Home / Example backtest
Crypto trend-following: 100-day breakout, vol-targeted
A real, unedited run from the pipeline — including the parts that didn't look perfect. This is the whole point of the tool: it reports what actually happened, then flags where it's fragile.
One command — /signalchain Time-series trend-following on BTC-USD, ETH-USD, SOL-USD: long when price above 100-day breakout, flat otherwise, volatility-targeted sizing. — and the pipeline researched the idea, set benchmarks, wrote and ran the VectorBT backtest, tuned it against those benchmarks, and graded itself. Here's exactly what it produced.
This page is a trimmed summary. The complete run — the full research trail, the runnable backtest code, the written report, the interactive chart, and the sweep files — is included with SignalChain.
The strategy
A long/flat time-series-momentum book on crypto majors. An asset is bought when its daily close breaks above the prior 100-day high (Donchian upper channel) and sold to flat when it breaks below the prior 50-day low. Every active position is sized by a 20-day realized-volatility target, so each sleeve carries a similar risk budget regardless of its own volatility. No leverage (per-asset and gross weights capped at 100%), 5 bps fees + 10 bps slippage per side, BTC buy-and-hold as the benchmark.
Results vs. the research benchmarks
| Metric | Value | Benchmark band | In band? |
|---|---|---|---|
| Sharpe | 1.04 | 0.4 – 1.4 | ✓ yes |
| CAGR | 32.2% | 5 – 50% | ✓ yes |
| Max drawdown | 44.7% | 15 – 45% | ✓ yes (near ceiling) |
| Total return | 990.5% | 40 – 600% | ▲ over |
| Trades | 1,685 | > 20 | ✓ yes |
| Win rate | 76.9% | — | — |
| Profit factor | 1.72 | — | — |
All three risk-adjusted gates (Sharpe, CAGR, Max DD) land in band, so the verdict is PASS. The report is candid about the two soft spots: max drawdown sits right against the 45% ceiling, and total return overshoots its band — a compounding artifact of the 2018–2026 crypto drift (the annualized 32.2% CAGR is the horizon-invariant read, and it's mid-band). The book trails BTC buy-and-hold on raw return (990% vs 2795%) exactly as a vol-scaled, long/flat trend book should: it trades upside capture for cutting spot-crypto's 80%+ drawdowns down to ~45%.
The interactive equity curve
This is the actual chart the tool wrote — strategy vs. BTC buy-and-hold, with the drawdown panel below. Hover and zoom.
The honest part: what the robustness sweep found
A single passing backtest proves nothing on its own — so the tool swept the two key parameters (breakout_lookback × target_vol) and reported the result without flattering it. Two findings:
1. The signal is robust. The Sharpe surface is smooth and monotone — no cliffs, no lone spikes — and the 100–150-day lookback dominates on both Sharpe and drawdown. The chosen window isn't a lucky, cherry-picked cell.
| lookback \ target_vol | 0.30 | 0.50 | 0.70 |
|---|---|---|---|
| 20 | 0.99 / 70.9% | 0.93 / 75.1% | 0.87 / 76.1% |
| 55 | 0.91 / 66.2% | 0.81 / 73.1% | 0.74 / 75.8% |
| 100 | 1.07 / 53.4% | 1.00 / 63.0% | 0.95 / 65.0% |
| 150 | 1.07 / 50.4% | 0.98 / 58.8% | 0.91 / 64.0% |
each cell: Sharpe / max drawdown
target_vol of 0.20, which sits below the swept range. So the tool's own conclusion was blunt: the drawdown pass is fragile, the recommended configs still breach the ceiling by ~5 points, and a denser, lower sweep is needed before trusting the drawdown. It told on itself — which is exactly what you want a research tool to do.
Caveats (straight from the report)
Costs and turnover are the biggest unknown — daily vol re-scaling churns the book (1,685 trades), and a no-trade band wasn't implemented, so real fee drag could be higher. The 100-day window is deliberately conservative (above the 10–30-day zone the literature favors). It's long/flat only, so it earns nothing in sustained downtrends beyond sitting in cash. SOL's history is short and bull-skewed. Every number here is a hypothetical, in-sample backtest — an optimistic upper bound, not a forward expectation.
Run your own — and get the same honest report
SignalChain does this for any strategy you can describe: research, benchmarks, code, backtest, PASS/FAIL verdict, and a robustness sweep — and you keep every file it generates. All inside Claude Code, no API key required.
Get SignalChain — $49 →Not financial advice. This page describes a hypothetical, backtested research result, not a recommendation. Backtested performance is not indicative of future results, and there is no guarantee of profitability. SignalChain is a research and educational tool; you are solely responsible for any decisions you make.