What does the Wing pricing — SHAPE × VOL predict?

It answers: Every other options number describes at-the-money vol. None prices the trade actually put on — a far-OTM option. Does this name reach 2 standard deviations more often than the wing is priced for? Output: wing_edge per name + one-word verdict — a pricing diagnostic, never a size. Horizon: the near expiry's remaining life (~18-33 sessions on the current panel). Coverage: US names carrying a live iv30 with ≥120 forward windows (126 rows on the 2026-08-31 run).

Do stocks make 2-standard-deviation moves more often than far out-of-the-money options price in?

Wing pricing — SHAPE × VOL (M-007). Short answer: the hub rates it FIT FOR LIMITED USE.

Data as of 6 Nov 2026.

Open M-007 in the live Model Hub All 30 models and verdicts

Verdict: Wing pricing — SHAPE × VOL (M-007) is rated “FIT FOR LIMITED USE” in its validation review dated 13 Aug 2026.
Key result: SHAPE 2.0sd, up wing (pooled): 2.07 [1.75, 2.46].
Key result: SHAPE 2.0sd, down wing (pooled): 1.91 [1.44, 2.46].
How it was tested: training data 2023-08→2026-08, test data not stated; out-of-sample design (how it was checked on data it never saw while being built): none.
Horizon: the near expiry's remaining life (~18-33 sessions on the current panel). Coverage: US names carrying a live iv30 with ≥120 forward windows (126 rows on the 2026-08-31 run).
Forward record: 4 of 23 published lists have reached their 21-session horizon. Average pick return minus SPY's over the same dates: -2.26 percentage points (standard error 1.25). 37% of the matured picks beat SPY over the horizon. Mean rank IC of the matured lists: -0.041.
Stage: Production. Version v1; the hub record is as of 6 Nov 2026.

The verdict

FIT FOR LIMITED USE as a pricing diagnostic. The pooled fat-tail claim is measured to this platform's bootstrap standard and survives the both-wings test; nothing here is a claim that buying wings makes money, and no per-name CI exists.

Development summary: Both wings are fat at 2sd and statistically indistinguishable, so the excess is a distributional property and not 2023-26 drift. Read wing_edge as a pricing diagnostic — dear or cheap versus the tape — never as an expected return or a size.

What the review allows

Reading whether a name's 2sd wing is dear or cheap versus its own tape, alongside the verdict word.
The structure nudge in Options Picks (M-005): outright vs spread at wing_edge 1.3 / 0.8.
The pooled SHAPE table as evidence that far-OTM wings on this panel were underpriced on 2023-26 data.

What the review rules out

Sizing, entering or exiting a trade off wing_edge — it is not an expected return.
Ranking a single name on its own wing_edge as if it carried a confidence interval; only the pooled figures are bootstrapped.
Reading the 1.5sd row as a tail statement — the down wing's CI includes 1, and that gap is drift.
Applying it to deal-pinned names or to drift_driven names, which the engine excludes for a reason.
Treating a European option as paid off by a TOUCH — the barrier statistic is American-style.

What it predicts

Item	Detail
Question it answers	Every other options number describes at-the-money vol. None prices the trade actually put on — a far-OTM option. Does this name reach 2 standard deviations more often than the wing is priced for?
Method	Empirical touch odds × vol
How it works	Empirical P(touch a 2sd barrier within the option's life) ÷ driftless-lognormal P(same), with sd measured at the vol prevailing then (SHAPE), × median realised vol over the life ÷ iv30 (VOL)
Inputs	2sd barrier touch rate (3y), Lognormal touch odds at iv30, Realised vol ÷ iv30 (VOL), Days to near expiry, Open interest & ATM spread
Output	wing_edge per name + one-word verdict — a pricing diagnostic, never a size
Horizon	the near expiry's remaining life (~18-33 sessions on the current panel)
Coverage	US names carrying a live iv30 with ≥120 forward windows (126 rows on the 2026-08-31 run)
Markets	US
Model type	Model · ML/Stat
Stage	Production
Version	v1, first recorded 13 Aug 2026

Validation evidence

Training data: 2023-08→2026-08. Test data: not stated. Out-of-sample design: none.

Headline results from the model's development record, as the hub publishes them.
Measure	Result	Compared with	Note
SHAPE at 2.0sd, up wing	2.07 [1.75, 2.46]	1.00 = lognormal	100,035 observations on 742 dates; moving-block bootstrap, 21-day blocks, clustered by date
SHAPE at 2.0sd, down wing	1.91 [1.44, 2.46]	up wing 2.07	the two wings overlap heavily and both exclude 1 — drift fattens only the upside, vol-clustering fattens both
SHAPE at 1.5sd, down wing	0.97 (CI includes 1)	up wing 1.45, CI excludes 1	the honest reason the 1.5sd row is not read as a tail statement — that gap is drift
Re-run 18 sessions later (2026-08-31)	2.05 [1.73, 2.39]	2.07 [1.75, 2.46] on 2026-08-13	the pooled bootstrap re-computes nightly; the headline is unchanged within noise
Net wing at 2sd (SHAPE × VOL)	≈1.7	1.00 = fairly priced	2.07 × 0.84 — on this sample the fat tail more than paid for the variance risk premium

The hub's standard metrics for this model class (only those the record measured).
Metric	Sample / window	Compared with	Value	Range (CI)
Observations	2023-08→2026-08 · 742 dates	n/a	100,035	n/a

Forward record: 4 of 23 published lists have reached their 21-session horizon. Average pick return minus SPY's over the same dates: -2.26 percentage points (standard error 1.25). 37% of the matured picks beat SPY over the horizon. Mean rank IC of the matured lists: -0.041.

