Who this is for — Anyone who wants to link size to measurable edge (win rate + payoff). Theoretical reference — not a direct operational command.
The Kelly criterion estimates the optimal capital fraction to risk for maximizing long-run logarithmic growth. Simplified formula: f = W − (1−W)/R* where W = win rate, R = payoff ratio (avg win / avg loss).
In plain terms — With statistical edge, Kelly says «how much to bet». In practice full Kelly is almost always too aggressive.
Correct use (risk module)
| Step | Rule |
|---|---|
| Estimate | W and R on robust sample (≥100 trades) |
| Calculate | f* as baseline, sensitivity on W±2% |
| Constraints | respect daily/weekly stops and max % capital |
| Output | upper ceiling — not default size |
High uncertainty → fractional Kelly.
Real limits
- Assumes stationarity — markets aren't
- Small-sample estimates → unstable f*
- Full Kelly → deep drawdowns and unsustainable psychology
Typical mistake — Applying full Kelly on optimistic backtest — live = different parameters, different account.
Example — W=52%, R=1.6 → f* ≈ 23%. Operational limit 2%/trade → use f* as benchmark, real size 1.5% (~1/15 Kelly).
Summary card
- Input: win rate, payoff ratio, sample size.
- Alert: sample < 50 trades → don't use Kelly.
- Next: always fraction in live.
Gold path — Risk control module. Index: Gold path.