Trend Ensemble (Risk-Managed)

After exhaustively proving that directional prediction has no tradeable edge (17 strategy families, beta-stripped, NO-GO every time), one approach survived — and it is not prediction. The Trend Ensemble does not try to be right about the next move. It controls how you hold crypto so that you survive crashes and compound through cycles.

We are blunt about what this is: it is not alpha. It does not beat the market in a bull run — it gives up some upside. What it does is shrink crash drawdowns. On Bitcoin alone (daily, 2019-09 → 2025-12) the ensemble cut the worst drawdown from −76.7% to −38.5%. In a later, shorter window (2021–2026) the cut was smaller — −76% held vs −58% for the ensemble on its own — and a 50/50 trend-and-cash portfolio reached −32%, with the cash half doing more than half of that work. That shrinkage is the difference between blowing up and compounding. The measured drawdown was smaller in these windows; leverage still carries liquidation risk.

1. Ensemble trend signal. Average the trend direction across many lookbacks (20, 40, 60, 80, 120, 160, 200 days) instead of betting on one. No single parameter is over-fit — the ensemble is the structure.
2. Exposure dial. When the ensemble is bullish, hold; when it turns down, scale into cash. A floor (e.g. 50% always invested) keeps you from missing the recovery.
3. Multi-asset. Run it across BTC, ETH, SOL equal-weight to diversify the crash protection.
4. Cash buffer. Holding ~40–50% cash damps volatility and, in the leverage test, keeps the book from being liquidated at 3× where buy-and-hold is. Read “survives” literally: as of 2026-06-23, the 40%-cash book at 3× returned CAGR +43% at a −84% maximum drawdown, and the finer-grid companion run (BTC only, 2019–2025) puts the ensemble’s own 3× drawdown at −88.9%. Not liquidated is not the same as tolerable. That test is a weekly grid and does not model intraday wick liquidation, so it is an upper bound on survival — and the portfolio bot this site offers is 1× long only, not a 3× product.

The edge is crash avoidance, not forecasting. Trend-following has no idea where price is going; it simply reduces exposure after a downtrend establishes and restores it after an uptrend does. You are always a little late — and that lateness is the price of never riding a crash to zero.

In every window we tested, the maximum drawdown came down — but by how much depends on the window: roughly halved on Bitcoin 2019-09 → 2025-12 (−76.7% → −38.5%) and in a 2017–2026 four-coin test that includes BNB (−81% → −46% with no exposure floor; with a 50% floor, −65%), but only about a quarter in 2021–2026 (−76% → −58%). The walk-forward check passed 2 of 4 folds on Calmar. The risk-adjusted return was better too, but its size depends on the period: in the 2021–2026 window (the one with options data, where Bitcoin buy-and-hold was unusually weak at Sharpe 0.28) the 50/50 trend-and-cash portfolio reached Sharpe ~0.82; in a strong bull cycle that advantage is much smaller. The honest cost: in a roaring bull market it lags buy-and-hold. The claim we stand behind is survival, not a return multiple.

Limit: this is BTC/ETH/SOL with enough history to include a bear market. It is risk management, not a money machine. Pair it with DCA for accumulation and — when validated — the volatility premium as an uncorrelated overlay.

This is the honest answer to “what can I actually do?” — not a prediction, a discipline. Backtest the trend logic on the simulator with real fees, and look at the drawdown, not just the return. Don’t believe us. Verify.