This action might not be possible to undo. Are you sure you want to continue?
Performance and persistence of Commodity Trading Advisors: Parametric Evidence Greg Gregoriou, Georges Hübner, Maher Kooli
September 06 / N° 200609/01
Performance and persistence of Commodity Trading Advisors: Parametric Evidence
Greg N. Gregoriou Georges Hübner Maher Kooli
September 2006 version
JEL Classification: G2, G11, G15 Keywords: CTA; Performance persistence; Multifactor models; managed futures
Greg N. Gregoriou is Associate Professor of Finance at State University of New York (Plattsburgh), 101 Broad Street, Plattsburgh, New York 12901, E-mail: firstname.lastname@example.org, Tel: (518)564-4202; Fax: (518) 564-4215. Georges Hübner is the Deloitte Professor of Financial Management at HEC Management School – University of Liège, Liège, Bld du Rectorat 7, B31, 4000 Liège, Belgium, E-mail: G.Hubner@ulg.ac.be. Tel : (32) 43662765.Fax: (32) 43662767; Associate Professor of Finance at the University of Maastricht; Senior Researcher at the Luxembourg School of Finance, University of Luxembourg, and Affiliate Professor of Finance at EDHEC.. Maher Kooli is Assistant Professor of Finance at University of Quebec at Montreal, School of Business and Management, Montreal, Quebec, P.O Box 8888, Station (Centre-Ville), H3C 3P8, E-mail, kooli,email@example.com. Tel: (514) 987-3000 ext. 2082. Fax: (514) 987-0422
Performance and persistence of Commodity Trading Advisors: Parametric Evidence
Abstract We re-examine the performance of Commodity Trading Advisors (CTAs) using a parametric calendar time approach during the 1996-2005 period. We compare abnormal performance based on a number of alternative existing models, as well as a category-specific model introducing asset-, option- and moments-based factors. Taking more factors into account significantly raises the explanatory power, but only four out of 7 CTA categories significantly outperform the market. We find that numerous CTAs show persistence over a horizon of at least three months and they also more likely to be persistent over a longer period. Yet, most of the persistence fades away upon the “acid test” of considering only the top and bottom quartiles of the managers.
Performance and persistence of Commodity Trading Advisors: Parametric Evidence
1. Introduction For individual or institutional investors that are both simultaneously performance-oriented and risk-conscious, the main question is how best to achieve a higher overall rate of return with acceptable level of risk. The answer could be a diversified investment portfolio with a certain portion of the total assets allocated to alternative investments, more specifically managed futures. A Commodity Trading Advisors (CTAs) is any money manager that accepts compensation for advising clients on either buying or selling commodity futures, forwards, options deemed as managed futures contracts. Managed futures provide access to investment not easily available through traditional stock and bonds and they have witnessed a dramatic increase in assets during the last decade. For instance, Managed Accounts Reports (MAR) cites an increase in managed futures1 from less than $1 billion in 1980 to almost $131 billion in 2004 (Schneeweis, 2005). However, there is still some confusion about the performance of managed futures as the number of CTAs has grown during this period. When using the composite performance of 15 trading advisors, Lintner (1983) shows that the risk-return ratio is greater than a well-diversified stock and bond portfolio. Furthermore, the author finds a low correlation between the returns of trading advisors and those of stocks, bonds, or a combined stock/bond investment portfolio. It is well-known that CTA returns produce negative correlation to stock and bond market indices. In addition, Schneeweis and Spurgin (1997) show that various CTA and hedge-fund-energy-based investment provide risk and return opportunities not available from a wide range of traditional commodity investments or real estate investments.
The term managed futures describes an industry made up of professional money managers known as commodity trading advisors (CTA). These trading advisors managed client assets on a discretionary basis using global futures markets as an investment medium.
During the one of the strongest stock market periods in U.S. history (1980-2005) managed futures (as measured by the Barclay CTA Index) had a compound annual return of about 12.76% comparing favorably with the 13.24% return the S&P 500 index produced during the same period. The Barclay CTA index exceeded the Smith-Barney U.S. Government Bond Index and returned 8.48% during the same period. Moreover, during the 1980-2005 time frame, analysis shows that a portfolio comprised of a partial allocation to managed futures had similar profitability with far less risk. Therefore, CTA can be considered as good hedging instruments in down markets for hedge funds, fund of funds, and the equity markets (Liang, 2003). Numerous academic studies during the last 20 years have examined the performance of managed futures with mixed results and with the lack of consensus. Using regression analysis, this study investigates the performance persistence issue as a well during both bull and bear market periods. We aim at providing a comprehensive set of evidence as to whether CTAs can truly add value and sustain their performance. Since the attrition rate of CTAs is typically higher than mutual funds it is an important subject to analyze during both the short and long-term periods. A multi-period time frame analysis differentiates between performance persistence due to luck and due to managerial skill. We assess and differentiate the results from a multi-period framework with the results from the conventional two-period analysis. Moreover, we provide a set of new “acid tests” on persistence by considering the ability of CTA managers to sustain performance over different horizons and with different ranking schemes. The procedure enables us to considerably deepen our insights over the durability of CTA performance per categories. With this approach, never previously used for CTAs, we shed new light on previous studies on CTA performance (see for example, Elton, Gruber and Rentzler, 1987; Liang, 2003) who conclude that investors cannot select top performing funds using historical performance.
The rest of the paper is organized as follows. Section 2 is the literature review. Section 3 sets out the data and the performance models considered. Section 4 brings some insights on CTA performance and reports the main results. Section 5 concludes the paper.
2. Literature review While examining the benefits of managed funds, Schneeweis and Georgiev (2002) demonstrate that CTAs reduce portfolio volatility risk and enhance portfolio returns in most economic environments where stock and bond offer limited opportunities. However, these authors sustain that for managed futures to grow as an investment alternative class, investors need to increase their knowledge and comfort level as to the use of managed futures in their investment portfolios. Irwin, Zulauf and Ward (1992a) study 363 CTAs during the 1979-1989 period and find that there is a lack of performance persistence when the authors looked at past CTA returns to predict future returns. Irwin, Krukemeyer, and Zulauf (1992b) examine commodity pools during the 19791990 period and use quintiles to separate CTAs but the results are not robust. Irwin et al. (1994) focus on individual CTAs, and their results concur with the previous findings of little evidence of predictability in average CTA returns. In an early study McCarthy, Schneeweis and Spurgin (1996) find some performance persistence. However; their sample size of CTAs used is relatively small with 56 CTAs using the 1985-1991 time frame. The same authors observe that multi-advisor managed futures funds have more persistence than single advisor CTAs (McCarthy, Schneeweis and Spurgin, 1997). Brorsen (1998) investigates data from private and public funds, and CTAs using various statistical methods such as regression analysis, Monte Carlo methods, and out-of-sample tests and finds that there exists limited evidence of performance persistence. The main drawback of the above studies is the
short examination period during a bull market, not encompassing any sustained bear market environment. One of the pioneering studies on managed futures by Elton, Gruber and Rentzler (1990) use 152 commodity pools during the 1990-1998 period to examine the performance issue. The authors find no evidence of performance and suggest that management fees and transaction costs must be reduced for commodity investments to be more appealing. Brown, Goetzmann and Park (1999) find no persistence in CTA returns using a non-parametric approach during the 1989-1998 period confirming the results of Edwards and Caglayan (2001). These authors, using Managed Account Reports data on 2,345 commodity funds during the January 1990 to August 1998 period, find no support regarding performance persistence either. However, using regression analysis may offer some more robust results than non-parametric analysis. Diz (1996) used 925 managed futures programs from the Barclay CTA database during the 1975-1995 and concludes that survival of CTAs is connected to performance but performance is not essentially linked with survival. Fung and Hsieh (1997) use 901 private CTA funds during the 1989-1996 period and find that the survivorship bias is 3.4% per year implying that CTA returns of live funds are on average 3.4% higher than the returns of both live and dead funds. Brorsen and Townsend (2003) use the 1990-1999 period and find a small amount of performance persistence using regression analysis. More recently, Capocci (2004) investigates the long-term performance persistence of live and dissolved CTAs by combining three databases, (1) the Barclay Trading Group (2) TASS and (3) Managed Account Reports during the 1985-2002 period (throughout both bull and bear markets) while encompassing numerous extreme market events. To examine the performance issue, the author ranks CTA strategies in deciles and finds that out of 11 strategies only 7 significantly outperform the CTA Global Index. In terms of performance persistence a large majority of deciles underperforms the CTA Global Index during the entire period.
However, sub-period analysis reveals that only one decile appears to outperform the CTA Global Index. Furthermore, the low significance level of his regressions does not lend strong economic support to his results; Financial research has further suggested that managed futures provide positive skewness and lower kurtosis when included with hedge funds in a traditional investment portfolio (Kat, 2005). Although managed futures are considered high risk investments, indices such as the S&P 500 have produced higher volatility than the Center for International Securities and Derivatives Markets (CISDM) CTA index during the 1990-2004 period. Yet, when managed futures are coupled with hedge funds, stocks and bonds, they are usually claimed to enhance the overall portfolio performance, provide instant diversification and reduce risk. Liang (2003) uses the Zurich database and examines 294 live CTAs and 1,216 defunct CTAs and concludes that they underperform hedge funds and funds of hedge funds due to their management fees, higher attrition rates and survivorship bias. To our knowledge, no comprehensive analysis of CTA performance and persistence using a similar approach to ours has been carried out so far. This is probably due to the fact that regression methods that have been tested so far have hardly produced any satisfactory significance levels. This has probably precluded the safe run of performance persistence analysis using in a multi-period framework.
