Investable Indices: A Viable Alternative to Funds of Funds? 2004 The success of investable indices has taken many hedge-fund professionals by surprise In the early 2000s, the announced market launch of the first investable hedge fund indices elicited sarcasm among hedge-fund investment specialists. In their view, investable indices were designed to lure certain institutional investors into believing that a traditional investment approach could be applied to hedge funds. These investors would, after a brief experience with investable indices, quickly opt for either direct hedge-fund exposures or a diversified approach via funds of funds. A few years on, these forecasts have been proved notoriously wrong. Since 2002, leading financial institutions such as Standard & Poor s, MSCI, Dow Jones and FTSE, as well as well-established players in the field of traditional investments, such as HFR or Van Hedge, each launched their own set of investable indices. In parallel, of course, these institutions also marketed a comprehensive range of investment vehicles aimed at generating these indices returns. The scheme was quickly successful and to date, according to the Tenessean.com website, more than USD 7bn are invested in such indexed vehicles. For the promoters of these investable indices, the enthusiasm generated by their products is mainly due to the alternative to funds of funds thus provided. If investable indices can be viewed as funds of funds, their optimality is confirmed neither empirically nor by theory This note examines the merits of a hedge-fund average-based investment approach. Initially, after a review of some research results among the numerous academic articles dedicated to the lack of representativeness characterising hedge fund indices, we shall empirically verify that these so-called indices are in reality no more than a particular type of funds of funds. In its second part, our study shall demonstrate, within the limitations of comparable elements, that investable indices have not delivered a more attractive performance than the average of funds of funds. Further, we shall remind readers of an elementary risk-management principle (the adequacy between portfolio liquidity and the investment vehicle s liquidity) applicable to hedge fund investments, and highlight that this principle would be largely flouted by using strategy-specific indices for tactical allocation purposes. Finally, we shall point out that a hedgefund-average approach is neither optimal nor by any means justified by theory. François Duc, Ph. D.
REPRESENTATIVENESS OF INVESTABLE INDICES Investable indices are derived from databases that do not provide an adequate representation of the hedge fund universe Contrary to funds of funds, hedge fund indices aim to represent a hedge fund universe. However, a comprehensive inventory of hedge funds is impossible to establish, for there is no regulatory requirement to report a hedge fund s existence or the returns it generates to any supervisory authority. The only means to access a sufficiently large group of hedge funds is by referring to one of the numerous commercial databases available. A global, non investable hedge fund index therefore consists in a central measurement of the return data on all hedge funds included in a given database, or in a subset thereof. An investable index provides estimated returns for a particular database, but exclusively refers to the open-ended hedge funds it contains. Thereby, an investable index is the constrained version of a non-investable index which, for its part, can also include closed-ended funds. Therefore, at best, an investable index can match a non-investable index in terms of representativeness; but it will never provide a better estimate 1. Furthermore, it inherits all representativeness biases of the underlying database. The lack of representativeness in non-investable indices is amply documented. Amin & Kat (2002), Brittain (2001), Brooks & Kat (2001), Fung & Hsieh (2000 and 2002), Naik (2000), Liang (1999), Park, Browman & Goetzman (1999), Posthuma & van der Sluis (2003), among others, have valued the various biases and analysed the problems encountered in interpreting the estimates provided by hedge fund indices. In an empirical approach, Amenc & Martellini (2003) demonstrate that over the same period, monthly returns of two indices representing the same strategy can reveal substantial differences. Thus, in May 2000, the Van Hedge Global Macro Index recorded a performance of -5.8%, whereas HedgeFund.net s Global Macro Index generated a return of +12% over the same period; the performance differential between both indices thus amounts to 17.8% in a single month. Among eleven strategies studied by the authors, seven reveal monthly return differentials in excess of 7% between indices dedicated to