The Williams VIX Fix (WVF) is a synthetic VIX built from price alone: take the highest close of the last 22 sessions, subtract today’s low, divide by that highest close and multiply by 100. On the S&P 500, its daily readings have had a 0.75 correlation with the Cboe VIX across 9,277 sessions from 1990 to November 6, 2026. After the VIX Fix spiked, S&P 500 returns over the next three months were a little better than average, but the gap was small and it changed from one era to the next.
Larry Williams laid out the indicator in “The VIX Fix,” in the December 2007 issue of Active Trader magazine. His aim was a fear gauge for markets that have no options-based VIX of their own: bonds, gold, grains, single stocks. Everything below was calculated from the daily data listed under Sources.
The formula and a worked example
The VIX Fix measures how far today’s low sits below the best close of roughly the past month, in percent.
WVF = (Highest close of the last 22 sessions − Today’s low) ÷ Highest close of the last 22 sessions × 100
The 22-session window includes today. Williams chose 22 because a month never has more than 22 trading days, and he said the number was not optimized. Because a low can never be above that day’s close, and the highest close is at least today’s close, the reading never goes below zero. It reaches zero only when the day’s low equals the month’s best close. A big number means the market traded well below its recent peak at some point during the day.
Here are the last four sessions in the data, in the week the S&P 500 fell 4.36% on November 5, 2026:
| Session | Highest close, last 22 sessions | Day low | Day close | WVF | VIX close |
|---|---|---|---|---|---|
| 2026-11-03 | 7818.93 (2026-10-06) | 7545.39 | 7568.92 | 3.50 | 20.83 |
| 2026-11-04 | 7818.93 (2026-10-06) | 7635.81 | 7688.62 | 2.34 | 16.61 |
| 2026-11-05 | 7811.54 (2026-10-09) | 7350.35 | 7353.73 | 5.90 | 28.18 |
| 2026-11-06 | 7811.54 (2026-10-09) | 7369.55 | 7376.86 | 5.66 | 25.64 |
Take November 5. The best close in the window was 7,811.54 from October 9. The day’s low was 7,350.35. So (7,811.54 − 7,350.35) ÷ 7,811.54 × 100 = 5.90. Notice the reference high changed that day: the October 6 close of 7,818.93 rolled out of the 22-session window, and the next-best close took its place. That rolling high is the indicator’s memory, and it explains a lot of its behavior.
How closely the VIX Fix tracks the VIX
Closely in shape, loosely in size. Daily levels correlate at 0.75 since 1990 and daily changes at 0.49, and the fit has tightened in recent decades.
| Years | Sessions | Correlation of levels | Correlation of daily changes | Median WVF | Median VIX |
|---|---|---|---|---|---|
| 1990-2026 | 9277 | 0.75 | 0.49 | 1.69 | 17.57 |
| 1990-1999 | 2524 | 0.62 | 0.44 | 1.63 | 17.09 |
| 2000-2009 | 2515 | 0.76 | 0.48 | 2.54 | 20.64 |
| 2010-2019 | 2516 | 0.77 | 0.53 | 1.29 | 15.47 |
| 2020-2026 | 1722 | 0.80 | 0.54 | 1.49 | 18.64 |
The scales are nothing alike. A VIX Fix of 2 is an ordinary day; a VIX of 2 has never happened. To translate a reading, this table shows the VIX level that went with each VIX Fix range:
| WVF reading | Sessions | Share of sessions | Median VIX | Lowest VIX | Highest VIX |
|---|---|---|---|---|---|
| 0 to 1 | 3005 | 32.4% | 13.96 | 9.14 | 41.67 |
| 1 to 2 | 2133 | 23.0% | 16.27 | 10.03 | 49.84 |
| 2 to 4 | 2023 | 21.8% | 18.60 | 10.77 | 55.73 |
| 4 to 6 | 997 | 10.7% | 21.72 | 13.27 | 58.91 |
| 6 to 10 | 768 | 8.3% | 26.16 | 15.24 | 59.98 |
| 10 to 15 | 256 | 2.8% | 32.29 | 20.10 | 66.31 |
| 15+ | 95 | 1.0% | 53.54 | 29.02 | 82.69 |
The ranges overlap a lot. A reading of 0 to 1 still came with a VIX above 40 on some days, which is the main limitation covered below.
