Engineering Higher-Dimensional Features to Anticipate Moving Average Crossovers
Summary
The article develops a short-period moving-average crossover idea that applies averages separately to open, high, low, and close prices, then explores whether engineered features can help identify crossover opportunities earlier. Its MQL5 data pipeline exports price and moving-average levels, historical changes, and relative differences among price channels into a 40-column dataset. The stated forecasting horizon is five bars, and the discussion describes training an ONNX model to classify future crossover behavior.
The central hypothesis is that expanding a feature space can make some classification problems easier, in contrast to dimensionality-reduction methods discussed elsewhere in the series. The article describes a demonstration notebook and a EURUSD daily model, but the supplied text gives no quantified out-of-sample results, trading costs, or robustness analysis. More features and earlier signals do not guarantee better generalization; any apparent lag reduction would need validation against leakage, noise, and realistic execution assumptions.
Key ideas
- The method applies same-period moving averages to multiple OHLC price series to construct crossover features.
- Historical changes and relative differences among price and moving-average channels expand the input dataset.
- A model is trained to classify whether a crossover will occur within a specified forecast horizon.
- Increasing feature dimensionality may improve separability, but the article does not provide quantified validation results.
- Out-of-sample testing and checks for leakage are needed before treating earlier predictions as tradable signals.
Tags
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.