P. 1
Risk of Hedge Funds

Risk of Hedge Funds

|Views: 11|Likes:
Published by Shamin Suvarna

More info:

Published by: Shamin Suvarna on Jan 04, 2011
Copyright:Attribution Non-commercial

Availability:

Read on Scribd mobile: iPhone, iPad and Android.
download as PDF, TXT or read online from Scribd
See more
See less

12/25/2013

pdf

text

original

The Society for Financial Studies

The Risk in Hedge Fund Strategies: Theory and Evidence from Trend Followers Author(s): William Fung and David A. Hsieh Source: The Review of Financial Studies, Vol. 14, No. 2 (Summer, 2001), pp. 313-341 Published by: Oxford University Press. Sponsor: The Society for Financial Studies. Stable URL: http://www.jstor.org/stable/2696743 . Accessed: 24/12/2010 04:20
Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in the JSTOR archive only for your personal, non-commercial use. Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at . http://www.jstor.org/action/showPublisher?publisherCode=oup. . Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed page of such transmission. JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.

Oxford University Press and The Society for Financial Studies are collaborating with JSTOR to digitize, preserve and extend access to The Review of Financial Studies.

http://www.jstor.org

The Risk in Hedge Fund Strategies: Theory and Evidence from Trend Followers
William Fung PI Asset Management,LLC David A. Hsieh Duke University

Hedge fund strategiestypically generate option-like returns.Linear-factormodels using benchmarkasset indices have difficulty explaining them. Following the suggestions in Glosten and Jagannathan (1994), this article shows how to model hedge fund returnsby focusing on the popular "trend-following" strategy.We use lookback straddlesto model trend-followingstrategies,and show that they can explain trend-followingfunds' returns better than standardasset indices. Though standardstraddles lead to similar empirical results, lookback straddlesare theoreticallycloser to the concept of trendfollowing. Our model should be useful in the design of performancebenchmarksfor trend-following funds.

The last decade has witnessed a growing interest in hedge funds from investors, academics, and regulators.Investors and academics are intrigued by the unconventionalperformancecharacteristicsin hedge funds, and regulators are concerned with the market impact of their reported speculative activities during major market events.' The near bankruptcyof Long-Term Capital Management (LTCM) in 1998 has further heightened attention on hedge fund risk. Because hedge funds are typically organized as private investmentvehicles for wealthy individuals and institutionalinvestors,2they do not disclose their activities publicly. Hence, little is known about the
This article received research support from the Foundation for Managed Derivatives Research and John W. Henry & Company. We acknowledge the New York Mercantile Exchange, the London International Financial Futures Exchange, the Sydney Futures Exchange, and the Tokyo Intemational Financial Futures Exchange for providing their historical data; thanks to James Cui and Guy Ingram for computational assistance. An earlier version benefited from comments by Todd Petzel, Jules Stanowicz, Richard Roll, Don Chance, Debra Lucas, and the participants of the CBOT Winter Research Seminar on December 8-9, 1997, in Chicago and the participants of the finance workshops at University of Washington at Seattle, Virginia Polytechnic Institute, University of California at Los Angeles, the NBER Asset Pricing Program in Chicago (October 30, 1998), University of Rochester, Washington University at St. Louis, and Carnegie-Mellon University. We are especially indebted to Joe Sweeney, Mark Rubinstein, and Ravi Jagannathan for valuable comments and suggestions. Address correspondence to David A. Hsieh, Fuqua School of Business, Duke University, Box 90120, Durham, NC 27708-0120 or e-mail: david.a.hsieh@duke.edu. 'See Fung and Hsieh (2000) and Fung et al. (1999) for analyses on the market impact of hedge fund activities.
2

See Fung and Hsieh (1999) for an overview for hedge fund organizational rationale.

structure and their economic

The Review of Financial Studies Summer 2001 Vol. 14, No. 2, pp. 313-341 C 2001 The Society for Financial Studies

The Review of Financial Studies / v 14 n 2 2001

risk in hedge fund strategies. This article illustrates a general methodology hedge fund risk by modeling a particulartrading strategy for understanding commonly referredto as "trendfollowing" by the investmentindustry. As documented in Fung and Hsieh (1997a), hedge fund managers typically employ dynamic trading strategies that have option-like returns with no models of investmentstyles using apparently systematicrisk. Linear-factor standardasset benchmarks,as in Sharpe (1992), are not designed to capture the nonlinearreturnfeatures commonly found among hedge funds. This may lead investors to conclude erroneouslythat there are no systematic risks. A remedy is in Glosten and Jagannathan(1994), where they suggested indices that have embeddedoption-like features.3This using benchmark-style is done in Fung and Hsieh (1997a) for hedge funds where they extractedstyle factors from a broad sample of hedge fund returns.By construction,these style factors capturedmuch of the option-like features while preserving the general lack of correlationwith standardasset benchmarks.To fully capture hedge fund risk, we must model the nonlinear relationshipsbetween these style factors and the marketsin which hedge funds trade.This is not a simple task. The lack of public disclosure makes it difficult to link hedge fund style factors to asset markets. Our task is furthercomplicatedby the fact that hedge fund managers,who are no strangersto risk, generally diversify their fund's performanceacross a variety of strategies. Consequently,the observed returnsand extracted style factors are generatedby portfolios of different strategies,each having a different type of risk. The combination of the dynamic allocation of capital resourcesto a portfolio of tradingstrategies,each with nonlinearreturncharacteristics, greatly limits the value of analyzing a general sample of many hedge funds. From a modeling perspective, it is useful to concentrateon a specific trading strategy that is identifiable with a reasonably large number of hedge funds, whose returnsare predominantlygeneratedby that strategy. In this article, we focus on a popular strategy commonly referred to as "trendfollowing."4Trendfollowing is a self-describedstrategyfor the majority of commodity trading advisors (CTAs), as shown in Billingsley and
3 Typically,

performance evaluation and attribution models rely on regressing a manager's historical returns on one or more benchmarks. The slope coefficients reflect benchmark-related performance, whereas the constant term ("alpha") measures performance "benchmark risk." This approach dates back to Jensen's (1968) original work. Unfortunately, this type of regression method is sensitive to nonlinear relationships between the manager's returns and the benchmarks and can result in incorrect inferences. For instance, Grinblatt and Titman (1989) showed that a manager can generate positive Jensen's alphas by selling call options on the underlying stocks of a given standard benchmark. Merton (1981) and Dybvig and Ross (1985) showed that a portfolio manager with market-timing ability can switch between stocks and bonds to generate returns with option-like features without explicitly trading options. Empirically, Lehman and Modest (1987) found that a number of mutual funds exhibited option-like return features. A standard way to deal with option-like return features is to add nonlinear functions of the benchmark return as regressors. This was done in Treynor and Mazuy (1966) and Henriksson and Merton (1981). modeling other trading styles have emerged. See, for example, pairs trading in Gatev et al. (1999), risk arbitrage in Mitchell and Pulvino (1999), and relative-value trading in Richards (1999).

4Studies

314

The return profile shown in Figure 1 indicates that the relationship between trend followers and the equity marketis nonlinear.Risk in Hedge Fund Strategies Chance (1996). traders regard their trading systems to be proprietaryand are reluctant to disclose them. who trade primarily futures contracts on behalf of a customer. In addition. currency. Furthermore. Not only are returns of trend-following funds uncorrelatedwith the standardequity.5% of its value.althoughwe use the term trendfollowers to describe a certainclass 5 CTAs are individuals or trading organizations.In fact. Fung and Hsieh (1997b) showed that the returns of CTA funds have one dominant style factor. 1998 provides an out-of-sample observation that substantiates this view. reproducedhere as Figure 1. 82% generated positive results in August. albeit in a nonlinear manner.and commodity indices. this model contributes to explaining the performanceof other hedge funds that use trend following as part of their portfolio of strategies.but understandably. commodity funds generally had positive returns.6 This is evident in Exhibit 2 of Fung and Hsieh (1997b). If the tradingrules used by trend followers are readily available.Note that. Unfortunately. with 46 of them posting returns of more than 10%. This implies that there is one dominanttradingstrategyin CTA funds. The monthly returnsof the world equity market." State 1 consists of the worst months. Jaye Scholl wrote. the returnpatternof trend-followingfunds in Figure 1 is similar to those of contingent claims on the underlying asset and must thereforehave systematic risk. We can thereforeonly theorize what trend followers do. Trend following is a particularlyinteresting trading strategy.in each state. the state-dependentbeta estimates tend to be positive in up markets and negative in down markets. In a Barron's September 9. article. Fung and Hsieh (1997b) noted that a similar patternholds for an equally weighted portfolio of all trend-followingfunds. in the presence of nonlinearity. as proxied by the Morgan Stanley (MS) World Equity Index. While the S&P 500 lost 14. 1998. Fung and Hsieh (1997b) found these returnsto exhibit option-like features-they tended to be large and positive during the best and worst performingmonths of the world equity markets. and that strategyis trendfollowing.Although returns of trend-following funds have a low beta against equities on average.The goal of this article is to model how trend followers achieve this unusual return characteristicin order to provide a frameworkfor assessing the systematic risk of their strategy. This figure graphsthe averagemonthly returnof an equally weighted portfolio of the six largest trend-followingfunds. bond. We thereforefocus our empiricalwork on the returnof CTA funds to develop a model that explains their returns. we can directly estimate their systematic risk. are sorted into five "states. "Of the 17 commodity trading advisors reporting to MAR. and State 5 the best months.betas from a standardlinear-factor model can either overstate the systematic risk or understateit (as in the case of LTCM). registered with the Commodity Futures Trading Conmmission (CFTC) through membership in the National Futures Association.5 Also." 6 August 315 . along with that of the world equity markets.

