can hedge-fund returns be replicated?: the...
TRANSCRIPT
JOIMwww.joim.com
JOURNAL OF INVESTMENT MANAGEMENT, Vol. 5, No. 2, (2007), pp. 5–45
© JOIM 2007
CAN HEDGE-FUND RETURNS BE REPLICATED?: THE LINEAR CASE�
Jasmina Hasanhodzica and Andrew W. Lob,∗
In contrast to traditional investments such as stocks and bonds, hedge-fund returns havemore complex risk exposures that yield additional and complementary sources of risk premia.This raises the possibility of creating passive replicating portfolios or “clones” using liquidexchange-traded instruments that provide similar risk exposures at lower cost and withgreater transparency. By using monthly returns data for 1610 hedge funds in the TASSdatabase from 1986 to 2005, we estimate linear factor models for individual hedge fundsusing six common factors, and measure the proportion of the funds’ expected returns andvolatility that are attributable to such factors. For certain hedge-fund style categories, wefind that a significant fraction of both can be captured by common factors corresponding toliquid exchange-traded instruments. While the performance of linear clones is often inferiorto their hedge-fund counterparts, they perform well enough to warrant serious considerationas passive, transparent, scalable, and lower-cost alternatives to hedge funds.
�The views and opinions expressed in this paper are those ofthe authors only, and do not necessarily represent the viewsand opinions of AlphaSimplex Group, MIT, or any of theiraffiliates and employees. The authors make no representationsor warranty, either expressed or implied, as to the accuracy orcompleteness of the information contained in this paper, norare they recommending that this paper can serve as the basisfor any investment decision—this paper is for informationpurposes only.aDepartment of Electrical Engineering and Computer Sci-ence; Laboratory for Financial Engineering, MIT, Cambridge,MA, USAbSloan School of Management; Laboratory for Financial Engi-neering, MIT, Cambridge, MA, USA and AlphaSimplexGroup, LLC
1 Introduction
As institutional investors take a more active inter-est in alternative investments, a significant gap hasemerged between the culture and expectations ofthose investors and hedge-fund managers. Pensionplan sponsors typically require transparency fromtheir managers and impose a number of restrictionsin their investment mandates because of regula-tory requirements such as ERISA rules; hedge-fundmanagers rarely provide position-level transparency
∗Corresponding author. MIT Sloan School, 50 Memo-rial Derive, E52–454, Cambridge 02142–1347, MA, USA.E-mail: [email protected]
SECOND QUARTER 2007 5
6 JASMINA HASANHODZIC AND ANDREW W. LO
and bristle at any restrictions on their investmentprocess because restrictions often hurt performance.Plan sponsors require a certain degree of liquidityin their assets to meet their pension obligations,and also desire significant capacity because of theirlimited resources in managing large pools of assets;hedge-fund managers routinely impose lock-ups of1–3 years, and the most successful managers havethe least capacity to offer, in many cases return-ing investors’ capital once they make their personalfortunes. And as fiduciaries, plan sponsors arehypersensitive to the outsize fees that hedge fundscharge, and are concerned about misaligned incen-tives induced by performance fees; hedge-fundmanagers argue that their fees are fair compensa-tion for their unique investment acumen, and atleast for now, the market seems to agree.
This cultural gap raises the natural question ofwhether it is possible to obtain hedge-fund-likereturns without investing in hedge funds. In short,can hedge-fund returns be “cloned”?
In this paper, we provide one answer to thischallenge by constructing “linear clones” of indi-vidual hedge funds in the TASS Hedge FundDatabase. These are passive portfolios of com-mon risk factors like the S&P 500 and theUS Dollar Indexes, with portfolio weights esti-mated by regressing individual hedge-fund returnson the risk factors. If a hedge fund generates partof its expected return and risk profile from certaincommon risk factors, then it may be possible todesign a low-cost passive portfolio—not an activedynamic trading strategy—that captures some ofthat fund’s risk/reward characteristics by taking onjust those risk exposures. For example, if a particularlong/short equity hedge fund is 40% long growthstocks, it may be possible to create a passive port-folio that has similar characteristics, for example,a long-only position in a passive growth portfoliocoupled with a 60% short position in stock-indexfutures.
The magnitude of hedge-fund alpha that can becaptured by a linear clone depends, of course, onhow much of a fund’s expected return is drivenby common risk factors versus manager-specificalpha. This can be measured empirically. Whileportable alpha strategies have become fashionablelately among institutions, our research suggeststhat for certain classes of hedge-fund strategies,portable beta may be an even more importantsource of untapped expected returns and diversifi-cation. In particular, in contrast to previous studiesemploying more complex factor-based models ofhedge-fund returns, we use six factors that corre-spond to basic sources of risk and, consequently,expected return: the stock market, the bond mar-ket, currencies, commodities, credit, and volatility.These factors are also chosen because, with theexception of volatility, each of them is tradable vialiquid exchange-traded securities such as futures orforward contracts.
Using standard regression analysis we decomposethe expected returns of a sample of 1610 indi-vidual hedge funds from the TASS Hedge FundLive Database into factor-based risk premia andmanager-specific alpha, and we find that for certainhedge-fund style categories, a significant fraction ofthe funds’ expected returns is due to risk premia.For example, in the category of Convertible Arbi-trage funds, the average percentage contribution ofthe US Dollar Index risk premium, averaged acrossall funds in this category, is 67%. While estimatesof manager-specific alpha are also quite significantin most cases, these results suggest that at least aportion of a hedge fund’s expected return can beobtained by bearing factor risks.
To explore this possibility, we construct linear clonesusing five of the six factors (we omit volatilitybecause the market for volatility swaps and futuresis still developing), and compare their performanceto the original funds. For certain categories such asEquity Market Neutral, Global Macro, Long/Short
JOURNAL OF INVESTMENT MANAGEMENT SECOND QUARTER 2007
REPLICATING HEDGE-FUND RETURNS 7
Equity Hedge, Managed Futures, Multi-Strategy,and Fund of Funds, linear clones have comparableperformance to their fund counterparts, but forother categories such as Event Driven and Emerg-ing Markets, clones do not perform nearly as well.However, in all cases, linear clones are more liquid(as measured by their serial correlation coefficients),more transparent and scalable (by construction),and with correlations to a broad array of marketindexes that are similar to those of the hedge fundson which they are based. For these reasons, weconclude that hedge-fund replication, at least forcertain types of funds, is both possible and, in somecases, worthy of serious consideration.
We begin in Section 2 with a brief review of theliterature on hedge-fund replication, and providetwo examples that motivate this endeavor. In Sec-tion 3, we present a linear regression analysis ofhedge-fund returns from the TASS Hedge FundLive Database, with which we decompose the funds’expected returns into risk premia and manager-specific alpha. These results suggest that for certainhedge-fund styles, linear clones may yield reason-ably compelling investment performance, and weexplore this possibility directly in Section 4. Weconclude in Section 5.
2 Motivation
In a series of recent papers, Kat and Palaro(2005, 2006a,b) argued that sophisticated dynamictrading strategies involving liquid futures con-tracts can replicate many of the statistical prop-erties of hedge-fund returns. More generally,Bertsimas et al. (2001) have shown that securi-ties with very general payoff functions (like hedgefunds, or complex derivatives) can be syntheti-cally replicated to an arbitrary degree of accuracyby dynamic trading strategies—called “epsilon-arbitrage” strategies—involving more liquid instru-ments. While these results are encouraging for the
hedge-fund replication problem, the replicatingstrategies are quite involved and not easily imple-mented by the typical institutional investor. Indeed,some of the derivatives-based replication strategiesmay be more complex than the hedge-fund strate-gies they intend to replicate, defeating the verypurpose of replication.1
The motivation for our study comes, instead, fromSharpe’s (1992) asset-class factor models in which heproposes to decompose a mutual fund’s return intotwo distinct components: asset-class factors such aslarge-cap stocks, growth stocks, and intermediategovernment bonds, which he interprets as “style,”and an uncorrelated residual that he interprets as“selection.” This approach was applied by Fung andHsieh (1997a) to hedge funds, but there the factorswere derived statistically from a principal compo-nents analysis of the covariance matrix of theirsample of 409 hedge funds and CTAs. While suchfactors may yield high in-sample R2s, they sufferfrom significant over-fitting bias and also lack eco-nomic interpretation, which is one of the primarymotivations for Sharpe’s (1992) decomposition.Several authors have estimated factor models forhedge funds using more easily interpretable factorssuch as fund characteristics and indexes (Schneeweisand Spurgin, 1998; Liang, 1999; Edwards andCaglayan, 2001; Capocci and Hubner, 2004; Hill,et al., 2004), and the returns to certain options-based strategies and other basic portfolios (Fungand Hsieh, 2001, 2004; Agarwal and Naik 2000a,b,2004).
However, the most direct application of Sharpe’s(1992) analysis to hedge funds is found in Ennis andSebastian (2003). They provide a thorough styleanalysis of the HFR Fund of Funds index, and con-clude that funds of funds are not market neutraland although they do exhibit some market-timingabilities, “…, the performance of hedge funds hasnot been good enough to warrant their inclusionin balanced portfolios. The high cost of investing
SECOND QUARTER 2007 JOURNAL OF INVESTMENT MANAGEMENT
8 JASMINA HASANHODZIC AND ANDREW W. LO
in funds of funds contributes to this result” (Ennisand Sebastian, 2003, p. 111). This conclusion is thestarting point for our study of linear clones.
Before turning to our empirical analysis of individ-ual hedge-fund returns, we provide two concreteexamples that span the extremes of the hedge-fundreplication problem. For one hedge-fund strategy,we show that replication can be accomplished eas-ily, and for another strategy, we find replication tobe almost impossible using linear models.
2.1 Capital Decimation Partners
The first example is a hypothetical strategy proposedby Lo (2001) called “Capital Decimation Partners”(CDP), which yields an enviable track record thatmany investors would associate with a successfulhedge fund: a 43.1% annualized mean return and20% annualized volatility, implying a Sharpe ratioof 2.15,2 and with only 6 negative months over the96-month simulation period from January 1992 toDecember 1999 (see Table 1). A closer inspection ofthis strategy’s monthly returns in Table 2 yields few
Table 1 Performance summary of simulatedshort-put-option strategy consisting of short-selling out-of-the-money S&P 500 put optionswith strikes approximately 7% out of the moneyand with maturities less than or equal to 3months, from January 1992 to December 1999.
Statistic S&P500 CDP
Monthly mean 1.4% 3.6%Monthly SD 3.6% 5.8%Minimum month −8.9% −18.3%Maximum month 14.0% 27.0%Annual Sharpe ratio 1.39 2.15# Negative months 36 6Correlation to S&P 500 100% 61%Return Since Inception 367% 2560%
surprises for the seasoned hedge-fund investor—themost challenging period for CDP was the summerof 1998 during the LTCM crisis, when the strategysuffered losses of −18.3% and −16.2% in Augustand September, respectively. But those investorscourageous enough to have maintained their CDPinvestment during this period were rewarded withreturns of 27.0% in October and 22.8% in Novem-ber. Overall, 1998 was the second-best year forCDP, with an annual return of 87.3%.
So what is CDP’s secret? The investment strategysummarized inTables 1 and 2 involves shorting out-of-the-money S&P 500 (SPX) put options on eachmonthly expiration date for maturities less thanor equal to 3 months, and with strikes approx-imately 7% out of the money. According to Lo(2001), the number of contracts sold each monthis determined by the combination of: (1) CBOEmargin requirements3; (2) an assumption that weare required to post 66% of the margin as collateral4;and (3) $10M of initial risk capital. The essenceof this strategy is the provision of insurance. CDPinvestors receive option premia for each option con-tract sold short, and as long as the option contractsexpire out of the money, no payments are necessary.Therefore, the only time CDP experiences lossesis when its put options are in the money, that is,when the S&P 500 declines by more than 7% dur-ing the life of a given option. From this perspective,the handsome returns to CDP investors seem morejustifiable—in exchange for providing downsideprotection, CDP investors are paid a risk premiumin the same way that insurance companies receiveregular payments for providing earthquake or hur-ricane insurance. Given the relatively infrequentnature of 7% losses, CDP’s risk/reward profile canseem very attractive in comparison to more tradi-tional investments, but there is nothing unusualor unique about CDP. Investors willing to takeon “tail risk”—the risk of rare but severe events—will be paid well for this service (consider howmuch individuals are willing to pay each month for
JOURNAL OF INVESTMENT MANAGEMENT SECOND QUARTER 2007
REPLICATING HEDGE-FUND RETURNS 9
Tab
le2
Mon
thly
retu
rns
ofsi
mul
ated
shor
t-pu
t-op
tion
stra
tegy
cons
isti
ngof
shor
tsel
ling
out-
of-t
he-m
oney
S&P
500
put
opti
ons
wit
hst
rike
sapp
roxi
mat
ely
7%ou
toft
hem
oney
and
wit
hm
atur
itie
sles
stha
nor
equa
lto
3m
onth
s,fr
omJa
nuar
y19
92to
Dec
embe
r19
99.
1992
1993
1994
1995
1996
1997
1998
1999
Mon
thSP
XC
DP
SPX
CD
PSP
XC
DP
SPX
CD
PSP
XC
DP
SPX
CD
PSP
XC
DP
SPX
CD
P
Jan
8.2
8.1
−1.2
1.8
1.8
2.3
1.3
3.7
−0.7
1.0
3.6
4.4
1.6
15.3
5.5
10.1
Feb
−1.8
4.8
−0.4
1.0
−1.5
0.7
3.9
0.7
5.9
1.2
3.3
6.0
7.6
11.7
−0.3
16.6
Mar
0.0
2.3
3.7
3.6
0.7
2.2
2.7
1.9
−1.0
0.6
−2.2
3.0
6.3
6.7
4.8
10.0
Apr
1.2
3.4
−0.3
1.6
−5.3
−0.1
2.6
2.4
0.6
3.0
−2.3
2.8
2.1
3.5
1.5
7.2
May
−1.4
1.4
−0.7
1.3
2.0
5.5
2.1
1.6
3.7
4.0
8.3
5.7
−1.2
5.8
0.9
7.2
Jun
−1.6
0.6
−0.5
1.7
0.8
1.5
5.0
1.8
−0.3
2.0
8.3
4.9
−0.7
3.9
0.9
8.6
Jul
3.0
2.0
0.5
1.9
−0.9
0.4
1.5
1.6
−4.2
0.3
1.8
5.5
7.8
7.5
5.7
6.1
Aug
−0.2
1.8
2.3
1.4
2.1
2.9
1.0
1.2
4.1
3.2
−1.6
2.6
−8.9
−18.
3−5
.8−3
.1Se
p1.
92.
10.
60.
81.
60.
84.
31.
33.
33.
45.
511
.5−5
.7−1
6.2
−0.1
8.3
Oct
−2.6
−3.0
2.3
3.0
−1.3
0.9
0.3
1.1
3.5
2.2
−0.7
5.6
3.6
27.0
−6.6
−10.
7N
ov3.
68.
5−1
.50.
6−0
.72.
72.
61.
43.
83.
02.
04.
610
.122
.814
.014
.5D
ec3.
41.
30.
82.
9−0
.610
.02.
71.
51.
52.
0−1
.76.
71.
34.
3−0
.12.
4
Year
14.0
38.2
5.7
23.7
−1.6
33.6
34.3
22.1
21.5
28.9
26.4
84.8
24.5
87.3
20.6
105.
7
SECOND QUARTER 2007 JOURNAL OF INVESTMENT MANAGEMENT
10 JASMINA HASANHODZIC AND ANDREW W. LO
their homeowner’s, auto, health, and life insurancepolicies). CDP involves few proprietary elements,and can be implemented by most investors, hencethis is one example of a hedge-fund-like strategythat can easily be cloned.
2.2 Capital Multiplication Partners
Consider now the case of “Capital MultiplicationPartners” (CMP), a hypothetical fund based on adynamic asset-allocation strategy between the S&P500 and 1-month US Treasury Bills, where thefund manager can correctly forecast which of thetwo assets will do better in each month and investsthe fund’s assets in the higher-yielding asset at thestart of the month.5 Therefore, the monthly returnof this perfect market-timing strategy is simply thelarger of the monthly return of the S&P 500 andT-Bills. The source of this strategy’s alpha is clear:Merton (1981) observes that perfect market-timingis equivalent to a long-only investment in the S&P500 plus a put option on the S&P 500 with a strikeprice equal to the T-Bill return. Therefore, the eco-nomic value of perfect market-timing is equal to thesum of monthly put-option premia over the life ofthe strategy. And there is little doubt that such astrategy contains significant alpha: a $1 investmentin CMP in January 1926 grows to $23,143,205,448
Table 3 Performance summary of simulated monthly perfect market-timingstrategy between the S&P 500 and 1-month US Treasury bills, and a passivelinear clone, from January 1926 to December 2004.
Statistic S&P 500 T-Bills CMP Clone
Monthly mean 1.0% 0.3% 2.6% 0.7%Monthly SD 5.5% 0.3% 3.6% 3.0%Minimum month −29.7% −0.1% −0.1% −16.3%Maximum month 42.6% 1.4% 42.6% 23.4%Annual Sharpe ratio 0.63 4.12 2.50 0.79# Negative months 360 12 10 340Correlation to S&P 500 100% −2% 84% 100%Growth of $1 since inception $3,098 $18 $2.3 × 1010 $429
by December 2004! Table 3 provides a more detailedperformance summary of CMP which confirmsits remarkable characteristics—CMP’s Sharpe ratioof 2.50 exceeds that of Warren Buffett’s BerkshireHathaway, arguably the most successful pooledinvestment vehicle of all time!6
It should be obvious to even the most naive investorthat CMP is a fantasy because no one can time themarket perfectly. Therefore, attempting to replicatesuch a strategy with exchange-traded instrumentsseems hopeless. But suppose we try to replicateit anyway—how close can we come? In particu-lar, suppose we attempt to relate CMP’s monthlyreturns to the monthly returns of the S&P 500by fitting a simple linear regression (see Figure 1).The option-like nature of CMP’s perfect market-timing strategy is apparent in Figure 1’s scatterof points, and visually, it is obvious that the lin-ear regression does not capture the essence of thisinherently nonlinear strategy. However, the formalmeasure of how well the linear regression fits the
data, the “R2,” is 70.3% in this case, which suggests
a very strong linear relationship indeed. But whenthe estimated linear regression is used to constructa fixed portfolio of the S&P 500 and 1-monthT-Bills, the results are not nearly as attractive asCMP’s returns, as Table 3 shows.
