Discussing the article: "Development and Forward Testing of an Autonomous LLM Agent for Trading with SEAL"

[MetaQuotes]

Check out the new article: Development and Forward Testing of an Autonomous LLM Agent for Trading with SEAL.

A hybrid architecture based on Llama 3.2 and SEAL is being tested on eight currency pairs (M15), with forward-period data isolation and information leakage control. The methodology combines adversarial self-play, curriculum learning, and class balancing to ensure stable training. The experiments confirm the gap between forecast accuracy and actual returns, providing readers with practical guidelines for testing strategies and accurately assessing their generalizability.

The aricle describes the development of a hybrid trading system that integrates a fine-tuned Llama 3.2 language model with the Self-Evolving Adversarial Learning architecture. The system is built on the MetaTrader 5 platform and demonstrates the ability to adapt continuously through adversarial training and curriculum learning. A critical component of the architecture is strict data separation, including a dedicated forward-test period, which prevents information leakage and ensures an objective assessment of the model's performance.

The system architecture consists of four main modules. The first module generates a balanced dataset from historical MetaTrader 5 data with automatic labeling of the price direction 24 hours later. The second module fine-tunes the base language model on the generated dataset using the Ollama library. The third module is a SEAL agent that implements adversarial self-play and evolutionary optimization to improve the robustness of trading decisions. The fourth module validates the system on strictly isolated forward-period data, without any possibility of information leakage from the test set.

The experimental validation was conducted on eight major currency pairs — EURUSD, GBPUSD, USDCHF, USDCAD, AUDUSD, NZDUSD, EURGBP, and AUDCHF — using the M15 timeframe. The training dataset covered a 30-day period, excluding the last 7 days for forward testing. The forecast horizon was 96 bars (24 hours), which corresponds to the typical time scale of position trading strategies on intraday timeframes. The base balance was set at $140, with an 8% risk per trade and a minimum model confidence threshold of 60% for opening a position.

Author: Yevgeniy Koshtenko