which we label as has the "primitivetrend-followingstrategy. a comparison of our model to the market-timing model of Merton (1981) can be found in Section 2 of this article. We posit that the simplest trend-following strategy. From a modeling perspective. Trend followers can converge onto the same "trend"for differentreasons and have very different"entryand exit" points. 316 .5% 0 -1s MSWolddEquitiesOSixTrend Followers -15% 1 2 3 Quintiles of MS World Equities 4 5 Figure 1 Average monthly returns of six large trend-following states Source: Fung and Hsieh (1997a). in practice their respective approachescan differ widely.7 The concept of a lookback option was first introduced in Goldman et al. trendfollowers should deliver returnsresembling those of a portfolio of bills and lookback straddles.The Review of Financial Studies / v 14 n 2 2001 15% 10% M 5% E 0% . which delivers the ex post maximumpayout of any trend-followingstrategy. a lookback put option allows the owner to sell at the highest price. This aspect is excluded in our simple model." the same payout as a structured The owner of a lookback call option option known as the "lookbackstraddle. However. Withinthis context." has the right to buy the underlying asset at the lowest price over the life of the option. (1979).we need a level of aggregationthat capturesthe essence of this tradingstyle but avoids some of the distractingidiosyncrasies of individualtrend followers. Similarly. The combinationof these two options is the lookback straddle.Unlike earlierstudies that 7 In reality. trend followers often make multiple entry and exit decisions over a sufficiently long investment horizon so long as there is sufficient volatility surrounding the underlying trend. funds in five different MS world equity market of traders.

Summary and conclusions are in Section 5.It confirmsthe intuitionthat trend-followingfunds' returns are similar to those of contingentclaims on standardasset indices.Z(t)}. Z(t) . We explore the similaritiesand differencesbetween trendfollowing and markettiming as trading strategies in the Merton (1981) framework. In a follow-up paper. For a perfect market timer. This provides the key link between the returnsof trend-followingfunds and standardasset markets. 1. the market timer's return is modified to reflect the short sale alternative. Section 4 discusses the question of performancebenchmarksfor trend followers. Otherwise. At the start of the period. 317 . as in Merton (1981). Rather. which is the returnof a portfolio of bills and a straddleon stocks. The rest of the articleis organizedas follows. The Primitive Trend-Following Strategy The convex returnpatternobserved in Figure 1 resembles the payout profile of a straddleon the underlyingasset.R(t)} + Max{O. only stocks will be held. Z(t) . These tradersapply capital resourcesto differentmarketsin a dynamic fashion and do so in a manner peculiar to their individual skill and technology.R(t)}. Merton (1981) showed that the return of his portfolio is given by R(t) + Max{O.only bills will be held. This implicitly assumes the presence of short sales constraints. Section 3 reportsthe improvements on explainingtrend-followingfunds' returnsusing our model versus standard asset benchmarks. it is designed to capture the general characteristicsof the entire family of trend-followingstrategies. In the absence of short sale constraints. which is the returnof a portfolio of bills and a call option on stocks.8 particularoption strategyis not designed to replicate any specific trend-following strategy. we show that the lookback straddleis better suited to capturethe essence of trend-followingstrategiesthan a simple straddle.Henriksson 8 See Alexander (1961). A simple strategythat yields the return patternof a straddleis that of a "markettimer" who can go long and short on the underlyingasset. We demonstrateempirically that lookback straddle returns resemble the returns of trend-following funds. if the markettimer forecasts stocks to outperformbills. R(t) .Risk in Hedge Fund Strategies explicitly specify "technicaltradingrules" to proxy a popularform of trendthis following strategy.let Z(t) denote the returnper dollar invested in the stock marketand R(t) the return per dollar invested in Treasurybills in period t.For a perfect markettimer.Section 2 details the data sample used to test our model. Section 1 sets out the theoretical foundationof the primitivetrend-followingstrategiesas lookback straddles. Here we note the opportunisticnatureof trendfollowers. Following his notation. Merton (1981) showed that the returnof his portfolio is given by R(t) +Max{O.Given any asset.

The distribution of capital resources between the respective trading strategy and the riskless asset will also depend on the investor's risk preference. but they do so in different ways. Both market timers and trend followers attempt to profit from price movements. 10 Therefore. The construction of the PTFS. they would capture the optimal payouts IS' . In reality.Smin. attempts to capture the largest price movement during the time interval. whereas trend followers trade 9 Note that we are not advocating that markets trend.The Review of Financial Studies / v 14 n 2 2001 test and Merton (1981) proposed a nonparametric on whether a markettimer had the ability to time the market. Smax' and Smin represent the initial asset price.-1Capital allocation to the PMTS or PTFS is determined by comparing the payout of the respective strategy to the return of the risk-free asset. without incurring any costs. We use a similar approach to model a trend follower. respectively."Trends are commonly related to serial correlationin price changes. A trend follower attemptsto capture "markettrends. Let S. these traders cannot perfectly anticipate price movements. generating the payout S'. The standard buy-and-hold strategy buys at the beginning and sells at the end of the period. S'. adds the possibility of trading on Smaxand Smin. we introduce the concepts of Primitive Market-TimingStrategy (PMTS) and Primitive Trend-FollowingStrategy (PTFS) as follows. A helpful distinction between their approaches can be made as follows. The PTFS. going long to capture a price increase. the optimal payout of the PTFS is Smax. Note that the PMTS is defined in a manner consistent with Merton (1981). In Merton (1981). It is helpful to begin with a qualitativecomparisonof markettiming and trend following as trading strategies. and the minimum price achieved over a given time interval. the ending price.11 If we are dealing with perfect market timers and perfect trend followers. a concept featuredprominentlyin the early tests of marketefficiency. and going short to capturea price decrease.where price changes exhibit positive serial correlation. on other hand. the optimal payout of the PMTS is IS' . Thus.A trend follower attemptsto identify developing price patternswith this propertyand trade in the directionof the trendif and when this occurs. on the other hand. a market timer forecasts the direction of an asset. we restrict our strategies to complete a single trade over the given time interval. Generally. A trend is a series of asset prices that move persistently in one directionover a given time interval.9 To provide a formal definition of these two trading strategies.SI. If S' is expected to be higher (lower) than S.Smin. the payout of the PTFS must equal that of the PMTS.SI and Snax . a long (short) position is initiated. 318 . Consistent with the Merton (1981) framework. market timers enter into a trade in anticipation of a price move over a given time period. Consequently. The PMTS attempts to capture the price movement between S and S'. The trade is reversed at the end of the period.S. the maximum price. That is an empirical issue best deferred to another occasion. if Merton's (1981) assumptions were strictly imposed.

and T the time to maturityof the option. In the next section. assume that Black and Scholes (1973) holds. we have some remarksregardingthe differences between the PTFS and the PMTS. and buying breakouts refers to the strategy of going long when a breakout happens. (6) I ).and al = [ln(S/X) + (r + Ia2 )T]/(aT1/2). in reality. In other words. (4) (5) - d = ln(S/Q) d. and the prices of lookback options can be found in Goldman et al. (1) where No()is the cumulative standardnormal distribution. we assume that the ex ante cost of the PMTS is the value of an at-the-moneystandardstraddle. We cannot estimate these costs without knowledge of their strategies.Ija2)T]/(aT - (r + 2a2)T]/(aT 1/2). = [ln(S/Q) + (r d3= [-ln(S/ Q) 12 42a)T]/(aT1/2).the delta of the lookback straddleis given by 8LB = [1 + -[1 a2/21r] N(-b3) + I + (u/a) exp(-rT - + 2rb/a2) N(b2) - a2/r] N(d3) (u/a) exp(-rT 2rd/a2) N(d2). (2) Here. (1979)."2 strategies. and that of the PTFS is the value of a lookback straddle. (3) where u = (r . The delta of the standardstraddleis given by S = 2N(al) - 1.Instead. For illustrativepurposes. and selling breakdowns refers to the strategy of going short when a breakdown happens.Before doing so. Also. the PMTS is a long position in a standardstraddle. we will empirically create returnsof the PTFS using lookback straddleson 26 different markets. market timers and trend followers do incur costs when they attemptto capturetheir respective optimal payouts.1 2). The terms buying breakouts and selling breakdowns are often used to describe These are very common characteristicsof trend-following trend followers. r the instantaneousinterest rate. As the PMTS and PTFS are option positions. we can illustratetheir theoretical differencevia their deltas.Risk in Hedge Fund Strategies only after they have observed certain price movements during a period. The price of a standard straddle is then well known. Breakdown means that the price moves below a recent low. Breakout means that the price of an asset moves above a recent high. In comparison. 319 . and the PTFS is a long position in a lookback straddle. (7) (8) d2= [-ln(S/ Q) + (r . S is the currentprice of the underlying asset with instantaneousvariance a.

