The study store did not answer; this is the study as last read.
FAILA MACD and RSI intraday momentum signal on SPY, QQQ and IWM, held to the close
When the 15-minute MACD trend and an RSI rejection agree, does SPY, QQQ or IWM keep moving that way until the 16:00 close by enough to beat trading costs?
No. Neither the call side nor the put side passed. From 2016 to April 2021, 5,790 call signals averaged -0.014% to the close after costs and 6,433 put signals -0.027%. Before costs the moves were small and positive (+0.036% and +0.024%), and costs of about 0.05% a signal took them below zero. Both failed the lower bound, the year-by-year check, double costs and the deflated Sharpe; the put signals also did worse than the MACD trend alone. The test measures the ETF's own move: trading it with same-day options adds spread and time decay, which can only make it worse. Because nothing passed discovery, the validation and holdout periods were never judged.
Backtested by QS Research
Pre-registration
Written and approved before any data for the counted test was read. On 15-minute regular-hours bars of SPY, QQQ and IWM, a signal is the first bar of a MACD(12,26,9) histogram trend that an RSI(14) rejection confirms. The outcome is the ETF's move from that bar's close to the 16:00 close, net of two times the liquidity-tier fee plus a day of borrow on puts. Signals after 15:55 and on half days are skipped. Calls and puts are two trials, judged at a family of 14 that counts 90 earlier looks at the same signal, on six checks including beating the MACD trend alone. The split is 60/20/20 by date, and the holdout opens only for a trial that passes validation.
Registered 2026-11-05A holdout sample is sealed: its results stay unread until the study is final.
Trials · 2 trials · 0 passed · judged at N = 14 trials; 1868 trials run in total
| Trial | Verdict | |
|---|---|---|
| Calls: upward signals | RESULT | counted |
| Puts: downward signals | RESULT | counted |
| Signals | RESULT | not counted |
| Mean net move | RESULT | not counted |
| Month-clustered lower bound (one-sided α = 0.025) | RESULT | not counted |
| Per-year: positive years / losing years (of 5) | RESULT | not counted |
| Stress mean (costs × 2) | RESULT | not counted |
| Beats the MACD-only baseline (−0.017% calls, −0.022% puts) | RESULT | not counted |
| Deflated Sharpe at N = 14 | RESULT | not counted |
| Verdict | RESULT | not counted |
Limitations
- It measures the ETF's move with ETF costs. Same-day options add spread and time decay, so an option version would do worse.
- The signal had been looked at 90 times before (45 cells, two sides); those looks are counted in the family.
- Half days are skipped, not traded.
- Costs are a model: about 2.5 basis points a side for these ETFs, plus a day of borrow on puts.
- QQQ calls are slightly positive after costs (+0.010%). That is a slice found after the test failed; it is not a tested result.
- The descriptive views are partial: the data ends on 2024-12-31, and one descriptive symbol (IBIT) is not in the pinned data.
The numbers: every trial's checks, metrics, curve and yearly results, and the provenance
3 published versions · current 38960d156eaa8421, published 2026-11-11