The volatility risk premium is the gap between the volatility option traders price in and the volatility that actually shows up. For the S&P 500 it has been there most of the time: from January 2, 1990 to October 8, 2026 (9,256 trading days), the VIX closed above the index’s realized volatility over the next 21 sessions on 86.78% of days, by 4.43 points on average (an average VIX of 19.43 against 15.00 realized).
When this series first ran in October 2014, it went straight to switching rules for volatility ETNs. This rebuild starts one step earlier, with the measurement those rules sit on. Part II covers measuring it in real time, Part III the risks, and the strategy test a hypothetical rule built on it.
What the volatility risk premium measures
The premium compares two numbers in the same unit, annualized volatility points. One is forward-looking and quoted every day; the other is only known a month later.
- Implied volatility: the VIX. Cboe describes the VIX as the market’s expectation of 30-day S&P 500 volatility, calculated from SPX option prices. The current method launched on September 22, 2003, and Cboe publishes values back to January 1990 (Cboe methodology).
- Realized volatility. The standard deviation of the S&P 500’s daily log returns over the next 21 trading days (about one calendar month), annualized with the square root of 252.
The gap is VIX minus realized. Two technical points make it run slightly wide. The VIX is the square root of expected variance, which is a bit higher than expected volatility whenever volatility itself moves. And the VIX covers 30 calendar days, while 21 sessions is the closest trading-day match.
| Sample | n | Avg VIX | Avg next-21-day realized vol | Avg gap (pts) | Median gap (pts) | % VIX above realized |
|---|---|---|---|---|---|---|
| Every trading day | 9256 | 19.43 | 15.00 | 4.43 | 4.95 | 86.78 |
| First trading day of each month (non-overlapping) | 442 | 19.48 | 14.96 | 4.52 | 5.10 | 85.52 |
Daily observations overlap (each 21-day window shares 20 days with the next), so 9,256 overstates how many independent tests this is. The one-per-month sample, 442 months that don’t overlap, tells the same story.
The premium year by year
The average gap was positive in 36 of the 37 calendar years (2026 counts through October 8). 2008 is the only negative year, and the share of days with VIX above realized ranged from 61.26% (2008) to 100% (2003, 2017 and 2023).
| Year | n days | Avg VIX | Avg next-21-day RV | Avg gap | Median gap | % days VIX above |
|---|---|---|---|---|---|---|
| 1990 | 253 | 23.06 | 15.03 | 8.03 | 8.74 | 96.44 |
| 1991 | 252 | 18.37 | 13.13 | 5.24 | 5.24 | 90.08 |
| 1992 | 254 | 15.45 | 9.26 | 6.19 | 6.07 | 99.61 |
| 1993 | 253 | 12.69 | 8.24 | 4.45 | 4.50 | 95.26 |
| 1994 | 252 | 13.93 | 9.37 | 4.55 | 4.79 | 94.84 |
| 1995 | 252 | 12.39 | 7.82 | 4.57 | 4.97 | 96.43 |
| 1996 | 254 | 16.44 | 11.10 | 5.34 | 6.26 | 88.98 |
| 1997 | 251 | 22.36 | 17.28 | 5.09 | 5.40 | 89.24 |
| 1998 | 252 | 25.60 | 18.15 | 7.45 | 8.35 | 86.11 |
| 1999 | 251 | 24.37 | 17.86 | 6.51 | 6.61 | 97.61 |
| 2000 | 252 | 23.32 | 21.31 | 2.00 | 3.10 | 68.65 |
| 2001 | 248 | 25.75 | 19.18 | 6.57 | 7.33 | 80.24 |
| 2002 | 252 | 27.29 | 24.52 | 2.78 | 3.44 | 71.83 |
| 2003 | 252 | 21.98 | 15.29 | 6.69 | 6.20 | 100.00 |
| 2004 | 252 | 15.48 | 10.74 | 4.74 | 4.41 | 98.02 |
| 2005 | 252 | 12.81 | 9.92 | 2.89 | 3.07 | 90.08 |
| 2006 | 251 | 12.81 | 9.28 | 3.53 | 3.86 | 93.23 |
| 2007 | 251 | 17.54 | 15.51 | 2.03 | 2.58 | 67.73 |
| 2008 | 253 | 32.69 | 34.77 | -2.07 | 1.16 | 61.26 |
| 2009 | 252 | 31.48 | 23.56 | 7.92 | 8.23 | 96.43 |
| 2010 | 252 | 22.55 | 16.08 | 6.46 | 7.74 | 86.51 |
| 2011 | 252 | 24.20 | 20.55 | 3.65 | 5.06 | 84.92 |
| 2012 | 250 | 17.80 | 12.29 | 5.51 | 5.24 | 99.20 |
| 2013 | 252 | 14.23 | 10.52 | 3.71 | 4.04 | 86.11 |
| 2014 | 252 | 14.18 | 10.82 | 3.35 | 3.72 | 78.17 |
| 2015 | 252 | 16.67 | 14.59 | 2.08 | 2.95 | 79.37 |
| 2016 | 252 | 15.83 | 11.05 | 4.77 | 5.49 | 87.30 |
| 2017 | 251 | 11.09 | 6.49 | 4.60 | 4.79 | 100.00 |
| 2018 | 251 | 16.64 | 15.68 | 0.96 | 3.61 | 61.35 |
| 2019 | 252 | 15.39 | 10.80 | 4.58 | 5.35 | 85.32 |
| 2020 | 253 | 29.25 | 26.50 | 2.75 | 8.96 | 76.68 |
| 2021 | 252 | 19.66 | 12.51 | 7.15 | 7.10 | 95.63 |
| 2022 | 251 | 25.62 | 23.28 | 2.35 | 2.48 | 67.33 |
| 2023 | 250 | 16.87 | 12.20 | 4.67 | 4.38 | 100.00 |
| 2024 | 252 | 15.61 | 12.15 | 3.45 | 3.91 | 82.14 |
| 2025 | 250 | 18.96 | 15.10 | 3.87 | 6.27 | 85.20 |
| 2026 (to Oct 8, 2026) | 193 | 18.17 | 12.53 | 5.64 | 5.53 | 95.34 |
Look at 2020: the median gap was 8.96 points, one of the widest in the table, yet the average was only 2.75. A few weeks in February and March pulled the average down. That pattern, a steady premium most days and a few days that wipe out a lot of it, is what Part III is about.
