- ago
Silly me. I just assumed WL7 would support assigning specific strategy parameter values to individual stocks. What is known in WL6 jargon as Preferred Values or PVs.

All my WL6 strategies employ specific strategy parameter assignments to each individual stock because different stocks behave differently and need their own parameter assignments to be optimal (e.g. volatility, confidence interval). And the parameter optimizer can set these strategy parameters uniquely for each individual stock.

In addition, when re-optimizing strategy parameters, the re-assignments should only be performed if they are an improvement over the existing parameter settings. For example, if the optimizer is trying to maximize the ScoreCard metric ProfitPerBar for individual stocks, and the new parameters setting yield a poorer ProfitPerBar than the previous parameter settings, then those settings should not be updated. This way, all subsequent optimizations can only improve the ScoreCard metric over the existing parameter values. This feature will be new for WL7; WL6 doesn't have this.

Numerical analysis note: Some parameter optimizers (like Particle Swarm) take a random guess to start their optimization. This can lead to a new local maxima, which can be a good thing. But to take advantage of this, the parameters must be selectively updated as discussed in the paragraph above.
10
261
4 Answers

Reply

Bookmark

Sort
- ago
#1
We'll likely need to add a new mode, and a new "Symbol by Symbol" tab in the Optimization control to do this. Unlike WL6, WL7 doesn't run the Strategy on one symbol at a time and then combine the results, it runs the Strategy on each symbol in parallel, bar by bar.

So, the Optimizer now in WL7 isn't capable of distinguising the best parameter values for each symbol. So we'll need a new mode and a new tab.
0
- ago
#2
Wow this request has risen quickly! Good thing I've been mentally preparing myself for the challenge!
2
- ago
#3
Nice, can't wait for this feature! :)
0
- ago
#4
Superticker,
QUOTE:
when re-optimizing strategy parameters, the re-assignments should only be performed if they are an improvement over the existing parameter settings.
This implies that the strategy has not been changed in any way. I suppose that the existing parameters could be retested to prove that they remain better than those newly found.
1

Reply

Bookmark

Sort