Twelve widely known intraday rules on SPY, QQQ and IWM: an exploratory, after-cost check
Does any of the 12 widely known intraday rules on SPY, QQQ and IWM make money after costs, and does a modeled same-day or next-day at-the-money option on the same signal keep any of it?
None of the 12 rules cleared the bar: the study records all 12 as NO EDGE. In the train period (2016-01-04 to 2021-05-25, 1,347 sessions) the mean net return per trade after a 5 bps (0.05%) round trip ran from -0.067% to +0.023% across the 12 rules, and the Holm-adjusted one-sided p was 1.0 for every rule. In the test period (2021-05-26 to 2023-03-10, 449 sessions) the means ran from -0.098% to +0.001%, and no test t reached 2 for the ETF, the 10 bps stress cost or either modeled option overlay (the highest was 0.02); the stress-cost mean was negative for every rule, and the modeled overlays were significantly negative, losing 9.1% to 47.4% of premium same-day and 4.9% to 16.4% next-day. Before costs the means were mostly within a few bps of zero; the largest was +7.3 bps, and the four gross t-statistics beyond +/-2 (three above +2, one below -2) never held on both train and test. The holdout from 2023-03-13 was never read by any rule. This is an exploratory, descriptive result, not a pre-registered test: the rules were written down in the study's own spec before the run, but no pre-registration was approved and the record was written after the results.
The study's own label is exploratory: it was not pre-registered. The rules were written in the study's own spec before the run, but that spec was not approved as a pre-registration, no trial was registered, and the record was written retrospectively on 2026-11-21, after the results. It is published as descriptive: none of the 12 rules cleared the declared bar after costs, and the 12 NO EDGE results mean no rule cleared that bar, not that every rule loses money before costs.
Backtested by QS Research
Pre-registration
No pre-registration existed. The 12 rules, the 5 bps (0.05%) round-trip cost, the 60/20/20 date split, the session-clustered t-statistic, the Holm correction over 12 rules (a multiple-testing correction) and the three verdict bars were written in the study's own spec before the runner existed and clarified before the first run on 2026-11-12; that spec was not approved as a pre-registration and no trial was registered. The record was written retrospectively on 2026-11-21, after every result was seen, and changes none of the rules or numbers. The data is a pinned 1-minute snapshot of SPY, QQQ and IWM from 2016-01-04 to 2024-12-31; every file read was checked against its recorded hash. The 12 rules are opening-range breakouts over 15 and 30 minutes, VWAP reclaim and rejection, gap fade and gap-and-go, first-hour momentum, RSI(2) both sides and long only, first half-hour to last half-hour, late-day continuation, and an unconditional last-half-hour long. Entry is at the signal bar's close; the ETF costs 5 bps (0.05%) round trip, stressed at 10 bps; the three names are pooled into one trial per rule. The split is 60/20/20 by session date, and no minute on or after the holdout's first session (2023-03-13) reached a rule. Each rule also gets a modeled at-the-money same-day and next-day option on the same signal, priced from realized volatility; the overlays are shown beside the verdict and are not counted in the 12-trial family. The bar: EDGE needs a Holm-adjusted train p below 0.05 and a test t of at least 2; UNDERPOWERED needs a positive train mean, a train 95% upper bound of at least +5 bps and a positive test mean; otherwise NO EDGE. The family is 12 trials; the 12 rules are recorded as prior looks for any later pre-registration on them.
Registered 2026-11-21Amendments
- 2026-11-21 The retrospective record was added after every result was seen: no pre-registration existed, and no rule, threshold, cost, split or verdict bar changed. approved: Recorded after the results on independent review, 2026-11-21; no pre-registration existed and no rule changed.
A holdout sample is sealed: its results stay unread until the study is final.
Trials · 12 trials · 0 passed · judged at N = 12 trials; 2965 trials run in total
| Trial | Verdict | |
|---|---|---|
| R1 ORB 15m | FAIL | counted |
| R2 ORB 30m | FAIL | counted |
| R3 VWAP reclaim | FAIL | counted |
| R4 VWAP rejection | FAIL | counted |
| R5 Gap fade | FAIL | counted |
| R6 Gap and go | FAIL | counted |
| R7 First-hour momentum | FAIL | counted |
| R8 RSI(2) both sides | FAIL | counted |
| R9 RSI(2) long only | FAIL | counted |
| R10 First 30m → last 30m | FAIL | counted |
| R11 Late-day continuation | FAIL | counted |
| R12 Last 30m long | FAIL | counted |
Limitations
- No pre-registration. The rules, costs, split and verdict bar were written in the study's own spec and clarified before the run, but no pre-registration was approved, no trial was registered, and the record was written retrospectively on 2026-11-21, after every result was seen. Nothing in this study is a pre-registered test; a later test on any of these rules must count all 12 as prior looks.
- The option results are modeled, not traded. The model prices an at-the-money option with volatility at 1.10 times the prior 20 sessions' realized volatility (floor 8%), a $1 strike grid, a half-spread of the larger of $0.02 and 3% of the model value, and $0.0065 a share each side. It shows how much of a direction call premium and option costs would keep, not what same-day options actually cost; before 2022 SPY did not have a same-day expiry every day. The next-day model gives the option a full session of time value even when the next session closes early (13 half days in the loaded period), so it is slightly generous for trades on the day before those.
- The three index ETFs are pooled into one trial per rule; no result by name is judged.
- Fills are at the signal bar's close. A stop fills at the stop level or at the bar's open if it gapped through; there is no queue, no market impact and no borrow. Prices are raw and unadjusted, so ex-dividend gaps (about four sessions a year a name) are not removed.
- Half days are excluded; 13 fall in the loaded period and the other 5 are in the holdout. A session missing the 15:59 bar is skipped and gives the next session no prior close and a broken volatility window (SPY and IWM: 2019-08-12). A session with fewer than 360 of 390 bars is also skipped (QQQ: 2016-02-22, which has the closing bar), so the three names do not trade exactly the same sessions.
- The holdout from 2023-03-13 was never read by any rule; nothing in the study says when or whether it would be opened.
- The 12 rules are one family of 12 trials. The two modeled option overlays are shown beside the verdict and are not counted in that family, so their numbers carry no multiple-testing correction.
- The result is judged at the study's 12 rules, not against the 2,965 trials QS Research had run when the report was built.
- One window and three index ETFs: train 2016-01-04 to 2021-05-25, test 2021-05-26 to 2023-03-10. It says nothing about other names, other rule variants or the years after 2023-03-10.
The numbers: every trial's checks, metrics, curve and yearly results, and the provenance
2 published versions · current 43c9bffbfa76d507, published 2026-11-21