Funding Rate Extremes: The Only Signal That Survived
After 40 hypotheses, one pattern held: LONG on extreme negative funding shows 77.6% win rate. SHORT on positive funding? 47.3% WR. Direction asymmetry is real. SOL >> ETH >> BTC for this signal.
AI agents that backtest ideas in plain English.
Surface alpha from noise. Ship signals, not spreadsheets.
> backtest momentum strategy on top 20 perps, 30d lookback
Running grid search... 200 parameter combinations
Train period: 2023-02 to 2025-03 | Test: 2025-03 to 2026-02
Best params: LB=30d, Hold=7d, N=5 coins
OOS Sharpe: 0.52 | Max DD: -69% | Win Rate: 47%
>
Get early access
Academic research shows exchanges systematically misreport open interest and delay liquidation data. We tested 40 hypotheses and found no edge. Here's why that makes sense.
After 40 hypotheses, one pattern held: LONG on extreme negative funding shows 77.6% win rate. SHORT on positive funding? 47.3% WR. Direction asymmetry is real. SOL >> ETH >> BTC for this signal.
Built a complete liquidation pipeline across Binance, HyperLiquid, OKX, Bybit. Tested cascade detection, z-score velocity, exhaustion patterns. 275+ parameter sets, 0 profitable configurations.
An AI research co-pilot for crypto. Ask it to backtest any idea in plain English. It writes the code, runs the analysis, and tells you if there's alpha or not.