This repository demonstrates how neural networks can be used inside MetaTrader 5 to enhance a traditional correlation based statistical arbitrage strategy. The project includes two expert advisors:
- MLP_Stat_Arb.mq5 – a multi‑layer perceptron (MLP) approach that scans every major currency pair combination and scores potential trades.
- EA_Correlation.mq5 – a simple Pearson/Spearman correlation strategy that operates on a fixed symbol pair.
The MLP EA shows how to embed and parameterise a neural network directly in MQL5. The weights included in experts/MLP_Stat_Arb.mq5 are placeholders only. The performance graphs in images/ were generated using a fully trained network that was forward and genetically optimised to reduce overfitting.
experts/ MQL5 source files for both strategies
images/ Back‑test results
The MLP approach trades all major currency crosses and therefore captures far more opportunities than the benchmark EA, which only trades a single predefined pair. In historical tests this resulted in larger net profits and more robust behaviour across different market regimes.
Open the .mq5 files in MetaEditor and press Compile. The MLP EA requires no additional DLLs – the neural network is implemented entirely in MQL5.
Planned features include:
- fixed exposure and spread leg risk management
- beta‑neutral portfolio constraints
- support for additional asset classes