the same strategy. Also, indices intended to represent the same strategy, which should therefore reveal a correlation coefficient in the vicinity of 1 can prove to be decorrelated, as is the case of the EACM Long/Short and the Zurich Long/Short indices. Finally, as we demonstrated in a recent study, the returns of global hedge fund indices also reveal substantial heterogeneity. Maximum differentials between such indices often exceed 4.5%. In October 1998, for example, the CSFB/Tremont and Barclay/GHS indices respective performances differed by 6.7%. Over 30 months, the performance differential between two indices can thus exceed 40%. On average, two hedge fund databases share a little less than 53% of common data; therefore, the estimators extracted from two different databases cannot statistically represent the same universe. 1 This constrained-estimate mechanism remains even when the data is dutifully purged. With the HFRX Index, for example, preliminary analysis has been conducted, not on the full database but on a subset of approximately 200 hedge funds. If this procedure reduces the classification bias and enhances data quality, the database s other biases remain and their impact is proportionally stronger as the database subset decreases in size. 2
Investable indices are characterised by a high degree of heterogeneity Thus a non-investable hedge fund index at best offers an apt representation of the database it is extracted from. Since investable indices are designed to duplicate their non-investable counterparts, but with the added constraint of exclusively including open-ended hedge funds, they are subject to the same lack of representativeness. Thus, as at November 2003, the comparative analysis of four investable indices (CSFB/Tremont, HFRX, MSCI, and S&P) with regard to three simple strategic categories (Macro and CTA; Arbitrage and relative-value strategies; Long/Short equity) reveals wide differences in strategic exposures. Long/Short exposures, for example, vary between 13% and 45%. It is therefore normal to observe such heterogeneous results. 80% 60% 40% 20% CSFB Investissable Hedge Fund Index HFRX Global Hedge Fund Index MSCI Hedge Invest Index S & P Hedge Funds Index 45% 39% 22% 17% 20% 17% 13% 13% 67% 61% 48% 38% 0% / h b l Figure 1: Between indices, strategic exposures vary significantly. Investable indices also differ substantially in terms of selected managers. Of the 159 managers covered in total by the S&P, CSFB and MSCI indices at November 2003, only 14 (i.e. 9%) were selected in more than one index, and only two are included in all three indices (see Figure 2). The small size of statistical intersects and the small numbers of index constituents are undoubtedly the most striking characteristics of investable indices compared to their non-investable counterparts. Figure 2: Manager exposures vary significantly between indices. 3
Hedge fund selection within investable indices does not aim for representativeness Although certain index promoters admit that representativeness is compromised by small index constituencies, all without exception claim that this shortfall is offset by their selection efforts. In their view, the combination of certain hedge funds enables a strategy s adequate representation. This argument raises doubts for at least three reasons. Firstly, a hedge fund index constructed by combining various managers aims to estimate two effects, pure strategy and average hedge fund manager skill. In this context, the requirement of representativeness creates a major conflict of interest for index promoters. As demonstrated by Amenc, Curtis & Martellini (2003), the hedge fund universe includes more than 50% of lesser-quality managers. For the sake of representativeness, investable indices would therefore either need to exclusively feature medium-quality managers or include ill-performing managers to offset their otherwise excessive exposure to strong performers. Since the promoters of investable indices have a direct or indirect financial interest in maximising the performance of their indexed investment vehicles, such assumptions hardly appear plausible 2. Secondly, the actual management of investable indices does not seem compatible with a pure representation of strategy. Thus, at the end of February 2003, the GLC Gestalt Europe Fund replaced the Jemmco Fund in the S&P Index. Both managers follow statistical arbitrage strategies, but SLC Gestalt Europe specialises in pair trading on European stocks, whereas Jemmco applies several statistical models (including one pair-trading model) to mainly US-based equities. Hence it is difficult to comprehend how these two hedge funds can represent the same mean approach to pure strategy or induce the same diversification effects. Thirdly, even optimally combining the underlying assets of two indices applying the same strategy remains a far cry from common representativeness. For example, one may consider the optimal retrospective combination of Long/Short hedge funds included in the S&P Index and subjected to annual rebalancing in order to provide the best possible duplication 3 of the CSFB Investable Long/Short Index over the period ranging from January 4 2000 to December 2003. At the end of the period, the total differential between both indicators exceeds 27%, whereas the maximum monthly differential amounts to 9% and correlation is less than 0.76 (see Figure 3). 