JOURNAL OF INVESTMENT MANAGEMENT SECOND QUARTER 2007
REPLICATING HEDGE-FUND RETURNS 11
Regression of CMP Returns on S&P 500 ReturnsJanuary 1926 to December 2004
y = 0.5476x + 0.0206
R2 = 0.7027
-20%
-10%
0%
10%
20%
30%
40%
50%
-40% -30% -20% -10% 0% 10% 20% 30% 40% 50%
S&P 500 Return
CM
P R
etu
rn
CMP vs. S&P 500 Returns Linear (CMP vs. S&P 500 Returns)
Figure 1 Scatter plot of simulated monthly returns of a perfect market-timing strategy between the S&P500 and 1-month US Treasury bills, against monthly returns of the S&P 500, from January 1926 toDecember 2004.
This example underscores the difficulty in replicat-ing certain strategies with genuine alpha using linear
clones, and cautions against using the R2
as the only
metric of success. Despite the high R2
achieved bythe linear regression of CMP’s returns on the mar-ket index, the actual performance of the linear clonefalls far short of the strategy because a linear modelwill never be able to capture the option-like payoffstructure of the perfect market-timer.
3 Linear regression analysis
To explore the full range of possibilities for repli-cating hedge-fund returns illustrated by the twoextremes of CDP and CMP, we investigated thecharacteristics of a sample of individual hedge fundsdrawn from the TASS Hedge Fund Database. The
database is divided into two parts: “Live” and“Graveyard” funds. Hedge funds that are in the“Live” database are considered to be active as of theend of our sample period, September 2005.7 Weconfine our attention to funds in the Live databasesince we wish to focus on the most current set ofrisk exposures in the hedge-fund industry, and weacknowledge that the Live database suffers fromsurvivorship bias.8
However, the importance of such a bias for ourapplication is tempered by two considerations.First, many successful funds leave the sample as wellas the poor performers, reducing the upward biasin expected returns. In particular, Fung and Hsieh(2000) estimated the magnitude of survivorshipbias to be 3.00% per year, and Liang’s (2000)estimate is 2.24% per year. Second, the focus of
SECOND QUARTER 2007 JOURNAL OF INVESTMENT MANAGEMENT
12 JASMINA HASANHODZIC AND ANDREW W. LO
our study is on the relative performance of hedgefunds versus relatively passive portfolios of liquidsecurities, and as long as our cloning process is notselectively applied to a peculiar subset of funds inthe TASS database, any survivorship bias shouldimpact both funds and clones identically, leavingtheir relative performance unaffected.
Of course, other biases plague the TASS databasesuch as backfill bias (including funds with returnhistories that start before the date of inclusion),9
selection bias (inclusion in the database is volun-tary, hence only funds seeking new investors areincluded),10 and other potential biases imparted bythe process by which TASS decides which funds toinclude and which to omit (part of this process isqualitative). As with survivorship bias, the hope isthat the impact of these additional biases will besimilar for clones and funds, leaving relative com-parisons unaffected. Unfortunately, there is little tobe done about such biases other than to acknowl-edge their existence, and to interpret the outcomeof our empirical analysis with an extra measure ofcaution.
Although theTASS Hedge Fund Live database startsin February 1977, we limit our analysis to the sam-ple period from February 1986 to September 2005because this is the timespan for which we havecomplete data for all of our risk factors. Of thesefunds, we drop those that: (i) do not report net-of-fee returns11; (ii) report returns in currenciesother than the US dollar; (iii) report returns lessfrequently than monthly; (iv) do not provide assetsunder management or only provide estimates; and(v) have fewer than 36 monthly returns. These filtersyield a final sample of 1610 funds.
3.1 Summary statistics
TASS classifies funds into one of the 11 differentinvestment styles, listed in Table 4 and described
in Appendix A, of which 10 correspond exactly tothe CSFB/Tremont sub-index definitions.12 Table 4also reports the number of funds in each categoryfor our sample, as well as summary statistics forthe individual funds and for the equal-weightedportfolio of funds in each of the categories. Thecategory counts show that the funds are not evenlydistributed across investment styles, but are con-centrated among five categories: Long/Short EquityHedge (520), Fund of Funds (355), Event Driven(169), Managed Futures (114), and Emerging Mar-kets (102). Together, these five categories accountfor 78% of the 1610 funds in our sample. Theperformance summary statistics in Table 4 under-score the reason for the growing interest in hedgefunds in recent years—double-digit cross-sectionalaverage returns for most categories with aver-age volatility lower than that of the S&P 500,implying average annualized Sharpe ratios rang-ing from a low of 0.25 for Dedicated Short Biasfunds to a high of 2.70 for Convertible Arbitragefunds.
Another feature of the data highlighted by Table 4is the large positive average return-autocorrelationsfor funds in Convertible Arbitrage (42.2%), Emerg-ing Markets (18.0%), Event Driven (22.2%),Fixed Income Arbitrage (22.1%), Multi-Strategy(21.0%), and Fund of Funds (23.2%) categories.Lo (2001) and Getmansky et al. (2004) haveshown that such high serial correlation in hedge-fund returns is likely to be an indication ofilliquidity exposure. There is, of course, nothinginappropriate about hedge funds taking on liq-uidity risk—indeed, this is a legitimate and oftenlucrative source of expected return—as long asinvestors are aware of such risks, and not mis-led by the siren call of attractive Sharpe ratios.13
But illiquidity exposure is typically accompaniedby capacity limits, and we shall return to thisissue when we compare the properties of hedgefunds to more liquid alternatives such as linearclones.
JOURNAL OF INVESTMENT MANAGEMENT SECOND QUARTER 2007
REPLICATING HEDGE-FUND RETURNS 13
Tab
le4
Sum
mar
yst
atis
tics
forT
ASS
Live
hedg
efu
nds
incl
uded
inou
rsa
mpl
efr
omFe
brua
ry19
86to
Sept
embe
r20
05.
Ann
ualiz
edpe
rfor
man
ceA
nnua
lized
Ann
ualiz
edSD
Ann
ualiz
edLj
ung-
Box
ofeq
ual-
wei
ghte
dm
ean
(%)
(%)
Shar
pera
tio
ρ1
(%)
p-va
lue
(%)
port
folio
offu
nds
Sam
ple
Cat
egor
ysi
zeM
ean
SDM
ean
SDM
ean
SDM
ean
SDM
ean
SDM
ean
(%)
SD(%
)Sh
arpe
Con
vert
ible
arbi
trag
e82
8.41
5.11
6.20
5.28
2.70
5.84
42.2
17.3
11.0
22.2
11.0
75.
362.
07D
edic
ated
shor
tbia
s10
5.98
4.77
28.2
710
.05
0.25
0.24
5.5
12.6
24.2
20.3
6.40
23.2
30.
28E
mer
ging
mar
kets
102
20.4
113
.01
22.9
215
.16
1.42
2.11
18.0
12.4
36.3
30.2
22.3
417
.71
1.26
Equ
ity
mar
ketn
eutr
al83
8.09
4.77
7.78
5.84
1.44
1.20
9.1
23.0
32.6
29.7
12.8
36.
232.
06Ev
entd
rive
n16
913
.03
8.65
8.40
8.09
1.99
1.37
22.2
17.6
27.0
29.3
13.4
74.
373.
08Fi
xed
inco
me
arbi
trag
e62
9.50
4.54
6.56
4.41
2.05
1.48
22.1
17.6
35.9
35.2
10.4
83.
582.
93G
loba
lmac
ro54
11.3
86.
1611
.93
6.10
1.07
0.58
5.8
12.2
43.1
32.5
14.9
18.
641.
73Lo
ng/S
hort
equi
tyhe
dge
520
14.5
98.
1415
.96
9.06
1.06
0.58
12.8
14.9
36.0
30.5
16.3
511
.84
1.38
Man
aged
futu
res
114
13.6
49.
3521
.46
12.0
70.
670.
392.
510
.240
.131
.515
.96
19.2
40.
83M
ulti
-str
ateg
y59
10.7
95.
228.
729.
701.
861.
0321
.020
.128
.230
.114
.59
5.78
2.52
Fund
offu
nds
355
8.25
3.73
6.36
4.47
1.66
0.86
23.2
15.0
27.1
26.3
11.9
37.
481.
59
SECOND QUARTER 2007 JOURNAL OF INVESTMENT MANAGEMENT
14 JASMINA HASANHODZIC AND ANDREW W. LO
3.2 Factor model specification
To determine the explanatory power of com-mon risk factors for hedge funds, we performa time-series regression for each of the 1610hedge funds in our sample, regressing the hedgefund’s monthly returns on the following six factors:(1) USD: the US Dollar Index return; (2) BOND:the return on the Lehman Corporate AA Inter-mediate Bond Index; (3) CREDIT: the spreadbetween the Lehman BAA Corporate Bond Indexand the Lehman Treasury Index; (4) SP500: theS&P 500 total return; (5) CMDTY: the GoldmanSachs Commodity Index (GSCI) total return; and(6) DVIX: the first-difference of the end-of-monthvalue of the CBOE Volatility Index (VIX).These sixfactors are selected for two reasons: They providea reasonably broad cross-section of risk exposuresfor the typical hedge fund (stocks, bonds, curren-cies, commodities, credit, and volatility), and eachof the factor returns can be realized through rel-atively liquid instruments so that the returns oflinear clones may be achievable in practice. In par-ticular, there are forward contracts for each of thecomponent currencies of the US Dollar index, andfutures contracts for the stock and bond indexesand for the components of the commodity index.Futures contracts on the VIX index were intro-duced by the CBOE in March 2004 and are notas liquid as the other index futures, but the OTCmarket for variance and volatility swaps is growingrapidly.
The linear regression model provides a simple butuseful decomposition of a hedge fund’s return Ritinto several components:
Rit = αi + βi1RiskFactor1t + · · ·+ βiK RiskFactorKt + εit (1)
From this decomposition, we have the followingcharacterization of the fund’s expected return and
variance:
E[Rit ] = αi + βi1E[RiskFactor1t ] + · · ·+ βiK E[RiskFactorKt ] (2)
Var[Rit ] = β2i1Var[RiskFactor1t ] + · · ·
+ β2iK Var[RiskFactorKt ]
+ Covariances + Var[εit ] (3)
where “Covariances” is the sum of all pairwisecovariances between RiskFactorpt and RiskFactorqtweighted by the product of their respective betacoefficients βipβiq .
This characterization implies that there are two dis-tinct sources of a hedge fund’s expected return: betaexposures βik multiplied by the risk premia asso-ciated with those exposures E[RiskFactorkt ], andmanager-specific alpha αi . By “manager-specific,”we do not mean to imply that a hedge fund’sunique source of alpha is without risk—we are sim-ply distinguishing this source of expected returnfrom those that have clearly identifiable risk fac-tors associated with them. In particular, it may wellbe the case that αi arises from risk factors otherthan the six we have proposed, and a more refinedversion of Eq. (1)—one that reflects the particu-lar investment style of the manager—may yield abetter-performing linear clone.
From Eq. (3), we see that a hedge fund’s vari-ance has three distinct sources: the variances of therisk factors multiplied by the squared beta coef-ficients, the variance of the residual εit (whichmay be related to the specific economic sources ofαi), and the weighted covariances among the fac-tors. This decomposition highlights the fact that ahedge fund can have several sources of risk, eachof which should yield some risk premium, that is,risk-based alpha, otherwise investors would not bewilling to bear such risk. By taking on exposureto multiple risk factors, a hedge fund can generate
JOURNAL OF INVESTMENT MANAGEMENT SECOND QUARTER 2007
REPLICATING HEDGE-FUND RETURNS 15
attractive expected returns from the investor’s per-spective (see, e.g. Capital Decimation Partners inSection 2.1).14
3.3 Factor exposures
Table 5 presents summary statistics for the beta coef-ficients or factor exposures in Eq. (1) estimatedfor each of the 1610 hedge funds by ordinaryleast squares and grouped by category. In partic-ular, for each category we report the minimum,median, mean, and maximum beta coefficient foreach of the six factors and the intercept, acrossall regressions in that category. For example, theupper left block of entries with the title “Inter-cept” presents summary statistics for the interceptsfrom the individual hedge-fund regressions withineach category, and the “Mean” column shows thatthe average manager-specific alpha is positive forall categories, ranging from 0.42% per monthfor Managed Futures funds to 1.41% per monthfor Emerging Markets funds. This suggests thatmanagers in our sample are, on average, indeed con-tributing value above and beyond the risk premiaassociated with the six factors we have chosen inEq. (1). We shall return to this important issue inSection 3.4.
The panel in Table 5 with the heading Rsp500 pro-vides summary statistics for the beta coefficientscorresponding to the S&P 500 return factor, and theentries in the “Mean” column are broadly consistentwith each of the category definitions. For example,funds in the Dedicated Short Bias category have anaverage S&P 500 beta of −0.88, which is consistentwith their shortselling mandate. On the other hand,Equity Market Neutral funds have an average S&P500 beta of 0.05, confirming their market neutralstatus. And Long/Short Equity Hedge funds, whichare mandated to provide partially hedged equity-market exposure, have an average S&P 500 betaof 0.38.
The remaining panels in Table 5 show that riskexposures do vary considerably across categories.This is more easily seen in Figure 2 which plots themean beta coefficients for all six factors, categoryby category. From Figure 2, we see that ConvertibleArbitrage funds have three main exposures (longcredit, long bonds, and long volatility), whereasEmerging Markets funds have four somewhat dif-ferent exposures (long stocks, short USD, longcredit, and long commodities). The category withthe smallest overall risk exposures is Equity MarketNeutral, and not surprisingly, this category exhibitsthe second lowest average mean return, 8.09%.
The lower right panel of Table 5 contains a sum-mary of the explanatory power of (1) as measured by
the R2
statistic of the regression (1). The mean R2s
range from a low of 10.4% for Equity Market Neu-tral (as expected, given this category’s small averagefactor exposures to all six factors) to a high of 40.4%for Dedicated Short Bias (which is also expectedgiven this category’s large negative exposure to theS&P 500).
To provide further intuition for the relation between
R2
and fund characteristics, we regress R2
on several
fund characteristics, and find that lower R2
fundsare those with higher Sharpe ratios, higher manage-ment fees, and higher incentive fees. This accordswell with the intuition that funds providing greater
diversification benefits, that is, lower R2’s, com-
mand higher fees in equilibrium. See Hasanhodzicand Lo (2006b) for further details.
3.4 Expected-return decomposition
Using the parameter estimates of (1) for the indi-vidual hedge funds in our sample, we can nowreformulate the question of whether or not a hedge-fund strategy can be cloned as a question abouthow much of a hedge fund’s expected return is due
SECOND QUARTER 2007 JOURNAL OF INVESTMENT MANAGEMENT
16 JASMINA HASANHODZIC AND ANDREW W. LO
Tab
le5
Sum
mar
yst
atis
tics
for
mul
tiva
riat
elin
ear
regr
essi
ons
ofm
onth
lyre
turn
sof
hedg
efu
nds
inth
eT
ASS
Live
data
base
from
Febr
uary
1986
toSe
ptem
ber
2005
onsi
xfa
ctor
s:th
eS&
P50
0to
talr
etur
n,th
eLe
hman
Cor
pora
teA
AIn
term
edia
teB
ond
Inde
xre
turn
,the
US
Dol
lar
Inde
xre
turn
,the
spre
adbe
twee
nth
eLe
hman
US
Agg
rega
teLo
ngC
redi
tB
AA
Bon
dIn
dex
and
the
Lehm
anTr
easu
ryLo
ngIn
dex,
the
first
-diff
eren
ceof
the
CB
OE
Vol
atili
tyIn
dex
(VIX
),an
dth
eG
oldm
anSa
chs
Com
mod
ity
Inde
x(G
SCI)
tota
lret
urn.
Inte
rcep
tR
sp50
0R
lbR
usd
Sam
ple
Cat
egor
ysi
zeSt
atis
tic
Min
Med
Mea
nM
axSD
Min
Med
Mea
nM
axSD
Min
Med
Mea
nM
axSD
Min
Med
Mea
nM
axSD
Con
vert
ible
arbi
trag
e82
beta
−0.5
20.
410.
431.
570.
37−0
.63
−0.0
1−0
.02
0.45
0.15
−0.0
80.
260.
301.
730.
29−0
.98
0.01
−0.0
20.
680.
28t-
stat
−1.5
62.
124.
5583
.10
11.3
5−2
.58
−0.1
40.
067.
651.
53−0
.52
1.60
1.60
4.50
1.12
−2.2
30.
150.
122.
911.
22D
edic
ated
shor
tbia
s10
beta
−0.0
40.
770.
671.
130.
38−1
.78
−1.0
1−0
.88
−0.1
10.
50−0
.60
0.18
0.25
0.96
0.48
−0.0
80.
730.
671.
250.
51t-
stat
−0.1
20.
730.
911.
830.
66−1
0.95
−3.2
9−3
.88
−0.4
82.
72−1
.37
0.24
0.17
1.05
0.70
−0.1
91.
261.
071.
990.
77E
mer
ging
mar
kets
102
beta
−0.7
51.
191.
416.
501.
08−0
.41
0.31
0.43
3.30
0.52
−4.5
30.
020.
012.
330.
77−4
.66
−0.3
9−0
.42
2.18
0.79
t-st
at−1
.03
1.83
2.74
44.6
74.
57−1
.77
1.69
1.65
5.46
1.61
−2.1
70.
090.
223.
711.
09−3
.74
−1.0
3−0
.97
2.53
1.20
Equ
ity
mar
ketn
eutr
al83
beta
−0.6
10.
