Thought Leadership Series

An emerging taxonomy for hedge funds

THE BANK OF NEW YORK MELLON AND OXFORD METRICA

The purpose of most classification schemes to date has been to identify the observable attributes of funds that are believed to determine return patterns. Then, using these proxies for returns, indices are created which group funds with similar attributes. The indices created are then put to various uses including benchmarking, portfolio construction and, where the funds are investable, direct trading. The most common proxy is a fund’s self-described strategy. Figure 1 provides an overview of how hedge funds currently are classified for index construction. The format is loosely based on the scheme developed by Morgan Stanley Capital International in the development of the MSCI family of hedge fund indices.


The wheel diagram provides a very useful insight into the hedge fund world. It takes the three primary attributes of hedge fund strategy as the first basis of classification; (1) Investment Process, (2) Asset Class and (3) Geography. Thus hedge funds are first labelled in these three dimensions (inner wheel) and then into further sub-categories; 18 by process, 13 by geography and 6 by asset class. The primary attribute classification results in 1404 unique categories (pigeon holes) being available (18x13x6). These 1404 pigeon holes are further compartmentalised by overlaying a secondary classification scheme based on such characteristics as size, focus and sector. Each “pigeon hole” can be compartmentalised a minimum of two ways by each of these characteristics which gives rise to a classification scheme with 11,232 (1,404x2x2x2) potential categories.


“This convergence between hedge fund returns and equity returns raises several issues for the sector and investors therein”


This is convenient, as it exceeds the number of hedge funds currently in existence. The problem is that a classification scheme is meant to be a process of reduction and, ideally, there would be fewer categories than there are elements to be classified. However, it could be argued that an excess of pigeon holes allows for growth.

The classification scheme is extremely useful as a way of depicting the universe of hedge funds and for understanding where individual funds fit into the picture. It does not really help the next level of analysis, which is the construction of performance indices. There are simply too many ways to slice and dice the industry and the result would be a bewildering plethora of indices. MSCI uses a five-label approach and forms an index of any category that has a minimum number of funds classified in the same way. MSCI currently offers 190 indices using a labelling system similar to the taxonomy shown in Figure 1. The MSCI approach deserves credit for being the most comprehensive. However, the emerging industry standard seems to be somewhat more simplified.

In order to follow the progress of the emerging taxonomy, we select here the FTSE series for analysis of risk and return. Figure 2 provides a summary that shows the core classification being adopted as an industry standard. The structure is more simplified than that of Figure 1. The composite hedge fund index is decomposed into three broad categories of investment process; directional, non-directional and event-driven. These, in turn, are divided into three sub-strategies. In all FTSE offer 12 indices; 8 sub-strategy, 3 strategy and 1 composite index1. The FTSE series is selected as being representative of all public indices.

Hedge fund performance analysis

This section seeks to provide some insight into the performance of alternative investments generally, and the relative performance of various sub-strategies in terms of risk, return and correlation from January 2000 to December 2006.

Figure 3 sets out the time series performance of the FTSE hedge fund composite (FT H) versus the MSCI World Index for the period under analysis. Three features of these series are worth pointing out. Firstly, over the full interval, the FTSE Hedge Index has outperformed the MSCI and this has been achieved with much less volatility. Secondly, there is a striking difference between the performance of the MSCI in the first half of the period and the second. In the first three years, the MSCI lost considerable value whilst the hedge funds turned in a positive performance. This is precisely what they were invented to do; generate an absolute return even when markets were declining. In this first sub-interval there is a low correlation between hedge funds and equity markets globally. Thirdly, this lack of correlation is significantly reversed in the second period and the equity markets outperform the hedge funds, albeit with alarger volatility.


Figure 4 illustrates the systematic rise in correlation between hedge funds generally and equity markets, as represented by the respective indices. It shows the rolling correlation between the FTSE Hedge Index and the MSCI World Index for the period under review.

This convergence between hedge fund returns and equity returns raises several issues for the sector and investors therein. Firstly, will investors accept absolute return fee structures when excess returns are low? The spread in return between the two sectors, often called alpha, is demonstrably diminishing while fund managers are largely continuing to enjoy performance fees based on total returns. Secondly, what is the cause of the convergence? Some commentators pessimistically argue that the industry has run out of ideas and all opportunities have been arbitraged. Alternatively, it may reflect simply a general ‘style drift’ as successful funds become larger. Thirdly, regardless of the causes of the general trend, it is of critical importance for investors in hedge funds to include the issue of style drift in their analysis of the individual hedge funds in which they intend to invest. Since the analysis is based on a considerable level of aggregation, generalisations are made with caution.