2 One could argue that a hedge fund index featuring several managers exclusively aims to measure the pure strategy s effects. In this case, however, one would need to combine average manager skills that completely offset one another. Moreover, it would be surprising if combining only a few managers did produce an optimal solution to a problem that has kept academic circles busy for several years now. Indeed, a number of financial researchers are currently working on modelling passive indices (i.e. indices that only represent the pure strategy s effects). To date, only certain systematic strategies and strategies characterised by a limited number of opportunities (M&A arbitrage, for example) can be efficiently represented by passive indices. 3 Different objectives can be pursued here: the minimisation of tracking error, the maximisation of correlation, etc. If final results differ depending on the objective chosen, result levels are essentially the same and prove the high degree of heterogeneity in combinations of underlying assets. 4 In this case, for illustration purposes, we backtrack on the CSFB/Tremont Investable Long/Short Index, even though this data series does not really exist. We thus create the most favourable conditions for a common representativeness of both indices (use of the CSFB s backtrack and retrospective optimisation of the S&P Index). 4
130 125 120 115 110 105 100 95 90 85 App S&P L/S Index CSFB Inv L/S Index Dec-99 Dec-00 Dec-01 Dec-02 Dec-03 Figure 3: Simulation covering the S&P s Long/Short managers and the (real and simulated) time series of the CSFB/Tremont Long/Short Investable Index. The indices are thus visibly not derived from a selection of representative managers. Index promoters choose to privilege superior-quality hedge funds. Furthermore, several index-constituent hedge funds are selectively closed, which means that they exclusively offer investment capacity to such investable indices that have selected them in advance. Investable index promoters not only select managers according to strongest performance, but also reserve capacity in advance. This is typically the working approach of all fund-of-funds managers. The only significant difference resides in the type of portfolio management applied at index level. Manager and strategy weightings are arrived at according to a systematic process (equal, capitalisation- or optimisation-based weightings) aiming to achieve database representativeness. Unfortunately, hedge fund databases cannot be relied upon to provide an objective representation of the hedge fund universe. These indices can thus be viewed as funds of funds subject to systematic management based on arbitrary principles. INVESTABLE INDEX RETURNS The valuation of investable index returns requires an examination of real returns, not of simulated data Since it is established that investable indices are a particular type of funds of funds, it can be stated that they do not provide a true alternative to these instruments. However, it is worth examining whether the type of portfolio management used in investable indices generates higher returns than average funds of funds. This, at least, is the conclusion certain studies tend to arrive at. In our view, however, reality tends to be biased in these studies, owing to the fact that they compare different types of performance. Investable indices produce four different types of returns: noninvestable index returns; the investable index simulated performance; reported returns since index inception; and the returns of the investment vehicle duplicating the index. The returns provided by the non-investable index cannot be used for comparative purposes, for they are impossible to duplicate: there is practically no likelihood of any similarity between the investable index returns and those of its non-replicable counterpart. In other words, investable-index returns are not representative of the likely performance of an indexed fund of funds. 5
