Scheduled events and peer spillover: which option moves are mispriced?
Does the option market systematically under- or over-price the move on scheduled events after costs, and do the peers of a company that reports price a spillover move that does not arrive?
No counted trade trial passed. All five pre-registered trades lost money after costs on the confirmation span. Buying the peer straddle on a large company's earnings night lost 11.5% a trade over 487 trades; the sector ETF straddle lost 14.9% over 23 trades, below the 30-trade minimum; the QQQ straddle lost 10.3% over 128; the event-variance calendar lost 238% of the premium it cost over 421; and the defined-risk short-vol calendar lost 55.8% over 450. Costs are the main economic reason for the peer and sector ETF straddles and the calendars: the median round trip was 9.6% of premium for the peer straddle, 14.8% for the sector ETF straddle and 72% to 292% for the calendars. The QQQ straddle cost only 2.6% round trip; there the move simply did not arrive. The event-variance calendar also fails the skew check and the short-vol calendar also fails the worst-5% stress check. The new finding is on the sell side. On a large company's earnings night, its peers' options priced a 1.95% implied move against a 1.25% realised move (variance ratio 0.31, control 0.61, 1,448 peer sides; 1 means fairly priced), and both the discovery and confirmation spans classified the class as overpriced. The trade structures tested cannot collect it after costs. Single-name earnings were fairly priced (8,166 sides, variance ratio 0.93, control 0.74). No scheduled event class was underpriced.
The overall verdict is fail: 0 of 5 counted trade trials passes. Two classes are recorded as sell-side candidate pockets only; neither has a working trade after costs.
Backtested by QS Research
Pre-registration
Written and frozen on 2026-11-18 before any option quote for the family was read. For each event the study finds one strike where both the first and second expiry have two-sided quotes, solves each expiry's total implied variance from its straddle, reads a base variance per session from the slope between the two expiries, and removes the base rate to leave the event-window variance. It compares that with the realised move; the class statistic is the ratio of summed squared moves to summed implied variance, where 1 is fair. Every class has a control window 21 sessions later. Discovery is the first 60% of each class's events by date; the last 40% is read only after the discovery calls are committed. For the spillover trades the pre-registration fixed the confirmation span as the last 40% of leader events. Counted trials were capped at 13: up to 9 event-class trades (a long straddle, a long calendar, or a defined-risk short-vol calendar) and 4 spillover trades; 5 ran. The bar is at least 30 trades, a positive mean after costs, a positive one-sided lower bound at 0.05 divided by the options family's trial count, a positive signal-minus-control bound, and a positive skew for long premium or a positive worst-5% stress mean for short premium. The family N was 1,868 prior options trials plus the 5 here, 1,873. Three amendments: procedural night lists and a code port; the pinned snapshot and the coverage gap; and a post-hoc own-event check on the spillover peers.
Registered 2026-11-18Amendments
- 2026-11-18 Night lists, extraction code and the split adjustment: FDA decisions, index calendar days, spillover peers and IPOs; the lists and extractor were built before any quote was read. approved: Recorded 2026-11-18, before any option quote was read
- 2026-11-18 The new event-vol snapshot pinned by content hash; the FDA-decision coverage gap stated; the five trade trials computed by the study command rather than the options engine. approved: Recorded 2026-11-18, before any quote in the new snapshot was read
- 2026-11-18 Post-hoc own-event check: split peer sides by whether the peer's own earnings fell inside the option window; reported as a diagnostic, no rule or trial changed. approved: Post hoc, 2026-11-18; no rule or trial changed
Trials · 5 trials · 0 passed · judged at N = 1873 trials; 2965 trials run in total
| Trial | Verdict | |
|---|---|---|
| Long straddle on same-industry peers on a large company's earnings night | FAIL | counted |
| Long straddle on the reporting company's sector ETF | FAIL | counted |
| Long straddle on QQQ on a large company's earnings night | FAIL | counted |
| Long event-variance calendar on the peers | FAIL | counted |
| Short event vol on the peers (defined-risk calendar) | FAIL | counted |
| Elections (SPY/QQQ/IWM pooled) | RESULT | not counted |
