■ SMA10 ■ SMA50 ■ SMA100 ■ SMA200 — line weight tracks measured usefulness, not convention.
| Average | Level | Price vs | Position | Measured as a signal (98 years) |
|---|---|---|---|---|
| SMA10 | 772.19 | +0.1% | above | 0.28 Sharpe, 39 trades/yr, −65% maxDD — noise: whipsaws, deeper drawdown |
| SMA50 | 768.34 | +0.6% | above | 0.33 Sharpe, 16 trades/yr, −50% maxDD — weak |
| SMA100 | 758.09 | +2.0% | above | 0.35 Sharpe, 10 trades/yr, −49% maxDD — useful |
| SMA200 | 726.29 | +6.4% | above | 0.37 Sharpe, 6 trades/yr, −49% maxDD — best of the family |
Longer is better, monotonically, and the fast averages are worse than nothing. Across 98 years price above SMA200 returned 7.90% at 6 trades a year; price above SMA10 returned 6.65% at 39 trades a year with a deeper drawdown (−65% vs −49%). The best variant found was price above a rising SMA200 — highest Sharpe (0.39) and the shallowest drawdown (−29.5%).
All of them lose to buy-and-hold since 2010. What holds in every era is drawdown reduction, not return.
For premium selling these filters add no return, but the full bullish stack cut max drawdown from −20.3% to −8.5% at essentially unchanged return (6.71% vs 6.73%). Risk reduction, not alpha. Method in research/SMA.md.
EMA vs SMA is a non-question. The paired difference across 20 era/window pairs was +0.015 Sharpe (t = 0.88, interval spanning zero), against a 0.097 spread between windows. SMA200 and EMA200 agree on 97.2% of months. Pick a window, not a weighting.