Automating Group Selection for Multi-Currency Strategy Optimization

Summary

This article describes automating the second stage of a multi-currency Expert Advisor optimization workflow: selecting groups of individually optimized strategy instances whose combined behavior may improve drawdown or balance-curve smoothness. Instead of reading a large results database on every test agent, the process creates a compact task database containing only the initialization strings needed for a particular symbol and timeframe. The task database is distributed to agents, while collected results can be written back to the main database.

The article explains changes to the database connection class, generating optimization tasks, and running group selection as a genetic optimization. It reports that combining selected strategy instances can be automated and that the workflow can assemble initialization strings for groups across symbols and timeframes. The implementation remains a draft: experiments covered a relatively short period, and clustering was not automated. Its practical value therefore lies mainly in the optimization workflow; the document does not establish that the selected groups will remain effective out of sample or in live trading.

Key ideas

• The second optimization stage searches for groups of individually optimized strategy instances that work well together.
• A task-specific database limits each test agent to data for the relevant symbol and timeframe.
• Compact task databases reduce the storage and transfer burden compared with distributing the full results database.
• Genetic optimization can automate selection of strategy-instance combinations for group testing.
• The reported work is preliminary and uses a relatively short test interval.

Tags

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