Parameter shift oddness...

Strategies might be attractive on the surface but fail moving forward. Why? Here we discuss the qualities of effective trading strategies, and the many traps one should avoid.

Parameter shift oddness...

Postby Jaded » Sat Nov 03, 2007 7:08 pm

OK.. So I am having weird parameter shifting issues.

In the oneclick setup for the strategy:

($F_1SAvgAnnReturn / $F_AvgAnnReturn) * 100
Best: 100
Worst: 75 (Reject checked)


In the detailed analysis of systems it is producing:

PortSize: 5
AvgAnnReturn: 166.37%
1SAvgAnnReturn: 9.61%


Shouldn't the 1SAvgAnnReturn be at least 75% of the AvgAnnReturn?

I wonder if I am overlooking something or do not understand how it is calculated, but from what I have setup, the 1SAvgAnnReturn should be 75% of what AvgAnnReturn is.

Let me know please,

Posts: 151
Joined: Mon May 14, 2007 6:49 pm

Postby Overload » Sat Nov 03, 2007 11:56 pm

Your OneClick Setup is used to filter individual strategies, while the Detailed Analysis is for the combined multi-system. Once the individual strategies are combined together, you can end up with a very different portfolio-selection. And that can lead to situation you've seen (i.e. where the individual strategies pass a filter but the combined multi-system does not).

If you already had strategies in place before you added the parameter shift filter, then that would be a simple explanation. If you wish to start the search over from scratch so that all strategies must pass that new filter, then you'll need to click the Reset Results button on the General tab of the OneClick Setup and restart the search.

If that parameter shift filter isn't new, and each of the strategies has actually passed it, then it may be a sign that the portfolio-selection isn't indicative of the overall trades. If that's the case, you may need to perform some additional tests, such as Monte Carlo, to ensure that the selected strategies will back test just as well being part of a multi-system as they do when back tested on their own.

Hopefully that makes sense, but let me know if you have additional questions on this.

Posts: 2246
Joined: Wed Nov 30, 2005 12:14 pm

Return to Curve-Fitting and Other Pitfalls