Professor Harry Kat of the Cass Business School at City University, London, has been a thorn in the side of the hedge fund industry for several years now. And yesterday, he released his latest attack on the sector, reserving his sharpest barbs for the hedge fund replicators themselves. In fact, he concludes his analysis by replicating both Goldman Sach’s (NYSE:GS) and Partners Groups own hedge fund replication offerings.
The latest paper is an extended version of Kat’s earlier work “Alternative Route to Hedge Fund Replication”. He adds over 20 pages of succinct and pointed commentary on everything from the research of Andrew Lo, Thomas Schneeweis, Lars Jaeger, Bill Fung and David Hsieh to the hedge fund replication offerings of Merrill Lynch (MER), Goldman Sachs, JP Morgan (NYSE:JPM) and Partners Group.
If you are new to the whole “hedge fund replication” shtick, this is a must-read. And if you have been following this story for some time, you will find Kat’s viewpoint typically concise - even if you don’t agree with it.
We asked Kat why he chose now to revise his earlier work. Using his characteristic Robin Hood bravado, he explains:
First, with the media attention given to the billion dollar bonuses that some hedge fund managers have taken home again, more and more investors are now realizing that, to a large extent, the hedge fund game is all about asset gathering and pocketing unjustifiable fees. Furthermore, investors realize that in the end, they are paying for it all.
He also told us that newly published results from various new hedge fund replication products now allow a more thorough analysis of their techniques. Kat tells us:
…some more return data has become available over the past few months, which for a first time allows us to properly analyze the various products’ performance and merits as portfolio diversifiers. The results are very interesting and confirm what I’ve suspected all along: that, despite the different marketing stories, the main product providers all offer more or less the same product. In addition, it shows that these products have a very high correlation with the stock market and therefore make lousy diversifiers.
The extended paper covers what Kat calls “a number of problems” with the factor modeling techniques studied by Hasanhodzic & Lo, Thomas Schneeweis, and Jaeger & Wagner: missing factors, normality assumption, execution costs, and lack of dynamic trading. While these issues were covered in the first version of this paper, what follows below was not.
Kat proposes a method of “scoring” the success of factor models. His criteria: model fit (r-squared), out-of-sample prediction, and back test fit. He proceeds to validate three prominent techniques. His conclusions:
Hasanhodzic & Lo (2006): “…only explains 15-20% of the variation in individual hedge fund returns. Obviously, a model that leaves 80%-85% of a fund’s return variability unexplained is unlikely to provide a very fruitful starting point…” Schneeweis et al (2003): “…despite a much higher systematic component in the index returns, the factor model used is unable to accurately replicate the returns of the (various hedge fund sub-) indices." Jaeger & Wagner (2005): “…straightforward strategies, like long/short equity score quite well, but more complex strategies, like managed futures and equity market neutral, come out a lot worse.”
Kat goes on to say that the Jaeger & Wagner technique “is not very accurate” when applied to out-of-sample data. (Although he does not seem to subject the other two techniques to this test).
Factor-Based Hedge Fund Replication Products
Finally, Kat determines if factor modeling techniques (encapsulated in products such as the Merrill Lynch Factor Index, the Goldman Sachs Absolute Return Tracker, and the JP Morgan Alternative Beta Index) could have predicted the returns actually recorded by the HFRI Index. In a departure from his overwhelming skepticism about these offerings, he acknowledges that their returns are indeed highly correlated with the actual results of the HFRI Fund of Funds Index. But his unilateral cease-fire is short-lived as he says that explaining such a broadly diversified index as the HFRI FoF Index is bound to produce a high r-squared since idiosyncratic elements cancel each other out.
