“Markets and LLMs together beat either one alone for predicting earnings call mentions”
This paper studies how to design input context for LLMs to forecast mention market outcomes, where contracts resolve based on whether a company mentions a specified keyword during its earnings call. The authors introduce Market-Conditioned Prompting (MCP), which treats the market price as a Bayesian prior and instructs the LLM to update it using textual evidence from news and prior transcripts. A mixture of the market price and MCP (MixMCP) outperforms the market baseline alone, suggesting LLMs can serve as complementary forecasters alongside live market prices.
Extensive technical background assumed