590.
592.
420.
41−1
.22
0.05
0.05
0.90
0.27
−1.1
60.
050.
020.
820.
33−2
.83
0.02
−0.0
41.
240.
44t-
stat
−1.4
02.
022.
8813
.89
3.00
−4.8
60.
750.
654.
161.
98−3
.74
0.30
0.27
2.67
1.09
−4.1
70.
080.
163.
651.
39Ev
entd
rive
n16
9be
ta−0
.12
0.78
0.93
6.18
0.78
−0.3
50.
080.
131.
170.
22−4
.23
0.08
0.04
1.31
0.46
−6.3
8−0
.05
−0.1
31.
460.
60t-
stat
−0.6
93.
383.
8821
.54
2.89
−2.8
01.
261.
3410
.87
1.88
−2.3
10.
400.
423.
211.
08−2
.86
−0.3
1−0
.14
3.40
1.35
Fixe
din
com
ear
bitr
age
62be
ta0.
000.
520.
582.
030.
42−0
.39
0.03
0.02
0.23
0.10
−0.5
50.
200.
271.
860.
40−0
.66
0.05
0.07
0.77
0.35
t-st
at0.
002.
853.
8524
.30
3.91
−2.4
20.
550.
443.
231.
25−2
.63
1.00
1.26
11.0
21.
99−3
.48
0.38
0.66
4.62
1.68
Glo
balm
acro
54be
ta−0
.79
0.63
0.59
1.75
0.54
−0.4
90.
010.
101.
140.
30−0
.74
0.21
0.34
2.03
0.56
−2.0
0−0
.23
−0.2
31.
350.
67t-
stat
−1.5
61.
531.
717.
661.
62−2
.97
0.19
0.59
6.16
1.84
−1.9
30.
710.
926.
051.
51−6
.51
−0.8
3−0
.73
4.52
1.95
Long
/Sho
rteq
uity
hedg
e52
0be
ta−1
.53
0.84
0.89
7.60
0.75
−1.3
70.
330.
383.
130.
44−3
.04
−0.0
10.
033.
490.
59−2
.57
−0.0
3−0
.09
2.45
0.60
t-st
at−1
.80
1.84
1.86
10.4
71.
38−3
.72
2.06
2.27
20.0
72.
50−3
.47
−0.0
10.
063.
331.
06−4
.60
−0.1
0−0
.19
3.41
1.18
Man
aged
futu
res
114
beta
−1.8
40.
480.
423.
690.
73−0
.81
−0.0
10.
032.
300.
37−0
.44
0.88
0.89
2.62
0.67
−2.6
5−0
.37
−0.3
91.
140.
63t-
stat
−2.3
60.
720.
654.
981.
08−2
.94
−0.0
50.
207.
881.
43−1
.70
1.46
1.60
4.34
1.22
−4.2
5−0
.83
−0.7
21.
990.
98M
ulti
-str
ateg
y59
beta
−0.4
10.
710.
712.
680.
47−0
.31
0.07
0.15
1.34
0.26
−1.8
10.
100.
122.
400.
51−1
.84
0.07
0.01
0.78
0.41
t-st
at−0
.43
3.22
3.41
10.5
12.
41−2
.22
1.27
1.37
5.98
1.68
−1.4
90.
580.
573.
491.
13−2
.78
0.36
0.39
3.19
1.34
Fund
offu
nds
355
beta
−0.7
70.
420.
431.
880.
34−0
.80
0.09
0.12
0.85
0.15
−0.5
00.
120.
182.
250.
29−1
.12
−0.0
7−0
.10
0.62
0.24
t-st
at−3
.55
2.34
2.67
10.5
12.
14−2
.65
1.56
1.84
9.44
1.80
−1.5
90.
830.
954.
841.
17−3
.63
−0.5
3−0
.42
3.32
1.28
JOURNAL OF INVESTMENT MANAGEMENT SECOND QUARTER 2007
REPLICATING HEDGE-FUND RETURNS 17
Tab
le5
(Con
tinue
d)
Rcs
�V
IXR
gsci
Sign
ifica
nce
(%)
Sam
ple
Cat
egor
ysi
zeSt
atis
tic
Min
Med
Mea
nM
axSD
Min
Med
Mea
nM
axSD
Min
Med
Mea
nM
axSD
Stat
isti
cM
inM
edM
ean
Max
SD
Con
vert
ible
arbi
trag
e82
beta
0.00
0.39
0.52
2.87
0.57
−0.2
50.
050.
050.
320.
08−0
.07
0.01
0.02
0.16
0.03
Adj
.R2
−11.
016
.017
.366
.215
.4
t-st
at0.
193.
062.
957.
721.
58−1
.41
0.50
0.66
3.56
0.98
−1.1
50.
520.
512.
170.
69p(
F)0.
01.
011
.897
.123
.6
Ded
icat
edsh
ortb
ias
10be
ta−0
.98
−0.2
6−0
.19
0.93
0.67
−0.2
60.
050.
040.
440.
23−0
.38
−0.1
1−0
.12
0.06
0.13
Adj
.R2
−3.5
39.7
40.4
79.5
25.4
t-st
at−2
.67
−0.6
8−0
.44
2.54
1.64
−1.1
10.
240.
232.
561.
10−2
.19
−0.8
6−0
.95
0.54
0.92
p(F)
0.0
0.0
8.3
83.0
26.2
Em
ergi
ngm
arke
ts10
2be
ta−0
.56
0.46
0.59
2.89
0.67
−1.4
1−0
.05
0.01
3.91
0.50
−0.3
40.
050.
060.
340.
09A
dj.R
2−4
.717
.419
.454
.714
.3
t-st
at−1
.97
1.32
1.33
4.82
1.36
−3.9
5−0
.35
−0.2
83.
881.
17−1
.46
0.68
0.60
2.40
0.79
p(F)
0.0
0.2
8.4
78.8
17.7
Equ
ity
mar
ketn
eutr
al83
beta
−1.7
8−0
.03
−0.0
60.
720.
31−1
.19
0.02
0.03
0.80
0.23
−0.1
20.
010.
020.
380.
07A
dj.R
2−8
.17.
210
.463
.213
.7
t-st
at−3
.83
−0.2
7−0
.35
3.34
1.44
−3.1
00.
220.
253.
951.
23−2
.05
0.48
0.43
2.80
1.11
p(F)
0.0
7.4
19.9
94.1
24.6
Even
tdri
ven
169
beta
−1.9
60.
250.
332.
010.
45−1
.81
0.02
0.05
1.19
0.26
−0.2
70.
010.
010.
270.
06A
dj.R
2−7
.515
.519
.568
.516
.4
t-st
at−1
.66
1.51
1.81
8.31
1.99
−2.7
60.
420.
364.
581.
17−2
.27
0.50
0.60
4.06
1.15
p(F)
0.0
0.3
11.1
88.6
20.0
Fixe
din
com
ear
bitr
age
62be
ta−0
.70
0.10
0.19
1.54
0.46
−0.7
10.
050.
070.
500.
18−0
.06
0.01
0.02
0.15
0.05
Adj
.R2
−8.9
12.8
14.9
78.9
15.9
t-st
at−3
.29
0.80
1.25
11.7
42.
56−3
.16
0.85
1.16
5.62
1.93
−1.7
60.
570.
522.
521.
10p(
F)0.
02.
117
.794
.626
.3
Glo
balm
acro
54be
ta−0
.61
0.13
0.18
1.73
0.42
−0.3
60.
030.
070.
550.
19−0
.09
0.02
0.04
0.27
0.08
Adj
.R2
−12.
68.
914
.874
.017
.3
t-st
at−1
.60
0.44
0.60
3.96
1.25
−3.0
80.
330.
343.
611.
11−1
.22
0.37
0.60
3.92
1.20
p(F)
0.0
4.9
16.8
97.0
24.3
Long
/Sho
rteq
uity
hedg
e52
0be
ta−1
.37
0.17
0.28
4.55
0.59
−1.6
70.
070.
072.
760.
33−0
.33
0.04
0.06
0.88
0.11
Adj
.R2
−13.
818
.821
.690
.219
.0
t-st
at−5
.28
0.58
0.69
4.94
1.36
−4.7
00.
460.
383.
671.
28−3
.31
0.74
0.77
5.91
1.13
p(F)
0.0
0.4
11.8
97.7
22.9
Man
aged
futu
res
114
beta
−5.9
8−0
.33
−0.3
53.
200.
82−0
.75
0.14
0.15
1.29
0.32
−0.3
10.
110.
130.
800.
15A
dj.R
2−6
.013
.315
.370
.013
.3
t-st
at−2
.85
−0.9
2−0
.73
2.56
1.04
−2.8
10.
730.
744.
361.
28−2
.15
1.32
1.36
5.25
1.22
p(F)
0.0
0.6
8.2
88.5
17.0
Mul
ti-s
trat
egy
59be
ta−0
.48
0.07
0.17
1.64
0.41
−0.3
80.
040.
090.
950.
19−0
.05
0.03
0.04
0.75
0.11
Adj
.R2
−13.
58.
912
.951
.715
.7
t-st
at−2
.20
0.72
1.21
6.34
2.12
−1.5
90.
680.
873.
721.
31−1
.34
0.87
0.81
2.90
0.97
p(F)
0.0
6.7
21.7
97.5
28.9
Fund
offu
nds
355
beta
−0.7
80.
170.
171.
410.
22−0
.32
0.06
0.07
0.48
0.09
−0.2
30.
030.
050.
350.
05A
dj.R
2−7
.220
.422
.372
.314
.9
t-st
at−3
.62
1.38
1.53
6.35
1.55
−2.7
40.
980.
984.
691.
12−3
.16
1.38
1.39
4.28
1.01
p(F)
0.0
0.2
5.7
84.0
14.3
SECOND QUARTER 2007 JOURNAL OF INVESTMENT MANAGEMENT
18 JASMINA HASANHODZIC AND ANDREW W. LO
-1.00
-0.50
0.00
0.50
1.00
1.50
2.00
ConvertibleArbitrage
DedicatedShort Bias
EmergingMarkets
Equity MarketNeutral
Event Driven Fixed IncomeArbitrage
Global Macro Long/ShortEquity
ManagedFutures
Multi-Strategy Fund ofFunds
Ave
rag
e R
egre
ssio
n C
oef
fici
ent
Manager-Specific Alpha S&P 500 BondsUSD Credit Spread DVIXCommodities
Figure 2 Average regression coefficients for multivariate linear regressions of monthly returns of hedgefunds in the TASS Live database from February 1986 to September 2005 on six factors: the S&P 500total return, the Lehman Corporate AA Intermediate Bond Index return, the US Dollar Index return, thespread between the Lehman US Aggregate Long Credit BAA Bond Index and the Lehman Treasury LongIndex, the first-difference of the CBOE Volatility Index (VIX), and the Goldman Sachs Commodity Index(GSCI) total return.
to risk premia from identifiable factors. If it is asignificant portion and the relationship is primar-ily linear, then a passive portfolio with just thoserisk exposures—created by means of liquid instru-ments such as index futures, forwards, and othermarketable securities—may be a reasonable alter-native to a less liquid and opaque investment inthe fund.
Table 6 summarizes the results of the expected-return decomposition (2) for our sample of 1610funds, grouped according to their style categoriesand for all funds. Each row of Table 6 containsthe average total mean return of funds in a given
category and averages of the percent contributionsof each of the six factors and the manager-specificalpha to that average total mean return.15 Notethat the average percentage contributions add upto 100% when summed across all six factors andthe manager-specific alpha because this decompo-sition sums to 100% for each fund, and when thisdecomposition is averaged across all funds, the sumis preserved.
The first row’s entries indicate that the most signifi-cant contributors to the average total mean return of8.4% for Convertible Arbitrage funds are CREDIT(27.1%), USD (67.1%), BOND (34.9%), and
JOURNAL OF INVESTMENT MANAGEMENT SECOND QUARTER 2007
REPLICATING HEDGE-FUND RETURNS 19
Table 6 Decomposition of total mean returns of hedge funds in the TASS Live database according topercentage contributions from six factors and manager-specific alpha, for 1610 hedge funds from February1986 to September 2005.
Average of percentage contribution of factors tototal expected return (%)
Category Sampledescription size Avg. E [R] CREDIT USD SP500 BOND DVIX CMDTY ALPHA
Convertible arbitrage 82 8.4 27.1 67.1 −19.3 34.9 −8.4 31.8 −33.3Dedicated short bias 10 6.0 12.2 19.4 −108.2 7.0 8.9 −64.9 225.6Emerging markets 102 20.4 −0.3 −3.2 19.3 0.1 −0.4 6.2 78.3Equity market neutral 83 8.1 0.2 3.6 4.0 3.9 1.3 6.3 80.8Event driven 169 13.0 2.1 3.0 4.3 9.4 −0.7 3.1 79.0Fixed income arbitrage 62 9.5 −1.4 3.3 2.7 18.5 −0.5 4.4 73.1Global macro 54 11.4 2.0 8.1 9.7 25.0 −3.3 10.0 48.6Long/Short equity hedge 520 14.6 1.1 1.9 17.8 2.1 −1.8 8.4 70.5Managed futures 114 13.6 1.9 23.4 −3.4 53.8 −1.5 53.2 −27.5Multi-strategy 59 10.8 0.5 3.5 5.7 10.1 −1.9 3.2 78.9Fund of funds 355 8.3 0.5 5.4 9.7 8.8 −2.8 7.3 71.1All Funds 1,610 11.3 2.3 7.8 8.5 11.3 −1.9 10.9 61.0
CMDTY (31.8%), and the average contributionof manager-specific alpha is −33.3%. This impliesthat on average, Convertible Arbitrage funds earnmore than all of their mean returns from the riskpremia associated with the six factor exposures, andthat the average contribution of other sources ofalpha is negative! Of course, this does not meanthat convertible-arbitrage managers are not addingvalue—Table 6’s results are averages across all fundsin our sample, hence the positive manager-specificalphas of successful managers will be dampenedand, in some cases, outweighed by the negativemanager-specific alphas of the unsuccessful ones.
In contrast to the Convertible Arbitrage funds,for the 10 funds in the Dedicated Short Bias cat-egory, the manager-specific alpha accounts for225.6% of the funds’ total mean return on average,and the contribution of the SP500 factor is neg-ative. This result is not as anomalous as it seems.The bull market of the 1990s implies a perfor-mance drag for any fund with negative exposureto the S&P 500, therefore, Dedicated Short Biasmanagers that have generated positive performance
during this period must have done so through othermeans. A concrete illustration of this intuition isgiven by the return decomposition of the annual-ized average return of the two most successful fundsin the Dedicated Short Bias category, funds 33735and 33736. These two funds posted annualized net-of-fee returns of 15.56% and 10.02%, respectively,but the contribution of the SP500 factor to theseannualized returns was negative in both cases (bothfunds had negative SP500 beta exposures, as theyshould, and the S&P 500 yielded positive returnsover the funds’ lifetimes).16
Between the two extremes of Convertible Arbi-trage and Dedicated Short Bias funds, Table 6shows considerable variation in the importanceof manager-specific alpha for the other categories.Over 80% of the average total return of EquityMarket Neutral funds is due to manager-specificalpha, but for Managed Futures, the manager-specific alpha accounts for −27.5%. For the entiresample of 1610 funds, 61% of the average totalreturn is attributable to manager-specific alpha,implying that on average, the remaining 39% is
SECOND QUARTER 2007 JOURNAL OF INVESTMENT MANAGEMENT
20 JASMINA HASANHODZIC AND ANDREW W. LO
due to the risk premia from our six factors. Theseresults suggest that for certain types of hedge-fundstrategies, a multi-factor portfolio may yield someof the same benefits but in a transparent, scalable,and lower-cost vehicle.
4 Linear clones
The multivariate regression results in Section 3 sug-gest that linear clones may be able to replicate someof the risk exposures of hedge funds, and in thissection we investigate this possibility directly byconsidering two types of clones. The first type con-sists of “fixed-weight” portfolios, where we use theentire sample of a given fund’s returns to estimatea set of portfolio weights for the instruments corre-sponding to the factors used in the linear regression.These portfolio weights are fixed through time foreach fund, hence the term “fixed-weight.”17 Butbecause this approach involves a certain degree of“look-ahead” bias—we use the entire history of afund’s returns to construct the portfolio weights thatare applied each period to compute the clone port-folio return—we also construct a second type oflinear clone based on rolling-window regressions.
4.1 Fixed-weight vs. rolling-window clones
To construct a fixed-weight linear clone for fund i,we begin by regressing the fund’s returns {Rit } onfive of the six factors we considered in Section 3(we drop the DVIX factor because its returns arenot as easily realized with liquid instruments),where we omit the intercept and constrain the betacoefficients to sum to one:
Rit = βi1SP500t + βi2BONDt
+ βi3USDt + βi4CREDITt
+ βi5CMDTYt + εit ,
t = 1, . . . , T (4a)
subject to 1 = βi1 + · · · + βi5 (4b)
This is the same technique proposed by Sharpe(1992) for conducting “style analysis,” however, ourmotivation is quite different. We omit the inter-cept because our objective is to estimate a weightedaverage of the factors that best replicates the fund’sreturns, and omitting the constant term forces theleast-squares algorithm to use the factor means tofit the mean of the fund, an important feature ofreplicating hedge-fund expected returns with factorrisk premia. And we constrain the beta coefficientsto sum to one to yield a portfolio interpretationfor the weights. Note that we do not constrain theregression coefficients to be non-negative as Sharpe(1992) does because unlike Sharpe’s original appli-cation to long-only mutual funds, in our context allfive factors correspond to instruments that can beshortsold, and we do expect to be shortselling eachof these instruments on occasion to achieve the kindof risk exposures hedge funds typically exhibit. Forexample, clones of Dedicated Short Bias funds willundoubtedly require shorting the SP500 factor.
