Discussing the article: "LLM-Based Trading Agent with Embedded Top Trader Philosophy"

New comment
 

Check out the new article: LLM-Based Trading Agent with Embedded Top Trader Philosophy.

The article provides a critical analysis of an LLM strategy in which forecasting the direction is separated from trading decisions, and demonstrates why this leads to a disconnect between metrics and PnL. We will describe procedures for dataset balancing, feature engineering, prompt and response preparation, fine-tuning configuration in Ollama, and reliable parsing. Backtesting and forward testing reveal systematic degradation. The practical conclusion is that the problem must be formulated as a direct optimization of trading outcomes.

The trader experiences classic cognitive dissonance: the model guesses correctly more than 50% of the time, the backtest looks great, but in reality the money either does not come in or is lost. Why this happens, and how to properly reformulate the problem so that the model focuses not on “guessing the direction” but on the actual profitability of the trade—that is the main topic of this article.

In its first version, the system was developed in accordance with all methodological requirements: strict data separation without look-ahead bias, the use of modern LLM architectures, the incorporation of the philosophy of elite traders, and honest forward testing on unseen data. Nevertheless, even with these improvements, when moving from theoretical metrics to actual PnL, the system demonstrates a systematic gap between predictive power and trading returns.

The second version is an attempt to partially bridge this gap through more rigorous dataset quality control, class balancing, and some improvements in prompting and parsing. But even after these revisions, the main problem remains—and this article is devoted to diagnosing precisely that problem. As a result, we are building a promising system that is already showing good potential in backtesting, and it will undergo thorough testing on real-world data.


Author: Yevgeniy Koshtenko

New comment