AMD earnings-day sign vs next-5-session drift, and scaling with EPS surprise (last ~3 years)
Does the sign of AMD’s earnings-day move predict the cumulative return over the next five sessions, and does any drift grow with the size of the EPS surprise? We looked at 11 quarterly prints, anchoring the first trade on-or-after the release and measuring day-0 return and the following five-session drift to test that exact thesis.
The headline: no clear pattern. Positive day-0 moves tended to be followed by negative five-session returns while negative day-0 moves tended to be followed by positive five-session returns, but the POS-versus-NEG gap is statistically indistinguishable from noise (Welch p = 0.6649). Correlation and regression tests likewise show essentially no explanatory power; the full methodology, charts and per-event numbers are below.
For AMD over the past ~3 years, does the sign of the earnings-day return predict the cumulative return over the following 5 sessions, and does that drift scale with the size of the EPS surprise? Thesis: stocks that jump on the print keep drifting higher while misses keep sliding, with the drift growing as surprise magnitude rises — the market systematically underreacts to AMD's earnings.
How this was measured
Resampled AMD minute bars to daily closes. For each quarterly earnings with a known reported_date and non-null surprise_percentage, anchored the first trading day on-or-after the release (t0). Computed the earnings-day return as close(t0)/close(t-1)-1 and the forward 5-session drift as close(t0+5)/close(t0)-1. Compared next-5 means after positive vs negative day-0 moves (Welch t-test), measured linear correlation between day-0 and next-5 returns, and tested whether drift scales with EPS surprise via a simple regression (next-5 ~ surprise_percentage). Surprise buckets (terciles) summarize drift by magnitude.
The key numbers
Reading the numbers
Across 11 earnings events, the average five-session return after negative day-0 prints is +2.94% (N=8) while after positive day-0 prints it is −1.66% (N=3), a −4.59 percentage-point gap that is not statistically significant (Welch p=0.6649). Correlation between day-0 and next-5 is mildly negative (r=−0.259, p=0.4417), so no clear continuation signal emerges.
The charts
This scatter compares each earnings-day return (x) to the five-session cumulative return that followed (y). Note the wide spread: day‑0 returns run from −11.3% to +7.31%, while next‑5 returns span −14.17% to +15.83%; points are scattered rather than forming an upward cloud. Statistically the Pearson r is −0.259 (p=0.4417) and the mean next‑5 is +2.94% after NEG day‑0 (N=8) versus −1.66% after POS day‑0 (N=3) with Welch p=0.6649, so there’s no reliable continuation by sign in this sample.
This plot puts EPS surprise (%) on the x-axis and the following five-session return on the y-axis; surprises range from 0.0% to 15.9091% (mean 3.385%). There is no visible slope in the cloud: the fitted slope is essentially zero (−0.0004876 return per 1%-point of surprise) with p=0.9360 and R²≈0.00076, meaning surprise magnitude does not explain subsequent drift. In plain terms, bigger surprises on these 11 events did not predict larger five‑session moves.
The box plots break next‑5 returns into Low, Mid, and High surprise buckets (N=4, 3, 4). Low surprise has mean +1.2679% (range −7.76% to 6.89%), Mid surprise has mean −5.357% (range −14.17% to 4.73%), and High surprise has mean +7.38% (range −3.74% to 15.83%); the High bucket shows the largest positive mean and max, but the Mid bucket is the weakest. Given the small bucket sizes and the nonsignificant slope and p-values, the buckets give mixed evidence rather than a clean, monotonic drift with surprise magnitude.
AMD earnings events — day-0 vs next-5 returns
| t0_trading_day | surprise_pct | day0_ret | next5_ret |
|---|---|---|---|
| 2023-08-01 | 1.75 | 0.0509 | -0.0663 |
| 2023-10-31 | 2.94 | 0.0166 | 0.1583 |
| 2024-01-30 | 0 | -0.1031 | 0.0373 |
| 2024-04-30 | 1.64 | -0.0838 | 0.0473 |
| 2024-07-30 | 1.47 | 0.0731 | -0.1417 |
| 2024-10-29 | 0 | -0.0476 | -0.0776 |
| 2025-02-04 | 0 | -0.0653 | 0.0221 |
| 2025-05-06 | 3.23 | -0.0044 | 0.1299 |
| 2025-08-05 | 0 | -0.0868 | 0.0689 |
| 2025-11-04 | 10.29 | -0.0767 | 0.0445 |
| 2026-02-03 | 15.91 | -0.113 | -0.0374 |
The takeaway
No — in this 11-event sample the sign of AMD's earnings-day return does not reliably predict the cumulative return over the next five sessions, and there is no clear scaling with EPS surprise. Over 11 earnings, the mean next-5 return after positive day-0 moves was −1.66% (N=3) versus +2.94% after negative day-0 moves (N=8), a POS−NEG gap of −4.59 percentage points, but the Welch test gives p = 0.6649 (about a 66-in-100 chance this gap is just noise). The day-0 vs next-5 correlation is r = −0.259 (p ≈ 0.442), and regressing next-5 on EPS surprise gives a slope of −0.000488 (p ≈ 0.936) with R² ≈ 0.00076, so surprise magnitude explains essentially none of the forward return. Surprise terciles show noisy means (low +1.27%, mid −5.36%, high +7.38%), but those bucket averages are based on only 3–4 events each and conflict with the regression’s null result. Bottom line: with only 11 quarterly cases and very high p-values, this looks inconclusive — more a coin flip than a reliable underreaction pattern. Practical takeaway: don’t count on a repeatable, surprise-scaled drift after AMD prints based on this sample.
The fine print
- Only 11 events (quarterly cadence) — thin statistical power.
- t0 maps to the first trading day on/after reported_date; after-close vs pre-open timing is simplified.
- Forward returns are close-to-close over five sessions; intraday post-print drift on day-0 is not captured.
- surprise_pct can be unstable when estimated EPS is near zero; extreme percentage values may skew results.