Capex Up, Stress Down: Reading 64,183 Calls Where Spend Was on the Rise
This study profiles earnings calls where capex was read as increasing, drawing on 64,183 calls (38.86% of a 165,182-call corpus spanning 1990-2026). Calls with rising capex read as more confident (7.45 vs. 7.21), more promotional (5.17 vs. 5.05), and less stressed (2.20 vs. 2.43) than the baseline. Guidance behavior differs too: 25.52% raised guidance versus 21.06% in the base, while only 8.32% lowered it versus 11.56%. Post-call return distributions are nearly indistinguishable: median -7.14% vs. -7.16%, and a 39.17% beat rate vs. 39.47% base. The annual share peaked at 50.04% in 2021 and fell to 29.96% in 2025.
- Calls with increasing capex score higher on confidence (7.45 vs. 7.21) and lower on stress (2.20 vs. 2.43) than the base corpus.
- Guidance raises are more common in this group (25.52% vs. 21.06%), while withdrawals are rarer (1.34% vs. 2.66%).
- Forward returns look essentially identical: median -7.14% vs. -7.16%, with beat rates of 39.17% vs. 39.47%.
- The annual share of such calls peaked at 50.04% in 2021, then declined to 35.05% in 2024 and 29.96% in 2025.
1Introduction
When a management team commits to higher capital expenditure, it is usually read as a statement of conviction: the business is strong enough to deserve more investment. But conviction is also a rhetorical posture, and earnings calls are performances as much as disclosures. Whether talk of rising capex carries behavioral or market-relevant signals is an open empirical question, not an article of faith. This study examines 64,183 earnings calls where capex was read as increasing, profiling their language, guidance behavior, annual prevalence, and post-call return outcomes against the broader corpus.
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 capex was read as increasing (n = 64,183; 38.9% of the reference set, 95% Wilson interval 38.6%–39.1%). 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 tilts upbeat: specificity runs 7.66 vs. 7.56, promotion 5.17 vs. 5.05, and confidence 7.45 vs. 7.21, while stress reads 2.20 vs. 2.43. Guidance skews positive, with raises at 25.52% vs. 21.06% and lowers at 8.32% vs. 11.56%. The annual share climbed from 34.71% in 2015 to a 50.04% peak in 2021 before easing to 29.96% in 2025. Yet market outcomes barely differ: median forward return -7.14% vs. -7.16% and a 39.17% vs. 39.47% beat rate.
| Meter | Study group | Baseline | Δ |
|---|---|---|---|
| Candor | 6.89 | 6.86 | +0.02 |
| Evasion | 2.68 | 2.70 | -0.02 |
| Specificity | 7.66 | 7.56 | +0.10 |
| Stress | 2.20 | 2.43 | -0.23 |
| Promotion | 5.17 | 5.05 | +0.12 |
| Confidence | 7.45 | 7.21 | +0.23 |
| Action | Study group | Baseline |
|---|---|---|
| Raised | 25.5% | 21.1% |
| Maintained | 51.7% | 48.8% |
| Lowered | 8.3% | 11.6% |
| Withdrawn | 1.3% | 2.7% |
| Statistic | Study group | Returns sample |
|---|---|---|
| Median excess return | -7.1% | -7.2% |
| Interquartile range | -25.4% to +11.2% | — |
| Share beating SPY | 39.2% (95% CI 38%–40%) | 39.5% |
| Observations | 10,332 | 22,449 |
| Ticker | Quarter | Call date | Call grade |
|---|---|---|---|
| DOC | Q2 2025 | 2025-07-25 | C |
| HCA | Q2 2025 | 2025-07-25 | C |
| AON | Q2 2025 | 2025-07-25 | C |
| NWG | Q2 2025 | 2025-07-25 | B+ |
| FFIC | Q2 2025 | 2025-07-25 | B+ |
| MOG.A | Q3 2025 | 2025-07-25 | B+ |
| VRTS | Q2 2025 | 2025-07-25 | C+ |
| HMDPF | Q2 2025 | 2025-07-25 | B |
4Discussion
The honest takeaway is contrast: rising-capex calls sound better and guide better, but their post-call return distributions look like everyone else's. A careful reader should treat the behavioral and guidance gaps as descriptive features of these calls, not as a recipe for outperformance. No causal claim is warranted, and the near-identical medians and beat rates argue against any simple directional interpretation.
5Limitations
AI-read fields like candor and stress are noisy measurements of tone, not ground truth. The returns sample covers 10,332 calls from a 22,449-call base skewed toward liquid names, so results may not generalize. Our own forward tests falsified directional prediction on these features. LLM scoring can partially remember famous stocks' histories, contaminating any backtest. The 2025 trend figure reflects a partial year of data. See the full methodology, including the C1 pattern’s forward-test failure and the LLM-memorization finding.