SPY turn-of-the-month seasonality — last ~3 years (L + F1–F3 vs the rest)
Contrary to the seasonality thesis, the last trading day plus the first three of the next month did not capture SPY’s recent gains. Over the trailing ~3 years those TOM sessions compounded to about 7.40% while the other trading days accounted for roughly 74.78% of the cumulative return — TOM contributed only a small share of the total log-return.
I resampled minute bars to daily closes across 751 sessions, separated the defined TOM window from the rest, and decomposed log-returns (with Welch’s t-test on mean daily returns). The report below lays out the full statistics, charts, and significance testing that overturn the idea that turn-of-month days drove SPY’s performance in 2023‑06 through 2026‑05.
For SPY over the past ~3 years, are gains concentrated in the 'turn-of-the-month' window (the last trading day of each month plus the first three of the next) while the other ~16 sessions are roughly flat? Thesis: that handful of turn-of-month days captured the bulk of SPY's cumulative return and the rest of the month added almost nothing, showing the seasonality is still alive.
How this was measured
Resampled SPY minute bars to daily closes and computed close-to-close returns over the trailing ~36 months (2023-06-01 to 2026-05-29). Turn-of-the-month (TOM) days are defined as the last trading day of each calendar month plus the first three trading days of the following month. Daily log-returns were summed within TOM versus non-TOM buckets to measure cumulative contribution (log-additivity ensures exact decomposition), then exponentiated back to simple returns. Welch's t-test compares mean daily returns between TOM and non-TOM days.
The key numbers
Reading the numbers
Across 751 trading days (141 TOM), TOM days generated a cumulative return of 0.074 while non‑TOM days generated 0.7478; mean daily returns were 0.0005484 (TOM) vs 0.0009637 (non‑TOM) and the two‑sample p=0.6332 shows no statistically‑clear mean difference.
The charts
This cumulative‑return line chart layers three paths: Total ends at 0.8772, the turn‑of‑month path ends at 0.074, and the rest‑of‑month path ends at 0.7478. Visually the TOM line stays near the bottom compared with the tall rise in the rest‑of‑month line — the big chunk of total cumulative gain is in the rest series. That directly challenges the thesis that a few TOM sessions captured most gains: numerically the rest of month accounts for far more of the total path.
The box plots show daily return dispersion for TOM (n=141) and rest (n=610): TOM daily returns range roughly −0.03 to 0.0344 with mean 0.0005, while the rest ranges about −0.0575 to 0.1125 with mean 0.001. The rest group has larger upside outliers and a wider spread; TOM days are tighter and lack the extreme positive moves that drive large cumulative gains. If you were expecting TOM to concentrate the big positive days, this plot shows those big positive outliers live mostly in the rest‑of‑month sample.
The two bars compare mean daily returns: TOM at 0.0005 versus Rest at 0.001. The rest mean is roughly double the TOM mean in these raw numbers, but both means are very small and the reported Welch t p‑value is 0.6332, indicating that difference is not statistically clear. So while the rest has a higher average daily return here, the evidence doesn’t show a reliable mean separation.
This decomposed cumulative bar chart makes the split explicit: TOM contributed 0.074, rest contributed 0.7478, and the total is 0.8772. Look at the tall rest bar versus the much smaller TOM bar — most of the cumulative return over the sample period came from non‑TOM days. That outcome contradicts the idea that a handful of turn‑of‑month sessions captured the bulk of SPY’s gains over these ~3 years.
TOM vs Rest summary (daily stats and cumulative over window)
| group | N_days | mean_daily_ret | median_daily_ret | std_daily_ret | cumulative_return | fraction_positive_days |
|---|---|---|---|---|---|---|
| Turn-of-month (L + F1–F3) | 141 | 0.0005 | 0.0009 | 0.0092 | 0.074 | 0.5603 |
| Rest of month | 610 | 0.001 | 0.0013 | 0.0099 | 0.7478 | 0.5639 |
| Total (sanity check) | 751 | 0.0009 | 0.0012 | 0.0097 | 0.8772 | 0.5632 |
The takeaway
No — over the last ~3 years the turn-of-month days did not concentrate SPY's gains; the rest of the month produced the lion's share. Across 751 trading days there were 141 TOM days: those TOM sessions compounded to about 7.40% while the other 610 days produced about 74.78%, so TOM contributed only ~11.3% of the total log-return. On a per-day basis the TOM mean was actually smaller: 0.0005484 per day (~0.055%) vs 0.0009637 (~0.096%) on non‑TOM days. The difference is not statistically meaningful: Welch's t = -0.4779 with p = 0.633, which is roughly a 63‑in‑100 chance the observed mean gap is just noise. With 751 days (141 TOM) we have reasonable recent coverage, and the data actively contradicts the thesis rather than merely failing to prove it. Bottom line: in 2023‑06 through 2026‑05 the turn‑of‑month seasonality is not driving SPY's returns — the other sessions accounted for almost all of the cumulative gain.
The fine print
- Window is limited to 2023-06-01 through 2026-05-29; seasonality can be regime-dependent.
- TOM windows are clipped at the analysis start/end, so the first/last month may be partial.
- Cumulative contributions are from log-sum decomposition; simple-return shares can differ slightly.
- Welch's t-test assumes independent, roughly normal returns; equity returns are heavy-tailed and can have short-term autocorrelation.