Where Have All the Roll-Ups Gone? Consolidation Talk Fades From Calls
What does it mean when management starts talking about industry consolidation? This study flags every call in the corpus (1990-2026) where the model answered yes to "Consolidation Among Peers": 42,662 of 165,182 calls, a 25.8% share (CI 25.6%-26.0%). These calls sound slightly sharper and calmer than average - confidence 7.32 vs 7.21, specificity 7.67 vs 7.56, stress 2.35 vs 2.43 - and lean toward maintained guidance (52.43% vs 48.79%), with fewer lowered (10.60% vs 11.56%). The footprint is shrinking: flagged calls fell from 34.54% of the 2015 sample to 21.57% in 2025. Outcomes look ordinary: across 6,003 calls with linked returns, the median is -6.86% versus -7.16% baseline, and the beat rate is 39.66% versus 39.47%. Consolidation talk marks a tone, not a tell.
- The model answered yes to "Consolidation Among Peers" on 42,662 of 165,182 earnings calls, a 25.8% share (CI 25.6%-26.0%).
- Flagged calls score higher on confidence (7.32 vs 7.21) and specificity (7.67 vs 7.56) and lower on stress (2.35 vs 2.43) and evasion (2.65 vs 2.70) than the corpus.
- Guidance on flagged calls is more often maintained (52.43% vs 48.79%) and less often lowered (10.60% vs 11.56%) or withdrawn (2.36% vs 2.66%).
- The flagged share fell from 34.54% of calls in 2015 to 21.57% in 2025, yet outcomes stay baseline-like, with a median return of -6.86% vs -7.16% and a beat rate of 39.66% vs 39.47%.
1Introduction
Every earnings season, someone asks the M&A question: is your industry consolidating, and where do you stand? Analysts read those answers for hints about pricing power, competitive pressure, and appetite for deal-making, and it is easy to over-read a scripted exchange. How often does consolidation talk actually surface, what does it sound like, and does it coincide with anything measurable? This study measures the pattern at scale: 42,662 calls from 1990 to 2026 where the model answered yes to the battery item "Consolidation Among Peers," compared against the full 165,182-call corpus on language traits, guidance actions, yearly prevalence, and post-call returns. The goal is description, not a trading signal.
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 "Consolidation Among Peers" (n = 42,662; 25.8% of the reference set, 95% Wilson interval 25.6%–26.0%). 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 language deltas are small but directionally consistent: flagged calls score higher on confidence (7.32 vs 7.21) and specificity (7.67 vs 7.56) and lower on stress (2.35 vs 2.43) and evasion (2.65 vs 2.70), while promotion sits at 5.05 on both sides. Guidance tilts conservative: 52.43% maintain guidance versus 48.79% corpus-wide, with lowered guidance at 10.60% versus 11.56% and withdrawals at 2.36% versus 2.66%. Prevalence moves the opposite way, from 34.54% of calls in 2015 to 21.3% in 2022 and 21.57% in 2025, so the topic is getting rarer even as it sounds steadier. Returns refuse to add drama: across 6,003 linked calls the median is -6.86% versus -7.16% baseline (quartiles -24.98% to 11.16%), and beat rates are 39.66% versus 39.47% over 22,449 baseline calls.
| Meter | Study group | Baseline | Δ |
|---|---|---|---|
| Candor | 6.93 | 6.86 | +0.07 |
| Evasion | 2.65 | 2.70 | -0.05 |
| Specificity | 7.67 | 7.56 | +0.11 |
| Stress | 2.35 | 2.43 | -0.08 |
| Promotion | 5.05 | 5.05 | -0.00 |
| Confidence | 7.32 | 7.21 | +0.11 |
| Action | Study group | Baseline |
|---|---|---|
| Raised | 21.8% | 21.1% |
| Maintained | 52.4% | 48.8% |
| Lowered | 10.6% | 11.6% |
| Withdrawn | 2.4% | 2.7% |
| Statistic | Study group | Returns sample |
|---|---|---|
| Median excess return | -6.9% | -7.2% |
| Interquartile range | -25.0% to +11.2% | — |
| Share beating SPY | 39.7% (95% CI 38%–41%) | 39.5% |
| Observations | 6,003 | 22,449 |
| Ticker | Quarter | Call date | Call grade |
|---|---|---|---|
| SBFG | Q2 2025 | 2025-07-25 | A |
| DOC | Q2 2025 | 2025-07-25 | C |
| USCB | Q2 2025 | 2025-07-25 | B+ |
| AON | Q2 2025 | 2025-07-25 | C |
| GBCI | Q2 2025 | 2025-07-25 | A |
| VRTS | Q2 2025 | 2025-07-25 | C+ |
| FFBC | Q2 2025 | 2025-07-25 | B+ |
| SMBC | Q4 2025 | 2025-07-25 | B+ |
4Discussion
The safe conclusions are descriptive: calls that touch on peer consolidation sound slightly more composed, lean toward maintained guidance, and have grown less common since 2015. The unsafe conclusion is causal or predictive; nothing here shows the topic causes steadier guidance, and the near-identical beat rates (39.66% vs 39.47%) point to no edge. One battery item also cannot separate genuine deal-making chatter from rehearsed industry commentary. Read the deltas as a fingerprint of when the subject surfaces, not as a forecast of what happens next.
5Limitations
All AI-read fields, including the yes/no flag and every tone score, are model judgments rather than ground truth, so measurement noise rides along with each delta. The returns comparison covers 22,449 calls skewed toward liquid names, leaving thinly traded issues underrepresented. Our own forward tests falsified directional prediction, and LLMs partially remember famous stocks' histories, which can contaminate any backtest of this kind. The 2025 bucket also reflects just 6,012 calls to date, so its 21.57% share is provisional. Treat every number as descriptive. See the full methodology, including the C1 pattern’s forward-test failure and the LLM-memorization finding.