Kelly Criterion in Trading: Practical Use and Hard Limits

The Kelly formula f* = (bp - q) / b gives the fraction of capital to bet for maximum long-term growth, where b is net odds (win/loss ratio), p is win probability, and q = 1 - p. In trading terms, b = average win / average loss. If your average win is $300, average loss $150, then b = 2. With a 55% win rate: f* = (2*0.55 - 0.45) / 2 = 0.325. Kelly says risk 32.5% of equity per trade — which is insane.

The math is correct; the inputs are the problem. Your edge and win rate are estimates from a finite sample. A 55% rate measured over 100 trades carries a 95% confidence interval of roughly ±10%. Feeding an overestimated edge into full Kelly produces catastrophic drawdowns — at full Kelly, a 50% drawdown is statistically routine.

Practical application rules

• Use fractional Kelly. Quarter-Kelly (0.25 * f*) is the professional default. The 32.5% above becomes ~8%, still high; most cap absolute risk at 1–2%.
• Re-estimate edge quarterly from at least 30–50 trades per setup type. Never blend setups — a trend system and a mean-reversion system have different b and p.
• Adjust for correlation. Kelly assumes independent bets. Correlated holdings (e.g. long five tech names) violate this — treat the cluster as one bet or scale f* down further.
• Variable losses break the model. Kelly models fixed odds. When stop distance varies, use the geometric mean of historical trade outcomes instead of the closed-form formula.

Fractional Kelly reference

The hard limit

Kelly optimizes terminal wealth, not the path to it. A 40% drawdown may be mathematically "optimal" yet unacceptable operationally and psychologically. Cap position size at 2% risk regardless of what Kelly computes. If Kelly outputs a number above 2%, your edge estimate is probably wrong — trust the cap, not the formula.