Slow and Steady Wins the Call: A Study of Compounding Evidence on Earnings Calls

1Introduction

Most earnings-call language decays fast: a claim is made, a quarter passes, and the narrative resets. The rarer pattern is the compounding call, where each statement appears to build on prior evidence — an early product gaining traction, a small share of a large market, volume ready to step up. For anyone who parses these transcripts, such calls are a distinct genre worth characterizing: they carry a recognizable tonal signature, a distinct guidance posture, and their own frequency over time. This study examines what the 84 compounding-evidence calls in a 470-call corpus spanning 2015–2024 look like across tone, guidance behavior, language themes, and subsequent returns.

2Data & methodology

The corpus comprises 470 earnings-call transcripts published between 2015 and 2024, each scored independently by a large language model on an identical 37-field battery: seven categorical business verdicts, eight 0–9 behavioral meters, and twenty yes/no judgments. The study group is defined as calls that answered YES to the research hypothesis "Compounding evidence" (n = 84; 17.9% of the reference set, 95% Wilson interval 14.7%–21.6%). Baseline figures use the set of calls on which this question was tested. Market outcomes join a fixed sample of 22,449 calls with twelve-month total returns in excess of SPY, measured from the first close after each call; this sample skews toward liquid U.S. names and is reported as descriptive history only.

3Results

The tonal profile of compounding-evidence calls skews promotional: promotion runs +0.48 above the base (5.58 vs 5.11) and confidence +0.31 higher (7.64 vs 7.33), while stress sits 0.26 lower (2.13 vs 2.39); candor (-0.11) and evasion (-0.02) barely move. Guidance leans slightly positive — 26.2% raised versus 21.9% in the base — with withdrawals essentially unchanged (1.2% vs 1.1%). The theme lifts are the most distinctive feature: "Early Products Growing Fast" appears at 1.68x base frequency, "A Tiny Fraction of the Market" at 1.59x, "Founder-Led Companies" at 1.5x, and "Volume About to Step Up" at 1.36x. The annual trend peaked at 0.11 in 2022, fell to 0.09 in 2023 and 0.06 in 2024, with no qualifying calls in the partial 2025 data.

4Discussion

A careful reader should treat this as a description of a recognizable call archetype, not a signal. Compounding-evidence calls do talk differently — more promotion, more confidence, less stress — and they lean on growth-narrative phrases at elevated rates. Their forward returns in this sample (median -7.1% vs -10.0% base) are less bad but still negative, and the returns subsample is small at 28 calls with overlapping confidence intervals on the beat rate (42.9%, CI 26.5%–60.9%). Nothing here establishes that the language caused any outcome, or that identifying this pattern in a new transcript would help an investor. The pattern is real as a description; its usefulness is unproven.

5Limitations

The tone, theme, and hypothesis labels are produced by AI reading call transcripts and are inherently noisy; misclassification at the call level is expected. The returns analysis covers only 28 of these calls within a 22,449-call base skewed toward liquid names, so the comparison is not representative of all listed companies. Our own forward tests falsified directional prediction from these features, and no claim of predictive power survives that result. Additionally, LLMs partially remember famous stocks' history, which contaminates any backtest of language against subsequent returns; measured lift may reflect memorized narratives rather than genuine inference from the text. See the full methodology, including the C1 pattern’s forward-test failure and the LLM-memorization finding.

Companion page: every company matching this hypothesis is listed at the question’s own page.