Momentum, Commentary Tone, and Earnings Surprise in LLM Financial Forecasts
Updated
Updated · arxiv.org · Oct 15
Momentum, Commentary Tone, and Earnings Surprise in LLM Financial Forecasts
1 articles · Updated · arxiv.org · Oct 15
Summary
A controlled study tested how large language models (LLMs) weigh momentum, commentary tone, and earnings surprise in financial forecasts.
Six open-weight LLM checkpoints were evaluated with 2,016 synthetic earnings prompts, revealing that tone and momentum often influenced predictions more than earnings surprise.
Results highlight that LLMs' forecast sensitivity varies by model and input, but do not confirm internal reasoning or real-world predictive power.