Show Me the Money You Haven't Collected Yet: Deferred Revenue on Earnings Calls
We asked a language model a single yes/no question about every earnings call in a 165,182-call corpus spanning 1990 to 2026: is deferred revenue growing? It answered yes on 14,649 calls, an 8.87% share (95% CI 8.73%-9.01%). Calls flagged this way show a distinctly promotional tone: promotion scores 5.57 versus 5.05 baseline and confidence 7.54 versus 7.21, with stress lower at 2.21 versus 2.43. These calls are also more likely to have raised guidance (33.02% vs 21.05%) and less likely to have lowered it (9.22% vs 11.56%). Among 2,336 calls with return data, the median next-quarter return was -0.46%, far above the -7.16% baseline, though the mean of 1.77% reflects wide dispersion (Q1 -21.55%, Q3 21.47%).
- Deferred-revenue-growth calls make up 8.87% of the corpus (14,649 of 165,182 calls), with a 95% confidence interval of 8.73% to 9.01%.
- Calls flagged for growing deferred revenue show higher promotion (5.57 vs 5.05) and confidence (7.54 vs 7.21), and lower stress (2.21 vs 2.43) than the corpus baseline.
- Guidance on these calls skews positive: 33.02% raised versus 21.05% in the baseline, while 9.22% lowered versus 11.56%.
- The phrase 'Early Products Growing Fast' appears 1.67x more often than baseline on these calls, and 'Founder-Led Companies' 1.54x more often.
- Among 2,336 calls with post-call returns, the median next-quarter return was -0.46% versus -7.16% for the 22,449-call baseline, with a 49.57% beat rate against a 39.47% base beat rate.
1Introduction
Deferred revenue is one of the few accounting lines management cannot easily talk around: cash has arrived before it can be recognized, and bookings are running ahead of revenue. Whether executives volunteer that fact on an earnings call, and how they talk when they do, says something about the tone of a growth story. A call where the model flags growing deferred revenue is a call sitting on a forward-looking tailwind, and the surrounding language, guidance behavior, and market reception are worth mapping. This study examines all calls in our 165,182-call corpus where the model answered yes to 'Deferred Revenue Growing', tracked by year.
2Data & methodology
The corpus comprises 165,182 earnings-call transcripts published between 1990 and 2026, 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 where the model answered YES to "Deferred Revenue Growing", tracked by year (n = 14,649; 8.9% of the reference set, 95% Wilson interval 8.7%–9.0%). Baseline figures use all scored calls. 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 behavioral profile of these calls is more promotional than the corpus: promotion scores 5.57 against 5.05, confidence 7.54 against 7.21, and stress is lower at 2.21 versus 2.43. Guidance skews positive, with 33.02% of these calls raising versus 21.05% of the baseline and only 9.22% lowering versus 11.56%. Lifted phrases fit the theme: 'Early Products Growing Fast' runs 1.67x baseline and 'Volume About to Step Up' 1.29x, while the only underrepresented phrase is 'The Finished-Story Tell' at 0.68x. Yearly frequency climbed from 7.21 per thousand calls in 2015 to a 2021 peak of 10.69 before easing to 8.78 in 2024. The returns sample (2,336 calls) shows a median next-quarter return of -0.46% versus -7.16% baseline.
| Meter | Study group | Baseline | Δ |
|---|---|---|---|
| Candor | 6.75 | 6.86 | -0.11 |
| Evasion | 2.73 | 2.70 | +0.03 |
| Specificity | 7.62 | 7.56 | +0.06 |
| Stress | 2.21 | 2.43 | -0.22 |
| Promotion | 5.57 | 5.05 | +0.52 |
| Confidence | 7.54 | 7.21 | +0.33 |
| Action | Study group | Baseline |
|---|---|---|
| Raised | 33.0% | 21.1% |
| Maintained | 44.5% | 48.8% |
| Lowered | 9.2% | 11.6% |
| Withdrawn | 1.9% | 2.7% |
| Signal | Lift | In group | Baseline |
|---|---|---|---|
| Early Products Growing Fast | 1.67× | 64.4% | 38.5% |
| Founder-Led Companies | 1.54× | 30.9% | 20.0% |
| A Tiny Fraction of the Market | 1.50× | 45.2% | 30.0% |
| Volume About to Step Up | 1.29× | 36.7% | 28.5% |
| The Finished-Story Tell | 0.68× | 3.0% | 4.4% |
| Statistic | Study group | Returns sample |
|---|---|---|
| Median excess return | -0.5% | -7.2% |
| Interquartile range | -21.5% to +21.5% | — |
| Share beating SPY | 49.6% (95% CI 48%–52%) | 39.5% |
| Observations | 2,336 | 22,449 |
| Ticker | Quarter | Call date | Call grade |
|---|---|---|---|
| MOG.A | Q3 2025 | 2025-07-25 | B+ |
| DBOEY | Q2 2025 | 2025-07-25 | B+ |
| CSBR | Q4 2025 | 2025-07-25 | C+ |
| COUR | Q2 2025 | 2025-07-24 | A |
| DLR | Q2 2025 | 2025-07-24 | B |
| ESRT | Q2 2025 | 2025-07-24 | C+ |
| SKYW | Q2 2025 | 2025-07-24 | B+ |
| BX | Q2 2025 | 2025-07-24 | B |
4Discussion
A careful reader should conclude that calls flagged for growing deferred revenue come with a measurably more promotional tone, more positive guidance activity, and a returns distribution whose median sits well above the broad baseline. What follows should not be a causal claim: we cannot say deferred revenue causes better outcomes, that tone signals future performance, or that the flag identifies better companies rather than companies in a particular phase or sector. The mean return of 1.77% alongside a median of -0.46% shows a heavily skewed distribution. These are descriptive associations in historical data, not a trading signal, and none of our forward tests demonstrated directional predictive value.
5Limitations
The yes/no fields are produced by an AI reader and are noisy: phrasing, hedging, and transcription artifacts can flip a label. The returns sample covers 22,449 baseline calls and 2,336 flagged calls, skewed toward liquid names, so the comparison inherits that selection. Our own forward tests falsified directional prediction from these signals. Additionally, language models partially remember the histories of famous stocks, which can contaminate any backtest that reads calls with an LLM; label quality and market-memory leakage both argue for treating these statistics as descriptive only. See the full methodology, including the C1 pattern’s forward-test failure and the LLM-memorization finding.