Registered: Sep 2011
09-11-12 09:50 PM
Quote from JasonEdinburgh:
Hi Wide Tailz,
You have some pretty excellent looking equity curves in this thread and I wondered if they turned out to be generally successful, or if they were the result of over-optimisation.
Thanks for sharing your thoughts and I'm sure if you post a few more times, your PM privileges will kick in. I have a different background from most system traders (mechanical / aerospace engineering) and it may be why I'm doing things they normally wouldn't. If you are afraid of curve fitting, my optimization efforts were probably a bit over the top for what you may consider valid.
The process I used was (from what little info I found) fringe industry standard: make the program, optimize it, test it out of sample, then dump it all into a walk forward tester. Finally, let it run forward for enough time to confirm that the equity curve characteristics are still there in the future, and keep re-optimizing it on a set schedule. Note what is missing: testing one program on several different stocks / futures / markets. I don't believe in this. I think it is an invalid assumption to believe that if a system works on several markets then it must also work the future. Working in the future is what is important, and has nothing to do with working on several unrelated instruments.
My 'over optimized' algos mostly passed this process, but some of the more filtered stock buyers are still in forward test because they don't trade as often as the SPY system.
In my opinion, the only rule that matters is that the algo must make money in the future in the actual market. You can try to go with the simplest thing that works if it helps you believe that it won't break down in the future...... or you can go with something that you have proven to work in the future, which was made to trade the particular instrument for which it is applied. And yes there are professional fund managers using trend following algos of this type.....
The way I see it, why use a pair of channel lock pliers when you could get much better results with a torque wrench and set of sockets?