LLMs Show High Accuracy in Financial Sentiment, But Fail to Predict Stock Returns

A new study benchmarks several large language models (LLMs) for their effectiveness in financial sentiment classification and return predictability. Researchers found that while models like Mistral-7B and QLoRA-adapted Qwen2.5-7B achieved high accuracy in classification tasks, their ability to predict stock market returns was minimal and not statistically significant after rigorous correction. The findings highlight a notable gap between linguistic performance and practical economic utility in financial applications. AI

IMPACT Highlights the limitations of current LLMs in translating classification accuracy to real-world financial prediction, suggesting further research is needed for economic utility.

RANK_REASON Academic paper detailing LLM performance on financial tasks. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• Benzinga
• Financial-RoBERTa
• FinBERT
• LLaMA3-8B
• Mistral-7B
• QLoRA
• Qwen2.5-7B
• TF--IDF Naive Bayes

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →