OmniVest Forum OmniVest Forum
forums calendars search
today this week
 
register logon control panel Forum Rules
You are currently browsing as a guest.
You should logon to access more features
A Self-Moderated Community - ALL MEMBERS, PLEASE READ!
Vote for Members who contribute the most to your trading, and help us moderate content within the Forums.


[Random Quote] -


Only Forum Moderators, Administrators, and the owner of this message may delete it.

 
Steve Mayo

Legend
100100100100
Posts: 414

Joined: 10/11/2012
Location: Austin, TX

User Profile
 
Subject : RE: The Learning Curve
Posted : 5/7/2014 10:55 AM
Post #30414 - In reply to #30409

Last year, I did a big chunk of work on trying to use correlation as a basis for port optimization, with the idea being it might help to "fill in the holes" in the %invested. This was under V1 when all we had were mostly the RTM strats. Mark took my spreadsheet and created a bunch of macros and put that concept through its paces, but found that just "solving" for ending equity was generally more effective than trying to find 'uncorrelated' strats to add to the mix. I think the reason is because the RTMs are all highly correlated to begin with.

With Condition manager, a wider range of trading strategies, and the ability now to use alternative asset classes (gold, real estate, bond, commodity, foreign ETFs, etc.) via the custom lists, I think the concept is worth revisiting. Hopefully, we can get that functionality (incrementally adding the next strat based on non-correlation) into Port Builder.

Jim, your comment about a post-build "rejiggering" of the strat mix after an iterative successive build is a good one -- the problem is that it's difficult, with any programmatic accuracy, to know WHICH strat to switch out. And with high-TTM systems like the RTMs, every little change can make a big difference. Even when the strat you are swapping out doesn't seem to be making a bunch of trades, what you don't know is what will happen when the other strats step up to fill that newly available daily equity.

I think the better approach might be to come up with better tools and metrics for evaluating the robustness of our ports -- things like Monte Carlo and non-parametric statistics that are, unfortunately, hard to calculate and even harder to interpret. Yes, we can rejigger to get a better equity curve on a simulation, but are we in-effect just"optimizing the simulation" to select out/in certain trades that are not representative of what the system will do in real-world trading? It's that big standard deviation/non-Gaussian distribution (bigger tails?) problem again! Is it better to have a system that can generate 150% in a year when the planets all align, or one that has a much lower potential but is more predictive? I think we'll need to look less at equity curves and start looking more at distribution curves, albeit not as much fun as seeing your simulated account grow to $1billion in 14 years! :-)
Deleting message 30414 : RE: The Learning Curve


Nirvana Systems
For any problems or issues please contact our Webmaster at webmaster@nirvsys.com.