V8 MOMENTUM — the same book, one more condition on when to hold it
V8 MOMENTUM trades the same basket and the same ranking as v8LONG. It adds a third condition to the gate, and the trade it makes is explicit: +405% instead of +1277% over 2020-10-23 … 2026-07-26, at a worst drawdown of −21.6% instead of −29.4%. Less than half the return, roughly half the pain.
What it does
- When to be in the market. Three conditions, all of them, or the book sits in cash: a breadth-and-momentum reading called heat above 74, bitcoin above its own long-term average, and bitcoin's 30-day return in the top third of its own history to date while it also trades above its 140-day average.
- How often that question is re-asked. Every 7 days, not every day. Re-deciding daily is a different strategy: different entries, different turnover, different result.
- When the third condition is ignored. Never — but it can shut the gate on its own. If the link between the trigger and what followed it is weaker than 0.05 on the history available at that moment, the gate is closed rather than trusted.
- What to buy. While the gate is open, the strongest names by risk-adjusted momentum, equal weight, capped per name, with a hysteresis band so a drifting holding is not churned. Identical to v8LONG.
The gate was open on 24% of the 2103 days in the window. The other 76% the book held cash and earned nothing, which is where the drawdown difference comes from.
Long only, spot, leverage 1.0: the book cannot lose more than the capital in it, and the backtest stops with an assertion if it ever tries to spend money it does not have. There is no short leg and no hedge — when the gate shuts, the position is sold, not offset. Nothing here is protected against a gap: the strategy exits on the next daily open, at whatever that open is.
The nulls
Two controls, both computed on this configuration and this window. The index null holds everything the gate allows, with no selection at all. The placebo keeps the days, the sizes and the costs, and scrambles which coin is which — 200 times, one fixed permutation per run.
| Version | Total | Sharpe |
|---|---|---|
| V8 MOMENTUM | +405% | 1.27 |
| Buy everything the gate lets through | +299% | 1.09 |
| Random coins, median | +249% | 0.95 |
| Random coins, the luckiest 5% | +469% | 1.24 |
The rule claims: тайминг, отбор. Each layer is checked against ITS OWN null — a name shuffle tests selection and says nothing about timing. The rule returns +405%. подтверждено: тайминг; не подтверждено: отбор. A name shuffle beat the rule in 26 of 200 seeds.
| Layer | Control | Control returns | State |
|---|---|---|---|
| тайминг — when to be in the market | hold the basket throughout | +322% | держится |
| отбор — which names to hold | equal weight, everything the gate allows | +248% | держится |
| no top-20 cut | +299% | держится | |
| name shuffle, p95 | +469% | проигрывает контролю |
By market phase: алгоритм фазы «бычка»
The phase is labelled by formula (BTC against its own long moving average and the slope of that average), not by hand. The best phase carries 94.6% of the return. The other rows are printed right here: an algorithm may work in one phase, it may not stay quiet about the rest.
| Market phase | Days | Total | Annualised | sharpe | Drawdown |
|---|---|---|---|---|---|
| бычка | 940 | +362% | +81% | 1.78 | −21.6% |
| медвежка | 588 | 0% | 0% | None | 0.0% |
| аккумуляция | 404 | +9% | +8% | 0.6 | −5.7% |
Against the alternatives
| Total | Annualised | maxDD | Sharpe | Trades | |
|---|---|---|---|---|---|
| V8 MOMENTUM | +405% | 36% | −21.6% | 1.27 | 134 |
| V8 LONG | +1277% | 58% | −29.4% | 1.35 | 288 |
| Hold BTC | +406% | 32% | −76.6% | 0.78 | — |
One window, one basket, one cost model for all three rows: 2020-10-23 … 2026-07-26, 0.30% round trip on spot, adverse slippage on every fill. Holding bitcoin returns less than either version here and costs a −76.6% drawdown to sit through.
| Year | V8 MOMENTUM | V8 LONG |
|---|---|---|
| 2021 | +42.6% | +135.9% |
| 2022 | 0.0% | 0.0% |
| 2023 | +129.4% | +151.5% |
| 2024 | +48.0% | +78.3% |
| 2025 | +4.3% | +16.9% |
| 2026 | 0.0% | 0.0% |
The flat years — 2022, 2026 — are cash, not losses: the gate never opened in them at all.
Which coins it traded
The universe allowed 34 names. The book touched 30 of them across 134 trades. The twelve that mattered most:
| Coin | Trades | Avg | Win rate | Total |
|---|---|---|---|---|
| SOL | 12 | +20.1% | 75% | +241% |
| INJ | 8 | +18.1% | 75% | +144% |
| AVAX | 5 | +23.8% | 60% | +119% |
| ADA | 11 | +7.4% | 55% | +82% |
| BTC | 12 | +5.6% | 50% | +67% |
| TRX | 8 | +8.0% | 75% | +64% |
| NEAR | 5 | +10.9% | 40% | +54% |
| SUI | 4 | +13.1% | 75% | +52% |
| BNB | 6 | +8.1% | 50% | +48% |
| XRP | 3 | +15.4% | 67% | +46% |
| ATOM | 3 | +9.8% | 100% | +29% |
| OP | 1 | +27.9% | 100% | +28% |
Never traded: ARB, FIL, TIA, TON.
Compounding the trades one after another and dropping the twenty best leaves 17% of that series. Read the ratio, not the level: that series is not the strategy's equity curve — the book holds several names at once. What it means for a live bot is that a skipped fill is not a rounding error.
Timing
What is enforced in the engine, and can be read in it: the percentile is computed on values strictly before the day being judged, the moving average on closes before that day, the correlation on returns that had all finished before it, the trigger is taken at a bar close and the fill at the next open. Win rate is 57%, below the threshold that would block publication pending an audit.
What has not been done: no separate archive-truncation audit has been run on this configuration. The one that returned clean on v8LONG — the raw price and funding archives cut at two past dates and everything replayed — covers the engine and the ranking they share, not the gate condition added here. A clean audit proves the absence of look-ahead and nothing else anyway; it is not evidence that a rule works.
What this leaves
The defensible sentence is narrow: on 2020-10-23 … 2026-07-26, adding a momentum condition to the regime gate cut the worst drawdown from −29.4% to −21.6% and gave up +872% of return to do it. Which of the two to run is a question about drawdown tolerance, not about which is better. The programme's own warning still applies to both: the universe holds coins still listed today, so every absolute number here is an upper bound.
The fine print
- Только дожившие. The archive holds 34 coins listed today. Coins that listed and died are absent.
- Costs. 0.30% round trip on spot, adverse slippage on every fill.
- Effective sample. 134 trades is not 134 independent observations: the gate was open on 24% of days, in a small number of episodes.
- Multiple testing. Both components of the added condition were measured on the desk's shared factor test before this strategy existed; the weaker entry threshold was tested and discarded there. That budget accumulates across sessions rather than resetting.
- Machine-readable. The rules ship as a generated spec file beside the strategy, written from the same constants the backtest imports, so a bot author never re-reads the engine by hand.