The estimated regression coefficients {β∗ik} are then
used as portfolio weights for the five factors, hencethe portfolio returns are equivalent to the fittedvalues R∗
it of the regression equation. However,we implement an additional renormalization sothat the resulting portfolio return R̂it has the samesample volatility as the original fund’s return series:
R∗it ≡ β∗
i1SP500t + β∗i2BONDt + β∗
i3USDt
+ β∗i4CREDITt + β∗
i5CMDTYt (5)
R̂it ≡ γi R∗it , γi ≡
√∑Tt=1 (Rit − Ri)2/(T −1)√∑Tt=1 (R∗
it − R∗i )2/(T −1)
(6)
Ri ≡ 1
T
T∑t=1
Rit , R∗i ≡ 1
T
T∑t=1
R∗it (7)
The motivation for this renormalization is to createa fair comparison between the clone portfolio and
JOURNAL OF INVESTMENT MANAGEMENT SECOND QUARTER 2007
REPLICATING HEDGE-FUND RETURNS 21
the fund by equalizing their volatilities. Renormal-izing (5) is equivalent to changing the leverage ofthe clone portfolio, since the sum of the renormal-ized betas γi
∑k β∗
ik will equal the renormalizationfactor γi , not one. If γi exceeds one, then pos-itive leverage is required, and if less than one,the portfolio is not fully invested in the five fac-tors. A more complete expression of the portfolioweights of clone i may be obtained by introducingan additional asset that represents leverage, that is,borrowing and lending, in which case the portfolioweights of the five factors and this additional assetmust sum to one:
1 = γi(β∗i1 + · · · + β∗
i5) + δi (8)
The clone return is then given by:
R̂it = γi(β∗i1 SP500t + · · ·
+ β∗i5 CMDTYt ) + δi Rl (9)
where Rl is the borrowing/lending rate. Since thisrate depends on many factors such as the credit qual-ity of the respective counterparties, the riskiness ofthe instruments and portfolio strategy, the size ofthe transaction, and general market conditions, wedo not attempt to assume a particular value for Rlbut simply point out its existence.18
As discussed above, fixed-weight linear clones areaffected by look-ahead bias because the entire histo-ries of fund and factor returns are used to constructthe clones’ portfolio weights and renormalizationfactors. To address this issue, we present an alter-nate method of constructing linear clones using arolling window for estimating the regression coeffi-cients and renormalization factors. Rolling-windowestimators can also address the ubiquitous issueof nonstationarity that affects most financial time-series studies; time-varying means, volatilities, andgeneral market conditions can be captured to somedegree by using rolling windows.
To construct a “rolling-window” linear clone, foreach month t , we use a 24-month rolling window
from months t −24 to t −1 to estimate the sameregression (4) as before19:
Rit−k = βit1SP500t−k + βit2BONDt−k
+ βit3USDt−k + βit4CREDITt−k
+ βit5CMDTYt−k + εit−k ,
k = 1, . . . , 24 (10a)
subject to 1 = βit1 + · · · + βit5 (10b)
but now the coefficients are indexed by both i andt since we repeat this process each month for everyfund i. The parameter estimates are then used in thesame manner as in the fixed-weight case to constructclone returns R̂it :
R∗it ≡ β∗
it1SP500t + β∗it2BONDt + β∗
it3USDt
+ β∗it4CREDITt + β∗
it5CMDTYt (11)
R̂it ≡ γit R∗it , γit ≡
√∑24k=1 (Rit−k − Rit )2/23√∑24k=1 (R∗
it−k − R∗it )2/23
(12)
Rit ≡ 1
24
24∑k=1
Rit−k , R∗it ≡ 1
24
24∑k=1
R∗it−k (13)
where the renormalization factors γit are nowindexed by time t to reflect the fact that they arealso computed within the rolling window. Thisimplies that for any given clone i, the volatilityof its returns over the entire history will no longerbe identical to the volatility of its matching fundbecause the renormalization process is applied onlyto rolling windows, not to the entire history ofreturns. However, as long as volatilities do not shiftdramatically over time, the rolling-window renor-malization process should yield clones with similarvolatilities.
Although rolling-window clones may seem morepractically relevant because they avoid the mostobvious forms of look-ahead bias, they have
SECOND QUARTER 2007 JOURNAL OF INVESTMENT MANAGEMENT
22 JASMINA HASANHODZIC AND ANDREW W. LO
drawbacks as well. For example, the rolling-window estimation procedure generates more fre-quent rebalancing needs for the clone portfolio,which is counter to the passive spirit of the cloningendeavor. Moreover, rolling-window estimators aretypically subject to greater estimation error becauseof the smaller sample size. This implies that at leastpart of the rebalancing of rolling-window clonesis unnecessary. The amount of rebalancing can,of course, be controlled by adjusting the lengthof the rolling window—a longer window impliesmore stable weights, but stability implies less flex-ibility in capturing potential nonstationarities inthe data.
Ultimately, the choice between fixed-weight androlling-window clones depends on the nature ofthe application, the time-series properties of thestrategies being cloned, and the specific goals andconstraints of the investor. A passive investor withlittle expertise in trading and risk management maywell prefer the fixed-weight clone, whereas a moreactive investor with trading capabilities and a desireto implement dynamic asset-allocation policies willprefer the rolling-window clone. For these rea-sons, we present results for both types of clonesin Sections 4.2–4.5.
4.2 Performance results
Table 7 contains a comparison between the perfor-mance of fixed-weight and rolling-window linearclones and the original funds from which the clonesare derived.20 The results are striking—for severalcategories, the average mean return of the clones isonly slightly lower than that of their fund counter-parts, and in some categories, the clones do better.For example, the average mean return of the Con-vertible Arbitrage fixed-weight clones is 7.40%, andthe corresponding figure for the funds is 8.41%.For Long/Short Equity Hedge funds, the aver-age mean return for fixed-weight clones and funds
is 13.12% and 14.59%, respectively. And in theMulti-Strategy category, the average mean returnfor fixed-weight clones and funds is 10.32% and10.79%, respectively.
In five cases, the average mean return of thefixed-weight clones is higher than that of thefunds: Dedicated Short Bias (6.70% vs. 5.98%),Equity Market Neutral (10.00% vs. 8.09%), GlobalMacro (15.54% vs. 11.38%), Managed Futures(27.97% vs. 13.64%), and Fund of Funds (9.29%vs. 8.25%). However, these differences are notnecessarily statistically significant because of thevariability in mean returns of funds and cloneswithin their own categories. Even in the case ofManaged Futures, the difference in average meanreturn between fixed-weight clones and funds—almost 15 percentage points—is not significantbecause of the large fluctuations in average meanreturns of the Managed Futures fixed-weight clonesand their corresponding funds (e.g. one standarddeviation of the average mean of the ManagedFutures fixed-weight clones is 16.32% and onestandard deviation of the average mean of the cor-responding sample of funds is 9.35%, accordingto Table 7). Nevertheless, these results suggest thatfor certain categories, the performance of fixed-weight clones may be comparable to that of theircorresponding funds.
On the other hand, at 9.84%, the average perfor-mance of the Event-Driven fixed-weight clones isconsiderably lower than the 13.03% average forthe funds. While also not statistically significant,this gap is understandable given the idiosyncraticand opportunistic nature of most event-drivenstrategies. Moreover, a significant source of the prof-itability of event-driven strategies is the illiquiditypremium that managers earn by providing capitalin times of distress. This illiquidity premium willclearly be missing from a clone portfolio of liquidsecurities, hence we should expect a sizable perfor-mance gap in this case. The same can be said for the
JOURNAL OF INVESTMENT MANAGEMENT SECOND QUARTER 2007
REPLICATING HEDGE-FUND RETURNS 23
Tab
le7
Perf
orm
ance
com
pari
son
offix
ed-w
eigh
tan
dro
lling
-win
dow
linea
rcl
ones
ofhe
dge
fund
sin
the
TA
SSLi
veda
taba
sean
dth
eir
corr
espo
ndin
gfu
nds,
from
Febr
uary
1986
toSe
ptem
ber
2005
.The
cate
gory
“Ded
icat
edSh
ortB
ias*
”ex
clud
esFu
nd33
735.
Fixe
d-w
eigh
tlin
ear
clon
es24
-mon
thro
lling
-win
dow
linea
rcl
ones
Ann
ualm
ean
Ann
ualS
D(%
)A
nnua
lSha
rpe
ρ1(%
)p-
valu
e(Q
12)
Ann
ualm
ean
Ann
ualS
D(%
)A
nnua
lSha
rpe
ρ1(%
)p-
valu
e(Q
6)
(%)
retu
rn(%
)(%
)re
turn
(%)
Cat
egor
ySa
mpl
ede
scri
ptio
nsi
zeM
ean
SDM
ean
SDM
ean
SDM
ean
SDM
ean
SDM
ean
SDM
ean
SDM
ean
SDM
ean
SDM
ean
SD
Fund
sC
onve
rtib
lear
bitr
age
828.
415.
116.
205.
282.
705.
8442
.217
.311
.022
.24.
047.
835.
764.
552.
318.
9642
.316
.212
.419
.4D
edic
ated
shor
tbia
s10
5.98
4.77
28.2
710
.05
0.25
0.24
5.5
12.6
24.2
20.3
2.58
7.19
25.9
114
.20
0.02
0.42
8.3
5.5
31.9
19.0
Ded
icat
edsh
ortb
ias*
94.
923.
5828
.75
10.5
30.
200.
203.
411
.325
.521
.11.
426.
5526
.21
15.0
3−0
.04
0.39
7.4
4.9
35.2
16.8
Em
ergi
ngm
arke
ts10
220
.41
13.0
122
.92
15.1
61.
422.
1118
.012
.436
.330
.221
.12
13.8
619
.95
14.0
61.
742.
5716
.014
.339
.228
.1E
quit
ym
arke
tneu
tral
838.
094.
777.
785.
841.
441.
209.
123
.032
.629
.75.
714.
146.
605.
911.
441.
685.
324
.040
.233
.9Ev
entd
rive
n16
913
.03
8.65
8.40
8.09
1.99
1.37
22.2
17.6
27.0
29.3
11.6
510
.45
7.62
7.68
2.01
1.43
17.2
17.8
31.1
29.8
Fixe
din
com
ear
bitr
age
629.
504.
546.
564.
412.
051.
4822
.117
.635
.935
.27.
807.
595.
734.
522.
171.
8123
.321
.430
.132
.4G
loba
lmac
ro54
11.3
86.
1611
.93
6.10
1.07
0.58
5.8
12.2
43.1
32.5
9.01
6.72
11.1
66.
500.
910.
736.
618
.944
.731
.2Lo
ng/S
hort
equi
tyhe
dge
520
14.5
98.
1415
.96
9.06
1.06
0.58
12.8
14.9
36.0
30.5
11.9
08.
9313
.90
8.69
1.04
0.77
9.8
16.7
42.0
28.5
Man
aged
futu
res
114
13.6
49.
3521
.46
12.0
70.
670.
392.
510
.240
.131
.511
.84
8.82
20.1
910
.94
0.66
0.52
4.0
14.9
37.0
28.3
Mul
ti-s
trat
egy
5910
.79
5.22
8.72
9.70
1.86
1.03
21.0
20.1
28.2
30.1
8.97
6.13
7.65
10.1
01.
861.
2518
.322
.529
.128
.6Fu
ndof
fund
s35
58.
253.
736.
364.
471.
660.
8623
.215
.027
.126
.37.
343.
955.
684.
291.
670.
9722
.616
.324
.026
.5A
llex
cept
fund
offu
nds
1255
13.2
98.
7113
.95
10.7
61.
391.
8815
.718
.333
.230
.911
.15
9.86
12.3
810
.12
1.38
2.64
13.5
19.8
36.6
29.8
Line
arcl
ones
Con
vert
ible
arbi
trag
e82
7.40
3.17
6.20
5.28
1.52
0.62
10.7
10.5
55.6
24.9
2.78
4.95
6.20
6.57
0.71
0.77
6.4
12.7
43.8
29.1
Ded
icat
edsh
ortb
ias
106.
7011
.59
28.2
710
.05
0.32
0.48
2.8
5.9
73.6
17.5
6.83
16.1
829
.31
15.6
10.
090.
450.
48.
836
.728
.7D
edic
ated
shor
tbia
s*9
3.61
6.61
28.7
510
.53
0.19
0.29
3.7
5.6
77.3
14.0
9.08
15.4
130
.00
16.3
90.
170.
40−0
.78.
536
.830
.4E
mer
ging
mar
kets
102
14.7
711
.47
22.9
215
.16
0.88
0.58
0.0
9.0
62.7
27.6
5.17
14.7
025
.04
17.9
40.
470.
667.
712
.442
.527
.3E
quit
ym
arke
tneu
tral
8310
.00
7.00
7.78
5.84
1.42
0.58
1.8
9.6
57.0
24.7
4.43
4.90
7.91
6.49
0.64
0.68
4.2
12.7
47.8
27.0
Even
tdri
ven
169
9.84
6.69
8.40
8.09
1.43
0.52
4.3
11.0
55.8
24.1
6.96
8.33
7.79
7.10
1.05
0.56
3.0
13.3
39.6
27.3
Fixe
din
com
ear
bitr
age
628.
355.
206.
564.
411.
480.
594.
18.
464
.429
.84.
474.
636.
855.
170.
840.
714.
39.
940
.830
.0G
loba
lmac
ro54
15.5
48.
3511
.93
6.10
1.41
0.55
2.6
8.3
52.2
23.8
12.9
78.
9012
.48
7.38
1.08
0.59
4.1
11.1
45.3
28.2
Long
/Sho
rteq
uity
hedg
e52
013
.12
8.68
15.9
69.
060.
980.
57−0
.110
.059
.726
.39.
0811
.03
15.8
310
.64
0.76
0.68
0.3
15.6
42.5
29.1
Man
aged
futu
res
114
27.9
716
.32
21.4
612
.07
1.36
0.40
4.7
8.1
61.4
29.0
19.2
413
.32
22.9
613
.71
0.91
0.57
5.5
10.9
46.5
27.7
Mul
ti-s
trat
egy
5910
.32
7.21
8.72
9.70
1.51
0.62
2.3
10.1
59.7
28.8
5.33
7.52
9.16
9.59
0.71
0.60
0.8
13.0
35.9
28.2
Fund
offu
nds
355
9.29
5.62
6.36
4.47
1.59
0.44
−0.1
11.1
50.5
24.7
5.67
4.57
6.22
5.40
1.11
0.54
0.0
13.0
39.8
28.5
All
exce
ptfu
ndof
fund
s12
5513
.27
10.4
813
.95
10.7
61.
190.
612.
210
.259
.226
.48.
4211
.06
14.2
012
.14
0.79
0.67
2.8
13.9
42.6
28.4
SECOND QUARTER 2007 JOURNAL OF INVESTMENT MANAGEMENT
24 JASMINA HASANHODZIC AND ANDREW W. LO
Emerging Markets fixed-weight clones (14.77%)versus their fund counterparts (20.41%).
For Dedicated Short Bias funds, the difference inaverage mean return between fixed-weight clonesand funds—6.70% and 5.98%, respectively—mayseem somewhat counterintuitive in the light of theexpected-return decomposition in Table 6, wherewe observed that Dedicated Short Bias funds wereresponsible for more than 100% of the averagetotal returns of funds in this category. The factthat Dedicated Short Bias clones have better aver-age performance than the corresponding funds isdue entirely to the clone of a single fund, 33735,and when this outlier is dropped from the sample,the average mean return of the remaining 9 clonesdrops to 3.61% (see the “Dedicated Short Bias∗”row).21 The intuition for the underperformance ofthe fixed-weight clones in this category is clear—given the positive trend in the S&P 500 duringthe 1980s and 1990s, a passive strategy of shortingthe S&P 500 is unlikely to have produced attractivereturns when compared to the performance of morenimble discretionary shortsellers.
The results for the rolling-window clones arebroadly consistent with those of the fixed-weightclones, though the average performance of rolling-window clones is typically lower than that of theirfixed-weight counterparts. For example, the aver-age mean returns of rolling-window clones for allcategories except Dedicated Short Bias are lowerthan their fixed-weight versions, in some cases bya factor of two or three. Part of these differencescan be explained by the different sample periods onwhich rolling-window clones are based—observethat the average mean returns of the underlyingfunds are lower in the rolling-window sample thanin the fixed-weight sample for all categories exceptEmerging Markets. But the more likely sourceof the performance difference between rolling-window and fixed-weight clones is the combinedeffects of look-ahead bias for the fixed-weight clones
and the increased estimation errors implicit in therolling-window clones.
Given these two effects, the performance of therolling-window clones is all the more remark-able in categories such as Dedicated Short Bias(6.83% average mean return vs. 2.58% averagemean return for the corresponding sample of funds),Equity Market Neutral (4.43% clones vs. 5.71%funds), Global Macro (12.97% clones vs. 9.01%funds), Long/Short Equity Hedge (9.08% clonesvs. 11.90% funds), and Managed Futures (19.24%clones vs. 11.84% funds). In the case of Dedi-cated Short Bias funds, it is not surprising thatrolling-window clones are able to outperform bothfixed-weight clones and the funds on which theyare based—the rolling-window feature providesadditional flexibility for capturing time-varyingexpected returns (such as the bull market of the1980s and 1990s) that a fixed-weight strategy sim-ply cannot. And in the case of Managed Futures,as with the fixed-weight clones, the rolling-windowclones exhibit considerable cross-sectional variationin their mean returns hence the superior perfor-mance of clones in this case may not be statisticallysignificant.
Nevertheless, as with fixed-weight clones, rolling-window clones also fall short substantially in thecategories of Emerging Markets (5.17% clonesvs. 21.12% funds), Event Driven (6.96% clonesvs. 11.65% funds), and Fixed Income Arbitrage(4.47% vs. 7.80%). Funds in these categoriesearn part of their expected return from bearingilliquidity risk, which is clearly absent from theclone portfolios constructed with the five factorswe employ. Therefore, we should expect clonesto underperform their fund counterparts in thesecategories.