Figure 5 decomposes the FTSE Index into the three major constituent elements, directional, non-directional and event-driven. In the second half of the period analysed, the directional strategies appear to have dominated whereas the non-directional strategies dominated the first half. Although the recent downturn in equity markets had most impact on the directional strategies, and these accounted for all of the downturn in the overall index, the event-driven strategies actually were the better performers in the second period after a poor start.

Figure 6 illustrates the risk and return for the various strategies. Return is plotted against standard deviation (volatility). It is clear that the non-directional strategies (red) tend to do better in terms of the return/risk ratio; that is, they tend to generate a better return for a given level of risk.

The event-driven strategies (black) tend to extend the risk range of the non-directional group. The directional strategies (grey) typically outperformed the event driven.

The highest risk strategy, as measured by volatility, is the CTA (directional). However, this group is the worst of all in terms of the return/risk or Sharpe ratio. Although there is some overlap, the non-directional occupies a lower range in the risk spectrum, followed by event-driven and directional.

In Figure 7 risk is measured in two ways; firstly, by standard deviation as in the previous figure and, secondly, as beta which is plotted on the vertical axis. Beta reports the covariance in returns relative to the MSCI World Index. A value of 1 indicates that a strategy’s return tends to fluctuate by the same amount as the market, a beta greater or less than 1 indicates that the strategy’s return is amplified or dampened in concert with market movements. Beta captures the correlation between the returns to a particular strategy and the market returns using the MSCI World Index as a proxy for the market.


All categories exhibit low beta values; the distressed and opportunity funds show the highest at around 0.3. It should be remembered that this is an average and masks the fact that correlation is increasing, which eventually will impact betas. The non-directional strategies exhibit a beta not significantly different from zero. This demonstrates that these strategiespursue investment opportunities that appear to produce absolute returns, or pure alpha.

Typically, a beta close to zero suggests little or no net exposure to equity markets. The event-driven strategies offer a range of beta typically higher than the non-directional strategies; the largest beta in the latter group is the convertible arbitrage strategy, which has a beta (0.07) slightly greater than the lowest of the former, namely merger arbitrage at 0.06, although this difference is not material. The directional strategies exhibit the greatest divergence in beta. The CTA/managed futures funds are negatively correlated with the MSCI World Index as demonstrated in a negative beta. This implies that, this group of funds is net short the equity markets. This latter group is an example of an investment opportunity, which, although the highest risk in terms of volatility, provides strong diversification of equity market risk.

There are a number of stereotypes about hedge fund investing that are challenged in analysing these data:

i. Hedge funds are high-volatility. This may not be the case as all but one index analysed has exhibited considerably lower volatility, (as measured by standard deviation in return), than the MSCI World Index. This equity index enjoys considerably reduced volatility due to the lack of correlation among its widely spread constituents. Caution is required in that individual hedge funds may be of considerable risk, which is not captured in the historic standard deviation metric; furthermore the diversification of manager risk is fairly limited in a number of these indices, which contain only a small number of funds.

Nevertheless, the composite index, which acts as a proxy for a well-diversified hedge fund portfolio, has a much lower volatility for the study period than the MSCI World Index, which is a proxy for a globally diversified equity portfolio.

ii. Hedge funds generate pure alpha. Although in aggregate hedge funds appear to have a low beta relative to equity markets over the full period, as reflected in the value of 0.1 for the FT HI, certain sub-strategies do have high betas. In addition, over recent sub-periods, the degree of correlation between hedge funds and equity markets has increased markedly.

iii. Hedge funds contribute little marginal risk to an all equity portfolio. As hedge fund and equity returns converge these vehicles are less effective diversification media.

Cluster analysis

Cluster analysis involves analysing in detail five different aspects of return for 5,282 funds, which allows one to map the hedge fund universe in terms of groupings or clusters. Each fund is either assigned to a specific stable cluster based on strict similarity criteria or it is identified as an outlier with no peers. A third possibility is for a fund to be a ‘drifter’ that moves between clusters over time. Figure 8 illustrates the process.

Figure 9 places fund 1 (ART Target Fund) at the centre and reports the distance to the other 3,018 funds analysed. Fund 2 (ABN Global Multi-Strategy Fund) represents ART’s closest neighbour whereas 464 (DB Torus Japan Fund) is its most distant and 1332 (P&A Balanced Fund) it’s second nearest and so on. Clusters are based on a full aggregation.

