Skip to content
All library documents

Automating Strategy Group Selection with Forward-Period Results

Article MQL5 articles
Read the original

Summary

This article describes extending a multi-currency Expert Advisor optimization workflow so strategy instances can be grouped using their behavior in a forward period. It contrasts selecting the strongest optimization performers with selecting instances whose results remain similar across optimization and held-out intervals. Earlier comparisons found no clear advantage for either method over a short history, while the similarity-based approach produced better group results on the selected interval than simply choosing the top optimization performers. The author suggests combining groups selected by both methods.

The proposed workflow runs optimization with a designated forward period so that the same parameter combinations are evaluated in both periods. It stores pass results in a database, calculates comparative measures and group scores, ranks candidate groups, and selects the top group across symbol and timeframe combinations. The article also documents database fixes needed to save optimization results correctly and describes preliminary automated testing on one symbol and timeframe. Results are limited by the short comparison period and small initial test setup; the excerpt does not provide a complete performance evaluation of the automated selection method.

Key ideas

  • Forward-period evaluation lets the same parameter combinations be compared across optimization and held-out periods.
  • The article compares selecting top optimization performers with selecting instances that behave consistently across periods.
  • Earlier results did not establish a clear winner, and the comparison covered too little history for strong conclusions.
  • The proposed automation scores and ranks strategy groups using database queries or a Python analysis step.
  • Database connection and task-record handling must be correct for automated optimization results to be stored reliably.

Tags

Multi-assetBacktestingPortfolio constructionStatistics

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