VIX — by volume, RSI, OBV and S&P divergence
Four daily readings of fear on one shared time axis: how high it is and how much is trading behind it (VIX option volume, VXX shares), whether it is overbought or oversold against its own history (RSI, percentile-triggered), whether volume confirms the move (VXX on-balance volume — the VIX index itself has no volume, so the short-term VIX-futures ETP stands in), and whether it is decoupling from the S&P 500 (normalized overlay, 20-day log-return correlation against its own history, descriptive swing breaks). Every number is shown against its own history with the sample size; nothing here predicts.
P1 · level + activity
P2 · momentum vs own history
P3 · volume confirmation
P4 · vs the S&P
Which of these readings is unusual right now, and how unusual?
every reading against its own history · n shown on every percentile · divergences within the 60-session live span · ranked by class, then distance from the middle, then recency — no composite score| # | Whatthe reading, in plain words | Vs whatthe history or the pair of swings it is measured against | Valuein its own unit | nobservations actually used / target (504 = two years); N/A under 60 | Agepivot date · confirmed date (sessions ago) | Own-history percentilep50 = ordinary · dot = the live bar · last 60 sessions | Readingwhat was observed — from the copy table, never a forecast |
|---|---|---|---|---|---|---|---|
| loading… | |||||||
Swing breaks between the S&P 500 and the VIX (descriptive, unranked)
the house swing detector pairs two same-type S&P pivots inside 60 bars and reads the VIX at those two dates — an observation about the pair, nothing more- loading…
Machine-readable ASCII — the numbers behind every chart and the table, as fixed-width text
loading…
How was this measured, and what are its limits?
The VIX is an index computed from S&P 500 option prices; nothing trades "the VIX". Two proxies carry the activity read: VIX option volume (contracts on the index itself, OPRA, shipped from the Windows archive — a bounded extract, hatched outside its coverage) and VXX shares (the short-term VIX-futures ETP, Yahoo consolidated, from 2018-01-25). The OBV on P3 is built from VXX only.
A2.rsi(VIX close, 14). The extreme trigger is where today's RSI sits among its own trailing 504 sessions (≥ p90 / ≤ p10, self-inclusive, n printed). The textbook 70/30 lines are drawn second, for reference only.
OBV − SMA50(OBV), divided by the rolling 252-session sigma of that spread, then ranked against its own trailing 504 sessions. The stretch is masked until 50 + 252 valid VXX sessions exist; the engine's own early sigma is never shown.
The house detector (OBVX.detect → A2.divergences, pivot widths 2/3/5/8/13, one closed bar confirms). Price vs oscillator only: VXX close vs VXX OBV, VIX close vs VIX RSI. Strength is normalised by the oscillator's range over the slice scanned, so chart lines use the visible slice (+60 bars) and the table uses a fixed 312-session tail. "Live" = confirmed within the last 60 closed sessions. A partial current session is cut from every detector input.
Pearson on 20 paired daily log returns (VIX vs ^GSPC), ranked against its own trailing 504 readings excluding the current one; paired n and reference n printed. Flagged only by percentile (≥ p90 = less inverse than usual, ≤ p10 = more inverse than usual). Overlapping windows make the percentile a rank, not a confidence.
P4 runs the same detector with price = S&P 500 and oscillator = −VIX, so a pair of S&P swings is read together with the VIX at those two dates. Four labels, all descriptive: SPX higher high + VIX higher high · SPX lower low + VIX lower low · SPX lower high + VIX lower · SPX higher low + VIX higher. Not ranked, not scored.
data.php is the single writer of data/bars_1d.json (10 years daily, Yahoo, upserted at most every 675 s, every old row kept, served from the record; a failed fetch serves the record unchanged and says so). Sessions are aligned on the ET date of the bar; vendor replays of a holiday are dropped and listed; nulls are never zeros. The option snapshot is written only by the Windows shipper.
3m · 6m · 1y (default) · 2y · 5y of the 10-year record; wheel zooms, drag pans, shift-drag brushes, double-click shows the whole record. All four charts share one X; every stamp names the window actually drawn with its real dates and session count.
Views only. Every sentence describes what was observed and the mechanics behind it; nothing names a side, a motive or what happens next. Percentiles are ranks against each series' own past, and every one prints the n it was computed from.