3. Data and method 3.1 Data The data set consists of 527 live and 1290 dead CTAs that reported monthly performance figures net of all fees to the Barclay Group database during the period January 1996 to December 2005, a total of 123 months. We select this time frame because previous studies examining the
performance of CTAs are during the bull market period (pre-1999). Because of their negative correlations, CTAs perform poorly in bull markets. To ratify this we use both the bull and bear market periods to properly determine how CTAs perform in these types of markets. Our study is the first to include the bear market starting with the bursting of the tech bubble in the Spring of 2000. When database vendor’s backfill CTA returns, data is upwardly biased and that this type of bias can add a yearly return of 1.4% (Fung and Hsieh, 2000). We remove CTAs born prior to January 1996 to eliminate backfill bias, and omit CTAs with missing classifications and no covariate data. In this paper, CTAs enter the Barclay database and remain alive until they stop reporting results for three consecutive months, at which time they are considered defunct and removed. They can, however, re-enter the database and backfill their unreported data. To qualify for inclusion in the database, a CTA must have at least four years of preceding historical performance. New programs introduced by CTAs are only added after the second year to the Barclay database. These limitations are enforced to compensate for the high turnover rates of CTAs because of their high short-term performance to guarantee precise and consistency of the Barclay database. Each of the seven CTA classifications is examined to obtain more informative and precise results, which would be beneficial to money managers and investors in terms of performance. The definitions of the classifications are presented in Table 1. Discretionary and systematic are trading styles, whereas the other classifications are usually considered as portfolios of options and futures.
Insert Table 1 about here***
3.2. Methods Following Fama and French (1993), we choose the value-weighted portfolio of all NYSE, Amex and NASDAQ stocks as a market index. We also consider using the Russell 3000 index as well.
The comparison of descriptive statistics of the two indices suggests that both market proxies are very similar. Thus, the results of the study should not be influenced by the market proxy chosen. We use the 1-month U.S. T-bill rate from Ibbotson Associates as the risk-free rate.
3.2.1 Traditional Models In the alternative investments literature, typically multifactor models have been preferred over the well-known CAPM for performance evaluation, but there is no universally accepted model. In this paper, we use several specifications in order to compare our results. We start with the market model derived from the CAPM. It provides a simple benchmark to assess the gains in significance levels obtained with more sophisticated specifications.
Market model: Assume that CTA returns follow the one-factor return generating process, Rjt =αj+βjRMt+εjt (1)
where Rjt is the rate of return of the CTA j on month t; RMt is the rate of return of a market index on month t; εjt is the error term.
The Fama and French (1993) three-factor model2: It takes the size and the book-to-market ratio of the firms into account and is estimated from the following extension of the CAPM regression Rjt−RFt=αj + βj1(RMt−RFt) + βj2SMBt + βj3HMLt + ε jt (2)
where SMBt=the factor-mimicking portfolio for size (small minus big) and HMLt=the factormimicking portfolio for book-to-market equity (high minus low).
See Fama and French (1993) for a complete description of the factor returns.
The Fama and French (1993) model with Ibbotson’s (1975) “Returns Across Time and Securities” (RATS) procedure: The RATS technique allows the estimate of beta to vary during returns window. In other words, it presumes that the betas of the portfolio should vary over time as new information becomes readily available. The Fama and French (1993) model with RATS procedure provides: Rjt−RFt=αj + βj1t(RMt−RFt) + βj2SMBt + βj3HMLt +
3.2.2 Adapted Models Unlike previous studies, rather than considering a general multifactor model for all CTA categories, we use a more rigorous approach that takes into account the many dimensions of financial risk inherent to CTA. More specifically, using a large set of potential variables, including moment factors and options factors, we choose first a subset of significant variables for each CTA category through a stepwise regression procedure. We calculate then the performance for each CTA category using a calendar time approach discussed later. To the best of our knowledge this the first global study that considers a specific model for each CTA category. The exposure of CTA returns to non-linear risk factors has been largely documented in the recent literature (see Martellini and Vaissié, 2004). Liang (2004) applies the equity-based option factors developed by Agarwal and Naik (2004) to a sample of CTAs and finds superior significance levels for this class of securities than for hedge funds or funds of funds, with an increase in adjusted R-square of 20.29% for the CTA index. On the other hand, the pervasive positive skewness and excess kurtosis of CTA strategies that makes them useful diversification tools (Kat, 2004) also explains their significant loading to the square and cube of market returns as shown by Hübner and Papageorgiou (2004).
We select eight financial indexes on which option-based factors are constructed: three exchange rates (USD/Yen, USD/Euro3, and USD/UK Pound Sterling), two interest rate indexes (Lehman High Yield Bond index and the Lehman US Treasury LT Bond Index), and three commodity indexes (GSCI Wheat, Gold and Crude Oil Indexes). We do not include an equity-based option as it overlaps with the higher moments variables constructed below. For each index, we compute the Black-Scholes (B-S) price of four six-week options (ATM Call and Put, and 5% OTM Call and Put) on a monthly basis. We use a rolling estimate of the 12-month volatility as input. At the end of the month, we compute the realized return with the new B-S price of the option. To account for the likelihood that most CTAs follow non-directional strategies, we complement this set of options with the corresponding long ATM straddles and OTM strangles constructed with the samples of calls and put. Even though these portfolios are less complex than the ones proposed by Fung and Hsieh (2001) in the context of hedge funds, they have the advantage of reasonably replicating simple trading strategies. Amongst the modeling approaches that aim at integrating moments of order three (skewness) and four (kurtosis) into the asset pricing model, we have to select a method that produces a tradable index. We adopt an approach advocated by Harvey and Siddique (2001), who suggest building these variables in the same way as the SMB and HML variables of Fama and French (1993), with stocks ranked on the basis of their coskewness with the stock index. We simplify this procedure by collecting the S&P sub-industry indices from July 1994 to 2005 as raw data (123 indices), and computing the lagged 12-month variance, and 6-month skewness and kurtosis.4 Then, for each month, we rank the indexes in turn by their variance, skewness and
Prior to the freezing of parities, we take the USD/German Mark. Our data show that variance shocks exhibit a longer memory than skewness or kurtosis shocks, and so the variance has to be computed on a longer time interval.
kurtosis and compute the corresponding risk premia by taking the difference between the top and bottom third selected (67th percentile minus 33rd percentile). We also include Carhart’s (1997) momentum risk premium.5 Consistently with its original interpretation, this factor is able to capture some persistence in the stock returns and does not overlap with the moment and option factors.6
3.2.3 Performance estimation We employ the calendar-time method developed by Jaffe (1974) and Mandelker (1974) which controls for cross-correlation. Lyon et al. (1999) show that the calendar time approach yields well specified test statistics, because it entails calculating average returns of rolling, calendar-time portfolios of event funds. Specifically, for each calendar month, we form an equally weighted τmonth portfolios set up to include any CTA strategy which has a return during the previous τmonths, for τ = 12, 24, and 36. The calendar time approach has the additional advantage that it provides a direct measure of the opportunities available to investors attempting to exploit any abnormal performance. For each model specification, we define the out-or under-performance relative to the market proxy used (i.e., the abnormal return) for the fund on month t as:
ˆ Ajt = Rjt − ∑ β kt Fkt
k =1 K
where Fkt represents the k-th factor in the regression. The average abnormal return AARt is the sample mean:
See Carhart (1997) for a description of the construction of PR1YR. We have considered other indices: the MSCI World Index excluding US, the return of the Salomon World Government Bond Index, the return of the Goldman Sachs Commodity Index (GSCI), but the stepwise regression procedure rejects all of them for each strategy. We have also tested Fama and French (1998) international value, JP Morgan Emerging Market Bond Index, the term spread, implicit volatility, and currency factor)s. However, due to their high collinearity with other factors, we decide to not test these indices further
Over an interval of two or more trading months beginning with month T1, and ending with T2, the cumulative average abnormal return (CAAR) is:
j =1 T1
To test the null hypothesis that the mean abnormal return is equal to zero for a sample of n CTAs, we first employ a cross-sectional t-statistic. The standard error for this test for each month is computed across funds, not across time. To eliminate the skewness bias when long-run abnormal returns are calculated, we also use the bootstrapped skewness-adjusted t-statistic.
4. Results 4.1 Basic performance
Table 2 reports means, standard deviations, medians Sharpe ratios, kurtosis and skewness, for the seven CTA categories and the four factors based on monthly data for the 1996-2005 period. The first part of Table 2 shows that the highest mean return was achieved by the Stock Index CTA (1.46%) followed by the Energy CTA (1.00%). The Arbitrage CTA offers the lowest mean return (0.34%). When standard deviation is taken into account through the Sharpe measure (the ratio of excess return and standard deviation), results confirm our previous observations. CTA offering the best trade-off between risk and return is the Index Stock CTA (0.49) and the worst Sharpe ratio is obtained by the Arbitrage CTA (0.01), which is also the worst performing index when risk is not taken into account. Table 2 also confirms the heterogeneity of CTA universe: some CTA categories have relatively high volatility while other CTAs have lower volatility.
