Originally Posted by DaveDr89
good point RE the perfect fit to the data. Portfolio optimization problems are well-defined and the hard part is the estimation of the inputs, viz., the covariance matrix and the expected returns of the universe of investment choices. How are you estimating the covariance matrix? If you are estimating it with the sample covariance matrix then that is a sign that you guys still have some homework to do.
RE leveraged ETFs, hopefully you guys also know that these are only designed to be in sync with the advertised multiplier on a daily basis and that for longer holding periods they may be in the wrong direction? There is a lot of literature on this.
Yeah, we're aware of that problem with leveraged ETFs. However, with our system we aren't holding anything for longer than three business days, so we anticipate any effect should be very very small.
As for covariance matrix, you'd need to talk to my partner on that one.
Edit: want to be clear on the issue of perfect fit -- the ground truth for our validation sets are usually somewhere in the 100,000,000% to 1,000,000,000% range. We are obviously not getting results anywhere near that, and we are certainly not getting perfect fits. We're getting decent fits that consistently offer solid returns in backtesting, but nothing even close to perfect. There's a lot of losses mixed in there.Edited by NameBack - 10/25/11 at 5:02pm