Does a higher VIX mean a bigger premium?
In points, yes. The average gap rises from 3.03 when the VIX is under 15 to 7.02 when it is between 30 and 40. But the premium also gets less reliable at the very top: with the VIX at 40 or higher, it beat realized volatility on 79.81% of days, the lowest of any band.
| VIX on the day | n days | Avg VIX | Avg next-21-day RV | Avg gap | Median gap | % VIX above |
|---|---|---|---|---|---|---|
| Under 15 | 2962 | 12.81 | 9.78 | 3.03 | 3.76 | 86.93 |
| 15 to 20 | 2870 | 17.30 | 12.87 | 4.43 | 5.30 | 88.36 |
| 20 to 25 | 1824 | 22.28 | 17.33 | 4.95 | 6.08 | 84.81 |
| 25 to 30 | 864 | 27.17 | 21.19 | 5.98 | 7.14 | 85.19 |
| 30 to 40 | 528 | 33.44 | 26.42 | 7.02 | 8.93 | 89.39 |
| 40 and up | 208 | 50.20 | 43.68 | 6.52 | 8.98 | 79.81 |
The 40-and-up band has only 208 days, 125 of them in 2008 and 2009 and 35 in 2020, so it is really three episodes. Treat it as a small sample.
The big exceptions
The premium fails worst when a shock arrives while the VIX is still calm. Below are the ten largest misses, keeping only the worst day of each episode (at least 42 sessions apart). The VIX number is that day’s close; the realized number is what the next 21 sessions delivered.
| Day (VIX close) | VIX | Realized vol, next 21 days | Gap (pts) |
|---|---|---|---|
| Feb 19, 2020 | 14.38 | 82.25 | -67.87 |
| Sep 12, 2008 | 25.66 | 73.42 | -47.76 |
| Mar 25, 2025 | 17.15 | 48.09 | -30.94 |
| Jul 25, 2011 | 19.35 | 46.07 | -26.72 |
| Aug 10, 2015 | 12.23 | 30.21 | -17.98 |
| Jun 28, 2002 | 25.40 | 41.98 | -16.58 |
| Oct 22, 1997 | 19.86 | 35.86 | -16.00 |
| Dec 3, 2018 | 16.44 | 30.94 | -14.50 |
| Jan 26, 2018 | 11.08 | 25.51 | -14.43 |
| Apr 22, 2010 | 16.47 | 29.59 | -13.12 |
And the stress years, counted by how many days the VIX ended up below what followed:
| Year | n days | Days VIX below realized | % of days | Avg gap | Worst day (gap) |
|---|---|---|---|---|---|
| 1998 | 252 | 35 | 13.89 | 7.45 | Aug 25, 1998 (-12.78) |
| 2002 | 252 | 71 | 28.17 | 2.78 | Jun 28, 2002 (-16.58) |
| 2008 | 253 | 98 | 38.74 | -2.07 | Sep 12, 2008 (-47.76) |
| 2011 | 252 | 38 | 15.08 | 3.65 | Jul 25, 2011 (-26.72) |
| 2018 | 251 | 97 | 38.65 | 0.96 | Dec 3, 2018 (-14.50) |
| 2020 | 253 | 59 | 23.32 | 2.75 | Feb 19, 2020 (-67.87) |
| 2022 | 251 | 82 | 32.67 | 2.35 | Apr 20, 2022 (-13.08) |
| 2025 | 250 | 37 | 14.80 | 3.87 | Mar 25, 2025 (-30.94) |
1998 is the surprise. The S&P 500 fell 19.34% from its high to its low within that year, yet the VIX averaged 25.60, well above its 19.43 long-run average, so it was already pricing a lot of movement. The average gap (7.45) is the third largest in the yearly table. The sign flips most often in long grinding years (2008, 2018, 2022, each above 32% of days), while the deepest single misses come from sudden shocks out of quiet markets (2020, 2015, early 2018).
Across all 9,256 days, the gap was positive 86.78% of the time and averaged 6.08 points when it was. On the other 13.22% of days it averaged -6.42, and the extremes are lopsided: +31.55 points at best, -67.87 at worst.
What this does and doesn’t show
It shows that SPX options have, on most days since 1990, priced more volatility than the index then delivered. It doesn’t show a return anyone earned. The VIX itself can’t be bought or sold, and the products that try to collect this gap (short options, short VIX futures, the old inverse-VIX notes) each add costs and risks of their own. Part II looks at how much of the premium you can see in advance, and what the VIX futures products were.
These are historical statistics, not a forecast or a recommendation. More studies on the Market Stats page, including the Williams VIX Fix.
Every number above was calculated from Cboe’s VIX history and S&P 500 closes, listed under Sources.