When CTA data are contrasted against the descriptive statistics of the market proxy, the Fama and French (1993) SML and HML, and Carhart (1997) momentum’s factor, the statistics are in favor of the managed futures strategies. The mean return of the Market Proxy is 0.26% per month (about 4.78% per year). The average SMB and HML returns are 0.29% and 0.46%, respectively. The highest mean return was obtained by the Momentum factor (0.87% per month). On average, the Sharpe ratio obtained by the seven CTA categories (0.23) is higher than the one for the Market Proxy (-0.01).
Insert Table 2 about here***
Table 3 reports correlation coefficients among CTA categories. There is a high variability between different categories, ranging from 0.83 (between Financial/ Metal CTA and Diversified CTA) to -0.28 (between Diversified CTA and Arbitrage CTA). Only one correlation coefficient is greater than 0.70 and six of them are negative.
Insert Table 3 about here***
Table 4 reports correlation coefficients between CTA categories and equity, bond and commodity indices. Correlation coefficients between CTA categories and the Market Proxy are, in seven cases, smaller than or equal to 0.39. Hence, the addition of CTAs to a traditional portfolio should improve its mean-variance trade-off. Liang (2003) also finds that returns from CTAs are negatively correlated with other instruments, making CTAs suitable for hedging against downside risk.
Insert Table 4 about here***
4.2 Calendar time abnormal performance
The discussion of performance models is performed for the whole 1996-2005 period. The first performance model used is the CAPM-based single index model, taken as the benchmark to assess the value-added of further improvements. Table 6 reports the results for the whole sample. We estimate each CTA category (portfolio) individually using calendar time approach. In other words, CTAs are grouped into portfolios by event date and by categories. A portfolio standard deviation is estimated from the time series of portfolio abnormal returns in the estimation period, and used to standardize the portfolio return. Starting from 1996, we note that all CTA categories seem to significantly outperform the market. In all cases, the alphas from CAPM (See Table 5) are significant at the 1% confidence level. For example, after 3 years, the cumulative abnormal return for Energy CTA is 2.81% per month (bootstrapped Skewness corrected t-statistic = 7.55).
Insert Table 5 about here***
In Table 6, we report the results for Fama and French (1993) three-factor model applied to all CTA categories. Table 6 tends to confirm the seemingly superior performance of CTAs. The positive and significant calendar time alphas range from a monthly percentage of 1.28% for Stock Index CTA to 2.10% for Discretionary CTAs. The table also reveals that the premia on the SMB and the HML factors are negative (except for HML factor for Energy CTAs). There are however significant only for Systematic CTAs. Systematic CTA managers seem to prefer larger stocks and those with low book-to-market ratios. The adjusted R2s range from 0.003 for Energy CTAs to 0.153 for Stock Index CTAs. As expected, the highest adjusted R2s are obtained when the alphas are the lowest. However, these low adjusted R-squared coefficients confirm all outstanding evidence about the powerlessness of most traditional asset pricing models.
Insert Table 6 about here***
Table 7 reports the result of Fama and French (1993) three-factor model applied to all CTA categories using RATS procedure for different calendar time periods. Again, we confirm that the all CTA categories outperform the market for the 1996-2005 period.
Insert Table 7 about here***
Table 8 reports the results of the multi-factor model. In a strong contrast to previous results, we find that taking more factors into account induces that only four out of seven CTA categories significantly outperform the market and one out of seven has positive but insignificant excess returns. The alphas of Arbitrage and Energy CTAs are negative but only significant for arbitrage CTAs. Based on alpha, the best strategy is Diversified (alpha = 4.03% per month). Interestingly, in only two out of seven CTA categories we retain the premium on the SMB factor but it is statistically insignificant. Also in one category, the premium on the Momentum factor is insignificantly negative, reflecting that CTAs are generally not arbitragers in the equity markets. Overall, the most significant factors for CTAs are option factors. Although managers may not trade in the option markets directly, their returns may show non-linear or option-like patterns due to some long-short combinations. Our observation is consistent with Fung and Hsieh (1997b) and Liang (2003) that CTAs exhibit option-like return patterns with respect to equity markets. Our multifactor models results provide some insight into the preferences of CTA managers: (1) we find that Energy, Financial/Metals, Diversified, and Stock Index CTAs managers invest in gold (as a proxy for metals); (2) Energy, Discretionary, and Stock Index CTAs managers invest in oil (as a proxy for fossil energy materials); (3) Diversified and Financial/ Metal CTAs managers invest in wheat (as a proxy for comestibles); (4) Diversified, Arbitrage, and Systematic CTAs managers invest in credit-risky bonds; (5) the best performing category (Diversified CTA) invest in government and corporate bonds, currency, wheat, and gold.
Overall, it seems that the multifactor models do a good job in explaining CTA behavior of various categories. The average adjusted R2 increases from 0.06 for the Fama & French three-factor model to 0.31 for the multifactor models (the adjusted R2s of the seven multifactor models range from 0.07 to 0.49). But at the same time this means that although we can explain a substantial part of the variation of CTA returns by the multifactor models, another, non-trivial part is still missing. The set of explanatory variables considered seems particularly adapted to Diversified (0.49) and Arbitrage (0.43) CTAs but does a poorer job with Discretionary CTA. However, we should mention that our adjusted R2 are higher that those obtained by Liang (2003). He reports adjusted R2s from 7.2% to 14.3%. Liang also concludes that CTAs are different from hedge funds or fund-of-funds in trading strategies and those multifactor models have very low explanatory powers for CTAs.
Insert Table 8 about here***
Table 9 reports the same analysis of performance over different subperiods. First, we subdivide the 1996-2005 period in two sub-periods of equal lengths (January 1, 1996-December 31, 2000 and January 1, 2001-December 31, 2005). Second, we consider the Asian crisis period. When the time period is divided in two, we notice that the significant underperformance of Arbitrage CTA for the 1996-2005 period is mainly due to the first sub-period. We also notice that Stock Index, Diversified and Financial/Metal CTAs significantly outperform in both subperiods. Moreover, during the Asian crisis, Energy and Stock Index CTAs took advantage of this period and earn positive and significant excess return. Four CTAs categories outperform the market, though insignificantly.
Insert Table 9 about here***
4.3 Relative performance persistence of CTAs.
The second objective of this study is to examine the performance persistence of CTAs over different time periods. We follow Brown, Goetzmann and Ibbotson (1999) and Agarwal and Naik (2002) and compare the performance measures in the current period on the performance measures in the previous period. We rely on alpha as a performance measure, defined as the return of a CTA following a particular strategy minus the average return for all CTAs following the same strategy. Moreover, to investigate the issue of persistence in two consecutive periods, we rely on a nonparametric method. More specifically, we construct a contingency table of winners and losers. A CTA fund is a winner if the alpha of that fund is greater than the median alpha of all the CTA funds following the same strategy in that period otherwise it is a loser. In this context, persistence refers to the existence of CTA funds that are winners in two consecutive periods (one month, three months, six months, and twelve months periods), denoted by WW, or losers in two consecutive periods, denoted by LL. Similarly, winners in the first period and losers in the second period are denoted WL and LW denote the reverse. In this paper, we use both cross-product ratio (CPR) and Chi-square statistic to detect persistence. The CPR test statistic is the ratio of the product of repeat winners (WW) and repeat losers (LL) divided by the product of winner-losers (WL) and loser-winners (LW), i.e. (WW×LL)/(LW×WL). A CPR of one would support the hypothesis that the performance in one period is unrelated to that in another. A CPR greater than one indicates persistence, while a value below one indicates that reversals in performance dominate the sample. We determine the statistical significance of the CPR by using the standard error of the natural logarithm of the CPR is given by (see Christensen, 1990)
1 1 1 1 + + + WW WL LW LL
We also conduct a chi-square test comparing the observed frequency distribution of WW, WL, LW, and LL with the expected frequency distribution. Carpenter and Lynch (1999) study the specification and power of various persistence tests and find that the chi-square test based on the number of winners and losers is well-specified, powerful, and more robust to the presence of survivorship bias compared to other test methodologies. The chi-square statistic is defined as: ˆ χ2 = where (WW + WL )(WW + LW ) N (WW + WL )(WL + LL ) D2 = N ( LW + LL )(WW + LW ) D3 = N ( LW + LL )(WL + LL ) D4 = N N = WW + WL + LW + LL (WW − D1 ) 2 (WL − D2 ) 2 ( LW − D3 ) 2 ( LL − D4 ) 2 + + + D1 D2 D3 D4 (8)
We test this statistic at the 5% significance level, which corresponds to a critical value of Chi-square statistic of 3.84 (one degree of freedom). Table 10 shows our empirical results. For each formation/holding period, the sample is split into overlapping (to maximize power) periods of the required frequency. Each panel of Table 10 (A to D) displays the percentage of occurrences of WW, WL, LW, and LL over the sample period, the cross-product ratio (CPR) with its Z-statistic, and the chi-square test result. For the latter statistic, we report the percentage of cases where statistically significant persistence was observed in each CTA category.