Another metric of comparison between clones andfunds is the average Sharpe ratio, which adjustsfor the volatilities of the respective strategies. Of
JOURNAL OF INVESTMENT MANAGEMENT SECOND QUARTER 2007
REPLICATING HEDGE-FUND RETURNS 25
course, given our renormalization process (7), thestandard deviations of the fixed-weight clones areidentical to their fund counterparts, so a compar-ison of Sharpe ratios reduces to a comparison ofmean returns in this case. However, the averageSharpe ratio of a category is not the same as theratio of that category’s average mean return to itsaverage volatility, so the Sharpe ratio statistics inTable 7 and Figure 3 do provide some incrementalinformation. Moreover, for rolling-window clones,there may be some differences in volatilities depend-ing on the time series properties of the underlyingfunds, which makes Sharpe ratio comparisons moreinformative.
The average Sharpe ratio of the fixed-weight sam-ple of Convertible Arbitrage funds is 2.70, which isalmost twice the average Sharpe ratio of 1.52 for thefixed-weight clones, a significant risk-adjusted per-formance gap. Noticeable gaps also exist for EventDriven, Emerging Markets, and Fixed Income Arbi-trage clones versus funds (recall that funds in thesecategories are likely to earn illiquidity risk premianot available to the corresponding clones). How-ever, there is virtually no difference in averageSharpe ratios between fixed-weight clones and fundsin the Dedicated Short Bias, Equity Market Neu-tral, Long/Short Equity Hedge, Multi-Strategy, andFund of Funds categories. And for Global Macroand Managed Futures, the average Sharpe ratiosof the fixed-weight clones are, in fact, higher thanthose of the funds in these categories.
The gaps between average Sharpe ratios of rolling-window clones and those of the underlying fundstend to be more substantial—for the twin reasonscited above—but with some notable exceptions. Onaverage, rolling-window clones in the DedicatedShort Bias, Global Macro, and Managed Futurescategories do better than their fund counterpartson a risk-adjusted basis, with average Sharpe ratiosof 0.09, 1.08, and 0.91, respectively, as com-pared to average Sharpe ratios of 0.02, 0.91, and
0.66, respectively, for the corresponding sampleof funds.
4.3 Liquidity
Table 7 provides another comparison worth noting:the average first-order autocorrelation coefficientsof clones and funds. The first-order autocorrela-tion ρ1 is the correlation between a fund’s currentreturn and the previous month’s return, and Lo(2001, 2002) and Getmansky et al. (2004) observethat positive values for ρ1 in hedge-fund returns isa proxy for illiquidity risk. Table 7 and Figure 4show that the clones have much lower averageautocorrelations than their fund counterparts, withthe exception of Managed Futures for which bothclones and funds have very low average autocorre-lations. For example, the average autocorrelationof Convertible Arbitrage funds in the fixed-weightsample is 42.2%, and the corresponding averagevalue for Convertible Arbitrage fixed-weight androlling-window clones is only 10.7% and 6.4%,respectively. The average autocorrelation of Fundof Funds is 23.2% in the fixed-weight sample, andthe corresponding value for the fixed-weight androlling-window clones is only −0.1% and 0.0%,respectively.
The last two columns of each of the two sub-panels in Table 7 provide a more formal measureof the statistical significance of autocorrelation inthe monthly returns of clones and funds, theLjung–Box Q -statistic, based on the first 12 auto-correlation coefficients in the fixed-weight case andon the first 6 autocorrelation coefficients in therolling-window case.22 Smaller p-values indicatemore statistically significant autocorrelations, andfor every single category, the average p-value ofthe funds is lower than that of the clones. Theseresults confirm our intuition that, by construction,clones are more liquid than their correspondingfunds, highlighting another potential advantage of
SECOND QUARTER 2007 JOURNAL OF INVESTMENT MANAGEMENT
26 JASMINA HASANHODZIC AND ANDREW W. LO
Performance Comparison of Clones vs. FundsFixed-Weight Linear Clones
2.70
0.25
0.20
1.42
1.44
1.99
2.05
1.07
1.06
0.67
1.86 1.66
1.39
1.52
0.32 0.19
0.88
1.42
1.43
1.48
1.41
0.98
1.36
1.51
1.59
1.19
0.00
0.50
1.00
1.50
2.00
2.50
3.00
ConvertibleArbitrage
DedicatedShort Bias
DedicatedShort Bias*
EmergingMarkets
EquityMarketNeutral
EventDriven
FixedIncome
Arbitrage
GlobalMacro
Long/ShortEquityHedge
ManagedFutures
Multi-Strategy
Fund ofFunds
All ExceptFund ofFunds
Av
era
ge
Sh
arp
e R
ati
o
Funds Linear Clones
Performance Comparison of Clones vs. Funds24-Month Rolling-Window Linear Clones
2.31
0.02
-0.04
1.74
1.44
2.01
2.17
0.91
1.04
0.66
1.86 1.67
1.38
0.71
0.09
0.17
0.47
0.64
1.05 0.84
1.08
0.76
0.91 0.71
1.11
0.79
-0.50
0.00
0.50
1.00
1.50
2.00
2.50
3.00
ConvertibleArbitrage
DedicatedShort Bias
DedicatedShort Bias*
EmergingMarkets
EquityMarketNeutral
EventDriven
FixedIncome
Arbitrage
GlobalMacro
Long/ShortEquityHedge
ManagedFutures
Multi-Strategy
Fund ofFunds
All ExceptFund ofFunds
Ave
rag
e S
har
pe
Rat
io
Funds Linear Clones
Figure 3 Comparison of average Sharpe ratios of fixed-weight and 24-month rolling-window linear clonesand their corresponding funds in the TASS Live database, from February 1986 to September 2005. Thecategory “Dedicated Short Bias∗” excludes Fund 33735.
JOURNAL OF INVESTMENT MANAGEMENT SECOND QUARTER 2007
REPLICATING HEDGE-FUND RETURNS 27
Comparison of Average Autocorrelations of Funds andFixed-Weight Linear Clones
42.2
5.5 3.4
18.0
9.1
22.2
22.1
5.8
12.8
2.5
21.0
23.2
15.7
10.7
2.8
3.7
0.0
1.8
4.3
4.1 2.6
-0.1
4.7 2.3
-0.1
2.2
-5.0
0.0
5.0
10.0
15.0
20.0
25.0
30.0
35.0
40.0
45.0
ConvertibleArbitrage
DedicatedShort Bias
DedicatedShort Bias*
EmergingMarkets
EquityMarketNeutral
EventDriven
FixedIncome
Arbitrage
GlobalMacro
Long/ShortEquityHedge
ManagedFutures
Multi-Strategy
Fund ofFunds
All ExceptFund ofFunds
Ave
rag
e A
uto
corr
elat
ion
(%
)
Funds Linear Clones
Comparison of Average Autocorrelations of Funds and24-Month Rolling-Window Linear Clones
42.3
8.3
7.4
16.0
5.3
17.2
23.3
6.6
9.8
4.0
18.3
22.6
13.5
6.4
0.4
-0.7
7.7
4.2 3.0
4.3
4.1
0.3
5.5
0.8
0.0
2.8
-5.0
0.0
5.0
10.0
15.0
20.0
25.0
30.0
35.0
40.0
45.0
ConvertibleArbitrage
DedicatedShort Bias
DedicatedShort Bias*
EmergingMarkets
EquityMarketNeutral
EventDriven
FixedIncome
Arbitrage
GlobalMacro
Long/ShortEquityHedge
ManagedFutures
Multi-Strategy
Fund ofFunds
All ExceptFund ofFunds
Ave
rag
e A
uto
corr
elat
ion
(%
)
Funds Linear Clones
Figure 4 Comparison of average first-order autocorrelation coefficients of fixed-weight and 24-monthrolling-window linear clones and their corresponding funds in the TASS Live database, from February1986 to September 2005. The category “Dedicated Short Bias∗” excludes Fund 33735.
SECOND QUARTER 2007 JOURNAL OF INVESTMENT MANAGEMENT
28 JASMINA HASANHODZIC AND ANDREW W. LO
clone portfolios over direct investments in hedgefunds. However, this advantage comes at a cost;as we saw in Section 4.2, the performance gapbetween clones and funds is particularly large forthose categories with the highest levels of illiquidityexposure.
4.4 Leverage ratios
Another consideration in evaluating the practicalsignificance of fixed-weight linear clones is themagnitudes of the renormalization factors γi . Asdiscussed in Section 4.1, these factors representadjustments in the clone portfolios’ leverage so asto yield comparable levels of volatility. If the mag-nitudes are too large, this may render the cloningprocess impractical for the typical investor, who maynot have sufficient credit to support such leverage.The summary statistics in the left panel of Table 8for the renormalization factors {γi} suggest thatthis is not likely to be a concern—the average γiacross all funds in the fixed-weight sample is 2.05,and the median value is 1.81, implying that thetypical amount of additional leverage required toyield fixed-weight clones of comparable volatilityis 81–105% on average, which is far less than theleverage afforded by standard futures contracts suchas the S&P 500.23 For the individual categories, theaverage value of γi for fixed-weight clones variesfrom a low of 1.69 for Fund of Funds to a highof 2.76 for Managed Futures. This accords wellwith our intuition that Fund of Funds is lowerin volatility because of its diversified investmentprofile, and Managed Futures is higher in volatil-ity given the leverage already incorporated into thefutures contracts traded by CTAs and CPOs. Infact, outside of the Managed Futures category, eventhe maximum values for γi are relatively mild—ranging from 2.92 for Convertible Arbitrage to 8.85for Dedicated Short Bias—and the maximum valueof 18.35 for Managed Futures is also reasonablyconservative for that category.
Developing intuition for the leverage ratios for therolling-window clones is slightly more challeng-ing because they vary over time for each clone,so the right panel of Table 8 reports the cross-sectional means and standard deviations of thetime-series means, standard deviations, first-orderautocorrelations, and Q -statistic p-values of eachclone i’s {γit }. For example, the value 1.62 is themean across all rolling-window clones of the time-series-mean leverage ratio of each clone, and thevalue 0.46 is the cross-sectional standard deviation,across all rolling-window clones, of those time-seriesmeans. Table 8 shows that the average time seriesmean leverage ratios for rolling-window clones aresomewhat lower than their fixed-weight counter-parts but roughly comparable, ranging from a lowof 1.38 for Fund of Funds to a high of 1.97 forManaged Futures, with standard deviations of thetime-series means ranging from 0.31 for Fund ofFunds to 0.75 for Dedicated Short Bias (recall thatDedicated Short Bias has only 10 funds, and thatits short-bias mandate during the bull market is alsolikely to create more active rebalancings, contribut-ing to more volatile leverage ratios). However, noneof these leverage ratios fall outside the realm of prac-tical possibility for the five instruments implicit inthe cloning process.
4.5 Equal-weighted clone portfolios
The results in Sections 4.2–4.4 suggest that cloneportfolios consisting primarily of futures and for-ward contracts, properly leveraged, can yield com-parable volatility levels and some of the same riskexposures as certain types of hedge-fund strategies.But these impressions are based on averages of the1610 funds in our sample and their correspond-ing clones, not on specific realizable portfolios. Toaddress this issue, in this section we report thecharacteristics of equal-weighted portfolios of allfixed-weight and rolling-window clones, and com-pare them to the characteristics of equal-weighted
JOURNAL OF INVESTMENT MANAGEMENT SECOND QUARTER 2007
REPLICATING HEDGE-FUND RETURNS 29
Tab
le8
Sum
mar
yst
atis
tics
for
reno
rmal
izat
ion
fact
orsγ
iof
fixed
-wei
ght
and
24-m
onth
rolli
ng-w
indo
wcl
ones
ofhe
dge
fund
sin
the
TA
SSLi
veda
taba
se,f
rom
Febr
uary
1986
toSe
ptem
ber
2005
.
Fixe
d-w
eigh
tlin
ear
clon
e24
-mon
thro
lling
-win
dow
linea
rcl
one
reno
rmal
izat
ion
fact
orre
norm
aliz
atio
nfa
ctor
TS-
mea
nT
S-SD
ρ1
p-va
lue(
Q6)
Cat
egor
ySa
mpl
ede
scri
ptio
nsi
zeM
inM
edM
ean
Max
SDM
ean
SDM
ean
SDM
ean
SDM
ean
SD
All
fund
s16
100.
111.
812.
0518
.35
0.96
1.62
0.46
0.36
0.28
80.7
14.6
1.9
9.4
Con
vert
ible
arbi
trag
e82
0.11
1.64
1.71
2.92
0.53
1.44
0.40
0.27
0.19
77.1
21.0
3.6
12.6
Ded
icat
edsh
ortb
ias
101.
131.
512.
268.
852.
341.
570.
750.
340.
3682
.412
.20.
92.
8E
mer
ging
mar
kets
102
0.26
1.95
2.13
5.17
0.82
1.68
0.45
0.45
0.38
83.8
10.4
0.5
2.8
Equ
ity
mar
ketn
eutr
al83
0.39
2.17
2.18
4.25
0.80
1.78
0.54
0.40
0.26
77.6
17.4
5.2
17.3
Even
tdri
ven
169
0.40
1.60
1.78
4.45
0.62
1.44
0.36
0.28
0.21
81.5
13.1
1.5
8.2
Fixe
din
com
ear
bitr
age
620.
401.
692.
075.
461.
011.
470.
400.
320.
2179
.014
.92.
912
.2G
loba
lmac
ro54
1.13
2.44
2.60
5.67
1.15
1.86
0.45
0.44
0.34
78.2
17.7
2.0
6.3
Long
/Sho
rteq
uity
hedg
e52
01.
021.
922.
186.
320.
891.
750.
460.
380.
2679
.315
.11.
68.
6M
anag
edfu
ture
s11
41.
172.
412.
7618
.35
1.76
1.97
0.46
0.56
0.37
82.5
12.4
1.3
8.9
Mul
ti-s
trat
egy
590.
451.
892.
043.
750.
751.
570.
460.
380.
3081
.811
.71.
69.
5Fu
ndof
fund
s35
50.
751.
581.
697.
530.
651.
380.
310.
270.
2082
.813
.31.
78.
8
SECOND QUARTER 2007 JOURNAL OF INVESTMENT MANAGEMENT
30 JASMINA HASANHODZIC AND ANDREW W. LO
portfolios of the corresponding funds. By includingall clones and funds in each of their respective port-folios, we avoid the potential selection biases thatcan arise from picking a particular subset of clones,
for example, those with particularly high R2s or
statistically significant factor exposures.
Figure 5 plots the cumulative returns of the equal-weighted portfolios of fixed-weight and rolling-window clones, as well as the equal-weightedportfolios of their respective funds and the S&P500. The top panel shows that the equal-weightedportfolio of all fixed-weight clones outperformsboth the equal-weighted portfolio of correspond-ing funds and the S&P 500 over the sample period.However, the bottom panel shows that the perfor-mance of the equal-weighted portfolio of 24-monthrolling-window clones is not quite as impressive,underperforming both the funds portfolio and theS&P 500. However, the clones portfolio underper-forms the S&P 500 only slightly, and apparentlywith less volatility as visual inspection suggests.
Tables 9 and 10 and Figure 6 provide a more detailedperformance comparison of the portfolios of clonesand funds. In particular, Figure 6, which plotsthe Sharpe ratios of the equal-weighted portfoliosof the two types of clones and their correspond-ing funds, for all funds and category by category,shows that for some categories the fixed-weightclone portfolio underperforms the fund portfolio,for example, Dedicated Short Bias (0.09 for theclone portfolio vs. 0.28 for the fund portfolio),Emerging Markets (0.71 clones vs. 1.26 funds),Event Driven (1.89 clones vs. 3.08 funds), FixedIncome Arbitrage (1.79 clones vs. 2.93 funds), andMulti-Strategy (1.62 clones vs. 2.52 funds). How-ever, in other categories, the fixed-weight cloneportfolios have comparable performance and, insome cases, superior performance, for example,Managed Futures, where the fixed-weight cloneportfolio exhibits a Sharpe ratio of 1.80 versus 0.83
for the corresponding fund portfolio. When allclones are used to construct an equal-weighted port-folio, Table 9 reports an annualized mean return of20.44% with an annualized standard deviation of10.23% over the sample period, implying a Sharperatio of 2.00. The annualized mean and standarddeviation for an equal-weighted portfolio of allfunds are 14.76% and 9.06%, respectively, yieldinga Sharpe ratio of 1.63.
The lower panel of Figure 6 provides a compar-ison of the Sharpe ratios of the rolling-windowclone portfolios with their fund counterparts, whichexhibits patterns similar to those of the fixed-weightfund and clone portfolios. The rolling-windowclone portfolios underperform in some categoriesbut yield comparable performance in others, andsuperior performance in the categories of DedicatedShort Bias (0.19 clones vs. 0.12 funds), Equity Mar-ket Neutral (1.42 clones vs. 1.04 funds), GlobalMacro (1.48 clones vs. 1.39 funds), and Fund ofFunds (1.68 clones vs. 1.61 funds). For the equal-weighted portfolio of all rolling-window clones, theaverage return is 12.69% with a standard devia-tion of 10.60%, yielding a Sharpe ratio of 1.20;by comparison, the equal-weighted portfolio of allfunds has a 13.71% average return with a standarddeviation of 8.51%, yielding a Sharpe ratio of 1.61.
Tables 9 and 10 also report skewness, kurtosis, andautocorrelation coefficients for the two types ofclone portfolios, which gives a more detailed char-acterization of the risks of the return streams. Forsome of the categories, the differences in these mea-sures between clones and funds are quite striking.For example, according to Table 9, the portfolio ofFixed Income Arbitrage funds in the fixed-weightcase exhibits a skewness coefficient of −6, a kurtosiscoefficient of 59, and a first-order autocorrelationcoefficient of 27%, implying a negatively skewedreturn distribution with fat tails and significantilliquidity exposure. In contrast, the portfolio ofFixed Income Arbitrage fixed-weight clones has a
JOURNAL OF INVESTMENT MANAGEMENT SECOND QUARTER 2007
REPLICATING HEDGE-FUND RETURNS 31
Cumulative Return of Equal-Weighted Portfolio of Funds and ClonesFixed-Weight Linear Clones
0
10
20
30
40
50
Feb-86 Feb-88 Feb-90 Feb-92 Feb-94 Feb-96 Feb-98 Feb-00 Feb-02 Feb-04
Funds Linear Clones SP500
Cumulative Return of Equal-Weighted Portfolio of Funds and Clones24-Month Rolling-Window Linear Clones
0
2
4
6
8
10
12
Feb-86 Feb-88 Feb-90 Feb-92 Feb-94 Feb-96 Feb-98 Feb-00 Feb-02 Feb-04
Funds Linear Clones SP500
Figure 5 Cumulative returns of equal-weighted portfolios of funds and fixed-weight and 24-month rolling-window linear clones, and the S&P 500 index, from February 1986 to September 2005.
