Selling the volatility risk premium wins small and often and loses big and rarely. On a non-overlapping monthly sample from 1990 to 2026 (442 months), a stylized one-month variance trade, collecting the VIX’s implied variance and paying the variance that followed, came out ahead in 85.52% of months. The other 64 months gave back 36.2% of everything the winning months had collected, and the single worst month (March 2020) lost about 48 times what a typical winning month earned.
Part I showed how often the premium is there and Part II how hard it is to time. This part, which completes the series, is about the months when it fails.
How lopsided the gap is
The gap between the VIX and the next 21 sessions of S&P 500 realized volatility (Part I’s measure) is mostly positive and occasionally very negative. Across 9,256 trading days from January 2, 1990 to October 8, 2026:
| Percentile | Gap (vol pts) |
|---|---|
| 1st | -18.81 |
| 5th | -4.31 |
| 10th | -1.00 |
| 25th | 2.41 |
| 50th | 4.95 |
| 75th | 7.51 |
| 90th | 10.22 |
| 95th | 12.30 |
| 99th | 17.12 |
Eight days in ten sit between -1.00 and +10.22 points, and the 1st and 99th percentiles are about the same distance from zero. The shape changes beyond them: the best day in the sample was +31.55 points and the worst was -67.87 (February 19, 2020). The skew of the daily gap is -3.36, a strongly left-leaning distribution.
Why variance makes the tail fatter
Cboe calculates the VIX as the square root of a 30-day expected variance built from SPX option prices (Cboe methodology). A trade settled on variance (volatility squared) is paid in squared terms, so a shock that doubles realized volatility costs four times as much. The table below uses the first trading day of each month and scores each month as (VIX² minus realized variance) / 100. It is an illustration with no costs, margin or product, not a return series.
| Month starting | VIX | Realized vol next 21 days | Variance gap |
|---|---|---|---|
| Mar 2, 2020 | 33.42 | 92.43 | -74.26 |
| Oct 1, 2008 | 39.81 | 81.40 | -50.41 |
| Nov 3, 2008 | 53.68 | 74.96 | -27.38 |
| Sep 2, 2008 | 21.99 | 53.83 | -24.14 |
| Apr 1, 2025 | 21.77 | 47.69 | -18.00 |
The 378 winning months summed to 853.0; the five months above summed to -194.2, or 22.8% of that total on their own. The median winning month was 1.55. Note that the VIX in March 2020 was already 33.42 when the month began. A high VIX is not protection if what follows is higher still.
The days that broke short-volatility products
The fastest damage comes from one-day jumps in implied volatility, which hit anything that is short VIX futures or short options marked to market every day.
| Date | VIX prior close | VIX close | % change |
|---|---|---|---|
| Feb 5, 2018 | 17.31 | 37.32 | 115.60 |
| Dec 18, 2024 | 15.87 | 27.62 | 74.04 |
| Nov 5, 2026 | 16.61 | 28.18 | 69.66 |
| Aug 5, 2024 | 23.39 | 38.57 | 64.90 |
| Feb 27, 2007 | 11.15 | 18.31 | 64.22 |
| Jan 27, 2021 | 23.02 | 37.21 | 61.64 |
February 5, 2018 is the day Credit Suisse’s inverse VIX note, XIV, fell to 20% or less of its prior close intraday and triggered the clause that ended it (Credit Suisse release, SEC; details in Part II). The arithmetic of a daily inverse product is unforgiving: a 100% one-day rise in the futures index it shorts takes it to zero. Two of the six largest jumps since 1990 came in 2024, and the third largest came on November 5, 2026, the second-to-last session in this data.