Secondly, once it is admitted that investable indices are funds of funds, utilising their data prior to index launch would be tantamount to referring to a fund of funds simulated returns an error any sophisticated investor would be loath to commit. Indeed, such proforma data either is the result of an optimisation process (whereby it always tends to be highly favourable) or has influenced hedge fund selection. Although all selectors are aware of the fact that past performance is no guarantee of future results, there remains a natural trend to privilege those hedge funds that have generated attractive returns in the past. Therefore, it is not surprising that, as shown in Figure 4, all investable indices are poorer performers, immediately after their launch, than their non-investable counterparts, whereas such is not the case for simulated periods 5. Valid comparisons can therefore only be performed among non-simulated returns. 20% Performance 21.1% Non-investable index Investable Index 15% 10% 5% 9.9% 8.9% 9.5% 5.6% 5.6% 0% CSFB/Tremont (Aug. 03 - Apr. 04) HFR (Apr. 03 - Apr. 04) MSCI (Jul. 03 - Apr. 04) Figure 4: Comparative analysis of the real returns respectively generated by the investable and non-investable versions of given indices To date, investable indices have not generated stronger returns than other funds of funds on average Although the two last levels of returns are real, only the returns of indexed funds are reported to investors. Compared to publicly reported performance data (e.g. on Bloomberg), these figures may incorporate tracking fees, management fees, entrance and redemption fees, as well as tracking error-related differences. A comparison of index returns as reported by index promoters generally favours investable index versions. However, even from this viewpoint, investable indices, considered since inception, do not outperform the average of funds of funds 6. The table below compares the reported returns of four investable indices with those of the EDHEC funds of funds index, which is found to outperform all investable indices. In other terms, if the net returns of index-linked investment vehicles were used, the comparison over real investment periods would at best be highly unlikely to reveal a systematic returns superiority of indices over the average of funds of funds. 5 Thus the investable CSFB/Tremont index outperforms its non-investable version every half year on returns preceding index construction (working hypothesis: a threemonth waiting period is required between the construction and launch of an index); thereupon, performance becomes far less attractive. In terms of monthly performance, prior to index construction, the investable index offers lower returns on 39% of months, whereas the non-investable index outperforms its investable counterpart every month following the index construction. 6 The EDHEC funds of funds index is the best estimator of average fund-of-funds performance (see Amenc and Martellini, 2003). 6
Performance CSFB/Tremont Investable Index +5.6% EDHEC Fund of Funds Index +8.3% HFRX Global Hedge Fund Index +8.9% EDHEC Fund of Funds Index +13.0% MSCI Invest Index +5.6% EDHEC Fund of Funds Index +8.6% S&P Hedge Funds Index +14.4% EDHEC Fund of Funds Index +15.5% Period August 2003 - April 2004 April 2003 - April 2004 July 2003 - April 2004 September 2002 - April 2004 Table 1: Comparison of non-simulated, reported investable index returns against average (net) fund-of-funds returns. INVESTABLE INDICES AND TACTICAL ALLOCATION The emergence of global investable hedge fund indices brought along the creation of strategy-specific investable indices (e.g. convertible arbitrage). One of the particular features of investable indices is their very attractive issue and redemption frequency. Depending on the index chosen, this frequency is daily, weekly or sometimes monthly, which compares favourably to the liquidity offered by most hedge funds or funds of funds. As a consequence, strategy-specific investable indices would seem to form an adequate tool for tactical allocations among various hedge fund strategies. This active hedge-fund investment practise gained significant popularity towards mid-2001, when large numbers of hedge fund players reduced their mean hedge fund return projections. To ensure a level of performance in line with historic average, hedge fund portfolio managers were forced to either increase their leverage or favour those strategies deemed the most profitable. The justification of tactical allocation was soon corroborated by academic research (such as, for example, Amenc, El Bied, Martellini, 2002). The key problem with tactical allocation resides in its initial setup. For example, the hedge funds offering the best tactical allocation opportunities are those applying arbitrage strategies (Duc, 2002), which however also offer the least attractive liquidity (quarterly frequency). One could therefore consider that investable indices provide a means to bypass the problem of restricted liquidity while offering exclusive exposure to a selected strategy without a need to account