SECOND QUARTER 2007 JOURNAL OF INVESTMENT MANAGEMENT
32 JASMINA HASANHODZIC AND ANDREW W. LOT
able
9Pe
rfor
man
ceco
mpa
riso
nof
equa
l-w
eigh
ted
port
folio
sof
all
fixed
-wei
ght
linea
rcl
ones
vers
usfu
nds
inth
eT
ASS
Live
data
base
,fro
mFe
brua
ry19
86to
Sept
embe
r20
05.
Ded
icat
edsh
ort
Em
ergi
ngE
quit
ym
arke
tA
llfu
nds
Con
vert
ible
arb
bias
mar
kets
neut
ral
Even
tdri
ven
Stat
isti
cFu
nds
Clo
nes
Fund
sC
lone
sFu
nds
Clo
nes
Fund
sC
lone
sFu
nds
Clo
nes
Fund
sC
lone
s
Ann
ualc
ompo
und
retu
rn15
.34
21.8
511
.50
10.1
73.
78−1
.01
22.8
312
.45
13.4
019
.99
14.2
211
.95
Ann
ualiz
edm
ean
14.7
620
.44
11.0
79.
876.
402.
4122
.34
13.6
912
.83
18.6
613
.47
11.5
3A
nnua
lized
SD9.
0610
.23
5.36
5.45
23.2
326
.45
17.7
119
.28
6.23
7.73
4.37
6.11
Ann
ualiz
edSh
arpe
1.63
2.00
2.07
1.81
0.28
0.09
1.26
0.71
2.06
2.41
3.08
1.89
Skew
ness
10
00
00
−1−1
11
−1−1
Kur
tosi
s5
03
22
15
34
18
2ρ
1(≥
20%
inre
d)11
−631
913
336
−31
1232
0ρ
2(≥
20%
inre
d)−1
05
67
−12
−87
−33
1410
4ρ
3(≥
20%
inre
d)−1
10
−4−2
−69
−36
2116
−12
Cor
rela
tion
sto
vari
ous
mar
keti
ndex
es(≥
50%
inre
d,≤
−25%
inre
d):
S&P
500
inde
x44
6948
63−6
6−9
047
886
3758
78M
SCI
Wor
ldin
dex
4059
4252
−71
−92
5181
923
4962
Rus
sell
1000
inde
x44
6849
63−7
0−9
048
886
3660
78R
usse
ll20
00in
dex
4549
5254
−84
−73
5273
−118
7164
NA
SDA
Q10
0st
ock
inde
x38
5342
48−7
8−7
440
743
2652
66B
BA
LIB
OR
USD
3−m
onth
−13
−37
−29
−35
2−1
2−8
−6−8
−29
−8−2
2D
JLe
hman
Bon
dC
omp
GLB
L13
5722
492
7−5
1715
52−6
26U
STr
easu
ryN
/B−1
0−4
6−1
0−3
5−1
6−3
412
7−9
−52
8−1
8G
old
(Spo
t$/o
z)5
−6−9
−1−1
5−8
06
11
−5−7
Oil
(Gen
eric
1st‘
CL’
Futu
re)
−516
−20
−9−1
4−9
−99
1629
−12
US
Dol
lar
spot
inde
x2
−10
61
721
9−1
5−1
52
189
Five
risk
fact
ors
CR
ED
IT3
−920
30−3
9−5
338
55−9
−15
3737
USD
−15
−13
−2−1
3559
−13
−44
−10
14−2
−1B
ON
D13
6324
5810
25−2
911
673
38SP
500
4469
4863
−66
−90
4788
638
5878
DV
IX−2
2−3
2−2
4−4
350
70−3
6−6
46
−21
−50
−50
CM
DT
Y5
27−1
4−1
−12
−13
016
1738
512
CSF
B/T
rem
onti
ndex
esA
llfu
nds
8256
4650
−65
−42
5547
3448
6357
Con
vert
ible
arbi
trag
e47
2679
37−1
9−8
3421
3824
5331
Ded
icat
edsh
ortb
ias
−69
−62
−44
−43
8477
−58
−77
−26
−31
−73
−68
Em
ergi
ngm
arke
ts72
4442
38−6
4−5
894
5521
2460
49E
quit
ym
arke
tneu
tral
5039
3625
−29
−39
2638
5129
3636
Even
tdri
ven
7757
6456
−57
−56
6963
3835
8964
Fixe
din
com
ear
bitr
age
3320
3930
−55
2410
2622
3322
Glo
balm
acro
5436
2636
−25
−628
1722
4331
35Lo
ng/S
hort
equi
tyhe
dge
8261
4347
−79
−58
5564
3639
7160
Man
aged
futu
res
140
−12
−410
17−1
3−1
11
9−1
7−7
Mul
ti-s
trat
egy
2313
2920
−22
−10
013
278
2116
JOURNAL OF INVESTMENT MANAGEMENT SECOND QUARTER 2007
REPLICATING HEDGE-FUND RETURNS 33T
able
9(C
ontin
ued
)Fi
xed-
inco
me
Long
/Sho
rtar
bitr
age
Glo
balm
acro
equi
tyhe
dge
Man
aged
futu
res
Mul
ti-s
trat
egy
Fund
offu
nds
Stat
isti
cFu
nds
Clo
nes
Fund
sC
lone
sFu
nds
Clo
nes
Fund
sC
lone
sFu
nds
Clo
nes
Fund
sC
lone
s
Ann
ualc
ompo
und
retu
rn10
.93
7.91
15.5
624
.90
16.7
917
.40
15.1
538
.37
15.4
315
.08
12.3
018
.74
Ann
ualiz
edm
ean
10.4
87.
7314
.91
22.8
716
.35
17.1
515
.96
34.7
314
.59
14.5
111
.93
17.6
1A
nnua
lized
SD3.
584.
318.
649.
4011
.84
14.0
119
.24
19.3
25.
788.
947.
487.
97A
nnua
lized
Shar
pe2.
931.
791.
732.
431.
381.
220.
831.
802.
521.
621.
592.
21Sk
ewne
ss−6
01
0−2
−11
02
12
0K
urto
sis
593
50
185
20
87
71
ρ1
(≥20
%in
red)
27−2
1717
14−6
59
22−7
205
ρ2
(≥20
%in
red)
135
39
−3−3
−15
−415
−2−8
8ρ
3(≥
20%
inre
d)1
1−7
11−6
2−1
31
1512
−10
−1C
orre
lati
ons
tova
riou
sm
arke
tind
exes
(≥50
%in
red,
≤−2
5%in
red)
:S&
P50
0in
dex
−5−4
1047
7795
−1−2
4371
3166
MSC
IW
orld
inde
x−6
−10
638
6378
−1−2
4460
2949
Rus
sell
1000
inde
x−4
−410
4779
95−3
−244
7131
65R
usse
ll20
00in
dex
22
730
8678
−7−1
345
5332
49N
ASD
AQ
100
Stoc
kin
dex
2−3
333
7278
−7−9
4158
2849
BB
ALI
BO
RU
SD3−
Mon
th−4
−25
−13
−35
−5−1
2−1
3−4
0−1
5−1
9−9
−35
DJ
Lehm
anB
ond
Com
pG
LBL
639
1366
−714
2781
1040
1045
US
Trea
sury
N/B
−8−4
8−1
3−5
88
−7−2
8−8
53
−28
−9−4
0G
old
(Spo
t$/o
z)−1
715
6−1
0−1
113
63
−410
−10
Oil
(Gen
eric
1st‘
CL’
futu
re)
1020
1717
−57
125
921
−217
US
Dol
lar
spot
inde
x11
17−5
−10
178
−14
−27
−6−6
95
Five
risk
fact
ors
CR
ED
IT19
192
029
28−2
4−6
023
182
0U
SD18
32−4
−7−7
−12
−11
−11
−8−1
0−9
7B
ON
D14
6017
782
2123
886
4112
60SP
500
−4−4
1048
7795
−1−1
4371
3166
DV
IX27
00
−28
−56
−61
1318
−28
−45
−9−3
4C
MD
TY
724
1825
318
734
1630
829
CSF
B/T
rem
onti
ndex
esA
llfu
nds
3729
6149
7255
2317
6255
9157
Con
vert
ible
arbi
trag
e48
3226
2837
211
746
2452
28D
edic
ated
shor
tbia
s4
7−2
4−3
5−8
2−7
413
13−5
2−5
5−5
7−5
3E
mer
ging
mar
kets
217
2426
6452
−9−1
150
4070
40E
quit
ym
arke
tneu
tral
−4−1
3628
4641
1812
4737
4835
Even
tdri
ven
2418
3842
7462
−12
−761
5176
54Fi
xed
inco
me
arbi
trag
e79
3718
2521
128
1325
2043
25G
loba
lmac
ro42
3658
4037
2935
2535
3768
40Lo
ng/S
hort
equi
tyhe
dge
128
4242
8965
47
6458
8356
Man
aged
futu
res
−39
408
−7−7
8434
62
162
Mul
ti-s
trat
egy
3522
3113
2011
7−1
3813
2714
SECOND QUARTER 2007 JOURNAL OF INVESTMENT MANAGEMENT
34 JASMINA HASANHODZIC AND ANDREW W. LOT
able
10Pe
rfor
man
ceco
mpa
riso
nof
equa
l-w
eigh
ted
port
folio
sof
all2
4-m
onth
rolli
ng-w
indo
wlin
ear
clon
esve
rsus
fund
sin
the
TA
SSLi
veda
taba
se,f
rom
Febr
uary
1986
toSe
ptem
ber
2005
.D
edic
ated
shor
tE
mer
ging
Equ
ity
mar
ket
All
fund
sC
onve
rtib
lear
bbi
asm
arke
tsne
utra
lEv
entd
rive
n
Stat
isti
cFu
nds
Clo
nes
Fund
sC
lone
sFu
nds
Clo
nes
Fund
sC
lone
sFu
nds
Clo
nes
Fund
sC
lone
s
Ann
ualc
ompo
und
retu
rn14
.21
12.8
311
.07
6.88
−0.6
41.
5316
.92
5.34
7.53
12.0
513
.64
10.1
8A
nnua
lized
mea
n13
.71
12.6
910
.71
6.84
3.40
5.92
17.4
38.
587.
5511
.77
12.9
59.
90A
nnua
lized
SD8.
5110
.60
5.79
5.85
27.8
630
.63
18.1
624
.15
7.23
8.26
4.23
5.80
Ann
ualiz
edSh
arpe
1.61
1.20
1.85
1.17
0.12
0.19
0.96
0.36
1.04
1.42
3.06
1.71
Skew
ness
10
01
01
−1−2
−21
−2−1
Kur
tosi
s8
45
36
66
157
212
6ρ
1(≥
20%
inre
d)5
212
−58
−230
73
1436
−6ρ
2(≥
20%
inre
d)−7
−713
9−8
−28
4−1
96
128
ρ3
(≥20
%in
red)
−16
54
−20
24−2
2−1
8−8
−4−2
Cor
rela
tion
sto
vari
ous
mar
keti
ndex
es(≥
50%
inre
d,≤
−25%
inre
d):
S&P
500
inde
x41
5240
61−5
7−7
853
7612
2452
65M
SCI
Wor
ldin
dex
2837
3950
−63
−79
5769
1125
4750
Rus
sell
1000
inde
x41
5242
61−6
1−7
854
7713
2454
66R
usse
ll20
00in
dex
4237
4946
−80
−53
5560
1512
6348
NA
SDA
Q10
0st
ock
inde
x35
4141
46−7
6−6
744
6214
1546
50B
BA
LIB
OR
USD
3-M
onth
−11
−13
−27
−21
3−2
4−5
−4−2
−30
−7−1
6D
JLe
hman
Bon
dC
omp
GLB
L6
2420
313
8−9
112
43−2
16U
STr
easu
ryN
/B−8
−17
−6−2
0−1
4−3
415
12−9
−38
9−1
3G
old
(Spo
t$/o
z)3
3−5
−6−1
47
62
−7−2
5−3
Oil
(Gen
eric
1st‘
CL’
Futu
re)
015
−8−2
−17
−16
−16
818
59
US
Dol
lar
spot
inde
x13
4−2
48
2011
3−1
0−5
1113
Five
risk
fact
ors
CR
ED
IT13
1623
36−3
4−4
542
50−1
1−3
4536
USD
−4−9
−7−3
3044
−20
−16
−75
−64
BO
ND
1430
2037
1229
−44
1051
428
SP50
041
5240
61−5
7−7
853
7612
2452
66D
VIX
−20
−38
−16
−33
4660
−44
−62
−6−1
2−4
1−4
7C
MD
TY
821
−24
−11
−20
76
1326
916
CSF
B/T
rem
onti
ndex
esA
llfu
nds
8356
4939
−81
−27
5847
3337
6357
Con
vert
ible
arbi
trag
e44
3181
30−1
1−3
3421
3120
5333
Ded
icat
edsh
ortb
ias
−67
−66
−44
−47
7956
−59
−69
−28
−32
−72
−63
Em
ergi
ngm
arke
ts69
5040
37−7
1−4
194
5325
2760
51E
quit
ym
arke
tneu
tral
4843
3019
−12
−30
2726
5226
3531
Even
tdri
ven
7561
6248
−60
−38
7067
3736
9064
Fixe
din
com
ear
bitr
age
3423
4526
−4−7
2720
2519
3328
Glo
balm
acro
5835
2926
−39
532
2118
3130
38Lo
ng/S
hort
equi
tyhe
dge
8059
4538
−83
−31
5757
3833
7057
Man
aged
futu
res
18−9
−14
−13
624
−15
−27
511
−17
−10
Mul
ti-s
trat
egy
213
3510
−40
−39
−13
218
209
JOURNAL OF INVESTMENT MANAGEMENT SECOND QUARTER 2007
REPLICATING HEDGE-FUND RETURNS 35T
able
10(C
ontin
ued
)Fi
xed-
inco
me
Long
/Sho
rtar
bitr
age
Glo
balm
acro
equi
tyhe
dge
Man
aged
futu
res
Mul
ti-s
trat
egy
Fund
offu
nds
Stat
isti
cFu
nds
Clo
nes
Fund
sC
lone
sFu
nds
Clo
nes
Fund
sC
lone
sFu
nds
Clo
nes
Fund
sC
lone
s
Ann
ualc
ompo
und
retu
rn8.
604.
5712
.74
17.9
217
.17
12.7
914
.07
14.8
811
.31
8.05
10.7
113
.84
Ann
ualiz
edm
ean
8.36
4.56
12.4
417
.26
16.4
212
.96
14.8
316
.04
10.9
18.
0610
.43
13.3
5A
nnua
lized
SD4.
114.
078.
9511
.66
9.82
13.0
418
.19
20.5
85.
457.
746.
497.
96A
nnua
lized
Shar
pe2.
031.
121.
391.
481.
670.
990.
820.
782.
001.
041.
611.
