This is a hypothetical, rules-based historical test, not a trading record and not a recommendation. A simple volatility-risk-premium rule, holding SPY only when the VIX sits above recent realized volatility, would have returned 14.75% a year from January 2015 to November 6, 2026, against 13.20% for buy and hold, with a maximum drawdown of -18.99% instead of -33.72%. That is before costs, with perfect execution at the close, and other reasonable settings of the same idea did worse.
The January 2015 post at this address reported “preliminary” results for a rule that switched between two volatility notes, XIV and VXX, on partly simulated data. Both notes are gone (see Part II). This rebuild tests the same signal on things that still exist and can be checked: SPY and Cboe’s PutWrite index. The years after the original post, 2015 to 2026, serve as the out-of-sample check.
The rules
All rules were fixed before the results were run. The main rule copies the version of the VRP signal that volatility blogs were discussing in 2014: VIX minus 10-day historical volatility, smoothed over 5 days.
- Signal at each close: the 5-day simple average of [VIX close minus the S&P 500’s 10-day realized volatility], where realized volatility is the annualized standard deviation of daily log returns (the same measure as Part I).
- Position: if the signal is above zero, hold the asset from that close to the next close; otherwise hold cash.
- Cash earns 0%. That understates the rule, since T-bills paid interest for much of the period.
- Costs: none in the main columns; one column charges 0.10% of equity on every switch in or out.
- Assets: SPY with dividends reinvested (adjusted close), from January 29, 1993; and the Cboe S&P 500 PutWrite Index (PUT), which sells one-month at-the-money S&P 500 puts against Treasury bills (Cboe methodology), from January 3, 2007.
Results on SPY
| Period | Dates | n days | Buy-and-hold CAGR % | Rule CAGR % | Rule CAGR after 0.10%/switch | Buy-and-hold max DD % | Rule max DD % | % time invested | Switches |
|---|---|---|---|---|---|---|---|---|---|
| Full: 1993-2026 | 1993-01-29 to 2026-11-06 | 8498 | 10.66 | 11.62 | 10.92 | -55.19 | -48.87 | 91.30 | 213 |
| 1993-2014 | 1993-01-29 to 2014-12-31 | 5518 | 9.31 | 9.97 | 9.31 | -55.19 | -48.87 | 92.01 | 132 |
| 2015-2026 (out of sample) | 2015-01-02 to 2026-11-06 | 2979 | 13.20 | 14.75 | 13.97 | -33.72 | -18.99 | 90.03 | 81 |
The rule was invested about 90% of the time and stepped aside for the rest. Before 2015, its extra return was entirely eaten by the 0.10% switch charge (9.31% either way). Its worst stretch then was March 2000 to March 2003 (-48.87%), and in the 2007-2009 bear market it fell 43.53% against buy and hold’s 55.19%, so stepping aside softened that crash without avoiding it. After 2015 it did better on both counts. Its deepest post-2015 drawdown came from February 19 to April 8, 2025 (-18.99%); buy and hold’s was the February 19 to March 23, 2020 crash (-33.72%).
Year by year after 2015, the lead came from a few years, not a steady drip:
| Year | Buy-and-hold % | Rule % | Difference (pts) |
|---|---|---|---|
| 2015 | 1.23 | 2.37 | 1.14 |
| 2016 | 12.00 | 10.76 | -1.24 |
| 2017 | 21.71 | 20.88 | -0.83 |
| 2018 | -4.57 | 3.15 | 7.72 |
| 2019 | 31.22 | 24.50 | -6.72 |
| 2020 | 18.33 | 34.68 | 16.35 |
| 2021 | 28.73 | 27.63 | -1.10 |
| 2022 | -18.18 | -11.36 | 6.81 |
| 2023 | 26.18 | 27.07 | 0.90 |
| 2024 | 24.89 | 23.56 | -1.33 |
| 2025 | 17.72 | 8.53 | -9.18 |
| 2026 | 8.67 | 11.67 | 3.00 |
2026 runs through November 6. The rule trailed in 6 of the 12 years, and three years (2018, 2020 and 2022) did most of the work. In 2025 it trailed by 9.18 points, and that year also held its deepest post-2015 drawdown.