What option-selling and short-VIX indices did
Cboe publishes benchmark indices for several ways of selling volatility. Three of them: PUT sells one-month at-the-money S&P 500 puts each month, fully backed by Treasury bills (PUT methodology). BXM holds the S&P 500 and sells a one-month at-the-money call against it (BXM methodology). VPD sells next-month VIX futures on a Treasury bill account, sized so that a 25-point rise in VIX futures would still leave 75% of the starting value (VPD methodology). From November 16, 2007, VPD’s first value, to November 6, 2026 (4,768 days, 228 months):
| Index | CAGR % | Ann. vol % | Max drawdown % | Best month % | Worst month % | Monthly skew | % months up | Worst day % |
|---|---|---|---|---|---|---|---|---|
| Cboe S&P 500 PutWrite (PUT) | 7.08 | 13.92 | -37.09 | 8.98 | -17.65 | -1.65 | 73.68 | -11.46 |
| Cboe S&P 500 BuyWrite (BXM) | 5.99 | 14.39 | -40.14 | 10.01 | -15.01 | -1.34 | 69.30 | -12.15 |
| Cboe VIX Premium Strategy (VPD, short VIX futures) | 10.48 | 24.23 | -61.73 | 13.77 | -41.34 | -2.92 | 64.91 | -31.16 |
| S&P 500 total return (SPY) | 10.91 | 19.71 | -53.89 | 12.70 | -16.52 | -0.54 | 65.35 | -10.94 |
PUT was up in more months than the S&P 500 and was less volatile, but its best month (+8.98%) was half the size of its worst (-17.65%). That is the capped-upside, open-downside payoff of a sold option. VPD shows the same shape in stronger form: a worst month of -41.34% and a worst day of -31.16%, with roughly the S&P 500’s long-run return at higher volatility.
Peak-to-trough inside each shock window:
| Episode | Window | VIX at start | Peak VIX close | S&P 500 TR drawdown % | PUT drawdown % | BXM drawdown % | VPD drawdown % |
|---|---|---|---|---|---|---|---|
| Fall 2008 | Sep 2, 2008 to Mar 31, 2009 | 21.99 | 80.86 (Nov 20, 2008) | -46.04 | -35.52 | -36.97 | -60.70 |
| Aug 2011 | Jul 1, 2011 to Oct 31, 2011 | 15.87 | 48.00 (Aug 8, 2011) | -18.37 | -14.65 | -14.36 | -30.77 |
| Aug 2015 | Aug 3, 2015 to Sep 30, 2015 | 12.56 | 40.74 (Aug 24, 2015) | -11.07 | -8.77 | -9.70 | -15.35 |
| Feb 2018 | Jan 22, 2018 to Mar 29, 2018 | 11.03 | 37.32 (Feb 5, 2018) | -10.10 | -7.64 | -7.72 | -21.37 |
| Mar 2020 | Feb 14, 2020 to Apr 30, 2020 | 13.68 | 82.69 (Mar 16, 2020) | -33.72 | -28.93 | -30.26 | -57.42 |
| Aug 2024 | Jul 10, 2024 to Aug 30, 2024 | 12.85 | 38.57 (Aug 5, 2024) | -8.41 | -4.96 | -5.07 | -18.30 |
| Apr 2025 | Mar 25, 2025 to May 30, 2025 | 17.15 | 52.33 (Apr 8, 2025) | -13.72 | -10.72 | -10.81 | -20.74 |
The windows are hand-picked around each spike, so read the rows as snapshots rather than a ranking. In all seven, the fully collateralized put-selling index fell less than the S&P 500 itself, and the short VIX futures index fell much further than either, from 15.35% (August 2015) to 60.70% (fall 2008). On February 5, 2018, VPD lost 18.29% in one session while PUT lost 3.86%; on August 5, 2024 the numbers were 11.20% and 2.65%; on March 16, 2020, 31.16% and 11.46%.
Why “selling the premium” has fat left tails
Three things stack up. The payoff of a sold option or a short volatility future is capped on the winning side and open on the losing side. Volatility losses scale with variance, so they grow faster than the move that causes them. And the worst shocks tend to start from a calm, low VIX (Part I found the deepest misses began with the VIX at 11 to 26), exactly when the premium being collected is smallest. How much of this a given position feels depends mostly on sizing and collateral: the same underlying premium produced a -28.93% drawdown in PUT and a -57.42% drawdown in VPD in early 2020.
That is what the data show. It doesn’t show whether the premium will be larger or smaller from here, and none of this is a recommendation to buy or sell any volatility product. The strategy test takes a hypothetical rule that tries to step aside when the premium looks thin, and checks how it held up after 2015. More studies on the Market Stats page.
Every number above was calculated from the Cboe and S&P 500 daily data listed under Sources.