The rollover process is much reminiscent of the buying breakouts and selling breakdowns characteristics of trend-following strategies. The key difference lies in the path-dependency of the lookback option.13 However. Furthermore. 320 . Empirically. b. )T]/(TT1/2). involving bills and lookback straddles. of the asset since the inception of the lookback straddle. this observation period may be too short to deliver a consistently dramatic payout difference between lookback straddles and standard straddles. that performs like trend-following funds. the difference between the PMTS and the PTFS is much more subtle. the lookback straddle can be replicated by dynamically rolling standard straddles over the life of the option. A derivation of Equation (3) is available from the authors on request. In the empirical implementation. Given any investment horizon. It is conceivable that.The Review of Financial Studies / v 14 n 2 2001 b = ln(M/S). (1979). as both the PMTS and PTFS make use of standard straddles on the same asset. We note here that the goal of this article is to show that there is at least one option portfolio. Several examples of the difference in the deltas are in Appendix A. albeit in a different manner. the payout of the two strategies will diverge. which tend to be the most liquid options. This issue is explored in the empirical sections of the article. 13 A more detailed description of this process can be found in Section 3. the payouts of the two strategies converge. because the probability of Smax and Smin being interior points to the investment horizon increases. as the investment horizon shrinks. we use three-month options. = [ln(M/S) b2 = - (9) (r - 2 )T]/(orT1/2) (10) [-ln(M/S) (r IU2 a2)T]/(T"1/2). respectively. As the investment horizon lengthens. Consequently. We do not attempt to answer the question of which option portfolio best describes the returns of trendfollowing funds. as pointed out in Goldman et al. their returns are likely to be correlated. depending on the data sample used. Given these two considerations. even though the PTFS better describes trend-following strategies theoretically. alternative strategies to the PTFS can better replicate trend-following funds' returns empirically. it may be difficult to distinguish between the PMTS and the PTFS empirically. and (11) (12) b3 = [ln(M/S) (r + Here Q and M denote the minimum and maximum prices. the payout of the PMTS equals that of the PTFS if and only if Smax and Smin occur at the beginning and end of the period in any order.

However. SFE. Option data on the CME were purchased from the FIl and updated by the CME. the Paris InterbankOffer Rate (PIBOR) (MATIF). FTSE 100 (LIFFE). We use the other straddle to lock in the lowest price of the underlying asset by rolling the straddleto a lower strike whenever the price of the underlying asset moves below the currentstrike."4 14 An alternative replication strategy is a delta-hedging strategy using the underlying asset. and Swiss franc on the CME. 30-year Treasurybonds (CBOT). silver. Instead. we used the futures contracts on soybean. Futures and option data on the DTB. MATIF. German Bunds (LIFFE). CME.and Osaka were purchased from the Futures Industry Institute (FII).3-month Sterling (LIFFE). For currencies.S. we used the futures contracts on the S&P 500 (CME). Futures data on the CBOT. For bonds. Deutschemark.Appendix B provides information on the data. (1979). Euro-Deutsche Mark (LIFFE). we used the futures contracts on the British pound. and TIFFE and option data on the CBOT and NYMEX were supplied by the respective exchanges. we replicatedthe payout of a lookback straddleby rolling a pair of standardstraddles.so we canreturns.First. Futures and option data on the LIFFE. this latter straddle's strike must equal the low price achieved by the underlyingasset since inception. At expiration. lookback options are not exchange-traded not directly observe their prices. the pair of standard straddlesmust pay the difference between the maximum and minimum price achieved by the underlyingasset from inception to expiration. and NYMEX came from Datastream. we used the futures contracts on the 3-month Eurodollar (CME). For commodities.Thus.which is exactly the payout of the lookback straddle. the French 10-year GovernmentBond (MATIF). we need the implied volatility of the option to calculate its delta. we used the futures contracts on the U.Risk in Hedge Fund Strategies 2. Constructing a Performance Database of PTFSs To verify if the PTFS can mimic the performanceof trend followers. Japanese yen. We use one straddle to lock in the high price of the underlyingasset by rolling this straddleto a higher strike whenever the price of the underlying asset moves above the currentstrike. and corn futures traded on the CBOT and gold. Nikkei 225 (Osaka). For stock indices. First. The replication process calls for the purchase of two at-themoney straddlesat inception using standardputs and calls. and the AustralianAll OrdinaryIndex (SFE).At expiration. For three-monthinterest rates. Euro-Yen (TIFFE). a deltahedging strategy has two problems. wheat.as describedin Goldman et al. and the Australian 10-year Government Bond (SFE). 321 . UK Gilts (LIFFE). A numberof technical complicationsarose in the constructionof the PTFS contracts.and the AustralianBankersAcceptance Rate (SFE).DAX 30 (DTB). we generatedthe historical returnsof the PTFS applied to the most active markets in the world.this straddle'sstrikemust equal the highest price achieved by the underlying asset since inception. and crude oil tradedon the NYMEX.

S. as it requires dynamically changing the amount of the underlying asset every time its price changes. The monthly returns of the PTFS from rolling the straddles are summarized in Table 1. As lookback options are not traded. we formed five portfolios of straddles. a delta-hedging strategy can incur substantial transaction costs. the straddle-rollingstrategy should be implemented continuously if we were to match the assumptionsin Goldman et al. Finally. our straddle-rollingstrategy ignores the fact that many exchangetradedoptions are not European-styleoptions. This biases downwardthe returnsof the PTFS.as it requires tick-by-tick data. exchanges are American-style options. (1979) exactly. quarterlyexpirationstend to have longer history and larger volume. Their correlationmatrix is given in panel G of Table 1. currencies. we have to select the horizon of the lookback straddle. 322 .The Review of Financial Studies / v 14 n 2 2001 Second. the straddlerolls may generate additional cash flows when straddles are rolled from one strike price to another.In practice. In the case of commodity options. it may not perfectly replicate all the cash flows of the lookback straddle. It is also unclearto what extent it is feasible to simulate straddle rolls on a tick-by-tick basis. We then aggregatedthe daily returnsup to monthly returnsto match the standardreportinginterval for hedge funds.Even when monthly expirations are available. Second. three-monthinterest rates. we will have to make some assumptions to obtain an implied volatility. bonds.The choice is primarilydictatedby availabilityand liquidity of the options in our data set. which have higher prices than European-styleoptions. and commodities. and December.We included these cash flows in calculating the returnsof the straddle-rollingstrategy.A lookback straddlehas only two cash flows: an upfrontpremium at inception and a payout equal to the maximum range of the price of the underlying asset at expiration. we rolled the straddlesonly at the end of each trading day using the settlement prices of the options and the underlying assets. All the financialoptions in our data set have quarterlyexpirations. There is no problem with options on the LIFFE. though the strategy of rolling standardstraddlescan replicate the payout of the lookback straddle. Based on these return series. Most of the options tradedon U. we use the nearest-to-themoney options to approximateat-the-money options. Fifth. due to the asynchronous natureof options trading (at different strikes) and the potential distortionof bid-offer spreads. This is impractical. The straddle-rolling strategy will incur many fewer transactions. The error is likely to be small. the majority have expirations in March. In replicating this. Fourth. To compare results across markets. which are futures-styleoptions. Because exchange-traded options have discrete strikes. we frequently do not observe at-the-money options. September.one each for stocks. June. we used lookback straddles with three months to expirationas close to the end of a quarteras possible to maintainconsistency. Third. which are costly to purchase and time-consumingto process.

