Discussion in 'Business, Careers & Education' started by NameBack, Oct 19, 2011.
What sort of return does it give over the last five years? Things like changes between high and low volatility, tranding/non-trending, and a heap of other conditions can render a winning algorithm into a serious loser. Watching some of the Collective2 algorithms fall under a bus was illuminating. You probably have all this covered already but if not...
This doesn't sound like it will end well.
How is that possible?
He's run a series of algorithms that tell him when to go long and when to go short. Mix that with some leverage and you have your 200%
I'm not a finance expert but isn't this something they refer to as "too good to be true"?
Oh definitely not. That's the big mistake, right there. No, we built it on a preceding 5-year period. We also validated it back into the past, 00-05, where it earned 650%.
When we validate it against the same dataset it's built in, it gets a basically perfect fit, so the results are somewhere in the millions of percent, so it's pretty easy to tell if you're validating on the wrong set or not.
We're sure the returns will go down when rubber meets road, but we're glad that they look this good on paper. And we're confident our methods for backtesting are sound -- no future data seeping into our calculations, taking into account transaction costs, etc.
It took a lot of work and some entirely ground-up creativity to get the model where it is today. It definitely wasn't validating this high when we started. When we started, it sucked. We just stuck with it.
Loves crashes. Hates slow steady growth. Performs great in 00, 08, current market. Performs medicore in basically all of the '90s, 2003-2007, etc.
It still makes money, but nowhere near as much.
Yeah we're definitely starting there. Looks like we might be getting 30k (measly for finance, I know) for a test fund soon.
Well, right now we're moving ahead with a portfolio simulator to build a track record for when we start trying to raise seed-capital for our first year or two. But once we're raising seed capital, honestly we'll probably be taking any and all amounts. Our plan is to run a small test fund for at least a year (not a huge sample, but a decent size), and use that track record to bring in larger investors going forward. We estimate our first year's operating costs at <$300k, including money for the test fund. But we do need a proper prospectus.
Edit: As I mentioned above, we may be getting money for a small test-fund ahead of schedule, which we weren't really looking for, but we have someone we know who wants to invest right now, for tax purposes.
Dude, I know very little about the nuts and bolts of this business but any results in the 200% or greater range would seem to indicate a flawed algorithm. Nobody is making that kind of money and to think you found the magic beans is a bit silly IMHO.
the people at Renaissance Technologies are.
Jim Simons is a baller. But he has gotten 38% annualized returns (since the mid 90s I think) after fees (5%, 36% probably the highest in the industry). Still nowhere close to 200%
It's not 200% annualized. Average annualized it's probably about in the 40-50% range -- which is still obviously very high -- because it performs better in these high-volatility markets and then slows down quite a bit when the market moves into a steady bull mode. And again, we do anticipate some falloff from the numbers on paper, just nothing incredibly dramatic. I'd be happy if we ended up doing half that in real life. 100% since 01/10 would be pretty darn nice.
And, I take your point -- believe me, we've thought about it -- but when you have numbers that validate consistently, that you've error checked, that you've had professionals look over, and that you can replicate across different periods of time, well, what are you supposed to think except that you have something that might work? Certainly I think it's worth the attempt. Of course it could all turn out to be useless and fall apart -- that's the nature of the business.
Separate names with a comma.