Year-out at-the-money calls bought when implied volatility is at most 0.9 times plain realized volatility, re-run with delisted names
Does buying a year-out at-the-money call when its implied volatility is at most 0.9 times the lower of one-year and three-month realized volatility (not adjusted for earnings), and holding it to expiry, still make money after costs when the universe includes names that later delisted, and does it beat buying every year-out at-the-money call?
The year-out rule survives the pre-registered bar, but smaller. Buying a year-out at-the-money call when its implied volatility is at most 0.90 times the lower of one-year and three-month realized volatility (not adjusted for earnings), and holding it to expiry, returned +0.92 per option after costs over 2,419 trades in 97 entry months, down from +1.30 in the earlier run. The one-sided 0.05 lower bound is +0.48, both halves are positive (+1.42 and +0.63), and the rule beat buying every year-out at-the-money call by +0.35 [+0.16]; that baseline made +0.57 on 15,049 trades. The median trade lost 0.18 and 48% of trades made money, so a minority of large winners carries the mean. About 29% of the mean and 38% of the excess over the baseline was survivorship. On the 565 added names that reach a row the rule alone has no edge: 645 trades, mean -0.12, only +0.05 better than every year-out call on those names [lower bound -0.08]. The pooled pass rests on the names in the original snapshot, which still supply 1,774 of the 2,419 trades; scored alone, those names reproduce the earlier run's rule and all six held cells exactly. Of the six held cells, the cheap year-out cell (+1.18) and the rich year-out cell (+0.10) pass; the cheap mid-dated cell (+0.42) is a candidate pocket because its six-cell lower bound is -0.02; the cheap short-dated cell (+0.14), the rich short-dated cell (-0.02) and the rich mid-dated cell (+0.03) show no edge. On the added names every held cell is no edge (cheap year-out -0.45 on 368 trades, rich year-out +0.01 on 299), so the held passes too come from the original snapshot. 58 of the earlier run's 66 candidate pockets survive, including all 22 rich-leg pockets; the 8 that drop are cheap-leg. The delisting rule closed 100 scored trades at the last close (82 year-out, 11 mid-dated, 7 short-dated); three more exit attempts were refused for a split. The run charged 465 cells in the trial record; the seven pre-registered cells are the judged family.
The pass is the year-out rule's. Only three of the seven judged cells pass: the rule, the held cheap year-out cell and the held rich year-out cell. The held cheap mid-dated cell is a candidate pocket because its six-cell lower bound is below zero, and the cheap short-dated cell, the rich short-dated cell and the rich mid-dated cell show no edge. The rule was judged as one trial; the six held cells were judged as a family of six. The run charged 465 cells in the trial record; the seven pre-registered cells are the judged family.
Backtested by QS Research
Pre-registration
Written on 2026-11-18 before any added option chain was fetched or any row scored, with one amendment before the run and one earlier pre-run correction. The earlier run's rules are kept unchanged: monthly entries on the first session of the month from February 2016 to November 2026; the prior session's 400 largest US stocks by dollar volume with a close of at least $5; near, mid and year-out expiries; at-the-money and 25-delta legs chosen by skew; a 10% spread cap; earnings-excluded realized volatility; buys at the ask and sells at the bid with $0.0065 a share; held to intrinsic; a credit vertical on maximum loss for rich sells; a moving-block bootstrap of 10,000 draws with blocks of 2, 5 and 13 months and seed 20261118; halves split at January 2021. The re-run adds the names that later delisted: the same point-in-time universe over 130 entry dates, with every eligible name-month the original snapshot lacked fetched into one research-only snapshot. A name whose price series ends more than 30 calendar days before 2026-11-13 exits a still-open trade at its last close; the horizon stays 2026-11-13. The year-out rule is one trial on the plain (not earnings-excluded) realized read, the six held cells are a family of six on the earlier run's earnings-excluded reads, and the earlier run's 66 candidate pockets are re-read as the same slice trials. The year-out rule survives if its cell works (mean positive, one-sided 0.05 lower bound positive, at least 300 trades on at least 36 entry months, both halves positive) and it beats every year-out at-the-money call with a positive paired lower bound; it is weakened if its excess over the baseline is below half the earlier run's. The original snapshot's names must reproduce the earlier run exactly. The run charged 465 cells in the trial record; only the seven pre-registered cells were judged.