3513 0.0097 0.3047 2.0375 0.2657 0.0157 0.0783 -0.0449c 0.0060 -0.We startby comparingthe two types of straddles on two quarterlyoptions on the Japaneseyen futures contract.4286 1.3117 2.0435c Soybean -0.0304c 0.6884 -0.2685 1.5119 1.20a 31 Euro-Yen 0.93a 44 0.4129 1.21a German Bund 0.1627 0.0174 0.Risk in Hedge Fund Strategies Table 1 Statistical properties of primitive trend-following strategy (PTFS) returns for 26 markets and 5 portfolios Panel A: PTFS monthly returns for stock markets S&P 500 Mean SD Maximum Minimum Skewness % positive -0.74a 45 0.3167 1.52a 33 Before proceeding to compare the PTFS returnsto trend-followingfunds' returns.5699 -0.0174 -0.2774 2.2977 1.83a FTSE 100 -0.54a 33 0.4982 1.3716 2.8859 -0.0170 0.0496a 0.4391 1.3912 -0.19a Nikkei 225 -0.16a French IOY 0.43a Australian l0Y 0.4464 1.3077 1.0355" Crude Oil 0.2927 1.2039 -0.37a 30 0.0453 0.3560 -0.3914 1.0437c 0.0136 0.73a 41 Deutsche Mark 0.2703 1.87a 37 0.55a 40 UK Gilt 0.0161 0.9642 -0.9989 -0.48" 38 JapaneseYen 0.3992 1.2577 1.0750c 0.4412 -0.2285 0.2411 0.4780 2.0232 0.38a Australia Bankers Acceptance 0.88a 33 44 32 Panel B: PTFS monthly returns for government bond markets US 30Y Mean SD Maximum Minimum Skewness % positive 0.0455b Gold -0.4003 3.72" Paris Interbank Rate 0.3286 -0.4444 3.3070 1.71a 41 DAX 30 0.70a 45 36 50 Panel D: PTFS monthly retums for currency markets British Pound Mean SD Maximum Minimum Skewness % positive 0.2685 1.4433 1.1845 0.3433 1.0539a Silver -0.0952 -0.5408 -0.4950 2.5556 1.1573 -0.3881 0.0189 0.4066 2.8883 -0.3110 1.0502b SD Maximum Minimum Skewness % positive 0.33a 40 3-month Sterling 0.3503 1.59" 39 Euro-DM -0.we examine the empirical difference between the standardstraddle and the lookback straddle.3867 0.99a 48 Panel E: PTFS monthly returns for commodity markets Com Mean -0.3001 1.1063 -0.0177 0.72a 33 Australian All Ordinary -0.2333 1.0513c 0.2932 -0.7667 1.2775 1.7349 -0.0321c 0.4999 -0.0266 -0.4223 1.3372 1.0470 0.2351 0.93a 40 49 49 39 Panel C: PTFS monthly returns for three-month interest rate markets Euro-Dollar Mean SD Maximum Minimum Skewness % positive 0.0455c 0.5000 1.3495 1.1054 -0.2661 -0.2788 1.4545 2.2455 0.3978 1.5313 -0.0135 Wheat 0.67a 44 Swiss Franc 0. .2051 -0.

00 Bond PTFS 0.1867 Currency PTFS 0. We initiated the straddles at the end of November 1993.06 0.0260). At that time.62a 0. aStatisticallydifferentfrom zero at the 1% one-tailed test.0072 0. These two graphs show that the two straddles can have different payouts over a given obser- 324 .4739 -0.16 0.8878 in early January 1994.6444 near the end of June.0193 0.2573 1.The Review of Financial Studies / v 14 n 2 2001 Table 1 (continued) Panel F: Monthly returns for trend followers Trend-Following Funds Mean SD 0. and then rose to a high of 0.9780 by mid-February.1310 Stock PTFS -0.7147 at the expiration of the contract.2094 0.0137a and five PTFS portfolios (1989-97) Bond PTFS 0. using the September 1990 Japanese yen futures contract. and ended at 0. It declined to a low of 0.1573 Interest Rate PTFS 0.0195 0.19a Panel G: Correlation matrix of the five PTFS portfolios Stock PTFS Stock PTFS Bond PTFS Interest rate PTFS Currency PTFS Commodity PTFS 1.6413 -0.21 0. we initiated the straddles at the end of May 1990.0006 -0. the payout of the lookback straddle (0.07 0.3240 -0. bStatisticallydifferentfrom zero at the 5% one-tailed test. using the March 1994 Japanese yen contract.0703) was much closer to that of the standard straddle (0.36 1.18 1.37 1.0491 Maximum Minimum Skewness 0.0177 0. %Positiverefers to the percentageof months with positive returns. It declined to a low of 0. The second comparison is graphed in Figure 3.9459 when the contract expired in the middle of March.00 Currency PTFS 0.12 0.0181 0. The sample periods for each marketis given in Appendix B.00 Interest Rate PTFS 0.0902) was substantially greater than that of the standard straddle (0.79a 1. rose to a high of 0.3013 1.00 Commodity PTFS 0.0820 0.0556).32 1.2305 Commodity PTFS -0.07a 0.8158 -0. In this case. so the payout of the lookback straddle (0. approximately three months prior to expiration.The five PTFS portfolios are equally weighted portfolios of the PTFSs in the five groups of markets(panels A throughE).00 The primitive trend-followingstrategy (PTFS) is a long position on three-monthlookback straddles. As the contract's minimum and maximum prices occurred in the middle of the observation period. the contract's minimum and maximum prices occurred near the beginning and the end of the observation period. the September yen futures price was 0. Like the first comparison. The first comparison is graphed in Figure 2. approximately three months prior to expiration. 'Statistically differentfrom zero at the 10% one-tailed test. the March yen futures price was 0.2497 1. At that time.46a 1.1837 -0. Trend-followingfunds' returns are based on an equally weighted portfolio of 407 defunct and operatingcommodity funds that had significantcorrelationwith the first principalcomponentfrom a principalcomponent analysis of 1304 defunct and operatingcommodity funds.2285 1.6591.68a 0.37 0.5172 2.9199.

62 150 0 0.72 300 0.00 180 0..84 11130193 0 12/30193 Time 1130/94 2128/94 Figure 2 Standardized cumulative returns of the lookback straddle and the standard straddle on the March 1994 0.80 50 6130/90 Tine 7131/90 8/31/90 5131190 Figure 3 Standardized cumulative returns of the lookhack straddle and the standard straddle on the September 1990 Japanese yen futures contract 325 .98 160 0.... JY FuturesPrice(LHS)- 40 20 0. W3/31319 .Risk in Hedge Fund Strategies 1.96 0.8 0.Standard Straddle(RHS) 0.88 11130193~~~.86 Looback Straddle(RHS) .70 250 a0. 300 0.94 Time 0.92 YFtrsPc LS-obc ~~~80~ tade(H)---Sadr ~~~~~ tade(H) /89 120 = ..68820- U-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~U (n0.669-i \110 Japanese yen futures contract 0x 0...

despite the theoretical differences in the strategies. we update the results of Fung and Hsieh (1997b) using the Tass CTA database as of March 1998. or operating. This is consistent with the notion that trend-following funds have systematic risks. in most cases. Fung and Hsieh (1997b) found a single dominant trading style.86. we are also empirically constrained to use monthly returns as higher frequency observations on the performance of trend-following funds are limited. 3. Fortunately. This dominant style was interpreted to be a trend-following style. and their monthly returns had a correlation of 0."5 The returns of the equally weighted Fung and Hsieh (1997b) noted that the inclusion of defunct funds helps guard against "survivorship bias" in their estimate of the returns of the trend-following trading style.39. The daily returns of the two types of straddles on the Japanese yen had a correlation of 0. we begin with a representative series of trend-following funds' returns. We are empirically constrained to use quarterly options because data for longer-dated options are generally unreliable and. funds are used to estimate the returns of a group of funds. 3.The Review of Financial Studies / v 14 n 2 2001 vation period. because the omitted defunct funds generally have poorer performance than surviving funds. even though this is not necessarily so at a different return interval. we apply the lookback straddle in our empirical tests given its superior theoretical properties. In this article. This may require a tailor-made benchmark for each fund. the difference between the PMTS and the PTFS may be hard to discern. In addition. we examine the entire data sample from March 1986 to December 1997. unavailable. contrary to the prediction of linear-factor models applied to standard asset benchmarks. '5 326 . This indicates that. Theoretically. in our empirical application using monthly returns. Survivorship bias comes about when only surviving. different trend-following funds may use different trading strategies. Applying principal components analysis on all defunct and operating CTA funds. Evaluating the Risk in Trend-Following Strategies In this section we show that the returns of trend-following funds are strongly correlated with the returns of the PTFSs. It is an empirical regularity that the standard straddle and the lookback straddle are highly correlated in our data sample. as shown in Fung and Hsieh (1997b). depending on when the maximum and minimum prices were reached. This is likely to result in an upward bias. Consequently. Next. 407 are strongly correlated to the first principal component. based on extensive interviews with the manager. which is the most popular self-described CTA trading style. This is a consequence of using monthly returns of options that expire quarterly.1 Standard benchmarks do not explain trend-following funds' returns To explore this issue. Out of 1304 defunct and operating CTA funds. there is a high degree of commonality in the returns of trend-following funds.