The topography of the hedge fund universe

The key result is that approximately 50% of the funds subjected to the cluster analysis were members of a stable cluster. The number of funds in each cluster varied from 2 to 90, approximately one third of the stable funds belonged to a cluster with more than 6 members, approximately one third reside in funds with between 4 and 6 members and the remaining third were assigned to clusters with fewer than 4 members. All further analysis will focus on the largest twenty clusters, all of which containat least six funds. In total, the largest 20 clusters contain 292 funds.


“The event-driven strategies tend to extend the risk range of the non-directional group”


Table 1 sets out the summary statistics for each of the 20 stable clusters. In addition, four indices are reported; all eligible funds (3,019); the top 20 by size (20); all outliers (895) and all drifters (522). Notice that each stable cluster is allocated a style based on the self-described style of the majority of its members. Although some clusters, such as Cluster 1 (fund of funds), are populated with funds which all have the same self-described style, this is not a necessary requirement in cluster analysis. The analysis is made without reference to self-described style. The result is that we have generated a mapping of the hedge fund universe in a risk-return plane in such a way that we are able to track performance for representative samples of the universe that have a proximity and stability in return behaviour.


Since stable clusters represent most of the major classifications, it turns out to be a reasonably straightforward task to generate indices that are composed of stable clusters for each strategy class. The risk/return plane depicted in Figure 10 is strikingly different from the pattern generated by the index approach shown in Figure 6. Aside from the obvious dominance in performance of the stable clusters of the indices in Figure 6, the stable clusters provide a sample across a wider range of return and risk than the indices. The indices are largely concentrated in the bottom left quadrant of the plane. Furthermore there is a more complete representation of the opportunities on offer in using clusters. In addition, the stable clusters represent funds that are very similar over a period of time. For this reason, we find it an extremely informative additional window into the general performance of the sector. Some features are worth highlighting. The largest cluster, which, as mentioned above, is made up of fund of funds, generates a better risk-adjusted return than most other clusters. This suggests that the genre generates good value for investors. The best risk/return ratio is generated by the outlier index. Thus it would be unwise for investors to stick to stable clusters only; in fact, the better performing hedge funds contain a significant proportion of outlier funds. Drifters, by contrast, under perform the other three indices in Figure 10 in terms of the risk/return ratio.


Interestingly, the top 20 clusters exhibit a similar risk/return ratio to the index of all stable clusters, suggesting that the size of the cluster is not a major factor in determining performance. The most important feature of the clusters is the degree of correlation among constituent members as measured by the average intra-cluster correlation coefficient reported in Table 1. Interestingly, but perhaps not surprisingly, there are a number of fund of funds clusters. This illustrates two important points.


Firstly, the analysis is able to uncover significant distinctions within the general fund of funds groupings. Fund of funds are arrayed into quite distinct clusters rather than being treated as an homogenous group. Secondly, fund of funds being made up of many funds appear more likely to rise to the surface as stable clusters than other funds. Notice that not all fund of funds have this quality, however they are represented in clusters in a greater proportion to other funds. Not surprisingly they constitute an important element of the hedge fund universe. Conversely multi-strategy funds areunder represented in stable clusters. Amaranth is notably absent from any stable cluster and it is identified as an outlier in our analysis, exhibiting a return to risk ratio below the average for this group. The OMCA tool identified Amaranth as requiring special evaluation. Its returns through June 2006 exhibit highly idiosyncratic behaviour which, although not a problem in and of itself, should have alerted investors to the divergence in Amaranth’s return over time.

The decomposition of risk into beta and standard deviation is shown in Figure 10. Again, clusters exhibit a wider range of betas than the hedge indices in Figure 7. The outliers tended to be lower beta funds than the drifters and stable clusters. The funds of funds were lower than most stable clusters. Notable exceptions are the convertible arbitrage cluster and an unusual cluster of mainly emerging market Eastern Europe funds, which obviously short the market.

The main advantage of adding cluster analysis to the evaluation of hedge funds is that, as a classification system, it is based on the consistent similarity in the observed return behaviour of funds. It adds a time dimension to the classification and thereby allows a robust means of evaluating any drift in style over time.

Additional copies of the full report are available on request from David Aldrich, Managing Director, The Bank of New York Mellon via email daldrich@bnymellon.com


Notes

styles1 The FTSE series combines special situations and distressed securities resulting in only two event-driven sub-strategy indices.