AI Research GOOGLGOOGL_earnings

GOOGL: buy‑the‑rumor‑sell‑the‑news? Pre‑earnings 20d rally vs earnings‑day return (N=87 events)

87
Events analysed

Over the past three years, GOOGL has staged only two earnings events where the stock rallied more than 13.7% in the prior 20 sessions. In both cases, the stock actually gained another 4% on earnings day — not exactly a sell-the-news pattern.

That small sample makes it impossible to draw a reliable conclusion. Across all 87 events, the overall correlation between the pre-earnings run and the earnings-day return is weakly positive (Spearman ρ = 0.24, p = 0.024), which runs directly counter to the idea that big rallies invite selling. The full analysis below unpacks the numbers and the methodology.

The research question

For GOOGL over the past ~3 years, does the stock perform worse on earnings day when it has already staged a top-decile rally over the prior 20 sessions — a buy-the-rumor-sell-the-news pattern? Thesis: a big pre-earnings run-up steals the upside surprise, so the actual report meets with selling.

How this was measured

Daily GOOGL close data; trailing 20‑trading‑session return computed as (close[t‑1] / close[t‑21] – 1) for every trading day. The 90th percentile of this distribution across ALL trading days defines the “top‑decile rally” threshold. Each quarterly earnings event is assigned the trading day on or after its reported date. Its pre‑earnings 20‑day rally is the trailing return as of that trading day; earnings‑day return is the close‑to‑close return on that same day. Events are split into “top‑decile pre‑run” and all others. A Welch t‑test compares the two groups’ earnings‑day returns. Additionally, Spearman and Pearson correlations between the pre‑earnings rally magnitude and the earnings‑day return are reported across all events.

The key numbers

Events analysed
87
quarterly reports with sufficient prior history
Top‑decile threshold (20d ret)
13.7193%
90th percentile across all trading days
Top‑decile events count
2
pre‑earnings 20d ret ≥ threshold
Top‑decile mean earn‑day ret
4.0338%
Other events mean earn‑day ret
-1.4043%
Top‑decile median earn‑day ret
4.0338%
Other events median earn‑day ret
-1.7319%
Welch t‑stat (top vs other)
1.762
positive = top‑decile day higher
Welch p‑value (two‑sided)
0.3257
p=0.3257 ≥ 0.05 → no statistically‑clear difference
Spearman ρ (pre‑run vs earn‑day)
0.2414
|ρ|=0.241 ≤ 0.3 → weak association
Spearman p‑value
0.0243
p=0.0243 < 0.05 → relationship detected
Pearson r
-0.1889

Reading the numbers

Across 87 earnings events, the typical GOOGL earnings-day return was slightly negative (-1.4%). In the 2 events with a top-decile pre-earnings run-up (above 13.7%), the stock actually gained about 4.0% on earnings day, but with only 2 such events and a p-value of 0.33, this pattern is not statistically reliable.

The charts

GOOGL pre‑earnings 20‑day rally vs earnings‑day return
What this chart says

The scatterplot plots each earnings event based on its pre-earnings 20-day return (x-axis) versus its earnings-day return (y-axis). Most observations are near the center, with a slight negative average on earnings day (around -1.3%). The two points farthest to the right, representing the top-decile pre-rallies, sit above the zero line, indicating they actually gained on earnings day. This is the opposite of a buy-the-rumor-sell-the-news pattern, but because there are only two such events, we cannot draw a strong conclusion from the chart alone.

