Using Kohonen Maps to Select Robust Trading Parameters and Forecast Prices

Summary

The article applies self-organizing Kohonen maps to two tasks: identifying robust expert-advisor settings from optimization results and forecasting time series. For parameter selection, it argues that traders should seek broad regions, or plateaus, of acceptable objective values rather than isolated peaks. It discusses how genetic optimization may undersample poor settings and nearby loss regions, and suggests fuller or staged searches, repeated runs with varied criteria, and feature masks that separate EA inputs from performance statistics when clustering.

For forecasting, the article describes training maps on selected market series and gives an example of predicting silver using a group of currency and metals data. It reports that accuracy sometimes reaches 60 percent in a particular test period, while emphasizing the need to choose inputs and history carefully and tune the model extensively. The result is an exploratory demonstration, not broad proof of forecasting reliability. The author also cautions that training inside an indicator can block it, recommending scheduled map generation outside the indicator.

Key ideas

• Robust parameter selection should favor wide regions of acceptable performance over a single best backtest result.
• Genetic optimization can leave nearby failure regions underrepresented, obscuring parameter instability.
• Repeated optimization runs and feature masking can help assess relationships between EA parameters and performance measures.
• Kohonen maps can be trained on selected market series to forecast another series, but input and history selection require care.
• The reported forecasting accuracy is occasional and tied to a specific test example, so it does not establish general predictive reliability.

Tags

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