Quantitative methods · Intermediate · Daily
Kelly Criterion
How it works, and when it doesn't.
Raw RSI divergence has ~52% accuracy. XGBoost improves this to ~63% by conditioning on contextual features — a meaningful edge when applied consistently over many trades.
Why it works
RSI divergence (price makes new high/low while RSI does not) is one of the most reliable reversal signals in technical analysis. This strategy uses XGBoost to classify which divergences are likely to follow through, filtering out false signals using 15 contextual features including volume, regime, and volatility.
The signal logic
- Compute RSI(14) and identify price vs RSI divergences
- Bullish divergence: price lower low + RSI higher low → Long candidate
- Bearish divergence: price higher high + RSI lower high → Short candidate
- XGBoost filters using 15 features (volume, regime, BB position, etc.)
- Only trade when XGBoost classifier confidence > 65%
When it fails
Strong trending markets override divergences. An RSI divergence in a powerful uptrend can persist for weeks before resolving.
- Best market
- Range-bound high-volatility markets with clear momentum extremes
- Worst market
- Parabolic trends where momentum extremes persist indefinitely
Backtest
This is a research tool rather than a directional strategy; it has no live signal of its own.
Kelly Criterion, live.
Re-scored after every close.
The platform runs this and 32 other strategies against the live tape each day — with the evidence behind every number.