A long-dated call held alone versus the same call with a one-month call sold against it every month
Does selling a one-month call against a year-out, deep in-the-money call every month beat simply holding that long call, after real bid/ask quotes and commissions?
No. Selling a one-month call against a held LEAPS (a long-dated call) did worse than holding it in every test. Across 1,897 trades on 20 large, liquid US names (17 stocks and 3 index ETFs), entered from 2016 to 2025 and held about a year, the overlay gave back 29.2% of the premium paid, with a recorded 98.75% interval of -33.8% to -24.2%. The subset where the short-expiry at-the-money implied volatility was above trailing realized volatility at entry lost 28.1% (-34.4% to -22.2%). The other two pre-registered variants lost 19.8% and 38.2%, and the two extra subsets those runs computed lost 19.1% and 36.8%; all six trials fail. The overlay helped only when the stock rose little or fell (+2.2% in the bottom third of stock returns) and hurt most when it rose (-69.7% in the top third). It lost in 9 of the 10 entry years; the only positive year was 2021 (+8.4%). Holding the LEAPS alone averaged +80.1% (median +54.9%), but that is 20 large names in a strong decade, not evidence of an edge.
Backtested by QS Research
Pre-registration
Written on 2026-10-31, before any code or priced trade. One trade per name per monthly snapshot: buy the listed call nearest 0.80 delta with 300 to 450 days to expiry at the ask, hold it to expiry, and compare it with the same call plus a one-month call sold at the 0.30-delta strike at the bid on the entry date and again each month after the previous short expires. Each leg pays $0.65 a contract; expiries settle at intrinsic on the closing price; returns are measured on the long call's premium. The tests fixed in advance were the all-trade mean and the subset where the short-expiry at-the-money implied volatility was above trailing 60-session realized volatility at entry, plus the same all-trade mean at 0.80/0.20 and 0.70/0.30. A test passes only if the lower bound of its 98.75% bootstrap interval, clustered by entry date with 2,000 resamples, is above zero. No holdout was reserved. A post-run erratum records that six trials were charged, not four (each variant also computed the implied-volatility subset), and that all six count; the stricter six-test bar is 99.17%. The recorded intervals are the original four-test 98.75% bounds.
Registered 2026-10-31Amendments
- 2026-11-03 Post-run correction of three statements the run disproved: overlapping holds make the date-clustered interval too narrow, six trials were run and recorded rather than four, and the result covers only the 20 liquid names with long-dated chains. The pre-registered rules are unchanged. approved: Recorded after the run; the pre-registered rules are unchanged
- 2026-11-21 Independent review before publication: the calendar-year block interval was withdrawn as unreproducible, and the four-test interval level, the six counted trials and the unrecorded skip counts are disclosed. The pre-registered rules are unchanged. approved: Recorded on independent review; the pre-registered rules are unchanged
Trials · 6 trials · 0 passed · judged at N = 6 trials; 2965 trials run in total
| Trial | Verdict | |
|---|---|---|
| T1: all trades | FAIL | counted |
| T2: short IV > trailing RV | FAIL | counted |
| 0.80 long / 0.20 short: all trades | FAIL | counted |
| 0.80 long / 0.20 short: implied vol above trailing realized vol | FAIL | counted |
| 0.70 long / 0.30 short: all trades | FAIL | counted |
| 0.70 long / 0.30 short: implied vol above trailing realized vol | FAIL | counted |
Limitations
- Only 20 large, liquid US names (17 stocks and 3 index ETFs) had long-dated option chains in the archive; the other names in the archive produced no trades. The 20 are today's list, so the result carries survivorship, and the span is a large-cap bull decade. No holdout was reserved, so the whole 2016-2025 span is in-sample.
- End-of-day bid/ask quotes, not fills: the long call is bought at the ask and each short sold at the bid, with $0.65 a contract a leg. A real fill could be worse.
- Entries are monthly and every position is held to expiry: no early exit, no rolling, no early assignment beyond settling at intrinsic, and no interest on the cash from short sales.
- One-year holds entered monthly overlap by about 11 months, so the intervals published here (clustered by entry date) are too narrow. An earlier draft reported a wider calendar-year block interval; it was withdrawn because no committed code or result produces it. The reading rests on every point estimate being -19% or worse and every upper bound below -15%.
- Six trials count, but the published intervals were computed at the four-test 98.75% level. The stricter six-test 99.17% bounds are slightly wider and were not recomputed; every upper bound is far below zero.
- The run is not tied to a pinned data snapshot or a recorded release, and the committed results keep summaries only, not the per-trade rows. The numbers cannot be re-run or independently verified from the published record.
- The three variant runs report different counts of name-month entries skipped for having no long-dated expiry (7,143, 7,153 and 7,173), so they were not all read from one archive state. The 1,897-trade set is the same in all three.
- Months where no short call could be sold are skipped; the results do not total them, although the pre-registration says they are counted. Skipping a short moves the overlay toward simply holding the long call. The split of the 966 trades outside the implied-volatility subset (volatility condition not met versus no usable short at entry) is also not reported.
- The backtest picks the nearest expiry with a quote and the nearest two-sided at-the-money call; if the nearest listed contract was unquoted, a slightly different contract was traded. How often this happened is not recorded.
The numbers: every trial's checks, metrics, curve and yearly results, and the provenance
2 published versions · current b9438fb3b772c7fd, published 2026-11-21