Long-dated momentum options with a take-profit: do the anecdote's rules beat buying the same contracts every month?
Do 270-day at-the-money calls on the 12-month momentum leaders, held about two weeks with a +50% or +100% take-profit, make money after costs on 17 large US names from 2017 to June 2024, and do they beat buying the same contracts on every scored date? Do puts on the weakest names, or a rule that switches side with the market's 200-day trend, do better?
No. None of the four pre-registered rules passed. The primary call rule's mean return after costs was +0.2% in discovery (587 trades) and +3.8% in validation (157), but the discovery and validation lower bounds were -3.9% and -3.3%. The +100% take-profit version was +0.9% and +3.9% (587 and 157 trades), with lower bounds -3.6% and -3.3%. Puts on the weakest names lost 3.7% in discovery (434 trades) and 9.3% in validation (129). The switching rule was -0.1% (631) and +1.3% (162). Buying the same contracts on every scored date did about as well: the whole-window signal-minus-baseline difference was about zero (its lower bound between -6.5% and -7.2%), inside the 13.7% difference the sample could detect, so it does not separate the rules from the baseline. In the 3 x $1,500 simulation of the primary rule, 10x in 63 sessions happened on 0.00-0.01% of paths.
Backtested by QS Research
Pre-registration
Four trials fixed before any option quote was read for a result: calls on the top five 12-month momentum leaders, take-profit at +50% or +100%; the mirror with puts on the five weakest names; and a fourth rule that holds the call list when the market is above its 200-day average and the put list otherwise. Contracts are 0.50-delta, 270-day options bought at the ask, with a 10-session time stop; the judge is every scored name bought on the same contract with no momentum condition. The 17 names, decision dates, entry lists and data snapshot were pinned before the run. Discovery is 2017-2022 and validation is 2023 to June 2024; the holdout from July 2024 is sealed. The family is four trials, so each one-sided lower bound is read at 0.05/4. Three dated amendments (expiry listed at entry, the fourth switching trial, and the data pin) were made before any result.
Registered 2026-11-03Amendments
- 2026-11-03 The expiry must be listed on the entry session; the fetch walks the listed monthlies. No other rule changed. approved: Approved 2026-11-03, before any option quote was read for a result
- 2026-11-04 A fourth trial switches between calls and puts with the market's 200-day trend; the family becomes four. approved: Recorded 2026-11-04, before any option quote was read for a result
- 2026-11-04 Regime timing, the up/down split, the underpowered put cell judged as a whole, and the counted data coverage. No rule changed. approved: Recorded 2026-11-04, before any option quote was read for a result
- 2026-11-06 The fetched path data is pinned; no rule changed. approved: Recorded 2026-11-06, before the counted run
- 2026-11-21 Errata before publication: the first decision date is 2017-01-03, not 2017-01-04, and the split-drop rule was applied from entry to the option's expiry rather than over the hold. No rule changed and no result number changed. approved: Recorded 2026-11-21 with the independent review, after the counted run
A holdout sample is sealed: its results stay unread until the study is final.
Trials · 4 trials · 0 passed · judged at N = 4 trials; 2965 trials run in total
| Trial | Verdict | |
|---|---|---|
| Trial 1 (primary): calls on the momentum leaders, take-profit 1.5x | FAIL | counted |
| Trial 2: calls on the momentum leaders, take-profit 2x | FAIL | counted |
| Trial 3: puts on the laggards, take-profit 1.5x | FAIL | counted |
| Trial 4: calls or puts by the market's 200-day trend, take-profit 1.5x | FAIL | counted |
| Trial 1 registration, charged before any quote was read (not a result) | RESULT | not counted |
| Trial 2 registration, charged before any quote was read (not a result) | RESULT | not counted |
| Trial 3 registration, charged before any quote was read (not a result) | RESULT | not counted |
| Trial 4 registration, charged before any quote was read (not a result) | RESULT | not counted |
Limitations
- The 17 names are today's liquid large caps, so every one survived to 2026 and the decade was mostly rising. The every-name baseline controls for both; the absolute returns do not.
- Contracts are priced at end-of-day bid and ask quotes, not intraday fills: bought at the ask, sold at the bid. A take-profit touched during the day and reversed by the close is not captured.
- The source's 'sell when momentum stalls' is replaced by the 10-session time stop; there is no hedge, no model picking and no position sizing. The pre-registered stall count was not computed; it is descriptive only.
- A trade is dropped when the stock splits any time before the option expires, not only during the 10-session hold. That removed 90 of the 883 call entries, but only 4 had a split inside the hold; about 86 of the 90 were closed before their split, and the effect on the means is untested. The call result is a selected sample and is not a clean read of the rule.
- The baseline comparison is imprecise: the sample could detect only about a 13.7% difference. The point estimates are close, and the put rule's point estimate was slightly above its baseline; the comparison is not measurable at this n.
- The mean itself is not precise enough to rule out a real edge; the sample could reliably detect only a mean of about 6.6% with trades grouped by day, and the pre-registered grouping by month gives a wider interval, which was not computed.
- The pre-registered family-N (0.05/4) lower bounds are a separate computation from the printed gate: the run judged at the count of all trials (1,772 at the run; larger when the report is built) and grouped the mean by entry day, while the 0.05/4 bounds group by entry month. The baseline columns are whole-window at that count. The separate computation's code and output are not published, so those bounds cannot be re-checked from the repository.
- The split-by-side coverage table that the pre-registration required before the fourth trial was made only after the run, and it splits counted from holdout, not discovery from validation.
- The validation stretch is 18 months (157 call trades), and the switching trial has only 10 validation put trades.
- The holdout from July 2024 is sealed; no holdout result is shown.
- The 3 x $1,500 simulation is descriptive: it resamples the counted trades and is not a forecast. 10x in three months was reached on essentially no path.
- No trade-level result file is published, and the counted section is transcribed from the results table; the data pin is carried by the trial-registration rows, not by a machine-readable counted result. The numbers cannot be independently recomputed from the repository.
The numbers: every trial's checks, metrics, curve and yearly results, and the provenance
2 published versions · current 13a2eac48c658f2d, published 2026-11-21