Cheap far-out-of-the-money options on 20 large US names: calls, puts and long-dated quiet-volatility buys
Does buying cheap out-of-the-money options pay after costs on 20 large US names, and do 12-24-month calls bought when a name's 60-day realised volatility is in its bottom decile do better?
Nine of the ten pre-registered trials failed. The one pass is calls 10% out of the money, 60-120 days, bought on the first trading day of each month from January 2019 to June 2025 on 20 large US names: 1,414 trades, +50.1% mean return on premium after costs, 99.5% band [+11.7%, +95.7%], 27% of trades profitable. That is near-the-money, not the 'cheap far out of the money' idea: calls 30% and 50% out of the money averaged +11.0% and -24.5%, puts lost 9.9% to 27.3%, and every one of those bands includes zero, so each fails on its lower bound. The top tenth of the 10%-out-of-the-money trades produced 1.74 times the total return, so the other nine-tenths lost money. The long-dated quiet-volatility calls (12-24 months, only when 60-day realised volatility was in the name's bottom decile) averaged +120.8% at 25% out of the money and +166.5% at 50%, on 34 trades over 16 entry dates, with bands that include zero; 9 of the 34 trades in each of the two long-dated call tests were still open when the data ended and were marked at their last quoted bid, not settled, and those marks include the largest returns. The partial put test had 3 trades, fewer than the 30 the bar needs, and lost 93.3%. The pass is judged at the ten pre-registered trials, not at the 2,965 trials QS Research had run when the report was built; at that count the same lower bound is -8.4%. The 20 names are survivors picked in 2026, and the window is mostly a bull market, so the 10% pass is leveraged long beta, not a trading conclusion.
One of the ten pre-registered trials passed: calls 10% out of the money, 60-120 days, on 1,414 trades from January 2019 to June 2025, at +50.1% mean return on premium with a 99.5% band of [+11.7%, +95.7%]. The study itself reports that as near-the-money leveraged long beta on a survivor list in a bull market, not a trading conclusion. The question was about cheap far-out-of-the-money options: calls 30% and 50% out of the money and puts 10-30% out of the money all failed, as did both long-dated quiet-volatility call tests and the partial put test. The headline verdict is fail.
Backtested by QS Research
Pre-registration
Ten trials were fixed before any option quote was read. Seven are monthly entries on the first trading day of January 2019 to June 2025, held to expiry: calls 10%, 20%, 30% and 50% out of the money and puts 10%, 20% and 30% out of the money, 60-120 days, bought at the ask plus $0.0065 a share, with asks under $0.05 skipped. Two are long-dated quiet-volatility call tests: 12-24-month calls 25% and 50% out of the money, entered only when the name's 60-day realised volatility was in the bottom tenth of its trailing three years, from 2016, held to expiry; the first such entry is August 2018, because the percentile needs three years of history. The tenth is a partial version of the same idea on puts 30% out of the money, only on names up at least 50% over the previous 12 months, with the valuation filter left out. Each trial is judged on at least 30 trades, a positive mean after costs, a positive one-sided bootstrap lower bound at 0.05/10 resampling entry dates, and positive skew. The universe is the 20 large US names still listed in 2026: a survivor list, which flatters calls, stated before the run.
Registered 2026-10-29Trials · 10 trials · 1 passed · judged at N = 10 trials; 2965 trials run in total
| Trial | Verdict | |
|---|---|---|
| quick-c10 | PASS | counted |
| quick-c20 | FAIL | counted |
| quick-c30 | FAIL | counted |
| quick-c50 | FAIL | counted |
| quick-p10 | FAIL | counted |
| quick-p20 | FAIL | counted |
| quick-p30 | FAIL | counted |
| r3-25 | FAIL | counted |
| r3-50 | FAIL | counted |
| r3b-partial | FAIL | counted |
Limitations
- Twenty large US names still listed in 2026, picked in 2026: a survivor list that flatters calls. The window is mostly a bull market, and the 10%-out-of-the-money pass is near-the-money long beta, not a trading conclusion.
- The one pass is judged at the ten pre-registered trials, not at the 2,965 trials QS Research had run when the report was built; at that count the same bootstrap's lower bound is -8.4%. It is also thin: without NVDA, TSLA and AMD the mean is +36.1% and the same 99.5% lower bound is +0.9%; on the three index ETFs alone the mean is +30.7% but the lower bound is -42.3% (QQQ +117.2%, SPY -24.0%, IWM +0.4%), so the breadth is one ETF.
- 9 of the 34 25%-out-of-the-money quiet-volatility trades, 9 of the 34 50%-out-of-the-money trades, and 1 of the 3 partial put trades were still open when the data ended. They were marked at their last quoted bid on 2026-10-01, not at expiry, and those marks include the largest returns (CAT +1,073% at 25% out of the money and +2,067% at 50%). Valued at zero instead, the means are +45.0% and +68.6% (and -94.2% for the partial put test); no verdict changes.
- Held-to-expiry trades are worth intrinsic value at the unadjusted close; exercise and settlement frictions are not modelled. The open marks above are the exception to 'held to expiry'.
- The engine takes the nearest quoted strike to the requested distance, with no cap. For the 50%-out-of-the-money calls, 119 of 1,076 trades are less than 35% out of the money and 52 less than 25%, so that row is not uniformly at 50%; the 10%-out-of-the-money pass is not materially affected (1 trade under 7%).
- Snapshots are monthly and cover entry days only: only hold-to-expiry is tested, and take-profit and time-stop exits were not fetched. META has 58 snapshots because before mid-2022 the ticker belonged to another issuer. No earnings adjustment.
- No pinned data snapshot and no recorded binary or release: the counted run cannot be replayed byte-for-byte. It was fetched before the fix that makes a failed name fail the run, so a fetch failure would look like an ordinary no-quote skip.
- The comparison with the earlier quick test on a second data source is descriptive only: that run committed no result file, so its numbers cannot be re-derived. The two sources agree within 5.4 points on six of the seven structures and differ by 12.4 points on the 30%-out-of-the-money puts (-9.9% here, -22.3% there).
- The engine accepts an entry ask up to one trading session old, while the pre-registration says the entry day's close quote. The result files do not record the quote's age, so how many entries used a one-session-old ask is not known.
- The pre-registration's prose does not state the partial put test's contract (12-24-month puts 30% out of the money on names up at least 50% over 12 months, with the quiet-volatility trigger); the spec file committed with it does. The full version with its valuation filter was not run.
- Skipped entries are counted in the result files but not shown on the page: 1-339 asks under $0.05 per call trial and 0-25 per put trial, 95 no-quote entries per quick trial, 29 no-expiry, 5 chain/spot mismatches and 16 splits; the long-dated tests add 2 chain/spot and 4 splits, and the partial put test adds 32 no-quote, 2 splits and 126 name-months below the 50% 12-month return.
The numbers: every trial's checks, metrics, curve and yearly results, and the provenance
1 published version · current e3f7fce408ce8e00, published 2026-11-21