Out-of-the-money option wings bought before FDA PDUFA decisions
If you buy the 0.20-delta call, the 0.20-delta put, or both three sessions before a PDUFA date and sell after the decision, does the premium pay after costs?
No. All three pre-registered structures lost money after costs. The 0.20-delta strangle lost 50.0% of premium on average over 721 events (median -97.1%; one-sided 95% lower bound -56.5%); the 0.20-delta call wing lost 55.1% over 726 (median -100%; lower bound -66.4%); the 0.20-delta put wing lost 46.4% over 734 (median -100%; lower bound -54.5%). Skew was positive (4.62, 7.21, 5.46) and every structure had at least 30 tradeable events, but the mean and its lower bound were far below zero, so all three fail. The descriptive core check points the same way: on the 753 events with the needed quotes, the at-the-money straddle's median implied move at entry was 14.7% against a median realised move of 7.2%, and realised exceeded implied on only 26.4% of them. Options priced a bigger move than the decision delivered, even for complete response letters (median implied 28.8%, realised 18.4%), so buying these wings was not cheap.
Backtested by QS Research
Pre-registration
Written before any event list or option or stock price was read, and amended once before any result. Events are US-listed common stocks with a PDUFA date announced at least five sessions ahead, from January 2018; the pre-registration called for one row per announcement, with a trade using the latest PDUFA date known at entry, but the frozen list, committed at 830 rows before pricing, is one row per ticker and PDUFA date and each trade uses the date in its own row (see limitations). Three structures are each a trial: buy the 0.20-delta call and put (strangle), the call only, or the put only, three sessions before the date at the ask. The expiry is the first listed expiry at least two sessions after the date. Exit is at the bid on the first close after the decision filing, or at intrinsic if no decision is found by expiry. Costs are $0.0065 a share on each fill; returns are per $1 of premium. A structure passes only if the mean return after costs is above zero, the one-sided 95% lower bound of a 10,000-draw bootstrap clustered by PDUFA month is above zero, skew is positive and there are at least 30 tradeable events. Three trials are counted; there is no holdout. Amendment 1 added a delta diagnostic and two exploratory variants with no pass verdict.
Registered 2026-10-29Amendments
- 2026-10-29 Added a delta diagnostic and two exploratory variants (bid-zero contracts admitted as candidates, and only legs with delta 0.10-0.30), reported beside the test with no pass verdict. The pre-registered test was unchanged. Recorded after one event's quotes had been read to build the parser and before any result. approved: Recorded before any result and reviewed
Trials · 3 trials · 0 passed · judged at N = 3 trials; 2965 trials run in total
| Trial | Verdict | |
|---|---|---|
| (a) long strangle (~0.20-delta call AND put) | FAIL | counted |
| (b) call wing only (~0.20-delta call) | FAIL | counted |
| (c) put wing only (~0.20-delta put) | FAIL | counted |
Limitations
- Entry is at the ask and exit at the bid on end-of-day quotes, with $0.0065 a share on each fill (no exit commission at expiry): real and large costs in thin small-cap option chains.
- 43 of the 830 events have no option data (18 where the source returned no quotes, 17 with no listed option root, 8 with no listed expiry late enough) and are counted as not tradeable, not dropped. Per structure, 37-53 more events have no contract meeting the rules, 4 have no stock close at entry and 1 no exit quote; 8-11 trades are dropped for a stock split.
- Exit timing comes from SEC 8-K filings: 39.0% of events have no decision filing that can be tied to the drug, a further 1.6% are ambiguous, and some filings lag the news by months. Those trades hold to expiry at intrinsic, so a trader on the wire could exit earlier; the study does not measure how much that would change the result.
- 792 candidate rows were excluded before the list was frozen: 487 with no day-level PDUFA date, 138 with fewer than five sessions of notice (the announcement session counts as one), 52 after the cutoff, 49 with no ticker, 31 before 2018, 30 with an ambiguous ticker and 5 not US-listed. Some reasons count filings and others ticker-date pairs, so the total is not a count of distinct events.
- Three events with PDUFA dates on 17 and 26 October 2026 have an expiry after the data end on 29 October 2026. They are valued at the last close before expiry, not at expiry, so those nine structure rows are censored at the data end, not complete outcomes; they stay in the counted sample, and the fail does not depend on them.
- The frozen event list is one row per ticker and PDUFA date, not one row per announcement as pre-registered: 525 of 830 rows merge more than one filing (1,877 filing mentions among them). A trade also uses the date in its own row, not a later moved date known at entry as the pre-registration required.
- Ticker resolution is as of the filing for 808 of the 830 rows; 22 use the issuer's current ticker because the filing text and cover page did not give a listed symbol. Some of those 22 have data missing under the current ticker: DMKPQ (then ADMP) has no option root, and two of the four 2018 BHC rows and SEEL (then APRI) have no contract.
- Returns can be below -100% of premium: when the exit bid is zero, the $0.0065 exit commission is still charged (185 of the 2,181 tradeable trades, minimum -106.1%). A holder would more likely let a worthless option lapse, so the loss on those trades is overstated; all three structures still fail by a wide margin.
- Only the event list is pinned by content hash. The option quotes, stock closes, decision filings and split table are not pinned in the report, and no release commit is recorded, so the run cannot be reproduced from the published provenance alone.
- The three structures are not independent (the call and put are the legs of the strangle), and no correction across the three trials is applied; all three fail by a wide margin.
The numbers: every trial's checks, metrics, curve and yearly results, and the provenance
1 published version · current 2a3f56025454d8c7, published 2026-11-21