Is there any evidence that hedge funds (any in particular, or the industry as a whole) actually offer any expertise?

Clearly, they claim that they do (that is what management fees are for after all), but after reading many things like:

“They’ll have to be very good at manager selection,” Mr. Lack said, “because there’s been very little return persistence. I looked at one-third of the hedge fund industry. Of those in top 40 percent of returns, only 7 percent stayed there throughout the period.”

NYTimes though I have seen statements like this in other places.

Now, I assume that one way of testing this would be to do P-Test and see what the significance is. Unfortunately, I do not know anything that I need to use the test. I don't know the performance of the funds, a reasonable benchmark, a real null hypothosis, etc.

  • Is this even a reasonable way to go about it?
  • Are there studies that address this in a rigorous manner?
  • On a more minor note, is my intuition generally speaking OK (that one would expect persistence in "games" of skill.)?
  • Possibly more on topic on Cross Validated? Not off-topic here, per se, though.
    – Sklivvz
    Oct 12, 2012 at 21:43
  • @Sklivvz I have no experience with either site really, whatever you think is better
    – soandos
    Oct 12, 2012 at 21:45
  • This is a slightly more specific version of this question: skeptics.stackexchange.com/questions/11226/… .
    – matt_black
    Oct 13, 2012 at 14:34
  • 1
    Are you sure you mean “p-test” rather than, say, “t-test”? I’ve never heard of the p-test, especially not in the context of p-values, and the only mathematical-looking Google hit refers to some other, unrelated concept of that name. Nov 30, 2012 at 21:48
  • @KonradRudolph I have no idea. I never took any stats
    – soandos
    Dec 1, 2012 at 23:28

1 Answer 1


Performance attribution is a very major field of ongoing research. The most simple method of skill gauging is to look at a linear model between the fund's periodic returns and a number of risk factors which are already known to be relevant to the cross-section of asset returns. If there is a statistically significant intercept estimate, then this is attributable to the skill of the manager. The reason for this is that the manager is seen as generating returns that are gained beyond how much they're exposed to the specified risk factors.

A typical specification for an equity fund would be:

enter image description here

Here's some evidence from a model similar to this one:

enter image description here

So if we accept the alpha generating model used in the table to be the equilibrium model, our answer is YES. However, the question always remains as to whether we've included all the relevant risk pricing factors into our model. If the manager was to concentrate on very high leverage and very illiquid firms then we might get a very large, positive and statistically significant alpha, even if this manager wasn't actually being skillful.

The problem with performance attribution to a single fund is that asset prices are very random. For example, if we think of stock prices as evolving according to some stochastic process (such as Geometric Brownian motion):

enter image description here

and we tell 1000 funds to invest randomly in a number of these processes ... Well, there will be a large proper subset of these that look "skillful" from the outside. Yet they just got lucky.

Future questions of this sort are better suited on quant.stackexchange.com.

Here's some reading material:


However I think it'd be more beneficial for you to simply google "performance attribution" (or an analogous string).

Also, hedge funds lack transparency, are very difficult to monitor, are very hard for retail clients to invest in (they cap the number of investors so that they can avoid SEC regulations), and so are not recommended for clients that aren't very sophisticated. They also attempt to gain large exposure to very specific risks so are not suitable for retail clients.

Other important points:

  • A single fund with a high alpha doesn't mean much. You have to make sure that the equilibrium model is correctly specified. Also you will have 5% type-1 errors due to the definition of p-values.

  • You have to take into account transaction costs.

  • There are many many other ways to measure performance. I've described only one way to measure performance from an external stand-point.

  • 1) Why is such a linear combination a good model? Are none of the factors something other than linear? 2) What is epsilon in the equation? 3) Is there a link to a paper that you can include here? 4) Doesn't your final point illustrate that the way you are determining statistical significance to be flawed?
    – soandos
    Nov 30, 2012 at 13:01
  • 1
    @soandos All good questions, and I haven't done any research on performance attribution so I am certainly unqualified to answer your queries properly. However, I'll try; (1) I think that the factor model representation follows from Arbitrage Pricing Theory, which purportedly has solid theoretical underpinnings. Also, this type of model is easier to estimate than non-linear models, which may also contribute to its prevalence. (2) The epsilon is a random disturbance; assumed to be distributed with mean 0 and constant variance (see Gauss-Markov assumptions of OLS). (cont'd)
    – Jase
    Nov 30, 2012 at 13:09
  • 1
    (cont'd) I recommend googling Ordinary Least Squares. Basically, we can say the fund's returns (R_t) are a mix of compensations for taking on known risks (e.g. R_m, R_mom, etc etc), "skill" (i.e., alpha) mixed with some randomness that can be diversified (epsilon). (4) Stocks don't actually follow Geometric Brownian motion, as numerous factors (such as value, momentum and size) have been shown to be relevant to stock returns. I was just making a point that stocks behave as if they were mostly random, and so inevitably there will be some funds that look great even if they don't have skill.
    – Jase
    Nov 30, 2012 at 13:13
  • (cont'd) For (3), see my edit.
    – Jase
    Nov 30, 2012 at 13:15

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .