AAPL: Do pre-print estimate revisions (last 4 weeks) predict the earnings-day reaction?
Surprisingly, AAPL’s final-month net EPS revisions carry almost no predictive signal for the earnings-day price move. Across 12 earnings events the 30-day relative drift versus the next-day close gave Pearson r = 0.187 (p = 0.5598) and Spearman rho = -0.069, and the drift sign matched the reaction only 58.3% of the time. A superficially wide gap in mean reactions (+0.55% vs −2.88%) collapses under scrutiny—the negative-drift bucket has a single event and a Welch test returns p = 1.00.
Below you’ll find the full methodology and charts: how we measured close-to-close earnings reactions, constructed 30-day consensus drift and revision counts, and tested correlations (including surprise vs reaction). The full evidence supports the working thesis: last-month net analyst revisions look largely baked into price, not predictive of the earnings-day move.
For AAPL over the past ~3 years, does the net direction of analyst EPS-estimate revisions in the four weeks before each quarterly report predict the sign or magnitude of the earnings-day price reaction? Thesis: pre-print revisions are already baked into the price and show essentially no relationship with the actual earnings-day move — only the surprise versus the latest estimate matters.
How this was measured
For each AAPL earnings over the last ~3 years, we measured the earnings-day price reaction as the close-to-close return from the last trading day before the first trading day on/after the reported_date to that first trading day's close (captures BMO/AMC reactions). Using AAPL_estimates, we took the latest snapshot on/before the day prior to each report and computed 30-day consensus drift (latest − 30d-ago, both absolute and relative) plus revision counts (up vs down in trailing 30 days). We tested whether these pre-print revision signals correlate with the earnings-day reaction via Pearson/Spearman correlations and a Welch test (positive- vs negative-drift buckets). As a benchmark, we also measured correlation between earnings surprise and the reaction.
The key numbers
Reading the numbers
Twelve earnings events studied. The relationship between 30‑day consensus drift and earnings‑day reaction is very weak (Pearson r=0.187, p=0.5598) and surprise% vs reaction is essentially zero (r=0.021, p=0.95).
The charts
The scatter plots each event’s 30‑day relative consensus drift (a very small range: min −0.0033 to max 0.0035, mean 0.0023) against the earnings‑day close‑to‑close move (much wider: min −0.0388 to max 0.077, mean 0.0026). There is no visible slope and the formal stats confirm it: Pearson r=0.187 (p=0.5598) and Spearman rho ≈ −0.069 (p=0.8303), so neither linear nor monotonic drift predicts the reaction. The takeaway for your question is that pre‑print revisions over the prior 30 days do not reliably explain the size or sign of the earnings‑day move.
This scatter shows earnings surprise percentage (all positive here, range 0.0185–0.0979, mean 0.0441) against the same earnings‑day reactions. Despite every event being a positive surprise, reactions still span negative to positive, and the correlation is essentially zero (Pearson r=0.021, p=0.95). In short, surprise magnitude in this sample does not map cleanly to the post‑earnings price move either.
The bar chart summarizes mean reaction by revision direction over the last 30 days: down revisions mean −0.0288, flat 0.0, up revisions mean 0.0054515630, and no‑estimate 0.0. Note the down‑revision mean comes from just one event (N=1) while up revisions cover the rest, so comparisons are noisy, and a Welch test reports p=1.0, indicating no clear difference. The sign‑match rate is only about 58.33%, so a pre‑print drift sign only slightly more often than not aligns with the earnings‑day sign.
Event-level summary (last ~3 years)
| reported_date | fiscal_quarter | est_snapshot_date | drift_rel_30d | rev_up_30d | rev_dn_30d | reaction_return | surprise_pct_dec |
|---|---|---|---|---|---|---|---|
| 2023-08-03 | Q2 2023 | 2022-09-30 | -0.0033 | 7 | 1 | -0.0288 | 0.0588 |
| 2023-11-02 | Q3 2023 | 2023-09-30 | 0.0035 | 1 | 1 | -0.0183 | 0.0504 |
| 2024-02-01 | Q4 2023 | 2023-09-30 | 0.0035 | 1 | 1 | -0.0228 | 0.0332 |
| 2024-05-02 | Q1 2024 | 2023-09-30 | 0.0035 | 1 | 1 | 0.077 | 0.02 |
| 2024-08-01 | Q2 2024 | 2023-09-30 | 0.0035 | 1 | 1 | -0.0185 | 0.0448 |
| 2024-10-31 | Q3 2024 | 2024-09-30 | 0.003 | 6 | 1 | -0.0388 | 0.0211 |
| 2025-01-30 | Q4 2024 | 2024-09-30 | 0.003 | 6 | 1 | 0.0231 | 0.0256 |
| 2025-05-01 | Q1 2025 | 2024-09-30 | 0.003 | 6 | 1 | -0.0182 | 0.0185 |
| 2025-07-31 | Q2 2025 | 2024-09-30 | 0.003 | 6 | 1 | 0.0185 | 0.0979 |
| 2025-10-30 | Q3 2025 | 2025-09-30 | 0.0016 | 9 | 3 | 0.0256 | 0.0511 |
| 2026-01-29 | Q4 2025 | 2025-09-30 | 0.0016 | 9 | 3 | 0.0113 | 0.0637 |
| 2026-04-30 | Q1 2026 | 2025-09-30 | 0.0016 | 9 | 3 | 0.021 |
The takeaway
No — in this 12-event sample, the net direction of analyst EPS revisions in the last ~30 days before an AAPL print does not predict the earnings-day price reaction. The Pearson correlation between 30d relative drift and the reaction is only r=0.187 (N=12, p=0.5598) and Spearman rho = -0.069 (p=0.8303), i.e., tiny and not statistically significant. Average reactions appear superficially different (+0.55% after positive drift vs -2.88% after negative drift), but the negative-drift bucket contains only one event and the Welch test gives p=1.00, so that difference is not credible. The drift sign matched the reaction sign just 58.3% of the time (N=12), and surprise percentage also showed no relationship with reaction (Pearson r=0.021, N=11, p=0.95). Bottom line: the data lean toward “no predictive power” for final-month net revisions, but the evidence is weak and not definitive given the small, imbalanced sample.
The fine print
- Sample is just 12 earnings events over ~3 years; statistical power is limited and results can flip with a few events.
- The earnings-day reaction is close-to-close across the report boundary; intraday timing (BMO vs AMC) and path are not separated.
- Some estimate snapshots may reflect inconsistent horizon labels; a few rows could target a different quarter than intended.
- Surprise percentage can be unstable or missing when estimated EPS is near zero; such rows were dropped from surprise tests.