Words That Matter

This chapter tests whether language in earnings-call prepared remarks contains tradable signals beyond standard disclosures. Using the first three 2023 calls from 93 NYSE firms (market cap ≥ USD 50B), we link textual features to announcement-day (day 0) and five-day cumulative returns, aligning pre/post-market timings. Call length is uninformative. Traditional lexicon sentiment is weak, but multi-emotion markers show “joy” predicts positive returns while “anticipation” predicts negative returns, with limited “fear” effects at longer horizons. We introduce a text-to-profit mapping that identifies broadly used, frequently occurring words that systematically forecast post-call drift. Terms tied to weight-loss drugs, post-pandemic travel, and “workload” foreshadow gains; autos, global logistics, and “commerce” presage losses. Many effects emerge after day 0, enabling short-horizon strategies. An out-of-sample rule that selects predictive words from first-quarter calls and trades around subsequent calls averages +0.51 percentage points over five days (buy side +0.89; sell side +0.11). Overall, the evidence suggests incomplete same-day assimilation of linguistic cues and highlights the value of carefully engineered NLP features for event-driven trading and investor-reaction research.