Topic: Best Method for Improving Generated Strategies

Some collections will have only a couple acceptable strategies. I want to work with others to improve them.

I am not sure as to how to do that and at the same time retain the integrity of the collection.

I wonder oif people have ideas as to what is best practise for that.

Thanks for comments.

My 'secret' goal is to push EA Studio until I can net 3000 pips per day....

Re: Best Method for Improving Generated Strategies


You can try to optimize your collections with the new Validator tool.
It will collect only the strategies passing the Acceptance Criteria, which will give you only the best strategies on the newest data.

You can actually drag / drop all collections and to test them on newer data from different markets. You will receive updated collection for these markets very easy.

3 (edited by qattack 2017-08-13 22:16:44)

Re: Best Method for Improving Generated Strategies

Dave, how much data and what time frame are you generating over? And how do you define an "acceptable" strategy?

Are you using OOS to visually validate the results?

I'm just starting to experiment with EA Studio. I am running eight instances on my rented server over a 14 year period on M30. 50% OOS. I am searching by SQN.

Acceptance Criteria:
Max Amb. Bars: 10
Min Net Profit: 10
Min trades: 600
Min SQN: 1.6
Min win/loss: 0.3
Max Balance Deviation: 15
Max Stagnation %: 15
Max Equity DD%: 12 (too high)

Some of these values will be radically adjusted when I better figure out what I'm doing. I am particularly interested in toying around with "Max Balance Deviation".

I include "Optimization" (defaults), "All data validation", and "Monte Carlo validation" (100 count; 85% validated tests).

Every so often, I sort strategies by Max DD% and go each manually, give the equity graph a visual check, and remove any offensive-looking strategies. I pay particular attention to the OOS period, giving a bit more weight to the most recent period. If it has stagnated for a longer period at the end, I will remove it. And of course if the OOS is bad in general, I remove it.

Many beautiful-looking strategies (as well as some on the borderline) remain after only one day's execution.

I then need to run the OOS periods through the Validation tool. Then, I will run the strategies through the optimizer, but I don't know whether I want to optimize over the entire data sample, only the OOS period, or perhaps even only a portion of the most recent OOS.

4 (edited by qattack 2017-08-14 00:14:16)

Re: Best Method for Improving Generated Strategies

Using the Validator on the OOS period is surprisingly easy. I just dropped multiple collections, changed a couple settings, and hit Start.

For my purposes, I had to make two runs. The first, I deleted Acceptance Criteria, Optimized, and ran MC validation. The second, I restored Acceptance Criteria and ran the optimized strategies to check vs. Acceptance Criteria. (I had to restart when I figured out that since I had 140 strategies, the bottom 40 would be pruned...I didn't want that to happen during my first run.)

Most of the OOS periods were being filtered by the Acceptance Criteria prior to optimization before doing this. Perhaps it's instead best to simply lower the Acceptance Criteria to a degree for the first run instead of deleting it. This would improve the quality of the remaining strategies and some balance could be found.