We first observe that, in general, both the CPR and chi-square tests indicate a greater extent of persistence, i.e. persistence is observable for longer holding periods. In particular for holding periods of three, six and twelve months, Diversified, Financial/Metal and Systematic CTAs are more persistent than the others categories in terms of the percentage of cases where persistence is statistically significant. For instance, for a holding period of twelve months, the percentage of cases where persistence is statistically significant is 90%, 90% and 92%, for Diversified, Systematic, and Financial/Metal CTAs, respectively. On the other hand, we find that Arbitrage CTA managers are less persistent than other CTA managers. Surprisingly, we find a weak presence of persistence when the holding period is one month. For all CTA categories, their corresponding CPRs are not statistically different from one. Finally, contrary to hedge funds evidence (Agarwal and Naik, 2002) the extent of CTA persistence does not seem to be related to return measurement intervals. Overall, we find that managers who show persistence over a short time horizon is more likely to be persistent over a longer one. In Table 11, we extend our investigation to examine what happens after periods during which some CTAs exhibited some persistence. In other words, we show among the periods for which we find 1 month persistence for instance, how much are still persistent after 3, 6 and 12 months and vice versa. This indicates whether the same periods yielded persistence at various horizons. For example, for Diversified CTAs, among the 45.11% results with a significant chisquare, 4.48%, 0.38% and 0.04% (percents based on the whole number of funds) were also significant with a 3, 6 and 9 months holding periods, respectively. The results show a striking difference between one-month persistence and persistence over longer periods. Conditionally on observing short-term persistence over a given month, the same sets of portfolios (based on a one-month formation period) show very weak persistence over longer periods. The loss in persistence when the holding period increases from one to several months is
especially strong for those strategies that exhibit the largest short term persistence, namely Diversified and Financial/Metal. The converse holds for the Arbitrage and Energy strategies, for which 10.02% and 9.09% of the periods sustain one-month and three-month persistence altogether. These strategies, that exhibit the weakest short-term persistence, seem able to sustain this persistence over longer periods. They also display low levels of medium- and long-term persistence. This indicates that, while there is scarcer evidence of short- and long-term persistence for these CTAs, persistence is more robust to the passage of time. For the other strategies, persistence over the very short term does not signal persistence over longer periods.
Insert Table 11 about here***
Table 12, shows persistence for different sub-periods. For all but the Stock Index and Energy sectors, persistence over very short (one month) and short periods is better during the first (bullish) sub-period. For these holding periods, managers the Energy sector particularly suffered from the Asian crisis, where very little persistence could be observed. For 12 month persistence however, all CTA strategies exhibit lower persistence in the second sub-period. The impact is particularly strong for Energy CTAs as well. This suggests that, for this particular strategy, persistence has significantly shifted from longer to shorter holding periods from one sub-period to another. Evidence reported in this Table suggests that persistence has consistently deteriorated over time, especially for longer holding periods. During the second sub-period there is a clear absence of longterm trends. The market has been moving sideways, with lots of minor bumps during the 2001-2005 period. Trend-follower models cannot cope with that, since they represent more than 70% of CTAs. A few trends (oil) still exist, but most markets are trend-less. Short-term persistence results of CTAs do not lead to significant evidence but when they are examined every two consecutive
yearly periods i.e, 1994-1995 , 1995-1996 up to 2002, performance persistence is strong and statistically significant. However, Vuille and Crisan (2004) observe that long-term
persistence does not led to significant persistence during June 1, 1994-May 31, 1998 and June 1, 1998 and May 3, 2002 periods.
This result is very consistent across strategies, unlike evidence presented in Table 9 showing that performance was not monotonically evolving over time for all strategies.
Insert Table 12 about here***
We perform additional tests to analyze the persistence in the extremities of the rankings. More specifically, our performance analysis award a “win” to those funds with a ranking in the top quartile (normalized ranking greater than or equal to 0.75), and a “lose” to those funds with a ranking in the bottom quartile (normalized ranking less than 0.25). If a CTA fund earns a ranking placing them in either of the middle quartiles (greater than 0.25 or less than 0.75) it is assigned a rating of NS. Only CTA funds in the top or bottom quartile are paired with a subsequent fund. As we are principally interested in the extreme performers, we only present results for WTWT (top “T” funds that are winners in two consecutive periods) and LBLB (bottom “B” funds that are losers in two consecutive periods); The results are shown in Table 13, which should be analyzed in conjunction with Table 10. In Panel A, we find percentages that are considerably below the expected frequencies under the null (25%) for all but the Arbitrage categories. Thus, in the short run, most extreme performers among the CTA managers tend to “join the pack”. The very high rejection rate using the khi-square statistics suggest that this behavior is hardly random, and sharply contrasts with evidence presented in Table 10 for the same period.
This gregarious behavior fades away with the three-month period, except for Discretionary CTAs. Next, in Panels C and D, we observe a reversal. There is again some evidence of persistence, but mostly located in the bottom quartile. Good managers manage to sustain their rank for Financial/Metals and, to a lesser extent (as indicated by the khi-square test) by Stock Index managers. Discretionary CTAs consistently exhibit a return-to-the-mean behavior across all test periods.
Insert Table 13 about here***
In this paper we have re-examined the performance of CTAs over the period 1996 to 2005. We assess abnormal performance using a number of alternative benchmarks and an approach, originally used in event studies. More specifically, the benchmarks employed allow for the standard CAPM, the Fama-French three-factor model, the RATS technique, and a customized multifactor model. In addition, we use the calendar-time approach as developed by Jaffe (1974) and Mandelker (1974). This approach has the added advantage that it provides a direct measure of the opportunities available to investors attempting to exploit any abnormal performance. We find that, using the market model or the Fama and French (1993) three factor model as benchmarks, CTA categories do outperform the market. However, taking more factors into account induces that only four CTA categories significantly outperform the market and only one has positive but insignificant excess returns. Based on alpha, the best strategy is Diversified Further, we find that the most significant risk drivers for CTAs are option-based factors, which confirms the empirical finding of. Fung and Hsieh (1997b) and Liang (2003). The second objective of this study is to examine the performance persistence of CTAs over different time periods. We first find that for Diversified, Systematic, and Financial/Metal CTAs are consistently beating the others categories in terms of the percentage of managers that are
performing above their median. Second, we find that the performance of Arbitrage CTA managers is always less persistent than the one of other CTA managers. Overall, a manager who shows persistence over a horizon of at least three months is more likely to be persistent over a longer one. However, most of these results do not stand the “acid test” of considering the ability of a manager to stay in a top quartile rather than the top half of a category. As noted by several academics, research in alternative investments, in general, is still at its infancy. Our studies primarily shows that CTAs, in particular, address new challenges to financial theory and a lot remains to be done to identify CTA performance drivers. Their return generating process shows more than half of total variance unexplained. This probably indicates a high instability in the CTA risk exposures. Furthermore, our study clearly shows that short- and longterm measurement of persistence, as well as the ranking scheme of the managers, induce interpretations that can greatly vary from one category to another.
Agarwal, V., Naik, N.Y., 2002. Multi-Period Peformance Persistence Analysis of Hedge Funds. Journal of Financial and Quantitative Analysis 35(3), 327-342. Agarwal, V., Naik, N.Y., 2004. “Risks and Portfolio Decisions Involving Hedge Funds”.
Review of Financial Studies 17, 63-98.
Brorsen, B.W., 1998. Performance Persistence for Managed Futures. Working Paper, Oklahoma State University. Brorsen, B.W., Townsend, J., (2002). Performance Persistence for Managed Futures.
Journal of Alternative Investments 4(4), 57-61.
Brown, S.J., Goetzmann, W.N., Ibbotson, R.G., 1999. Offshore Hedge Funds: Survival and Performance 1989-2005. Journal of Business 72 (1), 91–117. Brown, S.J., Goetzmann, W.N., Park, J., 2001. Careers and Survival: Competition and Risk in the Hedge Fund and CTA Industry. Journal of Finance 56(5), 1869-1886. Capocci, D., 2005. CTA Performance, Survivorship Bias, and Dissolution Frequencies. In Gregoriou, G.N, Karavas, V.N., Lhabitant. F-S., and Rouah, F. (eds.) Commodity Trading Advisors : Risk Performance Analysis and Selection. John Wiley and Sons Inc., Hoboken, NJ. Carhart, M., 1997. On Persistence in Mutual Fund Performance. Journal of Finance 52(1), 57-82. Diz, F., 1996. Is Performance Realted to Survival? Working Paper, Syracuse University and the Barclay Trading Group, Syracuse, NY.
Edwards, F.R., Ma, C., 1998. Commodity Pool Performance: Is the Information Contained in Pool Prospectuses Useful? The Journal of Futures Markets 8(5), 589616. Edwards, F.R., Caglayan, O., 2001. Survivorship Bias and Investment Style in the Returns of CTAs. Journal of Portfolio Management 27(4), 30-41. Elton, E.J., Gruber, M.J., and Rentzler, J.C., 1990. The Performance of Publicly Offered Commodity Funds. Financial Analysts Journal 46(4), 23-30. Elton, E.J., Gruber, M.J., and Rentzler, J.C., 1989. New Public Offerings, Information, and Investor Rationality: The Case of Publicly Offered Commodity Funds. Journal of
Business 62(1), 1-15.
Elton, E. J., Gruber M.J., Rentzler J.C., 1987. Professionally Managed Publicly Traded Commodity Funds. Journal of Business 60(2), 177-199. Fama, E.F., French, K.R., 1993, Common Risk Factors in the Returns on Stocks and Bonds. Journal of Financial Economics 33(1), 3-56. Fung, W., Hsieh, D.A., 1997. Survivorship Bias and Investment Style in the Returns of CTAs. Journal of Portfolio Management 24(1), 30-41. Fung, W., Hsieh, D.A., 2000. Performance Characteristics of Hedge Funds and Commodity Funds: Natural vs. Spurious Biases. Journal of Quantitative and Financial
Analysis 35(3), 291-307.
