Forecasting Moving Average Direction

Article MQL5 articles

Summary

The article compares machine-learning classification of future close-price direction with classification of a 60-period simple moving average's future direction. Using exported MetaTrader 5 data, it builds scaled OHLC and moving-average inputs, labels whether price or the indicator rises over the forecast horizon, and discusses principal-component transformation and time-series validation with a gap. The author frames indicator forecasting as potentially easier because the indicator is computed from observable price history, whereas many drivers of future price are unknown.

A neural-network example reports 49.9% accuracy for close direction and 68.8% for moving-average direction. The text also describes comparing several classifier families, but the supplied excerpt omits much of the evaluation and explanation. These accuracy figures are tied to one instrument, data period, feature set, and evaluation approach; they do not establish profitability or robust predictive power. The article acknowledges that moving-average direction can lag or diverge from price and calls for further research across indicators.

Key ideas

The study labels future price direction and future moving-average direction as separate classification targets.
A neural-network example reports higher accuracy for moving-average direction than for close-price direction on the described data.
The workflow uses historical terminal data, scaled OHLC and moving-average features, PCA, and time-ordered evaluation.
The author argues that indicator targets have a more defined relationship to observable inputs than future prices do.
Higher indicator accuracy does not guarantee profitable signals, and moving averages may lag or disagree with price.

Forecasting Moving Average Direction Instead of Price with Machine Learning

Summary

Key ideas

Tags