Earnings‑event details

reported_dateearn_datetrail_20earn_rettop_decile
2004-10-212023-08-010.1097-0.0173false
2018-07-232023-08-010.1097-0.0173false
2018-04-232023-08-010.1097-0.0173false
2018-02-012023-08-010.1097-0.0173false
2017-10-262023-08-010.1097-0.0173false
2017-07-242023-08-010.1097-0.0173false
2017-04-272023-08-010.1097-0.0173false
2017-01-262023-08-010.1097-0.0173false
2016-10-272023-08-010.1097-0.0173false
2016-07-282023-08-010.1097-0.0173false
2016-04-212023-08-010.1097-0.0173false
2016-02-012023-08-010.1097-0.0173false
2015-10-222023-08-010.1097-0.0173false
2015-04-232023-08-010.1097-0.0173false
2015-01-292023-08-010.1097-0.0173false
2014-10-162023-08-010.1097-0.0173false
2018-10-252023-08-010.1097-0.0173false
2019-02-042023-08-010.1097-0.0173false
2019-04-292023-08-010.1097-0.0173false
2019-07-252023-08-010.1097-0.0173false
2023-07-252023-08-010.1097-0.0173false
2023-04-252023-08-010.1097-0.0173false
2023-02-022023-08-010.1097-0.0173false
2022-10-252023-08-010.1097-0.0173false
2022-07-262023-08-010.1097-0.0173false
2022-04-262023-08-010.1097-0.0173false
2022-02-012023-08-010.1097-0.0173false
2014-07-172023-08-010.1097-0.0173false
2021-10-262023-08-010.1097-0.0173false
2021-04-272023-08-010.1097-0.0173false
2021-02-022023-08-010.1097-0.0173false
2020-10-292023-08-010.1097-0.0173false
2020-07-302023-08-010.1097-0.0173false
2020-04-282023-08-010.1097-0.0173false
2020-02-032023-08-010.1097-0.0173false
2019-10-282023-08-010.1097-0.0173false
2021-07-272023-08-010.1097-0.0173false
2014-04-162023-08-010.1097-0.0173false
2015-07-162023-08-010.1097-0.0173false
2013-10-172023-08-010.1097-0.0173false
2008-10-162023-08-010.1097-0.0173false
2008-07-172023-08-010.1097-0.0173false
2014-01-302023-08-010.1097-0.0173false
2008-01-312023-08-010.1097-0.0173false
2007-10-182023-08-010.1097-0.0173false
2007-07-192023-08-010.1097-0.0173false
2007-04-192023-08-010.1097-0.0173false
2009-01-222023-08-010.1097-0.0173false
2007-01-312023-08-010.1097-0.0173false
2006-07-202023-08-010.1097-0.0173false
2006-04-202023-08-010.1097-0.0173false
2006-01-312023-08-010.1097-0.0173false
2005-10-202023-08-010.1097-0.0173false
2005-07-212023-08-010.1097-0.0173false
2005-04-212023-08-010.1097-0.0173false
2005-02-012023-08-010.1097-0.0173false
2006-10-192023-08-010.1097-0.0173false
2009-04-162023-08-010.1097-0.0173false
2008-04-172023-08-010.1097-0.0173false
2009-10-152023-08-010.1097-0.0173false
2013-07-182023-08-010.1097-0.0173false
2013-04-182023-08-010.1097-0.0173false
2013-01-222023-08-010.1097-0.0173false
2009-07-162023-08-010.1097-0.0173false
2012-07-192023-08-010.1097-0.0173false
2012-04-122023-08-010.1097-0.0173false
2012-01-192023-08-010.1097-0.0173false
2012-10-182023-08-010.1097-0.0173false
2011-07-142023-08-010.1097-0.0173false
2011-04-142023-08-010.1097-0.0173false
2011-01-202023-08-010.1097-0.0173false
2011-10-132023-08-010.1097-0.0173false
2010-10-142023-08-010.1097-0.0173false
2010-07-152023-08-010.1097-0.0173false
2010-04-152023-08-010.1097-0.0173false
2010-01-212023-08-010.1097-0.0173false
2023-10-242023-10-240.0478-0.0523false
2024-01-302024-01-300.097-0.0717false
2024-04-252024-04-250.02340.126false
2024-07-232024-07-230.0134-0.0225false
2024-10-292024-10-290.01110.0718false
2025-02-042025-02-040.072-0.0628false
2025-04-242025-04-24-0.08780.071false
2025-07-232025-07-230.15030.0096true
2025-10-292025-10-290.10580.0933false
2026-02-042026-02-040.0802-0.0301false
2026-04-292026-04-290.28630.0711true

The takeaway

The data don't support a buy-the-rumor-sell-the-news pattern for GOOGL over the past three years. Across 87 earnings events, only two landed in the top decile of 20-day pre-earnings rallies (a gain of at least 13.7%). Those two cases actually posted an average earnings-day gain of +4.0%, while the other 85 events averaged a loss of -1.4%. But with just two observations in the "big rally" group, the difference is nowhere near statistically significant — the p-value of 0.33 means there's roughly a 1-in-3 chance you'd see this split if the true effect were zero. The overall correlation between pre-earnings run-up and earnings-day return is weakly positive (Spearman ρ = 0.24, p = 0.024), which runs counter to the hypothesis that big rallies invite selling. In plain terms: the sample is far too thin to conclude anything reliable about a sell-the-news effect after large pre-earnings gains.

The fine print