Selecting Multi-Strategy Groups with Forward-Period Testing
Summary
This article examines whether groups of trading-strategy parameter sets selected on historical data continue to perform in a later forward period. It treats the main optimization interval as in-sample and the reserved interval as out-of-sample, then compares manually chosen groups with groups selected by profit and by clustering. Initial forward results weaken or turn negative, with drawdowns rising, showing that strong in-sample selection alone does not ensure comparable later performance.
The author then runs optimization with a forward period and compares both segments using normalized profit, defined by scaling profit to a target drawdown. Because forward-period drawdown can differ from in-sample drawdown, the article adjusts the forward result to use the in-sample position-scaling factor. Some groups show positive forward results and drawdown closer to the earlier period, but the author raises concerns about selecting groups using forward data and the short comparison window. The evidence is exploratory and period-specific; the text does not establish that the selection procedure will remain effective in later periods.
Key ideas
- A tester forward period provides a separate performance segment that can expose deterioration after optimization.
- Profit-selected groups performed worse in the reported forward comparisons than their main-period results suggested.
- Normalized profit scales results to a chosen drawdown target, and comparisons require consistent position scaling.
- Selecting strategies using forward outcomes risks using evaluation data to guide selection.
- Short forward samples and strategy drawdown fluctuations limit confidence in the reported comparisons.
Tags
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.