Earnings Call Transcript Analysis

A 10-K is drafted by lawyers and revised for months. An earnings call transcript is much closer to real time — executives answering unscripted questions from analysts, live. That difference matters: unscripted speech leaks information a carefully lawyered filing doesn't, in word choice, hedging, and how a prepared-remarks tone compares to the tone of the unscripted Q&A that follows it.

The analogy: reading a scripted speech versus the press conference after it

A politician's prepared speech is polished and gives away little beyond what was intended. The unscripted press conference afterward, fielding reporters' questions, is where hesitation, evasive answers, and genuine confidence (or its absence) actually show up. An earnings call has the same two-part structure: prepared remarks (closer to a scripted speech) followed by a Q&A session (closer to the press conference), and the two halves are usually analyzed separately because they carry different kinds of signal.

What gets measured, piece by piece

• Prepared remarks tone. A Loughran-McDonald-style net tone score on the CEO/CFO's opening statement, comparable across quarters for the same company.
• Q&A tone and tone gap. The same tone score computed only on the Q&A section, then compared to the prepared-remarks tone. A large negative gap — upbeat prepared remarks followed by a much gloomier, more hedged Q&A — is a documented predictor of weaker subsequent returns, since analysts' questions tend to probe exactly the parts management would rather not dwell on.
• Hedging and uncertainty language. Counts of words like "approximately," "we believe," "it's hard to say" in Q&A answers, which tend to rise when management is less confident in the numbers they're discussing.
• Linguistic markers of evasion. Answer length relative to question length, and use of vague pronouns instead of specific figures, have been studied as markers correlated with subsequently restated or revised financials.

In plain English: the goal isn't just "was the call positive or negative," it's "did management's confidence hold up once analysts started asking pointed questions" — a signal that lives in the contrast between two parts of the same call, not in either part alone.

Worked example: the prepared-versus-Q&A tone gap

Suppose prepared remarks score a Loughran-McDonald net tone of $+0.35$ (clearly upbeat), while the Q&A section for the same call scores $-0.05$ (roughly neutral to slightly negative). The tone gap is

$$\Delta = \text{tone}_{QA} - \text{tone}_{\text{prepared}} = -0.05 - 0.35 = -0.40 .$$

A large negative gap like this flags a call where the scripted portion was noticeably rosier than what held up under unscripted questioning — historically, calls with the widest such gaps have shown, on average, weaker returns in the following weeks than calls where tone was consistent throughout.

Worked example: hedging word counts across two calls

Company X's Q&A contains 12 hedging phrases ("approximately," "we're not entirely sure," "it depends") across 40 analyst questions — a rate of 0.30 per question. The prior quarter's call had a rate of 0.12 per question for the same company. The near-tripling in hedging rate, even without any change in headline tone words, is itself a signal that management felt less certain answering the same category of questions this quarter.

What this means in practice

Transcript-based signals are attractive because they're available the same day as the call, well before the next filing, and they capture how management responds under unscripted pressure — something a written filing never has to face. The catch is that transcripts vary in length and structure across companies and vendors, tone dictionaries built for filings don't transfer perfectly to spoken language, and any signal built from analyst questions needs to account for the fact that different analysts ask harder or softer questions of different companies.