3. one-month Eurodollar) Five major stock indices (S&P 500.1 3.3m Sterling.9 k2 refers to adjusted R2 of the regressions of trend-followingfunds' returnson ten different sets of risk factors. Emerging marketequities. CurrencyPTFS.S. trend-following funds' returns have a higher mean and greater statistical significance than the PTFS returns.S.S. -1.1 6. French 10-year. Paris InterbankRate) Four currencymarkets (British pound. The regression of trend-following funds' returns against the eight major asset classes Table 2 Explaining trend-following funds' returns: The R2s of regressions on ten sets of risk factors Sets of Risk Factors 1. Bond PTFS.S. The difference is that trend-following funds' returns have a positive and statistically significant mean. whereas the PTFS portfolios and most of the individual PTFSs do not.Risk in Hedge Fund Strategies portfolio of these 407 funds are used as the representative trend-following funds' returns. 8. -2. 10.0 2. Australian All Ordinary) Five governmentbond markets (U. wheat.5 4.5 5. Eight major asset classes in Fung and Hsieh (1997a) (U. We start with a key distributional feature of trend-following funds' returns. We defer further analysis of this implicit alpha in trend-following funds' returns until Section 4. 9. DAX 30.7 -0. U. 7.2 Lookback straddle benchmarks explain trend-following returns funds' Next. 1. The returns of the five PTFS portfolios as well as all the individual PTFSs are also strongly positively skewed. three-monthinterest rate PTFS. With the exception of the PTFS for the Swiss franc. crude oil. gold. we document the apparent lack of systematic risk in trend-following funds' returns in standard linear-factor models in Table 2.Australian10-year) Six three-monthinterest rate markets (Eurodollar.2 -0. dollar index.S.S. FTSE 100. bonds.8 7. 327 . Nikkei 225. U. GermanBund. soybean. UK Gilt. Euro-Yen. silver) Goldman Sachs Commodity Index Commodity Research Bureau Index Mount Lucas/BARRATrend-FollowingIndex Five PTFS portfolios (Stock PTFS. Table 1 shows that the trend-following funds' returns have strongly positively skewed returns. equities.5 47. non-U. Euro-DM. equities. 7. gold. bonds. Swiss franc) Six commodity markets (coin. AustralianBankersAcceptance. Commodity PTFS) R2 of Regression (%) 1. non-U. deutschemark. 30-year. -3.Japaneseyen.

These results are summarizedin Table 2.the currencyPTFS also makes large profits.S. gold. with currenciesand commoditieshaving the largest explanatory power.The Review of Financial Studies / v 14 n 2 2001 (U. U. Doing this by groups. and the six commodities -3. In fact.S. currencies.1%. The first panel graphs the trend-following funds' returns against the stock PTFS returns. equities. three-monthinterest rates. with an R2 of 7.9%.Plots of the residuals of the regressionagainstthe explanatoryvariables(availablefrom the authors on request) do not show any remainingnonlinearrelationships. This indicates that trend followers do have systematic risks. the panel for currencyPTFS contains the most strikingscatterplot.5%. These risks are just not evident in a linear-factormodel applied to standardasset benchmarks. It basically shows that there is no apparentrelationshipbetween these two variableseither in the center of the distributionor at the extremes.8% and is also not significant. the five bond markets 7. bonds. U. bonds. one-month Eurodollar interest rate. The F-test that none of the PTFS portfolios is correlatedwith the trend-following funds' returns is rejected at any conventional significance level.We do so using scatterplots of the trend-followingfunds' returns againstthe five PTFS portfolios in the five panels in Figure 4. The Commodity Research Bureau (CRB) Index has an R2 of -0. we document the PTFS's ability to characterizethe performanceof trendfollowers using standardlinear statisticaltechniques.S. respectively. The regression of trend-following funds' returnson the GSCI Total Return Index has an R2 of -0.They show that the currencyand commodity PTFSs have the strongestrelationshipwith trend followers. Dollar Index. the four currencies -1.2%.0%.7% and is not statistically significant.1%.5%. Next. In particular. the six three-monthinterest rates 1. non-U. and it is statistically significant.S. Proper diagnostics are needed to guard against nonlinearrelationshipsin regressions. and emerging market equities) in Fung and Hsieh (1997a) has an R2 of 1. For completeness. Other indices frequentlyassociated with commodity tradersand trendfollowers produced similarly poor results. 328 .5%.S. non-U. when trendfollowers make large profits. It shows that. equities. the five stock marketshave an R2 of -2. An investor using a linear-factormodel on standard asset benchmarkswould have concluded that trend followers had no systematic risk.The regressionof the trend-followingfunds' returnson the five PTFS portfolios returnshas an k2of 47. and none of the variables are statistically significant. and commodities. The other panels in Figure 4 are the scatter plots of the trend-followingfunds' returns against the PTFS portfolios in bonds. we examined the 26 markets used to construct the PTFSs in Table 1. the relationshipbetween the currencyPTFS returnsand the trend-followingfunds' returnsappearreasonablylinear. The Mount Lucas/BARRA Trend-FollowingIndex is slightly better.

Theoretically.60%.16 It is perhaps not surprisingthat the PTFSs in bonds. which declined on average 4. and commodities during these states. The result is reportedin Table 4. State 1 represents the worst 30 months for world equities.it was not surprising that the PTFSs for stocks had positive returnsduringthese two down periods for equities. The PTFSs on bonds. 1997b) that the returnsof trend followers have option-like payouts relative to world equity markets.2% during these two periods. As expected. and currencies also had high and positive average returnsin states 1 and 5. and currencies) should respond to the same set of 16 The low average return for the commodity PTFS in state 1 was primarily due to one outlier. threemonth interest rates. based on the performanceof the world equity market.9%).Risk in Hedge Fund Strategies 3.4%) andAug-Sep 1990 (-18. which rose 6. respectively. bonds. and currenciesgeneratedhigh returnsduring extreme moves in the world equity markets. in which equities gained 3. three-monthinterest rates.and commodities.0% and 10. are: Sep-Nov 1987 (-20. Trend-following funds recordedlarge positive returnsof 13.59% per month.the returnsof the PTFSs tended to be high during extreme states 1 and 5 for stocks. three-monthinterest rates. as measured by the MS World Equity index. when equities fell an average of 0. However.3 Trend-following funds' returns are sensitive to large moves in world equity markets Next we examine anotherimportantcharacteristicof trend-followingfunds' returnscorrespondingto extreme moves in world equity markets.The two worst periods for world equities in the last 10 years. bonds. It was first observed in Fung and Hsieh (1997a.74% on average. State 2 consists of the next 30 worst months. financial assets (stocks. State 4 are the next best 30 months.State 3 contains the remaining34 months. However.36% on average. three-monthinterest rates. The less obvious results were the positive returnsfrom the PTFSs for most of the governmentbonds. Here we divided the longer time series for which there were data for all PTFSs (March 1985 to December 1997) into 5 states. 329 .we provide furthercollaboratingevidence on this relationshipbetween the returnsof trendfollowers and those of the world equity marketsover a large sample period. where we report the trend followers' sizable positive returns during downturnsin world equity markets. currencies.We begin with Table 3. To generalize these unusual returncharacteristics. less than half of the PTFSs for three-monthinterest rates were profitableduring these two periods. currencies.We extend this observationto encompass a largerdata set using the returnsfrom the PTFSs. We then report the average returnand the standarddeviation of the PTFSs for stocks. Given the lookback option's structure. Consistentwith the earlier observation. State 5 has the best 30 months for equities. the PTFSs on commodities did not exhibit this option-like behavior. the PTFSs on stocks had high and positive average returnsduring the two extreme states.

* * 20% * *-* -10% -5% 0% 5% Trend-Following Funds' Retums 10% 20% Figure 4 Scatter plots of monthly trend-following funds' returns versus five PTFS portfolio returns 330 .The Review of Financial Studies / v 14 n 2 2001 150% 5o%r E 100% 2~~~~~~ e 50% *g * -50% -100% .10% -5% 0% 5%Trend-Following Funds'Retums 1O% i5% 2 60% 50% 40% 30%*- & 20% -10 % 5 0 5 * 10 0 0% - * - * -10%~~~~~ -20% * -30% -10% -5% 0%/ 5% Trend-Following Funds' Retums i 0a/ 15S 20 100% 6 0% * .