What the validation review found

Is the design sound? Sound and unusually well-posed. The null (driftless GBM, reflection principle) is the model the wing is actually priced off, and normalising each observation by the vol that prevailed then is the correct answer to the regime confound — a naive empirical/implied ratio reports 'vol was higher in 2023', which is exactly the error that had ranked one name at 13×. The decomposition into SHAPE and VOL keeps the two forces readable separately rather than netting them into one opaque number.

Data quality. One panel, no joins, so few failure modes. Survivorship is total (today's listings only). Overlapping windows are acknowledged and handled in the CIs. A 300-bar minimum and MIN_OBS 120 keep thin names out. iv30 is a model IV on the underlying, not a chain-derived surface, and is applied at a dte it was not computed for.

Backtest. None. The measured object is a barrier-touch frequency, not a P&L. Establishing that buying these wings earns money needs historical option chains, which this project only began banking in 2026-08.

Against simple benchmarks. The lognormal itself is the benchmark, which is the right one for a pricing question. No alternative tail model (Student-t, jump-diffusion, empirical bootstrap of the whole path) was fitted for comparison.

Independent replication. The pooled bootstrap re-runs every night, which is a continuous replication. The 2026-08-31 run reproduces the 2026-08-13 headline within noise (up wing 2.05 [1.73, 2.39] vs 2.07 [1.75, 2.46]; down 1.94 vs 1.91).

Sensitivity. Three barriers are shipped and they behave as theory predicts (SHAPE rises with z). The 1.5sd result is the sensitivity check that fails, and the engine reports the failure rather than burying it. No sensitivity to LOOKBACK, MIN_OBS or the 20-day σ_then window has been run.

Monitoring. Pooled CIs and the VOL median print on every run; the verdict mix is visible in the output file. There is no stored history of the pooled figures, so drift over months has to be read from the log rather than from a series.

Limitations

Not a P&L backtest — touching a strike ignores the premium paid, the path and the exit.
One regime, 2023-08 → 2026-08; SHAPE's normalisation is the defence, not a claim about a crisis.
Overlapping forward windows: the nominal observation count flatters the sample, which is why the CIs are date-clustered rather than i.i.d.
Deal stocks break the premise entirely (step function, not a diffusion) and are excluded, not ranked; the rule is a heuristic on collapsed realised vol plus a tight price band.
A single name can carry a fat up-wing and a thin down-wing — a trend wearing a volatility costume. Those are labelled drift_driven and excluded from the ranked list.
The verdict order runs the liquidity gate before the drift test, so an illiquid drift name is labelled untradeable; the outputs exporter re-applies the drift rule so such a name cannot head the list.

Questions

Is the Wing pricing — SHAPE × VOL validated?: Its validation review dated 13 Aug 2026 rated it FIT FOR LIMITED USE as a pricing diagnostic. The pooled fat-tail claim is measured to this platform's bootstrap standard and survives the both-wings test; nothing here is a claim that buying wings makes money, and no per-name CI exists.
What does the Wing pricing — SHAPE × VOL predict?: It answers: Every other options number describes at-the-money vol. None prices the trade actually put on — a far-OTM option. Does this name reach 2 standard deviations more often than the wing is priced for? Output: wing_edge per name + one-word verdict — a pricing diagnostic, never a size. Horizon: the near expiry's remaining life (~18-33 sessions on the current panel). Coverage: US names carrying a live iv30 with ≥120 forward windows (126 rows on the 2026-08-31 run).
How was the Wing pricing — SHAPE × VOL tested?: Training data 2023-08→2026-08, test data not stated, out-of-sample design: none. SHAPE 2.0sd, up wing (pooled): 2.07 [1.75, 2.46]; SHAPE 2.0sd, down wing (pooled): 1.91 [1.44, 2.46]; VOL factor, median: 0.84. Forward record: 4 of 23 published lists have reached their 21-session horizon. Average pick return minus SPY's over the same dates: -2.26 percentage points (standard error 1.25). 37% of the matured picks beat SPY over the horizon. Mean rank IC of the matured lists: -0.041.

Source: the Model Hub's public model register, as of 6 Nov 2026. The M-007 page in the live Model Hub carries the full development and validation record, monitoring and open decisions. All 30 models and their verdicts.

Terms used on this page

Out-of-sample: results measured on data the model never saw while it was built or tuned.
Rank IC: information coefficient: the rank correlation between a model's scores and the returns that followed; 0 means no skill.
Standard error (s.e.): the typical size of the error in an estimate; a result more than 2 standard errors from zero is unlikely to be luck.
Confidence interval: the range that likely contains the true value (95% unless stated); [a, b] on this page is that range.
SPY: the SPDR S&P 500 ETF, the US market benchmark used here.
Implied volatility (IV): the volatility that an option's price implies.