Quantitative futures trader running a multi-timeframe breakout book on Hyperliquid perpetuals. Seven liquid instruments, three horizons each, every position sized by fixed-fractional risk and executed by code. No discretion. No revenge trading. No guessing.
Built on Hyperliquid L1 perpetuals. Sub-second settlement, deep orderbooks, no custody risk. The engine runs 24/7, scanning seven liquid instruments across three horizons simultaneously.
Each timeframe carries a different job. The 5m layer catches fast expansion. The 15m layer is the workhorse, balancing signal frequency with signal quality. The 1h layer trades only high-conviction structure. Position size scales with timeframe: smaller on noise, larger on conviction.
Stops are volatility-adaptive. A fixed percentage stop treats a 0.3% move the same on every asset, which is wrong. Every stop sits at ATR(14) multiplied by a per-instrument factor, so risk stays constant across instruments with wildly different volatility profiles.
Targets are R-multiples, not price guesses. Each setup defines its own reward ratio, validated against measured hit rate. A 1:4 target at 31% win rate beats a 1:1 target at 60%. The math is not close.
Exits stay untouched. I tested breakeven locks, partial take-profits, and trailing stops across 618 trades. All three destroyed performance. A trend-following book with 1:4 targets survives on a handful of large winners. Cutting them short removes the entire edge.
Correlation is treated as position sizing. Instruments that move together are capped together. Five correlated positions is one position at five times the size.
Results below are 90-day performance across three horizons, measured on five-minute, fifteen-minute, and hourly execution.
| Timeframe | Trades | Win Rate | Profit Factor | Total R | Max DD |
|---|---|---|---|---|---|
| 5m | 604 | 23.2% | 1.07 | +38.2 | -25.2 |
| 15m | 695 | 29.8% | 1.34 | +173.0 | -21.5 |
| 1h | 208 | 34.1% | 1.51 | +70.0 | -12.2 |
Robustness stack: split-sample validation, 24-configuration parameter sensitivity (all passed), and an execution model that accounts for next-bar entry plus realistic slippage. The edge survived all three.
Everything runs on a private VPS. No third-party signal services, no black-box subscriptions. Data straight from the exchange, execution via signed API calls, monitoring on a custom terminal.