TidelineStudy · generated 2026-10-11 · Rule v1
What Rule v1 did on crypto
A pre-registered replay of the rule exactly as it runs here — Supertrend(10, 3) on completed daily candles, long or cash — on a basket of the 20 most-traded crypto assets, rebuilt every quarter using only what was known at the time. 2019-01-01 to 2026-10-09, 7.78 years, 126 assets considered. Costs 0.30% per side. Every benchmark, cost level and weakness is shown; the wording below was fixed before the numbers existed.
In past data the rule had smaller falls; its return was similar overall and lower in recent years.
Max drawdown -46.1% against -85.1% for holding the same basket — 46% shallower. Compound return 42.5% a year against 38.5% (1.10× the basket's). This describes the past. It is not a forecast and not advice.
Headline numbers
| Rule v1 | Hold the basket | Hold Bitcoin | Hold less, always | |
|---|---|---|---|---|
| Compound return per year | 42.5% | 38.5% | 48.4% | 38.3% |
| Worst fall from a peak | -46.1% | -85.1% | -76.6% | -83.1% |
| Ulcer index (pain of drawdowns) | 26.69 | 59.01 | 37.28 | 56.38 |
| Longest time under water | 922 days | 1980 days | 846 days | 1980 days |
| Sharpe | 1.04 | 0.82 | 0.96 | 0.82 |
| Time invested | 95% | 100% | 100% | 100% |
Random timing: 1000 runs that hold the same stretches in a random order reach a worst fall of -47.4% at the 90th percentile (median -54.6%) and an Ulcer index of 23.22 at the best tenth (median 29.69). The rule does not beat that bar on both measures.
Growth of 1 (log scale)
Drawdown from the last peak
What it felt like
Worst six months against the basket: from 2020-11-10, the rule returned 317.8% while holding returned 775.6%.
Round-trip results
The big falls
Every Bitcoin drawdown of 30 % or more in the period: how much of the basket's fall the rule avoided, and how much of the rebound it missed.
| Peak → trough | Bitcoin fell | Basket fell | Rule fell | Fall avoided | Rebound missed |
|---|---|---|---|---|---|
| 2019-06-26 → 2020-03-12 | 62% | -54% | -3% | 95% | 72% |
| 2021-04-13 → 2021-07-20 | 53% | -42% | -25% | 41% | 49% |
| 2021-11-08 → 2022-11-21 | 77% | -81% | -41% | 49% | 46% |
| 2025-10-06 → 2026-06-30 | 53% | -50% | -16% | 69% | 38% |
Year by year
| Year | Rule v1 | Hold the basket |
|---|---|---|
| 2019 | 21.9% | 20.4% |
| 2020 | 170.9% | 241.2% |
| 2021 | 239.3% | 348.8% |
| 2022 | -34.0% | -75.7% |
| 2023 | 61.2% | 98.7% |
| 2024 | -1.3% | 35.2% |
| 2025 | -1.1% | -18.8% |
| 2026 | 35.2% | 29.0% |
Does it hold up?
- Costs. At zero cost 46.4% a year; at 0.30% 42.5%; at 0.60% 38.7% (worst fall -48.0%). Executing a day late: 41.5%, worst fall -49.1%.
- Halves of the period. first half: worst fall -43.6% vs -84.6% holding, return 69.2% vs 49.1%.second half: worst fall -45.9% vs -58.0% holding, return 20.1% vs 28.7%.
- Bitcoin alone. Worst fall -53.9% vs -76.6%; return 29.0% vs 48.4% a year.
- Altcoins without Bitcoin. Worst fall -46.7% vs -85.8%; return 42.8% vs 37.3%.
- Every eligible asset, not just the top 20. Worst fall -52.1% vs -84.0%; return 73.8% vs 55.0%.
- Without the largest fall (2021-11-08 to 2024-03-04): worst fall -45.9% vs -65.7% — still 30% shallower.
- Resampling the calendar (2000 draws): the drawdown advantage is 27% with a 95 % range of 7% to 45%; the return difference is 1% with a range of -58% to 40% — the return edge is not distinguishable from zero, the drawdown edge is.
- Nearby settings. Rule v1's settings (10, 3) are the conventional defaults and were not tuned on this data; among neighbouring settings it is not an isolated peak. The full grid is in the appendix.
Does the live record match?
The study's code is the production code. Replaying the production venues' own series reproduces 77,922 ledger rows with 0 mismatches, and every live row since launch (405) and every live flip (8 of 8, 1 of them late and shown separately). The first candles of each series depend on the window the venue serves and are excluded by construction (18,356 rows, listed in the appendix). The acceptance test passes.
Limitations
Crypto history is short and contains few full cycles. The assets move together, so the effective sample is far smaller than the asset count. Assets that delisted are not in the data, which flatters every line including the benchmarks. Prices are Yahoo's composite, not a venue. Results describe the past; real execution, fees and discipline will differ. 14 assets were excluded (too little history or unreliable data; listed in the appendix).
Appendix
Full results (JSON) · Technical report (Markdown) · snapshot fd1d67415b8f8408… · code and pre-registration.