GOOGL: Does 14-day pre-earnings news sentiment foreshadow EPS surprise?
Does the crowd’s mood into a report actually tip you off to the number? We tested that exact idea for GOOGL by measuring the average news sentiment in the 14 calendar days strictly before each quarterly report over the last ~3 years (12 usable events) and comparing it to the EPS surprise. The question matters because many models and traders treat pre-print buzz as a leading signal for beats or misses.
The result is blunt: there’s no reliable lead. The linear correlation is small and negative (Pearson r = -0.217, p ≈ 0.50), and rank tests and group comparisons show no consistent lift in sentiment ahead of beats — in fact, pre-print sentiment was slightly higher before misses. A naive sign rule looks impressive on the surface but is a misleading artifact; the full statistical breakdown and charts are below.
For GOOGL over the past ~3 years, does the tone of news sentiment in the two weeks before each quarterly report actually foreshadow the EPS surprise — is the crowd's pre-print mood a leading tell on the beat, or just positioning noise? Thesis: average pre-earnings sentiment shows essentially no correlation with the sign or size of the surprise, so bullish buzz into the print reflects crowd positioning rather than any real edge on the number.
How this was measured
For each GOOGL quarterly earnings reported in the last ~3 years, compute the mean news sentiment over the 14 calendar days strictly before the reported_date (using ticker_sentiment_score with a fallback to overall_sentiment_score). Join to the reported EPS surprise percentage and test: Pearson/Spearman correlation between sentiment and surprise magnitude; Welch t-test of pre-earnings sentiment for beats vs misses; and a simple sign rule (positive sentiment → predict beat). Day-of headlines are excluded to avoid post-release leakage.
The key numbers
Reading the numbers
Across 12 earnings events, there is no reliable relationship between pre-print sentiment and EPS surprise: Pearson r = -0.217 (p = 0.4978) and Spearman ρ = 0.0629 (p = 0.8459), so any linear or rank association is statistically weak or absent.
The charts
This scatter shows each quarter's 14‑day average sentiment (range 0.0579–0.2209) against the EPS surprise (range −100 to 39.801, mean 3.0481). There is no visible slope — similar sentiment levels map to both big beats and the one very large miss — and with n=12 the Pearson r = -0.217 (p = 0.4978) confirms no significant linear signal. Look at the extreme −100 surprise point: it dominates the vertical spread and illustrates how one outlier, not a consistent trend, creates much of the variation.
The box plots split average pre‑earn sentiment by outcome: the single miss has sentiment 0.2191 while the 11 beats span 0.0579–0.2209 with mean 0.1626. The miss actually sits higher than the beat distribution, which argues against a simple story that higher pre‑print sentiment predicts beats; with only one miss and a Welch p‑value of 1.0 there is no statistically meaningful difference. In short, the apparent overlap and the tiny miss sample make any outcome-based sentiment gap unreliable.
The bar chart shows mean EPS surprise by sentiment polarity: Negative = 0.0, Zero = 0.0, Positive = 3.0481. That pattern reflects that only the positive‑polarity group contributes a nonzero mean surprise in this sample, so the category averages are driven by uneven counts rather than a clean signal. Combine this with the sign‑rule caveat (sign(sent>0 → beat) = 91.67% but excludes zero‑sent events) and it becomes clear these polarity bars are not robust evidence that pre‑print mood forecasts the size or sign of surprises.
GOOGL earnings events — pre-earnings sentiment and surprise
| reported_date | fiscal_date_ending | surprise_pct | pre_sentiment_avg | n_headlines | beat |
|---|---|---|---|---|---|
| 2023-07-25 | 2023-06-30 | 7.46 | 0.1406 | 12 | true |
| 2023-10-24 | 2023-09-30 | 6.9 | 0.2209 | 17 | true |
| 2024-01-30 | 2023-12-31 | 3.14 | 0.0579 | 13 | true |
| 2024-04-25 | 2024-03-31 | 25.17 | 0.128 | 14 | true |
| 2024-07-23 | 2024-06-30 | 2.16 | 0.1767 | 27 | true |
| 2024-10-29 | 2024-09-30 | 14.59 | 0.2119 | 35 | true |
| 2025-02-04 | 2024-12-31 | 0.94 | 0.1273 | 32 | true |
| 2025-04-24 | 2025-03-31 | 39.8 | 0.2088 | 28 | true |
| 2025-07-23 | 2025-06-30 | 5.48 | 0.1635 | 41 | true |
| 2025-10-29 | 2025-09-30 | 23.71 | 0.1658 | 933 | true |
| 2026-02-04 | 2025-12-31 | 7.22 | 0.1868 | 503 | true |
| 2026-04-29 | 2026-03-31 | -100 | 0.2191 | 944 | false |
The takeaway
Short answer: no — 14-day average pre-earnings news sentiment did not foreshadow GOOGL EPS surprises over the last ~3 years. Across 12 usable quarters the linear correlation is small and negative (Pearson r = -0.217) with a p≈0.50 (about a 50-in-100 chance this is luck), and the rank correlation is essentially zero (Spearman ρ = 0.063, p≈0.85). The mean sentiment was actually a hair higher before misses (0.2191) than beats (0.1626), and a Welch test gives no difference (p = 1.00), so there’s no consistent lift in sentiment ahead of beats. A naive sign rule (predict a beat whenever sentiment > 0) shows 91.7% accuracy, but the point-biserial association is negative (r ≈ -0.338, p≈0.28) and non-significant — that high hit rate looks like a misleading artifact, not a robust signal. Bottom line: with only 12 events and large p-values, the evidence is weak-to-nonexistent — treat pre-print bullish buzz as positioning/noise, not a reliable edge on the number.
The fine print
- Tiny sample: just 12 usable quarterly events — statistical power is thin.
- Day-of headlines were excluded; late-evening pre-print pieces can still reflect whispers.
- Sentiment is an unweighted mean of headlines; source or relevance weighting could change results.
- Events with missing or zero estimated EPS were excluded from tests (surprise% NaN).