“Measuring Unobservable Risk”

That is a paraphrasing of the title of an article in an issue of CFA magazine I found around my apartment. It caught my attention. Particularly because I’m inclined to think anything that makes that sort of claim is either BS or dangerously misleading.

So here’s the mechanism to the ‘omega-score’:

  1. Correlate returns data to ‘operational issues’ (“These include criminal, civil, and regulatory issues.”) disclosed in a regulatory filing *some* hedge funds have to make.
  2. Establish that funds with these issues perform poorly. No surprise. They use something called ‘canonical correlation‘, which assigns weights to a set of variables (‘operational issues’) and tests their weighted effect on returns. Without reading the paper, it’s tough know what the ‘omega-score’ is, but it’s probably some kind of representation of the correlation.
  3. Now find some data that is publicly disclosed by all hedge funds. These are the new variables.
  4. Test various correlations that match the profile of the #2 variables. This is how we set some kind of statistical equivalence between the variables we know have an effect on the returns (from #1) and those we suppose have an effect (the ones that are always disclosed).

That ‘canonical correlation’ thing is a clever technique that I don’t know much about. One thing to watch out for is that we’re talking about establishing a statistical artifact that has no direct intuitive basis.

In other words, there’s a difference between saying “hey, I think that these two or three variables will have an effect on returns” and saying “hey, these variables might be related to these other variables that probably have an effect on returns”. Gets messy.

And the focus is a bit confusing. What we’re really doing is correlating ‘operational issues’ with variables that heretofore had not been associated with that kind of nastiness. If I’m right that that’s what we’re doing, I’d have two questions:

  • Why do we need to do this for each fund individually? If variable x is associated with bad behavior at one fund, wouldn’t it be similarly associated at another?
  • Aren’t the findings of the correlation study interesting in of themselves? If you can get better at figuring out which funds are in legal trouble, why risk building a statistical white elephant trying to link it to returns results?

Anyway, really interesting stuff. This kind of study demonstrates the power of even modest bits of information if they’re consistently and universally disclosed.

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.