.and the Sydney 90-day BankersAcceptance futures contract.. Historically.. lookback straddlereturnshad a correlationof 0..444 in the case of lookback 331 .being the least correlated. Both straddles had statistically significant correlationto trend-following funds' returns..We investigated this for two markets:the Swiss franc futures contract because its PTFS returns were most highly correlatedto trend-followingfunds' returns. we examine the possibility of using standard straddles as an alternative option-replicationstrategy. extreme moves tended to be caused by a common set of dramaticevents (such as the Persian Gulf War in 1990 and the attendantshock to the oil markets).. * .. -10% -5% 0% 5% Trend-$ollowing Funds' Retums 10% 15% 20% Figure 4 (continued) macroeconomicevents.0. To conclude this subsection..Risk in Hedge Fund Strategies 70% 60%. .834 with those of standardstraddles. For the Swiss franc. 50% 40% 30% ~20% c30% .leading all financialmarketsto move in concert. -30% -10% -0% 0% 5% Trend-Following Funds' Retums 10% 15% 20% 120% 100%E 80% 20% * | *u0 **I -20% * -40%. -10% -0% *- *** *5 ..

3187 0. lookback straddlereturnshad a correlationof 0.1889 Trend-Following Funds 0.however. we regressedthe trend-following funds' returnson the PTFSs' returnsduringthe extreme states 1 and 5.0882 Swiss Franc 0.2040 Soybean 0.5165 Panel F: PTFS returns for commodity markets Period 8709-8711 9008-9009 Commodity PTFS 0.1257 French 1OY 1.0753 -0. These results indicate that. for quarterlyexpirations.Neither straddle.4 Preferred habitat of trend followers Next we address the question of preferredhabitat or which markets trend followers were active in duringthe extreme equity marketmoves (i.3670 UK Gilt 0.006 in the case of lookback straddles and 0.2725 0.Our choice of the lookback straddlein the empirical applicationrests on its superiortheoreticalpropertiesgiven in Section 2.010 in the case of standard straddles. states 1 and 5 in Table4).4336 Wheat 0.1372 0.9578 Silver 0.0513 GermanBund 0.0052 -0.1060 0.1494 0. To answerthis question.0377 Australian 10Y Panel D: PTFS returns for three-month interest rate markets Period 8709-8711 9008-9009 Australia Paris Bankers Interbank Interest Rates Rate PTFS Euro-Dollar 3-month Sterling Euro-DM Euro-Yen Acceptance 0.e.The Review of Financial Studies / v 14 n 2 2001 Table 3 Large PTFS returns during the two worst periods for world equities: Sep-Nov 1987 and Aug-Sep 1990 Panel A: World equities and trend-following funds Period 8709-8711 9008-9009 MS World Equities -0.8194 -0.3368 0.7758 0.8344 -0. Given 332 .0599 JapaneseYen 0.7448 0.1299 0.924 with those of standardstraddles.1719 straddles and 0.1832 British Pound 1.standardstraddlesare similar to lookback straddlesfor the purpose of replicatingmonthly trend-followingfunds' returns.8525 Corn 0.7744 FTSE 100 DAX 30 Nikkei 225 AustralianAll Ordinary Panel C: PTFS returns for bond markets Period 8709-8711 9008-9009 Bond PTFS 0.433 for standardstraddles.0898 -0.3557 US 30Y 0.2891 Crude Oil -0. had statistically significant correlationwith trend-following funds' returns. 3..3871 5.0220 0.3937 0.4268 Panel E: PTFS retums for currency markets Period 8709-8711 9008-9009 CurrencyPTFS 0.7744 S&P 500 2.1019 Panel B: PTFS returns for stock markets Period 8709-8711 9008-9009 Stock PTFS 2. For the Sydney 90-day Bankers Acceptance.1887 Deutsche Dark 0.4406 Gold -0.3842 0.8194 -0. -.5401 -0.1060 0.3145 0.2036 -0.

0183 0.0076 0.For commodity PTFSs.5% trade commodities. bonds. Figure 5 provides a scatterplot of the fitted values of the regressionagainst the returns of trend-followingfunds. is the K2 9.7% with U.0039 State 5 consists of the worst 30 months of the MS WorldEquity Index.0362 0. 30. however.5%.1281 0.0108 0.0014 0.0355 0.0484 -0. statistical inference is not reliable. is the natureof ex post performance attribution.0518 0.0156 0. State 3 consists of the remaining30 months of the MS World Equity Index.0188 0. the R2 is 21.Risk in Hedge Fund Strategies Table 4 Option-like behavior of PTFS returns during five different states for equities: Mar 85-Dec 97 Interest Commodity Trend Following MS World State Stock PTFS Bond PTFS CurrencyPTFS Rate PTFS Funds PTFS Equities 1 2 3 4 5 0.1% with U.0059 0.0336 0.0245 0. 3.0058 0.0307 0.1017 0.0130 0.0439 0.0012 0.0149 -0.0009 0.0098 0.0674 0.5% with the deutschemark the Japaneseyen is being the significantvariables.9% trade currencies.0375 0.As Billingsley and Chance (1996) 333 .For and currencyPTFSs.0928 -0.0267 0.0375 0.0114 0.0832 0.we ran the regression five times using groups of PTFSs. State 2 consists of the next worst 30 months of the MS World Equity Index.0407 -0.0176 -0.0068 0. As the regressors were selected based on previous regressions.The final regressionis reportedin panelA of Table5.0470 0.0454 0.0264 -0. ex post.0941 0.0211 0. State 4 consists of the next best 30 months of the MS World Equity Index.0588 0.0323 -0.0237 -0.0112 0.S. the large numberof PTFSs and the small numberof observations. and silver being the is significant variables. For the three-monthinterest rate PTFSs.0453 0.0309 -0. where the significant variables are the Eurodollarand Short Sterling contracts. In the stock PTFS regression.0821 0.4% trade stock index futures. can account for trend followers' performanceduring extreme equity marketmoves.0560 0.0132 0.0203 0.where data-mining techniques are applied to determine which marketstrend followers were active in.0832 0.0284 -0.0306 0.0553 0.0591 0. These are the markets that.0816 0.2% trade bonds and threemonth interest rate futures. Billingsley and Chance (1996) found that among the CTAs trading only specialized markets.0146 0.0010 0.0244 -0.For bond PTFSs.0253 0.0121 0. using all five PTFSs (in panel B) and three statistically significantPTFSs (in panel C).0056 -0.0170 0. and 12.0436 0.0321 -0.0569 0.0827 0.4%.0390 -0. 15.5 Relationship to other empirical studies To complete our analysis.5% where the wheat and silver futurescontractswere the significantvariables. State 5 consists of the best 34 months of the MS World Equity Index. Standarderrors are its italics. the K2 30.wheat. This. bonds being the only significant variable. the K2is 10. 41. the K2 60.0246 -0. and none of the equity indices is statistically significant. deutschemark. the R2 is 39. it is helpful to incorporatequalitativeresults from other independentstudies. Using only the significantPTFSs. Table5 also providesinformationon the regressions for the overall sample.0460 0.0483 0.S.

0234 0. Although it is difficult to expect qualitative results to line up closely with quantitative observations.02468 k2 DW X2(5) Panel C: Regression of trend-following funds' returns on selected PTFS portfolios using the full sample Constant PTFS on stocks PTFS on bonds PTFS on currencies PTFS on interest rates PTFS on commodities R2 0.02096 0.13913 0.0513 0. See the note for Table 2 for the definition of markets.0255 0.10308 0.0180 0.05164 0.00332 0.44 (p-value: 0.01575 0.0169 0.02413 0. we assume that the group is evenly divided between the two instruments.01155 -0.0204 0.10994 -0.01713 0.03517 0.0000) 0.0128 Panel B: Regression of trend-following funds' returns on five PTFS portfolios using the full sample Constant PTFS on stocks PTFS on bonds PTFS on currencies PTFS on interest rates PTFS on commodities R2 0. did not report the split between bonds and three-monthinterest rates.0336 0.4788 2.presumably.0166 0.01594 0.0544 0.The Review of Financial Studies / v 14 n 2 2001 Table 5 Estimating trend-following funds' preferred babitat: January 1989-December 1997 Regressors Coefficient Estimate StandardErrorsa Panel A: Regression of trend-following funds' returns on selected PTFSs during two extreme states (1 and 5) for world equities Constant US bond Euro-$ Short sterling DM JY Wheat Silver R2 k2 0.5032 0.This means that the currencymarketis the most popularmarketfor trend followers attracting.0052 0.01229 0.607 0.50 (p-value: 0.0540 0.4666 2.0478 0.0000) 0. 'With correction for heteroskedasticity.0194 0.02933 0.4816 0.0123 0.31 163.00312 0.0259 0.02077 0.14999 0.674 0.02495 0. whereas the equity market is the least popular.the lion's share of available risk capital.28 123.02657 k5 DW X2(3) See the note for Table I for the definition of PTFS. 334 . and stock indices attractedthe least trend-following activities. both approaches came to a similar conclusion: Currencies appeared to be the instrumentof choice.