| Quarterly expiry / S&P rebalance, SPY | RESULT | not counted |
| Jobs report, SPY | RESULT | not counted |
| Jobs report, QQQ | RESULT | not counted |
| Earnings, ADV > $1B | RESULT | not counted |
| CPI, SPY | RESULT | not counted |
| Spillover onto QQQ (leader earnings) | RESULT | not counted |
| Single-name earnings, all | RESULT | not counted |
| FDA decision dates (PDUFA; coverage gap stated) | RESULT | not counted |
| CPI, QQQ | RESULT | not counted |
| FOMC, IWM | RESULT | not counted |
| Jobs report, IWM | RESULT | not counted |
| Spillover onto the leader's sector ETF | RESULT | not counted |
| CPI, IWM | RESULT | not counted |
| Monthly opex (pooled) | RESULT | not counted |
| Quarterly expiry, QQQ | RESULT | not counted |
| Mega-cap IPO day (QQQ, sector ETF, peers) | RESULT | not counted |
| FOMC, QQQ | RESULT | not counted |
| FOMC, SPY | RESULT | not counted |
| Spillover onto single-name peers | RESULT | not counted |
| Quarterly expiry, IWM | RESULT | not counted |
| Russell reconstitution, IWM | RESULT | not counted |
Limitations
- The pre-registration says each trade trial is recorded in the trial ledger before any quote of its span is read. No ledger record of these five trials is kept with the study, so that step is not evidenced. The five are the trials the pre-registration fixed, and the fail does not depend on when they were recorded.
- The spillover trade confirmation span was taken as the last 40% of the peer, sector ETF and QQQ sides by date, not the last 40% of leader events that the pre-registration fixed. Every trial fails by a wide margin, but the split is not exactly the registered one.
- The short-vol calendar also fails the pre-registered worst-5% stress check (mean with the worst 5% counted twice, -61.0%), and the event-variance calendar also fails the positive-skew check (-9.45). The sector ETF straddle has 23 trades, below the 30-trade minimum. Costs are not the only reason those fail.
- FDA decision dates: 47 symbols (75 of 830 nights, 9.0%) had no price series, and 214 fetched nights were ticker-reuse mismatches. The lost events are complete-response-letter heavy (22% vs 9% of kept), and the fixed window misses early decisions; both biases make the class look more overpriced than it is. It is no edge either way.
- Sides with non-positive implied event variance are dropped (24% of earnings control sides, 19% of spillover signal sides). That keeps the larger-variance sides and pushes the variance ratio down on both signal and control; the signal-minus-control comparison is the fairer read.
- The 547-name universe and the ETFs are survivors; single-name events start in 2021 because the event list does. The study is a prototype outside the signed engine, and the published report does not carry a pinned data snapshot; the pre-registration names three by content hash.
- The family N of 1,873 is a floor: it is the options family's prior count plus the five trials here. A later, higher count only makes the bar stricter, and no trial passed.
- Classes with fewer than 30 discovery events were reported as descriptive rather than classified, although the pre-registration designated only elections, Russell reconstitution and IPOs as descriptive. The other small classes are no edge either way.
- The ranked table shows 22 of the 33 pre-registered cells; the other 11 earnings tier and sector cells are in the result records and none classifies as buy or sell. The table is a selection, not the full ranking.
- The write-up reports round-trip costs for the five trade structures, but not the per-class entry and exit spread table the pre-registration asked for. The cost conclusion rests on the five trial structures.
- The own-event check on the spillover peers is post hoc: 362 peer sides had their own earnings between the window exit and the second expiry, and excluding them leaves 1,086 sides. No class-level summary of that clean set was kept with the study, so its reading rests on its slices only: all four option-cost tiers have a variance ratio below 1 (0.54 to 0.78), and the 10-20% cost tier (n = 284, variance ratio 0.778, t -1.67) does not pass the pocket rule. It is a diagnostic, not a pre-registered trial.
- The event-window split assumes a flat base variance between the two expiries and ignores dividends and early exercise; a noisy or inverted slope can distort the event variance.
- No holdout was opened. The last 40% is a confirmation span read after the discovery calls were committed, not a sealed holdout.
The numbers: every trial's checks, metrics, curve and yearly results, and the provenance
1 published version · current a0751fa30203e08f, published 2026-11-21