QQQ turn-of-the-month effect — last trading day + first 3 of next month (last ~3 years)
The turn-of-the-month pattern didn’t explain QQQ’s rally over the last three years. I tagged the last trading day of each month plus the first three of the next and compared their compounded contribution to the rest of the sessions; those 144 ToM days were 19.2% of the 752 trading days but accounted for only about 12.78% of the total cumulative return.
Across the sample the remainder of sessions produced most of the gain and average daily returns were actually a touch higher outside the turn; the difference is statistically indistinguishable. Below is the full decomposition, monthly windows, t-test results and charts that back this conclusion.
For QQQ over the past ~3 years, does the 'turn-of-the-month' flow effect still pay — do the last trading day of the month plus the first three of the next capture the bulk of QQQ's entire cumulative return, while the other ~16 sessions of the month net out to roughly nothing? Thesis: those four calendar-driven days, barely 20% of all sessions, account for most of the index's gains as payroll/401k and index-rebalance money hits, so the middle of the month is dead money and the 'buy the turn' anomaly is alive rather than arbitraged away.
How this was measured
Built a daily close-to-close return series for QQQ limited to the last 36 months ending on the latest available date. Tagged turn-of-the-month (ToM) days as the union of (i) the last trading day of each calendar month and (ii) the first three trading days of the subsequent month. Decomposed cumulative performance using log returns: sum of logs over ToM days vs mid-month days, with expm1 to report compounded returns. Also computed a per-boundary ToM window return (prev month’s last + next month’s first three) for each month. Welch’s t-test compares mean daily returns on ToM vs mid-month days.
The key numbers
Reading the numbers
Over the ~3-year window QQQ's cumulative all-day return is 1.0264 (about +102.6%). The four turn-of-month sessions (19.17% of days) produced cumulative 0.1278 (+12.78%) and account for 17.03% of total log-return — not the majority.
The charts
This cumulative-return line compares the three series from 2023-06-30 to 2026-06-30. Look at the end points: All-days finishes at 1.0264, mid-month at 0.7968 and turn-of-month at 0.1278 — the mid-month line carries the bulk of the upward movement. That directly contradicts a story that the four turn days alone drive most gains: visually and numerically the mid-month contribution is much larger.
The two bars show each subset's share of total log-return: turn-of-month 0.1703 versus mid-month 0.8297. Despite being ~19% of sessions, ToM captures only about 17% of log-return while mid-month captures roughly 83%. In log-return terms the middle-of-month dominates rather than the turn.
The box plots summarize daily returns for the two groups: means are almost the same (ToM mean 0.000912 vs mid-month mean 0.001057) and the distributions overlap a lot. ToM has a narrower observed range (min −0.0387, max 0.0434) while mid-month shows wider tails (min −0.0618, max 0.1362); fraction positive is 59.72% for ToM vs 56.84% mid, but the Welch test is tiny and p ≈ 0.9017 so there is no statistically clear mean advantage for the turn days.
Per-month ToM-window vs mid-month returns
| month | tom_window_ret | mid_month_ret | month_total_ret | tom_days_in_window |
|---|---|---|---|---|
| 2023-06 | 0 | 0 | 0 | 0 |
| 2023-07 | -0.0054 | 0.0451 | 0.0395 | 3 |
| 2023-08 | -0.0228 | 0.0081 | -0.0151 | 4 |
| 2023-09 | -0.0062 | -0.0414 | -0.05 | 4 |
| 2023-10 | 0.0037 | -0.0266 | -0.0238 | 4 |
| 2023-11 | 0.0558 | 0.0548 | 0.1092 | 4 |
| 2023-12 | -0.0046 | 0.0554 | 0.0561 | 4 |
| 2024-01 | -0.0367 | 0.0554 | 0.0232 | 4 |
| 2024-02 | 0.0161 | 0.0252 | 0.05 | 4 |
| 2024-03 | 0.0036 | 0.0187 | 0.0116 | 4 |
| 2024-04 | -0.001 | -0.0465 | -0.0469 | 4 |
| 2024-05 | 0.0077 | 0.0355 | 0.0672 | 4 |
| 2024-06 | 0.0299 | 0.0357 | 0.0638 | 4 |
| 2024-07 | 0.0171 | -0.0318 | -0.0092 | 4 |
| 2024-08 | -0.0397 | 0.0771 | 0.002 | 4 |
| 2024-09 | -0.0255 | 0.0621 | 0.025 | 4 |
| 2024-10 | -0.0091 | 0.008 | -0.0044 | 4 |
| 2024-11 | -0.0006 | 0.0341 | 0.0512 | 4 |
| 2024-12 | 0.0323 | -0.0197 | 0.0037 | 4 |
| 2025-01 | 0.02 | -0.0102 | 0.0161 | 4 |
| 2025-02 | 0.0039 | -0.0359 | -0.0216 | 4 |
| 2025-03 | -0.0003 | -0.0661 | -0.0805 | 4 |
| 2025-04 | -0.0362 | 0.0712 | 0.033 | 4 |
| 2025-05 | 0.0236 | 0.0698 | 0.0754 | 4 |
| 2025-06 | 0.0207 | 0.045 | 0.0652 | 4 |
| 2025-07 | 0.015 | 0.014 | 0.0218 | 4 |
| 2025-08 | -0.0274 | 0.0205 | 0.0108 | 4 |
| 2025-09 | -0.0008 | 0.0402 | 0.0522 | 4 |
| 2025-10 | 0.0076 | 0.0433 | 0.0507 | 4 |
| 2025-11 | -0.0172 | -0.0052 | -0.0159 | 4 |
| 2025-12 | 0.0125 | -0.0126 | -0.0065 | 4 |
| 2026-01 | 0.0096 | -0.0058 | 0.0097 | 4 |
| 2026-02 | -0.031 | -0.0048 | -0.024 | 4 |
| 2026-03 | 0.0097 | -0.0519 | -0.0404 | 4 |
| 2026-04 | 0.0557 | 0.1377 | 0.1512 | 4 |
| 2026-05 | 0.0307 | 0.0738 | 0.1049 | 4 |
| 2026-06 | 0.0027 | -0.0017 | -0.0023 | 4 |
The takeaway
No — the last trading day plus the first three of the next month did not capture most of QQQ’s gains over the past 36 months. Those ToM sessions were 144 of 752 trading days (19.2%) but compounded to only 0.127801 (≈12.78%) of the total, whereas the other sessions compounded to 0.796786 (≈79.68%); in log-return terms ToM contributed about 0.17029 (17.03%) of the total. Average daily returns were actually a hair higher outside the turn: 0.000912 (0.0912%) on ToM vs 0.001057 (0.1057%) mid-month, and the gap is statistically nil (Welch t = -0.124, two-sided p = 0.9017 with N=144 ToM days vs N=607 mid-month days). In short, this three-year sample gives no evidence that “buy the turn” captured the bulk of gains — the pattern is mixed month-to-month and not distinguishable from randomness. Practical takeaway: you wouldn’t have caught most of QQQ’s rally by trading only the turn days over this period; if anything, mid-month sessions drove more of the compounded gain.
The fine print
- Window is restricted to the most recent 36 months; longer samples can materially change contribution shares.
- Returns are close-to-close; this blends overnight and intraday moves and doesn’t isolate open-driven flows.
- Welch t-test assumes i.i.d. returns; heavy tails and volatility clustering in equity returns make p-values approximate.