Mean reversion · daily rule · Semiconductors
Bollinger Reversion on Nvidia
What the play says
NVDA with the play, last two years
Shading shows the play's call from the close that triggered it.
Growth of $10,000 since 2005
Log scale. After trading costs, with T-bill interest while out.
The record
The play against buying and holding NVDA over exactly the same dates, after costs.
| Play, since 2005 | Hold, since 2005 | Play, 5 yrs | Hold, 5 yrs | |
|---|---|---|---|---|
| Annual return | +10.6% | +38.9% | +26.7% | +61.8% |
| Total return | +800% | +127,041% | +225% | +996% |
| Worst drawdown | −60.6% | −85.1% | −39.4% | −66.3% |
| Volatility | 26.7% | 48.3% | 27.3% | 51.9% |
| Sharpe ratio | 0.45 | 0.89 | 0.87 | 1.11 |
| Time invested | 20% | 100% | ||
| Switches per year | 8.3 | 0 |
Every call, graded
Right on 117 of 180 closed calls (65%). In calls averaged +2.9%; out calls saw NVDA move +9.7% on average.
| Called | Call | Price | Until | NVDA moved | Result |
|---|---|---|---|---|---|
| Aug 3, 2026 | OUT | $206.41 | now | +10.7% | open |
| Jul 29, 2026 | IN | $189.80 | Aug 3, 2026 | +8.8% | ✓ right |
| Jul 8, 2026 | OUT | $203.89 | Jul 29, 2026 | −6.9% | ✓ right |
| Jun 5, 2026 | IN | $204.87 | Jul 8, 2026 | −0.5% | ✕ wrong |
| Apr 6, 2026 | OUT | $177.24 | Jun 5, 2026 | +15.6% | ✕ wrong |
| Mar 26, 2026 | IN | $170.85 | Apr 6, 2026 | +3.7% | ✓ right |
| Feb 6, 2026 | OUT | $184.98 | Mar 26, 2026 | −7.6% | ✓ right |
| Feb 4, 2026 | IN | $173.78 | Feb 6, 2026 | +6.4% | ✓ right |
| Jan 23, 2026 | OUT | $187.23 | Feb 4, 2026 | −7.2% | ✓ right |
| Jan 20, 2026 | IN | $177.65 | Jan 23, 2026 | +5.4% | ✓ right |
| Dec 19, 2025 | OUT | $180.57 | Jan 20, 2026 | −1.6% | ✓ right |
| Dec 17, 2025 | IN | $170.54 | Dec 19, 2025 | +5.9% | ✓ right |
An in call is right when the price rose before the next call; an out call is right when it fell. The latest 12 of 181 calls shown.
How the play works
- In on a close below the lower band (20-day average − 2 standard deviations).
- Out on a close above the 20-day average.
This runs against Bollinger's advice that tags of the bands are not signals. It is shown because so many traders use it.
The case against
Valeriy Zakamulin tested moving-average timing on up to 155 years of U.S. stock data and found no statistically significant edge in the second half of that record. The rules earned their keep in a few severe bear markets and lagged buy and hold for long stretches in between.