I'd like to add that difficulty in measuring independence of recommendations makes the task of trying to estimate the likelihood of a particular result for a pe

I'd like to add that difficulty in measuring independence of recommendations makes the task of trying to estimate the likelihood of a particular result for a person or fund being a real out performance extremely difficult. If independence is underestimated (or not accounted for at all), the performance will look far more statistically significant than it actually is. There are other issues too, like non-gaussian return distributions, but I think estimating independence is the killer. Note that dependencies between securities are non-linear, eg increased co-movement of securities during times of high volatility. We're trying to model all this at http://backrecord.com http://backrecord.com We're very close to releasing something..