The distribution of iR2 are given in the third column of Table 6. and 62 to commodities. There can be. we assessed the ability of standard asset benchmarks to explain returns of individual funds.e. fonned from their benchmark (i. however. Five funds had two significant coefficients. 33 to the 335 . The distribution of iR2 are given in the second column of Table 6. Benchmarking Individual Trend-Following Funds So far. with an average of 11%. three-month interest rates. We investigate individual funds in this section. Eighty-six funds had no regression coefficients significant at the 1% level. As a starting point. Seventy-two funds had one significant coefficient: 2 funds had exposure to stocks. First. and commodities.Risk in Hedge Fund Strategies 20% 15% 0 00 -10% 0 0 -1% 0% 5 10% 0 -10% 05% 00 Funds' 0 Returns 0 0 Trend-Following Figure 5 Scatter plots of monthly trend-following fund's returns versus PTFS replication portfolio returns 4. Next. we focus on 163 trend-following funds with at least 24 months of return information through the end of 1997. 1 to currencies. They ranged from -2% to 64%. with an average of 24%. we regressed the returns of each fund on the five PTFS portfolios. Ninety-six funds had one significant coefficient: 12 had exposure to the bond PTFS. We regressed each fund's returns on five portfolios of stocks.. we have examined the return characteristics of a portfolio of 407 trend-following funds. cuffencies. 7 to bonds. buy-and-hold) returns rather than the PTFS returns. They ranged from -9% to 58%. wide individual variations not reflected in this portfolio that merit documentation. Thirty-nine funds had no regression coefficients significant at the 1% level. bonds.

8 -0.The Review of Financial Studies / v 14 n 2 2001 Table 6 Explaining the monthly returns of 163 trend-following funds using the five PTFS portfolios and five buy-and-hold portfolios Number of Trend-FollowingFunds K2from.1 0. -0. to -1.0 0.9 Regressions Using 5 PTFSs Portfolios 0 Regressions Using 5 Buy-and-HoldPortfolios 0 -0. 0.4. Last.9.5 0. 11 to the bond and commodity PTFSs. three-month interestrate. These results indicate that the PTFS returns(particularlybond.9 0. The currencyPTFS exposure ranged from 3% to 44%.We examined 21 trend-following funds whose names imply they trade only currencies.2.2. and 51 to the commodity PTFS. averaging20%. we demonstratethat the PTFSs can provide reasonable results for identifying the preferredhabitatof traders.7 -0. 0.3. 0.3.4 -0. 0.-0. -0.5. currency PTFS. -0. 18 had statistically significant exposure to the currencyPTFS only.9. Last. -0. one fund had exposure to the bond.2 0. -0.6.7. the statistically significant exposure to the bond PTFS rangedfrom 4% to 65%.4. -0. 0. 0.6.8 0. 2 had exposure to the currencyPTFS along with either the three-monthinterest rate PTFS or the commodity PTFS. In terms of magnitudes. currency. 0. averaging26%.0 0 0 0 0 0 0 0 0 3 21 43 44 26 17 7 2 0 0 0 0 0 0 0 0 0 0 3 26 53 44 28 8 1 0 0 0 0 0 The R2s are based on regressions of 163 trend-followingfunds with 24 months of returnson the five PTFS portfolios and on five buy-and-holdportfolios based on the underlying marketsof the PTFS portfolios. 0.2 -0.0. -0.6 0.0. averaging20%. The commodity PTFS exposure rangedfrom 9% to 87%. and commodity) had much higher explanatory power than the benchmark asset returnseven at the level of individual trend-followingfunds.5 -0. Of these. 0. averaging 16%.3 0.3 -0.4 0. 1.6 -0. and commodity PTFSs.8. and 1 fund had no significantexposure to any PTFSs. -0.7 0. 1.1 -0. 0.5.1. and 1 to the currencyand three-month interest rate PTFS. The three-monthinterestrate PTFS exposure ranged from 18% to 22%. 10 to the currencyand commodity PTFSs. Twenty-seven funds had two regression coefficients significant at the 1% level: 5 were exposed to the bond and currencyPTFSs.7. It is worth noting that no fund had any significantexposure to the stock PTFS. They can 336 .8.

the results also indicate substantial differences in preferred habitats across trend-following funds. we label this the PrimitiveTrend-FollowingStrategy (PTFS) for that market. it would be difficult to find a single benchmarkfor monitoring the performanceof trend-following funds. Both tend to have positive returnsduring extreme up and down moves in the world equity markets. we show that these PTFSs capturethree essential performancefeatures of trend-followingfunds. when the S&P declined more than 10% and the 337 . this risk cannot be observed in the context of a linear-factormodel applied to standardasset benchmarks. but not the PTFSs on stock indices. The cost of implementing this strategy can be established using observable. In light of these results. One implication is that trend-following funds do have systematic risk. In fact. First. or a portfolio of lookback straddleson currencies. Second. For each asset market.The second implication is that trend followers. in determiningwhether a fundspecific benchmarkis necessary.the superiorexplanatorypower of the PTFSs over standard buy-andhold benchmarkssupportsour contentionthat trendfollowers have nonlinear. Glosten and Jagannathan (1994) recommendedextensive discussions with each fund manager to understandhow he or she operates. bonds. 5.Risk in Hedge Fund Strategies also help in performanceattribution. The implications of these performancefeatures are threefold. This is in agreementwith qualitativeresults in previous studies that indicate that stock indices are the least popular market to CTAs. However. three-month interest rates (Eurodollarand Short Sterling).They both have strong positive skewness.bonds. exchange-tradedoption prices. can reduce the volatility of a typical stock and bond portfolio duringextreme marketdownturns. Nonetheless.This strategy delivers the performanceof a perfect foresight trend follower. Third. as well as benchmarksused by the hedge fund industry. In addition. and U. option-liketradingstrategies." in the sense that they tend to deliver positive performancein extreme marketenvironments.However. trend followers tend to performas if they are long "volatility"and "marketevent risk. trend-followingfunds' returnsduring extreme marketmoves can be explained by a combination of PTFSs on currencies (deutschemarkand Japanese yen). the PTFS returns replicate key features of trend-following funds' returns. Conclusions In this article. we createda simple trend-followingstrategyusing a lookback straddle. commodities (wheat and silver).Empirically. our model contributesto the design of benchmarksfor trend-following funds by capturingthe nonlinear dynamics of their strategy. the PTFSs are better able to explain trend-followingfunds' returnsthan standardbuy-and-holdbenchmarkreturnson major asset classes. and commodities.S.Specifically.This view is corroborated the out-of-sample events in by the third quarterof 1998.

Consider a standardstraddleand a lookback straddlethat were purchased when the asset price was 100. Here. the asset price falls steadily from 100 to 40 over the life of the straddles. as well as any fund that uses trend-following strategies. In scenario C. Appendix A: The Illustrated Difference between the Deltas of the Lookback Straddle and the Standard Straddle This appendixcomparesthe replicationstrategyof the standardstraddleand the lookback straddle to gain furtherinsights into the difference between markettiming and trend following. Throughoutthis entire period. resembling a trend follower selling breakdown. Nonetheless. both were at-the-moneyoptions. in strongly trending markets. resembling a trend follower buying breakouts. the evidence indicates that there are substantial differences in trading strategies among trend-following funds. The delta in fact turns negative as the asset price returnsto 100. When the asset price declines from 130 toward 110. is assumed to be 20%.When the asset price subsequentlyrises from 110. the delta of the lookback straddle declines rapidly as the asset price falls back from the high of 130. a. and rose to 150. the delta of the lookback straddleturnspositive and rises sharplyonce more. as shown in panel D. Suppose the asset price stays at 100. the benchmarking of an individual fund's performance may need to incorporate specific aspects of the manager's operations. the payouts of the two straddlesare identical. These two cases show that. the two straddleshave different deltas but the same payouts. fell back to 110. Let S denote the price of the underlyingasset. Thus. is 6%.its delta can be calculated without knowing the path of the so underlyingasset's price.The Review of Financial Studies / v 14 n 2 2001 vast majority of trend-following funds made large gains. shown in panel B. However. the asset price rises steadily from 100 to 160 over the life of the straddles. Its instantaneousvolatility. Because the standardstraddle's payout is not path-dependent. Consider the following cases. past 130.Note that the payouts of the two straddles are also different: The standardstraddle pays out 0. 338 . It is similar to the delta of the standardstraddlewhen the asset price rises from 100 to' 130 for the first time. However. resembling a trendfollower selling breakdowns. the two straddleshave differentdeltas and differentpayouts. the delta of the lookback straddledeclines sharply and actually turns negative. At inception. the asset price rises to 130 and falls back to 100.The deltas of the two straddlesare similar and rise with the price. The interest rate. In scenario A. to 150. The delta of the standardstraddle stays positive over the entire period.the two straddlesare virtually identical. the lookback straddle'sdelta is quite different. The third implication is that the PTFSs are key building blocks for the construction of a performance benchmark for trend-following funds. their strike prices were 100. illustratedin panel C. However. the standardstraddlehas a positive delta. r. We consider their deltas in four more scenarios graphedin Figure 6. Both options have 60 days to expiration. and both payouts are also zero. Also. Here. In contrast. the lookback straddle'spayout is path-dependent. its delta depends on the ex post range of the underlyingasset's price. it may not be possible to find a single benchmark that can be used to monitor the performance of all trend followers.Again. Then the deltas of the standardstraddleand the lookback straddleare zero. In scenario D. the PTFSs are useful tools for the construction of benchmarks for trend-following funds. while the lookback straddlepays out 30. In scenario B. that is. the asset price rises to 130. illustratedin panel A. As suggested in Glosten and Jagannathan (1994). the two straddleshave similar deltas and identical payouts.