68Sk
ewne
ss−7
−10
00
−11
00
01
0K
urto
sis
684
31
14
31
22
53
ρ1
(≥20
%in
red)
26−1
48
1217
−11
212
17−1
35
1ρ
2(≥
20%
inre
d)2
−19
−10
83
1−1
4−1
115
91
2ρ
3(≥
20%
inre
d)−6
4−3
8−4
10−1
22
229
−513
Cor
rela
tion
sto
vari
ous
mar
keti
ndex
es(≥
50%
inre
d,≤
−25%
inre
d):
S&P
500
inde
x−9
320
2774
86−5
−844
6134
44M
SCI
Wor
ldin
dex
−9−3
2120
6470
−12
−15
4855
2932
Rus
sell
1000
inde
x−8
321
2777
86−6
−845
6135
45R
usse
ll20
00in
dex
211
2012
8865
−9−9
4644
3630
NA
SDA
Q10
0st
ock
inde
x−1
213
1574
69−1
0−8
4453
2935
BB
ALI
BO
RU
SD3-
Mon
th−6
−11
−17
−21
−9−6
−9−1
2−1
1−2
−11
−12
DJ
Lehm
anB
ond
Com
pG
LBL
512
2327
615
1934
417
925
US
Trea
sury
N/B
−7−2
1−1
9−3
22
−4−2
8−4
22
−13
−9−1
9G
old
(Spo
t$/o
z)2
1014
2−7
−89
108
−58
1O
il(G
ener
ic1s
t‘C
L’Fu
ture
)10
162
13−1
20
517
1528
719
US
Dol
lar
spot
inde
x13
15−1
010
178
−3−3
127
113
Five
Ris
kFa
ctor
sC
RE
DIT
2231
23
2829
−22
−28
2729
1413
USD
1914
−12
0−3
−9−2
−7−9
0−4
−2B
ON
D14
3125
3910
1623
417
2215
33SP
500
−93
2127
7486
−5−8
4361
3445
DV
IX27
−14
−5−1
8−4
8−5
817
8−1
7−4
1−1
3−3
6C
MD
TY
514
515
011
1021
2034
1326
CSF
B/T
rem
onti
ndex
esA
llfu
nds
3421
6143
7352
2514
5841
9055
Con
vert
ible
arbi
trag
e49
2825
3237
241
1244
1846
32D
edic
ated
shor
tbia
s6
−10
−26
−36
−83
−72
131
−43
−52
−56
−59
Em
ergi
ngm
arke
ts15
1329
3363
52−6
−251
3468
46E
quit
ym
arke
tneu
tral
−99
3339
4340
1621
3431
4941
Even
tdri
ven
2427
3540
7461
−12
058
4573
59Fi
xed
inco
me
arbi
trag
e82
2019
2323
1710
727
1539
26G
loba
lmac
ro42
1958
3739
2837
1834
2669
38Lo
ng/S
hort
equi
tyhe
dge
911
4435
8959
411
5844
8153
Man
aged
futu
res
−66
4312
−8−1
483
277
220
−9M
ulti
-str
ateg
y40
1128
518
74
−11
3717
254
SECOND QUARTER 2007 JOURNAL OF INVESTMENT MANAGEMENT
36 JASMINA HASANHODZIC AND ANDREW W. LO
Performance comparison of equal-weighted actual-fund vs. clone portfoliosFixed-weight linear clones
1.63
2.07
0.28 0.18
1.26
2.06
3.08 2.93
1.73
1.38
0.83
2.52
1.59
2.00 1.81
0.09
-0.09
0.71
2.41
1.89 1.79
2.43
1.22
1.80 1.62
2.21
-0.50
0.00
0.50
1.00
1.50
2.00
2.50
3.00
3.50
All Funds ConvertibleArbitrage
DedicatedShort Bias
DedicatedShort Bias*
EmergingMarkets
EquityMarketNeutral
Event Driven FixedIncome
Arbitrage
GlobalMacro
Long/ShortEquityHedge
ManagedFutures
Multi-Strategy
Fund ofFunds
An
nu
aliz
edS
har
pe
Funds Linear clones
Performance comparison of equal-weighted actual-fund vs. clone portfolios24-month rolling-window linear clones
1.61
1.85
0.12
0.09
0.96
1.04
3.06
2.03
1.39
1.67
0.82
2.00
1.61
1.20
1.17
0.19
0.26
0.36
1.42
1.71
1.12
1.48
0.99
0.78
1.04
1.68
0.00
0.50
1.00
1.50
2.00
2.50
3.00
3.50
All Funds ConvertibleArbitrage
DedicatedShort Bias
DedicatedShort Bias*
EmergingMarkets
EquityMarketNeutral
Event Driven FixedIncome
Arbitrage
GlobalMacro
Long/ShortEquityHedge
ManagedFutures
Multi-Strategy
Fund ofFunds
An
nu
aliz
edS
har
pe
Funds Linear clones
Figure 6 Comparison of Sharpe ratios of equal-weighted portfolios of funds versus fixed-weight and 24-month rolling-window linear clones of hedge funds in the TASS Live database, from February 1986 toSeptember 2005.
JOURNAL OF INVESTMENT MANAGEMENT SECOND QUARTER 2007
REPLICATING HEDGE-FUND RETURNS 37
skewness of 0, a kurtosis of 3, and a first-orderautocorrelation of −2%, and the portfolio of FixedIncome Arbitrage rolling-window clones has simi-lar characteristics, which is consistent with the factthat the clone portfolios are comprised of highlyliquid securities. Other examples of this differ-ence in liquidity exposure include the portfolios offunds in Convertible Arbitrage, Emerging Markets,Event Driven, Multi-Strategy, and Fund of Fundscategories, all of which exhibit significant posi-tive first-order autocorrelation coefficients (31%,36%, 32%, 22%, and 20%, respectively) in con-trast to their fixed-weight clone counterparts (9%,−3%, 0%, −7%, and 5%, respectively). Similarly,the first-order autocorrelations of the portfolios offunds in these five categories using the rolling-window sample (12%, 30%, 36%, 17%, and 5%,respectively) are all larger than their rolling-windowclone counterparts (−5%, 7%, −6%, −13%,and 1%).
While the statistical properties of clone portfoliosmay seem more attractive, it should be kept inmind that some of these characteristics are related toperformance. In particular, one source of negativeskewness and positive kurtosis is the kind of option-based strategies associated with Capital DecimationPartners (see Section 2.1), which is a legitimatesource of expected return. And liquidity exposure isanother source of expected return, as in the case ofFixed Income Arbitrage where one common themeis to purchase illiquid bonds and shortsell more liq-uid bonds with matching nominal cashflows. Byreducing exposures to these risk factors throughclones, we should expect a corresponding reduc-tion in expected return. For example, in the caseof Fixed Income Arbitrage, Table 9 reports thatthe portfolio of funds yields an average return of10.48% with a standard deviation of 3.58% fora Sharpe ratio of 2.93, as compared to the fixed-weight portfolio of clones’ average return of 7.73%with a standard deviation of 4.31% for a Sharperatio of 1.79.
In addition to its expected return and volatility, aportfolio’s correlation with major market indexesis another important characteristic that concernshedge-fund investors because of the diversificationbenefits that alternative investments have tradi-tionally provided. Tables 9 and 10 show that thefixed-weight and rolling-window clone portfoliosexhibit correlations that are similar to those oftheir matching portfolios of funds for a variety ofstock, bond, currency, commodity, and hedge-fundindexes.24 For example, the portfolio of ConvertibleArbitrage funds in the fixed-weight sample has a48% correlation to the S&P 500, a −29% cor-relation to 3-month LIBOR, a 6% correlation tothe US Dollar Index, and a 79% correlation tothe CSFB/Tremont Convertible Arbitrage Index. Incomparison, the portfolio of Convertible Arbitragefixed-weight clones has a 63% correlation to theS&P 500, a −35% correlation to 3-month LIBOR,a 1% correlation to the US Dollar Index, and a37% correlation to the CSFB/Tremont ConvertibleArbitrage Index.
However, some differences do exist. The equal-weighted portfolios of funds tend to have highercorrelation with the corresponding CSFB/TremontHedge-Fund Index of the same category than theequal-weighted portfolios of both types of clones.For example, the correlation between the portfo-lio of funds and the CSFB/Tremont Hedge-FundIndex in the fixed-weight sample is 82%, and thecorresponding correlation for the portfolio of fixed-weight clones with the same index is 56%, and thesame pair of correlations for the rolling-windowcase is 83% and 56%, respectively. This patternis repeated in every single category for both typesof clones, and is not unexpected given that theCSFB/Tremont indexes are constructed from thefunds themselves. On the other hand, the portfoliosof clones are sometimes more highly correlated withcertain indexes than their fund counterparts becauseof how the clones are constructed. For example,the correlation of the portfolio of Equity Market
SECOND QUARTER 2007 JOURNAL OF INVESTMENT MANAGEMENT
38 JASMINA HASANHODZIC AND ANDREW W. LO
Table 11 Comparison of signs and absolute differences of correlations of funds and clones to 28market indexes, where fixed-weight and 24-month rolling-window linear clones are constructedfrom hedge funds in the TASS Live database, from February 1986 to September 2005.
Fixed-weight linear clones Rolling-window linear clones
% Same Mean SD % Same Mean SDCategory sign |ρf − ρc| |ρf − ρc| sign |ρf − ρc| |ρf − ρc|All funds 86 19 11 93 12 7Convertible arbitrage 100 12 10 93 12 10Dedicated short bias 93 13 7 89 19 13Emerging markets 79 19 12 86 10 9Equity market neutral 86 19 13 96 12 10Event driven 89 12 10 86 11 6Fixed income arbitrage 86 13 13 71 14 13Global macro 100 20 17 93 10 7Long/Short equity hedge 89 11 6 96 10 6Managed futures 96 15 20 96 9 11Multi-strategy 93 14 10 93 12 6Fund of funds 89 23 11 96 13 9
Neutral fixed-weight clones with the BOND fac-tor is 67%, whereas the correlation of the portfolioof corresponding Equity Market Neutral funds isonly 11%. This difference is likely the result ofthe fact that the BOND factor is one of the fivefactors used to construct clone returns, so the cor-relations of clone portfolios to these factors willtypically be larger in absolute value than those ofthe corresponding fund portfolios.
A summary of the differences in correlation prop-erties between funds and clones is provided byTable 11. The first column in each of the twosub-panels labelled “% Same Sign” contains thepercentage of the 28 market-index correlations inTables 9 and 10, respectively, for which the fundcorrelation and the clone correlation are of the samesign. The next two columns of each sub-panel con-tain the mean and standard deviation of the absolutedifferences in fund- and clone-correlation across the28 market-index correlations. These results show
remarkable agreement in sign for both fixed-weightand rolling-window clones, ranging from 71% to100%, and mean absolute-differences ranging from9% to 23%. And even for the largest mean absolute-difference of 23% (fixed-weight clones in the Fundof Funds category), 89% of the correlations exhibitthe same sign in this category.
Overall, the results in Tables 9–11 show that thecorrelations of clone portfolios are generally com-parable in sign and magnitude to those of the fundportfolios, implying that portfolios of clones canprovide some of the same diversification benefits astheir hedge-fund counterparts.
5 Conclusion
A portion of every hedge fund’s expected return isrisk premia—compensation to investors for bearingcertain risks. One of the most important benefits of
JOURNAL OF INVESTMENT MANAGEMENT SECOND QUARTER 2007
REPLICATING HEDGE-FUND RETURNS 39
hedge-fund investments is the non-traditional typesof risks they encompass, such as tail risk, liquid-ity risk, and credit risk. Most investors would dowell to take on small amounts of such risks if theyare not already doing so because these factors usu-ally yield attractive risk premia, and many of theserisks are not highly correlated with those of tra-ditional long-only investments. Although talentedhedge-fund managers are always likely to outper-form passive fixed-weight portfolios, the challengesof manager selection and monitoring, the lack oftransparency, the limited capacity of such managers,and the high fees may tip the scales for the institu-tional investor in favor of clone portfolios. In suchcircumstances, portable beta may be a reasonablealternative to portable alpha.
Our empirical findings suggest that the possibil-ity of cloning hedge-fund returns is real. Forcertain hedge-fund categories, the average perfor-mance of clones is comparable—on both a raw-return and a risk-adjusted basis—to their hedge-fund counterparts. For other categories like EventDriven and Emerging Markets, the clones are lesssuccessful.
The differences in performance of clones acrosshedge-fund categories raise an important philo-sophical issue: What is the source of the clones’value-added? One possible interpretation is that thecloning process “reverse-engineers” a hedge fund’sproprietary trading strategy, thereby profiting fromthe fund’s intellectual property. Two assumptionsunderlie this interpretation, both of which arerather unlikely: (1) it is possible to reverse-engineera hedge fund’s strategy using a linear regression of itsmonthly returns on a small number of market-indexreturns; and (2) all hedge funds possess intellec-tual property worth reverse-engineering. Given theactive nature and complexity of most hedge-fundstrategies, it is hard to imagine reverse-engineeringthem by regressing their monthly returns on fivefactors. However, if such strategies have risk
factors in common, it is not hard to imagineidentifying them by averaging a reasonable cross-section of time-series regressions of monthly returnson those risk factors. As for whether all hedgefunds have intellectual property worth reverse-engineering, we have purposely included all theTASS hedge funds in our sample—not just the suc-cessful ones—and it is unlikely that all of the 1610funds possess significant manager-specific alpha.In fact, for our purposes, the main attractionof this sample of hedge funds is the funds’ betaexposures.
Our interpretation of the clones’ value-added is lessdevious: By analyzing the monthly returns of a largecross-section of hedge funds (some of which havegenuine manager-specific alpha, and others whichdo not), it is possible to identify common risk fac-tors from which those funds earn part, but notnecessarily all, of their expected returns. By takingsimilar risk exposures, it should be possible to earnsimilar risk premia from those exposures, hence atleast part of the hedge funds’ expected returns can beattained, but in a lower-cost, transparent, scalable,and liquid manner.
As encouraging as our empirical results may seem,a number of qualifications should be kept in mind.First, we observed a significant performance dif-ference between fixed-weight and rolling-windowclones, and this gap must be weighed carefully inany practical implementation of the cloning pro-cess. The fixed-weight approach yields better histor-ical performance and lower turnover, but is subjectto look-ahead bias so the performance may notbe achievable out-of-sample. The rolling-windowapproach yields less attractive historical perfor-mance, but the simulated performance may be moreattainable, and the flexibility of rolling-windowestimators may be critical for capturing nonsta-tionarities such as time-varying means, volatilities,and regime changes. The costs and benefits of eachapproach must be evaluated on a case-by-case basis
SECOND QUARTER 2007 JOURNAL OF INVESTMENT MANAGEMENT
40 JASMINA HASANHODZIC AND ANDREW W. LO
with the specific objectives and constraints of theinvestor in mind.
Second, despite the promising properties of lin-ear clones in several style categories, it is well-known that certain hedge-fund strategies containinherent nonlinearities that cannot be capturedby linear models (see, e.g., Capital Multiplica-tion Partners). Therefore, more sophisticated non-linear methods—including nonlinear regression,regime-switching processes, stochastic volatilitymodels, and Kat and Palaro’s (2005) copula-basedalgorithm—may yield significant benefits in termsof performance and goodness-of-fit. However, thereis an important trade-off between the goodness-of-fit and complexity of the replication process, andthis trade-off varies from one investor to the next.As more sophisticated replication methods are used,the resulting clone becomes less passive, requiringmore trading and risk-management expertise, andeventually becoming as dynamic and complex as thehedge-fund strategy itself.
Third, the replicating factors we proposed are only asmall subset of the many liquid instruments that areavailable to the institutional investor. By expandingthe universe of factors to include options and otherderivative securities, and customizing the set of fac-tors to each hedge-fund category (and perhaps toeach fund), it should be possible to achieve addi-tional improvements in performance, including theability to capture tail risk and other nonlinearitiesin a fixed-weight portfolio. In fact, Haugh and Lo(2001) show that a judiciously constructed fixed-weight portfolio of simple put and call options canyield an excellent approximation to certain dynamictrading strategies, and this approach can be adoptedin our context to create better clones.
Finally, we have not incorporated any transactionscosts or other frictions into our performance anal-ysis of the clone portfolios, which will clearly havean impact on performance. The more passive clones
will be less costly to implement, but they maynot capture as many risk exposures and nonlinear-ities as the more sophisticated versions. However,by construction, clones will have a significant costadvantage over a traditional fund of funds invest-ment, not only because of the extra layer of fees thatfunds of funds typically charge, but also becauseof the clone portfolio’s more efficient use of capi-tal due to the cross-netting of margin requirementsand incentive fees. For example, consider a fundof funds with equal allocations to two managers,each of which charges a 2% management fee and a20% incentive fee, and suppose that in a given year,one manager generates a 25% return and the othermanager loses 5%. Assuming a 1% managementfee and a 10% incentive fee for the fund of funds,and no loss carryforwards for the underlying fundsfrom previous years, this scenario yields a net returnof only 4.05% for the fund of funds investors. Inthis case, the fees paid by the investors amount to astunning 59.5% of the net profits generated by theunderlying hedge funds.
Of course, a number of implementation issuesremain to be resolved before hedge-fund clonesbecome a reality, for example, the estimation meth-ods for computing clone portfolio weights, theimplications of the implied leverage required by ourrenormalization process, the optimal rebalancinginterval, the types of strategies to be cloned, andthe best method for combining clones into a sin-gle portfolio. We are cautiously optimistic that thepromise of our initial findings will provide sufficientmotivation to take on these practical challenges.
Acknowledgments
Research support from AlphaSimplex Group andthe MIT Laboratory for Financial Engineering isgratefully acknowledged. We thank Tom Bren-nan, Nicholas Chan, John Cox, Arnout Eikeboom,Gifford Fong (the editor), Jacob Goldfield, Shane
JOURNAL OF INVESTMENT MANAGEMENT SECOND QUARTER 2007
REPLICATING HEDGE-FUND RETURNS 41
Haas, Charles Harris, Eric Rosenfeld, KendallWalker, an anonymous referee, and participants atthe CFA Institute Risk Symposium 2006, the Octo-ber 2006 MIT Sloan Leaders Network, and Le ClubB 2006 for many helpful comments.
Notes
1 Nevertheless, derivatives-based replication strategies mayserve a different purpose that is not vitiated by com-plexity: risk attribution, with the ultimate objectiveof portfolio risk management. Even if an underlyinghedge-fund strategy is simpler than its derivatives-basedreplication strategy, the replication strategy may still beuseful in measuring the overall risk exposures of the hedgefund and designing a hedging policy for a portfolio ofhedge-fund investments.
2 As a matter of convention, throughout this paper wedefine the Sharpe ratio as the ratio of the monthly averagereturn to the monthly standard deviation, then annualizeby multiplying by the square root of 12. In the originaldefinition of the Sharpe ratio, the numerator is the excessreturn of the fund, in excess of the risk-free rate. Giventhe time variation in this rate over our sample period, weuse the total return so as to allow readers to select theirown benchmarks.
3 The margin required per contract is assumed to be:
100 × {15% × (current level of the SPX)
− (put premium) − (amount out of the money)}where the amount out of the money is equal to the currentlevel of the SPX minus the strike price of the put.
4 This figure varies from broker to broker, and is meant tobe a rather conservative estimate that might apply to a$10M startup hedge fund with no prior track record.
5 This example was first proposed by Merton in his15.415 Finance Theory class at the MIT Sloan Schoolof Management.
6 During the period from November 1976 to December2004, the annualized mean and standard deviation ofBerkshire Hathaway’s Series A shares were 29.0% and26.1%, respectively, for a Sharpe ratio of 1.12 using 0%for the risk-free benchmark return.
7 Once a hedge fund decides not to report its performance,is liquidated, is closed to new investment, restructured,or merged with other hedge funds, the fund is transferred
into the “Graveyard” database. A hedge fund can only belisted in the “Graveyard” database after being listed in the“Live" database. Because the TASS database fully rep-resents returns and asset information for live and deadfunds, the effects of suvivorship bias are minimized.However, the database is subject to backfill bias—when afund decides to be included in the database, TASS addsthe fund to the “Live” database and includes all availableprior performance of the fund. Hedge funds do not needto meet any specific requirements to be included in theTASS database. Due to reporting delays and time lagsin contacting hedge funds, some Graveyard funds canbe incorrectly listed in the Live database for a period oftime. However, TASS has adopted a policy of transferringfunds from the Live to the Graveyard database if they donot report over an 8- to 10-month period.
