We show that the dispersion of “sentiment” in conference calls contains value-relevant information, incremental to the level. Using question-and-answer (Q&A) sessions of earnings conference calls, we measure both the level and dispersion of textual tone with FinBERT, and information quantities with a computational topic modeling approach aided by human interpretation. We find that tone dispersion is significantly related to the information quantities embedded in earnings calls. More tone-dispersed calls are associated with a larger sensitivity of stock price to earnings news. Moreover, it is the topic-related components of tone dispersion that drive such price sensitivity, and analyst tone dispersion playing a larger role than that of executives.

zlynne@ceibs.edu