Automating the Selection and Grouping of Trading Strategy Instances

Summary

This article explores automating the selection of optimized strategy instances for a multi-currency Expert Advisor. It first prepares optimization data, removes loss-making results, and ranks candidates with a weighted score built from scaled performance measures such as profit, recovery, Sharpe ratio, drawdown, and trade count. The author then manually constructs a baseline group, favoring instances with different parameters and drawdown timing so their risks may diversify when combined.

The proposed automation searches for groups and compares their normalized performance with that baseline. The reported example says a manually selected group of 16 instances had a smaller drawdown when position sizes were reduced across the group, and later automated selection produced profitability comparable to or better than manual selection. These are in-sample optimization results, not proof of durable performance. The article acknowledges that broader searches take time and leaves out-of-sample evaluation for future work; optimization settings, selection bias, and the quality of the test period limit what can be concluded.

Key ideas

• The workflow begins with exported optimization results, adds missing inputs, and filters out losing candidates.
• Candidate instances are ranked using a weighted sum of scaled performance metrics, with lower drawdown contributing positively.
• Groups are assembled to diversify drawdown timing rather than simply combining the individually highest-ranked instances.
• Position sizes and a scaling factor normalize a group to a chosen drawdown target.
• Automated selection showed promising profitability in the reported tests, but out-of-sample validation remained future work.

Tags

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.