for manager selection. The limited number of hedge funds within a strategic index amplifies the effects of selection as well as the exposure to process risk Alas, investable indices do not provide an adequate tactical-allocation tool. Firstly, as already indicated, manager selection is a key element in the construction of investable indices. This is all the more true as strategy-specific investable indices are small in size. The CSFB/Tremont convertible arbitrage investable index, for example, only includes six hedge funds and no extension of its constituency is planned in the foreseeable future. Assuming that one of these six constituents were to be affected by a fate similar to that of Lipper Convertibles 7, the fund would lose more than 40% in a few months, knocking approximately 7% off the index performance. 7 Lipper Convertibles was one of the CSFB/Tremont Convertible non-investable index constituents. 7
Certain strategy-specific investable indices are subject to substantial liquidity risk Secondly, the liquidity offered by investable indices is unrealistic and infringes a cardinal hedge fund risk-management rule, whereby all hedge funds must maintain a degree of adequacy between its portfolio s liquidity and that offered to investors. As an example, let us assume that a manager needs several months to sell his portfolio of financial assets at fair market prices. If this manager offers monthly liquidity to investors, a large number of which simultaneously request to sell out of the fund, he will be forced to liquidate assets at prices that will clearly disadvantage stakeholders interests. Admittedly, investable indices are generally not invested in funds but in managed accounts providing holders with daily liquidity. Nevertheless, a managed account is in reality made up of a portfolio where liquidity can be starkly different. Therefore, there is no true adequacy between the liquidity of investable indices and that of underlying managed-account positions. In other words, the exposure to liquidity risk is far greater than one may tend to believe, and it is when investors will need it most urgently that liquidity will most likely be refused them. This is the reason why certain index-linked investment vehicles offer attractive liquidity but often charge dissuasive redemption fees. SUBOPTIMALITY OF INDEXED HEDGE FUND MANAGEMENT An indexed approach to hedge funds has no fundamental justification Thus far, we have upheld the view that hedge fund indices are merely a particular type of funds of funds, characterised by lower returns as well as greater risks due to the illusory liquidity they normally offer. In the fourth part of this note, we shall endeavour to demonstrate why investable-index promoters would have no interest in creating true indices if they could, and why certain indices have deliberately forgone attempts at representativeness by maintaining small constituencies. Indexed hedge-fund management can be perceived as the application to hedge funds of a relatively common and approach in traditional financial markets, where its advantages are universally recognised. Under the assumptions of market efficiency, the price of an asset results from the mechanism involving offer and demand and thus reflects that asset s value as it is estimated by all market participants. Over the long term, investing in the market s portfolio corresponds to the most profitable form of passive management and reflects an optimal diversification of systematic risk. For hedge funds, however, there is no fundamental reason justifying passive investment in the average portfolio made up of all hedge funds. Creating a weighted portfolio including all managers is a venture devoid of sense. Likewise, applying weightings by market capitalisation is not adequate to capture returns on investments in hedge fund strategies. Firstly, hedge funds can be leveraged by means of capital borrowings; hence, two equally sized hedge funds can reveal highly different levels of exposure to a given strategy. Moreover, hedge fund strategies are not exclusively reserved to hedge fund operations: the trading desks of financial institutions 8, certain private investors and multi-strategy funds also make liberal use of hedge fund strategies. 8 Most hedge fund managers are former trading desk operatives. 8
