SPY pre-holiday session returns vs baseline (last ~3 years)
Despite the folklore that stocks tend to rally into holidays, SPY’s last ~3 years show only a tiny, statistically inconclusive bump. We compared 31 pre-holiday sessions to 720 other trading days; pre-holiday sessions averaged 0.1286% versus 0.0868% on other days, but the raw mean gap is only about 0.04 percentage points and the Welch t-test yields p = 0.655. That’s close to a coin flip, not a persistent anomaly.
The study used resampled minute bars to daily closes, labeled the last trading day before full non-trading weekdays as pre-holiday, and tested distributional differences formally. Read on for the full statistics, charts, and the detailed hypothesis test that underpin this result.
For SPY over the past ~3 years, do the trading sessions immediately before US market holidays deliver outsized average returns — the old 'pre-holiday rally' — versus all other days? Thesis: the pre-holiday sessions still post returns well above the everyday baseline, showing this seasonality has stubbornly survived rather than being arbitraged away.
How this was measured
Resampled SPY minute bars to daily closes over the last ~36 months and computed daily close-to-close returns. Identified US market holidays as WEEKDAYS with no trading by comparing a business-day calendar to the observed trading-day index. Labeled a 'pre-holiday session' as the last trading day immediately before each full non-trading weekday, deduplicated when multiple consecutive closure days map to the same prior session. Compared the return distribution of pre-holiday sessions to all other trading days using summary statistics and a Welch two-sample t-test for mean difference (unequal variances).
The key numbers
Reading the numbers
Across the 3-year window, the average close-to-close return on pre-holiday SPY sessions was 0.0012858 versus 0.0008685 on other days, a mean edge of 0.0004173; however the Welch p-value = 0.6551 indicates no statistically-clear difference.
The charts
The box plot stacks the distribution for 31 pre-holiday sessions against 720 other trading days. Look at the tighter vertical range for pre-holiday days (min −0.0071, max 0.0112) versus the much wider tails for other days (min −0.0575, max 0.1125) — pre-holiday returns are less extreme. The group means are 0.0013 (pre-holiday) and 0.0009 (other), but the medians tell a different nuance: median pre-holiday is 0.0009954 while median other days is 0.0012417, so the small mean advantage for pre-holiday sessions appears driven by a few positive moves rather than a clear shift in the middle of the distribution. With only 31 pre-holiday points, that overlap and small sample size argue caution in cl
The bar chart simply highlights the mean daily-return gap: 0.0013 for pre-holiday versus 0.0009 for other days, a mean difference of 0.0004173. That visual gap is real but tiny in absolute terms, and the statistical test (Welch t = 0.4499, two-sided p = 0.6551) shows the difference is not statistically clear. In plain terms, pre-holiday bars sit a bit higher, but given the noise and sample sizes the chart does not prove a reliably outsized pre-holiday rally.
Pre-holiday sessions (last 25 in window)
| pre_holiday_session_date | following_holiday_date | session_return |
|---|---|---|
| 2024-01-12 | 2024-01-15 | 0.0003 |
| 2024-02-16 | 2024-02-19 | -0.0065 |
| 2024-03-28 | 2024-03-29 | 0.001 |
| 2024-05-24 | 2024-05-27 | 0.0063 |
| 2024-06-18 | 2024-06-19 | 0.0024 |
| 2024-07-03 | 2024-07-04 | 0.0059 |
| 2024-08-30 | 2024-09-02 | 0.0084 |
| 2024-11-27 | 2024-11-28 | -0.0028 |
| 2024-12-24 | 2024-12-25 | 0.0112 |
| 2024-12-31 | 2025-01-01 | -0.0015 |
| 2025-01-08 | 2025-01-09 | -0.0022 |
| 2025-01-17 | 2025-01-20 | 0.0104 |
| 2025-02-14 | 2025-02-17 | 0.0002 |
| 2025-04-17 | 2025-04-18 | 0.0024 |
| 2025-05-23 | 2025-05-26 | -0.0071 |
| 2025-06-18 | 2025-06-19 | 0.0019 |
| 2025-07-03 | 2025-07-04 | 0.0076 |
| 2025-08-29 | 2025-09-01 | -0.0056 |
| 2025-11-26 | 2025-11-27 | 0.0066 |
| 2025-12-24 | 2025-12-25 | 0.0039 |
| 2025-12-31 | 2026-01-01 | -0.0048 |
| 2026-01-16 | 2026-01-19 | -0.0022 |
| 2026-02-13 | 2026-02-16 | 0.0004 |
| 2026-04-02 | 2026-04-03 | 0.0018 |
| 2026-05-22 | 2026-05-25 | -0.0005 |
The takeaway
Short answer: No — over the last ~3 years SPY does not show a reliable pre-holiday outperformance. Pre-holiday sessions averaged about 0.13% (31 sessions) versus about 0.09% on other days (720 days), and 64.5% of pre-holiday sessions were positive versus 56.0% for other days. The raw mean gap is tiny: the mean difference is 0.0004173 in returns (about 0.04 percentage points). The Welch test returns p = 0.655, which translates to roughly a 65-in-100 chance this observed gap is just random noise — in plain terms this is basically a coin flip, not a convincing edge. Practical takeaway: you can’t rely on a durable pre-holiday rally in this 36‑month sample — the small average bump seen here is weak and statistically inconclusive.
The fine print
- Holiday labeling is based on missing weekday trading dates and may include rare ad-hoc closures; holidays are not individually named.
- Only 31 pre-holiday sessions in the sample — thin evidence for a persistent effect.
- Window is limited to the most recent ~36 months; seasonality can be regime-sensitive and may differ over longer histories.
- Returns are close-to-close only; intraday patterns (open vs close) are not examined here.