Adding ONNX Deep Learning Filters to Three Trading Strategies

Summary

The article explores adding Python-trained deep learning models to three existing Expert Advisors based on causal network analysis, stochastic model optimal control, and Nash game theory. It describes using historical EURUSD prices to train a convolutional and recurrent neural network, evaluating prediction error, exporting the model to ONNX, and using its output as an additional condition for trade decisions in MQL5. The intended comparison is each strategy with and without the model.

Reported backtests give mixed outcomes: the deep learning addition is described as less profitable but more conservative for the causal strategy, more favorable for the stochastic control strategy, and weaker overall for the Nash strategy. The author cautions that these results depend on particular backtest periods and market conditions and calls for broader testing. The document gives limited methodological detail for assessing whether the comparisons avoid leakage or support generalization; its results therefore do not establish that deep learning improves trading performance reliably.

Key ideas

• The proposed workflow trains a sequence model in Python and exports it to ONNX for use in MQL5.
• The model acts as a filter on trade decisions in three strategies with different foundations.
• Reported performance effects vary by strategy, with no uniform benefit from adding deep learning.
• The comparison is limited to particular backtest samples and market conditions.
• Broader testing is needed before inferring that the approach generalizes.

Tags

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