Earnings Call Sentiment

Methodology

Data. The universe is the S&P 500. For each company, Alpha Vantage's EARNINGS endpoint identifies which fiscal-quarter label its most recent reported quarter carries in Alpha Vantage's numbering. That label follows each company's own fiscal year, so Apple's June-2025-ending quarter is "Q3 Fiscal Year 2025" since its fiscal year ends in September. It's derived from the fiscal year-end month (the most common month in the company's annualEarnings history) and each quarter's ending month. The label is then used to pull Alpha Vantage's EARNINGS_CALL_TRANSCRIPT, which has an LLM-derived sentiment score (0 to 1) for every speaker turn on the call.

Tone. Each turn is grouped by speaker title. Management is any CEO/CFO/COO/president/chair title, including variants like "CEO of Walmart U.S." Analyst is any title Alpha Vantage tags as Analyst. The operator and investor-relations host are excluded from both. A company's management tone (or analyst tone) is the unweighted average sentiment across its turns in that group, so a short answer and a long one count equally. The "gap" is management tone minus analyst tone for the same call.

Limitation. Alpha Vantage's transcript indexing appears to lag the earnings release by some number of weeks. A company that reported about a month earlier had no transcript yet for that quarter while its prior quarter did. This snapshot doesn't fall back to a prior quarter. A company whose latest quarter isn't indexed is left out and reappears after a later manual refresh.

Surprise % and sentiment scale. Earnings surprise % uses the same guardrails as Earnings Surprise History. A consensus estimate under a nickel is excluded, since a percentage surprise off a near-zero base is noise, and prints beyond ±200% are dropped. Alpha Vantage's sentiment field also never went negative in any transcript inspected, including calls around a double-digit-percent miss. It reads as 0 (neutral/factual) to 1 (highly positive), not a signed measure, so a low score means little positive framing, not confirmed negative framing.

Regressions. The two regressions (Pearson and Spearman, as on Factor Analysis) run once per company on its most recent usable quarter. It is a cross-sectional test, not a time series, since this page is a one-time snapshot. Leaderboards require at least 3 turns in the relevant group so a one-line answer doesn't stand in for a company's average tone.