Half Kelly applies the Kelly formula and multiplies the result by 0.5. Same inputs, half the position size.
Shape
"position_sizing": {
"method": "half_kelly",
"estimated_win_rate": 0.60,
"estimated_rr_ratio": 1.2,
"max_position_pct": 15
}Why half
Real-world W and R are noisy. Under noise, full Kelly has very high drawdowns (50%+ is common). Half Kelly captures roughly 75% of the long-run growth with about 25% of the drawdown variance.
When to pick this
- You have meaningful historical data but not perfect knowledge of the strategy's edge.
- You want positive expectancy from Kelly-style sizing without the full Kelly drawdown profile.
Pitfalls
Same as full Kelly - overestimating inputs is more dangerous than underestimating.