Who this is for — Anyone using quantitative models but wanting sustainable drawdowns. Operational version of Kelly in non-stationary markets.
Fractional Kelly applies only part of theoretical Kelly — typically ½, ⅓, or ¼ of f*. Slower growth, higher probability of staying in the game when real parameters diverge from estimates.
In plain terms — Take Kelly size and use a fraction. Less aggressive, more survival.
Practical rules
| Fraction | When |
|---|---|
| ½ Kelly | Solid data, medium DD tolerance |
| ⅓ Kelly | Prudent live default |
| ¼ Kelly | High uncertainty, leverage, unstable markets |
- Pair with hard daily/weekly limits
- Recalibrate quarterly on out-of-sample W and R
- Fraction fixed in playbook — not changed emotionally
Governance
- Document assumptions (W, R, sample) at each recalc
- If DD exceeds forecast despite fraction → ↓ fraction or pause
- Pair with drawdown control
Typical mistake — Switching to full Kelly after positive streak — parameters didn't improve, only luck.
Example — Theoretical Kelly 6%/trade. ⅓ Kelly → 2% max. Weak out-of-sample → 1.5% effective in playbook.
Summary card
- Live default: ⅓ Kelly or less.
- Review: quarterly + post drawdown.
- Cap: never exceed plan % limit regardless of f*.
Gold path — Risk control module. Index: Gold path.
Sources
- E. Busseti, E. Ryu, and S. Boyd, Risk-Constrained Kelly Gambling — formalizes the growth versus drawdown-risk trade-off and compares the constraint with fractional Kelly.
- MacLean, Thorp, and Ziemba, Long-term capital growth: the good and bad properties of the Kelly and fractional Kelly capital growth criteria — describes Kelly's growth benefit and the highly volatile paths of full Kelly.