Cause of death
Passed its test, no edge
One idea cleared the pre-registered test it was written for: a 200-day trend filter with a 10% volatility target that we call the drawdown brake. That test was against the plain 200-day rule, not against holding the market on the same days. We do not call that an edge.
- It passed vs the plain 200-day overlay, including 6 of 7 new market groups, mostly by cutting the worst drops.
- It did not beat simply holding SPY on the same invested days. Sharpe 1.00 vs 1.04 in 2017–2021, and −0.03 vs 0.02 in 2022–2023. There's no evidence it picked better days. The smaller drops come from sitting in cash.
- It earned far less than fully invested SPY: about 8% a year against 18.28% in 2017–2021.
- It failed the batch-level overfitting check. It was confirmed under its own pre-registered rule, not by clearing the deflated-Sharpe bar across the batch it came from.
- It missed one of seven groups (country ETFs, by 0.008 of Sharpe) and we left that as a miss.
- It is paper only. No real money has used it. Costs and fills are simulated.
- The samples are friendly or short. Crypto is a few boom-heavy years. The 2022–2023 window is about 480 trading sessions.
The official 2017–2021 run, against holding SPY on the same days and against staying fully invested. Sharpe and yearly return credit idle cash at the 3-month T-bill rate.
| 2017–2021 | Sharpe | Yearly return | Worst drop | Time invested |
|---|---|---|---|---|
| Drawdown brake | 1.00 | 8.05% | 11.3% | 57% |
| SPY, same invested days | 1.04 | 10.0% | 15.3% | 57% |
| SPY, fully invested | 0.92 | 18.28% | 33.7% | 100% |
Read the full write-up, with the official-run tables, the charts, and the market-by-market results.
If we ever trade it with real money, we'll say so here. Until then, this is a paper result that passed one test and failed a fairer comparison.