Results on the PutWrite index
| Period | Dates | n days | Buy-and-hold CAGR % | Rule CAGR % | Rule CAGR after 0.10%/switch | Buy-and-hold max DD % | Rule max DD % | % time invested | Switches |
|---|---|---|---|---|---|---|---|---|---|
| Full: 2007-2026 | 2007-01-03 to 2026-11-06 | 4993 | 7.10 | 9.27 | 8.48 | -37.09 | -24.81 | 89.47 | 143 |
| 2007-2014 | 2007-01-03 to 2014-12-31 | 2013 | 5.85 | 9.12 | 8.28 | -37.09 | -24.81 | 88.67 | 62 |
| 2015-2026 (out of sample) | 2015-01-02 to 2026-11-06 | 2979 | 7.94 | 9.37 | 8.62 | -28.93 | -14.93 | 90.03 | 81 |
On PUT the rule’s edge over buy and hold was larger in both halves and survived the switch charge. One reason to be careful: PUT is an index, not a fund, and switching in and out of a put-writing position daily would mean closing and reopening options at their bid-ask spreads, which a 0.10% charge may not cover. The index also assumes its puts are sold at a volume-weighted price on each monthly roll date.
How sensitive is it to the settings?
Quite. Same idea, different lookbacks, all on SPY:
| Rule | Period | CAGR % | Max DD % | % time invested | Switches |
|---|---|---|---|---|---|
| Primary: VIX minus 10-day RV, 5-day avg > 0 | 1993-2014 | 9.97 | -48.87 | 92.01 | 132 |
| Primary: VIX minus 10-day RV, 5-day avg > 0 | 2015-2026 | 14.75 | -18.99 | 90.03 | 81 |
| VIX minus 21-day RV, no avg > 0 | 1993-2014 | 8.98 | -56.26 | 90.11 | 203 |
| VIX minus 21-day RV, no avg > 0 | 2015-2026 | 13.05 | -24.15 | 86.77 | 133 |
| VIX minus 2-day RV, 5-day avg > 1.25 (the 2014 “revised” rule) | 1993-2014 | 10.39 | -47.77 | 98.46 | 63 |
| VIX minus 2-day RV, 5-day avg > 1.25 (the 2014 “revised” rule) | 2015-2026 | 15.96 | -24.08 | 98.46 | 39 |
| Primary, acting one day late | 1993-2014 | 9.14 | -48.94 | 92.02 | 132 |
| Primary, acting one day late | 2015-2026 | 14.49 | -24.51 | 90.03 | 81 |
| Buy and hold SPY | 1993-2014 | 9.31 | -55.19 | 100.00 | 0 |
| Buy and hold SPY | 2015-2026 | 13.20 | -33.72 | 100.00 | 0 |
The 21-day version trailed buy and hold in both periods, with a worse drawdown before 2015. The 2-day version (the shorter window the 2014 posts favored) was invested 98.46% of the time, so it was close to buy and hold with a few exits. Acting one session late, which is what you get if you wait for the close to confirm the signal and trade the next day, cut the pre-2015 return below buy and hold and gave back much of the post-2015 drawdown improvement (-24.51% instead of -18.99%).
What the rule was really doing
It mostly avoided high-volatility days, good and bad alike.
| Days | n | Avg daily SPY return % | Median % | % up | Worst day % | Best day % | Std dev % |
|---|---|---|---|---|---|---|---|
| Rule invested | 7759 | 0.05 | 0.07 | 54.21 | -7.42 | 10.50 | 1.05 |
| Rule in cash | 739 | -0.02 | 0.04 | 52.10 | -10.94 | 14.52 | 2.04 |
The 739 days in cash (1993-2026) were about twice as volatile as the invested days and held both SPY’s worst day in the sample (-10.94%) and its best (+14.52%). The average out-of-market day was slightly negative, which is where the rule’s edge came from, but 739 days with a 2.04% standard deviation is a noisy basis for that average. That fits Part II: a negative real-time premium mostly flags volatility that has already arrived.
Caveats, plainly
- Hypothetical. No account traded these rules. Results are simulated from index and ETF closes.
- Not tradeable as shown. Trading at the same close that sets the signal, no commissions, spreads, slippage or taxes, and no interest on cash.
- One sample, many choices. The lookback, the smoothing and the zero cutoff are choices; nearby choices gave weaker results (table above). The main rule was already being discussed publicly in 2014, after it had looked good on earlier data, so the 1993-2014 numbers aren’t an independent test.
- Products changed. The notes the original strategy used no longer exist, and PUT is a benchmark index, not something you can buy directly.
- Past results don’t predict future ones. The post-2015 outperformance rests mostly on three years, 2020 above all.
For the risks of the short-volatility side of this trade, see Part III. More studies on the Market Stats page.
Every number above was calculated from the Cboe and S&P 500 daily data listed under Sources.
