Tools, Techniques, Essays and News that will help you pick the best CTAs, Hedge Funds and Alternative Investment managers. No nonsense. No excuses. No easy answers.
Oxford Metrica and Cluster Analysis: Hedge Fund Style Analysis du Jour.
Some clusters are more interesting than others
Several websites have reported that The Bank of New York Mellon and independent research firm Oxford Metrica have released a study titled, “Rethinking performance in the hedge fund industry.” According to MoneyMarketing, “The convergence between hedge fund and equity market returns could cause confusion on how such funds should be used to diversify investment portfolios, says the Bank of New York Mellon.
“A study conducted by the bank and research firm Oxford Metrica shows that combined with inconsistencies in hedge fund classification, the confusion could result in unrealistic return expectations for investors.
“The report recommends that hedge funds be classified using cluster analysis instead of the traditional classification by strategy. Cluster analysis groups funds according to the observed behaviour in their returns, as opposed to management styles."
What exactly is cluster analysis and why is it such an improvement? Or is it an improvement? Cluster analysis is not a technique, but a group of statistical techniques that attempts to group or cluster things without being told how to group them. A statistician performing a cluster analysis points a program at a set of data and says, in effect, “Divide that data into groups.”
This approach has profound advantages and profound disadvantages over other approaches to investment style. On the one hand, cluster analysis has produced real and important results in non-investment research. Even better, this technique completely ignores the statistician’s preconceptions. Considering how often our preconceptions poison our investment work, this is no small advantage.
One of the more useful tools. No kidding.
On the other hand, cluster analysis is difficult to manage. Have you seen those late night TV advertisements for shrink-wrapped investment software that all but guarantee to make you rich? Strictly speaking, anyone who buys such software can call himself a quantitative analyst. The difference between buying such software and punching the keys almost at random and doing real quantitative analysis is great. But the gulf between buying a software package and performing cluster analysis and performing a real cluster analysis is greater still.
It is not too hard to find out what is going on beneath the hood of most shrink-wrapped investment software. Shrink-wrapped investment software usually comes with a manual which explains what a moving average is and how the software performs pro-forma accounting. But the manuals for cluster analysis software may only provide you with abstract descriptions or explanations in matrix algebra. Hire a statistician to explain cluster analysis and he might well say, “Let’s start by considering N-dimensional space…” Trust me on this one; it’s going to be a long night.
As I mentioned before, there are many types of cluster analysis. If you point a number of cluster analysis techniques at a set of data and they all come to the same conclusion, then those are your clusters. Your next task is to discover what the clusters mean. A cluster analysis program will typically label your clusters one, two, three and so forth, names that do not provide much insight; that is your problem. Worse, if the various clustering programs do not come to the same conclusion, what are you going to do then? I presume that Oxford Metrica would respond that this is not an issue because you can trust them to do the job right. Perhaps. The descriptions I’ve read of their work so far do not give me any comfort. I have not seen their report, understand. I have contracted Oxford Metrica, of course, but apparently I am not worthy of their notice. I assume here, and this is a strong assumption, that the reports I read in the press are accurate. I will discuss the technical issues further when I have seen a copy of their report.
I don’t mean to suggest that cluster analysis won’t help you find hedge fund styles. I do mean to suggest that cluster analysis is almost certainly of less value than the people at the Bank of New York Mellon and Oxford Metrica think it is. This isn’t the first time the use of this technique has been announced. I used the technique to group hedge funds some fifteen years ago and then again five years ago with some success for two separate clients. While I would like to claim to have introduced cluster analysis into the investment industry, I cannot. Someone else did that over twenty years ago, I think. The technique keeps getting discovered and forgotten. This is not because the technique is without value. It is because the technique is almost always oversold.
Comments and Critiques
"If your mother says she loves you, check it out." --Old reporters' motto; also our motto. Copyright (C) 2005, 2006, 2007, 2008, 2009 Fred Gehm. All rights reserved.