Finally, a hedge fund s capitalisation does not reflect its performance or financing needs, but is rather an indicator of capital flows into a given strategy. And these capital flows have a marked tendency to favour recent winners, as highlighted by Agarwal, Daniel & Naik (2003). Thus, in the field of arbitrage strategies, Figure 5 illustrates the strong correlation between a given year s capital flows and the previous year s strongest average performances. In other words, capital flows into specific strategies demonstrate that, on average, investors track past performance. Such an allocation process is clearly sub-optimal in that it runs contrary to the existence of strategy cycles and the proven absence of average sustainability in hedge fund returns. An indexed approach to hedge funds is contrary to investor interests An indexed approach to hedge funds is also sub-optimal from the viewpoints of performance, volatility and diversification, for it tends to negate the core driver of hedge fund returns: manager skill. 8000 6000 4000 2000 0-2000 Year's annual capital inflow (USD millions) Previous year's returns 1994 1995 1996 1997 1998 1999 2000 20% 15% 10% 5% 0% -5% Figure 5: Capital flows into arbitrage strategies and average strategy returns for the previous year Indeed, the hedge fund universe is made up, among others, of inferior-quality managers. Aiming to invest in that universe s mean portfolio is equivalent to seeking an exposure to these mediocre hedge funds. As demonstrated by Liew (2002), selection becomes rapidly profitable for a given universe 9 of hedge funds. Conversely, increasing the number of hedge funds within a portfolio has a drawback. As shown by Lhabitant & De Piante Vicin (2004), the marginal gain on portfolio volatility is relatively low beyond approximately ten to fifteen managers, whereas correlation with equity markets continues increasing even beyond 30 managers. Diversification among managers is quickly achieved owing to the large variety of hedge fund approaches. However, diversifying hedge funds further amounts to diluting manager skill and thus to reduce the diversification of a traditional portfolio. In theory, a portfolio including about ten well-selected hedge funds would be optimal. The general investment practice of holding between fifteen and thirty managers within a fund of funds thus represents a realistic compromise and accounts for non-market risks. Increasing the number of hedge funds beyond this level merely improves correlation, without any further advantages. For this reason, certain promoters deliberately maintain a low number of managers within their indices (40 to 60 funds), thus abandoning hypothetical representativeness to preserve some diversification. 9 The contrary perception investor may have when comparing non-investable index returns with fund-of-funds performances is due not only to all biases and methodology issues, but also to the fact that they do not operate in the same universe. 9
CONCLUSION It is commonly admitted that all indices should meet at least four requirements: to be representative, easily duplicable, transparent, and subject to little constituent turnover. Investable hedge fund indices do not meet these conditions. Owing to the biased databases they are derived from, they are not representative. They are only investable for their promoters, since certain constituent hedge funds do not accept fresh investor capital. They are not all transparent if one considers that, for example, it is quite difficult to obtain underlying HFRX index constituent names. Finally, if hedge fund indices were to be truly representative, manager turnover rates would need to be significant. Indeed, today s hedge fund universe includes 12% of funds that cease operating after one year s activity. Hence, hedge fund indices should reflect this aspect of the underlying universe by providing an at least equivalent turnover rate 10. As pointed out by Prof. Schneeweiss in the April, 2003 issue of Absolute return, Calling something an index does not make it an index. Furthermore, it is relatively clear that promoters have no interest whatsoever in creating true-to-form indices. Indeed, no theoretical or empirical argument can be found to support an indexed investment approach to hedge funds. On the contrary, it appears that for any given set of hedge funds, an indexed approach may be suboptimal both from a return and diversification viewpoint. Thus hedge fund indices are funds of funds in disguise. It is revealing, in this perspective, that investable indices are always created in conjunction with investment products for which they act as underlying assets. If non-investable indices were launched by more or less neutral database promoters, investable indices are products designed by financial institutions to be sold and profitable. To date, investable indices on average have not outperformed funds of funds as a whole. On can however expect these vehicles to be more profitable than average at certain times in the future. Nevertheless, given that the indexed approach is suboptimal for any given universe, it is erroneous to believe that these products may outperform other funds of funds over the long term. Still, investable indices do carry certain advantages. For hedge fund analysts, representativeness provides an efficient excuse