Testing Rounded Price-Shape Templates with Perceptrons and Neural Networks

Summary

The article explores using “fan” templates to turn recent candlestick closing prices into inputs for perceptrons and neural networks. Each input is a rounded point distance between the close of the oldest candle in a window and closes at selected positions within that window. The experiments vary the template length and the number of sampled distances, comparing a 24-candle window with a 48-candle window and four-value with eight-value inputs.

The author describes optimizing EURUSD expert advisors in MetaTrader 5, then combining multiple top optimization results in a one-year forward test. The setup uses several stop-loss and take-profit configurations and compares perceptron models with neural networks. The article frames these as experiments into input representation and model behavior; it does not establish that one template or architecture is reliably superior. Results are also limited by the chosen pair, dates, platform, optimization procedure, and the author’s use of repeated genetic optimization runs.

Key ideas

• Fan templates encode rounded price differences between selected candle closes as model inputs.
• Input count and historical window length are varied to study how template design affects learning.
• The experiments compare perceptrons with neural networks trained and optimized in MetaTrader 5.
• Forward testing combines multiple leading optimization results, so it evaluates a group of optimized systems.
• The article presents a specific EURUSD experiment rather than evidence of general trading performance.

Tags

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