile foragivenmonth. tstatisticsare indicated inparenthes
standarderrors. []t statisticsbasedonShankenstandarderrors. {}t statisticsbasedonGMMstandarderror5%,*10%.
8/10/2019 2014-033
42/44
39
Figure1:Return,standarddeviationandSharperatiostatisticsfromsortedportfoliosp1
top12
Notes: Thefigureshowsstatisticsfromthe12dynamicallyconstructed,technicaltrading
portfoliosinforeignexchangemarkets,basedonproceduresinNeelyandWeller(2013).
8/10/2019 2014-033
43/44
40
Figure2:Actualmeanannualreturnsvs.predictedreturnsforportfoliosp1throughp12
(crosssectionalregressionwithaconstant)19782010
Notes: Thefigureshowsscatterplotsofactualversuspredictedreturnsfromthe12dynamically
constructed,technicaltradingportfoliosinforeignexchangemarkets,basedonproceduresinNeelyandWeller(2013). Thepredictedreturnsareconstructedfromthebetarepresentationof
thelinearconsumptionmodelswithaconstantinthecrosssectionalequation.
0 1 2 3 4 50
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
1
2
3
4
5
6
7 8
9
10
11
12
Predicted R
ActualR
CCAPM
0 1 2 3 4 50
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
1
2
3
4
5
6
7 8
9
10
11
12
Predicted R
ActualR
DCAPM
0 1 2 3 4 50
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
1
2
3
4
5
6
7 8
9
10
11
12
Predicted R
ActualR
EZ-DCAPM
8/10/2019 2014-033
44/44
Figure3:Actualmeanannualreturnsvs.predictedreturnsforportfoliosp1throughp12
(crosssectionalregressionwithoutaconstant)19782010
Notes: Thefigureshowsscatterplotsofactualversuspredictedreturnsfromthe12dynamically
constructed,technicaltradingportfoliosinforeignexchangemarkets,basedonproceduresin
NeelyandWeller(2013). Thepredictedreturnsareconstructedfromthebetarepresentationof
thelinearconsumptionmodelswithoutaconstantinthecrosssectionalequation.
-2 -1 0 1 2 3 4 5-2
-1
0
1
2
3
4
51
2
3
4
5
6
7 8
910
11
12
Predicted R
ActualR
CCAPM
-2 -1 0 1 2 3 4-2
-1
0
1
2
3
4
5
1
2
3
4
5
6
7 8
910
11
12
Predicted R
ActualR
EZ-DCAPM
-2 -1 0 1 2 3 4 5-2
-1
0
1
2
3
4
51
2
3
4
5
6
7 8
910
11
12
Predicted R
ActualR
DCAPM