for selection errors. For the managers of indexed investment vehicles, the task consists in managing a fund which has its own performance for benchmark. For the distributors of index-linked products, the confusion created by the different levels of return, as well as the index hypothetical representativeness enable the alternative use of the pro-forma figures, gross data or even data pertaining to another product. Finally, to the managers of institutional portfolios, choosing an allocation to a product reputed to be an index provides an ideal disclaimer of all selection-related liability. If one excludes the advantages of responsibility delegation, investable indices, owing to 10 For indices featuring capitalisation-based weightings, instability is thought to be mainly due to constantly changing relative capital flows into strategies. 10
their heterogeneity, do not simplify institutional investors selection task (differences in construction methods, in strategic exposures, in selected managers, and among the databases they aim to represent). Moreover, this phenomenon cannot be discounted as an initial shortfall due to the small size of indices. Indeed, non-investable indices comprising more than 1 000 managers are also subject to a high degree of heterogeneity. However, the success of investable indices cannot be solely explained by marketing considerations and the opportunity to delegate responsibilities. There is real demand for an exposure to hedge fund strategies that does not depend on manager selection. Institutional investors could thus create a core portfolio that would, in a sense, provide them with the Beta of hedge fund strategies, and complement it with a portfolio including a few carefully selected hedge funds in order to capture their Alpha. Hedge fund selectors could strictly limit their exposure to manager skill by short-selling strategic indices and buying into the best hedge funds in the same strategy. Such legitimate approaches require the creation of so-called passive indices, i.e. indices that exclusively reflect a strategy and disregard the added value of manager skill. However, such an index cannot be obtained by aggregating allocations to a vast number of hedge funds. It rather requires a duplication of strategies by means of derivative products and systematic management. Unfortunately, the road remains long before such indices are correctly constructed and offered to investors; some even deem the task impossible. Meanwhile, investors in investable indices are choosing the wrong product. ACKNOWLEDGEMENTS The author wishes to thank all participants in 3A s annual conference on 12 February 2004. Their highly relevant input has added substantial value to this note s contents. 11
Bibliography Agarwal, V., Daniel, D., and N.Y. Naik, 2003, Flows and Performance in the Hedge Fund Industry, Working Paper, Centre for Hedge Fund Research and Education. Amenc, N., El Bied, S., and L. Martellini, 2003, Predictability in Hedge Fund Return, Financial Analysts Journal. Amenc, N., Curtis, S., and L. Martellini, 2003, The Alpha and Omega of Hedge Fund Performance Measurement, working paper, EDHEC-MISYS Risk and Asset Management Research Center. Amenc, N., and L. Martellini, 2003, The Brave New World of Hedge Fund Indices, working paper, EDHEC-MISYS Risk and Asset Management Research Center. Amin, G.S., and H. M. Kat, 2002, Welcome to the dark side: hedge fund attrition and survivorship bias over the period 1994-2001, working paper, Cass Business School. Brittain, B., 2001, Hedge Funds and the Institutional Investor, Journal Of International Financial Management and Accounting. Brooks, C., and H. M. Kat, 2001, The statistical Properties of Hedge Fund Index Returns and Their Implications for Investors, working paper, Cass Business School. Duc, F., 2003, Hedge Fund selection and Allocation by Investment Style : Empirical Evidence, Banque Syz & Co. Duc, F., 2004, Hedge fund Indices : status review and user guide Banque Syz & Co. Fung, W., and D. A. Hsieh, 2000, Performance Characteristics of Hedge Funds and Commodity Funds: Natural versus Spurious Biases, Journal of Financial and Quantitative Analysis. Fung, W., and D. A. Hsieh, 2002, Hedge fund benchmarks: information content and biases, Financial Analysts Journal. Park, J., Brown, S., and W Goetumann, 1999, Performance Benchmarks and Survivorship Bias for Hedge Funds and Commodity Trading Advisors, Hedge Fund News, August 1999. Lhabitant, F., and M. De Piante Vicin, 2004, Finding the sweet spot of hedge fund diversification, Capco Institute Journal. Posthuma, N., and P. J. van der Sluis, 2003, A reality check on Hedge Fund Returns, working paper, ABP Investments. 3A S.A. Alternative Asset Advisors Rue du Rhône 30 1204 Geneva Switzerland Tel: +41 (0)22 819 9800 Fax: +41 (0)22 819 0904 E-mail: 3a@3-a.ch 12