Forecasting Moving Average Direction and Price Divergence with Machine Learning

Summary

The article describes feature engineering for longer-range market forecasts by predicting the direction of a moving average rather than price directly, while also modeling when price and the average diverge. It compares logistic regression models using time-series cross-validation on M1 data across a large set of available symbols. The reported averages favor moving-average direction over price direction, and the divergence classifier also performs above chance. The author suggests that moving averages may be easier to predict because they aggregate past prices.

The article further discusses tailoring a moving-average feature for a selected market and evaluating a strategy on unseen data. Its conclusion distinguishes shorter forecast horizons, where direct price prediction may be preferable, from longer ones, where moving-average changes may help. These are reported classification accuracies, not proof of a tradable edge. Results are limited to the described data, timeframe, features, and model setup; the excerpt does not provide enough detail to independently assess trading costs, robustness, or the unseen-data strategy performance.

Key ideas

• The study compares forecasts of future price direction with forecasts of moving-average direction.
• It uses logistic regression and time-series cross-validation on M1 data across multiple symbols.
• The reported mean accuracy is higher for moving-average direction and divergence than for price direction.
• The author proposes that aggregation in moving averages may make them more predictable than raw price.
• Classification accuracy alone does not establish profitability after costs or demonstrate robustness beyond the tested setup.

Tags

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