If You Build It and They Don't Come: The Underused Fixed Costs Signal
We ask what management teams reveal when they admit their fixed costs are underused. Using a model-scored battery across 165,182 earnings-call transcripts from 1990 to 2026, we isolate the 68,779 calls (41.6%) where the model answered yes to "Underused Fixed Costs." These calls sound more promotional (5.17 vs 5.05) but less confident (6.97 vs 7.21) and more stressed (2.82 vs 2.43), with lower specificity (7.42 vs 7.56). Guidance tells a similar story: 15.6% raised versus 21.1% in the base, while 4.3% withdrew guidance versus 2.7%. The flag peaked at 48.47% of calls in 2020 and fell to 38.26% in 2025. Subsequent returns were weaker — median -0.112 versus -0.072 — and the beat rate was 36.4% against 39.5%. The most over-represented language: "Scale-Dependent Advantage Claims," at 1.97x.
- Guidance behavior skews cautious: 15.6% of flagged calls raised guidance versus 21.1% of the base, while 4.3% withdrew it against 2.7%.
- The tone gap runs through stress and confidence: 2.82 versus 2.43 on stress, and 6.97 versus 7.21 on confidence.
- Scale-Dependent Advantage Claims carries the largest phrase lift at 1.97x, while Skeptic Reassured under-indexes at 0.74x.
- Median returns came in at -0.112 versus -0.072 for the base, with a 36.4% beat rate against 39.5%.
1Introduction
Fixed costs are the promise every scaling story rests on: once volumes arrive, margins follow. So when a management team concedes that its fixed costs are underused, listeners lean in — is this a bargain on capacity waiting for demand, or a confession that demand never came? The same sentence can headline a bull case or quietly mark a slowdown. Earnings-call veterans know the tell, but intuition is not measurement. This study takes a systematic look: across 165,182 transcripts (1990-2026), we isolate the 68,779 calls (41.6%) where a model answered yes to "Underused Fixed Costs," and compare their tone, guidance behavior, characteristic phrases, prevalence over time, and subsequent return outcomes against the rest of the 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 the model answered YES to the battery item "Underused Fixed Costs" (n = 68,779; 41.6% of the reference set, 95% Wilson interval 41.4%–41.9%). 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 tone profile is uneasy rather than bullish: stress runs 2.82 versus 2.43, confidence 6.97 versus 7.21, and specificity 7.42 versus 7.56, while promotion edges up to 5.17 from 5.05. The phrase table explains the split — these teams pitch scale ("Scale-Dependent Advantage Claims" at 1.97x, "Volume About to Step Up" at 1.46x) but dodge reassurance ("Skeptic Reassured" at 0.74x). Guidance skews defensive: 15.3% lowered versus 11.6% in the base, and 4.3% withdrawn versus 2.7%. The trend is cyclical, peaking at 48.47% of calls in 2020 after a 2018 trough of 36.45%, and easing to 38.26% in 2025. In the returns sample, the median ran -0.112 versus -0.072, with a 36.4% beat rate against 39.5%.
| Meter | Study group | Baseline | Δ |
|---|---|---|---|
| Candor | 6.82 | 6.86 | -0.04 |
| Evasion | 2.79 | 2.70 | +0.09 |
| Specificity | 7.42 | 7.56 | -0.14 |
| Stress | 2.82 | 2.43 | +0.39 |
| Promotion | 5.17 | 5.05 | +0.12 |
| Confidence | 6.97 | 7.21 | -0.24 |
| Action | Study group | Baseline |
|---|---|---|
| Raised | 15.6% | 21.1% |
| Maintained | 44.9% | 48.8% |
| Lowered | 15.3% | 11.6% |
| Withdrawn | 4.3% | 2.7% |
| Signal | Lift | In group | Baseline |
|---|---|---|---|
| Scale-Dependent Advantage Claims | 1.97× | 21.8% | 11.1% |
| Volume About to Step Up | 1.46× | 41.5% | 28.5% |
| Results Worse Than Direction | 1.43× | 73.4% | 51.1% |
| A Tiny Fraction of the Market | 1.34× | 40.2% | 30.0% |
| The Hidden Segment | 1.32× | 27.9% | 21.1% |
| Skeptic Reassured | 0.74× | 49.4% | 66.4% |
| Statistic | Study group | Returns sample |
|---|---|---|
| Median excess return | -11.2% | -7.2% |
| Interquartile range | -33.5% to +11.7% | — |
| Share beating SPY | 36.4% (95% CI 35%–38%) | 39.5% |
| Observations | 6,316 | 22,449 |
| Ticker | Quarter | Call date | Call grade |
|---|---|---|---|
| SBFG | Q2 2025 | 2025-07-25 | A |
| DOC | Q2 2025 | 2025-07-25 | C |
| FLG | Q2 2025 | 2025-07-25 | B |
| FRST | Q2 2025 | 2025-07-25 | A |
| HMDPF | Q2 2025 | 2025-07-25 | B |
| ULH | Q2 2025 | 2025-07-25 | C+ |
| AN | Q2 2025 | 2025-07-25 | B |
| CHTR | Q2 2025 | 2025-07-25 | C+ |
4Discussion
A careful reader can conclude that calls admitting underused fixed costs travel with a recognizable package: more stress, less confidence, less specific detail, more scale talk, softer guidance, and modestly weaker measured returns. What you should not conclude is that the flag predicts anything. The median gap (-0.112 versus -0.072) is a descriptive difference in a sample skewed toward liquid names, and our own forward tests failed to confirm directional value. Treat the label as context — management leaning on a scaling story the numbers have not yet delivered — and weigh it alongside everything else on the call.
5Limitations
These fields are AI-read and noisy; the Underused Fixed Costs label is a model judgment, not an audited line item. The returns comparison covers 22,449 calls (6,316 of them flagged) and skews toward liquid names, so it may not generalize to the full corpus. Most importantly, our own forward tests falsified directional prediction from this signal — none of these gaps supports a trading rule. Finally, LLMs partially remember famous stocks' histories, which contaminates any backtest; read the numbers as descriptive language statistics, not evidence of an edge. See the full methodology, including the C1 pattern’s forward-test failure and the LLM-memorization finding.