Options History Analytics

Every other tab in the Options Viewer reads a single point-in-time chain. The History tab is the exception: it reads accumulated daily history, which is what makes it the only place the viewer can answer how does today compare to normal?

That distinction matters more than it sounds. The Volatility tab's own methodology notes that the informative signal in a risk reversal is its level against its own history, not its sign — a statement a single snapshot structurally cannot act on. Everything below exists to close that gap.

Why this needs an ETL rather than a live query

An option chain snapshot is not reproducible from the live feed. A FRED series can be re-fetched for any historical date, which is why most data on this platform tolerates a missed run. A chain quoted at the close on a given session ceases to exist when the session ends — Cboe publishes current state, not an archive.

A missed capture is therefore a hole no re-run can fill, not a delayed row. That single property is why the capture runs on a cloud scheduler after each close rather than on demand. The gap is not strictly irrecoverable — a data vendor that recorded OPRA at the time can serve historical chains — but recovering it costs money and effort that capturing it live does not.

Percentile ranks

Each headline metric — normalised skew, 25Δ risk reversal, 25Δ butterfly, ATM implied vol, net gamma exposure, put/call open interest ratio — is ranked against its own stored history.

This is what converts "risk reversal is +3.89 vol points" into "risk reversal sits in the 88th percentile of where this surface normally trades". The first number is uninterpretable without knowing the surface's habits; the second is the actual signal.

Ranks are withheld below 20 stored sessions. A percentile computed against four observations looks precise and carries no information, which is worse than showing nothing — so the tab reports how far along the series is instead. Raw values render correctly from the first session; only the ranks wait.

Skew Stickiness Ratio

How the surface actually moved, measured against how its own skew said it would:

$$\text{SSR} = \frac{\Delta \sigma_{\text{ATM}} \,/\, \Delta \ln S}{\partial \sigma / \partial \ln K \big|_{\text{ATM}}}$$

The numerator is the realised move in at-the-money implied vol per unit of log-spot. The denominator is the at-the-money skew that was already in place before the move — fit by least squares over out-of-the-money quotes within ±5% of spot. Both sides are vol points per log-unit, so the ratio is dimensionless.

Three choices in that slope fit are deliberate. The band is tight because the smile is curved, so a wide window measures curvature as much as slope when the quantity wanted is the derivative at the money. Only out-of-the-money quotes are used, since on each wing those are the tradeable side of the spread. And it is fit against log-moneyness rather than strike, which keeps the measure scale-free as the index level drifts over years and dimensionally matches the $\Delta \ln S$ it is divided into.

Why it matters beyond classification. The Gamma tab's zero-gamma flip level is computed by repricing every contract at candidate spot levels while holding implied vol constant — the sticky-strike assumption, stated explicitly in that tab's methodology. SSR is the only measurement on the platform that reports whether that assumption actually held on a given day. A reading far from 1 means the published flip level rests on shakier ground than its precision suggests.

SSR is unavailable on the first stored session, and on any day the index moved less than roughly 0.1%. In that second case the ratio is dividing by a near-zero denominator, which turns ordinary quote noise in ATM vol into an enormous and entirely meaningless number.

Convention warning. SSR normalisations differ across the literature — some scale so that sticky delta reads 1 and sticky strike reads 2. The definition above is this platform's. Check which convention a source uses before comparing its figures against this one. This is the same discipline the risk reversal sign convention requires elsewhere in the wiki.

Open interest flow

Volume says how much traded. It cannot say whether that trading opened or closed positions — and those are opposite signals wearing an identical signature. Rising open interest on heavy volume is conviction being built; falling open interest on the same volume is an unwind or a short cover.

Separating them requires differencing two sessions of stored chain, which is why no single snapshot can produce this read no matter how carefully it is analysed.

Each side is classified by its open-interest change relative to that day's own volume:

The threshold is a ratio rather than an absolute contract count on purpose: 10,000 contracts of open-interest change means something completely different on a 2,000-lot day than on a 2,000,000-lot one.

What "differencing two sessions" quietly assumes

That both sessions describe the same book. They do not, automatically. SPX lists new expiration cycles continuously, so a cycle can appear in today's stored chain having never been stored before — and a contract with no prior row cannot be differenced at all.

Treating that absence as zero prior open interest is the trap: it books the cycle's entire resting position as opened today, which is indistinguishable from real conviction and large enough to dominate the total. The pipeline handles this from both ends. A cycle, once stored, is kept until it expires, so the set only ever grows by genuinely new listings. And the change itself is measured only over contracts present in both sessions, with the coverage reported alongside the reading — if it drops materially below 100%, the panel says so rather than quietly reporting a partial book as a whole one.

This is the unglamorous half of any day-over-day signal: the arithmetic is trivial, and almost all the difficulty is in guaranteeing the two things being subtracted are actually comparable.

Skew versus price divergence

Over a trailing 20-session window, the joint direction of price and normalised skew:

The window is roughly one trading month: long enough for a trend to establish, short enough that the current regime has not been averaged away.

This is the sharpest thing stored history buys. The Volatility tab's morphology classifier can report that skew is flat — a level. Only this can report that skew is flattening while price rises — a transition. The transition is the signal; the level is only the state it arrived at.

Limitations

• The series cannot be backfilled from the live feed. History begins when capture began; closing an earlier gap requires a paid historical-chain vendor.
• Percentile ranks need 20 sessions; open-interest flow and SSR need 2. The tab degrades to a progress state rather than an error, but the reads are genuinely unavailable until then.
• Open interest publishes once daily, so ΔOI is strictly session-over-session and cannot resolve intraday positioning changes.
• One sample per session. Intraday skew dynamics are invisible; only session-to-session moves are captured.
• SSR uses a platform-specific normalisation and is not directly comparable to published figures scaled to a different convention.
• The per-strike history is a slice, not the whole chain — key expirations within ±25% of spot. Deep-wing per-strike history exists only in the raw archive, not in the queryable tables.
• Percentile ranks are only as representative as the window they cover. A young series ranks against a narrow slice of market conditions; a reading in the 90th percentile of two months is not the same claim as the 90th percentile of five years.

Related Reading

• Volatility Surface Analytics — the smile, skew and RND specs these metrics build on
• GEX Methodology — the gamma flip level whose sticky-strike assumption SSR tests
• Volume & Open Interest Analysis — what OI and volume each measure
• Volatility Smile & Skew Explained — the underlying skew concepts