Parameters: the lookback and the price you subtract
Williams used 22 sessions and the day’s low. Both can be changed, and on the S&P 500 the longer windows tracked the VIX a bit more closely.
| Lookback (sessions) | Price subtracted | Correlation with VIX |
|---|---|---|
| 10 | Low (original) | 0.71 |
| 10 | Close | 0.65 |
| 22 | Low (original) | 0.75 |
| 22 | Close | 0.72 |
| 44 | Low (original) | 0.80 |
| 44 | Close | 0.79 |
| 66 | Low (original) | 0.82 |
| 66 | Close | 0.81 |
A longer window keeps an old high in the formula for longer, so the reading stays raised for weeks after a selloff, which is also how the VIX tends to behave. Using the close instead of the low lets you build the indicator from closing prices only, at a small cost in fit.
To flag a “spike,” this page uses the defaults of CM_Williams_Vix_Fix, a widely copied charting script: the VIX Fix closes above its 20-day Bollinger band (20-day average plus 2 standard deviations), or at 85% or more of its highest reading in the last 50 sessions. On the S&P 500 since 1962, one of those rules was true on 16.9% of sessions, so spikes come in bunches. The “first spike” rows below keep only spikes with no other spike in the previous 10 sessions.
What the S&P 500 did after a VIX Fix spike
Somewhat better than an average day, mostly at the three-month horizon. These are S&P 500 price returns from the close of the spike day, 1962 to 2026:
| Signal | n | 5d avg % | 5d median % | 5d % up | 21d avg % | 21d median % | 21d % up | 63d avg % | 63d median % | 63d % up | n with 63d data |
|---|---|---|---|---|---|---|---|---|---|---|---|
| All sessions (baseline) | 16301 | 0.17 | 0.29 | 56.44 | 0.71 | 1.08 | 61.35 | 2.13 | 2.66 | 66.46 | 16238 |
| Bollinger rule | 1391 | 0.31 | 0.37 | 56.66 | 0.66 | 1.13 | 60.42 | 2.46 | 2.95 | 67.39 | 1386 |
| Range rule (85% of 50-day high) | 2116 | 0.31 | 0.42 | 57.34 | 1.06 | 1.34 | 61.47 | 2.87 | 3.10 | 68.56 | 2109 |
| Either rule (spike day) | 2751 | 0.34 | 0.44 | 57.59 | 0.91 | 1.19 | 60.50 | 2.64 | 2.94 | 67.74 | 2743 |
| First spike in 10 sessions | 416 | 0.36 | 0.40 | 59.38 | 0.91 | 1.23 | 61.20 | 2.65 | 3.04 | 70.53 | 414 |
Across 416 first spikes, the S&P 500 was higher 63 sessions later 70.5% of the time, against 66.5% for any session. The 21-day numbers barely differ from the baseline.
