Scenario disclosure and market expectations: insights from earnings conferences via large language models

A scenario described vividly and in detail is more easily imagined, leading individuals to overestimate the likelihood of its occurrence. We leverage a novel large-language-model (LLM) framework to analyze management speech style during unstructured earnings conferences and construct quantitative measures of scenario oral disclosure. Such disclosure triggers scenario thinking, inflating investors’ beliefs about future firm prospects, particularly when firm-specific information is scarce. Scenario disclosure is more pronounced when conveying positive information, with poor relative performance and under negative media sentiment, suggesting that management employs scenario framing to manage expectations. Finally, management scenario disclosure significantly increases stock price crash risk. This represents LLMs’ first application to identify and quantify scenario disclosure in earnings conferences, providing guidance for regulation and institutional design.