Fung, W., Hsieh, D.A., 2001. The Risk in Hedge Fund Strategies: Theory and Evidence from Trend Followers. Review of Financial Studies 14, 313-341. Harvey, C., Siddique, A., 2001, Conditional Skewness in Asset Pricing Tests. Journal of
Finance 55, 1263-1295.
Hübner, G., Papageorgiou, N., 2004, The Performance of CTAs in Changing Market Conditions. In Gregoriou, G.N, Karavas, V.N., Lhabitant. F-S., and Rouah, F. (eds.) Commodity Trading Advisors : Risk Performance Analysis and Selection. John Wiley and Sons Inc., Hoboken, NJ. Ibbotson, R.G., 1975. Price Performance of Common Stock New Issues. Journal of
Financial Economics 2(3), 235-272.
Irwin, S.H., 1994. Further Evidence on the Usefulness of CTA Performance Information in Public Commodity Pool Prospectuses and a Proposal for Reform. In Advances in Futures and Options Research, D.M. Chance and R.R. Trippi, (eds.), JAI Press, Greenwich, CT. Irwin, S., Zulauf, C.R., Ward, B., 1992a. The Predictability of Managed Futures Returns.
Journal of Derivatives, 2(2), 20-27.
Irwin, S., Krukemeyer, T., Zulauf, C.R., 1992b. Are Publicly Commodity Pools a Good Investment? In Peter, C.C., (ed.) Managed Futures: Performance evaluation and Analysis of Commodity Funds, Pools, and Accounts. Probus Publishing, Chicago, IL. Jaffe, J. F., 1974. Special Information and Insider Trading. Journal of Business 47(3), 410- 428. Kat, H.M., 2004. Managed Futures and Hedge Funds: A Match Made in Heaven. In Gregoriou, G.N, Karavas, V.N., Lhabitant. F-S., and Rouah, F. (eds.) Commodity Trading Advisors : Risk Performance Analysis and Selection. John Wiley and Sons Inc., Hoboken, NJ. Liang, B., 2003. The Accuracy of Hedge Fund Returns. Journal of Portfolio Management 29(3), 111-121.
Liang, B., 2004. Alternative Investments: CTAs, Hedge Funds, and Funds-of-Funds.
Journal of Investment Management 2(4), 1-18.
Carpenter, J. and Lynch A., 1999. Survivorship Bias and Attrition Effects in Measures of Performance Persistence, Journal of Financial Economics 54, 337-374. Christensen, R., 1990. Log-linear models . Springer-Verlag, New York. Lintner, J.V. 1983. The Potential Role of Managed Commodity-Financial Futures
Accounts in Portfolios of Stocks and Bonds. Annual Conference of the Financial
Analysts Federation, Toronto, Canada.
Lyon, J., Brad B., Tsai C., 1999. Improved Methods for Tests of Long-run Abnormal Stock Returns. Journal of Finance 54(1), 165-201. Mandelker, G., 1974. Risk and Return: The Case of Merging Firms. Journal of Financial
Economics 1(4), 303-336.
Martellini, L., Vaissié, M. 2004. Benchmarking the Performance of CTAs. In Gregoriou, G.N, Karavas, V.N., Lhabitant. F-S., and Rouah, F. (eds.) Commodity Trading Advisors : Risk Performance Analysis and Selection. John Wiley and Sons Inc., Hoboken, NJ. McCarthy, D., Schneeweis, T., Spurgin, R. 1996. Survivor Bias in Commodity Trading Performance. Journal of Futures Markets, 16(3), 757-772. Schneeweis, T., Spurgin, R., McCarthy, D. 1997. Informational Content in Historical CTA Performance Journal of Futures Markets 14(1), 51-78. Schneeweis, T., 2005. The Benefits of Managed Futures. CISDM Presentation, University of Massachusetts, Amherst, MA.
Schneeweis, T., Spurgin, R., 1997. Managed Futures, Hedge Funds and Mutual Fund Return Estimation: A Multifactor Approach. CISDM Working Paper, University of Massachusetts, Amherst, MA. Schneeweis T., Georgiev, G. 2002. The Benefits of Managed Futures. CISDM Working Paper, Isenberg School of Management, University of Massachusetts, Amherst, MA.
Vuille, S., and C. Crisan. (2004). A Quantitative Analysis of CTA Funds Possibly a reason for short and long-run persistence. HEC Lausanne Mater's Thesis, Lausanne.
Table 1: Definition of CTA Classifications Classification Definition
Traders who trade either intra-commodity or inter-commodity spread trading as their predominant trading method. Energy Traders who trade exclusively energy contracts such as crude oil, gasoline, natural gas, etc. Discretionary Traders who, in response to a questionnaire, indicated that at least 65% of their trading decision process was judgmental or discretionary Diversified Traders who trade a diversified portfolio, including most of the major sectors Financial/Metals Traders who trade at least two of the following: currencies, interest rates, stock indices, precious metals Stock Index Traders who trade stock indices Systematic Traders who, in response to a questionnaire, indicated that at least 95% of their trading decision process was systematic Source. The Barclay Group.
Table 2: Descriptive statistics for all CTA categories. This table shows mean returns, t-statistics, medians, standard deviations, Sharpe ratios, skewness, and kurtosis for all CTA categories and indexes for the 1996-2005 sample period. RMt = return of the Market Proxy on month t; SMBt = the factor-mimicking portfolio for size (small minus big); HMLt = the factormimicking portfolio for book-to-market equity (high minus low); PR1YRt = the factor-mimicking portfolio for the momentum effect. We calculate the Sharpe ratio considering Ibbotson Associates 1-month T-bills. All numbers in table are monthly percentages.
Arithmetic Mean (%)
CTAs Diversified Arbitrage Systematic Stock Index Energy Financial/ Metal Discretionary Indexes RMt SMBt HMLt PR1YRt
Standard Deviation (%) 3.43 3.87 1.72 2.38 5.17 1.99 2.13 4.67 4.35 3.95 5.75
0.94 0.34 0.78 1.46 1.00 0.87 0.85 0.26 0.29 0.46 0.87
3.01 0.96 5.00 6.70 2.11 4.78 4.35 0.60 0.74 1.27 1.66
0.57 0.37 0.71 1.54 0.53 0.56 0.75 0.94 0.19 0.53 1.09
0.19 0.01 0.28 0.49 0.13 0.29 0.26 -0.01 0.00 0.04 0.10
0.35 -0.47 0.11 -0.92 1.43 0.51 0.56 -0.73 0.75 0.02 -0.64
0.05 3.36 0.11 2.36 7.44 0.19 1.57 0.62 5.90 1.94 3.93
Table 3: Correlation among CTA categories This table shows the linear (Pearson) correlation coefficient among CTA categories for the 1996-2005 sample period.
Energy Systematic Stock Index Arbitrage Financial/Metal Diversified Discretionary
Energy 1 0.06 -0.06 -0.06 0.08 0.07 0.03
Systematic 1 0.21 -0.19 0.65 0.55 0.11
1 0.12 0.14 0.09 -0.07
1 -0.26 -0.28 0.02
1 0.83 0.07
Table 4: Correlation matrix among passive indexes and between all CTA categories and passive indexes. This table shows the linear (Pearson) correlation coefficient among passive indexes and between all CTA categories and passive indexes for the 1996-2005 sample period. RMt = return of the Market Proxy on month t; SMBt = the factor-mimicking portfolio for size (small minus big); HMLt = the factor-mimicking portfolio for book-to-market equity (high minus low); PR1YRt = the factor-mimicking portfolio for the momentum effect; MSWXUSt = return of the MSCI World Index excluding US; SWGBIt = return of the Salomon World Government Bond Index; GSCIt = return of the Goldman Sachs Commodity Index.