By era
The edge was not steady. This table sets first spikes against all sessions in the same years:
| Era / set | n | 5d avg % | 5d median % | 5d % up | 21d avg % | 21d median % | 21d % up | 63d avg % | 63d median % | 63d % up | n with 63d data |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 1962-1989 first spikes | 176 | 0.12 | 0.21 | 53.41 | 0.56 | 0.80 | 53.98 | 2.19 | 2.11 | 64.20 | 176 |
| 1962-1989 all sessions | 7020 | 0.14 | 0.24 | 54.86 | 0.57 | 0.81 | 57.98 | 1.73 | 1.82 | 61.45 | 7020 |
| 1990-1999 first spikes | 66 | 0.60 | 0.69 | 69.70 | 2.10 | 1.70 | 75.76 | 4.54 | 4.54 | 84.85 | 66 |
| 1990-1999 all sessions | 2528 | 0.29 | 0.36 | 58.98 | 1.27 | 1.25 | 65.39 | 3.84 | 3.39 | 77.85 | 2528 |
| 2000-2009 first spikes | 61 | 0.51 | 0.39 | 60.66 | 0.18 | 0.61 | 57.38 | 0.14 | 1.53 | 59.02 | 61 |
| 2000-2009 all sessions | 2515 | -0.01 | 0.11 | 52.21 | -0.06 | 0.63 | 56.62 | -0.16 | 0.96 | 54.51 | 2515 |
| 2010-2019 first spikes | 63 | 0.11 | 0.19 | 55.56 | 0.96 | 1.77 | 61.90 | 3.46 | 3.96 | 80.95 | 63 |
| 2010-2019 all sessions | 2516 | 0.23 | 0.39 | 60.69 | 0.96 | 1.40 | 67.25 | 2.68 | 3.39 | 74.88 | 2516 |
| 2020-2026 first spikes | 50 | 1.03 | 1.30 | 70.00 | 1.44 | 2.55 | 71.43 | 3.84 | 5.13 | 75.00 | 48 |
| 2020-2026 all sessions | 1722 | 0.28 | 0.46 | 59.17 | 1.19 | 1.74 | 67.55 | 3.87 | 4.75 | 75.59 | 1659 |
From 1962 to 1989 a spike told you almost nothing. In the 1990s and the 2020s the next week looked clearly better than usual. From 2010 to 2019 the next week was actually below average, even though the three-month figure was above it. Each era holds 50 to 176 signals, which is a small sample.
Low or close? The 2010 question
The original 2010 post on this URL tested whether swapping the day’s low for the close changed SPY results, and found the close-only version did better over 1990-2010. With data through 2026 and the spike rules above, the two versions come out close:
| WVF version | n | 5d avg % | 5d median % | 5d % up | 21d avg % | 21d median % | 21d % up | 63d avg % | 63d median % | 63d % up | n with 63d data |
|---|---|---|---|---|---|---|---|---|---|---|---|
| Low (original) | 240 | 0.54 | 0.53 | 63.75 | 1.18 | 1.61 | 66.53 | 2.99 | 3.92 | 75.21 | 238 |
| Close | 241 | 0.43 | 0.46 | 62.24 | 1.02 | 1.40 | 63.33 | 3.03 | 3.94 | 75.00 | 240 |
Limitations of the VIX Fix
It only sees one kind of fear: a drop from a recent high. The VIX also prices fear that hasn’t shown up in prices yet, and it stays high after a crash while prices go sideways at a lower level.
- It misses calm-price fear. Since 1990, 736 sessions closed with the VIX at 30 or more. On 108 of them (14.7%) the VIX Fix was below 3, in years including 1998, 2008, 2009 and 2020. Once the 22-session high rolls down to the new, lower prices, the VIX Fix falls back even if the VIX doesn’t. The reverse is rare: only 3 sessions had a VIX under 15 with a VIX Fix of 5 or more.
- Its memory is fixed. A single high close props the reading up for 22 sessions, then drops out all at once, as the November 2026 example shows.
- The top end compresses. The highest readings were 34.33 on March 23, 2020 (VIX 61.59), 34.02 on October 20, 1987, and 33.09 on October 10, 2008 (VIX 69.95). The 1987 crash has no VIX on Cboe’s 1990-onward file, which is exactly the gap the VIX Fix fills.
- These are statistics, not a strategy. Returns overlap, exclude dividends and costs, and the spike thresholds came from a charting script, not from Williams’ article, which described high readings as signs a low may be near.
As of November 6, 2026, the S&P 500 VIX Fix stood at 5.66 and both spike rules were true, with spike days on October 30 and November 2, 3, 5 and 6. The tables above describe what followed past spikes; they don’t say what happens this time.
For the other side of the same week, see what followed 20% one-day VIX jumps, and for the opposite mood, the level-of-bullishness streaks. The volatility risk premium series covers how the VIX compares with the volatility that actually follows. All posts are on the Market Stats page.
