Skip to content
All library documents

Separating Factual and Subjective Reasoning in LLM Crypto Trading

Article arXiv papers · Author: Qian Wang et al.
Read the original

Summary

The document examines why more capable large language models may sometimes trade cryptocurrency less effectively than weaker models. It attributes part of this pattern to stronger models favoring factual information over subjective interpretation. The proposed FS-ReasoningAgent framework separates reasoning into factual and subjective components so that trading decisions can draw on both types of input.

The reported experiments show profit improvements for BTC, ETH, and SOL, with stated gains of 7%, 2%, and 10%, respectively. An ablation study reports that subjective news performed better in bull markets, while factual information performed better in bear markets. These results suggest that the relative value of the two reasoning modes may depend on market conditions. The document does not specify the full experimental setup, evaluation period, trading costs, or risk-adjusted performance, limiting conclusions about reproducibility and live trading applicability.

Key ideas

  • The study reports that stronger LLMs can underperform weaker ones in cryptocurrency trading.
  • It links this pattern to stronger models' preference for factual information over subjective interpretation.
  • FS-ReasoningAgent separates factual and subjective reasoning within a multi-agent approach.
  • Reported results show gains across BTC, ETH, and SOL, though the experimental details are not provided here.
  • Subjective news reportedly worked better in bull markets, while factual inputs worked better in bear markets.

Tags

CryptoMachine learningSentimentBacktesting
Full text
# Exploring LLM Cryptocurrency Trading Through Fact-Subjectivity Aware Reasoning


# Exploring LLM Cryptocurrency Trading Through Fact-Subjectivity Aware Reasoning









While many studies show that more advanced LLMs excel in tasks such as mathematics and coding, we observe that in cryptocurrency trading, stronger LLMs sometimes underperform compared to weaker ones. To investigate this counterintuitive phenomenon, we examine how LLMs reason when making trading decisions. Our findings reveal that (1) stronger LLMs show a preference for factual information over subjectivity; (2) separating the reasoning process into factual and subjective components leads to higher profits. Building on these insights, we propose a multi-agent framework, FS-ReasoningAgent, which enables LLMs to recognize and learn from both factual and subjective reasoning. Extensive experiments demonstrate that this fine-grained reasoning approach enhances LLM trading performance in cryptocurrency markets, yielding profit improvements of 7\% in BTC, 2\% in ETH, and 10\% in SOL. Additionally, an ablation study reveals that relying on subjective news generates higher returns in bull markets, while focusing on factual information yields better results in bear markets. Code is available at https://github.com/Persdre/FS-ReasoningAgent.

Shown in full with attribution under the source's licence. Licence: abstract CC0

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