Standrd(Staddle-LHSLpSna(RdHOS)adl(LS--Ldac ae(H) .-. ofthe.5 1 1...Risk in Hedge Fund Strategies Delts 6 -1.0 1.0.Lookback SO4dd10 4 (RHS) -Lookbck tradle-LHSkbo - Pricdl(RHS) (LHS)-Price(RHS) A 11 an 1 11.......sraddetannloobackstradlelnesenaros _os H B.tandrd. 10 o0OoNsbs)OyS4 tadrdSradl -1. Doootheundelyinrasst's llOl ricepat Stadc(LHSHS . 15 *H-Lokac s0 11~~~~~~~~~~~~-1(1(00 go~~~~~~~No6llooio 100 .5 D0o5 I --Slandard Figure 2 6RH) tnadStrad4dl HS) o .0 1. SO40Ile(LHS) 1 -Price okbackHS .....S0oldolOSt002d.5120 0 110de(LS 400 130 339 ...

495 368. CME. and Osaka were purchased from the Futures Industry Institute (FII). dPortions of data missing during 1987 and 1988.The following table provides informationon the option data. Futures and option data on the LIFFE.065 213.385 45. 'Portions of data missing during 1993.275 519.785 162.622 227.680 70.665 118. and option data on the CBOT and NYMEX. The traderis long (short) when the currentprice is near the maximum (minimum) price.970 80. and NYMEX came from Datastream.247 443.483 406. and TIFFE. 340 . of Observations Option Contract S&P 500 FTSE 100 DAX 30 Nikkei 225 All Ordinary 30-yearUS bond 10-year Gilts 10-year Bund 10-year French bd 10-yearAus.The Review of Financial Studies / v 14 n 2 2001 To summarize. MATIF.894 214.070 SFE FII FIT& CME FII & CME FII & CME FII & CME CBOT CBOT CBOT NYMEX NYMEX NYMEX 85/02/25c 85/02/27 88/01/04 84/10/31 86/11/14 82/10/04 88/01/04 'All samples end on December 31.960 56.970 71.883 110.473 52.295 153. Acc.679 286.693 348.398 68. Appendix B: Data Description and Data Sources Futures and option data on the DTB. Option data on the CME were purchasedfrom the FII and updatedby the CME.497 333. The lookback straddle's delta mimics a trader whose actions depend on the relationship between the current price to the maximum and minimum prices since inception.352 341.700 26. Futuresdata on the CBOT. bData missing from many dates.124 56.800 63.015 120.410 91.505 267.558 324. PIBOR British pound Deutschemark Japaneseyen Swiss franc Coin Wheat Soybean Crude oil Gold Silver SFE MATIF CME CME CME CME CBOT CBOT CBOT NYMEX NYMEX NYMEX 92/01/02 90/03/01 85/03/01 84/01/24c 86/03/06d 72 84 154 168 142 149 155 120 158 134 180 120 81.250 15.200 230.175 269.651 245. This is what a market timer would do.469 359.338 408. The traderis long if the currentprice is above the inception price and short otherwise.309 622.282 309.470 Options 497.335 70. No. were supplied by the respective exchanges. SFE.368 Source FII & CME LIFFE FII FII SFE CBOT LIFFE LIFFE FII SFE FII & CME LIFFE LIFFE TIFFE Aus.the standardstraddle'sdelta mimics a traderwhose actions depend only on the relationshipbetween the currentprice and the inception price of 100.348 332. NA = not applicablefor cash options. 1997. but not any intermediate prices. Bank.560 160.525 38.310 NA NA 51.545 61.090 626.545 34. This is what a trend follower would do.714 94. bd Euro-$ Short Sterling Euro-DM Euro-Yen Exchange CME LIFFE DTB Osaka SFE CBOT LIFFE LIFFE MATIF SFE CME LIFFE LIFFE TIFFE Start Date' 83/01/28 92/03/13 92/01/02 89/06/12b 92/01/02 82/10/04 86/03/13 89/05/02 90/01/02 92/01/02 85/03/21 87/12/01 90/03/01 91/08/01b Month 180 70 72 96 72 183 142 104 92 72 154 121 94 77 Futures 79.578 463.

. Fung. Edward. and S. Fung. 1999. "The Performance of Mutual Funds in the Period 1945-1964. A. 377-421. "A Contingent Claim Approach to Performance Evaluation." Review of Finiancial Studies. 1979. Ross. "Survivorship Bias and Investment Style in the Returns of CTAs. 7-26. 16. Billingsley. A. 1999.. M." working paper. 363-407. H." Journal of Finance. 131-136.. Yale University. "Do Managed Futures Make Good Investments?" Jourtnal of Flutures Markets. 7. Pulvino. "On Market Timing and Investment Performance II: Statistical Procedures for Evaluating Forecasting Skills." Journlal of Enmpirical Finance. 1997b. and K.Risk in Hedge Fund Strategies References Alexander. P. 55. 1994. "Path Dependent Options: 'Buy at the Low. Fung. and D." Joutrnial of Portfolio Management. 1. H. 10. 1111-1127. and T. Goetzmann. 2000. 30-41. Journ-lal of Finianice." Industrial Management Review. Chance." Journal of Finianice. Merton. Park. 233-251." 389-416. 867-887." Journal of Empirical Finance. W. 275-302. "Idiosyncratic Risks: An Empirical Analysis. H. and D. 1985. Jensen.. 1981. 2000.. W.. R. 34. and K. 1996. W. and D. 133-160. Jagannathan. 2000. Rouwenhorst. W. 7-19.. D.. "Portfolio Performance Evaluation: Old Issues and New Insights. J. Henriksson.. 6. "Pairs Trading: Performance of a Relative Value Goldman. with Implications for the Risk of RelativeValue Trading Strategies. R. 1999. W. "Benefits and Limitations of Diversification among Commodity Trading Advisors.. 1973. Titman. "Asset Allocation: Management Style and Performance Measurement. C. 23. Scholes. Treynor. "Can Mutual Funds Outguess the Market?" Harvard Blusiniess Review. and K. A. M. D. Grinblatt. M. 1968. 341 ." Journal of Political Dybvig. R. 309-331. Sell at the High. Sosin. 1-36. "Beyond Mean-Variance: Performance Measurement in a Nonsymmetrical World."' Journal of Finance. 1999. 1966. Hsieh. Glosten. A. Gatev. 1999." Journal of Businiess. and R. F.. 81. and D. and J. Hsieh. Mazuy. Modest. and M. 54. Leland. Richards. 1-26. Mitchell. 27-36. A. 55. and R. 2. Fung. Merton. and D." Journlal of Emtipirical Finiance. M. "A Primer on Hedge Funds. Hsieh.. C. "Characteristics of Risk in Risk Arbitrage. "Price Movements in Speculative Markets: Trends or Random Walks. 23. "Measuring the Market Impact of Hedge Funds. 1987. B. Tsatsaronis. 1992. Arbitrage Rule. "On Market Timing and Investment Performance I: An Equilibrium Theory of Value for Market Forecasts. Gatto. C. "The Pricing of Options and Corporate Liabilities. 393-421.. 1997a. and D. 18." working paper. A. 44. Northwestern University and Harvard University. 1996. 2.. W. and S. 65-80." Finianicial Analyst Journal. 1981. 233-265.. 23.." working paper. "Mutual Fund Performance Evaluation: A Comparison of Benchmarks and Benchmark Comparisons." Review of Financial Stuidies. Black.. A." Journial of Portfolio Managemenit. S. Fung. 1989. "Do Hedge Funds Disrupt Emerging Markets?" BiookinigsWharton Papers on Financial Services. Hsieh. Sharpe. International Monetary Fund. E." Journal of Portfolio Maniagement. 41. Hsieh. 475-517. 1961. Economy." Journal of Business. "Empirical Characteristics of Dynamic Trading Strategies: The Case of Hedge Funds... 42. Lehman. R.. and M. "Differential Information and Performance Measurement Using a Security Market Line... W.. L. M.

You're Reading a Free Preview

Download
scribd
/*********** DO NOT ALTER ANYTHING BELOW THIS LINE ! ************/ var s_code=s.t();if(s_code)document.write(s_code)//-->