8 For studies attempting to quantify the degree and impactof survivorship bias, see Baquero et al. (2005), Brownet al. (1992), Brown et al. (1999), Brown et al. (1997),Carpenter and Lynch (1999), Fung and Hsieh (1997b,2000), Hendricks et al. (1997), Horst et al. (2001), Liang(2000), and Schneeweis and Spurgin (1996).
9 See Posthuma and van der Sluis (2003).10 We are not aware of any studies focusing on this type of
bias, but the impact of selection on statistical inferencehas been explored in some detail by several authors. See,for example, Leamer (1978) and the citations within.
11 TASS defines returns as the change in net asset valueduring the month (assuming the reinvestment of anydistributions on the reinvestment date used by the fund)divided by the net asset value at the beginning of themonth, net of management fees, incentive fees, and otherfund expenses. Therefore, these reported returns shouldapproximate the returns realized by investors.
12 This is no coincidence—TASS is owned by TremontCapital Management, which created the CSFB/Tremontindexes in partnership with Credit Suisse First Boston.
13 It is no coincidence that the categories with the highestdegree of average positive serial correlation are also thecategories with the highest average Sharpe ratios. Smoothreturn series will, by definition, have higher Sharpe ratiosthan more volatile return series with the same mean.
14 Litterman (2005) calls such risk exposures “exotic betas”and argues that “[t]he adjective ‘exotic’ distinguishes itfrom market beta, the only beta which deserves to get paida risk premium.” We disagree—there are several well-established economic models that illustrate the possibilityof multiple sources of systematic risk, each of which com-mands a positive risk premium, for example, Merton
SECOND QUARTER 2007 JOURNAL OF INVESTMENT MANAGEMENT
42 JASMINA HASANHODZIC AND ANDREW W. LO
(1973) and Ross (1976). We believe that hedge fundsare practical illustrations of these multi-factor models ofexpected returns.
15 Throughout this paper, all statistics except for thoserelated to the first-order autocorrelation have been annu-alized to facilitate interpretation and comparison.
16 In fact, a decomposition of the total mean returns ofthese two funds shows that the six factors account for verylittle of the two funds’ performance, hence the manager-specific alphas are particularly significant for these twomanagers (see Hasanhodzic and Lo, 2006b, Table 8 fordetails).
17 In Hasanhodzic and Lo (2006a) and in a previous draftof this paper, we used the term “buy-and-hold” todescribe this type of clone. Although this is consistentwith the passive nature of the clone portfolio, it is notstrictly accurate because a portfolio with fixed weightsdoes require periodic rebalancing to maintain those fixedweights. Moreover, if a clone is to be implemented viafutures and forward contracts as we propose, then even inthe absence of portfolio rebalancings, some trading willbe necessary as maturing contracts are “rolled” into thoseof the next maturity date. For these reasons, we now referto clone portfolios with constant portfolio weights overtime as “fixed-weight” clones.
18 However, it should be kept in mind that the futures andforward contracts corresponding to the five factors inEq. (5) have sizable amounts of leverage built into thecontracts themselves, so that for reasonable values of γi ,we can re-write Eq. (9) as:
R̂it = β∗i1 (γiSP500t ) + · · ·
+ β∗i5
(γiCMDTYt
) + δi Rl (14)
= β∗i1 SP500∗
t + · · · + β∗i5 CMDTY∗
t (15)
where we have redefined five new instruments in Eq. (15)that can achieve γi times the leverage of the originalinstruments in (9) at no additional cost. Since the coeffi-cients {β∗
ik} sum to one by construction, there is no needfor any additional borrowing or lending because eachredefined instrument is already leveraged by the factorγi , hence δi ≡0 in Eq. (15).
19 If there are missing observations within this 24-monthwindow, we extend the window backwards in time untilwe obtain 24 datapoints for our regression. Our choiceof 24 months for the rolling window was a compromisebetween the desire to capture nonstationarities in the dataand the need for a sufficient number of observations toestimate the parameters of the clone. We did not try
other window lengths because we wished to reduce theimpact of “backtest bias” or over-fitting on our empiricalresults.
20 Note that each type of clone has its own set of matchingresults for the funds. This is due to the fact that the first 24months of each fund’s history are used to calibrate the ini-tial estimates of the rolling-window clones, hence they arenot included in the rolling-window dataset from whichfund and clone performance statistics are computed.
21 Specifically, fund 33735 has a small but positive SP500beta coefficient, hence the clone portfolio for this fundis long the S&P 500 throughout the sample period fromApril 1997 to March 2005, implying a strong positivecontribution for the SP500 factor. Because the SP500beta coefficient is small and the SP500 factor is the mostvolatile of the five factors, the un-renormalized volatilityof the clone portfolio is considerably smaller than thevolatility of the fund, which causes our renormalizationprocess to magnify the positive mean return of the clonesubstantially.
22 Ljung and Box (1978) propose the following statisticto gauge the significance of the first m autocorrelationcoefficients of a time series with T observations:
Q = T (T +2)m∑
k=1
ρ̂2k/(T −k) (16)
which is asymptotically χ2m under the null hypothesis
of no autocorrelation. By forming the sum of squaredautocorrelations, the statistic Q reflects the absolute mag-nitudes of the ρ̂ks irrespective of their signs, hence fundswith large positive or negative autocorrelation coefficientswill exhibit large Q -statistics.
23 As of July 28, 2006, the initial margin requirement ofthe S&P 500 futures contract that trades on the ChicagoMercantile Exchange is $19,688, with a maintenancemargin requirement of $15,750. Given the contract valueof $250 times the S&P 500 Index and the settlement priceof 1284.30 on July 28, 2006 for the September 2006 con-tract, the initial and maintenance margin requirementsare 6.1% and 4.9% of the contract value, respectively,implying leverage ratios of 16.3 and 20.4, respectively(see http://www.cme.com for further details).
24 Except for the SP500 and DVIX factors, the index returnsused to compute the correlations in Tables 9 and 10are derived solely from the indexes themselves in theusual way (Returnt ≡ (Indext −Indext−1)/Indext ), withno accounting for any distributions. The SP500 factordoes include dividends, and the DVIX factor is the firstdifference of the month-end VIX index.
JOURNAL OF INVESTMENT MANAGEMENT SECOND QUARTER 2007
REPLICATING HEDGE-FUND RETURNS 43
A Appendix
The following is a list of category descriptions, takendirectly from TASS documentation, that define thecriteria used by TASS in assigning funds in theirdatabase to one of 11 possible categories:
Convertible Arbitrage This strategy is identifiedby hedge investing in the convertible securities of acompany. A typical investment is to be long the con-vertible bond and short the common stock of thesame company. Positions are designed to generateprofits from the fixed income security as well as theshort sale of stock, while protecting principal frommarket moves.
Dedicated Short Bias Dedicated short sellerswere once a robust category of hedge funds beforethe long bull market rendered the strategy difficultto implement. A new category, short biased, hasemerged. The strategy is to maintain net short asopposed to pure short exposure. Short biased man-agers take short positions in mostly equities andderivatives. The short bias of a manager’s portfoliomust be constantly greater than zero to be classifiedin this category.
Emerging Markets This strategy involves equityor fixed income investing in emerging marketsaround the world. Because many emerging mar-kets do not allow short selling, nor offer viablefutures or other derivative products with which tohedge, emerging market investing often employs along-only strategy.
Equity Market Neutral This investment strategyis designed to exploit equity market inefficienciesand usually involves being simultaneously long andshort matched equity portfolios of the same sizewithin a country. Market neutral portfolios aredesigned to be either beta or currency neutral, orboth. Well-designed portfolios typically control forindustry, sector, market capitalization, and other
exposures. Leverage is often applied to enhancereturns.
Event Driven This strategy is defined as “spe-cial situations” investing designed to capture pricemovement generated by a significant pendingcorporate event such as a merger, corporate restruc-turing, liquidation, bankruptcy, or reorganiza-tion. There are three popular sub-categories inevent-driven strategies: risk (merger) arbitrage,distressed/high yield securities, and Regulation D.
Fixed Income Arbitrage The fixed income arbi-trageur aims to profit from price anomalies betweenrelated interest rate securities. Most managers tradeglobally with a goal of generating steady returnswith low volatility. This category includes interestrate swap arbitrage, US and non-US govern-ment bond arbitrage, forward yield curve arbitrage,and mortgage-backed securities arbitrage. Themortgage-backed market is primarily US-based,over-the-counter, and particularly complex.
Global Macro Global macro managers carry longand short positions in any of the world’s major cap-ital or derivative markets. These positions reflecttheir views on overall market direction as influ-enced by major economic trends and/or events. Theportfolios of these funds can include stocks, bonds,currencies, and commodities in the form of cash orderivatives instruments. Most funds invest globallyin both developed and emerging markets.
Long/Short Equity Hedge This directional strat-egy involves equity-oriented investing on both thelong and short sides of the market. The objective isnot to be market neutral. Managers have the abilityto shift from value to growth, from small to mediumto large capitalization stocks, and from a net longposition to a net short position. Managers may usefutures and options to hedge. The focus may beregional, such as long/short US or European equity,or sector specific, such as long and short technologyor healthcare stocks. Long/Short equity funds tend
SECOND QUARTER 2007 JOURNAL OF INVESTMENT MANAGEMENT
44 JASMINA HASANHODZIC AND ANDREW W. LO
to build and hold portfolios that are substantiallymore concentrated than those of traditional stockfunds.
Managed Futures This strategy invests in listedfinancial and commodity futures markets and cur-rency markets around the world. The managersare usually referred to as Commodity TradingAdvisors, or CTAs. Trading disciplines are gener-ally systematic or discretionary. Systematic traderstend to use price and market specific information(often technical) to make trading decisions, whilediscretionary managers use a judgmental approach.
Multi-Strategy The funds in this category arecharacterized by their ability to dynamically allocatecapital among strategies falling within severaltraditional hedge-fund disciplines. The use of manystrategies, and the ability to reallocate capitalbetween them in response to market opportunities,means that such funds are not easily assigned to anytraditional category.
The Multi-Strategy category also includes fundsemploying unique strategies that do not fall underany of the other descriptions.
Fund of Funds A “Multi Manager” fund willemploy the services of two or more trading advi-sors or Hedge Funds who will be allocated cash bythe Trading Manager to trade on behalf of the fund.
References
Agarwal, V. and Naik, N. (2000a). “Generalised Style Analysisof Hedge Funds.” Journal of Asset Management 1, 93–109.
Agarwal, V. and Naik, N. (2000b). “On Taking the Alterna-tive Route: Risks, Rewards, and Performance Persistence ofHedge Funds.” Journal of Alternative Investments 2, 6–23.
Agarwal, V. and Naik, N. (2004). “Risk and Portfolio Deci-sions Involving Hedge Funds.” Review of Financial Studies17, 63–98.
Baquero, G., Horst, J. and Verbeek, M. (2005). “Sur-vival, Look-Ahead Bias, and Persistence in Hedge Fund
Performance.” Journal of Financial and Quantitative Analysis40, 493–517.
Bertsimas, D., Kogan, L., and Lo, A. (2001). “Hedg-ing Derivative Securities and Incomplete Markets: Anε-Arbitrage Approach.” Operations Research 49, 372–397.
Brown, S., Goetzmann, W., Ibbotson, R., and Ross, S.(1992). “Survivorship Bias in Performance Studies.” Reviewof Financial Studies 5, 553–580.
Brown, S., Goetzmann, W., and Ibbotson, R. (1999).“Offshore Hedge Funds: Survival and Performance 1989–1995.” Journal of Business 72, 91–118.
Brown, S., Goetzmann, W., and Park, J. (1997). “Conditionsfor Survival: Changing Risk and the Performance of HedgeFund Managers and CTAs.” Yale School of ManagementWorking Paper No. F-59.
Brown, S., Goetzmann, W., and Park, J. (2000). “HedgeFunds and the Asian Currency Crisis.” Journal of PortfolioManagement 26, 95–101.
Capocci, D. and Hubner, G. (2004). “Analysis of Hedge FundPerformance.” Journal of Empirical Finance 11, 55–89.
Carpenter, J. and Lynch, A. (1999). “Survivorship Bias andAttrition Effects in Measures of Performance Persistence.”Journal of Financial Economics 54, 337–374.
Edwards, F. and Caglayan, M. (2001). “Hedge Fund Perfor-mance and Manager Skill.” Journal of Futures Markets 21,1003–1028.
Ennis, R. and Sebastian, M. (2003). “A Critical Look at theCase for Hedge Funds: Lessons from the Bubble.” Journalof Portfolio Management 29, 103–112.
Fung, W. and Hsieh, D. (1997a). “Empirical Characteristicsof Dynamic Trading Strategies: The Case of Hedge Funds.”Review of Financial Studies 10, 275–302.
Fung, W. and Hsieh, D. (1997b). “Investment Style and Sur-vivorship Bias in the Returns of CTAs: The InformationContent of Track Records.” Journal of Portfolio Management24, 30–41.
Fung, W. and Hsieh, D. (2000). “Performance Characteristicsof Hedge Funds and Commodity Funds: Natural vs. Spu-rious Biases.” Journal of Financial and Quantitative Analysis35, 291–307.
Fung, W. and Hsieh, D. (2001). “The Risk in Hedge FundStrategies: Theory and Evidence from Trend Followers.”Review of Financial Studies 14, 313–341.
Fung, W. and Hsieh, D. (2004). “Hedge Fund Benchmarks:A Risk-Based Approach.” Financial Analysts Journal 60,65–80.
Getmansky, M., Lo, A., and Makarov, I. (2004). “An Econo-metric Analysis of Serial Correlation and Illiquidity inHedge-Fund Returns.” Journal of Financial Economics 74,529 –609.
JOURNAL OF INVESTMENT MANAGEMENT SECOND QUARTER 2007
REPLICATING HEDGE-FUND RETURNS 45
Getmansky, M., Lo, A., and Mei, S. (2004). “SiftingThrough the Wreckage: Lessons from Recent Hedge-Fund Liquidations.” Journal of Investment Management 2,6–38.
Hasanhodzic, J. and Lo, A. (2006a). “Attack of the Clones.”Alpha, June, 54–63.
Hasanhodzic, J. and Lo, A. (2006b). “Can Hedge-FundReturns Be Replicated?: The Linear Case” (August 16,2006). Available at SSRN: <http://ssrn.com/abstract=924565>.
Haugh, M. and Lo, A. (2001). “Asset Allocation andDerivatives.” Quantitative Finance 1, 45–72.
Hendricks, D., Patel, J., and Zeckhauser, R. (1997). “The J-Shape of Performance Persistence Given Survivorship Bias.”Review of Economics and Statistics 79, 161–170.
Hill, J., Mueller, B., and Balasubramanian, V. (2004). “The‘Secret Sauce’ of Hedge Fund Investing—Trading RiskDynamically.” Goldman Sachs Equity Derivatives Strategy,November 2.
Horst, J., Nijman, T., and Verbeek, M. (2001). “EliminatingLook-Ahead Bias in Evaluating Persistence in Mutual FundPerformance.” Journal of Empirical Finance 8, 345–373.
Kat, H. and Palaro, H. 2005, “Who Needs HedgeFunds? A Copula-Based Approach to Hedge Fund ReturnReplication.” Alternative Investment Research CentreWorking paper No. 27, Cass Business School, City Uni-versity, London.
Kat, H. and Palaro, H. (2006a). “Replication and Evaluationof Fund of Hedge Funds Returns.” Alternative InvestmentResearch Centre Working paper No. 28, Cass BusinessSchool, City University, London.
Kat, H. and Palaro, H. (2006b). “Superstars or AverageJoes?: A Replication-Based Performance Evaluation of 1917Individual Hedge Funds.” Alternative Investment ResearchCentre Working paper No. 30, Cass Business School, CityUniversity, London.
Leamer, E. (1978). Specification Searches. New York: JohnWiley and Sons.
Liang, B. (1999). “On the Performance of Hedge Funds.”Financial Analysts Journal, 72–85.
Liang, B. (2000). “Hedge Funds: The Living and the Dead.”Journal of Financial and Quantitative Analysis 35, 309–326.
Litterman, R. (2005). “Beyond Active Alpha.” GoldmanSachs Asset Management.
Ljung, G. and Box, G. (1978). “On a Measure of Lack of Fitin Time Series Models.” Biometrika 65, 297–304.
Lo, A. (2001). “Risk Management For Hedge Funds: Intro-duction and Overview.” Financial Analysts Journal 57,16–33.
Lo, A. (2002). “The Statistics of Sharpe Ratios.” FinancialAnalysts Journal 58, 36–50.
Merton, R. (1973). “An Intertemporal Capital Asset PricingModel.” Econometrica 41, 867–887.
Merton, R. (1981). “On Market Timing and Investment Per-formance I: An Equilibrium Theory of Value for MarketForecasts.” Journal of Business 54, 363–406.
Posthuma, N. and van der Sluis, P. (2003). “A Reality Checkon Hedge Funds Returns” (July 8, 2003). Available atSSRN: <http://ssrn.com/abstract=438840>.
Ross, S. (1976). “The Arbitrage Theory of Capital AssetPricing.” Journal of Economic Theory 13, 341–360.
Schneeweis, T. and Spurgin, R. (1996). “Survivor Biasin Commodity Trading Advisor Performance.” Journal ofFutures Markets 16, 757–772.
Schneeweis, T. and Spurgin, R. (1998). “Multi-Factor Analy-sis of Hedge Fund, Managed Futures, and Mutual FundReturn and Risk Characteristics.” Journal of AlternativeInvestments 1, 1–24.
Sharpe, W. F. (1992). “Asset Allocation: ManagementStyle and Performance Measurement.” Journal of PortfolioManagement 18, 7–19.
Keywords: Hedge fund; portfolio management; riskmanagement
SECOND QUARTER 2007 JOURNAL OF INVESTMENT MANAGEMENT