The study behind every reading

The ratio of touch to close beyond is 1.95 at the 68% edge and 2.05 at the 90% edge, and it stayed between 1.86 and 2.19 in every year, every liquidity and implied-volatility tercile and on both edges. Earnings sessions and funds touch less often for the same close because more of their move arrives overnight. The sample is a survivor panel.

Last week, October 26–October 30, 2026. Single names: of 7,340 sessions, 33.3% closed beyond the 68% band and 61.4% touched it. Funds (ETF): of 2,085 sessions, 32.6% closed beyond the 68% band and 55.8% touched it.

Measured as the touch-odds study above: the band is the chain's own on the entry date, the outcome is the next trading session's high, low and close, and a session counts in the month its outcome falls in. Sessions are dropped where the chain's spot is more than 1% from that day's close or the next session has no bar. A frequency, not a direction. Updated each Saturday.

What was measured. 91.4% of option prints fill at the bid or the ask with no improvement; the median listed spread is 18.6% of mid, and choosing by spread removes about a third of the cost.

How. Effective against quoted spread on sampled sessions of our own tape, priced against the end-of-day chain.

What it does not claim. A quote, not a fill: the spread can widen before an order arrives, and a cheaper contract is a different position.

What was measured. The move forecast ranked the reaction with IC 0.44 against 0.37 for the market's own straddle, 10 of 10 years; the crush forecast IC 0.56, 9 of 9.

How. Walk-forward by year, trained on prior years only; symbol hold-out 7 of 7.

What it does not claim. An unsigned size. It is not a straddle trade: the quoted spread is 4.7× the whole earnings variance premium.

What was measured. The realised move was 1.00–1.05× the forecast at every horizon from 1 to 21 sessions, in all eight held-out years.

How. Walk-forward 2019–2026: total variance interpolated across the expiry ladder in trading sessions, the earnings jump split out, one multiplier per horizon fitted on prior years only.

What it does not claim. Size, not direction. The bands are close-to-close; how often a session only touches an edge is measured on the odds card.

What was measured. No signed-flow feature predicted the next 1, 5, 10 or 21 sessions' return: the best of 399 signed-return tests reached |t| 3.99, inside the shuffled-label bar of 5.59, while 29 volatility pairs cleared it.

How. Daily cross-sectional rank IC, both sides residualised within each session on prior realised vol, ATM IV, dollar volume and price; Newey-West t; the same run on shuffled labels sets the bar.

What it does not claim. Describes today's prints. Bullish and bearish premium move with price, not ahead of it — read it as the mood, not a forecast.

What was measured. An untouched 68% edge on a single name is touched by the close with odds 2·(1−Φ(z / (0.85·√V))), z its distance in the band's sigma and V the share of the day's intraday variance still to come; out of sample every probability bin was within 2.4 points, and touch ran 1.94–2.06× close-beyond at every half hour.

How. Point-in-time: the previous close's one-session band against the session's minute path. Variance clock and k fitted Jan–Jun 2026, tested Jul–Oct 2026; Wilson and date-clustered bootstrap intervals; split by month, liquidity tercile, edge, band width and time of day.

What it does not claim. A frequency, not a direction. Not shown on earnings sessions or for funds. A month quieter or busier than options priced moved it by up to 5 points (Sep 2026: −5). Ten months of one regime of the 150 most active names.

What was measured. On single names the next session touched a 68% edge 59.5% of the time and closed beyond it 30.5% — 1.95×, and 2.05× at the 90% edge — within 1.86–2.19 in every year 2016–2026, every liquidity and IV tercile and both edges.

How. Point-in-time: the chain's band on the entry date against the next session's bars; date-clustered bootstrap on the ratio. Earnings sessions (1.73×) and funds (1.71×) measured as their own rows.

What it does not claim. A frequency, not a direction: a touch says the priced move arrived, not which way the day ends. Survivor panel. How the odds change through the session is its own record (touch odds through the session).

What was measured. The 10-session IV path beat its best stored rule in 8 of 8 walk-forward folds (IC 0.44 against 0.38), and the premium over realised did the same.

How. Yearly walk-forward folds 2019–2026, rank IC against the realised path.

What it does not claim. A volatility forecast, not a return. Selling the premium is charged at a median spread and the 2025–26 stretch was the decade's weakest.

What was measured. The remaining-move cone held its 68% and 90% bands in every out-of-sample year, and a strike is touched about twice as often as it is closed beyond.

How. Calibrated by hour and distance on SPY and SPX 0DTE sessions, each year held out in turn; FOMC days measured separately.

What it does not claim. No direction. SPY's option-implied odds ran high in every year, so SPY shows the measured frequency and SPX the market's own.