META earnings-event realized range: pre-10d trend, event-day spike, and post-week behavior (≈3y window)
Traders often price in a slow buildup of volatility into earnings; for META over the last ~36 months the data do not bear that out. We measured realized intraday range ((high−low)/close) around 12 quarterly prints (t−10…t+7) and compared the pre-10d window, the event day, and the following week to an unconditional baseline. The 10 trading days before reports averaged 11.03% versus a baseline of 11.81%, the earnings session spiked to 21.95% (p=0.0468), and the subsequent week averaged 17.26% (p=0.0111).
The short thesis: there is no statistically supported gradual ramp into prints — volatility is essentially normal heading into t0, detonates on the earnings session, and then stays elevated for several days. The detailed tests, charts, and robustness checks follow below.
For META over the past ~3 years, does its daily high-low range actually build up across the ten sessions before each quarterly earnings report the way traders assume, or does volatility stay near-normal right into the print and then detonate almost entirely in the single earnings session before collapsing below baseline over the following week? Thesis: realized range is flat and unremarkable heading into the report, spikes overwhelmingly on the earnings day itself, and then gets crushed to below-average afterward, so the move is a one-session shock rather than a gradual pre-print build.
How this was measured
Resampled META minute bars to daily OHLCV and computed realized intraday range as (high−low)/close. Identified quarterly earnings dates from META_earnings and snapped each to the first trading day on-or-after the reported_date (captures pre-open and after-close releases cleanly). For each event, collected a window of t−10…t+7 trading days and aggregated realized range by offset. Compared pre-10d (−10…−1), earnings day (0), and post-week (+1…+5) distributions against the unconditional baseline via Welch t-tests. Tested pre-window build-up by fitting a linear slope of mean range vs days-to-event (10→1). Window capped to the last 36 months of price history (2023-06-30 to 2026-06-30).
The key numbers
Reading the numbers
Across 12 earnings events, the pre-10-day mean range sits near baseline (0.110289 vs baseline 0.118053), the earnings-day mean nearly doubles the range to 0.219460 (~1.86x baseline), and the first post-week stays elevated at 0.172554 (~1.46x baseline).
The charts
The line plots mean realized range from t-10 through t+7 against a flat baseline at 0.118053. Look for the flat band before the event: the pre-window averages are close to baseline (pre-10d mean 0.110289) and the pre-window slope is slightly negative (-0.002666) and not statistically meaningful (p=0.5194), so there is no steady build into the print. Instead the series shows a clear single-session spike around the event (earnings-day mean 0.219460, with the plot's max at 0.2404) followed by a decline; the week after remains higher than baseline on average (post-week mean 0.172554) rather than showing a gradual pre-print accumulation.
The box plots compare distributions for pre (-10..-1), event-day (0), post (+1..+5) and the unconditional baseline. The event-day group has a much higher average (mean 0.219460, min 0.0965, max 0.6481, n=12) than the pre-window (mean 0.110289, min 0.0134, max 0.8231, n=120) while the post-week sits between event and baseline (mean 0.172554, min 0.0291, max 0.7048). Statistical tests mirror that picture: pre vs baseline is not different (p=0.4692), but the earnings-day and the post-week both differ from baseline (p=0.0468 and p=0.0111), indicating a concentrated jump at the print and a persistent elevated distribution afterward rather than a gradual pre-print ramp.
Per-offset realized range summary (META earnings windows)
| offset | N | mean | median | std |
|---|---|---|---|---|
| t-10 | 12 | 0.0827 | 0.0706 | 0.0475 |
| t-9 | 12 | 0.1633 | 0.0903 | 0.2204 |
| t-8 | 12 | 0.0794 | 0.0564 | 0.0568 |
| t-7 | 12 | 0.1007 | 0.0995 | 0.0626 |
| t-6 | 12 | 0.0829 | 0.0752 | 0.0536 |
| t-5 | 12 | 0.082 | 0.0809 | 0.0468 |
| t-4 | 12 | 0.1129 | 0.074 | 0.1133 |
| t-3 | 12 | 0.1413 | 0.1001 | 0.1098 |
| t-2 | 12 | 0.1682 | 0.1105 | 0.1505 |
| t-1 | 12 | 0.0894 | 0.0701 | 0.0564 |
| t+0 | 12 | 0.2195 | 0.1677 | 0.1566 |
| t+1 | 12 | 0.2178 | 0.1429 | 0.2048 |
| t+2 | 12 | 0.1676 | 0.1235 | 0.1369 |
| t+3 | 12 | 0.2201 | 0.1246 | 0.2053 |
| t+4 | 12 | 0.1649 | 0.1366 | 0.1255 |
| t+5 | 12 | 0.0924 | 0.0841 | 0.0605 |
| t+6 | 12 | 0.1148 | 0.1069 | 0.0648 |
| t+7 | 12 | 0.2404 | 0.1069 | 0.2543 |
The takeaway
Short answer: the tape into META’s prints shows no gradual volatility build — range is near-normal before the report, it detonates on the earnings session, and then stays elevated for several days rather than collapsing below baseline. The numbers: the 10 trading days before reports averaged 11.03% realized range versus a baseline of 11.81% (pre-window N=120; ratio 0.93; Welch p=0.469), while the earnings day jumped to 21.95% (N=12; 1.86× baseline; Welch p=0.0468) and the following week averaged 17.26% (N=60; 1.46× baseline; Welch p=0.0111). The pre-window slope is slightly negative (-0.002666) and non-significant (p=0.519), so there’s no statistical evidence of a rising build into t0. How confident should you be: the event-day spike is statistically significant but rests on only 12 events (about a 4.7-in-100 chance this is luck), while the post-week elevation is a stronger signal (about a 1.1-in-100 chance of luck with 60 observations); the pre-print null is essentially a coin flip. Practical takeaway: don’t expect a slow ramp — for META over this period volatility tends to arrive as a one-session shock and then linger above normal for days, not as a gradual pre-print buildup.
The fine print
- Earnings anchor snaps to the first trading day on/after reported_date; precise release time (pre-open vs after-close) isn’t used.
- Baseline is the unconditional daily realized range; a regime- or volatility-adjusted baseline could change the multiples.
- Event counts are small: 12 distinct earnings days (pre-window N=120; post-week N=60), so single events carry weight.
- Tests assume day-level independence; overlapping windows or nearby regime shifts could bias results.