RMt RMt SMBt HMLt PR1YRt Energy Systematic Stock Index Arbitrage Financial/Metal Diversified Discretionary 1 0.21 -0.55 -0.23 0.07 0.24 0.39 0.13 -0.02 -0.14 -0.02
1 -0.50 0.17 0.06 -0.04 0.17 0.14 -0.03 0.00 -0.17 1 -0.06 0.00 -0.19 -0.30 -0.17 0.06 0.11 -0.02 1 -0.05 0.03 0.03 -0.07 0.06 0.16 0.04 1 0.06 -0.05 -0.05 0.07 0.07 0.03 1 0.19 -0.18 0.63 0.54 0.08 1 0.12 0.11 0.07 -0.08 1 -0.23 -0.26 0.02 1 0.83 0.05 1 0.20 1
Table 5: Abnormal performance using the market model Each panel displays the calendar time alpha in percentage using the one-factor market model for different portfolio formation periods during the 1996-2005 sample period. Bootstrapped Skewness Corrected tstatistics are in parentheses. ***Significant at the 1% level. Calendar months 12 months 24 months 2.31%*** 1.85%*** (17.12) (19.25) 1.62%*** 1.03%** (2.90) (2.55) 0.92%*** 1.71%*** (8.40) (6.85) 1.64%*** 1.29%*** (5.00) (5.24) 3.41%*** 3.71%*** (5.85) (7.08) 2.20%*** 1.75%*** (12.86) (14.28) 1.59%*** 2.34%*** (6.78) (6.86)
Diversified Arbitrage Systematic Stock Index Energy Financial/ Metal Discretionary
36 months 1.60%*** (20.53) 1.37%*** (3.65) 0.63%*** (5.82) 1.07%*** (5.02) 2.81%*** (7.55) 1.42%*** (14.93) 1.61%*** (7.67)
Table 6: Performance measurement using the Fama and French (1993) three-factor model Each panel displays the monthly calendar time alpha in percentage using the Fama and Fench (1993) three factor model for the 1996-2005 sample period. Heteroscedasticity consistent t-statistics are in parentheses. * Significant at the 10% level. **Significant at the 5% level. ***Significant at the 1% level. Alpha (%) 2.02%*** (5.41) 1.36%*** (3.04) 0.87%*** (5.23) 1.28%*** (4.20) 1.89%*** (4.55) 2.05%*** (6.90) 2.10%*** (4.72) RMt - RFt -0.06 (-0.6) 0.007 (0.05) 0.09* (1.81) 0.14 (0.93) 0.10 (0.94) -0.12 (-1.09) -0.07 (-0.52) SMBt -0.15 (-1.55) -0.06 (-0.33) -0.19*** (-3.52) -0.13 (-1.29) -0.03 (-0.23) -0.10 (-1.07) -0.17 (-1.33) HMLt -0.06 (-0.47) -0.38 (-1.38) -0.15*** (-2.54) -0.18 (-1.55) 0.03 (0.24) -0.12 (-1.05) -0.22 (-1.13) R-squared 0.048 0.028 0.129 0.153 0.003 0.028 0.018
Diversified Arbitrage Systematic Stock Index Energy Financial/ Metal Discretionary
Table 7: Performance measurement using the Fama and French (1993) three-factor model with Ibbotson (1975) RATS procedure Each panel displays the monthly calendar time alpha in percentage using the Fama and Fench (1993) three factor model with Ibbotson (1975) “Returns Across Time and Securities” (RATS) procedure, for different portfolio formation periods during the 1996-2005 sample period. With the RATS technique, we presume that the betas of each calendar time portfolio vary over time as new information becomes readily available. Heteroscedasticity consistent t-statistics are in parentheses. *Significant at the 10% level, **significant at the 5% level, ***significant at the 1% level. . Calendar time period 12 months 24 months 36 months Diversified 2.31%*** 3.69%*** 4.79%*** (17.12) (19.25) (20.53) Arbitrage 1.62%*** 2.05%*** 4.12%*** (2.90) (2.55) (3.65) *** *** Systematic 1.71% 1.85% 1.90%*** (8.40) (6.85) (5.82) Stock Index 1.64%*** 2.58%*** 3.20%*** (5.01) (5.24) (5.03) 6.82%*** 8.44%*** Energy 3.71%*** (5.85) (7.08) (7.55) Financial/ Metal 2.20%*** 3.51%*** 4.25%*** (12.86) (14.29) (14.93) 3.18%*** 4.84%*** Discretionary 2.34%*** (6.78) (6.86) (7.67)
Table 8: Performance measurement using a specific multifactor model for each CTA categories This table presents the results of the estimation of the multifactor models on monthly data for the 1996-2005 sample period. Mkt = excess return of the Market Proxy; SMB = the factor-mimicking portfolio for size (small minus big); HML = the factor-mimicking portfolio for book-to-market equity (high minus low); PR1YR = the factor-mimicking portfolio for the momentum effect; Var, Skew, Kurt = the factor-mimicking portfolios for variance, skewness and kurtosis (higher 33% minus lower 33%), respectively. For each underlying, the reported coefficient of option-based factors corresponds to the double-letter code to the left: “A” means ATM, “O” means OTM (+5% for calls, - 5% for puts), “C” means call, “P” means put, “S” means straddle or strangle. t-statistics (not reported) are heteroskedasticity consistent. *Significant at the 10% level, **significant at the 5% level, ***significant at the 1% level. Diversified 4.03%*** Arbitrage -5.19%*** Financial/metal 2.19%*** 0.011 Systematic 0.45% 0.0259** 0.0006 -0.0535 0.0001 -0.0432 -0.0419 OC AP OP OS OS OP AP OP -0.0014*** 0.0084*** 0.0154*** -0.0008* -0.2538** 0.009*** 0.0142*** -0.0184** OS AP 0.0027 -0.4762* 1.2938*** -0.0341 -0.0533*** AS OS OS OS AC -1.0452 -0.0496 -0.0197 -0.0005 0.0183* -0.1314* 0.0031** 0.0012 OS OC AP AC OS AS -0.0008*** AC AP AC 0.0373*** -0.0233*** 0.0861 0.0025* 0.0029* 0.0104** -0.0575 AC OC AC OS AS AC 0.0686 0.2914 0.3233 0.0710 0.2454 0.0098 0.0112 0.299*** -0.2852*** -0.1161** OP 0.2628** OP 0.0057 0.0098 -2.3815** -0.1174* 0.0296 OC 0.0022 -0.0868 0.1017 0.168 AP 0.128 0.0957* 0.1323 0.0079 Energy -0.17% Discretionary 1.06%*** Stock index 1.68%*** 0.0077 -0.5009
Indexes Mkt SMB PR1YR Moments Var Skew Kurt Options FX-$/€ FX-$/£ FX-$/¥ IR-Credit IR-Term
0.0738 -0.0051 -0.0248
AP OC AP OS
0.0504*** -0.034** -0.0288*** -0.07***
-0.0077*** -0.0609* 0.4939
Table 9: Performance of CTAs using the multifactor model in different subperiods This table presents the performance of CTAs using the multifactor model for different subperiods. tstatistics (not reported) are heteroskedasticity consistent. *Significant at the 10% level, **significant at the 5% level, ***significant at the 1% level. Whole period 1996-2005 4.03%*** -5.19%*** 0.45% 1.68%*** -0.17%*** 2.19%*** 1.06%*** 2 subperiods 1/1996- 12/2000 1/2001- 12/2005 4.14%*** 3.38%*** ** -1.40% -0.13% 0.07%*** 0.01% 1.82%*** 0.08%*** 1.72% 1.17% 2.46%*** 3.20%*** 0.18% 0.97% Asian crisis 1/1997-6/1998 10.56% -5.32% 0.31% 2.14%** 7.05%* 3.32% 0.25%
CTA Category Diversified Arbitrage Systematic Stock Index Energy Financial/Metal Discretionary
Table 10: Performance persistence of CTAs For each formation/holding period, the sample is split into overlapping (to maximize power) periods of the required frequency during the 1996-2005 sample period. WW denotes funds that are winners in two consecutive periods; LL denotes funds that are losers in two consecutive periods; WL denotes funds that are winners in the first period and losers in the second period; and LW denotes the reverse. Each panel displays the percentage of occurrences of WW, WL, LW, and LL over the sample period, the chi-square test result (the percentage of cases where statistically significant persistence was observed in each CTA category), and the cross-product ratio (CPR). The statistical significance of the CPR is tested with a Zstatistic, which measures the ratio of the natural log of CPR to its standard error. The CPR and chi-square statistic are calculated as per Section 4.3. Panel A: 1 month Diversified Arbitrage Systematic Stock Index Energy Financial/ Metal Discretionary Panel B: 3 months Diversified Arbitrage Systematic Stock Index Energy Financial/ Metal Discretionary Panel C: 6 months Diversified Arbitrage Systematic Stock Index Energy Financial/ Metal Discretionary Panel D: 12 months Diversified Arbitrage Systematic Stock Index Energy Financial/ Metal Discretionary WW 25% 25% 25% 24% 22% 25% 25% 35% 33% 34% 31% 34% 35% 35% 37% 37% 36% 34% 36% 38% 37% 39% 40% 38% 35% 38% 40% 39% WL 25% 19% 23% 20% 23% 24% 23% 14% 11% 13% 11% 13% 13% 13% 10% 6% 12% 7% 10% 10% 10% 7% 5% 8% 5% 6% 8% 7% LW 25% 18% 23% 19% 23% 23% 22% 15% 11% 14% 11% 13% 15% 14% 12% 5% 11% 8% 10% 12% 11% 10% 4% 9% 7% 6% 10% 10% LL 25% 40% 27% 36% 32% 27% 29% 35% 45% 37% 46% 41% 37% 37% 40% 51% 39% 49% 44% 40% 41% 42% 49% 42% 52% 48% 41% 43% CPR 1.00 2.92 1.28 2.27 1.33 1.22 1.43 5.83 12.27 6.91 11.79 8.25 6.64 7.12 12.33 62.90 10.64 29.75 15.84 12.67 13.79 23.40 98.00 22.17 52.00 50.67 20.50 23.96 Z-stat 0.03 0.54 0.51 1.11 0.19 0.58 0.51 6.35 1.02 2.90 2.73 1.12 4.67 2.51 7.50 1.16 3.35 3.16 1.25 5.61 2.51 7.59 1.15 3.54 3.19 1.32 5.96 2.54 χ2 45% 14% 24% 24% 17% 41% 12% 90% 26% 79% 71% 45% 85% 62% 91% 32% 88% 87% 54% 90% 80% 90% 33% 89% 92% 50% 90% 83%
Table 11: Performance persistence of CTAs For each period we report the (n+1) period persistence conditional on (n) period persistence for the 1996-2005 sample period. Each panel displays the chi-square test result (the percentage of cases where statistically significant persistence is observed in each CTA category), The chi-square statistic is calculated as per Section 4.3.
