Discussing the article: "Neural Networks in Trading: A Multi-Agent System with Conceptual Reinforcement (Final Part)"

[MetaQuotes]

Check out the new article: Neural Networks in Trading: A Multi-Agent System with Conceptual Reinforcement (Final Part).

We continue to implement the approaches proposed by the authors of the FinCon framework. FinCon is a multi-agent system based on Large Language Models (LLMs). Today, we will implement the necessary modules and conduct comprehensive testing of the model on real historical data.

It should be noted that the implementation presented here differs significantly from the original one, which naturally affects the results. Therefore, we can speak only about evaluating the efficiency of the implemented approaches, not about reproducing the original results.

For model training, we used H1 EURUSD data from 2024. The parameters of the analyzed indicators were left unchanged to focus exclusively on evaluating the algorithmic performance.

The training dataset was formed from multiple runs of several models with randomly initialized parameters. In addition, we included successful runs derived from available market signal data using the Real-ORL method. This enriched the dataset with positive examples and expanded the coverage of possible market scenarios.

During training, we used an algorithm that generates "near-perfect" target actions for the Agent. This enables model training without the need for continuous dataset updates. However, we recommend periodic data updates, which can further improve learning outcomes by expanding state-space coverage.

The final testing was conducted using available data from January 2025, with all other parameters unchanged. The results are presented below.

Author: Dmitriy Gizlyk