Walk-Forward Evaluation of Optimized EAs and Oscillator Signals
Summary
The article warns against treating the best historical optimization result as a forecast of future performance. It proposes repeatedly optimizing over a rolling historical window, then testing the selected settings on the next, separate period. Recording each run’s net profit, gains, losses, profitability, drawdown, and trade count provides a way to judge how optimization choices behave across successive out-of-sample periods. The author presents this as a more efficient historical check than waiting for prolonged demo trading, while noting that results still depend on the available history and tested selection approach.
The second topic is an oscillator system with entries when an indicator crosses into overbought or oversold territory, plus signals when it exits those zones into a neutral area. The implementation combines analogous entry cases around configurable levels. The article cautions that oscillator signals can produce false entries against the prevailing trend, suggesting use in flat conditions or for entries aligned with trend. It supplies no performance results establishing the system’s profitability.
Key ideas
- A strong in-sample optimization result alone does not show that an EA will perform well later.
- Rolling optimization windows followed by separate test windows provide repeated out-of-sample checks.
- Tracking profit, loss, drawdown, and trade counts helps compare optimization approaches across test periods.
- The described oscillator system uses crossings into and out of overbought and oversold zones.
- Oscillator signals can fail against the prevailing trend, and the article provides no proof of profitability.
Tags
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.