Period CTA with 1 month persistence 3 month persistence conditional on 1 month persistence 6 month persistence conditional on previous period persistence 12 month persistence conditional on previous period persistence CTA with 3 month persistence 6 month persistence conditional on 3 month persistence 12 month persistence conditional on previous period persistence CTA with 6 month persistence 12 month persistence conditional on 6 month persistence Diversified 45.11% 4.48% 0.38% 0.04% 89% 87% 84% 91% 88% Arbitrage 13.53% 10.02% 6.73% 4.47% 25% 19% 15% 32% 28% Systematic 23.31% 4.80% 0.56% 0.06% 79% 69% 60% 88% 82% Stock Index 24.06% 6.98% 0.87% 0.07% 70% 49% 36% 87% 79% Energy 16.54% 9.09% 4.19% 2.06% 44% 28% 13% 53% 31% Financial/ Metal 40.60% 5.89% 0.60% 0.06% 85% 79% 74% 89% 88% Discretionary 12.03% 4.50% 0.91% 0.16% 62% 43% 31% 79% 69%
Table 12: Performance persistence of CTAs in different subperiods This table presents the performance persistence of CTAs for different subperiods for the 1996-2005 sample period. Persistence is based on the chi-square test result (the percentage of cases where statistically significant persistence was observed in each CTA category), The chi-square statistic is calculated as per Section 4.3. Whole period 1996-2005 45% 14% 24% 24% 17% 41% 12% 90% 26% 79% 71% 45% 85% 62% 91% 32% 88% 87% 54% 90% 80% 90% 33% 89% 92% 50% 90% 83% 2 subperiods 1/1996- 12/2000 1/2001- 12/2005 47% 44% 15% 12% 27% 20% 23% 26% 11% 23% 44% 38% 15% 9% 97% 24% 85% 62% 40% 92% 70% 97% 42% 94% 86% 56% 97% 87% 97% 42% 97% 94% 70% 97% 91% 83% 27% 73% 80% 50% 78% 55% 85% 21% 82% 89% 51% 82% 72% 82% 22% 80% 90% 27% 82% 72% Asian crisis 1/1997-6/1998 37% 16% 32% 26% 5% 47% 15% 100% 55% 90% 68% 32% 84% 68% 100% 63% 95% 85% 53% 100% 95% 100% 47% 100% 95% 57% 100% 95%
Panel A: 1 month Diversified Arbitrage Systematic Stock Index Energy Financial/ Metal Discretionary Panel B: 3 months Diversified Arbitrage Systematic Stock Index Energy Financial/ Metal Discretionary Panel C: 6 months Diversified Arbitrage Systematic Stock Index Energy Financial/ Metal Discretionary Panel D: 12 months Diversified Arbitrage Systematic Stock Index Energy Financial/ Metal Discretionary
Table 13: Performance persistence of CTAs in the extremities of the rankings We use data for the 1996-2005 sample period. The performance analysis award a “win” to those funds with a ranking in the top quartile “T”(normalized ranking greater than or equal to 0.75), and a “lose” to those funds with a ranking in the bottom quartile “B” (normalized ranking less than 0.25). If a CTA fund earns a ranking placing them in either of the middle quartiles (greater than 0.25 or less than 0.75) it is assigned a rating of NS. Only CTA funds in the top or bottom quartile are paired with a subsequent fund. As we are principally interested in the extreme performers, we only present results for WTWT (top “T” funds that are winners in two consecutive periods) and LBLB (bottom “B” funds that are losers in two consecutive periods); Each panel displays the percentage of occurrences of WTWT and LBLB over the sample period and the chi-square test result (the percentage of cases where statistically significant persistence was observed in each CTA category). The chi-square statistics are calculated as per Section 4.3. Panel A: 1 month Diversified Arbitrage Systematic Stock Index Energy Financial/ Metal Discretionary Panel B: 3 months Diversified Arbitrage Systematic Stock Index Energy Financial/ Metal Discretionary Panel C: 6 months Diversified Arbitrage Systematic Stock Index Energy Financial/ Metal Discretionary Panel D: 12 months Diversified Arbitrage Systematic Stock Index Energy Financial/ Metal Discretionary WT WT 13% 22% 16% 17% 18% 12% 17% 23% 25% 23% 26% 23% 24% 18% 28% 27% 30% 30% 23% 27% 21% 8% 29% 14% 32% 28% 31% 12% LBLB 10% 20% 13% 14% 16% 11% 14% 22% 26% 24% 25% 20% 22% 12% 27% 29% 28% 30% 24% 27% 12% 48% 30% 35% 33% 28% 31% 23% χ2 92% 1% 76% 50% 5% 98% 58% 48% 0% 4% 7% 2% 46% 49% 47% 0% 12% 15% 1% 27% 40% 70% 0% 39% 19% 0% 54% 54%
HEC-ULg Working Paper Series
September 06 / N° 200609/06 Accounting Standardization and Governance Structures Elena Chane-Alune September 06 / N° 200609/05 Investor Sentiment, Mutual Fund Flows and its Impact on Returns and Volatility Aline Muller , R. Beaumont , B. Frijns and T. Lehnert September 06 / N° 200609/04 Identities and change in public administrations: which interactions over time? Giseline Rondeaux September 06 / N° 200609/03 The evolution of the organizational identity of a Federal Public Service in a context of change. Giseline Rondeaux September 06 / N° 200609/02 El diccionario jurídico bilingüe,puente entre dos mundos doblemente extraños Bernard Thiry September 06 / N° 200609/01 Performance and persistence of Commodity Trading Advisors: Parametric Evidence Greg Gregoriou , Georges Hübner and Maher Kooli August 06 / N° 200608/15 Le processus d’apprentissage des entrepreneurs naissants Zineb Aouni August 06 / N° 200608/14 The Impact of Corporate Derivative Usage on Foreign Exchange Risk Exposure Aline Muller and W.F.C. Verschoor August 06 / N° 200608/13 Foreign Trade and Exchange Risk Exposure: A Firm and Industry Analysis of Eastern European Multinationals Aline Muller and W.F.C. Verschoor August 06 / N° 200608/12 La création d'activités par les femmes : freins et leviers Christine Delhaye , Christina Constantinidis , Manal El Abboubi , Anne-Marie Dieu , Caroline Pâques , Stéphanie Gerkens and Annie Cornet August 06 / N° 200608/11 Brand Concept Maps: A New Way of Segmenting the Market Céline Brandt and Charles Pahud de Mortanges August 06 / N° 200608/10 Growth, longevity and public policy Gregory Ponthière August 06 / N° 200608/09 Une classification de fonctions analytique : une condition pour une politique d'équité salariale Jacques Hodeige , Katrien Baert , Mieke Van Haegendoren and Annie Cornet August 06 / N° 200608/08 L’audit social : un outil d’amélioration de la qualité du pilotage social à l’heure des reformes des entreprises publiques au maroc Fatima El Kandoussi and Manal El Abboubi August 06 / N° 200608/07Deregulation and supplier relations in automotive distribution : A cross-national study Veronica Liljander , Janjaap Semeijn , Allard C.R. van Riel and Pia Polsa August 06 / N° 200608/06 Optional and Conditional Components in Hedge Fund Returns Georges Hübner and Nicolas Papageorgiou August 06 / N° 200608/05 Une vision positiviste de nombres hyperréels au moyen d’angles Jacques Bair and Valérie Henry August 06 / N° 200608/04 Les femmes repreneuses d’une entreprise familiale : difficultés et stratégies Christina Constantinidis and Annie Cornet
HEC-ULg Working Paper Series
August 06 / N° 200608/03 L’impact d’une formation à distance sur le sentiment d'efficacité personnelle: étude d’un cas concret Christina Constantinidis August 06 / N° 200608/02 Measuring operational risk in financial institutions:Contribution of credit risk modelling Georges Hübner , Jean-Philippe Peters and Séverine Plunus August 06 / N° 200608/01 Fondements et implications de la diversité organisationnelle au sein du commerce équitable Benjamin Huybrechts July 06 / N° 200607/02 Développement d'un modèle de microsimulation pour analyser l'efficience globale de stratégies récentes de prise en charge de l'ostéoporose Mickaël Hiligsmann July 06 / N° 200607/01 Propiedad y contagio en el comparatismo y traducción jurídica Bernard Thiry June 06 / N° 200606/04 How do performance measures perform ? Georges Hübner June 06 / N° 200606/03 Counting Aggregate Classifers Jan Adem , Yves Crama , Willy Gochetz and Frits C.R. Spieksma June 06 / N° 200606/02 Simulation and optimization of a shipbuilding workshop Frédéric Bair , Yves Langer , Thomas Richir and Philippe Rigo June 06 / N° 200606/01 Analyse socio-économique comparée des acteurs du commerce équitable Benjamin Huybrechts May 06 / N° 200605/02 Au-delà des labels : Du management de la RSE au management par la RSE Christine Delhaye , Manal El Abboubi and Virginie Xhauflair May 06 / N° 200605/01 Corporate International Diversification and the Cos t of Equity: European Evidence Robert Joliet and Georges Hübner April 06 / N° 200604/01 Les enjeux du management responsable dans le secteur agroalimentaire au Maroc : cas de la région Sous Massa Manal El Abboubi and Fatima El Kandoussi March 06 / N° 200603/02 Polynomial Regression with Censored Data Based on Preliminary Nonparametric Estimation Cédric Heuchenne and Ingrid Van Keilegom March 06 / N° 200603/01 L’analyse équilibrée des symptômes de déséquilibre de la PME à reprendre, facteur-clé du succès du processus de reprise : légitimation théorique et première validation empirique Didier Van Caillie , Sarah Santin , Nathalie Crutzen and Charles Kabwigiri February 06 / N° 200602/02 Penalties and Optimality in Financial Contracts : Taking Stock Michel Robe , Eva-Maria Steiger and Pierre Armand Michel February 06 / N° 200602/01 Managing the Waiting Experience at Checkout Allard C.R. van Riel , Janjaap Semeijn , Dina Ribbink and Yvette Peters January 06 / N° 200601/04 Les entreprises de survie et le développement de certains pays africains : Le cas de la République Démocratique du Congo Kéké Edgar Makunza
