There is a formula that tells you the mathematically optimal fraction of your account to risk on each trade. It is also a formula that, followed literally, will hand you drawdowns so violent you will quit before it ever pays off.
That formula is the Kelly criterion, and understanding both halves of that sentence is what separates traders who use it from traders who get hurt by it.
The Kelly formula and what each term means
Kelly answers one question: given a repeatable edge, what bet size maximizes the long-run growth rate of your capital? The classic form is:
f* = W − (1 − W) / R
- f* is the fraction of your bankroll to risk on the next bet.
- W is your probability of winning (your win rate as a decimal).
- R is your payoff ratio — average win size divided by average loss size, in money or in R-multiples.
The intuition is clean. The first term rewards you for winning often. The second term penalizes you for losing, scaled down when your winners are larger than your losers. When the two balance out to zero or below, Kelly tells you the honest truth: there is no edge here, so bet nothing.
Turning win rate and payoff ratio into a Kelly %
Say you win 45% of the time and your average winner is twice your average loser (R = 2). Plug it in:
f* = 0.45 − (0.55 / 2) = 0.45 − 0.275 = 0.175
Kelly says risk 17.5% of your account per trade. That number should make you flinch, and it should. It is a growth-maximizing answer to a math problem, not a survival plan for a real account with a drawdown limit.
Two inputs drive everything, and both come from your own trade history rather than a hunch. If you have not measured your true win rate and payoff, an expectancy calculator will turn a batch of closed trades into the W and R that Kelly actually needs. Garbage in, catastrophic bet size out.
Why full Kelly produces stomach-churning drawdowns
Full Kelly maximizes growth in theory but ignores the path you take to get there. Along that path, full Kelly regularly produces drawdowns on the order of 50% or more. The math of growth-optimality does not care that a 50% drawdown feels like an emergency, or that on a prop-firm account it is a hard failure long before you reach it.
Two things make full Kelly unlivable in practice:
- The volatility is enormous. Growth-optimal and comfort-optimal are different objectives. Kelly targets the former and pays for it in swing size.
- It assumes your edge is exact. Kelly presumes you know W and R precisely. You do not — you have a noisy estimate from a limited sample, and that gap is where the danger lives.
Half- and quarter-Kelly: the practical compromise
The standard fix is to bet a fraction of the Kelly number — most commonly half, sometimes a quarter. The trade-off is remarkably favorable:
| Fraction | Approx. growth retained | Approx. volatility |
|---|---|---|
| Full Kelly | 100% | Highest |
| Half Kelly | ~75% | Roughly halved |
| Quarter Kelly | ~44% | Roughly quartered |
Half-Kelly keeps about three-quarters of the theoretical growth rate while cutting the volatility of your equity curve roughly in half. You give up a little compounding and buy a lot of survivability. For a trader — especially one operating inside someone else’s rules — that is close to a free lunch.
How estimation error makes over-betting dangerous
Here is the part the formula hides. Kelly is exquisitely sensitive to the accuracy of your inputs, and the error is asymmetric: overestimating your edge and over-betting hurts you far more than underestimating and under-betting.
If your real win rate is 45% but a lucky 60-trade sample flattered you into believing it was 52%, full Kelly on the wrong number can push you past the growth-optimal point into the region where your long-run growth actually turns negative. You would be betting more aggressively than a system with no edge at all would justify. Because your measured W and R are always estimates from a finite sample, you should treat full Kelly as if it were computed on numbers that are optimistically wrong — and fractional Kelly is exactly the margin of safety that assumption demands.
This is also why a live, honest read on your edge matters more than a one-time calculation. Shibiki’s edge health wraps every strategy’s win rate in a Wilson confidence interval, so you can see not just your point estimate but how much sample you actually have behind it. When the interval is wide, your Kelly number is a guess wearing a lab coat — and you size down accordingly.
Using Kelly as a ceiling, not a target
The healthiest way to hold Kelly is to invert its purpose. Do not treat it as the number to hit. Treat it as the number you must never exceed.
- Kelly is a ceiling. If the formula says 17.5%, that is the line past which you are objectively over-betting, no matter how good the setup feels.
- Your working risk lives well below it. Most durable traders operate at a small fixed percentage that happens to sit at a quarter of Kelly or less — which is also why the conventional 1–2% rules survive contact with reality.
- Re-derive it as your data grows. A 40-trade edge and a 400-trade edge deserve different confidence, and therefore different fractions.
Kelly is not a trade signal. It is a governor on your ambition — a way to know, precisely, the point at which confidence becomes recklessness.
Related: Trading expectancy · Position size calculator · Expectancy calculator