ORCL post-earnings drift: does the day-0 reaction keep going over the next 20 sessions? (N=11 events)
A modest continuation shows up after Oracle’s quarterly shocks: over 11 earnings events spanning roughly three years, the average signed post‑earnings drift is +3.39% and a solid plurality (about 64%) of day‑0 moves keep heading the same way through the next 20 trading sessions. The test asks whether the earnings‑day reaction is just the opening installment of a larger repricing — greens grinding higher, reds continuing to bleed.
That pattern is suggestive but far from conclusive. Statistical tests fail to reach conventional significance (one‑sample t p ≈ 0.26) and the sample is small, so treat the result as an empirical hint rather than proof. The full methodology, event table, charts and tests follow below.
For ORCL over the past ~3 years, does the market underreact to Oracle's earnings — does the SIGN of the earnings-day reaction keep drifting the same direction over the next 20 sessions, so the day-one move is just the opening installment rather than the full repricing? Thesis: green reactions keep grinding higher and red ones keep bleeding, so forward 20-day returns line up with the initial move's direction and beat the everyday baseline, meaning post-earnings drift is alive in ORCL rather than arbitraged away.
How this was measured
Resampled ORCL minute bars to daily closes. For each quarterly earnings with a known reported_date, set t0 as the first trading day on-or-after that date. Day-0 reaction = close(t0)/close(t−1)−1; forward-20 return = close(t0+20)/close(t0)−1. Classified events by the sign of the day-0 reaction and computed mean forward-20 returns for positive vs negative reactions. Measured 'signed drift' as sign(day-0)×forward-20 so values >0 indicate drift in the same direction as the initial move. Compared signed-drift mean to 0 via one-sample t-test; also compared positive vs negative groups via Welch's test. Baseline is the unconditional 20-day forward return averaged across trading days spanning the event window.
The key numbers
Reading the numbers
Across 11 earnings since 2023-09-11, the sign of the day‑0 move matched the sign of the subsequent 20‑day return about 63.64% of the time, and the average signed drift (sign×fwd20) was +3.394%. That positive mean is not statistically significant (t=1.200, p=0.2577).
The charts
This scatter plots each event's day‑0 close‑to‑close reaction (x from -0.1166 to 0.2889, mean 0.0268) against the 20‑day return after that event (y from -0.1138 to 0.2152, mean -0.0059). What to look at is the cloud of points — there is no clear upward line; the Pearson r is just 0.090 (p=0.791), so day‑0 size and the subsequent 20‑day return show essentially no linear relationship. For your thesis, the scatter does not offer strong visual evidence that the initial move reliably continues in the same direction over 20 trading days.
This histogram shows the signed drift distribution (n=11, min -0.1138, max 0.2152, mean +0.033947). You can see there are more positive than negative values on average, but the spread crosses zero and the one‑sample t (t=1.200) yields p=0.2577, meaning the positive average is not statistically distinguishable from zero. In plain terms, the signed‑drift tilt exists but is weak and not conclusive for a persistent post‑earnings grind.
The three bars compare mean 20‑day returns after positive reactions (+0.0257), after negative reactions (−0.0439), and the baseline 20‑day mean (+0.0194). Positive reactions show a slightly higher 20‑day gain than baseline, and negative reactions show a larger loss, but sample sizes are small (N=6 pos, N=5 neg) and the Welch test gives t=1.261 with p=0.2506, so the pos/neg separation is not statistically clear. This pattern is directionally consistent with your thesis but not robust enough to claim the market systematically underreacts for ORCL over this sample.
Event-level summary — ORCL earnings (last ~3y)
| reported_date | t0_anchor | day0_return | fwd20_return | aligned | surprise_pct |
|---|---|---|---|---|---|
| 2023-09-11 | 2023-09-11 | -0.0918 | -0.0416 | true | 3.48 |
| 2023-12-11 | 2023-12-11 | -0.0832 | -0.0061 | true | 0.75 |
| 2024-03-11 | 2024-03-11 | 0.161 | -0.0533 | false | 2.17 |
| 2024-06-11 | 2024-06-11 | 0.0796 | 0.056 | true | -1.21 |
| 2024-09-09 | 2024-09-09 | 0.0764 | 0.1152 | true | 4.51 |
| 2024-12-09 | 2024-12-09 | -0.0921 | -0.0732 | true | -0.68 |
| 2025-03-10 | 2025-03-10 | -0.0734 | -0.1035 | true | -1.34 |
| 2025-06-11 | 2025-06-11 | 0.069 | 0.2152 | true | 3.66 |
| 2025-09-09 | 2025-09-09 | 0.2889 | -0.0654 | false | -0.68 |
| 2025-12-10 | 2025-12-10 | -0.1166 | 0.005 | false | 38.65 |
| 2026-03-10 | 2026-03-10 | 0.0769 | -0.1138 | false | 4.07 |
The takeaway
Short answer: there’s a mild tendency for the initial earnings-day move in ORCL to continue over the next 20 trading days, but the evidence is far from conclusive. The average signed drift (sign(day‑0) × fwd20) is +3.39%, and about 63.6% of events show the forward 20-day return in the same direction as day‑0. Breaking it down, fwd20 after positive day‑0s averaged +2.57% (N=6) while after negative day‑0s averaged −4.39% (N=5), versus an unconditional 20-day baseline of +1.94%. Statistical tests do not clear the bar: the one‑sample t gives t=1.20 with p≈0.258 (roughly a 1-in-4 chance this pattern is luck), the pos vs neg comparison has p≈0.251, and the correlation between day‑0 and fwd20 is near zero (r≈0.09, p≈0.79). Practical takeaway: you can point to a suggestive pattern, but with only 11 events and non‑significant tests it’s not a reliable, tradeable signal without more data or a cleaner excess‑return test.
The fine print
- Sample is tiny: only 11 earnings events across the ~3-year window, so estimates are noisy.
- These are raw price returns, not excess returns versus a market benchmark.
- Day‑0 is measured close‑to‑close (includes post‑close moves into T0 open), which affects the magnitude.
- 20‑day windows can overlap other news and events; edge events were dropped if T‑1 or T+20 were missing.