Registered 2026-11-18Amendments
- 2026-11-18 Before any added option chain was fetched: the delisted-name gap widened from 7 to 30 calendar days, and a check was added that the original snapshot's names must reproduce the earlier run's rule and held cells exactly. No result had been read. approved: Recorded before the fetch; no result had been read
- 2026-11-18 The option snapshot was pinned and its coverage counts recorded before the run: 1,846 names, 10,687 of 11,283 added name-months with a chain, 43,971 of 45,431 eligible name-months covered. No quote was read for a result. approved: Recorded before the counted run; no quote was read for a result
Trials · 7 trials · 3 passed · judged at N = 7 trials; 2965 trials run in total
| Trial | Verdict | |
|---|---|---|
| Year-out ATM call when IV ≤ 0.90× the lower realized vol, held | PASS | counted |
| Held cheap buy, short-dated | FAIL | counted |
| Held cheap buy, mid-dated | FAIL | counted |
| Held cheap buy, year-out | PASS | counted |
| Held rich sell (credit vertical), short-dated | FAIL | counted |
| Held rich sell (credit vertical), mid-dated | FAIL | counted |
| Held rich sell (credit vertical), year-out | PASS | counted |
Limitations
- The remaining gaps are not random. 3.2% of eligible name-months still have no option chain: 1.9% because there is no price series under the ticker of the day (mostly renamed companies and a few delisted names) and 1.3% because the chain had no monthly quotes. Some large names are missing their older months under an old ticker. The residual can still flatter or penalise the result.
- End-of-day bid and ask quotes, not fills: buys pay the ask and sells receive the bid, with $0.0065 a share commission. Real fills and size are not tested.
- One option data source, monthly entries only, one decade (2016-2026), one mostly rising market. The every-month baseline itself made +0.57 per option.
- Held results are direction, not volatility pricing. The version that hedges the stock exposure shows no edge in any of its six cells, so the held pass is not evidence that the implied-volatility read is right.
- Most picks lose: the median year-out trade lost 0.18 and only 48% made money; a minority of large winners carries the mean.
- The added names alone show no edge (645 trades, mean -0.12), and every held cell is no edge there too (cheap year-out -0.45, rich year-out +0.01). The pooled pass rests on the names in the original snapshot (1,774 of 2,419 trades), so it is a statement about adding delisted names to a survivor list, not a rule that works across the added names.
- The delisting rule values a still-open trade at intrinsic on the last close: a cash takeover settles near the deal price, a collapse near its last print. Deal terms, the delisting return and the path between the last close and the delisting are not modelled. 100 scored trades exited this way; three more exit attempts were refused for a split.
- A ticker reused by a later company could carry the wrong price series. The pre-registration says a strike-bracket guard is applied, but the counted code path does not implement it; a reused ticker with a plausible price is not caught, and how many chains would fail the guard is not measured.
- The year-out rule is a third look: it passed in two earlier runs. This run judges it as one trial at 0.05, not corrected for those earlier looks; the 465 charged cells are not the judged family. The six held cells are judged as a family of six.
- The 58 surviving pockets are the earlier run's own slices re-read on a superset that includes the same rows. They are not an out-of-sample confirmation; the original-snapshot half is identical, so the exercise measures dilution by the added names, not fresh evidence. One rich mid-dated slice rose slightly; the rest fell. One new pocket appears only in this run (rich short-dated sells when the VIX is 20 to 30) and is a lead, not a survivor.
- The names added to the universe with no matching company filing are treated as non-filers: the plain realized volatility read and a combined ETF/non-filer sector. This touches the cheap ratio of the held cells and the sector slices, but not the year-out rule, which uses the plain read for every name. About 30% of the added rows are in that bucket. The sector slices use the company reference panel's SIC division, not its market-capitalisation field.
- A volatility-rank measure in the study's other tests needs 24 prior monthly snapshots; the added names carry only their eligible months, so it is undefined for most of them. The other pre-registered tests were computed and charged but not judged here.
- No holdout: every year from 2016 to 2026 was read. The original-snapshot half reproduces the earlier run exactly, which is a consistency check, not an out-of-sample test.
- The reproduction check is a comparison of the two result files, not a hard stop in the code; a future re-run would not fail closed on a mismatch.
The numbers: every trial's checks, metrics, curve and yearly results, and the provenance
1 published version · current e62c7e364f13730e, published 2026-11-21