Price Discovery Within Earnings Calls

Abstract

This paper studies how earnings-call information is incorporated into prices in real time by separating "what is said" (content) from "how it is said" (delivery). Using 3,415 earnings call recordings, I extract 1.33 million time-stamped audio segments, defined as contiguous speech between detected pauses, and link each segment to subsequent NBBO quote-based returns. Negative textual sentiment predicts quote-based returns within the first minute, and the effect persists for up to two minutes, consistent with rapid incorporation of content. Reactions to text are faster and larger for stocks with higher turnover, greater analyst coverage, and larger earnings surprises. In contrast, delivery contributes mainly through cross-modal alignment: conditional on text, negative tone is not statistically distinguishable from zero, whereas conditional on negative tone, negative text predicts an additional downward reaction in the 90-120 second window. Thus, content is the primary driver, while delivery contributes later through this alignment channel; in Q&A, the alignment effect is most immediate on the ask side. Finally, poorer audio quality attenuates the fast text response and weakens the delayed text-voice alignment effect. The results provide segment-level evidence on price discovery during earnings calls and document distinct incorporation dynamics for content and delivery.

Keywords: Audio Analysis, Price Discovery, Earnings Call