While Kat only provides correlations between these products and the HFRI Composite, there appears to be a very high correlation between each product and the HFRI FoF Index too. Below is Kat’s comparison of the HFRI Indices and the Merrill Lynch Factor Index (others contained in paper):
But how can Merrill, Goldman and JP Morgan hit the mark so well when Lo, Schneeweis and Jaeger can only explain a portion of hedge fund returns? The answer, says Kat, lies in the fact that the recent crop of hedge fund replication products aim to track only a broad index - not any particular sub-indices. Says Kat:
The largest obstacle standing in the way of accurate factor model-based replication is the fact that hedge funds dynamically trade in and out of asset classes…Funds following different strategies will trade differently and their trades may (at least partially) cancel each other out. As a result, dynamic trading becomes less of an issue. It is exactly this phenomenon, which allows Merrill, Goldman and JP Morgan to obtain such accurate replication. The HFRI Composite is a basket of over 2000 hedge funds, following a large variety of strategies. Since it is extremely diversified, the peculiarities of the various strategies will largely diversify away. The result is an index with little hedge fund-like properties left, containing mainly equity and credit risk.
While these broadly diversified replication products diversify-away the dynamic trading element of hedge fund returns, Kat wonders whether the so-called “mechanical trading rule approach” embodied by products such as the Merrill Lynch Equity Volatility Arbitrage Index, the Merrill Lynch FX Clone, the Deutsche Bank Currency Return Index, and the Bear Stearns “Mast” (Fixed Income) Index might yield fruit.
To answer this question, he picks on Partners Group’s ABS Index because it “is a collection of 18 mechanical rules-based strategies, similar to a multi-strategy hedge fund”. While the ABS has beaten the HFRI Composite handily since January 2000, Kat says it has a volatility that is a third higher than the Index. To adjust for this, he de-levers ABS for an apples-to-apples comparison. He concludes that the ABS has tracked the HFRI Composite quite closely since the turn of the century.
However, he points out, the delevered ABS under-performs the Index since 2004 - about the same time, says Kat with tongue in cheek, that the ABS was launched. His conclusion:
From an intellectual perspective, the ABS fund makes for a more appealing story than factor model-based products. Unfortunately, this does not appear to translate in more appealing returns. In a way this makes sense. The ABS fund mechanically replicates a number of hedge fund trades and then combines these into one single portfolio. Because it mixes a variety of strategies, however, the ABS portfolio will be quite similar to the hedge fund portfolios that factor model-based products aim to replicate.
The “Fund Creator Approach”
In 2006, Kat and partners launched a software tool he says will revolutionize the hedge fund industry. Unfortunately, given the proprietary nature of the tool, its inner workings aren’t publicly available. However, Kat explains the basic principles and illustrative results in this paper. As Kat points out frequently, his approach does not aim to replicate the month-to-month returns of hedge funds. Instead, it aims only to replicate the statistical properties of the returns (the volatility, skew & correlation to a predefined benchmark). It just happens that the resulting mean monthly returns of such statistical distributions are very close to those produced by the actual funds his approach aims to replicate. In other words, the market compensates hedge funds for their statistical qualities by simply paying a fair market price for them.
He includes an example taken from the Fund Creator website showing the resulting mean return of distributions with various statistical qualities. He shows that increasing the standard deviation ceteris paribus yields a higher mean return (in keeping with CAPM). But he also calculates the effective ”cost” of increasing the skew ceteris paribus (in terms of a decrease in mean return). In theory, we suppose this effect can also be described using options pricing formulae. But most interesting, in our opinion, is that he essentially calculates a “cost” for an increasing negative correlation with a 50/50 stock/bond portfolio.
In any case, the bottom line with the Fund Creator approach is that it can produce return distributions very close to the HFRI.
The paper includes some illustrative examples of the impact of a low correlation to a fictitious 50/50 portfolio.
Replicating the Replicators
To conclude, Kat turns the tables on the hedge fund replicators - actually trying to replicate them. He compares Goldman’s ART Index and Partners Groups’ ABS Fund to Fund Creator’s best attempt in order to settle this once and for all. Unfortunately, while Kat finds the market correlation of ART and ABS to be too high for comfort, the results of this comparison are somewhat less than clear:
…over the period studied, the ART Index and the ABS fund have generated neither superior nor inferior returns.