HEC-ULg Working Paper Series
January 06 / N° 200601/03 Impact of the collection threshold on the determination of the capital charge for operational risk Yves Crama , Georges Hübner and Jean-Philippe Peters January 06 / N° 200601/02 Ombre et cône asymptotique d’une courbe algébrique plane André Antibi , Jacques Bair and Valérie Henry January 06 / N° 200601/01 Social Responsibility among Fair Trade Actors : Extent and Prospects Benjamin Huybrechts December 05 / N° 200512/06 Foreign Exchange Risk Exposure: Survey and Suggestions Aline Muller and W.F.C. Verschoor December 05 / N° 200512/05 Mean Preservation in Nonparametric Regression with Censored Data Cédric Heuchenne and Ingrid Van Keilegom December 05 / N° 200512/04 Working or schooling: what determine children’s time allocation in Senegal? Eric Bonsang and Ousmane Faye December 05 / N° 200512/03 On the generation of a regular multi-input multi-output technology using parametric output distance functions Sergio Perelman and Daniel Santin December 05 / N° 200512/02 Product specialization, efficiency and productivity change in the Spanish insurance industry Hugo Fuentes , Emili Griffel-Tatjé and Sergio Perelman December 05 / N° 200512/01 CAP reforms and total factor productivity growth in Belgian agriculture: a Malmquist index approach Tim Coelli , Sergio Perelman and Dirk Van Lierde November 05 / N° 200511/03 Projet EVA - Equité salariale : aperçu de la littérature Annie Cornet , Anne-Marie Dieu , Christine Delhaye and Anne-Sophie Crahay November 05 / N° 200511/02 Measuring educational efficiency at student level with parametric stochastic distance functions: An application to Spanish PISA results Sergio Perelman and Daniel Santin November 05 / N° 200511/01 Multiplicity and complexity issues in contemporary production scheduling Nadia Brauner , Yves Cama , Alexander Grigoriev and Joris Van de Klundert October 05 / N° 200510/03 La Terminología a la luz de una investigación en Derecho Bernard Thiry October 05 / N° 200510/02 Presentación del Diccionario jurídico : Terminología de la Responsabilidad civil (español-francés y francés-español) Bernard Thiry October 05 / N° 200510/01 Tool Switching Problem Revisited Yves Crama , Linda Moonen , Friets Spieksma and Ellen Talloen September 05 / N° 200509/07 Asian Foreign Exchange Risk Exposure Aline Muller and W.F.C. Verschoor September 05 / N° 200509/06 Foreign Exchange Risk Exposure: evidence from U.S. multinationals Aline Muller and W.F.C. Verschoor
HEC-ULg Working Paper Series
September 05 / N° 200509/05 Gestion des compétences, impératifs économiques et préoccupations sociales: un mariage forcé ? Mélanie Antoine , Dimitri Deflandre and François Pichault September 05 / N° 200509/04 Construire l’Europe sociale dans l’entreprise : conditions et enjeux de la Flexicurité Isabelle Donnay and Virginie Xhauflair September 05 / N° 200509/03 Enjeux de la Société de la Communication Michel Hermans September 05 / N° 200509/02 Une étude didactique sur les quartiles d'une sériestatistique univariée Valérie Henry September 05 / N° 200509/01 Grafting Information in Scenario Trees: Application to Option Prices Michael Schyns , Yves Crama and Georges Hübner August 05 / N° 200508/02 Strong Uniform Consistency Results of the Weighted Average of Conditional Artificial Data Points Cédric Heuchenne August 05 / N° 200508/01 L’impact economique des intérêts notionnels. Première partie: Références à la théorie financière classique Bruno Colmant and Georges Hübner July 05 / N° 200507/03 A multiple criteria performance dynamic analysis oftelecommunications operators in the Maghreb countries Karim Sabri , Gérard Colson and Mapapa Mbangala July 05 / N° 200507/02 A multiple criteria analysis for household solidwaste management in the Urban Community of Dakar Ka-Mbayu Kapepula , Gérard Colson , Karim Sabri and Philippe Thonard July 05 / N° 200507/01 Radiographie du langage de l'économie Bernard Thiry June 05 / N° 200506/04 Financing of women-owned ventures:the impact of gender and other owner- and firmrelated variables Christina Constantinidis and Annie Cornet June 05 / N° 200506/03 Faut-il un ou plusieurs indicateurs d'exclusion sociale? Bernadette Biatour , Sergio Perelman and Pierre Pestieau June 05 / N° 200506/02 Départ à la retraite et taxation optimale Helmuth Cremer , Jean-Marie Lozachmeur and Pierre Pestieau June 05 / N° 200506/01 La chaîne de valeur integrée: le profil du nouveau cadre en logistique et quelques recommendations pour sa formation Allard C.R. van Riel May 05 / N° 200505/03 Estimation in Nonparametric Location-Scale Regression Models with Censored Data Cédric Heuchenne and Ingrid Van Keilegom May 05 / N° 200505/02 Space allocation optimization using heuristic approaches Yves Langer May 05 / N° 200505/01 Contrevérités sur le départ à la retraite Pierre Pestieau
HEC-ULg Working Paper Series
April 05 / N° 200504/01 Enquête sur les pratiques de gestion du temps de travail dans lesemployeurs de la Province du Luxembourg Annie Cornet , Christine Delhaye and Isabelle Cardol March 05 / N° 200503/02 Asian Crisis Exchange Risk Exposure of U.S. Multinationals Aline Muller and W.F.C. Verschoor March 05 / N° 200503/01 Funds of hedge funds: bias and persistence in returns Daniel Capocci and Georges Hübner January 05 / N° 200501/04 European Foreign Exchange Risk Exposure Aline Muller and W.F.C. Verschoor January 05 / N° 200501/03 The Latin American Exchange Rate Exposure of U.S. Multinationals Aline Muller and W.F.C. Verschoor January 05 / N° 200501/02 Décalage interdisciplinaire dans l'enseignement universitaire en physique : un modèle Valérie Henry January 05 / N° 200501/01 Les stocks options en période de décroissance boursière – une réflexion sur les conflits d’agence Andrée DIGHAYE 2004 /10 Analyse classique et analyse non standard : deux contextes pour définir la notion de limite Valérie Henry 2004 /09 Control and voting power in corporate networks:Concepts and computational aspects Yves CRAMA and Luc LERUTH 2004 /08 Une application de la Gestion des Risques aux risques environnementaux liés aux émissions de CO2 Gérard COLSON and Isabelle HENNEN 2004 /07 The inconsistencies between the traditional structures of social dialogue and the network organizations’ reality Virginie Xhauflair 2004 /06 The Generalized Treynor Ratio Georges HÜBNER 2004 /05 Concentrated Announcements on Clustered Data: An Event Study on Biotechnology Stocks Véronique BASTIN and Georges HÜBNER 2004 /04 Comparaison multicritère entre trois sites d’inter emulti-modalité : Liège – Duisburg – Montréal Gérard COLSON and Isabelle HENNEN 2004 /03 Corporate International Diversification and StockReturns: European Evidence Robert JOLIET and Georges HÜBNER 2004 /02 Les facteurs clés du succès des entreprises africaines :Le cas des facteurs de la performance des entreprises manufacturières de la République Démocratique du Congo Kéké Edgar Makunza
HEC-ULg Working Paper Series
2004 /01 Venture Capital Financing in Biotechnologies : A Prospective Assessment Véronique BASTIN, Michel BERGERON, Georges HÜBNER, Pierre-Armand MICHEL and Mélanie SERVAIS 2003 /07 L'analyse sexuée des budgets des politiques publiques Annie Cornet, Isabelle Cecchini and Nathalie Holvoet 2003 /06 Restructurations et nouveaux périmètres organisationnels : effets induits sur le travailleur Virginie Xhauflair 2003 /05 Supply and Demand of Venture Capital for Biotech Firms: The Case of Wallonia and Brussels Véronique BASTIN, Georges HÜBNER, Pierre-Armand MICHEL and Mélanie SERVAIS 2003 /04 Hedge Fund Performance and Persistence in Bull and BearMarkets Daniel CAPOCCI, Albert CORHAY and Georges HÜBNER 2003 /03 Grafting Information in Scenario Trees - Application to Option Prices Michael SCHYNS, Yves CRAMA and Georges HÜBNER 2003 /02 L'intervention en organisation : Revue critique des principaux facteurs de succès François PICHAULT 2003 /01 Reconsidering the Question of Coordination in a Process of Organizational Change: from Managerial Toolkits towards Social Constructions François PICHAULT and Éric ALSÈNE
This action might not be possible to undo. Are you sure you want to continue?
We've moved you to where you read on your other device.
Get the full title to continue reading from where you left off, or restart the preview.